YITP-99-67

arXiv:hep-th/9912205 21 Dec 1999

FIELDS
Warren Siegel
C. N. Yang Institute for Theoretical Physics
State University of New York at Stony Brook
Stony Brook, New York 11794-3840 USA

mailto:[email protected]
http://insti.physics.sunysb.edu/˜siegel/plan.html
 i

CONTENTS
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv Some field theory texts . . . . . . . . . xviii
.................. ..................
. . . . . . . . . . . . . . . . . . PART ONE: SYMMETRY . . . . . . . . . . . . . . . . . .
I. Global III. Local
A. Coordinates A. Actions
1. Nonrelativity . . . . . . . . . . . . . . 3 1. General . . . . . . . . . . . . . . . . . 115
2. Fermions . . . . . . . . . . . . . . . . . . 7 2. Fermions . . . . . . . . . . . . . . . . 119
3. Lie algebra . . . . . . . . . . . . . . . 11 3. Fields . . . . . . . . . . . . . . . . . . . 120
4. Relativity . . . . . . . . . . . . . . . . 15 4. Relativity . . . . . . . . . . . . . . . 122
5. Discrete: C, P, T . . . . . . . . . 20 5. Constrained systems . . . . 128
6. Conformal . . . . . . . . . . . . . . . 23 B. Particles
B. Indices 1. Free . . . . . . . . . . . . . . . . . . . . 132
1. Matrices . . . . . . . . . . . . . . . . . 28 2. Gauges . . . . . . . . . . . . . . . . . 136
2. Representations . . . . . . . . . . 30 3. Coupling . . . . . . . . . . . . . . . . 137
3. Determinants . . . . . . . . . . . . 35 4. Conservation . . . . . . . . . . . . 138
4. Classical groups . . . . . . . . . . 38 5. Pair creation . . . . . . . . . . . . 141
5. Tensor notation . . . . . . . . . . 40 C. Yang-Mills
C. Representations 1. Nonabelian. . . . . . . . . . . . . .144
1. More coordinates . . . . . . . . . 45 2. Lightcone . . . . . . . . . . . . . . . 148
2. Coordinate tensors . . . . . . . 47 3. Plane waves . . . . . . . . . . . . . 152
3. Young tableaux . . . . . . . . . . 51 4. Self-duality . . . . . . . . . . . . . 153
4. Color and flavor . . . . . . . . . . 53 5. Twistors . . . . . . . . . . . . . . . . 156
5. Covering groups . . . . . . . . . . 58 6. Instantons . . . . . . . . . . . . . . 159
7. ADHM . . . . . . . . . . . . . . . . . 163
II. Spin 8. Monopoles . . . . . . . . . . . . . . 165
A. Two components
1. 3-vectors . . . . . . . . . . . . . . . . . 61 IV. Mixed
2. Rotations . . . . . . . . . . . . . . . . 64 A. Hidden symmetry
3. Spinors . . . . . . . . . . . . . . . . . . 66 1. Spontaneous breakdown . 171
4. Indices . . . . . . . . . . . . . . . . . . . 67 2. Sigma models . . . . . . . . . . . 173
5. Lorentz . . . . . . . . . . . . . . . . . . 70 3. Coset space . . . . . . . . . . . . . 176
6. Dirac . . . . . . . . . . . . . . . . . . . . 76 4. Chiral symmetry . . . . . . . . 177
7. Chirality/duality . . . . . . . . . 78 5. Stückelberg . . . . . . . . . . . . . 180
B. Poincaré 6. Higgs . . . . . . . . . . . . . . . . . . . 182
1. Field equations . . . . . . . . . . . 80 B. Standard model
2. Examples . . . . . . . . . . . . . . . . 83 1. Chromodynamics. . . . . . . . 185
3. Solution. . . . . . . . . . . . . . . . . . 84 2. Electroweak . . . . . . . . . . . . . 189
4. Mass . . . . . . . . . . . . . . . . . . . . . 89 3. Families. . . . . . . . . . . . . . . . . 193
5. Foldy-Wouthuysen . . . . . . . 92 4. Grand Unified Theories. .195
6. Twistors . . . . . . . . . . . . . . . . . 96 C. Supersymmetry
7. Helicity . . . . . . . . . . . . . . . . . . 98 1. Chiral . . . . . . . . . . . . . . . . . . 200
C. Supersymmetry 2. Actions . . . . . . . . . . . . . . . . . 202
1. Algebra . . . . . . . . . . . . . . . . . 103 3. Covariant derivatives . . . . 204
2. Supercoordinates . . . . . . . . 104 4. Prepotential. . . . . . . . . . . . .207
3. Supergroups . . . . . . . . . . . . 106 5. Gauge actions . . . . . . . . . . . 209
4. Superconformal . . . . . . . . . 109 6. Breaking . . . . . . . . . . . . . . . . 211
5. Supertwistors . . . . . . . . . . . 110 7. Extended . . . . . . . . . . . . . . . 214
 ii

.................... ....................
.................... PART TWO: QUANTA ....................
V. Quantization VII. Loops
A. General A. General
1. Path integrals . . . . . . . . . . . 221 1. Dimensional renormaliz’n350
2. Semiclassical expansion . . 225 2. Momentum integration . . 353
3. Propagators . . . . . . . . . . . . . 229 3. Modified subtractions . . . 357
4. S-matrices . . . . . . . . . . . . . . 231 4. Optical theorem . . . . . . . . . 361
5. Wick rotation . . . . . . . . . . . 235 5. Power counting. . . . . . . . . .363
B. Propagators 6. Infrared divergences . . . . . 367
1. Particles . . . . . . . . . . . . . . . . 239 B. Examples
2. Properties. . . . . . . . . . . . . . .242 1. Tadpoles . . . . . . . . . . . . . . . . 371
3. Generalizations. . . . . . . . . .245 2. Effective potential . . . . . . . 374
4. Wick rotation . . . . . . . . . . . 248 3. Dimensional transmut’n . 377
C. S-matrix 4. Massless propagators . . . . 378
1. Path integrals . . . . . . . . . . . 253 5. Massive propagators. . . . . 381
2. Graphs . . . . . . . . . . . . . . . . . 257 6. Renormalization group . . 385
3. Semiclassical expansion . . 262 7. Overlapping divergences . 388
4. Feynman rules . . . . . . . . . . 266 C. Resummation
5. Semiclassical unitarity . . . 272 1. Improved perturbation . . 395
6. Cutting rules . . . . . . . . . . . . 274 2. Renormalons . . . . . . . . . . . . 400
7. Cross sections . . . . . . . . . . . 277 3. Borel . . . . . . . . . . . . . . . . . . . 403
8. Singularities. . . . . . . . . . . . .280 4. 1/N expansion . . . . . . . . . . 406
9. Group theory . . . . . . . . . . . 282
VIII. Gauge loops
VI. Quantum gauge theory A. Propagators
A. Becchi-Rouet-Stora-Tyutin 1. Fermion. . . . . . . . . . . . . . . . . 412
1. Hamiltonian . . . . . . . . . . . . 288 2. Photon . . . . . . . . . . . . . . . . . 415
2. Lagrangian. . . . . . . . . . . . . . 292 3. Gluon. . . . . . . . . . . . . . . . . . . 416
3. Particles . . . . . . . . . . . . . . . . 295 4. Grand Unified Theories. .422
4. Fields . . . . . . . . . . . . . . . . . . . 296 5. Supermatter . . . . . . . . . . . . 425
B. Gauges 6. Supergluon. . . . . . . . . . . . . . 427
1. Radial . . . . . . . . . . . . . . . . . . 300 7. Bosonization . . . . . . . . . . . . 432
2. Lorentz . . . . . . . . . . . . . . . . . 303 8. Schwinger model . . . . . . . . 435
3. Massive . . . . . . . . . . . . . . . . . 305 B. Low energy
4. Gervais-Neveu. . . . . . . . . . .307 1. JWKB . . . . . . . . . . . . . . . . . . 441
5. Super Gervais-Neveu . . . . 310 2. Axial anomaly . . . . . . . . . . 444
6. Spacecone . . . . . . . . . . . . . . . 313 3. Anomaly cancelation . . . . 447
7. Superspacecone . . . . . . . . . 317 4. π 0 → 2γ . . . . . . . . . . . . . . . . 450
8. Background-field . . . . . . . . 319 5. Vertex . . . . . . . . . . . . . . . . . . 452
9. Nielsen-Kallosh . . . . . . . . . 325 6. Nonrelativistic JWKB . . . 455
10. Super background-field . . 327 C. High energy
C. Scattering 1. Conformal anomaly . . . . . 460
1. Yang-Mills . . . . . . . . . . . . . . 331 2. e+e− → hadrons . . . . . . . . 463
2. Recursion . . . . . . . . . . . . . . . 335 3. Parton model . . . . . . . . . . . 465
3. Fermions . . . . . . . . . . . . . . . . 337
4. Masses . . . . . . . . . . . . . . . . . . 339
5. Supergraphs . . . . . . . . . . . . 345
 iii

.............. ..............
. . . . . . . . . . . . . . PART THREE: HIGHER SPIN . . . . . . . . . . . . . .
IX. General relativity XI. Strings
A. Actions A. Scattering
1. Gauge invariance . . . . . . . . 473 1. Regge theory . . . . . . . . . . . . 595
2. Covariant derivatives . . . . 476 2. Classical mechanics . . . . . 598
3. Conditions . . . . . . . . . . . . . . 481 3. Gauges . . . . . . . . . . . . . . . . . 601
4. Integration . . . . . . . . . . . . . . 484 4. Quantum mechanics . . . . . 606
5. Gravity . . . . . . . . . . . . . . . . . 488 5. Anomaly . . . . . . . . . . . . . . . . 609
6. Energy-momentum . . . . . . 491 6. Tree amplitudes . . . . . . . . . 611
7. Weyl scale . . . . . . . . . . . . . . 494 B. Symmetries
B. Gauges 1. Massless spectrum. . . . . . .618
1. Lorentz . . . . . . . . . . . . . . . . . 500 2. Reality and orientation . . 620
2. Geodesics . . . . . . . . . . . . . . . 502 3. Supergravity . . . . . . . . . . . . 621
3. Axial . . . . . . . . . . . . . . . . . . . 505 4. T-duality . . . . . . . . . . . . . . . 622
4. Radial . . . . . . . . . . . . . . . . . . 507 5. Dilaton . . . . . . . . . . . . . . . . . 624
5. Weyl scale . . . . . . . . . . . . . . 512 6. Superdilaton . . . . . . . . . . . . 626
C. Curved spaces 7. Conformal field theory . . 628
1. Self-duality . . . . . . . . . . . . . 517 8. Triality . . . . . . . . . . . . . . . . . 633
2. De Sitter . . . . . . . . . . . . . . . . 518 C. Lattices
3. Cosmology . . . . . . . . . . . . . . 521 1. Spacetime lattice . . . . . . . . 637
4. Red shift . . . . . . . . . . . . . . . . 523 2. Worldsheet lattice . . . . . . . 641
5. Schwarzschild . . . . . . . . . . . 525 3. QCD strings . . . . . . . . . . . . 643
6. Experiments . . . . . . . . . . . . 531
7. Black holes. . . . . . . . . . . . . . 534 XII. Mechanics
A. OSp(1,1|2)
X. Supergravity 1. Lightcone . . . . . . . . . . . . . . . 649
A. Superspace 2. Algebra . . . . . . . . . . . . . . . . . 652
1. Covariant derivatives . . . . 538 3. Action . . . . . . . . . . . . . . . . . . 655
2. Field strengths . . . . . . . . . . 543 4. Spinors . . . . . . . . . . . . . . . . . 657
3. Compensators . . . . . . . . . . . 547 5. Examples . . . . . . . . . . . . . . . 659
4. Scale gauges . . . . . . . . . . . . 549 B. IGL(1)
B. Actions 1. Algebra . . . . . . . . . . . . . . . . . 664
1. Integration . . . . . . . . . . . . . . 555 2. Inner product . . . . . . . . . . . 665
2. Ectoplasm . . . . . . . . . . . . . . 558 3. Action . . . . . . . . . . . . . . . . . . 667
3. Component transform’ns 561 4. Solution. . . . . . . . . . . . . . . . .670
4. Component approach . . . . 562 5. Spinors . . . . . . . . . . . . . . . . . 673
5. Duality . . . . . . . . . . . . . . . . . 565 6. Masses . . . . . . . . . . . . . . . . . . 674
6. Superhiggs . . . . . . . . . . . . . . 569 7. Background fields . . . . . . . 675
7. No-scale . . . . . . . . . . . . . . . . 571 8. Strings . . . . . . . . . . . . . . . . . . 677
C. Higher dimensions 9. Relation to OSp(1,1|2) . . 682
1. Dirac spinors . . . . . . . . . . . . 574 C. Gauge fixing
2. Wick rotation . . . . . . . . . . . 577 1. Antibracket . . . . . . . . . . . . . 685
3. Other spins . . . . . . . . . . . . . 580 2. ZJBV . . . . . . . . . . . . . . . . . . . 688
4. Supersymmetry . . . . . . . . . 582 3. BRST . . . . . . . . . . . . . . . . . . 692
5. Theories . . . . . . . . . . . . . . . . 585
6. Reduction to D=4. . . . . . . 588

AfterMath . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696
 iv

............................... ...............................
............................... PREFACE ...............................

Scientific method
Although there are many fine textbooks on quantum field theory, they all have
various shortcomings. Instinct is claimed as a basis for most discussions of quantum
field theory, though clearly this topic is too recent to affect evolution. Their subjectiv-
ity more accurately identifies this as fashion: (1) The old-fashioned approach justifies
itself with the instinct of intuition. However, anyone who remembers when they first
learned quantum mechanics or special relativity knows they are counter-intuitive;
quantum field theory is the synthesis of those two topics. Thus, the intuition in this
case is probably just habit: Such an approach is actually historical or traditional,
recounting the chronological development of the subject. Generally the first half (or
volume) is devoted to quantum electrodynamics, treated in the way it was viewed in
the 1950’s, while the second half tells the story of quantum chromodynamics, as it
was understood in the 1970’s. Such a “dualistic” approach is necessarily redundant,
e.g., using canonical quantization for QED but path-integral quantization for QCD,
contrary to scientific principles, which advocate applying the same “unified” methods
to all theories. While some teachers may feel more comfortable by beginning a topic
the way they first learned it, students may wonder why the course didn’t begin with
the approach that they will wind up using in the end. Topics that are unfamiliar
to the author’s intuition are often labeled as “formal” (lacking substance) or even
“mathematical” (devoid of physics). Recent topics are usually treated there as ad-
vanced: The opposite is often true, since explanations simplify with time, as the topic
is better understood. On the positive side, this approach generally presents topics
with better experimental verification.
(2) In contrast, the fashionable approach is described as being based on the in-
stinct of beauty. But this subjective beauty of art is not the instinctive beauty of
nature, and in science it is merely a consolation. Treatments based on this approach
are usually found in review articles rather than textbooks, due to the shorter life ex-
pectancy of the latest fashion. On the other hand, this approach has more imagination
than the traditional one, and attempts to capture the future of the subject.
A related issue in the treatment of field theory is the relative importance of con-
cepts vs. calculations: (1) Some texts emphasize the concepts, including those which
have not proven of practical value, but were considered motivational historically (in
the traditional approach) or currently (in the artistic approach). However, many ap-
proaches that were once considered at the forefront of research have faded into oblivion
not because they were proven wrong by experimental evidence or lacked conceptual
v

attractiveness, but because they were too complex for calculation, or so vague they
lacked predicitive ability. Some methods claimed total generality, which they used to
prove theorems (though sometimes without examples); but ultimately the only useful
proofs of theorems are by construction. Often a dualistic, two-volume approach is
again advocated (and frequently the author writes only one of the two volumes): Like
the traditional approach of QED volume + QCD volume, some prefer concept volume
+ calculation volume. Generally, this means that gauge theory S-matrix calculations
are omitted from the conceptual field theory course, and left for a “particle physics”
course, or perhaps an “advanced field theory” course. Unfortunately, the particle
physics course will find the specialized techniques of gauge theory too technical to
cover, while the advanced field theory course will frighten away many students by its
title alone.
(2) On the other hand, some authors express a desire to introduce Feynman graphs
as quickly as possible: This suggests a lack of appreciation of field theory outside of
diagrammatics. Many essential aspects of field theory (such as symmetry breaking
and the Higgs effect) can be seen only from the action, and its analysis also leads to
better methods of applying perturbation theory than those obtained from a fixed set
of rules. Also, functional equations are often simpler than pictorial ones, especially
when they are nonlinear in the fields. The result of over-emphasizing the calculations
is a cookbook, of the kind familiar from some lower-division undergraduate courses
intended for physics majors but designed for engineers.
The best explanation of a theory is the one that fits the principles of scientific
method: simplicity, generality, and experimental verification. In this text we thus
take a more economical or pragmatic approach, with methods based on efficiency
and power. Unattractiveness or counter-intuitiveness of such methods become ad-
vantages, because they force one to accept new and better ways of thinking about
the subject: The efficiency of the method directs one to the underlying idea. For
example, although some consider Einstein’s original explanation of special relativity
in terms of relativistic trains and Lorentz transformations with square roots as be-
ing more physical, the concept of Minkowski space gave a much simpler explanation
and deeper understanding that proved more useful and led to generalization. Many
theories have “miraculous cancelations” when traditional methods are used, which
led to new methods (background field gauge, supergraphs, spacecone, etc.) that not
only incorporate the cancelations automatically (so that the “zeros” need not be cal-
culated), but are built on the principles that explain them. We place an emphasis
on such new concepts, as well as the calculational methods that allow them to be
compared with nature. It is important not to neglect one for the sake of the other,
artificial and misleading to try to separate them.
 vi

As a result, many of our explanations of the standard topics are new to textbooks,
and some are completely new: For example, (1) we derive the Foldy-Wouthuysen
transformation by dimensional reduction from an analogous one for the massless case
(subsections IIB3,5). (2) We derive the Feynman rules in terms of background fields
rather than sources (subsection VC1); this avoids the need for amputation of exter-
nal lines for S-matrices or effective actions, and is more useful for background-field
gauges. (3) We obtain the nonrelativistic QED effective action, used in modern treat-
ments of the Lamb shift (because it makes perturbation easier than the older Bethe-
Salpeter methods), by field redefinition of the relativistic effective action (subsection
VIIIB6), rather than fitting parameters by comparing Feynman diagrams from the
relativistic and nonrelativistic actions. (In general, manipulations in the action are
easier than in diagrams.) (4) We present a somewhat new method for solving for
the curvature in general relativity that is slightly easier than all previous methods
(subsections IXA2,C5). There are also some completely new topics, like: (1) the anti-
Gervais-Neveu gauge, where spin in U(N) Yang-Mills is treated in almost the same
way as internal symmetry — with Chan-Paton factors (subsection VIB4); (2) the
superspacecone gauge, the simplest gauge for QCD (subsection VIB7); and (3) a new
“(almost-)first-order” superspace action for supergravity, analogous to the one for
super Yang-Mills (subsection XB1).
We try to give the simplest possible calculational tools, not only for the above
reasons, but also so group theory (internal and spacetime) and integrals can be per-
formed with the least effort and memory. (Some traditionalists may claim that the
old methods are easy enough, but their arguments are less convincing when the order
of perturbation is increased. Even computer calculations are more efficient when left
as a last resort.) We give examples of (and excercises on) these methods, but not
exhaustively. We also include more recent topics (or those more recently appreciated
in the particle physics community) that might be deemed non-introductory, but are
commonly used, and are simple and important enough to include at the earliest level.
For example, the related topics of (unitary) lightcone gauge, twistors, and spinor
helicity are absent from all field theory texts, and as a result no such text performs
the calculation of as basic a diagram as the 4-gluon tree amplitude. Another missing
topic is the relation of QCD to strings through the random worldsheet lattice and
large-color (1/N) expansion, which is the only known method that might quantita-
tively describe its high-energy nonperturbative behavior (bound states of arbitrarily
large mass).
This text is meant to cover all the field theory every high energy theorist should
know, but not all that any particular theorist might need to know. It is not meant as
an introduction to research, but as a preliminary to such courses: We try to fill in the
vii

cracks that often lie between standard field theory courses and advanced specialized
courses. For example, we have some discussion of string theory, but it is more oriented
toward the strong interactions, where it has some experimental justification, rather
than quantum gravity and unification, where its usefulness is still under investigation.
We do not mention statistical mechanics, although many of the field theory methods
we discuss are useful there. Also, we do not discuss any experimental results in detail;
phenomenology and analysis of experiments deserve their own text. We give and apply
the methods of calculation and discuss the qualitative features of the results, but do
not make a numerical comparison to nature: We concentrate more on the “forest”
than the “trees”.
Unfortunately, our discussions of the (somewhat related) topics of infrared-diver-
gence cancelation, Lamb shift, and the parton model are sketchy, due to our inability
to give fully satisfying treatments — but maybe in a Second Edition?
Unlike all previous texts on quantum field theory, this one is available for free over
the Internet (as usual, from xxx.lanl.gov and its mirrors), and may be periodically
updated. Errata, additions, and other changes will be posted on my web page at
http://insti.physics.sunysb.edu/˜siegel/plan.html until enough are accumulated for
a new edition. Electronic distribution not only makes it more available, but easier
to update to new editions, and you don’t have to bother to lug your copy with you
when you go anywhere (like home) that has a computer (so you can access it from
a cartridge/diskette or the internet). It also offers the option of reading it on the
computer, which saves space and trees (and the figures are nicer in color). The
PDF version allows searches that are more general than the Index, and includes an
“outline” window with clickable “bookmarks” that is more convenient than the Table
of Contents, as well as the usual web links to xxx.lanl.gov (and a couple of other
places).

Highlights
This text also differs from others in most of the following ways: (1) We place a
greater emphasis on mechanics in introducing some of the more elementary physical
concepts of field theory: (a) Some basic ideas, such as antiparticles, can be more sim-
ply understood already with classical mechanics. (b) Some interactions can also be
treated through first-quantization: This is sufficient for evaluating certain tree and
one-loop graphs as particles in external fields. Also, Schwinger parameters can be
understood from first-quantization: They are useful for performing momentum inte-
grals (reducing them to Gaussians), studying the high-energy behavior of Feynman
graphs, and finding their singularities in a way that exposes their classical mechanics
 viii

interpretation. (c) Quantum mechanics is very similar to free classical field the-
ory, by the usual “semiclassical” correspondence between particles (mechanics) and
waves (fields). They use the same wave equations, since the mechanics Hamiltonian
or Becchi-Rouet-Stora-Tyutin operator is the kinetic operator of the corresponding
classical field theory, so the free theories are equivalent. In particular, (relativistic)
quantum mechanical BRST provides a simple explanation of the off-shell degrees
of freedom of general gauge theories, and introduces concepts useful in string theory.
As in the nonrelativistic case, this treatment starts directly with quantum mechanics,
rather than by (first-)quantization of a classical mechanical system. Since supersym-
metry and strings are so important in present theoretical research, it is useful to have
a text that includes the field theory concepts that are prerequisites to a course on
these topics. (For the same reason, and because it can be treated so similarly to
Yang-Mills, we also discuss general relativity.)
(2) We also emphasize conformal invariance. Although a badly broken sym-
metry, the fact that it is larger than Poincaré invariance makes it useful in many
ways: (a) General classical theories can be described most simply by first analyzing
conformal theories, and then introducing mass scales by various techniques. This
is particularly useful for the general analysis of free theories, and for constructing
actions for supergravity theories. (b) Quantum theories that are well-defined within
perturbation theory are conformal (“scaling”) at high energies. (A possible excep-
tion is string theories, but the supposedly well understood string theories that are
finite perturbatively have been discovered to be hard-to-quantize membranes in dis-
guise nonperturbatively.) This makes methods based on conformal invariance useful
for finding classical solutions, as well as studying the high-energy behavior of the
quantum theory, and simplifying the calculation of amplitudes. (c) Theories whose
conformal invariance is not (further) broken by quantum corrections avoid certain
problems at the nonperturbative level. Thus conformal theories ultimately may be
required for an unambiguous description of high-energy physics.
(3) We make extensive use of two-component (chiral) spinors, which are ubiqui-
tous in particle physics: (a) The method of twistors (more recently dubbed “spinor
helicity”) greatly simplifies the Lorentz algebra in Feynman diagrams for massless
(or high-energy) particles with spin, and it’s now a standard in QCD. (Twistors
are also related to conformal invariance and self-duality.) On the other hand, most
texts still struggle with 4-component Dirac (rather than 2-component Weyl) spinor
notation, which requires gamma-matrix and Fierz identities, when discussing QCD
calculations. (b) Chirality and duality are important concepts in all the interactions:
Two-component spinors were first found useful for weak interactions in the days of
4-fermion interactions. Chiral symmetry in strong interactions has been important
ix

since the early days of pion physics; the related topic of instantons (self-dual solutions)
is simplified by two-component notation, and general self-dual solutions are expressed
in terms of twistors. Duality is simplest in two-component spinor notation, even when
applied to just the electromagnetic field. (c) Supersymmetry still has no convincing
experimental verification (at least not at the moment I’m typing this), but its the-
oretical properties promise to solve many of the fundamental problems of quantum
field theory. It is an element of most of the proposed generalizations of the Standard
Model. Chiral symmetry is built into supersymmetry, making two-component spinors
unavoidable.
(4) The topics are ordered in a more pedagogical manner: (a) Abelian and non-
abelian gauge theories are treated together using modern techniques. (Classical grav-
ity is treated with the same methods.) (b) Classical Yang-Mills theory is discussed be-
fore any quantum field theory. This allows much of the physics, such as the Standard
Model (which may appeal to a wider audience), of which Yang-Mills is an essential
part, to be introduced earlier. In particular, symmetries and mass generation in the
Standard Model appear already at the classical level, and can be seen more easily from
the action (classically) or effective action (quantum) than from diagrams. (c) Only
the method of path integrals is used for second-quantization. Canonical quantization
is more cumbersome and hides Lorentz invariance, as has been emphasized even by
Feynman when he introduced his diagrams. We thus avoid such spurious concepts as
the “Dirac sea”, which supposedly explains positrons while being totally inapplica-
ble to bosons. However, for quantum physics of general systems or single particles,
operator methods are more powerful than any type of first-quantization of a classical
system, and path integrals are mainly of pedagogical interest. We therefore “review”
quantum physics first, discussing various properties (path integrals, S-matrices, uni-
tarity, BRST, etc.) in a general (but simpler) framework, so that these properties need
not be rederived for the special case of quantum field theory, for which path-integral
methods are then sufficient as well as preferable.
(5) Gauge fixing is discussed in a way more general and efficient than older meth-
ods: (a) The best gauge for studying unitarity is the (unitary) lightcone gauge. This
rarely appears in field theory texts, or is treated only half way, missing the important
explicit elimination of all unphysical degrees of freedom. (b) Ghosts are introduced
by BRST symmetry, which proves unitarity by showing equivalence of convenient
and manifestly covariant gauges to the manifestly unitary lightcone gauge. It can be
applied directly to the classical action, avoiding the explicit use of functional determi-
nants of the older Faddeev-Popov method. It also allows direct introduction of more
general gauges (again at the classical level) through the use of Nakanishi-Lautrup
fields (which are omitted in older treatments of BRST), rather than the functional
 x

averaging over Landau gauges required by the Faddeev-Popov method. (c) For non-
abelian gauge theories the background field gauge is a must. It makes the effective
action gauge invariant, so Slavnov-Taylor identities need not be applied to it. Beta
functions can be found from just propagator corrections.
(6) Dimensional regularization is used exclusively (with the exception of one-loop
axial anomaly calculations): (a) It is the only one that preserves all possible sym-
metries, as well as being the only one practical enough for higher-loop calculations.
(b) We also use it exclusively for infrared regularization, allowing all divergences to
be regularized with a single regulator (in contrast, e.g., to the three regulators used
for the standard treatment of Lamb shift). (c) It is good not only for regularization,
but renormalization (“dimensional renormalization”). For example, the renormaliza-
tion group is most simply described using dimensional regularization methods. More
importantly, renormalization itself is performed most simply by a minimal prescrip-
tion implied by dimensional regularization. Unfortunately, many books, even among
those that use dimensional regularization, apply more complicated renormalization
procedures that require additional, finite renormalizations as prescribed by Slavnov-
Taylor identities. This is a needless duplication of effort that ignores the manifest
gauge invariance whose preservation led to the choice of dimensional regularization in
the first place. By using dimensional renormalization, gauge theories are as easy to
treat as scalar theories: BRST does not have to be applied to amplitudes explicitly,
since the dimensional regularization and renormalization procedure preserves it.
(7) Perhaps the most fundamental omission in most field theory texts is the
expansion of QCD in the inverse of the number of colors: (a) It provides a gauge-
invariant organization of graphs into subsets, allowing simplifications of calculations
at intermediate stages, and is commonly used in QCD today. (b) It is useful as a
perturbation expansion, whose experimental basis is the Okubo-Zweig-Iizuka rule.
(c) At the nonperturbative level, it leads to a resummation of diagrams in a way that
can be associated with strings, suggesting an explanation of confinement.

Notes for instructors

This text is intended for reference and as the basis for a full-year course on rela-
tivistic quantum field theory for second-year graduate students. A preliminary version
of the first two parts was used for a one-year course I taught at Stony Brook. The
chapter on gravity and pieces of early chapters cover a one-semester graduate rela-
tivity course I gave several times here — I used most of the following: IA, IB3, IC2,
IIA, IIIA-C5, VIB1, IX, XIA2-4, XIB4-5. The prerequisites (for the quantum field
xi

theory course) are the usual first-year courses in classical mechanics, classical elec-
trodynamics, and quantum mechanics. For example, the student should be familiar
with Hamiltonians and Lagrangians, Lorentz transformations for particles and elec-
tromagnetism, Green functions for wave equations, SU(2) and spin, and Hilbert space.
Unfortunately, I find that many second-year graduate students (especially many who
got their undergraduate training in the USA) still have only an undergraduate level
of understanding of the prerequisite topics, lacking a working knowledge of action
principles, commutators, creation and annihilation operators, etc. While most such
topics are briefly reviewed here, they should be learned elsewhere.
There is far more material here than can be covered comfortably in one year,
mostly because of included material that should be covered earlier, but rarely is.
Ideally, a modern curriculum for field theory students should include: (1) courses on
classical mechanics, nonrelativistic quantum mechanics, and classical electrodynamics
in the first semester of graduate study, without overly reviewing aspects that should
have been covered in undergraduate study (and in particular avoiding the enormous
overlap of the last two subjects due to both covering primarily the solution of wave
equations); (2) in the second semester, statistical mechanics as the sequel to clas-
sical, relativistic quantum mechanics as the sequel to nonrelativistic, and classical
nonabelian field theory (Yang-Mills and gravity) as the sequel to classical electrody-
namics; (3) in the second year, one year of quantum field theory, and at least one
semester on “phenomenology” (model-building and direct comparison with observa-
tions, including those for general relativity and cosmology); and (4) in the third year,
more specialized courses, such as a semester on supersymmetry and strings. Unfor-
tunately, in practice little of relativistic quantum mechanics and classical field theory
(other than electromagnetism) will have been covered previously, which means they
will comprise half of the “quantum” field theory course, while the true quantum field
theory will be squeezed into the last half.
One way to cut the material to fit a one-year course is to omit Part Three, which
can be left for a third semester on “advanced quantum field theory”; then the first
semester (Part One) is classical while the second (Part Two) is quantum. Further-
more, the ordering of the chapters is somewhat flexible: The “flow” is indicated by
the following “3D” plot:
classical → quantum
 symmetry fields quantize loop
 lower spin Bose I III V VII
&.
 ↓ IX XI
higher spin
X XII
Fermi II IV VI VIII
 xii

where the 3 dimensions are spin (“j”), quantization (“h̄”), and statistics (“s”): The
three independent flows are down the page, to the right, and into the page. (The third
dimension has been represented as perpendicular to the page, with “higher spin” in
smaller type to indicate perspective, for legibility.) To present these chapters in the
1 dimension of time we have classified them as jh̄s, but other orderings are possible:

jh̄s : I II III IV V VI VII VIII IX X XI XII

jsh̄ : I III V VII II IV VI VIII IX XI X XII
h̄js : I II III IV IX X V VI XI XII VII VIII
h̄sj : I II III IX IV X V XI VI XII VII VIII
sjh̄ : I III V VII IX XI II IV VI VIII X XII
sh̄j : I III IX V XI VII II IV X VI XII VIII
(However, the spinor notation of II is used for discussing instantons in III, so some
rearrangement would be required, except in the jh̄s, h̄js, and h̄sj cases.) For exam-
ple, the first half of the course can cover all of the classical, and the second quantum,
dividing Part Three between them (h̄js or h̄sj). Another alternative (jsh̄) is a one-
semester course on quantum field theory, followed by a semester on the Standard
Model, and finishing with supergravity and strings. Although some of these (espe-
cially the first two) allow division of the course into one-semester courses, this should
not be used as an excuse to treat such courses as complete: Any particle physics
student who was content to sit through another entire year of quantum mechanics in
graduate school should be prepared to take at least a year of field theory.

Notes for students

Field theory is a hard course. (If you don’t think so, name me a harder one at this
level.) But you knew as an undergraduate that physics was a hard major. Students
who plan to do research in field theory will find the topic challenging; those with less
enthusiasm for the topic may find it overwhelming. The main difference between field
theory and lower courses is that it is not set in stone: There is much more variation in
style and content among field theory courses than, e.g., quantum mechanics courses,
since quantum mechanics (to the extent taught in courses) was pretty much finished
in the 1920’s, while field theory is still an active research topic, even though it has had
many experimentally confirmed results since the 1940’s. As a result, a field theory
course has the flavor of research: There is no set of mathematically rigorous rules to
solve any problem. Answers are not final, and should be treated as questions: One
should not be satisfied with the solution of a problem, but consider it as a first step
toward generalization. The student should not expect to capture all the details of
xiii

field theory the first time through, since many of them are not yet fully understood
by people who work in the area. (It is far more likely that instead you will discover
details that you missed in earlier courses.) And one reminder: The only reason for
lectures (including seminars and conferences) is for the attendees to ask questions
(and not just in private), and there are no stupid questions (except for the infamous
“How many questions are on the exam?”). Only half of teaching is the responsibility
of the instructor.

Outline
The preceding Table of Contents lists the three parts of the text: Symmetry,
Quanta, and Higher Spin. Each part is divided into four chapters, each of which has
three sections, divided further into subsections. Each section is followed by references
to reviews and original papers. Excercises appear throughout the text, immediately
following the items they test: This purposely disrupts the flow of the text, forcing
the reader to stop and think about what he has just learned. These excercises are
interesting in their own right, and not just examples or memory tests. This is not a
crime for homeworks and exams, which at least by graduate school should be about
more than just grades.
The first part of the text focuses on symmetry: The Poincaré group is special
relativity, and is sufficient to find all free equations of motion for particles and fields.
Internal symmetries include both global ones, used for classifying particles, and local
ones, which describe the interactions of fields.
The first chapter discusses global symmetry, both spacetime and internal. Space-
time symmetries covered include not only Poincaré but also Galilean (i.e., nonrel-
ativistic, used as an introduction) and conformal (broken in nature, but still very
useful). Parity, time reversal, and even charge conjugation are described simply in
terms of classical mechanics. Lightcone bases are introduced. Some general proper-
ties of Lie algebras are summarized, including fermions and anticommuting numbers.
Classical groups are described using tensor methods and index notation, including
Young tableaux. Dirac gamma matrices appear as coordinates for orthogonal groups.
The color and flavor symmetries of the particles of the Standard Model, and observed
light hadrons, are given as examples.
The second chapter extends the first chapter’s treatment of spacetime symmetry
to include spin. The methods introduced in this chapter are the most efficient ones for
handling Lorentz indices in QCD (or even pure Yang-Mills theory). Two-component
spinor notation is introduced by the rotation group in three (space) dimensions: Ten-
sor notation avoids Clebsch-Gordan-Wigner coefficients. The study of the simple
 xiv

algebraic properties of 2×2 matrices is extended straightforwardly from three dimen-

sions to four, and applied to simple examples in free field theory. The conformal
group gives an easy and unified way to find massless free field equations in general,
and dimensional reduction does the same for masses. As an application, we discuss
the Foldy-Wouthuysen transformation (and its massless analog) for arbitrary spin,
with minimal electromagnetic coupling to spin 1/2 as an example. (The case of non-
minimal coupling will be useful in chapter VIII for the Lamb shift.) Twistors, related
to conformal invariance and self-duality, yield a convenient and covariant method to
solve the massless equations, and also explain helicity. The chapter concludes with a
discussion of the general properties of supersymmetry and its representations, using
superspace and supertwistors.
Local symmetries are covered in the third chapter. It begins with a discussion
of the action principle, including fermions and constrained systems, applied to the
simple free examples given earlier (spins 1/2 and 1). The concepts of gauge invariance
and gauge fixing are introduced through the simple case of the (spinless) relativistic
particle, whose classical mechanics will prove useful later in understanding several
features of Feynman diagrams. As an example, classical pair creation and annihilation
is examined. Finally, pure Yang-Mills theory is analyzed, including some solutions to
the classical field equations. Twistors are used to study self-duality and instantons.
The lightcone gauge is used as a unitary gauge, and in combination with self-duality.
Gauge symmetry is coupled to lower spins in chapter four. Following an intro-
duction to spontaneous symmetry breakdown of global symmetries, nonlinear sigma
models are considered as low-energy theories, particularly in the study of chiral sym-
metry, and gauge invariance is used in their general construction. The use of scalars
to generate mass for vectors is illustrated first by the free case of the Stückelberg
model, generalized to the Higgs model, and applied to the Standard Model. Families
and Grand Unified Theories are also described, as well as the basics for construct-
ing actions for supersymmetric theories in superspace (including a brief discussion of
extended supersymmetry).
The second part of the text covers the quantum aspects of field theory, as revealed
through perturbation theory. Although some have conjectured that nonperturbative
approaches might solve the renormalization difficulties found in perturbation, all ev-
idence indicates these features survive in the complete theory.
Chapter five focuses on the method of quantization of classical theories based
on path integrals. The chapter begins by considering various properties of quantum
physics in a general context — relation to canonical quantization, Wick rotation,
unitarity, and causality — so that these items need not be repeated in the more
xv

specialized and complicated cases of field theory. Path integrals are then applied
to classical mechanics to explain the Stückelberg-Feynman propagator. Path integra-
tion of field theory produces a generating functional of background fields for Feynman
diagrams, as well as its connected and one-particle-irreducible parts. We use back-
grounds fields instead of sources exclusively: All uses of Feynman diagrams involve
either the S-matrix or the effective action, both of which require the removal of ex-
ternal propagators, which is equivalent to replacing sources with fields. The classical
(tree) graphs are shown to give the perturbative solution to the classical field equa-
tions. Also described are the properties of the classical action needed for unitarity, the
diagrammatic translation of unitarity and causality, the definition of cross sections,
the relation of Landau singularities to classical mechanics, and the use of the quark
line rules for dealing with group theory in graphs easily. Throughout the chapter
simple examples are given from scalar theories.
Complications that arise from quantization of gauge theories are described in the
sixth chapter. BRST symmetry is the easiest way to gauge fix, and makes unitarity
clear by relating general gauges to unitary gauges. Again a general discussion is
given in the framework of quantum physics and canonical quantization, including
the relation of Hamiltonian and Lagrangian approaches, so that later field theory
can be addressed covariantly with path integrals. Various gauges are considered for
Yang-Mills fields: radial, Lorentz, Landau, Fermi-Feynman, unitary, renormalizable,
Gervais-Neveu, anti-Gervais-Neveu, super Gervais-Neveu, spacecone, superspacecone,
background-field, and Nielsen-Kallosh. The spacecone gauge is used as the simplest
method to calculate graphs in massless theories, with examples given from (massless)
QCD, including the 4-gluon and 5-gluon tree amplitudes. The Fermi-Feynman gauge
is used to calculate all the 4-point tree amplitudes of QED, and their differential
cross sections. The supergraph rules are derived for supersymmetric theories, and
the locality of the effective action in the anticommuting coodinates is shown to imply
nonrenormalization theorems.
General features of higher orders in perturbation theory due to momentum inte-
gration are examined in chapter seven. Renormalization is explained (but not proven),
and dimensional regularization is applied. Tadpole integrals are used to explain di-
mensional transmutation through the example of the effective potential. The running
of couplings with energy is shown through the evaluation of one-loop massless and
massive propagator corrections. Some simple overlapping two-loop divergences are
used to illustrate renormalization of subdivergences. The renormalization group equa-
tions are introduced via dimensional regularization. The problems solved perturba-
tively by renormalization are shown to reappear upon resummation of the expansion.
Instantons and IR and UV renormalons are analyzed through a Borel transform in
 xvi

the coupling, and the resultant ambiguities are related to nonperturbative vacuum
values of composite fields. The expansion in the inverse of the number of colors is an
approach to this problem, related to string theory, that also has uses at finite orders
of perturbation.
Chapter eight applies these methods to gauge theories. Propagator corrections in
QED and QCD are used to analyze the one-loop conformal anomaly and its relation to
asymptotic freedom. Finite N=1 supersymmetric theories are considered as a solution
to the renormalon problem. The Schwinger model is given as another example from
two dimensions, illustrating interesting features at one loop such as bound states,
bosonization, and the axial anomaly. The axial anomaly is then evaluated more
generally, and considered in four dimensions in relation to constraints on electroweak
models and electromagnetic pion decay. The nonrelativistic form of the effective
action useful for finding the Lamb shift (including the anomalous magnetic moment) is
given as an example of vertex corrections. Finally, the production of hadrons through
electron-positron annihilation, deep inelastic scattering, and Drell-Yan scattering are
briefly described as applications of perturbative QCD.
Part Three treats general spins, particularly spin 2, which ultimately must be
included in any complete theory of nature. Such spins are observed experimentally
for bound states, but may be required also as fundamental fields.
Gravity is described through the theory of general relativity in chapter nine.
The classic experimental tests are described, including cosmology. The treatment
used is closely related to that applied to Yang-Mills theory, and differs from that
of most texts on gravity: (1) We emphasize the action for deriving field equations
for gravity (and matter), rather than treating it as an afterthought. (2) We make
use of local (Weyl) scale invariance for cosmological solutions, gauge fixing, field
redefinitions, and studying conformal properties. In particular, other texts neglect the
(unphysical) dilaton, which is crucial in such treatments (especially for generalization
to supergravity and strings). (3) While most gravity texts leave spinors till the end,
and treat them briefly, our discussion of gravity is based on methods that can be
applied directly to spinors, and therefore to supergravity and superstrings. (4) Our
method of calculating curvatures for purposes of solving the classical field equations
is somewhat new, but probably the simplest, and is directly related to the simplest
methods for super Yang-Mills theory and supergravity.
The approach of the previous chapter is generalized straightforwardly to super-
gravity in chapter ten. Actions with matter are constructed, and are analyzed in
superspace, and in terms of component fields using component expansion and sep-
aration of superconformal breaking (“compensator”) terms. The spin-3/2 particle
xvii

is given mass by the superhiggs effect, and no-scale supergravity provides a model
whereby the would-be resultant cosmological constant vanishes naturally. Extended
supergravity is analyzed through general properties of extended supersymmetry and
by reduction from higher dimensions.
Strings are proposed in chapter eleven as an approach to studying the most im-
portant yet least understood property of QCD: confinement. Other methods have
been proposed to study this phenomenon, but none have achieved explicit results for
more than low hadron energy, which no more exhibits confinement than chemistry
disproves the existence of free nuclei. Known string theories are not suitable for de-
scribing hadrons quantitatively, but are useful models of observed properties, such
as Regge behavior. The classical and quantum theory of the simplest such model is
analyzed, and qualitative features expected of general theories are described. The dis-
cretization of the worldsheet of the string into a sum of Feynman diagrams is shown
to exhibit features relevant to a string theory of hadrons.
The final chapter gives a general derivation of free actions for any gauge theory,
based on adding equal numbers of commuting and anticommuting ghost dimensions
to the lightcone formulation of the Poincaré group. The usual ghost fields appear as
components of the gauge fields in anticommuting directions, as do necessary auxiliary
fields like the determinant of the metric tensor in gravity. Gauge fixing to the Fermi-
Feynman gauge is automatic. The “antifields” and “antibracket” of the Zinn-Justin-
Batalin-Vilkovisky method appear naturally from the anticommuting coordinate that
is the first-quantized ghost of the Klein-Gordon equation.
Following the body of the text (and preceding the Index) is the AfterMath, con-
taining conventions and some of the more important equations.

Acknowledgments
I thank everyone with whom I have discussed field theory, especially Gordon
Chalmers, Marc Grisaru, Marcelo Leite, Martin Roček, Jack Smith, George Sterman,
and Peter van Nieuwenhuizen. More generally, I thank the human race, without
whom this work would have been neither possible nor necessary.

December 20, 1999

 xviii

............. .............
............. SOME FIELD THEORY TEXTS .............

Comprehensive, traditional
Complete texts; use canonical quantization for QED, then path integrals for QCD
1 S. Weinberg, The quantum theory of fields, 3 v. (Cambridge University, 1995,6,9?)
609+489+c.500 pp.:
First volume just QED; second volume contains many interesting topics; third
volume supersymmetry. By one of the developers of the Standard Model.
2 M.E. Peskin and D.V. Schroeder, An introduction to quantum field theory
(Addison-Wesley, 1995) 842 pp.:
Many applications; style similar to Bjorken and Drell.
3 M. Kaku, Quantum field theory: a modern introduction (Oxford University, 1993)
785 pp.:
Includes introduction to supergravity and superstrings.
4 C. Itzykson and J.-B. Zuber, Quantum field theory (McGraw-Hill, 1980) 705 pp.
(but with lots of small print):
Emphasis on QED.

Somewhat specialized
Basics, plus thorough treatment of an advanced topic
5 J. Zinn-Justin, Quantum field theory and critical phenomena, 3rd ed. (Clarendon,
1996) 1008 pp.:
First 1/2 is basic text, with interesting treatments of many topics, but no S-matrix
examples or discussion of cross sections; second 1/2 is statistical mechanics.
6 G. Sterman, An introduction to quantum field theory (Cambridge University,
1993) 572 pp.:
First 3/4 can be used as basic text, including S-matrix examples; last 1/4 has
extensive treatment of perturbative QCD, emphasizing factorization.

Basic; few S-matrix examples

Should be supplemented with a “QED/particle physics text”
7 L.H. Ryder, Quantum field theory, 2nd ed. (Cambridge University, 1996) 487 pp.:
Includes introduction to supersymmetry.
8 D. Bailin and A. Love, Introduction to gauge field theory, 2nd ed. (Institute of
Physics, 1993) 364 pp.:
All the fundamentals.
9 P. Ramond, Field theory: a modern primer , 2nd ed. (Addison-Wesley, 1989)
329 pp.:
Short text on QCD: no weak interactions or Higgs.
xix

Classics
Older but unconventional treatments from their originators; no Yang-Mills or Higgs
10 N.N. Bogoliubov and D.V. Shirkov, Introduction to the theory of quantized fields,
3rd ed. (Wiley, 1980) 620 pp.:
Ahead of its time (1st English ed. 1959); early treatments of path integrals, causal-
ity, background fields, and renormalization of all field theories (not just QED).
11 R.P. Feynman, Quantum electrodynamics: a lecture note and reprint volume (Ben-
jamin, 1961) 198 pp.:
Original treatment of quantum field theory as we know it today, but from me-
chanics; includes reprints of original articles (1949).

QED/particle physics
Numerous Feynman diagram calculations; no Yang-Mills or Higgs
12 B. de Wit and J. Smith, Field theory in particle physics, v. 1 (Elsevier Science,
1986) 490 pp.:
Oriented toward experimentalists (but wait till v. 2, due any day now...).
13 A.I. Akhiezer and V.B. Berestetskii, Quantum electrodynamics (Wiley, 1965) 868
pp.:
Extensive examples of lower-order QED diagrams.

Advanced topics
For further reading; including brief reviews of some standard topics
14 Theoretical Advanced Study Institute in Elementary Particle Physics (TASI) pro-
ceedings, University of Colorado, Boulder, CO (World Scientific):
Annual collection of summer school lectures on recent research topics.
15 W. Siegel, Introduction to string field theory (World Scientific, 1988) 244 pp.:
Reviews lightcone, BRST, gravity, first-quantization, spinors, twistors, strings;
besides, I like the author.
16 S.J. Gates, Jr., M.T. Grisaru, M. Roček, and W. Siegel, Superspace: or one thou-
sand and one lessons in supersymmetry (Benjamin/Cummings, 1983) 548 pp.:
Covers supersymmetry, spinor notation, lightcone, Stückelberg fields, gravity,
Weyl scale, gauge fixing, background-field method, regularization, and anoma-
lies; same author as previous, plus three other guys whose names sound familiar.
May soon be available for free where you found this book.
 1

PART ONE: SYMMETRY

The first four chapters present a one-semester course on “classical field theory”.
Perhaps a more accurate description would be “everything you should know before
learning quantum field theory”. This is basically a study of global and local symme-
tries: Classical dynamics represents only a certain limit of quantum dynamics, and not
the one usually emphasized, but most of the symmetries of classical physics survive
quantization. The phenomenon of symmetry breaking, and the related mechanisms of
mass generation, can also be seen at the classical level. In perturbative quantum field
theory, classical field theory is simply the leading term in the perturbation expansion.
Continuous symmetry is one of the most fundamental and important concepts
of physics. In the framework of an action principle (which is required in quantum
physics), it is equivalent to conservation laws, which have been a cornerstone of physics
since Newton. From a practical viewpoint, it simplifies calculations by relating differ-
ent solutions to equations of motion, and allowing these equations to be written more
concisely by treating independent degrees of freedom as a single entity. In particular,
local (“gauge”) symmetries, which allow independent transformations at each coordi-
nate point, are basic to all the fundamental interactions: All the fundamental forces
are mediated by particles described by Yang-Mills theory and its generalizations.
Symmetries are the result of a redundant, but useful, description of a theory.
(Note that here we refer to symmetries of a theory, not of a solution to the theory.) For
example, translation invariance says that only differences in position are measurable,
not absolute position: We can’t measure the position of the “origin”. There are two
ways to deal with this: (1) Choose an origin; i.e., make a “choice of coordinates”.
For example, place an object at the origin; i.e., choose the position of an object at
a certain time to be the origin. (2) Work only in terms of differences of coordinates,
which are “translationally invariant”. Although the latter choice is more physical,
the former is usually more convenient: The use of redundant variables, together with
symmetry, often gives a simpler description of a theory. Another example is quantum
mechanics, where the arbitrariness of the phase of the wave function can be considered
a symmetry: Although quantum mechanics can be reformulated in terms of phase-
invariant probabilities, currents, or density matrices instead of wave functions, and
this can be useful for some purposes of exposing physical properties, formulating and
solving the Schrödinger equation is simpler in terms of the wave function. The same
applies to “local” symmetries, where there is an independent symmetry at each point
of space and time: For example, quarks and gluons have a local “color” symmetry,
2

and are not (yet) observed independently in nature, but are simpler objects in terms
of which to describe strong interactions than the observed hadrons (protons, neutrons,
etc.), which are described by color-invariant products of quark/gluon wave functions,
in the same way that probabilities are phase-invariant products of wave functions.
(Note that in quantum mechanics there is a subtle distinction between observed and
observer that can obscure this symmetry if the observer is not invariant under it. This
can always be avoided by choosing to define the observer as invariant: For example,
the detection apparatus can be included as part of the quantum mechanical system,
while the observer can be defined as some “remote” recorder, who may be abstracted
as even being translationally invariant. In practice we are less precise, and abstract
even the detection apparatus to be invariant: For example, we describe the scattering
of particles in terms of the coordinates of only the particles, and deal with the origin
problem as above in terms of just those coordinates.)
Note that “global” (time-, and usually space-independent) symmetries can elim-
inate a variable, but not its time derivative. For example, translation invariance
allows us to fix (i.e., eliminate) the position of the center of mass of a system at some
initial time, but not its time derivative, which is just the total momentum, whose
conservation is a consequence of that same symmetry. A local symmetry, being time
dependent, may allow the elimination of a variable at all times: The existence of this
possibility depends on the dynamics, and will be discussed later.
Of particular intrerest are ways in which symmetries can be made manifest. Fre-
quently in the literature “manifest” is used vacuously; a “manifest symmetry” is an
obvious one: If you know the group, the representation under consideration doesn’t
need to be stated, but can be seen from just the notation. (In fact, one of the main
uses of index notation is just to manifest the symmetry.) Formulations where global
and local symmetries are manifest simplify calculations and their results, as well as
clarifying their meaning.
One of the main uses of manifest symmetry is rarely needing to explicitly perform
a specific symmetry transformation. For example, one might need to examine a rela-
tivistic problem in different Lorentz frames. Rather than starting with a description
of the problem in one frame, and then explicitly transforming to another, it is much
simpler to start with a manifestly covariant description, make one choice of frame,
then make another choice of frame. One then never uses the messy square roots of the
familiar Lorentz contraction factors (although they may appear at the end from kine-
matic constraints). A more extreme example is the corresponding situation for local
 A. COORDINATES 3

symmetries, where such transformations are intractable in general, and one always
starts with the manifestly covariant form.

I. GLOBAL
In the first chapter we study symmetry in general, concentrating primarily on
spacetime symmetries, but also discussing general properties that will have other
applications in the following chapter.
....................... .......................
....................... A. COORDINATES .......................

In this section we discuss the Poincaré (and conformal) group as coordinate trans-
formations. This is the simplest way to represent it on the physical world. In later
sections we find general representations by adding spin.

1. Nonrelativity
We begin by reviewing some general properties of symmetries, including as an
example the symmetry group of nonrelativstic physics. In the Hamiltonian approach
to mechanics, both symmetries and dynamics can be expressed conveniently in terms
of a “bracket”: the Poisson bracket for classical mechanics, the commutator for quan-
tum mechanics. In this formulation, the fundamental variables (operators) are some
set of coordinates and their canonically conjugate momenta, as functions of time.
The (Heisenberg) operator approach to quantum mechanics then is related to classi-
cal mechanics by identifying the semiclassical limit of the commutator as the Poisson
bracket: For any functions A and B of p and q, the quantum mechanical commutator
is  
∂A ∂B ∂B ∂A
AB − BA = −ih̄ − + O(h̄2 )
∂pm ∂q m ∂pm ∂q m
In other words, the true classical limit of AB − BA is zero, since classically functions
commute; thus the semiclassical limit is defined by
 
1
lim (AB − BA)
h̄→0 h̄

(which is really a derivative with respect to h̄). We therefore define the bracket for
the two cases by
  

 −i ∂A ∂B − ∂B ∂A semiclassically
[A, B] ≡ ∂pm ∂q m ∂pm ∂q m


AB − BA quantum mechanically
4 I. GLOBAL

The semiclassical definition of the bracket then can be applied to classical physics
(where it was originally discovered). Classically A and B are two arbitrary functions
of the coordinates q and momenta p; in quantum mechanics they can be arbitrary
operators. We have included an “i” in the classical normalization so the two agree
in the semiclassical limit. We generally use (natural/Planck) units h̄ = 1, so mass is
measured as inverse length, etc.; when we do use an explicit h̄, it is a dimensionless
parameter, and appears only for defining Jeffries-Wentzel-Kramers-Brillouin (JWKB)
expansions or (semi)classical limits.
Our indices may appear either as subscripts or superscripts, with preferences to
be explained later: For nonrelativistic purposes we treat them the same. We also use
the Einstein summation convention, that any repeated index in a product is summed
over (“contracted”); usually we contract a superscript with a subscript:
X
Am Bm ≡ Am Bm
m

The definition of the bracket is equivalent to using

[pm , q n] = −iδm
n

n
(where δm is the “Kronecker delta function”: 1 if m = n, 0 if m 6= n) together with
the general properties of the bracket

[A, B] = −[B, A], [A, B]† = −[A†, B †]

[[A, B], C] + [[B, C], A] + [[C, A], B] = 0

[A, BC] = [A, B]C + B[A, C]

The first set of identities exhibit the antisymmetry of the bracket; next are the “Ja-
cobi identities”. In the last identity the ordering is important only in the quantum
mechanical case: In general, the difference between classical and quantum mechanics
comes from the fact that in the quantum case operator reordering after taking the
commutator results in multiple commutators.
Infinitesimal symmetry transformations are then written as

δA = i[G, A], A0 = A + δA

where G is the “generator” of the transformation. More explicitly, infinitesimal gen-

erators will contain infinitesimal parameters: For example, for translations we have

G = i pi ⇒ δxi = i[G, xi] = i

 A. COORDINATES 5

where i are infinitesimal numbers.

The most evident physical symmetries are those involving spacetime. For nonrel-
ativistic particles, these symmetries form the “Galilean group”: For the free particle,
those infinitesimal transformations are linear combinations of
p2i
M = m, Pi = pi , Jij = x[ipj] ≡ xi pj − xj pi , E=H= , Vi = mxi − pi t
2m
in terms of the position xi (i = 1, 2, 3), momenta pi , and (nonvanishing) mass m,
where [ij] means to antisymmetrize in those indices, by summing over all permuta-
tions (just two in this case), with plus signs for even permutations and minus for odd.
(In three spatial dimensions, one often writes Ji = 12 ijk Jjk to make J into a vector.
This is a peculiarity of three dimensions, and will lose its utility once we consider rela-
tivity in four spacetime dimensions.) These transformations are the space translations
(momentum) P , rotations (angular momentum — just orbital for the spinless case)
J, time translations (energy) E, and velocity transformations (“Galilean boosts”)
V . (The mass M is not normally associated with a symmetry, and is not conserved
relativistically.)
Excercise IA1.1
Let’s examine the Galilean group more closely. Using just the relations for
[x, p] and [A, BC] (and the antisymmetry of the bracket):
a Find the action on xi of each kind of infintesimal Galilean transformation.
b Show that the nonvanishing commutation relations for the generators are
[k
[Jij , Pk ] = iδk[i Pj] , [Jij , Vk ] = iδk[i Vj] , [Jij , J kl ] = iδ[i Jj] l]

[Pi , Vj ] = −iδij M, [H, Vi ] = −iPi

For more than one free particle, we introduce an m, xi, and pi for each particle
(but the same t), and the generators are the sums over all particles of the above ex-
pressions. If the particles interact with each other the expression for H is modified, in
such a way as to preserve the commutation relations. If the particles also interact with
dynamical fields, field-dependent terms must be added to the generators. (External,
nondynamical fields break the invariance. For example, a particle in a Coulomb po-
tential is not translation invariant since the potential is centered about some point.)
Note that for N particles there are 3N coordinates describing the particles, but still
only 3 translations: The particles interact in the same 3-dimensional space. We can
use translational invariance to fix the position of any one particle at a given time, but
not the rest: The differences in position are translationally invariant. On the other
6 I. GLOBAL

hand, it is often useful not to fix the position of any particle, since keeping this invari-
ance (and the corresponding redundant variables) allows all particles to be treated
equally. We might also consider using the differences of positions themselves as the
variables, allowing a symmetric treatment of the particles in terms of translationally
invariant variables: However, this would require applying constraints on the variables,
since there are 3N(N-1)/2 differences, of which only 3(N-1) are independent. We will
find similar features later for “local” invariances: In general, the most convenient
description of a theory is with the invariance; the invariance can then be fixed, or
invariant combinations of variables used, appropriately for the particular application.
The rotations (or at least their “orbital” parts) and space translations are exam-
ples of coordinate transformations. In general, generators of coordinate transforma-
tions are of the form

G = λi (x)pi ⇒ δφ(x) = i[G, φ] = λi ∂i φ

where ∂i = ∂/∂xi and φ(x) is a “scalar field” (or “spin-0 wave function”), a function
of only the coordinates.
In classical mechanics, or quantum mechanics in the Heisenberg picture, time
development also can be expressed in terms of the Hamiltonian using the bracket:
 
d ∂ ∂
A= + iH, A = A + i[H, A]
dt ∂t ∂t

(The middle expression with the commutator of ∂/∂t makes sense only in the quantum
case, and is not defined for the Poisson bracket.) Again, this general relation is
equivalent to the special cases, which in the classical limit are Hamilton’s equations
of motion:
dq m ∂H dpm ∂H
= i[H, q m] = , = i[H, pm ] = − m
dt ∂pm dt ∂q
The Hamiltonian has no explicit time dependence in the absence of time-dependent
nondyamical fields (external potentials whose time dependence is fixed by hand,
rather than by introducing the fields and their conjugate variables into the Hamilto-
nian). Consequently, time development is itself a symmetry: Time translations are
generated by the Hamiltonian; the ∂/∂t term in d/dt term can be dropped when
acting on operators without explicit time dependence.
Invariance of the theory under a symmetry means that the equations of motion
are unchanged under the transformation:
 0
dA dA0
=
dt dt
 A. COORDINATES 7

To apply our above translation of infinitesimal transformations into bracket language,

we define δ(d/dt) by    
d d d
δ A =δ A + δA
dt dt dt
In the quantum case we can write
    
d ∂
δ = iG, + iH ,
dt ∂t

which follows from the Jacobi identity using B = iG and C = ∂/∂t+iH, and inserting
A into the blank spaces of the commutators above. (The classical case can be treated
similarly, except that the time derivatives are not written as brackets.) We then find
that the generator of a symmetry transformation is conserved (constant), since
     
d ∂G dG
0=δ = −i − [G, H], = −i ,
dt ∂t dt

Excercise IA1.2
Show that the generators of the Galilean group are conserved, using the rela-
tion d/dt = ∂/∂t + i[H, ] for the hamiltonian H of a free particle. Solve the
equations of motion for x(t) and p(t) in terms of initial conditions, and substi-
tute into the expression for the generators to give an independent derivation
of their time independence.
In the cases where time dependence is not involved, symmetries can be treated
in almost exactly the same way either classically or quantum mechanically using the
corresponding bracket (Poisson or commutator), by using the properties that they
have in common. In particular, the fact that a symmetry generator G = λm (x)pm
is conserved means that we can solve for a component of p in terms of the constant
G, and substitute the result into the remaining equations of motion. For example,
translation invariance of a potential in a particular direction means that component
of the momentum is a constant (dp1 /dt = −∂H/∂q 1 = 0), rotational invariance
about some axis means that component of angular momentum is a constant (dJ/dt =
−∂H/∂θ = 0), etc.

2. Fermions
As we know experimentally, and we will see follows from relativistic field theory,
particles with half-integral spins obey Fermi-Dirac statistics. Let’s therefore consider
the classical limit of fermions: This will lead to generalizations of the concepts of
brackets and coordinates. Bosons obey commutation relations, such as [x, p] = ih̄;
8 I. GLOBAL

in the classical limit they just commute. Fermions obey anticommutation relations,
such as {ζ, ζ † } = h̄ for a single fermionic harmonic oscillator, where

{A, B} = AB + BA

is the anticommutator. So, in the truly classical (not semiclassical) limit they an-
ticommute, ζζ † + ζ † ζ = 0. Actually, the simplest case is a single real (hermitian)
fermion: Quantum mechanically, or semiclassically, we have

h̄ = {ξ, ξ} = 2ξ 2

while classically ξ 2 = 0. There is no analog for a single boson: [x, x] = x2 − x2 = 0.

This means that classical fermionic fields must be “anticommuting”: Two such objects
get a minus sign when pushed past each other. As a result, the product of two
fermionic quantities is bosonic, while fermionic times bosonic gives fermionic.
Excercise IA2.1
Show

[B, C] = [A, D] = 0 ⇒ [AB, CD] = 12 {A, C}[B, D] + 12 [A, C]{B, D}

To work with wave functions that are functions of anticommuting numbers, we

first must understand how to define general properties of functions of anticommuting
variables. For instance, given a single anticommuting variable ψ, we need to be able
to Taylor expand functions in ψ, e.g., to find a basis for the states. We therefore have
an anticommuting derivative ∂/∂ψ, satisfying
 2
∂
=0
∂ψ
from either anticommutativity or the fact functions of ψ terminate at first order in ψ.
We also need a ψ integral to define the inner product; indefinite integration turns out
to be enough. The most important property of the integral is integration by parts;
then, when acting on any function of ψ,
Z Z
∂ ∂
dψ =0 ⇒ dψ =
∂ψ ∂ψ
where the normalization is fixed for convenience. This also implies

δ(ψ) = ψ

Excercise IA2.2
Prove this is the most general possibility for anticommuting integration by
 A. COORDINATES 9

considering action of integration and differentiation on the most general func-

tion of ψ (which has only two terms).
In general, when Taylor expanding a function of anticommuting variables we must
preserve the statistics: If we Taylor expand a quantity that is defined to be commuting
(bosonic), then the coefficients of even powers of anticommuting variables will also be
commuting, while the coefficients of odd powers will be anticommuting (fermionic),
to maintain the commuting nature of that term (the product of the variables and
coefficient). Similarly, when expanding an anticommuting quantity the coefficients of
even powers will also be anticommuting, while for odd powers it will be commuting.
We can now consider operators that depend on both commuting (φm ) and anti-
commuting (ψ µ ) classical variables,

ΦM = (φm , ψ µ )

Classically they satisfy the “graded” commutation relations (anticommutation if both

elements are fermionic, commutation otherwise), not to be confused with the Poisson
bracket,

classically [ΦM , ΦN } = 0 : φm φn −φnφm = φm ψ ν −ψ ν φm = ψ µ ψ ν +ψ ν ψ µ = 0

This relation is then generalized to the the graded quantum mechanical commutator
or Poisson bracket by

[ΦM , ΦN } = h̄ΩM N , ΩM N ΩP N = δPM

where Ω is constant, hermitian, and “graded antisymmetric”:

Ω(M N ] = 0 : Ω(mn) = Ω[µν] = Ωmν + Ωνm = 0

For the standard normalization of canonically conjugate pairs of bosons

φm = φiα = (q i, pi )

and self-conjugate fermions, we choose

0 i


Ωµν = δ µν ; Ωiα,jβ = δ ij C αβ , C αβ = −i 0

Because of signs resulting from ordering anticommuting quantities, we define

derivatives unambiguously by their action from the left:
∂
ΦN = δM
N
∂ΦM
10 I. GLOBAL

The general Poisson bracket then can be written as

←
∂ ∂
semiclassically [A, B} ≡ −A M ΩN M B
∂Φ ∂ΦN
Since derivatives are normally defined to act from the left, there is a minus sign from
pushing the first derivative to the left if A and that particular component of ∂/∂ΦM
are both fermionic.
Excercise IA2.3
Let’s examine some properties of fermionic oscillators:
a For a single set of harmonic oscillators we have

{a, a†} = 1, {a, a} = {a†, a† } = 0

Show that the “number operator” a† a has the property

†
{a, eiπa a } = 0

(Hint: Since this system has only 2 states, the easiest way is to check the
action on those states.)
b Define eigenstates of the annihilation operator (“coherent states”) by

a|ζi = ζ|ζi

where ζ is anticommuting. Show that this implies

∂ † 0 † †
a† |ζi = − |ζi, |ζi = eζa |0i, eζ a |ζi = |ζ + ζ 0 i, xa a|ζi = |xζi,
∂ζ
Z
0 ζ*ζ 0
hζ|ζ i = e , 1= dζ*dζ e−ζ*ζ |ζihζ|

Define wave functions in this space, Ψ(ζ) = hζ|Ψi. Taylor expand them in
ζ, and compare this to the usual two-component representation using |0i and
a†|0i as a basis.
c Define the “supertrace” by
Z
str(A) = dζ*dζ e−ζ*ζ hζ|A|ζi

Find the relation between any operator in this space and a 2×2 matrix, and
find the expression for the supertrace in terms of this matrix.
d Repeat part b for the bosonic oscillator ([a, a†] = 1), where the Hilbert space
is infinite-dimensional, paying attention to signs, etc. Show that the analog
of part c defines the ordinary trace.
 A. COORDINATES 11

e For two sets of fermionic oscillators, we define

{a1, a†1 } = {a2 , a†2 } = 1, other { , } = 0

Show that the new operators

†
ã1 = a1, ã2 = eiπa 1 a1
a2

(and their Hermitian conjugates) are equivalent to the original ones except
that one set of the new oscillators commutes (not anticommutes) with the
other ([ã1, ㆠ2 ] = 0, etc.), even though each set satisfies the same anticom-
mutation relations with itself ({ã1 , ㆠ1} = 1, etc.). Thus, choice of statistics
is relevant only for particles in the same state: at most one fermion, but
unlimited bosons. (This change of oscillator basis is called a “Klein trans-
formation”. It can be useful for discrete sets of oscillators, but not for those
labeled by a continuous parameter, because of the discontinuity in the com-
mutation relations when the two labels are equal.)

3. Lie algebra
Since the same symmetries can be expressed in terms of different kinds of brackets
for classical and quantum theories, it can be useful to work with just those properties
that the Poisson bracket and commutator have in common, i.e., those that involve
only the bracket of two operators, not just their ordinary product:

[αA + βB, C] = α[A, C] + β[B, C] for numbers α, β (distributivity)

[A, B] = −[B, A] (antisymmetry)

[A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0 (Jacobi identity)

with similar expressions (differing only by signs) for anticommutators or mixed com-
mutators and anticommutators.
Excercise IA3.1
Find the generalizations of the Jacobi identity using also anticommutators,
corresponding to the cases where 2 or 3 of the objects involved are considered
as fermionic instead of bosonic.
These properties also give an abstract definition of a form of multiplication, the
“Lie bracket”, which defines a “Lie algebra”. (The first property is true of algebras
in general.) Other Lie brackets include those defined by another, associative, form of
12 I. GLOBAL

multiplication, such as matrix multiplication, or operator (infinite matrix) multipli-

cation as in quantum mechanics: In those cases we can write [A, B] = AB − BA, and
use the usual properties of multiplication (distributivity and associativity) to derive
the properties of the Lie bracket. (Another familiar example in physics is the “cross”
product for three-vectors; however, this can also be expressed in terms of matrix
multiplication.) The most important use of Lie algebras for physics is for describing
(continuous) infinitesimal transformations, especially those describing symmetries.
Excercise IA3.2
Using only the commutation relations of the generators of the Galilean group
(excercise IA1.1), check all the Jacobi identities.
For describing transformations, we can also think of the bracket as a derivative:
The “Lie derivative” of B with respect to A is defined as

LA B = [A, B]

As a consequence of the properties of the Lie bracket, this derivative satisfies the
usual properties of a derivative, including the Leibniz rule. (In fact, for coordinate
transformations the Lie derivative is really a derivative with respect to the coordi-
nates.)
We can now define finite transformations by exponentiating infinitesimal ones:

A0 ≈ (1 + iLG)A ⇒ A0 = lim(1 + iLG)1/ A = eiLG A

→0

In cases where we have [A, B] = AB − BA, we can also write

eiLG A = eiGAe−iG

This follows from replacing G on both sides with αG and taking the derivative with
respect to α, to see that both satisfy the same differential equation with the same
initial condition. We then can recognize this as the way transformations are performed
in quantum mechanics: A linear transformation that preserves the Hilbert-space inner
product must be unitary, which means it can be written as the exponential of an
antihermitian operator.
Just as infinitesimal transformations define a Lie algebra with elements A, finite
ones define a “Lie group” with elements

g = eiG

The multiplication law of two group elements follows from the fact the product of
two exponentials can be expressed in terms of multiple commutators:
1
eAeB = eA+B+ 2 [A,B]+...
 A. COORDINATES 13

We now have the mathematical properties that define a group, namely: (1) a product,
so that for two group elements g1 and g2 , we can define g1 g2 , which is another element
of the group (closure), (2) an identity element, so gI = Ig = g, (3) an inverse, where
gg −1 = g −1 g = I, and (4) associativity, g1 (g2 g3 ) = (g1 g2 )g3 . In this case the identity
is 1 = e0, while the inverse is (eA)−1 = e−A.
Since the elements of a Lie algebra form a vector space (we can add them and
multiply by numbers), it’s useful to define a basis:
i
G = αi Gi ⇒ g = eiα Gi

The parameters αi then also give a set of (redundant) coordinates for the Lie group.
(Previously they were required to be infintesimal, for infintesimal transformations;
now they are finite, but may be periodic, as determined by topological considerations
that we will mostly ignore.) Now the multiplication rules for both the algebra and
the group are given by those of the basis:

[Gi , Gj ] = −ifij k Gk

for the (“structure”) constants fij k = −fji k , which define the algebra/group (but are
ambiguous up to a change of basis). They satisfy the Jacobi identity

[[G[i, Gj ], Gk] ] = 0 ⇒ f[ij l fk]l m = 0

A familiar example is SO(3) (SU(2)), 3D rotations, where fij k = ijk if we use Gi =

1
2 ijk Jjk .

Another useful concept is a “subgroup”: If some subset of the elements of a group

also form a group, that is called a “subgroup” of the original group. In particular, for
a Lie group the basis of that subgroup will be a subset of some basis for the original
group. For example, for the Galilean group Jij generate the rotation subgroup.
Excercise IA3.3
Let’s examine the subgroup of the Galilean group describing (spatial) coor-
dinate transformations — rotations and spatial translations:
a Show that the infinitesimal transformations are given by

δxi = xj j i + ˆi , ij = −ji

where the ’s are constants.

b Exponentiate to find the finite transformations

x0i = xj Λj i + Λ̂i
14 I. GLOBAL

c Show that Λi j must satisfy

Λi k Λj l δkl = δij

both to preserve the scalar product, and as a consequence of exponentiating.

(Hint: Use matrix notation, and find the equivalent relation between Λ and
Λ−1 .)
d Show that the last equation implies det Λ = ±1, while exponentiating can
give only det Λ = 1 (since +1 can’t change continuously to −1). What is the
physical interpretation of a transformation with det Λ = −1? (Hint: Consider
a simple example.)
These results can be generalized to include anticommutators: When some of the
basis elements Gi are fermionic, the corresponding parameters αi are anticommuting
numbers, the structure constants are defined by [Gi , Gj }, etc.. Then G = αi Gi is
bosonic term by term, as is g, so bosons transform into bosons and fermions into
fermions, but Taylor expansion in the α’s will have both bosonic and fermionic co-
efficients. (For example, for δA = B, if A is bosonic, then so is B, but if also  is
fermionic, then B will also be fermionic.)
For some purposes it is more convenient to absorb the “i” in the infinitesimal
transformation into the definition of the generator:

G → −iG ⇒ δA = [G, A] = LG A, g = eG , [Gi , Gj ] = fij k Gk

This affects the reality properties of G: In particular, if g is unitary (gg† = I), as

usually required in quantum mechanics, g = eiG makes G hermitian (G = G† ), while
g = eG makes G antihermitian (G = −G† ). In some cases anithermiticity can be
an advantage: For example, for translations we would then have Pi = ∂i and for
rotations Jij = x[i∂j] , which is more convenient since we know the i’s in these (and
any) coordinate transformations must cancel anyway. On the other hand, the U(1)
transformations of electrodynamics (on the wave function for a charged particle) are
just phase transformations g = eiθ (where θ is a real number), so clearly we want the
explicit i; then the only generator has the representation Gi = 1. In general we’ll find
that for our purposes absorbing the i’s into the generators is more convenient for just
spacetime symmetries, while explicit i’s are more convenient for internal symmetries.
 A. COORDINATES 15

4. Relativity
The Hamiltonian approach singles out the time coordinate. In relativistic theories
time can be treated on equal footing with space, and it is useful to take advantage of
this fact, so that the full Poincaré invariance is manifest. So, we treat the time t and
spatial position xi together as a four-vector (or D-vector in D−1 space and 1 time
dimension)
xm = (x0 , xi) = (t, xi)
where m = 0, 1, ..., 3 (or D−1), i = 1, 2, 3. Since the energy E and three-momentum
pi are canonically conjugate to them,

[pi , xj ] = −iδ ij , [E, t] = +i

we define the 4-momentum as

pm = (E, pi ) = η mn pn , pm = ηmn pn ; [pm , xn ] = −iη mn, [pm , xn] = −iδm

n

where we raise and lower indices with the “Minkowski metric”, in an “orthonormal
basis”,
 0 1 2 3
0 −1 0 0 0
 
1 0 1 0 0 
ηmn =  
 ⇒ p0 = −p0 = −E

2 0 0 1 0 
3 0 0 0 1
in four spacetime dimensions, with obvious generalizations to higher dimensions.
(Sometimes the metric with signs + − −− is used; we prefer − + ++ because it
is more convenient for quantum calculations.) Therefore, we now distinguish upper
and lower indices in general: At least for position and momentum, the upper-indexed
xm and pm have the usual physical interpretation (so xm and pm have extra signs).
This is consistent with our previous nonrelativistic notation, since 3-vector indices do
not change sign upon raising or lowering.
Of course, we could have done that much nonrelativistically. Relativity is a
symmetry of kinematics and dynamics: In particular, a free, spinless, relativistic
particle is completely described by the constraint

p2 + m2 = 0

where we define the covariant square

p2 = pm pm = pm pn ηmn = −(p0 )2 + (p1 )2 + (p2 )2 + (p3 )2

16 I. GLOBAL

Our relativistic symmetry must leave this constraint invariant: Thus the metric de-
fines the norm of a vector (and an invariant inner product). Therefore, to preserve
Lorentz invariance it is important that we contract only an upper index with a lower
index. For similar reasons, we have
∂
∂m = , ∂m x n = δ m
n
∂xm
so quantum mechanically pm = −i∂m .
Unlike the positive-definite nonrelativistic norm of a 3-vector V i , for an arbitrary
4-vector V m we can have
  
<  timelike
2
V = 0: lightlike/null
  
> spacelike
In particular, the 4-momentum is timelike for massive particles (m2 > 0) and lightlike
for massless ones (while “tachyons”, with spacelike momenta and m2 < 0, do not exist,
for reasons that are most clear from quantum field theory).
The quantum mechanics will be described later, but the result is that this con-
straint can be used as the wave equation. The main qualitative distinction from the
nonrelativistic case in the constraint

nonrelativistic : − 2mE + p~ 2 = 0
relativistic : − E 2 + m2 + ~p 2 = 0

is that the equation for the energy E ≡ p0 is now quadratic, and thus has two
solutions:
p
p0 = ±ω, ω = (pi )2 + m2
Later we’ll see how the second solution is interpreted as an “antiparticle”. We also
use (natural/Planck) units c = 1, so length and duration are measured in the same
units; c then appears only as a parameter for defining nonrelativistic expansions and
limits.
The translations and Lorentz transformations make up the Poincaré group, the
symmetry that defines special relativity. (The Lorentz group in D−1 space and 1
time dimension is the “orthogonal” group “O(D−1,1)”. The “proper” Lorentz group
“SO(D−1,1)”, where the “S” is for “special”, transforms the coordinates by a matrix
whose determinant is 1. The Poincaré group is ISO(D−1,1), where the “I” stands
for “inhomogeneous”.) For the spinless particle they are generated by coordinate
transformations GI = (Pa , Jab):

Pa = pa , Jab = x[a pb]

 A. COORDINATES 17

(where also a, b = 0, ..., 3). Then the fact that the physics of the free particle is
invariant under Poincaré transformations is expressed as

[Pa , p2 + m2] = [Jab , p2 + m2] = 0

Writing an arbitrary infinitesimal transformation as a linear combination of the gen-

erators, we find
δxm = xn nm + ˆm , mn = −nm
where the ’s are constants. Note that antisymmetry of mn does not imply antisym-
metry of m n = mp η pn, because of additional signs. (Similar remarks apply to Jab.)
Exponentiating to find the finite transformations, we have

x0m = xnΛn m + Λ̂m , Λm p Λnq ηpq = ηmn

The same Lorentz transformations apply to pm , but the translations do not affect
it. The condition on Λ follows from preservation of the Minkowski norm (or inner
product), but it is equivalent to the antisymmetry of m n by exponentiating Λ = e
(compare excercise IA3.3).
Since dxa pa is invariant under the coordinate transformations defined by the Pois-
son bracket (the chain rule, since effectively pa ∼ ∂a ), it follows that the Poincaré
invariance of p2 is equivalent to the invariance of the line element

ds2 = −dxm dxnηmn

which defines the “proper time” s. Spacetime with this indefinite metric is called
“Minkowski space”, in contrast to the “Euclidean space” with positive definite metric
used to describe nonrelativistic length measured in just the three spatial dimensions.
For the massive case, we also have
dxa
pa = m
ds
For the massless case ds = 0: Massless particles travel along lightlike lines. However,
we can define a new parameter τ such that
dxa
pa =
dτ
is well-defined in the massless case. In general, we then have

s = mτ

While this fixes τ = s/m in the massive case, in the massless case it instead restricts
s = 0. Thus, proper time does not provide a useful parametrization of the world
18 I. GLOBAL

line of a classical massless particle, while τ does: For any piece of such a line, dτ is
given in terms of (any component of) pa and dxa . Later we’ll see how this parameter
appears in relativistic classical mechanics, and is useful for quantum mechanics and
field theory.
Excercise IA4.1
The relation between x and p is closely related to the Poincaré conservation
laws:
a Show that
dPa = dJab = 0 ⇒ p[a dxb] = 0

and use this to prove that conservation of P and J imply the existence of a
parameter τ such that pa = dxa /dτ .
b Consider a multiparticle system (but still without spin) where some of the
particles can interact only when at the same point (i.e., by collision; they
P P
act as free particles otherwise). Define Pa = pa and Jab = x[a pb] as the
sum of the individual momenta and angular momenta. Show that momentum
conservation implies angular momentum conservation,

∆Pa = 0 ⇒ ∆Jab = 0

where “∆” refers to the change from before to after the collision(s).
Special relativity can also be stated as the fact that the only physically observable
quantities are those that are Poincaré invariant. (Other objects, such as vectors,
depend on the choice of reference frame.) For example, consider two spinless particles
that interact by collision, producing two spinless particles (which may differ from the
originals). Without loss of generality, we can describe this process in terms of just
the momenta. (Quantum mechanically, this is automatically a complete description;
classically, the position is found by p = dx/dτ .) All invariants can be expressed in
terms of the masses and the “Mandelstam variables” (not to be confused with time
and proper time)

s = −(p1 + p2 )2, t = −(p1 − p3 )2 , u = −(p1 − p4 )2

where we have used momentum conservation, which shows that even these three
quantities are not independent:

X
4
p2I = −m2I , p1 + p2 = p3 + p4 ⇒ s+t+u= m2I
I=1
 A. COORDINATES 19

(The explicit index now labels the particle, for the process 1+2→3+4.) The simplest
reference frame to describe this interaction is the center-of-mass frame (actually the
center of momentum, where the two 3-momenta cancel). In that Lorentz frame, using
also rotational invariance, momentum conservation, and the mass-shell conditions, the
momenta can be written in terms of these invariants as
p1 = √1 ( 1 (s
s 2
+ m21 − m22 ), λ12, 0, 0)
p2 = √ ( 1 (s
1
s 2
+ m22 − m21 ), −λ12, 0, 0)
p3 = √1 ( 1 (s
s 2
+ m23 − m24 ), λ34 cos θ, λ34 sin θ, 0)
p4 = 1
s 2
+ m24 − m23 ), −λ34 cos θ, −λ34 sin θ, 0)
√ ( 1 (s
P
s2 + 2st − ( m2I )s + (m21 − m22)(m23 − m24 )
cos θ =
4λ12 λ34
λ2IJ = 14 [s − (mI + mJ )2 ][s − (mI − mJ )2]
The “physical region” of momentum space is then given by s ≥ (m1 + m2 )2 and
(m3 + m4 )2, and |cos θ| ≤ 1.
Excercise IA4.2
Derive the above expressions for the momenta in terms of invariants in the
center-of-mass frame.
Excercise IA4.3
Find the conditions on s, t and u that define the physical region in the case
where all masses are equal.
For some purposes it will prove more convenient to use a “lightcone basis”

 + − 2 3
+ 0 −1 0 0
 
− −1 0 0 0 
p± = √1 (p0 ±p1 )
2
⇒ ηmn = 
,
 p2 = −2p+ p− +(p2 )2 +(p3)2
2 0 0 1 0
3 0 0 0 1
and similarly for the “lightcone coordinates” (x±, x2, x3 ). (“Lightcone” is an unfor-
tunate but common misnomer, having nothing to do with cones in most usages.) In
this basis the solution to the mass-shell condition p2 + m2 = 0 can be written as
(pi )2 + m2
p± = −p∓ =
2p∓
(where now i = 2, 3), which more closely resembles the nonrelativistic expression.
(Note the change on indices + ↔ − upon raising and lowering.) A special lightcone
basis is the “null basis”,
p± = √1 (p0
2
± p1 ), pt = √1 (p2
2
− ip3), p̄t = √1 (p2
2
+ ip3 )
20 I. GLOBAL

 + − t t̄ 
+ 0 −1 0 0
 
−  −1 0 0 0 
,
⇒ ηmn =   p2 = −2p+ p− + 2pt p̄t
t  0 0 0 1
t̄ 0 0 1 0
where the square of a vector is linear in each component. (We often use “ ” to
indicate complex conjugation.)
Excercise IA4.4
Show that for p2 + m2 = 0 (m2 ≥ 0, pa 6= 0), the signs of p+ and p− are
always the same as the sign of the canonical energy p0 .
Excercise IA4.5
Consider the Poincaré group in 1 extra space dimension (D space, 1 time)
for a massless particle. Interpret p+ as the mass, and p− as the energy.
Show that the constraint p2 = 0 gives the usual nonrelativistic expression
for the energy. Show that the subgroup of the Poincaré group generated by
all generators that commute with p+ is the Galilean group (in D−1 space
and 1 time dimensions). Now nonrelativistic mass conservation is part of
momentum conservation, and all the Galilean transformations are coordinate
transformations. Also, positivity of the mass is related to positivity of the
energy (see excercise IA4.4).

5. Discrete: C, P, T
By considering only symmetries than can be obtained continuously from the iden-
tity (Lie groups), we have missed some important symmetries: those that reflect some
of the coordinates. It’s sufficient to consider a single reflection of a spacelike axis,
and one of a timelike axis; all other reflections can be obtained by combining these
with the continuous (“proper, orthochronous”) Lorentz transformations. (Spacelike
and timelike vectors can’t be Lorentz transformed into each other, and reflection of
a lightlike axis won’t preserve p2 + m2.) Also, the reflection of one spatial axis can
be combined with a π rotation about that axis, resulting in reflection of all three
spatial coordinates. (Similar generalizations hold for higher dimensions. Note that
the product of an even number of reflections about different axes is a proper rotation;
thus, for even numbers of spatial dimensions reflections of all spatial coordinates are
proper rotations, even though the reflection of a single axis is not.) The reversal of the
spatial coordinates is called “parity (P)”, while that of the time coordinate is called
“time reversal” (“T”; actually, for historical reasons, to be explained shortly, this is
 A. COORDINATES 21

usually labeled “CT”.) These transformations have the same effect on the momen-
tum, so that the definition of the Poisson bracket is also preserved. These “discrete”
transformations, unlike the proper ones, are not symmetries of nature (except in cer-
tain approximations): The only exception is the transformation that reflects all axes
(“CPT”).
While the metric ηmn is invariant under all Lorentz transformations (by defini-
tion), the “Levi-Civita tensor”

mnpq totally antisymmetric, 0123 = −0123 = 1

is invariant under only proper Lorentz transformations: It has an odd number of

space indices and of time indices, so it changes sign under parity or time reversal.
Consequently, we can use it to define “pseudotensors”: Given “polar vectors”, whose
signs change as position or momentum under improper Lorentz transformations, and
scalars, which are invariant, we can define “axial vectors” and “pseudoscalars” as

Va = abcd B b C c Dd , φ = abcd AaB b C c Dd

which get an extra sign change under such transformations (P or CT, but not CPT).
There is another such “discrete” transformation that is defined on phase space,
but which does not affect spacetime. It changes the sign of all components of the
momentum, while leaving the spacetime coordinates unchanged. This transformation
is called “charge conjugation (C)”, and is also only an approximate symmetry in
nature. (Quantum mechanically, complex conjugation of the position-space wave
function changes the sign of the momentum.) Furthermore, it does not preserve the
Poisson bracket, but changes it by an overall sign. (The misnomer “CT” for time
reversal follows historically from the fact that the combination of reversing the time
axis and charge conjugation preserves the sign of the energy.) The physical meaning
of this transformation is clear from the spacetime-momentum relation of relativistic
classical mechanics p = m dx/ds: It is proper-time reversal, changing the sign of s.
The relation to charge follows from “minimal coupling”: The “covariant momentum”
m dx/ds = p + qA (for charge q) appears in the constraint (p + qA)2 + m2 = 0 in an
electromagnetic background; p → −p then has the same effect as q → −q.
In the previous subsection, we mentioned how negative energies were associated
with “antiparticles”. Now we can better see the relation in terms of charge conjuga-
tion. Note that charge conjugation, since it only changes the sign of τ but does not
effect the coordinates, does not change the path of the particle, but only how it is
parametrized. This is also true in terms of momentum, since the velocity is given by
22 I. GLOBAL

pi /p0 . Thus, the only observable property that is changed is charge; spacetime prop-
erties (path, velocity, mass; also spin, as we’ll see later) remain the same. Another
way to say this is that charge conjugation commutes with the Poincaré group. One
way to identify an antiparticle is that it has all the same kinematical properties (mass,
spin) as the corresponding particle, but opposite sign for internal quantum numbers
(like charge). (Another way is pair creation and annihilation: See subsection IIIB5
below.) Quantum mechanically, we can identify a particle with its antiparticle by
requiring the wave function or field to be invariant under charge conjugation: For
example, for a scalar field (spinless particle), we have the reality condition

φ(x) = φ*(x)

or in momentum space, by Fourier transformation,

φ̃(p) = [φ̃(−p)]*

which implies the particle has charge zero (neutral).

All these transformations are summarized in the table:

C CT P T CP P T CP T
s − + + − − − +
t + − + − + − −
~x + + − + − − −
E − − + + − + −
p~ − + − − + + −

(The upper-left 3×3 matrix contains the definitions, the rest is implied.)
However, from the point of view of the “particle” there is some kind of kinematic
change, since the proper time has changed sign: If we think of the mechanics of a
particle as a one-dimensional theory in τ space (the worldline), where x(τ ) (as well
as any such variables describing spin or internal symmetry) is a wave function or field
on that space, then τ → −τ is T on that one-dimensional space. (The fact we don’t
get CT can be seen when we add additional variables: For example, if we describe
internal U(N) symmetry in terms of creation and annihilation operators a†i and ai,
then C mixes them on both the worldline and spacetime. So, on the worldline we
have the “pure” worldline geometric symmetry CT times C = T.) Thus, in terms of
“zeroth quantization”,
worldline T ↔ spacetime C
 A. COORDINATES 23

On the other hand, spacetime P and CT are simply internal symmetries with respect
to the worldline (as are proper, orthochronous Poincaré transformations).

6. Conformal
Although Poincaré transformations are the most general coordinate transforma-
tions that preserve the mass condition p2 + m2 = 0, there is a larger group, the
“conformal group”, that preserves this constraint in the massless case. Transforma-
tions λ that satisfy
[λa (x)pa , p2 ] = ζ(x)p2

for some ζ also preserve p2 = 0, although they don’t leave p2 invariant. Equivalently,
we can look for coordinate transformations that scale

dx02 = ξ(x)dx2

Excercise IA6.1
Find the conformal group explicitly in two dimensions, and show it’s infinite
dimensional (not just the SO(2,2) described below). (Hint: Use lightcone
coordinates.)
This symmetry can be made manifest by starting with a space with one extra
space and time dimension:

y A = (y a, y +, y −) ⇒ y 2 = y Ay B ηAB = (y a )2 − 2y + y −

where (y a)2 = y ay b ηab uses the usual D-dimensional Minkowski-space metric ηab ,
and the two additional dimensions have been written in a lightcone basis (not to
be confused for the similar basis that can be used for the Minkowski metric itself).
With respect to this metric, the original SO(D−1,1) Lorentz symmetry has been
enlarged to SO(D,2). This is the conformal group in D dimensions. However, rather
than also preserving (D+2)-dimensional translation invariance, we instead impose the
constraint and invariance
y 2 = 0, δy A = ζy A

This reduces the original space to the “projective” (invariant under the ζ scaling)
lightcone (which in this case really is a cone).
These two conditions can be solved by

y A = ewA, wA = (xa , 1, 12 xaxa )

24 I. GLOBAL

Projective invariance then means independence from e (y +), while the lightcone con-
dition has determined y −. y 2 = 0 implies y · dy = 0, so the simplest conformal
invariant is
dy 2 = (edw + wde)2 = e2dw2 = e2dx2

where we have used w2 = 0 ⇒ w · dw = 0. This means any SO(D,2) transformation

on y A will simply scale dx2 , and scale e2 in the opposite way:
 2
02 e
dx = dx2
e02
in agreement with the previous definition of the conformal group.
The explicit form of conformal transformations on xa = y a/y + now follows from
their linear form on y A, using the generators

GAB = y [ArB], [rA, y B ] = −iδAB

of SO(D,2) in terms of the momentum rA conjugate to y A . (These are defined the

same way as the Lorentz generators J ab = x[apb] .) For example, G+− just scales xa.
(Scale transformations are also known as “dilatations”.) We can also recognize G+a as
generating translations on xa. The only complicated transformations are generated
by G−a , known as “conformal boosts” (acceleration transformations). Since they
commute with each other (like translations), it’s easy to exponentiate to find the
finite transformations:
y 0 = eGy, G = vay [−∂ a]

for some constant D-vector v a (where ∂A ≡ ∂/∂y A). Since the conformal boosts act
as “lowering operators” for scale weight (+ → a → −), only the first three terms in
the exponential survive:

Gy − = 0, Gy a = v ay −, Gy + = v aya ⇒

y 0− = y −, y 0a = y a + v ay −, y 0+ = y + + v aya + 12 v 2y − ⇒
xa + 12 v ax2
x0a =
1 + v · x + 14 v 2x2
using xa = y a/y + , y − /y + = 12 x2 .
Excercise IA6.2
Make the change of variables to xa = y a/y + , e = y +, z = 12 y 2. Express
rA in terms of the momenta (pa , n, s) conjugate to (xa, e, z). Show that the
conditions y 2 = y ArA = r2 = 0 become z = en = p2 = 0 in terms of the new
variables.
 A. COORDINATES 25

Excercise IA6.3
Find the generator of infinitesimal conformal boosts in terms of xa and pa .
We actually have the full O(D,2) symmetry: Besides the continuous symmetries,
and the discrete ones of SO(D−1,1), we have a second “time” reversal (from our
second time dimension):
xa
y + ↔ −y − ⇒ xa ↔ − 1 2
2x

This transformation is called an “inversion”.

Excercise IA6.4
Show that a finite conformal boost can be obtained by performing a transla-
tion sandwiched between two inversions.
Excercise IA6.5
The conformal group for Euclidean space (or any spacetime signature) can be
obtained by the same construction. Consider the special case of D=2 for these
SO(D+1,1) transformations. (This is a subgroup of the 2D superconformal
group: See excercise IA6.1.) Use complex coordinates for the two “physical”
dimensions:
z = √12 (x1 + ix2 )

a Show that the inversion is

1
z↔−
z*
b Show that the conformal boost is (using a complex number also for the boost
vector)
z
z→
1 + v*z
Excercise IA6.6
Any parity transformation (reflection in a spatial axis) can be obtained from
any other by a rotation of the spatial coordinates. Similarly, when there is
more than one time dimension, any time reversal can be obtained from another
(but time reversal can’t be rotated into parity, since a timelike vector can’t be
rotated into a spacelike one). Thus, the complete orthogonal group O(m,n)
can be obtained from those transformations that are continuous from the
identity by combining them with 1 parity transformation and 1 time reversal
transformation (for mn6=0). For the conformal group, find the rotation (in
terms of an angle) that rotates between the two time directions, and express
its action on xa . Show that for angle π it produces a transformation that is
26 I. GLOBAL

the product of time reversal and inversion. Use this to show that inversion is
related to time reversal by finding the continuum of conformal transformations
that connect them.
Although conformal symmetry is not observed in nature, it is important in all
approaches to field theory: (1) First of all, it is useful in the construction of free the-
ories (see subsections IIB1-4 below). All massive fields can be described consistently
in quantum field theory in terms of coupling massless fields. Massless theories are a
subset of conformal theories, and some conditions on massless theories can be found
more easily by finding the appropriate subset of those on conformal theories. This is
related to the fact that the conformal group, unlike the Poincaré group, is “simple”:
It has no nontrivial subgroup that transforms into itself under the rest of the group
(like the way translations transform into themselves under Lorentz transformations).
(2) In interacting theories at the classical level, conformal symmetry is also important
in finding and classifying solutions, since at least some parts of the action are confor-
mally invariant, so corresponding solutions are related by conformal transformations
(see subsections IIIC5-7). Furthermore, it is often convenient to treat arbitrary theo-
ries as broken conformal theories, introducing fields with which the breaking is asso-
ciated, and analyze the conformal and conformal-breaking fields separately. This is
particularly true for the case of gravity (see subsections IXA7,B5,C2-3,XA3-4,B5-7).
(3) Within quantum field theory at the perturbative level, the only physical quantum
field theories are ones that are conformal at high energies (see subsection VIIIC1).
The quantum corrections to conformal invariance at high energy are relatively sim-
ple. (4) Beyond perturbation theory, the only quantum theories that are well defined
may be just the ones whose breaking of conformal invariance at low energy is only
classical (see subsections VIIC2-3,VIIIA5-6). Furthermore, the largest possible sym-
metry of a nontrivial S-matrix is conformal symmetry (or superconformal symmetry
if we include fermionic generators). (5) Self-duality (a generalization of a condition
that equates electric and magnetism fields) is useful for finding solutions to classical
field equations as well as simplifying perturbation theory, and is closely related to
“twistors” (see subsections IIB6-7,C5,IIIC4-7). In general, self-duality is related to
conformal invariance: For example, it can be shown that the free conformal theories
in arbitrary even dimensions are just those with (on-mass-shell) field strengths on
which self-duality can be imposed. (In arbitrary odd dimensions the free conformal
theories are just the scalar and spinor.)

REFERENCES
1 F.A. Berezin, The method of second quantization (Academic, 1966):
 A. COORDINATES 27

calculus with anticommuting numbers.

2 P.A.M. Dirac, Proc. Roy. Soc. A126 (1930) 360:
antiparticles.
3 E.C.G. Stückelberg, Helv. Phys. Acta 14 (1941) 588, 15 (1942) 23;
J.A. Wheeler, 1940, unpublished:
the relation of antiparticles to proper time.
4 S. Mandelstam, Phys. Rev. 112 (1958) 1344.
5 P.A.M. Dirac, Ann. Math. 37 (1936) 429;
H.A. Kastrup, Phys. Rev. 150 (1966) 1186;
G. Mack and A. Salam, Ann. Phys. 53 (1969) 174;
S. Adler, Phys. Rev. D6 (1972) 3445;
R. Marnelius and B. Nilsson, Phys. Rev. D22 (1980) 830:
conformal symmetry.
6 S. Coleman and J. Mandula, Phys. Rev. 159 (1967) 1251:
conformal symmetry as the largest (bosonic) symmetry of the S-matrix.
7 W. Siegel, Int. J. Mod. Phys. A 4 (1989) 2015:
equivalence between conformal invariance and self-duality in all dimensions.
28 I. GLOBAL

............................. .............................
............................. B. INDICES .............................

In the previous section we saw various spacetime groups (Galilean, Poincaré,

conformal) in terms of how they acted on coordinates. This not only gave them a
simple physical interpretation, but also allowed a direct relation between classical
and quantum theories. However, as we know from studying rotations in quantum
theory in terms of spin, we will often need to study symmetries of quantum theories
for which the classical analog is not so useful or perhaps even nonexistent.
We therefore now consider some general results of group theory, mostly for con-
tinuous groups. We use tensor methods, rather than the slightly more powerful but
greatly less convenient Cartan-Weyl-Dynkin methods. Much of this section should
be review, but is included here for completeness; it is not intended as a substitute for
a group theory course, but as a summary of those results commonly useful in field
theory.

1. Matrices
Matrices are defined by the way they act on some vector space; an n×n matrix
takes one n-component vector to another. Given some group, and its multiplication
table (which defines the group completely), there is more than one way to represent
it by matrices. Any set of matrices we find that has the same multiplication table as
the group elements is called a “representation” of that group, and the vector space on
which those matrices act is called the “representation space.” The representation of
the algebra or group in terms of explicit matrices is given by choosing a basis for the
vector space. If we include infinite-dimensional representations, then a representation
of a group is simply a way to write its transformations that is linear: ψ 0 = Mψ is
linear in ψ. More generally, we can also have a “realization” of a group, where the
transformations can be nonlinear. These tend to be more cumbersome, so we usually
try to make redefinitions of the variables that make the realization linear. A precise
definition of “manifest symmetry” is that all the realizations used are linear. (One
possible exception is “affine” or “inhomogeneous” transformations ψ 0 = M1 ψ + M2,
such as the usual coordinate representation of Poincaré transformations, since these
transformations are still very simple, because they are really still linear, though not
homogeneous.)
For convenience, we write matrices with a Hilbert-space-like notation, but unlike
Hilbert space we don’t necessarily associate bras directly with kets by Hermitian
conjugation, or even transposition. In general, the two spaces can even be different
 B. INDICES 29

sizes, to describe matrices that are not square; however, for group theory we are
interested only in matrices that take us from some vector space into itself, so they
are square. Bras have an inner product with kets, but neither necessarily has a norm
(inner product with itself): In general, if we start with some vector space, written
as kets, we can always define the “dual” space, written as bras, by defining such an
inner product. In our case, we may start with some representation of a group, in
terms of some vector space, and that will give us directly the dual representation. (If
the representation is in terms of unitary matrices, we have a Hilbert space, and the
dual representation is just the complex conjugate.)
So, we define column vectors |ψi with a basis |I i, and row vectors hψ| with a
basis hI |, where I = 1, ...,n to describe n×n matrices. The two bases have a relative
normalization defined so that the inner product gives the usual component sum:

|ψi = |I iψI , hχ| = χI hI |; hI |J i = δIJ ⇒ hχ|ψi = χI ψI ; hI |ψi = ψI , hχ|I i = χI

These bases then define not only the components of vectors, but also matrices:

M = |I iMI J hJ |, hI |M|J i = MI J

where as usual the I on the component (matrix element) MI J labels the row of the
matrix M, and J the column. This implies the usual matrix multiplication rules,
inserting the identity in terms of the basis,

I = |K ihK | ⇒ (MN)I J = hI |M|K ihK |N|J i = MI K NK J

Closely related is the definition of the trace,

tr M = hI |M|I i = MI I ⇒ tr(MN) = tr(NM)

(We’ll discuss the determinant later. The bra-ket notation is really just matrix nota-
tion written in a way to clearly distinguish column vectors, row vectors, and matrices.)
Thus, for example, we can easily translate transformation laws from matrix no-
tation into index notation just by using a basis for the representation space:

g|I i = |J igJ I , G|I i = |J iGJ I

G = αi Gi , δ|ψi = iG|ψi = |I iiαi (Gi )I J ψJ ⇒ δψI = iαi (Gi )I J ψJ

The dual space isn’t needed for this purpose. However, for any representation of a
group, the transpose
(M T )I J = MJ I
30 I. GLOBAL

of the inverse of those matrices also gives a representation of the group, since

g1 g2 = g3 ⇒ (g1 )T −1(g2 )T −1 = (g3 )T −1

[G1, G2 ] = G3 ⇒ [−GT1 , −GT2 ] = −GT3

This is the dual representation, which follows from defining the above inner product
to be invariant under the group:

δhψ|χi = 0 ⇒ δψ I = −iψ J αi (Gi )J I

The complex conjugate of a complex representation is also a representation, since

g1 g2 = g3 ⇒ g1 *g2 * = g3 *

[G1 , G2 ] = G3 ⇒ [G1 *, G2 *] = G3 *

From any given representation, we can thus find three others from taking the dual
and the conjugate: In matrix and index notation,

ψ 0 = gψ : ψI0 = gI J ψJ
ψ 0 = (g −1 )T ψ : ψ 0I = g −1 J I ψ J
.
ψ 0 = g*ψ : ψ 0. = g* .J ψ .
I I J
. . .
ψ 0 = (g −1 )† ψ : ψ 0I = g*−1 .I ψ J
J

since (g −1 )T , g*, and (g −1 )† (but not g T , etc.) satisfy the same multiplication algebra
as g, including ordering. We use up/down and dotted/undotted indices to denote
the transformation law of each type of index; contracting undotted up indices with
undotted down indices preserves the transformation law as indicated by the remaining
indices, and similarly for dotted indices. These four representations are not necessarily
independent: Imposing relations among them is how the classical groups are defined
(see subsections IB4-5 below).

2. Representations
For example, we always have the “adjoint” representation of a Lie group/algebra,
which is how the algebra acts on its own generators:

G = αi Gi , A = β iGi ⇒ δA = i[G, A] = β j αi fij k Gk

⇒ δβ i = −iβ k αj (Gj )k i , (Gi )j k = ifij k

 B. INDICES 31

This gives us two ways to represent the adjoint representation space: as either the
usual vector space, or in terms of the generators. Thus, we either use the matrix
A = β iGi (for arbitrary representation of the matrices Gi , or treating Gi as just
abstract generators), and write δA = i[G, A], or we can write A as a row vector,

hA| = β ihi | ⇒ δhA| = −ihA|G ⇒ δβ ihi | = −iβ k αj (Gj )k i hi |

The adjoint representation also provides a convenient way to define a (symmetric)

group metric invariant under the group, the “Cartan metric”:

ηij = trA(Gi Gj ) = −fik l fjl k

For “Abelian” groups the structure constants vanish, and thus so does this metric.
“Semisimple” groups are those where the metric is invertible (no vanishing eigenval-
ues). A “simple” group has no nontrivial subgroup that transforms into itself under
the rest of the group: Semisimple groups can be written as “products” of simple
groups. “Compact” groups are those where it is positive definite (all eigenvalues pos-
itive); they are also those for which the invariant volume of the group space is finite.
For simple, compact groups it’s convenient to choose a basis where

ηij = cA δij

for some constant cA (the “Dynkin index” for the adjoint representation). For some
general irreducible representation R of such a group the normalization of the trace is
cR
trR(Gi Gj ) = cR δij = ηij
cA
Now the proportionality constant cR /cA is fixed by the choice of R (only), since we
have already fixed the normalization of our basis.
In general, the cyclicity property of the trace implies, for any representation, that

0 = tr([Gi , Gj ]) = −ifij k tr(Gk )

so tr(Gi ) = 0 for semisimple groups. Similarly, we find

fijk ≡ fij l ηlk = i trA([Gi , Gj ]Gk )

is totally antisymmetric: For semisimple groups, this implies the total antisymmetry
of the structure constants fij k , up to factors (which are absent for compact groups in
a basis where ηij ∼ δij ). This also means the adjoint representation is its own dual.
32 I. GLOBAL

(For example, for the compact group SO(3), we have ηij = −ikl jlk = 2δij .) Thus,
we can write A in a third way, as a column vector

|Ai = |i iβi ≡ |i iβ j ηji

We can also do this for Abelian groups, by defining an invertible metric unrelated to
the Cartan metric: This is trivial for Abelian groups, since the generators themselves
are invariant, and thus so is any metric on them.
An identity related to the trace one is the normalization of the value kR of the
“Casimir operator” for any particular representation,

η ij Gi Gj = kR I

Its proportionality to the identity follows from the fact that it commutes with each
generator:
[η jk Gj Gk , Gi ] = −if j i k {Gj , Gk } = 0
using the antisymmetry of the structure constants. (Thus it takes the same value on
any component of an irreducible representation, since they are all related by group
transformations.) By tracing this identity, and contracting the trace identity,
cR
dA = trR(η ij Gi Gj ) = kR dR
cA
cR dA
⇒ kR =
cA dR
where dR ≡ trR(I) is the dimension of that representation.
Although quantum mechanics is defined on Hilbert space, which is a kind of com-
plex vector space, more generally we want to consider real objects, like spacetime
vectors. This restricts the form of linear transformations: Specifically, if we absorb
i’s as g = eG, then in such representations G itself must be real. These represen-
tations are then called “real representations”, while a “complex representation” is
one whose representation isn’t real in any basis. A complex representation space can
have a real representation, but a real representation space can’t have a complex rep-
resentation. In particular, coordinate transformations (of real coordinates) have only
real representations, which is why absorbing the i’s into the generators is a useful
convention there. For semisimple unitary groups, hermiticity of the generators of the
adjoint representation implies (using total antisymmetry of the structure constants
and reality of the Cartan metric) that the structure constants are real, and thus the
adjoint representation is a real representation. More generally, any real unitary rep-
resentation will have antisymmetric generators (G = G* = −G† ⇒ G = −GT ). If
 B. INDICES 33

the complex conjugate representation is the same as the original (same matrices up
to a similarity transformation g* = MgM −1 ), but the representation is not real, then
it is called “pseudoreal”. (An example is the spinor of SU(2), to be described in the
next section.)
For any representation g of the group, a transformation g → g0 gg0−1 on every
group element g for some particular group element g0 clearly maps the algebra to
itself, and preserves the multiplication rules. (Similar remarks apply to applying the
transformation to the generators.) However, the same is true for complex conjugation,
g → g*: Not only are the multiplication rules preserved, but for any element g
of that representation of the group, g* is also an element. (This can be shown,
e.g., by defining representations in terms of the values of all the Casimir operators,
contructed from various powers of the generators.) In quantum mechanics (where
the representations are unitary), the latter is called an “antiunitary transformation”.
Although this is a symmetry of the group, it cannot be reproduced by a unitary
transformation, except when the representation is (pseudo)real.
A very simple way to build a representation from others is by “direct sum”. If we
have two representations of a group, on two different spaces, then we can take their
direct sum by just putting one column vector on top of the other, creating a bigger
vector whose size (“dimension”) is the sum of that of the original two. Explicitly, if
0
we start with the basis |ι i for the first representation and |ι i for the second, then
0 0
the union (|ιi, |ι i) is the basis for the direct sum. (We can also write |I i = (|ιi, |ι i),
where ι = 1, ..., m; ι0 = 1, ..., n; I = 1, ..., m, m + 1, ..., m + n.) The group then acts
on each part of the new vector in the obvious way:
0 0 0 0
ψ = |ιiψι , χ = |ι iχι0 ; g|ι i = |κ igκ ι ,
g|ι i = |κ igκ0 ι
 
ι0 ψ
⇒ |Ψi = | iψι ⊕ | iχι0 = |ψi ⊕ |χi or (Ψ) =
ι
χ
 κ 
κ0 ι0 gι 0
g|Ψi = | igκ ψι ⊕ | igκ0 χι0 or (g) =
κ ι
0
0 gι0 κ
(We can replace the ⊕ with an ordinary + if we understand the basis vectors to be
now in a bigger space, where the elements of the first basis have zeros for the new
components on the bottom while those of the second have zeros for the new compo-
nents on top.) The important point is that no group element mixes the two spaces:
The group representation is block diagonal. Any representation that can be written
as a direct sum (after an appropriate choice of basis) is called “reducible”. For exam-
ple, we can build a reducible real representation from an irreducible complex one by
34 I. GLOBAL

just taking the direct sum of this complex representation with the complex conjugate
representation. Similarly, we can take direct sums of more than two representations.
A more useful way to build representations is by “direct product”. The idea there
is to take a colummn vector and a row vector and use them to construct a matrix,
where the group element acts simultaneously on rows according to one representation
0
and columns according to the other. If the two original bases are again |ιi and |ι i,
0
the new basis can also be written as |I i = |ιι i (I = 1, ..., mn). Explicitly,

0 0 0 0 0 0
|ψi = |ι i ⊗ |ι iψιι0 , g(|ι i ⊗ |ι i) = |κ i ⊗ |κ igκ ιgκ0 ι ⇒ gιι0 κκ = gι κ gι0 κ

or in terms of the algebra

0 0 0
Gιι0 κκ = Gι κ δι0 κ + δικ Gι0 κ

A familar example from quantum mechanics is rotations (or Lorentz transformations),

where the first space is position space (so ι is the continuous index x), acted on by
the orbital part of the generators, while the second space is finite-dimensional, and is
acted on by the spin part of the generators. Direct product representations are usually
reducible: They then can be written also as direct sums, in a way that depends on
the particulars of the group and the representations.
Consider a representation constructed by direct product: In matrix notation

Ĝi = Gi ⊗ I 0 + I ⊗ G0i

Using tr(A ⊗ B) = tr(A)tr(B), and assuming tr(Gi ) = tr(G0i ) = 0, we have

tr(Ĝi Ĝj ) = tr(I 0)tr(Gi Gj ) + tr(I)tr(G0i G0j )

For example, for SU(N) (see subsection IB4 below) we can construct the adjoint rep-
resentation from the direct product of the N-dimensional, “defining” representation
and its complex conjugate. (We also get a singlet, but it will not affect the result for
the adjoint.) In that case we find

cD 1
trA(Gi Gj ) = 2N trD (Gi Gj ) ⇒ =
cA 2N

For most purposes, we use trD (Gi Gj ) = δij (cD = 1) for SU(N), so cA = 2N.
 B. INDICES 35

3. Determinants
We now “review” some properties of determinants that will prove useful for the
group analysis of the following subsections. Determinants can be defined in terms of
the Levi-Civita tensor . As a consequence of its antisymmetry,

 totally antisymmetric, 12...n = 12...n = 1 ⇒ I1

J1 ...Jn I1 ...In = δ[J1
· · · δJInn ]

since each possible numerical index value appears once in each , so they can be
matched up with δ’s. By similar reasoning,
1

m! K1...Km J1 ...Jn−m
K1...Km I1 ...In−m I1
= δ[J1
· · · δJIn−m
n−m ]

where the normalization compensates for the number of terms in the summation.
This tensor is used to define the determinant:

det MI J = 1

n! J1 ...Jn
I1 ...In MI1 J1 · · · MIn Jn ⇒ J1 ...Jn MI1 J1 · · · MIn Jn = I1 ...In det M

since anything totally antisymmetric in n indices must be proportional to the  tensor.

This yields an explicit expression for the inverse:

(M −1 )J1 I1 = 1

(n−1)! J1 ...Jn
I1 ...In MI2 J2 · · · MIn Jn (det M)−1

From this follows a useful expression for the variation of the determinant:
∂
det M = (M −1 )J I det M
∂MI J
which is equivalent to
δ ln det M = tr(M −1 δM)
Replacing M with eM gives the often-used identity

δ ln det eM = tr(e−M δeM ) = tr δM ⇒ det eM = etr M

where we have used the boundary condition for M = 0. Finally, replacing M in

the last identity with ln(1 + L) and expanding both sides to order Ln gives general
expressions for determinants of n × n matrices in terms of traces:

det(1 + L) = etr ln(1+L) ⇒ det L = 1

n!
(tr L)n − 1
2(n−2)!
(tr L2 )(tr L)n−2 + · · ·

Excercise IB3.1
Use the definition of the determinant (and not its relation to the trace) to
show
det(AB) = det(A)det(B)
36 I. GLOBAL

These identities can also be derived by defining the determinant in terms of a

Gaussian integral. We first collect some general properties of (indefinite) Gaussian
integrals. The simplest such integral is
Z Z 2π Z ∞ Z ∞
d2 x −x2 /2 dθ −r 2 /2
e = dr re = du e−u = 1
2π 0 2π 0 0

Z Z D Z D/2
dD x dx −x2 /2 d2 x −x2 /2
e−x /2 =
2
⇒ D/2
√ e = e =1
(2π) 2π 2π
The complex form of this integral is
Z D
d z* dD z −|z|2
e =1
(2πi)D
√
by reducing to real parameters as z = (x + iy)/ 2. These generalize to integrals
involving a real, symmetric matrix S or a Hermitian matrix H as
Z Z D
dD x −xT Sx/2 −1/2 d z* dD z −z† Hz
e = (det S) , e = (det H)−1
(2π)D/2 (2πi)D

by diagonalizing the matrices, making appropriate redefinitions of the integration

variables, and identifying the determinant of a diagonal matrix. Alternatively, we
can use these integrals to define the determinant, and derive the previous definition.
The relation for the symmetric matrix follows from that for the Hermitian one by
separating z into its real and imaginary parts for the special case H = S. If we treat
z and z* as independent variables, the determinant can also be understood as the
Jacobian for the (dummy) variable change z → H −1 z, z* → z*. More generally, if
we define the integral by an appropriate limiting procedure or analytic continuation
(for convergence), we can choose z and z* to be unrelated (or even separate real
variables), and S and H to be complex.
Excercise IB3.2
Other properties of determinants can also be derived directly from the integral
definition:
a Find an integral expression for the inverse of a (complex) matrix M by using
the identity Z
∂ †
0= (zJ e−z M z )
∂zI

b Derive the identity δ ln det M = tr(M −1 δM) by varying the Gaussian defi-
nition of the (complex) determinant with respect to M.
 B. INDICES 37

An even better definition of the determinant is in terms of an anticommuting

integral (see subsection IA2), since anticommutativity automatically gives the anti-
symmetry of the Levi-Civita tensor, and we don’t have to worry about convergence.
We then have, for any matrix M,
Z
†
dD ζ † dD ζ e−ζ M ζ = det M

where ζ † can be chosen as the Hermitian conjugate of ζ or as an independent variable,

whichever is convenient. From the definition of anticommuting integration, the only
terms in the Taylor expansion of the exponential that contribute are those with the
product of one of each anticommuting variable. Total antisymmetry in ζ and in ζ †
then yields the determinant; we define “dD ζ † dD ζ” to give the correct normalization.
(The normalization is ambiguous anyway because of the signs in ordering the dζ’s.)
This determinant can also be considered a Jacobian, but the inverse of the commuting
result follows from the fact that the integrals are now really derivatives.
Excercise IB3.3
Divide up the range of a square matrix into two (not necessarily equal) parts:
In block form,  
A B
M=
C D
and do the same for the (commuting or anticommuting) variables used in
defining its determinant. Show that
 
A B
det = det D · det(A − BD−1 C) = det A · det(D − CA−1 B)
C D

a by integrating over one part of the variables first (this requires off-diagonal
changes of variables of the form y → y + Ox, which have unit Jacobian), or
b by first proving the identity
     
A B I BD−1 A − BD−1 C 0 I 0
=
C D 0 I 0 D D−1 C I

We then have, for any anti symmetric (even-dimensional) matrix A,

Z
d2D ξ e−ξ Aξ/2 = Pf A,
T
(Pf A)2 = det A

by the same method as the commuting case (again with appropriate definition of the
normalization of d2D ξ; the determinant of an odd-dimensional antisymmetric matrix
vanishes, since det M = det M T ). However, there is now an important difference: The
38 I. GLOBAL

“Pfaffian” is not merely the square root of the determinant, but itself a polynomial,
since we can evaluate it also by Taylor expansion:

Pf AIJ = 1
D!2D
I1 ...I2D AI1 I2 · · · AI2D−1 I2D

which can be used as an alternate definition. (Normalization can be checked by

examining a special case; the overall sign is part of the normalization convention.)

4. Classical groups
The rotation group in three dimensions can be expressed most simply in terms
of 2×2 matrices. This description is the most convenient for not only spin 1/2, but
all spins. This result can be extended to orthogonal groups (such as the rotation,
Lorentz, and conformal groups) in other low dimensions, including all those relevant
to spacetime symmetries in four dimensions.
There are an infinite number of Lie groups. Of the compact ones, all but a finite
number are among the “classical” Lie groups. These classical groups can be defined
easily in terms of (real or complex) matrices satisfying a few simple constraints. (The
remaining “exceptional” compact groups can be defined in a similar way with a little
extra effort, but they are of rather specialized interest, so we won’t cover them here.)
These matrices are thus called the “defining” representation of the group. (Sometimes
this representation is also called the “fundamental” representation; however, this term
has been used in slightly different ways in the literature, so we will avoid it.) These
constraints are a subset of:

volume: Special: det(g) = 1


 hermitian: Unitary: gΥg † = Υ (Υ† = Υ)

metric: (η T = η)
 (anti)symmetric: Orthogonal: gηg T = η
Symplectic: gΩg T = Ω (ΩT = −Ω)

Real: g* = ηgη −1
reality:
pseudoreal (*): g* = ΩgΩ−1

where g is any matrix in the defining representation of the group, while Υ, η, Ω are
group “metrics”, defining inner products (while the determinant defines the volume,
as in the Jacobian). For the compact cases Υ and η can be chosen to be the identity,
but we will also consider some noncompact cases. (There are also some uninteresting
variations of “Special” for complex matrices, setting the determinant to be real or its
magnitude to be 1.)
 B. INDICES 39

Excercise IB4.1
Write all the defining constraints of the classical groups (S, U, O, Sp, R,
pseudoreal) in terms of the algebra rather than the group.
Note the modified definition of unitarity, etc. Such things are also encountered
in quantum mechanics with ghosts, since the resulting Hilbert space can have an
indefinite metric. For example, if we have a finite-dimensional Hilbert space where
the inner product is represented in terms of matrices as

hψ|χi = ψ † Υχ

then “observables” satisfy a “pseudohermiticity” condition

hψ|Hχi = hHψ|χi ⇒ ΥH = H † Υ

and unitarity generalizes to

hUψ|Uχi = hψ|χi ⇒ U † ΥU = Υ

Similar remarks apply when replacing the Hilbert-space “sesquilinear” (vector times
complex conjugate of vector) inner product with a symmetric (orthogonal) or anti-
symmetric (symplectic) bilinear inner product. An important example is when the
wave function carries a Lorentz vector index, as expected for a relativistic description
of spin 1; then clearly the time component is unphysical.
The groups of matrices that can be constructed from these conditions are then:

GL(n,C) [SL(n,C)] U: [S]U(n+,n−) U R *

O: [S]O(n,C)
O [S]O(n+,n−) SO*(2n)
Sp: Sp(2n,C)
Sp Sp(2n) USp(2n+,2n− )
R: GL(n) [SL(n)]
*: [S]U*(2n)

Of the non-determinant constraints, in the first column we applied none (“GL” means
“general linear”, and “C” refers to the complex numbers; the real numbers “R” are
implicit); in the second column we applied one; in the third column we applied three,
since two of the three types (unitarity, symmetry, reality) imply the third. (The
corresponding groups with unit determinant, when distinct, are given in brackets.)
These square matrices are of size n, n+ +n−, 2n, or 2n+ +2n− , as indicated. n+ and
n− refer to the number of positive and negative eigenvalues of the metric Υ or η.
O(n) differs from SO(n) by including “parity”-type transformations, which can’t be
40 I. GLOBAL

obtained continuously from the identity. (SSp(2n) is the same as Sp(2n).) For this
reason, and also for studying “topological” properties, for finite transformations it
is sometimes more useful to work directly with the group elements g, rather than
parametrizing them in terms of algebra elements as g = eiG. U(n) differs from SU(n)
(and similarly for GL(n) vs. SL(n)) only by including a U(1) group that commutes
with the SU(n): Although U(1) is noncompact (it consists of just phase transforma-
tions), a compact form of it can be used by requiring that all “charges” are integers
(i.e., all representations transform as ψ 0 = eiqθ ψ for group parameter θ, where q is an
integer defining the representation).
Of these groups, the compact ones are just SU(n), SO(n) (and O(n)), and USp(2n)
(all with n−=0). The compact groups have an interesting interpretation in terms of
various number systems: SO(n) is the unitary group of n×n matrices over the real
numbers, SU(n) is the same for the complex numbers, and USp(2n) is the same for
the quaternions. (Similar interpretations can be made for some of the noncompact
groups.) The remaining compact Lie groups that we didn’t discuss, the “exceptional”
groups, can be interpreted as unitary groups over the octonions. (Unlike the classical
groups, which form infinite series, there are only five exceptional compact groups,
because of the restrictions following from the nonassociativity of octonions.)

5. Tensor notation

Although historically group representations have usually been taught in the no-
tation where an m-component representation of a group defined by n×n matrices is
represented by an m-component vector, carrying a single index with values 1 to m,
a much more convenient and transparent method is “tensor notation”, where a gen-
eral representation carries many indices ranging from 1 to n, with certain symmetries
(and perhaps tracelessness) imposed on them. (Tensor notation for a covering group
is generally known as “spinor notation” for the corresponding orthogonal group: See
subsection IC5.) This notation takes advantage of the property described above for
expressing arbitrary representations in terms of direct products of vectors. In terms
of transformation laws, it means we need to know only the defining representation,
since the transformation of this representation is applied to each index. There are
at most four vector representations, by taking the dual and complex conjugate; we
use the corresponding index notation. Then the group constraints simply state the
invariance of the group metrics (and their complex conjugates and inverses), which
thus can be used to raise, lower, and contract indices:
 B. INDICES 41

volume: Special: I1 ...In

 .
 hermitian: Unitary: ΥIJ

metric: Orthogonal: η IJ
 (anti)symmetric:
Symplectic: ΩIJ

Real: η .J
reality: I
pseudoreal (*): Ω .J
I

As a result, we have relations such as

. .
h | i=η
I J IJ
or IJ
Ω , h | i=Υ
I J IJ

We also define inverse metrics satisfying

.
KI KI KI
η ηKJ = Ω ΩKJ = Υ Υ . = δJI
KJ

(and similarly for contracting the second index of each pair). Therefore, with uni-
tarity/(pseudo)reality we can ignore complex conjugate representations (and dotted
indices), converting them into unconjugated ones with the metric, while for orthogo-
nality/symplecticity we can do the same with respect to raising/lowering indices:
. .
Unitary: ψ I = ΥIJ ψJ
Orthogonal: ψ I = η IJ ψJ
Symplectic: ψ I = ΩIJ ψJ
Real: ψ . = η .J ψJ
I I
pseudoreal (*): ψ . = Ω .J ψJ
I I

For the real groups there is also the constraint of reality on the defining representation:

ψ̄ . ≡ (ψI )* = ψ . ≡ η .J ψJ
I I I

Excercise IB5.1
As an example of the advantages of index notation, show that SSp is the same
as Sp. (Hint: Write one  in the definition of the determinant in terms of Ω’s
by total antisymmetrization, which then can be dropped because it is enforced
by the other . One can ignore normalization by just showing det M = det I.)
For SO(n+ ,n−), there is a slight modification of a sign convention: Since then
indices can be raised and lowered with the metric, I... is usually defined to be the
result of raising indices on I..., which means

12...n = 1 ⇒ 12...n = det η = (−1)n−

42 I. GLOBAL

Then I... should be replaced with (−1)n− I... in the equations of subsection IB3: For
example,
I1
J1 ...Jn I1 ...In = (−1)n− δ[J1
· · · δJInn ]

We now give the simplest explicit forms for the defining representations of the
classical groups. The most convenient notation is to label the generators by a pair
of fundamental indices, since the adjoint representation is obtained from the direct
product of the fundamental representation and its dual (i.e., as a matrix labeled by
row and column). The simplest example is GL(n), since the generators are arbitrary
matrices. We therefore choose as a basis matrices with a 1 as one entry and 0’s
everywhere else, and label that generator by the row and column where the 1 appears.
Explicitly,
GL(n) : (GI J )K L = δILδKJ
⇒ GI J = |J ihI |
This basis applies for GL(n,C) as well, the only difference being that the coefficients
α in G = αI J GJ I are complex instead of real. The next simplest case is U(n): We can
again use this basis, although the matrices GI J are not all hermitian, by requiring
that αI J be a hermitian matrix. This turns out to be more convenient in practice
than using a hermitian basis for the generators. A well known example is SU(2),
where the two generators with the 1 as an off-diagonal element (and 0’s elsewhere)
are known as the “raising and lowering operators” J± , and are more convenient than
their hermitian parts for purposes of contructing representations. (This generalizes
to other unitary groups, where all the generators on one side of the diagonal are
raising, all those on the other side are lowering, and those along the diagonal give the
maximal Abelian subalgebra, or “Cartan subalgebra”.)
Representations for the other classical groups follow from applying their defini-
tions to the GL(n) basis. We thus find

SL(n) : (GI J )K L = δILδK

J
− n1 δIJ δK
L
⇒ GI J = |J ihI | − n1 δIJ |K ihK |
SO(n) : (GIJ )KL = δ[IK δJ]
L
⇒ GIJ = |[I ihJ] |
Sp(n) : (GIJ )KL = δ(I
K L
δJ) ⇒ GIJ = |(I ihJ) |

As before, SL(n,C) and SU(n) use the same basis as SL(n), etc. For SO(n) and Sp(n)
we have raised and lowered indices with the appropriate metric (so SO(n) includes
SO(n+ ,n−)). For some purposes (especially for SL(n)), it’s more convenient to impose
tracelessness or (anti)symmetry on the matrix α, and use the simpler GL(n) basis.
Excercise IB5.2
Our normalization for the generators of the classical groups is the simplest,
and independent of n (except for subtracting out traces):
 B. INDICES 43

a Find the commutation relations of the generators (structure constants) for the
defining representation of GL(n) as given in the text. Note that the values of
all the structure constants are 0, ±i. Show that

cD = 1

(see subsection IB2).

b Consider the GL(m) subgroup of GL(n) (m<n) found by restricting the range
of the index of the above defining representation. Show the structure con-
stants are the same as those given by starting with the above representation
of GL(m).
c Find the structure constants for SO(n) and Sp(n).
d Directly evaluate kD cA (= δ ij Gi Gj ) for SL(n), SO(n), and Sp(n), and compare
with cD dA /dD .
Excercise IB5.3
A tensor that pops up in various contexts is

dijk = tr(Gi {Gj , Gk })

It takes a very simple form in terms of defining indices:

a Show that for SU(n) this tensor is determined to be, up to an overall normal-
ization (that depends on the representation),

2 4
tr GI1 J1 GI2 J2 , GI3 J3 ∼ [(231)+(312)]− [(132)+(213)+(321)]+ 2 (123)
n n
(ijk) ≡ δIJi1 δIJj2 δIJk3
from just the total symmetry of dijk (and GI I = 0), since the only invariant
tensor available is δIJ . (If IJ... were used, IJ... would also be required, to
balance the number of subscripts and superscripts; but their product can be
expressed in terms of just δ’s also.)
b Check this result by using the explicit G’s for the defining representation, and
determine the proportionality constant for that representation.
With the exception of the “spinor” representations of SO(n) (to be discussed
in subsection IC5, section IIA, and subsection XC1), general representations can be
obtained by reducing direct products of the defining representations. This means they
can be described by objects with multiple indices (up/down, dotted/undotted), where
each index is that of a defining representation, and satisfying various (anti)symmetry
and tracelessness conditions on the indices.
44 I. GLOBAL

Excercise IB5.4
Consider the representations of SU(n) obtained from the symmetric and an-
tisymmetric part of the direct product of two defining representations. For
simplicity, one can work with the U(n) generators, since the U(1) pieces will
appear in a simple way.
a Using tensor notation for the generators (GI J )KL M N , find their explicit rep-
resentation for these two representations.
b By evaluating the trace, show that the Dynkin index for the two cases is

ca = n − 2, cs = n + 2

c Show the sum of these two is consistent with the argument at the end of
subsection IB2. Show each case is consistent with n=2, and the antisymmetric
case with n=3, by relating those cases to the singlet, defining, and adjoint
representations.

REFERENCES
1 H. Georgi, Lie algebras in particle physics: from isospin to unified theories
(Benjamin/Cummings, 1982):
best book on Lie groups; unlike other texts, covers not only more powerful Cartan-
Weyl methods, but also more useful tensor methods; also has useful applications to
nonrelativistic quark model.
2 S. Helgason, Differential geometry and symmetric spaces (Academic, 1962);
R. Gilmore, Lie groups, Lie algebras, and some of their applications (Wiley, 1974):
noncompact classical Lie groups.
 C. REPRESENTATIONS 45

................... ...................
................... C. REPRESENTATIONS ...................

We now consider some of the more useful representations, as explicit examples

of the results of the previous section. In particular, we consider symmetries of the
quark model.

1. More coordinates
We began our “review” of group theory by looking at how symmetries were rep-
resented on coordinates. We now return to coordinates as a special case (particular
representation) of the general results of the previous section. The idea is that the
coordinates themselves are already a representation of the group, and the wave func-
tions are functions of these coordinates. For example, for ordinary rotations we use
wave functions that depend on position or momentum, which transforms as a vec-
tor. (This is not always the case: For example, in our description of the conformal
group the usual space and time coordinates transformed nonlinearly, and not just
by multiplication by constant matrices unless the extra two coordinates were intro-
duced.) This is the basic distinction between classical mechanics and classical field
theory: Mechanics uses the coordinates themselves as the basic variables, while field
theory uses functions of the coordinates. (Similarly, in quantum mechanics the wave
functions are functions of the coordinates, while in quantum field theory the wave
functions are “functionals” of functions of the coordinates.)
In general, the construction of such a “coordinate representation” starts with a
given matrix representation (usually finite dimensional) (Gi )I J and then defines a
new representation

Ĝi = q I (Gi )I J pJ ; [pI , q J } = δIJ , [q, q} = [p, p} = 0

for some objects q and p, which are interpreted as either coordinates and their con-
jugate momenta (up to a factor of i), or as creation and annihilation operators: The
latter nomenclature is used when the boundary conditions allow the existence of a
state |0i called the “vacuum”, satisfying p|0i = 0, so we can define the other states
as functions of q acting on |0i. (If the coordinates are fermionic, the distinction is
moot, since by the usual Taylor expansion the Hilbert space is finite dimensional. See
excercise IA2.3.) It is easy to check that Ĝi satisfy the same commutation relations
as Gi . In particular, if the matrices are in the adjoint representation, q i can be inter-
preted as the group coordinates themselves: This follows from considering the action
i
of an infinitesimal transformation on the group element g(q) = eiq Gi (or just the Lie
algebra element G(q) = q i Gi ).
46 I. GLOBAL

If we write these results in bra/ket notation, since

[Ĝi , q I ] = q J (Gi )J I , [Ĝi , pI ] = −(Gi )I J pJ

it is more natural to look at the action on bras:

hq| = q I hI |, |pi = |I ipI ⇒ Ĝi hq| = hq|Gi , Ĝi |pi = −Gi |pi

Note that this vector space is coordinate space itself, not the space of functions of
the coordinates; it is the same space on which Gi is defined. (Of course, Ĝi is defined
on arbitrary functions of the coordinates; it has a reducible representation bigger
than (Gi )I J . Effectively, (Gi )I J is represented on the space of functions linear in the
coordinates.) Then, for example

Ĝ1 Ĝ2 hq| = Ĝ1 hq|G2 = hq|G1 G2

is obviously equivalent, while (ignoring any extra signs for fermions)

Ĝ1 Ĝ2 |pi = −Ĝ1 G2 |pi = −G2 Ĝ1 |pi = G2 G1 |pi

at least gives an equivalent result for the commutator algebra [Ĝ1 , Ĝ2 ]. This is the
expected result for the dual representation Gi → −GTi .
Interesting examples are given by using the defining representation for G. For
example, the commonly used oscillator representation for U(n) is

U(n) : GI J = a†J aI , [aI , a†J } = δIJ

where the oscillators can be bosonic or fermionic. For the SO and Sp cases, because
we can raise and lower indices, and because of the (anti)symmetry on the indices, the
interesting possibility arises to identify the coordinates with their momenta, with the
statistics appropriate to the symmetry:

Sp(n) : GIJ = 12 z(I zJ) , [zI , zJ ] = ΩIJ

SO(n) : GIJ = 12 γ[I γJ] , {γI , γJ } = ηIJ

For SO(n) the representation is finite dimensional because of the Fermi-Dirac statis-
tics, and is called a “Dirac spinor” (and γ the “Dirac matrices”). If the opposite
statistics are chosen, the coordinates and momenta can’t be identified: For example,
bosonic coordinates for SO(n) give the usual spatial rotation generators GIJ = x[I ∂J] .
Excercise IC1.1
Use this bosonic oscillator representation for U(2)=SU(2)⊗U(1), and use the
 C. REPRESENTATIONS 47

SU(2) subgroup to describe spin. Show that the spin s (the integer or half-
integer number that defines the representation) itself has a very simple ex-
pression in terms of the U(1) generator. Show this holds in the quantum
mechanical case (by interpreting the bracket as the quantum commutator),
giving the usual s(s+1) for the sum of the squares of the generators (with ap-
propriate normalization). Use this result to show that these oscillators, acting
on the vacuum state, can be used to construct the usual states of arbitrary
spin s.
Excercise IC1.2
Considering SO(2n), divide up γI into pairs of canonical (and complex) con-
√
jugates a1 = (γ1 + iγ2 )/ 2, etc., so {a, a†} = 1. Write the SO(2n) generators
in terms of aa, a†a†, and a†a. Show that the a† a’s by themselves generate a
U(n) subgroup. Decompose the Dirac spinor into U(n) representations. Show
that the product of all the γ’s is related to the U(1) generator, and commutes
with all the SO(2n) generators. Show that the states created by even or odd
numbers of a† ’s on the vacuum don’t mix with each other under SO(2n), so
the Dirac spinor is reducible into two “Weyl spinors”.

2. Coordinate tensors

Many groups can be represented on coordinates. Depending on the choice of

coordinates, the coordinates may transform nonlinearly (i.e., as a realization, not a
representation), as for the D-dimensional conformal group in terms of D (not D+2)
coordinates. However, given the nonlinear transformation of the coordinates, there
are always representations other than the defining one (scalar field) that we can im-
mediately write down (such as the adjoint). We now consider such representations:
These are useful not only for the spacetime symmetries we have already considered,
but also for general relativity, where the symmetry group consists of arbitrary coor-
dinate transformations. Furthermore, these considerations are useful for describing
coordinate transformations that are not symmetries, such as the change from Carte-
sian to polar coordinates in nonrelativistic theories.
When applied to quantum mechanics, we write the action of a symmetry on a
state as δψ = iGψ (or ψ 0 = eiGψ), but on an operator as δA = i[G, A] (or A0 =
eiGAe−iG). In classical mechanics, we always write δA = i[G, A] (since classical
objects are identified with quantum operators, not states). However, if G = λm ∂m is
a coordinate transformation (e.g., a rotation) and φ is a scalar field, then in quantum
48 I. GLOBAL

notation we can write

δφ(x) = [G, φ] = Gφ = λm ∂m φ (φ0 = eGφe−G = eG φ)

since the derivatives in G just differentiate φ. (For this discussion of coordinate

transformations we switch to absorbing the i’s into the generators.) The coordinate
transformation G has the usual properties of a derivative:

[G, f(x)] = Gf ⇒ Gf1 f2 = [G, f1 f2] = (Gf1 )f2 + f1Gf2

eGf1 f2 = eGf1 f2 e−G = (eGf1 e−G)(eG f2e−G ) = (eGf1)(eG f2)

and similarly for products of more functions.

The adjoint representation of coordinate transformations is a “vector field” (in the
sense of a spatial vector), a function that has general dependence on the coordinates
(like a scalar field) but is also linear in the momenta (as are the Poincaré generators):

G = λm (x)∂m, V = V m (x)∂m ⇒ δV = [G, V ] = (λm ∂m V n − V m ∂m λn )∂n

⇒ δV m = λn ∂nV m − V n ∂nλm

The same result follows if we use the Poisson bracket instead of the quantum me-
chanical commutator, replacing ∂m with ipm in both G and V .
Finite transformations can also be expressed in terms of transformed coordinates
themselves, instead of the transformation parameter:

φ(x) = e−λ φ0 (x) = φ0(e−λ

m∂ m∂
m m
x)

as seen, for example, from a Taylor expansion of φ0 , using e−G φ0 = e−Gφ0 eG. We then
define
φ0 (x0) = φ(x) x0 = e−λ
m∂
⇒ m
x

This is essentially the statement that the active and passive transformations cancel.
However, in general this method of defining coordinate transformations is not con-
venient for applications: When we make a coordinate transformation, we want to
know φ0 (x). Working with the “inverse” transformation on the coordinates, i.e., our
original e+G ,
φ0 (x) = eGφ(x) = φ(x̃(x))
m∂
x̃ ≡ e+λ m
x ⇒

So, for finite transformations, we work directly in terms of x̃(x), and simply plug this
into φ in place of x (x → x̃(x)) to find φ0 as a function of x.
 C. REPRESENTATIONS 49

Similar remarks apply for the vector, and for derivatives in general. We then use

x0 = e−G x ⇒ ∂ 0 = e−G ∂eG

where ∂ 0 = ∂/∂x0, since ∂ 0x0 = ∂x = δ. This tells us

V m (x)∂m = e−G V 0m (x)∂m eG = V 0m(x0 )∂m

0

or V 0 (x0) = V (x). Acting with both sides on x0m ,

∂x0m
V 0m (x0) = V n (x)
∂xn
On the other hand, working in terms of x̃ is again more convenient: Since for ∂˜ = ∂/∂ x̃
" −1 # n
∂ x̃(x)
∂˜m = eG∂m e−G = ∂n
∂x
m

we have, for example,

" −1 # n
m 0 G m m ∂ x̃(x)
(V ∂m φ) (x) = e (V ∂m φ)(x) = V (x̃(x)) ∂n φ(x̃(x))
∂x
m

which yields an explicit expression for the transformed fields.

A “differential form” is defined as an infinitesimal W = dxm Wm (x). Its transfor-
mation law under coordinate transformations, like that of scalar and vector fields, is
defined by W 0 (x0) = W (x). For any vector field V = V m (x)∂m , V m Wm transforms as
a scalar, as follows from the “chain rule” d = dx0m ∂m0
= dxm ∂m . Explicitly,
∂xn
Wm0 (x0) = Wn (x) 0m
∂x
or in infinitesimal form
δWm = λn ∂n Wm + Wn ∂m λn

Thus a differential form is dual to a vector, at least as far as the matrix part of coor-
dinate transformations is concerned. They transform the same way under rotations,
because rotations are orthogonal; however, more generally they transform differently,
and in the absence of a metric there is not even a way to relate the two by raising or
lowering indices.
Higher-rank differential forms can be defined by antisymmetric products of the
R
above “one-forms”. These are useful for integration: Just as the line integral W =
R m
dx Wm is invariant under coordinate transformations by definition (as long as we
choose the curve along which the integral is performed in a coordinate-independent
50 I. GLOBAL

way), so is a totally antisymmetric Nth-rank tensor (“N-form”) Wm1 ···mN integrated

on an N-dimensional subspace as
Z
∂xp1 ∂xpN
dxm1 · · · dxmN Wm1 ···mN : Wm0 1 ···mN (x0) = Wp1 ···pN (x) 0m1 · · · 0mN
∂x ∂x
where the surface element dxm1 · · · dxmN is interpreted as antisymmetric. (The signs
come from switching initial and final limits of integration, as prescribed by the “ori-
entation” of the hypersurface.) This is clear if we rewrite the integral more explicitly
in terms of coordinates σ i for the subspace: Then
Z Z Z
iN c ci1 ···i (σ)
dx · · · dx Wm1 ···mN (x) = dσ · · · dσ Wi1 ···iN (σ) = dN σ i1 ···iN W
m1 mN i1
N

where m1 mN
ci1 ···i (σ) = ∂x · · · ∂x Wm1 ···m (x)
W N N
∂σ i1 ∂σ iN
is the result of a coordinate transformation that converts N of the x’s to σ’s, an
interpretation of the functions x(σ) that define the surface. Then any coordinate
transformation on x → x0 (not on σ) will leave W c (σ) invariant. In particular, if the
R
subspace is the full space, so we can look directly at dN x m1 ···mN Wm1 ···mN , we see
that a coordinate transformation generates from W an N-dimensional determinant
exactly canceling the Jacobian resulting from changing the integration measure dN x.
Excercise IC2.1
For all of the following, use the exponential form of the finite coordinate trans-
formation: Show that any (local) function of a scalar field (without explicit x
dependence additional to that in the field) is also a scalar field (i.e., satisfies
the same coordinate transformation law). Show that the transformation law
of a vector field or differential form remains the same when multiplied by a
scalar field (at the same x). Show that V φ = V m ∂m φ is a scalar field for any
scalar field φ and vector field V . Show that [V, W ] is a vector field for any
vector fields V and W .
Excercise IC2.2
Examine finite coordinate transformations for integrals of differential forms
in terms of x̃ rather than x0 . Find the explicit expression for W 0(x) in terms
of W (x̃(x)), etc., and use this to show invariance:
Z Z
0
dx · · · dx Wm1 ···mN (x) = dx̃m1 · · · dx̃mN Wm1 ···mN (x̃)
m1 mN

Z
= dxm1 · · · dxmN Wm1 ···mN (x)
 C. REPRESENTATIONS 51

where in the last step we have simply substituted x̃ → x as a change of integra-

tion variables. Note that, using the x̃ form of the transformation rather than
x0, the transformation generates the needed Jacobain, rather than canceling
one.
From the above transformation law, we see that the curl of a differential form is
also a differential form:
0
∂[m1
Wm0 2 ···mN ] (x0) = ∂[m
0
1
0
(∂m 2
0
xp2 ) · · · (∂m N]
xpN )Wp2 ···pN (x)
0 0
= [∂[p1 Wp2 ···pN ] (x)](∂m 1
xp1 ) · · · (∂m N
xpN )
because the curl kills ∂ 0∂ 0x terms that would appear if there were no antisymmetriza-
tion. Objects that transform “covariantly” under coordinate transformations, without
such higher derivatives of x (or λ in the other notation), like scalars, vectors, differen-
tial forms and their products, are called (coordinate) “tensors”. Getting derivatives of
tensors to come out covariant in general requires special fields, and will be discussed
in chapter IX. An important application of the covariance of the curl of differential
forms is the generalized Stokes’ theorem (which includes the usual Stokes’ theorem
and Gauss’ law as special cases):
Z I
dx · · · dx
m1 mN +1 1
∂ W
(N +1)! [m1 m2 ···mN +1 ]
= dxm1 · · · dxmN Wm1 ···mN

where the second integral is over the boundary of the space over which the first is
H
integrated. (We use the symbol “ ” to refer to boundary integrals, including those
over contours, which are closed boundaries of 2D surfaces.)

3. Young tableaux
We now return to our discussion of finite-dimensional representations. In the
previous section we gave the machinery for describing them using index notation,
but examined only the defining representation in detail. Now we analyze general
irreducible representations.
All the irreducible finite-dimensional representations of the groups SU(N) can be
described by tensors with lower N-valued indices with various (anti)symmetrizations.
(An upper index can be replaced with N−1 lower indices by using the Levi-Civita
tensor.) Although detailed calculations require explicit use of these indices, three
properties can be more conveniently discussed pictorially: (1) the (anti)symmetries
of the indices, (2) the dimension (number of independent components) of the repre-
sentation, and (3) the reduction of the direct product of two representations (which
irreducible representations result, and how many of each).
52 I. GLOBAL

A “Young tableau” is a picture representing an irreducible representation in terms

of boxes arranged in a regular grid into rows and columns, such that the columns are
aligned at the top, and their depths are nonincreasing to the right: for example,

Each box represents an index, with antisymmetry among indices in any column, and
symmetry among indices in any row. More precisely, since one can’t simultaneously
have these symmetries and antisymmetries, it corresponds to the result of taking any
arbitrary tensor with that many indices, first symmetrizing the indices in each row,
and then antisymmetrizing the indices in each column (or vice versa; symmetrizing
and then antisymmetrizing and then symmetrizing again gives the same result as
skipping the first symmetrization, etc.). This gives a simple way to classify and
symbolize each representation. (We can denote the singlet representation, which has
no boxes, by a dot.) Note that the deepest column should have no more than N−1
boxes for SU(N) because of the antisymmetry.
To calculate the dimension of the representation for a given tableau, we use the
“factors over hooks” rule: (1) Write an “N” in the box in the upper-left corner, and
fill the rest of the boxes with numbers that decrease by 1 for each step down and
increase by 1 for each step to the right. (2) Draw (or picture in your mind) a “hook”
for each box — a “Γ” with its corner in the box and lines extending right and down
out of the tableau. (3) The dimension is then given by the formula
Y integer written there
dimension =
# boxes intersected by its hook
each box
For the previous example, we find (listing boxes first down and then to the right)
N N −1 N −2 N −3 N +1 N N −1 N +2 N +1 N +3 N +2 N +4
· · · · · · · · · · ·
8 6 3 1 6 4 1 4 2 3 1 1

The direct product of two Young tableaux A⊗B is analyzed by the following rules:
First, label all the boxes in B by putting an “a” in each box in the top row, “b” in
the second row, etc. Then, take the following steps in all possible ways to find the
Young tableaux resulting from the direct product: (1) Add all the “a” boxes from B
to the right side and bottom of A, then “b” to the right and bottom of that, etc., to
make a new Young tableaux. Any two tableaux constructed in this way with the same
arrangement of boxes but different assignment of letters are considered distinct, i.e.,
multiple occurences of the same representation in the direct product. (2) No more
than 1 “a” can be in any column, and similarly for the other letters. (3) Reading from
 C. REPRESENTATIONS 53

right to left, and then from top to bottom (i.e., like Hebrew/Arabic), the number of
a’s read should always be ≥ the number of b’s, b’s ≥ c’s, etc. For example,
a
⊗ aa
b = b
aa
⊕ a
ab ⊕ a
b

Note that A⊗B always gives the same result as B⊗A, but one way may be simpler
than the other. For a given value of N, a column of N boxes is equivalent to none
(again by antisymmetry), while more than N boxes in a column gives a vanishing
tableau.
Excercise IC3.1
Calculate
⊗

Check the result by finding the dimensions of all the representations and
adding them up.
These SU(N) tableaux also apply to SL(N): Only the reality properties are dif-
ferent. Similar methods can be applied to USp(2N) (or Sp(2N)), but tracelessness
(with respect to the symplectic metric) must be imposed in antisymmetrized indices,
so these trace pieces must be separated out when considering the above rules. (I.e.,
consider USp(2N)⊂SU(2N).) Similar remarks apply to SO(N), which has a symmet-
ric metric, but there are also “spinor” representations (see below). The additional
irreducible representations then can be constructed from taking direct products of
the above with the smallest spinors, and removing the “gamma-matrix” traces.

4. Color and flavor

We now consider the application of these methods to “internal symmetries” (those
that don’t act on the coordinates) in particle physics. The symmetries with experi-
mental confirmation involve only the unitary groups (U and SU) of small dimension.
However, we will find later that larger unitary groups can be useful for approxima-
tion schemes. (Also, larger unitary and other groups continue to be investigated for
unification and other purposes, which we consider in later chapters.)
The “Standard Model” describes all of particle physics that is well confirmed
experimentally (except gravity, which is not understood at the quantum level). It
includes as its “fundamental” particles: (1) the spin-1/2 quarks that make up the
observed strongly interacting particles, but do not exist as asymptotic states, (2) the
weakly interacting spin-1/2 leptons, (3) the spin-1 gluons that bind the quarks to-
gether, which couple to the charges associated with SU(3) “color” symmetry, but also
54 I. GLOBAL

are not asymptotic, (4) the spin-1 particles that mediate the weak and electromag-
netic interactions, which couple to SU(2)⊗U(1) “flavor”, and (5) the yet unobserved
spin-0 Higgs particles that are responsible for all the masses of these weakly interact-
ing particles. These particles, along with their masses (in GeV) and (electromagnetic)
charges (Q = Q̄ + ∆Q), are:

1
s= 2 s=1
color: → quark (3) lepton (1) color: → gluon electroweak
flavor (∆Q) (Q̄ = 16 ) (Q̄ = − 12 ) flavor (Q) (8) (1)
+ 12 u (.005) νe (< 10−8 ) 0 g (0) γ (< 2 · 10−25 )
− 12 d (.009) e (.0005109991) 0 Z (91.187)
+ 12 c (1.4) νµ (<.00017) ±1 W (80.4)
− 12 s (.18) µ (.10565839)
s=0
+ 12 t (174) ντ (<.0182)
− 12 b (4.7) τ (1.7770) (Q = 0) H (>77.5)

The quark masses we have listed are the “current quark masses”, the effective masses
when the quarks are relativistic with respect to their hadron, and act as almost free.
Nonrelativistic quark models use instead the “constituent quark masses”, which in-
clude potential energy from the gluons. This extra potential energy is about .30
GeV per quark in the lightest mesons, .35 GeV in the lightest baryons; there is also
a contribution to the binding energy from spin-spin interaction. Unlike electrody-
namics, where the potential energy is negative because the electrons are free at large
distances, where the potential levels off (the top of the “well”), in chromodynamics
the potential energy is positive because the quarks are free at high energies (short
distances, the bottom of the well), and the potential is infinitely rising. Masslessness
of the gluons is implied by the fact that no colorful asymptotic states have ever been
observed. We have divided the spin-1/2 particles into 3 “families” with the same
quantum numbers (but different masses). Within each family, the quarks are similar
to the leptons, except that: (1) the masses and average charges (Q̄) are different,
(2) the quarks come in 3 colors, while the leptons are colorless, and (3) the neutrinos,
to within experimental error, are massless, so they have half as many components
as the massive fermions (1 helicity state each, instead of 2 spin states each). This
means that each lepton family has 1 SU(2) doublet and 1 SU(2) singlet. For symme-
try (and better, quantum mechanical, reasons to be explained later), we also assume
the quarks have 1 SU(2) doublet, but therefore 2 SU(2) singlets. (Some experiments
have indicated small masses for neutrinos: This would require generalization of the
 C. REPRESENTATIONS 55

Standard Model, such as models with parity broken by interactions. Some examples
of such theories will be discussed in subsection IVB4.)
We first look at the color group theory of the physical states, which are color
singlets. The fundamental unobserved particles are the spin-1 “gluons”, described by
the Yang-Mills gauge fields, and the spin-1/2 quarks. Suppressing all but color indices,
we denote the quark states by q i , and the antiquarks by q †i , where the indices are those
of the defining representation of SU(n), and its complex conjugate. The quarks also
carry a representation of a “flavor” group, unlike the gluons. The simplest flavorful
states are those made up of only (anti)quarks, with indices completely contracted by
one factor of an SU(n) group metric: From the “U” of SU(n), we can contract defining
indices with their complex conjugates, giving the “mesons”, described by q †i q i (quark-
antiquark), which are their own antiparticles. From the “S” of SU(n), we have the
“baryons”, described by i1 ...in q i1 ...q in (n-quark), and the antibaryons, described by
the complex conjugate fields. All other colorless states made of just (anti)quarks
can be written as products of these fields, and therefore considered as describing
composites of them. Thus, we can approximate the ground states of the mesons by
q †i (x)q i(x), which describe spins 0 and 1 because of the various combinations of spins
↔
(from 12 ⊗ 12 = 0 ⊕ 1). The first excited level will then be described by q † ∂ q (where
↔
A ∂ B ≡ A∂B − (∂A)B

and picks out the relative momentum of the two quarks): It includes spins 0, 1, and 2,
etc., where each derivative introduces orbital angular momentum 1. (Similar remarks
apply to baryons.) We can also have flavorless states made from just gluons, called
“glueballs”: The ground states can be described by Fi j Fj i , where each F is a gluon
state (in the adjoint representation of SU(n)), and includes spins 0 and 2 (from the
symmetric part of 1 ⊗ 1). Because of their flavor multiplets and (electroweak) inter-
actions, many mesons and baryons corresponding to such ground and excited states
have been experimentally identified, while the glueballs’ existence is still uncertain.
Actually, quarks and gluons can almost be observed independently at high energies,
where the “strong” interaction is weak: The energetic particle appears as a “jet” —
a particle of high energy accompanied by particles of much lower energy (perhaps
too small to detect) in color-singlet combinations. (Depending on the available decay
modes, the jet might not be observed until after decaying, but still within a small
angle of spread.)
We now look at the flavor group theory of the physical hadronic states. In contrast
to the previous paragraph, we now suppress all but the flavor indices. Mesons Mi j =
56 I. GLOBAL

q †i q j are thus in the adjoint representation of flavor U(m) (m ⊗ m̄, where m is the
defining representation and m̄ its complex conjugate), for both the spin-0 and the
spin-1 ground states. The baryons are more complicated: For simplicity we consider
SU(3) color, which accurately describes physics at observed energies. Then the color
structure described above results in total symmetry in combined flavor and Lorentz
indices (from the antisymmetry in the color indices, and the overall antisymmetry for
Fermi-Dirac statistics). Thus, for the 3-quark baryons, the Young tableaux

⊕ ⊕

for SU(m) flavor are accompanied by the same Young tableaux for spin indices: In
nonrelativistic notation, the first tableau, being totally antisymmetric in flavor in-
dices, is also totally antisymmetric in the three two-valued spinor indices, and thus
vanishes. Similarly, the last tableau describes spin 3/2 (total symmetry in both types
of indices), while the middle one describes spin 1/2. Since only 3 flavors of quarks
have small masses compared to the hadronic mass scale, hadrons can be most conve-
niently grouped into flavor multiplets for SU(3) flavor: The ground states are then,
in terms of SU(3) flavor multiplets, 8⊕1 for the pseudoscalars, 8⊕1 for the vectors, 8
for spin 1/2, and 10 for spin 3/2.
Excercise IC4.1
What SU(flavor) Young tableaux, corresponding to what spins, would we have
for mesons and baryons if there were 2 colors? 4 colors?
However, the differing masses of the different flavors of quarks break the SU(3)
flavor symmetry (as does the weak interaction). In particular, the mass eigenstates
tend to be pure states of the various combinations of the different flavors of quarks,
rather than the linear combinations expected from the flavor symmetry. Specifically,
the linear combinations predicted by an 8⊕1 separation for mesons (trace and traceless
pieces of a 3⊗3 matrix) are replaced with particles that are more accurately described
by a particular flavor of quark bound to a particular flavor of antiquark. (This is
known as “ideal mixing”.) The one exception is the lighest mesons (pseudoscalars),
which are more accurately described by the 8⊕1 split, for this restriction to the 3
lighter flavors of quarks, but the mass of the singlet differs from that naively expected
from group theory or nonrelativistic quark models. (This is known as the “U(1)
problem”.) The solution is probably that the singlet mixes strongly with the lightest
psuedoscalar glueball (described by tr abcd Fab Fcd ); the mass eigenstates are linear
combinations of these two fields with the same quantum numbers. In any case, the
most convenient notation for labeling the entries of the matrix Mi j representing the
 C. REPRESENTATIONS 57

various meson states for any particular spin and angular momentum of the quark-
antiquark combination is that corresponding to the choice we gave earlier for the
generators of U(n): Label each entry by a separate name, where the complex conjugate
appears reflected across the diagonal. These directly correspond to the combination
of a particular quark with a particular antiquark, and to the mass eigenstates, with
the possible exception of the entries along the diagonal for the 3 lightest flavors,
where the mass eigenstates are various linear combinations. (However, the SU(2) of
the 2 lightest flavors is only slightly broken by the quark masses, so in that case the
combinations are very close to the 3⊕1 split of SU(2).)
For example, for the lightest multiplet of mesons (spin 0, and relative angular mo-
mentum 0 for the quark and antiquark, but not all of which have yet been observed),
we can write the U(6) matrix (for the 6 flavors of the 3 known families)
 
ūu ūd ūc ūs ūt ūb
 d̄u dd
¯ dc¯ d̄s dt¯ db¯ 
 
 
 c̄u c̄d c̄c c̄s c̄t c̄b 
Mi = 
j 
 s̄u s̄d s̄c s̄s s̄t s̄b 
 
 
 t̄u t̄d t̄c t̄s t̄t t̄b 
b̄u b̄d b̄c b̄s b̄t b̄b

u d c s t b
 
ū ηu π −(.1395700) D0 (1.8646) K −(.49368) T0 B −(5.279)
 + 
d̄  π (") ηd D+ (1.8693) K̄ 0 (.49767) T+ B̄ 0 (5.279) 
 0 
c̄  D̄ (") D− (") ηc (2.980) Ds− (1.9685) Tc0 Bc− 
=  
 + 
s̄  K (") K 0 (") Ds+ (") ηs Ts+ B̄s0 
 0 
t̄  T̄ T− T̄c0 Ts− ηt Tb− 
b̄ B +(") B 0 (") Bc+ Bs0 Tb+ ηb

where (approximately)

ηu = √1 π 0 (.1349764)
2
+ 12 [η 0(.9578) + η(.5473)]

ηd = − √12 π 0 + 12 (η 0 + η), ηs = √1 (η 0
2
− η)

in terms of the mass eigenstates (observed particles), with masses again in GeV,
and ditto marks refer to the transposed entry. (We have neglected the important
58 I. GLOBAL

contribution from the glueball.) For the corresponding spin-1 multiplet,

u d c s t b
 
ū ωu ρ− (.7700) D*0 (2.0067) K*− (.8917) T *0 B*−(5.325)
 
d¯  ρ+(") ωd D*+ (2.0100) K̄*0 (.8961) T *+ B̄*0 (5.325) 
 0 
c̄  D̄* (") D*− (") J/ψ(3.09688) D*− T *0c B*− 
fi =
M j  s c 
 
s̄  K*+ (") K*0 (") D*+ φ(1.019413) T *+ B̄*0s 
 s s

t̄  T̄ *0 T *− T̄ *0c T *−
s θ −
T *b 
b̄ B*+(") B*0 (") B*+ c B*0s T *+b Υ(9.4604)

where
ωu = √1 [ω(.7819)
2
+ ρ0(.7700)], ωd = √1 (ω
2
− ρ0 )

(with s̄s = φ, ideal mixing, also approximate).

5. Covering groups
The orthogonal groups O(n+,n− ) are of obvious interest for describing Lorentz
symmetry in spacetimes with n+ space and n− time dimensions, or conformal sym-
metry in spacetimes with n+−1 space and n−−1 time dimensions. This means we
should be interested in O(n) for n≤6, and their “Wick rotations”: transformations
that put in extra factors of i to change some signs on the metric. Coincidentally, these
are just the cases where the Lie algebras of the orthogonal groups are equivalent to
those of some algebras for smaller matrices. The smaller representation then can be
identified as the “spinor” representation of that orthogonal group. Since the “vec-
tor”, or defining representation space of the orthogonal group, itself is represented as
a matrix with respect to the other group (i.e., the state carries two spinor indices),
the other group may include certain phase transformations (such as −1) that cancel
in the transformation of the vector. The other group is then called the “covering”
group for that orthogonal group, since it includes those missing transformations in
its defining representation. (As a result, its group space also has a more interesting
topology, which we won’t discuss here.)
One way to discover these covering groups is to first count generators, then try
to construct explicitly the orthogonal metric on matrices. SO(n) has n(n−1)/2 gen-
erators (antisymmetric matrices), Sp(n) has n(n+1)/2 (symmetric), and SU(n) has
n2 −1 (traceless). (These are hermitian generators, since we applied reality or her-
miticity.) So, for some group SO(n), we look for another group that has the same
number of generators. Then, if the new group is defined on m×m matrices, we look
for conditions to impose on an m×m matrix (not necessarily the adjoint) to get an
 C. REPRESENTATIONS 59

n-component representation. This is easy to do by inspection for small n; for large n

it’s easy to see that it can’t work, since m will be of the order of n, and the simple
constraints will give of the order of n2 components instead of n. We then construct
the norm of this matrix M as tr(M † M), which is just the sum of the absolute value
squared of the components, for SO(n), and the other orthogonal groups by Wick
rotation. (Wick rotation affects mainly the reality conditions on M.)
The identifications for the Lie algebras are then:
SO(2) = U(1), SO(1,1) = GL(1)
SO(3) = SU(2) = SU*(2) = USp(2), SO(2,1) = SU(1,1) = SL(2) = Sp(2)
SO(4) = SU(2)⊗SU(2), SO(3,1) = SL(2,C) = Sp(2,C), SO(2,2) = SL(2)⊗SL(2)
SO(5) = USp(4), SO(4,1) = USp(2,2), SO(3,2) = Sp(4)
SO(6) = SU(4), SO(5,1) = SU*(4), SO(4,2) = SU(2,2), SO(3,3) = SL(4)
Note that the Euclidean cases are all unitary, while the ones with (almost) equal
numbers of space and time dimensions are all real. There are also some similar
relations for the pseudoreal orthogonal groups:
SO*(2) = U(1), SO*(4) = SU(2)⊗SL(2), SO*(6) = SU(3,1), SO*(8) = SO(6,2)
The norm and conditions for an m-spinor of SO(n+ ,n−) are:

n− ⇒ 0 1 2 3
T
m n norm symmetry : z = reality : z* =
1 2 z 0z z0 z (z 0* = z 0 )
2 3 z αβ z γδ γαδβ z −z z
0 0
4 z αβ z γδ γαδ0 β 0 −z zT z
4 5 αβ γδ
z z δγβα −z (z Ωβα = 0)
αβ 1
2 z
1
2 (ΥzΥ) z
6 z αβ z γδ δγβα −z 1
2 z −ΩzΩ 1
2 (ΥzΥ) z

Note that in all but the 2D cases the norms are associated with determinants: For
D=3 and 4 the norm is given by the determinant, while for D=5 and 6 we use the
fact that the determinant of an antisymmetric matrix is the square of the Pfaffian.
Excercise IC5.1
Show that for D=5 zz and zzΩΩ give the same norm. (Hint: Consider
Ω[αβ Ωγδ] .)
Unfortunately, for SO(n) for larger n, the spinor is as least as large as, and usu-
ally larger than, the vector. In general, the spinor is like the “square root” of the
vector, in that the vector can be found by taking the direct product of two spinors.
60 I. GLOBAL

It is impossible to find the spinor representation by taking direct products of vec-

tors. This situation occurs only for orthogonal groups: In all other classical groups,
all (finite-dimensional) representations are among those obtained from multiple di-
rect products of vectors. Furthermore, in those cases the “irreducible” representa-
tions (those that can’t be divided into smaller representations) can be picked out by
(anti)symmetrization, and by separating trace and traceless pieces (where traces are
taken with the group metrics). Fortunately, for the above cases of orthogonal groups,
we can perform the same construction starting with the spinor representations, since
those are the “vectors” of non-orthogonal groups.

REFERENCES
1 J. Schwinger, On angular momentum, Quantum theory of angular momentum: a collec-
tion of reprints and original papers, eds. L.C. Biedenharn and H. Van Dam (Academic,
1965) p. 229:
spin using spinor oscillators.
2 Georgi, loc. cit. (IB).
3 P.A.M. Dirac, Proc. Roy. Soc. A117 (1928) 610.
4 H. Weyl, Z. Phys. 56 (1929) 330.
5 M. Hamermesh, Group theory and its application to physical problems (Addison-Wesley,
1962):
detailed discussion of Young tableaux.
6 M. Gell-Mann, Phys. Lett. 8 (1964) 214;
G. Zweig, preprints CERN-TH-401 and 412 (1964); Fractionally charged particles and
SU6 , in Symmetries and elementary particle physics, proc. Int. School of Physics “Ettore
Majorana”, Erice, Italy, Aug.-Sept., 1964, ed. A. Zichichi (Academic, 1965) p. 192:
quarks.
7 Particle Data Group (in 1998, C. Caso, G. Conforto, A. Gurtu, M. Aguilar-Benitez, C.
Amsler, R.M. Barnett, P.R. Burchat, C.D. Carone, O. Dahl, M. Doser, S. Eidelman, J.L.
Feng, M. Goodman, C. Grab, D.E. Groom, K. Hagiwara, K.G. Hayes, J.J. Hernández,
K. Hikasa, K. Honscheid, F. James, M.L. Mangano, A.V. Manoha, K. Mönig, H. Mu-
rayama, K. Nakamura, K.A. Olive, A. Piepke, M. Roos, R.H. Schindler, R.E. Shrock,
M. Tanabashi, N.A. Törnqvist, T.G. Trippe, P. Vogel, C.G. Wohl, R.L. Workman, and
W.-M. Yao), http://pdg.lbl.gov/pdg.html (published biannually, including European
Phys. J. C3 (1998) 1):
Review of Particle Physics, a.k.a. Review of Particle Properties, a.k.a. Rosenfeld tables;
tables of masses, decay rates, etc., of all known particles, plus useful brief reviews.
 A. TWO COMPONENTS 61

II. SPIN
Special relativity is simply the statement that the laws of nature are symmetric
under the Poincaré group. Free relativistic quantum mechanics or field theory is then
equivalent to a study of the representations of the Poincaré group. Since the con-
formal group is a classical group, while its subgroup the Poincaré group is not, it is
easier to first study the conformal group, which is sufficient for finding the massless
representations of the Poincaré group. The massive ones then can be found by di-
mensional reduction, which gives them in the same form as occurs in interacting field
theories. In four spacetime dimensions we use the covering group of the conformal
group, which is the easiest way to include spinors. These methods extend straight-
forwardly to supersymmetry, a symmetry between fermions and bosons that includes
the Poincaré group.

.................. ..................
.................. A. TWO COMPONENTS ..................

Although we have already specialized to spacetime symmetries, we have consid-

ered arbitrary spacetime dimensions. We have also noted that many of the lower-
dimensional Lie groups have special properties, especially with regard to covering
groups. In this section we will take advantage of those features; specifically, we ex-
amine the physical case D=4, where the rotation group is SO(3)=SU(2), the Lorentz
group is SO(3,1)=SL(2,C), and the conformal group is SO(4,2)=SU(2,2).

1. 3-vectors
The most important nontrivial Lie group in physics is the rotations in three
dimensions. It is also the simplest nontrivial example of a Lie group. This makes
it the ideal example to illustrate the properties discussed in the previous chapter,
as well as lay the groundwork for later discussions. We have already mentioned the
orbital part of rotations, i.e., the representation of rotations on spatial coordinates.
In this chapter we discuss the spin part; this is really the same as finding all (finite
dimensional, unitary) representations.
A useful way to understand spin, or general representations of the rotation group
in three dimensions of space, is to consider properties of 2×2 matrices. (This way
also generalizes in a very simple way to relativity, in three space and one time dimen-
sions.) Consider such matrices to be hermitian, which is natural from the quantum
mechanical point of view. Then they have four real components, one too many for a
62 II. SPIN

three-vector (but just right for a relativistic four-vector), so we restrict them to also
be traceless:
V = V †, tr V = 0
The simplest way to get a single number out of a matrix, besides taking the trace, is
to take the determinant. By expanding a general matrix identity to quadratic order
we find an identity for 2×2 matrices

det(I + M) = etr ln(I+M ) ⇒ −2 det M = tr(M 2 ) − (tr M)2

It is then clear that in our case −det V is positive definite, as well as quadratic, so
we can define the norm of this 3-vector as

|V |2 = −2 det V = tr(V 2 )

This can be compared easily with conventional notation by picking a basis:

 
V1 V 2 − iV 3
V = 21
√ = V~ · ~σ ⇒ det V = − 12 (V i )2
V 2 + iV 3 −V 1
where ~σ are the Pauli σ matrices, up to normalization. As usual, the inner product
follows from the norm:

|V + W |2 = |V |2 + |W |2 + 2V · W

⇒ V · W = det V + det W − det(V + W ) = tr(V W )

Applying our previous identities for determinants to 2×2 matrices, we have

MCM T C = I det M, M −1 = CM T C(det M)−1

where we now use the imaginary, hermitian matrix

0 −i


C= i 0

If we make the replacement M → eM and expand to linear order in M, we find

M + CM T C = I tr M

This implies
tr V = 0 ⇔ (V C)T = V C
i.e., the tracelessness of V is equivalent to the symmetry of V C. Furthermore, the
combination of the trace and determinant identities tell us

M 2 = M tr M − I det M ⇒ V 2 = −I det V = I 12 |V |2
 A. TWO COMPONENTS 63

Here by “V 2 ” we mean the square of the matrix, while “|V |2”= (V i )2 is the square
of the norm (neither of which should be confused with the component V 2 = V i δi2.)
Again expressing the inner product in terms of the norm, we then find

{V, W } = (V · W )I

Furthermore, since the trace of a commutator of two finite matrices is traceless, and
picks up a minus sign under hermitian conjugation, we can define an outer product
(vector×vector = vector) by
√
[V, W ] = 2iV × W

Combining these two results,

V W = 12 (V · W )I + √1 iV
2
×W

In other words, the product of two traceless hermitian 2×2 matrices gives a real trace
piece, symmetric in the two matrices, plus an antihermitian traceless piece, antisym-
metric in the two. Thus, we have a simple relation between the matrix product, the
inner (“dot”) product and the outer (“cross”) product. Therefore, the cross product
is a special case of the Lie bracket, or commutator. This way of treating vectors
(except for factors of “i”) is basically Hamilton’s “quaternions”.
Excercise IIA1.1
Check this result in two ways:
a Show the normalization agrees with the usual outer product. Using only the
above definition of V × W , along with {V, W } = (V · W )I, show

−I|V × W |2 = ([V, W ])2 = −I[|V |2|W |2 − (V · W )2]

b Use components, with the above basis.

Excercise IIA1.2
Write an arbitrary two-dimensional vector in terms of a complex number as
V = √12 (vx − ivy ).
a Show that the phase (U(1)) transformation V 0 = V eiθ generates the usual
rotation. Show that for any two vectors V1 and V2 , V1 *V2 is invariant, and
identify its real and imaginary parts in terms of well known vector prod-
ucts. What kind of transformation is V → V *, and how does it affect these
products?
64 II. SPIN

b Consider two-dimensional functions in terms of z = √12 (x + iy) and z* =

√1 (x − iy). Show by the chain rule that ∂z = √1 (∂x − i∂y ). Write the real
2 2
and imaginary parts of the equation ∂z* V = 0 in terms of the divergence and
curl. (Then V is a function of just z.)
c Consider the complex integral
I
dz
V
2πi
H
where “ ” is a “contour integral”: an integral over a closed path in the
complex plane defined by parametrizing dz = du(dz/du) in terms of some
real parameter u. This is useful if V can be Laurent expanded as V (z) =
P∞
n=−∞ cn (z − z0 ) inside the contour about a point z0 there, since by con-
n

sidering circles z = z0 + reiθ we find only the 1/(z − z0) term contributes.
Show that this integral contains as its real and imaginary parts the usual line
integral and “surface” integral. (In two dimensions a surface element differs
from a line element only by its direction.) Use this fact to solve Gauss’ law
in two dimensions for a unit point charge as E = 1/z.
Excercise IIA1.3
Consider electromagnetism in 2×2 matrix notation: Define the field strength
√
as a complex vector F = 2(E + iB). Write partial derivatives as the sum
of a (rotational) scalar plus a (3-)vector as ∂ = √12 I∂t + ∇, where ∂t = ∂/∂t
is the time derivative and ∇ is the partial space derivatives written as a
traceless matrix. Do the same for the charge density ρ and (3-)current j as
J = − √12 Iρ + j. Using the definition of dot and cross products in terms of
matrix multiplication as discussed in this section, show that the simple matrix
equation ∂F = −J, when separated into its trace and traceless pieces, and
its hermitian and antihermitian pieces, gives the usual Maxwell equations

∇ · B = 0, ∇ · E = ρ, ∇ × E + ∂tB = 0, ∇ × B − ∂t E = j

(Note: Avoid the Pauli σ-matrices and explicit components.)

2. Rotations
One convenience of representing three-vectors as 2×2 instead of 3×1 is that rota-
tions are easier to write. Since vectors are hermitian, we expect their transformations
to be unitary:
V 0 = UV U † , U † = U −1
 A. TWO COMPONENTS 65

It is easily checked that this preserves the properties of these matrices:

(V 0 )† = (UV U † )† = V 0 , tr(V 0) = tr(UV U −1 ) = tr(U −1 UV ) = tr(V ) = 0

Furthermore, it also preserves the norm (and thus the inner product):

det(V 0 ) = det(UV U −1 ) = det(U)det(V )(det U)−1 = det V

Unitary 2×2 matrices have 4 parameters; however, we can elimimate one by the
condition
det U = 1
This eliminates only the phase factor in U, which cancels out in the transformation
law anyway. Taking the product of two rotations now involves multiplying only 2×2
matrices, and not 3×3 matrices.
We can also write U in exponential notation, which is useful for going to the
infinitesimal limit:
U = eiG ⇒ G† = G, tr G = 0
This means that G itself can be considered a vector. Rotations can be parametrized
√
by a vector whose direction is the axis of rotation, and whose magnitude is (1/ 2×)
the angle of rotation:
√
V 0 = eiGV e−iG ⇒ δV = i[G, V ] = − 2G × V

We also now see that the Lie bracket we previously identified as the cross product is
the bracket for the rotation group.
Excercise IIA2.1
Evaluate the elements of the matrix eiG in closed form for a diagonal generator
G. Generalize this result to arbitrary G. (Hint: Use rotational invariance.)
The hermiticity condition on V can also be expressed as a reality condition:

V = V† and tr V = 0 ⇒ V * = −CV C, (V C)* = C(V C)C

where “ * ” is the usual complex conjugate. A similar condition for U is

U † = U −1 and det U = 1 ⇒ U* = CUC

which is also a consequence of the fact that we can write U in terms of a vector as
U = eiV . As a result, the transformation law for the vector can be written in terms
of V C in a simple way, which manifestly preserves its symmetry:

(V C)0 = UV U −1 C = U(V C)U T

66 II. SPIN

Excercise IIA2.2
Write an arbitrary rotation in two dimensions in terms of the slope (dy/dx) of
the rotation (the slope to which the x-axis is rotated) rather than the angle.
(This is actually more convenient to measure if you happen to have a ruler,
which you need to measure lengths anyway, but not a protractor.) This avoids
trigonometry, but introduces ugly square roots. Show that the square roots
can be eliminated by using the slope of half the angle of transformation as
the variable. Show the relation to the variables used in writing 3D rotations
in terms of 2×2 matrices. (Hint: Consider U and V C diagonal.)

3. Spinors
Note that the mapping of SU(2) to SO(3) is two-to-one: This follows from the
fact V 0 = V when U is a phase factor. We eliminated continuous phase factors from
U by the condition det U = 1, which restricts U(2) to SU(2). However,

det(Ieiθ) = e2iθ = 1 ⇒ eiθ = ±1

for 2×2 matrices. More generally, for any SU(2) element U, −U is also an element of
SU(2), but acts the same way on a vector; i.e., these two SU(2) transformations give
the same SO(3) transformation. Thus SU(2) is called a “double covering” of SO(3).
However, this second transformation is not redundant, because it acts differently on
half-integral spins, which we discuss in the following subsections.
The other convenience of using 2×2 matrices is that it makes obvious how to
introduce spinors — Since a vector already transforms with two factors of U, we
define a “square root” of a vector that transforms with just one U:

ψ 0 = Uψ ⇒ ψ †0 = ψ †U −1

where ψ is a two-component “vector”, i.e., a 2×1 matrix. The complex conjugate of

a spinor then transforms in essentially the same way:

(Cψ*)0 = CU*ψ* = U(Cψ*)

Note that the antisymmetry of C implies that ψ must be complex: We might think
that, since Cψ* transforms in the same way as ψ, we can identify the two consistently
with the transformation law. But then we would have

ψ = Cψ* = C(Cψ*)* = CC*ψ = −ψ

 A. TWO COMPONENTS 67

Thus the representation is pseudoreal. The fact that Cψ* transforms the same way
under rotations as ψ leads us to consider the transformation

ψ 0 = Cψ*

Since a vector transforms the same way under rotations as ψψ †, under this transfor-
mation we have
V 0 = CV *C = −V

which identifies it as a reflection.

Another useful way to write rotations on ψ (like looking at V C instead of V ) is

(ψ T C)0 = (ψ T C)U −1

This tells us how to take an invariant inner product of spinors:

ψ 0 = Uψ, χ0 = Uχ ⇒ (ψ T Cχ)0 = (ψ T Cχ)

In other words, C is the “metric” in the space of spinors. An important difference of

this inner product from the familiar one for three-vectors is that it is antisymmetric.
Thus, if ψ and χ are anticommuting spinors,

ψ T Cχ = −χT C T ψ = χT Cψ

where one minus sign comes from anticommutativity and another is from the anti-
symmetry of C. Thus, it makes sense to take the norm of an anticommuting spinor
as ψ T Cψ, which would vanish if ψ were commuting. Of course, since rotations are
unitary, we also have the usual ψ † ψ as an invariant, positive definite, inner product.

4. Indices
The best way to discuss general spins is to use index notation, rather than matrix
notation. Then a spinor rotates as

ψα0 = Uα β ψβ

with two-valued indices α = ⊕, . The inner product is defined by

ψ αχα = ψ αCβαχβ = −ψαχα

where we have defined raising and lowering of indices by

ψα = ψ β Cβα, ψ α = C αβ ψβ
68 II. SPIN

0 −i


Cαβ = −Cβα = −C αβ = C βα = i 0

paying careful attention to signs. (In general, we fix signs by using a convention of
contracting indices from upper-left to lower-right.) Then objects with many indices
transform as the product of spinors:

A0αβ...γ = Uα δ Uβ  ...Uγ ζ Aδ...ζ

An infinitesimal transformation is then a sum:

−iδAαβ...γ = Gα δ Aδβ...γ + Gβ δ Aαδ...γ + ... + Gγ δ Aαβ...δ

This is also true for Cαβ , even though it is an invariant constant:

0
Cαβ = Uαγ Uβ δ Cγδ = Cαβ det U = Cαβ

A more interesting case is the vector: The transformation law is

0
Vαβ = Uαγ Uβ δ Vγδ

where Vαβ is the symmetric V C considered earlier (in contrast to the antisymmetric
C).
There is basically only one identity in index notation, namely

0 = 12 C[αβ Cγ]δ = Cαβ Cγδ + Cβγ Cαδ + CγαCβδ

The expression vanishes because it is antisymmetric in those indices, and thus the
indices must all have different values, but there are three two-valued indices. Another
way to write this identity is to use the definition of C αβ as the inverse of Cαβ :

Cαγ C βγ = δαβ ⇒ γ δ
Cαβ C γδ = δ[α δβ] ≡ δαγ δβδ − δβγ δαδ

This tells us that antisymmetrizing in any pair of indices automatically contracts

(sums over) them: Contracting this identity with an arbitrary tensor Aγδ ,

A[αβ] = Cαβ C γδ Aγδ = −Cαβ Aγ γ

That means that we need to consider only objects that are totally symmetric in their
free indices. This gives all spins: Such a field with 2s indices describes spin s; we have
already seen spins 0, 1/2, and 1.
We have defined the transformation law of all fields with lower indices by con-
sidering the direct product of spinors. Transformations for upper indices follow from
multiplication with C αβ : They all follow from

ψ 0α = ψ β (U −1 )β α
 A. TWO COMPONENTS 69

Since the vertical position of the index indicates the form of the transformation law,
we define
ψ̄α = (ψ α)*

where the “ ” indicates complex conjugation. Thus, a hermitian matrix is written

as
β
Mα β = (Mβ α )* = M α ⇒ Mαβ = M βα

So, for a vector we have

Vαβ = V αβ = Vβα

Spin s is usually formulated in terms of a (2s+1)-component “vector”. Then

one needs to calculate Clebsch-Gordan-Wigner coefficients to construct Hamiltonians
relating different spins. For example, to couple two spin-1/2 objects to a spin-1 object,
one might write something like V ~ · ψ †~σχ. The matrix elements of the Pauli matrices
~σ are the CGW coefficients for the spin-1 piece of 12 ⊗ 12 = 1 ⊕ 0. This method gets
progressively messier for higher spins. On the other hand, in spinor notation such a
term would be simply V αβ ψ̄αχβ ; no special coefficients are necessary, only contraction
of indices. Similarly the decomposition of products of spins involves only the picking
out of the various symmetric and antisymmetric pieces: For example, for 12 ⊗ 12 ,

ψαχβ = 12 (ψ(αχβ) + ψ[αχβ]) = 12 ψ(αχβ) − Cαβ ψ γ χγ = Vαβ + Cαβ S

where (αβ) means to symmetrize in those indices, by adding all permutations with
plus signs. We have thus explicitly separated out the spin-1 and spin-0 parts V and
S of the product. The square roots of various integers that appear in the CGW
coefficients come from permutation factors that appear in the normalizations of the
various fields/wave functions that appear in the products: For example,

Aαβγ Āαβγ = |A⊕⊕⊕|2 + 3|A⊕⊕ |2 + 3|A⊕ |2 + |A |2

In the spinor index method, the square roots never appear explicitly, only their squares
appear in normalizations: For example, in calculating a probability for A ⊗ B → C,
we evaluate
hA ⊗ B|CihC|A ⊗ Bi
hA|AihB|BihC|Ci
where A, B, and C each have 2s indices for spin s, and h|i means contracting all
indices (with the usual complex conjugation).
70 II. SPIN

5. Lorentz
Consider now the general 2×2 hermitian matrix
   
† V+ Vt * 1 V0 + V1 V2 − iV3
(V ) . = V = = √2 = Va (σ a ) .
αβ Vt V− V2 + iV3 V0 − V1 αβ

where we distinguish the right spinor index by a dot because it will be chosen to
transform differently from the left one. For comparison, raising both spinor indices
with the matrix C as for SU(2), and the vector indices with the Minkowski metric (in
either the orthonormal or null basis, as appropriate — see subsection IA4), we find
another hermitian matrix
.  V + V t*   0
V + V 1 V 2 + iV 3
 .
αβ √1 a αβ
(V ) = = = V (σ a )
Vt V− 2 V 2 − iV 3 V 0 − V 1
In the orthonormal basis, σa are the Pauli matrices and the identity, up to normal-
ization. They are also the Clebsch-Gordan-Wigner coefficients for spinor⊗spinor =
vector. In the null basis, they are completely trivial: 1 for one element, 0 for the rest,
the usual basis for matrices. In other words, they are simply an arbitrary way (ac-
cording to choice of basis) to translate a 2×2 (hermitian) matrix into a 4-component
vector.
Examining the determinant of (either version of) V , we find the correct Minkowski
norms:

−2 det V = −2V +V − + 2V t V t * = −(V 0)2 + (V 1 )2 + (V 2 )2 + (V 3 )2 = V 2

Thus Lorentz transformations will be those that preserve the hermiticity of this matrix
and leave its determinant invariant:

V 0 = gV g † , det g = 1

(det g could also have a phase, but that would cancel in the transformation.) Thus g
is an element of SL(2,C). In terms of the representation of the Lie algebra,

g = eG , tr G = 0

Thus the group space is 6-dimensional (G has three independent complex compo-
nents), the same as SO(3,1) (where gηg T = η ⇒ (Gη)T = −Gη).
Excercise IIA5.1
SL(2,C) also can be seen (less conveniently) from vector notation:
a Consider the generators
(±)
Jab = 12 (Jab ± i 12 abcd J cd )
 A. TWO COMPONENTS 71

of SO(3,1). Find their commutation relations, and in particular show

(±) (±) (±)
[J (+), J (−) ] = 0. Express J0i in terms of Jij . Show Jij have the same
commutation relations as Jij . Finally, take a general infinitesimal Lorentz
(±)
transformation in terms of Jab and rewrite it in terms of Jij , paying special
attention to the reality properties of the coefficients. This demonstrates that
the algebra of SO(3,1) is the same as that of SU(2)⊗SU(2), but Wick rotated
to SL(2,C).
b Apply the same procedure to SO(4) and SO(2,2) to derive their covering
groups.
Excercise IIA5.2
Consider relativity in two dimensions (one space, one time):
a Show that SO(1,1) is represented in lightcone coordinates by

x0+ = Λx+, x0− = Λ−1 x−

for some (nonvanishing) real number Λ, and therefore SO(1,1) = GL(1). Write
this one Lorentz transformation, in analogy to excercise IIA1.2a on rotations
in two space dimensions, in terms of an analog of the angle (“rapidity”) for
those transformations that can be obtained continuously from the identity.
Do the relativistic analog of excercise IIA2.2.
b Still using lightcone coordinates, find the parity and time reversal transfor-
mations. Note that writing Λ as an exponential, so it can be obtained con-
tinuously from the identity, restricts it to be positive, yielding a subgroup of
GL(1). Explicitly, what are the transformations of O(1,1) missing from this
subgroup? Which of P, (C)T, and (C)PT are missing from these transforma-
tions, and which are missing from GL(1) itself?
In index notation, we write for this vector
.
0 γδ
V . = gα g* V .
.
αβ β γδ

while for a (“Weyl”) spinor we have

ψα0 = gα β ψβ

The metric of the group SL(2,C) is the two-index antisymmetric symbol, which is
also the metric for Sp(2,C): In our conventions,

0 −i


Cαβ = −Cβα = −C αβ = C . . = i 0
αβ
72 II. SPIN

We also have the identities

det Lαβ = 12 C αβ Cγδ Lα γ Lβ δ = 12 (tr L)2 − 12 tr(L2), (L−1 )δ β = C αβ Cγδ Lα γ (det L)−1

A[αβ] = Cαβ C γδ Aγδ , A[αβγ] = 0

discussed earlier in this section. As there, we use the metric to raise, lower, and
contract indices: .
ψα = ψ β Cβα, ψα. = ψ β C . .
βα
.
V · W = V αβ W .
αβ

These results for SO(3,1) = SL(2,C) generalize to SO(4) = SU(2)⊗SU(2) and

SO(2,2) = SL(2)⊗SL(2). As described earlier, the reality conditions change, so now
0 0 0 0
SO(4) : (V αβ )* = V γδ Cγα Cδ0 β 0 , SO(2, 2) : (V αβ )* = V γδ

consistent with the (pseudo)reality properties of spinors for SU(2) and SL(2), where
we now use unprimed and primed indices for the two independent group factors
(V → gV g 0 ).
Excercise IIA5.3
Take the explicit 2×2 representation for a vector given above, change the
factors of i to satisfy the new reality conditions for SO(4) and SO(2,2), and
show the determinant gives the right signatures for the metrics.
A common example of index manipulation is to use antisymmetry. . whenever pos-
αβ γ δ . .
sible to give vector products. For example, from the fact that V V C is antisym-
δβ
metric in αγ we have that .
V αβ V . = 12 δγαV 2
γβ
where the normalization follows from tracing both sides. Similarly,
. .
V W + W V . = δγαV · W
αβ . αβ
γβ γβ

It then follows that

. . . . . .
V αβ W .V γ δ = (δγαV · W − W αβ V .)V γ δ = V · W V αδ − 12 V 2 W αδ
γβ γβ

Antisymmetry in vector indices also implies some antisymmetry in spinor indices.

For example, the antisymmetric Maxwell field strength Fab = −Fba, after translating
vector indices into spinor, can be separated into its parts symmetric and antisymmet-
ric in undotted indices; antisymmetry in vector indices (now spinor index pairs) then
implies the opposite symmetry in dotted indices:
.
F . . = −F . . = 14 (F
αγ,β δ β δ,αγ
.. + F . . ) = C̄ . .fαβ + Cαβ f¯. .,
(αβ)[γδ] [αβ](γδ) γδ
. γ
fαβ = 12 Fαγ,β
γδ
 A. TWO COMPONENTS 73

Thus, an antisymmetric tensor also can be written in terms of a (complex) 2×2

matrix.
We also need to define complex (hermitian) conjugates carefully because C is
imaginary, and uses indices consistent with transformation properties:
. .
ψ̄ α = (ψ α)* ⇒ ψ̄α. = −(ψα )*, (ψ α ψα)† = ψ̄ α ψ̄α.
. . . .
αβ
V = (V βα
)* ⇒ x̄ αβ
=x αβ

where we have used the spacetime coordinates as an example of a real vector (her-
mitian 2×2 matrix). In general, hermitian conjugation properties for any Lorentz
representation are defined by the corresponding product of spinors: For example,
. . . . ..
(ψ (αχβ))† = χ̄(αψ̄ β) = −ψ̄ (αχ̄β) ⇒ f¯αβ ≡ −(f αβ )*

More generally, we find

. . . .
(T (α1...αj )(β1 ...βk ) )† ≡ (−1)j(j−1)/2+k(k−1)/2T̄ (β1...βk )(α1 ...αj )

As we’ll see later, most spinor algebra involves, besides spinors, just vectors and
antisymmetric tensors, which carry only two spinor indices, so matrix algebra is often
useful. When using bra-ket notation for 2-component spinors, it is often convenient
to distinguish undotted and dotted spinors. Furthermore, since spinor indices can
be raised and lowered, we can always choose the bras to carry upper indices and the
kets lower, consistent with our index-contraction conventions, to avoid extra signs
and factors of C. We therefore define (see subsection IB1)
. .
hψ| = ψ αhα |, |ψi = |αiψα ; [ψ| = ψ α[α.|, |ψ] = |α ]ψα.
. . . .
V = | iVα [ .|
α β
⇒ V * = −| ]V α.hβ |,
α β
f = | ifα hβ |
α β
⇒ f* = | ]fα [ |
α
. β .
β β

As a result, we also have

.
hψχi = hχψi = ψ αχα, [ψχ] = ψ αχα.; hψχi† = [ψχ]
.
hψ|V |χ] = ψ α Vα β χ ., hψ|f|χi = ψ α fαβ χβ
β

V W * + W V * = (V · W )I

where we have used the anticommutativity of the spinor .fields. From now on, we use
this notation for the matrix representing a vector V (Vα β ), rather than the one with
.
which we started (V αβ ).
74 II. SPIN

Excercise IIA5.4
Consider the generators
.
Gα β = xβ γ ∂αγ. + |β ihα |
.
and their Hermitian conjugates, where ∂ . = ∂/∂xαβ . Show their algebra
αβ
closes. What group do they generate? Find a subset of these generators that
can be identified with (a representation of) the Lorentz group.
Since we have exhausted all possible linear transformations on spinors (except for
scale, which relates to conformal transformations), the only way to represent discrete
Lorentz transformations is as antilinear ones:
√ . √
ψα0 = 2nα β ψ̄ . (ψ 0 = − 2nψ*)
β

From its index structure we see that n is a vector, representing the direction of the
reflection. The product of two identical reflections is then, in matrix notation

ψ 00 = 2n(nψ*)* = n2 ψ ⇒ n2 = ±1

where we have required closure on an SL(2,C) transformation (±1). Thus n is a unit

vector, either
. spacelike or timelike.
. Applying the same transformation to a vector,
αβ α β
where V transforms like ψ χ , we write in matrix notation

V 0 = −2nV *n = n2 V − 2(n · V )n

(The overall sign is ambiguous, and depends on whether it is a polar or axial vector.)
This transformation thus describes parity (actually CP, because of the complex con-
jugation). In particular, to describe purely CP without any additional rotation (i.e.,
exactly reflection of the 3 spatial axes), in our basis we must choose a unit vector in
the time direction,
√ αβ. . .
2n = δ αβ ⇒ ψ 0α = ψ̄α. (ψ̄ 0α = −ψα)
.
⇒ V 0αβ = −Vβ α.
which corresponds to the usual in vector notation, since in our basis
.
σaαβ = σβaα.

To describe time reversal, we need a transformation that does not preserve the com-
plex conjugation properties of spinors: For example, CPT is
. .
ψ 0α = ψ α , ψ̄ 0β = −ψ̄ β ⇒ V 0 = −V
 A. TWO COMPONENTS 75

(The overall sign on V is unambiguous.)

In principle, whenever we work on a problem with both spinors and vectors we
could use a mixed vector-spinor notation, converting between the usual basis for
vectors and the spinor-index basis with identities such as
. . .
σ .σbαβ
a
= δab , σ .σaγ δ
a
= δαγ δ δ.
αβ αβ β

However, in practice it’s much simpler to use spinor indices exclusively, since then
one needs no σ-matrix identities at all, but only the trivial identities for the matrix
C that follow from its antisymmetry. For example, converting the vector index on
.
the σ matrices themselves into spinor indices (a → αβ), they become trivial:
. .
.αβ α β
(σ ) = δγ δ .
γδ δ

(This is the same as saying an orthonormal basis of vectors has the components
(V a )b = δba when the components are defined with respect to the same basis.)
Thus, the most general irreducible (finite-dimensional) representation of SL(2,C)
(and thus SO(3,1)) has an arbitrary number of dotted and undotted indices, and is
totally symmetric in each: A . . . Treating a vector index directly as a
(α1...αm)(β 1 ...βn )
.
dotted-undotted pair of indices (e.g., a = αα, which is just a funny way of labeling
a 4-valued index), we can translate into spinor notation the two constant tensors of
SO(3,1): Since the only constant tensor of SL(2,C) is the antisymmetric symbol, they
can be expressed in terms of it:

η . . = Cαβ C . .,  . . . . = i(Cαβ Cγδ C . .C . . − Cαδ Cβγ C . .C . .)

αα,β β αβ αα,β β,γ γ,δ δ αδ β γ αβ γ δ

When we work with just vectors, these can be expressed in matrix language:

V · W = tr(V W *)

abcd V aW b X c Y d ≡ (V, W, X, Y ) = i tr(V W *XY * − Y *XW *V )

Excercise IIA5.5
Prove this expression for the  tensor (in either index or matrix version)
by (1) showing total antisymmetry, (2) explicitly evaluating a nonvanishing
component.
76 II. SPIN

6. Dirac
The Dirac spinor we encountered earlier is a 4-component reducible representation
in D=4: in terms of two (“left” and “right”) two-component spinors,
 
ψLα
Ψ=
ψ̄Rα.
The Hermitian metric Υ that defines the (Lorentz-invariant) Dirac spinor inner prod-
uct
.
Ψ̄Ψ = Ψ† ΥΨ = ψLα ψRα + h.c., Ψ̄ ≡ Ψ† Υ = (ψRα ψ̄Lα)
takes the simple form  .. 
0 C̄ αβ √
Υ= = 2γ0
C αβ 0
The Dirac matrices are given by
 .  
0 Vα β 0 V
V/ ≡ γ · V = =
V β α. 0 −V * 0
where the indices have been chosen to insure that the γ matrices always take a Dirac
spinor to the same type of spinor. Since {V/ , W
/ } = −V · W , the γ matrices satisfy

{γ a , γ b } = −η ab

The extra sign is the result of normalizing the γ’s to be pseudohermitian with respect
to the metric: Υγ † Υ−1 = +γ. This Dirac spinor can be made irreducible by imposing
a reality condition that relates ψL and ψR: The resulting “Majorana spinor” is then
 
1 ψα
Ψ = √2
ψ̄α.

The product of all the γ’s is a pseudoscalar:

!
√ −iδαβ 0
γ−1 = 2 2
 γ aγ bγ c γ d = √1 .
4! abcd 2
0 iδ β.
α

(This is usually called “γ5 ” in the literature for D = 4, or “γD ” for D 6= 4. We have
renamed it for consistency with dimensional reduction.) It can be used to project a
Dirac or Majorana spinor onto its two two-component spinors:
√


Π± = 12 (I ± 2iγ−1 ) = I0 00 , 00 0I

Various identities for these matrices can be derived directly from the anticommu-
tation relations: For example,

γ a γa = −2, γ a a/γa = a/, γ a a//bγa = a · b, γ a a//b/cγa = c//ba/

 A. TWO COMPONENTS 77

tr(I) = 4, tr(a//b) = −2a · b, tr(a//b/c/d) = a · b c · d + a · d b · c − a · c b · d

The trace identities follow from the fact that the only way to get a nonvanishing trace
out of a product of γ matrices is when there are terms proportional to the identity;
since {γ a , γ b } = −η ab , this only happens when the indices are pairwise identical. The
above results then follow from examination of relevant special cases. (Traces of odd
numbers of γ matrices vanish. An exception is γ−1 , until it is rewritten in terms of
its definition as the product of the other γ-matrices.)
Although use of the anticommutation relations is convenient for generalization of
such identities to arbitrary dimensions, 2-spinor bra-ket notation is easier for deriving
4D identities. Since a Dirac spinor is the direct sum of a Weyl spinor and its complex
conjugate, we write
. .
Ψ = |αiψLα + |α ]ψ̄Rα., Ψ̄ = ψRα hα | + ψ̄Lα [α.|

In this notation, there is no need to use a spinor metric Υ, just as in Minkowski

4-vector bra-ket notation there is no need for an explicit matrix to represent the
Minkowski metric: It is included implicitly in the definition of the inner product
for the basis elements (ha |b i = ηab or hα |β i = Cαβ ). Thus hermitian conjugation is
automatically pseudohermitian conjugation, etc.: Ψ̄ is Ψ†, from the effect of hermitian
conjugating the basis vectors along with the components of the spinors they multiply.
(See subsections IB4-5.) We then have simply
.
γ . = −|αi[ .| − | .]hα|; Π+ = |α ihα |, Π− = |α ihα.|
αβ β β
.
where we have replaced the vector index a → αβ on γa .
Excercise IIA6.1
Use this representation for the γ matrices and projection operators Π± for all
of the following:
a Derive .
γ αβ a/1 · · · a/2n+1 γ . = a/2n+1 · · · a/1
αβ
.
γ αβ a/1 · · · a/2nγ . = − 12 I tr(a/1 · · · a/2n ) − γ−1 tr(γ−1 a/1 · · · a/2n )
αβ

b Rederive the trace identities above. (Hint: For the last identity, use the
identity C[αβ Cγ]δ = 0 repeatedly.)
c Show that
tr[(Π+ − Π−)γαα.γ .γγ γ.γ .] = −i . . . .
ββ δδ αα,β β,γ γ,δ δ
by comparison with the expression of the previous subsection for .
78 II. SPIN

7. Chirality/duality
Π± are often called “chiral projectors”; 2-component spinors (not paired into
Dirac spinors) are often called “chiral spinors”, and appear in “chiral theories”; the
two 2-component spinors of a Dirac spinor are often labeled as having left and right
“chirality”; etc. When these two halves decouple, a theory can have a “chiral sym-
metry”
ψα0 = eiθ ψα
Since chirality is closely related to parity (chiral spinors can represent CP, but need to
be doubled to allow C, and thus P), Dirac spinors are often used to describe theories
where parity is preserved, or softly broken, or to analyze parity violation specifically,
using γ−1 to identify it.
A similar feature appears in electrodynamics. We first translate the theory into
spinor notation: The Maxwell field strength Fab is expressed in terms of the vector po-
tential (“gauge field”) Aa, with a “gauge invariance” in terms of a “gauge parameter”
λ with spacetime dependence. The gauge transformation δAa = −∂aλ becomes

A0 . = A . − ∂ .λ
αβ αβ αβ
.
where ∂ . = ∂/∂xαβ . It leaves invariant the field strength Fab = ∂[aAb] :
αβ

F . . = ∂αγ.A . − ∂ .Aαγ. = 14 (F . . + F[αβ](γ.δ)

.)
αγ,β δ βδ βδ (αβ)[γδ]
.
= C̄ . .fαβ + Cαβ f¯. ., fαβ = 12 ∂(αγ.Aβ)γ
γδ γδ

Maxwell’s equations are

∂ β γ.fβα ∼ Jαγ.
They include both the field equations (the hermitian part) and the “Bianchi identi-
ties” (the antihermitian part).
Excercise IIA7.1
We already saw V W * + W V * gave the dot product; show how V W * − W V *
is related to the “cross product” V[a Wb] .
Excercise IIA7.2
Write Maxwell’s equations, and the expression for the field strength in terms
of the gauge vector, in 2×2 matrix notation, without using C’s. Combine
them to derive the wave equation for A.
Maxwell’s equations now can be easily generalized to include magnetic charge by
allowing the current J to be complex. (However, the expression for F in terms of A is
 A. TWO COMPONENTS 79

no longer valid.) This is because the “duality transformation” that switches electric
and magnetic fields is much simpler in spinor notation: Using the expression given
above for the 4D Levi-Civita tensor using spinor indices,

0 0
Fab = 12 abcd F cd ⇒ fαβ = −ifαβ

More generally, Maxwell’s equations in free space (but not the expression for F in
terms of A) are invariant under the continuous duality transformation

0
fαβ = eiθ fαβ

(and J 0 . = eiθ J . in the presence of both electric and magnetic charges).

αβ αβ
Excercise IIA7.3
Prove the relation between duality in vector and spinor notation. Show that
Fab + i 12 abcd F cd contains only fαβ and not f . ..
αβ
Excercise IIA7.4
How does complexifying J . modify Maxwell’s equations in vector notation?
αβ
In even time dimensions, Wick rotation kills the i (or −i) in the spinor-index
expression for αβ 0,γδ0 ,ζ 0 ,ηθ0 . Since the (discrete and continuous) duality transformation
now contains no i, we can impose self-duality or anti-self-duality; i.e., that fαβ or fα0 β 0
vanishes, since they are now independent and real instead of complex conjugates.
These continuous chirality and duality symmetries on the field strengths generalize
to the free field equations for arbitrary massless fields in four dimensions. For reasons
to be explained in the following section, they distinguish the two polarizations of the
waves described by such fields. They are closely related to conformal invariance: In
higher dimensions, where not all free, massless theories are conformal (even on the
mass shell), these symmetries exist exactly for those that are conformal.

REFERENCES
1 L.D. Landau and E.M. Lifshitz, Quantum mechanics, non-relativstic theory, 2nd ed.
(Pergamon, 1965) ch. VIII:
review of SU(2) spinor notation.
2 B. van der Waerden, Göttinger Nachrichten (1929) 100:
SL(2,C) spinor notation.
3 E. Majorana, Nuo. Cim. 14 (1937) 171.
4 A. Salam, Nuo. Cim. 5 (1957) 299;
L. Landau, JETP 32 (1957) 407, Nucl. Phys. 3 (1957) 27;
T.D. Lee and C.N. Yang, Phys. Rev. 105 (1957) 1671:
identification of neutrino with Weyl spinor.
80 II. SPIN

............................ ............................
............................ B. POINCARÉ . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The general procedure for finding arbitrary representations of the Poincaré group
relevant to physics is to: (1) Describe spin 0. As we have seen, this means starting
with the coordinate representation, which is reducible, and apply the constraint p2 +
m2 = 0 to get an irreducible one. (2) Find arbitrary, finite-dimensional, irreducible
representations of the Lorentz group. This we have done in the previous section.
(3) Take the direct product of these two representations of the Poincaré group, which
give the orbital and spin parts of the generators. (The spin part of translations
vanishes.) We then need a further constraint to pick out an irreducible piece of this
product, which is the subject of this section.

1. Field equations
We first show how the derivation above of the massless particle from the confor-
mal particle for spin 0 can be generalized to all “spins”, i.e., all representations of
the Poincaré group in arbitrary dimensions. There is a way to do this in terms of
classical mechanics for all representations of the conformal group, by generalizing the
description of the classical spinning particle. However, by analyzing the conformal
particle quantum mechanically instead, applying a set of constraints, it will be clear
how to generalize from conformal particles to general massless particles by weakening
the constraints. The general idea is that the symmetry group for massive particles is
the Poincaré group, while that for massless particles includes also scale transforma-
tions, and finally conformal particles have also conformal boosts. So, starting with
the conformal group and dropping anything to do with conformal boosts will give
massless particles. Massive particles then follow from dimensional reduction: adding
a further spatial dimension and fixing its component of momentum to a constant,
the mass, so p2 → p2 + m2. The equation p2 + m2 = 0, as an operator equation
acting on a field or wave function is the “Klein-Gordon (or relativistic Schrödinger)
equation”. States or fields that satisfy this (and the other) field equation are called
“on-(mass-)shell”, while those that don’t (or for which the equations haven’t been
imposed) are “off-shell”.
We begin with a general representation of the conformal group SO(D,2) in terms
of generators GAB , where A, B are D+2-component vector indices. We then impose
constraints that are the conformally covariant form of p2 = 0: Identifying

(G+a , Gab , G+− , G−a ) = (P a , J ab, ∆, K a)

 B. POINCARÉ 81

(where A = (±, a)) as the generators for translations, Lorentz transformations, di-
latations, and conformal boosts, we see that

G AB = 12 GC(A GC B) − 1
D+2
η AB GCD GCD =0

is an irreducible piece of the product GG (symmetric and traceless) and includes:

(G ++, G +a , G ab, G +− , G −a , G −−) = (P 2, 12 {J ab, Pb } + 12 {∆, P a }, ...)

where “...” all have terms containing K a .

Excercise IIB1.1
Work out all the G’s in terms of P , J, ∆, and K.
In general theories, even massless ones, it is not always possible to have invariance
under conformal boosts. (We’ll see examples of this insubsection IXA7.) However, all
massless theories are scale invariant, at least at the free level. (In D=4, free massless
theories can always be made conformal on shell. However, the fact that even these
theories can have actions that are not invariant under conformal boosts proves that it
is sufficient to add just dilatations to the Poincaré group. Furthermore, the fact that
conformal boosts are not always an invariance in D>4 means that dropping them will
give results in a dimension-independent form.) Therefore, only G ++ and G +a can be
defined in general massless theories, but we’ll see that these are sufficient to define
the kinematics. The former is just the masslessness condition, which we used to pick
the constraints in the first place.
As we saw earlier, ∆ just scales xa: We can therefore write the relevant generators
as

P a = ∂ a, J ab = x[a∂ b] + S ab , ∆ = 12 {xa, ∂a } + w − 1 = xa∂a + w + D−2

2

(We have used the antihermitian form of the generators.) The “scale weight” w + D−2 2
is the real “spin” part of ∆, just as Sab is the spin part of the angular momentum
Jab . To preserve the algebra it must commute with everything, and thus we can
set it equal to a constant on an irreducible representation. We’ll see shortly that
its value is actually determined by the spin Sab . It is the engineering dimension of
the corresponding field. It has been normalized for later convenience; the value of
w depends on the representation of S ab , but is independent of D. The dilatation
generator ∆ is not exactly antihermitian because the integration measure dD x isn’t
invariant under scaling. This is another reason w is determined, by the free action.
The form we have given preserves reality of fields. The commutation relations for the
spin parts, and the total generators, are the same as those for the orbital parts; e.g.,
[c
[Sab, S cd ] = −δ[aSb] d]
82 II. SPIN

(A convenient mnemonic for evaluating this commutator in general is to use S ab →

x[a∂ b] instead.)
Excercise IIB1.2
Find an expression for K a in terms of x, ∂, S, and w that preserves the
commutation relations. Evaluate all the constraints G, and express the inde-
pendent ones in terms of just ∂, S, and w (no x).
Substituting the explicit representation of the generators into the constraint G +a,
and using the former constraint P 2 = 0 (when acting on wave functions on the right),
we find that all x dependence drops out, leaving for G +a the condition

Sa b ∂b + w∂a = 0

(paying careful attention to quantum mechanical ordering). This equation is the

general field equation for all spins (acting on the field strength), in addition to the
Klein-Gordon equation (which is redundant except for spin 0).
Excercise IIB1.3
Define spin for the conformal group by starting in D+2 dimensions: In terms
of the (D+2)-dimensional coordinates y A and their derivatives ∂A ,

GAB = y[A∂B] + SAB

Besides the previous conditions

y 2 = ∂ 2 = {y A, ∂A} = 0

impose the constraints, in analogy to the D-dimensional field equations, and

taking into account the symmetry between y and ∂,

SAB yB + wyA = SAB ∂B + w∂A = 0

Show that the algebra of constraints closes, if we include the additional con-
straint
1 C D
2 S(A SB)C + w(w + 2
)ηAB =0

Solve all the constraints with explicit y’s for everything with an upper “−”
index, reducing the manifest symmetry to SO(D−1,1), in analogy to the way
y 2 = 0 was solved to find y −. Write all the conformal generators in terms of
xa , ∂a , Sab , and w.
 B. POINCARÉ 83

2. Examples
We now examine the constraints Sab ∂b + w∂a = 0 in more detail. We begin by
looking at some simple (but useful) examples. The simplest case is spin 0:

Sab = 0 ⇒ w=0

The next simplest case (for arbitrary dimension) is the Dirac spinor:

Sab = − 12 [γa , γb ] ⇒ Sa b ∂b + w∂a = −γa γ b ∂b + (w − 12 )∂a

⇒ γ a ∂a = 0, w= 1
2

where we have separated out the pieces of the constraint that are irreducible with
respect to the Lorentz group (e.g., by multiplying on the left with γ a ). This gives
the (massless) “Dirac equation” ∂/ Ψ = 0. The next case is the vector: In terms of the
basis |V i = V a |a i, the spin is
Sab = |[aihb] |
However, the vector yields just another description of the scalar:
Excercise IIB2.1
Apply the field equations for general field strengths to the case of a vector
field strength.
a Find the independent field equations (assuming the field strength is not just
a constant)
∂[a Fb] = 0, ∂ aFa = 0, w=1
Note that solving the first equation determines the vector in terms of a scalar,
while the second then gives the Klein-Gordon equation for that scalar, and
the third fixes the weight of the scalar to be the same as that found by starting
with a scalar field strength.
b Solve the second equation first to find a gauge field that is not a scalar.
All other representations can be built up from the spinor and vector. As our final
example, we consider the case where the field is a 2nd-rank antisymmetric tensor,
[c
(Sab F )cd = δ[aFb] d] ⇒

(Sa b ∂b + w∂a )Fcd = 12 ∂[a Fcd] − ηa[c ∂ b Fd]b + (w − 1)∂a Fcd

⇒ ∂[aFbc] = ∂ bFab = 0, w=1
which are Maxwell’s equations, again separating out irreducible pieces (e.g., by tracing
and antisymmetrizing).
84 II. SPIN

Excercise IIB2.2
Verify the representation of Lorentz spin given above for Fab by finding the
commutation relations implied by this representation.
Excercise IIB2.3
Consider the field equations in 4D spinor notation for a general field strength,
totally symmetric in its m undotted indices and n dotted indices,
.
Sα ∂β γ − m∂αγ = S̄α ∂ . − n∂γ α. = 0,
β . . . β
w = 12 (m + n)
γβ

where the spin acts on spinors as

(Sαβ ψ)γ = Cγ(α ψβ), (S̄ . .ψ)γ. = C̄γ(

. α.ψ . ((Sαβ ψ)γ. = (S̄ . .ψ)γ = 0)
αβ β) αβ

a Show this implies .

.
∂ αγ ψ . = ∂ γβ ψ . =0
α...β... α...β...

b Translate the field equations into vector notation (in terms of Sab ), finding
Sa b ∂b + w∂a = 0 and an axial vector equation.
c Show that the two equations are equivalent by deriving the equations of part
a from Sa b ∂b + w∂a = 0 alone, and from the axial equation alone.
In each case, choosing the wrong scale weight w would imply the field was con-
stant. Note that we chose the field strength Fab to describe electromagnetism: The
arguments we used to derive field equations were based on physical degrees of free-
dom, and did not take gauge invariance into account. In chapter XII we use more
powerful methods to find the gauge covariant field equations for the gauge fields, and
their actions.

3. Solution
Free field equations can be solved easily in momentum space. Then the simplest
way to do the algebra is in the “lightcone frame”. This is a reference frame, obtained
by a Lorentz transformation, where a massless momentum takes the simple form

pa = δ+
a +
p

(using only rotations), or the even simpler form pa = ±δ+

a
(using also a Lorentz
boost), where again ± is the sign of the energy. In that frame the general field
equation Sa b ∂b + w∂a = 0 reduces to

S −i = 0, w = S +−
 B. POINCARÉ 85

The constraint S −i = 0 determines S +− to take its maximum possible value within

that irreducible representation, since S −i is the raising operators for S +− : For any
eigenstate of S +− ,

S +−|hi = h|hi ⇒ S +− (S −i |hi) = (S −i S +− + [S +− , S −i ])|hi = (h + 1)(S −i |hi)

The remaining constraint then determines w: It is the maximum value of S +− for

that representation. By parity (+ ↔ −), −w is the minimum, so

w ≥ 0; w=0 ⇔ S ab = 0

since if S +− = 0 for all states then S ab = 0 by Lorentz transformation. As we have

seen by other methods (but can easily be derived by this method), w = 12 for the
Dirac spinor and w = 1 for the vector; since general representations can be built from
reducing direct products of these, we see that w is an integer for bosons and half-
integer for fermions. If we describe a general irreducible representation by a Young
tableau for SO(D−1,1) (with tracelessness imposed), or a Young tableau times a
spinor (with also γ-tracelessness γ a ψa...b = 0), then it is easy to see from the results
for the spinor and vector, and antisymmetry in rows, that w is simply the number
of columns of the tableau (its “width”), counting a spinor index as half a column:
S +− just counts the maximum number of “−” indices that can be stuck in the boxes
describing the basis elements. (In fact, Dirac spinor ⊗ Dirac spinor gives just all
possible 1-column representations.)
This leaves undetermined only S ij and S +i . However, S +i (“creation operator”)
is canonically conjugate to S −i (“annihilation operator”), so its action has also been
fixed:
[S −i , S +j ] = δ ij S +− + S ij

(S ij vanishes for i = j, so S +i and S −i are conjugate, though not “orthonormal”. The

constant S +− was fixed above to be nonvanishing, except for the trivial case of spin
0.)
Thus only the “little group” SO(D−2) spin S ij remains nontrivial: The original
irreducible representation of SO(D−1,1) Lorentz spin S ab was a reducible representa-
tion of SO(D−2) spin S ij ; the irreducible SO(D−2) representation with the highest
value of S +− is picked out of this SO(D−1,1) representation. This solution also gives
the field strength in terms of the gauge field: Working with just the highest-S +− -
weight states is equivalent to working with the gauge field, up to factors of ∂ +.
86 II. SPIN

As an explicit example, for spin 1/2 we have simply γ − Ψ = 0, which kills half the
components, leaving the half given by γ +Ψ. For spin 1, we find

pb Fab = 0 ⇒ F −a = 0
⇒ only F +i 6= 0
p[a F bc] = 0 ⇒ only F +a 6= 0

In the “lightcone gauge” A+ = 0, we have F +i = ∂ +Ai , so the highest-weight part of

F ab is the transverse part of the gauge field. The general pattern, in terms of field
strengths, is then to keep only pieces with as many as possible upper + indices and
no upper − indices (and thus highest S +− weight). In terms of the vector potential,
we have
F ab ∼ p[a Ab] ⇒ only Ai 6= 0

The general rule for the gauge field is to drop ± indices, so the field becomes an
irreducible representation of SO(D−2). All + indices on the field strength are picked
up by the momenta, which also account for the scale weight of the field strength: All
gauge fields have w = 0 for bosons and w = 12 for fermions.
Excercise IIB3.1
Using only the anticommutation relations {γ a , γ b } = −η ab, construct projec-
tion operators from γ ± : These are operators ΠI that satisfy
 X X
ΠI ΠJ = δIJ ΠI no , ΠI = 1

Because of time reversal symmetry γ + ↔ −γ − (or parity γ + ↔ γ −), these

project onto two subspaces equal in size.
A method equivalent to using the lightcone frame is to perform a unitary trans-
formation U on the spin that is the inverse of the transformation on the coordi-
nates/momentum that would take us to the lightcone frame: We want a Lorentz
transformation Λa b on the field equations, which are of the form

Oab pb = 0, Oa b = Sab + wδab

that has the effect

UOab U −1 = Λac Oc d Λb d , Λb apb = p0a , p0a = δ+

a +
p ⇒

0 = UOab pb U −1 = Λa c Oc d Λb d pb = Λa c Oc d p0d ⇒ Oa b p0b = 0

If |ψi satisfies the original constraint, then U|ψi will satisfy the new one. If we like,
we can always transform back at the end. This is equivalent to a gauge transformation
in the field theory.
 B. POINCARÉ 87

It is easy to check that the appropriate operator is

+i pi /p+
U = eS

Any operator V a that transforms as a vector under S ab ,

[S ab, V c ] = V [a η b]c

but commutes with p, is transformed by U into UV U −1 = V 0 as

pi pi i 2
+ (p )
V 0+ = V +, V 0i = V i + V + , V 0− = V − + V i + V
p+ p+ 2(p+ )2
as follows from explicit Taylor expansion, which terminates because S +i act as low-
ering operators (as for conformal boosts in subsection IA6). This yields the desired
result
V+
V 0apa = V ap0a + + p2
2p
when we impose the field equation p2 = 0.
Excercise IIB3.2
Check this result by performing the transformation explicitly on the con-
straint. Before the transformation, the lightcone decomposition of the con-
straint is
(−S +− + w)p+ + S +i pi = 0

−S i− p+ + S ij pj + wpi − S i+ p− = 0

S −i pi + (−S −+ + w)p− = 0
Show that after this transformation, the constraint becomes

(−S +− + w)p+ = 0
p2
−S i−p+ + (−S +− + w)pi − 12 S +i =0
p+
p2 1 +i i p
2
S −i pi + (−S +− + w)p− − S +− − S p =0
p+ 2 p+2
Clearly these imply
w = S +− , S −i = 0

with p2 = 0.
On the other hand, if instead of using the lightcone identification of x+ as “time”,
we choose to use the usual x0 for purposes of finding the evolution of the system, then
we want to consider transformations that do not involve p0 , instead of not involving
88 II. SPIN

the “energy” p− . Thus, by p0 -independent rotations alone, the best we can do is to

choose
pi = 0, p1 = ω

i.e., we can fix the value of the spatial momentum, but not in a way that relates to
the sign of the energy. The result is then

p0 > 0 : pa = δ+
a +
p
a −
p0 < 0 : pa = δ− p

The result is similar to before, but now the positive and negative energy solutions are
separated: In this frame the field equations reduce to

p0 > 0 : S −i = 0, S +− = w
p0 < 0 : S +i = 0, S +− = −w

Thus, while w takes the same value as before, now the positive-energy states are
associated with the highest weight of S +− , while the negative-energy ones go with
the lowest weight (and nothing between). The unitary transformation that achieves
this result is a spin rotation that rotates S ab in the field equations with the same
effect as an orbital transformation that would rotate (p1 , pi ) → (ω, 0). By looking at
the special case D = 3 (where there is only one rotation generator), we easily find
the explicit transformation
  i  
−1 |p | 1i p
i
U = exp tan S
p1 |pi |

Excercise IIB3.3
Perform this transformation:
a Find the action of the above transformation on an arbitrary vector V a. (Hint:
Look at D = 3 to get the transformation on the “longitudinal” part of the
vector.) In particular, show that

V 0apa = V a p0a , p0a = δ0a p0 + δ1aω

b Show the field equations are transformed as

p0 10
S 0apa + wp0 ⇒ −ωS 10 + wp0 = p0 (w − ω
S ) − ω1 S 10p2
p0 10
S 1apa + wp1 ⇒ 1 i
ω
p (ωS 1i − p0 S 0i ) + p1 (w − ω
S )
p0 10
S iapa + wpi ⇒ −[δ ij − 1
ω(ω+p1 )
pi pj ](ωS 1j − p0 S 0j ) + pi (w − ω
S )
 B. POINCARÉ 89

Note that the first equation gives the time-dependent Schrödinger equation,
with Hamiltonian

H = w1 (S 10p1 − S 0ipi ) → 1 10
w
S ω

This diagonalizes the Hamiltonian H (in a representation where S 10 is diag-

onal). Thus the only independent equations are

p2 = 0, S 10 = (p0 )w, S 1i − (p0 )S 0i = 0

leading to the advertised result.

c Find the transformation that rotates to the pi direction instead of the 1 di-
rection, so
pi
H → − w1 S 0i j ω
|p |
4. Mass
So far we have considered only massless theories. We now introduce masses
by “dimensional reduction”, identifying mass with the component of momentum in
an extra dimension. As with the extra dimensions used for describing conformal
symmetry, this extra dimension is just a mathematical construct used to give a simple
derivation. (Theories have been postulated with extra, unseen dimensions that are
hidden by “compactification”: Space curls up in those directions to a size too small
to detect with present experiments. However, no compelling reason has been given
for why the extra dimensions should want to compactify.)
The method is to: (1) extend the range of vector indices by one additional spatial
direction, which we call “−1”; (2) set the corresponding component of momentum to
equal the mass,
p−1 = m
and (3) introduce extra factors of i to restore reality, since ∂−1 = ip−1 = im, by
a unitary transformation. Since all representations can be constructed by direct
products of the vector and spinor, it’s sufficient to define this last step on them. For
the scalar this method is trivial, since then simply p2 → p2 + m2 .
For the spinor, since any transformation on the spinor index can be written in
terms of the gamma matrices, and the transformation must affect only the −1 direc-
tion, we can use only γ−1 . (For even dimensions, we can identify the γ−1 of dimen-
sional reduction with the one coming from the product of all the other γ’s, since in
odd dimensions the product of all the γ’s is proportional to the identity.) We find
√ √
U = exp(−πγ−1/2 2) : γ−1 → γ−1 , γa → − 2γ−1 γa
90 II. SPIN

We perform this transformation directly on the spin operators appearing in the con-
straints, or the inverse transformation on the states. Dimensional reduction, followed
by this transformation, then modifies the massless equation of motion as
√
i∂/ → i∂/ − mγ−1 → − 2γ−1 (i∂/ + √m2 )

so i∂/ Ψ = 0 → (i∂/ + m
√
2
)Ψ = 0.
The prescription for the vector is

U = exp( 12 iπ|−1ih−1 |) : |−1 i → i|−1 i, h−1 | → −ih−1| (h−1 |−1 i = 1)

with the other basis states unchanged. This has the effect of giving each field a −i
for each (−1)-index. For example, for Maxwell’s equations
 
∂[aFbc] ∂[aFbc] (redundant)
∂[aFbc] → →
∂[aFb]−1 + imFab −i(∂[aFb]−1 − mFab)
 b  b
∂ Fab + imFa−1 ∂ Fab + mFa−1
∂ Fab →
b
→
a
∂ F−1a −i∂ aF−1a (redundant)
Note that only the mass-independent equations are redundant. Also, Fa−1 appears
explicitly as the potential for Fab, but without gauge invariance. Alternatively, we
can keep the gauge potential:
 
Fab = ∂[a Ab] Fab = ∂[aAb]
Fab = ∂[aAb] → →
Fa−1 = ∂a A−1 − imAa −iFa−1 = −i(∂aA−1 + mAa)
This is known as the “Stückelberg formalism” for a massive vector, which maintains
gauge invariance by having a scalar A−1 in addition to the vector: The gauge trans-
formations are now
 
δAa = −∂aλ δAa = −∂aλ
δAa = −∂a λ → →
δA−1 = −imλ −iδA−1 = −imλ

Excercise IIB4.1
Consider the general massive field equations that follow from the general
massless ones by dimensional reduction. One of these is

S−1 a ∂a + wim = 0

(before restoring reality). This scalar equation alone gives the complete field
equations for w=1/2 and 1 (antisymmetric tensors), 0 being trivial.
a Show that for w=1/2 it gives the (massive) Dirac equation.
b Expanding the state over explicit fields, find the covariant field equations it
implies for w=1. Show these are sufficient to describe spins 0 (vector field
 B. POINCARÉ 91

strength: see excercise IIB2.1) and 1 (Fab and Fa−1 ). Note that S −1a act as
generalized γ matrices (the Dirac matrices for spin 1/2, the “Duffin-Kemmer
matrices” for w=1), where

S ab = −[S −1a , S −1b ]

c Show that these covariant field equations imply the Klein-Gordon equation
for arbitrary antisymmetric tensors. Show that in D=4 all antisymmetric
tensors (coming from 0-5 indices in D=5) are equivalent to either spin 0 or
spin 1, or trivial. (Hint: Use abcd .)
d Consider the reducible representation coming from the direct product of two
Dirac spinors, and represent the wave function itself as a matrix:

S ij Ψ = S̃ ij Ψ + ΨS̃ ij

where i = (−1, a) and S̃ ij is the usual Dirac-spinor representation. Using the

fact that any 4×4 (in D=4) matrix can be written as a linear combination of
products of γ-matrices (antisymmetric products, since symmetrization yields
anticommutators), find the irreducible representations of SO(4,1) in Ψ, and
relate to part c.
Excercise IIB4.2
Solve the field equations for massive spins 1/2 and 1 in momentum space by
going to the rest frame.
The solution to the general massive field equations can also be found by going
to the rest frame: The combination of that and dimensional reduction is, in terms of
the massive analog of lightcone components,
√
p+ = √1 (p0
2
+ p−1 ) = 2m, p− = √1 (p0
2
− p−1 ) = 0, pi = 0

where pi are now the other D−1 (spatial) components. This fixing of the momentum
is the same as the lightcone frame except that p1 has been replaced by p−1 , and
thus pi now has D−1 components instead of D−2. The solution to the constraints is
thus also the same, except that we are left with an irreducible representation of the
“little group” SO(D−1) as found in the rest frame for the massive particle, vs. one
of SO(D−2) found in the lightcone frame for the massless case.
92 II. SPIN

5. Foldy-Wouthuysen
The other frame we used for the massless analysis, which involved only energy-
independent rotations, can also be applied to the massive case by dimensional reduc-
tion. The result is known as the “Foldy-Wouthuysen transformation”, and is useful for
analyzing interacting massive field equations in the nonrelativistic limit. Replacing
p1 → p−1 = m in our previous result, we have for the free case
   
−1 |~p | −1i p
i
U = exp tan S , UHU −1 = w1 S −10 ω
m |~p |
For purposes of generalization to interactions, it was important that in the free trans-
formation (1) we used only the spin part of a rotation, since the orbital part could
introduce explicit x dependence, and (2) we used only rotations, since a Lorentz
boost would introduce p0 dependence in the “parameters” of the transformation,
which could generate additional p0 (time derivative) terms in the field equation.
Excercise IIB5.1
Perform this transformation for the Dirac spinor, and then apply the reality-
restoring transformation to obtain
√
H → 2γ0 ω

√
We then can use the diagonal representation γ0 = I0 −I 0
/ 2. (We can de-
fine this representation, up to phases, by switching γ0 and γ−1 of the usual
representation.) In general the reality-restoring transformation will be unnec-
essary for any spin, since applying the field equation S −10 = ±w picks out a
representation of the “little group” SO(D−1).
In the interacting case the result generally can’t be obtained in closed form, so it is
derived perturbatively in 1/m. The goal is again a Hamiltonian diagonal with respect
to S −10 , to preserve the separation of positive and negative energies; we then can set
S −10 = w to describe just positive energies. We thus choose the transformation to
cancel any terms in H that are off-diagonal, which come from odd total numbers of
“−1” and “0” indices from the spin factors in any term: i.e., odd numbers of S 0i
and S −1i (e.g., the S 0i pi term in the original H). For example, for coupling to an
electromagnetic field, the exponent of U is generalized by covariantizing derivatives
(minimal coupling ∂ → ∇ = ∂ +iA), but also requires field-strength (E and B) terms
to cancel certain ones of those generated from commutators of these derivatives in
the transformation:

∇a = ∂ a + iAa ⇒ [∇a, ∇b] = iF ab

 B. POINCARÉ 93

Before performing this transformation explicitly for the first few orders, we con-
sider some general properties that will allow us to collect similar terms in advance.
(Few duplicate terms would appear to the order we consider, but they breed like
rabbits at higher orders.) We start with a field equation F that can be separated into
“even” terms E and “odd” ones O, each of which can be expanded in powers of 1/m:
X
∞ X
∞
−n
F =E +O : E= m En , O= m−n On
n=−1 n=0

Note that the leading (m+1 ) term is even; thus we choose only odd generators to
transform away the odd terms in F , perturbatively from this leading term:
X
∞
F 0 = eGF e−G , G= m−n Gn
n=1

Since F 0 is even while G is odd, we can separate this equation into its even and odd
parts as

F 0 = cosh(LG )E + sinh(LG )O
0 = sinh(LG )E + cosh(LG )O

(with LG = [G, ] as in subsection IA3). Since we can perturbatively invert any

Taylor-expandable function of LG that begins with 1, we can use the second equation
to give a recursion relation for Gn : Separating the leading term of F ,

E = mE−1 + ∆E, −m[G, E−1] = [G, ∆E] + LG coth(LG )O

which we can expand in 1/m [after Taylor expanding LG coth(LG )] to give an expres-
sion for [Gn , E−1 ] to solve for Gn . We can also use the implicit solution for [G, E]
directly to simplify the expression for F 0:

F 0 = E + tanh( 12 LG )O

For example, to order 1/m2 we have for F 0

0
F−1 = E−1 , F00 = E0 , F10 = E1 + 12 [G1 , O0]

F20 = E2 + 12 [G2 , O0] + 12 [G1 , O1]

To this order we therefore need to solve

−[G1 , E−1 ] = O0 , −[G2 , E−1 ] = O1 + [G1 , E0 ]

94 II. SPIN

For our applications we will always have

E−1 = − w1 S −10

unchanged by interactions. We have oversimplified things a bit in the above deriva-

tion: For general spin we need to consider more than just even and odd terms; we
need to consider all eigenvalues of S −10 :

[S −10, Fs ] = sFs

and find the transformation that makes F 0 commute with it (s = 0). The procedure
is to first divide into even and odd values of s, as above, then to divide the remaining
even terms in F 0 into twice even values of s (multiples of 4) as the new E 0 and twice
odd as the new O0, which are transformed away with the new twice odd G0 , and so
on. This very rapidly removes the lower nonzero values of |s| (1 → 2 → 4 → ...),
which has a maximum value of 2w (from the operators that mix the maximum value
S −10 = w with the minimum S −10 = −w). For example, for the case of most interest,
the Dirac spinor, the only eigenvalues (for operators) are 0 and ±1, so the original
even part does commute with S −10 , and the procedure need be applied only once.
Furthermore, terms in F of eigenvalue s can be generated only at order m1−s or
higher; so at any given order the procedure rapidly removes all undesired terms for
any spin.
Since the terms we want to cancel are exactly the ones with nonvanishing eigen-
values of S −10, they can always be written as [G, S −10 ] for some G, so we can always
find a transformation to eliminate them:

[S −10 , Gsn ] = sGsn ⇒ Gsn = − ws {[G, ∆E] + LG coth(LG )O}sn

(This is just diagonalization of a Hermitian matrix in operator language.) In partic-

ular for the Dirac spinor, since E−1 has only ±1 eigenvalues, it’s easy to see that not
only do all even operators commute with it, but all odd operators anticommute with

0 b

it. (Consider the diagonal representation of E−1 : { 10 −1
0
, a 0 } = 0.) We then have
simply

w= 1
2 ⇒ (E−1 )2 = 1 ⇒ [E−1 , ∆E] = {E−1 , O} = {E−1 , G} = 0

⇒ mG = − 12 {[G, ∆E] + LG coth(LG )O}E−1

As a final step, we can apply the usual transformation

−10 /w
U0 = eimtS
 B. POINCARÉ 95

which commutes with all but the p0 term in E0 to have the sole effect of canceling
E−1 , eliminating the rest-mass term from the nonrelativistic-style expression for the
energy.
For the minimal electromagnetic coupling described above, we have besides E−1

E0 = π 0, O0 = w1 S 0iπ i

where we have written π a = pa + Aa (instead of π a = −i∇a, to save some i’s). There

are no additional terms in F for minimal coupling for spin 1/2, but later we’ll need to
include nonminimal effective couplings coming from quantum (field theoretic) effects.
There are also extra terms for spins 0 and 1 because the field strength is not the same
as the fundamental field, so we’ll treat only spin 1/2 here, but we’ll continue to use
the general notation to illustrate the procedure. Using the above results, we find to
order 1/m2 for F 0
G1 = S −1i π i, G2 = wS 0iiF 0i

in agreement with with the free case up to field strength terms. The diagonalized
Schrödinger equation is then to this order, including the effect of U0 ,
0
F−1 = 0, F00 = π 0, F10 = − 2w [ 2 {S −1i , S 0j }iF ij + S −10 (π i)2 ]
1 1

F20 = − 14 [{S 0i, S 0j }(∂ iF 0j ) − S ij {iF 0i, π j }]

For spin 1/2 we are done, but for other spins we would need a further transformation
(before U0 ) to pick out the part of F20 that commutes with S −10 (by eliminating the
twice odd part); the final result is

F20 = 14 [ 12 ({S −1i , S −1j } − {S 0i , S 0j })(∂ i F 0j ) + S ij {iF 0i, π j }]

It can also be convenient to translate into ± notation (as for the massless case, but
with index 1 → −1): We then write
0
F−1 = 0, F00 = π 0, F10 = − 2w [ 2 {S +i , S −j }iF ij + S +− (π i )2]
1 1

F20 = − 14 [ 12 {S +(i , S −j) }(∂ i F 0j ) − S ij {iF 0i, π j }]

In this notation the eigenvalue of S +− = S −10 for any combination of spin operators
can be simply read off as the number of − indices minus the number of +.
Excercise IIB5.2
Find the Hamiltonian for spin 1/2 in background electromagnetism, expanded
nonrelativistically to this order, by substituting the appropriate expressions
for the spin operators in terms of γ matrices, and applying S +− = ±w on
96 II. SPIN

the right for positive/negative energy. (Ignore the reality-restoring transfor-

mation.) γ-matrix algebra can be performed directly with the spin operators:
For the Dirac spinor we have the identities

b d
S(a (b Sc) d) = 12 δ(a δc) − ηac η bd ⇒ {S +i , S −j } = 12 δ ij − 2S ij S +−

6. Twistors
Besides describing spin 1/2, spinors provide a convenient way to solve the condi-
tion p2 = 0 covariantly: Any hermitian matrix with vanishing determinant must have
a zero eigenvalue (consider the diagonalized matrix), and so such a 2×2 matrix can
be simply expressed in terms of its other eigenvector. Absorbing all but the sign of
the nontrivial eigenvalue into the normalization of the eigenvector, we have
. .
p2 = 0 ⇒ pαβ = ±pα pβ

for some spinor pα . Since p0 is the (canonical) energy, the ± is the sign of the energy.
This explains why time reversal (actually CT in the usual terminology) is not a linear
transformation. Note that pα is a commuting object, while most spinors are fermionic,
and thus anticommuting (at least in quantum theory). Such commuting spinors are
called “twistors”.
Excercise IIB6.1
Show that, in terms of its energy E and the angular direction (θ, φ) of its
3-momentum, a massless particle is described by the twistor
p
pα = 21/4 |E|(cos θ2 e−iφ/2 , sin θ2 eiφ/2)

One useful way to think of twistors is in terms of the lightcone frame. In spinor
notation, the momentum is .

p = ± 10 00
αβ

If we write an arbitrary massless momentum as a Lorentz transformation from this

lightcone frame, then the twistor is just the part of the SL(2,C) matrix that con-
tributes: . . . . . . .
p0αβ = pγ δ gγ α ḡ .β = ±δ⊕ δ . gγ αḡ .β = ±g⊕α ḡ⊕. β = ±pα pβ
γ δ
δ ⊕ δ

For this reason, the twistor formalism can be understood as a Lorentz covariant form
of the lightcone formalism.
The twistor construction thus gives a covariant way of constructing wave functions
satisfying the mass-shell condition (Klein-Gordon equation) for the massless case,
 B. POINCARÉ 97

ψ = 0, where = ∂ 2 = −p2 . We simply Fourier transform, and use the twistor

expression for the momentum, writing the momentum-space wave function in terms
of twistor variables (“Penrose transform”):
Z . .
ψ(x) = d2 pα d2 p̄α.[exp(ixαβ pα p̄ .)χ+(pα , pα.) + exp(−ixαβ pα p̄ .)χ−(pα , p̄α.)]
β β

where χ± describe the positive- and negative-energy states, respectively. (The integral
over p̄α. can be performed also, effectively taking the Fourier transform with respect
.
to that variable only, treating ±xαβ pα as the conjugate.)
We can extend the matrix notation of subsection IIA5-6 to twistors:
.
hp| = pα , |pi = pα ; [p| = pα , |p] = pα.

P = |pi[p|, −P * = |p]hp|
As a result, we also have for twistors

hpqi = −hqpi, [pq] = −[qp]; hpqihrsi + hqrihpsi + hrpihqsi = 0; hpqi* = [qp]

These properties do not apply to physical, anticommuting spinors, where hψχi =

+hχψi, and hψψi =6 0.
Another natural way to understand twistors is through the conformal group. We
have already seen that the conformal group in D dimensions is SO(D,2). Since this
group in four dimensions is the same as SU(2,2), it’s simpler to describe its general
representations (and in particular spinors) in SU(2,2) spinor notation. Then the
simplest way to generate representations of this group is to use spinor coordinates:
We therefore write the generators as (see subsection IC1)

GA B = ζ B ζ̄A − 14 δA
B C
ζ ζ̄C

where we have subtracted out the trace piece to reduce U(2,2) to SU(2,2) and, consis-
tently with the group transformation properties under complex conjugation, we have
chosen the complex conjugate of the spinor to also be the canonical conjugate: The
Poisson bracket is defined by
[ζ̄A , ζ B ] = δA
B

To compare with four-dimensional notation, we reduce this four-component spinor by

recognizing it as a particular use of the Dirac spinor. Using the same representation
as in subsection IIA6, we write
.  .. 
. 0 C̄ αβ
ζ A = (pα , ω̄α), ζ̄A = (ωα , p̄α.); ΥAB =
C αβ 0
98 II. SPIN

. .
β β
ωα , p δαβ , .
ωα , pβ
δ.
α

.
pα p ., ωα ω ., p(αωβ) , p(α.ω . , pα ωα pα ωα. −
β β β)

y2 1

4 ABCD
y AB y CD ⇒ y AB z Aα z B α

A α z Aα

y −
z · −
y

. .
A µ, µ , z Aα λα ν δνµ , xν µ ⇒ SL(2) gauge λα ν λδαν

e λ2
Excercise IIB6.2
z λ, x y ∼ z2
y e, x

7. Helicity
S ij
1 1
Wabc 2 P[a Jbc] 2 P[a Sbc]

S ij
Sab
 B. POINCARÉ 99

with the field equations is proportional to the field equations (i.e., it preserves the
constraints). In D=4 this is the “Pauli-Lubański (axial) vector”

Wa = 16 W bcd bcda

We can choose our states to be eigenstates of a component of it: For example, for
massless states W 0/P 0 is called the “helicity”. For massive states the helicity is
defined as W 0/|P~ |, but is less useful, especially since it is undefined (0/0) in the rest
frame. In that case one instead chooses a component in terms of a (momentum-
dependent axial) vector sa as sa Wa , where sa Pa = 0 and s2 = 1/m2 .
Excercise IIB7.1
Show in both the massless and massive cases that Wabc reduces to the little
group generators on shell by going to the appropriate reference frame.
The twistor representation of the conformal group does not give the most general
representation, but it does give all the (free) massless ones. The reason it gives
massless ones is that this representation satisifies the constraint (see subsection IIB1)

G[AB][CD] = G[A [C GB] D] − traces = 0

which includes p2 = 0 as well as all the equations that follow from p2 = 0 by conformal
transformations. As a consequence, this representation also satisfies

GA C GC B − trace = hGA B

where h is the helicity. This equation may be more recognizable in SO(4,2) notation,
as
1 ABCDEF
8
 GCD GEF = ihGAB

This equation includes, as its lowest mass-dimension part (as defined by dilatations),
the Pauli-Lubański vector

W a = 12 bcda Pb Jcd = ihP a

(The “i” appears in the last two equations only when we use the antihermitian form of
the generators GAB and Jab .) Although any massless representation of the conformal
group satisfies the above conditions (see excercise IIB2.3), the twistor representation
satisfies the unusual property that helicity is realized as a linear transformation on the
coordinates: For the twistors the implicit definition of helicity can be solved explicitly
to give
.
h = 14 {ζ̄ A , ζA } = 12 ζ̄ A ζA + 1 = 12 (pα ωα − p̄α ω̄α.)
100 II. SPIN

which is exactly the U(1) transformation of U(2,2)=SU(2,2)⊗U(1). (This is similar

to SU(2) in terms of “twistors”: See excercise IC1.1.) On functions of pα and p̄α., it
effectively just counts half the number of pα ’s minus p̄α. ’s.
Excercise IIB7.2
These results are pretty clear from symmetry, but we should do some algebra
to check coefficients:
a Use the definition of the action of the Lorentz generators on a vector operator
in vector and spinor notations,

[Jab, Vc ] = V[a ηb]c ; [Jαβ , V .] = V .Cβ)γ , [J¯. ., V .] = Vγ(α.C̄ . .

γδ (αδ αβ γδ β)δ

to derive
J . . = − 12 (Cαβ J¯. . + C̄ . .Jαβ )
ααβ β αβ αβ

b Express Jab and Pa in terms of the twistors pα , p̄α. , ωα, ω̄α. (with normalization
of Jab fixed by its action on the twistors themselves), and plug into P J = ihP
to derive the above expression of h in terms of twistors.
The simple form of the helicity in the twistor formalism is another consequence
of it being a covariantized lightcone formalism. In the lightcone frame, there is still
a residual Lorentz invariance; in particular, a rotation about the spatial direction in
which the momentum points leaves the momentum invariant. This is another defini-
tion of the helicity, as the part of the angular momentum performing that rotation.
(Only spin contributes, since by definition the momentum is not rotated.) Since the
product of two Lorentz transformations is another one, this rotation can be inter-
preted as a transformation acting on the Lorentz transformation to the lightcone
frame, i.e., on the twistor, such that the momentum is invariant. This is simply a
phase transformation:
 iθ γ
0 β e 0
gα = −iθ
gγ β ⇒ p0α = eiθ pα
0 e α

We can generalize the Penrose transform in a simple way to wave functions car-
rying indices to describe spin:
Z
ψ . . (x) = d2 pα d2 p̄α. pα1 · · · pαm p̄ . · · · p̄ .
α1 ...αmβ 1 ...β n β1 βn
. .
× [exp(ix pα p̄ )χ+(pα , pα ) + exp(−ix pα p̄ .)χ−(pα , pα. )]
αβ . . αβ
β β

For the integral to give a nonvanishing result, the integrand must be invariant under
the U(1) transformation generated by the helicity operator h: In other words, χ± must
 B. POINCARÉ 101

have a transformation under h, i.e., a certain helicity, that is exactly the opposite that
of the explicit p factors that carry the external indices to give a contribution to the
integral, since otherwise integrating over the phase of pα would average it to zero.
(Explicitly, if we derive the helicity by acting on the Penrose transform, this minus
sign comes from integration by parts.) This means that ψ(x) automatically has a
certain helicity, half the number of dotted minus undotted indices:

h = 12 (n − m) [w = 12 (m + n)]

as given by the above twistor operator expression acting on χ±. (Alternatively, com-
paring the x-space form of the Pauli-Lubańsky vector, its action plus that of the
twistor-space one must vanish on |α ipα , so the helicity is again minus the twistor-
space helicity operator acting on the prefactor.)
Since, after restricting to the appropriate helicity, the integral over this phase is
trivial, we can also eliminate it by replacing the “volume” integral over the twistor
or its complex conjugate (but not both) with a “surface” (boundary) integral:
Z I
d pα → pα dpα
2

(Alternatively, we can insert a δ-function in the helicity.) The result is equivalent to

the usual integral over the three independent components of the momentum.
This generalization of the Penrose transform implies that ψ(x) satisfies some
equations of motion besides p2 = 0, namely
. .
pαα ψ . . = pγ γ ψ . .=0
α...β γ...δ α...β γ...δ

which are also implied by Sa b ∂b + w∂a = 0 (see excercise IIB2.3). Besides Poincaré
invariance, these equations are invariant under the phase transformation

ψ0 i2hθ
. δ.
. . = e ψα...β γ...
α...β γ...δ

that generalizes duality and chiral transformations. We also see that (anti-)self-
duality and chirality are related to helicity. Another way to understand the twistor
result is to remember its interpretation. as a Lorentz transformation from the light
.
cone: In the light cone frame, where p++ is the only nonvanishing component of. pαα. ,
the above. equations of motion imply the only nonvanishing component of ψ α. 1 ...αmβ1 ...βn
. .
is ψ +...++...+, which can be identified with χ+ (for p++ > 0) or χ− (for p++ < 0).

REFERENCES
1 E. Wigner, Ann. Math. 40 (1939) 149;
V. Bargmann and E.P. Wigner, Proc. Nat. Acad. Sci. US 34 (1946) 211:
102 II. SPIN

little group; more general discussion of Poincaré representations and relativistic wave
equations for D=4.
2 A.J. Bracken, Lett. Nuo. Cim. 2 (1971) 574;
A.J. Bracken and B. Jessup, J. Math. Phys. 23 (1982) 1925;
W. Siegel, Nucl. Phys. B263 (1986) 93:
conformal constraints.
3 E.C.G. Stückelberg, Helv. Phys. Acta 11 (1938) 299.
4 P.A.M. Dirac, Rev. Mod. Phys. 21 (1949) 392:
lightcone gauge.
5 G. Nordström, Phys. Z. 15 (1914) 504:
dimensional reduction.
6 O. Klein, Z. Phys. 37 (1926) 895;
V. Fock, Z. Phys. 39 (1927) 226:
mass from dimensional reduction.
7 R.J. Duffin, Phys. Rev. 54 (1938) 1114;
N. Kemmer, Proc. Roy. Soc. A173 (1939) 91.
8 M.H.L. Pryce, Proc. Roy. Soc. A195 (1948) 62;
S. Tani, Soryushiron Kenkyu 1 (1949) 15 (in Japanese):
free version of Foldy-Wouthuysen.
9 L.L. Foldy and S.A. Wouthuysen, Phys. Rev. 78 (1950) 29.
10 M. Cini and B. Touschek, Nuo. Cim. 7 (1958) 422;
S.K. Bose, A. Gamba, and E.C.G. Sudarshan, Phys. Rev. 113 (1959) 1661:
“ultrarelativistic” version of Foldy-Wouthuysen transformation.
11 R. Penrose, J. Math. Phys. 8 (1967) 345, Int. J. Theor. Phys. 1 (1968) 61;
M.A.H. MacCallum and R. Penrose, Phys. Rep. 6 (1973) 241:
twistors.
12 W. Pauli, unpublished;
J.K. Lubański, Physica IX (1942) 310, 325.
 C. SUPERSYMMETRY 103

.................... ....................
.................... C. SUPERSYMMETRY ....................

We’ll see later that quantum field theory requires particles with integer spin to be
bosons, and those with half-integer spin to be fermions. This means that any sym-
metry that relates bosonic wave functions/fields to fermionic ones must be generated
by operators with half-integer spin. The simplest (but also the most general, at least
of those that preserve the vacuum) is spin 1/2. Supersymmetry is a major ingredi-
ent in the most promising generalizations of the Standard Model. In this section we
look at representations, generalizing the results of the previous sections for Poincaré
symmetry.

1. Algebra
From quantum mechanics we know that for any operator A
X
hψ|{A, A†}|ψi = (hψ|A|nihn|A†|ψi + hψ|A†|nihn|A|ψi)
n
X
= (|hn|A†|ψi|2 + |hn|A|ψi|2) ≥ 0
n

from inserting a complete set of states. In particular,

{A, A†} = 0 ⇒ A=0

from examining the matrix element for all states |ψi. This means the anticommuta-
tion relations of the supersymmetry generators must be nontrivial.
We are then led to anticommutation relations of the form, in Dirac (Majorana)
notation,
√
{q, q̄} = p/ or {q, q †} = pa γa 2γ0

(We use translations instead of internal symmetry or Lorentz generators because of

dimensional analysis: Bosonic fields differ in dimension from fermionic ones by half
integers.) Note that this implies the positivity of the energy:
√ a √
tr{q, q †} = 2p tr(γa γ0 ) = 2pa tr( 12 {γa , γ0 }) = p0 √12 tr I

Similar arguments imply that the supersymmetry generators are constrained, just
as the momentum is constrained by the mass-shell condition. For example, in the
massless case,
{p/q, q̄p/} = p//p/p = − 12 p2 p/ = 0 ⇒ p/q = 0
104 II. SPIN

In four dimensions the commutation relations can be written in terms of irre-

ducible spinors as
. .
{q α, q̄ β } = pαβ , {q, q} = {q̄, q̄} = 0

This generalizes straightforwardly to more than one spinor, carrying a U(N) index:
. .
{qi α , q̄ j β } = δij pαβ

2. Supercoordinates
Since the momentum is usually represented as coordinate derivatives, we natu-
rally look for a similar representation for supersymmetry. We therefore introduce
an anticommuting spinor coordinate θα. Because of the anticommutation relations q
can’t be simply ∂/∂θ, but the modification is obvious:

∂ .
1 β ∂ ∂ ∂
qα = −i α + 2 θ̄ ., q̄α. = −i 1 β
. + 2θ .
∂θ ∂xαβ ∂ θ̄α ∂xβ α
We can also express supersymmetry in terms of its action on the “supercoordinates”:
.
Using the hermitian infinitesimal generator αqα + ¯αqα.,
. . . . .
δθα = α , δ θ̄α = ¯α , δxαβ = 12 i(αθ̄β + ¯β θα )
.
Note that (q α)† = q̄ α, (qα)† = −q̄α..
We can also define “covariant derivatives”: derivatives that (anti)commute with
(are invariant under) supersymmetry. These are easily found to be

∂ . ∂
1 β 1 β
dα = α + 2 θ̄ p ., d̄α. = . + 2 θ pβ α.
∂θ αβ
∂ θ̄α
Besides overall normalization factors of i, leading to the opposite hermiticity condition
.
(dα )† = −d̄α , these differ from the q’s by the relative sign of the two terms. These
changes combine to preserve
. .
{dα , d¯β } = pαβ

as a result of which p is also a covariant derivative as well as being a symmetry

generator (as for the Poincaré group), but now (dα )† = +d¯α..
In classical mechanics, the fact that ∂/∂x commutes with translations is “dual” to
the fact that the infinitesimal change dx, or the finite change x − x0, is also invariant
under translations. Furthermore, the d’Alembertian = (∂/∂x)2 being Poincaré
invariant is dual to the line element ds2 = −(dx)2 being invariant. This allows the
 C. SUPERSYMMETRY 105

.
construction of the action from x2 . In the supersymmetric case the infinitesimal
invariants under the q’s (and therefore p) are
. . . .
α α αβ 1 α β 1 β α
dθ , dθ̄ , dx + 2 i(dθ )θ̄ + 2 i(dθ̄ )θ

and the corresponding finite ones (by integration) are

. . . . . .
θα − θ0α, θ̄α − θ̄0α , xαβ − x0αβ + 12 iθαθ̄0β + 12 iθ̄β θ0α

Although these can be used to construct classical mechanics actions, their quantiza-
tion is rather complicated. Just as for particles of one particular spin, direct treatment
of the quantum mechanics has proven much simpler than deriving it by quantization
of a classical system.
Excercise IIC2.1
Check explicitly the invariance of the above infinitesimal and finite differences
under supersymmetry.
Now that we have a (super)coordinate representation of the supersymmetry gen-
erators, we can examine the wave functions/fields that carry this representation. Such
“superfields” can be Taylor expanded in the θ’s with a finite number of terms, with
ordinary fields as the coefficients. For example, if we expand a real (hermitian) scalar
superfield
.
Φ(x, θ, θ̄) = φ(x) + θα ψα(x) + θ̄α ψ̄α.(x) + ...

and also expand its supersymmetry transformation

. . .
δΦ =  ψα + ¯ ψ̄α + 2 i θ̄ ∂ .φ + 12 iαθβ ∂β α.φ + ...
α α . 1 α β
αβ

we find the component field transformations

. .
δφ = α ψα + ¯α ψ̄α., δψα = − 12 i¯β ∂ .φ + ..., δ ψ̄α. = − 12 iβ ∂β α.φ + ..., ...
αβ

which mix the different spins.

An alternative, and more convenient, way to define the θ expansion is by use of
the covariant derivatives. Using “|” to mean “|θ=0 ”, we can define

φ = Φ|, ψα = (dα Φ)|, ψ̄α. = (d¯α.Φ)|, ...

There is some ambiguity at higher orders in θ because the d’s don’t anticommute, and
this can be resolved according to whatever is convenient for the particular problem,
avoiding field redefinitions in terms of fields appearing at lower order in θ: Since
the field equations must be covariant under supersymmetry (otherwise there is no
106 II. SPIN

advantage to using superfields), they must be written with the covariant derivatives.
Then one defines the component expansions by choosing the same ordering of d’s as
appear in the field equations (where relevant), which gives the component expansion
of the field equations the simplest form. It also gives a convenient method for deriving
supersymmetry transformations, since the d’s anticommute with the q’s:

δ[(d...dΦ)|] = [d...d(δΦ)]| = [d...d(iqΦ)]| = [(iq)d...dΦ]| = [(d)d...dΦ]|

where we have used the fact that q = −id + θ-stuff, where the θ-stuff is killed by
evaluating at θ = 0, once it has been pulled in front of all the θ-derivatives. Covariant
R
derivatives can also be used for integration, since dθ = ∂/∂θ = d up to an x-
R
derivative, which can be dropped when also integrating dx.

3. Supergroups
We saw certain relations between the lower-dimensional classical groups that
turned out to be useful for just the cases of physical interest of rotational (SO(D−1)),
Lorentz (SO(D−1,1)), and conformal (SO(D,2)) groups. In particular, the Poincaré
group, though not a classical group, is a certain limit (“contraction”) of the groups
SO(D,1) and SO(D−1,2), and a subgroup of the conformal group. Similar remarks
apply to supersymmetry, but because of its relation to spinors, these classical “su-
pergroups” (or “graded” classical groups) exist only for certain lower dimensions, the
same as those where covering groups for the orthogonal groups exist. In higher di-
mensions the supergroups do not correspond to supersymmetry, at least not in any
way that can be represented on physical states.
We’ll consider only the graded generalization of the classical groups that appear
in the bosonic case. The basic idea is then to take the group metrics and combine
them in ways that take into account the difference in symmetry between bosons and
fermions:
.
Unitary: ΥAB
OrthoSymplectic: M AB
Real: η .B
A
pseudoreal (*): Ω .B
A

where η is symmetric and Ω antisymmetric, as before, while M is graded symmetric:

For A = (a, α) with bosonic indices a and fermionic ones α,

M [AB) = 0 : M ab − M ba = M aβ − M βa = M αβ + M βα = 0
 C. SUPERSYMMETRY 107

Again we have inverse metrics, e.g.,

M KI MKJ = δJI

With respect to the usual index-contraction convention (no extra grading signs when
superscript is contracted with subscript immediately following), we should take the
ordering of indices on δ as δJ I .
There is no analog of the  tensor, at least for finite-dimensional groups, since it
would have an infinite number of indices when totally symmetric. However, “special”
supergroups can still be defined by generalizing the definition of trace and determinant
to supermatrices. One convenient way to do this is by using Gaussian integrals, since
this is a common way that such expressions will arise. As a generalization of the
bosonic and fermionic identities we therefore define the “superdeterminant”
Z
†
(sdet M) = N dz † dz e−z M z
−1

where “N” is a normalization factor defined so sdet I = 1. By explicitly evaluating

the integral, separating out the commuting and anticommuting parts, we find (see
excercise IB3.3)
 
A B det A det(A − BD−1 C)
sdet = =
C D det(D − CA−1B) det D
The “supertrace” (see also excercise IA2.3c) then can be defined by generalizing the
bosonic identity det(eM ) = etr M :

sdet(eM ) = estr M

str(MAB ) = (−1)A MAA = Ma a − Mα α = tr A − tr D

follows, as in the bosonic case, from δ ln sdet M = str(M −1 δM), which is derived by
varying the Gaussian definition.
A useful identity for superdeterminants can be derived by starting with the fol-
lowing identity for the inverse of a matrix for which the range of the indices has been
divided into two pieces:
   
a b −1 (a − bd−1 c)−1 (c − db−1 a)−1
=
c d (b − ac−1d)−1 (d − ca−1 b)−1
We have assumed all the submatrices are square and invertible; equivalent expressions,
which are more useful in other cases, can be derived easily by multiplying and dividing
by the submatrices: For example,
   
a b −1 (a − bd−1 c)−1 −a−1b(d − ca−1 b)−1
=
c d −d−1 c(a − bd−1 c)−1 (d − ca−1 b)−1
108 II. SPIN

From either of these we immediately see

     
A B −1 Ã B̃ A B 1
= ⇒ sdet = det A det D̃ =
C D C̃ D̃ C D det D det Ã

The graded generalizations of the classical groups are then

GL(m|n,C) [SL(m|n,C),SSL(n|n,C)]

U: [S]U(m+,m−|n) [SSU(n+,n−|n+ +n−)]

OSp: OSp(m|2n,C)
R: GL(m|n) [SL(m|n),SSL(n|n)]
*: [S]U*(2m|2n) [SSU*(2n|2n)]

U & OSp
R: OSp(m+,m−|2n)
*: OSp*(2m|2n)

where “(m|n)” refers to m bosonic and n fermionic indices, or vice versa. In the
matrices of the defining representation, the elements with one bosonic index and one
fermionic are anticommuting numbers, while those with both indices of the same kind
are commuting. In particular, the commuting parts give the bosonic subgroups:

GL(m|n,C) ⊃ GL(m,C)⊗GL(n,C)
SL(m|n,C) ⊃ GL(m,C)⊗SL(n,C)
SSL(n|n,C) ⊃ SL(n,C)⊗SL(n,C)
U(m+,m−|n) ⊃ U(m+,m−)⊗U(n)
SU(m+,m−|n) ⊃ U(m+,m−)⊗SU(n)
SSU(n+,n−|n+ +n−) ⊃ SU(n+ ,n−)⊗SU(n++n−)
OSp(m|2n,C) ⊃ SO(m,C)⊗Sp(2n,C)
GL(m|n) ⊃ GL(m)⊗GL(n)
SL(m|n) ⊃ GL(m)⊗SL(n)
SSL(n|n) ⊃ SL(n)⊗SL(n)
U*(2m|2n) ⊃ U*(2m)⊗U*(2n)
SU*(2m|2n) ⊃ U*(2m)⊗SU*(2n)
SSU*(2n|2n) ⊃ SU*(2n)⊗SU*(2n)
OSp(m+,m−|2n) ⊃ SO(m+ ,m−)⊗Sp(2n)
OSp*(2m|2n) ⊃ SO*(2m)⊗USp(2n)
 C. SUPERSYMMETRY 109

When the commuting and anticommuting dimensions are equal, we can impose trace-
lessness conditions on both bosonic parts of the generators separately (“SS”: tr A =
tr D = 0). This is related to the fact str(I) = 0 in such cases.

4. Superconformal
Since the conformal group is a classical group, its supersymmetric generalization
should be a classical supergroup. Because the fermionic generators must include the
supersymmetry generators, which are spinors, the representation of the conformal
group that appears in the defining representation of the supergroup must be the spinor
representation. However, we have seen that only for n≤6 (where covering groups exist)
and n=8 (where the spinor of SO(8) is another of its defining representations) can
the spinor representation of SO(n) be defined by classical group restrictions. This
implies that the superconformal group exists only in D≤4 and D=6.
The relevant supergroups can be identified easily by looking at the bosonic sub-
groups:

D = 3 : OSp(N|4)
4 : SU(2,2|N) (or SSU(2,2|4))
6 : OSp*(8|2N)

(We consider only D>2, since the conformal group is infinite-dimensional in D≤2.)
These three cases of D=3,4,6 are special for a number of reasons: In particular,
these three supergroups can be related to SU(N|4) over the division algebras: the
real numbers, complex numbers, and quaternions, respectively. (Similar remarks
apply to their important classical bosonic subgroups: the conformal, Lorentz, and
rotation groups. Attempts have been made to extend these results to the octonions
for D=10, but with less success, and there seems to be no superconformal group for
that case.) However, just as in the case of the Hilbert space of quantum mechanics,
the complex numbers seems to be the best of these “division algebras”, having the
analytic properties the real numbers lack, while avoiding the noncommutativity of
the quaternions. We’ll see later that nontrivial interacting (local, classical) conformal
field theories exist only in D≤4.
For example, for D=4 we find that the bosonic generators are the conformal group
and the internal symmetry group U(N) (or SU(4) for N=4), while the fermionic gener-
ators include supersymmetry (N spinors) and its fraternal twin, “S-supersymmetry”.
As supersymmetry is the “square root” of translations, so S-supersymmetry is the
square-root of conformal boosts.
110 II. SPIN

Excercise IIC4.1
For D=4, write the (graded) commutation relations of the superconformal
generators. Decompose them into representations of the Lorentz group, and
find their commutation relations.

5. Supertwistors
We saw that a simple way to find representations of SO(4,2) was to use the
coordinate representation for SU(2,2): The resulting twistors gave all massless rep-
resentations (p2 = 0 for all helicities). This method generalizes straightforwardly to
the superconformal groups: The generators are

GA B = ζ B ζ¯A

(For the SU case we should also subtract out the trace, but that generator commutes
with the rest anyway.) The coordinates and their conjugate momenta satisfy

[ζ̄A, ζ B } = δAB

ζ A is then in the defining representation of the supergroup, while the wave function,
which is a function of ζ, contains more general representations.
¯ so the ζ’s are the graded generalization
For D=3, the reality condition sets ζ = ζ,
of Dirac γ matrices. In fact, the anticommuting ζ’s are the γ matrices of the SO(N)
subgroup of the OSp(N|4). On the other hand, the commuting ζ’s carry the index of
the defining representation of Sp(4), so they are a spinor of SO(3,2), the 3D conformal
group: They are the bosonic twistor, and can be used in a similar way to the 4D
twistors discussed earlier.
For D=4, there is a U(1) symmetry acting on ζ under which GA B is invariant,
generated by (−1)A GAA , as in the bosonic case: This is the “superhelicity”.
For D=6, ζ is pseudoreal. In general, for pseudoreal representations of groups it is
often convenient to introduce a new SU(2) under which the . pseudoreal representation
ζ A and its equivalent complex conjugate representation ζ¯B Ω . A transform as a doublet
B
(SU(2) spinor). This is also obvious from construction, since half of the components
are related to the complex conjugate of the other half. We then can write
. ..
ζ Ai = (ζ A , iζ̄ B Ω . A ) = ζ¯B k Ω . .Ai ; Ω . .Ai = Ω . A C ki , M Ai,Bk = M AB C ik
B Bk Bk B

GA B = ζ BiζAi
 C. SUPERSYMMETRY 111

(Thus, OSp*(2m|2n)⊂OSp(4n|4m), and SO*(2m)⊂Sp(4n), USp(2n)⊂SO(4m).) This

means there is now an SU(2) symmetry on ζ, generated by

Gij = ζ A(i ζAj)

under which GA B is invariant. This is the 6D version of superhelicity. In the D=6

light cone, the manifest part of Lorentz invariance is SO(D−2)=SO(4)=SU(2)⊗SU(2).
This is one of those SU(2)’s.
We now concentrate on D=4 (although our methods generalize straightforwardly
to D=3 and 6). The simplest way to find (massless) representations of 4D super-
symmetry is to generalize the Penrose transform. Just as twistors automatically
satisfy the massless field equations in D=4, supertwistors automatically satisfy their
supersymmetric generalization. The supertwistor is the defining representation of
SU(2,2|N). The SU(2,2) part is the usual twistor, while the SU(N) part is the usual
fermionic creation and annihilation operators for SU(N). Thus, to relate superspace
to supertwistors, we write
p . = −i∂ . → ±pα p̄ .
αβ αβ β
.
qiα = −i∂iα + 12 θ̄iβ ∂ . → ±aipα , q̄αi. = −i∂¯αi. + 12 θiβ ∂β α. → ±a†i p̄α.
αβ

This determines the Penrose transform from superspace to supertwistors:

Z
φ . (x, θ) = d2 pα d2 p̄α.dN ai pα · · · p̄ . · · · [eiϕχ+(pα , p̄α., ai ) + e−iϕ χ−(pα , p̄α., ai )]
α...β... β

. .
ϕ = (xαβ − i 12 θiαθ̄iβ )pα p̄ . + θiα ai pα
β

where we have used “chiral superfields” (trivial dependence on θ̄) without loss of
generality. (Instead of treating ai as coordinates to be integrated, we can also treat
them as operators; we then make the χ’s functions of a†, and replace the integration
with vacuum evaluation h0| |0i.) As for ordinary twistors, this result can be related
to the lightcone: For given momentum, we can choose the lightcone frame pα = δ⊕ α
;
⊕
then qi = ±ai, while qi = 0 is a result of the supertwistor formalism automatically
incorporating p/q = 0.
Excercise IIC5.1
Find the Penrose transform for D=3. (Warning: The anticommuting part
of the twistor is now like Dirac matrices rather than creation/annihilation
operators.)
112 II. SPIN

Taylor expanding in ai (and thus θiα , producing terms antisymmetric in i...j and
symmetric in α...β), the states then carry the index structure φ, φi , φij , ..., φ̃i, φ̃, totally
antisymmetric, and terminating with another singlet, where

1 i1 ···iN
φ̃ = N!
 φi1 ···iN , φ̃i1 = 1
(N −1)!
i1 ···iN φi2 ···iN , ...

From our discussion of helicity in subsection IIB7, we see that the states also decrease
in helicity by 1/2 for each a (i.e., ignoring θ̄, each θiα comes with a pα , simply because
it adds an undotted index). Taking the direct product with any helicity (coming from
the explicit pα ’s and p̄α. ’s carrying the external Lorentz indices), we see that the states


have helicity h, h − 1/2, h − 1, ..., h − N/2, with multiplicty Nn for helicity h − n/2:

state helicity (Poincaré) multiplicity [SU(N)]

φ h 1
φi h − 12 N
N (N −1)
φij h−1 2
.. .. ..
. . .


φi1 ···in h− n
2
N
n
.. .. ..
. . .
φ̃i h − N2 + 1
2 N
φ̃ h − N2 1

This multiplet structure is carried separately by χ+ and by χ− , which are related

by charge (complex) conjugation, one describing the antiparticles of the other, as for
ordinary twistors. (The existence of both multiplets also follows from CPT invariance,
which is required for local actions, to be discussed in subsection IVB1. Here we
generalized from the Penrose transform, which contained both terms as a consequence
of being the most general solution to S abpb + wpa = 0, which is CPT invariant.)
Because of the values of the helicities, we can impose a reality condition, identifying
all states with helicity j as the complex conjugates of those with −j, only for −h =
h−N/2 → h = N/4, when N is a multiple of 4. We can also get larger representations
by taking the direct product of these smallest representations of supersymmetry with
representations of U(N), in which case the fields will carry those additional SU(N)
indices.
Excercise IIC5.2
For D=4, show that “supergravity”, the supersymmetric theory with helicities
of magnitude 2 and lower, can exist only for N≤8. Show that the relevant
 C. SUPERSYMMETRY 113

representation for N=8, if real, is the same as the one (complex) for N=7.
Find the analogous statements for “super Yang-Mills”, with helicities 1 and
less.
The explicit form of the reality condition is somewhat complicated in terms of
the chiral superfields, because they are really field strengths of real gauge fields.
(Consider, for example, expressing reality of A . in terms of fαβ in the case of elec-
αβ
tromagnetism.) However, in terms of the twistor variables, charge conjugation can
be expressed as
C: χ± → χ∓*, ai → (ai )†

where the transformation on ai is required because it carries the SU(N) “charge”.

Since this violates “chirality” in these variables (dependence on a and not a† ), it is
accomplished by Fourier transformation:
Z
†i
C: χ±(ai ) → C dã†i eã ai [χ∓(ãi )]*C −1

for some “charge conjugation matrix” C (in case the field carries an additional index).

REFERENCES
1 P. Ramond, Phys. Rev. D3 (1971) 86;
A. Neveu and J.H. Schwarz, Nucl. Phys. B31 (1971) 86, Phys. Rev. D4 (1971) 1109:
2D supersymmetry in strings.
2 Yu.A. Gol’fand and E.P. Likhtman, JETP Lett. 13 (1971) 323;
D.V. Volkov and V.P. Akulov, Phys. Lett. 46B (1973) 109;
J. Wess and B. Zumino, Nucl. Phys. B70 (1974) 39:
4D supersymmetry.
3 A. Salam and J. Strathdee, Nucl. Phys. B76 (1974) 477;
S. Ferrara, B. Zumino, and J. Wess, Phys. Lett. 51B (1974) 239:
superspace.
4 Gates, Grisaru, Roček, and Siegel, loc. cit.;
J. Wess and J. Bagger, Supersymmetry and supergravity, 2nd ed. (Princeton University,
1992);
P. West, Introduction to supersymmetry and supergravity, 2nd ed. (World-Scientific,
1990);
I.L. Buchbinder and S.M. Kuzenko, Ideas and methods of supersymmetry and super-
gravity, or a walk through superspace (Institute of Physics, 1995).
5 Berezin, loc. cit. (IA):
superdeterminant.
6 P. Ramond, Physica 15D (1985) 25:
supergroups and their relation to supersymmetry.
7 J. Wess and B. Zumino, Nucl. Phys. B70 (1974) 39:
superconformal group in D=4.
8 R. Haag, J.T. Lopuszański, and M. Sohnius, Nucl. Phys. B88 (1975) 257:
superconformal symmetry as the largest symmetry of the S-matrix.
114 II. SPIN

9 A. Ferber, Nucl. Phys. B132 (1978) 55:

supertwistors.
10 T. Kugo and P. Townsend, Nucl. Phys. B221 (1983) 357;
A. Sudbery, J. Phys. A17 (1984) 939;
K.-W. Chung and A. Sudbery, Phys. Lett. 198B (1987) 161:
spacetime symmetries and division algebras.
 A. ACTIONS 115

III. LOCAL
In the previous chapters we considered symmetries acting on coordinates or wave
functions. For the most part, the transformations we considered had constant param-
eters: They were “global” transformations. In this chapter we will consider mostly
field theory. Since fields are functions of spacetime, it will be natural to consider
transformations whose parameters are also functions of spacetime, especially those
that are localized in some small region. Such “local” or “gauge” transformations are
fundamental in defining the theories that describe the fundamental interactions.

............................ ............................
............................ A. ACTIONS ............................

A fundamental concept in physics, of as great importance as symmetry, is the

action principle. In quantum physics the dynamics is necessarily formulated in terms
of an action (in the path-integral approach), or an equivalent Hamiltonian (in the
Heisenberg and Schrödinger approaches). Action principles are also convenient and
powerful for classical physics, allowing all field equations to be derived from a single
function, and making symmetries simpler to check.

1. General
We begin with some general properties of actions. (For this subsection we’ll re-
strict ourselves to bosonic variables; however, in the following subsection we’ll find
that the only modification for fermions is a more careful treatment of signs.) Gen-
erally, equations of motion are derived from actions by setting their variation with
respect to their arguments to vanish:

δS[φ] ≡ S[φ + δφ] − S[φ] = 0

Here the variables φ are themselves functions of time; thus, S is a function of functions,
a “functional”. A general principle of mechanics is “locality”, that events at one time
directly affect only those events an infinitesimal time away. (In field theory these
events can be also only an infinitesimal distance away in space.) This means that the
action can be expressed in terms of a Lagrangian:
Z
S[φ] = dt L[φ(t)]

where L at time t is a function of only φ(t) and a finite number of its derivatives. For
more subtle reasons, this number of time derivatives is restricted to be no more than
116 III. LOCAL

two for any term in L; after integration by parts, each derivative acts on a different
factor of φ. The general form of the action is then
. . .
L(φ) = − 12 φm φn gmn (φ) + φm Am (φ) + U(φ)
.
where “ ” means ∂/∂t, and the “metric” g, “vector potential” A, and “scalar po-
tential” U are not to be varied independently when deriving the equations of motion.
(Specifically, δU = (δφm)(∂U/∂φm ), etc. Note that our definition of the Lagrangian
differs in sign from the usual.) The equations of motion following from varying an
action that can be written in terms of a Lagrangian are
Z
δS δS
0 = δS ≡ dt δφm m ⇒ =0
δφ δφm
.
where we have eliminated δ φm terms by integration by parts (assuming δφ = 0 at the
boundaries in t), thereby defining the “functional derivative” δS/δφm , and used the
fact that δφ(t) is arbitrary at each value of t. For example,
Z Z Z
1 .2 . . .. δS ..
S = − dt 2 q ⇒ 0 = δS = − dt qδ q = dt (δq) q ⇒ =q=0
δq

Excercise IIIA1.1
Find the equations of motion for φm from the above general action in terms
of the external fields g, A, and U (and their partial derivatives with respect
to φ).
Sometimes the functional derivative is defined in terms of that of the variable
itself:
δφm(t)
= δnm δ(t − t0)
δφn(t0)
where δnm is the usual Kronecker delta function, while δ(t − t0) is the “Dirac delta
function”. It’s not really a function, since it takes only the values 0 or ∞, but a
“distribution”, meaning it’s defined only by integration:
Z
dt0 f(t0 )δ(t − t0) = f(t)

If we apply this definition of the Dirac δ to δφ/δφ, we obtain the previous definition
R
of the functional derivative. (Consider, e.g., S = dt fφ.)
Such actions can be reduced to ones that are only linear in time derivatives
by introducing additional variables. First, separate out the subspace where g is
invertible, with coordinates q (φm = (q i, ψ µ )); the Lagrangian is then written as
. . . .
L(q, ψ) = − 12 q i qj gij (q, ψ) + qi Ai(q, ψ) + ψµ Aµ (q, ψ) + U(q, ψ)
 A. ACTIONS 117

This Lagrangian gives equivalent equations of motion to

. .
L0 (q, p, ψ) = [−qi pi + ψ µ Aµ ] + [ 12 g ij (pi + Ai)(pj + Aj ) + U]

where g ij is the inverse of gij . (Many other forms are possible by redefinitions of
p.) Eliminating the new variables p by their equations of motion gives back L(q, ψ).
Note that this works only because p’s equations of motion are algebraic: For example,
. .
eliminating x from the Lagrangian −xp+ 12 p2 by the equation of motion x = p is illegal
R
(it would give the trivial action S = dt 12 p2 ), since it would require solving for the
time dependence of x. On the other hand, p is given explicitly in terms of the other
variables by its equations of motion without inverting time derivatives, so eliminating
it does not lose any of the dynamics. (It is an “auxiliary variable”.)
The result is a Hamiltonian form of the Lagrangian:
.
LH (Φ) = iΦM AM (Φ) + H(Φ)

in terms of the Hamiltonian H, where Φ = (q, p, ψ). It has the “gauge invariance”

δAM = ∂M Λ(Φ)
.
(where ∂M = ∂/∂ΦM ), since that adds only a total derivative term iΛ. Clearly A
will introduce a modification of the Poisson bracket if it is not linear in Φ (e.g., as
when we make independent nonlinear redefinitions of coordinates and momenta on
the usual form of the Lagrangian). To determine this modification we compare the
equation of motion as defined by a Poisson bracket,
.
ΦM = −i[ΦM , H] = −i[ΦM , ΦN ]∂N H

with that following from varying the action,

.
−iΦN FN M + ∂M H = 0, FM N = ∂[M AN ]

to find
[ΦM , ΦN ] = (F −1)N M

where “F −1 ” is the inverse on the maximal subspace where F is invertible. The

variables in the directions where F vanishes are “auxiliary”, since they appear without
time derivatives: Their equations of motion are not described by the Poisson bracket.
In particular, if they appear linearly in H they are “Lagrange multipliers”, whose
variation imposes algebraic constraints on the rest of Φ.
118 III. LOCAL

Finally, we can make redefinitions of the part of Φ describing the invertible sub-
space so that A is linear:
.
AM = 12 ΦN ΩN M ⇒ LH (Φ) = 12 iΦM ΦN ΩN M + H(Φ)

where Ω is a constant, hermitian, antisymmetric (and thus imaginary) matrix. For

some purposes it is more convenient to assume this Hamiltonian form of the action
as a starting point. We now have the canonical commutation relations as

[ΦM , ΦN ] = ΩM N

where ΩM N is the inverse of ΩN M on the maximal subspace:

ΩM N ΩP N = ΠP M

for the projection operator Π for that subspace.

Excercise IIIA1.2
~ =E
For electromagnetism, define ψ ~ + iB.
~ Show that Maxwell’s equations
~ Interpret
(in empty space) can be written as two equations in terms of ψ.
the equation involving the time derivative as a Schrödinger equation for the
~ and find the Hamiltonian operator. Define the obvious inner
wave function ψ,
R 3
product d x ψ* ~ · ψ:
~ What physical conserved quantity does this represent?
(Note that, unlike electrons, the number of photons is not conserved.)
Note that the requirement of the existence of a Hamiltonian formulation deter-
.
mines that the kinetic term for a particle in the Lagrangian formulation go as x2
..
and not xx. Although such terms give the same equations of motion, they are not
equivalent quantum mechanically, where boundary terms (dropped when using inte-
gration by parts for deriving the equations of motion) contribute. Furthermore, the
Hamiltonian form of the action
Z
S= dt H − dxi pi

shows that the energy H relates to the time in the same way the momentum relates to
the coordinates, except for an interesting minus sign that is explained only by special
relativity.
 A. ACTIONS 119

2. Fermions
In nonrelativistic quantum mechanics, spin is usually treated as a quantum effect,
rather than being derived from classical mechanics. Although it is possible to derive
spin from classical mechanics, in general it is rather cumbersome, and involves first
introducing a large number of spins and then constraining away all the undesired ones,
whereas in the quantum mechanics one can just directly introduce some particular
representation of the spin angular momentum operators. The one nontrivial exception
is spin 1/2.
We know from quantum mechanics that the spin variables for spin 1/2 are de-
scribed by the Pauli σ matrices. Since they satisfy anticommutation relations, and
are represented by finite-dimensional matrices, they are interpreted as fermionic. We
have already seen that classical fermions are described by anticommuting numbers,
so we begin by considering general quantization of such objects.
We can now consider actions that depend on both commuting and anticommuting
classical variables, ΦM = (φm , ψ µ ), where now φ refers to the bosonic variables and ψ
to the fermionic ones. The Hamiltonian form of the Lagrangian can again be written
as
.
LH (Φ) = 12 iΦM ΦN ΩN M + H(Φ)
When Ω is invertible, the graded bracket is defined by (see subsection IA2)

[ΦM , ΦN } = h̄ΩM N , ΩM N ΩP N = δPM

To describe spin 1/2, we therefore look for particle actions of the form
Z .
.
SH = dt[−xi pi + 12 iψi ψi + H(x, p, ψ)]

This corresponds to using

ΦM = (φm ; ψ µ ) = (φiα ; ψ i) = (xi , pi ; ψ i)

Ωmn = Ωiα,jβ = δij Cαβ , Ωµν = Ωij = δij , Ωmν = Ωµn = 0

The fundamental commutation relations are then

[xi, pj ] = ih̄δji , {ψ i , ψ j } = h̄δ ij ([x, x] = [p, p] = [x, ψ] = [p, ψ] = 0)

We recognize ψ i as the Pauli σ matrices (the Dirac matrices of subsection IC1 for the
√
special case of SO(3)), ψ i = h̄σ i . The free Hamiltonian is just
p2
H=
2m
120 III. LOCAL

as for spin 0: Spin does not affect the motion of free particles.
A more interesting case is coupling to electromagnetism: Quantum mechanically,
the Hamiltonian can be written in the simple form

{ψ i[pi + qAi(x)]}2
H= − qA0(x)
mh̄
in terms of the vector and scalar potentials Ai and A0. The classical expression is not
as simple, because the commutation relations must be used to cancel the 1/h̄ before
taking the classical limit. This is an example of “minimal coupling”,

H(pi ) → H(pi + qAi) − qA0

However, this prescription works only if H for spin 1/2 is written in the above form:
Using the commutation relations before or after minimal coupling gives different re-
sults. The form we have used is justified only by considering the nonrelativistic limit
of the relativistic theory.
Excercise IIIA2.1
Use the multipication rules of the σ matrices to show that the quantum me-
chanical Hamiltonian for spin 1/2 in an electromagnetic field can be written
as a spin-independent piece, identical to the spin-0 Hamiltonian, plus a term
coupling the spin to the magnetic field.

3. Fields
The field equations for all field theories (e.g., electromagnetism) are wave equa-
tions. Wave equations also follow from mechanics upon quantization. Although
classical field theory and quantum mechanics are not equivalent in their physical in-
terpretation, they are mathematically equivalent in that they have identical wave
equations. This is true not only for the free theories, but also for particles in external
fields, and without direct self-interactions. This is no accident: Classical field theory
and classical mechanics are two different limits of quantum field theory. They are
both called classical limits, and written as h̄ → 0, but since h̄ is really 1, this limit
depends on how one inserts h̄’s into the quantum field theory action.
The wave equation in quantum mechanics is the Schrödinger equation. The cor-
responding field theory action is then simply the one that gives this wave equation
as the equation of motion, where the wave function is replaced with the field:
Z
Sf t = d4 x ψ*(−i∂t + H)ψ
 A. ACTIONS 121

As usual (cf. electromagnetism), the field is a function of space and time; thus, we
integrate d4 x = dt d3 x over the three space and one time dimensions. The Hamilto-
nian is some function of coordinates and momenta, with the replacement pi → −i∂i,
where ∂i = ∂/∂xi are the space derivatives and ∂t = ∂/∂t is the time derivative.
The Hamiltonian can contain coupling to other fields. For a general Hamilto-
nian quadratic in momenta, in a notation implied by the corresponding Lagrangian
quadratic in time derivatives,

H = 12 g ij (−i∂i + Ai )(−i∂j + Aj ) + U

where g ij , Ai , and U are now interpreted as fields, and thus depend on both xi and t,
as does ψ. In the case g ij = δ ij , we can identify Ai and U as the three-vector and scalar
potentials of electromagnetism, and we can add the usual action for electromagnetism
to the action for ψ. The action then can be varied also with respect to A and U to
obtain Maxwell’s equations with a current in terms of ψ and ψ*. We can also treat
g ij as a field, in which case it and parts of A and U are the components of the
gravitational field.
Field theory actions can be quantized in the same ways as mechanics ones. In
. .
this case, we recognize the ψ*ψ term as a special case of the ΦΦΩ term in the
generic Hamiltonian form of the action discussed earlier. Thus, ψ(xi) and ψ*(xi )
have replaced xi and pi as the variables; xi is now just an index (label) on ψ and ψ*,
just as i was an index on xi and pi . The field-theory Hamiltonian is then identified
as Z
Hf t [ψ, ψ*] = d3 x H, H = ψ*Hψ

In field theory the Hamiltonian will always be a space integral of a “Hamiltonian

density” H.
We can now define the two classical limits of quantum field theory. If we put in
h̄ in the generic way for actions,

Sf t → h̄−1 Sf t

then we define the classical limit h̄ → 0 as classical field theory, since in that limit
the classical field equations are preserved. On the other hand, if we put in h̄’s as

∂i → h̄∂i , ∂t → h̄∂t

which gives the usual h̄ dependence associated with the Schrödinger equation, then
the classical limit h̄ → 0 gives classical mechanics. This defines classical mechanics
as the macroscopic limit, the limit of large distances and times.
122 III. LOCAL

A convenient way to implement this limit is to introduce the mechanics action

R .
S = dt(−xi pi + H) into the field theory, and then take the limit h̄ → 0 after the
replacement
S → h̄−1 S
on the mechanics action instead of on the derivatives. The mechanics action can be
introduced when solving the field equations: The solution to the wave equation can
be expressed in terms of the propagator, which in turn can be written in terms of the
mechanics action or Hamiltonian.
More generally, we can define actions that are not restricted to be quadratic in
any field. The Hamiltonian density H(t, xi ) or Lagrangian density L(t, xi ),
Z
S[φ] = dt d3 x L[φ(t, xi)]

should be a function of fields at that point, with only a finite number (usually no
more than two) spacetime derivatives. This is the definition of locality used for gen-
eral quantum systems in subsection IIIA1, but extended from derivatives in time to
also those in space. Although this condition is not always used in nonrelativistic
field theory (for example, when long-range interactions, such as Coulomb or gravita-
tional, are described without attributing them to fields), it is crucial in relativistic
field theory. For example, global symmetries lead by locality to local (current) conser-
vation laws. Locality is also the reason that spacetime coordinates are so important:
Translation invariance says that the position of the origin is an unphysical, redundant
variable; however, locality is most easily used with this redundancy.
Field equations are derived by the straightforward generalization of the variation
of actions defined in subsection IIIA1: As follows from treating the spatial coordinates
in the same way as discrete indices,
Z
δS
δS ≡ dt d3 x δφm(t, xi ) m
δφ (t, xi)
For example, Z . ..
δS
S=− dt d3 x 12 φ2 ⇒ =φ
δφ
4. Relativity
Generalization to relativistic theories is straightforward, except for the fact that
the Klein-Gordon equation is second-order in time derivatives; however, we are fa-
miliar with such actions from nonrelativistic quantum mechanics. As usual, we need
to check the sign of the terms in the action: Checking the positivity of the Hamil-
tonian (i.e., the energy), we see from the general relation between the Lagrangian
 A. ACTIONS 123

and Hamiltonian (subsection IIIA1) that the terms without time derivatives must be
positive; the time-derivative terms are then determined by Lorentz covariance.
At this point we introduce some normalizations and conventions that will prove
convenient for Fourier transformation and other reasons to be explained later. When-
ever D-dimensional integrations are involved (as should be clear from context), we
use Z Z Z Z
dD x dD p
dx ≡ , dp ≡
(2π)D/2 (2π)D/2
δ(x − x0) ≡ (2π)D/2δ D (x − x0 ), δ(p − p0 ) ≡ (2π)D/2 δ D (p − p0 )
In particular, this normalization will be used in Green functions and actions. For
example, these implicit 2π’s appear in functional variations:
Z
δS δ
δS ≡ dx δφ ⇒ φ(x0) = δ(x − x0 )
δφ δφ(x)

The action for a real scalar is then

Z
S = dx L, L = 14 (∂φ)2 + V (φ)

where V (φ) ≥ 0, and we now write L for the Lagrange density. In particular, V =
m φ for the free theory. The free field equation is then p2 + m2 = − + m2 = 0,
1 2 2
4
replacing the nonrelativistic −i∂t+H = 0. For a complex scalar, we replace 12 φφ → χ̄χ
in both terms.
We know from previous considerations (subsection IIB2) that the field equation
for a free, massless, Dirac spinor is γ · ∂Ψ = 0. The generalization to the massive case
(subsection IIB4) is obvious from various considerations, e.g., dimensional analysis;
the action is Z
S = dx Ψ̄(i∂/ + √m2 )Ψ

in arbitrary dimensions, again using the notation ∂/ = γ · ∂. In four dimensions, we

can decompose the Dirac spinor into its two Weyl spinors (see subsection IIA6):
. . .
L = Ψ̄(i∂/ + √m2 )Ψ = (ψ̄Lα i∂ αα.ψLα + ψ̄Rα i∂ αα.ψRα) + √m2 (ψLαψRα + ψ̄Lαψ̄Rα.)

For the case of the Majorana spinor, the 4D action reduces to that for a single Weyl
spinor, Z . .
S = dx [−iψ̄ β ∂ .ψ α + √m2 12 (ψ αψα + ψ̄ αψ̄α.)]
αβ

Note that in our conventions σ 0 . = √1 δαβ (and similarly for the opposite indices, since
αβ 2
.
σ a . = σaβ α), so that the time derivative term is always proportional to ψ †(−i∂0)ψ, as
αβ
nonrelativistically (previous subsection).
124 III. LOCAL

A scalar field must be complex to be charged (i.e., a representation of U(1)):

From the gauge transformation
χ0 = eiλχ

we find the minimal coupling (for q = 1)

Z
Sχ = dx [ 12 |(∂ + iA)χ|2 + 12 m2|χ|2]

This action is also invariant under charge conjugation

C : χ → χ*, A → −A

which changes the sign of the charge, since χ*0 = e−iλ χ*.
Excercise IIIA4.1
Let’s consider the semiclassical interpretation of a charged particle as de-
scribed by a complex scalar field ψ, with Lagrangian

L = 12 (|∇ψ|2 + m2|ψ|2)

a Use the semiclassical expansion in h̄ defined by

√ −iS/h̄
∇ → h̄∂ + iqA, ψ→ ρe

Find the Lagrangian in terms of ρ and S (and the background field A), order-
by-order in h̄ (in this case, just h̄0 and h̄2 ).
b Take the semiclassical limit by dropping the h̄2 term in L, to find

L → ρ 12 [(−∂S + qA)2 + m2]

Vary with respect to S and ρ to find the equations of motion. Defining

p ≡ −∂S

show that these field equations can be interpreted as the mass-shell condition
and current conservation. Show that A couples to this current by varying L
with respect to A.
The spinor field also needs doubling for charge. (Actually, the doubling can be
avoided in the massless case; however, problems show up at the quantum level, related
to the fact that there is no charge conjugation transformation without doubling.) The
gauge transformations are similar to the scalar case, and the action again follows from
 A. ACTIONS 125

minimal coupling, to an action that has the global invariance (λ = constant in the
absence of A):
ψL0α = eiλ ψLα, ψR0α = e−iλ ψRα
Z . . .
Se = dx [ψ̄L(−i∂ . + A .)ψL + ψ̄R(−i∂ . − A .)ψRα + √m2 (ψLαψRα + ψ̄Lαψ̄Rα.)]
β α β
αβ αβ αβ αβ

The current is found from varying with respect to A:

. . .
J = ψ̄LψL − ψ̄RψR
αβ β α β α

Charge conjugation
C : ψLα ↔ ψRα , A → −A

(which commutes with Poincaré transformations) changes the sign of the charge and
current.
Excercise IIIA4.2
Show that this action can be rewritten in Dirac notation as
Z
Se = dx Ψ̄(i∂/ − A / + √m2 )Ψ

and find the action of the gauge transformation and charge conjugation on
the Dirac spinor.
As a last example, we consider the action for electromagnetism itself. As before,
we have the gauge invariance and field strength

A0 . = A . − ∂ .λ
αβ αβ αβ

.
F . . = ∂αγ.A . − ∂ .Aαγ. = Cαβ f¯. . + C̄ . .fαβ , fαβ = 12 ∂(αγ.Aβ)γ
αγ,β δ βδ βδ γδ γδ

We can write the action for pure electromagnetism as

Z Z .. Z
αβ 1 ¯αβ ¯ .
1
SA = dx 2e2 f fαβ = dx 2e2 f f . = dx 1
8e2
F abFab
αβ

dropping boundary terms, with the overall sign again determined by positivity of the
Hamiltonian, where e is the electromagnetic coupling constant, i.e., the charge of the
proton. (Other normalizations can be used by rescaling A ..) Maxwell’s equations
αβ
follow from varying the action with a source term added:
Z .
S = SA + dx Aαβ J . ⇒ e12 ∂ β γ.fβα = Jαγ.
αβ

Excercise IIIA4.3
By plugging in the appropriate expressions in terms of Aa (and repeatedly
126 III. LOCAL

integrating by parts), show that all of the above expressions for the electro-
magnetism action can be written as
Z
SA = − dx 4e12 [A · A + (∂ · A)2 ]

Excercise IIIA4.4
Find all the field equations for all the fields, found from adding to SA all the
minimally coupled matter actions above.
The energy-momentum tensor for electromagnetism is much simpler in this spinor
notation, and follows (up to normalization) from gauge invariance, dimensional anal-
ysis, Lorentz invariance, and the vanishing of its trace. It has a form similar to that
of the current in electrodynamics:

T . . = − e12 fαβ f¯γ.δ.

αβ γδ

Note that it is invariant under the duality transformations of subsection IIA7 (as is
the electrodynamic current under chirality).
We have used conventions where e appears multiplying only the action SA, and
not in the “covariant derivative”

∇ = ∂ + iqA

where q is the charge in units of e: e.g., q = 1 for the proton, q = −1 for the electron.
Alternatively, we can scale A, as a field redefinition, to produce the opposite situation:
Z
A → eA : SA → dx 18 F 2, ∇ → ∂ + iqeA

The former form, which we use unless noted otherwise, has the advantage that the
coupling appears only in the one term SA , while the latter has the advantage that the
kinetic (free) term for A is normalized the same way as for scalars. The former form
has the further advantage that e appears in the gauge transformations of none of the
fields, making it clear that the group theory does not depend on the value of e. (This
will be more important when generalizing to nonabelian groups in section IIIC.)
Note that the massless part of the kinetic (free) terms in these actions are scale
invariant (in arbitrary dimensions, when the dimension-independent forms are used),
when the fields are assigned the scale weights found from conformal arguments in
subsection IIB2.
Excercise IIIA4.5
Using vector notation, minimal coupling, and dimensional analysis, find the
 A. ACTIONS 127

mass dimensions of the electric charge e in arbitrary spacetime dimensions,

and show it is dimensionless only in D = 4.
An interesting distinction between gravity and electromagnetism is that static
bodies always attract gravitationally, whereas electrically they repel if they are like
and attract if they are opposite. This is a direct consequence of the fact that the
graviton has spin 2 while the photon has spin 1: The Lagrangian for a field of integer
spin s coupled to a current, in an appropriate gauge and the weak-field approximation,
is
− 4s!
1 a1 ...as
φ φa1 ...as + s!1 φa1 ...as J a1...as
where the sign of the first term is fixed by unitarity in quantum field theory. (Clas-
sically the sign can also be related to positivity of the energy.) From a scalar field in
the semiclassical approximation (see excercise IIIA4.1 above), starting with
↔ ↔
J a1 ...as ∼ ψ* ∂ a1 · · · ∂ as ψ
↔
where “A ∂ B” means “A∂B − (∂A)B”, we see that the current will be of the form
J a1 ...as ∼ ρpa1 · · · pas
for a scalar particle, for some ρ. (The same follows from comparing the expressions
for currents and energy-momentum tensors for particles as in subsection IIIB4 below.
The only way to get vector indices out of a scalar particle, to couple to the vector
indices for the spin of the force field, is from momentum.) In the static approximation,
only time components contribute: We then can write this Lagrangian as, taking into
account η00 = −1,
−(−1)s 4s!1
φ0...0 φ0...0 + s!1 φ0...0ρ(p0 )s
where E = p0 > 0 for a particle and < 0 for an antiparticle. Thus the spin-dependence
of the potential/force between two particles goes as (−E1E2 )s . It then follows that
all particles attract by forces mediated by even-spin particles, and a particle and
its antiparticle attract under all forces, while repulsion will occur for odd-spin forces
between two identical particles. (We can substitute “particles of the same sign charge”
for “identical particles”, and “particles of opposite sign charge” for “particle and its
antiparticle”, where the charge is the coupling constant appropriate for that force.)
Excercise IIIA4.6
Show that the above current is conserved,
∂a1 J a1···as = 0
(and the same for the other indices, by symmetry) if ψ satisfies the free Klein-
Gordon equation (massless or massive).
128 III. LOCAL

5. Constrained systems

Constraints not only frequently appear in nonrelativistic physics, but are a general
feature of relativistic particles, so we now give a brief description of how they are
incorporated into actions. Consider a general action, with constraints, in Hamiltonian
form: Z
.
S = dt(−qm pm + H), H = Hgi (q, p) + λi Gi (q, p)

(For simplicity, we consider all physical variables to be bosonic for this subsection,
but the method generalizes straightforwardly paying careful attention to signs.) This
action is a functional of q m , pm , λi , which are in turn functions of t, where m and i run
over any number of values. We can think of this as describing a nonrelativistic particle
with coordinates q and momenta p in terms of time t, but the form is general enough to
.
apply to relativistic theories. The qp term tells us p is canonically conjugate to q; the
rest of the action gives the Hamiltonian, usually quadratic in momenta. The variables
λi are “Lagrange multipliers”, whose variation in the action implies the constraints
Gi = 0. We then can interpret Hgi as the usual (“gauge invariant”) Hamiltonian. We
also require that the transformations generated by the constraints close, and that the
Hamiltonian be invariant:

[Gi , Gj ] = −ifij k Gk , [Gi , Hgi ] = 0

(More generally, we can allow [Gi , Hgi ] = −ifi j Gj .) This says that the constraints
don’t imply any new constraints that we might have missed, and that the “energy”
represented by Hgi is invariant under these transformations.
We then find that the action is invariant under the canonical transformations
∂Gi ∂Gi
δ(q, p) = i[ζ iGi , (q, p)] ⇒ δq m = ζ i , δpm = −ζ i
∂pm ∂q m
   
d ∂ .
0=δ =δ + iH = i(δλi )Gi − iζ i Gi + [λj Gj , ζ i Gi ]
dt ∂t
.
⇒ δλi = ζ i + ζ j λk fkj i

(with δ(d/dt) defined as in subsection IA1), where ∂/∂t acts on the “explicit” t
dependence (that in everything except q and p): For general expressions, the total
time derivative and total variation are given by commutators as

d ∂
A = A + i[H, A], δA = δ0A + i[ζ iGi , A]
dt ∂t
 A. ACTIONS 129

where δ0 acts on everything except q and p. The action then varies under these
transformations as the integral of a total derivative, which vanishes under appropriate
boundary conditions:
Z
d
δSH = dt [−(δq m)pm + ζ i Gi ] = 0
dt
The simplest example is the case with one constraint, which is linear in the vari-
ables: If the constraint is p, the gauge transformation is δq = λ, so we gauge q = 0
and use the constraint p = 0. In general, this means that for every degree of freedom
we can gauge away, the conjugate variable can be fixed by the constraint. Thus,
for each constraint we eliminate 3 variables: the variable fixed by the constraint, its
conjugate, and the Lagrange multiplier that enforced the constraint, which has no
conjugate. (In the Lagrangian form of the action the conjugate may not appear ex-
plicitly, so only 2 variables are eliminated.) As another example, for a nonrelativistic
particle constrained to a sphere, G = (xi )2 −1, we can change to spherical coordinates,
apply the constraint to eliminate the radial coordinate, and use the gauge invariance
to eliminate the radial component of the momentum, leaving an unconstrained the-
ory in terms of angles and their conjugates. In most cases in field theory a similar
procedure can be applied: The result is called a “unitary gauge”.
The standard example of a relativistic constrained system is in field theory —
electromagnetism. Its action can be written in “first-order (in derivatives) formalism”
by introducing an auxiliary field Gab :

F 2 → F 2 − G2 → F 2 − (G − F )2 = 2GF − G2

where in the first step we added a trivial term for G and in the second step made
a trivial redefinition of G, so elimination of G by its algebraic equation of motion
returns the original Lagrangian. The Hamiltonian form comes from eliminating only
Gij by its field equation, since only F0i contains time derivatives:

2GF − G2 → (Fij )2 − 4G0i F0i + 2(G0i )2

.
= −4Ai G0i + [2(G0i )2 + (Fij )2 ] − 4A0 ∂i G0i
which we recognize as the three generic terms for the action in Hamiltonian form,
with G0i as the canonical momenta for Ai , and A0 as the Lagrange multiplier. The
constraint is Gauss’ law, and it generates the usual gauge transformations.
Thus λi are also gauge fields for the gauge (time-dependent) transformations ζ i (t).
They allow construction of the gauge-covariant time derivative
d
∇ = ∂t + iλi Gi , = ∇ + iHgi ⇒ δ∇ = i[ζ i Gi , ∇]
dt
130 III. LOCAL

It is convenient to transform the gauge fields away using these gauge transformations,
R∞
so H = Hgi . However, with the usual boundary conditions −∞ dt λi is gauge invariant
under the linearized transformations, so the most we could expect is to gauge λi to
constants. More precisely, the group element
  Z ∞ 
T exp −i i
dt λ (t)Gi
−∞

is gauge invariant, where “T ” is time ordering, meaning we write the exponential of

the integral as the product of exponentials of infinitesimal integrals, and order them
with respect to time, later time intervals going to the left of earlier ones. (We treat
Gi quantum mechanically or use Poisson brackets when combining the exponentials.)
This is the quantum mechanical version of the time development resulting from the
corresponding term in the classical action. It is also the phase factor coming from
the infinite limit of the covariant time translation
  Z t 
−k∇(t)
e = T exp −i dt λ (t )Gi e−k∂t
0 i 0
t−k

as seen from reordering the time derivatives when writing e−k∇(t) as the product of
exponentials of infinitesimal exponents. This allows us to write the explicit gauge
transformation
  Z t 
−iΛ(t) 0 i 0
e = T exp −i dt λ (t )Gi = e−(∆t)∇(t)e(∆t)∂t , ∆t = t − t0
t0

⇒ ∇0(t) = eiΛ(t)∇(t)e−iΛ(t) = e−(∆t)∂t ∇(t)e(∆t)∂t = ∇(t0) = ∂t + iλi (t0)Gi

(where we define ∂t to vary t while keeping t − t0 fixed). Thus, we can gauge λ to

its value at a fixed time t0. Another way to see this is that varying λi in the action
at a fixed time gives Gi = 0 at that time, but the remaining field equations imply
.
Gi = 0, so Gi = 0 always, and λi is redundant at other times. This means that if we
carelessly impose λi = 0 at all times, we must also impose Gi = 0 at some fixed time.
Note that this special gauge transformation itself has a very simple gauge trans-
formation: Transforming the λ in Λ by an arbitrary finite transformation ζ i (t),
0
e−iΛ (t) = e−iζ e−iΛ(t) eiζ
i (t)G i (t
i 0 )Gi

consistent with the transformation law of ∇0(t) above. Thus, applying the trans-
formation Λ to any gauge-dependent quantity φ gives a gauge-independent quantity
φ0 (φ, λ), which is invariant under the local transformations ζ(t) and transforms only
under the “global” transformations ζ(t0 ). Thus, fixing the gauge λ(t) = 0 is equivalent
to working with gauge-invariant quantities.
 A. ACTIONS 131

Fixing an invariance of the action is not unique to gauge invariances: Global

invariances also need to be fixed, although the procedure is so trivial we seldom
discuss it. For example, even in nonrelativistic systems Galilean invariance needs to
be fixed: When analyzing a specific problem, we often choose some object to be at
rest (velocity transformations), choose another to be oriented or moving in a specific
direction (rotations), and choose a specific event to happen at the origin of space and
time (translations). Alternatively, we can work with Galilean invariants, just as in
gauge theories we can work with gauge invariants; however, in practice, for explicit
calculations (as opposed to discussing general properties), it is more convenient to fix
the invariance, as this allows simplification of the equations.

REFERENCES
1 M.B. Halpern and W. Siegel, Phys. Rev. D16 (1977) 2486:
Classical mechanics as a limit of quantum field theory.
2 P.A.M. Dirac, Proc. Roy. Soc. A246 (1958) 326;
L.D. Faddeev, Theo. Math. Phys. 1 (1969) 1:
Hamiltonian formalism for constrained systems.
132 III. LOCAL

........................... ...........................
........................... B. PARTICLES ...........................

The simplest relativistic actions are those for the mechanics (as opposed to field
theory) of particles. These also give the simplest examples of gauge invariance in rela-
tivistic theories. Later we will find that various properties of the quantum mechanics
of these actions help to explain some features of quantum field theory.

1. Free
For nonrelativistic mechanics, the fact that the energy is expressed as a function of
the three-momentum is conjugate to the fact that the spatial coordinates are expressed
as functions of the time coordinate. In the relativistic generalization, all the spacetime
coordinates are expressed as functions of a parameter τ : All the points that a particle
occupies in spacetime form a curve, or “worldline”, and we can parametrize this curve
in an arbitrary way. Such parameters generally can be useful to describe curves: A
circle is better described by x(θ), y(θ) than y(x) (avoiding ambiguities in square roots),
and a cycloid can be described explicitly only this way.
The action for a free, spinless particle then can be written in relativistic Hamil-
tonian form as Z
.
SH = dτ [−xm pm + v 21 (p2 + m2)]

where v is a Lagrange multiplier enforcing the constraint p2 + m2 = 0. This ac-

tion is very similar to nonrelativistic ones, but instead of xi (t), pi (t) we now have
xm (τ ), pm (τ ), v(τ ) (where “ . ” now means d/dτ ). The gauge invariance generated by
p2 + m2 is
.
δx = ζp, δp = 0, δv = ζ

A more recognizable form of this invariance can be obtained by noting that any
action S(φA ) has invariances of the form
δS
δφA = AB , AB = −BA
δφB
which have no physical significance, since they vanish by the equations of motion. In
this case we can add
. .
δx = (x − vp), δp = p, δv = 0

and set ζ = v to get

. . .
δx = x, δp = p, δv = (v)
 B. PARTICLES 133

We then can recognize this as a (infinitesimal) coordinate transformation for τ :

x0(τ 0) = x(τ ), p0 (τ 0) = p(τ ), dτ 0 v 0(τ 0) = dτ v(τ ); τ 0 = τ − (τ )

The transformation laws for x and p identify them as “scalars” with respect to these
“one-dimensional” (worldline) coordinate transformations (but they are vectors with
respect to D-dimensional spacetime). On the other hand, v transforms as a “density”:
The “volume element” dτ v of the world line transforms as a scalar. This gives us
a way to measure length on the worldline in a way independent of the choice of τ
parametrization. Because of this geometric interpretation, we are led to constrain

v>0

so that any segment of the worldline will have positive length. Because of this re-
striction, v is not a Lagrange multiplier in the usual sense. This has significant
physical consequences: p2 + m2 is treated neither as a constraint nor as the Hamil-
tonian. While in nonrelativistic theories the Schrödinger equation is (E − H)ψ = 0
and Gi ψ = 0 is imposed on the initial states, in relativistic theories (p2 + m2)ψ = 0
is the Schrödinger equation: This is more like Hψ = 0, since p2 already contains the
necessary E dependence.
The Lagrangian form of the free particle action follows from eliminating p by its
.
equation of motion vp = x:
Z
.
SL = dτ 12 (vm2 − v −1 x2 )
.
For m 6= 0, we can also eliminate v by its equation of motion v −2x2 + m2 = 0:
Z p . Z √ Z
S = m dτ −x = m 2 −dx = m ds = ms
2

The action then has the purely geometrical interpretation as the proper time; how-
ever, this last form of the action is awkward to use because of the square root, and
doesn’t apply to the massless case. Note that the v equation implies ds = m(dτ v),
relating the “intrinsic” length of the worldline (as measured with the worldline vol-
ume element) to its “extrinsic” length (as measured by the spacetime metric). As a
consequence, in the massive case we also have the usual relation between momentum
and “velocity”
dxm
pm = m
ds
(Note that p0 is the energy, not p0 .)
Excercise IIIB1.1
Take the nonrelativistic limit of the Poincaré algebra:
134 III. LOCAL

a Insert the speed of light c in appropriate places for the structure constants of
the Poincaré group (guided by dimensional analysis) and take the limit c → 0
to find the algebra of the Galilean group.
b Do the same for the representation of the Poincaré group generators in terms
of coordinates and momenta. In particular, take the limit of the Lorentz
boosts to find the Galilean boosts.
c Take the nonrelativistic limit of the spinless particle action, in the form ms.
(Note that, while the relativistic action is positive, the nonrelativistic one is
negative.)
Excercise IIIB1.2
Consider the following action for a particle with additional fermionic variables
γ and additional fermionic constraint γ · p:
Z
. .
SH = dτ (−xm pm − 12 iγ m γm + 12 vp2 + iλγ · p)

where λ is also anticommuting so that each term in the action is bosonic.

Find the algebra of the constraints, and the transformations they generate
on the variables appearing in the action. Show that the “Dirac equation”
γ · p|Ψi = 0 implies p2 |Ψi = 0. Find the Lagrangian form of the action as
usual by eliminating p by its equation of motion. (Note λ2 = 0.)
Excercise IIIB1.3
Consider a “supercoordinate” X m that is a function of both a fermionic vari-
able ζ and the usual τ :

X m (τ, ζ) = xm (τ ) + iζγ m (τ )

where the Taylor expansion in ζ terminates because ζ 2 = 0. Identify x with

the usual x, and γ with its fermionic partner introduced in the previous
problem. In analogy to the way γ · p was the square root of the τ -translation
generator 12 p2 , we can define a square root of ∂/∂τ by the “covariant fermionic
derivative”
∂ ∂ ∂
D= + iζ ⇒ D2 = i
∂ζ ∂τ ∂τ
We also want to generalize v in the same way as x, to make the action inde-
pendent of coordinate choice for both τ and ζ. This suggests defining

E = v −1 + iζλ

and the gauge invariant action

Z
SL = dτ dζ 12 E(D2 X m )DXm
 B. PARTICLES 135

Integrate this action over ζ, and show this agrees with the action of the
previous problem after suitable redefinitions (including the normalization of
R
dζ).
The (D+2)-dimensional (conformal) representation of the massless particle (sub-
section IA6) can be derived from the action
Z
.
S= dτ 12 (−y 2 + λy 2 )

where λ is a Lagrange multiplier. This action is gauge invariant under

. . . . ...
δy = y − 12 y, δλ = λ + 2λ + 12 

If we vary λ to eliminate it and y − as in subsection IA6, the action becomes

Z
.
S = − dτ 12 e2x2

which agrees with the previous result, identifying v = e−2 , which also guarantees
v > 0.
Excercise IIIB1.4
Find the Hamiltonian form of the action for y: The constraints are now y 2,
r2 , and y · r, in terms of the conjugate r to y (see excercise IA6.2). Find
the gauge transformations in the standard way (see subsection IIIA5). Show
how the above Lagrangian form can be obtained from it, including the gauge
transformations.
Using instead the corresponding twistor (subsection IIB6) to satisfy y 2 = 0, the
massless, spinless particle now has a single term for its mechanics action:
Z
. .
S = dτ 14 ABCD z Aα z B α z Cβ z D β

Unlike all other relativistic mechanics actions, all variables have been unified into just
z, without the introduction of square roots.
Excercise IIIB1.5
.
Expressing z in terms of λα µ and xµµ as in subsection IIB6, show this action
reduces to the previous one.
136 III. LOCAL

2. Gauges
Rather than use the equation of motion to eliminate v it’s more convenient to use
a gauge choice: The gauge v = 1 is called “affine parametrization” of the worldline.
.
Note that the gauge transformation of v, δv = ζ, has no dependence on the coordi-
nates x and momenta p, so that choosing the gauge v = 1 avoids any extraneous x
or p dependence that could arise from the gauge fixing. (The appearance of such de-
R
pendence will be discussed in later chapters.) Since T = dτ v, the intrinsic length,
is gauge invariant, that part of v still remains when the length is finite, but it can be
.
incorporated into the limits of integration: The gauge v = 1 is maintained by ζ = 0,
and this constant ζ can be used to gauge one limit of integration to zero, completely
RT
fixing the gauge (i.e., the choice of τ ). We then integrate 0 , where T ≥ 0 (since
originally v > 0), and T is a variable to vary in the action. The gauge-fixed action is
then Z T
.
SH,GF = dτ [−xm pm + 12 (p2 + m2 )]
0
In the massive case, we can instead choose the gauge v = 1/m; then the equations
of motion imply that τ is the proper time. The Hamiltonian p2 /2m + constant then
resembles the nonrelativistic one.
Another useful gauge is the “lightcone gauge”
x+
τ=
p+
which, unlike the Poincaré covariant gauge v = 1, fixes τ completely; since the gauge
variation δ(x+/p+ ) = ζ, we must set ζ = 0 to maintain the gauge. Also, the gauge
transformation is again x and p independent. In lightcone gauges we always assume
p+ 6= 0, since we often divide by it. This is usually not too dangerous an assumption,
since we can treat p+ = 0 as a limiting case (in D>2).
We saw from our study of constrained systems that, for every degree of freedom we
can gauge away, the conjugate variable can be fixed by the constraint that generates
that gauge invariance: In the case where the constraint is p, the gauge transformation
is δq = λ, so we gauge q = 0 and use the constraint p = 0. In lightcone gauges the
constraints are almost linear: The gauge condition is x+ = p+τ and the constraint is
p− = ..., so the Lagrange multiplier v is varied to determine p− . On the other hand,
varying p− gives
δp− ⇒ v = 1

so this gauge is a special case of the gauge v = 1. An important point is that we used
only “auxiliary” equations of motion: those not involving time derivatives. (A slight
 B. PARTICLES 137

trick involves the factor of p+ : This is a constant by the equations of motion, so we

.
can ignore p+ terms. However, technically we should not use that equation of motion;
.
instead, we can redefine x− → x− + ..., which will generate terms to cancel any p+
terms.) The net result of gauge fixing and the auxiliary equation on the action is
Z ∞
. .
SH,GF = dτ [x−p+ − xi pi + 12 (pi2 + m2 )]
−∞

where xa = (x+, x−, xi ), etc. In particular, since we have fixed one more gauge degree
of freedom (corresponding to constant ζ), we have also eliminated one more constraint
variable (T , the constant part of v). This is one of the main advantages of lightcone
gauges: They are “unitary”, eliminating all unphysical degrees of freedom.
Excercise IIIB2.1
Another obvious gauge is τ = x0, which works as well as the lightcone gauge
as far as eliminating worldline coordinate invariance is concerned. (The same
is true for τ = n · x for any constant vector n.) Unfortunately, the same is not
true for the auxiliary equations of motion: After using the gauge condition,
p0 appears without time derivatives, so it and v can be eliminated by their
equations of motion. Show this gauge is consistent only for p0 > 0. The
resulting square root is awkward except in the nonrelativistic limit: Take it,
and compare with the usual nonrelativistic mechanics.

3. Coupling
One way to introduce external fields into the mechanics action is by considering
the most general Lagrangian quadratic in τ derivatives:
Z
. . .
SL = dτ [− 12 v −1gmn (x)xm xn + Am (x)xm + vφ(x)]

In the free case we have constant fields gmn = ηmn , Am = 0, and φ = 12 m2. The v
dependence has been assigned consistent with worldline coordinate invariance. The
curved-space metric tensor gmn describes gravity, the D-vector potential Am describes
electromagnetism, and φ is a scalar field that can be used to introduce mass by
interaction.
Excercise IIIB3.1
Use the method of the problem IIIB1.3 to write the nonrelativistic action
for a spinning particle in terms of a 3-vector (or (D−1)-vector) X i (τ, ζ) and
the fermionic derivative D. Find the coupling to a magnetic field, in terms
of the 3-vector potential Ai (X). Integrate the Lagrangian over ζ. Show
138 III. LOCAL

that the quantum mechanical square of ψ i[pi + Ai (x)] is proportional to the

Hamiltonian.
Excercise IIIB3.2
Derive the Lorentz force law by varying the Lagrangian form of the action for
the relativistic particle, in an external electromagnetic field (but flat metric),
with respect to x.
This action also has very simple transformation properties under D-dimensional
gauge transformations on the external fields:

δgmn = p ∂p gmn + gp(m ∂n) p, δAm = p∂p Am + Ap∂m p − ∂m λ, δφ = p∂p φ

⇒ SL [x] + δSL [x] = SL [x + ] − λ(xf ) + λ(xi )

Rτ
where we have integrated the action τif dτ and set x(τi ) = xi , x(τf ) = xf . These
transformations have a very natural interpretation in the quantum theory, where
Z
Dx e−iS = hxf |xi i

Then the λ transformation of A is canceled by the U(1) (phase) transformation

ψ 0(x) = eiλ(x)ψ(x)

in the inner product

Z Z
hψf |ψii = dxf dxi hψf |xf ihxf |xiihxi |ψii = dxf dxi ψf *(xf )hxf |xiiψi (xi )

while the  transformation associated with gmn is canceled by the D-dimensional

coordinate transformation
ψ 0(x) = ψ(x + )

4. Conservation
There are two types of conservation laws generally found in physics: In mechanics
.
we usually have global conservation laws, of the form Q = 0, associated with a
symmetry of the Hamiltonian H generated by a conserved quantity Q:
.
0 = δH = i[Q, H] = −Q

On the other hand, in field theory we have local conservation laws, since the action
R
for a field is written as an integral dD x of a Lagrangian density that depends only
 B. PARTICLES 139

on fields at x, and a finite number of their derivatives. The local conservation law
implies a global one, since
Z Z .
dD x d dD−1 x 0
∂m J = 0 ⇒ 0 =
m
∂ m J m
∼ J = Q =0
(2π)D/2 dt (2π)D/2
where we have integrated over a volume whose boundaries in space are at infin-
ity (where J vanishes), and whose boundaries in time are infinitesimally separated.
Equivalently, the global symmetry is a special case of the local one.
A simple way to derive the local conservation laws is by coupling gauge fields: We
couple the electromagnetic field Am to arbitrary charged matter fields φ and demand
gauge invariance of the matter part of the action, the matter-free part of the action
being separately invariant. We then have
Z  
δSM δSM
0 = δSM = dx (δAm) + (δφ)
δAm δφ
using just the definition of the functional derivative δ/δ. Applying the matter field
equations δSM /δφ = 0, integration by parts, and the gauge transformation δAm =
−∂m λ, we find
Z  
δSM δSM
0 = dx Λ ∂m ⇒ Jm = , ∂m J m = 0
δAm δAm
Similar remarks apply to gravity, but only if we evaluate the “current”, in this case the
energy-momentum tensor, in flat space gmn = ηmn , since gravity is self-interacting.
We then find 
δSM 
T mn
= −2  , ∂m T mn = 0
δgmn gmn =ηmn
where the normalization factor of −2 will be found later for consistency with the
particle. In this case the corresponding “charge” is the D-momentum:
Z D−1
m d x 0m
P = D/2
T
(2π)
In particular we see that the condition for the energy in any region of space to be
nonnegative is
T 00 ≥ 0

To apply this to the action for the particle in external fields, we must first dis-
tinguish the particle coordinates X(τ ) from coordinates x for all of spacetime: The
particle exists only at x = X(τ ) for some τ , but the fields exist at all x. In this
notation we can write the mechanics action as
Z  Z . .
SL = dx − gmn (x) dτ δ(x − X) 12 v −1 X m X n
Z Z 
.m
+Am (x) dτ δ(x − X)X + φ(x) dτ δ(x − X)v
140 III. LOCAL

R
using dx δ(x − X(τ )) = 1. We then have
Z .
m
J (x) = dτ δ(x − X)X m

Z . .
T mn
= dτ δ(x − X)v −1 X m X n

Note that T 00 ≥ 0 (since v > 0). Integrating to find the charge and momentum:
Z .0 Z .
Q= dτ δ(x − X )X = dX 0 (X 0)δ(x0 − X 0 ) = (p0 )
0 0

Z .0 .m Z . .
P m
= dτ δ(x − X )v X X = dX 0 (X 0 )δ(x0 − X 0 )v −1X m = (p0 )pm
0 0 −1

.
where we have used p = v −1X (for the free particle), where p is the momentum
conjugate to X, not to be confused with P . The factor of (p0 ) ((u) = u/|u| is the
sign of u) comes from the Jacobian from changing integration variables from τ to X 0 .
The result is that our naive expectations for the momentum and charge of the
particle can differ from the correct result by a sign. In particular p0 , which semi-
classically is identified with the angular frequency of the corresponding wave, can
be either positive or negative, while the true energy P 0 = |p0 | is always positive, as
physically required. (Otherwise all states could decay into lower-energy ones: There
would be no lowest-energy state, the “vacuum”.) When p0 is negative, the charge Q
and dX 0 /dτ are also negative. In the massive case, we also have dX 0 /ds negative.
This means that as the proper time s increases, X 0 decreases. Since the proper time
is the time as measured in the rest frame of the particle, this means that the particle
is traveling backward in time: Its clock changes in the direction opposite to that of
the coordinate system xm . Particles traveling backward in time are called “antiparti-
cles”, and have charges opposite to their corresponding particles. They have positive
true energy, but the “energy” p0 conjugate to the time is negative.
Excercise IIIB4.1
Compare these expressions for the current and energy-momentum tensor to
those from the semiclassical expansion in excercise IIIA4.1. (Include the in-
verse metric to define the square of −∂m S + qAm there.)
 B. PARTICLES 141

5. Pair creation
Free particles travel in straight lines. Nonrelativistically, external fields can alter
the motion of a particle to the extent of changing the signs of spatial components of
the momentum. Relativistically, we might then expect that interactions could also
change the sign of the energy, or at least the canonical energy p0 . As an extreme case,
consider a worldline that is a closed loop: We can pick τ as an angular coordinate
around the loop. As τ increases, X 0 will either increase or decrease. For example, a
circle in the x0-x1 plane will be viewed by the particle as repeating its history after
some finite τ , moving forward with respect to time x0 until reaching a latest time tf ,
and then backward until some earliest time ti . On the other hand, from the point of
view of an observer at rest with respect to the xm coordinate system, there are no
particles until x0 = ti, at which time both a particle and an antiparticle appear at
the same position in space, move away from each other, and then come back together
and disappear. This process is known as “pair creation and annihilation”.

tf
x0
x1
τ
ti

Whether such a process can actually occur is determined by solving the equations
of motion. A simple example is a particle in the presence of only a static electric
field, produced by the time component A0 of the potential. We consider the case of a
piecewise constant potential, vanishing outside a certain region and constant inside.
Then the electric field vanishes except at the boundaries, so the particle travels in
straight lines except at the boundaries. For simplicity we reduce the problem to two
dimensions:
A0 = −V for 0 ≤ x1 ≤ L, 0 otherwise
for some constant V . The action is, in Hamiltonian form,
Z
.
SH = dτ {−xm pm + v 21 [(p + A)2 + m2 ]}

and the equations of motion are

.
pm = −v(p + A)n∂m An ⇒ p0 = E
p
(p + A)2 = −m2 ⇒ p1 = ± (E + A0 )2 − m2
. . .
v −1 x = p + A ⇒ v −1 x1 = p1 , v −1x0 = E + A0
142 III. LOCAL

.
where E is a constant (the canonical energy at x1 = ∞) and the equation p1 = ...
is redundant because of gauge invariance. We assume E > 0, so initially we have a
particle and not an antiparticle.
We look only at the cases where the worldline begins at x0 = x1 = −∞ (lower
left) and continues toward the right till it reaches x0 = x1 = +∞ (upper right), so
.
that p1 = v −1x1 > 0 everywhere (no reflection). However, the worldline might bend
.
backward in time (x0 < 0) inside the potential: To the outside viewer, this looks
like pair creation at the right edge before the first particle reaches the left edge; the
antiparticle then annihilates the original particle when it reaches the left edge, while
the new particle continues on to the right. From the particle’s point of view, it has
simply traveled backward in time so that it exits the right of the potential before it
enters the left, but it is the same particle that travels out the right as came in the
left. The velocity of the particle outside and inside the potential is
√

 E 2 − m2

 outside
dx1 E
= p
dx0   (E − V )2 − m2

 inside
E−V
From the sign of the velocity we then see that we have normal transmission (no
antiparticles) for E > m + V and E > m, and pair creation/annihilation when
V −m>E >m ⇒ V > 2m
The true “kinetic” energy of the antiparticle (which appears only inside the potential)
is then −(E − V ) > m.
Excercise IIIB5.1
This solution might seem to violate causality. However, in mechanics as well
as field theory, causality is related to boundary conditions at infinite times.
Describe another solution to the equations of motion that would be inter-
preted by an outside observer as pair creation without any initial particles:
What happens ultimately to the particle and antiparticle? What are the al-
lowed values of their kinetic energies (maximum and minimum)? Since many
such pairs can be created by the potential alone, it can be accidental (and not
acausal) that an external particle meets up with such an antiparticle. Note
that the generator of the potential, to maintain its value, continuously loses
energy (and charge) by emitting these particles.

REFERENCES
1 A. Barducci, R. Casalbuoni, and L. Lusanna, Nuo. Cim. 35A (1976) 377;
L. Brink, S. Deser, B. Zumino, P. DiVecchia, and P. Howe, Phys. Lett. 64B (1976) 435;
 B. PARTICLES 143

P.A. Collins and R.W. Tucker, Nucl. Phys. B121 (1977) 307:
world-line metric; classical mechanics for relativistic spinors.
2 R. Marnelius, Phys. Rev. D20 (1979) 2091;
W. Siegel, Int. J. Mod. Phys. A 3 (1988) 2713:
conformal invariance in classical mechanics actions.
3 W. Siegel, hep-th/9412011, Phys. Rev. D52 (1995) 1042:
twistor formulation of classical mechanics.
4 R.P. Feynman, Phys. Rev. 74 (1948) 939:
classical pair creation.
144 III. LOCAL

......................... .........................
......................... C. YANG-MILLS .........................

The concept of a “covariant derivative” allows the straightforward generalization

of electromagnetism to a self-interacting theory, once U(1) has been generalized to a
nonabelian group. Yang-Mills theory is an essential part of the Standard Model.

1. Nonabelian
The group U(1) of electromagnetism is Abelian: Group elements commute, which
makes group multiplication equivalent to multiplication of real numbers, or addition
if we write U = eiG. The linearity of this addition is directly related to the linearity
of the field equations for electromagnetism without matter. On the other hand, the
nonlinearity of nonabelian groups causes the corresponding particles to interact with
themselves: Photons are neutral, but “gluons” have charge and “gravitons” have
weight.
In coupling electromagnetism to the particle, the relation of the canonical mo-
mentum to the velocity is modified: Classically, the covariant momentum is dx/dτ =
p + qA for a particle of charge q (e.g., q = 1 for the proton). Quantum mechanically,
the net effect is that the wave equation is modified by the replacement

∂ → ∇ = ∂ + iqA

which accounts for all dependence on A (“minimal coupling”). This “covariant deriva-
tive” has a fundamental role in the formulation of gauge theories, including gravity.
Its main purpose is to preserve gauge invariance of the action that gives the wave
equation, which would otherwise be spoiled by derivatives acting on the coordinate-
dependent gauge parameters: In electromagnetism,

ψ 0 = eiqλψ, A0 = A − ∂λ ⇒ (∇ψ)0 = eiqλ(∇ψ)

or more simply
∇0 = eiqλ∇e−iqλ
(More generally, q is some Hermitian matrix when ψ is a reducible representation of
U(1).)
Yang-Mills theory then can be obtained as a straightforward generalization of elec-
tromagnetism, the only difference being that the gauge transformation, and therefore
the covariant derivative, now depends on the generators of some nonabelian group.
We begin with the hermitian generators

[Gi , Gj ] = −ifij k Gk , Gi † = Gi
 C. YANG-MILLS 145

and exponentiate linear combinations of them to obtain the unitary group elements

g = eiλ, λ = λi Gi ; λi * = λi ⇒ g† = g−1

We then can define representations of the group (see subsection IB1)

ψ 0 = eiλψ, ψ †0 = ψ †e−iλ ; (Gi ψ)A = (Gi )AB ψB

For compact groups charge is quantized: For example, for SU(2) the spin (or, for
internal symmetry, “isospin”) is integral or half-integral. On the other hand, with
Abelian groups the charge can take continuous values: For example, in principle the
proton might decay into a particle of charge π and another of charge 1 − π. The
experimental fact that charge is quantized suggests already semiclassically that all
interactions should be descibed by (semi)simple groups.
If λ is coordinate dependent (a local, or “gauge” transformation), the ordinary
partial derivative spoils gauge covariance, so we introduce the covariant derivative

∇a = ∂a + iAa, Aa = Aa i Gi

Thus, the covariant derivative acts on matter in a way similar to the infinitesimal
gauge transformation,

δψA = iλi GiA B ψB , ∇a ψA = ∂aψA + iAai GiA B ψB

Gauge covariance is preserved by demanding it have a covariant transformation law

∇0 = eiλ∇e−iλ ⇒ δA = −[∇, λ] = −∂λ − i[A, λ]

The gauge covariance of the field strength follows from defining it in a manifestly
covariant way:

[∇a, ∇b ] = iFab ⇒ F 0 = eiλF e−iλ, Fab = Fabi Gi = ∂[aAb] + i[Aa, Ab ]

⇒ Fab i = ∂[aAb]i + Aaj Ab k fjk i

The Jacobi identity for the covariant derivative is the Bianchi identity for the field
strength:
0 = [∇[a, [∇b, ∇c] ]] = i[∇[a, Fbc] ]
(If we choose instead to use antihermitian generators, all the explicit i’s go away;
however, with hermitian generators the i’s will cancel with those from the derivatives
when we Fourier transform for purposes of quantization.) Since the adjoint represen-
tation can be treated as either matrices or vectors (see subsection IB2), the covariant
146 III. LOCAL

derivative on it can be written as either a commutator or multiplication: For example,

we may write either or [∇, F ] or ∇F , depending on the context.
Actions then can be constructed in a manifestly covariant way: For matter, we
take a Lagrangian LM,0(∂, ψ) that is invariant under global (constant) group trans-
formations, and couple to Yang-Mills as

LM,0(∂, ψ) → LM,A = LM,0 (∇, ψ)

(This is the analog of minimal coupling in electrodynamics.) The representation we

use for Gi in ∇a = ∂a + iAiaGi is determined by how ψ represents the group. (For
an Abelian group factor U(1), G is just the charge q, in multiples of the g for that
factor.) For example, the Lagrangian for a massless scalar is simply

L0 = 12 (∇aφ)† (∇aφ)

(normalized for a complex representation).

For the part of the action describing Yang-Mills itself we take (in analogy to the
U(1) case)
j
LA (Aia) = 8g12 F iabFab ηij
A

where ηij is the Cartan metric (see subsection IB2). This way of writing the action
is independent of our choice of normalization of the structure constants, and so gives
one unambiguous definition for the normalization of the coupling constant g. (It is
invariant under any simultaneous redefinition of the fields and the generators that
leaves the covariant derivative invariant.) Generally, for simple groups we can choose
to (ortho)normalize the generators Gi with the condition

ηij = cA δij

for some constant cA; for groups that are products of simple groups (semisimple),
we might choose different normalization factors (but, of course, also different g’s) for
each simple group. For Abelian groups (U(1) factors) ηij = 0, but then the gauge
field has no self-interactions, so the normalization of the coupling constant is defined
only by matter terms in the action, and we can replace ηij with δij in the above.
Usually it will prove more convenient to use matrix notation: Choosing some
convenient representation R of Gi (not necessarily the adjoint), we write

LA(Aia ) = 1
2 trR
8gR
F abFab

The normalization of the trace is determined by R, and thus so is the normalization

convention for the coupling constant; a change in the representation used in the action
 C. YANG-MILLS 147

can also be absorbed by a redefinition of the coupling. For example, comparing the
defining and adjoint representations of SU(N) (see subsection IB2),

LA = 1
2 trD
8gD
F abFab = 1
2 trA
8gA
F abFab ⇒ gA2 = 2NgD
2

In general, we specify our normalization of the structure constants by fixing cR for

some R, and our normalization of the coupling constant by specifying the choice of
representation used in the trace (or use explicit adjoint indices). As a rule, we find
the most convenient choices of normalization are

cD = 1, g = gD

(see subsection IB5).

Excercise IIIC1.1
Write the action for SU(N) Yang-Mills coupled to a massless (2-component)
spinor in the defining representation. Make all (internal and Lorentz) indices
explicit (no “tr”, etc.), and use defining (N-component) indices on the Yang-
Mills field.
We have chosen a normalization where the Yang-Mills coupling constant g appears
only as an overall factor multiplying the F 2 term (and similarly for the electromagnetic
coupling, as discussed in previous chapters). An alternative is to rescale A → gA and
F → gF everywhere; then ∇ = ∂ + igA and F = ∂A + ig[A, A], and the F 2 term has
no extra factor. This allows the Yang-Mills coupling to be treated similarly to other
couplings, which are usually not written multiplying kinetic terms (unless analogies to
Yang-Mills are being drawn), since (almost) only for Yang-Mills is there a nonlinear
symmetry relating kinetic and interaction terms.
Current conservation works a bit differently in the nonabelian case: Applying
the same argument as in subsection IIIB4, but taking into account the modified
(infinitesimal) gauge transformation law, we find
δSM
Jm = , ∇m J m = 0
δAm
Since ∂m J m 6= 0, there is no corresponding covariant conserved charge.
Excercise IIIC1.2
Let’s look at the field equations:
a Using properties of the trace, show the entire covariant derivative can be
integrated by parts as
Z Z Z Z
dx tr(A[∇, B]) = − dx tr([∇, A]B), dx ψ ∇χ = − dx (∇ψ)†χ
†
148 III. LOCAL

for matrices A, B and column vectors ψ, χ.

b Show
δFab = ∇[aδAb]

c Using the definition of the current as for electromagnetism (subsection IIIB4),

derive the field equations with arbitrary matter,

g2 2
∇ Fba
1 1 b
= Ja

d Show that gauge invariance of the action SA implies

∇a(∇b Fba) = 0

Also show this is true directly, using the Jacobi identity, but not the field
equations. (Hint: Write the covariant derivatives as commutators.)
e Expand the left-hand side in the field, as

1 1 b
g2 2
∇ Fba = 1 1 b
g2 2
∂ ∂[bAa] − ja

where j contains the quadratic and higher-order terms. Show the noncovari-
ant current
Ja = Ja + ja

is conserved. The j term can be considered the gluon contribution to the

current: Unlike photons, gluons are charged. Although the current is gauge
dependent, and thus physically meaningless, the corresponding charge can
be gauge independent under situations where the boundary conditions are
suitable.

2. Lightcone
Since gauge parameters are always of the same form as the gauge field, but with
one less vector index, an obvious type of gauge choice (at least from the point of view
of counting components) is to require the gauge field to vanish when one vector index
is fixed to a certain value. Explicitly, in terms of the covariant derivative we set

n·∇=n·∂ ⇒ n·A=0

for some constant vector na . We then can distinguish three types of “axial gauges”:
(1) “Arnowitt-Fickler”, or spacelike (n2 > 0), (2) “lightcone”, or lightlike (n2 = 0),
and “temporal”, or timelike (n2 < 0). By appropriate choice of reference frame, and
 C. YANG-MILLS 149

with the usual notation, we can write these gauge conditions as ∇1 = ∂ 1, ∇+ = ∂ +,

and ∇0 = ∂ 0.
One way to apply this gauge in the action is to keep the same set of fields, but
have explicit n dependence. A much simpler choice is to use a gauge choice such as
A0 = 0 simply to eliminate A0 explicitly from the action. For example, for Yang-Mills
we find
. .
A0 = 0 ⇒ F0i = Ai ⇒ 18 (Fab )2 = − 14 (Ai )2 + 18 (Fij )2
.
where “ ” here refers to the time derivative. Canonical quantization is simple in
this gauge, because we have the canonical time-derivative term. However, the gauge
condition can’t be imposed everywhere, as seen for the corresponding gauge for the
one-dimensional metric in subsection IIIB2, and in our general discussion in subsec-
tion IIIA5: Here we can generalize the time-ordered integral for the temporal gauge
to an integral path-ordered with respect to a straight-line path in the n direction:
  Z x 
−kn·∇(x) −iΛ(x,x−kn) −kn·∂ −iΛ(x,x−kn) 0 0
e =e e , e = P exp −i dx · A(x )
x−kn

Applying this gauge transformation to n · ∇, as in subsection IIIA5, fixes n · A to a

constant with respect to n · ∂; the effect on all of ∇ is:

∇0(x) = eiΛ(x,x−kn) ∇(x)e−iΛ(x,x−kn) ⇔ ∇0(x + kn) = ekn·∇(x) ∇(x)e−kn·∇(x)

For example, for the temporal gauge, if we choose “x” to be on the initial hypersurface
x0 = t0, then we can choose k = t − t0 so that ∇0 is evaluated at arbitrary time t:

∇0a(x) = [(ek∇0 (x)∇a(x)e−k∇0 (x))|x0 =t0 ]|k=x0

By Taylor expanding in k, this gives an explicit expression for Aa at all times in terms
of Aa, and Fab and its covariant time-derivatives, evaluated at some initial time, but
with simply A0(t, xi) = A0(t0, xi ). Thus, we still need to impose the A0 field equation
.
[∇i, Ai ] = 0 as a constraint at some initial time.
Excercise IIIC2.1
Show explicitly that the field equations for Ai following from the action for
Yang-Mills in the gauge A0 = 0 (see excercise IIIC1.2 for J = 0) imply that
.
the time derivative of the constraint [∇i, Ai ] = 0 vanishes.
In the case of the lightcone gauge we can carry this analysis one step further.
In subsection IIB3 we saw that lightcone formalisms are described by massless fields
with (D−2)-dimensional (“transverse”) indices. In the present analysis, gauge fixing
alone gives us, again for the example of pure Yang-Mills,

A+ = 0 ⇒ F +i = ∂ +Ai , F +− = ∂ +A−, F −i = ∂ −Ai − [∇i, A−]

150 III. LOCAL

⇒ 1
8
(F ab)2 = − 14 (∂ +A− )2 − 12 (∂ +Ai)(∂ −Ai − [∇i, A−]) + 18 (F ij )2

In the lightcone formalism ∂ − (−∂+) is to be treated as a time derivative, while

∂ + can be freely inverted (i.e., modes propagate to infinity in the x+ direction, but
boundary conditions set them to vanish in the x− direction). Thus, we can treat A−
as an auxiliary field. The solution to its field equation is
1
A− = [∇i, ∂ +Ai ]
∂ +2
which can be substituted directly into the action:
1
1
8
(F ab)2 = 12 Ai∂ +∂ −Ai + 18 (F ij )2 − 14 [∇i, ∂ +Ai ] [∇j , ∂ +Aj ]
∂ +2
1
= − 14 Ai Ai + i 12 [Ai, Aj ]∂ iAj + i 12 (∂ iAi ) + [Aj , ∂ +Aj ]
∂
+ i 1
− 8 [A , A ] + 4 [A , ∂ A ] +2 [A , ∂ A ]
1 i j 2 1 i j + j
∂

We can save a couple of steps in this derivation by noting that elimination of any
auxiliary field, appearing quadratically (as in going from Hamiltonian to Lagrangian
formalisms), has the effect

L = 12 ax2 + bx + c → − 12 ax2|∂L/∂x=0 + L|x=0

In this case, the quadratic term is (F −+)2 , and we have

1
8
(F ab )2 = 18 (F ij )2 − 12 F +i F −i − 14 (F +−)2 → 18 (F ij )2 − 12 (∂ +Ai )(∂ −Ai ) + 14 (F +−)2

where the last term is evaluated at

1
0 = [∇a, F +a] = −∂ +F +− + [∇i , F +i] ⇒ F +− = [∇i, F +i ]
∂+
1
⇒ L = 18 (F ij )2 + 12 Ai ∂ +∂ −Ai − 14 [∇i, ∂ +Ai ] [∇j , ∂ +Aj ]
∂ +2
as above.
In this case, canonical quantization is even simpler, since interpreting ∂ − as the
time derivative makes the action look like that for a nonrelativistic field theory, with
a kinetic term linear in time derivatives (as well as interactions without them). The
free part of the field equation is also simpler, since the kinetic operator is now just
. (This is true in general in lightcone formalisms from the analysis of free theories
in chapter XII.) In general, lightcone gauges are the simplest for analyzing physical
degrees of freedom (within perturbation theory), since the maximum number of de-
grees of freedom is eliminated, and thus kinetic operators look like those of scalars.
 C. YANG-MILLS 151

On the other hand, interaction terms are more complicated because of the nonlo-
cal Coulomb-like terms involving 1/∂ + : The inverse of a derivative is an integral.
(However, in practice we often work in momentum space, where 1/p+ is local, but
Fourier transformation itself introduces multiple integrals.) This makes lightcone
gauges useful for discussing unitarity (they are “unitary gauges”), but inconvenient
for explicit calculations. However, in subsection VIB6 we’ll find a slight modification
of the lightcone that makes it the most convenient method for certain calculations.
(In the literature, “lightcone gauge” is sometimes used to refer to an axial gauge
where A+ is set to vanish but A− is not eliminated, and D-vector notation is still
used, so unitarity is not manifest. Here we always eliminate both components and
explicitly use (D − 2)-vectors, which has distinct technical advantages.)
Although spin 1/2 has no gauge invariance, the second step of the lightcone
formalism, eliminating auxiliary fields, .can also be applied there: For example, for a
massless spinor in D=4, identifying ∂ = ∂ − as the lightcone “time” derivative, we
.
vary ψ̄ (or ψ ) as the auxiliary field:
. . . . . . . .
−iL = ψ̄ ∂ ⊕⊕ ψ + ψ̄ ⊕∂ ψ ⊕ − ψ̄ ∂ ⊕ ψ ⊕ − ψ̄ ⊕∂ ⊕ ψ
1 .
⇒ ψ = . ∂ ⊕ ⊕
ψ
∂ ⊕⊕
. 1
⇒ L = ψ̄ ⊕ 2 . ψ ⊕
i∂ ⊕⊕
This tells us that a 4D massless spinor, like a 4D massless vector (or a complex scalar)
has only 1 complex (2 real) degree of freedom, describing a particle of helicity +1/2
and its antiparticle of helicity −1/2 (±1 for the vector, 0 for the scalar), in agreement
with our general discussion of helicity in subsection IIB7. On the other hand, in the
massive case we can always go to a rest frame, so the analysis is in terms of spin
(SU(2) for D=4) rather than helicity. For a massive Weyl spinor we can perform the
same analysis as above, with the modifications

im . . . 1( − m2 ) ⊕
L → L + √ (ψ ⊕ψ + ψ̄ ⊕ψ̄ ) ⇒ L = ψ̄ ⊕ 2 . ψ
2 i∂ ⊕⊕
where we have dropped some terms that vanish upon using integration by parts and
the antisymmetry of the fermions. So now we have the two states of an SU(2) spinor,
but these are identified with their antiparticles. This differs from the vector: While
for the spinor we have 2 states of a given energy for both the massless and massive
cases, for a vector we have 2 for the massless but 3 for the massive, since for SU(2)
spin s has 2s+1 states.
152 III. LOCAL

Excercise IIIC2.2
Show that integration by parts for 1/∂ gives just a sign change, just as for ∂.
In general dimensions, massless particles are representations of the “little group”
SO(D−2) (the helicity SO(2) in D=4), as described in subsection IIB3. Massive
particles represent the little group SO(D−1), corresponding to dimensional reduction
from an extra dimension, as described in subsection IIB4.

3. Plane waves
The simplest nontrivial solutions to nonabelian field equations are the general-
izations of the plane wave solutions of the free theory. We begin with general, free,
massless theories, as analyzed in subsection IIB3. In the lightcone frame only p+ is
nonvanishing. In position space this means the field strength depends only on x−.
This describes a wave traveling at the speed of light in the positive x1 direction, with
no other spatial dependence (i.e., a plane wave). We allow arbitrary dependence on
x−, corresponding to a superposition of waves with parallel momenta (but different
values of p+ ). While its dependence on only x− solves the Klein-Gordon equation,
Maxwell’s equations are solved by giving the field strength as many upper + indices
as possible, and no upper −’s.
Generalizing to interactions, we notice that the Yang-Mills field equations and
Bianchi identities differ from Maxwell’s equations only by the covariantization of the
derivatives (at least for pure Yang-Mills). Because Maxwell’s equations were satisfied
by just restricting the index structure, we can do the same for the covariant derivatives
by assuming that only ∇+ is novanishing on the field strengths. In other words, we
can solve the field equations and Bianchi identities by choosing the only nontrivial
components of the gauge fields to be those in ∇+.
The final step is to solve the relation between covariant derivative and field
strength. This is simple because the index structure we found implies the only non-
trvial commutators are
[∂ i, ∇+] = iF i+, [∂ −, ∇+] = 0
In particular, this implies that the gauge fields have no x+ dependence, and only a
very simple dependence on xi . We find directly
A+ = xi F i+(x−)
where F i+(x−) is unrestricted (other than the explicit index structure and coordinate
dependence). Of course, this result can also be used in the free theory, although it
differs from the usual lightcone gauge.
 C. YANG-MILLS 153

4. Self-duality
The simplest and most important solutions to the field equations are those that
are invariant under the “duality” symmetry that relates electric and magnetic charge:

[∇a, ∇b] = ± 12 abcd [∇c, ∇d ]

Applying the self-duality condition twice, we find

abef ef cd = +δ[a

c d
δb]

which requires an even number of time dimensions. For example, since the action
is usually Wick rotated anyway for perturbative purposes, we might assume that
we should do the same for classical solutions that are not considered as “small”
fluctations about the usual vacuum. (Such a Euclidean definition of field theory
has been considered for a mathematically rigorous formalism, called “constructive
quantum field theory”, since the Gaussian path integrals for scalars and vectors are
then well-defined and convergent. However, other spins, such as for fermions or
gravity, are a problem in this approach.) Alternatively, we can replace abcd with
iabcd and complexify our fields. The self-duality condition, when combined with the
Bianchi identities, implies the field equations: For Yang-Mills,

∇[a Fbc] = 0 ⇒ 0 = ± 12 abcd ∇aFbc = 14 abcd ∇a bcef F ef = ∇aF ad

Since the self-duality condition is only first-order in derivatives, it’s easier to solve
than the usual field equations.
Plane wave solutions provide a simple example of self-duality, since the field
strengths can easily be written as the sum of self-dual and anti-self-dual parts: In
Minkowski space we define the self-dual part as helicity +1 (f . .), and anti-self-dual
αβ
as −1 (fαβ ). For example, for a wave traveling in the “1” direction, the F +2 ∓ iF +3
components give the two self-dualities for Yang-Mills, describing helicities ±1 (the
two circular polarizations).
Before further analyzing solutions to the self-duality condition, we consider ac-
tions that use self-dual fields directly. This will allow us to describe not only theories
whose only solutions are self-dual, but also more standard theories as perturbations
about self-duality, and even massive theories. The most unusual feature of this ap-
proach is that complex fields are used without their complex conjugates, since this
is implied in D=3+1 by self-duality. (Alternatively, we can Wick rotate to 2+2 di-
mensions, where all Lorentz representations are real.) There are two stages to this
154 III. LOCAL

approach: (1) Use a first-order formalism where the auxiliary field is self-dual. The
usual first-order actions for spin 1/2 (Weyl or Dirac) already can be interpreted in
this way, where “self-duality” means “chirality”. (2) For the massive theory, elimi-
nate the non-self-dual field (as an auxiliary field, as allowed by the mass term), so
that the dynamics is described by the self-dual field, which was formerly considered
as auxiliary. The massless theory then can be treated as a limiting case.
The simplest (and perhaps most useful) example is massive spin 1/2 coupled in
a real representation to Yang-Mills fields:
. .
L = ψ T αi∇αα ψ̄α. + 1
√
2 2
m(ψ T αψα + ψ̄ T αψ̄α.)

where the transposition (“T ”) refers to the Yang-Mills group index (with respect to
which the spinors are column vectors). Note that ψ must be a real representation of
this group (AT = −A) for the mass term to be gauge invariant (unless the mass term
includes scalars: see the following chapter). Even though ψ and ψ̄ are complex conju-
gates, they can be treated independently as far as field equations are concerned, since
they are just different linear combinations of their real and imaginary parts. (Com-
plex conjugation can be treated as just a symmetry, related to unitarity.) Noticing
that the quadratic term for ψ̄ has no derivatives, we can treat it as an auxiliary field,
and integrate it out (i.e., eliminate it by its equation of motion, which gives an explicit
local expression for it):
√
L → − m2 [ 14 ψ T α( − m2)ψα + 12 ψ T αifαβ ψβ ]

where we have used the identity

. . .
∇α γ ∇β γ. = 12 {∇αγ , ∇β γ.} + 12 [∇αγ , ∇β γ.] = − 12 δαβ − ifαβ

whose simplicity followed from ψ being a real representation of the Yang-Mills group.
(Of course, we could have eliminated ψ instead, but not both.) For convenience we
also scale ψ by a constant
√
ψ → 2−1/4 mψ

to find the final result

L → − 14 ψ T α( − m2)ψα − 12 ψ T αifαβ ψβ

Now the massless limit can be taken easily. This action resembles that of a scalar, plus
a “magnetic-moment coupling”, which couples the “(anti-)self-dual” (chiral) spinor
ψα to only the (anti-)self-dual part fαβ of the Yang-Mills field strength.
 C. YANG-MILLS 155

For the same reason, the kinetic operator can be written in terms of just the
self-dual part Sαβ of the spin operator:

L = − 14 ψ T ( − m2 − if αβ Sβα)ψ

This operator is of the same form found by squaring the Dirac operator:

−2∇
/ 2 = −2(γ · ∇)2 = −({γ a , γ b } + [γ a, γ b ])∇a∇b = − iF abSba

except for the self-duality. The simple form of this result again depends on the
reality (parity invariance) of the Yang-Mills representation; although this squaring
trick can be applied for complex representations (parity violating), the coupling does
not simplify. This is related to the fact that real representations are required for our
derivation of the self-dual form.
In the special case where the real representation is the direct sum of a complex
one ψ+α with its complex conjugate ψ−α (as for quarks in the Standard Model, or
electrons in electrodynamics), we can rewrite the Lagrangian as

Lc = − 12 ψ+
Tα
( − m2)ψ−α − ψ+
Tα
ifαβ ψ−β

The method can also be generalized to the case of scalar couplings, but the action
becomes nonpolynomial.
For spin 1, we start with the massless case. We can write the Lagrangian for
Yang-Mills as
L = tr(Gαβ fαβ − 12 g 2 G2αβ )
where Gαβ is a (anti-)self-dual auxiliary field. Although this action is complex, elim-
inating G by its algebraic field equation gives the usual Yang-Mills action up to a
total derivative term (abcd Fab Fcd ), which can be dropped for purposes of perturbation
theory. For g = 0, this is an action where G acts as a Lagrange multiplier, enforcing
the self-duality of the Yang-Mills field strength.
If we simply add a mass term

Lm = 14 ( mg )2 A2

then A can be eliminated by its field equation, giving a nonpolynomial action of the
form
L + Lm → − 12 (∂G)[( m g
)2 + G]−1 (∂G) − 12 g 2 G2
Just as the spin-1/2 action contained only a 2-component spinor describing the 2
polarizations of spin 1/2, this action contains only the 3-component Gαβ , describing
the 3 polarizations of (massive) spin 1.
156 III. LOCAL

Excercise IIIC4.1
Find the Abelian part of this action. Show the free field equation is
( − m2)Gαβ = 0 (without gauge fixing).

5. Twistors
In four dimensions with an even number of time dimensions, the “Lorentz” group
factorizes (into SU(2)2 for D=4+0 and SL(2)2 for D=2+2). This makes self-duality
especially simple in spinor notation: For Yang-Mills (cf. electromagnetism in subsec-
tion IIA7),
0 0 0 0
[∇αβ , ∇γδ ] = iC αγ f β δ (f αβ = 0)
where we have written primes instead of dots to emphasize that the two kinds of
indices transform independently (instead of as complex conjugates, as in D=3+1). For
purposes of analyzing self-duality within perturbation theory, we can use a lightcone
method that breaks only one of the two SL(2)’s (or SU(2)’s), by separating out its
indices into the ⊕ and components:
0 0 0 0
[∇⊕α , ∇⊕β ] = 0 ⇒ ∇⊕α = ∂ ⊕α

where we have chosen a lightcone gauge: The vanishing of all field strengths for the
0
covariant derivative ∇⊕α says that it is pure gauge (as seen by ignoring all but the
0
x α coordinates). We now solve
0 0 0 0 0
[∇[⊕α , ∇ ]β ] = 0 ⇒ ∇ α = ∂ α + i∂ ⊕α φ
0 0
i.e., ∇ α − ∂ α has vanishing curl, and is therefore a gradient. We therefore have
0 0 0 0 0 0 0
A⊕α = 0, A α = ∂ ⊕α φ; f α β = −i∂ ⊕α ∂ ⊕β φ

These can also be written in terms of an arbitrary constant twistor α (= δ

α
above)
as
0 0 0 0 0 0
Aαβ = ∂ γβ (−iαγ φ), f α β = ∂ γα ∂ δβ (iγ δ φ)
0 0
The final self-duality condition [∇ α , ∇ β ] = 0 then gives the equation of motion
0
1
2 φ + (∂ ⊕α φ)(∂ ⊕α0 φ) = 0

Excercise IIIC5.1
Show that the sign convention for Wick rotation of the Levi-Civita tensor
consistent with the above equations is

Fab = 12 abcd F cd , Fαα0 ββ 0 = Cαβ fα0β 0 ⇒

 C. YANG-MILLS 157

αα0 ββ 0γγ 0 δδ0 = Cαβ Cγδ Cα0 δ0 Cβ 0γ 0 − Cαδ Cβγ Cα0 β 0 Cγ 0 δ0

Excercise IIIC5.2
Look at the action Gαβ fαβ for self-dual Yang-Mills in the lightcone gauge,
using the results above. Show that this action is equivalent to the lightcone
action for ordinary Yang-Mills (subsection IIIC2), with some terms in the
interaction dropped.
At least for 4D Yang-Mills, advantage can be taken of the conformal invariance
of the classical interacting theory by using a formalism where this invariance is man-
ifest. We saw in subsection IA6 that classical mechanics could be made manifestly
conformal by use of extra coordinates. Covariant derivatives can be defined in terms
of projective lightcone coordinates, but the twistor coordinates z Aα (see subsections
IIB6 and IIIB1) are more useful. The self-dual covariant derivatives then satisfy

[∇Aα, ∇Bβ ] = iCαβ fAB

in direct analogy to 4D spinor notation. This equation also can be solved by the
lightcone method used above, but now this method breaks only the internal SL(2)
symmetry, leaving SL(4) conformal symmetry manifest. More general self-dual field
strengths in this twistor space are also of the form fA...B , totally symmetric in the
indices. We also need to impose the constraint on the field strength

z Aα fAB = 0

(and similarly for the more general case) to restrict the range of indices to the usual
4D spinor indices (in which the field strengths are totally symmetric). Self-duality
implies the Bianchi identity
∇[AαfB]C = 0

which also generalizes to the other field strengths, and is the equivalent of the usual
first-order differential equations (Dirac, Maxwell, etc.) satisfied by 4D field strengths.
As usual, it in turn implies the interacting Klein-Gordon equation, which in the
Yang-Mills case is
2 ∇[A ∇B]α fCD = −i[fC[A, fB]D ]
1 α

The Bianchi together with the z index constraint imply the constraint on coordinate
dependence
(z Aα ∇Aβ + δβα)fBC = 0

which eliminates dependence on all but the usual 4D coordinates. These four equa-
tions are generically satisified by self-dual field strengths. The self-duality itself of the
158 III. LOCAL

field strengths is a consequence of their total symmetry in their indices, and the fact
that they are all lower (SL(4)) indices. (The z index constraint then reduces them to
SL(2) Weyl indices all of the same chirality.)
Excercise IIIC5.3
Derive the last three equations from the previous two (self-duality and zf=0).
Excercise IIIC5.4
Show that non-self-dual Yang-Mills is conformally invariant in D=4 by extend-
ing the (4+2)-dimensional formalism of subsections IA6 and IIIB1 (especially
excercise IIIB1.4): Show the field strength

FABC = −i 12 y[A[∇B , ∇C] ]

satisfies the gauge covariances

δAA = −[∇A, λ] − yA λ̂

and Bianchi identities

y[AFBCD] = ∇[AFBCD] = 0

The duality transformation

FABC → 16 ABCDEF F DEF

then suggests the field equations

y AFABC = ∇AFABC = 0

in addition to the usual constraint

y 2 FABC = 0

By reducing to D=4 coordinates with the aid of the above yF conditions, show
F reduces to the usual field strength, and the remaining equations reduce to
the usual gauge transformation, Bianchi identity, duality, and field equation.
 C. YANG-MILLS 159

6. Instantons
Another interesting class of self-dual solutions to Yang-Mills theory are “instan-
tons”, so called because the field strength is maximum at points in spacetime, unlike
the plane waves, whose wavefronts propagate from and toward timelike infinity. A
particular subset of these can be expressed in a very simple form by the ’t Hooft
ansatz in terms of a scalar field: In twistor notation, choosing the Yang-Mills gauge
group GL(2) (in 2+2 dimensions, or SU(2)⊗GL(1) for 4+0),

iAAαικ = −δακ ∂Aι ln φ ⇒ iAAαιι = −∂Aα ln φ

so the GL(1) piece is pure gauge, and has been included just for convenience. Note
that this ansatz ties the SL(2) twistor index with the SL(2) gauge group indices (ι, κ),
but in this notation the index that carries the spacetime (conformal) symmetry is free.
Imposing the self-duality condition on the field strength, and separating out the terms
symmetric and antisymmetric in AB, we find

ifABικ = − 12 φ∂(A κ ∂B)ιφ−1

φ−1 ∂A α ∂Bαφ = 0

The “field equation” for φ is just the twistor version of the (free) Klein-Gordon
equation, and its solution is the projective lightcone version of 4D point sources (see
subsection IA6): Since for any two 6D lightlike vectors y and y 0

y = e(x, 1, 12 x2) ⇒ y · y 0 = − 12 ee0(x − x0 )2

we have the solution

X
k+1
φ= (y · yi)−1 , yi2 = 0
i=1

with y given in terms of z as before, and yi are constant null vectors. “k” is the number
of instantons. (The one term for k = 0 is pure gauge.) The usual singularities in
the Klein-Gordon equation at y = yi are killed by the extra factor of φ−1 in the field
equation.
Excercise IIIC6.1
Let’s check the Klein-Gordon equation for y 6= yi directly in twistor space.
We will need the identity
!
X
yi2 = 0 ⇒ yiA[B yiCD] = 0 no
i
160 III. LOCAL

in the product
y · yi = 12 y AB yiAB
Prove this identity in two ways:
a Show it follows from the definition

yi2 = 14 ABCD yiAB yiCD

b Show it follows from plugging in the solution to the lightlike condition,

yiAB = ziA α ziBα

c Now use the identity to show the above solution satisfies its field equation by
evaluating the z derivatives.
We can rewrite this in the usual 4D coordinates by transforming from z Aα to λαµ
0 0
and xµν as z A α = λα ν (δνµ , xν µ ) (see subsection IIB6):
0 0
dz Aα AAαικ = dxµν Aµν 0 ικ + [(dλα ν )λ−1ν β ]z Aβ AAαικ = dxµν Aµν 0 ι κ − iλ−1ν ιdλκ ν

where in the first step we have used the expression for z in terms of λ and x, and in
the second we used the result that

(z Aα ∂Aβ + δβα)φ = 0

We now recognize that the gauge transformation that gets rid of all but the “x
components” of A (whose existence is guaranteed by the condition z Aα fAB = 0) uses
λ itself as the gauge parameter:

dz Aα AAαικ = −iλ−1ν ι dλκ ν + λ−1ν ι(dz Aα A0Aαν µ )λκ µ

The net result is that A can be reduced to an ordinary 4-dimensional expression by

just setting λ = δ in the original expression. Then
X 1
iAµµ0 ι κ = −δµκ∂ιµ0 ln φ, φ=
i
ei (x − xi )2

with y in terms of x and a scale factor (worldline metric) e as in subsections IA6

and IIIB1 (and dropping an overall factor that doesn’t contribute to A). Note that,
unlike the expression in twistor space, where conformal invariance is manifest, here
Lorentz invariance is tied to the Yang-Mills symmetry.
Excercise IIIC6.2
Show in 4D coordinates that the gauge-invariant quantity tr(f 2 ) is finite at
 C. YANG-MILLS 161

the points x = xi , where A is singular. (This means that the gauge choice is
singular, not physical quantities.)
Another important property of instantons is that they give finite contributions to
the action. In vector notation, we have
Z Z
d4 x d4 x abcd
F = 2  Fcd ⇒ S = 8g2 tr
ab 1 abcd 1 ab 1
F Fab = 16g2 tr  Fab Fcd
(2π)2 (2π)2
The last expression can be reduced to a boundary term, since

1
8
tr F[abFcd] = 16 ∂[a Bbcd]

in terms of the “Chern-Simons form”

Babc = tr( 12 A[a∂b Ac] + i 13 A[aAb Ac] )

Excercise IIIC6.3
Although the Chern-Simons form is not manifestly invariant, its variation is,
up to a total derivative:
a Show that its general variation is

δBabc = 1
2 tr[(δA[a)Fbc] − ∂[aAb δAc]]

b Show the gauge transformation of B is

δBabc = − 12 ∂[a λbc] , λab = 12 tr(λ∂[a Ab] )

If we assume boundary conditions such that F drops off rapidly at infinity, then
A must drop off to pure gauge at infinity:

iAm → g−1 ∂m g

Since instantons always deal with an SU(2) subgroup of the gauge group, we’ll assume
now for simplicity that the whole group is itself SU(2). Then the action can be given
a group theory interpretation directly, since the integral over the surface at infinity
is an integral over the 3-sphere, which covers the group space of SO(3), and thus half
the group space of SU(2). Explicitly,
I
S = 8π2 g2 d3 σm 16 mnpq Bnpq
1

I
→ 1
8π2 g 2
d3 σm 16 mnpq tr(g−1 ∂n g)(g−1 ∂pg)(g−1 ∂q g)
162 III. LOCAL

I
= 1
8π2 g 2
d3 x 16 ijk tr(g−1 ∂i g)(g−1∂j g)(g−1 ∂k g)
where in the last step we have switched to coordinates for the 3-sphere, using the fact
R 4 mnpq
dx fmnpq is independent of coordinate choice. In fact, in the case where g is
a one-to-one map between the 3-sphere and the group SO(3), this last expression is
just the definition of the invariant volume of the SO(3) group space. In that case, the
integral gives just the volume of the 3-sphere (2π 2 ). In general, the map g will cover
the SU(2) group space an integer number q of times, and thus cover the SO(3) group
space 2q times, so the result will be
|q|
S=
2g 2
where we have used the fact that self-dual solutions have q > 0 while anti-self-dual
have q < 0.
(Anti-)self-dual solutions give relative minima of the action with respect to more
general field configurations:
Z Z
0 ≤ tr 2 (Fab ± 2 abcd F ) = tr (F 2 ± 12 abcd FabFcd )
1 1 cd 2

|q|
⇒ S≥
2g 2
q is an integer, and thus can’t be changed by continuous variations: It is a topological
property of finite-action configurations. Thus the self-dual solutions give absolute
minima for a given topology. (All these solutions will be given implicitly by twistor
construction in the following subsection. Note that our normalization for the structure
constants of SU(2) differs from the usual, since we use effectively tr(Gi Gj ) = δij
instead of the more common tr(Gi Gj ) = 12 δij , which would normalize the structure
constants as in SO(3): fijk = ijk . The net effect is that our g 2 contains a relative
extra factor of 1/2, in addition to the effective extra factors coming from our different
normalization of the action.)
Excercise IIIC6.4
Explicitly evaluate the integral for the instanton number q for the solutions
of the ’t Hooft ansatz. Show that the asymptotic form can be expressed in
0
terms of (g)λ κ = xκλ . (det g 6= 1 because of the GL(1) piece.) Note that
there are boundary contributions not only at x = ∞ but also around the
singular points x = xi, which are of the same form but opposite sign. (Since
the singular parts of A are pure gauge, they cancel in F .)
At the quantum level, instantons are important mostly because they are an ex-
R
ample of fields that don’t fall off rapidly at infinity, and thus contribute to F F .
 C. YANG-MILLS 163

However, once the restriction on boundary conditions is relaxed, there can be many
such field configurations. The instantons are then distinguished by the fact that they
are the minimal action solutions for a given topology; this makes them important for
describing low-energy behavior.

7. ADHM
Much more general solutions of this form can be constructed using twistor meth-
ods. (In fact, they can be shown to be the most general self-dual solutions that fall
off fast enough at infinity in all directions.) The first step of the Atiyah-Drinfel’d-
Hitchin-Manin (ADHM) construction is to introduce a scalar square matrix in a larger
group space
0
UI I = (uI ι, vIiα ) (I 0 = (ι, iα))

The index α is the usual two-valued twistor index, for SU(2) in Euclidean space or
SL(2) in 2+2 dimensions. The other indices are

ι (H) I (G) i
SO(N) SO(N+4k) GL(2k)
SU(N) (SL(N)) SU(N+2k) (SL(N+2k)) GL(k,C)
USp(2N) (Sp(2N)) USp(2N+2k) (Sp(2N+2k)) GL(k)

The index ι is for the defining representation of the Yang-Mills group H, which is
any of the compact classical groups for Euclidean space, but is its real Wick rotation
for 2+2 dimensions. The index I is for the defining representation of the group G,
a larger version of H, where k is the instanton number. Finally, the index i is for a
general linear group. We also have the matrix

U I I 0 = (uI ι, v I i0 α )

For the SO and (U)Sp cases both U matrices are real, for the SU case they are complex
conjugates of each other, and for the SL case they are real and independent. We next
relate the two U’s by

uI ι uI κ = δικ , uI ι vIiα = v I i0 α uI ι = 0, v I i0 α vIiβ = Cβαgii0

so they are almost inverses of each other, except that the “metric” g is not constrained
to be a Kronecker δ. We then write the gauge field as a generalization of pure gauge:

iAAαικ = uI ι ∂Aα uI κ
164 III. LOCAL

(This is similar to the method used for nonlinear σ models of coset spaces G/H as
discussed in subsection IVA3 below, except for g.)
Self-duality then follows from requiring a certain coordinate dependence of the
U’s: This is fixed by giving the explicit dependence of the v’s as

vIiα = bIiA z A α , v I i0 α = bI i0 A z A α

where the b’s are constants. The orthonormality conditions on the U’s then implies
the constraint on the b’s
bI i0 (A bIiB) = 0

as well as determining the u’s in terms of the b’s (with much messier dependence than
the v’s), and thus A. Note that the z dependence of u can be written in terms of just
x, as follows from rewriting the uv orthogonality as (after multiplying by z)

uI ι bIiA y AB = uI ιbI i0 A y AB = 0

and noting scale invariance. Then the x components of A can also be written in
terms of just x. We then can check the self-duality condition by calculating f: The
orthonormality condition on the U’s can be written as

0
δIJ = uI ιuJ ι + vIi αg ii v J i0 α

0
where g ii is the inverse of gii0 . Then schematically we have

iF = ∂iA + iAiA
= (∂ ū)(∂u) − (∂ ū)uū(∂u)
= (∂ ū)vgv̄(∂u)
= ū(∂v)g(∂ v̄)u
= ūbg b̄u

or more explicitly
0
ifABικ = −(uI ιbIi(A )g ii (uJ κ bJ i0 B) )

where self-duality is FAα,Bβ = Cαβ fAB . We can also directly show z Aα fAB = 0.
 C. YANG-MILLS 165

8. Monopoles
Instantons are essentially 0-dimensional objects, localized near a point in 4-
dimensional spacetime (or many points for multi-instanton solutions). Another type
of solution is 1-dimensional; this represents a particle (with a 1D worldline). Unlike
the plane-wave solutions, which represent the massless particles already described
explicitly by fields in the action, we now look for time-independent solutions, which
describe massive (since they have a rest frame), bound-state particles.
Looking at time-independent solutions is similar to the dimensional reduction
that we considered in subsection IIB4 to introduce masses into free theories, only
(1) this mass vanishes, and (2) we reduce the time dimension, not a spatial one. In
our case, the dimensional reduction of a 4-vector (the Yang-Mills potential) gives a
3-vector and a scalar, both in the adjoint representation of the group. Let’s consider
the reduction in Euclidean space, so the scalar kinetic term comes out with the right
sign. Then the 4D Yang-Mills action reduces as
1 2
F
8 ab
→ 18 Fij2 + 14 [∇i, φ]2

where we have labeled the scalar A0 = φ and by dimensional reduction ∂0 → 0. Note

that this is the same action that would have been obtained by starting out with Yang-
Mills coupled to an adjoint scalar in four dimensions, either Minkowski or Euclidean,
and choosing the gauge A0 = 0. Thus, time-independent solutions to Euclidean
Yang-Mills theory are also time-independent solutions to Minkowskian Yang-Mills
coupled to an adjoint scalar (although not the most general, since the gauge A0 = 0
is not generally possible globally, especially when we assume time independence of
even gauge-dependent quantities). In particular, this means that time-independent
solutions to self-dual Yang-Mills are also solutions of Minkowskian Yang-Mills cou-
pled to an adjoint scalar. This allows us to use the first-order differential equations
and topological properties of self-dual Yang-Mills theory to find physical bound-state
particles in this vector-scalar theory.
Dimensionally reducing the (Euclidean) self-duality condition, we have

−[∇i, φ] = 12 ijk Fjk

As for instantons, the simplest solutions are for SU(2). As for the ’t Hooft ansatz,
we look for a solution that is covariant under the combined SU(2) of the gauge group
and 3D rotations: In SO(3) vector notation for both kinds of indices (using the SO(3)
normalization of the structure constants [iGi , iGj ] = ijk iGk ),

φi = xi ϕ(r), (Ai)j = ijk xk A(r)

166 III. LOCAL

(We know to use an  tensor in A because of covariance under parity.) The self-
duality equation then reduces to two nonlinear first-order differential equations (the
coefficients of δij and xi xj /r2 ):

−ϕ − r2 Aϕ = 2A + rA0 , −rϕ0 + r2 Aϕ = −rA0 + r2 A2

After some massaging, we find the change of variables

ϕ̃ = 1
r
+ rϕ, Ã = 1
r
+ rA

leads to the simplification

ϕ̃0 = −Ã2, Ã0 = −Ãϕ̃

ϕ̃ then can be eliminated, giving an equation for Ã. Making a final change of variables,

ψ = Ã−1 ⇒ ψψ 00 − (ψ 0)2 = −1

we can guess the solution (with regularity at r = 0)

 
−1 1 kr 1
ψ = k sinh(kr) ⇒ A = 2 −1 , ϕ= [kr coth(kr) − 1]
r sinh(kr) r2

Excercise IIIC8.1
Repeat this calculation in spinor notation:
0 ∼ Cαβ 0 . Show that we can then write
0
a In Euclidean space we can choose σαβ
the 4-vector potential for the monopole as

i(Aαβ )γδ = δαγ xβ δ A+(r) + δβδ xαγ A− (r)

which is symmetric in neither αβ nor γδ. (Compare the ’t Hooft ansatz in

subsection IIIC6.) However, xαβ is now symmetric from dropping x0.
b Impose self-duality, where
γ δ
∂αβ xγδ = 12 δ(α δβ) = δαγ δβδ − 12 Cαβ C γδ

from subtracting out the ∂0x0 piece. Derive the resulting equations for A±,
and show they agree with the above for

A± = − 12 (A ± ϕ)

In general, the Lagrangian of a Euclidean theory is the Hamiltonian of the

Minkowskian theory (with the sign conventions we introduced in subsection IIIA1),
 C. YANG-MILLS 167

since Wick rotation changes the sign of the kinetic energy and not the potential en-
ergy. In our case, this means the Minkowskian energy of the Yang-Mills + adjoint
scalar theory can be evaluated in terms of the same topological expression we used for
R
instantons: From the previous subsection, using S = dt E and ∂0 = 0, we evaluate
in Euclidean space
I
E = 16π2 g2 d2 σi i0jk B0jk , i0jk B0jk → −ijk tr(φFjk ) = 2 tr(φ[∇i, φ]) = ∂i tr(φ2 )
1

where we have used an integration by parts to simplify B. (Compare excercise

IIIC6.3a.) Since at spatial infinity
 
|k| 1 1
φi → xi − 2 , Aij → −ijk xk 2
r r r
H 2
and effectively d σi → 4πrxi , we find
|k|
E=
2πg 2
Also by similar arguments to those used for instantons, we see that any solutions with
boundary conditions A → 0, |φ| → |k| as r → ∞ have energy at least as great as
this. There is also a topological interpretation to this energy: Writing it as
I
E = − 16π2 g2 d2 σi ijk tr(hφiFjk )
1

we see that the energy is proportional to the magnetic flux, i.e., the “magnetic charge”
of the monopole. (The asymptotic value hφi of φ picks out a direction in isospace,
reducing SU(2) to U(1).) As in electromagnetism, magnetic charge is quantized in
terms of electric charge. However, for compact gauge groups, electric charge is also
quantized. (For the usual U(1), charges are arbitrary, but for SU(2), any component
of the isospin is quantized.) The energy is thus quantized in terms of k: It is a
multiple of the energy we found for the single monopole above.
Excercise IIIC8.2
Perform a singular gauge transformation that makes hφi point in a constant
(rather than radial) direction in isospin (SU(2)) space. Show that the isospin
component of the asymptotic form of A describes a U(1) magnetic monopole:
magnetic flux radiating outward from the origin.

REFERENCES
1. O. Klein, On the theory of charged fields, in New theories in physics, proc. Warsaw,
May 30 - June 3, 1938 (International Institute of Intellectual Co-operation, Paris, 1939)
p. 77;
168 III. LOCAL

W. Pauli, unpublished (1953):

early work on “Yang-Mills” theory.
2 C.N. Yang and R.L. Mills, Phys. Rev. 96 (1954) 191;
R. Shaw, The problem of particle types and other contributions to the theory of elemen-
tary particles, Cambridge University Ph.D. thesis (1955);
R. Utiyama, Phys. Rev. 101 (1956) 1597:
complete formulation of the (classical) theory.
3 L. O’Raifeartaigh, The dawning of gauge theory (Princeton University, 1997):
early history of Yang-Mills theory, including reprints and previously unpublished ma-
terial.
4 R.L. Arnowitt and S.I. Fickler, Phys. Rev. 127 (1962) 1821;
W. Kummer, Acta Phys. Austriaca 14 (1961) 149.
5 S. Weinberg, Phys. Rev. 150 (1966) 1313;
J.B. Kogut and D.E. Soper, Phys. Rev. D1 (1970) 2901:
lightcone field theory.
6 S. Coleman, Phys. Lett. 70B (1977) 59:
Yang-Mills plane waves.
7 R. Feynman and M. Gell-Mann, Phys. Rev. 109 (1958) 193;
L.M. Brown, Phys. Rev. 111 (1958) 957;
M. Tonin, Nuo. Cim. 14 (1959) 1108;
G. Chalmers and W. Siegel, hep-ph/9708251, Phys. Rev. D59 (1999) 045012;
M. Veltman, hep-th/9712216, Acta Phys. Polon. B 29 (1998) 783:
description of spin 1/2 with only undotted spinors.
8 A. Ashtekar, Phys. Rev. Lett. 57 (1986) 2244, Phys. Rev. D36 (1987) 1587;
T. Jacobson and L. Smolin, Phys. Lett. 196B (1987) 39, Class. Quant. Grav. 5 (1988)
583;
J. Samuel, Pramana 28 (1987) L429:
first-order actions with self-dual auxiliary fields.
9 C.N. Yang, Phys. Rev. Lett. 38 (1977) 1377:
reduction of self-dual Yang-Mills field to single component.
10 A.N. Leznov, Theor. Math. Phys. 73 (1988) 1233,
A.N. Leznov and M.A. Mukhtarov, J. Math. Phys. 28 (1987) 2574;
A. Parkes, hep-th/9203074, Phys. Lett. 286B (1992) 265:
lightcone gauge for self-dual Yang-Mills.
11 A.A Belavin, A.M. Polyakov, A.S. Shvarts, and Yu.S. Tyupkin, Phys. Lett. 59B (1975)
85:
instantons.
12 G. ’t Hooft, unpublished;
R. Jackiw, C. Nohl, and C. Rebbi, Phys. Rev. D15 (1977) 1642:
’t Hooft ansatz for multi-instanton solutions.
13 S.-S. Chern and J. Simons, Ann. Math. 99 (1974) 48.
14 M.F. Atiyah and R.S. Ward, Comm. Math. Phys. 55 (1977) 117;
M.F. Atiyah, V.G. Drinfel’d, N.J. Hitchin, and Yu.I. Manin, Phys. Lett. 65A (1978)
185;
E. Corrigan, D. Fairlie, P. Goddard, and S. Templeton, Nucl. Phys. B140 (1978) 31;
N.H. Christ, E.J. Weinberg, and N.K. Stanton, Phys. Rev. D18 (1978) 2013;
M.F. Atiyah, Geometry of Yang-Mills fields (Scuola Normale Superiore, Pisa, 1979);
 C. YANG-MILLS 169

V.E. Korepin and S.L. Shatashvili, Math. USSR Izvestiya 24 (1985) 307:
general multi-instanton construction, using twistors.
15 G. ’t Hooft, Nucl. Phys. B79 (1974) 276;
A.M. Polyakov, JETP Lett. 20 (1974) 194:
monopoles in nonabelian theories.
16 E.B. Bogomol’nyi, Sov. J. Nucl. Phys. 24 (1976) 449;
M.K. Prasad and C.M. Sommerfeld, Phys. Rev. Lett. 35 (1975) 760:
exact solution for monopole.
17 W. Nahm, Phys. Lett. 90B (1980) 413:
general monopole construction based on ADHM instanton construction.
170 IV. MIXED

IV. MIXED
In this chapter we consider ways in which gauge symmetry combines with global
symmetries for new effects. The interplay between global internal symmetries of scalar
and spinor theories and local symmetries of Yang-Mills is important for understanding
mass generation for all spins, and is fundamental for the Standard Model.

.................. ..................
.................. A. HIDDEN SYMMETRY ..................

Symmetries, especially local ones, are clearly very important in the formulation of
interactions. However, symmetries are not always apparent in nature: For example,
while most symmetries prefer massless particles, of all the observed particles the only
massless ones are the graviton, photon, and neutrinos. Furthermore, of the massive
ones, none with different properties have the same mass, although some are close (e.g.,
the proton and neutron). There are three solutions to this problem: (1) The symmetry
is not a property of nature, but only an approximate symmetry. Some terms in the
action are invariant under the symmetry, but other terms violate it. We can treat such
“explicit symmetry breaking” by first studying the symmetry for the invariant terms,
and then treating the breaking terms as a perturbation. (2) Although the laws of
physics are symmetric, nature is an asymmetric solution to them. In particular, such
a solution is the “vacuum”, or state of lowest energy, with respect to which all other
states are defined. This case is called “spontaneous symmetry breaking”. (3) The
particles in terms of which these laws are formulated are not those observed in nature.
For example, the hydrogen atom is most conveniently described in terms of a proton
and an electron, but in its low-energy physics only the atom itself is observed as a
separate entity: The U(1) symmetry related to charge is not seen from the neutral
atoms. The more extreme case where such particles always appear in bound states is
known as “confinement”.
Generally, such broken symmetries are at least partially restored at high energies.
For example, if the symmetry breaking introduces masses, or mass differences between
related particles, then the symmetry may become apparent at energies large with
respect to those masses. Similarly, a hydrogen atom excited to an energy much larger
than its lower energy levels will ionize to reveal its constituent particles.
It often is possible to change to a set of variables that are invariant under a
local symmetry. For example, if we can define everywhere a variable that transforms
as δφ(x) = λ(x), then it can be used to everywhere undo the invariance. We can
choose the “gauge” λ = −φ, transforming φ to 0 everywhere, leaving no residual
 A. HIDDEN SYMMETRY 171

invariance, or we can work with composite, invariant variables: E.g., ψ 0 = ψeiλ is

replaced (invertibly) with ψ̂ = ψe−iφ , so ψ̂ 0 = ψ̂.

1. Spontaneous breakdown
We first consider symmetry breaking by the vacuum, known as “spontaneous
breakdown”. The action is invariant under the symmetry, but the vacuum state is not:
Thus, the symmetry acting on the vacuum produces other zero-energy solutions to
the field equations, but this symmetry is not apparent when considering perturbation
about the vacuum. In this case, although the symmetry is broken, there are obvious
residual effects, particularly if the breaking can be considered as “small” with respect
to some other effects.
The “Goldstone theorem” is an important statement about the effect of symme-
try breakdown: If a continuous global symmetry is spontaneously broken, then there
is a corresponding massless scalar. The proof is simple: Consider a (relative) mini-
mum of the potential, as the vacuum. By definition, we have spontaneous symmetry
breaking if this minimum is not invariant under the continuous symmetry: i.e., ap-
plying infintesimal symmetry transformations gives a curve of nearby states, which
have the same energy, because the transformations are a symmetry of the theory. But
the mass of a scalar, by definition, is given by the quadratic term in its potential,
i.e., the second derivative of the potential evaluated at the vacuum value. (The first
derivative vanishes because the vacuum is a minimum.) So, if we look at the scalar
defined to parametrize this curve of constant energy in field space, its mass vanishes.
(This field may be a function of the given fields, such as an angle in field space.)
We can also formulate this more mathematically, for purposes of calculation:
Consider a theory with potential V (φi ). (The Lagrangian is V plus derivative terms.
For simplicity we consider just scalars.) The masses of the scalars are defined by
the quadratic term in the potential, expanding about a minimum, the vacuum. The
statement of symmetry of the potential means that

symmetry δφi = ζ i (φ) ⇒ 0 = δV = ζ i ∂i V for all φi

where we allow nonlinear symmetries, and ∂i = ∂/∂φi . Differentiating, and then

evaluating at this minimum,

h∂i V i = 0 at minimum φ = hφi

⇒ 0 = h∂j (ζ i ∂i V )i = h(∂j ζ i )(∂i V )i + hζ i ∂i ∂j V i = hζ i ih∂i ∂j V i

172 IV. MIXED

where here the vacuum value h i classically means to just evaluate at φ = hφi. (So
classically hABi = hAihBi.) Spontaneous symmetry breaking means the vacuum
breaks the symmetry: If this symmetry is broken, then hζ i i = 6 0, so it is a nontrivial
eigenvector of h∂i ∂j V i (the mass matrix) with vanishing eigenvalue. So, we can write
φi = hφi i + χhζ i i + ...
where χ is a massless field.
The simplest example is a single free, massless field, V = 0. Then ζ is simply a
constant. The simplest choice of vacuum is just hφi = 0, which breaks the symmetry:
L = 14 (∂φ)2, δφ = constant, hφi = 0
Then φ is a “Goldstone boson”.
The simplest nontrivial example, and a useful one, is a complex scalar with the
potential
V (φ) = 14 λ2 (|φ|2 − 12 m2)2
This is invariant under phase transformations δφ = iζφ. There is a continuous set of
√ √
minima at |φ| = m/ 2. We choose hφi = m/ 2; then the Goldstone theorem tells
us that the imaginary part of φ is the Goldstone field. Explicitly, separating the field
into its real and imaginary parts,
φ= √1 (m
2
+ ψ + iχ) ⇒ V = 14 λ2 m2 ψ 2 + 14 λ2 mψ(ψ 2 + χ2 ) + 1 2
16
λ (ψ 2 + χ2 )2
where hψi = hχi = 0. We could also use the nonlinear separation of the field into
√
magnitude and phase, φ = (m + ρ)eiθ / 2: Then θ drops out of the potential, and
its transformation (ρ is invariant) is the same as that of the free massless scalar. If
φ had been real, then only the discete symmetry φ ↔ −φ would have been broken,
and there would be no Goldstone boson.
Excercise IVA1.1
Write the complete action in terms of ρ and θ.
Note that this model would naively seem to have a tachyon (state with negative
(mass)2 ) if we had expanded about hφi = 0. However, since the vacuum is defined
always as a minimum in the potential (or the energy), the true states always have
nonnegative (mass)2 . This is the case for positive spins for similar reasons: We saw in
subsection IIB4 that free massive theories follow from massless ones by dimensional
reduction from one extra spatial dimension. If we had used an extra time dimension
instead, as required for the “wrong sign” for the mass term in p2 + m2 , there would
also be wrong signs for Lorentz indices, resulting in kinetic terms with arbitrarily
negative energy.
 A. HIDDEN SYMMETRY 173

2. Sigma models
The Goldstone mechanism thus produces massive particles as well as massless
ones, at least for polynomial potentials, to which we are restricted by quantum con-
siderations, to be discussed later. We now look for approximations to polynomial
scalar actions that eliminate the massive fields, but still take them into account
through their equations of motion, in the limit where their masses tend to infinity.
For example, in the above simple model, we can take the limit λ → ∞, which takes
the ψ mass (λm) to infinity. In this limit, the potential energy can remain finite only
if it vanishes: |φ|2 = 12 m2. (In quantum language, the potential’s contribution to the
R
path integral is just δ( V ) in that limit. Alternatively, we can neglect the kinetic
energy for |φ| in comparison to the mass or potential, and then eliminate |φ| through
its equation of motion in this approximation.) We can also enforce this limit directly
by using a Lagrangr multiplier field Λ:

L = 12 |∂φ|2 + Λ(|φ|2 − 12 m2)

m iθ
√
The solution to the constraint is φ = 2
e , and the action then describes just a free,
real scalar θ.
A less trivial example is a nonabelian generalization of this example: Consider
φ as a vector of an internal SO(n) symmetry. (The previous example was the case
SO(2).) The Lagrangian is then

L = 14 (∂φ)2 + 12 Λ(φ2 − m2)

The usual way to solve quadratic constraints without introducing square roots is to
use the identity

|(1 + ix)2|2 = (|1 + ix|2)2 ⇒ (2x)2 + (1 − x2)2 = (1 + x2)2

This is often used for trigonometric substitutions or simplifying integrals. For exam-
√
ple, when an integrand has a 1 − x2 , substituting x = sin θ eliminates the square
root at the price of requiring trigonometric identities, which in turn are usually solved
by making a second variable change to y = tan(θ/2). On the other hand, the above
identity suggests making instead the variable change x = 2y/(1 + y 2), which actually
gives the same result, more directly, as the previous two-step method. (This identity
can also be used for finding integer solutions to the Pythagorean theorem: A right
triangle with two shorter sides of integer lengths 2mn and m2 −n2 has the hypotenuse
m2 + n2 , where m, n are integers.)
174 IV. MIXED

We then can solve the constraint φ2 = m2 with the coordinates for the sphere in
terms of an SO(n−1) vector χ,
 
χ 1 − 14 χ2
φ=m ,
1 + 14 χ2 1 + 14 χ2

Then the kinetic term (now the whole action) becomes

(∂χ)2
1
(∂φ)2 = 14 m2
4
(1 + 14 χ2)2

Another way to obtain this result is to use the solution of subsection IA6 to the
constraint
0 = y 2 = (y a )2 − 2y + y − ⇒ y = e(xa, 1, 12 x2)

Then the desired constraint

(y a )2 + (y 1)2 = 1

follows from further constraining

√
2
1=y =0
e √12 (1 + 1 2
2x ) ⇒ e=
1 + 12 x2

2dx2
⇒ dy 2 = e2 dx2 =
(1 + 12 x2 )2
√
yielding the above result for x = χ/ 2. We thus have a nonpolynomial action, each
term having derivatives. The original SO(n) symmetry is nonlinearly realized on the
“angle” variables χ, and the vacuum (hχi = 0) spontaneously breaks the symmetry
to SO(n-1). The constant m acts as a dimensionful coupling, as seen by scaling
χ → χ/m to give the kinetic term the standard normalization.
A complex generalization of this model is described by the Lagrangian

L = 12 |∇φ|2 + Λ(|φ|2 − m2 )

where φ is now a complex n-component vector, ∇ is a U(1)-covariant derivative

(∇φ = (∂ + iA)φ), and Λ is a Lagrange multiplier enforcing that φ has magnitude
m. This model thus has a U(n) symmetry. Since A has no kinetic term (F 2), we can
eliminate it by its algebraic field equation:
↔
L → 12 |∂φ|2 + 1
8m2
(φ† ∂ φ)2 + Λ(|φ|2 − m2 )

where we have applied the constraint |φ|2 = m2 (or shifted Λ to cancel terms propor-
tional to |φ|2 − m2). Since the U(1) gauge was not fixed yet, we still have local U(1)
 A. HIDDEN SYMMETRY 175

invariance even without an explicit gauge field. We can use this invariance to fix the
phase of one component of φ, and use the constraint from Λ to fix its magnitude. In
terms of the remaining (n−1)-component complex vector χ,
 
1 − 14 |χ|2
χ
φ=m ,
1 + 14 |χ|2 1 + 14 |χ|2

↔
2 |∂χ| + 14 (χ† ∂ χ)2
2
⇒ L = 12 m
(1 + 14 |χ|2)2
(Alternatively, we can solve the constraint and fix the gauge first, then eliminate
A by its field equation.) This model is known as the CP(n−1) model (“complex
projective”).
Another example that will prove more relevant to physics is to generalize φ to an
n⊗n matrix: We then consider the Lagrangian

L = tr[ 12 (∂φ)† · (∂φ) + 12 λ2 (φ†φ − m2 I)2]

(where I is the identity matrix). Since φ† φ is hermitian and positive definite, the
minimum of the potential is at φ† φ = m2 I, and we can choose

hφi = mI

using the SU(n)⊗SU(n)⊗U(1) invariance

φ0 = UL φUR −1

(We can include the U(1) in either UL or UR .) The vacuum then spontaneously breaks
this invariance to SU(n):

hφ0 i = hφi ⇒ UL = UR

In the large-mass limit, we get the constraint

φ† φ = m 2 I ⇒ φ = mU, U † U = I, hUi = I, L → 12 m2tr[(∂U)†(∂U)]

so the field U itself is now unitary.

176 IV. MIXED

3. Coset space
The appearance of the scalar fields (Goldstone bosons) as group elements can
be generalized directly in terms of the effective theory, without reference to massive
fields. Such a theory should be considered as a low-energy approximation to some
unknown theory. Although the unknown theory may be better behaved at high
energies quantum mechanically (see later), the low-energy effective theory can be
determined from just (broken) symmetry. We therefore assume a symmetry group G
that is broken down to a subgroup H by the vacuum. (I.e., the vacuum is invariant
under the subgroup H, but not the full group G.) We are interested in only the
Goldstone bosons, associated with all the generators of the group G less those of
H. These fields are thus coordinates for the “coset space” G/H: They correspond to
elements of the group G, but elements related by the subgroup H are identified.
Explicitly, we first write the field g as an element of the group G, either by
choosing a matrix representation of the group (as in the U(N) example above), or
explicitly expanding over the group generators GI :

g = eiφ , φ = φI (x)GI

We then “factor” out the subgroup H by introducing a gauge invariance for that
subgroup:
g 0 = gh,
ι
h = eih (x)Hι
in terms of the H generators Hι , which are a subset of GI :

GI = (Hι , Ti )

where Ti are the remaining generators, corresponding to G/H. In particular, we can

choose
gauge φι = 0 ⇒ φ = φi Ti
However, G should still be a global invariance of the theory, though not of the vacuum.
We therefore assume the global transformation

g 0 = g0 g

where g0 is an element of the full group G, but is constant in x. The vacuum hgi = I
is then invariant under the global subgroup g0 = h−1 , where thus h is constant and
g0 ∈H (i.e., G is spontaneously broken to H).
We then attempt to construct a field strength invariant under the global group
as an element of the Lie algebra of G:

g −1 ∂ag = ∂a + iAιaHι + iFaiTi = ∇a + iFaiTi

 A. HIDDEN SYMMETRY 177

This can be evaluated in the φ parametrization as multiple commutators, as usual: A

and F are both nonpolynomial functions of φ, but with only one derivative. We have
absorbed A into a covariant derivative ∇ because of the remaining transformation
law under the local group H:

(∇ + iF )0 = h−1 (∇ + iF )h ⇒ ∇0 = h−1 ∇h, F 0 = h−1 F h

where we have assumed [Hι, Ti ] ∼ Tj . (In particular, this is true for compact groups,
where the structure constants are totally antisymmetric: Then fικi = 0 ⇒ fιiκ = 0.)
Then the action invariant under global and local transformations can be chosen as

L = 14 m2tr(F 2)

In particular, the real vector model we gave in the previous subsection describes the
coset space SO(n)/SO(n−1), the complex vector describes SU(n)/U(n−1), and the
matrix model describes U(n)⊗U(n)/U(n).
Excercise IVA3.1
Use the coset-space construction to derive the specific σ models explicitly
given in the previous subsection, as just identified.
We have described how nonpolynomial actions quadratic in derivatives can arise
as a low-energy approximation to polynomial theories. Further nonpolynomial terms
quartic in derivatives (but no more than quadratic in time derivatives) can be useful
for certain applications, but these arise from polynomial actions quadratic in deriva-
tives (which are preferred for quantum reasons) only by quantum effects.

4. Chiral symmetry
Later we’ll examine a description of the strongly interacting particles (“hadrons”)
in which they are all considered as composites (bound states) of fermionic “quarks”.
However, this theory is extremely difficult to solve, so we first consider treating the
hadrons as fundamental instead. Since there are probably an infinite number of kinds
of hadrons (or at least some integer power of 1040 , considering the (Planck mass)2 ),
this would require a formulation in terms of a “string” that treated all “mesons”
(bosonic hadrons) as a single entity. That possibility also will be considered later;
for now, we look at the simpler possibility of studying just the low-energy physics of
hadrons by using fields for just the lightest particles.
So far, the only observed scalar particles have been strongly interacting ones.
Some of the scalar mesons, especially the “pions”, are not only the lightest hadrons,
178 IV. MIXED

but can be considered close to massless on the hadronic scale. We therefore look for
a description of pions (and some close relatives) in the massless approximation; then
mass-generating corrections can be considered.
Normally, quantum corrections can affect masses. The only way to guarantee
masslessness at the quantum level is through some symmetry; we then can study this
symmetry already at the classical level. We have seen that (unbroken) gauge invari-
ance can require masslessness for all fields except the scalar and spinor. Masslessness
for a spinor can be enforced by “chiral symmetry”: If there is a U(1) symmetry for
all irreducible spinors ψα , then no mass terms (bilinears ψ1αψ2α) can be constructed.
(Generally, each spinor can have different U(1) charges, as long as no two charges add
to zero. Of course, this U(1) can be a subgroup of a larger chiral symmetry group.)
The only way a scalar can be guaranteed masslessness is if it is a Goldstone boson. We
therefore look for a description of pions as Goldstone bosons of some spontaneously
broken symmetry. (Supersymmetry is another possibility to enforce massless scalars,
but only if there are also massless fermions, which is not the case for hadrons.)
For simplicity, we consider the coupling of scalar mesons to quarks. We could
instead couple mesons to “baryons” (fermionic hadrons), thus treating only hadrons,
but the principles would be the same, only the indices would be messier. Combining
C invariance with chiral symmetry, and including a meson potential for spontaneous
symmetry breaking, we can write the action for just the quarks and scalar mesons as
Z
S = dx tr L
. .
L = [q †αL i∂ β α.qLβ + qRT β i∂β α q*Rα.] + [ 12 (∂φ)† · (∂φ) + 14 λ2 (φ† φ − 12 m2 I)2]
.
+Λ[qLαφqRα T
+ q*αR φ† q †Lα.]
where φ is an m⊗m matrix (m “flavors”), qL and qR are n⊗m matrices (n “colors”),
and Λ is the “Yukawa coupling”. Besides color symmetry (local if we had bothered
to write in the Yang-Mills fields for the “gluons”, by ∂ → ∇ on the quarks), we have
the (global) U(m)L⊗U(m)R chiral (flavor) symmetry

qL0 = qL UL , qR0 = qRU*R , φ0 = UL−1 φUR

including the (global) U(1) “baryon number” symmetry

UL = UR = eiθ ⇒ qL0 = eiθ qL , qR0 = e−iθ qR , φ0 = φ

If we think of baryon number as an SO(2) symmetry, then charge conjugation is

just the reflection that completes this to an O(2) symmetry (see excercise IIA1.2):

C: qL ↔ qR , φ → φT
 A. HIDDEN SYMMETRY 179

From this, and the usual CP

CP : qL → qL *, qR → qR *, φ → φ*

we find the parity symmetry

P : qL ↔ qR*, φ → φ†

(where for CP and P we also transform the coordinates as usual).

As before, the vacuum hφi = √m2 I breaks the flavor symmetry to the diagonal
subgroup UL = UR , which commutes with parity (and is therefore no longer “chiral”).
It also gives masses to the quarks (since chiral symmetry is broken); this is a general
feature of spinors coupled to scalars under spontaneous breakdown. In the limit
λ → ∞ (where the mass of all bosons but the Goldstones becomes infinite, but the
quark mass M = Λm is fixed), the Goldstone bosons are described by the unitary
matrix U, which transforms as U 0 = UL−1 UUR .
An interesting special case is m=1 (one flavor). The Goldstone boson of axial
U(1) can be identified with the π 0. In the limit λ → ∞, the Lagrangian becomes
(with a tr no longer needed)
. . .
L = (q †αL i∂ β α.qLβ + qRT β i∂β α q*Rα.) + 14 m2(∂π)2 + √M2 (eiπ qRT αqLα + e−iπ q †αL q*Rα.)

writing π for the neutral pion field. Λ = M/m is still the coupling of the pion to
the quarks, as can be seen by rescaling π → π/m to give the kinetic term the usual
normalization. (The coupling m is known as the “pion decay constant”, and is usually
denoted fπ . If we include leptons with the quarks, then this coupling also describes
the decay of the pion into two leptonic fermions.)
In this case, the (broken) axial U(1) transformations are

qL0 = eiθ qL , qR0 = eiθ qR, π 0 = π − 2θ

The corresponding axial current (determined, e.g., by coupling a gauge vector) is

. . . .
†β α
JA = (q L qL − qR q*R) − m ∂ π
αβ Tα β 2 αβ

This current is still conserved, since the field equations aren’t changed by the prop-
erties of the vacuum. The linear term is characteristic of expanding the Goldstone
field about the spontaneously broken vacuum; it corresponds to the fact that that
field has an inhomogeneous transformation under the broken symmetry.
However, in reality the pion is not exactly massless, so we should add to the
previous action a mass term for the pion, which explicitly violates the symmetry. (It
180 IV. MIXED

is then a “pseudogoldstone boson”.) In the general chiral symmetry model, where the
Goldstone bosons are described by a unitary matrix, a simple term that gives them
masses while preserving the polar (parity-preserving) diagonal symmetry UL = UR of
the vacuum is, for some constant ξ,
√
Lm = −ξ tr(φ + φ† − 2mI)

Since this explicitly breaks the axial U(m) symmetries, the corresponding currents
are no longer conserved. In the U(1) case, we can also add just a mass term

Lm = 14 ζπ 2 , ζ = m2m2π

(for some constant ζ), which is the leading contribution from the general term above.
The change in the field equation for π now violates the conservation law as

∂ · JA = −ζπ

This explicitly broken conservation law is known as “Partially Conserved Axial Cur-
rent” (PCAC).

5. Stückelberg
By definition, only gauge-invariant variables are observable. Although in general
a change of variables to gauge-invariant ones can be complicated and impractical,
there are certain theories where such a procedure can be implemented very simply
as part of the normal gauge-fixing. Not surprisingly, the only nonlinearity in these
redefinitions involves scalars.
The simplest cases of such redefinitions are free theories, and are thus contained
in our earlier discussion of general free, massive gauge theories. The simplest of these
is the massive vector. As described in subsection IIB4, the Lagrangian and gauge
invariance are
L = 18 F 2 + 14 (mA + ∂φ)2
δA = −∂λ, δφ = mλ
where Fab is the Abelian field strength. Note that the scalar is pure gauge: It is
called a “compensator” for this gauge invariance. Since it has a nonderivative gauge
transformation, it can easily be gauged to zero at each point, by just choosing λ =
−φ/m. This means that without loss of generality we can consider the theory in
terms of just the gauge-invariant field

A0 = A + 1
m
∂φ
 A. HIDDEN SYMMETRY 181

This “composite” field can also be considered as a field redefintion or gauge transfor-
mation on A. The lagrangian simplifies to
L = 18 F 02 + 14 m2A02
Later we’ll see that it is often more useful to keep φ as an independent field.
Excercise IVA5.1
Choose the gauge A0 = φ. Show that φ then can be eliminated by its equa-
tion of motion, leaving only the transverse 3-vector Ai , with Lagrangian
− 14 Ai ( − m2)Ai . Show the relation to the lightcone gauge of subsection
IIIC2, using the dimensional reduction langauge of subsection IIB4.
The original Lagrangian can also be considered an unusual coupling of a massless
vector to a massless scalar: Remember that the massless scalar is the simplest example
of a Goldstone boson, with the spontaneously broken global symmetry
δφ = T φ, Tφ = 1
where we have defined the symmetry generator T to act inhomogeneously on φ. We
then couple the “photon” to this charge: After a trivial rescaling of the gauge field,
L= 1
8m2
F2 + 14 (∇φ)2, ∇ = ∂ + AT
where m is the “charge” with which A couples to φ, which in this case happens to
have dimensions of mass. The electromagnetic current in this case is simply J = 12 ∇φ,
whose conservation is the scalar field equation φ = 0 (with gauge-covariantized ).
Because the spontaneously broken symmetry of the corresponding Goldstone
model is now gauged, expanding about hφi = 0 is no longer a physical statement
about the vacuum, since φ is no longer gauge invariant. (As we saw, we can even
choose φ = 0 as a gauge condition.) Therefore, from now on, when we make a
statement such as “hφi = 0” in such a case, it will be understood to refer to choos-
ing φ = 0 as the value about which to perform perturbation expansions (e.g., for
separating actions into kinetic terms and interactions).
Note that the Stückelberg action can be generated starting from the action with
just A0 , and performing a gauge transformation that is not an invariance:
A0 → A0 + 1
m
∂φ
Dropping the prime from A, this transformation is just the inverse of the one we used
to eliminate the scalar. If we start from an action that has also a coupling of A0 to
matter, we see that conserved currents decouple from φ:
Z Z Z
0
A · J → A · J − m φ∂ · J
1
182 IV. MIXED

More precisely, if the only term in the action for vector + matter that is not gauge
invariant is the vector mass term ( 14 m2 A02), then the above gauge transformation
affects only that term.

6. Higgs
We have seen that spontaneous symmetry breakdown can generate masses for
spinors. We also saw how a massless vector could become massive by “eating” a
would-be Goldstone scalar, in the simplest case of a scalar without self-interactions.
We’ll now examine more interesting models: Yang-Mills theories, which describe self-
interacting vectors, must couple to self-interacting scalars to become massive.
We can expect, by considering the linearization of any Yang-Mills theory coupled
to scalars, that we will need more scalars than massive vectors, since each vector needs
to eat a scalar to become massive, and some scalars will become massive and thus
uneaten. (Only would-be Goldstone bosons can be eaten, as seen by linearization to
the Stückelberg model.) For the simple (and most useful) example of U(n) for the
gauge group, an obvious choice for the scalar “Higgs” field is an n⊗n matrix. (SU(n)
can be treated as a slight modification.) The simplest such model is the one studied
in subsection IVA2: We now consider one of the SU(n) symmetries (together with
the U(1)) as the local “color” symmetry to which the Yang-Mills fields couple, and
the other SU(n) as the global “flavor” symmetry (where we use the names “color”
and “flavor” to distinguish local and global symmetries, not necessarily related to
chromodynamics).
The Lagrangian for this “Gervais-Neveu model” is then

L = tr[ 8g12 F 2 + 12 (∇φ)† · (∇φ) + 14 λ2 (φ†φ − 12 m2I)2]

where ∇ = ∂ + iA, and Aa and φ are n⊗n matrices (but Aa are hermitian). Now φ†φ
is gauge invariant (although not invariant under the flavor group), so we still have

hφ† φi = 12 m2I

as a gauge-invariant statement (but hφi = √m2 I, or hφφ†i = 12 m2I, still makes sense
only for purposes of gauge-dependent perturbation expansions).
√
Since any complex matrix can be written as φ = UH/ 2, where U is unitary and
H is hermitian, we can choose the “unitary gauge” U = I (i.e., φ = φ†). As for the
Stückelberg case, this is equivalent to working in terms of the gauge-invariant fields
(defined by using this U as a gauge transformation)

A0 = U −1 (−i∂ + A)U, φ0 = √1 H
2
= U −1 φ
 A. HIDDEN SYMMETRY 183

where U can be defined by

p √
√1 H
2
= φ† φ, U = φ 2H −1

This is well-defined as long as H is invertible, which is true for small perturbations

about its vacuum value
hHi = mI

If the perturbation is so large that H has vanishing eigenvalues, then this is equivalent
to looking at states so far away from the vacuum that some of the broken symmetry
is restored. Expanding about the vacuum (H → mI + H), the Lagrangian is now

L = tr[ 8g12 F 02 + 14 m2 A02 + 14 (∂H)2 + 14 λ2 m2H 2

↔
+A0 · 14 (Hi ∂ H) + 12 mA02H + 14 λ2 mH 3 + 14 A02H 2 + 1 2 4
16
λH ]

Thus all particles are now massive. As for the Goldstone case, we can take the limit
λ → ∞ to get rid of all the massive scalars, which in this case leaves just the massive
vectors, adding only the mass term to the original Yang-Mills action. This was clear
from the nonlinear σ model that resulted from that limit, by coupling that field (U)
to Yang-Mills directly.
Excercise IVA6.1
Find the chiral action for this model of the type described in subsection IIIC4,
where the massive vectors are described by self-dual tensors instead of vectors.
Excercise IVA6.2
Consider again this model, for the case n=2. We modify this example by
dropping the U(1) gauge field, so we have just SU(2). Since SU(2) is pseu-
doreal, we can further restrict the Higgs field to satisfy the reality condition
φ* = CφC. Thus, both color and flavor groups are SU(2), and φ is the usual
matrix representation of the 4-vector of SO(4)=SU(2)⊗SU(2) (see subsection
IIA5). Repeat the analysis given above.
Excercise IVA6.3
Consider again the gauge group SU(2), but now take the Higgs field in the
adjoint representation, with no flavor group (i.e., a real 3-vector). Show that
only 2 of the 3 vectors get mass, leaving a residual U(1) gauge invariance.
Explain this in terms of the gauge transformations of the 3-vector. (Hint:
think 3D rotations.)
184 IV. MIXED

REFERENCES
1 Y. Nambu, Phys. Rev. Lett. 4 (1960) 380;
Y. Nambu and G. Jona-Lasinio, Phys. Rev. 122 (1961) 345, 124 (1961) 246:
introduced into relativistic physics massless bosons associated with a broken symmetry.
2 J. Goldstone, Nuo. Cim. 19 (1961) 154;
J. Goldstone, A. Salam, and S. Weinberg, Phys. Rev. 127 (1962) 965:
found the theorem relating the two.
3 M. Gell-Mann and M. Lévy, Nuo. Cim. 16 (1960) 705:
σ -models.
4 S. Weinberg, Phys. Lett. 18 (1967) 188:
nonlinear σ -models.
5 H. Eichenherr, Nucl. Phys. B146 (1978) 215;
V.L. Golo and A.M. Perelomov, Lett. Math. Phys. 2 (1978) 477, Phys. Lett. 79B (1978)
112;
E. Cremmer and J. Scherk, Phys. Lett. 74B (1978) 341:
classical CP(n).
6 T.H.R. Skyrme, Proc. Roy. Soc. A260 (1961) 227;
J. Wess and B. Zumino, Phys. Lett. 37B (1971) 95;
E. Witten, Nucl. Phys. B223 (1983) 422, 433:
higher-derivative terms.
7 H. Yukawa, Proc. Phys.-Math. Soc. Japan 17 (1935) 48.
8 R. Marshak and E.C.G. Sudarshan, Proc. Padua-Venice conference on mesons and
recently discovered particles, September, 1957; Phys. Rev. 109 (1958) 1860;
Feynman and Gell-Mann, loc. cit. (IIIC):
chiral symmetry in weak interactions (for coupling to vector currents).
9 Gell-Mann and Levy, loc. cit.;
Nambu, loc. cit.;
K.-C. Chou, Sov. Phys. JETP 12 (1961) 492:
PCAC.
10 Stückelberg, loc. cit. (IIB).
11 P.W. Anderson, Phys. Rev. 112 (1958) 1900, 130 (1963) 439:
nonrelativistic “Higgs” effect in condensed matter theory.
12 F. Englert and R. Brout, Phys. Rev. Lett. 13 (1964) 321;
P.W. Higgs, Phys. Lett. 12 (1964) 132;
G.S. Guralnik, C.R. Hagen, and T.W.B. Kibble, Phys. Rev. Lett. 13 (1964) 585;
A.A. Migdal and A.M. Polyakov, Sov. Phys. JETP 24 (1967) 91:
“Higgs” effect.
13 J.L. Gervais and A. Neveu, Nucl. Phys. B46 (1972) 381.
 B. STANDARD MODEL 185

................... ...................
................... B. STANDARD MODEL ...................

In this section we discuss the “Standard Model”, the minimal theory that de-
scribes all the observed particles and forces (except gravity). We also consider some
features of “Grand Unified Theories” (GUTs), an extension of the Standard Model
that uses fewer multiplets.

1. Chromodynamics

One way in which particles naively described by the action can be hidden from
observation is if the force binding them is too strong to allow them to exist freely.
Such a condition is often called “infrared slavery” since this alleged property of the
force is a long-range feature, preventing the constituent particles from escaping to
infinity. This “confinement” is not a classical phenomenon, and its occurence even
at the quantum level has not yet been proven. Therefore, in this section we’ll simply
assume confinement, and describe the resultant symmetry properties, leaving the
dynamical properties for later chapters.
The assumption of “color” confinement is that the color forces are so strong that
they bind any objects of color to other such objects; thus, only “colorless” states, those
that are singlets under the color gauge group, can exist freely. Composite fields that
are invariant under the local group transformations can be obtained by multiplying
matter fields or Yang-Mills field strengths, perhaps using also covariant derivatives,
and contracting all color indices. The color gauge group is generally assumed to be
SU(n): usually SU(3), but sometimes larger n for purposes of perturbation in 1/n.
Larger n is also used for unification, but in that case the Higgs mechanism is used to
reduce the group of the massless vectors to SU(3) (times Abelian factors).
Another feature of these confined states, to be considered later, is their geomet-
rical structure. The observed spectrum and scattering amplitudes of the “hadrons”
(strongly interacting particles) indicates a stringlike identification of at least the ex-
cited states. (The ground states may behave more like “bags”.) This picture also
fits in with confinement, since the spatial separation of the quarks and antiquarks in
excited states would force the gluons that convey their interactions (and self-interact)
to confine themselves as much as possible by collapsing into “strings” connecting the
quarks. Thus, we describe a meson with an “open string”, with a quark at one end
and an antiquark at the other. Similarly, an excited glueball would no longer be a
ball, but rather a “closed string”, forming a closed loop.
186 IV. MIXED

We will need to reconsider also the discrete symmetries, C, P, and T, and their
combinations. First of all, we note the “CPT theorem”: All local, hermitian, Poincaré
invariant actions are CPT invariant. This is easy to see from the fact that CPT only
changes the overall sign of the coordinates, which is effectively the same as changing
the sign of each derivative, as well as giving a −1 for each vector index on a field. Since
CPT also gives signs to dotted spinors and not undotted ones, we also get −1’s for
.
vector combinations of indices on spinors (ψ α ψ̄ α; signs cancel when contracting spinor
indices on pairs of dotted or undotted spinors). Thus, all these signs cancel because
Poincaré invariance requires an even number of vector indices (in even numbers of
dimensions, from contracting with ηab and abcd ). Alternatively, and even more simply,
in D=4 we can attribute it to having even numbers both of undotted spinor indices
and of dotted spinor indices, since we can define CPT by associating a sign with each
dotted index (including those that appear as part of a vector index). Consequently,
from now on we ignore T and consider only C, P, and CP.
Although we have considered C (and thus CP) in the context of electromagnetism,
nonabelian gauge fields require some (simple) generalization, since they carry charge
themselves. We start with the general coupling of massless fermions to nonabelian
gauge fields:
.
†β
L = ψ (−i∂ . + A .)ψ α
αβ αβ

where ψ is a column vector with respect to the gauge group, and A a hermitian
matrix. The CP transformation of the fermions then determines that of the vectors,
needed for invariance:
. . .
CP : ψ 0α = ψ*α., ψ 0*α = −ψα, ∂ 0 . = −∂ β α, A0 . = AT β α
αβ αβ

.
(remember (ψ α )* ≡ ψ̄ α, but (ψα)* = −ψ̄α. because of the factor of Cαβ ), where we
have chosen to represent parity on the coordinates as acting on the explicit ∂ rather
than on the arguments of the fields. The transformation on the vector is thus parity
on the vector index, combined with charge conjugation A0a = −ATa = −Aa*: The
minus sign can be associated with change in sign of the coupling (as for the Abelian
case), while the complex conjugation takes into account the charge of the vector
fields themselves. (As discussed in subsection IB2, G → −G* is an invariance of the
algebra, where g → g* and g = eiG.) Although these terms, as well as the F 2 term
for the vectors, are CP invariant, this invariance can be broken by coupling to scalars:
The Yukawa coupling
LY = ψ T αφψα + h.c.
 B. STANDARD MODEL 187

for some matrix φ of scalars, would require the CP transformation

φ0 = φ*

(up to perhaps some unitary transformation), but unlike the vectors there is no guar-
antee that under complex conjugation the matrix φ = φi Mi for real scalars φi and
constant matrix (Yukawa couplings) Mi will preserve this form, i.e., satisfy

φ0i Mi = φi Mi *

since the matrices Mi can be complex.

The basic assumption of “chromodynamics”, or in the quantized version “quan-
tum chromodynamics (QCD)”, is that we have a nonabelian gauge theory without
fundamental scalars that couple directly (but scalars will show up when we intro-
duce electroweak interactions). Namely, we assume only Yang-Mills for the “color”
gauge group SU(n), specifically n=3, with the usual action, minimally coupled to
spin-1/2 “quarks” in the defining representation of the color group, which may have
masses. (These masses are actually generated by weakly interacting Higgs bosons,
whose coupling we consider in the next subsection; for now we include just the re-
sulting mass terms.) Such an action is automatically invariant under CP and T.
We furthermore assume invariance under charge conjugation: Just as an irreducible
real scalar describes particles that are their own antiparticles, and needs doubling (or
complexification) to define charge, an irreducible (massive) spinor cannot describe
distinguishable particle and antiparticle. Therefore, for every quark field qLα (“L”
for “left”) we have an “antiquark” field qRα (“R” for “right”), and they transform
into each other under charge conjugation, just as a scalar transforms into its complex
conjugate. (A spinor can’t transform into its complex conjugate under C, since C
commutes with spacetime symmetries, like Lorentz transformations.) Besides this
doubling, and the n colors of the quarks, we also assume a further multiplicity of m
different “flavors” of quarks. Gauge invariance requires the quark masses be indepen-
dent of color, and C invariance requires the mass terms couple qL to qR, but these
terms violate an otherwise global U(m)⊗U(m) flavor symmetry.
The action is then of the form

tr[ 18 F 2 + (q †L i∇qL + q †R i∇qR) + ( √

M T
q q + h.c.)]
2 R L

where we have written qL and qR as matrices with respect to SU(n) color (Uc ) and
U(m)⊗U(m) flavor (Uf L and Uf R ) such that they transform as

qL0 = Uc qL Uf L , qR0 = Uc *qR Uf R*

188 IV. MIXED

and thus the covariant derivatives can be written as

∇aqL = (∂a + iAa)qL , ∇a qR = (∂a − iAa*)qR

where Aa are hermitian, traceless, n×n matrices. (By definition, charge conjuga-
tion takes a representation of an internal symmetry into the complex-conjugate one.)
Charge conjugation is then

C: qLα ↔ qRα, Aa → −Aa*

while parity is
P : α
qL,R → q̄L,Rα., Aa → −Aa

While the color symmetry is a local symmetry, the flavor symmetry is broken,
inducing the transformation on the mass matrix

M 0 = Uf L MUf−1
R

This transformation allows the mass matrix M to be chosen real and diagonal: Any
matrix can be written as a hermitian one times a unitary one. A Uf R transformation,
as a field redefinition, then can be made to cancel the unitary factor in M; then
a unitary transformation Uf L = Uf R can be made to diagonalize M (while keeping
it hermitian). These diagonal elements are then simply the masses of the m dif-
ferent quark flavors. The most symmetric case is M = 0, which leaves the entire
U(m)⊗U(m) chiral symmetry unbroken. (See subsection IVA4.) The least symmet-
ric case is where all the masses are nonzero and unequal, leaving as unbroken only
the subgroup U(1)m , with Uf L = Uf R . (In general, Uf L = Uf R if all masses are
nonvanishing.)
The minimal form of this action, besides making CP and T automatic, also au-
tomatically extends the discrete symmetry C to an O(2) symmetry, whose “parity”
transformation is C and whose continuous SO(2)=U(1) symmetry is the U(1) part
of the U(m) flavor symmetry, which is not broken by the mass term. It corresponds
to a charge called “baryon number”: Up to an overall normalization factor, it simply
counts the number of quarks qLα , q̄Rα. (which form a Dirac spinor) minus the number
of “antiquarks” qRα, q̄Lα.. However, such an O(2) symmetry can be defined separately
for each flavor, since (after M has been diagonalized) the action can be written as
a sum of independent terms for each flavor. In particular, each flavor has its own
separately conserved quark number. These flavor conservation laws, at the classical
level, are broken only by the weak interactions, which we have not included in the
above action. (Gravity and electromagnetism do not violate them.)
 B. STANDARD MODEL 189

Since confinement is a quantum effect, the details of hadronic scattering cannot be

discussed within classical field theory. However, we saw that low-energy properties of
mesons (and similarly for baryons) could be described by effective Lagrangians. The
fact that hadrons are made of quarks can be used to obtain a bit more information
even at the classical level, especially if the relevant quarks are heavy. (Heavy with
respect to what is unfortunately also a question that can be answered only at the
quantum level.) For example, in a nonrelativistic approximation, low-energy proper-
ties of hadrons can be found from just the quantum numbers, spin-spin interactions,
and masses of the quarks, while their velocities are ignored, and the gluons are ne-
glected altogether. In such an approximation, reasonably accurate predictions are
made for the masses and magnetic moments of the ground-state hadrons.
Actually, the claim that color nonsinglet states can never be observed needs a
bit of stipulation: There may be a “quark-gluon plasma” phase of hadronic matter
that can exist only at extremely high temperatures or pressures. Thus, a hypothet-
ical observer during the first moments of the universe might observe “free” quarks
and gluons. Similarly, a small enough observer, living inside an individual hadron,
might see individual quarks and gluons, since the size of his equipment would be
much smaller than what we consider “asymptotic” distances. Conversely, we could
consider the possibility of a new chromodynamic force, other than the one respon-
sible for the hadrons of which we are composed, that has a confinement scale that
is astronomical (extremely low energy), so that earthly laboratories would fit inside
the new “hadrons”. Thus, any statement about the observability of color must be a
dynamical one, and does not follow as an automatic consequence of the appearance of
a nonabelian group: Just as for the Higgs effect, confinement can be repealed under
appropriate circumstances, and the observability of color depends on the details of
the dynamics, and in particular on the values of the various parameters (momenta
and couplings).

2. Electroweak

The weak and electromagnetic interactions are mediated by observed spin-1 par-
ticles, some of which have charge and mass. Specifically (see subsection IC4), the
massive vectors form a triplet (W +, W −, Z), while there is only one massless vector
(the photon). This suggests a gauge group of SU(2)⊗U(1), with a Higgs effect that
leaves only U(1) unbroken. From the table of known fundamental fermions, we can see
that they fall into doublets and singlets of this SU(2), with the U(1) charge being that
190 IV. MIXED

of electromagnetism. (This SU(2)⊗U(1) unification of the weak and electromagnetic

interactions is called the “Glashow-Salam-Weinberg” model.)
We saw in subsection IVA4 a very simple model of spontaneously broken chiral
U(m)⊗U(m) symmetry where masses were generated for quarks. In subsection IVA6
we saw how the same scalars could generate masses for vectors, by coupling to one
of the U(m)’s. We now combine those two models, specializing to the case m=2,
but with two slight modifications: (1) Since the defining representation of SU(2) is
pseudoreal, we can impose a reality condition on the Higgs field, which is in the ( 12 , 12 )
representation of SU(2)⊗SU(2):
φ* = CφC
This makes it a vector of SO(4)=SU(2)⊗SU(2) (See excercise IVA6.2.) It’s also the
reality condition satisfied by an element of (the defining representation of) SU(2). (See
subsection IIA2.) This is not surprising, since the group product U 0 = UL UUR allows
the interpretation of a group element itself as a representation of chiral symmetry.
This is the situation described in subsection IVA2 (φ → U in the large-mass limit),
but in this case φ† φ is automatically proportional to the identity (it gives the square
of the 4-vector), so in general an SO(4) 4-vector can be written as the product of a
scalar with an SU(2) element. This reality condition breaks the chiral U(1)⊗U(1) to
the diagonal U(1) that leaves the Higgs invariant.
(2) The gauged SU(2) is still one of the two chiral SU(2)’s, but the gauged U(1)
must now be a subgroup of the other SU(2), since the Higgs is now invariant under
the usual U(1)’s. Thus, the ungauged SU(2) is explictly broken, and this accounts
for the mass splittings in the doublets of known fundamental fermions. Remember
that observables are singlets of gauged nonabelian groups (except perhaps for Abelian
subgroups), so any observed internal SU(2) must be a global symmetry, even when
it’s broken. As described in subsection IVA6, these singlets can be constructed as
composite fields resulting from the gauge transformation obtained from the SU(2)
part of φ.
Using the electromagnetic charges of the various particles, we thus determine
their SU(2)⊗U(1) representations: For spin 1, we have W=(1,0) and V=(0,0), where
the first entry is the “isospin” and the second is the U(1) charge. For spin 0, we
have φ=( 12 ,± 12 ), choosing the U(1) generator as the diagonal one from UR . Finally,
for spin 1/2, we have for the leptons lL=( 12 ,− 12 ), which combines with φ to pro-
duce (0,0)⊕(0,−1), and lR=(0,1). Similarly, for the quarks we have qL =( 12 , 16 ), and
qR=(0,− 16 ± 12 ). The Lagrangian is then

L = L1 + L0 + L1/2
 B. STANDARD MODEL 191

L1 = 1
8g 02
F 2(V )+ 1
8g 2
tr F 2(W )

L0 = tr[ 14 (∇φ)†(∇φ) + 14 λ2 (φ†φ − 12 m2)2 ]

    
† Λ+ 0 T T 1
L1/2 = tr(ψ i∇ψ) + tr q qL φ + ΛlRlL φ + h.c.
0 Λ− R 0
where the fermions ψ = (qL , qR, lL , lR), and the SU(2)⊗U(1) covariant derivative acts
as


∇φ = ∂φ + iW φ − i 12 V φ 10 −10

∇qL = ∂qL − iqLW + i 16 V qL



∇qR = ∂qR + i 12 V qR − 13 I + 10 −1
0

∇lL = ∂lL − ilL W − i 12 V lL

∇lR = ∂lR + iV lR

For simplicity we have ignored the indices for color (and its gauge fields, treated in
the previous section), families (treated in the following subsection), and spin. We
have also used matrix notation with respect to the local SU(2) (gauged by W ) and
the global SU(2) (explicitly broken in L1/2 by the gauging of a U(1) subgroup, the
Yukawa couplings, and the chirality of the massless neutrinos): Thus W is a traceless
hermitian 2×2 matrix, φ is also 2×2 but satisfying the “reality” condition given above
(traceless antihermitian plus real trace), qL , qR, and lL are complex 2×2 matrices,
and lR is a 2-component column. (By definition, the diagonal parts of W and φ are
electromagnetically neutral.) The quark Yukawa coupling is diagonal in the broken
SU(2) to preserve the local U(1) symmetry. (The tr here is trivial for the lepton
Yukawa term, but we have left it for generalization to more than one family.)
In the unitary gauge for the local SU(2),

φ= √1 ϕI,
2
hϕi = m

where ϕ is a single real scalar, the simplifications to the Lagrangian are

2
L0 → 14 (∂ϕ)2 + 18 ϕ2 tr{[W − 12 V 10 −1
0
] } + 18 λ2 (ϕ2 − m2 )2
    
† Λ+ 0 1
L1/2 → tr(ψ i∇ψ) + √2 ϕ tr
1 T T
q qL + ΛlRlL + h.c.
0 Λ− R 0
We then can expand ϕ about its vacuum value m: The lowest order terms give masses
for most of the vectors and fermions: The massless fermions are the neutrinos, while
the massless vector gauging the unbroken U(1) (a combination of the original U(1)
with a U(1) subgroup of the SU(2)) is the photon (of electromagnetic fame). The
192 IV. MIXED

mass of the remaining vectors accounts for the weakness and short range of the “weak”
interactions.
Excercise IVB2.1
Diagonalize this Lagrangian with respect to the mass eigenstates. For conve-
nience, normalize

g= √1 g0
2
cos θW , g 0 = g0 sin θW

where θW is the “weak mixing (Weinberg) angle”.

a Find explicitly the masses for all the particles in the Standard Model (first
family for fermions) in terms of the couplings m, λ, g0 , θW , Λ±, Λ. Show from
the experimental values for the vector masses given in subsection IC4 that
sin2θW ≈ .223.
b Find all the other couplings of the mass eigenstates. Show that, with the
conventional electric charge assignments,
1 1 1
2
= 2 + 02
e 2g g
(Hint: Rather than rescaling the vectors, note that the generated mass term,
and the given couplings of V and W , suggest defining

√
W 0 = W − 12 V 10 −1
0
⇒ V = γ + k1 Z, 2W 0 = γ + k2 Z

in the conventions of subsection IIA1, for the new fields Z and photon γ, for
appropriate constants ki .)
Note that, unlike the strong (chromodynamic) or purely electromagnetic (or even
gravitational) interactions, the weak interactions break every discrete spacetime sym-
metry possible. (The others break none. CP violation will be discussed in the
following subsection. Of course, CPT is always preserved.) Sometimes this is at-
tributed to the presence of a chiral symmetry, used to reduce 4-component spinors to
2-component; however, we have already seen that in general chiral and parity symme-
tries are unrelated. (You can have either without the other. This fact will be further
discussed in subsections IVB4 and VIIIB3.) A better explanation is to attribute
P and C to doubling, which converts spinors from 2-component to 4-component:
2-component spinors are the simplest description of helicity/spin 12 ; 4-component
spinors are useful only to manifest a larger symmetry, when it exists. The weak inter-
actions violate parity because the neutrino is not doubled, and because the fermions
that are doubled no longer have a symmetry relating their two halves.
 B. STANDARD MODEL 193

3. Families
In the Standard Model (and its simpler generalizations) there is no explanation
for the existence of more than one family of fermions. However, the existence of 3
families does have interesting consequences. Most of these follow from the form of the
Yukawa couplings, and thus the fermion masses. In subsection IVB1 we considered
redefinitions of the fermion fields as unitary flavor transformations. These allowed us
to obtain the simplest form of the mass matrices, since they were not flavor singlets,
and thus transformed. We now perform similar transformations, but only on the
family indices, since transformations that don’t commute with the gauge symmetries
would complicate the other terms in the action. Now ignoring spin, color, and local
flavor indices, and using matrix notation for the family indices, the fermions transform
as
qL0 = qL UqL , qR±
0
= qR±UqR± *, lL0 = lL UlL , lR0
= lR UlR *
where qL , qR±, lL, and lR have m components for the m families. qR± are the 2
components of the (explicitly broken) global flavor doublet qR. We thus have 5 U(m)
symmetries, all broken by the Yukawa couplings: These field redefinitions induce
transformations on them,

Λ0± = UqL Λ±UqR±

−1
, Λ0 = UlL ΛUlR
−1

As in subsection IVB1, UqR± and UlR can be used to make Λ± and Λ hermitian.
Then UlL can be used to make Λ diagonal, also as in subsection IVB1, leaving a U(1)m
symmetry UlL = UlR , corresponding to separate conservation laws for electron number
(including its neutrino), muon number, and tauon number for the 3 known flavors.
However, the quark sector works a bit differently: We can use UqL to diagonalize Λ+
or Λ−, but not both. This leaves another U(1)m symmetry UqL = UqR+ = UqR−. If Λ+
has been diagonalized, then 1 of the m U(1)’s, corresponding to total quark number
(baryon number) conservation, leaves Λ− invariant, while the remaining m−1 U(1)’s
can be used to eliminate some of the phases of the complex off-diagonal components
of Λ−.
The remaining global flavor symmetries are thus m lepton U(1)’s and 1 quark
U(1). The remaining Yukawa couplings are the real, diagonal Λ, describing the m
masses of the massive leptons (the neutrinos remain massless), the real, diagonal
Λ+, giving the m masses of half of the quarks, and the hermitian Λ−, consisting
of m diagonal components, describing the masses of the other quarks, m(m−1)/2
magnitudes of the off-diagonal components, and (m−1)(m−2)/2 phases of the off-
diagonal components. These phases violate CP invariance: CP, besides its affect on
194 IV. MIXED

the coordinates, switches each spinor field with its complex conjugate. Since the
complex conjugate term in the action uses the complex conjugates of the Λ’s, this
symmetry is violated whenever any of the components have imaginary parts (after
taking into account all possible symmetries that could compensate for this, as we
have just done). Note that CP is violated only for 3 families or more. (C and P
are separately violated for any number of families by the SU(2)⊗U(1) coupling: As
discussed in subsection IVB1, C invariance of the strong interactions is the symmetry
qL ↔ qR.) Since we can choose to transform away the phases in the subsector of the
2 lighter quark families, the large masses of the heavier quarks suppress this effect,
accounting for the smallness of CP violation.
Since observed particles are mass eigenstates, it’s convenient to perform a further
unitary transformation (the “Cabibbo-Kobayashi-Maskawa matrix”) that diagonal-
izes the mass matrix. Although this is clearly possible by the arguments of subsec-
tion IVB1, it is not part of the unitary transformations considered in this subsection
because it does not commute with the SU(2) gauge symmetry: After such a trans-
formation, we find that the components of each SU(2) quark multiplet are linear
superpositions of different families.
Excercise IVB3.1
Perform this diagonalization explicitly for the case m=2 (two families), using
the two lightest families of quarks and leptons as listed in subsection IC4.
Which particles mix? Parametrize this mixing by an angle θc (the “Cabibbo
angle”).
An important experimental result with which the Standard Model is consistent is
the suppression of “flavor-changing neutral currents (FCNC)”. The two electrically
neutral gauge fields in this model, the Z and the photon γ, couple to currents that are
neutral with respect to the U(1) symmetries associated with each of the quark (flavor)
numbers. This is true by construction before the unitary CKM transformation, but
this transformation also leaves these two currents invariant (the “Glashow-Iliopoulos-
Maiani mechanism”). Thus, at the classical level we do not see effects such as the
decay K 0 → Z → µ+ µ− , which would violate this “conservation law”. Furthermore,
the quantum corrections are suppressed (though nonvanishing) for similar reasons:
For example, the lowest-order nonvanishing quantum correction comes from replacing
the Z with a W + W − pair. Without the CKM matrix, this contribution would vanish;
treating CKM, and its resulting contribution to quark masses, as a perturbtation, the
resulting contribution is suppressed by a factor of m2q /m2W . The absence of FCNC is
an important constraint on generalizations of the Standard Model.
 B. STANDARD MODEL 195

4. Grand Unified Theories

The Standard Model gives a description of the weak and electromagnetic inter-
actions that describes the spin-1 particles in terms of gauge fields, and accounts for
all masses by the Higgs effect. However, it does not give any unification, in the sense
that we still have 3 groups (SU(3), SU(2), and U(1)) for 3 interactions (strong, weak,
and electromagnetic), and a large variety of spin-1/2 fields that are unrelated ex-
cept by color and broken SU(2) flavor. Grand Unified Theories unify this symmetry
by forcing all 3 gauge groups to be subgroups of a simple group, which is broken to
SU(3)⊗SU(2)⊗U(1) by Higgs (and then broken to SU(3)⊗U(1) by more Higgs). This
means introducing new spin-1 particles that are unobserved so far because of their
very large masses. On the other hand, the known fermions are then grouped together
in a small number of multiplets without introducing new fermions (except perhaps
partners for the neutrinos to allow them to have small masses). Unfortunately, this
requires a more complicated (and ambiguous) Higgs sector, with separate spin-0 mul-
tiplets and couplings for first breaking to SU(3)⊗SU(2)⊗U(1) and then breaking to
SU(3)⊗U(1); we won’t discuss those Higgs fields here.
The simplest such model uses the group SU(5). Recall the SU(3)⊗SU(2)⊗U(1)
representations of each family of fermions:

qL = (3, 12 , 16 ), qR+ = (3̄, 0, 13 ), qR− = (3̄, 0, − 23 ), lL = (1, 12 , − 12 ), lR = (1, 0, 1)

where the first argument is the dimension of the SU(3) representation (3̄ being the
complex conjugate of the 3), the second is the SU(2) isospin, and the third is the U(1)
charge. An SU(3)⊗SU(2)⊗U(1) subgroup of SU(5) can be found easily by taking
the 5-component defining representation and picking 3 components as the defining
representation of SU(3) and the other 2 for that of SU(2): I.e., consider a traceless
hermitian 5×5 matrix as an element of the SU(5) Lie algebra, and take
 
SU(3) − 13 I × U(1) 0
( SU(5) ) →
0 SU(2) + 12 I × U(1)
or in other words
5 → (3, 0, − 13 ) ⊕ (1, 12 , 12 )

From this we recognize the fermions as falling into a 5̄ ⊕ 10, where the 10 is the
antisymmetric product of two 5’s, which consists of the antisymmetric product of the
two 3’s (a 3̄), the antisymmetric product of the two SU(2) doublets, and the product
of one of each:
5̄ → (3̄, 0, 13 ) ⊕ (1, 12 , − 12 ) = qR+ ⊕ lL
196 IV. MIXED

10 → (3̄, 0, − 23 ) ⊕ (1, 0, 1) ⊕ (3, 12 , 16 ) = qR− ⊕ lR ⊕ qL

A more unifying model is based on SO(10). A U(5) subgroup can be found from
the spinor representation by dividing up the set of 10 Dirac γ matrices into two
halves, and taking complex combinations to get 5 sets of anticommuting creation and
annihilation operators. (See excercise IC1.2.) The Dirac spinor is then

(1, − 52 ) ⊕ (5, − 32 ) ⊕ (10, − 12 ) ⊕ (10, 12 ) ⊕ (5̄, 32 ) ⊕ (1, 52 )

in terms of the SU(5) representation and the U(1) charge. This Dirac spinor is
√
reducible into Weyl spinors 16 ⊕ 16; in fact, i 2γ−1 is just (−1)Y +1/2 in terms of the
U(1) charge Y . (The SO(10) generators are even in oscillators, and thus do not mix
even levels with odd.) We then have

16 → (1, − 52 ) ⊕ (10, − 12 ) ⊕ (5̄, 32 )

Ignoring the U(1) charge, these are the multiplets found for each family in the SU(5)
GUT, plus an extra singlet.
A simple way to understand this extra singlet is to look at a different path of
breaking to SU(3)⊗SU(2)⊗U(1): Looking at the vector (defining) representation of
SO(10), we can break it up as 6+4 (in the same way we broke up the 5 of SU(5)
as 3+2) to get the subgroup SO(6)⊗SO(4)=SU(4)⊗SU(2)⊗SU(2). We can also see
that a Dirac spinor of SO(10) (16⊕16) will be a Dirac spinor of SO(6) (4⊕4̄) times
(not plus) a Dirac spinor of SO(4), while the Dirac spinor of SO(4) is a defining
representation of one SU(2) (( 12 ,0)) plus a defining representation of the other SU(2)
((0, 12 )). Thus,
16 → (4, 12 , 0) ⊕ (4̄, 0, 12 )

where we have used the fact that γ−1 (used for projection to Weyl spinors) of SO(10)
is proportional to the product of all the γ-matrices, and thus the product of γ−1 ’s for
SO(6) and SO(4).
Looking at this model (“Pati-Salam model”) as an alternative to SU(5) (but with
a semisimple, rather than simple, group, so it unifies only spin 1/2, not spin 1), we
now look at breaking SU(4)→U(3)= SU(3)⊗U(1) (using 4=3+1, as we did 5=3+2
for SU(5)), and breaking one SU(2)→ U(1). We then find

(4, 12 , 0) → (3, − 13 , 12 , 0) ⊕ (1, 1, 12 , 0) = qL ⊕ lL

(4̄, 0, 12 ) → (3̄, 13 , 0, 12 ) ⊕ (3̄, 13 , 0, − 12 ) ⊕ (1, −1, 0, 12 ) ⊕ (1, −1, 0, − 12 )

= qR+ ⊕ qR− ⊕ lR ⊕ lR−

 B. STANDARD MODEL 197

where the arguments are the SU(3) representation, the U(1) charge from SU(4),
the SU(2) isospin, and the U(1) charge from the broken SU(2). If we choose the
U(1) charge of SU(3)⊗SU(2)⊗U(1) as −1/2 times the former of these two U(1)
charges plus 1 times the latter, this agrees with the result obtained by way of SU(5).
However, we now see that all the left-handed fermions are contained within one
SU(4)⊗SU(2)⊗SU(2) multiplet, and the right-handed within another, but with a
partner for the neutrino. Also, one of the SU(2)’s is that of SU(3)⊗SU(2)⊗U(1),
while the other is the other SU(2) of the Standard Model, which was broken explic-
itly there to U(1), whereas here it is broken spontaneously. Thus, there is a local
chiral SU(2)⊗SU(2) flavor symmetry.

SO(10)
. &
SU(5) SU(4)⊗SU(2)L ⊗SU(2)R
& . ≈ 1016 GeV?
SU(3)⊗SU(2)L ⊗U(1)R
↓ ≈ 100 GeV
SU(3)⊗U(1)

Furthermore, the SU(4)⊗SU(2)⊗SU(2) model is invariant under C: In general, C

is just a permutation symmetry. In this case, it simply switches the two multiplets
of each family,
C: (4, 12 , 0) ↔ (4̄, 0, 12 )

Combining with the usual CP, this model is thus also invariant under P:

P : (4, 12 , 0) ↔ (4̄, 0, 12 )* i.e., ψ α(4, 12 , 0) ↔ ψ̄α.(4, 0, 12 )

But both C and P are broken spontaneously on reduction to the Standard Model.
However, SO(10) lacks C and P invariance (contrary to some statements in the liter-
ature), since there is only a single complex representation for each family of fermions
(and thus no nontrivial C; of course, there is still CP, at least for the vector-spinor
coupling, as always). In fact, the C of SU(4)⊗SU(2)⊗SU(2) is just an SO(10) trans-
formation: Although SO(10) is not O(10) (which is why it lacks a C), it still includes
reflections in an even number of “axes”, since reflection in any pair of axes is a π rota-
tion (just as for SO(2)). Thus, breaking 10 → 6 + 4 includes not only SO(6)⊗SO(4),
but also the reflection of an odd number of the “6” axes together with an odd number
of the “4” axes — a combined “parity” of both SO(6) and SO(4). (They are all the
198 IV. MIXED

same up to continuous SO(6)⊗SO(4) transformations.) This parity of the internal

space is the C given above. (We saw a similar situation for O(2) in subsection IVB1.)
Since GUTs unify quarks and leptons, they allow decay of the proton. However,
since this requires simultaneous decay of all 3 quarks into 3 leptons, it is an extremely
unlikely (i.e, slow) decay, but barely within limits of experiment, depending on the
model. Proton decay is still unobserved: This eliminates the simplest version of the
SU(5) GUT.

REFERENCES
1 The rise of the standard model: Particle physics in the 1960s and 1970s, eds. L. Hod-
deson, L. Brown, M. Riordan, and M. Dresden (Cambridge University, 1997):
interesting accounts of the development of the Standard Model from many of the people
responsible.
2 J.J.J. Kokkedee, The quark model (Benjamin, 1969);
J.L. Rosner, Plenary report on recursive spectroscopy (theory), in Proc. of the XVII Int.
Conf. on High Energy Theory, London, July, 1974, ed. J.R. Smith (Science Research
Council, 1974) p. II-171;
A. De Rújula, H. Georgi, and S. Glashow, Phys. Rev. D12 (1975) 147:
nonrelativistic quark model.
3 H. Fritzsch and M. Gell-Mann, Proc. XVI International Conference on High Energy
Physics, eds. J.D. Jackson and A. Roberts, Chicago and Batavia, Sep. 6-13, 1972 (Na-
tional Accelerator Laboratory, 1972) v. 2, p. 135;
H. Fritzsch, M. Gell-Mann, and H. Leutwyler, Phys. Lett. 47B (1973) 365;
S. Weinberg, Phys. Rev. Lett. 31 (1973) 494, Phys. Rev. D8 (1973) 4482:
QCD, with confinement.
4 W. Pauli, Niels Bohr and the development of physics (McGraw-Hill, 1955) p. 30;
G. Lüders, Ann. Phys. 2 (1957) 1:
CPT theorem.
5 E. Fermi, Ric. Scientifica 4 (1933) 491; Nuo. Cim. II (1934) 1; Z. Phys. 88 (1934) 161:
weak interactions, without vector bosons.
6 T.D. Lee and C.N. Yang, Phys. Rev. 104 (1956) 254:
parity violation in weak interactions.
7 S.L. Glashow, Nuc. Phys. 22 (1961) 579;
A. Salam and J.C. Ward, Phys. Lett. 13 (1964) 168;
S. Weinberg, Phys. Rev. Lett. 19 (1967) 1264:
unification of weak and electromagnetic interactions.
8 N. Cabibbo, Phys. Rev. Lett. 10 (1963) 531;
M. Kobayashi and K. Maskawa, Prog. Theor. Phys. 49 (1972) 282.
9 S.L. Glashow, J. Iliopoulos, and L. Maiani, Phys. Rev. D2 (1970) 1285.
10 J. Pati and A. Salam, Phys. Rev. Lett. 31 (1973) 275:
earliest GUT, but semisimple.
11 H. Georgi and S. Glashow, Phys. Rev. Lett. 32 (1974) 438:
SU(5) GUT; earliest with simple group.
12 H. Georgi, in Particles and Fields — 1974 , proc. AIP conference, Division of Particles
and Fields, Sep. 5-7, 1974, Williamsburg, ed. E. Carlson (American Institute of Physics,
 B. STANDARD MODEL 199

1975) p. 575;
H. Fritzsch and P. Minkowski, Ann. Phys. 93 (1975) 193:
SO(10) GUT.
13 G.G. Ross, Grand unified theories (Benjamin/Cummings, 1984).
200 IV. MIXED

.................... ....................
.................... C. SUPERSYMMETRY ....................

In section IIC we studied some general properties of supersymmetry in arbitrary

dimensions, and its representations in D=4. We now consider 4D interactions, by
introducing gauge fields defined on superspace, and their actions. A complete dis-
cussion of supersymmetry would require (at least) a semester; but here we give more
than just an overview, and include the basic tools with examples, which is enough for
many applications. Quantum aspects of supersymmetry will be discussed in chapters
VI and VIII, supergravity in chapter X, and some aspects of superstrings in chapter
XI.

1. Chiral
We first consider some field equations that appear in all free, massless, super-
symmetric theories. Of course, since the theory is massless it satisfies the massless
Klein-Gordon equation by definition: Φ = 0. From. our earlier . discussion of general
αβ . αβ
properties of supersymmetry, we also know that p q̄ Φ = p qαΦ = 0. These don’t
β
look covariant, but noticing that pq differs from pd only by θ terms (because of the
index contraction), which already vanishes, we have the field equations
. .
αβ ¯. αβ
p d Φ = p dα Φ = 0
β

These equations imply the Klein-Gordon equation, as seen by hitting them with
another d and using the anticommutation relations {dα , d¯.} = p .. They imply
β αβ
stronger equations: By evaluating at θ = 0, we find
. .
pαβ ψ̄ . = pαβ ψα = 0
β

the usual for massless spin 1/2.

Another equation that can be imposed is the “chirality” condition

d̄α.φ = 0

This requires that φ be complex, otherwise we would also have dα φ = 0 and thus pφ =
0 by the anticommutation relations. The component expansion is given completely
¯
by just the d’s and not the d’s:

φ| = A, (dα φ)| = ψα, (d2 φ)| = B

where A and B are complex scalars, and we use the normalization

d2 = 12 dα dα
 C. SUPERSYMMETRY 201

¯ can be pushed past the

All other components are x-derivatives of these, since the d’s
d’s (producing p’s) until they annihilate φ. Another way to state this is to use the
fact
.
dα = e−U/2∂α eU/2, d¯α. = eU/2∂¯α.e−U/2; U = θα θ̄β p .
αβ

to solve the chirality constraint as

φ(x, θ, θ̄) = eU/2φ̂(x, θ)

where φ̂ is independent of θ̄: It is defined on “chiral superspace”. (In this equation U

generates a complex coordinate transformation.) Another way to solve the chirality
constraint is to use the covariant derivatives: Since dα dβ dγ = 0 by anticommutativity
(and similarly for d̄’s),
d¯α. φ = 0 ⇒ φ = d¯2 ψ

where ψ is a “general” (unconstrained) complex superfield. It is the “prepotential”

for the field φ.
From the anticommutation relations we find
. .
[d̄α , d2 ] = pβ α dβ

Since this must vanish on φ, we find

dα d2 φ = d¯α. d2 φ = 0 ⇒ p .d2 φ = 0 ⇒ d2 φ = constant

αβ

(We can safely ignore this constant, at least when considering the free theory: It
corresponds to a term in the action linear in the fields.) This field equation, together
with the chirality constraint, is sufficient to determine the theory: A is the usual free
(complex) scalar, ψα is the usual free spinor, and B is a constant.
To describe interactions of this (“scalar”) multiplet, we keep the chirality condi-
tion, since that greatly simplifies the field content of the superfield. In fact, this is
clearly the simplest off-shell superfield we can define, since it already has the smallest
number of fermions (as do the coordinates of chiral superspace). (“Off shell” means
all components less gauge degrees of freedom.) This means that the equation d2 φ = 0
will be generalized, since it implies the Klein-Gordon equation. The simplest way to
do this is by constructing an explicit action, our next topic.
202 IV. MIXED

2. Actions
The construction of actions in superspace is different from ordinary theories be-
cause the geometrically simple objects, the potentials, are constrained, while the un-
constrained objects, the prepotentials, can be awkward to work with directly. (This
problem is magnified with extended supersymmetry, whose actions we don’t consider
here.)
We start with the simplest supermultiplet, the chiral superfield. Since chiral
superfields are defined on chiral superspace, a natural generalization of a potential
(nonderivative) term in the action to superspace is
Z
S1 = dx d2 θ f(φ) + h.c.

in terms of some function (not functional) f of chiral superfields φ (the “superpoten-

tial”). We can ignore any θ̄ dependence because it contributes only total derivatives:

φ = eU/2φ̂(x, θ) ⇒ f(φ) = eU/2f(φ̂)

Integration over θ is defined as in subsection IA2; however, now we can replace partial
derivatives with covariant ones, since the modification is again only by total deriva-
tives: Z Z
dx d θ = dx d2
2

with an appropriate normalization. This turns out to be the most convenient one,
since it allows covariant manipulations of the action, and the θ integration can be per-
formed covariantly: Since we know that the result of θ integration gives a Lagrangian
that depends only on x, up to total derivative terms, we can evaluate it as
Z Z
dx d θ f(φ) = dx [d2 f(φ)]|
2

Z
= dx [f 0 (φ|)(d2φ)| + f 00 (φ|) 12 (dα φ)|(dα φ)|]

(suppressing indices on multiple φ’s). This gives the result directly in terms of com-
ponent fields, using the covariant method of defining the component expansion: In
the conventions of the previous subsection, this part of the action becomes
Z
dx [f 0(A)B + f 00(A) 12 ψ α ψα]

We now consider integration over the full superspace. As a generalization of the

above, we can write Z Z
dx d θ K(φ, φ̄) = dx (d2 d̄2 K)|
4
 C. SUPERSYMMETRY 203

Supersymmetric versions of nonlinear σ models can be written in this way; here we

consider just the case where K is quadratic, which is the one interesting for quantum
theory. Since a function of just φ (or just φ̄) will give zero in the d4 θ integral, we
choose Z
K = −φ̄φ ⇒ S0 = − dx d4 θ φ̄φ

Explicitly,
.
L0 = d2 d¯2 (−φ̄φ) = −d2 (d̄2 φ̄)φ = −( 12 φ̄)φ + (i∂ αβ d¯.φ̄)dα φ − (d̄2 φ̄)(d2 φ)
β
.
→ −A 12 Ā + ψ α i∂αβ ψ̄ . − B B̄
β

where we have used the commutation relations of the covariant derivatives to push all
¯ to hit φ̄. Clearly, this term by itself reproduces the results derived in the
d’s past d’s
previous subsection based on kinematics, so it is the desired massless kinetic term.
This could also be seen by deriving the superfield equations of motion by varying
the action. Since φ is constrained, it can’t be varied arbitrarily; varying instead the
prepotential ψ (φ = d̄2 ψ), we find d2 φ = 0 (and the complex conjugate).
We can now see the influence of adding the superpotential term to the action:
The result of combining the two terms, and then eliminating the auxiliary field B by
its equation of motion, is
.
S0 + S1 → L = −A 12 Ā + ψ αi∂αβ ψ̄ . + |f 0 (A)|2 + [f 00(A) 12 ψ αψα + h.c.]
β

For example, a quadratic f gives mass to the physical scalar and spinor. This action is
invariant under modified supersymmetry transformations, where the auxiliary fields
are replaced by their equations of motion there also; those transformations then
become nonlinear in the presence of interactions. Note that the scalar potential is
positive definite; this is a consequence of supersymmetry, since it implies that the
energy is always positive.
Excercise IVC2.1
Find the explicit form of the component-field action for arbitrary K(φi , φ̄i )
and f(φi ) for an arbitrary number of chiral superfields φi , including all in-
dices. Eliminate the auxiliary fields from the action, and find the modified
supersymmetry transformations. Show by direct evaluation that the action is
still invariant.
As a notational convenience, we can drop the “|” after expanding a superspace
action in components: For example, we can write simply

ψα = dα φ, B = d2 φ
204 IV. MIXED

After performing the θ-integration as above by using derivatives d and d,¯ and then
“evaluating” these derivatives on φ by writing ψ and B, the component action is
expressed completely in terms of such superfields and only spacetime derivatives ∂ ..
αβ
This component action is independent of θ (the Lagrangian is independent up to total
spacetime derivatives): This is the statement of supersymmetry invariance. Thus, we
can choose to evaluate at θ = 0, or θ = , or whatever; it is irrelevant. It is then
understood that the relation to the usual component actions is simply to treat the
R
superfield as a component field, since the θ-derivatives (in d and dθ) have been
eliminated. From now on we will generally drop the |’s.
Since chiral superfields are essentially independent of θ̄, not only integration is
modified, but also (functional) variation. Since a chiral superfield is (up to a trans-
formation) an arbitrary function on chiral superspace, we define
Z
δS
δS[φ] = dx d2 θ (δφ)
δφ

for an arbitrary variation of a chiral superfield φ, and similarly for varying φ̄. In
evaluating such variations, we make use of the identities
Z Z
dx d θ L = dx d2 θ d¯2 L
4

d¯2 d2 φ = 1
2 φ (d̄2 d2 d̄2 = 1
2 d̄2 )

Thus, to vary a general action, it is convenient to first integrate over θ̄, and then vary
in the naive way: For example,
Z Z 
S = − dx d θ φ̄φ +
4 2
dx d θ f(φ) + h.c.

δS
⇒ 0= = −d̄2 φ̄ + f 0 (φ)
δφ
Excercise IVC2.2
Check for this action that the component expansion of the superfield equations
of motion agree with the variation of the corresponding component action.

3. Covariant derivatives
The supersymmetric generalization of nonabelian gauge theories can be derived by
similar methods. We first write the supersymmetry covariant derivatives collectively
as
dA = (dα , d̄α., ∂αα.) = EA M ∂M
 C. SUPERSYMMETRY 205

.
∂M = (∂µ , ∂µ., ∂m ) = ∂/∂z M , z M = (θµ , θ̄µ , xm )
Unlike the nonsupersymmetric case, the “vielbein” EAM has θ dependence even in
“flat” superspace, and thus the “torsion” T is nonvanishing:

[dA , dB } = TAB C dC
. . .
T .γ γ = T . γ γ = −iδαγ δ γ., rest = 0
αβ βα β

We now gauge-covariantize all the supersymmetry-covariant derivatives:

∇A = dA + iAA

The covariant field strengths are then defined as

[∇A, ∇B } = TAB C ∇C + iFAB

From our analysis of general representations of supersymmetry in D=4 in subsection

IIC5, we know that the simplest supersymmetrization of Yang-Mills is to include a
spinor with the vector, in terms of physical degrees of freedom. (The spinor and
vector each have two physical degrees of freedom, one for each sign of the helicity.)
Off shell, Fermi and Bose components must still balance, so there must also be an
auxiliary scalar. From dimensional analysis, the field strengths must therefore satisfy

Fαβ = F . . = F . = 0; F . = −iCαβ W ., F . . = −iC . .Wβ

αβ αβ α,β β β α,β β αβ

where Wα | is the physical spinor.

The constant piece of the torsion implies stronger relations among the field
strengths than in nonsupersymmetric theories. For super Yang-Mills we find from
the Jacobi identity for the covariant derivatives the Bianchi identity for the field
strengths
∇[AFBC) = T[AB|D FD|C)
Specifically, the dimension-1 constraints above on the field strengths imply the dimen-
sion-3/2 algebraic constraint that defines Wα , as well as

F . . = Cαβ 12 ∇(α.W . + C . . 21 ∇(αWβ)

αα,β β β) αβ

They also imply that Wα is covariantly chiral and satisfies a “reality” condition,
.
∇α.Wβ = 0, ∇αWα + ∇αWα. = 0

The most straightforward way to derive these results is to just evaluate the Jacobi
identities directly. We begin with a weaker set of conditions, both of dimension 1, that
206 IV. MIXED

will be found (in the following subsection) to be necessary and sufficient for solving
explicitly. One directly determines the vector derivative in terms of the spinor ones:

F .=0 ⇒ −i∇αα. = {∇α, ∇α.}

αβ

Since one could always define the vector covariant derivative this way, imposing this
condition simply eliminates redundant degrees of freedom.
The remaining constraint (including its complex conjugate) allows coupling of
super Yang-Mills to the chiral superfield:

∇α.φ = 0 ⇒ 0 = {∇α., ∇ .}φ = iF . .φ

β αβ

It also implies the maintenance of certain free identities, such as

∇α∇β = 12 [∇α, ∇β ] + 12 {∇α, ∇β } = Cβα ∇2

(Such constraints appear also for first quantization, e.g., in superstring theory, when-
ever a supersymmetric system is put in a background of a supersymmetric gauge field
of higher superspin. This should not be confused with background field equations
imposed by any gauge system put in a background of the same type: see subsection
VIB8.)
Thus, our minimal set of constraints can be written directly in terms of the field
strengths as
Fαβ = F . . = F . = 0
αβ αβ

but for our purposes it will prove more convenient to write them directly as (anti)com-
mutators:
{∇α, ∇β } = {∇α., ∇ .} = 0, {∇α, ∇ .} = −i∇ .
β β αβ

The solution to the dimension-3/2 Jacobis are then

[∇(α, {∇β , ∇γ)}] = 0 ⇒ trivial

[∇(α, {∇β), ∇γ.}] + [∇γ., {∇α, ∇β }] = 0 ⇒ [∇α, ∇β γ.] = Cαβ W γ.

for some field W , simply applying the constraints to drop {∇α, ∇β } and replace
{∇α, ∇ .} with ∇ .. Similarly, we find from the dimension-2 Jacobis
β αβ

{∇(α, [∇β), ∇γ γ.]} + [∇γ γ., {∇α, ∇β }] = 0 ⇒ ∇α W γ. = 0

[∇αα., {∇β , ∇γ.}] + {∇β , [∇γ., ∇αα.]} + {∇γ., [∇β , ∇αα.]} = 0

 C. SUPERSYMMETRY 207

.
⇒ [∇αα., ∇ .] = i(Cαβ f¯. . + C . .fαβ ), fαβ = 12 ∇(αWβ), ∇α Wα + ∇αW α. = 0
ββ αβ αβ

where we separated the last equation into its (Lorentz) irreducible pieces. (The
dimension-5/2 and 3 identities are redundant.)
Excercise IVC3.1
Explicitly evaluate all the remaining Jacobi identities, and show that they
imply no further conditions on Wα.

4. Prepotential
We saw in the previous subsection that coupling super Yang-Mills to matter gave
directly one of the minimal constraints on the super Yang-Mills fields themselves.
Hence, as for ordinary Yang-Mills, the definition of the gauge theory follows from
considering the transformation of matter, and generalizing it to a local symmetry. As
for self-dual Yang-Mills (see subsection IIIC5), the vanishing of some field strengths
implies that part of the covariant derivative is pure gauge:

{∇α, ∇β } = 0 ⇒ ∇α = e−Ω dα eΩ

However, since {∇α, ∇ .} 6= 0, this gauge transformation Ω (“prepotential”) is com-

β
plex. We therefore have the covariantly chiral superfield

∇α.φ = 0, ∇α. = eΩ̄dα.e−Ω̄ ⇒ φ = eΩ̄φ̂, dα.φ̂ = 0

Alternatively, we could combine this exponential with that already contained in the
free spinor derivative:

∇α = e−U/2−Ω∂α eU/2+Ω, φ = eU/2+Ω̄φ̂, ∂α.φ̂ = 0

U + 2Ω is the analog of the covariant derivative for the Yang-Mills prepotential. This
is a hint at supergravity: U is just the flat piece of the supergravity prepotential.
We thus see that supersymmetry automatically gives gravity the interpretation of the
gauge theory of translations.
Component expansions are now defined with Yang-Mills-covariant derivatives:

∇ α φ = ψα , ∇2 φ = B
.
∇αWβ = fαβ + iCαβ D, ∇ Wα = −i∇α W .
2 β
β

where we have used the Bianchi identities for W , and fαβ (not to be confused with Fαβ )
is the usual Yang-Mills field strength (in spinor notation). The “vector multiplet”
208 IV. MIXED

thus consists of the component fields Aa (the gauge field whose strength is f), Wα,
and D (auxiliary). (As explained earlier, we drop all |’s.)
Note that the gauge parameter is real, while the matter multiplet is (covariantly)
chiral. The resolution of this apparent inconsistency is that solving the constraints
introduces a new gauge invariance:
0
∇0A = eiK ∇Ae−iK , ∇α = e−Ωdα eΩ ⇒ eΩ = eiΛ̄ eΩe−iK , dα Λ̄ = 0

φ0 = eiK φ, φ = eΩ̄φ̂ ⇒ φ̂0 = eiΛφ̂

This suggests the definition of a new (“chiral”) representation, where we use the
obvious field φ̂ and the chiral gauge parameter Λ replaces the real one K: Making a
nonunitary similarity transformation,
b A = e−Ω̄∇AeΩ̄
∇ ⇒ b . = d .,
∇ b α = e−V dα eV ,
∇ eV = eΩeΩ̄
α α

φ̂ = e−Ω̄φ, ˆ = φ̄eΩ̄
φ̄ ⇒ dα.φ̂ = 0, ˆ = (φ̂)† eV
φ̄
b 0A = eiΛ∇
∇ b Ae−iΛ , 0
eV = eiΛ̄eV e−iΛ
Alternatively, we can also include U in the transformation as above; then U and V
appear only in the combination U + V .
Excercise IVC4.1
Show that the explicit expression for the field strength Wα in terms of the
prepotential V in the chiral representation is

Wα = −id̄2 (e−V dα eV )

Show this expression is chiral.

Excercise IVC4.2
In the Abelian case, give an explicit component expansion of the prepotential
V , such that the vector potential Aa, the physical spinor Wα , and the auxiliary
field D appear as independent components. Note that the other components
do not appear explicitly in component expansions when gauge-covariant ex-
pansion (∇...|) is used. The component (nonsupersymmetric) gauge where
these components are set to vanish is the “Wess-Zumino gauge”, and is the θ
part of the radial gauge of subsection VIB1 below.
Excercise IVC4.3
For some purposes (like quantization) we need the explicit form of an in-
finitesimal gauge transformation of V . Show this can be written as

δV = −iLV /2[(Λ + Λ̄) + coth(LV /2 )(Λ − Λ̄)]

 C. SUPERSYMMETRY 209

(Hint: Consider e−V δeV , and think of δ as an operator, as for the expansion
of ∇α = e−V dα eV . LA was defined in subsection IA3.)

5. Gauge actions
Generalization of actions to super Yang-Mills theory is straightforward. Matter
coupling is achieved simply by replacing the chiral superfields of the matter multiplets
with Yang-Mills-covariantly chiral superfields. The coupling can be seen explicitly in
the chiral representation: In the kinetic term,

φ̄φ = (φ̂)† eV φ̂
R
while in the d2 θ term all V -dependence drops out because of gauge invariance. (The
superpotential is a gauge invariant function of the φ’s, and the transformation to the
chiral representation is a complex gauge transformation. The fact that the gauge
transformation is complex is irrelevant, since the superpotential depends only on φ
and not φ̄.) Component expansion can be performed covariantly by replacing d’s with
∇’s in the definition of θ integration: Since the Lagrangian is a gauge singlet, this is
the same acting on it, although individual terms in the expansion differ because the
2
fields are not singlets. Similarly, d¯2 can be replaced with ∇ also when performing θ̄
integration for purposes of varying an action with respect to a chiral superfield. This
is equivalent to gauge covariantizing the functional derivative (e.g., by transforming
from a chiral representation) as
δφ(x, θ) 2
0 0
= ∇ δ(x − x0 )δ 4(θ − θ0 )
δφ(x , θ )
Usually we will drop the “ b ”’s on φ and φ̄, when the representation is clear from the
context by the use of explicit V ’s.
The action for super Yang-Mills itself follows from dimensional analysis: Since
each θ integral is really a θ derivative, d2 θ integration has mass dimension +1, the
same as a spacetime derivative. Since the Lagrangian for a physical spinor, in this
case W α , has a single such derivative, dimensional analysis says the action must be
Z
SsY M = − g2 tr dx d2 θ 12 W αWα
1

where the (covariant) chirality of W α allows integration over chiral superspace. (Sim-
R
ilar analysis applies to the matter multiplet, where d4 θ takes the place of a for
R 2
the scalar φ.) Replacing d θ → ∇ , we evaluate the component expansion as
2

Z .
SsY M = g2 tr dx ( 2 f fαβ + W i∇α W . − D2 )
1 1 αβ α β
β
210 IV. MIXED

Another term we can write, for superelectromagnetism (supersymmetrization of an

Abelian gauge theory) is the “Fayet-Iliopoulos term”
Z Z
4
SF I = dx d θ V = dx D

which involves only the auxiliary field D. (The analog for the chiral scalar superfield
R
is dx d2 θ φ.)
Excercise IVC5.1
Derive the supersymmetric analog of the Stückelberg model of subsection
IVA5, by coupling an Abelian vector multiplet to a massless chiral scalar
multiplet using the symmetry generator T defined there. (G → −iT in trans-
formation laws, covariant derivatives, etc., on φ, where T φ = 1 ⇒ T 2φ = 0.)
a To couple the gauge field it is necessary to start, as usual, with a (quadratic)
matter action that is globally invariant under this symmetry:
Z
S0 = dx d4 θ 12 (φ − φ̄)2

(At this point this is the usual, since only the cross-term survives, but this
will not be the case for the covariantly chiral superfields.) Find the super-
symmetric gauge coupling, and express the resulting action in terms of V and
φ̂.
b Use this result to find the mass term for V in the gauge φ̂ = 0.
Another interesting form of the action uses a generalization of the Chern-Simons
form defined in the discussion of instantons in subsection IIIC6. In superspace, the
calculation of the field strength with curved indices is modified to

∇M = EM A∇A = ∂M + iAM , −i[∇M , ∇N } = FM N = EM AEN B FAB

where we have left sign factors from index reordering in the last equation implicit.
Although the curved-index expressions are not as useful (for example, for seeing which
components vanish by constraints), we can see easily that some arguments used in
nonsupersymmetric theories carry over to superspace. Thus, we can define the super
Chern-Simons form by
1
8
tr F[M N FP Q) = 16 ∂[M BN P Q)

BM N P = tr( 12 A[M ∂N AP ) + i 13 A[M AN AP ))

Converting to flat superspace (again with some implicit sign factors),

BABC = EA M EB N EC P BM N P = tr( 12 A[AdB AC) − 14 A[ATBC) D AD + i 13 A[AAB AC) )

 C. SUPERSYMMETRY 211

In terms of this expression, the super Yang-Mills action can be written simply in
terms of the spinor-spinor-vector part B . of BABC as
αβc
Z .
SsY M,1 = 1 1
2 i g 2 tr dx d4 θ Bα,α.αα

Note that the fact that the curl of B is gauge invariant implies that B transforms
under a gauge transformation as the curl of something, and thus the integral of any
part of B is gauge invariant (up to possible torsion terms: see the excercise below).
Furthermore, we can drop the F . = 0 constraint on the A in this action; it follows
αβ
from variation with respect to A .. One simple way to check this action is to use the
αβ
.↔
chiral representation Aα. = 0: Then only the Aαβ d .Aα and (A .)2 terms contribute,
R β αβ R
and A . = id¯.Aα, while Wα = d̄2 Aα, so d2 θ̄ integration gives − dx d2 θ W 2 .
αβ β
Excercise IVC5.2
Derive the expression for BABC directly using only flat indices:
a Start with F[AB FCD) expressed in terms of T and A, and write it as a total
derivative plus torsion terms.
b Do the same for the gauge transformation of B. Show that the torsion terms
.
do not contribute to δB αα, α,α..
The multiplets and couplings we have considered are sufficient to write a super-
symmetric generalization of the Standard Model. Unfortunately, supersymmetry pro-
vides no unification. To get the right symmetry breaking, it turns out to be necessary
to provide a supersymmetry multiplet for each particle of the Standard Model: The
spin-1 gauge bosons are accompanied by spin-1/2 “gauginos” (“gluinos”, “photino”,
“Wino”, “Zino”), the spin-1/2 leptons by spin-0 “sleptons”, the quarks by “squarks”,
and the spin-0 Higgs’ by spin-1/2 “Higgsinos”. Furthermore, since a reality condition
can’t be imposed on chiral scalar multiplets, the Higgs scalars are themselves doubled.
Ultimately, the success of supersymmetry depends on the experimental detection of
these particles.

6. Breaking
The methods of the section IVA can be generalized straightforwardly to supersym-
metric theories: Goldstone bosons and Higgs fields become supermultiplets, etc. How-
ever, to obtain realistic models supersymmetry itself must be broken, since fermions
and bosons with similar mass and other properties are not observed in nature. More
specifically, since gravity is observed, any supersymmetric theory of the world must
212 IV. MIXED

include supergravity, and thus the breaking must be spontaneous. (Explicit breaking
would violate gauge invariance.) Then the gravitino, which gauges supersymmetry,
will become massive by a superhiggs mechanism, by eating a Goldstone fermion. (See
subsections XB6-7. If the graviton and gravitino are treated as composites, then this
fermion could also be a composite.)
We saw in subsection IIC1 that energy is always nonnegative in supersymmetric
theories. In particular, from the same arguments used there we see that a state can be
invariant under supersymmetry (q|ψi = q †|ψi = 0) if and only if it has zero energy.
Any such state can be identified as the vacuum, since no state has lower energy.
This means that the only way to guarantee spontaneous supersymmetry breaking is
to choose a theory which has no zero-energy state. (Note that energy is uniquely
defined by the supersymmetry algebra; there is no possibility of adding a constant
as in nonsupersymmetric theories.) In theories with extended supersymmetry, the
relation between supersymmetry and energy applies for each supersymmetry; thus
supersymmetry is either completely broken spontaneously or completely unbroken.
(An exception is central charges, which modify the supersymmetry algebra: See the
following subsection.)
Furthermore, physical scalars appear at θ = 0 in matter multiplets, while auxiliary
fields appear at higher order. Since supersymmetry breaking requires θ dependence
in a vacuum value of a superfield, this means an auxiliary field must get a vacuum
value.
A simple example of spontaneous supersymmetry breaking is the O’Raifeartaigh
model; it has the Lagrangian
Z X
3 Z 
LO’R = − 4
dθ Φ̄i Φi + 2
d θ λ(ζΦ1 + mΦ2 Φ3 + Φ1 Φ22 ) + h.c.
i=1

To study symmetry breaking we ignore derivative terms, since vacuum values are
constants. Then the scalar field equations are:
δ
→ −B̄i + ∂i f = 0 : −B̄1 + ζ + A22 = −B̄2 + mA3 + 2A1A2 = −B̄3 + mA2 = 0
δBi
δ
→ Bj ∂i∂j f = 0 : 2A2B2 = mB3 + 2A2 B1 + 2A1 B2 = mB2 = 0
δAi
(where ∂i = ∂/∂Ai on the superpotential f(A)). Since there is no solution for Bi = 0,
supersymmetry breaking is required. In general, for superpotential f(Φ), the field
equations for B = 0 are f 0 (A) = 0, so a linear term is always needed for supersym-
metry breaking.
 C. SUPERSYMMETRY 213

R
With Abelian vector multiplets, a Fayet-Iliopoulos term d4 θ V can also generate
such breaking, since it also is a linear term of an auxiliary field.
Excercise IVC6.1
R
Evaluate the Lagrangian − d4 θ φ̄φ for covariantly chiral φ by using covariant
R 2
θ-integration, d4 θ = ∇2 ∇ . For the case of U(1) gauge theory, add the
action for the gauge superfield with a Fayet-Iliopoulos term, and find the
potential for the physical scalars by eliminating the auxiliary field D by its
field equation.
For simplicity (as in this chapter), we may want to ignore supergravity; however,
we still need to take account of its contribution to breaking global supersymmetry via
the superhiggs effect. The net low-energy contribution from the supergravity fields
(assuming no cosmological constant is generated) is to introduce effective explicit
supersymmetry breaking: Although the original theory is locally supersymmetric, we
neglect the supergravity fields but not their vacuum values (in particular, those of
the auxiliary fields). In particular, if the supergravity fields are bound states, then
this procedure is essentially the classical introduction of nonperturbative quantum
effects.
Thus we consider adding terms to the classical action that break supersymmetry
explicitly. The easiest way to do this is to introduce constant superfields (“spurions”);
this allows us to continue to take advantage of the superspace formalism (at both
the classical and quantum levels). Since we are neglecting (super)gravity, and in
particular its nonrenormalizability (see chapter VII), we consider only terms that will
preserve the quantum properties of the unbroken theories. This will clearly be the
case if we consider only the usual terms, with some fields replaced by spurions: This
is equivalent to using background (fixed) fields, in addition to (but in the same way
as) the usual field variables, performing all (classical/quantum) calculations as usual,
and then setting the background fields (specifically, the auxiliary fields, which are
responsible for breaking supersymmetry) to constants.
Thus, introducing constant (in x) chiral and real spurion fields

ϕ = θ2 c, V = θ2θ̄2 r

in terms of complex and real parameters c and r, in addition to the true fields φ and
V , we have terms of the form
Z Z
d θ [ϕφ, ϕφ , ϕφ , ϕW , (d¯ d V)φWα],
2 2 3 2 2 α
d2 θ d2 θ̄ V φ̄eV φ
214 IV. MIXED

(and complex conjugates). These terms can preserve the usual gauge invariances, and
can be shown to also preserve the desirable quantum properties of supersymmetry:
The condition is that replacing the spurion field by 1 (instead of its above value)
gives either 0 or a conventional term (one with coupling constant of nonnegative
mass dimension). Another way to introduce these spurions (except perhaps for the
φV crossterm, which is less useful) is as coupling constants, rather than as fields:
Instead of introducing new terms to the action, we generalize the old ones, so the
constant part of each coupling is the usual coupling, while its θ-dependent terms
produce the breaking.
Excercise IVC6.2
Find the component expansions of the above explicit breaking terms. What
are the mass dimensions of the constants c and r in the various cases?
Excercise IVC6.3
Expand the Lagrangian
Z Z 
L = − d θ φ̄φ +
4 2 1 3
d θ ( 6 φ + ϕφ) + h.c.

in components. Find the masses.

7. Extended
The supersymmetry we discussed earlier in this chapter, with a single spinor coor-
dinate, is called “simple (N=1) supersymmetry”; the generalization to many spinors is
called “extended (N>1) supersymmetry” (for N spinor coordinates). N=1 supersym-
metric theories, at least for spins≤1, are most conveniently described by superspace
methods. (There are also some definite advantages for N=1 supergravity at the quan-
tum level.) On the other hand, the technical difficulties of extended superspace often
outweigh the advantages. (The main advantage of extended superspace is proving cer-
tain properties of the quantum theories. Of course, extended supersymmetric theories
are complicated in any case.) Alternative formulations of extended supersymmetry
are either (1) on shell, (2) in terms of components (ordinary spacetime, not super-
space), or (3) in simple superspace (manifesting only one of the supersymmetries).
By going half way, using N=1 superfields to describe extended supersymmetry,
some of the advantages of the superspace approach can be retained. In this subsection
we will list some of the extended supersymmetric actions for lower spins in N=1
superspace form. These actions can be obtained by: (1) using extended superspace
to derive the component field equations (usually using dimensional reduction: see
 C. SUPERSYMMETRY 215

subsections XC5-6), and combining components into N=1 superfields, or (2) writing
the extra supersymmetries in N=1 superspace form, and using them to determine the
action.
The simplest example is N=2 supersymmetry. As for any extended supersymme-
try, its algebra can be modified by including Abelian generators Z (with dimensions
of mass), called “central charges”:

{qiα , q̄ j.} = δij p ., {qiα , qjβ } = Cαβ Cij Z, {q̄αi., q̄ j.} = C . .C ij Z; [Z, q] = [Z, q̄] = 0
β αβ β αβ

(where i = 1, 2). In terms of dimensional reduction (for N=2, from D=5 or 6; see
subsections XC5-6), the origin of these generators can be understood as the higher-
dimensional components of the momentum. N=2 supersymmetry is sometimes called
“hypersymmetry”, and N=2 supermultiplets, “hypermultiplets”.
Our first example is the free, massive N=2 scalar multiplet: Since we already
know the field content (see subsection IIC5), it’s easy to write the free Lagrangian
Z Z 
i0 i0j 0
Lsm,N =2 = − d θ φ̄ φi0 + 2
4 1 2
d θ m φi0 φj 0 + h.c.

where the index “i0” is for an extra SU(2) (not the one acting on the supersymmetry
0 0
generators), broken by the mass term, and the mass matrix mi j is symmetric while
0 0 0
mi0 j = Ck0 i0 mk j is hermitian. In other words, it represents a 3-vector of this SU(2),
and thus a generator of the preserved U(1) subgroup, which we have used to define
the central charge:
0
Zφi0 = mi0 j φj 0

The other N=2 multiplet of low spin is the vector multiplet. It also has a simple
Lagrangian, Z Z 
LsY M,N =2 = − g2 tr
1 2 2 4
d θ W + d θ φ̄φ

where φ is covariantly chiral and in the adjoint representation of the Yang-Mills gauge
group. In the Abelian case, we can also add an N=2 Fayet-Iliopoulos term,
Z Z 
4 2
LF I,N =2 = d θ c0V + d θ c+φ + h.c.

where (c0 , c+, c−) (c− = c+ *) is a constant 3-vector of the SU(2) of the N=2 super-
symmetry: The 3 scalar auxiliary fields of this N=2 multiplet form a 3-vector of the
SU(2). Unlike the previous example, this multiplet has all the auxiliary fields needed
for an off-shell N=2 superspace formulation: Not only do the physical components
balance between bosons and fermions (4 of each), but also the auxiliary ones (also 4
of each).
216 IV. MIXED

These 2 N=2 multiplets can be coupled: The scalar multiplet action is modified
to Z Z 
i0 i0 j 0
Lsm,N =2 = − 4
d θ φ̄ φi0 + 1
2
2
dθτ φi0 (φ + M)φj 0 + h.c.

where now φi0 is also a representation of the Yang-Mills group (not necessarily ad-
joint), with respect to which it is covariantly chiral. However, the same SU(2) matrix
τ that appears in the mass matrix mi0 j 0 = Mτi0 j 0 now also appears with the N=2
super Yang-Mills fields,
0
∇Aφi0 = dA φi0 + iAnAGn τi0 j φj 0 , φ = φnGn

where Gn are the usual Yang-Mills group generators. (Without loss of generality, we
0

can choose τi0 j = 10 −1
0
; then φ+0 is some arbitrary representation of the Yang-Mills
group, while φ−0 is the complex conjugate.) Note that the mass term appears in
exactly the same way as an Abelian N=2 vector multiplet that has been replaced by
a vacuum value for its physical scalars. This can also be seen from the commutation
relations for the N=2 super Yang-Mills covariant derivatives (see below), since the
scalars appear in exactly the same way as the central charge.
By our earlier helicity arguments, the only N=3 supersymmetric theory with spins
≤ 1 is N=3 super Yang-Mills. The analogous statement also holds for N=4, while
no such theories exist for N>4. Since theories with N supersymmetries are a subset
of those with only N−1 supersymmetries, N=3 and N=4 super Yang-Mills must be
the same: Counting states of supersymmetry representations, we see that this theory
is the same as N=2 super Yang-Mills coupled to one N=2 scalar multiplet in the
adjoint representation (in direct analogy to N=2 super Yang-Mills in terms of N=1
multiplets). In terms of N=1 multiplets, this is super Yang-Mills plus 3 adjoint scalar
multiplets. The action then follows from the above results (without central charges
and Fayet-Iliopoulos terms):
 Z Z Z 
LsY M,N =4 = g2 tr − d θ W − d θ φ̄ φI +
1 2 2 4 I 2 1 IJK
d θ 6  φI [φJ , φK ] + h.c.

where “I” is a U(3) index. (The U(1) part of the U(3) symmetry involves also a phase
transformation of the θ’s.)
As for off-shell N=1 supersymmetry, much information on extended supersym-
metric gauge theories can be gained by examining the properties of the covariant
derivatives and their field strengths. In fact, this is more true in the extended case,
where the “obvious” constraints often imply field equations (which is more than one
would want for an off-shell formulation). The empty-space covariant derivatives are
 C. SUPERSYMMETRY 217

the direct generalization of N=1: Introducing N θ’s as θiα (and complex conjugate
.
θ̄i α ), where “i” is an N-valued index with as much as a U(N) symmetry,
.
dA = (diα , d¯i α., ∂αα.); diα = ∂iα − i 12 θ̄i α∂αα., d¯i α. = ∂¯i α. − i 12 θiα ∂αα.
. . .
Tiα,j .γ γ = T j .,iα γ γ = −iδij δαγ δ γ., rest = 0
β β β

Excercise IVC7.1
Find the superspace representation of the extended supersymmetry generators
(which anticommute with these covariant derivatives). For N=2, include the
central charge.
By definition, extended super Yang-Mills has only spins 1 and less. Dimensional
analysis then gives the unique result, including physical fields only,

{∇iα, ∇j .} = −δij i∇ .
β αβ

{∇iα, ∇jβ } = Cβαiφ̄ij

[∇i α., −i∇ .] = C . .iW i β
ββ βα

[∇αα., ∇ .] = Cαβ if¯. . + C . .ifαβ

ββ αβ αβ

(and complex conjugates of some of these equations). This corresponds directly to

our discussion in subsection IIC5, where we saw that a general representation looked
like antisymmetric tensors φ, φi, φij , ... of U(N), corresponding to helicities h, h−1/2,
h−1,... . In this case, h=1, and these helicities come from the surviving on-shell
components of fαβ , W i α , φij ,... . For N=4 we have self-duality with respect to charge
conjugation (see also subsection IIC5),

φij = 12 ijkl φ̄kl

Excercise IVC7.2
Analyze the Bianchi identities of these covariant derivatives. Show that for
N>2 they imply the field equations. Find a component action that yields
these field equations for N=4.
An interesting simplification of extended superspace occurs for self-duality: Con-
straining
f αβ = W iα = φij = 0
and dropping the self-duality condition for N=4 (so φ̄ij 6= 0), we find all commutators
i
involving ∇ α. are trivial:
i j
{∇iα, ∇j .} = −δij i∇ ., {∇ α., ∇ .} = [∇i α., ∇ .] = 0
β αβ β ββ
218 IV. MIXED

while all the remaining commutators have a similar form:

{∇iα, ∇jβ } = Cβαiφ̄ij , [∇iα, −i∇ .] = CβαiW ., [∇αα., ∇ .] = Cαβ if¯. .

ββ iβ ββ αβ

The latter result suggests we combine the internal and dotted spinor indices as
.
A = (α, i)

so that we can combine the nontrivial equations as

[∇Aα, ∇Bβ } = iCαβ fAB

The former equations then allow us to interpret the remaining covariant derivatives
i
∇ α. as a subset of the SL(2|N) generators that rotate the A index, which form a
subgroup of the superconformal group (S)SL(4|N). We therefore restrict ourselves to
the chiral superspace described by the coordinates
.
z Aα = (xαα, θiα )

The net result is that we have a superspace with no torsion, with coordinates that
represent half of the supersymmetries as translations and the other half as rotations.
By comparison with our treatment of the self-dual bosonic theory in subsections
IIIC5-7, we see that we can extend trivially all our results for the bosonic case to the
(extended) supersymmetric case by simply extending the range of the indices. In par-
ticular, we also have a chiral twistor superspace: Extending the range on the twistor
coordinates z Aα used there so A is now an SL(4|N) index, the superconformal group
is now manifest, and all the methods and results there (e.g., the ADHM construction)
apply automatically to the supersymmetric case.

REFERENCES
1 Wess and Zumino, loc. cit. (IIC, ref. 2):
free scalar and vector multiplets, Wess-Zumino gauge.
2 J. Wess and B. Zumino, Phys. Lett. 49B (1974) 52:
interacting chiral scalar multiplets.
3 J. Wess and B. Zumino, Nucl. Phys. B78 (1974) 1:
coupling Abelian vector multiplet to scalar multiplets.
4 S. Ferrara and B. Zumino, Nucl. Phys. B79 (1974) 413;
A. Salam and J. Strathdee, Phys. Lett. 51B (1974) 102:
nonabelian vector multiplets.
5 J. Wess and B. Zumino, Phys. Lett. 66B (1977) 361;
J. Wess, Supersymmetry-supergravity, in Topics in quantum field theory and gauge
theories, VIII GIFT Int. Seminar on Theoretical Physics, Salamanca, Spain, Jun 13-19,
 C. SUPERSYMMETRY 219

1977 (Springer-Verlag, 1978) p. 81:

covariant derivatives for super Yang-Mills.
6 P. Fayet and J. Iliopoulos, Phys. Lett. 51B (1974) 461.
7 S.J. Gates, Jr. and H. Nishino, Int. J. Mod. Phys. A 8 (1993) 3371:
Chern-Simons superform.
8 L. O’Raifeartaigh, Nucl. Phys. B96 (1975) 331.
9 L. Girardello and M.T. Grisaru, Nucl. Phys. B194 (1982) 65:
general soft explicit supersymmetry breaking, with spurions.
10 L.A. Avdeev, D.I. Kazakov, and I.N. Kondrashuk, hep-ph/9709397, Nucl. Phys. B510
(1998) 289;
N. Arkani-Hamed, G.F. Giudice, M.A. Luty, and R. Rattazzi, hep-ph/9803290, Phys.
Rev. D58 (1998) 115005:
spurions in couplings.
11 H.P. Nilles, Phys. Rep. 110 (1984) 1:
review of supersymmetry phenomenology.
12 P. Fayet, Nucl. Phys. B113 (1976) 135:
extended supersymmetry.
13 P. Fayet, Nucl. Phys. B149 (1979) 137:
central charges.
14 F. Del Aguila, M. Dugan, B. Grinstein, L. Hall, G. Ross, and P. West, Nucl. Phys.
B250 (1985) 225;
L. Girardello, M. Porrati, and A. Zaffaroni, hep-th/9704163, Nucl. Phys. B505 (1997)
272:
semi-realistic N=2 models.
15 F. Gliozzi, J. Scherk, and D.I. Olive, Nucl. Phys. B122 (1977) 253;
L. Brink, J.H. Schwarz, and J. Scherk, Nucl. Phys. B121 (1977) 77:
N=4 super Yang-Mills.
16 R. Grimm, M. Sohnius, and J. Wess, Nucl. Phys. B133 (1978) 275;
M.F. Sohnius, Nucl. Phys. B136 (1978) 461;
E. Witten, Phys. Lett. 77B (1978) 394:
covariant super Yang-Mills derivatives for extended supersymmetry.
17 W. Siegel, hep-th/9205075, Phys. Rev. D46 (1992) R3235; hep-th/9207043, Phys. Rev.
D47 (1993) 2504; loc. cit. (IIIB, ref. 3):
conformal chiral superspace.
18 Gates, Grisaru, Roček, and Siegel, loc. cit.
220 V. QUANTIZATION

PART TWO: QUANTA

Many important new features show up in field theory at the quantum level. Prob-
ably the most important is “renormalizability”, which states that all the parameters
(masses and couplings) that appear as coefficients of terms in the action must have
nonnegative mass dimension (when the massless part of the kinetic term has no di-
R
mensionful coefficient). Since the action is dimensionless, d4 x has dimension −4,
and the fields have positive dimension, this allows only a small number of terms for
any given set of fields. This one condition gives relativistic quantum field theory more
predictive power than any known alternative.
There are many perturbation expansions that can be applied to quantum field
theory. One is the mechanical JWKB expansion, which is an expansion in derivatives.
Of the inherently field theoretical expansions, the simplest is to expand directly in
fields, or equivalently, in the coupling constants. This expansion is the basis of per-
turbative quantum field theory. However, this expansion does not preserve gauge
invariance term by term. On the other hand, the terms in this expansion can be
collected into small subsets that do preserve gauge invariance. There are three such
regroupings, discussed in the four following chapters, and they are based on pertur-
bation expansions: (1) the field theoretic JWKB (“loop”) expansion, (2) expansions
in spin or helicity, and (3) expansions in internal symmetry (color or flavor).

V. QUANTIZATION
For the most part, integrals are hard to evaluate, in particular the path integrals of
exponentials that appear in quantum theory. The only exponentials that are generally
easy to integrate are Gaussians, and the products of them times polynomials, which
can in turn be evaluated as derivatives of Gaussians. Such integrals are the basis of
perturbation theory: We keep the quadratic part of the action, but Taylor expand the
exponential of higher-order terms. Effectively, this means that we not only expand
in orders of h̄ to perturb about the classical theory, but also expand in orders of
the coupling constants to perturb about the free theory. This makes particularly
useful our analysis of relativistic quantum mechanics (as free field theory). The
JWKB expansion for the wave function (or S-matrix) expands the exponent in powers
of h̄, dividing it onto three qualitatively different parts: (1) negative powers of h̄
(generally 1/h̄ only), which describe the classical theory (they dominate the classical
limit h̄ → 0), whose physical implications have been considered in previous chapters;
 A. GENERAL 221

(2) h̄-independent, where almost all of the important (perturbative) quantum features
appear (including topological ones, and quantum breaking of classical symmetries);
and (3) positive powers, which give more quantum corrections, but little new physics,
except when summed to all orders. These are generally known as “trees”, “one-loop”,
and “multiloop”, because of their graphical interpretation.

............................ ............................
............................ A. GENERAL ............................

In the Schrödinger approach to quantum mechanics one solves a differential equa-

tion. The Feynman approach is complementary: There one performs an integral.
R
Integrals are solutions to differential equations (e.g., f 0 = g ⇒ f = g), but usually
differential equations are easier to solve than integral equations. However, there is an
important exception: Gaussian integrals are easy, and so are their boundary condi-
tions. In field theory the most important approximation is one where the integrand
is approximated as a Gaussian, and the exact integral is evaluated as a perturbation
about that Gaussian. Of course, solving the corresponding differential equation is
also easy, but in that case the integral is easier because it corresponds to working
with the action, while the differential equation corresponds to working with the field
equations.
A major advantage of Feynman’s approach is that it allows space and time to
be treated on an equal footing. For example, as in classical electrodynamics, we
can solve the wave equation inside a spacetime volume in terms of conditions on the
boundary of that volume: It is not necessary to choose the spatial boundary at infinity
so that it can be ignored, and divide the temporal boundary into its “future” and
“past” halves so that all conditions are “initial” ones imposed at the past boundary.
It is not even necessary to distinguish between preparation (“if”) and measurement
(“then”) when describing probabilities: We can instead ask the probability of a given
wave function describing the whole boundary. This is a particular advantage for
relativistic quantum field theory, where space and time are more closely related than
in nonrelativistic theories. We now “review” Feynman’s approach for general quantum
systems, and quantum mechanics in particular, so that it can be applied without
further explanation when we come to quantum field theory.

1. Path integrals
In Feynman’s path integral approach to quantum mechanics (based on an analogy
of Dirac), the action is the starting point for quantization. The basic idea is to begin
222 V. QUANTIZATION

with the basic quantity in quantum mechanics, the transition amplitude, and write
it as an integral of the action
Z
hf|ii = Dφ e−iS[φ]
R
where Dφ is a “functional integral”: Integrate over φ(t) for each t (with some
appropriate normalization). The boundary conditions in t are defined by the choice
of initial and final states. In this subsection we will define this integral in a more
explicit way by breaking up the time interval into discrete points and taking the
continuum limit; in the next subsection we will study ways to evaluate it using its
general properties.
The path integral can be derived from the usual Hamiltonian operator formal-
ism. Considering for simplicity a single coordinate q, the wave function is given in
coordinate space by
Z
dq
ψ(q) = hq|ψi, |ψi = √ ψ(q)|qi
2π
where we use the convenient normalizations
Z Z h √ √ i
dq dp
√ |qihq| = 1 = √ |pihp| hq|q 0i = 2πδ(q − q 0), hp|p0 i = 2πδ(p − p0 )
2π 2π
for coordinate and momentum space. To describe time development, we work in
the Heisenberg picture, where time dependence is in the operators (and thus their
eigenstates):
ψ(q, t) = hq, t|ψi
Time development is then given completely by the “propagator” or “Green function”
Z
dqi
G(qf , tf ; qi , ti) ≡ hqf , tf |qi , ti i ⇒ ψ(qf , tf ) = √ G(qf , tf ; qi , ti)ψ(qi, ti )
2π
Excercise VA1.1
Let’s review the relationship between time development in the Heisenberg and
Schrödinger pictures. Using the usual relation

hψ|Q(t)|χi ≡ hψ(t)|Q|χ(t)i

between the time-independent states |ψi and time-dependent operators Q(t)

of the Heisenberg picture and the time-dependent states |ψ(t)i and time-
independent operators Q of the Schrödinger picture, define time-dependent
eigenstates in two ways:

hq(t)|ψ(t)i ≡ hq|ψi
Q|qi = q|qi ⇒
ψ(q, t) ≡ hq|ψ(t)i ≡ hq, t|ψi
 A. GENERAL 223

Given the time development of a state

|ψ(t)i = U(t)|ψi

find the development of O(t), |q(t)i, and |q, ti, and show in particular that
|q(t)i =
6 |q, ti. Which is the eigenstate of Q(t)?
In general, even for time-dependent Hamiltonians, we can find the infinitesimal
time development explicitly from the definition of the time derivative and the time-
dependent Schrödinger equation:

[i∂t − H(−i∂q , q, t)]hq, t| = 0

⇒ hq, t + | = hq, t|{1 − iH[P (t), Q(t), t]} ≈ hq, t|e−iH[P (t),Q(t),t]
and similarly for hp, t + | (where P and Q are the Hilbert-space operators). To
derive the path-integral formalism, we then iterate this result to obtain finite time
development by inserting unity infinitely many times, alternating between coordinate
and momentum,
Z
dp dq dp
hqf , tf |qi , tii = √ 0 √ 1 √ 1 ...hqf , tf |...
2π 2π 2π
...|p1, ti + 3ihp1 , ti + 3|q1, ti + 2ihq1 , ti + 2|p0 , ti + ihp0 , ti + |qi, ti i
to obtain successive infinitesimal exponentials,
Z
dp0 dq1 dp1
√ √ √ ...hqf |e−iH ...e−iH |p1 ihp1 |e−iH |q1 ihq1 |e−iH |p0 ihp0 |e−iH |qi i
2π 2π 2π
where the time dependence follows from the previous equation. However, note that
all the implicit time dependence of the Heisenberg picture drops out, because we
extracted the e−iH ’s, putting all the factors of each matrix element at the same
time: Although each matrix element is evaluated at time  earlier than the one to its
immediate left, each is of the form

ha, t|e−iH[P (t),Q(t),t]|b, ti = ha|e−iH[P,Q,t]|bi

(where |bi ≡ |b, tii, etc.), leaving only any explicit time dependence that may appear
in the Hamiltonian, effectively translating the other t’s → ti . Then we only need to
know
hq|pi = eipq , hp|qi = e−ipq
to evaluate the matrix elements in the path integral as
Z
dp dq dp
√ 0 √ 1 √ 1 ...exp{−i[qip0 +H(p0 , qi )−q1p0 +H(p0 , q1)+q1 p1 +H(p1 , q1)+...]}
2π 2π 2π
224 V. QUANTIZATION

More explicitly, this result is

Z Y dpn Y dqn
N −1 N −1
hqf , tf |qi , tii = Dp Dq e−iS , Dp Dq = √ √
n=0
2π n=1 2π
X
N −1
S= {−(qn+1 − qn)pn + [H(pn, qn ) + H(pn , qn+1 )]}
n=0
q0 = qi , qN = qf ; tf − ti = 2N

The classical picture is a segmented path, with the particle traveling along a
straight line segment from point qn to point qn+1 with momentum pn : Each q is
associated with a point, while each p is associated with the line segment connecting
two consecutive points. In the “continuum” limit  → 0, N → ∞, tf − ti fixed,
Z tf
.
S= dt[−qp + H(p, q, t)]
ti
(We have dropped some terms in hq|H|pi and hp|H|qi from reordering the operators
Q and P in H(P, Q) to apply P |pi = p|pi and Q|qi = q|qi. These commutator terms
alternate in sign, combining to give terms of order 2 , and can be dropped in the
continuum limit.)
More generally, we can evaluate an arbitrary transition amplitude as
Z Z
dqf dqi
A = hf|ii = √ √ ψf *(qf )hqf , tf |qi, ti iψi (qi ) = Dp Dq ψf *(qf )e−iS ψi (qi)
2π 2π
where now
Y
N −1
dpn Y dqn
N
Dp Dq = √ √
n=0
2π n=0 2π
Note that we can combine the initial and final wave function, as
Z
Ψ(qi, qf ) ≡ ψf *(qf )ψi (qi) ⇒ A = Dp Dq Ψ(qi , qf )e−iS

The complex conjugation of ψf vs. ψi is due to the complex conjugation involved in

time reversal (as seen, e.g., when comparing an eigenstate of p at the inital time to
the same eigenstate at the final time). In field theory, where the “p’s and q’s” are
functions of space as well as time, if we choose the boundary in space also to be finite,
so that the space and time boundaries form a single connected and closed boundary,
then Ψ is simply a function of the q’s over all that boundary.
Finite time development can also be written in operator form: From the above
derivation of the path integral, by integrating back out the insertions of unity im-
mediately after extracting the infinitesimal exponentials and translating the time of
each matrix element to zero, we find
hq, tf |q, tii = hqf |U(tf , ti )|qii
 A. GENERAL 225

  Z tf 
−iH(tf −) −iH(ti+) −iH(ti)
U(tf , ti ) = e ···e e ≡ T exp −i dt H(t)
ti
which defines the “time-ordered product” T . This is effectively a Schrödinger-picture
expression (all the P ’s and Q’s are at the initial time), and can also be derived in
that picture by solving for the time dependence of any state |ψ(t)i. If H is time
independent and we have a (orthonormal) basis of eigenstates of H, we can write
X 0
H|Ii = EI |Ii ⇒ U(t, t0) = |IihI|e−i(t−t )EI
I

2. Semiclassical expansion
The path integral formulation is especially suited for semiclassical approxima-
tions: The Bohr-Sommerfeld quantization rule follows from the fact that the func-
tional integral is invariant under S → S + 2πn, since S appears only as e−iS ; in that
sense the action is more like an angle than a single-valued function. The JWKB ex-
pansion follows from S → S/h̄ and expanding in h̄. This expansion can be interpreted
as an expansion in (space and time) derivatives, since it leads in the usual way to
the identification p = −ih̄∂/∂x and E = ih̄∂/∂t. One way to apply it is: (1) Find
a classical solution to the equations of motion. This gives the leading contribution
√
in h̄. (2) Expand about the classical solution as φ = φcl + h̄∆φ. Expanding in
∆φ, the first term gives the classical contribution, while the linear term vanishes by
the equations of motion. The quadratic term gives an h̄-independent contribution
to the exponential, which is easy to integrate since it is a Gaussian. The boundary
conditions are ∆φ = 0 at ti and tf . The normalization of the integral can be deter-
mined by comparing the free case, or considering the limit where the initial and final
times converge. (3) Higher orders in h̄ can be evaluated by perturbation about the
Gaussian integral: Polynomials times Gaussians are also easy to integrate.
As an example, consider the free particle. The separability of the action translates
into factorization of the functional integral, so the result can be found from the one-
dimensional case. As usual,
. xf − xi
L = − 12 mx2 ⇒ xcl (t) = xi + (t − ti)
tf − ti
where we have written the classical solution in terms of the variables appropriate to
the initial and final states, namely xi for an initial state localized there at time ti,
and xf , tf for the final state. Since the classical action is itself quadratic, so is its
expansion:
Z
(xf − xi )2 .
S = Scl + ∆S, Scl = − 2 m
1
, ∆S = − dt 12 m(∆x)2
tf − ti
226 V. QUANTIZATION

Since ∆S is independent of xf , xi, the functional integral of e−i∆S can be determined

easily by checking the limit tf → ti . The final one-dimensional result is then
s
−im im(xf −xi )2 /2(tf −ti )
hxf , tf |xi, ti i = e
tf − ti

where we have used

√
2πδ(x) = lim √1 e−x /2
2

→0

(one way of defining a Dirac δ function) to normalize

√
hxf , t|xi, ti = 2πδ(xf − xi )

Note that we have been sloppy about the definition of the “integration measure”:
In going from the Hamiltonian form of the action to the Lagrangian form, we ignored
some m dependence. Specifically, if we start with the Hamiltonian form, as derived
in the previous subsection, and derive the Lagrangian form by integrating out p, we
find the 1/m in H = p2 /2m leads to

Y
N −1
dp Y dxn
N −1 Y
N −1
dx
√n √ → mN/2 √ n
n=0
2π n=1 2π n=1
2π

The m(N −1)/2 then cancels similar factors from the N − 1 x-integrals, while the re-
√
maining m is that found in the final result above.
If we had considered a more general Hamiltonian, as in subsection IIIA1, where
p2 appeared as 12 g ij (x)pi pj , then we would have obtained a measure of the form (for
i = 1, ..., D)
Y
N −1
dD xn
[det g(x0 )det g(xN )]−1/4 p
(2π)D/2 det g(xn)
n=1
p
(We have averaged g as g(x)p2 → g(xn)g(xn+1 )p2n , since xn is associated with the
point n while pn is associated with the link from n to n + 1.) Such measure factors
are easy to recognize, since they are always local: If we included it in the action, it
would be a term proportional to
Y X Z
ln det g(xn ) =  1
 ln det g(xn ) ∼ δ(0) dt ln det g(x(t))
n n

In practice we just drop all such factors throughout the calculation, and fix the
normalization at the end of the calculation. Since the Lagrangian form follows from
the Hamiltonian form, which was properly normalized, we know such factors will
cancel anyway. Auxiliary fields can require similar factors for proper normalization;
 A. GENERAL 227

then such factors are simply the Jacobians from the field redefinitions from a form
where they appeared with trivial quadratic terms.
The Gaussian integral for the free particle can also be performed explicitly, by
using the discretized Hamiltonian path integral of the previous subsection. Besides
the Gaussian integrals considered in subsection IB3, we will also need to evaluate
Gaussians with linear terms:
Z
dD x −xT Sx/2+j T x T −1

D/2
e = (det S)−1/2 ej S j/2
(2π)
Z D
d z* dD z −z† Hz+z† j+j † z † −1

D
e = (det H)−1 ej H j
(2πi)
from shifting the integration variables (x → x + S −1 j, etc.) to eliminate the linear
terms, then using the previous results. These identities are sufficient for explicit eval-
uation of path integrals for quadratic actions. They are also useful for perturbation
about Gaussians: For functions multiplying the Gaussians, x can be replaced with
∂/∂j (and similarly for z) and then pulled outside the integral. (If a linear term is
not included, it can be introduced, and the result can be evaluated at j = 0.)
Excercise VA2.1
Generalize the above results for integration of Gaussians with linear terms to
the cases with fermionic and mixed (subsection IIC3) integration variables.
Excercise VA2.2
The path integral for the free, nonrelativstic particle can be evaluated much
more easily using the Hamiltonian form of the action. First consider the
Gaussian integral Z ∞
2 /2
dx eipx−x
−∞
as a special case of the Gaussians already evaluated, and use it to derive the
identity Z ∞
dx eipx = 2πδ(p)
−∞
(The  thus acts as a regulator to make the integral well defined.) Then use
the discretized expression of subsection VA1, and evaluate the x integrals
first. All but one of the p integrals then can be trivially evaluated, the last
giving a Fourier transform.
Excercise VA2.3
Evaluate the path integral for the harmonic oscillator, to find the result
r ( )
−imω imω[ 12 (x2f + x2i )cos ωt − xf xi ]
hxf , ti + t|xi , tii = exp
sin ωt sin ωt
228 V. QUANTIZATION

The two lowest orders in h̄ are then given by the above Gaussian integrals, the
classical contribution coming from the classical action S(x0) evaluated at the classical
solution S 0(x0) = 0, and the first quantum correction coming from the determinant
of S 00(x0 ). (The fact that the “S” in this exponent is imaginary will be treated in
subsection VA5.) Higher orders in h̄ come from expanding the exponential in the
S 000(x0) and higher-derivative terms.
Excercise VA2.4
Consider the nonrelativistic JWKB expansion for the propagator (for an ar-
bitrary Hamiltonian H) to the first two orders in h̄, writing it as
√
G ≈ ρe−iS/h̄

a Show the corresponding orders in the time-dependent Schrödinger equation

at t > 0 can be written as the classical equation of motion for the action S
and the (probability) current conservation law for the (probability) density ρ
(“Hamilton-Jacobi equations”),
 
. ∂ ∂H .
H = S, i
ρ +ρ=0
∂q ∂pi
when the argument p of H is evaluated at
∂S
pi = − i
∂q
(Assume a symmetric ordering of p’s and q’s in the quantum H.) Compare
the relativistic case examined in excercise IIIA4.1.
b The propagator is expressed in terms of q and q0 , where G(q, q0, t) ∼ δ(q − q0)
at t = 0, so the first order in h̄ is found by using the solution to the Hamilton-
Jacobi equations to write the classical action in terms of the “final” position
q and initial position q0. (In principle; in general even the classical equations
may be too difficult to solve analytically.) However, the Hamiltonian is given
as a function of p and q. Show that the change in variables from q, p to q, q0
gives
∂H ∂ 2S ∂ 2S
= −(M −1 )ij j , (M)ij = i j
∂pi ∂q0∂t ∂q0∂q
Show that
ρ = det (−i h̄1 M)
(the “van Vleck determinant”) solves the current conservation law, using the
explicit expression for (det M)M −1 given in subsection IB3. Check the nor-
malization, using the initial condition for propagators (or comparing to the
free case).
 A. GENERAL 229

c Consider the special case of the harmonic oscillator, and show the result agrees
with that of the previous problem. (Hint: First solve the classical equations
of motion for x(t), then rewrite it in terms of xi = q0 and xf = q; plug into
Scl = S and apply the above.)

3. Propagators
This free amplitude we just evaluated is the free propagator or Green function
for the Schrödinger equation. Explicitly, we define

G(q, t; q 0, t0) ≡ θ(t − t0)hq, t|q 0, t0i

where we have included the factor θ(t − t0) (1 for t > t0, 0 otherwise) to enforce that
the final time is later than the initial time (retarded propagator). This satisifies the
free case of the general defining equation of the propagator

[∂t + iH(−i∂q , q)]G(q, t; q 0, t0) = [−∂t0 + iH(i∂q0 , q 0)]G(q, t; q 0, t0)

√
= 2πδ(q − q 0)δ(t − t0 )

where we have used

∂t θ(t − t0) = δ(t − t0 )

and the facts that G without the θ factor is a homogeneous solution of the Schrödinger
equation (no δ’s) and becomes a δ in x for small times. The propagator then gives a
general solution of the Schrödinger equation as
Z Z
dq 0 0 0 0 0 dq 0
hq, t| = √ hq, t|q , t ihq , t | ⇒ ψ(q, t) = √ G(q, t; q 0, t0)ψ(q 0, t0)
2π 2π
√
In particular, for ψ(q, t0) = 2πδ(q − q 0) at some time t0 for some point q 0, ψ(q, t) =
G(q, t; q 0, t0) at all later times. For the example of the free particle in one dimension,
r
0 0 0 −im im(x−x0 )2 /2(t−t0)
G(x, t; x , t ) = θ(t − t ) e
t − t0

This solution for the propagator is not unique; as usual, a first-order differential
equation needs one boundary condition. Another way to say it is that the inhomoge-
neous differential equation is arbitrary up to a solution of the homogeneous equation.
We have eliminated the ambiguity by requiring that the propagator be retarded, as
incorporated in the factor θ(t − t0); using instead −θ(t0 − t) would give the advanced
propagator. This has an interesting translation in terms of the Fourier transform,
230 V. QUANTIZATION

which is another way to solve differential equations, and which replaces the so-called
“time-dependent” Schrödinger equation with the “time-independent” one. Defining
Z
dq dt
ψ̃(p, E) = √ √ e−i(qx−Et) ψ(q, t)
2π 2π

and similarly for G̃, we have, for any H(q, p) without explicit time dependence,
√
−i(E − H)G̃(p, E; p0 , E 0 ) = 2πδ(p − p0 )δ(E − E 0 )
i √
⇒ G̃ = 2πδ(p − p0 )δ(E − E 0)
E−H

t<0 E

H+iε
H−iε

t>0

Now inverse Fourier transforming, we have an ambiguity in integrating E past

the pole at E = H. We therefore shift the pole slightly off the real axis, so we can
integrate exactly on the real axis. Closing the contour by adding to the real axis a
semicircle of infinite radius in either the complex upper- or lower-half-plane, wherever
convergent, we find
Z
dE −iEt i
e = ±θ(±t)e−iHt
2π E − H ± i
which gives either the retarded or advanced propagator depending on the choice of
sign for the infintesimal constant  (retarded for E − H + i). To perform the inverse
Fourier transform, we note that the exponent needs an infinitesimal negative part to
make the integral convergent:
Z
i
dt eiEt(±)θ(±t)e−iHt∓t =
E − H ± i

Excercise VA3.1
Show that
i i
− = 2πδ(x)
x + i x − i
by three methods:
a Use the above result for the Fourier transform.
 A. GENERAL 231

b Show that this is the contour integral definition of the δ function, which is
actually a distribution, by integration, multiplying by an arbitrary (nonsin-
gular) function and integrating along the real axis. (Hint: Push the poles
onto the real axis, shifting the contours along with them, to find the integral
of a single function along the difference of two contours.)
c Prove the identity (checking the normalization)
2
lim = 2πδ(x)
→0 x2 + 2

After doing the E integrals, we have the half-transformed propagator for the
particle
√ 0 2
Ĝ(p, t; p0 , t0) = θ(t − t0 ) 2πδ(p − p0 )e−i(t−t )p /2m

in the retarded case, which we also could have found easily by solving the differential
equation directly. The remaining p integrals are then simple Gaussians.

4. S-matrices
In the interacting case, the amplitude we get from the path integral is the inter-
acting propagator. However, to be able to take the limit describing time development
between infinite initial and final times, we need to choose boundary conditions such
that the initial and final basis states have the time dependence of free particles, de-
scribed by H0 , assuming that the particle behaves freely at such asymptotically large
times. This is called the “interaction picture”, to distinguish from the Heisenberg
picture, where the states have no time dependence, and the Schrödinger picture,
where the states have the complete interacting time dependence. We thus evaluate
the limiting amplitude
Z
dq dq
A = t lim hψ f (tf )|ψi(ti )i = lim √ f √ i ψf *(qf , tf )hqf , tf |qi , tiiψi (qi , ti )
i →−∞ ti →−∞
2π 2π
tf →+∞ tf →+∞

for the interaction-picture states |ψ(t)i, relating the interaction-picture coordinate

basis 0hq, t| to the Heisenberg-picture basis hq, t| (with initial conditions 0 hq, 0| =
hq, 0| ≡ hq|):
 R tf 
−itH0 −i dt H
0 hq, t| = hq|e ⇒ hqf , tf |qi , tii = 0 hqf , tf |e itf H0
T e ti
e−iti H0 |qi , ti i0

 R tf 
−i dt H
ψ(q, t) = 0 hq, t|ψi ⇒ A = hψf |S|ψii, S = t lim
→−∞
e itf H0
T e ti
e−iti H0
i
tf →+∞
232 V. QUANTIZATION

The “S(cattering)-matrix” S then describes the time development between infinite

initial and final times, appropriate for describing scattering from a potential of finite
spatial extent.
The fact that time development conserves probability (H = H † ) is reflected in
the corresponding unitarity condition for the S-matrix:

S †S = 1

A more complicated condition is causality: The basic idea is that interactions take
place in chronological order. (A stronger statement of causality will be found in the
relativistic case: that any interaction should take place at a spacetime point, rather
than just at a single time. It follows from this weaker one in relativistic theories,
since event B is later than event A in every Lorentz frame only when B is in A’s
lightcone.) Causality is the condition that the Hamiltonian at any time involves only
variables evaluated at that time. (H(t) is a function of only φ(t), all at the same
time t, where φ = p, q are the quantum variables appearing in the Hamiltonian.) A
nice way to describe the interactions is by introducing a classical background as we
did for the semiclassical expansion of path integrals, such as by φ(t) → φ(t) + χ(t),
where χ is just some function. The important point is that we have shifted φ(t) by
χ(t) at the same t, so as not to disturb causality. We then consider the effect on the
S-matrix of modifying the background χ by a function δχ localized (nonvanishing) at
some particular time t, and a function δχ0 localized at t0, such that t > t0. Picking
out the δχ pieces in the time-ordered product, we can therefore write

S[χ + δχ + δχ0] = U(f, t)V(t)U(t, t0)V(t0)U(t0 , i)

S[χ + δχ] = U(f, t)V(t)U(t, t0)U(t0 , i)

S[χ + δχ0] = U(f, t)U(t, t0)V(t0)U(t0 , i)

S[χ] = U(f, t)U(t, t0)U(t0 , i)

where U(t0, i) is the time-development operator from time ti to time t0 (including the
factor with H0 ), V(t0) is the extra factor in the time development at time t0 resulting
from the function δχ0 localized there, etc. Then we easily find

S[χ + δχ + δχ0] = S[χ + δχ]S −1[χ]S[χ + δχ0]

⇒ (S −1 [χ + δχ]S[χ + δχ + δχ0] − I) − (S −1 [χ]S[χ + δχ0] − I) = 0

 
δ † δ
⇒ S[χ] S[χ] = 0 for t > t0
δχ(t) δχ(t0)
 A. GENERAL 233

using the infinitesimal functions δχ and δχ0 to define derivatives.

In general, it is not possible to solve the Schrödinger equation for the propagator
or the S-matrix exactly. One approximation scheme is the perturbation expansion in
orders of the interaction:
R Z tf
−i dt H −i(tf −ti )H0
H = H0 + V ⇒ T (e )=e + dt e−i(tf −t)H0 [−iV (t)]e−i(t−ti)H0
ti
Z tf Z t
0 0
+ dt dt0 e−i(tf −t)H0 [−iV (t)]e−i(t−t )H0 [−iV (t0)]e−i(t −ti )H0 + ...
ti ti
Z ∞
⇒ Sf i ≡ hf|S|ii = hf|ii + dt hf, t|[−iV (t)]|i, ti
−∞
Z ∞ Z t
0
+ dt dt0 hf, t|[−iV (t)]e−i(t−t )H0 [−iV (t0)]|i, t0i + ...
−∞ −∞

The first term in S is just the identity (i.e., the free piece). All the other terms
consist of a string of interactions (−iV ) connected by free propagators (e−itH0 , where
t is the time between the interactions), with each interaction integrated over all
time (subject to time-ordering of the interactions), and the initial/final state (wave
function) evaluated at the initial/final interaction time.
Excercise VA4.1
Assume the initial and final states are eigenstates of the free Hamiltonian:

H0 |ii = Ei |ii, H0 |fi = Ef |fi

Assuming V has no explicit time dependence, explicitly evaluate the time

integrals in the S-matrix, effectively Fourier transforming from time to energy,
to find
1
Sf i = hf|ii − 2πiδ(Ef − Ei )hf|(E − H0 ) (E − H0 )|ii|E=Ei
E − H + i
(Hints: Redefine the integration variables to be the times between interac-
tions. Taylor expand 1/(E − H + i) in V for comparison.)
In field theory we want to express any state in terms of a basis of products of
1-particle states, so we can calculate the behavior of these specified particles. We
try to do this by using field variables (the “q’s” of field theory): Each field operator
should produce a single particle. Unfortunately, this is not the case: An asymptotic
state of given 3-momentum created by such a field operator is not necessarily an
eigenstate of the energy, because such a state can be either 1-particle or n-particle,
due to interactions. The propagator for the field is then of the form
X 0
Ĝ(p, t; p0 , t0) ∼ δ(p − p0 ) ψ*I (p)ψI (p)e−i(t−t )EI (p)
I
234 V. QUANTIZATION

X
nI X
nI
EI (p) = EI,i (pi ), pi = p
i=1 i=1
P
where “EI,i (pi )” is the energy of a 1-particle state (the I will include an integral in
general). However, as long as all particles have masses, such an asymptotic 1-particle
state is distinguishable as that of lowest energy E0: The higher-energy states are
n-particle states to which this particle can couple. (If some of the n-particle states
were lower energy, the 1-particle state could decay into them, and thus the 1-particle
state would be unstable, and not asymptotic. With massless particles things are more
complicated: Then 1-particle states are more difficult to define and to measure.) In
principle, we could define the 1-particle states by constructing the corresponding
operator, consisting of the field plus terms higher order in the fields; in practice, this
is rather complicated.
A simpler way to make the asymptotic states unambiguous is by modifying the
definition of the S-matrix:
 R tf 
−i dt H
S= lim e itf H0
T e ti
e−iti H0
ti →−∞(1+i)
tf →+∞(1+i)

introducing factors of 1 + i for some positive , which may be chosen small for
convenience. (Actually, we can generally replace 1 + i with just i if it is not too
confusing: The result is the same.) The effect is seen by considering a matrix element
of particular fields that may be a superposition of different energies E, but evaluated
between states of energy E0 . Since E ≥ E0 , the time dependence of any such matrix
element is proportional to

i(tf −ti )(E−E0 ) 1 for E = E0
Sf i ∼ lim e =
ti →−∞(1+i)
t →+∞(1+i)
0 for E > E0
f

(For simplicity we have assumed energy conservation, so initial and final energies are
the same.) Alternatively, we can simply impose E = E0 directly in the definition:
 R tf 
−i dt H(t)
S = t lim
→−∞
e itf H0
δH(tf ),H0 T e ti δH(ti),H0 e−iti H0
i
tf →+∞

where the free Schrödinger equation H0 = E0 defines E0 for the initial and final states,
and δH,H0 is evaluated by examining the asymptotic time-dependence of the time-
development operator with respect to ti and tf : Normally field theory is calculated in
energy-momentum space, working with the spacetime Fourier transform of the above,
where this amounts to simply comparing energies Ef = Ei = E0.
 A. GENERAL 235

If we know some details of the interaction, this modification may be irrelevant: In

particular, in local quantum field theory interactions happen at a point in space and
time. For example, consider the inner product between a 1-particle state in its rest
frame and a related n-particle state, which appears in the same propagator. Because
of locality, the wave function for the n-particle state, when evaluated in position space
(which is where the theory is local) is simply the product of n 1-particle wave functions
evaluated at the same point. But we know that for small relative momenta (where a
nonrelativistic approximation holds) that the individual wave functions propagate as

|ψ| ∼ |t − t0|−(D−1)/2

from the form of the free 1-particle propagator. (Or, we can use dimensional analysis,
and consider the spread of a particle of restricted range of momenta from a confined
region: Then |ψ|2 ∼ 1/V and the volume V ∼ |t − t0|D−1 .) This implies that the
n-particle wave function will fall off as the nth power of that, so in the limit of large
times the 1-particle state will dominate. In a relativistic theory the length scale
associated with this fall-off will be associated with the masses involved, and thus at
a subatomic scale.

5. Wick rotation
In the previous subsection we ensured convergence in the definition of the S-matrix
by effectively making the “coordinate change”

t → (1 − i)t = e−i t

in the definition of the limit

(1 − i)t → ∞ ⇒ t → (1 + i)∞

This affected the time-development operator as

e−iHt → e−iHt−t

for H > 0 to pick out the ground state H = 0. The same effective substitution was
made in subsection VA3 in defining the contour integral for the propagator:
Z Z
dE −iE(1−i)t i dE −iEt i
e = e
2π E−H 2π (1 + i)E − H
Z Z
dE −iEt i dE −iEt i
= e = e
2π E − (1 − i)H 2π E − H + i
236 V. QUANTIZATION

which is the same as the substitution

E → (1 + i)E = ei E

since essentially E = i∂/∂t.

In general, having to do contour integrals and keep track of i’s in propagators is
inconvenient. Fortunately, there is a simple way in practical calculations to get rid
of not only the i’s but (almost) all the other i’s as well. The method is known as
“Wick rotation”. The basic idea is to extend the above complex rotation from angle
 to angle π/2:
t → −it = e−iπ/2 t, E → iE

pushing the contour even farther away from the singularities. Thus, the Schrödinger
equation is changed to a “diffusion equation” (to describe, e.g., Brownian motion):

(i∂t − H)ψ = 0 ⇒ (∂t + H)ψ = 0

For example, for the free particle the resulting equation has no i’s. The time-
independent Schrödinger equation then becomes

(E − H)ψ = 0 ⇒ (iE − H)ψ = 0

The result for the propagator is then

Z ∞
dE −iEt 1
e = θ(t)e−Ht
−∞ 2π H − iE
Now no i prescription is needed, since the pole was moved away from the real axis.
Similar remarks apply to the inverse Fourier transform
Z ∞
1
dt eiEtθ(t)e−Ht =
−∞ H − iE

Excercise VA5.1
Find the Wick-rotated retarded propagator G(x0 , t0; x, t) for the free (1D)
particle, satisfying
√
(∂t + H)G = (−∂t0 + H 0 )G = 2πδ(x − x0 )δ(t − t0 )

Furthermore, if we define the S-matrix directly in this Wick-rotated space

 R tf 
− dt H
S = t lim
→−∞
e tf H0
T e ti
e−ti H0
i
tf →+∞
 A. GENERAL 237

then the limiting procedure is unambiguous even in field theory, since


−(tf −ti )(E−E0 ) 1 for E = E0
lim
ti →−∞
e =
t →+∞
0 for E > E0
f

Another important effect is on actions. For example, in the mechanics path

.
integral for a particle with kinetic term T = 12 mx2 in a potential U(x), we integrated
Z
−iS
e : S = dt(U − T )

Upon Wick rotation, this becomes

Z
−S
e : S= dt(U + T )

The major change on the exponent −S is that it is now not only real, but negative
definite. (For physical purposes, we assume the potential has a lower bound, which
can be defined to be nonnegative without loss of generality.) Thus, the semiclassical
approximation we made earlier, called the “stationary phase” approximation, has now
become the “steepest descent” approximation, namely fitting e−S/h̄ to a Gaussian,
which is approximating the integral by the places where the integrand is largest. We
thus write

S(x) = S(x0 ) + 12 (x − x0)2 S 00(x0 ) + ..., S 0(x0 ) = 0, S 00(x0 ) > 0

for one variable, with the obvious generalization to many variables. Explictly, we
then have Z 

dx −S(x)/h̄ 1 
√ e ≈ p e−S(x)/h̄ 
2πh̄ 00
S (x)  0
S (x)=0

plus higher orders in h̄, expressed in terms of higher derivatives of S. In the case
of many variables, S 00 is replaced with a determinant, as for the Gaussian integrals
of subsection IB3, and for functional integrals, with a functional determinant. (But
sometimes the functional determinant can be replaced with an ordinary determinant:
See excercise VA2.4.)
So now we can first calculate everything in Wick-rotated spacetime, where everyt-
ing is real (more precisely, classical reality properties are preserved quantum mechan-
ically), and then Wick rotate back to find the correct result in physical spacetime. In
particular, the appropriate ’s, still needed to correctly position the singularities in
physical spacetime, can be restored by rotating back through an angle 12 π − :

inverse W ick : t → (i + )t = ei(π/2−)t, E → (−i + )E = e−i(π/2−) E

238 V. QUANTIZATION

REFERENCES
1 N. Wiener, J. Math. and Phys. Sci. 2 (1923) 132:
Euclidean path integrals for Brownian motion.
2 P.A.M. Dirac, Phys. Z. der Sowj. 3 (1933) 64:
proposed path integrals for quantum mechanics.
3 R.P. Feynman, Rev. Mod. Phys. 20 (1948) 367:
formulated path integral approach to quantum mechanics.
4 R.P. Feynman, Phys. Rev. 84 (1951) 108:
path integrals in phase space.
5 R.P. Feynman and A.R. Hibbs, Quantum mechanics and path integrals (McGraw-Hill,
1965):
review of path integrals; interesting side stuff, like D=2.
6 R. Shankar, Principles of quantum mechanics, 2nd ed. (Plenum, 1994):
good quantum mechanics text that includes path integrals.
7 J.H. van Vleck, Proc. Natl. Acad. Sci. USA 14 (1928) 178.
8 P.A.M. Dirac, Proc. Roy. Soc. A136 (1932) 453, P.A.M. Dirac, V.A. Fock, and B.
Podolosky, Phys. Z. Sowj. 2 (1932) 468:
interaction picture.
9 J.A. Wheeler, Phys. Rev. 52 (1937) 1107;
W. Heisenberg, Z. Phys. 120 (1943) 513, 673:
S-matrix.
10 E.C.G. Stückelberg, Helv. Phys. Acta 19 (1946) 242;
E.C.G. Stückelberg and D. Rivier, Helv. Phys. Acta 24 (1949) 215;
E.C.G. Stückelberg and T. Green, Helv. Phys. Acta 24 (1951) 153:
causality in quantum field theory.
11 N.N. Bogoliubov, Doklady Akad. Nauk USSR 82 (1952) 217, 99 (1954) 225:
explicit condition of causality on S-matrix, using functionals.
12 L.S. Brown, Quantum field theory (Cambridge University, 1992) p. 293:
decoupling of asymptotic multiparticle states in field-theory propagators.
13 M. Kac, On some connections between probability theory and differential and integral
equations, in Proc. 2nd Berkeley Symp. Math. Stat. Probability, ed. J. Neyman, 1950
(University of California, 1951) p. 189;
E. Nelson, J. Math. Phys. 5 (1964) 332:
Wick rotation of path integral in quantum mechanics.
14 G.C. Wick, Phys. Rev. 96 (1954) 1124:
Wick rotation in quantum field theory.
 B. PROPAGATORS 239

....................... .......................
....................... B. PROPAGATORS .......................

Classically we distinguish between particles and fields (waves). This can be con-
sistent with a classical limit of a quantum theory if there is a conserved charge associ-
ated with the classical particles, with respect to which the classical fields are neutral.
Then we can have continuous worldlines for the particles: The statement that the
worldlines do not end or split is associated with charge conservation. The interaction
between the particles and fields is described by modifying the particle (mechanics)
action (and not the action for the fields). If we look at just the mechanics action,
the modification is the same as considering external fields (like external potentials in
nonrelativistic mechanics), since we are ignoring the field action, which is needed for
the field equations.
The action for the fields then can be added separately. Coupling to such external
fields is a simple way to study properties of particles without applying field theory.
For example, in nonrelativistic mechanics it helps to explain charge and spin, which
don’t appear explicitly in the free Schrödinger equation.

1. Particles
All the information in quantum mechanics is contained in the propagator, which
gives the general solution to the Schrödinger equation, and can be obtained by the
Feynman path integral. Here we discuss the free propagator for the spinless particle
(whose classical description was given in section IIIB), which is the starting point for
relativistic perturbation theory.
We consider quantization first in the Lorentz covariant gauge v = 1. Except for
the T integration, the same methods can be applied as in the nonrelativistic case.
The simplest expression (and ultimately the most useful one) is obtained by Fourier
transforming with respect to x: In comparison to the multidimensional nonrelativistic
result
0 2
Ĝ(pi , t; p0i , t0) = δ(p − p0 )θ(t − t0)e−i(t−t )p /2m
(where here δ(p − p0 ) = (2π)(D−1)/2δ D−1(pi − p0i ) for D − 1 spatial dimensions), the
relativistic result is
Z
G̃(p, p ) = dT δ(p − p0 )θ(T )e−iT (p +m )/2
0 2 2

(where now δ(p − p0 ) = (2π)D/2δ D (pa − p0a ) for D spacetime dimensions).

There are several simple yet important differences from the nonrelativistic case:
(1) The dependence on the mass m is different. In particular, we can set m = 0 only in
240 V. QUANTIZATION

R
the relativistic case. (2) There is an additional integration dT , because the variable
T , which is the remaining part of v, survives the gauge v = 1. This is analogous to
the time integral in the nonrelativistic case for G̃(pi , E; p0i , E 0 ), if we set the energy
to zero. This is as expected, since the relativistic classical mechanics differs from the
nonrelativistic one mainly by constraining the “Hamiltonian” 12 (p2 + m2) to vanish.
This interpretation also leads to the “zero-energy” version of the inhomogeneous
(proper-)time-independent Schrödinger equation for this case,

−i 12 ( − m2 )G(x, x0) = δ(x − x0)

(3) The propagator is automatically “retarded” in the “proper time” T , as a conse-

quence of the positivity condition v > 0, which was motivated by the geometrical
interpretation of v as the worldline metric. When used in this manner to write the
propagator in terms of a Gaussian, T is known as a “Schwinger parameter”.
Generally, it is convenient to remove the momentum δ-function (which resulted
from translational invariance) as

G(x, x0) = ∆(x − x0) ⇒ G̃(p, p0 ) = δ(p − p0 )∆(p)

Z
∆(p) = dT θ(T )e−iT (p +m )/2
2 2

where we have simply written ∆(p) for the Fourier transform of ∆(x) (dropping the
tilde). Performing the T integral, using the same methods as for the t integral in the
nonrelativistic case, we have the final result
−i
∆(p) = 1 2
2 (p + m2 − i)

Actually, this result is almost obvious from solving the relativistic wave equation. The
only part that is not obvious is the “i prescription”: how to perform the contour
integration upon Fourier transformation. In the nonrelativistic case, we saw two
obvious choices, corresponding to retarded or advanced propagators; the classical
action did not distinguish between the two, although the retarded propagator has the
obvious convenience of determining later events from earlier ones. On the other hand,
in the relativistic case the choice of propagator was fixed from classical considerations.
T is restricted to be positive, and the i is needed to make the T integral converge.
Excercise VB1.1
Take the nonrelativistic limit of the relativistic propagator, and compare with
the propagator of nonrelativistic quantum mechanics. Explain the difference
in terms of the nonrelativistic limit of the classical mechanics action.
 B. PROPAGATORS 241

Excercise VB1.2
Perform the analysis of excercise VA2.2 for the relativistic particle. First
replace the integration over T by a sum: Instead of dividing up the time into
2N intervals of length  and taking the limit N → ∞,  → 0, with 2N fixed,
P
sum 2 ∞ N =0 , and then take the limit  → 0. (2N is now T instead of tf − ti,
and we integrate over it instead of keeping it fixed.) Perform all x integrals
and then all but the last p integral before summing over N. Again, the entire
calculation is much easier than using the Lagrangian (second-order) form of
the path integral.
To understand this point better, we examine the Fourier transformation with
respect to time. In contrast to the nonrelativistic case, there are now two poles, at
 
p i 1 1
p = ±ω, ω = (pi )2 + m2 :
0
∆= −
ω p0 − (ω − i) p0 + (ω − i)
where now pa = (p0 , pi ). These are also the two classical values of the canonical energy
(as opposed to the true energy, which is the absolute value), which we saw previously
corresponded to particles and antiparticles. With our prescription for integrating
around the poles, using the same methods as in the nonrelativistic case, we then find
1 0
Ĝ(pi , t; p0i , t0) = (2π)D/2 δ D−1(pi − p0i ) e−iω|t−t |
ω
1 0 0
= (2π)D/2δ D−1(pi − p0i ) [θ(t − t0)e−iω(t−t ) + θ(t0 − t)eiω(t−t )]
ω
We now see that the particles (p0 = ω) have a retarded propagator, while the antipar-
ticles (p0 = −ω) have an advanced propagator. This is the quantum version of the
classical result we saw earlier, that particles travel forward in time, while antiparticles
travel backward.
We next compare quantization in the lightcone gauge. Whereas in the covariant
gauge the analog to the nonrelativistic time t was the “proper time” T , the analog
is now the lightcone “time” τ . Since τ = x+/p+ , we have E = p− p+ (E = i∂/∂τ
vs. p− = i∂/∂x+), and thus
i −i
∆(p) = = 1 2
E − 2 (p + m ) + i
1 i2 2
2 (p + m − i)
2

as before. Note that this derivation was almost identical to the nonrelativistic one:
Unlike the covariant gauge, we did not have to add in T as a separate variable of
integration (but not path integration). However, this Schwinger parameter is useful
for evaluating momentum integrals and analyzing momentum dependence. This is a
typical characteristic of unitary gauges: They are more useful for keeping track of
degrees of freedom.
242 V. QUANTIZATION

2. Properties
As in electrodynamics, the free scalar satisifes a differential equation second-order
in time, so the propagator is used differently from nonrelativistic quantum mechanics
to give a general solution to the wave equation. We use the identity
I ↔
Z ↔
Z
d σ A ∂ m B = d x ∂ · (A ∂ B) = dD x [A( − m2)B − B( − m2)A]
D−1 m D

H
where “ dD−1 σ m” is the integral over the closed surface bounding the volume inte-
R
grated over in dD x. In practice we take the volume to encompass all spacetime in
the limit, neglect the part of the boundary at spacelike infinity, and choose the parts
of the boundary at timelike infinity to be surfaces at constant time, so the boundary
integrals are over just space:
I ↔
Z ↔
Z ↔
Z ↔
∞
dD−1 m
σ A ∂ mB = d D−1
x A ∂ t B|−∞ = d D−1
x A ∂ t B|∞ − dD−1 x A ∂ tB|−∞

We then have the solution for the wave function inside the volume in terms of that
on the boundary:

( − m2)ψ = 0,−i 12 ( − m2)G(x, x0) = −i 12 ( 0 − m2 )G(x, x0) = δ(x − x0 )

I D−1 0m ↔
d σ 0 1 0 0
⇒ D/2
G(x, x ) 2 i ∂ m ψ(x ) = ψ(x)
(2π)
where the wave equation for ψ is the Klein-Gordon equation.
Similarly, this defines a conserved current from any two wave functions
↔
∂ · (ψ1* ∂ ψ2) = ψ1*( − m2)ψ2 − ψ2( − m2)ψ1* = 0

or, evaluating the integral over a volume infinite in space but infinitesimal in time,
the conserved charge Z ↔
d
dD−1 x ψ1 * ∂ t ψ2 = 0
dt
This leads to the covariant inner product
Z ↔
dD−1 x
h1|2i = (p )
0
ψ 1 * 1
2 i ∂ t ψ2
(2π)D/2
where the (p0) appears because the contour integral gives a + at later times (positive
energy) and a − at earlier times (negative energy). Explicitly, we find for the inner
product of plane waves
ψp (x) = hx|pi = eip·x

⇒ hp|p0 i = (2π)D/2−1δ D−1(pi − p0i )(p0 ) 12 (p0 + p00)

 B. PROPAGATORS 243

We have used p2 + m2 = p02 + m2 = 0, which also implies that |p0 | = |p00|: Thus,
the inner product vanishes if the waves have opposite-sign energy, while for the same
sign (p0 ) 12 (p0 + p00 ) = |p0 |. The result then can be written manifestly covariantly as
δ(p − p0 )
hp|p0 i = = θ(p0 p00)ω(2π)D/2−1 δ D−1(pi − p0i )
2πδ[ 12 (p2 + m2)]
Similarly, the solution for the wave function in terms of the Green function gives
only positive-energy contributions from the part of the surface at earlier times, and
only negative-energy contributions from the part of the surface at later times. More
general on-shell wave functions, since they depend on only D − 1 spatial momenta
and the sign of the energy, can be written as a restricted Fourier transform
Z
ψ(x) = dp 2πδ[ 12 (p2 + m2)]eip·xψ̃(p)
Z
⇒ h1|2i = dp 2πδ[ 12 (p2 + m2 )]ψ̃1(p)*ψ̃2 (p)

(Here ψ̃(p)* means to complex conjugate after Fourier transforming to p-space; oth-
erwise, we need to change the sign of the argument.) In particular, for a plane wave
we have
δ(p − p0 )
ψ̃p0 (p) =
2πδ[ 12 (p2 + m2)]
It will prove useful later to have a collection of solutions to the homogeneous and
inhomogeneous Klein-Gordon equations, and compare them in 4-momentum space
and time-3-momentum space. Using the previous nonrelativistic and relativistic re-
sults, we find
∆: −i/(p2 + m2 − i) ⇒ θ(t)e−iωt +θ(−t)eiωt = e−iω|t|
∆*: i/(p2 + m2 + i) ⇒ θ(t)eiωt +θ(−t)e−iωt = eiω|t|
∆R : −i/(p2 + m2 − ip0) ⇒ θ(t)e−iωt −θ(t)eiωt = −2iθ(t)sin(ωt)
∆A : i/(p2 + m2 + ip0 ) ⇒ θ(−t)eiωt−θ(−t)e−iωt = 2iθ(−t)sin(ωt)
∆+ : θ(p0 )2πδ(p2 + m2 ) ⇒ θ(t)e−iωt +θ(−t)e−iωt = e−iωt
∆− : θ(−p0 )2πδ(p2 + m2) ⇒ θ(t)eiωt +θ(−t)eiωt = eiωt

where we have omitted certain common factors (see above). ∆± satisfy the homo-
geneous equation, while the rest satisfy the inhomogeneous one. (These are easily
checked in the mixed space, where the Klein-Gordon operator is −(∂t2 + ω 2 ).) This
table makes explicit which sign of the energy propagates in which time direction, as
well as the linear relations between the momentum-space expressions. In particular,
we see that ∆+ propagates just the positive-energy states, while ∆− propagates just
the negative-energy ones.
244 V. QUANTIZATION

Excercise VB2.1
The relativistic propagator uses a particular choice for integrating around the
two poles in the complex energy plane, as encoded in the i prescription. If
we ignored the classical determination of that prescription, there would be
four simple choices, integrating either above or below the two poles. Show
these four choices can be enforced by replacing i in p2 + m2 − i with

i, −i, ip0 , −ip0

and derive the results of the table above. Give explicit expressions for the
four propagators in position space in four dimensions for the massless case.
We can check the propagator’s behavior by explicit evaluation, using plane waves:
Z
dD−1 x0 0 1
↔
0 0 1 −iω|t−t0 | i~p·~x−ip0 t0
0
(p ) D/2
∆(x − x ) 2 i ∂ 0 1 0
t ψp (x ) = (p ) 2 (p + i∂t ) e e
(2π) ω

= θ[p0 (t − t0)]ψp(x)

where we have used the previous result for Ĝ(~p, t; ~p0 , t0) (and thus ∆(~p, t)). Again
we see that the propagator propagates positive-energy solutions forward in time and
negative-energy backward.
This propagator also applies to relativistic field theory. (See subsection IIIA3 for
nonrelativistic field theory.) In comparison to the nonrelativistic case, the propagator
is now −i/ 12 (p2 + m2) instead of −i/( 2m
1
~p 2 − E), and this determines the kinetic term
in the field theory action:
Z Z
S0 = − dx 12 φ 12 ( − m )φ =
2
dx 14 [(∂φ)2 + m2 φ2]

To make the functional integral of e−iS0 converge, we replace m2 → m2 − i, which

is the same i prescription found in first-quantization. Note that we have used a real
√
field φ* = φ. (A complex field can be used by doubling ψ = (φ1 + iφ2)/ 2.) This is
possible only in the relativistic case because we have both positive-energy solutions
e−iEt as well as negative ones e+iEt . (In other words, the relativistic Schrödinger
equation is a second-order differential equation, so we get two i’s to make the kinetic
operator real.) Reality simply means identifying particles with antiparticles. (E.g.,
there is no “antiphoton” distinct from the photon.)
 B. PROPAGATORS 245

3. Generalizations
More generally, we will find propagators of the form (in momentum space)
i
∆ = − , K = K†
K
corresponding to free actions Z
S0 = dx 12 φKφ

where K = − 12 ( −m2) in the case just considered. Then the inner product is defined
as above in terms of the Green function by using the relation
I Z
0
(p ) d σ A Mm B = i dD x [(KA†)B − A† KB]
D−1 m †

to define Z
dD−1 σ m †
h1|2i = ψ1 Mm ψ2
(2π)D/2
and thus I
dD−1 σ 0m
ψ(x) = (p ) 0
D/2
G(x, x0)Mm
0
ψ(x0)
(2π)
↔
where Mm = (p0 ) 12 i ∂ m above. (For the usual equal-time hypersurfaces, we use
M0 = −M 0. There may be additional implicit matrix factors in the Lorentz-invariant
inner product A† B.) This inner product gives a nonnegative norm on physical bosonic
states, but on physical fermionic states it is negative for negative energy, because
ordering the initial state to the left of the final state (the wrong ordering for quantum
mechanics) produces a minus sign from the anticommutativity of the fermions. (From
the explicit integral, this appears because K is generally second-order in derivatives
for bosons, but first-order for fermions, so Mm has one factor of p0 for bosons and
none for fermions.)
For physical fields, the (free) field equation will always imply the Klein-Gordon
equation (after gauge fixing for gauge fields). Thus, the propagator can always be
written as
i N(p)
∆ = − = −i 1 2
2 (p + m − i)
K 2

in terms of some matrix kinematic factor N(p). Using this expression for the propa-
gator in our above position-space inner product, this implies

ψp (x) = ψ̂(p)eip·x ⇒ N(p)Mm (p)ψ̂(p) = −(p0 )pm ψ̂(p)

If we choose a basis that is orthonormalized with respect to all quantum numbers

other than momenta (spin/helicity and internal symmetry), we then have
δ(p − p0 )
hp, i|p0 , ji = δij
2πδ[ 12 (p2 + m2 )]
246 V. QUANTIZATION

If we ignore coordinate/momentum dependence and focus on just the other quantum

P
numbers, then it is clear that the extra factor of N is just the sum i |iihi| (so
N|ii = |ii, etc.). The positive-energy propagator is then given by a sum over all
positive-energy states:
X
∆+ = N(p)θ(p0 )2πδ[ 12 (p2 + m2 )] ⇒ N(p) = [(p0)]2s ψ̂ †i (p)ψ̂i (p)
i

where we have included an extra sign factor for negative energy and half-integer spin
from the reordering of states, as explained above. The fact that K is not simply the
Klein-Gordon operator is a consequence of unphysical (gauge/auxiliary) degrees of
freedom appearing in the action: Then N is a projection operator that projects out
the auxiliary degrees of freedom on shell, and the gauge degrees of freedom on and
off (in unitary gauges), as represented above by a sum over physical states. However,
more general N’s are sometimes used that include unphysical degrees of freedom;
these must be canceled by “ghosts”, similar unphysical degrees of freedom of the
opposite statistics.
Excercise VB3.1
1
Demonstrate all these properties for spin (helicity) 2 (see chapter II):
a For the massless case, use twistors for the solution to the field equation to
find . .
(N) .(p) = pα p̄ . = (p0 )p ., (M .)γ δ = δαγ δ δ.
αβ β αβ αβ β

b Do the same for the massive Dirac spinor, to find

N = (p0 )(p/ + m
√
2
), Mm = γm

(Hint: Consider the rest frames for p0 > 0 and < 0.)
Excercise VB3.2
Use the construction of excercise VB1.2 to define the path integral for the
spinning particle of excercise IIIB1.2. Show that in the covariant gauge v = 1,
λ = constant, the propagator can be written as
Z 1 2 −iγ · p i
∆(p) ∼ dξ dT θ(T )e−T 2 p −iξγ·p ∼ 1 2 =
2p γ ·p
R
up to some arbitrary normalization factor, where ξ = dτ λ is the only gauge
R
invariant part of λ (as T = dτ v is for v).
We will find that quantum corrections modify the form of the propagator. In
particular, it may modify the position m2 of the pole in −p2 and its residue, as well
 B. PROPAGATORS 247

as adding terms that are analytic near that pole. For example, consider a scalar
propagator of the form
N
∆(p) = −i 1 +R
2 (p2 + m2 − i)

where N is a constant and R is analytic in p. By the procedure of “renormalization”,

N can be set to 1, and m2 can be set to its original value (see chapter VII). Alter-
natively, we can cancel N in the normalization of external states, and redefine the
masses of these states to coincide with what appears in the propagator.
Excercise VB3.3
Use this propagator to define the inner product between two plane waves,
and evaluate it explicitly. Show that R gives no contribution, and the plane
√
waves need factors of N to maintain their normalization. (Hint: What is
the wave equation corresponding to ∆, and how is it related to Mm ? You can
also consider the relation of N to ∆+ .)
Away from the pole, at higher values of −p2 than m2 , there will also be cuts cor-
responding to mutiparticle states. Although these higher-energy intermediate states
in the propagator will contribute to the time development even for on-shell states
(those satisfying p2 + m2 = 0 asymptotically), in S-matrix elements we can ignore
such contributions on external lines, using our modified definition for evaluating the
asymptotic limit for the S-matrix (see subsection VA4).
Excercise VB3.4
Consider the general scalar propagator
Z ∞
ρ(µ)
∆(p) = −i dµ 1 2 , ρ(µ) = δ(µ − m) + θ(µ − 2m)σ(µ)
2 (p + µ − i)
2
0

which contains a pole at mass m and contributions from multiparticle states

at mass 2m and higher. Fourier transform from energy to time. Use this
propagator to define the time development of a momentum eigenstate satis-
fying the free wave equation asymptotically, using the 1 + i prescription of
subsection VA4 to define the asymptotic limit:
Z ↔
ψ(t) ≡ lim G(t, ti ) 12 i ∂ ti ψ0 (ti)
ti →−∞(1+i)

and show that σ(µ) does not contribute:

( − m2)ψ0 = 0 ⇒ ψ(t) = ψ0(t)

248 V. QUANTIZATION

4. Wick rotation
As in the nonrelativistic case, the i prescription can also be fixed by the infinites-
imal Wick rotation (see subsection VA5)
1 1
t → (1 − i)t, E → (1 + i)E ⇒ →
p2 + m2 p2 + m2 − i
However, in the relativistic case, a finite Wick rotation gets rid of not only i’s but
also the annoying minus signs associated with the Minkowski metric. We now replace
all timelike coordinates, including proper time, with spacelike coordinates:

t → −it, τ → −iτ

In addition, for every vector field V a we replace

V 0 → −iV 0 (V i → V i )

and similarly for tensor fields. (Here we have defined Wick rotation in the first-
quantized sense: on all explicit coordinates and momenta, as well as on explicit
Lorentz indices. For example, in the field theory action we rotate the explicit deriva-
tives and the integration measure, rather than the arguments of the fields.)

Euc.
p0
−ω+iε
Min.
+ω−iε

Note that E as defined in the nonrelativistic case was −p0 , which is the same as
p0 only in Minkowski space:

p0 → −ip0 , p0 → +ip0

Furthermore, there is some apparent ambiguity in how to change the integration

measure, corresponding to how the integration contours are rotated (i.e., changes
in the limits of integration). In particular, we see from subsection VA5 that the
contour rotation for E is actually in the opposite direction of that for t, consistent
R
with Fourier transformation. (Effectively, we keep the extra i for dp from rotating
p0 , while dropping the −1 from p0 ↔ p0 because of the usual absolute value in the
 B. PROPAGATORS 249

R
Jacobian in real changes of variables.) The net result is the naive change for dx,
R
while that for dp preserves the inverse Fourier transform:
Z Z Z Z
dx → −i dx, dp → i dp; δ(x) → iδ(x), δ(p) → −iδ(p)

When manipulating explicit expressions, the factors of i on coordinates/momenta

and fields can be transferred to the constant tensors contracting their indices: The
net effect is that the Wick rotation is equivalent to changing just τ , the integration
measures, the flat-space metric, and the Levi-Civita tensor:
∂ ∂
τ → −iτ, →i
∂τ ∂τ
Z Z Z Z Z Z
dτ → −i dτ, dx → −i dx, dp → i dp

ηmn → δmn, abcd → −iabcd

So now the inner product is positive definite: We have gone from Minkowski space
to Euclidean space.
For example, for the relativistic particle in the gauge v = 1, the propagator is
now Z
1
∆(p) = dT θ(T )e−T (p +m )/2 = 1 2
2 2

2
2 (p + m )

The integral is automatically convergent because p2 +m2 is now positive definite. If we

examine our transformation of timelike components in terms of the complex p0 plane,
we see that we have just rotated the contour from the real axis to the imaginary axis,
through an angle π/2, which avoids the poles at p0 = ±(ω − i). (There is actually
a slight cheat in the massless case, since the two poles converge near vanishing 3-
momentum, where ω = 0, but this problem can be avoided by an appropriate limiting
procedure.) Note that in the relativistic case the Euclidean propagator is completely
real also in momentum space; even the overall −i has been killed. (Compare the
nonrelativistic case in subsection VA5, where −i/(H−E) → 1/(H−iE).) This follows
from first Wick rotating in position space, then performing the Fourier transform as
R
usual (avoiding an extra −i from rotating the dt in the Fourier transform).
Excercise VB4.1
Find the propagator in time-3-momentum space, Ĝ(pi , t; p0i, t0), after Wick
rotation. (I.e., Wick rotate ∆(p) first, then Fourier transform.)
Excercise VB4.2
Find the massless propagator in 4D Minkowski coordinate space, including
the i, by
250 V. QUANTIZATION

a Fourier transforming the Schwinger-parametrized momentum-space propaga-

tor in Minkowski space (including the “Minkowski” τ ),
b doing the same entirely in Euclidean space, and then Wick rotating the time
back to Minkowski space, and
c Fourier transforming both of the above cases without using the Schwinger
parameter, first doing the energy integrals as in the previous section. (Hint:
Use rotational invariance to point ~x in a particular direction to simplify the
angular integration.)
Excercise VB4.3
Although the propagator in momentum space is most useful for scatter-
ing of plane waves, its position-space dependence is more useful for bound
states/scattering of localized sources:
a Evaluate the Wick-rotated propagator in arbitrary dimensions D for large
√
x = x2 by (1) Fourier transforming the Schwinger-prametrized form, (2) per-
forming the (Gaussian) p integration before the T , and (3) using the steepest
descent approximation on the exponential to approximate the T integral (but
don’t bother sticking the power of T multiplying the exponential in it as a
log). Why does this approximation correspond to large x? (It may be useful
to make the redefinition T → (x/m)τ .) You should find the result
√
∆(x) ≈ 2πm(D−3)/2x−(D−1)/2e−mx

After Wick rotating back, the exponential becomes a phase inside the light-
cone, while outside it gives exponential damping, as in quantum mechanical
barrier penetration. (In the massless case, there is damping away from the
lightcone both inside and outside, but only by powers, determined by dimen-
sional analysis.)
b Show the above result is exact in D=1 and 3 by (1) Fourier transforming
without the Schwinger parameter, (2) performing a contour integral over the
magnitude of p by closing it appropriately and picking up the contributions at
the poles at p = ±im, and (3) for D=3, fixing xm in a particular direction and
doing the angular part of the p integration. (This method can also be used
to obtain the exact corrections to the above in terms of elementary functions
for higher odd D.) For the physical case of the potential produced by a static
√
point source in 3 spatial dimensions, we find ∆(x) = 2πe−mx /x, so at short
range the Coulomb potential is unmodified, while it is exponentially damped
at range 1/m.
 B. PROPAGATORS 251

c Check the results for b by taking the massless limit, and comparing to the
analogous result using the method of a, but doing the T integral exactly
for those cases. (Warning: The result is infinite in D=1, and some type of
“regularization” must be used to subtract an infinite constant, leaving a finite
x-dependent remainder.)
The corresponding effect on the action, where we path-integrated
Z
−iS . .
e : S = dτ 12 (vm2 − v −1xm xn ηmn )

is to integrate the Wick rotated expression

Z
−S . .
e : S = dτ 12 (vm2 + v −1xm xnδmn )

This method also applies to relativistic field theory. Wick rotation t → −it of the
kinetic term Z
−iS0
e : S0 = dx 14 [η mn (∂m φ)(∂nφ) + m2φ2]

now gives Z
−S0
e : S0 = dx 14 [δ mn(∂m φ)(∂n φ) + m2 φ2]

This has an interesting consequence in the complete action (including interac-

tions). Positivity of the energy implied the non-time-derivative terms in the action
had to be positive, but the time-derivative terms are now the same sign in the Wick-
rotated action, effectively the same as adding an extra spatial dimension: The energy
T + U is the same as the Wick-rotated Lagrangian (−T + U → T + U). So we can
replace the condition of positivity of the energy with positivity of the Wick-rotated
action, which we need anyway for path-integral quantization. (Note that for the
particle this again requires v ≥ 0.)
Unfortunately, the simple results for Wick rotation obtained here for spin 0,
although they generalize to spin 1, do not work so simply for other spins. (Consider,
e.g., trying it in spinor notation.) However, the method can still be applied to the
coordinates, and simplifies momentum integrals, since one can avoid contours, i’s,
and Minkowski minus signs.

REFERENCES
1 V.A. Fock, Phys. Z. Sowj. 12 (1937) 404;
E.C.G. Stückelberg, Helv. Phys. Acta 14 (1941) 322, loc. cit. (IA);
Y. Nambu, Prog. Theo. Phys. 5 (1950) 82,
R.P. Feynman, Phys. Rev. 80 (1950) 440:
quantization of relativistic mechanics in terms of proper time.
252 V. QUANTIZATION

2 J. Schwinger, Phys. Rev. 75 (1949) 651, 82 (1951) 664:

his parameters.
3 J. Polchinski, Commun. Math. Phys. 104 (1986) 37:
gives a more rigorous discussion of path-integral quantization of the relativistic particle.
4 Stückelberg, loc. cit. (IA);
R.P. Feynman, Phys. Rev. 76 (1949) 749, 769:
relativistic propagator.
 C. S-MATRIX 253

............................ ............................
............................ C. S-MATRIX ............................

A relativisitic quantum field theory is defined by three properties: (1) Poincaré

invariance (“relativistic”) is the basic result of special relativity. Its consequences
have been accurately observed both macroscopically and (sub)microscopically, and
no violations are known. (2) Unitarity (“quantum”) is the main mathematical result
of quantum mechanics: Any quantum theory can be considered as the corrections
to the classical theory implied by unitarity. This is one way to define perturbation
theory, and is equivalent to the usual (JWKB) expansion in h̄. (The other major
axioms of quantum mechanics are concerned with the physical interpretation of the
quantities calculated, such as the preparation and measurement of states.) Quantum
mechanics also has been accurately verified, with no observed violations. (3) Causal-
ity (“field theory”) appears in many areas of physics, formulated in many ways. The
strongest way to state causality, in a way independent of special relativity and quan-
tum mechanics, is as locality: All interactions happen at a point; there is no action at
a distance. This means that any force applied by an object at one point in spacetime
on another elsewhere(/when) must be mediated by yet another object that carries the
effect of that force between the two. The most accurate verifications of this principle
have been through the predictions of relativistic quantum field theory.
We now define the perturbation expansion of the S-matrix, and give its general
properties.

1. Path integrals
Since general transition amplitudes involve also wave functions, we also Taylor
expand them: In field theory, this is an expansion in the number of particles. We
therefore write
X∞ Z
1 dσ1m ...dσN
n
Ψ[φ] = N!
ψN (x1, ..., xN )M1m ...MN nφ(x1)...φ(xN )
N =0
(2π)N D/2

where we have used the covariant inner product of subsection VB3. (The surfaces of
integration are at t = ±∞.) We have also used the free N-particle wave function ψN ,

K1 ψN = ... = KN ψN = 0

which is sufficient to describe particles at t = ±∞. Amplitudes for such asymptotic

states are elements of the S-matrix, as defined in the interaction picture (see subsec-
tion VA4). In practice we choose a particular value of N, and use a basis element for
254 V. QUANTIZATION

ψN , namely the product of N 1-particle wave functions:

Y
N Y
N
ψN (x1 , ..., xN ) = ψN i(xi ) + permutations ⇒ Ψ[φ] = hψN i |φi
i=1 i=1

In principle, if there are bound states in the theory, we can consider similar wave
functions, but besides φ we expand in the composite field describing the bound state.
It should be possible to discover such states by looking at the properties of the
amplitudes of the φ states. (For example, a two-particle bound state would show
up in the amplitude describing the scattering of two particles.) Since in a covariant
approach we treat particles and antiparticles on an equal footing, Ψ[φ] should include
both the initial and final wave function.
We therefore want to evaluate the path integral
Z
A = Dφ e−iS[φ] Ψ[φ]

Separating out the free and interacting pieces of the (gauge-fixed) action,
Z
S = S0 + SI = dx 12 φKφ + SI

and using the integration identity

Z Z
du −uM u/2 du −uM u/2 u∂v −1
√ e f(u + v) = √ e e f(v) ∼ e∂vM ∂v /2 f(v)
2π 2π

at v = 0, we can evaluate the path integral as

 Z  
1 δ 1 δ

A = exp −i dx 2 e −iSI [φ]
Ψ[φ]
δφ K δφ φ=0
Z Z
δ 1 δ δ δ
−i dx 12 = dx dx0 12 ∆(x − x0)
δφ K δφ δφ(x) δφ(x0)
We have absorbed the proportionality constant into the definition of Dφ, as we did
for semiclassical expansion in subsection VA2. In this case, this normalization is fixed
by the “free” part of the S-matrix.
It will prove convenient to distinguish the ends of propagators that attach to SI
from those that attach to Ψ, so using the identity

f(∂x )g(x, x + y) = f(∂x0 + ∂y0 )g(x0, y 0) (x0 = x, y 0 = y + x)

evaluated at y = 0, we rewrite this expression as

 Z  
δ 1 δ δ 1 δ δ 1 δ 
A = exp −i 12 + + 12 e−iSI [φ] Ψ[ϕ]
δφ K δφ δφ K δϕ δϕ K δϕ φ=ϕ=0
 C. S-MATRIX 255

We then evaluate the derivatives that act on only one or only the other:
 Z  
δ 1 δ  A
A = exp −i Z[φ]Ψ̃[ϕ] Ψ Z
δφ K δϕ
  Z  φ=ϕ=0 
δ 1 δ
Ψ̃[φ] = exp −i 2 1
Ψ[φ] Ψ e-iSI
δφ K δφ
  Z  
1 δ 1 δ −iSI [φ]
Z[φ] = exp −i 2 e
δφ K δφ

Then we move the differential operators into the wave functional:

  
δ 
A = Ψ̂ Z[φ]
δφ φ=0

   Z  
δ δ 1 δ 
Ψ̂ = exp −i Ψ̃[ϕ]
δφ δφ K δϕ ϕ=0

Z[φ] (the “generating functional” for the S-matrix) contains all propagators with
SI [φ]’s attached at both ends, and forms the basis of the perturbation expansion.
From the integration identity above, we can also write it as
Z
Z[ϕ] = Dφ e−i(S0 [φ]+SI [φ+ϕ])

R
Effectively, we have just taken the functional integral Dφ e−iS[φ] and separated the
field into a “quantum field” φ (the integration variable) and a “background field” ϕ,
where ϕ includes the asymptotic states, which propagate to infinity, while φ vanishes
at infinity (or at least goes to a constant) fast enough to allow the usual integration
by parts in performing the functional integral. Thus ϕ gives the boundary value of
the field. This is essentially the same as the general prescription for path integrals
given in subsection VA2, except that we take ϕ to be arbitrary for convenience of
functional differentiation, and we drop free ϕ terms, which were incorporated into Ψ̃
(i.e., we expand just SI ).
Ψ̃[φ] is the result of contracting some pairs of the one-particle wave functions
with propagators. This gives the usual inner product of those one-particle states:
Since for any one-particle wave function satisfying the free field equations we have
the propagator identity
Z
dσ m
ψ(x)Mm ∆(x − x0) = ψ(x0)
(2π)D/2
256 V. QUANTIZATION

such contractions give

Z Z
dσ m dσ 0n 0 0 0 0 dσ m
ψ(x)Mm ∆(x − x )Mn ψ (x ) = ψ(x)Mm ψ 0(x) = hψ|ψ 0i
(2π)D (2π)D/2
where the inner product vanishes unless the states have opposite energies (i.e., one is
incoming and one is outgoing). This is the boring part of the S-matrix element: It
represents the corresponding particles not interacting at all. (For example, in the free
case SI = 0 we have Z = 1, and A = Ψ̃|φ=0 consists of only such inner products.) For
most purposes we just factor out such free inner products, and consider only processes
where all particles interact.
Finally, the conversion from Ψ̃ to Ψ̂ replaces all the on-shell inner products with
integrals over all spacetime: Using the propagator identity from above,
  X ∞ Z
δ 1 dN D x δ δ
Ψ̂ = N! N D/2
ψ̃N (x1 , ..., xN ) ...
δφ (2π) δφ(x1) δφ(xN )
N =0

where ψ̃N appears in Ψ̃ exactly as ψN in Ψ:

X∞ Z
1 dσ1m ...dσN
n
Ψ̃[φ] = N! N D/2
ψ̃N (x1, ..., xN )M1m ...MN nφ(x1)...φ(xN )
(2π)
N =0

Usually we represent ψN as the product of single-particle wave functions,

  Y 
N Z  Y N Z  
δ δ δ 
Ψ̂N = ψN i ⇒ AN = ψN i Z[φ]
δφ δφ δφ 
i=1 i=1 φ=0

in which case Ψ̂ simply replaces each field φ in Z with one of these wave functions.

A
〈|〉 Z 1/K



 ∆ : SI ( )SI

∆→ 1 : ψ( )SI

 e-iSI
 h | i : ψ( )ψ
ψ
1
SI
 C. S-MATRIX 257

Thus, of the three types of propagators, only the ones that connected two factors
of SI remain, all inside Z; the ones that connect the wave functions to Z have been
replaced with just spacetime integrals, while those connecting the wave functions to
each other have become the usual spatial integrals for the Hilbert-space inner product.
Excercise VC1.1
Show that the amplitude can also be written as
  
δ 
A = Ψ̌ Z[φ]Ψ0[φ] , Ψ0 = e−hφ|φi/2
δφ φ=0

  X ∞ Z
δ 1 dN D x δ δ
Ψ̌ = N! N D/2
ψN (x1 , ..., xN ) ...
δφ N =0
(2π) δφ(x1) δφ(xN )
making all inner products of non-scattering particles explicit. We can inter-
pret Ψ0 as the free “vacuum wave functional” and ZΨ0 as the interacting
vacuum wave functional.

2. Graphs
Before giving applications of these rules, we consider a few general properties.
A convenient way to describe the terms in the expansion of the two exponentials in
Z is pictorially, by “Feynman diagrams/graphs”. Each factor of SI is a “vertex” of
the graph, represented by a dot in the diagram; each factor of the propagator 1/K
is a “link” in the graph, represented by a line connecting the two dots representing
the two factors of SI on which each δ/δϕ acts. (Both derivatives can also act on
the same factor of SI , giving a loop.) So any term in the expansion of Z is repre-
sented by a diagram consisting of a bunch of dots (interaction vertices) connected by
lines (propagators). When we want to “draw” the amplitude A itself, we also draw
additional lines, each with one end attached to a vertex and one end unattached.
These “external lines” represent the one-particle wave functions coming from ΨN ,
and not propagators (“internal lines”). While in the diagram for Z each vertex can
have ϕ dependence, in the diagram for A there is none, and the number of lines
(internal and external) coming from any vertex explicitly indicates the order in ϕ of
the corresponding term in SI .
The physical interpretation of these diagrams is simple: The lines represent the
paths of the particles, where they act free, while the vertices represent their collisions,
where they interact. These diagrams are generally evaluated in momentum space: We
then can associate a particular momentum with each line (propagator), and momen-
tum is conserved at each vertex. An arrow is drawn on each line to indicate the
258 V. QUANTIZATION

direction of “flow” of the momentum. (Otherwise there is a sign ambiguity, since

complex conjugation in position space changes the sign of the momentum.) Then the
sum of all momenta flowing into (or all out of) a vertex vanishes. The momentum
associated with a line is then interpreted as the momentum of that particle, with
the arrow indicating the direction of flow of the proper time τ . (p changes sign with
τ .) When evaluating an S-matrix element, the fact that the external-line wave func-
tions satisfy the free wave equation means the external momenta are on-(mass-)shell
(p2 + m2 = 0); on the other hand, this is not true of the momenta on the internal
lines, even though those particles are treated as free.
Some graphs in the S-matrix are disconnected: They can be divided into separate
parts, each with a subset of the particles that interact with each other but not with
the other subsets. For convenience, we consider only the connected graphs: If we
write  
−iW [ϕ] δ
Z[ϕ] = e , Ac = Ψ̃ −i (−i)W [ϕ]|ϕ=0
δϕ
then W is the generating functional for the connected S-matrix Ac . To prove this
relation between Z and W , we first note that it is just the combinatorics of the
graphs, and has nothing to do with spacetime. Therefore, it is sufficient to consider
the simple (unphysical) case where the action has no derivatives. Since the propagator
is then local, connectedness is equivalent in this case to locality. We then observe
that the lack of derivatives allows the functional integral to be factorized explicitly
into ordinary integrals at each point in spacetime:
Z R YZ
−i dx L(φ(x),ϕ(x))
Z[ϕ] = Dφ e = dφ(x) e−iL(φ(x),ϕ(x))
x
Y R
= Z(ϕ(x)) = e−i dx W(ϕ(x))
= e−iW [ϕ]
x
Thus, this W is local, and therefore connected; this implies W is connected in the
general case. The simplest kind of connected graph is a “tree” graph, which is a graph
that has no closed paths; the rest are called “loop” graphs.
“One-particle-irreducible” (1PI) graphs are defined to be those connected graphs
that can’t be disconnected by severing a single propagator. It then follows that any
connected graph can be represented as a generalized tree graph, whose “vertices”
(including two-point vertices) are actually 1PI graphs. We then define the “effective
action” Γ[φ] to be the classical action plus all 1PI loop graphs. Note that the vertices
of the original action are the 1PI tree graphs; thus Γ is also the classical kinetic term
plus all (tree and loop) 1PI graphs. Actually, since the 1PI tree graphs are −iSI , we
define the classical part of Γ to be S, but the quantum part to be the quantum 1PI
 C. S-MATRIX 259

part of W . Of course, the effective action is nonlocal. However, the tree graphs that
follow from this action are exactly all the connected graphs of the original action:
This is clear for all but the 2-point vertices from the definition. For the propagator
and its relation to the 2-point 1PI loop graphs, we simply compute the expression
following from Γ: Denoting the 2-point 1PI loop operator as A, the kinetic operator
of Γ is K + A. The propagator following from Γ is then
1 1 1 1 1 1 1
= − A + A A − ...
K +A K K K K K K
But this is exactly the result of the complete propagtor (including loop graphs) fol-
lowing from the original action.

Z
Γ
W
Γ
Γ

W W

This quantum modification of the propagator leads us to reanalyze our prescrip-

tion for evaluating the S-matrix: For example, even in the simplest case, where this
A is just a constant, the full propagator differs from the free propagator by a change
in the mass. This means the mass of asymptotic states should also be changed, which
invalidates part of our evaluation of the S-matrix in the previous subsection. Similar
problems occur when A is proportional to K, which changes the normalization of
asymptotic states. There are two ways to fix these problems: (1) We compensate by
modifying the kinetic term in the classical action, replacing K with K minus such
local contributions from A. Treating these new terms as part of the interaction in
our derivation of the S-matrix, so our normalization and mass in the free propagator
are unchanged, these “interaction” terms cancel the unwanted terms in the quantum
propagator, so it then has the same residue and pole as the free one. This procedure
is known as “renormalization”, and will be discussed further in chapter VII, primarily
260 V. QUANTIZATION

for the purpose of eliminating infinities. (2) Alternatively, we modify our derivation
of the S-matrix to take the full propagator into account. The easiest way to see this
change is to remember that by definition the S-matrix follows from treating the effec-
tive action as a classical action (except for its nonlocality and nonhermiticity), but
keeping only the “tree” graphs. Then clearly (a) the quadratic part Γ0 of Γ is used
to define the asymptotic states, and (b) instead of eliminating all free propagators
except those connecting factors of SI , we eliminate all full propagators (found from
Γ0 ) except those connecting factors of ΓI , the nonquadratic part of Γ. In other words,
we modify our earlier definition of the S-matrix by dropping all graphs that have any
quantum correction to external lines. Thus, this procedure can also be applied in
the case of renormalization; in fact, it should be applied in general, simply because
it allows us to immediately ignore many graphs. It also allows us to avoid confusion
resulting from attaching wave functions of the wrong mass to propagator corrections:
E.g., in momentum space, we would have to interpret ambiguous factors such as
δ(K)A(1/K)..., where the factor of δ(K) comes from a plane-wave wave function.

Γ Γ
Γ Γ Γ Γ Γ Γ
Γ Γ

This analysis of the quadratic part of Γ also leads us to examine the terms of
lower order: constant and linear. The constant term is just a normalization, and
should be dropped. (This is not true in the case of gravity, where a constant term in
the Lagrangian is not gauge invariant by itself.) The linear term describes the decay
of a particle into the vacuum: It implies we have the wrong vacuum. A linear term
necessarily has no derivatives (otherwise it is a boundary term, which vanishes by our
boundary conditions); it is part of the “effective potential” (a generalization of the
potential energy, whose contribution to a classical mechanics Lagrangian contains no
time derivatives; see subsection VIIB2). The existence of a linear term means that
the minimum of the effective potential, i.e., the true vacuum, is not described by
vanishing fields. To correct this situation we therefore apply the same procedure as
for the classical action (chapter IV): (1) Shift the appropriate fields by constants,
to put us at the minimum of the potential, and (2) use the new quadratic terms in
the potential to determine the true masses of states defined by perturbation about
this new vacuum. Again eliminating any constant terms, the resulting Γ has only
quadratic and higher-order terms.
 C. S-MATRIX 261

To summarize, the general procedure for calculating Feynman graphs is: (1) Cal-
culate the effective action, i.e., the 1PI graphs. (2) Shift the scalars to put them at the
minimum of the effective potential, dropping the resultant constant, to reveal the true
masses of all particles. (3) Calculate the S-matrix from diagrams without external-
line corrections, with external wave functions whose normalization and masses are
determined by the zeroes of the kinetic operators in the shifted Γ.
Another use of the effective action, besides organizing the calculation of the S-
matrix, is for studying low-energy behavior: This means we apply an expansion in
derivatives, as in first-quantized JWKB (see subsection VA2). Of most interest is the
lowest order in the approximation, where all fields are effectively constant: This gives
the effective potential. (In practice, “all fields” means just the scalars, since constant
spinor fields are not generally useful, while higher spins are described by gauge fields,
whose constant pieces can be set to vanish in an appropriate formulation: E.g., the
constant piece of the metric tensor can be attributed to a scalar — see subsection
IXA7.) However, the definition of “1PI” graphs is ambiguous, depending on how we
define “particle”: For example, if we include auxiliary fields in the effective action (as
in supersymmetry, but also for bound-state problems: see subsections VIIB3 and 6),
the result at fixed order in any expansion parameter (h̄, coupling, etc.) is different,
since the auxiliaries get contributions at each order, so elminating them by their
effective-action field equations mixes orders. (E.g., B 2 + h̄Bf(A) → h̄2 f 2 .) This is
crucial when the composite fields defined by these auxiliaries, and thus the auxiliaries
themselves, obtain vacuum values. Therefore, the effective action is most useful for
these purposes when, for appropriate choice of fields and definition of h̄, a useful
first-quantized semiclassical expansion can be found. Another important use of the
effective action is that it is gauge invariant (even in the nonabelian case, when using
the background-field gauge; see subsection VIB8): Sometimes simplifications due to
gauge invariance are thus easier to see in the effective action than in the S-matrix.
We now consider an interesting topological property of graphs: For any graph, if
we draw an extra propagator from a vertex to itself or another, that gives an extra
“loop” (closed circuit) and no extra vertices. Adding a 2-point vertex to the middle
of a propagator gives an extra propagator and no extra loops. Adding an external
line to a vertex changes nothing else. Since any nontrivial (not a lone propagator)
connected graph can be built up this way from a lone vertex, we find

P −V =L−1

for P propagators, V vertices, and L loops (and E external lines). The same result
follows from counting momentum integrals: In momentum space there is an inter-
262 V. QUANTIZATION

nal momentum, and corresponding integral, for each propagator, and a momentum
conservation condition, and corresponding δ function, for each vertex. The only in-
dependent momenta are the external ones (associated with each ϕ in Z[ϕ]) and one
momentum vector for each loop. Thus, after integrating out all the delta functions,
except for an overall momentum conservation δ function for each connected graph,
we are left with integrations over only the loop momenta. So, we are again led to the
above result.

3. Semiclassical expansion
We can define perturbations by inserting h̄’s in various ways, as discussed in sub-
section IIIA3. The h̄ that defines classical mechanics yields an expansion in derivatives
on “matter” fields (those that describe classical particles in the limit h̄ → 0, as op-
posed to the “wave” fields). This expansion is covariant as long as the h̄ multiplies
covariant derivatives. However, it can’t be applied to Yang-Mills fields, and it doesn’t
correspond to a diagrammatic expansion. On the other hand, the h̄ that defines clas-
sical field theory is an expansion in the number of “loops”, and allows us to group
graphs in gauge-invariant sets, since gauge transformations are not h̄-dependent: As
in quantum mechanics, we can perform a JWKB expansion by appropriately inserting
h̄:
Z  
i
Z[ϕ] = Dφ exp − (S0 [φ] + SI [φ + ϕ])
h̄
 Z 
1 δ 1 δ
= exp −ih̄ dx 2 e−iSI [ϕ]/h̄
δϕ K δϕ
The order in h̄ has a simple graphical interpretation. We see that there is a factor
of h̄ for each propagator and a factor of 1/h̄ for each vertex. Thus, by the above
topological identity, for each connected graph the power in h̄ is one less than the
number of loops. We therefore write
X
∞
Z[ϕ] = e−iW [ϕ]/h̄ , W= h̄L WL
L=0

where W0 generates the connected “tree” graphs, which have no loops.

We know that the leading term in the JWKB expansion is associated with the
classical theory. We can make this more explicit in the field theory case by finding the
general classical (perturbative) solution to the field equations from the tree graphs.
Graphically the solution is very simple: We replace one ϕ on each tree graph with a
propagator, and associate the end of the propagator with the position of the classical
 C. S-MATRIX 263

field Φ(x). If we then act on this Φ, which is a sum over all tree graphs, with K,
it cancels the propagator, leaving a bunch of Φ’s (also sums over all tree graphs)
connected at x, with the appropriate vertex factor. In other words, we find KΦ =
−δSI [Φ]/δΦ, the classical field equations.

φ
φ Φ
φ φ
K Φ =
φ φ
Φ
φ

To prove this, it’s convenient to again use functionals, to automatically keep track
of all combinatorics. The quantum field equations can be derived from the general
identity Z
δ
Dφ f[φ] = 0
δφ
since we only integrate functionals f that are assumed to fall off fast enough as
φ → ∞ to kill all boundary terms. (This follows from the perturbative definition of
the functional integral.) In particular, for any action S̃,
Z Z
δ −iS̃/h̄ δ S̃ −iS̃/h̄
0 = Dφ ih̄ e = Dφ e
δφ δφ
For our present purposes, we choose

S̃ = S0 [φ] + SI [φ + ϕ]
Z   Z  
δS0[φ] δSI [φ + ϕ] −iS̃/h̄ δ
⇒ 0 = Dφ + e = Dφ Kφ + ih̄ e−iS̃/h̄
δφ δφ δϕ
δW [ϕ]
⇒ Khφiϕ + =0
δϕ
where “hφiϕ ” is the expectation value of the field in a background:
R
Dφ φe−iS̃/h̄
hφiϕ = R
Dφ e−iS̃/h̄
R
(and, of course, Dφ e−iS̃/h̄ = e−iW/h̄ ).
We now examine the classical limit h̄ → 0 of this result by noting that if we
impose the free field equation on the background
δ S̃ δS[φ + ϕ]
Kϕ = 0 ⇒ =
δφ δφ
264 V. QUANTIZATION

where S[φ] = S0 [φ] + SI [φ] is the usual action: In other words, the field equations
following from S̃ are just the usual field equations for the complete field

φ̃ = φ + ϕ, ϕ = lim φ̃
x→∞

since we chose our boundary conditions so φ → 0 as x → ∞ (including |t| → ∞,

whereas ϕ → 0 only at spatial infinity). We then apply the stationary-phase approx-
imation (or, after Wick rotation, the steepest-descent approximation)
Z  

lim Dφ f[φ]e−iS̃/h̄ = f[φ]e−iS̃/h̄ 
h̄→0 δ S̃/δφ=0

for f = φ and 1 to find

δW0[ϕ] 1 δW0[ϕ]
KΦ + =0 ⇒ Φ=ϕ−
δϕ K δϕ
Φ ≡ limhφ̃iϕ = φ̃|δS[φ̃]/δφ̃=0
h̄→0
Thus, Φ is the solution to the classical field equations with boundary condition Φ → ϕ,
and can be found directly from the classical (0-loop) part of W by replacing one field
with a propagator.
A similar result holds for W0 itself, by taking the classical limit as above for
f[φ] = 1:
W0 [ϕ] = S0 [φ] + SI [φ + ϕ]
when evaluated at the result of varying the above with respect to either argument:
δ δS[φ + ϕ]
⇒ = Kϕ = 0
δφ δφ
δ δ δW0[ϕ]
− ⇒ = −Kφ
δϕ δφ δϕ
The former follows directly from the limiting procedure; the latter we have just proven
equivalent by evaluating the limit for hφ̃iϕ . Since by definition the effective action Γ
is related to W in exactly the same way that S is related to W0 (the trees from Γ
give the full W ), we also have

W [ϕ] = Γ0 [φ] + ΓI [φ + ϕ]
δΓ[φ + ϕ] δW [ϕ]
= K̃ϕ = 0 ⇔ = −K̃φ
δφ δϕ
where K̃ is the kinetic operator appearing in Γ0 , the quadratic part of Γ. (Some care
must be taken for the fact that the poles and residues of K̃ in p2 may differ from
those of K, as discussed in the previous subsection.)
 C. S-MATRIX 265

In practice, if one wants to make use of the classical field equations perturbatively,
one looks at tree graphs with a specific number of external lines: For example, in a
scalar theory with φ3 interaction (assuming hΦi = 0),
X
∞ X
n−1
KΦ + 1 2
2Φ = 0, Φ= Φn ⇒ KΦn + 1
2 Φm Φn−m = 0
n=1 m=1

gives a recursion relation for the term Φn that is nth order in ϕ.

Excercise VC3.1
Consider the relativistic Schrödinger (Klein-Gordon) equation for a scalar
wave function ψ in an external scalar potential φ:

(K + φ)ψ = 0

(If you find it less confusing, you can consider the nonrelativistic case K =
~p 2 /2m − E, where H = ~p 2/2m + φ.) Find the perturbative solution for
the quantum mechanical (one-particle) S-matrix for ψ (see subsection VA4).
Show that this agrees with the contribution to the field-theoretic S-matrix for
the Lagrangian
L(ψ, φ) = ψ*(K + φ)ψ + Lφ (φ)
coming from tree graphs with an external ψ line, an external ψ* line, and an
arbitrary number of external φ lines.
Excercise VC3.2
Consider, instead of the background field ϕ, a “current” source J that attaches
propagators to external lines. The current is effectively just a one-point in-
teraction (it caps loose ends of propagators), so it can be introduced into the
generating functional by the modification
Z
SI [φ] → SI [φ] + dx Jφ

We now have
Z   Z 
−iW [J]/h̄ i
e = Dφ exp − S0[φ] + SI [φ] + dx Jφ
h̄

Z[J] is thus the Fourier transform of e−iS[φ]/h̄ with respect to the conjugate
variables φ and J.
a Derive the “Schwinger-Dyson equations”
 !
δS[φ] 
+ J(x) e−iW [J]/h̄ = 0
δφ(x) φ=ih̄δ/δJ
266 V. QUANTIZATION

b Find the classical limit

Z
δW0[J] δS[φ]
W0 [J] = S[φ] + Jφ at φ = ⇔ J=−
δJ δφ
(i.e., W0 [J] is the “Legendre transform” of S[φ]). Find the corresponding
relation for Γ[φ] and W [J].
c Show that the free part of W [J] is given by
Z
1 1
Wf ree = − 2 dx J J ⇒ ϕ ≡ φf ree = − J
1
K K
Note that J can be replaced with ϕ in W [J] everywhere except the free term,
since in all other terms J appears only in the combination (1/K)J. Show
how this can be done in such a way as to reproduce the results above for the
solution of the classical field equations in terms of W [ϕ]. Warning: Before
this substitution we use Kϕ + J = 0, but afterwards we apply Kϕ = 0; also
beware of integration by parts, since ϕ does not vanish at ∞, so the naive
substitution φ → φ + ϕ is not very helpful. (Historically, Z was introduced as
a functional of J. However, the only two applications of Feynman diagrams,
the S-matrix A and the effective action Γ, both required that the external-
line propagators resulting from the Feynman rules for Z[J] be “amputated”.
Therefore, we use background fields exclusively. The resulting derivations,
generalities, and applications are at least as simple as and often a little simpler
than the corresponding ones with current sources.)

4. Feynman rules
It is usually most convenient to calculate Feynman diagrams in Wick-rotated
(to eliminate i’s) momentum space (where massive propagators are simpler). The
“Feynman rules” are then read off of the action as
Z Z 
−W [φ] 1 δ 1 δ
S = dx 2 φKφ + SI [φ] ⇒ Z[φ] = e
1
= exp 2 e−SI [φ]
δφ K δφ

where in Z[φ] we simply replace each field φ with a single-particle wave function in
all possible permutations, since for the case of an N-particle amplitude we usually
write the wave function as the product of N single-particle wave functions (although
more generally it is a linear combination of these):
        Y N Z 
δ δ δ δ
AN = Ψ̂N Z[φ] AN,c = −Ψ̂N W [φ] , Ψ̂N = ψN i
δφ δφ δφ i=1
δφ
 C. S-MATRIX 267

We Fourier transform as
Z Z
φ(x) = dp eip·xφ(p), φ(p) = dx e−ip·xφ(x)

(Of course, φ(x) and φ(p) are different functions, but the distinction should be clear
from context.) In practice we choose the single-particle wave functions to be eigen-
states of the momentum, so

ψi (x) = eipi ·x ψ̂i(pi ), ψi (p) = δ(p − pi )ψ̂i (pi ) (p2i + m2 = 0)

where ψ̂ is some simple factor (e.g., 1 for a scalar). Then the external line factor
terms become Z
δ δ
dx ψi(x) = ψ̂i(pi )
δφ(x) δφ(pi )
while for propagator terms
Z Z
1 δ 1 δ δ 1 δ
dx 2 = dp 12
δφ(x) K(−i∂) δφ(x) δφ(p) K(p) δφ(−p)

and for vertex terms

Z Z
dx φ1(x)...φn(x) = dp1 ...dpn φ1(p1 )...φn(pn )δ(p1 + ... + pn )

where each of the φ’s in the vertex may represent a field with derivatives; then we
replace −i∂ on φ(x) with p on φ(p). Thus, e.g., we have
Y 
δ
AN = ψ̂N i(pi ) Z[φ]
δφ(pi )

Excercise VC4.1
Use the definition
δ
φ̃(p0 ) = δ(p − p0 )
φ̃(p)
to show that g
δ δ
(p) =
δφ δ φ̃(−p)
where we now use tildes to indicate Fourier transformation.
Note that there is some ambiguity in the normalization of external line factors,
associated with the numerator factor in the propagator

1 N(p)
∆= = 1 2 2
K 2 (p + m )
268 V. QUANTIZATION

For nonzero spin (or internal symmetry), we have already discussed the normalization
analogous to that for scalars, namely
X
ψ̂† (p)ψ̂(p) = ±N(p)

(with − for negative energy and half-integer spin). However, there is already some
freedom with respect to coupling constants: Even for scalars, if a coupling appears in
the kinetic term as a factor of 1/g 2 , then effectively the kinetic operator is K = K0 /g 2 ,
where K0 is the usual (coupling-independent) one. Thus N = g2N0 , so ψ̂ = g ψ̂0,
meaning coupling dependence in external lines. Alternatively, this external-line fac-
tor of the coupling can be included in the definition of probabilities in terms of
amplitudes, which already includes nontrivial factors because of the use of (non-
normalizable) plane waves. Furthermore, quantum effects modify the form of the
propagator: Such effects can be absorbed near the pole p2 = −m2 by a field redef-
inition, but often it is more convenient to leave them. Then N will again have a
constant, (but more complicated) coupling-dependent factor, which must be canceled
in either the external-line factors or probabilities. (However, note that these questions
do not arise in calculations of the functionals W [φ] or Γ[φ].)
In tree graphs all momentum integrals are trivial, with the momentum conserva-
tion δ functions at each vertex, and the δ functions of the external lines, determining
internal momenta in terms of external momenta. In loop graphs there is a momentum
integral left for each loop, over the momentum of that loop. The amplitude will al-
ways have an overall δ function for momentum conservation for each connected piece
of the graph. Since we are always interested in just the connected graphs, we pull
this conservation factor off to define the “T-matrix”: Including the factor of “i” from
Wick rotating back to Minkowski space,
X 
Sconnected = iδ p T

In general there will be combinatoric factors associated with a graph. These

follow automatically from the functional expressions, but can also be seen from the
symmetries of the graph. Here “symmetries” means ways in which the graph can
be twisted, with external lines fixed, such that the graph looks the same, including
the types of particles propagating along the lines. For example, a graph with 2
vertices that are connected by n identical propagators would get a factor of 1/n! for
that symmetry. There are also sign factors from fermions: Permutation of external
fermion lines gives minus signs, because it involves permutation of anticommuting
fields in the functionals. Each fermion loop gets a minus sign for the same reason.
 C. S-MATRIX 269

(This is related to the fact that fermionic integration gives determinants instead of
inverse determinants.) Explicitly, it comes from evaluating expressions of the form
(ignoring momentum dependence and external fields)
   
δ δ δ δ
··· (ψ̄ψ) · · · (ψ̄ψ)
δ ψ̄ δψ δ ψ̄ δψ
       
δ δ δ δ δ δ
= ψ̄ ψ ψ̄ · · · ψ ψ̄ ψ
δ ψ̄ δψ δψ δ ψ̄ δψ δ ψ̄
where the propagator derivatives (δ/δ ψ̄)(δ/δψ) give no signs connecting up successive
vertex factors ψ̄ψ, but the last one does in closing the loop.
The general rules for contributions to the (unrenormalized) effective action Γ[φ]
are then:
R
(A1) 1PI graphs only, plus S0 ( 12 φKφ).
R
(A2) Momenta: label consistently with conservation, with dp for each loop.
(A3) Propagators: 1/ 12 (p2 + m2 ), or 1/K, for each internal line.
(A4) Vertices: read off of −SI .
R P
(A5) External lines: attach the appropriate (off-shell) fields and dp, with δ( p).
(A6) Statistics: 1/n! for n-fold symmetry of internal/external lines;
−1 for fermionic loop; overall −1.
(If we want to calculate W [φ] instead, then simply replace step 1 with “Connected
graphs only”.) The next step is to analyze the vacuum:
(B1) Find the minimum of the effective potential (for scalars).
(B2) Shift (scalar) fields to perturb about minimum; drop constant in potential.
(B3) Find resulting masses; find wave function normalizations.
Renormalization is performed either before or after this step, depending on the
scheme. Finally, the trees from Γ are identified with the complete amplitudes from
S. Thus, T-matrix elements are given by:
(C1) Connected “trees” of (shifted, renormalized) Γ: (A2-4) for L=0 with S → Γ.
(C2) “Amputate” external Γ0 -propagators.
P
(C3) External lines: 1, or appropriate to Γ0 wave equation K̃ψ = 0 ( ψ̂ †ψ̂ = ±N).
(C4) External-line statistics: No symmetry factors; −1 for fermion permutation.

Note that Γ is usually simpler than T with respect to treatment of external

lines: In T we often have contributions from graphs which are identical except for
permutation of external lines from identical fields. In Γ only one such graph need
be considered, since the statistics of the attached external fields automatically takes
care of this symmetry. (We then also drop the 1/n!, or at least reduce it.)
270 V. QUANTIZATION

The simplest nontrivial tree graphs are 4-point amplitudes. These are conve-
niently expressed in terms of the Mandelstam variables (see subsection IA4). For
symmetry, it is now more convenient to label all external momenta as flowing out of
the diagram (or all flowing in); then

s = −(p1 + p2 )2 , t = −(p1 + p3 )2 , u = −(p1 + p4 )2

Now positive energies travel to x0 = +∞, while negative energies go to −∞. We

also use the convention that s is defined in terms of the momenta of the two initial
particles (and we also use this same definition when there are more than two final
particles); t and u are then more or less interchangable, but if initial and final particles
are pairwise related we choose t in terms of the momenta of such a pair.
For example, the simplest nontrivial theory is φ3 theory, with

K = 12 (− + m2 ) → 12 (p2 + m2)
Z Z X 
3
SI = dx 6 gφ = dp1 dp2 dp3 16 gφ(p1 )φ(p2 )φ(p3 )δ
1
p

The four-point S-matrix amplitude at tree level (order g2 ) then comes from using the
following contributions to the factors in A (calculated in Euclidean space):
δ δ δ δ
Ψ̂4 =
δφ(p1) δφ(p2 ) δφ(p3) δφ(p4 )
R Z
(δ/δφ)(1/K)(δ/δφ)/2 δ 1 δ
e = ... + dk 12 1 2 2
+ ...
δφ(k) 2 (k + m ) δφ(−k)
e−SI = ... + 12 SI2 + ...
Using (δ/δφ(p))φ(k) = δ(p − k), keeping only the connected part, and integrating out
P
the δ functions (except δ( pexternal )), we are left with
 
1 1 1
Sc = δ(p1 + p2 + p3 + p4 )g 2
2 − s)
+ 1 2 + 1 2
1
2 (m 2 (m − t) 2 (m − u)

(There would be an extra i for the δ in Minkowski space.) Note the symmetry factor
1/3! for the φ3 coupling, which is canceled upon taking 3 functional derivatives for
the vertex factor. The T-matrix then comes from just factoring out the δ:
 
2 1 1 1
T =g + 1 2 + 1 2
2 (m − s) 2 (m − t) 2 (m − u)
1 2

But this contribution to W is given by a single term:

Z
1
W = −g 2
dp1 dp2 dp3 12 [ 12 φ(p1 )φ(p2 )] 1 2 [ 12 φ(p3 )φ(−p1 − p2 − p3 )]
2 (m − s)
 C. S-MATRIX 271

or, in position space,

Z
1
W = −g 2
dx 12 ( 12 φ2 ) 1 ( 12 φ2 )
2 (m2 − )

(The 12 ’s correspond to the various symmetries: switching a pair connecting to the

same vertex, or switching the two pairs.)

1 3 1 3
1 3
s
+ t 1+3 + 1+4 u
1+2
2 4
2 4 2 4

Excercise VC4.2
Find the 5-point tree amplitude for φ3 theory. What order in g is the n-point
tree?
The momentum integrals are real in Euclidean space: There are no singularities
in the integrand, since p2 + m2 is always positive (although there are some subtleties
in the massless case). Thus, all these integrals are most conveniently performed in
Euclidean space. However, eventually the result must be analytically continued back
to Minkowski space: x0 → ix0, which means p0 → ip0 (but p0 → −ip0 , being also
n
careful to distinguish δm → δm
n
and δmn → ηmn for indices on fields) via a 90◦ rotation.
This returns some of the i dependence. There are also i’s associated with integration
measures: Since all the momentum integrals for the S-matrix elements have already
P
been performed, all that remains is a factor of i to go with δ( p) for each connected
graph. There can also be i dependence in external line factors.
Remember that negative p0 indicates a particle traveling backward in time; the
true motion of such a particle is opposite to that of the arrow indicating momentum
flow. Thus, external lines with arrows pointing into the diagram and positive p0 , or
out of the diagram and negative p0 , both indicate initial states, arriving from t = −∞.
Conversely, external lines with arrows pointing into the diagram and negative p0 , or
out of the diagram and positive p0 , both indicate final states, departing to t = +∞.
A related issue is particles vs. antiparticles. If a particle is described by a real
field, it is identified as its own antiparticle; but if it is described by a complex field,
then it is identified as a particle if it has a certain charge, and as an antiparticle if
it has the opposite charge. (For example, a proton is positively charged while an
272 V. QUANTIZATION

antiproton is negatively.) Of course, this is convention, since a complex field can

always be replaced by two real fields, and we can always relabel which is the field
and which the complex conjugate; generally there should be a useful conservation law
(symmetry) associated with these complex combinations (usually electric charge),
and the one called “particle” is the one more common to the observer. For example,
suppose we have a complex scalar external field/wave function φ(p). For p0 > 0 this
describes a particle propagating to x0 = +∞. Similarly, φ*(p) for p0 > 0 describes
an anti particle propagating to x0 = +∞. On the other hand, φ* for p0 < 0 describes
a particle propagating from x0 = −∞, while φ for p0 < 0 describes an antiparticle
propagating from x0 = −∞.

5. Semiclassical unitarity
As in nonrelativistic quantum mechanics, the only conditions for unitarity are
that: (1) the metric (inner product) on the Hilbert space is positive definite (so all
probabilities are nonnegative), and (2) the Hamiltonian is hermitian (so probabilities
are conserved). Both of these conditions are statements about the classical action.
The second is simply that the action is hermitian, which is easy to check. The first is
that the kinetic (quadratic) terms in the action, which define the (free) propagators,
have the right sign. This can be more subtle, since there are gauge and auxiliary
degrees of freedom. Therefore, the simplest way to check is by using the lightcone
formalism.
We see from the analysis of subsection IIB3 for field equations, or for actions in
subsection IIIC2 (for spins ≤ 1, and more generally below in chapter XII) that after
lightcone gauge fixing and elimination of auxiliary degrees of freedom the kinetic
terms for physical theories always reduce to − 14 φ φ for bosons and 14 ψ( /i∂ +)ψ for
fermions, where there is a sum over all bosons and fermions, and each term in the
sum has a field with a single hermitian component. Complex fields can multiply their
complex conjugates, but these can always be separated into real and imaginary parts.
There are never crossterms like A B, since after field redefinition, i.e., diagonalization
of the kinetic operator, this gives A0 A0 − B 0 B 0, so one term has the wrong sign.
Similar remarks apply to massive fields, but with replaced by − m2 (as seen, e.g.,
by dimensional reduction), or we can treat the mass term as part of the interactions.
Now we only need to check that the single-component propagators of these two
cases define positive-definite inner products. Since multiparticle inner products are
products of uniparticle inner products, it’s sufficient to look at one-particle states.
We therefore examine the S-matrix defined in subsection VC1 for the special case of
 C. S-MATRIX 273

1 particle at t = −∞ going to 1 particle at t = +∞, using the free, massless lightcone

Lagrangians given above. For the boson we found that this matrix element was simply
R
the inner product between the two states, appearing in the form dσ dσ 0 ψ∆ψ 0. This
worked only because the propagator had the right sign. (It is essentially +e−iω|t|.)
Thus the sign we use is required for unitarity.
A simple way to treat the fermion is to use supersymmetry: Since the boson and
fermion kinetic terms are spin independent in the lightcone formalism, we can look at
any supersymmetric theory, and check the boson and fermion kinetic terms there. If
the boson term agrees with the one we just checked, then the fermion term is OK by
supersymmetry (which preserves unitarity by the lightcone-like supertwistor analysis
of subsection IIC5). Alternatively, we can use the same method applied to the boson:
The propagator now has an extra factor of 1/(−i∂ + ), or 1/p+ in momentum space.
Since p+ is always positive for positive energy, these states also appear with the
correct-sign norm. To analyze negative energy (antiparticles), we note that in the
derivation of the path integral final states always appear to the left and initial states
to the right. In the fermionic case this is important because it will introduce an extra
sign: Since
ψ1 ψ2 = +ψ2 + ψ1
∂+ ∂
(with two signs canceling from reordering fermions and integration by parts), the
right sign will always be produced with correct ordering of initial vs. final states
(i.e., positive vs. negative energy), independent of the helicity (or whether electron
vs. positron, etc.)
For similar reasons, it is clear that integral spin is always described by commut-
ing (bosonic) fields, while half-integral spin is always described by anticommuting
(fermionic) fields: The number of undotted minus dotted indices on a field is always
odd for half-integer spin, even for integer, and a derivative carries one dotted and one
undotted index, so contraction of all indices means an even number of derivatives for
integer spin and odd for half-integer. Without loss of generality, we then can separate
each field into its real and imaginary parts. Then for each real field integration by
parts gives

φ(−i∂)nφ = (−1)n [(−i∂)nφ]φ = (−1)n+φ φ(−i∂)n φ ⇒ (−1)n+φ = 1

where (−1)φ is the statistical factor for φ (1 for bosons, −1 for fermions), and we
have inclueed the appropriate i’s for hermiticity of the action. Thus, integer spin is
associated with bosons ((−1)n = 1 = (−1)φ ), and half-integer with fermions ((−1)n =
−1 = (−1)φ ). This is the “spin-statistics theorem”. By using real fields with a
274 V. QUANTIZATION

diagonal kinetic term, we have implicitly assumed kinetic terms appear only with the
correct sign: For example, for a complex bosonic field
↔ ↔
φ = A + iB ⇒ φ† ∂φ = Ai ∂ B = −Bi ∂ A

and thus has indefinite sign. Thus, spin and statistics follows from Poincaré invari-
ance, locality, and unitarity. If we drop unitarity, we get “ghosts”: We’ll see examples
of such wrong-statistics fields when quantizing gauge theory.
Note that demanding unitarity (the right sign of the kinetic term) is the same as
demanding positivity of the true energy, as least as far as the kinetic term is concerned:
The energy is given by the Hamiltonian of the field theory; if the kinetic term changes
sign, the corresponding contribution to the Hamiltonian does also. (Compare the
discussion of Wick rotation of the action in subsection VB4.)
Using anticommuting fields to describe fermions is more than a formality. In gen-
eral, the significance of describing states by quantizing classical fields that commute
or anticommute has two purposes: (1) to avoid multiple counting for indistinguish-
able particles, and (2) to insure that two identical fermions do not occupy the same
state. Thus, when describing two particles in different states, the phase associated
with (anti)commutation is irrelevant: A “Klein transformation” can be made that
makes anticommuting quantities commute for different states, and anticommute (i.e.,
square to zero) only for the same state. However, such transformations are nonlocal,
and locality is crucial in relativistic field theory. (See excercise IA2.3e.)

6. Cutting rules
For some purposes it is useful to translate the three defining properties of relativis-
tic quantum field theory into graphical language. Poincaré invariance is trivial, since
the propagators and vertices are manifestly Poincaré covariant in covariant gauges.
Unitarity and causality can also be written in a simple way in functional notation. We
first note that the inner product for free multiparticle wave functions can be written
very simply in momentum space as
Z ←

 δ δ
hψ|χi = ψ [φ]e χ[φ] φ=0 ,
† D+
D+ = 12 dp ∆+(p)
δφ(p) δφ(−p)
−iN(p)
∆(p) = ⇒ ∆+(p) = θ(p0 )2πδ[ 12 (p2 + m2)]N(p)
1 2
2 (p+ m2 − i)
where ψ and χ are products of positive-energy single-particle states, and ∆+ projects
onto the positive-energy mass shell. (The exponential takes care of the usual combi-
natoric factors.) We have written a generic propagator, with numerator factor N (=1
 C. S-MATRIX 275

for scalars). (Without loss of generality, we have assumed a real basis for the fields,
so N can be taken as real.) The S-matrix amplitude then can be written in operator
language as
Z  
† −iS[φ] δ
Ψ[φ] = ψ [φ]χ[φ] ⇒ A = Dφ Ψ[φ]e =Ψ Z[φ] = hψ|S|χi
δφ
We have used the fact that positive-energy states propagate forward in time and
negative backwards to write the usual Hilbert-space inner product in terms of initial
and final states of positive energy. The S-matrix operator S appears because χ and
ψ satisfy the free equations of motion, and S performs time translation from t = −∞
for χ to t = +∞ for ψ to include interactions.
The unitarity condition is then (see subsection VA4)

S †S = 1 ⇒ Z[φ]†eD+ Z[φ] = 1

while causality is
   
δ † δ δ † D+ δ
S[φ] S[φ] = 0 ⇒ Z[φ] e Z[φ] = 0 for x0 > y 0
δφ(x) δφ(y) δφ(x) δφ(y)
This causality relation, which already holds in nonrelativistic field theory, can be
strengthened by using Lorentz invariance: If (x − y)2 > 0 (spacelike separation), then
x0 < y 0 can be Lorentz transformed to x0 > y 0. Thus, the above expression vanishes
everywhere except on or inside the backward lightcone with respect to x − y. These
functional forms of unitarity and causality (and Poincaré invariance) can also be used
as a basis for the derivation of the functional integral form of Z[φ] in terms of the
action, rather than relying on its relation to the Hamiltonian formalism.
The fact that these conditions are satisfied by Feynman diagrams follows easily
from inspection. We examine them using the explicit expression for Z following from
the functional integral:

Z[φ] = eD e−iSI , Z[φ]† = eD*eiSI

Z
1 δ δ −iN
D= dp ∆(p) , ∆(p) =
2
δφ(p) δφ(−p) 1 2
2 (p + m2 − i)
Z
1 δ δ iN
D* = 2 dp ∆(p)* , ∆(p)* = 1 2
δφ(p) δφ(−p) 2 (p + m2 + i)
These expressions can be translated straightforwardly into position space as
Z Z
δ δ δ δ
dp ∆(p) = dx dx0 ∆(x − x0)
δφ(p) δφ(−p) δφ(x) δφ(x0)
276 V. QUANTIZATION

etc.
From the results at the end of subsection VB2, we see that the propagators satisfy
the relations
∆+(x) = ∆(x) − ∆A(x) = ∆*(x) + ∆R (x)

∆+(x) = ∆−(−x), ∆(x) = ∆(−x), ∆R(x) = ∆A(−x)

and, of course, ∆R (x) = 0 for x0 < 0. We will now see that the cancelations in
the unitarity and causality relations occur graph by graph: There are contributions
consisting of a sum of terms represented by exactly the same diagram, with each term
differing only by whether each vertex comes from Z or Z † . First, this affects the sign
of the diagram, since each vertex from Z † gets an extra sign from the eiSI in Z † , as
compared to the e−iSI in Z. Second, this affects which propagators appear:

(x, y) in : (Z, Z) → ∆(x − y)

(Z † , Z) → ∆+ (x − y)
(Z, Z † ) → ∆− (x − y)
(Z †, Z † ) → ∆*(x − y)

Now if we sum over two diagrams differing only by whether x appears in Z or Z †,

the result is proportional to
Y Y Y Y
∆(x − yi ) ∆−(x − zj ) − ∆+(x − yi ) ∆*(x − zj )

where yi are from Z and zj are from Z † . However,

∆(x − y) − ∆+(x − y) = ∆−(x − y) − ∆*(x − y) = ∆R(y − x) = 0 for x0 > y 0

and therefore the two diagrams cancel if x0 is the latest of all the vertices. In the
unitarity relation we sum over whether a vertex occurs in Z or Z † for each vertex,
including the latest one, so that condition is easily satisfied. (The only diagram that
survives is the one with no particles, which gives 1.) In the causality relation we
perform this sum for each vertex except y (since δ/δφ(y) acts only on Z), but since
y 0 < x0, y 0 is not the latest vertex, so again the latest one is summed over.
Excercise VC6.1
Consider an arbitrary 1-loop graph. Why would replacing all the propagators
∆ with advanced propagators ∆A (or all with retarded propagators ∆R ) all the
way around the loop in the same direction give zero? Use this result, and the
relation between the various propagators, to show that any one-loop diagram
(with normal propagators) can be expressed as a sum of products of tree
 C. S-MATRIX 277

graphs, with some summations of external states (“Feynman tree theorem”).

How does this differ from the cutting rule for unitarity? (Hint: Look at the
signs of the energies of external states.)
There is one fine point in this construction: We may use Feynman rules from a
complex action, such as those used for massive theories in subsection IIIC4, or when
using complex gauges (see section VIB). In that case, since the S-matrix S is gauge-
independent, and the original action S was real (before eliminating complex fields
or choosing complex gauges), it is legal to use the Hermitian conjugate action S † to
define the Feynman rules for S † (and Z †): When multiplying S † S, we use the usual
rules to find the second factor S, and the conjugate rules to find the S used in the
first factor S † :
S † S = [S(S †)]†S(S)

The result of conjugating the S-matrix then will be to complex conjugate twice, and
return rules identical to those used for S, except for the differences noted above for
real actions. That means that the above proof of the cutting rules goes through
unmodified, where we use the same complex rules in the entire diagram, regardless of
whether they are associated with Z or Z † . In particular, this means that vertices from
the two parts of the graph will differ only by sign (conjugating just the i in eiSI , not the
SI ), and propagators will differ only by their (momentum-space) denominators, not
their numerators. This is particularly important for the complex fields of subsection
IIIC4, since otherwise even the types of indices carried by the fields would differ.

7. Cross sections

In quantum physics, the only measurables are probabilities, the squares of abso-
lute values of amplitudes. Since we calculate amplitudes in momentum space, proba-
bilities are expressed in terms of scattering of plane waves. They are more naturally
normalized as probabilities per unit 4-volume (or D-volume in arbitrary dimension),
since plane waves are uniformly distributed throughout space. This can be seen ex-
plcitly from the amplitudes: Because of the total momentum conservation δ-function
that appears with each connected S-matrix element Sf i , we have for the probability
P
X 
Sf i = iδ p Tf i
X  X  VD
⇒ P = |Sf i |2 = |Tf i|2 δ p δ(0) = |Tf i|2 δ p
(2π)D/2
278 V. QUANTIZATION

where we have found the coordinate D-volume by the Fourier-transform definition of

the δ-function: Z Z
dD x VD
δ(0) = dx 1 = 1 =
(2π)D/2 (2π)D/2
A “cross section” is defined as a probability for the scattering of two incoming
particles into some number of outgoing particles. The scattering is “elastic” if the
two final particles are the same as the two initial particles (they exchange only 4-
momentum), “inelastic” otherwise. Generally one particle is in a beam directed at a
target (at rest in the lab frame) containing the other “incoming” particle, but in some
experiments two beams are directed at each other. In either case the cross section is
defined by the rate at which one particle interacts divided by the flux of the other
particle, where
rate of arrival
flux = = (density) × (relative velocity)
area
and thus the “differential cross section” (yet to be integrated/summed over final
states) is
P 1
dσ = ×
VD ρ1ρ2 v12
The spatial density ρ is the integrand of the spatial integral that defines the inner
product: From subsection VB2, for bosonic plane waves we have
Z Z
dD−1 x 1 ↔ dD−1 x
hψ|ψi = (p )
0
ψ i ∂ t ψ = ω
(2π)D/2 2 (2π)D/2
ω
⇒ ρ=
(2π)D/2
where ω = |p0 |. (The same result can be obtained for fermions, when their external
line factors are appropriately normalized.)
The expression for dσ is actually independent of the frame, as long as the 3-
momenta of the two particles are parallel. This is the case for the most frequently
used reference frames, the center-of-mass frame and the “lab frame” for either particle
(where that particle is at rest, as is the lab if that particle is part of a target). Then
 2
 p
~ p
~ 
(ω1 ω2 v12)2 = ω12ω22  −  = |ω2 ~p1 − ω1 p~2 |2 = − 12 (p1[a p2b] )2 = λ212
1 2
ω1 ω2

λ212 ≡ 14 [s − (m1 + m2)2 ][s − (m1 − m2 )2]

using again the Mandelstam variables and λij introduced in subsection IA4. Finally,
we include the “phase space” for the final states to obtain
D D
P Y Y 1
2 (2π) δ ( p) dD−1 p
dσ = |Tf i|
λ12 f
(2π)D/2−1ω f :n ident n!
 C. S-MATRIX 279

where the first product is over all final one-particle states, and the second is over
each set of n identical final particles. The normalization again follows from the inner-
product for plane waves: By Fourier transformation, dD−1 x/ρ → dD−1 p/(2π)D−1 ρ.
It also appears in the “cut propagator” ∆+ used in unitarity, as in the previous
subsection: Z Z
dD p 0 1 2 2 dD−1 p
θ(p )2πδ[ 2 (p + m )] =
(2π)D/2 (2π)D/2−1ω
The simplest and most important case is where two particles scatter to two parti-
cles. (This includes elastic scattering.) The “differential cross section” dσ/dΩ, where
dΩ is the angular integration element for p3 , is found by integrating dσ (in D=4) over
d3 p4 and d|~p3 |. The former integration is trivial, using the δ function for 3-momentum
conservation. The latter integration is almost as trivial, integrating the remaining δ
function for energy conservation:
X   P 0 −1

2 ∂ p 
3
d p3 δ p = dΩ d|~p3 | (~p3 ) 
0
δ(|~p3 | − |~p3 |0 )
∂|~p3| 
X q q
p0 = −ω1 − ω2 + (~p3 )2 + m23 + (~p1 + p~2 − p~3 )2 + m24
P
∂ p0 1
(s − m23 − m24 )ω3 − m23ω4
⇒ = 2
∂|~p3 | |~p3 |ω3 ω4
P
where |~p3 |0 is |~p3 | evaluated as a function of the remaining variables at p0 = 0. We
then find
dσ |~p3 |3
= (2π)2 |Tf i|2
dΩ λ12 [ 12 (s − m23 − m24)ω3 − m23ω4 ]

The center-of-mass frame (see subsection IA4) is the simplest for computations.
In that frame the differential cross section simplifies to
dσ λ34
= (2π)2|Tf i|2
dΩ λ12 s
Another convenient form for the differential cross section is dσ/dt, trading θ for t and
R
integrating out the trivial dependence dφ = 2π. In the center-of-mass frame we
have
s
dΩ = 2πd(cos θ) = π dt
λ12 λ34
Since t is Lorentz invariant, we therefore have in all frames (that conform to our
earlier requirement for the λ12 factor in dσ)

dσ 1
= 12 (2π)3 |Tf i|2 2
dt λ12
280 V. QUANTIZATION

For example, for the 4-point scalar example considered in subsection VC4, we
have  2
dσ (4π)3 g 4 1 1 1
= + +
dt s(s − 4m2 ) s − m2 t − m2 u − m2

∆+
S S†
∆+
∆ ∆*
∆_
∆_

For some purposes (such as considerations of unitarity and causality, as in the

previous subsection) it is useful to draw the Feynman diagrams for the cross section
itself (or actually |T |2). In such a diagram we draw one of the diagrams from S
and one from S † , separating the two by a line (dashed, zig-zag, or shaded on one
side, according to your preference), and connecting all the external lines (initial and
final) on one side to the corresponding ones on the other. The result is a bubble
diagram with a “cut”: The “cut propagators” are ∆+ = 2πθ(p0 )Nδ[ 12 (p2 + m2)]
(or ∆−, depending on how we label the momenta), corresponding to the propagator
∆ = N/ 12 (p2 + m2 ), if we sum over all polarizations; otherwise N is replaced by a
P †
term in the sum N = ψ̂ ψ̂. The momenta of the cut propagators may not be
integrated over, depending on whether they represent final states whose momenta
are summed over (i.e., not measured; in practice the momenta of initial particles
are always measured). The only other difference in the Feynman rules from the S-
matrix is that in Wick rotating back S gets the usual m2 → m2 − i while S † gets
P
m2 → m2 + i, and each connected graph in S gets an iδ( p) while each in S †
P
gets a −iδ( p). The algebra for the cross section is thus identical to that of a
vacuum bubble (although the momentum integration is not, and the cut propagators
lack the usual denominators). In particular, instead of summing over just physical
polarizations in a cut vector propagator, which corresponds to using a unitary gauge,
we can include ghosts in the external states, and use any gauge: This follows from
the cutting rules derived for unitarity.

8. Singularities
We know from free theories that any propagator has a pole at the classical value of
the square of the mass. This statement can be extended to the interacting theory: The
 C. S-MATRIX 281

(“Landau”) singularities in any Feynman diagram are exactly at classically allowed

(on shell) values of the momenta.
The simplest way to see this is to write the propagators in a way reminiscent of
the classical theory, where the appearance of the worldline metric in the action results
in the (Wick-unrotated) Schwinger parametrization of the propagator,
Z ∞
−i
dτ e−iτ (p +m )/2
2 2
1 2 2
=
2 (p + m ) 0

For simplicity we consider a scalar field theory with nonderivative self-interactions.

The corresponding form of a Feynman diagram, written in momentum space by
Fourier transformation, is then
Z P
−i [τij (p2ij +m2 )/2−(xi−xj )·pij ]
dx0i dpij dτij e hiji

where i, j label vertices (and external endpoints), hiji labels links (propagators), and
dx0 indicates integration over just vertices and not free ends of external lines.
Integration over these x0 ’s produces δ functions for momentum conservation:
X
pij = 0
j

i.e., the sum of all momenta flowing into any vertex vanishes. (Note pij = −pji .) This
constraint can be solved by replacing momenta associated with each propagator with
momenta associated with each loop (and keeping momenta associated with external
lines). For example, for planar diagrams, we can write

pij = pIJ = kI − kJ

where I, J label loops: pIJ labels the propagator by the two loops on either side,
rather than the vertices at the ends (and similarly for τij = τIJ ). Similar remarks
apply to nonplanar diagrams, but the parametrization in terms of loop momenta kI is
more complicated because of the way external momenta appear. (In the planar case,
the above parametrization can also be used for external momenta, and automatically
enforces overall momentum conservation.) The Feynman integral then becomes
Z P
0 −i τIJ [pIJ (k)2 +m2 ]/2
dkI dτIJ e hIJi

where k are loop (k 0) and external momenta after solving the conservation conditions.
We can now treat the exponent of the Feynman diagram in the same way as a
classical mechanics action, and find the corresponding classical equations of motion by
282 V. QUANTIZATION

varying it. By the stationary phase approximation (or steepest descent, after Wick
rotation of τ ), the classical solutions give the most important contribution to the
integrals (at least for weak coupling). This approximation is related to long-distance
(i.e., infrared) behavior (see excercise VB4.3a). Taking acount of the fact that the τ ’s
are constrained to be positive (by treating τ = 0 separately or making a temporary
change of variables τ = β 2 or eβ to an unconstrained variable), we find
X
τIJ (p2IJ + m2) = 0, τIJ pIJ = 0
J

or, in terms of the original variables,

xi − xj X
τij (p2ij + m2) = 0, pij = , pij = 0
τij j

These are known as the “Landau equations”. Their correspondence with classical
configurations of particles follows from treating τ as the proper time, as seen from
p = ∆x/∆τ . (Actually, it is the generalization discussed in section IIIB of proper time
to include the massless case; in the massive case s = mτ .) The equation τ (p2 +m2 ) = 0
says that either the particle for the line is on-shell or there is no such line (it has
P
vanishing proper length), while the equation J τ p = 0 says that the sum of τ p
around a loop vanishes, another statement that p∆τ = ∆x.
Note that these physical singularities are all in physical Minkowski space: In
Euclidean space, p2 + m2 is positive definite, so it never vanishes (except for constant,
massless fields p = m = 0). Furthermore, in Euclidean space, one can always rotate
any momentum to any direction, whereas in Minkowski space one can never Lorentz
transform to or through either the forward or backward lightcone. Thus, calculating
in Euclidean space makes it clear that S-matrix elements for positive-energy states
are given by the same expressions as those for negative-energy states: To compare
two amplitudes that are the same except for some final particles being replaced with
initial antiparticles, or vice versa, we just change the sign of the energy. This is
called “crossing symmetry”. In the case where all particles are reversed, it is CPT
invariance (“CPT theorem”).

9. Group theory
Although the manipulation of spin indices in Feynman diagrams is closely tied to
momentum dependence, the group theoretic structure is completely independent, and
can be handled separately. Therefore, it is sufficient to consider the simple example
of scalars with a global symmetry. (The more physical case of chromodynamics will
 C. S-MATRIX 283

be the same with respect to group theory, but will differ in dependence on momentum
and spin.) The simplest case is the U(N) family of groups. We choose an action of
the form
L= 1
g2
tr[ 14 (∂φ)2 + V (φ)]

where φ is a hermitian N×N matrix. This action is invariant under the global U(N)
symmetry
φ0 = UφU −1

A simplification we have chosen is that the interaction has only a single trace; this
is the case analogous to pure Yang-Mills theory. (We also included the coupling
constant as in Yang-Mills.)

color

flavor

gluon quark

When we draw a Feynman diagram for this field theory, instead of a single line for
each propagator, we draw a double (parallel) line, each line corresponding to one of
the two indices on the matrix field. Because of the trace in the vertex, the propagator
lines connect up there in such a way that effectively we have continuous lines that
travel on through the vertices, although the two lines paired in a propagator go their
separate ways at the vertex. These lines never split or join, and begin or end only on
external fields. We can also draw arrows on the lines, pointing in the same direction
everywhere along a single line, but pointing in opposite directions on the two lines
in any propagator pair: This keeps track of the fact that φ appears in the trace
always multipled as φφ and never as φφT . A physical picture we can associate with
this is to think of the scalar as a bound-state of a quark-antiquark pair, with one
line associated with the quark and another with the antiquark; the arrows are then
oriented in the direction of time of the quark (which is the opposite of the direction
of time for the antiquark). The quark is thus in the defining representation of U(N)
(and the antiquark in the complex conjugate representation). Group theory factors
284 V. QUANTIZATION

are trivial to follow in these diagrams: The same color quark continues along the
extent of a “quark line”; thus, there is a Kronecker δ for the two indices appearing
at the ends of the quark line (at external fields); each quark is conserved.
Excercise VC9.1
Using the quark-line notation, where the lines now represent flavor, draw all
4-point tree graphs, with 3-point vertices, representing scattering of K + K − →
π +π − (see subsection IC4) via exchange of other mesons in that U(6) multi-
plet. What are the intermediate states (names of mesons) in each channel?
Note that the “flavor flow” of both diagrams can be represented by a single
diagram, by separating all pairs of intermediate lines to leave a square gap in
the middle: This “duality diagram” represents the mesons as strings; the gap
between quarks represents the “worldsheet”.
Excercise VC9.2
Consider quark-line notation for doing group theory in general: (Calculate
using only graphs, with numerical factors — no explicit indices or Kronecker
δ’s, except for translating definitions.)
a Write the structure constants for U(N) as the difference of two diagrams, by
considering a vertex of the form tr(A[B, C]).
b Find pictorially the resulting expression for the Cartan metric (see subsection
IB2): Show that it is the identity times 0 for the U(1) subgroup and 2N for
the SU(N) subgroup (as found previously in subsection IIIC1).
c Also use these diagrams to prove the Jacobi identity (i.e., find the resulting
6 diagrams and show they cancel pairwise).
d Derive diagrammatically the value of dijk of excercise IB5.3b for the defining
representation.
Excercise VC9.3
Consider the group theory factors for the above scalar theory, with only a
cubic interaction. Draw all the 1PI 1-loop diagrams with 4 external double-
lines, and rewrite the corresponding factors in terms of traces and products
of 4 fields, including factors of N. (Be careful to include all permutations of
connecting propagators to vertices.)
In general, we find that connected Feynman diagrams may include diagrams that
are disconnected with respect to the above group-theory diagrams, where we consider
the two group-theory lines on an external line to be connected. Such diagrams corre-
spond to multiple traces: There is a factor of the form tr(Gi Gj ...Gk ) corresponding
 C. S-MATRIX 285

to each connected group-theory graph, where i, j, ..., k are the group-theory indices
of the external lines (actually double indices in the previous notation). However, all
connected trees are group-theory connected. Furthermore, they are all “planar”: Any
connected U(N) tree can be drawn with none of the quark lines crossing, and all the
external lines on the outside of the diagram, if the external lines are “color-ordered”
appropriately. Of course, one must sum over permutations of external lines in the
T-matrix because of Bose symmetry. However, in calculating W [φ] one need consider
only one such planar graph, with the external lines color-ordered; the Bose symmetry
of the fields in W automatically incorporates the permutations. (Similar remarks
apply to loop and nonplanar graphs.)
As an example, consider the U(N) generalization of the 4-point tree example of
subsection VC4. The Lagrangian is now (scaling the coupling back into just the
vertex)
L = tr{ 14 [(∂φ)2 + m2φ2 ] + 13 gφ3 }
where the interaction term now has a combinatoric factor of 13 instead of 16 because
it is symmetric only under cyclic permutations. The result for W is now modified to
Z
1
W = −g tr dx 12 φ2 1 2
2
φ2
2 (m − )

This analysis for U(N) can be generalized to SU(N) by including extra diagrams
with lines inside propagators short-circuited, representing subtraction of traces. It can
also be generalized to SO(N) and USp(2N): In those cases, antisymmetry or symmetry
of the matrices means the lines no longer have arrows, and we include diagrams
where the lines inside the propagators have been “twisted”, with signs appropriate
to symmetrization or antisymmetrization. Generalization for all the above groups
to include defining representations is also straightforward: Such fields, like the true
quarks in QCD, carry only a single group-theory line. For example, sticking with
our simpler scalar model, we can generalize to a scalar theory with the same group
theory as chromodynamics, with “free” quark fields, appearing as scalar fields in the
defining representation of the group:

L → L + tr[ 12 (∂ψ)† · (∂ψ) + ψ †f(φ)ψ], ψ 0 = UψUf−1

where ψ is an N×M matrix with M flavors, and Uf is the flavor symmetry. (This
color+flavor symmetry was treated in subsections IC4 and IVA4.) This field has a
propagator with a single color line (with an arrow); however, we can also use another
double-line notation, where ψ propagators carry one line for color and another for
flavor. This method can be generalized to arbitrary representations obtained by
286 V. QUANTIZATION

direct products of defining representations, (anti)symmetrizations, and subtractions

of traces, by giving the propagators the corresponding number of lines (though usually
two lines are sufficient for the interesting cases).

REFERENCES
1 P.A.M. Dirac, Proc. Roy. Soc. A114 (1927) 243;
W. Heisenberg and W. Pauli, Z. Phys. 56 (1929) 1, 59 (1930) 168;
E. Fermi, Rev. Mod. Phys. 4 (1932) 87:
birth of (interacting) quantum field theory (QED).
2 S. Tomonaga, Prog. Theor. Phys. 1 (1946) 27;
J. Schwinger, Phys. Rev. 74 (1948) 1439, loc. cit. (VB, ref. 2, first ref.), 76 (1949) 790:
gave relativistic procedure for treating quantum field theory.
3 Feynman, loc. cit. (VB, ref. 4):
same as above, with diagrams.
4 F.J. Dyson, Phys. Rev. 75 (1949) 486, 1736:
related Tomonaga-Schwinger and Feynman approaches; gave procedure to all loops.
5 J. Schwinger, Proc. Natl. Acad. Sci. USA 37 (1951) 452, 455:
generating functionals for field theory.
6 Feynman, loc. cit. (VA):
path-integrals for field theory.
7 J. Schwinger, Quantum electrodynamics (Dover, 1958):
reprints of most of the important quantum field theory papers of the 40’s and 50’s.
8 S.S. Schweber, QED and the men who made it: Dyson, Feynman, Schwinger, and
Tomonaga (Princeton University, 1994).
9 J. Goldstone, A. Salam, and S. Weinberg, Phys. Rev. 127 (1962) 965;
G. Jona-Lasinio, Nuo. Cim. 34 (1964) 1790:
effective action.
10 Y. Nambu, Phys. Lett. 26B (1968) 626:
tree graphs are classical field theory.
11 D.G. Boulware and L.S. Brown, Phys. Rev. 172 (1968) 1628:
expression for classical field in terms of tree graphs.
12 S. Coleman, Secret symmetry: An introduction to spontaneous symmetry breakdown and
gauge fields, in Laws of hadronic matter , proc. 1973 International School of Subnuclear
Physics, Erice, July 8-26, 1973, ed. A. Zichichi (Academic, 1975) part A, p. 139:
relation of Legendre transform to 1PI graphs via path integrals.
13 Dyson, loc. cit. (ref. 2);
Schwinger, loc. cit. (ref. 4).
14 W. Pauli, Phys. Rev. 58 (1940) 716:
spin-statistics theorem.
15 R.E. Cutkosky, J. Math. Phys. 1 (1960) 429:
cutting rules for unitarity.
16 T. Veltman, Physica 29 (1963) 186:
cutting rules for causality.
17 R.P. Feynman, Acta Phys. Polon. 24 (1963) 697:
tree theorem.
18 L.D. Landau, Nucl. Phys. 13 (1959) 181.
 C. S-MATRIX 287

19 S. Coleman and R.E. Norton, Nuo. Cim. 38 (1965) 438:

classical interpretation of Landau singularities.
20 W. Siegel, hep-th/9601002, Int. J. Mod. Phys. A 13 (1998) 381:
Landau rules from classical mechanics action.
21 R.J. Eden, P.V. Landshoff, D.I. Olive, and J.C. Polkinghorne, The analytic S-matrix
(Cambridge University, 1966):
singularities of Feynman diagrams.
22 M. Gell-Mann and M.L. Goldberger, proc. Fourth International Conference on High
Energy Physics (University of Rochester, 1954);
J.M. Jauch and F. Rohrlich, The theory of photons and electrons, 1st ed. (Springer-
Verlag, 1955) p. 161:
crossing symmetry.
23 H. Harari, Phys. Rev. Lett. 22 (1969) 562;
J. Rosner, Phys. Rev. Lett. 22 (1969) 689;
T. Matsuoka, K. Ninomiya, and S. Sawada, Prog. Theor. Phys. 42 (1969) 56:
quark lines for group theory factors.
24 J.E. Paton and Chan H.-M., Nucl. Phys. B10 (1969) 516:
traces for group theory in planar graphs.
25 G. ’t Hooft, Nucl. Phys. B72 (1974) 461:
double-line notation in field theory.
288 VI. QUANTUM GAUGE THEORY

VI. QUANTUM GAUGE THEORY

We now consider special features of spin and gauge invariance, and introduce some
special methods for dealing with them. In quantum theory, gauge fixing is necessary
for functional integration: Gauge invariance says that the action is independent of
some variable; integration over that variable would thus give infinity when evaluating
amplitudes for gauge-invariant states. Eliminating that variable from the action (a
“unitary gauge”) solves the problem, but not always in the most convenient way. (If
it were, we wouldn’t have introduced such a redundant description in the first place.)
Note that such infinities already appear for global symmetries: For example, the
functional integral with wave function 1 (vacuum-to-vacuum amplitude) is infinite by
translation invariance. This infinity is easier to understand for a nontrivial amplitude
in momentum space, as a factor of a momentum-conservation δ-function (which is
either ∞ or 0, but necessarily ∞ for the vacuum amplitude, which has vanishing
momentum because the vacuum is translationally invariant).
....... .......
....... A. BECCHI-ROUET-STORA-TYUTIN .......

We have seen the relationship of gauge invariances to constraints in subsection

IIIA5. In this section we consider the quantization of constrained systems, and its
application to gauge theories. The Becchi-Rouet-Stora-Tyutin (BRST) method is
not only the most powerful, but also the easiest way to gauge fix: It replaces the
gauge symmetry with an unphysical, fermionic, global symmetry that acts only on
unphysical degrees of freedom.

1. Hamiltonian
Physical observables commute with the constraints. Thus, time development is
described by the gauge-invariant Hamiltonian Hgi , or we can set the gauge fields λi
equal to some arbitrary functions f i as a “gauge choice”:

λi = f i ⇒ H = Hgi + f i Gi

In quantum mechanics, physical states should be annihilated by the constraints. How-

ever, more generally we can require only that these states satisfy the constraints
through expectation values:
hψ|Gi |χi = 0
This condition is satisfied by dividing up the constraints into: (1) a subalgebra G0
that annihilates all physical states, (2) complex “lowering operators” G− that also
 A. BECCHI-ROUET-STORA-TYUTIN 289

annihilate these states, and are a representation of the subgroup generated by G0 ,

and (3) their hermitian conjugate “raising operators” G+ = G− †. (We treat +, −,
and 0 here as multivalued indices.) Thus,

G0 |ψi = G− |ψi = hψ|G0 = hψ|G+ = 0

In the Abelian case, it then follows that, although G+ do not annihilate these states,
they generate gauge invariances:

δ|ψi = G+ |ζ +i, G0 |ζ +i = G− |ζ + i = 0

preserves the inner product of such states as well as the constraints on |ψi.
Unfortunately, things get more complicated in the nonabelian case. For example,
the gauge invariance and constraints above are no longer compatible:

0 = G− δ|ψi = G− G+ |ζ + i = [G−, G+ ]|ζ +i = −if−++ G+ |ζ +i =

6 0

A convenient way to deal with this problem is to replace the nonabelian algebra Gi
with a single operator, which is therefore Abelian. We define a BRST operator Q
that imposes all constraints Gi by adding a classical anticommuting “ghost” variable
ci , and its canonical conjugate bi,

{bi , cj } = δij

for each constraint:

Q = ci Gi − i 12 ci cj fji k bk
The second term has been added to insure the Poisson bracket or commutator

{Q, Q} = 0

so that its crossterm cancels the square of the first term, while its own square vanishes
by the Jacobi identity f[ij l fk]l m = 0. Quantum mechanically, the BRST operator is
nilpotent:
quantum mechanically {Q, Q} ≡ 2Q2 = 0
We can also describe the ghost dependence by the “ghost number”

J = ci bi ⇒ [J, Q] = Q

(Quantum mechanically, we need to normal order these expressions for Q and J.)
The BRST operator provides a convenient method to treat more general gauges
than λ = f, such as ones where the gauge fields become dynamical, which will prove
290 VI. QUANTUM GAUGE THEORY

useful particularly in relativistic theories. Now the original physical observables A

will satisfy

[Gi , A] = [bi , A] = [ci , A] = 0 ⇒ [Q, A] = [J, A] = 0

and similarly the physical quantum mechanical states |ψi will satisfy

(G0 , G− ; b0, b− ; c+)|ψi = 0 ⇒ Q|ψi = J|ψi = 0

where we have used the fact the only nonvanishing structure constants are f000 , f0++,
f0−− , f++ + , f−− −, and f+− k . (In the quantum case there are also some subtleties
due to normal ordering.) We also have the gauge invariances

δA = {Q, Λ}, δ|ψi = Q|λi

for arbitrary operators Λ and (unrelated) states |λi, since the Q-terms won’t con-
tribute when evaluating matrix elements with states annihilated by Q:

(hψ1 | + hλ1 |Q)(A + {Q, Λ})(|ψ2i + Q|λ2 i) = hψ1 |A|ψ2i

The gauge invariances are consistent with the constraints because of the nilpotence
of the BRST operator. (So Q(|ψi + Q|λi) still vanishes, etc.) States satisfying

Q|ψi = 0, δ|ψi = Q|λi

(i.e., we identify states that differ by Q on something) are said to be in the “coho-
mology” of Q (“BRST cohomology”), and operators satisfying

[Q, A] = 0, δA = {Q, Λ}

are said to be in its “operator cohomology”. (The latter cohomology also has a
classical analog.)
Excercise VIA1.1
Assume that each physical state can be represented as a physical observable
(Hermitian operator) acting on a ground state, which is itself physical:

|ψi = A|0i, A = A† , Q|0i = 0

Show how this relates the gauge parameters and cohomologies of |ψi and Q.
The BRST operator incorporates the ghosts that are necessary to generalize treat-
ment of the constraints to the nonabelian case: For example, to reproduce the gauge
transformations of the Abelian case, we choose

|λi = b+ |ζ +i, Q|ζ +i = 0 ⇒ δ|ψi = Q|λi = Ĝ+ |ζ +i

 A. BECCHI-ROUET-STORA-TYUTIN 291

where
Ĝi = {Q, bi} = Gi − icj fji k bk
are the “gauge-fixed” constraints, which include an extra term to transform the ghosts
as the adjoint representation. They reduce to just Gi in the Abelian case, but add
ghost terms to the gauge transformation law otherwise.
In particular, the Hamiltonian is a physical operator describing the energy and
the time development, so we can write
 R   R 
−i dt H
H = Hgi + {Q, Λ}, [Q, Hgi ] = 0 ⇒ T e = T e−i dt Hgi + {Q, κ}

for some κ. This includes gauge fixing for the gauge λi = f i discussed above, using:
. .
Λ = f i bi ⇒ H = Hgi + f i Ĝi , L = −(qm pm − ici bi) + H

The ghost terms in Ĝi only affect the time development of unphysical states in this
gauge.
For example, when calculating S-matrix elements, the result is independent of
the gauge choice Λ, as long as both the gauge-invariant Hamiltonian Hgi and the
states are BRST invariant. (Hgi commutes with Q, the initial and final states are
annihiliated by it.) It is also independent of the gauge choice |λi for |ψi → |ψi+Q|λi.
(Such a “residual gauge invariance” persists even though the asymptotic states satisfy
the free field equations.)
In the cases of interest in relativistic physics, the constraints always consist of a
linear term depending only on the canonical momenta p (conjugate to the fundamental
variables q), at least after some redefinitions, plus higher-order terms, which can be
treated perturbatively. Therefore, as the simplest nontrivial example, we consider a
model with a single variable q, with

Hgi = 0, G=p ⇒ Q = cp

If we assume boundary conditions on the wave functions such that they can be Taylor
expanded in q (they can always be expanded in c), we can write
X
∞
ψ= (αn + cβn ) n!1 q n
n=0

and similarly for Λ. Examining δψ = QΛ, we see that we can easily gauge βn = 0 for
all n. Looking at Qψ = 0, we then find that αn = 0 for all n except n = 0, so only
the constant piece of ψ survives. In other words, the cohomology is given by

Qψ = 0, δψ = QΛ ⇒ pψ = bψ = 0
292 VI. QUANTUM GAUGE THEORY

So, solving for the cohomology of Q = cp is the same as solving the constraint pψ = 0
without ghosts.
Excercise VIA1.2
Find the cohomology of the BRST operator

Q = ca† + c† a

for creation and annihilator operators satisfying

[a, d†] = [d, a†] = {c, b† } = {b, c†} = 1

(the other commutators vanishing), by expanding in creation operators a†, b†,

c† , d† about a vacuum state destroyed by the annihilation operators a, b, c, d.
(This is the common alternative to the boundary conditions used for Q = cp
above.)

2. Lagrangian
To obtain more interesting gauges we need some extra bosonic variables, such
as the gauge fields λi that we lost along the way, and their canonical conjugates
(“Nakanishi-Lautrup fields”) Bi ,

[Bi , λj ] = −iδij

as well as their corresponding ghosts (“antighosts”) c̃i . We can do this in a trivial

way by including constraints that set B to zero:

Q = ci Gi − i 12 ci cj fji k bk + b̃i Bi , J = ci bi − c̃i b̃i

where c̃i is conjugate to b̃i ,

{c̃i , b̃j } = δij
As a simple example, consider

Λ = λi bi ⇒ {Q, Λ} = λi Ĝi − ib̃ibi

The action now includes the gauge fields and all the ghosts as dynamical variables:
. .
. .
L = −(qm pm + λi Bi − ici bi − ib̃i c̃i ) + Hgi + {Q, Λ}

For this gauge we can eliminate b and b̃ by their equations of motion; assuming Gi
is only linear in p, we then can eliminate p to return completely to a Lagrangian
formalism:
. .
L = Lgi (q, λ) − λi Bi − i(∇tci )c̃i
 A. BECCHI-ROUET-STORA-TYUTIN 293

where Lgi represents the original gauge-invariant action (which depended on both q
and λ, including time derivatives), and ∇t is the covariant (time) derivative:
.
∇tci = ci + cj λk fkj i
.
The gauge condition (from varying B) is now λ = 0, generalizing the non-derivative
gauges found without the antighosts and Nakanishi-Lautrup fields. Correspondingly,
the ghost term is now second order in derivatives.
Excercise VIA2.1
Consider the general gauge choice

Λ = λi bi + [F i(q, p) + E i (B)]c̃i

⇒ {Q, Λ} = λi Ĝi − ib̃i bi + (F i + E i )Bi + ci [Gi , F j ]c̃j

where F i are some arbitrary functions of the original variables, and E i are
functions that effectively average over the types of gauges produced by F i.
Find the gauge-fixed Hamiltonian and Lagrangian. In the case where E is
linear in B, eliminate B, b, and b̃ from the Lagrangian by their algebraic
equations of motion.
Now that we understand the principles, all these manipulations can be performed
directly in the Lagrangian formalism. This will have the advantage that in field
theory the Lagrangian is manifestly Lorentz covariant, while the Hamiltonian (or the
.
Lagrangian in the Hamiltonian form −qp+H) is not, because of the way it singles out
time derivatives (and not spatial ones). (Consider, e.g., electromagnetism.) Similarly,
the gauge Gi = 0 is usually not Lorentz covariant. We can work with just the original
variables q, λ plus the new variables B, c, c̃, and define Q by the transformation it
induces (as derived from the Hamiltonian formalism):
.
Qq m = ci δi q m, Qλi = −i(ci +cj λk fkj i ), Qci = −i 12 cj ck fkj i , Qc̃i = Bi , QBi = 0

where δi is the gauge transformation induced by Gi ([Gi , ] in the Hamiltonian for-

malism). Note that the BRST transformations of the original variables are exactly
the same as the gauge transformations, with the gauge parameters replaced with
the corresponding ghosts. We can also consider the Nakanishi-Lautrup fields B as
original variables, with the fact that they don’t occur explicitly implying they have
constraints B = 0. Alternatively, we can treat the antighosts c̃ as pure gauge degrees
of freedom, with their own nonderivative gauge transformation δc̃ = λ̃ that allows
them to be completely gauged away.
294 VI. QUANTUM GAUGE THEORY

The Lagrangian can be gauge-fixed directly as

L = Lgi + QΛL

where in the case just considered

.
ΛL = −λi c̃i
.
gives the same L as above for the λ = 0 gauge. In the simpler case described earlier
(the gauge λ = constant)
ΛL = (λi − f i )c̃i

This gives the result, for the simplest choice f = 0,

. .
L = Lgi (q, λ) + λi Bi − i(ci + cj λk fkj i )c̃i → Lgi (q, 0) − ici c̃i

after eliminating the Lagrange multipliers B and λ by their algebraic equations of

motion. Note that ΛL = λi c̃i corresponds to the Hamiltonian formalism’s Λ = 0.
Thus, in the Hamiltonian formalism we never quantize with H = Hgi + λG, but
can use just Hgi and Λ = 0, which is equivalent to using Hgi + {Q, Λ} for any Λ,
while in the Lagrangian formalism we can never quantize with just Lgi (q, λ), or even
Lgi (q, 0), and ΛL is never zero, but must be chosen so as to break the gauge invariance.
.
However, the extra term for Lgi (q, 0) is just the cb term found from converting Hgi to
the Lagrangian formalism.
Excercise VIA2.2
Repeat excercise VIA2.1 directly in the Lagrangian formalism. (Find ΛL,
etc.)
All our results for quantization apply equally well in the path-integral formalism,
which can be applied to either the Hamiltonian or Lagrangian. (Of course, for field
theory we will be interested in applying BRST to path integrals for Lagrangians.)
We then evaluate matrix elements as
Z
A = Dφ Ψ[φ]e−iS[φ]; S = Sgi + QΛ, Ψ = Ψgi + QΛΨ; QSgi = QΨgi = 0

Sgi and Ψgi depend on just the physical fields (no ghosts), so they are gauge invari-
ant as well as BRST invariant. For S-matrices, since Ψ is an asymptotic state, the
BRST operator used for its constraint and gauge invariance can be reduced to its
free part: Q then acts on only the gauge fields. The statement of gauge invariance of
Ψgi is then equivalent to the requirement that gauge fields appear in it only as their
 A. BECCHI-ROUET-STORA-TYUTIN 295

Abelian field strengths. For example, the usual gauge vector A describing electro-
Q
magnetism appears in single-particle factors in the wave functional (Ψ[φ] = Ψ1 [φ]
as in subsection VC1) only as:

Ψ1 [A] = hψ a |Aai, δAa = −∂aλ, ψa = λ=0

⇒ 0 = δΨ1 = hψ a |δAai = h∂a ψ a |λi

using ∂0hψ|λi = 0 (where the relativistic inner product h | i was defined in subsection
VB2). The transversality of ψ a is equivalent to coupling to the Abelian field strength,
since
∂aψ a = 0 ⇒ ψ a = ∂b ψ ab ⇒ hψ a |Aai = 12 hψ ab|Fab i
in terms of an antisymmetric-tensor external-line factor ψ ab.

3. Particles
We have seen that the relativistic particle (with or without spin) is a simple
example of a contrained system. For the simplest case, spin 0, the BRST operator
follows simply from the single constraint:

Q = c 12 ( − m2 )

Unlike the nonrelativistic case, the relativistic “Hamiltonian” is identified with this
constraint. Since we know constraints are treated by the BRST operator, we can
consider writing the field theory action in terms of it:
Z
S0 = − dx dc 12 ΦQΦ

Using the explicit c dependence of the field Φ = φ−icψ, we find the usual scalar kinetic
term. φ is thus the usual field, while ψ is an “antifield”, which has opposite statistics
to φ (fermion instead of boson). We’ll see in chapter XII that Q can be constructed
straightforwardly for arbitrary spin, and has a simple expression in term of generalized
spin operators. (As in nonrelativistic theories, spin is easier to treat directly in
quantum mechanics rather than by first-quantization of a classical system.) The
kinetic term then generally can be written as a slight modification of the above. Then
the antifields will be found to play a nontrivial function, rather than just automatically
dropping out as in this case.
From the constraints and their algebra for spin 1/2 (see also excercise IIIB1.2)
we find the BRST and ghost-number operators:

Q = c 12 ( ˜
− m2) − ξ(γ · ∂ − i √m2 ) − ξ 2 b + ξµ, J = cb + ξζ + ξ˜ζ̃
296 VI. QUANTUM GAUGE THEORY

where ξ and its conjugate ζ are bosonic ghosts, and we have added a nonminimal
term with boson ξ˜ (conjugate ζ̃) and fermion µ (conjugate κ) to allow gauges general
enough for first-quantization:

˜ = {κ, µ} = 1
[ζ, ξ] = [ζ̃, ξ]

For convenience, we also have chosen ξ (and ζ) to anti commute with γ,

{ξ, γ a } = {ζ, γ a } = 0

to avoid having to replace −i √m2 with γ−1 im; this has the natural interpretation of
treating ξ and ζ as bosonic (ghost) components of the γ matrices (see subsections
XIIA4-5,B5).
Note that [Q, ξ] = 0, but {Q, A} 6= ξ for any A, so ξ is in the operator coho-
mology of Q. Normally, this would imply infinite copies of the physical states in
the cohomology, since applying a “translation” with the ghost variable ξ gives a new
state in the cohomology from any given one. The nonminimal variables allow us to
avoid this problem by combining ξ with ξ˜ to produce harmonic oscillator creation
and annihilation operators:

ξ= √1 (a
2
+ a† ), ξ˜ = √1 i(a†
2
− a), ζ= √1 (ã
2
− ㆠ), ζ̃ = − √12 i(ㆠ+ ã)

[a, ㆠ] = [ã, a† ] = 1, rest = 0

This allows us to define a ground state

a|0i = ã|0i = 0

which breaks the translation symmetry of ξ. In chapter XII we’ll show in a more
general framework how the ΦQΦ type of action then reproduces the Dirac action.

4. Fields
As described in subsection VIA2, we can perform gauge fixing through BRST,
including the introduction of ghosts, directly on the Lagrangian at the classical level.
Also, the BRST transformations on the physical fields are just the gauge transforma-
tions with the gauge parameters replaced by ghosts, and the BRST transformation on
the ghosts is quadratic in ghosts times the structure constants, while on the antighosts
it gives the Nakanishi-Lautrup fields, and it annihilates the NL fields. In the case of
Yang-Mills we then have

QAa = −[∇a, C], QC = iC 2, QC̃ = −iB, QB = 0

 A. BECCHI-ROUET-STORA-TYUTIN 297

while for matter transforming as δφ = iλφ we have

Qφ = iCφ

where we have used matrix notation for the group algebra, as usual. There are two
minor differences from the transformation rules we used in our general discussion
previously: (1) We have included an extra “i” in our definition of the relativistic Q,
for a convenience that will become apparent only when we relate relativistic first- and
second-quantization (see chapter XII). (2) There is a relative sign difference for QC
because now Q is second-quantized while Gi is still first-quantized (i.e., matrices).
More explicitly, we have, e.g.,

C 2 = 12 {C, C} ⇒ QC i = − 12 C j C k fkj i ; Qφ = iC iGi φ

The gauge-fixed action is then the gauge-invariant action plus the BRST trans-
formation of some function Λ:

Sgf = Sgi − iQΛ

For example, consider Yang-Mills in the most common type of gauge, where some
function of A is fixed:
Z
∂f
Λ = tr 12 C̃[f(A) + 12 αB] ⇒ Lgf = Lgi − 12 B[f(A) + 12 αB] − 12 iC̃ · [∇, C]
∂A
for some constant α. For α = 0, B is a Lagrange multiplier, enforcing the gauge
f(A) = 0, while for α 6= 0, we can eliminate B by its auxiliary field equation:

− 12 B[f(A) + 12 αB] → 1 2
4α
f

Examples will be given in the following section.

In field theory gauge-fixing functions always have linear terms, as do gauge trans-
formations. Furthermore, there always exist “unitary gauges”, where no ghosts are
required. The ghost terms in general gauges serve simply to provide the appropriate
Jacobian factor for the field redefinition that transforms from the general gauge to
the unitary gauge, which appears at the quantum level from functionally integrating
out the ghosts. The simplest example is the trivial gauge invariance that occurs in
the Stückelberg model:

QA = −∂C, Qφ = C, QC̃ = −iB, QC = QB = 0

which we can fix with

−iQ(C̃Oφ) = −BOφ + iC̃OC
298 VI. QUANTUM GAUGE THEORY

for some field-independent operator O. Functionally integrating out B still sets φ = 0,

but produces an inverse functional determinant of O (from redefinition of φ, or from
δ(Oφ)), canceled by that from integrating out the ghosts. The advantage of BRST
is that all this can be treated at the classical level, in terms of the classical action,
without regard to functional integration, while directly giving a solution that can be
expressed immediately in terms of Feynman rules.
Excercise VIA4.1
Show that the gauge fixing

−iQ(C̃Oφ + C̃AB + CBφ + CCB)

where O, A, B, and C are field-independent operators, gives a result equivalent

to the previous, by considering functional determinants or field redefinitions.
Excercise VIA4.2
Show that the Lagrangian

AAB + CBD → AABD

by the field redefinition

D → D + B −1 B

for bosons A, B, C, D and operators A, B. This is the classical equivalent of

det(AB) = det(A)det(B).
These methods apply straightforwardly to supersymmetric theories in superspace:
From the gauge transformations of subsection IVC4,

QeV = iC̄eV − ieV C; QC = iC 2, QC̄ = iC̄ 2

QC̃ = −iB, QC̃ = −iB̄; QB = QB̄ = 0

where C, C̃, and B are chiral superfields, and C̄, C̃, and B̄ their hermitian conjugates.
In practice, this BRST approach is sufficient for gauge fixing. In particular, this
is true for the fundamental fields used in the standard model (including gravity),
which have spin≤2. Therefore, we’ll use mostly this approach in the rest of this text.
However, some observed hadrons have much higher spin. The first-quantized approach
of chapter XII gives a natural and direct way of understanding ghosts and BRST for
the fields describing such particles, and translates directly into the treatment of Zinn-
Justin, Batalin, and Vilkovisky (ZJBV) for field theory.
 A. BECCHI-ROUET-STORA-TYUTIN 299

REFERENCES
1 Feynman, loc. cit. (VC, ref. 17);
B. DeWitt, Phys. Rev. 162 (1967) 1195, 1239;
L.D. Faddeev and V.N. Popov, Phys. Lett. 25B (1967) 29;
S. Mandelstam, Phys. Rev. 175 (1968) 1580:
ghosts.
2 N. Nakanishi, Prog. Theor. Phys. 35 (1966) 1111;
B. Lautrup, K. Dan. Vidensk. Selsk. Mat. Fys. Medd. 34 (1967) No. 11, 1.
3 C. Becchi, A. Rouet, and R. Stora, Phys. Lett. 52B (1974) 344, Ann. Phys. 98 (1976)
287;
I.V. Tyutin, Gauge invariance in field theory and in statistical physics in the operator
formulation, Lebedev preprint FIAN No. 39 (1975), in Russian, unpublished;
M.Z. Iofa and I.V. Tyutin, Theor. Math. Phys. 27 (1976) 316;
T. Kugo and I. Ojima, Phys. Lett. 73B (1978) 459;
L. Baulieu, Phys. Rep. 129 (1985) 1:
BRST.
4 E.S. Fradkin and G.A. Vilkovisky, Phys. Lett. 55B (1975) 224;
I.A. Batalin and G.A. Vilkovisky, Phys. Lett. 69B (1977) 309;
E.S. Fradkin and T.E. Fradkina, Phys. Lett. 72B (1978) 343;
M. Henneaux, Phys. Rep. 126 (1985) 1:
Hamiltonian BRST.
300 VI. QUANTUM GAUGE THEORY

............................. .............................
............................. B. GAUGES .............................

There are two important properties of gauges we have examined: (1) Gauges
which eliminate some degrees of freedom, such as lightcone or unitary gauges, are
simpler classically, which makes them easier to understand physically. (2) Gauges that
manifest as many global invariances as possible, such as the Fermi-Feynman gauge,
will be found later to simplify quantum calcuations, because the explicit momentum
dependence of the propagator or vertices is simpler, and keeping a symmetry manifest
makes it unnecessary to check. In this section we’ll examine these gauges in greater
detail, especially as they relate to intereacting theories.
We’ll study also some special gauges, with nontrivial interaction terms, that have
both of these properties to some extent. In particular, they are manifestly Lorentz
covariant, but avoid many of the complications associated with ghosts.

1. Radial
We know from nonrelativistic classical and quantum mechanics that the equations
of motion can be solved exactly only for certain simple external field configurations.
One particular case we have already emphasized is that of an action quadratic in the
dynamical variables, i.e., the harmonic oscillator and its generalizations. Higher-order
terms are then treated as perturbations about the exact solution. Such an expansion
in the coordinates x is the particle version of the JWKB expansion in h̄: Calling the
√
“classical” part of x “y”, we substitute x → y + h̄x and Taylor expand in x. (From
now on we’ll drop the h̄’s, and just remember to perturb about the quadratic terms.)
For the scalar field we write

φ → φ + x · ∂φ + 12 xm xn∂m ∂nφ + ...

where ∂...∂φ is implicitly evaluated at y.

For the gauge fields we would like to be a bit more clever: For example, for the
electromagnetic potential Am we know we can always add a constant, so Am(y) is
irrelevant, while for ∂A only Fmn = ∂[m An] is gauge invariant. This means we want to
choose a gauge best suited to this calculation: a gauge that both eliminates as many
as possible of the lower-order terms, and expresses A(y + x) in terms of only F (y)
and its derivatives. Similarly, we should have a Taylor expansion for charged fields
in terms of covariant derivatives. The appropriate gauge, which easily can be found
explicitly, is the “radial gauge”

x · A(y + x) = 0
 B. GAUGES 301

(Note that, unlike F , A depends on x independently of y, not just as y + x, since the

gauge condition itself is x-dependent. We write A(y + x) only to indicate that A is
evaluated at position y + x.) One way to solve this condition is to use the identity

xnFnm = (x · ∂ + 1)Am − [∇m, x · A] (∂ = ∂/∂x)

which follows from the definition of F . Using the gauge condition, we then can write
1
Am = xnFnm
x·∂ +1
Alternatively, we can replace x everywhere (including the argument y + x) by τ x, and
then identify x · ∂ = τ ∂/∂τ to find

τ xn Fnm (y + τ x) = ∂τ τ Am (y + τ x)

Integrating both sides over τ from 0 to 1, we find

Z 1
Am (y + x) = dτ τ xnFnm (y + τ x)
0

Note in particular that A(y) = 0.

Another way to define this gauge is to consider gauge covariant translation from
y to y + x to produce a gauge transformation from an arbitrary gauge to the radial
gauge. Writing the covariant derivative at y as

D = D + iA(y), D = ∂/∂y

we know from subsection IIIC2 that

ψ 0(y + x) ≡ ex·D ψ(y) = eiΛex·D ψ(y) = eiΛψ(y + x)

so that covariant translation produces a ψ 0(y + x) that is the same as ψ(y + x) up to

a gauge transformation. Thus the gauge-transformed ψ can be written as a covariant
Taylor expansion (for purposes of perturbation) about y:
X
∞
0
ψ (x + y) = 1 a1
n!
x · · · xan (Da1 · · · Dan ψ 0 )(y)
n=0

In particular, ψ 0 (y) = ψ(y).

However, we want to define a covariant derivative with respect to x (not y), so
that

∇ = ∂ + iA0(x + y), ∇ψ 0(y + x) = (Dψ)0 (y + x) = ex·D (Dψ)(y)

302 VI. QUANTUM GAUGE THEORY

Using
∂ψ(y) = 0

we find the solution

∇ = ex·D De−x·D mod ex·D ∂e−x·D

where the latter term vanishes on ψ 0 (y + x), so the right amount of it can be added
to the former expression to cancel any D terms. The result is

∇ = ex·D (∂ + D)e−x·D

This implies x · A0 (y + x) = 0 directly: Contracting both sides with x, the Taylor

expansion of the right-hand side terminates after the first couple of terms. Taylor
expanding the uncontracted expression, we have
X
∞
A0a(y + x) = 1 1
n! n+2
(x · D)n xbFba (y)
n=0

We can also write

∂ + iA0(x + y) = ∇ = ex·D e−x·D [∂ + iA(x + y)]ex·De−x·D = eiΛ[∂ + iA(x + y)]e−iΛ

and
∇ = ex·D e−y·∂ Dey·∂ e−x·D

Excercise VIB1.1
Show this Taylor expansion is equivalent to that obtained from the first
method used in this section to solve the gauge condition. (Hint: Look out for
hidden x and y dependence — How does x · ∂ on ψ 0 or F 0 relate to x · D?
Also beware of notation: In the first construction we did not use a gauge
transformation, so no primes were used.)
Thus, to just quadratic order in x, the mechanics action for a relativistic particle
in external fields (subsection IIIB3) becomes
Z
. . .
SL ≈ dτ {− 12 v −1 ηmn xm xn + 12 xm xnFnm (y)

+v[φ(y) + xm (∂m φ)(y) + 12 xm xn (∂m ∂nφ)(y)]}

To this approximation the classical equations of motion are linear and can be solved
exactly. It can also be used to find exact solutions for constant electromagnetic fields.
 B. GAUGES 303

2. Lorentz
For purposes of explicit calculations in perturbation theory, it’s more convenient
to use gauges where Lorentz covariance is manifest. “Lorentz gauges” are a class of
gauges using
f = ∂ ·A

(and similarly for other gauge fields) as the gauge-fixing function. From the discussion
of subsection VIA4, we have from the usual BRST as applied to Yang-Mills

2
Lgf = 18 Fab − iQ 12 [C̃(∂ · A + 12 αB)]

= − 14 A · A − 14 (∂ · A)2 − 12 [Aa, Ab](−i∂aAb + 12 AaAb ) − 12 B∂ · A − 14 αB 2 − 12 iC̃∂ · [∇, C]

After eliminating B by its field equation, the kinetic terms are

− 14 A · A − 14 (∂ · A)2 + 1
4α
(∂ · A)2 − 12 iC̃ C

In particular, for α = 1 we have the “Fermi-Feynman” gauge, which gives the nicest
propagators. (It is also the gauge that follows automatically from a first-quantized
BRST construction, which will be described in chapter XII.) More generally, we find
the propagator from inverting the kinetic operator: For the ghosts this is always 2/p2 ,
but for Aa,  
a b −1 ηab pa pb
2[η p + ( α − 1)p p ] = 2 2 + (α − 1) 2 2
ab 2 1
p (p )
For α = 0 this is the “Landau gauge”, which has the advantage that the propagator
is proportional to the transverse projection operator. (It kills terms proportional to
pa .) However, α = 1 is clearly the simplest, and the 1/p4 term can cause problems in
perturbation theory.
Excercise VIB2.1
In the Abelian case, consider making a gauge transformation on the gauge-
fixed action (including matter), with λ ∼ −1 B. Show that the only effect
is to change the value of the coefficient α of the B 2 term. Find a similar
transformation for the form of the action where B has been eliminated. This
shows explicitly the decoupling of the longitudinal mode of the photon.
Excercise VIB2.2
Show that for general A and B
 
−1 1 B
(η A + p p B)
ab a b
= ηab − pa pb
A A + p2 B
304 VI. QUANTUM GAUGE THEORY

Note that in the Abelian case the lightcone gauge is a special case of the Landau
gauge. (An analogous situation occurs in the classical mechanics of the particle for
the gauges of the worldline metric, as discussed in subsection IIIB2.) Here we have

0 = na (∂ b Fab) = n · ∂(∂ · A) − (n · A)

In the lightcone formalism, this is the field equation that comes from varying the
auxiliary field. In the lightcone gauge n · A = 0, it implies ∂ · A = 0 (and thus also
Aa = 0), since n · ∂ is invertible.
This is particularly useful in D=4, where we can generalize from the lightcone to
a Lorentz-covariant form by using twistors: From subsection IIB6,
. . . .
p2 = 0 ⇒ pαα = (p0 )pα p̄α , n2 = 0 ⇒ nαα = (n0 )nαn̄α

Massless spinors are described on shell in momentum space by

. .
ψ α = pα φ, ψ̄ α = p̄α φ̄

where external-line factors for Feynman diagrams are given by setting φ = 1. For
massless vectors, we have p · A = n · A = 0 (but n · p 6= 0), so depending on whether
the helicity is +1 (self-dual field strength) or −1 (anti-self-dual field strength), we
find, respectively,
.
.αβ. . . . nα p̄β
f¯ ∼ p̄α p̄β , f αβ = 0 ⇒ Aαβ = γ φ̄
n pγ
.
. . . pα n̄β
f αβ ∼ pα pβ , f¯αβ = 0 ⇒ Aαβ = . φ
n̄γ p̄γ.
The normalization of A has been chosen compatible with |φ| = 1 and AaA*a = 1 for
evaluating cross sections. In a general Landau gauge the arbitrary gauge-dependent
.
polarization spinors nα , n̄α can be chosen independently for each external line, since
gauge invariance means independent gauge parameters for different momenta. (This
method is known as “spinor helicity”.) However, in a lightcone gauge the polarization
spinors are constant.
The lightcone gauge condition is thus again a stronger gauge condition than
Lorentz gauges, as expected from the fact that it has fewer derivatives. This differ-
ence shows itself in various ways: (1) In perturbation theory on shell, in the lightcone
frame the Landau gauge condition 0 = p · A = −p+ A− kills A− but says nothing
about A+, which can be eliminated by the residual gauge invariance δA+ = p+ λ to
obtain the lightcone gauge. (2) In perturbation theory off shell, more derivatives in
 B. GAUGES 305

the gauge transformation imply more derivatives in the ghost kinetic operator. Thus,
more ghost degrees of freedom are introduced to cancel the extra unphysical degrees
of freedom in the gauge field. (3) Lorentz gauges also have a nonperturbative ambi-
guity (the “Gribov ambiguity”) that axial gauges avoid: Nonperturbative solutions
to the gauge condition can be found that differ from the perturbative one, in the
nonabelian case. Specifically, it is possible to find a nontrivial gauge transformation
g (∇0 = g−1 ∇g) such that

0 = ∂ · A0 = −i∂ · g−1 (∇g) for ∂ · A = 0

even when g is required to satisfy boundary conditions that it approach the identity
at infinity (except in the Abelian case, where g = eiλ ⇒ λ = 0 ⇒ λ = 0). This is
not the case for axial gauges, where

0 = n · A0 = n · g−1 (∇g) for n·A =0 ⇒ g−1 (n · ∂g) = 0 ⇒ g=I

by simply integrating from infinity.

3. Massive
In subsection IIB4 we described the introduction of mass for the vector by dimen-
sional reduction, giving the Stückelberg formalism for a massive (Abelian) gauge field.
The gauge-invariant action (subsection IVA5) and BRST transformation laws (sub-
section VIA4) followed from adding an extra dimension and setting the corresponding
component of the momentum equal to the mass:
2
Lgi = 18 Fab + 14 (mAa + ∂a φ)2

QAa = −∂aC, Qφ = mC, QC̃ = −iB, QB = 0

where the scalar is the extra component of the vector.
There are two obvious covariant gauges for such a vector: (1) The “unitary gauge”

f =φ

simply gauges away the scalar. Since the scalar has a nonderivative gauge transfor-
mation, the ghosts do not propagate: The gauge-fixing term

−iQ(C̃φ) = −Bφ + imC̃C

simply eliminates the scalar and ghosts as auxiliary fields. The net result is that we
could have simply chosen
gauge φ = 0
306 VI. QUANTUM GAUGE THEORY

and ignored ghosts because of φ’s nonderivative transformation law. Thus the gauge-
fixed Lagrangian is just the result of adding a mass term to the massless Lagrangian:

Lgf = 18 Fab
2
+ 14 m2A2

But the propagator is

 
a b −1 ηab pa pb
2[η (p + m ) − p p ]
ab 2 2
=2 2 2
+ 2 2
p +m m (p + m2)
Notice that the second term is higher in derivatives than the first; this can cause some
technical problems in perturbation theory.
(2) The Fermi-Feynman gauge works similarly to the massless case. We then
modify the gauge-fixing function to

f = ∂ · A + mφ

so

−iQ[ 12 C̃(∂ · A + mφ + 12 B)] = − 12 B(∂ · A + mφ + 12 B) − 12 iC̃( − m2)C

⇒ Lgf = 18 F 2 + 14 (mA + ∂φ)2 + 14 (∂ · A + mφ)2 − 12 iC̃( − m2 )C

= − 14 A · ( − m2)A − 14 φ( − m2)φ − 12 iC̃( − m2 )C

The propagators are again simpler. The vector has D propagating components instead
of just the D−1 physical ones; the 2 ghosts cancel φ and the extra component in A.
Excercise VIB3.1
Generalize the Fermi-Feynman gauge for the Stückelberg formalism to the
“renormalizable gauges” with gauge-fixing function
m
f= ∂ · A + µφ
µ
Show that the ghosts and φ have mass µ, while the vector propagator has the
form  
pa pb  1 pa pb 1
2 ηab + 2 − 2 2
m p2 + m2 m p + µ2
This shows explicitly the second unphysical bosonic mode of mass µ to cancel
the 2 ghosts, as well as the 3 transverse physical modes of mass m. Look at
the cases 
 0 (Landau gauge)
µ = m (Fermi-Feynman gauge)

∞ (unitary gauge)
 B. GAUGES 307

These two choices of gauge also exist for Yang-Mills theories exhibiting the Higgs
mechanism, since those models give the Stückelberg model when linearized about
the vacuum values of the fields. The advantages are the same: The unitary gauge
eliminates as many unphysical degrees of freedom as possible (see subsection IVA6
for an example), while the Fermi-Feynman gauge gives the simplest propagators.
Excercise VIB3.2
Work out the Fermi-Feynman gauge for an arbitrary Higgs model, generalizing
the analysis for the Stückelberg case.

4. Gervais-Neveu
We next consider pure Yang-Mills theory for the gauge group U(N), but use a
complex gauge-fixing function

f0 = ∂ · A + iA2

where Aa is a vector of hermitian N×N matrices, and A2 ≡ Aa Aa. (The hermitian

conjugate, i → −i, gives similar results.) The gauge-fixed Lagrangian (in the action
R
S = g −2 tr L) is then

LA = 18 F 2 + 14 f02 = − 14 A · A − iAaAb ∂b Aa − 14 AaAb AaAb

(where is the free D’Alembertian) while the ghost action can be written as

LC = − 12 iC̃∇2 C − 12 C̃Cf0

where ∇ acts on C as if it were in the defining representation (i.e., ∇C = ∂C + iAC,

not [A, C]). This “Gervais-Neveu gauge” already has the simplification that some of
the terms in the Yang-Mills self-interaction have been canceled.
Excercise VIB4.1
Consider the “anti-Gervais-Neveu gauge”, where the same gauge-fixing term
is added with opposite overall sign.
a Show the resulting Lagrangian can be written as

LA ∼ tr[(∂/ A / 2)2]
/ + iA

where the trace is with respect to both (N×N) internal and (4×4) Dirac
matrices. Thus, spin can be treated in a manner closely analogous to internal
symmetry.
308 VI. QUANTUM GAUGE THEORY

b Show the propagator can be written in the form of the product of 2 (massless)
Dirac-spinor propagators.
c Starting with the complex first-order formulation of Yang-Mills of subsection
IIIC4, show that the action can be written in a way that replaces the above
4×4 matrices with 2×2 matrices, as

LA ∼ tr[Ĝ2 + Ĝ(∂A* + iAA*)]

in first-order form, where now Ĝ is neither traceless nor symmetric in spinor

indices (its trace is the Nakanishi-Lautrup field), or in second-order form as

LA ∼ tr[(∂A* + iAA*)2]

(Note that this differs from the above Dirac form, as expanded in 2×2 matri-
ces, because it includes the Chern-Simons term.)
Next, consider a model where the Yang-Mills fields couple to scalars that are also
represented by N×N matrices, but that are in the defining (N-component) represen-
tation of the gauge (“color”) U(N), while also being in the defining representation of
a second, global (“flavor”) U(N). (See subsection IVA6.) This complex field thus has
2N2 real components compared to the N2 gauge vectors, and the 2N2 ghosts. We also
choose a Higgs potential such that the masses of the scalar and vector come out the
same (but we can also specialize to the massless case). The scalar Lagrangian is then
(again with g −2 tr in the action)

Lφ = − 12 φ†∇2 φ + 14 R2 , R = φ† φ − 12 m2

Finally, we modify the gauge-fixing function to

f = f0 + iR

With this choice, the ghost terms are unmodified (R is gauge invariant), but the
scalar self-interaction is completely canceled (including the mass term). The total
Lagrangian is then

L = (LA + 14 m2 A2) + (− 12 φ† ∇2φ + i 12 φ† φf0 ) + i(− 12 C̃∇2C + i 12 C̃Cf0)

Since the scalar Lagrangian is identical in form to that of the ghosts, and neither
has self-interactions, functional integration over them will produce canceling func-
tional determinants, because they have opposite statistics. This is a reflection of the
fact that both sets of fields now describe unphysical polarizations, since both describe
 B. GAUGES 309

massless states in a theory where all physical states are massive (as seen, e.g., in a
unitary gauge). This has the great advantage that, for this particular model, both
the scalar fields and the ghosts can be dropped altogether, while the Lagrangian

L → LA + 14 m2A2

completely describes the physical massive vector and scalar states. This was possible
only because of the use of a complex gauge condition: The longitudinal component of
the vector is now imaginary, which fixes the wrong sign associated with the Minkowski
metric. A related characteristic of this gauge is that we nowhere needed to change
the vacuum value of any field, unlike other gauges for actions where there is a Higgs
effect.
We now note that this result for the massive case (and its massless limit) actually
can be obtained more easily than the result for the pure Yang-Mills case: Since the
final result has no ghosts, it is in a unitary gauge, where the vector not only “eats”
the usual compensating scalar, but “overeats” by absorbing the physical scalar. The
appropriate gauge condition is still complex and involves the scalars, but is now linear :

gauge φ = hφi = √1 mI
2

where φ†, treated as independent, is unfixed. (As for the usual unitary gauge Im φ =
0, i.e., φ = φ† , there are no propagating ghosts, since the gauge transformation of φ
has no derivatives.) In this gauge the action becomes quadratic in φ† :

Lφ → − 12 φ† i(∂ · A + iA2) √12 m + 14 (φ† √12 m − 12 m2)2

In fact, φ† appears as an auxiliary field (taking the place of the Nakanishi-Lautrup

field), so we can eliminate it by its equation of motion:
δ √
⇒ φ† = m
√ + 2
if0 ⇒ L = LA + 14 m2 A2
δφ† 2 m

This procedure is analogous to that used for the lightcone gauge, where one compo-
nent of the gauge field is fixed and one is eliminated as an auxiliary field: A closer
analogy will be found in subsection VIB6.
Of course, such gauges generalize to other Higgs models, but results will not be
as simple when the vector and scalar masses differ:
Excercise VIB4.2
Make the coefficient of the R2 term in Lφ arbitrary, so the masses of the
vector and scalar are unequal, but choose the same gauge φ = hφi. Find the
propagator, and compare with that of excercise VIB3.1.
310 VI. QUANTUM GAUGE THEORY

5. Super Gervais-Neveu
Nonhermitian gauges are also useful in supersymmetric theories: Here we consider
the supersymmetric analog of the massive model of the previous section. Although
we work in N=1 superspace, the model turns out to automatically have an N=2
supersymmetry. Just as the bosonic model ended with only a vector field describing
only physical polarizations, we now want a real scalar superfield to have only physical
polarizations. Since such a superfield has 8 bosonic components and 8 fermionic, while
massless N=1 multiplets have 2+2 physical polarizations, we need 1 vector multiplet
plus 3 scalar multiplets. Since the bosonic model had a complex scalar representation,
2 of these scalar multiplets must form the analogous defining ⊗ defining representation
of local ⊗ global groups, so the last must be a real (adjoint) representation of the
R
local group. The model is then given by (where again S = g−2 tr dx L)
Z Z
Lgi = − d θ W − d4 θ (e−V φ̄0eV φ0 + φ̄+eV φ+ + φ−e−V φ̄−)
2 2

Z 
− d θ (φ+φ− −
2 1 2
4
m )φ0 + h.c.

where we have included the only possible scale-invariant potential term, and intro-
duced a Higgs mechanism by an N=2 Fayet-Iliopoulos term, which we chose to write
in terms of the chiral scalar. (See subsection IVC7.)
The BRST transformations (which also imply the gauge transformations) for this
action are (see subsection VIA4)

QeV = iC̄eV − ieV C, Qe−V = −e−V (QeV )e−V = iCe−V − ie−V C̄

QC = iC 2, QC̄ = iC̄ 2; QC̃ = −iB, ˜ = −iB̄;

QC̄ QB = QB̄ = 0
Qφ+ = iCφ+, Qφ0 = i[C, φ0], Qφ− = −iφ−C
Qφ̄+ = −iφ̄+C̄, Qφ̄0 = i[C̄, φ̄0 ], Qφ̄− = iC̄ φ̄−

Our nonhermitian choice for the BRST gauge-fixing function is

Z Z
¯2 −V
Λ = − d θ C̃(d e + φ0 ) − d2 θ̄ C̄
2 ˜ (d2 eV + φ̄ )
0

Note that eV is an element of the algebra as well as a “nonunitary element” of the

group, only because we chose the group U(N) (as was the case for A2 in the bosonic
version). The gauge-fixing and ghost terms are then
Z Z
¯2 −V
−iQΛ = d θ B(d e + φ0) + d2 θ̄ B̄(d2 eV + φ̄0)
2

Z Z Z
−V ˜
− d θ (C̃e C̄ + C̄ e C) + d θ C C̃φ0 + d2 θ̄ C̄ C̄
4 V 2 ˜ φ̄
0
 B. GAUGES 311

where we have used the field equation enforced by the Lagrange multipliers B and
B̄ (or, equivalently, made field redefinitions of the Lagrange multipliers to generate
terms proportional to their constraints).
Excercise VIB5.1
Make a component analysis of this theory:
a Expand the gauge-invariant action in components.
b Do the same for the gauge-fixing terms.
c Compare the bosonic part of both the gauge-invariant and gauge-fixed actions
to those of the previous subsection, after elimination of auxiliary fields.
We now see that the ghost terms are identical in form to those for φ±, under the
identification
˜ , C̄)
(φ+ , φ−, φ̄+, φ̄−) ↔ (C, C̃, C̄
(but beware signs from ordering of ghosts). So again the ghosts cancel the (N=2)
matter fields, leaving only the N=2 vector multiplet. But we can also eliminate
the N=1 matter half of this N=2 multiplet using the Nakanishi-Lautrup Lagrange
multiplers: The final simple result for the gauge-fixed action is thus
Z Z
L = − d θ W − d4 θ [e−V (d2 eV )eV d¯2 e−V + 14 m2(eV + e−V )]
2 2

A further simplification results from the redefinition (again possible only for U(N))

eV → 1 + V

This also simplifies the BRST (and gauge) transformation for V :

QV = i(C̄ − C) + i(C̄V − V C)

whose linear form resembles the bosonic case. Using the expression (see excercise
IVC4.1)
1
Wα = −id̄2e−V dα eV → −id̄2 dα V
1+V
for the field strength, the Lagrangian becomes
Z 
1 1 1 1
L = − d θ − 12
4
(dα V )d̄2 dα V + (d2 V )(1 + V )d̄2
1+V 1+V 1+V 1+V
 
1
+ 14 m2 V +
1+V
Although the nonabelian vector multiplet has nonpolynomial self-interactions in
any gauge, this gauge simplifies the lower-point interactions, which are the ones more
312 VI. QUANTUM GAUGE THEORY

frequently used for a fixed number of external lines. Expanding this action to cubic
order, we use the identity
.
dα d¯2 dα = d¯α d2 d¯α. = − 12 + {d2 , d̄2 }
for the kinetic term, and
d¯α.d2 = d2 d¯α. + i∂αα.dα
for the gauge-fixing part of the cubic term, with integration by parts. (For the gauge-
2
invariant term, some work can be saved by using the equivalent W form.) The result
is Z .
L = d4 θ { 14 V ( − m2)V + [ 14 m2V 3 + (d¯α V )V i∂αα.dα V ] + O(V 4 )}
Not only are there fewer terms than with linear gauge conditions, but these terms have
fewer spinor derivatives, which yields viewer nonvanishing contributions in loops (see
subsection VIC5). As for the bosonic model of the previous subsection, this analysis
also applies for the unbroken case m = 0.
Excercise VIB5.2
Find the corresponding form of the kinetic and cubic terms without the re-
definition eV → 1 + V .
Excercise VIB5.3
Gauge fix by using the unitary gauge
m
√
φ+ = φ̄− = 2

to obtain the same result.

Excercise VIB5.4
Look at the super anti -Gervais-Neveu gauge, or super anti -Fermi-Feynman
gauge, changing the sign of the gauge-fixing term for the vector multiplet (see
excercise VIB4.1).
a Show that in the massless case the kinetic operator becomes, instead of ,
K ∼ d4 ≡ 1 αβγδ
4!
 dα dβ dγ dδ
where we now use 4-component spinor indices.
b Show that the resulting propagator is of the form, in supercoordinate space,
d4
∆(x, θ; x0, θ0 ) ∼ 2 δ 4(θ − θ0)δ 4 (x − x0) ∼ ln[(x − x0 − i 12 θγθ0 )2 ]
where “x − x0 − i 12 θγθ0” (see subsection IIC2) is the supersymmetry invariant.
(Hint: Consider a “supersymmetry-covariant” Fourier transform in both x
and θ. Warning: If derived by Fourier transformation, the integral is infrared
divergent, and requires dropping an infinite constant.)
 B. GAUGES 313

6. Spacecone
We have just seen that gauge independence allows complex gauge conditions,
which make the action complex. (In subsection IIIC4, we also used complex auxiliary
fields, with a similar effect.) In this subsection we introduce a complex analog of the
lightcone, the “spacecone”, which will greatly simplify Feynman diagram calculations
with massless fields. The spacecone gauge condition is

A2 − iA3 = 0

or more generally
n · A = 0, n2 = 0, n · n* > 0

(but only na , not n*a , appears in the action). While this gauge is spacelike (in the
sense that only spatial components of the gauge field are fixed), it is also null, by
virtue of being complex. Thus, although algebraically like the lightcone, it allows
canonical quantization with the usual time coordinate. In fact, it is just a Wick
rotation of the lightcone. We then eliminate A2 + iA3 as an auxiliary field.
The spacecone is a new gauge to add to our list of axial gauges n · A = 0 from
subsection IIIC2, and the related gauges for scalars from subsections IVA5-6, VIB3-4:

axial gauges non-null null (+ auxiliary field eq.)

(partly) temporal timelike : A0 = 0 lightcone : A+ = 0, δ/δA−
1
spacelike Arnowitt-F ickler : A = 0 spacecone : At = 0, δ/δ Āt
scalar unitary : φ = φ† Gervais-Neveu : φ = hφi, δ/δφ†

In fact, at least for the free theories, the gauges for the scalars can be considered
as dimensional reductions (from 1 or 2 extra dimensions) of those for the vector, as
used for deriving the Stückelberg formalism in subsection IIB4, where the spacelike
components of the vector associated with gauge fixing become scalars: Arnowitt-
Fickler → unitary, spacecone → Gervais-Neveu.
The main advantages of the spacecone over the lightcone are special to D=4, so
we now review the lightcone in a way specialized to physical spacetime. Starting with
the gauge condition (see subsections IA4 and IIA3 for notation)

At = 0

and eliminating Āt by its field equation, the Lagrangian for pure Yang-Mills becomes

L = A+∂ t∂¯tA− − 14 (F +−)2 + 14 (F tt̄)2

314 VI. QUANTUM GAUGE THEORY

F +− = ∂ +A− − ∂ −A+ + i[A+, A−]

1
F tt̄ = ∂ −A+ + ∂ +A− + i t ([A+, ∂ tA−] + [A−, ∂ tA+])
∂
We simplify the Lagrangian by using the self-dual and anti-self-dual combinations:
Dropping also the t superscripts on ∂ for simplicity,
1
F ± = 12 (F tt̄ ± F +−) = ∂ ±A∓ + i [A±, ∂A∓]
∂
+ ¯ − + −
L = A ∂ ∂A + F F   + 
− ∂− + − ∂ −
= A 2 A −i
+1
A [A , ∂A ] − i
+
A [A−, ∂A+]
∂ ∂
1
+ [A+, ∂A−] 2 [A−, ∂A+]
∂

Excercise VIB6.1
Label all the fields and derivatives in the above forms of the Lagrangian, F
in particular, in spinor notation.
Excercise VIB6.2
Show that the field redefinitions A± → (∂)±1 φ±, when applied to just the first
two terms of the above Lagrangian, produce a local action describing the self-
dual field equations of the lightcone formalism of subsection IIIC5 (taking into
account the difference between the lightcone and spacecone). Compare the
results of excercise IIIC5.2. Thus, by treating the latter two terms separately
from the former two, Yang-Mills can be treated as a perturbation about self-
dual Yang-Mills.
Another simplification for massless D=4, and closely related to the use of helicity,
is twistors. For our Feynman diagram calculations for spins ≤ 1, almost all spinor
algebra involves objects carrying at most two spinor indices (spinors, vectors, self-dual
tensors), so we use the twistor matrix notation of subsection IIB6. In particular, in a
general class of gauges the external line factors for Yang-Mills fields in this notation
(see subsection VIB2) read

|i[p|
A = + = ⇒ f* ∼ |p][p|, f =0
hpi
for + helicity or
|pi[|
A = − = ⇒ f ∼ |pihp|, f* = 0
[p]

for −, where (±)a are the polarization 4-vectors for helicity ±1 in terms of a twistor
.
α, ¯α , which can vary from line to line, and whose choice defines the gauge, as a special
 B. GAUGES 315

case of the Landau gauge. (Positive helicity is the same as self-duality, negative is
anti-self-dual. The Landau gauge condition is generally applied in arbitrary Lorentz
gauges to external lines, to eliminate the redundant longitudinal degrees of freedom.)
One special case is the lightcone gauge

n · A = 0, n = |i[|

in terms of an arbitrary constant lightlike vector n. A more convenient gauge is the

spacecone gauge, which can be written in terms of two twistors:

n = |+i[−|

These two twistors are sufficient to define a complete reference frame: We can convert
all spinor indices into this basis, as

ψ α = ψ ± ± α

etc. This corresponds to using two lightlike vectors to define the spacecone gauge,
n± = |±i[±|. For simplicity, we write |±i = |±i; then a vector in this basis can be
written as
p = p+ |+i[+| + p− |−i[−| + pt |−i[+| + p̄t |+i[−|

if we use the normalization

h+−i = [−+] = 1

E.g., for massless momentum p = |pi[p|,

p+ = hp−i[−p], p− = h+pi[p+], pt = h+pi[−p], p̄t = hp−i[p+]

We will also drop the superscript t in contexts where there is no ambiguity. This
basis is related to our previous spinor basis up to phase factors, |±i ∼ |±i, |±] ∼ |±. ],
and we assume them to be commuting (rather than anticommuting); these changes
are more convenient for dealing with twistors (commuting spinors).
The advantage of the spacecone is that we can Lorentz covariantize the Feyn-
man rules by identifying these two lightlike vectors with physical on-shell massless
momenta. We need two such “reference” vectors because we are not allowed to have
n = p on any line. Since only |+i appears in the external line factors for helicity
+1, and only |−] in those for −1, the simplest choice is to pick the momentum of
one external line with helicity +1 to define |−] for all lines with helicity −1, and pick
the momentum of one line with helicity −1 to define |+i for lines with helicity +1.
(In the presence of massless external spinors, we can also choose a helicity +1/2 line
316 VI. QUANTUM GAUGE THEORY

to define |−], etc.) Our above normalization means that we have chosen the phase
h+−i/[−+] = 1 as allowed by the usual ambiguity of twistor phases, while our choice
of the magnitude h+−i[−+] = −h+|[+| · |−i|−] = 1 is a choice of (mass) units. In
explicit calculations, we restore generality (in particular, to allow momentum integra-
tion) by inserting appropriate powers of h+−i and [−+] at the end of the calculations,
as determined by simple dimensional and helicity analysis. (This avoids a clutter of
p
normalization factors h+−i[−+] at intermediate stages.) For example, looking at
the form of the usual spinor helicity external line factors, and counting momenta in
the usual Feynman rules, we see that any tree amplitude (or individual graph) in pure
Yang-Mills must go as
h i2−E+ [ ]2−E−

where E± is the number of external lines with helicity ±.

We now return to external line factors. The naive factors for the above Lagrangian
are 1, since the kinetic term resembles that of a scalar. However, this would lead to
unusual normalization factors in probabilities, which are not obvious in this complex
gauge. Therefore, we determine external line factors from the earlier spinor helicity
expressions for external 4-vectors. In Lorentz gauges (±)a would be the polarization
for helicity ±1 for the complete 4-vector, but in the spacecone formalism only A±
appear. Furthermore, in the spacecone we find
|+i[p|
(+)− = −|+i[+| · =0
h+pi

|pi[−|
(−)+ = −|−i[−| · =0
[−p]
since by antisymmetry h++i = [−−] = 0, so that A+ carries only helicity +1 and
A− only −1. (This statement has literal meaning only on shell, but we can make this
convenient identification more general by using it as a definition of helicity off shell.)
The appropriate external line factors for these fields are thus
|+i[p| [−p]
(+)+ = −|−i[−| · =
h+pi h+pi

|pi[−| h+pi
(−)− = −|+i[+| · =
[−p] [−p]
Note that these factors are inverses of each other, consistent with leaving invariant
(the inner product defined by) the kinetic term.
An exception is the external line factors for the reference momenta themselves,
where |pi = |∓i for helicity ± gives vanishing results. However, examination of the
 B. GAUGES 317

Lagrangian shows this zero can be canceled by a 1/∂ in a vertex, since p = p̄ = 0 for
the reference momenta by definition. (Such cancelations occur automatically from
field redefinitions in the lightcone formulation of the self-dual theory.) The actual
expressions we want to evaluate, before choosing the reference lines, are then
p− h+pi[p+] [−p] [p+]
(+)+ = =
p h+pi[−p] h+pi h+pi

p+ hp−i[−p] h+pi hp−i

(−)− = =
p h+pi[−p] [−p] [−p]
Evaluating the former at |pi = |−i and the latter at |pi = |+i, we get 1 in both cases.
In summary, for reference lines: (1) use only the 3-point vertex of the corresponding
self-duality (± ± ∓ for helicity ±), and use only the term associating the singular
factor with the reference line (the other term and the other vertices give vanishing
contributions); (2) including the momentum factors on that line from the vertex, the
external line factor is 1.

7. Superspacecone
To generalize these results to high-energy (massless) QCD, we consider supersym-
metric QCD, i.e., Yang-Mills coupled to massless fermions in the adjoint representa-
tion. For tree graphs, this is equivalent to ordinary massless QCD except for group
theory, which can be evaluated separately. We first apply the spacecone approach to
the component action for supersymmetric QCD: The modification of this action for
ordinary massless QCD is trivial (replacing the adjoint quark current with defining).
From this we derive the “superspacecone” formalism, rewriting the action more sim-
ply in terms of spacecone superfields. (This form can also be derived from the usual
superspace, but we will not consider that here.)
We now combine the spacecone approach to pure Yang-Mills of subsection VIB6
with the spacecone version of the lightcone treatment of the massless spinor in sub-
section IIIC2. We modify
. the lightcone to the spacecone for the quarks by instead
⊕
eliminating ψ and ψ̄ as auxiliary. For later convenience, we also write the remaining
fermionic fields as .
ψ̄ ⊕ → ψ +, ψ → −ψ −

Then we directly find the terms in the Lagrangian

¯ − + F + F − + iψ +(∂¯ − ∇− 1 ∇+)ψ −
L = A+∂ ∂A ∂

F ± = ∂ ±A∓ − ∂1 ([A±, −i∂A∓] + {ψ +, ψ −})

318 VI. QUANTUM GAUGE THEORY

where the quark term in F ± comes from the quark coupling to At̄ when using its
equation of motion to solve for F tt̄. Collecting terms, we have

L = L2 + L3 + L4
1
−
L2 = A+ 21 A +ψ ψ−
+ 2
−i∂
 ∓   ∓ 
∂ ± ± ∓ − ∂ ±
L3 = A ([A , −i∂A ] + {ψ , ψ }) +
+
ψ [A±, ψ ∓]
∂ ∂
1 1
L4 = −([A+, −i∂A−]+{ψ +, ψ −}) 2 ([A−, −i∂A+]+{ψ +, ψ −})−[A+, ψ −] [A−, ψ +]
∂ −i∂
where terms with ± have only a single sum over it. Although this Lagrangian is
much messier than the original covariant one, one again saves work by expanding
terms once in the action rather than repeatedly for each Feynman diagram.
External-line factors for the spinors follow from the covariant ones of subsection
VIB2 as they did for the spacecone vectors of subsection VIB6. We thus have

ψ + = [−p], ψ − = h+pi

Compared with those for A±, we see ψ +ψ − has an extra factor of p = h+pi[−p] as
compared with A+A−, as expected from the extra factor of 1/(−i∂) in the kinetic
operator. Similarly, if we choose to use external quark lines as reference lines, we use
p− + p+ −
ψ = [p+], ψ = hp−i
p p
which also reduce to 1 for the appropriate reference momenta.
Noting that the bosonic and fermionic terms are the same except for factors of
−i∂, we can combine them into chiral superfields
. that depend on only two anticom-
+ − ⊕
muting coordinates, θ and θ (really θ and θ̄ ):
Z Z
−
S = g2 tr dx dθ dθ L,
1 +
dθ+ dθ− = d+ d− or − d− d+ , {d+ , d− } = −i∂

d± φ∓ = 0; φ±| = A±, d± φ±| = ψ ±

(no sum on ±). These spinor derivatives (and their corresponding supersymmetry
generators) describe only spatial supersymmetry, since they contain no time deriva-
tives. Then, using the identity

d± d∓ [φ±, φ∓]| = [A±, −i∂A∓] + {ψ +, ψ −}

we easily combine the terms in the Lagrangian L into the superspacecone Lagrangian
L: 1  ∓ 
− ∂ ± 1
L=φ + 2
φ + φ [φ±, φ∓] + [φ+, d− φ− ] 2 [φ−, d+ φ+]
−i∂ ∂ ∂
 B. GAUGES 319

The last term can also be written as

d+d− + −
−[φ+, φ−] [φ , φ ]
∂2
Excercise VIB7.1
Introduce another pair of chiral superfields as auxiliary. Show the above
L then can be rewritten in local form, with no spinor derivatives, where the
kinetic term resembles the covariant one for a massless spinor, while the inter-
action term contains no derivatives and is only cubic. (Hint: d± d∓ /(−i∂) are
projection operators for chiral superfields.) Thus this Lagrangian resembles
the Chern-Simons one that appears on the 3D boundary for the topological
term in Yang-Mills (see subsection IIIC6). Expand the action in components,
and separate out the pure Yang-Mills part.
Excercise VIB7.2
Repeat excercise VIB6.2 to obtain the superspacecone action for selfdual super
Yang-Mills, quadratic in φ+ and linear in φ−. Use the field redefinition

φ− → d+ ϕ−, d− ϕ− = 0

and integrate the action over just θ− (by acting with d− ) to obtain a “chiral”
action, with no spinor derivatives, and superfields that are functions of just
θ+ integrated over just θ+ . After further redefinitions as in VIB6.2, obtain an
action identical to the nonsupersymmetric one obtained there, except for the
R +
dθ . Expand in components, and relate to the nonsupersymmetric case.

8. Background-field
A more general type of gauge choice is the background field gauge. As we saw in
subsection VC1, the generating functional can be written in a form where the quan-
tum field is expanded about a background field in the interaction part of the classical
action. The basic steps of the background field gauge method are: (1) choose gauge
fixing that is gauge invariant in the background gauge field, (2) show that the quan-
tum/background splitting of the entire gauge-invariant action is also gauge invariant
in the background gauge field, and (3) show that the effect of splitting the kinetic
term in the gauge-invariant action can be neglected. (Only the interaction terms
should have been split.) The result is then that the effective action Γ, which depends
only on the background fields, is gauge invariant in them. This gauge invariance is
a strong condition which not only simplifies the effective action but allows a “back-
ground gauge” to be chosen for it that is independent of the “quantum gauge” applied
320 VI. QUANTUM GAUGE THEORY

to the path integral: The background fields and quantum fields can be in different
gauges. For example, for a relativistic treatment of the quantum corrections to bound
states whose constitutents are nonrelativistic (such as the hydrogen atom), it is con-
venient to use a Fermi-Feynman gauge (convenient for relativistic matter coupling to
electromagnetism or chromodynamics) for the quantum fields and a Coulomb gauge
(convenient for static or nonrelativistic matter) for the background fields.
A simple way to formulate the background expansion is in terms of the covariant
derivative:

A → Ã = A + A ⇒ ∇ → D + iA, D = ∂ + iA

where A is the quantum field (the variable of path integration) and D is the “back-
ground covariant derivative” in terms of the background field A. We then find for
the field strength

Fab → −i[Da + iAa, Db + iAb] = Fab + D[a Ab] + i[Aa, Ab]

and similarly for the action. Matter fields are split as usual,

φ → φ̃ = ϕ + φ

We now have two gauge invariances, corresponding to the two gauge fields. Both
transformations are defined to have the same, usual form on ∇ = D + iA (and on
φ̃ = ϕ + φ), and thus both leave the action inert, but (1) the “background gauge
invariance” is defined to transform the background fields covariantly

background : D0 = eiλDe−iλ (ϕ0 = eiλϕ), ∇0 = eiλ∇e−iλ (φ̃0 = eiλ φ̃)

⇒ A0 = eiλAe−iλ (φ0 = eiλφ)

and thus the quantum field transforms as a matter (non-gauge) field, while (2) the
“quantum gauge invariance” is defined to leave the background fields inert

quantum : D0 = D (ϕ0 = ϕ), ∇0 = eiλ∇e−iλ (φ̃0 = eiλ φ̃)

⇒ A0 = eiλ[(−iD + A)e−iλ] [φ0 = eiλ(ϕ + φ) − ϕ]

The latter then determines the new BRST transformations

QAa = −[Da + iAa, C], QC = iC 2, QC̃ = −iB, QB = 0 [Qφ = iC(ϕ + φ)]

The key to the background field gauge is to break the quantum invariance, so
a propagator can be defined, but preserve the background invariance, so the path
 B. GAUGES 321

integral is gauge invariant. Since Q, the BRST operator for the quantum gauge
invariance, is now background gauge invariant, we need only choose a gauge-fixing
function Λ that is also background gauge invariant. Many gauges are possible: The
basic rule is to modify any normal gauge condition simply by replacing any partial
derivatives ∂ with background covariant derivatives D. For example, for a Lorentz
covariant gauge
∂·A→D·A
We then gauge fix in the usual way, and now the gauge-fixing terms and the ghost
terms are background gauge invariant, as long as we define the ghosts to transform
covariantly:

background : C 0 = eiλCe−iλ, C̃ 0 = eiλC̃e−iλ , B 0 = eiλBe−iλ

For example, for Lorentz gauges the ghost term is modified, by the modification of
the gauge condition and the quantum BRST transformation, as

C̃∂ · ∇C → C̃D2 C + C̃D · i[A, C]

Furthermore, even axial gauges are modified: For example, even though the gauge
condition A0 = 0 allows elimination of a component of the quantum field, it doesn’t
affect the background field, which now appears in the ghost Lagrangian

C̃∂0C → C̃D0 C

Since the S-matrix is gauge-independent (when BRST is used to perform gauge

fixing, as we have), we can use the background field gauge version of the generating
functional (now using φ to represent all quantum fields and ϕ all background),
Z
Z[ϕ] = Dφ e−iS̃ , S̃ = S0[φ] + SI [φ + ϕ] − iQΛ[φ, ϕ] = Ŝ − (S0 [φ + ϕ] − S0 [φ])

Ŝ = S[φ + ϕ] − iQΛ, S[φ] = S0 [φ] + SI [φ]

where S[φ] is the original gauge-invariant action, QΛ is the gauge-fixing as de-
scribed above, and Ŝ is the sum of this gauge fixing and the background-expanded
gauge-invariant action. We have thus separated the total action S̃ appearing in the
background-gauge-fixed generating functional into the background-gauge-invariant
part Ŝ minus the noninvariant terms S0 [φ + ϕ] − S0 [φ].
As usual, the classical part of the effective action Γ[ϕ] is given by adding the
kinetic term S0 [ϕ] of the (gauge-invariant) classical action to the 1PI tree graphs,
which are just the vertices for the background fields. Thus,

Γclass [ϕ] = S̃|φ=0 + S0 [ϕ] = Ŝ|φ=0 = S[ϕ]

322 VI. QUANTUM GAUGE THEORY

We now note that, as far as calculating just the effective action is concerned,
we can drop all terms in the gauge-fixed action independent of or linear in φ: Any
independent term contributes only classically; any linear term will generate one-
particle reducible graphs (“tadpoles”). This means we can drop the noninvariant
terms S0 [φ + ϕ] − S0 [φ] from S̃. Thus, the Feynman rules for calculating Γ are:
(1) Use the classical gauge-invariant action S[ϕ] for the classical contribution to Γ;
and (2) for the quantum contribution, use all the 1PI loop graphs coming from Ŝ.
The result is background gauge invariant, since Ŝ is manifestly so.
Another important feature of the quantum-gauge-fixed background field action is
that it is background-gauge-invariant order-by-order in the quantum fields. In fact,
every term in the corresponding ordinary gauge action has been replaced by one (or
more, if there are F terms) background-gauge-invariant term.
Excercise VIB8.1
Consider the Fermi-Feynman background-field gauge for the quantum field
of pure Yang-Mills theory. Write all terms (both gauge-invariant and gauge-
fixed) quadratic in the quantum field. Show that these combine as

− 14 A · A − i 12 Aa [Fab, Ab]

where = (D)2 (and “DA” means “[D, A]”, etc.).

Since all external lines are associated with background fields, if we draw graphs
in such a way as to exhibit only the quantum fields, they will all look like vacuum
graphs: graphs with no external lines. However, any particular such vacuum graph
will represent many of the original graphs, since the background lines can be attached
in many ways. Furthermore, in background field gauges any such vacuum graph, con-
sidered as a contribution to the effective action, will be gauge invariant with respect
to the background gauge transformations, since it results from the non-background
gauge true vacuum graph by the replacement of the ordinary derivative with the
background covariant derivative ∂ → D (plus perhaps some noniminimal F terms),
including in the propagator. In particular, the complete one-loop contribution to Γ
is given by the vacuum graph with no quantum interactions: It can be obtained from
just the part of the Ŝ that is quadratic in the quantum fields.
 B. GAUGES 323

Excercise VIB8.2
Consider an arbitrary gauge-invariant Yang-Mills action S[Ã] with à = A+A
in terms of the background field A and quantum field A. Taylor expand the
action in A as
δS[A]
S[Ã] = S[A] + A + ...
δA
The infinitesimal quantum gauge transformation mixes different-order terms
in the expansion. Show that the term quadratic in A is invariant under an
Abelian quantum gauge transformation only if the background satisfies the
field equations, δS[A]/δA = 0. Similar remarks apply to BRST transforma-
tions and the gauge-fixed action. (Since quadratic actions, even in background
fields, yield only a propagator, they can be described by first-quantization:
Thus gauge invariance implying background field equations occurs whenever
a gauge field appears as both a quantum mechanical state and a background
field, for example in string theory. See subsection XIIB7 for a simpler exam-
ple.)
The S-matrix is then given in the usual way from Γ[ϕ], after adding another
gauge-fixing term for the background gauge invariance. Since the total S-matrix is
given by just the trees following from treating Γ as a classical action, we need only a
gauge-fixing term for the physical fields, and we can ignore background ghosts. (Of
course, quantum ghosts were already used to calculate Γ.) This background gauge
fixing is independent of the quantum gauge fixing. In particular, we can choose
different quantum and background gauges: For example, when treating spontaneously
broken gauge theories, it’s often more convenient to choose a Fermi-Feynman quantum
gauge and a unitary background gauge; i.e., we expand the background fields about
the physical vacuum to make the physical states obvious, but leave the quantum
fields unexpanded to avoid complicating the Feynman rules. This also avoids the
complication of having to expand about the vacuum twice, since vacuum values get
quantum corrections to those appearing in the classical action.
The gauge invariance of the effective action in the background-field formalism is
a big advantage over other quantum gauges, where the effective action is only BRST
invariant, since gauge invariance is a much stronger constraint than BRST invariance:
Gauge symmetry is local, while BRST is only global. Thus, the background-field
gauge produces a much simpler effective action. In other words, the background-
field gauge produces an effective action without ghosts: Although we can drop ghost
terms from the effective action in general, because there are no physical external
ghost states (since we calculate only the “tree” graphs of the effective action), the
324 VI. QUANTUM GAUGE THEORY

result is not normally BRST invariant; but in the background-field gauge it is still
BRST invariant, since it is gauge invariant. This means that the background-field
gauge yields not only simpler results, but fewer calculations: Many terms can be
determined by “gauge covariantization”.
Excercise VIB8.3
Consider an effective action for Yang-Mills plus matter in a background-field
gauge. Its gauge invariance can be used to derive “Ward-Takahashi identi-
ties”. (These were originally expressed as properties of the S-matrix, but are
much simpler to understand in terms of the effective action.)
a Show that the part of the effective action quadratic in the Yang-Mills fields,
and independent of the matter fields, is invariant under the Abelian gauge
transformations. (Hint: Taylor expand.) Thus, in such gauges the quantum
correction to the gluon propagator is transverse.
b By the same method, find a relation between any quantum 3-point vertex
coupling matter to Yang-Mills and the corresponding matter propagator cor-
rection. (Note a simpler case: Since the renormalization counterterms are
local, gauge invariance just says that the coefficients of the two corresponding
counterterms are the same, i.e., they occur in the combination ∼ Ψ̄∇ / Ψ.)
However, this does not mean we can completely ignore BRST and ghosts by using
background-field gauges: Although the effective action is gauge invariant and ghost
free, ghosts and BRST still appear in the (quantum-gauge-fixed) classical action. In
practice this means, as far as calculating the Feynman graphs that contribute to the
effective action, that in the background-field gauge calculations are about one loop
simpler than in other gauges. For example, for one-loop graphs we effectively cal-
culate free one-loop vacuum bubbles (including ghosts) covariantly coupled to back-
ground fields: There are fewer of the complications of nonabelian theories, since the
quantum fields appear only as non-gauge fields with covariant couplings and no self-
interactions. However, already at two loops we have self-interactions of the quantum
fields, which include the same kinds of terms that would have appeared had we not
used a background-field formalism.
Another complication is that BRST invariance is not as restrictive as gauge invari-
ance: It can be shown that in general gauges at the quantum level BRST invariance
is preserved only up to “wave-function renormalizations” (rescalings) of the quan-
tum fields. However, in the background-field gauge wave-function renormalizations
of the quantum fields can be ignored, since the quantum field is a dummy vari-
able: There are no external quantum fields, so all such factors cancel. (Actually, we
 B. GAUGES 325

can also ignore wave function renormalization counterterms in non-background-field

gauges, since when calculating S-matrix elements such divergences will be canceled
by corresponding divergences in the external-line factors. In general, external-line
normalization factors may be nontrivial even when wave-function renormalization is
performed, depending on the renormalization scheme.)
An exception is Abelian gauge theories, such as QED: Because the gauge-invariant
action for just the gauge fields is free, background field gauges are identical to ordinary
gauges. Also, the ghosts decouple (for linear gauge conditions).

9. Nielsen-Kallosh
So far we have considered only gauges where the gauge-fixing term is the square
of the gauge-fixing function. More generally, we’ll need gauge-fixing terms of the form
fOf for some operator O. Straightforwardly, we can write
∂f
iQ 12 C̃[f(A) − 12 αO−1 B] = 12 B(f − 12 αO−1 B) + 12 iC̃ · [∇, C]
∂A
However, B is no longer auxiliary, so we can’t eliminate it by its field equation. But
we can diagonalize the Lagrangian by the corresponding redefinition,

B → B + α1 Of

(The Jacobian of such redefinitions is unity, the determinant of a triangular matrix

of the form 10 x1 .) The gauge-fixing terms are then

− 14 αBO−1 B + 1
4α
fOf

The inverse operator is inconvenient for Feynman rules. We know that integrating
out B gives a functional determinant, so O−1 can be replaced by an O if we change
the statistics of the Nakanishi-Lautrup field. However, this is a bit formal, since tech-
nically O must be symmetric between the two B’s, while it should be antisymmetric
between two fermions.
A useful example is gauge fixing for super Yang-Mills in superspace. Gauge fixing
for massless Yang-Mills is actually more difficult than for the massive (Higgs) case,
considered in subsection VIB5. We’ll look at the Abelian theory, to determine what
kind of gauge fixing we need to define the propagator. (With slight generalization,
this is also sufficient for the background-field gauge: See the following subsection.)
In that case the BRST transformations are

QV = i(C̄ − C), QC̃ = −iB, ˜ = −iB̄,

QC̄ QC = QB = QC̄ = QB̄ = 0
326 VI. QUANTUM GAUGE THEORY

where C, C̃, and B are chiral. Using the result (the Abelian case of excercise IVC4.1)

Wα = −id̄2 dα V

the gauge-invariant kinetic term is (rearranging derivatives and using integration by

parts; see subsection VIB5)
Z Z Z
α ¯2
L0 = − d θ 2 W Wα = − d θ 2 V d d dα V = d4 θ V ( 14 − d̄2 d2 )V
2 1 α 4 1

To gauge-fix to the Fermi-Feynman gauge we choose

Z
L1 = −iQ d4 θ [(C̃ + C̄ ˜ )V + C̃( 1 )−1 B̄]
2
Z
˜ C − C̃ C̄) − (B + B̄)V − B( 1 )−1 B̄]
= d4 θ [(C̄ 2

(dropping d4 θ integrals of totally chiral or totally antichiral terms, which vanish). If

we were to simply redefine B by

B → B − d¯2 d2 V, B̄ → B̄ − d2 d¯2 V

the gauge-fixing terms would diagonalize as (using d̄2 d2 d¯2 = 1

2 d̄2 )

−(B + B̄)V − B( 12 )−1 B̄ → V d¯2 d2 V − B( 12 )−1 B̄

giving the desired result for V : At this stage the total result is
Z
L = L0 + L1 → d4 θ [ 14 V V + C̄ ˜ C − C̃ C̄ − B( 1 )−1 B̄]
2

Because B is complex, the replacement of B with a fermionic superfield can

be performed classically, just like the rest of the gauge-fixing procedure. We thus
introduce ghosts for a trivial gauge invariance as described in subsection VIA4:

QD = E, QĒ = −iD̄, QE = QD̄ = 0

We have treated the ghosts and their hermitian conjugates independently; alterna-
tively, we can consider D̄ and Ē as not being the conjugates of D and E. The gauge
fixing is simply
L2 = −iQ(−ĒD) = D̄D − iĒE

We next make the redefinition

D → D + ( 12 )−1 d̄2 B̄, D̄ → D̄ + ( 12 )−1 d2 B

 B. GAUGES 327

which has the effect

Z Z Z Z
−1
d θ [D̄D − B( 2 ) B̄] → d θ D̄D + d θ DB + d2 θ̄ D̄B̄
4 1 4 2

which vanishes, after using the now-algebraic field equations from varying B and
B̄. Alternatively, we can make this field redefinition instead of the previous field
redefinition: We then have the terms
Z Z Z Z
d θ [D̄D − (B + B̄)V ] + d θ DB + d θ̄ D̄B̄ → d4 θ V d¯2 d2 V
4 2 2

after using the still-algebraic B equations.

The net result
Z
L = L0 + L1 + L2 → d4 θ ( 14 V ˜ C − C̃ C̄ − iĒE)
V + C̄

is that the original nonlocal B term has been replaced classically with the local
ĒE term, which yields the same determinant upon quantization, but gives simple
Feynman rules more directly. (The determinant is nontrivial in the background-field
gauge. A similar procedure can be applied to gauge fixing for spin 3/2.)
Excercise VIB9.1
Perform the analogous quantization for the nonabelian case of pure super
Yang-Mills (no matter), using the super Gervais-Neveu gauge. Compare with
the limit m → 0 of the model considered in subsection VIB5, and show the
V part of the action agrees.
Excercise VIB9.2
Use this method to produce a gauge-fixing term α(n · A) (n · A) for a gauge
vector A in terms of a parameter α and constant vector n. Find all propaga-
tors. Look for simplifying special cases of α and n.

10. Super background-field

Although in principle the background-field formalism is the same for supersym-
metric theories as nonsupersymmetric, there are some technical differences because of
the nonlinearity in the prepotentials. (Similar remarks apply to nonlinear σ models.)
The basic idea is that we want to expand the full covariant derivative in quantum
fields about background-covariant derivatives: As for the nonsupersymmetric case,
∇ → D + iA, but now A is further expressed in terms of D and the prepotential
because of the constraints. The generalization in this case (and for nonlinear σ mod-
els) is easy because the solution to the constraints makes the prepotentials appear
328 VI. QUANTUM GAUGE THEORY

as (complex) group elements: Because of the closure of group multiplication, we can

write
g → gB gQ

in terms of quantum (gQ) and background (gB ) group elements (fields). More explic-
itly, for our case we write (see subsection IVC4)

eΩ → eΩB eΩQ

and thus for the covariant derivatives

∇α → e−ΩQ Dα eΩQ

absorbing the background prepotential completely into the background covariant

derivative
Dα = e−ΩB dα eΩB

In other words, as the name suggests, the full covariant derivative ∇ has been ex-
panded about an arbitrary background, described by ΩB . (This is even clearer in the
supergravity case, where we simply replace the flat-space dα with the curved-space
Dα , since dα is more than a partial derivative, and already contains the flat-space
part of the metric tensor.) For purposes of quantization, it is most convenient to go
to a chiral representation for the quantum field. For the background field we need not
be so specific, since it is hidden in the background covariant derivatives. The result
is then
∇α → e−V Dα eV , ∇α. → D̄α., ∇αα. → i{D̄α., e−V Dα eV }

where V is the quantum field.

Excercise VIB10.1
Solve the rest of the commutator algebra to find expressions for all the field
strengths in terms of V and DA.
The rest of the quantization procedure then follows as for the nonsupersymmetric
case, except for the Nielsen-Kallosh ghost described in the previous subsection. In
particular, for the terms in the gauge-fixed classical action quadratic in the quantum
field V ,
.
Wα → −i 12 [Dα , {Dα., e−V Dα eV }] = Wα − iD̄2 Dα V + i 12 D̄2 [V, Dα V ] + O(V 3 )
Z
⇒ S2V = dx d4 θ V (− 12 Dα D̄2 Dα + i 12 W α Dα + D̄2 D2 )V
 B. GAUGES 329

Pushing the D in the first term to the right, we find

. .
− 12 Dα D̄2 Dα = i 14 (Dαα D̄α. Dα + D̄α.Dαα Dα ) − D̄2 D2

Using integration by parts on all the derivatives in the second term so they act to the
left, then switching the V ’s so they again act to the right,
. . . .
Dαα D̄α.Dα + D̄α. Dαα Dα → Dαα D̄α.Dα + Dα Dαα D̄α.
. . .
= Dαα D̄α.Dα + Dαα Dα D̄α. + [Dα, Dαα ]D̄α.
α
.
= −i + 2W D̄α.

where = Da Da . The final result is similar to the bosonic case (excercise VIB8.1):
Z .
α
S2V = dx d4 θ 14 V ( + 2iW α Dα + 2iW D̄α.)V

(This result is invariant under integration by parts because of the Bianchi identity
.
α
Dα W α + D̄α.W = 0.)
Ghosts and matter are quantized straightforwardly: For matter we have

∇α.φ = ∇αφ̄ = 0 ⇒

φ → ϕ + φ, φ̄ → eV (ϕ̄ + φ̄); D̄α.ϕ = Dα ϕ̄ = 0

The action thus looks the same as usual ((ϕ̄+ φ̄)eV (ϕ+φ), etc.), except that all chiral
superfields are now background-chiral. For the standard ghosts we have for the ghost
R
action SC = dx d4 θ LC (remembering there are no background ghosts, and using
the full nonlinear transformation law from excercise IVC4.3)
˜ )L [coth(L )(C − C̄) + (C + C̄)] = (C̃ + C̄
LC = (C̃ + C̄ ˜ )(C − C̄) + O(V )
V /2 V /2

˜ C − C̃ C̄) + O(V )
→ (C̄
the same as in non-background gauges, except again the ghosts are background-chiral.
Now the Nielsen-Kallosh ghost of the previous subsection is nontrivial: We again have

LN K = −iĒE

but these ghosts also are background-chiral. This means they contribute to the effec-
tive action only at one loop, through “vacuum bubbles”.

REFERENCES
1 Fock, loc. cit. (VB);
Schwinger, loc. cit. (VB, ref. 2, second ref.):
radial gauge.
330 VI. QUANTUM GAUGE THEORY

2 Fermi, loc. cit. (VC);

Feynman, loc. cit. (VB, ref. 4).
3 A.A. Abrikosov, I.M. Khalatnikov, and L.D. Landau, Doklady Akad. Nauk USSR 95
(1954) 773:
Landau gauge.
4 P. De Causmaecker, R. Gastmans, W. Troost, and T.T. Wu, Nucl. Phys. B206 (1982)
53;
F.A. Berends, R. Kleiss, P. De Causmaecker, R. Gastmans, W. Troost, and T.T. Wu,
Nucl. Phys. B206 (1982) 61;
Z. Xu, D.-H. Zhang, and L. Chang, Nucl. Phys. B291 (1987) 392;
J.F. Gunion and Z. Kunszt, Phys. Lett. 161 (1985) 333;
R. Kleiss and W.J. Sterling, Nucl. Phys. B262 (1985) 235:
spinor helicity.
5 V.N. Gribov, Nucl. Phys. B139 (1978) 1.
6 G. ’t Hooft, Nucl. Phys. B35 (1971) 167;
K. Fujikawa, B.W. Lee, and A.I. Sanda, Phys. Rev. D6 (1972) 2923;
Y.-P. Yao, Phys. Rev. D7 (1973) 1647:
renormalizable gauges.
7 Gervais and Neveu, loc. cit. (IVA).
8 G.F. Chew and M. Levinson, Z. Phys. C 20 (1983) 19:
proposed Feynman rules similar to anti-Gervais-Neveu gauge in a context related to
string theory.
9 W. Siegel, hep-th/9502163, Phys. Rev. D52 (1995) 1035:
super Gervais-Neveu.
10 M. Mangano and S.J. Parke, Phys. Rep. 200 (1991) 301:
review of modern methods for tree graphs.
11 E.T. Newman and R. Penrose, J. Math. Phys. 3 (1962) 566:
use of “null tetrad” in gravity, complex basis vectors in terms of constant spinors.
12 G. Chalmers and W. Siegel, hep-ph/9801220, Phys. Rev. D59 (1999) 045013:
spacecone.
13 W. Siegel and S.J. Gates, Jr., Nucl. Phys. B189 (1981) 295;
S. Mandelstam, Nucl. Phys. B213 (1983) 149;
L. Brink, O. Lindgren, and B.E.W. Nilsson, Nucl. Phys. B212 (1983) 401:
lightcone superfields.
14 Feynman, loc. cit. (VC, ref. 17);
DeWitt, loc. cit. (VIA);
J. Honerkamp, Nucl. Phys. B48 (1972) 269;
G. ’t Hooft, The background field method in gauge field theories, in Functional and
probabilistic methods in quantum field theory, proc. 12th Winter School of Theoretical
Physics, Karpacz, Feb. 17-Mar. 2, 1975, v. 1, Acta Univ. Wratislav. 368 (1976) 345;
L.F. Abbott, Nucl. Phys. B185 (1981) 189:
background-field gauge.
15 J.C. Ward, Phys. Rev. Lett. 78 (1950) 182;
Y. Takahashi, Nuo. Cim. 6 (1957) 370.
16 N.K. Nielsen, Nucl. Phys. B140 (1978) 499;
R.E. Kallosh, Nucl. Phys. B141 (1978) 141.
17 M.T. Grisaru, W. Siegel, and M. Roček, Nucl. Phys. B159 (1979) 429:
super background-field gauge.
 C. SCATTERING 331

......................... .........................
......................... C. SCATTERING .........................

We have seen how covariant expansions of the S-matrix can be based on various
definitions of h̄. Covariant expansions can also be based on spacetime quantum
numbers: For example, we can perturb in mass; this is equivalent to adding low-
energy corrections to the high-energy approximation. Also, the first-quantized version
of the h̄ expansion, which expands in powers of momenta, is effectively an expansion
in inverse powers of mass (low-energy approximation).
The only other spacetime property of a particle is spin, or helicity for massless
particles in D=4. It is possible to define expansions in terms of it by describing the
leading order by a complex action. This violates semiclassical unitarity at that order;
however, the loop expansion violates unitarity at tree order also, so the expansion
is still useful as long as unitarity returns once the expansion has been summed.
Furthermore, we have already seen that gauges where unitarity is not manifest have
some advantages over unitary gauges. In particular, the Gervais-Neveu gauge uses a
complex gauge condition.

1. Yang-Mills

We first consider calculations for massless theories; these are simpler than massive
ones in D=4 because the little group of the Lorentz group is SO(D−2) instead of
SO(D−1), and is thus Abelian: We can label the spin of a state by an integer or half-
integer, the helicity, by use of the spacecone formalism. To simplify notation, we drop
the transverse index (pt → p), and distinguish 4-momentum P from its transverse
component p by using upper- and lower-case. We also use color ordering; i.e., we
examine only planar diagrams for each permutation of external lines.
We begin by summarizing the spacecone rules for pure Yang-Mills found in sub-
R
section VIB6: The Lagrangian appearing in the action S = g−2 tr dx L, writing
derivatives as momentum operators for later convenience, is
−
L = A+(− 12 P 2 )A− + ( pp A+)[A+, pA−] + ( pp A−)[A−, pA+] + [A+, pA− ] p12 [A−, pA+]
+

Twistor notation (see subsection IIB6) is used:

p+ = hp−i[−p], p− = h+pi[p+], p = h+pi[−p], p̄ = hp−i[p+]

h+−i = [−+] = 1
332 VI. QUANTUM GAUGE THEORY

The propagator and vertices are read from L in the usual way, but in addition we
have further simplification from the choice of external line factors
[−p] h+pi p− p+
+ = , − = ; ⊕ =  = 1
h+pi [−p] p p
where ⊕ and are the reference lines, with + and − helicity, respectively (not to
be confused with the earlier notation for spinor indices α = (⊕, )). However, the
reference momenta for helicities ± are taken from lines with helicities ∓:
P⊕ = |−i[−|, P = |+i[+|
The reference external line factors occur only in the above combinations, because only
1 term of 1 of the 3-point vertices contributes to each.
The simplest examples are classes of diagrams that vanish by virtue of their
“maximal helicity violation”: By simple counting of +’s and −’s, we see that the tree
graphs with the fewest external −’s, those with only self-dual vertices (++−), have a
single external −. Thus the all + amplitude vanishes automatically. Furthermore, the
diagrams with a single external − must have that line chosen as one of the reference
lines. However, by the above rules that line can carry only the anti -self-dual vertex
(−−+), so those amplitudes also vanish.
The simplest nonvanishing amplitude is ++−−. We consider the case where the
helicities are cyclically ordered as ++−−; we label them 1234, and choose 1 and
4 as the reference lines; this amplitude can be denoted as ⊕+− . (P4 = |+i[+],
P1 = |−i[−|: The positive-helicity reference line gives the reference momentum for
negative helicity, and vice versa.) We label all external momenta as flowing outward.
There are only three diagrams; however, the + reference line uses only the ++−
vertex, while the − reference line uses only the −−+ vertex, so the 4-point-vertex
diagram vanishes, as does the diagram with both reference lines at the same vertex.
Thus, we are left with only 1 graph.

2+ 3
+

1 4

Furthermore, we know that the 3-point vertices contribute only 1 term to the
reference line, so this graph has only 1 term. This means we can immediately write
down the answer (dropping the factors of −g at each vertex):
1 [−2] h+3i 1 1
2+3−p2 p3 1 = h+2i[−2]h+3i[−3]
2 (P3 + P4 )
2 h+2i [−3] h34i[34] h+−i[−+]
 C. SCATTERING 333

[12]2 h34i
=
[34][41]h14i
where we have restored helicity and dimensions, and used p1 = p4 = 0. We have
omitted the usual group theory factor (see subsection VC9). (Note that the propaga-
tor is −1/ 12 P 2 , because of the signature for the ± spacecone components. This extra
sign cancels that coming from the fact that one vertex has cyclic ordering and one
anticyclic with respect to group theory, i.e., the commutators in the action.) Using
the identities, following from overall momentum conservation,

(P1 + P4 )2 = (P2 + P3 )2 ⇒ [41]h14i = [23]h32i

X
|pi[p| = 0 ⇒ h34i[14] = −h32i[12]
this can be put in the standard form
[12]4
[12][23][34][41]

Excercise VIC1.1
Using similar manipulations, cast it into the form
h34i4
h12ih23ih34ih41i

Excercise VIC1.2
Repeat the calculation for the + − +− (color-ordered) amplitude.
The corresponding differential cross section is very simple: Using

hpqi* = −[pq] ⇒ |hpqi|2 = |[pq]|2 = −P · Q

and momentum conservation, we find (after including the coupling g)

s2 4t
2
|T |2 = g 4 or g
t2 s2
depending on the orientation of the diagram with respect to time, for this color-
ordered contribution. (Depending on the color quantum numbers of the external
334 VI. QUANTUM GAUGE THEORY

⊕++− . Again dropping all graphs with a reference line at a 4-point vertex or 2
references lines at a 3-point, all 5 graphs with a 4-point vertex are killed, and only
3 of the remaining 5 survive. (We also need to consider various combinations of +
and − indices, but only 1 survives for each graph because of the chirality of 3-vertices
with reference lines.)

2+ 3+ 3+ 3+ 4
2+ 4 +
4 + 2+
+ + + + +
1 + 5
5 1 5 1

Since 3-point vertices with (without) a reference line have 1 (2) terms, we are left
with only 6 terms. The initial result for the amplitude is then
  −   −   − 
3 p2 p− 2 p2 +1 p− p2 +1 p−
p4 p2 − p3 3
p2 p4 p2 − p3 3
p2 p4 p2 − p3
2 3

2+3+4−  − + 
(P2 · P3 )(P4 · P5 ) (P1 · P2 )(P4 · P5 ) (P1 · P2 )(P3 · P4 )

where we have used the fact that the reference lines have trivial momenta: 1 for the
component with ± index opposite to its helicity, 0 for the remaining components.
The two terms for each diagram simplify to one, using
p− [p+] p− p− [2+][−3] − [3+][−2] [23]
= ⇒ 2
− 3 = =
p [−p] p2 p3 [−2][−3] [−2][−3]
(applying the cyclic identity) with our normalization. Using this result, we find the
similar result
p−
2 +1 p− h+−i[−3] + h+2i[23] h+4i[34]
− 3 = =
p2 p3 h+2i[−2][−3] h+2i[−2][−3]
applying momentum conservation. We next translate the momentum denominators
into twistor notation, and also substitute the spacecone expressions for the polariza-
tions and numerators. canceling identical factors in numerator and denominator (but
no further use of identities), the amplitude becomes (+ = 5, − = 1)
 
h+4i3 [−4]2 [−4][34] h+2i[−2]
− + +
h+2ih+3i h23i[4+] h2−i[4+] h2−ih34i
 
h+4i3 h+2i[−4] h+2i[−2] h+4i3 h45i4
=− − + = =
h+2ih+3i h2−ih23i h2−ih34i h2−ih23ih34i h12ih23ih34ih45ih51i
 C. SCATTERING 335

applying momentum conservation twice, restoring normalization, and replacing the

numerals for ±.
Excercise VIC1.3
Using the spacecone gauge, evaluate all diagrams contributing to the six-point
gluon (Yang-Mills) scattering tree amplitude (T-matrix) with color-ordered
helicities ++++−−, that correspond to the symmetric diagram with a central
3-point vertex each of whose legs is connected to another 3-point vertex, each
of which carries 2 of the external lines.
These results can be generalized to arbitrary (color-ordered) n-point tree ampli-
tudes with two − helicities, labeled i and j, and the rest + (“Parke-Taylor ampli-
tudes”): The result is (in an obvious notation), including now the coupling (−g)n−2 ,

hiji4
h+1 · · · +i−1 −i +i+1 · · · +j−1 −j +j+1 · · · +n i = (−g)n−2
h12ih23i · · · hn − 1, nihn1i

2. Recursion
A simple way to derive higher-point amplitudes is using the classical field equa-
tions. (See subsection VC3. In the literature, the field has often been mistaken for
the current, since φ ∼ δ/δJ, J ∼ δ/δφ. As usual, these are distinguished by the
fact the field always has an external propagator, while the current has it amputated,
since Kφ + ... = −J.) The steps are: (1) Calculate the first few terms in the series
(enumerated by the number of external lines). (2) Guess the general result. (3) Prove
that it is correct by induction, using the classical field equations. Of course, the sec-
ond part is the hardest in general (at least when one simplifies the third step by using
spacecone methods), and has been possible for just a couple of cases, only because
the results for those cases are so simple. Since these results are for off-shell fields, and
not S-matrix elements, they are gauge dependent: For example, if they are inserted
into larger diagrams, the same choice of reference lines must be used.
The solution to the classical field equations is given by tree graphs with all external
lines but one (the field itself) amputated and put on shell. (The usual external-line
wave functions describe the asymptotic field, which is free.) The two cases with
known solutions are those where all the on-shell lines have the same helicity, or one
different. Note that the field A± has a ∓ associated with the opposite end of its
external propagator. We then see in the former case, with all +’s on on-shell lines,
that A− vanishes because there are no fully-amputated diagrams, even off-shell, with
only +’s externally (again counting +’s and −’s on vertices). Similarly, for the latter
case, with only one − on an on-shell line, we see that A− has only ++− vertices;
336 VI. QUANTUM GAUGE THEORY

but setting that one on-shell − to be a reference line (which by definition must be
on-shell), it is not allowed such a vertex, so A− vanishes also in this case. By similar
reasoning, we see that A+ in the former case consists entirely of ++− vertices; and
in the latter case consists of all ++− except for one −−+ (no ++−−), which must
have the − reference line directly attached.
The appearance of only the self-dual field (A+) and almost only the self-dual
vertex (++−) means that in both cases one is essentially solving equations in the
self-dual theory: If we take just the kinetic term and ++− vertex from the action,
and make the field redefinitions (see excercise VIB6.2)

A+ = pφ, A− = p−1 φ̂

we obtain (after integration by parts and rearrangement inside the trace)

L+−,++− = φ̂( 12 P 2 φ + [p− φ, pφ])

These redefinitions make the ++− vertex local. φ̂ appears only as a Lagrange mul-
tiplier, and its variation gives the self-dual field equation
.
1
2 φ + i(∂ α φ)(∂ α.φ) = 0

(which differs from the result of subsection IIIC5 by an i from the use of p instead of
∂ in the field redefinition, and ⊕ → from the use of the spacecone instead of the
lightcone).
We now consider in more detail the simpler (former) example (the one which
does not directly give a nontrivial scattering amplitude). As a slight simplification,
we look at the recursion relation for the field φ as defined in the self-dual theory. The
recursion relation is now (see subsection VC3), scaling the coupling out of the kinetic
term,

g Xn−1
φ(1, n) = − 1 2 φ(1, i)φ(i + 1, n)[p−(1, i)p(i + 1, n) − p(1, i)p− (i + 1, n)]
2 P (1, n) i=1

X
k
P (j, k) ≡ Pm
m=j

where we again use color ordering, number the external lines cyclically, and φ(j, k)
denotes the field with on-shell lines with momenta Pj through Pk . (Thus, on the
left-hand side of the equation the field has n on-shell lines, while on the right-hand
 C. SCATTERING 337

side the two fields have i and n− i.) Plugging in the twistor expressions for the vertex
momenta, we find

X
i X
n
− −
p (1, i)p(i + 1, n) − p(1, i)p (i + 1, n) = h+ji[jk]h+ki
j=1 k=i+1

If we are clever we can guess the general result from explicit evaluation of the
lower-order graphs; instead we find in the literature, after the above redefinition,
1
φ(i, j) = (−g)N −1
h+iihi, i + 1i · · · hj − 1, jih+ji

where N is the number of background momenta (Pi , ..., Pj ) for φ(i, j). For the initial-
condition case N = 1 this is simply the statement that the external line factor for φ
is now
+ 1
φ = =
p h+pi2
The induction hypothesis is also easy to check: The product of the two φ’s from the
induction hypothesis gives the desired result by itself up to a simple factor:
hi, i + 1i
φ(1, i)φ(i + 1, n) = − 1g φ(1, n)
h+iih+, i + 1i

(The algebra of the color indices works as usual.) We then perform the sum over i
before that over j and k (the complete sum is over all i, j, k with 1 ≤ j ≤ i < k ≤ n),
making use of the identity

habi hbci haci X

k−1
hi, i + 1i hjki
+ = ⇒ =
h+aih+bi h+bih+ci h+aih+ci i=j
h+iih+, i + 1i h+jih+ki

Multiplying this by the vertex momentum factor gives a sum over j < k of hjki[jk] =
Pj · Pk , canceling the external propagator, yielding the desired result.

3. Fermions
We have seen in subsection VIB7 how these methods can be applied to massless
spinors. Rather than applying the rules directly, in this subsection we examine the
relation of the results in QCD to those in pure Yang-Mills theory. We also saw in
subsection VIB7 how supersymmetry could be used to relate different QCD ampli-
tudes. However, in practice supersymmetry relations give only a few useful relations,
and only ones that can already be seen directly from the spacecone rules, which give
more results than can be seen by supersymmetry alone.
338 VI. QUANTUM GAUGE THEORY

The simplest relations that follow from supersymmetry are the vanishing of tree
graphs with fewer than two negative helicities, which we saw in subsection VIC1
follows automatically from the spacecone rules. The remaining useful supersymme-
try relation for tree graphs is the relation between Parke-Taylor amplitudes for pure
Yang-Mills and those with one external line each of positive and negative helicity
replaced with spinors or scalars. The easiest way to see this result is to make use
of the conventions of the selfdual theory, as in the preceeding subsection. In Parke-
Taylor amplitudes only one vertex is a non-selfdual vertex, which accounts for the
simplicity of these amplitudes. (Tree amplitudes with only selfdual vertices van-
ish.) Furthermore, after transforming to the selfdual conventions, all (nonvanishing)
selfdual vertices are identical — independent of spin. Furthermore, the nonselfdual
3-point vertex with one negative-helicity gluon chosen as a reference line (the only
non-selfdual vertex we’ll need for this relation) is independent of the spins of the
remaining two lines. Consequently, the only difference between the two amplitudes
we are relating comes from the difference in normalization of external line factors for
gluons and quarks (and scalars).
We will not review the superspace formulation of selfdual supersymmetric theories
here. The main features will be evident from the example of supersymmetric QCD
that we now examine in more detail. The main result follows from treating the
selfdual field of the nonsupersymmetric theory as a spacecone (or lightcone) superfield.
Dimensional analysis then tells us that the field of helicity h has dimension 1 − h.
The appropriate redefinitions of the spacecone fields are then
1
A+ → pA+, ψ + → pψ +, φ → φ, ψ − → ψ −, A− → A−
p
for the Yang-Mills fields A±, spinors ψ ±, and scalars φ. The resulting external line
factors are then simply
h+pi−2h
After these redefinitions, the kinetic terms, selfdual (++−) vertices, and antiselfdual
vertices for − gluon reference line (referencing positive helicity) are

L2 = A+ 12 P 2 A− + ψ + 21 P 2 ψ −

L3,sd = (p− A+)([pA+, A−] + {pψ + , ψ −}) + (p− ψ +)[pA+, ψ −]

 + 
p −
L3,sd,⊕ = A ([pA+, A−] + {pψ +, ψ −})
p2
for supersymmetric QCD. (In the A3 term in the last line we have used integration
by parts, and dropped a (p+ /p)A− term that vanishes for the reference line: There
(p+ /p2 )− = 1 now, so (p+ /p)− = 0 vanishes for that line since p → 0.)
 C. SCATTERING 339

We now see easily that the terms L2 and L3,sd that define the selfdual theory are
independent of whether boson or fermion is chosen for the positive helicity fields and
the negative helicity one (only the helicities of the fields must add up to 0 for L2 and 1
for L3,sd for Lorentz invariance). Thus, supersymmetry is a much stronger restriction
in a selfdual theory than a nonselfdual one. Finally, the current that couples to the
reference line (p+ /p2 )A− is also the same for bosons and fermions. We therefore have,
for example, the relation
h13i
(−, − 12 , + 12 , + · · · +) = (− − + · · · +)
h12i

for the color-ordered tree amplitudes (where we have labeled helicities ±1 by ±).
This follows from choosing line 1 as reference line (for positive helicity, from a line
with negative helicity). For example, from our result for the 4-gluon tree, we have
the 2-quark, 2-gluon tree
h12i2 h13i
(−, − 12 , + 12 , +) =
h23ih34ih41i

Excercise VIC3.1
Repeat these calculations using scalars in place of the spinors.

4. Masses
The spacecone formalism yields the simplest method for deriving S-matrix ele-
ments in massless theories (at least for trees; for loops it may be preferable to use
background field gauges, with a Lorentz gauge, like Gervais-Neveu, for the quantum
gauge and spacecone for the background gauge). The analogous method for the mas-
sive case is to use actions based on self-dual fields, as described in subsection IIIC4.
The advantage of these two methods is that they use fields that are representations of
the little group, so in the massive case fields have 2s+1 components and only undotted
spinor indices (SO(3)=SU(2)), while in the massless case they have only 2 components
and no indices (SO(2)=U(1)). Although the actions used are more complicated, this
is just a reflection of the fact that algebra that is usually done repetitively in graphs
has been performed once and for all in the action.
However, in the massive case the simplification is not as drastic as in the massless
one: S-matrix elements are just simpler in massless theories, with many vanishing;
the spacecone method takes advantage of this simplification in the final results by
simplifying the intermediate steps. The massive examples we will consider in this
subsection, taken from QED, are somewhat simple in any case, so we will stick to
340 VI. QUANTUM GAUGE THEORY

the older methods (although the uses of methods based on self-duality are still being
explored).
The major difference in simplicity between the massless and massive cases (in any
approach) is in the external line factors. The ambiguity in the explicit expressions
for the external line factors is just the little group: In the massless case the solutions
to the field equations (one solution and its complex conjugate) are unique up to a
phase factor, which is why the twistor formalism is so useful. In the massive case the
solutions (2s+1) are ambiguous up to an SU(2) transformation, which means they are
messy for any choice. Just as in the massless case the twistor is part of the Lorentz
transformation from an arbitrary frame to the lightcone frame, in the massive case
the solutions are part of the transformation to the rest frame. In other words, the
external line factors simply convert Lorentz indices to little-group indices; this makes
indices trivial for the massless case (in D=4), and not as nice for the massive.
The result is that in practice whenever any of the external particles are massive
their external line factors are left as implicit in S-matrix elements, and only their
squares are explicitly evaluated, in cross sections. This was common in older experi-
ments (especially QED), since recent experiments are mostly at energies so high that
masses of external, stable particles are usually neglected. This adds to the algebra,
since it means that Lorentz algebra is performed in each of n2 terms in the cross
section rather than n terms in the S-matrix.
Furthermore, the algebra is usually simplified by considering experiments where
polarization is determined in neither the preparation of the initial states nor the
measurement of the final states. This was also common in older experiments, when
devices for polarization were not well developed. The result is that one averages over
initial states and sums over final states, producing the same algebraic factors that
appear in the propagator, as described in subsection VB3: ∆ is replaced with ∆+.
One then applies the rules for Feynman diagrams for cross sections, as described in
subsection VC7.
The standard S-matrices in QED are the 4-point tree graphs, with 2 3-point
vertices and 1 internal propagator. There are just 2 graphs to consider, with various
labelings of momenta: (1) The graph with 4 external fermions (electrons/positrons)
connected by 1 internal photon describes both Møller (electron-electron) and Bhabha
(electron-positron) scattering, 2 labelings each. (2) The graph with 2 external photons
and 2 external fermions, as a continuous line that includes the 1 internal fermion,
describes Compton (electron-photon) scattering as well as electron-positron creation/-
annihilation, also 2 labelings each. In each case, the 1 S-matrix diagram results in 2
 C. SCATTERING 341

cross section diagrams, each with 2 momentum labelings (for a total of 2×2=4): 1
diagram from multiplying similar terms and 1 from cross-terms.
In Dirac spinor notation the Lagrangian for QED is (see subsection IIIA4)

1 2
8
F + Ψ̄(i∂/ − eA
/+ m
√
2
)Ψ

where we have scaled the “e” into the vertex. The Feynman rules are now (Fermi-
Feynman gauge):

Photon propagator: ηab / 12 p2

Fermion propagator: (p/ + √m2 )/ 12 (p2 + m2)
Vertex: eγa

(We use the Fermi-Feynman-gauge propagator also for defining the cut propagator;
ghosts decouple in QED.)
The cross section diagrams contain closed fermion loops, resulting in traces of
products of γ matrices (with a −1 for each loop by Fermi-Dirac statistics). The
algebra is manageable for the present case, using the 4D γ-matrix identities from
subsection IIA6:

γ a γa = −2, γ a a/γa = a/, γ a a//bγa = a · b, γ a a//b/cγa = c//ba/

tr(I) = 4, tr(a//b) = −2a · b, tr(a//b/c/d) = a · b c · d + a · d b · c − a · c b · d

The traces encountered in the above processes are of the form

N1 = tr(γ a Aγ bB)tr(γa Cγb D)

N2 = tr(γ a AγaBγ b Cγb D), N3 = −tr(γ aAγ b Bγa Cγb D)

m
√
where A = a/ + 2
, etc. Using the above identities, these are evaluated as

N1 = 4m4 + 2m2 (a · b + c · d) + 2(a · c b · d + a · d b · c)

N2 = 4m4 + m2 [2(a + c) · (b + d) − a · c − 4b · d] + (a · b c · d + a · d b · c − a · c b · d)

N3 = 2m4 + m2(a · b + a · c + a · d + b · c + b · d + c · d) + 2a · c b · d

Excercise VIC4.1
Generalize the above identities and expressions for the N’s to arbitrary di-
mension D.
342 VI. QUANTUM GAUGE THEORY

3 4 3 4
t
s

1 2 1 2

Our first example is e+e− → e+e− (“Bhabha scattering”). We have aligned

all momenta to be that of the electrons (i.e., minus that of the positrons), so that
all numerator factors are p/ + √m2 without signs. Specifically, we have chosen p1 for
the (positive-energy) momentum for the incoming electron, −p2 for the incoming
positron, p3 for the outgoing electron, and −p4 for the outgoing positron. With these
conventions,

−s = (p1 − p2 )2 = (p3 − p4 )2 , −t = (p1 − p3 )2 = (p2 − p4 )2

−u = (p1 + p4 )2 = (p2 + p3 )2 (p2i = −m2, s + t + u = 4m2)

⇒ p1 ·p2 = p3 ·p4 = 12 s−m2, p1 ·p3 = p2 ·p4 = 12 t−m2, p1 ·p4 = p2 ·p3 = − 12 u+m2

l k

k j
a b a b
j l
i i

N1(ijkl) N3(ijkl)

For the squared amplitude we have for the average over initial polarizations and
sum over final
X N1(1342) N1 (1243) N3 (1243) + N3(1342)
1
4
|T | 2
= + +
t2 s2 st
pol

f(s) + f(u) 1 f(t) + f(u) f(u)

1
= 2 +2 +
t2 s2 st
4
not including the overall factor of e , where

f(x) ≡ (x − 2m2 )(x − 6m2 )

Every other N term is the result of switching s ↔ t (p2 ↔ p3 , or p1 ↔ p4 ) in the

previous, since that is the relation of the 2 Feynman graphs contributing to the S-
matrix. The N1 terms are the squared diagrams, while the N3’s are the cross terms.
The “−” in N3 comes from Fermi-Dirac statistics, switching two fermion lines.
 C. SCATTERING 343

Finally, adding the appropriate factors to get the differential cross section (see
subsection VC7)
   
dσ (2π)3e4 f(s) + f(u) f(t) + f(u) f(u)
= + +2
dt Bhabha s(s − 4m2) t2 s2 st
The probabilities |T |2 for e− e− → e−e− (“Møller scattering”), or e+e+ → e+e+, are
related by crossing symmetry s ↔ u (p1 ↔ −p3 or p2 ↔ −p4 ):
   
dσ (2π)3 e4 f(s) + f(u) f(s) + f(t) f(s)
= + +2
dt Møller s(s − 4m2 ) t2 u2 tu

A convenient frame for any of these cross sections is the center-of-mass frame
(subsection IA4). In these cases all the external masses are equal, so the Mandelstam
variables have simple expressions in terms of the energy (which is the same for all 4
particles) and the scattering angle:

s = 4E 2 , t = −4(E 2 − m2)sin2 2θ , u = −4(E 2 − m2)cos2 θ

2

2 4 2 4

s u

1 3 1 3

Another famous example is e−γ → e−γ (“Compton scattering”). Now we label

p1 for the incoming electron, p3 for the incoming photon, p2 for the outgoing electron,
and −p4 for the outgoing photon, so the Mandelstam variables are

−s = (p1 + p3 )2 = (p2 − p4 )2 , −t = (p1 − p2 )2 = (p3 + p4 )2

−u = (p1 + p4 )2 = (p2 − p3 )2 (p21 = p22 = −m2, p23 = p24 = 0; s + t + u = 2m2 )

⇒ p1 · p2 = 12 t − m2, p1 · p3 = − 12 (s − m2), p1 · p4 = − 12 (u − m2 )
p2 · p3 = 12 (u − m2), p2 · p4 = 12 (s − m2), p3 · p4 = − 12 t

k k
b b
j
j
a a
l
i
i
N2(ijkj) – N3(ijkl)
344 VI. QUANTUM GAUGE THEORY

The probability is
X N2(1, 1 + 3, 2, 1 + 3) N2 (1, 1 + 4, 2, 1 + 4)
1
|T |2 = +
4
(s − m2)2 (u − m2)2
pol

N3(1, 1 + 4, 2, 1 + 3) + N3 (1, 1 + 3, 2, 1 + 4)
−
(s − m2)(u − m2)
1m
4
+ m2(3s + u) − su 1 m4 + m2(3u + s) − su m2(t − 4m2)
= + −
2
(s − m2)2 2
(u − m2)2 (s − m2)(u − m2)
where now every other term comes from switching s ↔ u (p3 ↔ p4 ), and the cross
section is, after some rearrangement,
  "  2  
dσ (2π)3e4 4 1 1 2 1 1
= 4m + + 4m +
dt Compton (s − m2)2 s − m2 u − m2 s − m2 u − m2
 
u − m2 s − m2
− +
s − m2 u − m2
A useful frame is the lab frame (i.e., the rest frame of the electron), where in
terms of the initial and final (positive) energies (E and E 0 ) and scattering angle of
the photon we have

1 1 2 sin2 2θ
s = m2 + 2mE, u = m2 − 2mE 0 , t = 2m(E 0 − E); − =
E0 E m

By crossing symmetry, s ↔ t, we get e+e− → 2γ (“pair annihilation”) and

2γ → e+e− (“pair creation”):
  "  2  
dσ (2π)3 e4 4 1 1 2 1 1
= 4m + + 4m +
dt annihil. s(s − 4m2 ) t − m2 u − m2 t − m2 u − m2
 
u − m2 s − m2
− +
t − m2 t − m2
  "  2  
dσ (2π)3 e4 4 1 1 2 1 1
= 4m + + 4m +
dt creation s2 t − m2 u − m2 t − m2 u − m2
 
u − m2 s − m2
− +
t − m2 t − m2
Excercise VIC4.2
Calculate all the corresponding massless cross sections using the spacecone
gauge. Show they agree with the m = 0 case of the above.
 C. SCATTERING 345

5. Supergraphs
In supersymmetric theories the easiest way to calculate Feynman diagrams is in
superspace. Supersymmetric cancelations are then automatic, and new special prop-
erties of supersymmetric theories are revealed. The derivation of the “supergraph”
rules is similar to that of subsection VC1, except for some fine points in the treat-
ment of chiral superfields. The path integral required the explicit evaluation of only
a Gaussian for perturbation. Since we dropped proportionality constants, this was
equivalent to substituting the solution to the classical, free field equations back into a
quadratic action. For real scalar superfields (used for super Yang-Mills) this is trivial,
but chiral scalar superfields (used for scalar multplets) satisfy the chirality constraint,
R
and have superpotential terms: integrals over chiral superspace ( dx d2 θ), not the
R
full superspace ( dx d4 θ).
We want to make use of the identity for evaluating the path integral (see subsec-
tion VC1)
Z Z
du −uM u/2 du −1
√ e f(u + v) = √ e−uM u/2 eu∂v f(v) ∼ e∂v M ∂v /2f(v)
2π 2π
Then the “action” we need to integrate is
Z Z  Z 
δ
S̃ = − dx d θ φ̄φ −
4
dx d θ ( √2 ) 2 φ + h.c. −
2 m 1 2 2
dx d θ φ + h.c.
δϕ

consisting of the (derivative part of the) kinetic term, mass term, and (minus the)
term that acts on e−SI [ϕ]. Solving the field equations (see subsection IVC2)
δ δ
d̄2 φ̄ + m
√
2
φ + = d2 φ + m
√
2
φ̄ + =0
δϕ δ ϕ̄
we find
   
1 2 δ δ 1 δ δ
φ= 1 d̄ − √2
m
, φ̄ = 1 d2
− √2
m

2 (− + m2 ) δ ϕ̄ δϕ 2 (− + m2 ) δϕ δ ϕ̄
R
The propagator exponent 12 (δ/δϕ)(1/K)(δ/δϕ) thus becomes (putting back ϕ → φ)
Z   Z   
δ 1 δ 4 1 δ 1 δ
4
dx d θ −1 + dx d θ 2 m
√
2 1 (−
+ h.c.
δ φ̄ 2 (− + m2) δφ δφ 2 + m2) δφ

Before writing the Feynman rules, we first note that functional differentiation
with respect to a chiral superfield, as follows from the above variation, gives
δ
φ(x0, θ0 ) = d¯2 δ 4(θ − θ0)δ(x − x0 )
δφ(x, θ)
346 VI. QUANTUM GAUGE THEORY

This means that there will be an extra d̄2 at the φ end of any chiral propagator and
an extra d2 at the φ̄ end. We could associate these directly with the propagator, but
R R
we will use one factor of d̄2 to convert a d2 θ into d4 θ at any superpotential vertex,
and similarly for the complex conjugate. Therefore, we include such factors explicitly
as a separate Feynman rule for the ends of chiral propagators. However, this means
the φφ propagator (and similarly for φ̄φ̄) gets an extra factor of d2 / 12 to compensate
for the fact that we include two d̄2 factors, whereas it really had only one because
its integral was only d2 θ. Furthermore, we Fourier transform x as usual, but not θ,
basically because there is no translation invariance in θ, but also for a better reason
to be explained soon. The Feynman rules of subsection VC4 are then modified as:
R
(A2 12 ) Theta’s: one for each vertex, with an d4 θ.
(A30 ) Propagators:
1
VV : 1 2 δ 4(θ − θ0 )
2 (p + m2 )
1
φ̄φ : − 1 2 2
δ 4(θ − θ0 )
2 (p + m )
 
d2 1
φφ : m
√ δ 4(θ − θ0 )
2 −2p
1 2 1 2
(p + m 2)
 ¯2  2
d 1
φ̄φ̄ : m
√ δ 4(θ − θ0 )
2 − 2 p2 2 (p2 + m2)
1 1

(A4 12 ) Chiral vertex factors: d¯2 on the φ end(s) of every chiral propagator,
d2 on the φ̄ end(s), but drop any one such factor at superpotential vertex.

We next explain how θ integrations are performed on any connected graph. Con-
sider any two vertices directly connected by a propagator. All the spinor derivatives
acting on its δ 4(θ − θ0) can be removed from that propagator by integration by parts.
R
We then can use that δ function to trivially integrate over θ0 , removing the d4 θ0 and
that δ 4 (θ − θ0), and replacing θ0 everywhere with θ. Effectively, those two vertices
have been contracted to the same point in θ space, eliminating that propagator as
far as θ dependence is concerned. We can repeat this procedure until all vertices are
contracted to a single point. However, we are then left with a “tadpole” for each
loop: Contracting propagators this way sequentially around a loop identifies all the
vertices of that loop, and leaves the loop as a single propagator with both ends at
that point. To evaluate this tadpole, we note that
[d̄2d2 δ 4 (θ − θ0)]|θ0 =θ = 1
(θ0 derivatives can be converted into minus θ derivatives when acting directly on
the δ; this is basically integration by parts.) Fewer derivatives give 0; more can be
 C. SCATTERING 347

reduced to terms of 4 or less. This completes all the evaluation in θ space, leaving an
expression in terms of fields (some with d’s acting on them) with different momenta,
times the usual momentum factors, with the usual momentum integrals, but all at
R
the same θ, with a single d4 θ. This means that the generating functionals W and
Γ are completely local in θ.
There is a further consequence of this evaluation. We have obtained terms with
R R
d θ, but none with d2 θ. However, to do it we had to introduce the factors
4

d2 /(− 12 p2 ) into the φφ propagators. On the other hand, such a factor can easily
be killed by a d¯2 from a vertex: We sandwich the d2 between a d¯2 from each vertex,
using the identity d¯2 d2 d¯2 = − 1 p2 d¯2 , and return the d¯2 to one vertex. The only time
2
we can’t do that everywhere is if every vertex is a superpotential (so every propagator
in the graph is φφ and every external field is φ), since otherwise we can inductively
R
borrow d¯2 ’s from some non- d2 θ vertex. Any such 1PI graph always vanishes, be-
cause there are exactly enough d’s left to make the tadpoles nonvanishing, leaving an
R 4
d θ of a product of φ’s with no d’s, which vanishes. On the other hand, for a tree
R R
graph there is exactly one d2 /p2 left, which converts the d4 θ to an d2 θ.
The net result is that not only are W and Γ local in θ, but only their classical
R
parts can have d2 θ terms, and the spurious d2 /p2 factors (which should not appear
in massive theories) are always canceled. In particular, this implies that all UV
R
divergences are d4 θ terms: All terms in the superpotential are unrenormalized (no
loop corrections) to all orders in perturbation theory.
Excercise VIC5.1
Calculate all the contributions to W [φ, φ̄] from 4-point trees in massive super-
φ3 theory, and write the result in both p- and x-space (in analogy to the
nonsupersymmetric example at the end of subsection VC4).
Improvements again result from using background-field gauges. We have already
seen in subsection VIB10 the modification to the quantization for supersymmetric
background-field gauges. The background-field expansion can be defined by solving
the constraints on the full covariant derivatives in terms of quantum prepotentials
but background potentials (AA, not V ), essentially by covariantizing dA to the back-
ground ∇A. Then ∇α can be manipulated (integration by parts, etc.) in the graphs
in the same way as dα was, leaving only Aa (not Aα) and its derivatives (Wα , etc.) as
background fields. This leads to improved power counting, and can be used to prove
“superrenormalizability” (finiteness beyond one loop) for N=2 extended supersym-
metric theories, and finiteness for N=4.
348 VI. QUANTUM GAUGE THEORY

As for other gauges (background-)chiral superfields need special treatment, now

to get the most out of background gauge invariance. Variation can be defined in the
obvious way, but now we also need the covariantized identity

D̄2 D2 φ = 12 ( + i[W α, Dα ])φ

from pushing the D̄’s to the right and using the commutation relations. The func-
tional integral over the quantum background-chiral superfields can also be performed
in the same way as for other gauges, the only modifications being background co-
variantization (including the above “nonminimal” term for ), and the fact that
we can no longer neglect the “vacuum” contribution (one-loop diagrams with only
background fields externally). Specifically, if we look at the general derivation of
the Feynman rules in subsection VC1, we see it gave rules for all graphs except the
one-loop vacuum bubble, since this graph has no (quantum) vertices. These rules, as
adapted to superspace earlier in this subsection, are now modified only by the covari-
antization just discussed, which only adds background potentials (not prepotentials)
and field strengths to propagators and vertices. The background-covariantized prop-
agators then can be further expanded about the free . The net result is that in
all diagrams except (perhaps) these chiral one-loop vacuum bubbles the background
fields appear only in the form of potentials and field strengths. These vacuum bubbles
then can be evaluated by the usual methods, since the formerly neglected Gaussian
path integral for these “quantum-free” fields is just the usual one-loop path integral
for a chiral superfield with external Yang-Mills superfields, only the external fields are
now identified as background instead of quantum. In some cases, this last calculation
can be further simplified to again yield an expression directly in terms of potentials
without explicit prepotentials (see subsection VIIIA6 below).

REFERENCES
1 M.T. Grisaru, H.N. Pendleton, and P. van Nieuwenhuizen, Phys. Rev. Lett. 15 (1977)
996;
M.T. Grisaru and H.N. Pendleton, Nucl. Phys. B124 (1977) 333:
supersymmetry identities for trees.
2 M.T. Grisaru and W. Siegel, Phys. Lett. 110B (1982) 49:
generalization to one loop.
3 S.J. Parke and T. Taylor, Nucl. Phys. B269 (1986) 410, Phys. Rev. Lett. 56 (1986)
2459;
F.A. Berends and W.T. Giele, Nucl. Phys. B306 (1988) 759:
Parke-Taylor amplitudes.
4 W.A. Bardeen, Prog. Theor. Phys. Suppl. 123 (1996) 1;
D. Cangemi, hep-th/9605208, Nucl. Phys. B484 (1997) 521;
 C. SCATTERING 349

G. Chalmers and W. Siegel, hep-th/9606061, Phys. Rev. D54 (1996) 7628:

relation of Parke-Taylor amplitudes to self-dual Yang-Mills.
5 R. Brooks (May, 1993), unpublished;
N. Berkovits and W. Siegel, hep-th/9703154, Nucl. Phys. B505 (1997) 139:
use of first-order actions with self-dual auxiliary fields for perturbation about self-
duality.
6 Feynman, loc. cit. (VB, ref. 4):
Feynman diagrams for QED, from the horse’s mouth. The original derivation, not count-
ing his later-published first-quantized path-integral approach. Basically a mechanics
point of view.
7 H.J. Bhabha, Proc. Roy. Soc. A154 (1936) 195.
8 C. Møller, Ann. Phys. 14 (1932) 531.
9 O. Klein and Y. Nishina, Z. Phys. 52 (1929) 853:
Compton scattering.
10 A. Salam and J. Strathdee, Phys. Rev. D11 (1975) 1521;
K. Fujikawa and W. Lang, Nucl. Phys. B88 (1975) 61;
J. Honerkamp, M. Schlindwein, F. Krause, and M. Scheunert, Nucl. Phys. B95 (1975)
397;
S. Ferrara and O. Piguet, Nucl. Phys. B93 (1975) 261;
D.M. Capper, Nuo. Cim. 25A (1975) 259;
R. Delbourgo, Nuo. Cim. 25A (1975) 646, J. Phys. G 1 (1975) 800;
R. Delbourgo and M. Ramón Medrano, Nucl. Phys. B110 (1976) 473;
W. Siegel, Phys. Lett. 84B (1979) 193, 197:
early supergraphs.
11 Wess and Zumino, loc. cit. (IVC, ref. 2);
J. Iliopoulos and B. Zumino, Nucl. Phys. B76 (1974) 310;
Capper, loc. cit.;
Delbourgo, loc.cit. (first ref. above);
B. Zumino, Nucl. Phys. B89 (1975) 535;
P.C. West, Nucl. Phys. B106 (1976) 219;
D.M. Capper and M. Ramón Medrano, J. Phys. G 2 (1976) 269;
S. Weinberg, Phys. Lett. 62B (1976) 111:
nonrenormalization theorems for chiral superfields, from components or old-fashioned
supergraphs.
12 Grisaru, Siegel, and Roček, loc. cit. (VB, ref. 14);
M.T. Grisaru and W. Siegel, Nucl. Phys. B201 (1982) 292;
Gates, Grisaru, Roček, and Siegel, loc. cit.;
M.T. Grisaru and D. Zanon, Phys. Lett. 142B (1984) 359, Nucl. Phys. B252 (1985)
578, 591:
supergraphs as done today; more general nonrenormalization theorems.
350 VII. LOOPS

VII. LOOPS
Although our analysis so far is sufficient to evaluate the lowest-order term in the
h̄ expansion (“trees”), certain new features arise at higher orders.

............................ ............................
............................ A. GENERAL ............................

When infinities were first found in perturbative quantum field theory, they were
thought to be a serious problem. A prescription can be given to remove these in-
finities, called “regularization”. It was later shown that this regularization can be
defined in such a way as to preserve all the desirable physical properties of the theory,
called “renormalization”. Unfortunately, it seems that the original infinities, exiled
by renormalization from any finite order of perturbation theory, return to plague field
theory when all orders of perturbation theory are summed. Therefore, renormaliza-
tion should not be considered a cure to the disease of infinities, but only a treatment
that allows divergent theories to be more useful.

1. Dimensional renormalization
“Perturbative renormalization” is defined to preserve the three properties that
define relativistic quantum field theory (Poincaré invariance, unitarity, causality).
The most general prescription is to start with a classical theory that is causal, Poincaré
invariant, and satisfies the semiclassical part of unitarity (as described in subsection
VC5). This gives the tree graphs of the theory. One then applies unitarity to define
a perturbation expansion, determining the higher orders (loop diagrams) from the
lowest (trees). Although the usual loop diagrams are divergent, there is enough
ambiguity in the unitarity condition to allow removal of the divergences.
The only practical way to implement this procedure is to slightly modify the di-
vergent graphs one obtains from the naive use of the Feynman graph rules (obtained,
e.g., from path integral quantization of a classical action). The steps are: (1) “Reg-
ularize” each divergent graph by modifying the momentum integrals, introducing a
parameter(s), the “regulator(s)”, giving a finite result that reproduces the original
divergent integral in a certain limit. (2) “Renormalize” each regularized graph by
subtracting out the “divergent part” of the regularized graph, keeping only the “fi-
nite part”. Once the graph has been rendered finite, the regulator can be dropped.
One then has to check that the method of removing divergences, order-by-order
in the perturbation expansion, preserves the three properties of relativistic quantum
 A. GENERAL 351

field theory. The easiest way to do this is to use both a regularization scheme and a
subtraction scheme that preserve these properties manifestly. The standard way for
the subtraction scheme to do this is to change the coefficients of terms in the classical
action by (real) constants that depend on both the regulator and h̄. Since the classical
action already satisfies Poincaré invariance, causality, and the semiclassical part of
unitarity, this will automatically preserve all the desired properties. In this manner
of renormalization, there thus remains only two steps to prove that a theory can be
renormalized: (1) the existence of a regularization that manifestly preserves the three
properties, and (2) that the modification of the action by making the coefficients
regulator- and h̄-dependent is sufficient to cancel all divergences that might reappear
in the limit as the regulator is removed.
The latter step, discussed in subsection VIIA5 below, can be further divided into
substeps, proving: (a) The ultraviolet divergence in any graph corresponding to scal-
ing all internal (integration) momenta to infinity (the “superficial” divergence) comes
from a term in that graph polynomial in external momenta, and can therefore be can-
celed by a local term from the action; and (b) recursively in the number L of loops,
if the renormalization procedure has already been successfully applied through L−1
loops, the resulting modification of the action is sufficient to cancel all divergences
appearing at L loops except the superficial ones. The former substep is the one that
detemines whether the theory can be renormalized.
The former step is satisfied by dimensional regularization, the standard method of
regularization in relativistic quantum field theory (and for practical purposes beyond
one loop, the only one). It is defined by writing the theory under consideration in
arbitrary dimensions D, and treating integrals over loop momenta as analytic func-
tions of D. These integrals are then analytically continued from lower D, where they
are (ultraviolet) convergent. The resulting expressions have pole singularities in D at
integer D, so these poles can be subtracted out as the divergent parts.
There are two main reasons why dimensional regularization is so useful: (1) Most
classical actions can be written as easily in arbitrary dimensions as in D=4. (The
important exception is those that in some way involve the Levi-Civita tensor ab...c.
The difficulty with such theories is not a drawback of dimensional regularization, but
a general property of quantum field theory, and is related to the quantum breakdown
of classical symmetries, to be discussed later.) In particular, this means it manifestly
preserves gauge invariance (which is a part of unitarity), the property of relativistic
quantum field theory most difficult to preserve. It is also the only workable scheme
of regularization to do so.
352 VII. LOOPS

(2) It requires only one regulator, the number of dimensions D itself. (Most other
regularization schemes require at least one regulator for each loop.) This is the main
reason why this scheme is the only practical method of regularization at higher loops.
(An enormous number of regularization schemes have been proposed, almost all of
which work well at one loop, but all of them are more difficult than dimensional
regularization already at two loops, and even worse at three loops and higher.)
The renormalization scheme based on this regularization is also very simple:
Defining D = D0 −2 (where D0 is the physical dimension, usually 4), we can Laurent
expand any L-loop amplitude in , starting at 1/L . These 1/n terms arise from n
or more divergent D-momentum integrals. If we cancel all the negative powers of ,
we can take the limit D → D0 ( → 0) by just dropping all the positive powers of ,
i.e., evaluating the remainder at  = 0. The procedure is to modify the coefficients
of the terms in the “classical” action (couplings, masses, and field normalizations) by
making them - and h̄-dependent, giving them h̄L−1 /n (“counter”)terms. Such terms
can cancel any local divergence at L loops. One then has to show that they also can-
cel all nonlocal divergences at higher loops resulting from this divergence appearing
in an L-loop subgraph. Thus, the procedure is recursive: (1) apply the counterterms
obtained from calculations at less than L loops to cancel subdivergences; (2) cancel
the remaining, local, superficial divergences by introducing new L-loop counterterms.
The form of the superficial divergence can be determined by evaluating the divergence
coming from the region of momentum space where all loop momenta go to infinity
at the same rate. The superficial divergence is determined by 1/ terms of this loop
and of subloop divergences; however, if the 1/ piece of a prospective counterterm
vanishes at a certain loop order, so do all higher powers at that loop order. Thus, sim-
ple power counting (as well as global and local symmetries) is sufficient to determine
what counterterms can appear at any particular loop order.
These rules are sufficient for evaluating momentum integrals to the point where
renormalization can be applied. However, further simplifications are possible where
spin is involved: Techniques specific to D=4 are useful to simplify algebra in general,
and required to preserve manifest supersymmetry in particular. These methods treat
spin indices as 4D, in contrast to the vector indices on momenta (and coordinates),
which are analytically continued away from D=4 by the definition of dimensional
regularization. This is natural in 4D N=1 supersymmetric theories formulated in
superspace (or 4D N>1 in N=1 superspace), since there spins ≤ 1 are described by
scalar prepotentials: There the simple prescription is to continue in the dimension
of the commuting coordinates (x), while fixing the dimension of the anticommuting
 A. GENERAL 353

coordinates (θ). These rules, for either supersymmetric or nonsupersymmetric the-

ories, have a simple physical interpretation for integer D<4: dimensional reduction.
The reduced theories differ from those produced by simple dimensional regularization:
Vectors get (4−D) extra scalars; spinors may become multiple spinors. For super-
symmetric theories this is again natural, since vector and scalar multiplets remain
irreducible after dimensional reduction. Unfortunately, vectors in nonsupersymmet-
ric theories reduce to vectors plus scalars that are not related by any symmetry, so
their renormalization is independent. However, the complications produced by the
extra renormalizations are usually smaller than the algebraic simplifications resulting
from the restriction to 4D spin algebra, especially for lower loops. Another compli-
cation is that Levi-Civita () tensors can’t be treated consistently in the dimensional
reduction scheme. Although serious in principle, in practice this is not a problem as
long as axial anomalies cancel, which is required anyway for unitarity. (See subsection
VIIIB3. Also, axial anomalies are easier to calculate using Pauli-Villars regularization
than with any form of dimensional regularization.)

2. Momentum integration

The first step in performing momentum integration is to make all integrals Gaus-
sian by exponentiating propagators using Schwinger parameters
Z ∞
1 −τ (p2+m2 )/2
1 2 2
= dτ e
2 (p + m ) 0

The general momentum integral in an arbitrary Feynman diagram is then

Z Z 
dLD k 
−kT ·A(τ )k/2−kT ·B(τ,p)+ikT ·x−C(τ,p,m) 
A = N(p, −i∂x) LD/2
P
d τ e 
(2π) x=0

where A, B, C are first-order in τ ; B is first-order in p while C is second-order in p, m;

and L is the number of loops and P the number of propagators. Also, we have used
matrix notation with repsect to the P -dimensional τ -space, with respect to which A is
a matrix, B is a vector, and C is a scalar. Finally, N represents all “numerator” factors
(propagator numerators and vertices; everything but propagator denominators) and
has been brought outside the integral by Fourier transformation (i.e., introducing x
to produce a generating functional for all numerators). The momentum integrals are
now Gaussian, and can be evaluated by the methods of subsection IB3 (and VA2):
Z 

−D/2 (B−ix)T A−1 (B−ix)/2−C 
A = N(p, −i∂x) d τ (det A)
P
e 
x=0
354 VII. LOOPS

Some of the τ integrations can be performed by various scalings of subsets of the

τ ’s. For example, to see the superficial divergence of the graph we scale all of the τ ’s
and insert the identity as
Z ∞ !
X
1= dλ δ λ − τi , τi = λαi
0 i

where α are “Feynman parameters”. The amplitude then becomes

Z  X 
A = N(p, −i∂x) dλ λP −1−LD/2 dP α δ 1 − α [det A(α)]−D/2

T A(α)−1 B(α)/2−C(α)]−iB(α)T ·A(α)−1x−xT ·A(α)−1x/2λ

×eλ[B(α) |x=0

(This method of introducing Feynman parameters is equivalent to directly changing

variables from τ ’s to λ and one less α, and finding the Jacobian.)
The x derivatives in N must now be taken. For a contribution from these deriva-
tives of order λ−n , we have λ integrals of the form
Z
T −1
dλ λP −1−LD/2−n eλ[B A B/2−C] = Γ(P − 12 LD − n)[C − 12 B T A−1 B]−P +LD/2+n

where we have used the definition of the Γ function

Z ∞
Γ(z) = dλ λz−1 e−λ
0

This integral converges only for Re z > 0, but we can extend it to (almost) all z by
analytic continuation: Using integration by parts,
Z ∞ Z ∞
−λ d z −λ z ∞
zΓ(z) = dλ e λ = (e λ )|0 + dλ λz e−λ = Γ(z + 1)
0 dλ 0

in the convergent region. Analytic continuation then says to evaluate the integral for
Γ(z) in the region 0 ≥ Re z > −1 as Γ(z + 1)/z in terms of the integral for Γ(z + 1),
and so on: Z ∞ !Z
Yn
1 ∞
dλ λz−1 e−λ = dλ λz+n e−λ
0 z + k 0
k=0

Thus, Γ(z) has simple poles in z at all the nonpositive integers.

Excercise VIIA2.1
Using the identity Γ(z + 1) = zΓ(z), derive the following special cases for
nonnegative integer n:
√ (2n)! √
Γ(n + 1) = n!, Γ(n + 12 ) = (n − 12 )(n − 32 )... 12 π = π
n!22n
 A. GENERAL 355

Feynman parameter integrations give more complicated functions. A simple but

common example is the Beta function
Z 1
Γ(x)Γ(y)
B(x, y) = dα αx−1 (1 − α)y−1 =
0 Γ(x + y)
The latter expression for the Beta function can itself be derived by similar methods:
Excercise VIIA2.2
Derive the following B function identities:
a Use the integral representation of the Γ function to write an expression for
Γ(a)Γ(b) as an integral over two Schwinger parameters, and introduce a scal-
ing parameter (as with general two-propagator Feynman graphs, except here
there is no momentum). Show the result is Γ(a + b)B(a, b), where B is given
by the integral definition, thus proving the Beta function can be expressed in
terms of Gamma functions.
b Use the integral definition of B to prove
√
Γ(x) 21−2x π
B(x, x) = 2 1−2x
B( 12 , x) ⇒ =
Γ(2x) Γ(x + 12 )

Use this result to find the expression in excercise VIIA2.1 for Γ(n + 12 ).
c Derive the identity
Z ∞
dτ τ a−1 (1 + τ )−a−b = B(a, b)
0

from the substitution τ = z/(1 − z). Use this to show

Γ(z)Γ(1 − z) = π csc(πz)

by changing variables τ = eu, and closing the contour in the complex plane
to pick up the contributions from the poles. We thus have

Γ(z + n)Γ(1 − z − n) = (−1)n Γ(z)Γ(1 − z)

(which is also seen inductively from Γ(z + 1) = Γ(z)).

In general UV divergences will come from powers of 1/ in a Γ resulting from
integration over some scaling parameter. We thus need an expression for the Laurent
expansion of Γ(z). This can be obtained directly from the integral expression: Using
the definition of e as a limit,
Z n Yn
1
Γ(z) = lim dλ λz−1 (1 − nλ )n = lim nz B(z, n + 1) = 1z lim nz
n→∞ 0 n→∞ n→∞
m=1
1 + mz
356 VII. LOOPS

from the change of variables λ = nα, and the above identities. Defining the “Euler-
Mascheroni constant” γ by
!
Xn
γ = lim −ln n + 1
m
= 0.5772156649...
n→∞
m=1

Y
n
−γz
⇒ e = lim n z
e−z/m
n→∞
m=1

we then can write

Y∞
ez/n
1 −γz
Γ(z) = e
z
n=1
1 + nz
which is an alternate definition of Γ. We then have
X
∞
ln Γ(1 − z) = γz + [−ln(1 − nz ) − nz ] ⇒
n=1

X
∞
ζ(n) X∞
1
ln Γ(1 − z) = γz + zn, ζ(y) =
n=2
n m=1
my
by Taylor expansion of the ln, where ζ is the “Riemann zeta function”.
Excercise VIIA2.3
Derive the following Γ identities from the previous:
a Find the first two terms in the Laurent expansion of Γ(z):
Z ∞
lim Γ(z) = lim z Γ(z + 1) = z − γ + O(z),
1 1
γ=− dλ (ln λ)e−λ
z→0 z→0 0

b Do the same for expansions about other integers:

" ! #
Xn
1
Γ(n + 1 + ) = n! 1 +  −γ + + O(2)
m=1
n
" ! #
1 1 Xn
1
Γ(−n + ) = (−1)n 1 +  −γ + + O(2)
n!  m=1
n

c Using the csc relation in excercise VIIA2.2c and the above expansion of
ln Γ(1 − z), show that
!
1 X∞
ζ(2n) 2n
cot z = 1−2 z
z n=1
π 2n

and thus ζ(2n) can be written as a rational number times π 2n.

 A. GENERAL 357

3. Modified subtractions
The convenient normalization for the quadratic part of the gauge-invariant action
for an arbitrary field theory we use is
Z
( 12 µ2 )(D−4)/2 dD x 1
S0 = φKφ
g2 (2π)D/2 2
for a real field φ and some coupling constant g, where for bosons
K = 12 (− + m2) + ...
The explicit factor of 1/2 cancels the factor of 2 obtained when varying the action with
respect to φ. (Similar permutation factors are used for interaction terms.) Equiva-
lently, it gives the natural normalization for Gaussian functional integration over φ
in the field theoretic path integral. For complex fields we instead have φ̄Kφ without
the 1/2, since then φ and φ̄ can be varied (or integrated over) independently.
The “renormalization mass scale” µ has been introduced to preserve the mass
dimension of g in arbitrary spacetime dimension; it appears naturally with the nor-
malization 12 µ2 because the kinetic operator contains 12 p2 and 12 m2. Generally there
will be more than one coupling, but only one µ. For some purposes it may be con-
venient to scale the fields so that the coupling and µ dependence appear only in the
interaction terms. We will usually suppress the µ dependence, since it is determined
by dimensional analysis, and is relevant only for quantum corrections, where D 6= 4
becomes important.
Note that our normalization differs from that normally chosen in the literature. It
has been chosen to give the normalization appropriate for Gaussian integrals, which
appear in both the first- and second-quantized theories. The first difference is the
factor of (2π)−D/2 for coordinate and momentum integration (rather than the usual 1
for coordinate and (2π)−D for momentum); the second is the factor of 1/2 multiplying
the kinetic operator p2 + m2 (contained in K), rather than 1. Our normalization is
more natural not only for Gaussian integration and Fourier transformation, but also
slightly simplifies perturbative field theory calculations, allowing one to ignore spuri-
ous factors of 2 and especially 2π. The net effect is only to change the normalization
of coupling constants, since such factors can be absorbed into the 1/g2 sitting in front.
For example, the most accurately experimentally verified prediction of quantum
theory is the anomalous magnetic moment of the electron, to be discussed later. The
result for the total magnetic moment, in various normalizations, to second order in
perturbation theory, is
e2 e2
µmag = 1 + e2 = 1 + m = 1 + f t2
2π 8π
358 VII. LOOPS

where “e” is for our normalization (obviously the simplest), “em ” is the normalization
you first learned in classical mechanics (the one that gives e2m /r2 as the electrostratic
force between two electrons), and “ef t ” is the one you would see in other quantum
electrodynamics courses. To complicate matters, you may have also seen the definition
α = e2m = e2f t/4π, of which the only merit is supposed to be that 1/α is very close to
the integer 137, which is silly since 1/e2 is even closer to the integer 861. (Actually,
α is the natural expansion parameter for nonrelativistic quantum mechanics, which
is basically 3D, but our e2 is more natural for the loop expansion, which is inherently
4D.) We have also used units h̄ = 1; restoring it introduces the further complication
α = e2m /h̄m = e2f th̄f t/4π because of the difference in the semiclassical expansions for
quantum mechanics and quantum field theory.
Furthermore, for nonabelian groups we have an extra factor of 12 compared to the
standard normalization because we normalize trD (Gi Gj ) = δij instead of trD (Gi Gj ) =
1
√
2 δij : The latter originated from the case of SU(2), where it cancels the 2 in the
diagonal generator (and in the others, if one uses hermitian ones rather than raising
and lowering). Unfortunately, for SU(N) with N ≥ 3 this historical normalization
√ √
introduces a 2, while not canceling factors like N (and making trD (Gi Gj ) N-
dependent would wreak havoc when considering subgroups, as well as for raising and
lowering operators). Thus, in the above notation,
gf2t
nonabelian : g2 =
16π 2
Although we have chosen a normalization for the coupling constants that is nat-
ural for Gaussian momentum integration, and for symmetry with respect to Fourier
transformation, in divergent integrals it has the disadvantage of having γ’s (Euler-
Mascheroni constant) in the finite parts. Another natural normalization that gets
rid of this irrational number from all graphs is to divide out the angular part of the
integrals: The volume of the unit D−1-dimensional sphere (the surface of unit radius
in D-dimensional Euclidean space), is easily evaluated with Gaussians:
Z Z Z ∞
dD k −k2 /2 −D/2
dk k D−1 e−k /2
D−1 2
1= D/2
e = (2π) d Ω
(2π) 0
Z
= 12 Γ( D2 )π −D/2 dD−1 Ω
We might therefore choose our normalization to cancel this factor in momentum
integrals, along with the (2π)−D from Fourier transformation. Then the action, e.g.,
for a massless scalar, might be normalized as (with conventional kinetic term)
Z R D−1 Z
µD−4 D ( d Ω) 1 2 µD−4 dD x
S0 = 2 d x (∂φ) = (∂φ)2
g (2π)D 2 g2 (4π)D/2Γ( D2 )
 A. GENERAL 359

This differs from our previous normalization by a factor of 1/Γ( D2 ), which is 1 in

exactly D=4, but differs infinitesimally far away. The result is that the two schemes
will differ effectively by finite renormalizations: For example, in divergent one-loops
graphs the 1/ divergences will have the same coefficient in the two schemes, but
the finite remainders will differ by constants, since effectively the coupling has been
redefined by a factor of 1+O(). Thus, the same result can be achieved by modifiying
the counterterm to be proportional to 1/ + constant. Hence, the earlier version of
dimensional regularization is called “minimal subtraction (MS)”, while the modifi-
cation inspired by the volume of the sphere is called “modified minimal subtraction
(MS)”.
We now examine explicitly the difference between the two schemes. As we can
see from our previous example, momentum integrals from scaling various subsets of
Schwinger parameters in a multiloop diagram will produce Γ function factors of the
form, with this new normalization,
Y
[Γ( D2 )]Li Γ(ni − Li D2 )
i
P
for some integers ni , where i Li = L. In even dimensions D0 (especially 4), we
can use Γ(z + 1) = zΓ(z) to write each Li -loop factor as a rational function of
 = (D0 − D)/2 times

[Γ(1 − )]Li Γ(1 + Li ) ∼ (eγ )Li e−Li γ = 1

where we have written only the γ dependence. Thus all γ’s cancel. This modifica-
tion allows some further simplification by eliminating extra terms arising at 2 loops
involving ζ(2) (but this is only π 2/6; see excercise VIIA2.3). Similar results can be
obtained by using Γ(1 − ) instead of Γ( D2 ).
Another subtraction scheme, the “G scheme”, is defined by normalizing momen-
tum integration so that the coefficient of the 1-loop massless propagator correction
in φ3 theory in 4 dimensions (see subsection VIIB4 below) is exactly 1/ (without
extra finite terms) times a power of 12 p2 , or that up to a sign in higher even dimen-
sions. (The normalization factor must be positive, and also finite and nonvanishing
as  → 0.) As for MS, we can also pull out rational factors to get just the Γ(1 + n)’s.
The net effect of these two schemes, as compared to MS, is to modify h̄, which appears
R R
as dx/h̄ in the classical action or h̄ dp for loop integrals, as

MS : h̄ → Γ( D2 )h̄ or Γ(1 − )h̄

(−1)D0 /2 Γ(1 − 2)
G: h̄ → h̄ or h̄
Γ(2 − 2 )B( 2 − 1, 2 − 1)
D D D
Γ(1 + )[Γ(1 − )]2
360 VII. LOOPS

(If we want to be picky, we can also normalize the former forms appropriately for
D = D0 , by including an extra factor of 1/Γ( D20 ) for MS and Γ( D20 − 1)/Γ(D0 − 2)
for G.)
This particular fix for eliminating irrational numbers works only for those arising
at one or two loops: In general, because subdivergences produce expressions of the
form Γ(1 + )/L at L loops, we encounter finite terms involving ζ(L) at L loops,
√
which is irrational (worse than just π’s, e’s, 2’s, etc.) for odd L. So, for example,
ζ(3) appears at 3 loops, and it can be shown that ζ(3) (and higher ζ(n)) can’t always
be canceled.
The “momentum subtraction scheme (MOM)”, rather than simplifying numer-
ically, is designed to give a more physical interpretation of the coupling constants
appearing in the action: It is defined so that they take their on-shell values. Thus,
it is particularly suited to low-energy calculations, which involve an expansion about
the mass shell. For example, consider the quantum kinetic operator K + ∆K, ap-
pearing in the quadratic part of the effective action. It depends on the momentum
through one variable: p2 or p/, etc. We then can consider Taylor expanding it in this
variable about its classical on-shell value. (For reasons to be explained later, this
can be dangerous for massless fields, when it requires infrared regularization.) This
is equivalent to expanding in powers of the classical kinetic operator K itself. The
MOM prescription is then to use subtraction terms −δK to cancel the terms in the
quantum correction ∆K to the kinetic operator that are linear in K:

∆K = a + bK + O(K 2 ) ⇒ δK = −a − bK
1 1
⇒ Kren ≡ K + ∆K + δK = K + O(K 2 ) ⇒ = + O(K 0)
Kren K
The two renormalizations are related directly to the “wave function” and mass renor-
malizations, one being proportional to the entire kinetic term, the other to the con-
stant (mass) term. The result is that the renormalized propagator has the same pole
and residue as the classical one.
Note that the MOM scheme, unlike the others, does not introduce an independent
mass scale µ: Only physical masses set the scale. This is a consequence of the fact
that the MOM scheme is designed for studying low-energy (near-mass-shell) behavior,
while the others are more suited for studying high-energy behavior. This will be
important for our explicit calculations later, when we see that MOM is more useful for
QED, which is better defined (and thus more useful), in terms of perturbation theory,
at low energies, while QCD is better defined at high energies. More precisely, the on-
shell values of QED masses and couplings are observed experimentally, whereas those
 A. GENERAL 361

of QCD are almost meaningless, since the corresponding particles are not observed
as asymptotic states. On the other hand, in QCD the introduction of the arbitrary
scale µ allows the definition of a more physical mass scale, and its arbitrariness can
actually improve the accuracy of perturbative calculations.

4. Optical theorem
Writing the S-matrix as S = 1 + iT , unitarity can be written as

1 = S †S = (1 − iT †)(1 + iT ) = 1 + i(T − T † ) + T † T ⇒ T † T = i(T † − T )

(Actually, the more useful statement is in terms of S = eiT , since then T represents
the connected graphs, but the result of the argument is the same.) Summation of a
probability over final states then yields the “optical theorem”:
X X
|Tf i |2 = hi|T † |fihf|T |ii = hi|T †T |ii = 2 Im Tii
f f

Applying unitarity in terms of the cutting rules (subsection VC6), we see that this
condition can be applied to Tii diagram by diagram, using any combination of parts
of Tf i and T † f i that fit together to form the graph considered for Tii . (For example,
the probability coming from a tree graph with two final states is the imaginary part
of a one-loop graph with two intermediate states.) Separating out the momentum-
conservation δ-function for a connected S-matrix element, we have finally
X  X X 
T =δ p T ⇒ δ p |Tf i|2 = 2 Im Tii
f

The simplest example of an experimental measurement of an interaction is a

decay rate. (The only particle properties contained in the free Lagrangian are mass
and spin.) At the tree (classical) level, this is given directly by a cubic (for decay
into two particles) or higher-order term in the Lagrangian. More generally, by the
optical theorem the decay probability is given by the imaginary part of the propagator
correction, evaluated on shell.
We then find the total decay probability per unit time by dividing the probabil-
ity by the spatial density ρ times the spatial volume times the time duration, and
summing over final states:
dP X P 2 Im Tii
= =
dt f
ρVD ω
362 VII. LOOPS

using the expressions for ρ, and P in terms of |Tf i|2, given in subsection VC7. The
optical theorem can be applied similarly for the total cross section:
(2π)D/2
σ = 2(Im Tii )
λ12

The decay rate for a particle is frame dependent, but we usually pick the rest
frame for massive particles, where ω = m. Alternatively, we can define the total
decay probability per unit proper time:
dt dP 2 Im Tii
ω=m ⇒ =
ds ds m
(where s and t should not be confused with the Mandelstam variables). For the
massless case, we can use instead the parameter τ , as it appears in classical mechanics
in the gauge v = 1, or as the classical value of the Schwinger parameter from the
Landau equations:
dxa dP
pa = ⇒ = 2 Im Tii
dτ dτ
Now the decay rate of a particle can also be associated with the imaginary part
of the mass, since

M = m − ir ⇒ |ψ|2 ∼ |e−iM t|2 = e−2rt

so the wave function for the particle at rest automatically includes a decay factor.
The probability of decay, normalized by dividing by |ψ|2, is thus
dP d(1 − |ψ|2)
= = 2r
dt |ψ|2dt
The analogous statement in momentum space is found by Fourier transforming the
propagator/wave function from time to energy:
1 1 1 1
e−iM t → = + (−ir) + ...
E −M E−m E−m E−m
This expansion is in terms of the free propagator 1/(E −m) and the connected graphs
−ir.
We now check that this result agrees with that obtained from the effective action.
The quantum propagator has a pole at p2 = −M 2 for some complex constant M:
1
lim ∆ = 1 2
p2 →−M 2
2 (p + M 2)
We have normalized the propagator at the pole by rescaling the field; we also keep
the real part of the mass the same as the classical value through renormalization of
 A. GENERAL 363

the mass term. (Only the real part can be renormalized consistently with unitarity.)
The on-shell condition is then at physical (real) momentum p2 = −m2, so the kinetic
operator on-shell is
1 2
2 (p + M 2 ) = − 12 m2 + 12 (m − ir)2 = − 12 r2 − imr

Remembering that “interaction” terms from Γ contribute with a minus sign to am-
plitudes, we then have
dP 2 Im Tii 2mr
= = = 2r
dt m m
5. Power counting
We now consider why subtracting out divergences, as poles in D, can be imple-
mented by giving singular D-dependence to the coupling constants. This is based on
dimensional analysis, which tells us how divergent a graph is at large momenta: the
“ultraviolet” (UV) divergence. (There are also infrared divergences, which occur for
physical reasons, and do not require renormalization. They occur only for massless
particles, and will be considered later.) Consider first any 1-loop 1PI graph. In mo-
R
mentum space, it has an integral over the loop momentum, dD p. It will diverge
if the integrand (before introducing Schwinger parameters) goes as p−D or slower to
infinite momentum (UV limit). In this limit we can ignore masses. If we differentiate
this graph with respect to any of the external momenta, it will become more conver-
gent, since the power of momenta in the integrand decreases. (The numerator of the
integrand is a polynomial, while each factor in the denominator depends on the loop
momentum.) With enough derivatives, it becomes convergent. This means that the
divergent part of the graph is a polynomial in the external momenta. Similar remarks
apply to any 1PI graph, if we consider the divergence coming from letting all loop mo-
menta go to infinity, known as the “superficial divergence”. Of course, the superficial
divergence is also polynomial in the coupling constants, as is the graph as a whole;
but the superficial divergence is also polynomial in the masses, since differentiation
with respect to them has the same effect as with respect to external momenta.
We can determine several more properties of this local, but divergent, contribution
to the effective action. First of all, it is Poincaré invariant and invariant under all
global symmetries of the classical action, since the effective action is invariant for all
values of D, so poles in D are also. (Consider, e.g., contour integration in D to pick
out the pole.) If we use the background field gauge (or consider only Abelian gauge
fields), then it is also gauge invariant. (However, possible exceptions are conformal
invariance and invariances involving γ−1 , since those are not invariances of the classical
action for all D.) The other property we need is that the coefficient of the divergence
364 VII. LOOPS

is real. This follows from the fact that the S-matrix satisfies unitarity, as preserved
manifestly by dimensional regularization. As discussed in the previous subsection, we
have unitarity

S † S = I, S = I + iT ⇒ i(T † − T ) = T † T

As we saw in subsection VC6, this identity actually can be applied to a single graph,
where the element of T on the left-hand side of the equation is that graph, while on
the right-hand side the summation over intermediate states gives a sum where each
term divides the graph into two parts, one for T and one for T † . We then see that
at any loop the imaginary part of a 1PI graph in T is given by “sewing” together
diagrams from lower loops. This means that any new divergence at any number of
loops must be real, since sewing doesn’t introduce new (UV) divergences: Sewed lines
are on shell, and phase space for on-shell states is always finite. (The 3-momentum
of each state is bounded by the energy, and each positive energy of an outgoing state
is bounded by the total energy of the system.)
Since Poincaré and gauge invariances, locality, and semiclassical unitarity were
used as properties to determine the classical action, this suggests that the divergent
terms in the effective action might all be of the form of terms already in the classical
action. Such a property is called (perturbative) “renormalizability”. When it holds,
all infinities can be absorbed by a redefinition of the coupling constants (and masses)
appearing in the classical action. This is physically important because all infinities
are defined only up to finite pieces: For example, in dimensional regularization, we
saw in subsection VIIA3 that the D-dependent normalization of the classical action
is ambiguous, resulting in an ambiguity in the finite pieces left over after subtrac-
tion of 1/ terms. Since we now know that superficial divergences are local, we see
that renormalization can produce arbitrary finite, local terms in the effective action,
corresponding to the divergent terms. But if the divergent terms are all of the same
form as in the classical action, all such finite terms can be absorbed by a redefintion
of the coupling constants. On the other hand, if such finite terms did not already
appear in the classical action, we would be forced to introduce them, to make the
renormalization procedure unambiguous. (Of course, we could give an unambiguous
prescription by definition, but from the point of view of another prescription this
would be the same as including the extra terms in the classical action, and using the
first prescription to arbitrarily fix the nonzero values of the couplings.) Thus, the
condition of renormalizability is necessary to prevent the appearance of an infinite
number of coupling constants, which would result in the loss of predictability. (If
divergences require only a finite number of such couplings to be added, we simply
 A. GENERAL 365

include those, to obtain a renormalizable theory with a number of couplings that is

finite, although larger than that with which we started.)
Since dimensional analysis determines the form of the divergent terms of any mo-
mentum integral, it also detemines which theories are renormalizable. By appropriate
rescaling of fields by constants, write the classical action in a form where the deriva-
tive parts of the kinetic terms have no dependence on any couplings. Then define the
couplings to be the coefficients of the interaction terms. It is easy to see that the renor-
malizable theories are the ones that include all terms which satisfy all the properties
required of the classical action (including preservation of all appropriate symmetries
that are manifestly preserved by dimensional regularization), where all couplings have
engineering dimensions that are nonnegative powers of mass: Consider first the case
where all couplings are dimensionless (and there are no masses, or at least we ignore
them at high energies for purposes of considering UV divergences). Then the theory
is renormalizable simply because there are no dimensionful parameters around, so
any local term must be of the form of those originally in the classical action. If we
now introduce couplings with positive mass dimension, then perturbatively they can
occur only to nonnegative power in any diagram, so any divergence thus produced
again has a coefficient with nonnegative mass dimension. Since the fields themselves
have positive mass dimension, there are only a finite number of such terms possible.
On the other hand, if we were to allow couplings with negative dimension, then terms
with arbitrarily high powers of such couplings would also allow arbitrarily high pow-
ers of fields, and thus lead to nonrenormalizability. (By similar arguments, theories
with only couplings of positive mass dimension, called “superrenormalizable”, can
have divergences only to a certain finite number of loops.)
In particular, in D=4 the derivative part of the kinetic term for bosons is of the
R R
form d4 x φ∂∂φ, and for fermions d4 x ψ∂ψ, so bosonic fields have dimension 1 and
fermions 3/2. That means that bosons can appear only quartically and fermions only
quadratically. More specifically, renormalizable theories can only have terms of the
form
φ, φ2, φ3, φ4 , φ∂φ, φ2∂φ, φ∂∂φ; ψ 2, ψ∂ψ; φψ 2

(where each φ can be any boson with any spin, and each ψ any fermion). There can
also be constant terms (field-independent), which we always drop, since they don’t
contribute to perturbative amplitudes (after appropriate normalization). The terms,
and their relations, are restricted by Lorentz, gauge, and internal symmetries. The
potential for scalar fields must also be bounded from below, to allow the existence of
a vacuum (state with lowest energy); otherwise nothing would be stable, continually
366 VII. LOOPS

decaying into states of lower energy (i.e, the energy of the scalars converting to other
particles): Thus, φ3 terms for scalars requires also φ4 terms.
Spin 1 can’t couple minimally to spins >1. (One way to show this is to covari-
antize the general field equation of IIB1 to Sab ∇b + k∇a , and show the commutator
algebra of this constraint, and + ..., doesn’t close unless the spin ≤1 or the ex-
ternal field strength vanishes.) Furthermore, gauge invariance for spins >1 prevents
them from having renormalizable gauge couplings in D=4: For example, we saw in
subsection IIIA4 that spin-2 (gravity) couplings include terms of the form φ∂φ∂φ.
Renormalizability therefore restricts us to spins 0, 1/2, and 1. Using Poincaré and
gauge invariance, the most general action is then of the form
n . o
1 2 α α . 1 2 1 α m
√
L = tr 8g2 F + ψ i∇α ψ̄α + 4 (∇φ) + V (φ) + [ 2 ψ (φ + 2 )ψα + h.c.]

where all group matrices are implicit: They may appear in all fields, in m, and even
in g, which has independent values for the different factors of the Yang-Mills gauge
group. Also, the matrices may differ for the same field in different terms: A (in ∇)
has different Yang-Mills representations on different scalar and spinor fields, and φ
appears with some matrix in its Yukawa coupling ψφψ. (Of course, all matrices must
be chosen consistently with gauge and global invariances.) The potential V (φ) is no
higher than quartic. Note that the Higgs mechanism is required to give nonabelian
gauge fields mass: A2 is not gauge invariant, and (∇φ)2 is the only way to dress it up
with scalars in a renormalizable way. (For the Abelian case we can use Stückelberg
fields, with ∇ = ∂ + mAT as in subsection IVA5. In the nonabelian case, introducing
scalars by a gauge transformation, as for Stückelberg, results in a nonrenormalizable
(e−iφ ∇eiφ)2 term.) If we ignore gauge invariance, the A2 term produces unitary-gauge
propagators with bad high-energy behavior (see subsection VIB3), which leads to the
same nonrenormalizable behavior in the absence of a Higgs mechanism.
Excercise VIIA5.1
Show by power counting that interacting renormalizable theories with poten-
tials that are bounded from below exist only in D≤4. Show that for D≤2
there are an infinite number of possible renormalizable terms in the action.
What are the kinds of renormalizable terms possible in D=3?
Excercise VIIA5.2
Superrenormalizable theories aren’t realistic, but they give oversimplified ex-
amples of many quantum features of field theory.
a What theories are superrenormalizable in D=3? Show that the only super-
renormalizable interaction in D=4 is (scalar) φ3 .
 A. GENERAL 367

b Let’s do power counting (dimensional analysis) for 4D φ3 theory. Write the

action in the form
Z
S = g2 dx [− 14 φ( − m2)φ + φ3 ]
1

so g 2 counts loops. (φ → gφ gives the form where g counts vertices.) What

are the dimensionless terms
Z
2 L−1
∆S = (g ) dx φn

for all n (including the vacuum bubbles n = 0; of course, L ≥ 0)? Since super-
ficial divergences are polynomial in everything (fields, momenta, couplings,
masses), this gives the maximum number of loops L(n) for a superficial di-
vergence to appear in an n-point 1PI amplitude. Make a similar analysis for
3D φ4 theory.
c Find all the divergent 1PI diagrams in 4D φ3 theory. (Hint: There are 5, or
6 if we include the trivial 1-loop graph with no vertices.)
Such global and local symmetry requirements can also be applied to the effective
action. In background-field gauges Γ is gauge invariant (see subsection VIB8), which
restricts the form of the effective potential, and even nonlocal terms. In QED charge
conjugation, in addition to switching 2-spinors of opposite charge, changes the sign
of the electromagnetic potential: Consequently, any pure-A term in Γ must be even
in A’s (“Furry’s theorem”). Such classical symmetries can be applied at the quantum
level only in the absence of “anomalies”, quantum violations (discussed in chapter
VIII below). However, even the anomalies themselves are restricted by symmetries:
Anomalies occur only in symmetries that can’t be manifestly preserved by regular-
ization, which means only conformal or axial symmetries. Thus, when the couplings
of gauge vectors are parity invariant, the axial anomaly (which violates parity by
definition) is irrelevant.

6. Infrared divergences
Although ultraviolet (UV) divergences represent a serious problem, in the sense
that they strongly restrict which theories can be useful, and require renormalization,
infrared (IR) divergences are merely a consequence of poor semantics: The definition
of the S-matrix assumes the existence of well-defined one-particle asymptotic states.
Unfortunately, these do not exist when massless particles are present, even in classical
mechanics: (1) Any particle can be accompanied by an arbitrary number of massless
368 VII. LOOPS

particles with vanishing 4-momentum, and such a collection of particles can be indis-
tinguishable from the lone particle if the (measured) quantum numbers are the same.
(These are physical states, since pa = 0 → p2 = 0.) (2) Any massless particle can
be indistinguishable from an arbitrary number of massless particles, with the same
total 4-momentum, and each with the same sign energy, if their 4-momenta are all
proportional, since then they are all traveling in the same direction at the same speed.
(This situation is not important for QED, since the photon can’t decay directly into
two photons.)
Experimentally, because detectors have finite accuracy, in the first case there can
be such “soft” particles with total energy below some small upper limit, and in the sec-
ond case there can be such “colinear” particles within some small angle of resolution.
In principle this means we should change our definition of asymptotic states accord-
ingly; in practice this is too complicated, but to any particular order in perturbation
theory only a finite number of such additional massless particles will couple. The
procedure is then to (1) infrared regularize the S-matrix amplitudes (by dimensional
regularization, or introducing masses for all particles, or keeping massless particles
off-shell); (2) calculate probabilities/cross sections, including contributions from soft
and colinear particles, as a function of some upper limit on their energy/angle (rep-
resenting the experimental accuracy); and (3) remove the regularization. No infrared
renormalization is necessary. (Examples will be given in later sections. Of course, for
total cross sections, all energies and angles are integrated over anyway.) In general,
such a procedure must be applied to both initial and final states (the“Kinoshita-Lee-
Nauenberg theorem”), but in QED (as opposed to QCD) it is sufficient to treat only
the final ones in cases of physical interest.
The reason why infinities appear in cross sections if we ignore this careful pre-
scription, and in S-matrix elements in any case, is the long range of forces mediated
by massless particles. A cross section, although it represents a probability, is nor-
malized in such a way that it is an area, representing the effective cross-sectional
area of a particle being targeted by another particle. The range of the interaction
sets the scale of this area; this is related to the mass of the particle mediating the
interaction. Since massless particles produce infinite-range forces, the result is infi-
nite cross sections. One might expect that these infinities would appear only in total
cross sections, where the momenta of particles in the final state are integrated over.
However, by the optical theorem, this total cross section is given by the imaginary
part of an S-matrix amplitude, which thus must also have this infinity.
 A. GENERAL 369

The fact that these infrared divergences are physical also follows from the fact
that these kinematic situations occur in classical mechanics: In subsection VC8 we
saw that physical singularities occur in S-matrices for momenta that are allowed
classically.

REFERENCES
1 P.A.M. Dirac, Proc. Cambridge Phil. Soc. 30 (1934) 150;
W. Heisenberg, Z. Phys. 92 (1934) 692;
V. Weisskopf, Kon. Dan. Vid. Sel. Mat.-fys. Medd. XIV (1936) 1:
charge renormalization.
2 W. Pauli and M. Fierz, Nuo. Cim. 15 (1938) 167;
H. Kramers, Nuo. Cim. 15 (1938) 134:
mass renormalization.
3 Dyson, loc. cit. (VC):
renormalization to all loops.
4 A. Salam, Phys. Rev. 82 (1951) 217, 84 (1951) 426:
completed Dyson’s proof.
5 Bogoliubov and Shirkov, loc. cit.:
general approach to renormalization.
6 G. ’t Hooft and M. Veltman, Nucl. Phys. B44 (1972) 189;
C.G. Bollini and J.J. Giambiagi, Nuo. Cim. 12B (1972) 20;
J.F. Ashmore, Lett. Nuo. Cimento 4 (1972) 289;
G.M. Cicuta and E. Montaldi, Lett. Nuo. Cim. 4 (1972) 329:
dimensional regularization.
7 G. ’t Hooft and M. Veltman, Diagrammar, in Particle interactions at very high energies,
proc. 2nd Summer Institute on Particle Interactions at Very High Energies, Louvain,
Belgium, Aug 12-25, 1973, eds. D. Speiser, F. Halzen, J. Weyers (Plenum, 1974) part B,
p. 177; and in Under the spell of the gauge principle, ed. G. ’t Hooft (World Scientific,
1994) p. 28:
dimensional renormalization.
8 Siegel, loc. cit. (VIC, ref. 10, next to last), W. Siegel, Phys. Lett. 94B (1980) 37;
Gates, Grisaru, Roček, and Siegel, loc. cit.:
regularization by dimensional reduction.
9 J. Polchinski, Nucl. Phys. B231 (1984) 269:
simplest proof of renormalization; uses renormalization group, doesn’t use geometry of
graphs (“forests” or “skeletons”, etc.) or “Weinberg’s theorem”; not yet simply gener-
alized to gauge theories.
10 Feynman, loc. cit. (VB, ref. 4):
Feynman parameters.
11 E.T. Whittaker and G.N. Watson, A course of modern analysis, 4th ed. (Cambridge
University, 1927) p. 235:
detailed discussion of Γ functions.
12 A.A. Vladimirov, Theor. Math. Phys. 36 (1978) 732;
W.A. Bardeen, A.J. Buras, D.W. Duke, and T. Muta, Phys. Rev. D18 (1978) 3998:
MS scheme.
370 VII. LOOPS

13 K.G. Chetyrkin, A.L. Kataev, and F.V. Tkachov, Nucl. Phys. B174 (1980) 345:
G scheme.
14 E. Feenberg, Phys. Rev. 40 (1932) 40;
N. Bohr, R.E. Peierls, and G. Placzek, 1947 manuscript, in Niels Bohr collected works,
v.9, ed. S.R. Peierls (North-Holland, 1986) p. 487:
optical theorem.
15 W.H. Furry, Phys. Rev. 51 (1937) 125.
16 F. Bloch and A. Nordsieck, Phys. Rev. 52 (1937) 54;
D.R. Yennie, S.C. Frautschi, and H. Suura, Ann. Phys. 13 (1961) 379:
infrared divergences.
17 T. Kinoshita, J. Math. Phys. 3 (1962) 650;
T.D. Lee and M. Nauenberg, Phys. Rev. 133B (1964) 1549.
18 R. Gastmans and R. Meuldermans, Nucl. Phys. B63 (1973) 277;
W.J. Marciano and A. Sirlin, Nucl. Phys. B88 (1975) 86:
earliest applications of dimensional regularization to infrared divergences.
 B. EXAMPLES 371

........................... ...........................
........................... B. EXAMPLES ...........................

We now give some explicit examples of the evaluation of S-matrices and contri-
butions to the effective action — momentum integration, regularization, and renor-
malization — and some examples of their application.

1. Tadpoles

...

The simplest examples of dimensional regularization are one-loop “tadpoles”,

graphs with only one external line. By the Schwinger parameter method described
in subsection VIIA2, we find
Z Z ∞
1
dτ τ −D/2e−(τ m +x /τ )/2
2 2
A1 (x, m ) = dk e 1 2
2 ik·x
2
=
2 (k + m ) 0

Further evaluation requires Taylor expansion in x (which we’ll need anyway to eval-
uate a specific integral of k...k/(k 2 + m2)):
X
∞
Γ(1 − D2 − n)
A1 = 1
(− 1 2 n
2 x )
n=0
n!
( 12 m2)1−D/2−n

The mass dependence, as well as the argument of the Γ function, are as expected by
R
dimensional analysis: dD k k 2n /k 2 is ultraviolet divergent (large k) for D ≥ 2(1− n),
and infrared divergent (small k) in the limit m →0 for D ≤ 2(1 − n). The ultraviolet
divergence is reflected in Γ(z), which has poles at the nonpositive integers.
To analyze the massless case, we evaluate the τ integral for D < 2(1 − n) and
m > 0, where it is finite and well-defined, analytically continue to the region Re D >
2(1−n) (but not exactly at the points where D is an even integer), take the limit m →
0 there, and finally analytically continue this vanishing result to all D. Therefore, all
massless tadpoles can be taken to vanish in dimensional regularization:
Z
ka ...kb
dk 1 2 = 0
2k

or more generally Z
ka ...kb
dk =0
( 12 k 2 )a
372 VII. LOOPS

This includes negative a, particularly integrals of polynomials of momenta. Such

an integral can result from “measure factors”, as discussed in subsections VA2 and
VC1: For example, if an auxiliary field appears in the action with its quadratic term
multiplied by a function of other fields, then functionally integrating it out of the
action results in a functional determinant (in addition to replacing it in the classical
action by the solution to its field equation). This is represented in terms of Feynman
graphs as one-loop diagrams whose propagators are all those of the auxiliary field,
namely 1. The result is then regularized as
Z Z
dx 1 ∼ δ(0) → dk 1 = 0

consistent with the fact that such factors would cancel corresponding factors we should
include in the functional integration measure. (In other words, since we can always
arrange to have all δ(0) factors cancel, we ignore them.)
On the other hand, massive tadpoles contribute both divergent and finite pieces
under minimal subtraction: For example, for D = 4 − 2,
Z
1
A1(0, m ) = dk 1 2
2
2
= Γ(1 − D2 )( 12 m2)D/2−1
2 (k + m )

= − 12 m2 { 1 + [−γ + 1 − ln( 12 m2 )]}

(see excercise VIIA2.3b), using A− = e− ln(A) . The γ can be killed by using an MS or
G scheme (see subsection VIIA3): At 1-loop order any version of those schemes has the
effect of just canceling the γ (but differences appear at 2 loops: see subsection VIIB7
below). To include the µ dependence of the coupling, we just replace everywhere (see
also subsection VIIA3)  2
m
ln( 2 m ) → ln
1 2
µ2
(and similarly for any momentum factors such as ln( 12 p2 ) that might appear more
generally); effectively we are using units 12 µ2 = 1. Note that we are not allowed to
Taylor expand in m: Doing so before integration would give an incorrect result; after
integration it’s impossible. Similar remarks apply to the exponential eik·x in A1 if we
interpret it as the definition by Fourier transformation of the propagator in position
space.
Excercise VIIB1.1
Find the 2D massless propagator in position space by Fourier transformation.
(But don’t Taylor expand in x.) Note that this Fourier transform is infinite,
and requires “renormalization” (of a constant of integration). Compare this
 B. EXAMPLES 373

with the result obtained by solving the integral form of the Klein-Gordon
(Laplace) equation (i.e., Gauss’ law in D=2).

Two-loop tadpole integrals are not much more difficult if one line is massive, or
two are massive with the same mass. (Again, tadpoles with only massless lines can be
taken to vanish in dimensional regularization.) If two propagators are massless, then
they can be treated first as a one-loop propagator graph: By dimensional analysis, the
result of that one-loop subintegral must be a power of the momentum squared. (The
explicit result will be calculated in the following subsection.) We therefore consider
more general one-loop tadpole integrals with more complicated propagators that may
result from subintegrations in a higher-loop graph. For example, we consider
Z
Γ(a)
Â1 (a, x, m ) = dk eik·x 1 2
2
[ 2 (k + m2)]a
Using the definition of the Γ function, we can write
Z ∞
Γ(a) a−1 −τ (k2 +m2 )/2
1 = dτ τ e
[ 2 (k 2 + m2 )]a 0

Performing the resultant Gaussian momentum integration and Taylor expanding in

x, we easily find
X
∞
Γ(a − D2 − n)
Â1 (a, x, m ) =
2 1
n!
1 2 n
(− 2 x ) 1 2 a−D/2−n
n=0
(2m )

A more complicated example is

Z
Γ(a) Γ(b)
Ǎ1(a, b, m ) =
2
dk
1 2 a 1
( 2 k ) [ 2 (k + m2)]b
2

Z ∞ Z
dk e−[τ1 k +τ2 (k +m )]/2
a−1 b−1 2 2 2
= dτ1 dτ2 τ1 τ2
0
Z ∞
dτ1 dτ2 τ1a−1 τ2b−1 (τ1 + τ2 )−D/2 e−τ2 m /2
2
=
0
We then introduce a scaling parameter λ (also described in VIIA2), scaling τi = λαi
in the insertion
Z ∞ Z ∞
1= dλ δ(λ − τ1 − τ2 ) = dλ λ−1 δ(1 − α1 − α2)
0 0
374 VII. LOOPS

and integrating the δ over α2 to get α2 = 1 − α1 . This gives (with α1 = α)

Z 1 Z ∞
dλ λa+b−D/2−1 e−λ(1−α)m /2
2
Ǎ1 (a, b, m ) =
2
dα α (1 − α)
a−1 b−1
0 0

Γ(a + b − D2 )
= 1 2 a+b−D/2 B(a, D2 − a)
(2m )

When two of the propagators in the two-loop tadpole graph have the same non-
vanishing mass, we consider directly the two-loop integral
Z
Γ(a) Γ(b) Γ(c)
A1,2(a, b, c, m ) = dk1 dk2 1
2
1 2 1 2
[ 2 (k1 + k2 )2 ]a [ 2 (k1 + m2)]b [ 2 (k2 + m2)]c

(This integral also represents the physically less interesting 2-loop “vacuum bubble”:
no external lines, and thus field independent.) Introducing the Schwinger parameters
and performing the momentum integration, we find
Z ∞
d3 τ τ1a−1τ2b−1 τ3c−1 [τ2τ3 + τ1 (τ2 + τ3 )]−D/2e−(τ2 +τ3 )m /2
2

Since τ1 does not appear in the exponential we integrate over it first directly, using
the second integral form for the Beta function, from excercise VIIA2.2c. Then τ2 and
τ3 can be handled by introducing a scaling parameter for them only, leading to the
previous types of integrals. The result is then
Γ(a + b + c − D)
A1,2(a, b, c, m2) = B(a + b − D
,a +c− D
)B(a, D2 − a)
( 12 m2)a+b+c−D 2 2

2. Effective potential
A propagator in an external field represents a certain class of Feynman tree dia-
grams. Thus, some tree graphs can be described by quantum mechanics. (In principle
this means we can start from classical mechanics and first-quantize, by either operator
or path-integral methods. However, as we’ll see in chapter XII, in practice we save
some effort if we start directly with the quantum mechanics.) If we take the ends
of such a propagator and sew them together, we can describe arbitrary 1PI 1-loop
graphs by the background field method. While tree graphs describe classical field
theory, one-loop graphs contain many of the important quantum properties, partly
because they are the lowest-order quantum correction, and partly because they are
associated with the functional determinant part of the (second-quantized) path inte-
gral. (In terms of the exponent, classical is the only negative power in h̄, 1-loop is
h̄-independent, and higher loops are positive powers.)
 B. EXAMPLES 375

In quantum mechanics, the expansion in h̄ is an expansion in derivatives (since

it appears only as pa = −ih̄∂a ). In terms of the contribution of one-loop graphs to
the effective action, this means an expansion in the number of derivatives acting on
the fields. This definition can be applied in general in quantum field theory, without
reference to quantum mechanics. However, the simplest one-loop calculations of this
expansion are most easily expressed in quantum mechanical terms. In practice, this
means expanding the external fields in x about some fixed point, expanding the
exponentiated (by a Schwinger parameter) propagator about the part Gaussian in p
and x, and using any of the usual methods to exactly evaluate the matrix element of
a polynomial times a Gaussian.
Since we generally want arbitrary orders in a field and some of its lower derivatives
for this method to have any advantage over the usual diagrammatic methods, in this
approach one generally cuts off the expansion at the approximation that gives just
the Gaussian. This means we can keep up to two derivatives of an external scalar, but
only a constant field strength for an external gauge vector. (See subsection VIB1.)
The simplest, and most useful, example is a constant scalar field. The part of the
effective action that consists of only scalars without derivatives is called the “effective
potential”, since it generalizes the potential term of the classical action. This potential
determines the quantum corrections to spontaneous symmetry breaking and the Higgs
effect, and this is important for describing mass generation for all spins.
Consider a complex scalar running around a loop, under the influence of an ex-
ternal real scalar. The Lagrangian is

L = ψ*[ 12 (− + m2) + φ]ψ + Lφ

where the form of Lφ won’t be important for calculating the ψ loop. A constant
external scalar field is effectively the same as a mass term, modifying m2 → m2 + 2φ.
Thus the effective potential in this case can be evaluated by summing tadpoles:
X∞ Z
V =− n
1 n n
(−1) φ dp [ 12 (p2 + m2)]−n
n=1

for our complex scalar; for a real scalar running around the loop there would be an
extra factor of 1/2. We can integrate before summing:
X∞
Γ(n − D
) 1 2 −n+D/2
V =− (−1)nφn 2
(2m )
n=1
n!
Using the identities (from Taylor expansion in a/b, and Γ(z + 1) = zΓ(z))
X∞    
x x n x−n x Γ(x + 1) Γ(n − x)
(a + b) = a b , = = (−1)n
n=0
n n n!Γ(x + 1 − n) n!Γ(−x)
376 VII. LOOPS

we have
 
V = −Γ(− D2 ) ( 12 m2 + φ)D/2 − ( 12 m2)D/2
We can also integrate after summing: Using the identities
Z a Z a  x n
1X X
∞ ∞
du
ln(a + b) − ln b = = du (−1)n =− 1
n
(−1)nan b−n
0 u + b 0 b b
Z a Z a Z ∞ n=0
Z ∞ n=1
du dτ

= du dτ e−τ (x+b) = − e−τ (a+b) − e−τ b
0 u+b 0 0 0 τ
we have Z ∞ Z  
dτ
dp e−τ [φ+(p +m )/2] − e−τ (p +m )/2
2 2 2 2
V =−
0 τ
which gives the same result.
For D=4, we find (after subtracting divergent counterterms, and some correspond-
ing finite pieces, corresponding to a MOM type of subtraction)
 
1 1 2 2 2φ
V = 2 ( 2 m + φ) ln 1 + 2
m
Since this modifies the classical potential, it demonstrates that quantum effects can
generate spontaneous symmetry breaking where there was none classically, or vice
versa (the “Coleman-Weinberg mechanism”).
For more complicated cases we need a more general procedure: The basic idea
is that any Gaussian integral gives a (inverse) determinant, of which we must take
(minus) the logarithm for the effective action, and we use ln det = tr ln. (The trace
includes integration over x or p.) After subtracting out the field-independent part
(vacuum bubble), this gives an expression as above: For a general kinetic operator
H = H0 + ... (generally H0 = 12 (p2 + m2)), we want
Z ∞ Z
 −1 −1
 dτ
Γ = − tr ln(H ) − tr ln(H0 ) = − dx hx|e−τ H − e−τ H0 |xi
0 τ
H (and H0 ) is now treated as an operator, in terms of the coordinate operator X
and momentum operator P , and X|xi = x|xi. External fields depend on X, but are
Taylor expanded about x: e.g.,
φ(X) = φ(x) + (X − x) · ∂φ(x) + ...
We then can use translation invariance to write
hx|e−τ H[P,X−x,φ(x)]|xi = h0|e−τ H[P,X,φ(x)]|0i
When H is quadratic in P and X, we can use (see excercise VA2.4)
s
∂ 2 (−S) −S
hx|e−τ H |yi = det e
∂x∂y
where S is the classical “action” corresponding to the “Hamiltonian” H. (Further
examples will be given in subsection VIIIB1.)
 B. EXAMPLES 377

3. Dimensional transmutation
The 2D version of the CP(n) model described in subsection IVA2 is an interesting
model in that it demonstrates generation of bound states at the one-loop level. Its
Lagrangian is:
L = 12 |∇φ|2 + Λ(|φ|2 − g12 )

where g is now dimensionless. For the effective potential for the Lagrange multiplier
Λ from a φ loop, we find (modifying the calculation of the previous subsection for
D=2)    
Λ
V1 = −Λ ln 1 2 − 1
2µ
after including the renormalization mass scale µ to make the argument of the loga-
rithm dimensionless, and the coupling dimensionless in all dimensions.
Now the coupling can be absorbed into the definition of this scale: Adding to the
classical term V0 = −Λ/g 2 , the total effective potential for Λ up to one loop is
       
1 Λ Λ
V = −Λ 2 + ln 1 2 − 1 = −Λ ln 1 2 − 1
g 2µ 2M

where M is the “renormalization group invariant mass scale”:

M 2 = µ2 e−1/g
2

Since this was the only place the coupling g appeared in the action, the mass scale M
has now replaced it completely. This replacement of a dimensionless coupling (g) with
a dimensionful one (M) is called “dimensional transmutation”. It is also a common
feature of quantum high-energy behavior (see below); its importance at low energies
depends on whether the classical theory already has dimensionful parameters (like
masses).
Varying the effective potential to find the minimum, which we identify as the
(quantum) vacuum value of the field Λ,
 
hΛi
ln 1 2 = 0 ⇒ hΛi = 12 M 2
2M

Because Λ has a vacuum value, φ now has a mass (as seen by expanding Λ about
its vacuum value). Furthermore, since Λ now has more than just linear terms in
the effective action, it is no longer a Lagrange multiplier. In fact, by calculating a
massive φ loop with two external Λ’s, we see that Λ is now a massive physical scalar
also. Without a Lagrange multiplier, φ is now unconstrained, so it has an additional
physical degree of freedom. This leads to a restoration of the spontaneously broken
378 VII. LOOPS

U(N) symmetry. This is related to φ gaining mass, since we no longer have Goldstone
bosons associated with the symmetry breaking. Finally, if we calculate a massive φ
loop with two external gauge vectors, we see that at low energies there is an F 2 term,
so A is now a physical, massive vector instead of an auxiliary field.

4. Massless propagators

For the massless one-loop propagator corrections, we also introduce a scaling

parameter to convert to Feynman parameters (see the examples of subsection VIIB1,
or the general method in subsection VIIA2), with the result
Z
1
A2(x, p ) = dk eik·x 1
2

2 (k + 2 p) 2 (k − 2 p)
1 21 1 2
Z ∞ Z 1
= dλ λ1−D/2 dα1 dα2 δ(1 − α1 − α2 )×
0 0
× exp{−λ 18 p2 [1 − (α1 − α2 )2 ] − i 12 (α1 − α2 )p · x − λ−1 21 x2 }

Making the change of variables

α1 = 12 (1 + β), α2 = 12 (1 − β)

the amplitude takes the form

Z 1 Z ∞
−iβp·x/2
A2 = 1 iβp·x/2
dβ 2 (e +e ) dλ λ1−D/2exp[−λ 18 (1 − β 2)p2 − λ−1 12 x2]
0 0

The integrals can be simplified if we make use of gauge invariance: For example,
↔
the electromagnetic current for a complex scalar is of the form φ* ∂ φ, so the gauge
field couples to the difference of the momenta of the two scalar lines, which is 2k for
the above as applied to the scalar-loop correction to the photon propagator. On the
other hand gauge invariance, or equivalently current conservation, says that such a
vertex factor should give a vanishing contribution when contracted with the external
momentum, which is p in that case. Checking this explicitly, we do in fact find
Z Z  
k·p 1 1
dk 1 1 2 = dk 1 1 2 − 1 1 2 = 0
2 (k + 2 p) 2 (k − 2 p) 2 (k − 2 p)
1 21
2 (k + 2 p)

(even with an arbitrary additional polynomial factor in the numerator), using the
facts that the integral of the sum is the sum of the integrals when regularized, and
 B. EXAMPLES 379

that massless tadpoles vanish. (This also tells us that replacing the numerator k · p
with k 2 + 14 p2 gives 0. Furthermore, without an extra numerator factor the integral
vanishes by antisymmetry under k → −k.) Thus, if x is proportional to p in A2, the
only contribution is from the x = 0 term in the Taylor expansion. This implies that
the dependence on x is only through the combination

u = (p · x)2 − p2 x2

so we can evaluate the integral by either of the substitutions

√
x2 → 0, p · x → u or p · x → 0, x2 → −u/p2

We’ll consider now the latter choice. (The former gives the same result: See the
excercise below.) Again, since we need to Taylor expand in x anyway to find the
result for a particular numerator, we expand and perform the λ integration:
X∞   Z 1
u n 1 2 n+D/2−2
A2 = 1
n! 2
(8p ) Γ(2 − 2 − n)
D
dβ (1 − β 2)n+D/2−2
n=0
2p 0

Performing the change of variables β 2 = γ to convert the remaining integral to a Beta

function, and using the identities
√
Γ( 12 ) = π, Γ(z)Γ(1 − z) = π csc(πz)

(see the excercises in subsection VIIA2), the final result is

X
∞
1
A2 = − 12 π 3/2csc(D π2 )( 18 p2 )D/2−2 { 1 [p2 x2 − (p · x)2 ]}n
n=0
n!Γ(n + D2 − 12 ) 16
From the csc factor we see the integral is divergent for all even D: These are ultraviolet
divergences for D ≥ 4 and infrared ones for D ≤ 4; dimensional regularization does
not carefully distinguish between the two, although the difference can usually be told
by examining momentum dependence (here from the exponent D/2 − 2). Also notice
that the two can be mixed up by the conversion to Feynman parameters.
Excercise VIIB4.1
Evaluate the general massless one-loop propagator correction using x2 → 0,
√
p · x → u.
a Show it gives the same result as p · x → 0, x2 → −u/p2 by using the Γ and B
identities in subsection VIIA2.
b Show it can also be written as (for convenience of expansion about D=4)
X∞
Γ(n + D2 − 1) 1 2 2
A2 = ( 12 p2 )D/2−2 Γ( D2 − 1)Γ(2 − D
) { 4 [p x − (p · x)2 ]}n
2
n=0
n!Γ(2n + D − 2)
380 VII. LOOPS

As discussed in subsection VIIB1, sometimes certain subdiagrams of higher-loop

diagrams can be evaluated explicitly, particularly propagator corrections that them-
selves involve only massless propagators. Furthermore, such a formula might be used
recursively in appropriate diagrams. For example, a higher-loop diagram that is itself
a propagator correction might reduce, as a final integration, to something of the form
Z
Γ(a) Γ(b) Γ(a + b − D2 ) D
dk 1 2 a 1 = B( 2 − a, D2 − b)
( 2 k ) [ 2 (k + p)2 ]b ( 12 p2 )a+b−D/2
again using the above methods, finding similar integrals to the previous.
Excercise VIIB4.2
Let’s examine this integral more carefully.
a Evaluate it in two different ways: first, by the method used above; second, by
Fourier transforming each factor using
Z
Γ(a) Γ( D − a)
dk eik·x 1 2 a = 1 22 D/2−a
(2k ) (2x )
(derive this also) and its inverse, simply multiplying the resulting factors in
x space, and inverse transforming.
b Show that the MS scheme cancels γ’s and ζ(2)’s in iterated massless propa-
gator corrections to all orders in  by examining
Z
D 1
Γ( 2 ) dk 1 2 n +L  1
(2k ) 1 1 [ 2 (k + p)2 ]n2 +L2 
where Li are the numbers of loops in the propagator subgraphs (show this by
dimensional analysis) and ni are other integers. Show the G scheme does the
same.
Excercise VIIB4.3
Calculate the “phase space” for n massless particles
Z "Y n
#
X
n
!
dD−1 pi
VP = D/2−1
(2π)D/2δ D p − pi
(2π) ωi

where p is the total momentum of the n particles, by using the optical theorem:
a Consider the scalar graph with n massless propagators connecting 2 vertices.
Show, both by induction in the number (n− 1) of loops, and by Fourier trans-
formation (as in the previous problem), that this graph (for distinguishable
particles) gives
[Γ( D2 − 1)]n Γ[n − (n − 1) D2 ]
Γ[n( D2 − 1)] ( 12 p2 )n−(n−1)D/2
 B. EXAMPLES 381

b Wick rotate back to Minkowski space (p2 < 0) and take the imaginary part
to obtain the result for continuous real D > 2
[Γ( D2 − 1)]n
VP = 2π (− 1 p2 )−n+(n−1)D/2
Γ[n( D2 − 1)]Γ[(n − 1)( D2 − 1)] 2

which simplifies in D=4 to

1
VP = 2π (− 1 p2 )n−2
(n − 1)!(n − 2)! 2

(Hint: ( 12 p2 − i)r = (− 12 p2 )r e−iπr .)

5. Massive propagators
Another way to distinguish infrared divergences is by introducing masses (being
careful not to break any invariances, or restoring them in the massless limit). For
example, we again evaluate the one-loop propagator correction, without numerator
factors, but with different masses on the internal propagators. By the same steps as
before, the Feynman parameter integral is
Z
1
Â2 (p , m1, m2) = dk 1
2 2 2

2 [(k + 2 p) + m1 ] 2 [(k − 2 p) + m2 ]
1 2 2 1 1 2 2

Z 1
= Γ(2 − D 1
)
2 2
dβ B D/2−2, B = 18 p2 (1 − β 2) + 14 (m21 + m22) + 14 β(m21 − m22 )
−1

Now the β integral is harder for all D, but the masses eliminate the IR divergences
(and the UV divergences are already explicit in the Γ), so we immediately expand
about D = 4 − 2: Z 1
Â2 ≈ Γ() 12 dβ (1 −  ln B)
−1

We then use integration by parts

Z 1 Z 1
d
dβ ln B = (β ln B)|−1 −
1
dβ β ln B
−1 −1 dβ
√
−b ± b2 − 4ac m2 − m22 ± 2λ12
B = aβ + bβ + c = a(β − β+)(β − β−),
2
β± = = 1
2a p2
d β β β+ β−
⇒ β ln B = + =2+ +
dβ β − β+ β − β− β − β+ β − β−
in terms of λ12 (s) of subsection IA4 for s = −p2 . Note that in Euclidean space
q q
2λ12 = (p2 + m21 − m22)2 + 4m22 p2 = (p2 + m22 − m21)2 + 4m21 p2 ≥ p2 + |m21 − m22|
382 VII. LOOPS

⇒ ±β± ≥ 1
where the strict inequality holds for both masses nonvanishing. The integrals then
take the simple form
Z 1      
β+ β− β+ − 1 β− − 1
dβ + = β+ln + β−ln
−1 β − β+ β − β− β+ + 1 β− + 1
Putting it all together,
    
1 β+ − 1 β− − 1
Â2 = Γ(1 + ) − ln( 2 m1m2 ) + 2 + 2 β+ln
1 1 1
+ 2 β−ln
 β+ + 1 β− + 1
(We can cancel the Γ(1 + ) by nonminimal subtraction.) By analytic continuation
from Euclidean space, taking p2 from positive to negative along the real axis, we see
there is no ambiguity at p2 = 0 or −(m1 − m2)2 , and Â2 remains real until we reach
p2 = −(m1 + m2 )2, where it gets an imaginary part (whose sign is determined by
(m1 + m2 )2 → (m1 + m2)2 − i), corresponding to the possibility of real 2-particle
intermediate states.
Excercise VIIB5.1
Let’s consider some special cases:
a Show for equal masses m1 = m2 = m that this result simplifies to
  
1 β−1
Â2(p , m , m ) = Γ(1 + )
2 2 2
− ln( 2 m ) + 2 + β ln
1 2
 β+1
s
p2 + 4m2
β=
p2
b Consider the case with one internal particle massless, m1 = m, m2 = 0, and
find
  2 
1 p2 + m2 p + m2
Â2 (p , m , 0) = Γ(1 + )
2 2
− ln( 2 m ) + 2 −
1 2
ln
 p2 m2

c Show both these results agree with the previously obtained massless result in
the limit m → 0. However, note that both these cases, unlike the massless
case, are IR convergent at p2 = −(m1 + m2 )2.
Excercise VIIB5.2
Find the phase space for 2 massive particles, again using the optical theorem
(as in excercise VIIB4.3). The calculation is easier if one takes the imaginary
part before performing the Feynman parameter integration: Show the result
is then Z β−
1
VP = π D dβ (−B)D/2−2
Γ( 2 − 1) β+
 B. EXAMPLES 383

which simplifies in D=4 to

λ12
VP = 2π
− 12 p2
In particular, show from the explicit parameter integral expression for the
propagator that the only cut is at −p2 ≥ (m1 + m2)2 , as expected from the
optical theorem. Evaluate in general D for the massless case and show it
agrees with excercise VIIB4.3.
In subsection VIIA3 we considered the application of the MOM subtraction
scheme to propagator corrections. We assumed the propagator corrections were Tay-
lor expandable in the classical kinetic operator. From the above explicit expression
for the 1-loop correction in scalar theories, we see this is possible except near the
branch point at p2 = −(m1 + m2)2 , i.e., when the external particle (whose propaga-
tor we’re correcting) has a mass equal to the sum of the internal ones. To analyze
this more carefully, let’s recalculate the propagator correction, performing the Taylor
expansion before evaluating the integrals. We consider the case with one vanishing
mass, m1 = m, m2 = 0, to generate an IR divergence. Assuming the external mass
is also m, we expand around the branch point in p2 + m2 . The Feynman parameter
integral is then, to linear order in p2 + m2, in terms of α = 12 (1 + β),
Z   −
1
1 − α p2 + m2
Â2 (p , m , 0) = Γ()
2 2
dα 1 2 2
2m α 1+
0 α m2
Z 1  2 2

1 −2 −1−2 p + m
≈ Γ(1 + )( 12 m2)− dα α − (1 − α)α
0  m2
   2 
1 2 − 1 1 1 1 p + m2
= Γ(1 + )( 2 m ) − −
 1 − 2 −2 1 − 2 m2
    2 
1 2 − 1 1 p + m2
≈ Γ(1 + )( 2 m ) +2 + 2 1
+2
U V IR m2
where we have distinguished the UV divergence (in the λ integral for  ≥ 0) from the
IR one (in the α integral for  ≤ 0). After including the ( 12 µ2 ) in the coupling, the
( 12 m2)− converts each 1/ into a 1/ − ln(m2/µ2 ). (Of course, we can choose µ = m
for convenience.) Note that this infrared divergence was a consequence of trying to
Taylor expand about a branch point due to a massless particle.
Excercise VIIB5.3
Do MOM subtraction for external mass M = m1 + m2, with neither internal
mass vanishing, and show there is no divergence other than the UV divergence
of the minimal scheme.
384 VII. LOOPS

Later, we will encounter propagator corrections in gauge theories with massive

internal lines, and with various numerators. Here, we examine these purely from the
point of view of the integrals. First, consider
Z
ka
Aa = dk 1
2 [(k − 2 p) + m1 ] 2 [(k + 2 p) + m2 ]
1 2 2 1 1 2 2

Since p is the only external momentum for a propagator, by Lorentz invariance we

have
1
Aa = pa 2 p · A
p
so it is sufficient to evaluate the integral of p · A. In analogy with the earlier massless
expression, we look at
1 1 k · p + 12 (m22 − m21)
− =
2 [(k − 2 p) + m1 ] 2 [(k − 2 p) + m1 ] 2 [(k + 2 p) + m2 ]
1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2
2 [(k + 2 p) + m2 ]

from which we find

m21 − m22
Aa = pa [Â2(p2 , m21, m22) − Â2(0, m21 , m22)]
2p2
in terms of our result Â2 above for the integral without numerator.
As a more complicated (but important) example, we examine
Z
ka kb
Aab = dk 1
2 [(k + 2 p) + m ] 2 [(k − 2 p) + m ]
1 2 2 1 1 2 2

Following our procedure of the previous example, we note

Z
(p · k)k
dk 1
2 [(k + 2 p) + m ] 2 [(k − 2 p) + m ]
1 2 2 1 1 2 2

Z
k k
= dk 1 − 1
2 [(k − 2 p) + m ]
1 2 2 1 2 2
2 [(k + 2 p) + m ]
Z Z
k + 12 p k − 12 p 1
= dk 1 2 2
− 1 2 2
= p dk 1 2 2
2 (k + m ) 2 (k + m ) 2 (k + m )
Thus transversality again determines the amplitude in terms of a scalar:
Z
ka kb ηab
Âab = dk 1 − 1 2
2 [(k + 2 p) + m ] 2 [(k − 2 p) + m ]
1 2 2 1 1 2 2 2
2 (k + m )

= (ηab p2 − pa pb )Ǎ(p2 , m2)

(This amplitude actually will be more useful than Aab .) We also have the identity
Z 1 2 2 1 2
2 (k + m ) + 8 p
dk 1
2 [(k + 2 p) + m ] 2 [(k − 2 p) + m ]
1 2 2 1 1 2 2
 B. EXAMPLES 385

Z Z
1 1 1 1
= dk 1 + 1 = dk
2 [(k − 2 p) + m ]
2 1 2 2 1 2 2 1 2
2 [(k + 2 p) + m ] 2 (k + m2 )
Taking the trace of the previous expression,
Z
k2 D
(D − 1)p Ǎ(p ) = dk 1
2 2
− 1 2
2 [(k + 2 p) + m ] 2 [(k − 2 p) + m ]
1 2 2 1 1 2 2 2
2 (k + m )
Z Z
1 1
= −( 4 p + m ) dk 1
1 2 2
− (D − 2) dk
2 [(k + 2 p) + m ] 2 [(k − 2 p) + m ]
1 2 2 1 1 2 2 1 2 2
2 (k + m )

= −( 14 p2 + m2)Â2 (p2 , m2, m2) − (D − 2)A1 (0, m2)

in terms of the φ3 propagator and tadpole graphs evaluated earlier. This result can
be reorganized if we make use of the p = 0 case:

0 = −m2Â2 (0, m2 , m2) − (D − 2)A1 (0, m2 )

(which also follows easily from the earlier explicit expression for Â1 (a, 0, m2)). We
then find

2 Â2 (p , m , m ) − Â2(0, m2 , m2)

2 2 2
Ǎ = − 4(D−1)
1
Â2 (p2 , m2, m2) − 1
D−1
m
p2

Excercise VIIB5.4
Check that these results are consistent in the massless limit with the expres-
sions obtained in the previous subsection, by relating the first two terms in
A2 (x, p2) for arbitrary D.
Excercise VIIB5.5
Calculate the one-loop propagator corrections for Λ and A in the 2D CP(n)
model.

6. Renormalization group
An interesting, useful, and simple application of the propagator correction is to
study the high-energy behavior of coupling constants. For example, we have seen that,
by a change in normalization of gauge fields A → A/g, gauge couplings can be moved
from the covariant derivative to the kinetic term: ∇ = ∂ +igA → ∂ +iA, L0 = 18 (∂A+
igAA)2 → 8g12 (∂A + iAA)2. Thus, quantum corrections to gauge couplings can be
found from just the propagator (kinetic-operator) correction. A simpler example is a
scalar field; a φ4 self-interaction has a dimensionless coupling in D=4, like Yang-Mills.
However, unlike Yang-Mills, this model has no cubic coupling, and thus no 1-loop
propagator correction. Furthermore, in Yang-Mills the one-loop propagator correction
386 VII. LOOPS

contribution to the effective action gives a multiloop contribution to the propagator

itself, from the expansion of 1/(K + A). This corresponds to the graph consisting of
a long string of these corrections connected by free propagators. There is a 1-loop
4-point correction in φ4 theory, and this graph resembles a propagator correction,
but with two external lines at each vertex instead of one. Such corrections can also
be strung together, resembling the Yang-Mills string, but with no free propagators
inserted. Since all the intermediate states in this graph are 2-particle, it is 1PI, so the
effect of this string is not contained in just the 1-loop effective action, even though it
is an iteration of a 1-loop effect.
This difficulty can be avoided by introducing the φ4 interaction through an aux-
iliary field, just as it appears in supersymmetric theories (see subsection IVC2):

L = 12 φ(− 12 + χ)φ − 1 2
2g
χ

where we have neglected the mass term since we will be concentrating on the high-
energy behavior. Here the coupling is introduced through the auxiliary-field “kinetic”
term. The diagrams just discussed now appear through the 1-loop correction to the
auxiliary-field propagator: Since its free propagator is just a constant, it can be
contracted to a point in these multiloop diagrams. The definition of 1PI graphs
has now changed, since we can now cut auxiliary-field propagators, which would not
exist in the usual φ4 form of the action. This modification of the effective action
simplifies the analysis of quantum corrections to the coupling, as well as making it
more analogous to gauge theories. Note in particular the change in interpretation
already at the tree level: We have used the conventional normalization of 1/n! for
factors of φn in the potential, since canceling factors of n! arise upon functional
differentiation. However, the result of eliminating χ from the classical action produces
1 4 1 4
8
φ instead of 24
φ. The reason is that in the diagrams with χ there are 3 graphs
contributing to the 4-φ-point tree, corresponding to χ propagators in the s, t, and u
“channels”. (See subsection VC4.) Although this is a trivial distinction for the trees,
this is not the case for the loops, where the propagator string consists of pairs of φ
particles running in one of these three channels.
The contribution to the 1-loop effective action for χ is then given by the above
calculations, after including the factors of 1/2 for symmetries of the internal and
external lines, and the usual −1 for the effective action:
 
L0 = − 14 χ Γ()B(1 − , 1 − )(− 12 )− χ
 B. EXAMPLES 387

where as usual  = 2 − D/2. Expressing the Beta function in terms of the Gamma
function, and expanding as in previous subsections,

Γ()B(1 − , 1 − )(− 12 )− ≈ 1


+ [−γ + 2 − ln(− 12 )]

Renormalizing away the constant pieces, we find for the classical action plus this part
of the 1-loop effective action
h  i

L + L0 = 12 φ(− 12 + χ)φ − 12 χ 1
g
− 12 ln − µ2 χ = 12 φ(− 12 + χ)φ + 14 χ ln − M 2 χ

where the renormalization group invariant mass scale M is given by

M 2 = µ2 e2/g

Thus, the constant coupling 1/g has been replaced by an effective “running coupling”
− 12 ln(p2 /M 2 ), with energy dependence set by the scale M. (This is sometimes called
the “renormalization group”, the group being related to scale invariance, which is
broken by the introduction of the mass scale M.)
We saw the same dimensional transmutation occuring in the effective potential
in massless theories in subsection VIIB3. The form is similar because both are re-
lated to the appearance of the renormalization mass scale µ from the breaking of
scale invariance by quantum corrections, at either low or high energy: In both cases
dimensional transmutation comes from a finite ln µ2 term arising from the infinite
renormalization. The difference is that in the effective potential case we ignore higher
derivatives, so the µ2 must appear in a ratio to scalar fields, while in the high energy
case we look at just the propagator correction, so it appears in the combination µ2 /p2 .
(More complicated combinations will appear in more general amplitudes.)
Excercise VIIB6.1
Generalize this model to include internal symmetry: Write an analog to the
scalar analog to QCD discussed in subsection VC9, where the “quark” φ now
carries color and flavor indices, while the “gluon” χ (classically auxiliary)
carries just color. Find M, especially its dependence on the numbers n of
colors and m of flavors. Write the same model with the gluons replaced
by “mesons” carrying just flavor indices (so that classical elimination of the
auxiliary fields yields the same action), and repeat the calculation. What are
the different approximation schemes relevant to the two approaches?
388 VII. LOOPS

7. Overlapping divergences

We now perform some 2-loop renormalizations. Our first example is part of the
propagator correction in φ4 theory. By restricting ourselves to the mass renormal-
ization (coefficient of the mass term), we need evaluate the graph only at vanishing
external momentum. (It is then equivalent to a vacuum bubble in φ3 theory, or a
tadpole graph in the mixed theory.) Furthermore, we consider the case where some of
the fields are massless. In such a theory, we encounter (a special case of) the 2-loop
graph of subsection VIIB1, where 1 propagator is massless and 2 are massive. Ex-
panding in , and keeping only the divergent terms (1/2 and 1/), we find (including
a symmetry factor of 1/2 for the 2 massive scalar lines for real scalars)

1 Γ(3 − D)
( 12 m2)1−2
T2 = 2 1 2 3−D B(2 − D
,2 − D
)B(1, D2 − 1) = − [Γ()]2
(2m ) 2 2
2(1 − )(1 − 2)

≈ 14 m2 [Γ()]2[−1 − 3 + 2 ln( 12 m2)]

To this we need to add the counterterm graph, coming from inserting into the
1-loop massive tadpole T1 (with 2 external lines) the counterterm ∆4 (for renormaliz-
ing the φ4 term) from the 1-loop divergence in the 4-point graph with 1 massive and
1 massless propagator. (Since the massless tadpole vanishes in dimensional regular-
ization, we need not consider the counterterm from the 4-point graph with 2 massive
propagators.) From section VIIB5, we use the corresponding integral for a 1-loop
propagator correction A, which is

A = Γ() + finite ⇒ ∆4 = −Γ()

We use a “modified minimal subtraction”, using the Γ() as the subtraction instead
of just the 1/ part of Γ() ≈ 1/ − γ.
 B. EXAMPLES 389

The 1-loop massive tadpole without coupling is

Γ(1 − D
) ( 12 m2)1−
T1 = 2
=− Γ() ≈ − 12 m2 Γ()[1 +  −  ln( 12 m2)]
D 1−
( 12 m2 )1− 2

Combining these results, the divergent part of the 2-loop propagator correction,
with 1-loop coupling counterterm contributions included, is

T2 + ∆4 T1 = 14 m2[Γ()]2[−1 − 3 + 2 ln( 12 m2 ) + 2 + 2 − 2 ln( 12 m2)]

= [Γ()]2(1 − ) 14 m2

Thus, the ln m2 divergences cancel, as expected. (Divergences must be polynomial in

masses as well as couplings.) If we had kept momentum dependence, we would have
seen ln p2 ’s associated with the ln m2’s: So, we have also canceled nonlocal diver-
gences. The surviving local divergence is the superficial divergence, to be canceled
by the 2-loop mass counterterm.
Excercise VIIB7.1
Calculate the p2 part of the 2-loop kinetic counterterm by writing the above
2-loop propagator graph with nonvanishing external momentum, introducing
the Schwinger parameters, doing the loop-momentum integration, taking the
derivative with respect to p2 , and then evaluating at p = 0. Why is there no
subdivergence (1/2 )?
Excercise VIIB7.2
Calculate the complete (all graphs, infinite and finite parts of the) 2-loop
propagator correction for massless φ4 . (See excercise VIIB4.3a.)

For our next example we consider massless φ3 theory, and work in 6 dimensions,
where the theory is renormalizable (instead of superrenormalizable, as in 4 dimen-
sions). For the 2-loop propagator correction, there are only two graphs (plus 1-loop
390 VII. LOOPS

graphs with 1-loop counterterm insertions), one of which is simply a 1-loop propagator
graph inserted into another. The other graph is
Z
1
P = dk dq 1
2 (k + q) 2 (k + 2 p) 2 (k − 2 p) 2 (q + 2 p) 2 (q − 2 p)
2 1 1 21 1 21 1 21 1 2

(with a symmetry factor of 12 for real scalars). This graph can be rewritten as iterated
propagator corrections by use of integration by parts in momentum space. This is
legalized by dimensional regularization, since boundary terms vanish in low enough
dimensions. All invariants can be expressed as linear combinations of the propagator
denominators (there are 5 of each, not counting the square of the external momentum
p2 ), so any product of momentum times derivative acting on the integrand will give
terms killing one denominator and squaring another, except for p2 terms, which can
be canceled by appropriate choice of the momentum multiplying the derivative:
Z
∂ k+q
dk dq · 1 =0
∂k 2 (k + q)2 2 (k + 2 p)2 2 (k − 12 p)2 12 (q + 12 p)2 12 (q − 12 p)2
1 1 1

This operation effectively gives the factor

∂ (q − 12 p)2 (q + 12 p)2 (k + q)2 (k + q)2
· (k + q) → (D − 4) + + − −
∂k (k + 12 p)2 (k − 12 p)2 (k + 12 p)2 (k − 12 p)2
We thus have
( D2 − 2)P = P1 − P2
Z Z
1 1
P1 = dk 1 dq 1
[ 2 (k + 2 p) ] 2 (k − 2 p)
1 2 21 1 2
2 (q + 2 p) 2 (q − 2 p)
1 21 1 2
Z Z
1 1
P2 = dk 1 dq 1
[ 2 (k + 2 p) ] 2 (k − 2 p)
1 2 21 1 2 2 1 1 2
2 (k + q) 2 (q + 2 p)
The former term is the product of two 1-loop propagator graphs, the latter is the
insertion of one 1-loop propagator graph into another.
Both graphs can be evaluated by repeated application of the generalized massless
one-loop propagator correction (with arbitrary powers of free propagators) given at
the end of subsection VIIB4. The result can be expressed as
(P0)2
P1 = −(D − 3) 1 2
2p

[Γ(D − 3)]2 Γ(5 − D)

P2 = cP1, c=
[Γ(3 − D2 )]2Γ( 3D
2
− 5)Γ( D2 − 1)
in terms of the 1-loop propagator correction P0 . We therefore modify our minimal
subtraction so that P0 has the simplest form (G scheme):

P0 = − 16 1 ( 12 p2 )1−
 B. EXAMPLES 391

where D = 6 − 2, and we calculated the coefficient of the 1/ term and threw in a
normalization factor that canceled the rest:

h̄ → N h̄
1 Γ(1 − 2)
N = = (1 − 23 )(1 − 2)
3(D − 6)Γ(2 − D
2
)B( D2 − 1, D
2
− 1) Γ(1 + )[Γ(1 − )]2
Further evaluating c, we find

1 − 2 [Γ(1 − 2)]2Γ(1 + 2)

c = − 13 
(1 − 32 )(1 − 3) [Γ(1 + )]2Γ(1 − 3)Γ(1 − )

Using the expansion of ln Γ(1 − z) in terms of γ and ζ(n), it is easily checked that
this combination of Γ’s is 1 + O(3 ), so we can just drop them. Collecting our results,
we have
 
1−c 1 1 3 − 2 1 − 2
P= D P1 = − 36 2 1
1 + 3 ( 12 p2 )1−2
2
− 2 1 −  (1 − 3
2
)(1 − 3)

Excercise VIIB7.3
Calculate the same graph in four dimensions. It’s finite there, so no countert-
erms are necessary. However, in this case integration by parts gives a factor
of 1/, and each of the two resulting graphs has an additional factor of 1/2 .
The result then has a factor of 1 minus the previously obtained combination
of Γ’s, which we already saw was of order 3. The final result is thus obvious
except for a factor of a rational number:
1
6ζ(3) 1 2
2p

(The on-shell infrared divergence is as expected from power counting.)

We next calculate the counterterm graphs. These are the ones that cancel the
subdivergences coming from the 1-loop 3-point subgraphs. We therefore need the
divergent part of this subgraph. This is easy to evaluate by our previous methods:
The result of Schwinger parametrization, scaling, etc., doing all integration exactly
except over the Feynman parameters is
Z
1
dq 1 2 1
2 q 2 (q + k + 2 p) 2 (q + k − 2 p)
1 21 1 2

Z 1 X
= d3 α δ(1 − α)Γ(3 − D2 )[ 12 α+(1 − α+)(k + 12 p)2 + 12 α− (1 − α−)(k − 12 p)2 ]D/2−3
0
11
= 2 + finite
392 VII. LOOPS

by simply replacing the factor in brackets by 1 (since it is raised to the − power),

where we have used
Z 1 X Z 1 Z 1−α+
d α δ(1 −
3
α) = dα+ dα− = 12
0 0 0

Since we know the divergence is momentum-independent, we can obtain the same

result from a (infrared regularized) tadpole graph with its propagator raised to the
third power: In the notation of subsection VIIB1,
1 Γ(3 − D2 ) 1 2 D/2−3
Â1 (3, 0, m2) = (2m ) = 11
2 + finite
Γ(3) Γ(3)
The contribution of the 2 counterterm graphs (or one for the effective action if we
drop the symmetry factor) is thus

2∆3 P0 = 2(− 12 1 )P0

Collecting terms, we have

 
1 − 23  1 − 2
P + 2∆3 P0 = − 12
1 1 1
1 + 3 ( 12 p2 )1−2 + 1 1 1 2 1−
( p)
2
1− (1 − 2 )(1 − 3)
3 6 2 2

After a little algebra, dropping terms that vanish as  → 0, we find

P + 2∆3 P0 = ( 12 p2 )[ 12
1 1
2
− 1 1
18 
− 1
12
(ln 12 p2 )2 + 19 ln 12 p2 − 23
216
]

Note that modifying minimal subtraction is equivalent to redefining 12 µ2 , which

we have set to 1, but which appears only in the ln’s, as ln( 12 p2 ) → ln(p2 /µ2 ). Thus,
modifying N , which appears only in the combination N ( 12 p2 )− , is the same as shifting
ln 12 p2 :
N → N ea ⇒ ln 12 p2 → ln 12 p2 − a
For example, choosing a = − 23 ,

P + 2∆3 P0 → ( 12 p2 )[ 12
1 1
2
− 1 1
18 
− 1
12
(ln 12 p2 )2 − 5
72
]

Only the O() part of the normalization factor affects the final result: More generally,

( 12 p2 ) → N ( 12 p2 )− ⇒ ln 12 p2 → ln 12 p2 − 1 ln N

Since after adding counterterms, which cancel nonlocal divergences arising from subdi-
vergences, ln’s appear only in finite terms, only the O() part of ln N will contribute.
Thus, we can approximate any normalization factor as

N ≈ ea, a = γ + rational
 B. EXAMPLES 393

as far as the renormalized results are concerned. (The γ identifies this normalization
as modified minimal subtraction, as the MS or G schemes.) This is just the statement
of the renormalization group, that the final result in minimal subtraction schemes
depends only on the choice of scale: The complete normalization factor is really

Ntotal = N ( 12 µ2 ) ⇒ ln 12 µ2 → ln 12 µ2 + 1 ln N

However, the higher-order terms can be convenient for intermediate stages of the
calculation. In this particular case, the nonlocal divergences appearing before cance-
lation are of the form (1/)ln p2 , so the O(2) part of N contributes at intermediate
stages. For example, replacing the original N with

N 0 = (1 − 23 )(1 − 2)Γ(1 − )

would have given the same result even before cancelation, since the change is by
another combination of Γ’s that give 1 + O(3).
Excercise VIIB7.4
Complete the 6D calculation of the exact 2-loop propagator correction in
φ3 theory, including the missing graph and counterterms, to find the total
renormalized 2-loop propagator and its 2-loop counterterms.

REFERENCES
1 Goldstone, Salam, and Weinberg; Jona-Lasinio; loc. cit. (VC):
effective potential.
2 S. Coleman and E. Weinberg, Phys. Rev. D7 (1973) 1888.
3 A. D’Adda, M. Lüscher, and P. Di Vecchia, Nucl. Phys. B146 (1978) 63;
E. Witten, Nucl. Phys. B149 (1979) 285:
quantum CP(n).
4 L.M. Brown and R.P. Feynman, Phys. Rev. 85 (1952) 231;
G. Passarino and M. Veltman, Nucl. Phys. B160 (1979) 151;
W.L. van Neerven and J.A.M. Vermaseren, Phys. Lett. 137B (1984) 241:
reduction of 1-loop integrals with numerators to scalar 1-loop integrals.
5 A.C.T. Wu, Mat. Fys. Medd. Dan. Vid. Selsk. 33 (1961) 72:
scalar box (1-loop 4-point) graph.
6 R.P. Feynman, unpublished;
L.M. Brown, Nuo. Cim. 22 (1961) 178;
F.R. Halpern, Phys. Rev. Lett. 10 (1963) 310;
B. Petersson, J. Math. Phys. 6 (1965) 1955;
D.B. Melrose, Nuo. Cim. A 40 (1965) 181:
reduction of scalar 1-loop graphs to ≤D-point graphs.
7 G. ’t Hooft and M. Veltman, Nucl. Phys. B153 (1979) 365;
G.J. van Oldenborgh and J.A.M. Vermaseren, Z. Phys. C 46 (1990) 425;
394 VII. LOOPS

A. Denner, U. Nierste, and R. Scharf, Nucl. Phys. B367 (1991) 637:

further evaluation and simplification of 1-loop graphs.
8 K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B192 (1981) 159:
integration by parts in momentum space for exact evaluation of higher-loop massless
propagators.
9 D.I. Kazakov, Phys. Lett. 133B (1983) 406, Theor. Math. Phys. 58 (1984) 223, 62
(1985) 84; D.I. Kazakov and A.V. Kotikov, Theor. Math. Phys. 73 (1988) 1264:
more techniques for exact evaluation of such graphs.
10 Dyson, loc. cit. (VC):
treatment of lowest-order overlapping divergences.
11 Salam, loc. cit. (VIIA):
overlapping divergences to all loops.
 C. RESUMMATION 395

...................... ......................
...................... C. RESUMMATION ......................

So far we have only assumed that confinement arises nonperturbatively in 4D

QCD. However, to connect with known, successful results of perturbation theory, we
need to understand how the same methods used to give these perturbative results
can be generalized to include the nonperturbative ones. The simplest method would
be to take the perturbation expansion as is, and find a good method for evaluating
(or perhaps redefining) its sum, with the hope that summation to all orders by itself
would reveal features invisible at finite orders.
Besides the technical difficulties associated with such an approach, the main prob-
lem is that the summation of the perturbation expansion does not converge. Parts of
this problem can be solved by appropriate redefinitions, but other parts indicate a se-
rious problem with perturbation theory, caused by the very renormalization that was
supposed to solve the main problem of finite-order perturbation theory (infinities).

1. Improved perturbation
We saw in the previous section that dimensional transmutation replaced the di-
mensionless coupling constant with a mass scale. In principle, we would like to
explicitly make this replacement as the basis of our perturbation expansion, not
only to make the perturbative parameter physical, but also to take into account
the running of the original coupling. Unfortunately, this is not possible in practice;
however, we can choose the arbitrary (unphysical) renormalization scale µ to be in
the range of energies in the problem at hand, so that the ln(p2 /µ2 ) corrections are
small. A change in scale from one value of µ to another is related to a resummation
of graphs: Although the one-loop term in the effective action containing ln(p2 /µ2 )
comes from a single 1PI amplitude, it contributes an infinite number of terms at dif-
ferent loop orders to the propagator when inserted into any higher-loop 1PI graph,
as 1/(K + A) = 1/K − (1/K)A(1/K) + ... . Although K + A depends only on M, K
depends only on g and A depends only on µ. Thus, any redefinition of µ that leaves
the physical quantity M unchanged requires a corresponding redefinition of g:
1
M 2 = µ2 e−1/g  2
2
⇒ g 2 (µ2 ) =
µ
ln M2

and thus changing µ redistributes the contributions to 1/(K + A) (and therefore to

the summation of graphs in any amplitude) over the different loop orders. For exam-
ple, if the amplitude is most sensitive to the momentum in a particular propagator
(independent of loop momenta), and we choose µ2 ≈ p2 , then although we can’t use
396 VII. LOOPS

the resummed perturbation expansion directly, we can at least push most of it into
the lower orders.
Things get more complicated at higher loops: It becomes difficult to associate the
running of the coupling with the resummation of a particular subset of all the graphs.
However, we already know that this effect can be derived from the breaking of scale
invariance by renormalization. For example, let’s consider Yang-Mills theory, since
gauge invariance restricts it to have only a single coupling parameter. (This makes
it the simplest case conceptually, although not computationally. Here we use only
the fact that it has a single coupling; its explicit renormalization constants won’t be
considered until the next chapter. As an alternative, we can consider the scalar QCD
analog of subsections VC9 and VIIB6, a φ4 theory with an auxiliary field, if we ignore
mass renormalization, or arbitrarily renormalize the mass to zero.) For convenience
of dimensional analysis, we use only coupling constants that are dimensionless in all
dimensions, by scaling with an appropriate power of µ. (In general, we can do this
even for masses.) The classical Yang-Mills action, before and after the addition of
counterterms, is then
Z Z
1 1 2 1
Sclass = 2 1 2  dx tr 8 F , Sclass + ∆S = 2 dx tr 18 F 2
g (2µ ) ĝ

where " #
1 1 1 X∞
1 X
∞

2
= 1 2  2+ n
cn (g 2 ) , cn (g 2 ) = (g 2 )L−1 cnL
ĝ (2µ ) g n=1
 L=n

for some numerical constants cnL . (We can also include h̄’s as g 2 → g 2h̄.) We use
1 2 1 2 1 2 −
2 µ to produce the combination ( 2 p / 2 µ ) in graphs. (In practice, one uses units
1 2
2 µ = 1 until the end of the calculation, and restores units.) The µ dependence is
then given by varying µ for fixed ĝ:
∂ 2 ∂ 2
µ2 2
g ≡ −g 2 − β(g 2), µ2 ĝ ≡ 0
∂µ ∂µ2

where the g 2 term is the classical contribution, β will be found to be independent of

 (except indirectly through g 2 ), and ĝ’s independence from µ is the statement that
the physics is independent of the choice of µ (i.e., ĝ depends on only M and ). We
then find
∂ 1 ∂ 2 ∂ 2 ∂
0 = µ2 µ2 ⇒ β = g4 (g c1 ), (g cn+1 ) = −β 2 cn
∂µ2 ĝ 2 ∂g 2 ∂g 2 ∂g
Thus, the coefficients of the 1/ terms determine those of both the higher order terms
and β.
 C. RESUMMATION 397

This gives us an expression for β,

X
∞
β= (g 2 )L+1 βL, βL = Lc1L
L=1

Since g 2 is itself unphysical, the information we can get from analyzing the running
of this coupling is arbitrary up to redefinitions. For example, assume that all βL are
nonvanishing, and write the definition of β as (in D = 4 →  = 0)

∂ 1 1 X ∞
2
µ 2 2
= f = 4
β = (g 2)L βL+1
∂µ g g L=0

Then under a redefinition g 2 → g 2 (g 02) we have

 −1
∂ 1 ∂(1/g 2 )
µ 2
= f(g 2 (g 02 )) ≡ f 0 (g 02 )
∂µ2 g 02 ∂(1/g 02)

Now we consider a “perturbative” type of redefinition, as results from changing renor-

malization prescriptions, so g 2 gets only “O(h̄)” corrections: Taylor expanding

g 2 = g 02 + k1 g 04 + +k2 g 06 + O(g 08 )

1 1
⇒ 2
= 02 + constant + O(g 02 )
g g
we find
∂(1/g 2 )
= 1 + O(g 04), f(g 2 (g 02)) = f(g 02) + O(g 04 )
∂(1/g 02 )
⇒ f 0 (g 02 ) = f(g 02 ) + O(g 04)

Thus, the first two coefficients of β (β1 and β2) are unaffected, while terms found at 3
loops and beyond can be modified arbitrarily, and even be set to vanish. In the more
general case of more than 1 coupling, it is sometimes possible to eliminate also some
of the 2-loop contributions.
Therefore, to consider the general behavior of the coupling as a function of energy
(µ2 ), it is sufficient to solve the equation

∂ 2
µ2 g = −β1g 4 − β2 g 6
∂µ2

(using, e.g., the change of variables t = ln µ2 and u = 1/β1 g 2 ) as

 −β2 /β12
µ2 2 1 β2
2
= e1/β1g 2
+
M g β1
398 VII. LOOPS

with M 2 as the constant of integration. Using an allowed type of redefinition for g2,
and also redefining the arbitrary constant of integration M 2 , we can simplify this to
1 1 β2 2

2
→ 2− , M 2 → M 2 eβ2 /β1
g g β1
β1 g 4 µ2 2 2
⇒ β= , = e1/β1g (g 2 )β2 /β1
1 − ββ21 g 2 M 2

(This redefinition changes the range of what g 2 is called negative and what positive.
However, g 2 is just a parameter, not a physical coupling: As far as the unitarity of
the kinetic term is concerned, only the residues near the poles of the propagator are
relevant. Also, our allowed class of redefinitions do not affect behavior for small g2,
and thus perturbation theory.)
Excercise VIIC1.1
Let’s analyze this solution in more detail:
a Graph the function y(x) = eax xb (or graph ln y to make it simpler) for a
and b positive, negative, and vanishing, to study the behavior of the function
µ2 (g 2 ). The analysis can be simpliifed (and the behavior for different values of
a and b related) by considering g 2 positive and negative, and the symmetries

a → −a, x → −x, y → (−1)b y

1
b → −b, x → −x, y → (−1)b
y
Note that g 2 can be nonpositive for some values of µ2 : For example, even for
β2 = 0, we have g 2 = 1/β1 ln(µ2/M 2 ), which is negative for µ < M or for
µ > M. What happens for β2 6= 0?
b After applying the above redefinition, apply the second redefinition
1 1 β2
→ + g
g g β1
Find the new β and µ2 (g 2). Compare to the behavior of µ2 (g 2 ) before this
redefinition, for the cases β2 /β1 < 0, noting the “duality” symmetry g ↔
(−β1/β2 )/g.
Excercise VIIC1.2
Consider some theory with a single dimensionless coupling g2 , but now also
a single mass m. By the above methods we find
µ2 ∂ 2
2
m = m2 [−1 − βm (g 2 )]
∂µ
 C. RESUMMATION 399

(The m dependence follows from dimensional analysis.) Solve for m2 as a

function of g 2 , as an integral over g 2 in terms of β and βm . Show that after
an appropriate redefinition
µ2 ∂ 2
2
m = m2(−1 − βm1 g 2 )
∂µ
for some constant βm1. Solve for m2 explicitly in terms of g 2, when we have
also redefined β to β1g 4 + β2 g 6. Then make the final redefinition 1/g2 →
1/g 2 − β2/β1 used to simplify M 2 .
For purposes of perturbation theory, it is useful to invert this: For small g2, we
have approximately  
1 µ2 β2 µ2
≈ β1ln 2 + ln β1ln 2
g2 M β1 M
This implies that the terms in the effective action that carry the M dependence are
given by Z   
− β2 −
ΓM ≈ tr dx 8 F β1ln 2 + ln β1ln 2
1
F
M β1 M
(We can also replace − → in this limit, ignoring iπ’s in comparison to ln’s.)
The general class of coupling redefinitions we considered are allowed by pertur-
bation theory: If we knew the exact solution to a field theory, we would be more
restrictive, requiring invertibility. However, in perturbation theory, given two renor-
malization prescriptions related by some such coupling redefinition, we might know
this redefinition only perturbatively, and perhaps only to a few orders. Even if we
knew it exactly, and knew it to be noninvertible, it still might not be clear which of
the two prescriptions was the correct one, if either. Therefore, the renormalization
group alone is sufficient to draw conclusions about the behavior of a theory only at
“small” ( 1) coupling.
Similar remarks apply to propagators, S-matrix elements, etc. Consider any func-
tion Gn appearing as the coefficient of n fields in a term in the effective action.
Comparing the unrenormalized Ĝn to the renormalized Gn ,
µ2 ∂
Gn (g 2, µ2 ) = Z −n (ĝ 2 ( 12 µ2 )− , )Ĝn (ĝ 2, ), Ĝ = 0
∂µ2
 2 
µ2 d µ∂ ∂ µ2 ∂
⇒ G = + β + nγ G = 0, γ = ln Z
dµ2 ∂µ2 ∂g 2 ∂µ2
where Z is a wave-function renormalization factor (not required for pure Yang-Mills
in the background gauge; or we can examine ratios of such quantities where the Z’s
cancel, which are more physical, such as S-matrix elements).
400 VII. LOOPS

2. Renormalons
The perturbation expansion in general can’t be resummed in the naive way be-
cause the number of diagrams increases as n!(constant)n at n loops. The simplest
example of this is a self-interacting scalar in D=0:
Z ∞ X ∞
dφ 1 2 1 2 4
Z= √ e− 2 φ − 4 g φ = g 2n Zn
−∞ 2π n=0

Since there is no momentum integration, each diagram is just 1 (times some permu-
tation factors), so Zn just counts the number of diagrams at n loops. We use g2 so
the coupling is similar to that in Yang-Mills: As usual, we can rescale φ → φ/g to
recognize g 2 as h̄:

φ0 = gφ ⇒ 1 2
2φ + 14 g 2φ4 = 1 1 02
( φ
g2 2
+ 14 φ04)

(Of course, we can be more explicit by writing h̄g2 in place of just g 2 or h̄, but the
effect is identical, since they both appear only in that combination.) This integral
can be evaluated exactly at any order of perturbation theory:
Z ∞
dφ 1 2 1
Zn = √ n!1 (− 14 φ4 )n e− 2 φ = n!1 (−1)n √1π Γ(2n + 12 ) ≈ √ (n − 1)!(−4)n
−∞ 2π 2π
where we have used the Stirling approximation for Γ(z) at large z.
Excercise VIIC2.1
Find the following properties of the Γ function for large argument:
a Derive the Stirling approximation
r
2π  z z
lim Γ(z) ≈
z→∞ z e
by applying the method of steepest descent to the integral definition of
Γ(z + 1). (See subsections VA2 and VA5.)
b Use this approximation, and limz→∞ (1 + 1z )z = e, to show
√
lim Γ(az + b) ≈ 2π(az)az+b−1/2 e−az
z→∞

Thus we might as well apply the steepest descent approximation directly to the
original integral: Using also an integral for h̄ (= g2 in this case),
I Z  
dh̄ φ
Zn = n+1 D √ e−S/h̄
2πih̄ h̄
 C. RESUMMATION 401

we first apply steepest descent to the φ integral, yielding the usual first two terms in
the JWKB expansion. Then the h̄ integral can be approximated as Γ(n) by keeping
only the part of the contour on the positive real axis:
I   n 
dh̄ 
−S/h̄  1 1 
n+1 e  ≈ Γ(n)
2πih̄ δS/δφ=0 2πi S δS/δφ=0

X   2 −1/2  n
1 δ S 1
⇒ Zn ≈ det (n − 1)!
2πi δφ2 S
δS/δφ=0
S6=0

(S = 0 solutions contribute only to Z0 . A similar result can be obtained by simulta-

neously using steepest descent for the h̄ integral, yielding a “classical value” of h̄ in
terms of S.) In the present case, the nontrivial classical solutions are

S = 12 φ2 + 14 φ4 ⇒ φ = ±i

which gives the same Zn as previously (being careful to sum the two terms for the
P
two solutions). Thus, we see that in general we have to sum ∞ n
n=0 n!(h̄/S) , which
does not converge. Furthermore, this divergence is associated with finite-action (“in-
stanton”) solutions to the classical equations of motion.
The simplest example of a resummation problem is the one-loop propagator cor-
rection. We have seen that the classical and one-loop kinetic terms can be combined
to give a kinetic operator of the form β1K(p2 )ln(−p2/M 2 ) in massless theories, or
at high energy in massive theories, where K is the classical kinetic operator. The
free (or asymptotic) theory has solutions where this kinetic operator has a zero (the
propagator blows up). Besides the classical solution at K(p2 ) = 0, there is another
at p2 = −M 2 :
1 1 M2 1
 =  ≈− × 2
2
β1 K(p2 ) ln − Mp 2 β1K(p2 ) ln 1 − p2 +M 2
β1K(−M ) p + M 2
2
M2

This might be expected to be a bound state, called a “renormalon” because of its rela-
tion to the renormalization group. However, the residue of this pole in the propagator
can have the wrong sign, indicating the appearance of a ghost (“Landau ghost”), and
thus a violation of unitarity.
Excercise VIIC2.2
The Landau ghost itself is not necessarily a problem in quantum field theory,
although it indicates the possibility of such problems. Examine the behavior
of this ghost after taking into account the 2-loop correction (β2 ), before and
after the simplifying redefinition of the previous subsection, for all the various
402 VII. LOOPS

signs of β1 and β2 . Since the expression for µ2 (g 2) can’t be inverted, use the
fact that the propagator follows from the coupling g2(µ2 ) as

g 2 (p2 )
∆∼
p2

(Field redefinitions can’t remove the momentum dependence of couplings.)

Then new poles (or other singularities) in the propagator correspond to the
limit g 2 → ∞, so find p2 (g 2 ) there.

...

...

This causes problems similar to those from instantons when the quantum prop-
agator is inserted into another graph. We set external momenta to vanish, as an
approximation for high energy for the loop momenta, or to evaluate low-energy quan-
tities such as anomalous magnetic moments. In any one-loop 1PI graph with n 1-loop
propagator insertions and l external lines, we get an integral at high energy of the form
(e.g., in QCD or the scalar analog with auxiliaries of subsections VC9 and VIIB6)
Z   2 n Z ∞  
2 −l k −(l−2)u n β1 n
d k (k ) −β1 ln
4
∼ (−β1) n
du e u ∼ n! −
µ2 0 l−2

where we changed variables to u = ln(k 2/µ2 ) (remembering Γ(n + 1) = n!). We

have used effectively an infrared cutoff by approximating the u-integral from 0 to ∞
instead of −∞ to ∞. If we look instead at the low-energy (of the loop momentum)
behavior, now taking l − 1 massive classical propagators with 1 massless propagator
(to insure IR convergence) with n insertions, we find
Z   2 n Z 0
2 −(l−1) 2 −1 k
4 2
d k (k + m ) (k ) −β1 ln 2
∼ β1
n
du eu(−u)n = n!β1n
µ −∞

Since the former comes from UV behavior it’s called a “UV renormalon”, while the
latter coming from IR behavior is called an “IR renormalon”. The essential difference
is the relative factor of (−1)n . In fact, the former expression is also the high-energy
limit of the latter (neglecting masses then), so the complete integral (u from −∞ to
∞, so k 2 from 0 to ∞) can be approximated as the sum of the UV renormalon and
IR renormalon contributions.
 C. RESUMMATION 403

3. Borel
Since renormalons and instantons cause the perturbation expansion to diverge by
a factor of n!, we look for a method to formally sum such series, by relating them to
series that do converge. In general, we consider the series
X
∞
A(h̄) = h̄n an
n=0

and define the “Borel transform” as:

Z r+i∞
d(1/h̄) z/h̄
Ã(z) = e A(h̄)
r−i∞ 2πi

(for some real number r to the right of all singularities of A) in anticipation of

instanton-like contributions. The inverse is
Z ∞
A(h̄) = dz e−z/h̄ Ã(z)
0

The inverse Borel transform is related to the Laplace transform (with the variable
change x = 1/h̄) and the Mellin transform (x = 1/h̄ and y = ez ). Evaluating explicitly
for the above series,
X
∞
Ã(z) = δ(z)a0 + z n n!1 an+1
n=0

So the Borel-transformed sum comverges faster by a factor of n!, which is just what
we need for perturbation theory. The idea for resumming the perturbation expansion
is to first do the Borel sum, then inverse Borel transform the resulting function. Of
course, this procedure does not necessarily fix the original problem, which might
merely be translated into problems of convergence or ambiguity for integration of the
inverse transform. In particular, we need Ã(z) to be well defined along the positive
real axis.
We saw that generically the sums involved were approximately of the form
X
∞
A(h̄) ∼ h̄n(n − 1)!(−k)n
n=1

In that case
X
∞
1
Ã(z) ∼ (−1)nz n k n+1 = 1
n=0
z+ k

When k < 0, this leads to a singularity in the integral defining the inverse Borel
transform. It can be “regularized” by choosing a contour that goes around the pole,
404 VII. LOOPS

but the choice of contour is ambiguous, and choosing an arbitrary linear combination
of the two contours introduces a free parameter. Explicitly, we have

A(h̄) = A0(h̄) + ζe−1/|k|h̄

where A0 is the result of a particular prescription (e.g., principal value), and ζ is the
new parameter. The ζ term is clearly nonperturbative, since each term in its Taylor
expansion in h̄ vanishes. This new parameter can be interpreted as a new (nonper-
turbative) coupling constant in the theory, just like ambiguities in renormalization of
new counterterms in perturbatively nonrenormalizable theories.
Now we more carefully analyze the explicit sums we found in the previous sub-
√
section. The first example is h̄Z for D=0:
Z r+i∞ Z Z ∞
d(1/h̄) z/h̄ ∞ dφ −S/h̄ dφ 1 X 0 −1
Ã(z) = e √ e = √ δ(z − S) = √ (S )
r−i∞ 2πi −∞ 2π −∞ 2π 2π S=z
√
(The contribution from S = 0 is artificial, coming from our using A = h̄Z instead
of Z.) So, this integral can be explicitly evaluated. (For example, for the action we
used above, we can explicitly solve for φ at S = z.) However, there is then a problem
in inverting the Borel transform: Near z = z0 ≡ S(φ0 ) for classical solutions φ0 , we
have

S(φ) ≈ S(φ0 )+ 12 S 00(φ0)(φ−φ0 )2 ⇒ S 0 (φ) ≈ S 00(φ0 )(φ−φ0 ), z−z0 ≈ 12 S 00(φ0)(φ−φ0 )2

⇒ (S 0)−1 ≈ [2S 00(φ0)]−1/2(z − z0)−1/2

Therefore, there are cuts with branchpoints at classical values of the action, leading
to ambiguities in the result for A(h̄). We thus see that new coupling constants are
introduced for each solution to the classical field equation with positive action. (For
our D=0 example S < 0, and there is no problem, but more realistic examples, like
Yang-Mills instantons, have S > 0.)
In the case of renormalons, we see from the previous subsection that the large-n
behavior gives singularities at z = N/β1 for positive integer N. This integral also is
easier to evaluate after Borel transforming: We consider a one-loop graph, but replace
one internal line with the “full” quantum propagator coming from the 1-loop effective
action (the same as summing a string of 1-loop propagator insertions), while using
massive propagators for the remaining lines. We thus examine first the transform of
the quantum propagator
Z  −β1 z
d(1/h̄) z/h̄ 1 1 1 k2
e = 2
2πi k 2 h̄1 + β1 ln(k 2 /µ2 ) k µ2
 C. RESUMMATION 405

by closing the contour on the left. Then inserting this transformed propagator into
the complete diagram,
Z  −β1 z  2 −β1 z
1 1 k2 m
4
dk 2 2 l−1 2 2
∼ Γ(1 − β1z)Γ(l − 2 + β1z)
(k + m ) k µ µ2
using the integrals of the subsection VIIB1. This expression is the sum over n of the
UV/IR renormalon example at the end of the previous subsection, except that we have
done the summation over n as the first step (and used the Borel transform to assist
in the evaluation). The first Γ has poles at z = N/β1 for positive N, representing
the IR renormalon, which are relevant for β1 > 0, but the second Γ has poles at
z = −(N + l − 3)/β1 for positive N (and l ≥ 3 for the original diagram to be UV
convergent), representing the UV renormalon, which are relevant for β1 < 0. To the
one-loop approximation for the β-function we have used, the singularities are just
poles, but if the two-loop propagator insertions are used, these singularities become
the branchpoints for cuts.
The new coupling constants that appear nonperturbatively can be given a physical
interpretation in terms of vacuum values of polynomials of the fields. The basic idea
is analogous to perturbative tadpoles: In that case corrections to S-matrices due to
vacuum expectation values of scalar fields can be expressed by propagators that end
at a “one-point vertex”, whose coefficient is the vacuum value of the field:
Z
hφ(x)i = Dφ e−iS φ(x) = c

in position space for some constant c, or in momentum space as

Z
Dφ e−iS φ(p) = cδ(p)

Similarly, we could expect graphs to have two propagators that end at a two-point
vertex representing the vacuum value of the product of the two fields associated with
the ends of the two propagators, and so on for higher-point vertices. For example,
for a φ2 vertex in a scalar theory, it would correspond to a contribution of the form
Z
Dφ e−iS φ(x)φ(y) = (c2 + c0 ) + ...

in position space, or in momentum space

Z
Dφ e−iS φ(p)φ(q) = (c2 + c0 )δ(p)δ(q) + ...

where c2 is the contribution from hφi2 , so c0 represents (∆φ)2 = hφ2 i − hφi2 . Such
vacuum values do not appear in perturbation theory for higher than one-point; we get
406 VII. LOOPS

only one δ(p) for each connected part of any graph. However, such contributions would
be expected to give similar contributions to those we have found for renormalons: By
dimensional transmutation, a contribution to an amplitude of the form ζe−n/β1 h̄ must
appear in the combination
 2 n  2 n
−n/β1 h̄ −n/β1 h̄ µ M
ζe → ζe 2
=ζ
p p2
This is the type of contribution expected from a propagator with tadpole insertions,
or in the same way from any other type of vacuum value. In particular, in QCD
there are no fundamental scalar fields, but only scalar fields can get vacuum values,
by Lorentz invariance. Thus, the vacuum values come from composite scalars, like
tr(F 2), q̄q, etc.
Note that renormalons are a feature of renormalizable theories: They do not
appear in superrenormalizable or finite theories. In particular, the path-integral
methods of “constructive quantum field theory” have been used to show that cer-
tain interacting field theories in lower dimensions can be proven rigorously to exist
— superrenormalizable theories with unique vacuua.

4. 1/N expansion
Perturbation theory is insufficient to evaluate all quantities in quantum physics,
since (1) such expansions don’t always converge; (2) if they do converge, they might
not converge to the complete result; and (3) even if they do give the complete answer,
their summation might not be practical.
There are many perturbative expansions in quantum field theory. When we say
“perturbation theory” in this context, we generally mean an expansion in the number
of fields (or, in diagrammatic terms, number of vertices), since in the path integral
we kept the exact quadratic part of the action but expanded in powers of the inter-
action terms (cubic and higher). (This is usually also an expansion in the coupling
constants, depending on how we define the fields, which can be redefined by factors
of the couplings.) One disadvantage of this expansion is that it violates manifest
gauge invariance: Nonabelian gauge transformations are nonlinear in the fields, and
thus mix diagrams with different numbers of fields. (These are the internal fields;
external fields are asymptotic, and approximated as free.) Graphs that are related by
gauge transformations must be added together to obtain gauge-invariant, and thus
physically meaningful, expressions. Also, in practice individual graphs contain “gauge
artifacts” that complicate them in certain gauges, but cancel in gauge-invariant ob-
jects, like S-matrix elements.
 C. RESUMMATION 407

There can be a large number of graphs contributing to a particular physical

process (given set of external states) at any particular loop order. There is another
gauge-invariant expansion that can be applied to Yang-Mills theory to subdivide these
sets of graphs, based on the freedom of choice of the Yang-Mills group itself: We
have seen that the classical groups are defined in terms of N×N matrices, where N is
arbitrary. Clearly, S-matrix elements must depend on N, even if the external states are
restricted to be group singlets or representations of an N-independent subgroup, since
the number of internal states increases as some polynomial in N. We now examine
how this can be used to define a perturbation expansion in terms of N.
We have already seen in subsection VC9 that the group theory of any graph can
be detached from the momentum and spin (so we considered there a simple model
of scalars φ). We also saw there that the group theory of such matrices is most con-
veniently graphed by a double-line notation, where each line acts group-theoretically
as a bound (anti)quark, reducing the group theory to trivial Kronecker δ’s. We now
notice that in some loop graphs, depending on how the lines are connected, some of
the quark lines form closed loops. Again the group theory is trivial: There is a factor
of N for each such loop, from the sum over the N colors. We can also give a physical
picture to these numerical factors: Since we draw the scalar propagator as quark and
antiquark lines with finite separation, think of the scalar as a (very short) string,
with a quark at one end and antiquark at the other. This gives a two-dimensional
structure to the diagram, by associating a surface with the area between the quarks
and antiquarks (including the area at the vertices). We can extend this picture by
associating a surface also with the area inside (i.e., on the other side of) each closed
quark loop. In particular, for any “planar” diagram, i.e., any diagram that can be
drawn on a sheet of paper without crossing any lines, and with all external lines on
the outside of the diagram, the entire diagram forms an open sheet without holes,
and with the topology of a disc (simply connected). It is also clear that, for a fixed
number of loops and a fixed number of external lines, a planar diagram has the great-
est number of factors of N, since crossing lines combines quark loops and reduces the
power of N.
We can be more quantitative about this N dependence, and relate it to the topol-
ogy of the graph. In subsection VC2 we saw the number of propagators, vertices, and
loops were related by P − V = L − 1. This relation treats a Feynman diagram as
just a graph, points connected by lines. We now consider a diagram as a polyhedron,
with propagators as the edges, and closed quark loops as the faces, as defined by our
use of matrices for fields. We then have as an additional relation for closed surfaces
408 VII. LOOPS

“Euler’s theorem”,
F = P − V − 2(H − 1)

(in terms of the “Euler number” χ ≡ −2(H − 1) = V − P + F ), where F is the

number of faces and H is the number of “handles”: 0 for the sphere, 1 for the torus
(doughnut), etc. This follows from the previous relation: First combining them as

L = F + 2H − 1

we note that “cutting” any handle along a loop (without separating the pieces) pro-
duces 2 faces; in other words, introducing two faces (as a “lens”) into a loop that
circles a handle changes the surface without changing the diagram, replacing 1 han-
dle with 2 faces. The last relation then follows from the case with no handles, where
each face gives a loop, except that the no-loop case corresponds to 1 face (or start
with a less trivial case, like a cube, if that’s easier to picture and count momenta for).
Using the fact that the g 2 appears in Yang-Mills the same way as h̄, and that
each face gets a factor of N, we find the g and N dependence of any graph is

(g 2 )L−1 N (L−1)−2(H−1) = (Ng 2 )L−1 N −2(H−1)

We thus see that effectively Ng 2 is the coupling squared suited to planar graphs,
counting the number of loops, while 1/N 2 is a new coupling squared, counting the
number of handles. Therefore, we can sum over both Ng2 and 1/N 2 : Each Feynman
graph is a particular order in each of these two couplings. The sum of all graphs at
fixed orders in both couplings gives a gauge-invariant subset of the graphs contribut-
ing to a particular S-matrix element. (This is sometimes called “color decomposition”.
Note that g 2 is the coupling normalized for matrices of the defining representation,
which was required here to define the 1/N expansion, while Ng2 = 12 gA2 is the coupling
normalized for the adjoint: If we had used matrices for the adjoint representation, a
factor 1/gA2 would appear in front of the action, because of the difference in normal-
ization of the trace of the matrices.)
Excercise VIIC4.1
Consider φ4 theory in D=4, where φ is now an N×N hermitian matrix. Gener-
alize the auxiliary-field propagator correction calculation of subsection VIIB6
to leading order in 1/N, showing the N-dependence at all steps. Show that
now, to this leading order, both the N- and g-dependence of the effective
action can be absorbed into M.
We can also consider more complicated models, such as chromodynamics, with
fields appearing in the defining representation of the group, such as quarks. When
 C. RESUMMATION 409

a quark field makes a closed loop, it looks like a planar loop of a gluon, except that
the closed quark line is missing, along with a corresponding factor of N. Thus, there
is effectively a “hole” in the surface. Since only one factor of N is missing, a hole
counts as half a handle. We can also draw a flavor-quark line for the quark propagator
alongside the color-quark line. Since this line closes in quark-field loops, we also get
a factor of M (for M flavors) for each quark loop.
The fact that the 1/N expansion is topological (the power of 1/N is the number
of holes plus twice the number of handles) closely ties in with the experimental ob-
servation that hadrons (in this case, mesons) act like strings. Thus, we can expand
in 1/N as well as in loops. While the leading order in the loop (Ng2 ) expansion is
classical (particle) field theory, the leading order in the 1/N expansion is classical
open-string theory (planar graphs). However, seeing the dynamical string properties
requires summing to all orders in Ng 2 for leading order in 1/N.
Thus, 1/N acts as the string coupling constant. (N appears nowhere else in the
action describing string states, since they are all color singlets.) The experimental fact
that the hadronic spectrum and scattering amplitudes follow so closely that of a string
(more on this later) indicates that the perturbative expansion in 1/N is accurate, i.e.,
that quantum corrections are “small” in that sense. One application of the smallness
of 1/N (largeness of N) is the “Okubo-Zweig-Iizuka rule”: A planar graph describes
classical scattering of open strings (mesons). It corresponds topologically to a disc,
which is a sphere with one hole, and is therefore order 1/N. Compare this to two
planar graphs connected by a handle. It describes classical scattering of open strings
with one intermediate closed string (glueball), where the handle is a closed-string
propagator connecting two otherwise-disconnected classical open-string graphs. It
corresponds to a cylinder, which is a sphere with two holes, and is therefore order
1/N2 . In terms of flavor lines, the latter graph differs from the former in that it
has an intermediate state (the glueball) with no flavor lines. The OZI rule is that
amplitudes containing an intermediate glueball are always smaller than those with an
intermediate meson. This rule also has been verified experimentally, giving a further
justification of the 1/N expansion (though not necessarily of string behavior).
Generalizing to groups SO(N) and USp(2N) gives more varied topologies: Since
the left and right sides of propagators are no longer distinguishable, the string surface
is no longer orientable (the surface no longer has two distinguishable sides), so we
can also have unorientable surfaces such as Möbius strips and Klein bottles. One can
also perform a separate expansion in the number M of flavors.
410 VII. LOOPS

The fact that the leading (planar) contributions are of order (Ng2 )L−1 requires a
modification of the Borel transform of the previous subsection: We now identify

h̄ = Ng 2

instead of just h̄ = g 2 , so we can use the 1/N expansion in conjunction with the Borel
transform. In particular, this means removing the factor of N from β1 and absorbing
it into h̄. The result is that the position of the renormalon singularities in the z plane
is independent of N. However, the same is not true for the instantons: A one-instanton
solution corresponds to choosing a single component of φ nonvanishing in our scalar
model, so that the classical solution φ0 for the action S[φ0 ] has no N-dependence.
(Choosing φ proportional to the identity matrix yields an N-instanton solution.) The
analog in the Yang-Mills case is using just a (S)U(2) subgroup of the full U(N) to
define the instanton. (Note that the structure constants for U(N) are N-independent
for the defining representation: See excercise IB5.2.) Then S/g2 = NS/h̄. The result
for the positions of the singularities in z is then at integer multiples (positive or
negative, depending on considerations given in the previous subsection) of z0, where

1/β1 for renormalons
z0 =
NS[φ0] for instantons

where β1 and the one-instanton action S are N-independent.

The net result is that instantons are unimportant for large N. Thus, if we take
the 1/N approach of using a resummation to define a string theory, the instantons
do not take a role in defining the string. (They might return in another form when
considering classical solutions to the string theory, or their contribution might be
just a small part of the total nonperturbative contribution.) On the other hand,
approaches that analyze just the low-energy behavior of a theory can make use of the
instantons: If the physical value of N is small, or the U(N) theory is spontaneously
broken to give a small effective N at low energies (as in GUTs), then instantons may
be treated as the dominant nonperturbative contribution to low-energy effects such
as chiral symmetry breaking. This can be sufficient for studying low-energy bound
states, but is insufficient for studying confinement, whose physical definition is the
existence of bound states of very high energy.

REFERENCES
1 E.C.G. Stückelberg and A. Petermann, Helv. Phys. Acta 26 (1953) 499;
M. Gell-Mann and F.E. Low, Phys. Rev. 95 (1954) 1300;
C.G. Callan, Phys. Rev. D2 (1970) 1541;
 C. RESUMMATION 411

K. Symanzik, Commun. Math. Phys. 18 (1970) 227:

renormalization group.
2 G. ’t Hooft, Nucl. Phys. B61 (1973) 455:
renormalization group via dimensional regularization.
3 G. ’t Hooft, Can we make sense out of “quantum chromodynamics”, in The whys of
subnuclear physics, proc. 1977 Int. School of Subnuclear Physics, Erice, ed. A. Zichichi
(Plenum, 1979) p. 943:
arbitrariness of all but first two nonvanishing coefficients in β .
4 L.D. Landau and I. Pomeranchuk, Doklady Akad. Nauk USSR 102 (1955) 489:
Landau ghost.
5 F.J. Dyson, Phys. Rev. 85 (1952) 631:
divergence of perturbation expansion in quantum field theory.
6 L.N. Lipatov, Soviet Physics JETP 45 (1977) 216:
D=0 renormalon.
7 G. ’t Hooft, loc. cit. (ref. 3), Phys. Lett. 109B (1982) 474;
B. Lautrup, Phys. Lett. 69B (1977) 109;
G. Parisi, Phys. Lett. 76B (1978) 65, Nucl. Phys. B150 (1979) 163;
Y. Frishman and A. White, Nucl. Phys. B158 (1979) 221;
C. DeCalan and V. Rivasseau, Comm. Math. Phys. 82 (1981) 69;
F. David, Nucl. Phys. B234 (1984) 237;
A.H. Mueller, Nucl. Phys. B250 (1985) 327, Phys. Lett. 308B (1993) 355;
L.S. Brown, L.G. Yaffe, and C.-X. Zhai, hep-ph/9205213, Phys. Rev. D46 (1992) 4712:
renormalons.
8 A.S. Wightman, Should we believe in quantum field theory, in The whys of subnuclear
physics, ibid. p. 983:
review of the relation of renormalons to constructive quantum field theory.
9 ’t Hooft, loc. cit. (VC):
1/N.
10 S. Okubo, Phys. Lett. 5 (1963) 165;
Zweig, loc. cit. (IC);
J. Iizuka, Prog. Theor. Phys. Suppl. 37-38 (1966) 21;
J. Iizuka, K. Okada, and O. Shito, Prog. Theor. Phys. 35 (1966) 1061.
11 G. Veneziano, Phys. Lett. 52B (1974) 220:
expansion in (1/) the number of flavors.
412 VIII. GAUGE LOOPS

VIII. GAUGE LOOPS

Gauge invariance plays an important role in quantum corrections. It not only
simplifies their form, but leads to new effects. In particular, it not only improves
high-energy behavior, but can eliminate divergences altogether, in the presence of
supersymmetry.
In general, the first thing to calculate in quantum field theory is the effective
action. Once this has been calculated, other properties can be determined: the
vacuum, S-matrix, etc. In particular, in spontaneously broken theories, the effective
action should be calculated with the symmetric (unbroken) vacuum, which has simpler
Feynman rules; once the effective action has been calculated, vacuum values of the
fields can be determined, and the S-matrix can be calculated as a perturbation about
this quantum vacuum. (The alternative of defining Feynman rules for the classical
broken vacuum and then calculating quantum corrections doubles the work in finding
vacuum values.)

....................... .......................
....................... A. PROPAGATORS .......................

We first consider propagator corrections in some specific theories with spin. In

the following calculations we assume the gauge coupling appears only as an overall
factor in the classical action: It thus also counts loops, so our 1-loop graphs are
coupling-independent. All the integrals have been performed in subsections VIIB4-5;
all that remains is the numerator algebra, which follows the examples of subsection
VIC4. As we have seen, such corrections are important in analyzing high-energy
behavior; as we’ll see in the following section, they are also important for low energy.
(Of course, for massless particles the two are related by conformal invariance, even
when quantum corrections break it.)

1. Fermion

Our first calculation is the one-loop correction to the electron kinetic operator in
QED: The S-matrix element is
Z γ a (k/ + 12 /p + m
√ )γ
2 a
A2e = dk 1
2 (k − 12 p)2 21 [(k + 1 2
2 p) + m2 ]
 A. PROPAGATORS 413

At this loop level the only difference between using D-dimensional γ-matrix algebra
(dimensional regularization) and 4-dimensional (dimensional reduction) is an unphys-
ical finite renormalization, so for simplicity we’ll use the latter method. Then the
numerator is
√
k/ + 12 p/ − 2m
The result of the integral is then
m2 √
A2e = −p/ 2 [Â2(p2 , 0, m2) − Â2 (0, 0, m2)] + ( 12 p/ − 2m)Â2(p2 , 0, m2 )
2p
in the notation of subsection VIIB5. The UV divergent part follows from

Â2 (p2 , 0, m2) = 1


+ finite

The contribution to Γ is minus the S-matrix element, but the counterterm has a
second minus sign to cancel the divergence:
Z √
∆S =  dx Ψ̄(− 12 i∂/ − 2m)Ψ
h̄

The calculation for the quark self-energy in QCD is the same except for group-theory
factors (see subsection VIIIA5).
Excercise VIIIA1.1
Repeat the calculation with D-dimensional γ-matrix algebra. What is the
difference in the finite part, and why doesn’t it matter?
In subsection VIIB5 we considered MOM subtraction (see subsection VIIA3) for
scalar propagators. The analysis in this case is similar, but now we expand in p/
instead of p2 :
∆K = a + b( √m2 − p/) + O[( √m2 − p/)2 ]
However, since ∆K is normally expressed as functions of p2 times 1 and p/, we need
to translate: Using ( √m2 + p/)( √m2 − p/) = 12 (p2 + m2 ),

∆K = a + b0( √m2 − p/) + c 12 (p2 + m2 ) + O[( √m2 − p/)(p2 + m2), (p2 + m2 )2 ]

= a + (b0 + 2 √m2 c)( √m2 − p/) + O[( √m2 − p/)2 ]

We next reevaluate the fermion propagator correction, to linear order in √m2 − p/.
Starting with
Z Z
1
Ã(x, p , m1, m2) = dk e 1
2 2 2 ix·k
= d2 τ λ−D/2 e−E
2 [(k + 1 2
2 p) + m 2 1
]
1 2 [(k − 1 2
2 p) + m 2
2 ]

E= 11 2
2 λx + ix · 12 βp + 18 λ(1 − β 2)p2 + 14 λ[(m21 + m22 ) + β(m21 − m22 )]
414 VIII. GAUGE LOOPS

we keep only linear order in x and p2 + m2, and set m1 = m, m2 = 0 (switching back
to α = 12 (1 + β)):

E ≈ ix · (α − 12 )p + 12 λα(1 − α)(p2 + m2 ) + 12 λm2 α2

To clearly separate UV divergences (from λ ≈ 0) and IR divergences (from α ≈ 0),

we scale

λ
λ→ ⇒ E ≈ (α − 12 )ix · p + 12 λ( α1 − 1)(p2 + m2) + 12 λm2
α2
Z ∞ Z 1
−1 −λm2 /2
à ≈ dλ λ e dα α−2 [1 − (α − 12 )ix · p][1 − 12 λ( α1 − 1)(p2 + m2)]
0 0

The integrals are easily performed in either order:


1 2 − 1 1 1
à ≈ Γ(1 + )( 2 m ) + 12 ix · p
U V 1 − 2 (1 − )(1 − 2)
    
1 1 p2 + m2 1 1 1 p2 + m2
+ 1
+ + 1
+ 3
− 1
ix · p
2
IR 1 − 2 m2 4
IR 1 − 2 2 1 − 
2
m2
and in the limit  → 0,
   2
1 1 p + m2
à ≈ Γ(1 + )( 12 m2)− +2+ 1
2 ix ·p+ 1
2 +1
U V IR m2
  
1 p2 + m2
+ 1
4
+ 1 ix · p
IR m2
The electron propagator correction to linear order in m
√
2
− p/ is then
 −    2 
m2 1 p + m2
A2e ≈ Γ(1 + ) 1
2 + 1
4
+1 p/
µ2 IR m2
  2  
1 p + m2 1 1
+ +2+ +1 ( 2 p/ − 2 √2 )
1
2
m
U V IR m2
 2 −      
m 1 1 1
≈ Γ(1 + ) 1 m
(− 2 ) √2 3 +5 + +2 + 5 ( √2 − p/)
m
µ2 U V U V IR
The 1/U V terms are the same as the 1/ terms obtained above for minimal subtrac-
tion. In the MOM scheme, this entire contribution (O(K 0 ) and O(K 1 )) is canceled
by counterterms.
 A. PROPAGATORS 415

2. Photon

We next calculate the spin-1/2 contribution to the photon (or gluon) self energy:
The S-matrix element is
Z tr[−γa(k/ − 12 p/ + √m2 )γb (k/ + 12 p/ + √m2 )]
dk 1
2 [(k − 2 p) + m ] 2 [(k + 2 p) + m ]
1 2 2 1 1 2 2

The result of the trace (again using 4-dimensional algebra) is

−2[kakb − ηab 21 (k 2 + 14 p2 + m2)] − 12 (ηab p2 − pa pb )

The first part is the expression appearing in Âab in subsection VIIB5, once we recog-
nize its ηab terms as the average of the denominator factors, yielding tadpoles. The
integral thus gives

(ηab p2 − pa pb )(−2Ǎ − 12 Â2) ≈ ( 1 − ln p2 )(− 13 )(ηab p2 − pa pb )

for the divergent and high-energy terms. Using

Aa (−p)(ηab p2 − pa pb )Ab (p) = 12 F ab (−p)Fab(p)

in terms of the linearized field strength F , the corresponding contributions to the

unrenormalized one-loop effective action are (including a factor of 12 for identical
external lines) Z
Γ1 ≈ h̄ 23 dx 18 F ab( 1 − ln )Fab

(neglecting the “−1” part of ln(− )) and the counterterm is thus

Z
∆S =  (− 3 ) dx 18 F abFab
h̄ 2

in the case of QED. For QCD, we must include the group-theory factor tr(Gi Gj )
j
multiplying F iabFab . (Examples will be given in the following subsections.)
This propagator correction is easier to analyze in the MOM scheme than the
electron propagator, since there are no internal massless particles, and thus no IR
divergence to distinguish from the UV one. We therefore just take the explicit ex-
pressions for the integrals from subsection VIIB5 and Taylor expand in p2 about 0
(or actually in 1/β of VIIB5.1a, substituting for p2 only at the end). The low-energy
416 VIII. GAUGE LOOPS

part of the renormalized effective action for the photon, exhibiting the momentum
dependence of the coupling, is then
Z  
1
Γ0+1,2γ,r ≈ dx 8 F
1 ab
+ 2
Fab
e2 15 m2
where we have applied MOM subtraction by canceling constant (infinite and finite)
constributions to the coupling.
Excercise VIIIA2.1
Evaluate this contribution to the unrenormalized effective action to this or-
der. Show that the constant contributions to the coupling (to be canceled by
renormalization) are
1
e2
→ 1
e2
+ 23 [ 1 − γ − ln( 12 m2)]

3. Gluon
The most interesting case is the propagator of the Yang-Mills field, in a theory
of Yang-Mills coupled to lower spins. There is an important simplification in this
calculation in the background field gauge: Writing the classical Yang-Mills Lagrangian
as tr F 2/g 2 , the covariant derivative appears as ∇ = ∂+iA without coupling constant,
so the gauge transformation of A is coupling independent, as in general for the matter
fields. (In terms of a group element g, φ0 = gφ and ∇0 = g∇g−1.) The effective action
is gauge invariant, which means the only divergent terms involving the Yang-Mills
field are the gauge-covariantized kinetic (less mass) terms of the various fields. The
divergences for the non-gauge fields are not so interesting, since they can be absorbed
by rescaling those fields (“wave-function renormalization”), but the divergence of
the tr F 2/g 2 term can be absorbed only by rescaling the coupling g itself. (On the
other hand, if we use ∇ = ∂ + igA, then renormalization of g requires the opposite
renormalization of A to preserve gauge invariance.) Thus this divergence is related
to the UV behavior of this coupling (as discussed in subsection VIIB6, and further
later). The important point is that there is no wave-function renormalization for the
Yang-Mills field (since there is no corresponding gauge-invariant counterterm), so the
coupling-constant renormalization (like mass renormalizations) can be found from
just the propagator correction, while in other gauges one would need also a much
messier vertex (3-point) correction: BRST invariance is not enough to give the result
from a single graph.
 A. PROPAGATORS 417

We now consider the contributions of spins 0 (including ghosts) and 1 (including

gluon self-interactions), and redo the spin-1/2 contribution in a way that resembles
the bosons. It is based on the observation that there is a universal form for the gauge-
covariantized Klein-Gordon equation for spins 0,1/2,1, which can also be shown by
supersymmetry. The kinetic operator in a background Yang-Mills field is

K = − 12 ( − iF abSba )

where now = (∇)2 is gauge covariantized. This form is true in arbitrary dimensions.
For spin 0 it is obvious. For spin 1/2, we use the fact that the one-loop contribution
to the functional integral is the trace of the logarithm of the propagator, as follows
from Gaussian integration,
Z
Dψ Dψ̄ e−ψ̄Kψ = det K = etr ln K

where the trace is over all indices, including the coordinates. Then the contribution
to the effective action from kinetic operator K is 1/2 the contribution from K 2 . (See
also excercise VIA4.2.) We then use (see subsection IIIC4)

−2∇
/ 2 = −2(γ · ∇)2 = −({γ a, γ b } + [γ a, γ b ])∇a∇b = + iS ab Fab

where we have used

(1/2)
Sab = − 12 γ[aγb] , {γa , γb } = −ηab

In the case D=4, this is equivalent to the result obtained in subsection IIIC4 in terms
of just the undotted spinor, but there the 1/2 is automatically included because there
are half as many fields, so the range of the trace is half as big. For spin 1, we use
the result of the background-field version of the Fermi-Feynman gauge: At quadratic
order in the quantum fields, from excercise VIB8.1 we have
1 1
1 2
8
F + 14 (∂ · A)2 → 8
(D[a A b] )2
+ i 4
F
1 ab
[A a , A b ] + 4 (D · A)2

= − 14 A · A − i 12 Aa[Fab, Ab ] = − 14 A · ( − iF abSba )A

where = (D)2 contains only the background gauge field, and in the last step we
have written the quantum field A as a column vector in the group space and the
background fields (like F ) as matrices for the adjoint representation (which replaces
commutators with multiplication), and used the explicit expression

(1)
Sab = |[a ihb] |, ha |b i = ηab
418 VIII. GAUGE LOOPS

To this order in the quantum fields, the kinetic operator for the two ghosts looks just
like that for two physical scalars, but gives a contribution to the effective action of
opposite sign because of statistics.
This method can be used for arbitrary one-loop graphs with external gluons,
and easily generalizes to massive fields; we now specialize to propagator corrections.
There are two kinds of vertices, the spin-0 kind and the vertex with the spin operator.
Since tr Sab = 0, we get only graphs with either 2 spin vertices or none. There is
only one spin graph, with 2 internal free propagators; the 2 spinless graphs include
such a graph but also a tadpole, which vanishes by dimensional regularization in the
massless case. Since the spinless graphs give the complete result for internal spin-0,
their sum is separately gauge invariant; the spin graph is obviously so, since it is
expressed directly in terms of the field strength. (We refer here to the Abelian part
of the gauge invariance, which is all you can see from just 2-point graphs.) As far
as Lorentz index algebra is concerned, we need to evaluate only tr(SabScd ). For the
vector, we have
(1) (1)
tr(Sab Scd ) = 2ηb[c ηd]a

For spin 1/2, the traces are the same as in D=4 except for overall normalization;
using earlier identities, or using the same methods for this case directly,
(1/2) (1/2)
tr(Sab Scd ) = 14 tr(I)ηb[cηd]a

where tr(I) is the size of the spinor.

Excercise VIIIA3.1
Let’s look at other ways to interpret the last two identities:
a Use the double-line notation (subsection VC9) for the defining representation
of the orthogonal group to derive the above expression for the trace of two
S (1)’s.
b Use the fermion action of IIIC4 in terms of just undotted spinors for D=4.
Evaluate
(1/2) (1/2)
tr(Sαβ Sγδ )

using both bra-ket notation and double-line notation for SL(2,C). Show the
result is the same as from vector notation (by relating Fab and fαβ ).
All diagrams will also have a group-theory factor of tr(Gi Gj ) ∼ δij . We’ll be
interested mostly in SU(N) for Yang-Mills theory (as appropriate to describe color in
the Standard Model for N=3, or arbitrary N for applying the 1/N expansion). Then
the most interesting representations are the adjoint (for the gluons and their ghosts)
 A. PROPAGATORS 419

and the defining (for the quarks). As explained in subsection IB2, or as follows from
the double-line notation of subsection VC9, we use the normalization

trD (Gi Gj ) = δij ⇒ trA(Gi Gj ) = 2Nδij

Finally, there are the momentum-space integrals, which have already been eval-
uated in subsection VIIB4 for the massless case (which is sufficient for determining
the high-energy behavior, and thus the UV divergences) and VIIB5 for the massive
case. The integral for the spin graph is the same as that for φ3 theory (using the Sab
vertex from − 12 − 12 iF abSab ). As labeled there, the external line has momentum p
and the internal lines k ± 12 p. Then the vertex factors in the spinless graph with two
propagators are both simply −k (from − 12 = − 12 ∂ 2 + 12 A · (−i∂) + 12 (−i∂) · A + 12 A2),
giving Aab, while the addition of the tadpole, with vertex factor −η, converts it to
Âab . (By comparison, the tadpole graph that was apparently avoided in the Dirac-
spinor calculation of the previous subsection appeared anyway after evaluating the
trace algebra.) This contribution also gets an overall tr(I) factor, simply counting
the number of degrees of freedom. Note that the scalar factor Ǎ that appears in Âab
is the sum of a divergent term proportional to the φ3 graph and a convergent term
that vanishes in the massless case.
We now combine all factors to obtain the contributions to the two-gluon part of
the unrenormalized 1-loop effective action (including the −1 for getting the effective
action from the S-matrix, a −1 for internal fermions, either spin 12 or ghost, the 12
for identical external gluon lines, the 12 for the spinor to compensate for squaring the
propagator, and yet another 12 for identical internal lines if the group representation
was real.) The result is the sum of contributions of the form
Z
Γ1,2g = h̄ tr dx 18 F ab( 12 cR)(−1)2s [ D−1
1
B1(− ) − 4s2 B2(− )]Fab

where cR is the group theory factor from the trace, which for the interesting cases is

2 for N ⊕ N̄ (defining)
cR =
2N for adjoint (real)

This result applies to spins s = 0, 12 , 1, with the understanding that it is the result
for two polarizations, so there is an implicit extra factor of 12 for a single scalar,
while for massive spin 1 (spontaneously broken gauge theories) the third polarization
in the (background-field) Fermi-Feynman gauge is carried by a scalar field. (The
result above for s=1 is the sum of the contributions from the vector field and the two
420 VIII. GAUGE LOOPS

fermionic ghosts.) The functions B1 and B2 are the spinless and spin contributions,
related to the massive φ3 propagator correction Â2 (p2 , m2, m2) as
B2 (p2 ) − B2(0)
B2(p2 ) = Â2(p2 , m2, m2), B1(p2 ) = B2(p2 ) + 4m2
p2

Note that
B1 ≈ B2 ≈ 1

− ln p2
as far as divergent (at D=4) or high-energy (i.e., massless) terms are concerned. Also
note that all contributions exactly cancel if all spins are in the adjoint and have the
same mass, and appear in the ratio 1:4:6 for spins 1 (including ghosts), 12 , 0: For the
massless case, this is N=4 super Yang-Mills, which is also the the massless sector of
the dimensional reduction of the open superstring from D=10. The massless sector of
the reduction of the open bosonic string from D=26 yields Yang-Mills plus 22 adjoint
scalars, which cancels near D=4 up to finite terms.
In examining the contribution of this term to the running of the coupling constant
with energy, we see that the vectors contribute with opposite sign to lower spins. In
particular, in terms of the coefficient β1 (of subsection VIIC1), only nonabelian vec-
tors make positive contributions (since Abelian vectors are neutral). This means that
nonabelian vectors are responsible for any weakening in a coupling at high energies,
known as “asymptotic freedom”, an important experimental feature of the strong
interactions (see section VIIIC). Note that while the sign of β1 for φ4 theory, using
the method of subsection VC9, is independent of the coupling (since all 1-loop correc-
tions are coupling-independent when the coupling appears as an overall factor in the
classical action, like h̄), changing the sign of the coupling changes its sign relative to
β1: The result is that this theory can be made asymptotically free only if its potential
has the wrong sign (negative for large φ). Thus, although nonabelian vectors are
required for asymptotic freedom in physical theories, “wrong-sign φ4 ” can be used
as a toy model for studying features associated with asymptotic freedom (especially
resummation of the perturbation expansion: see section VIIC).
Note that for (massless) fermions that couple chirally to vectors (as in electroweak
interactions), cR consists of the contribution from a complex representation but not its
complex conjugate: Only one of the two Weyl spinors of the Dirac spinor contributes.
The result is that the contribution to the vector propagator is half that of the parity-
invariant case. This fact follows from comparing the calculations of the chiral and
nonchiral cases without the squared-propagator trick: In Dirac (4-component) spinor
notation, the γ−1 ’s drop out of the calculation; in Weyl (2-component) spinor notation,
the left- and right-handed-spinor diagrams are identical except for (internal) group
 A. PROPAGATORS 421

theory. (Things are more complicated for higher-point functions, because the group
theory gives more than just trR(Gi Gj ): See subsection VIIIB3.)
Excercise VIIIA3.2
Find the conditions for exact cancelation if spins 12 and 0 include both adjoint
and defining representations. Find the weaker conditions if only the divergent
(and therefore also high-energy) terms cancel.
Excercise VIIIA3.3
Use the optical theorem to find the decay rate for a massive vector (e.g., Z
boson) into massive particle-antiparticle pairs of various spins.
Excercise VIIIA3.4
Find the propagator correction for internal particles of different masses on
each of the two lines (e.g., for a W boson propagator).
In the case of QCD, with color gauge group SU(Nc ) and Nf flavors of quarks in
the defining representation of color, the divergent and high-energy contributions to
this term in the unrenormalized 1-loop effective action are
Z
Γ1,2g,QCD ≈ h̄ tr dx 18 F ab 13 (2Nf − 11Nc )( 1 − ln )Fab

At higher loops the effective action will still be gauge invariant in background-field
gauges (for the quantum fields), so the renormalization of the Yang-Mills coupling
can still be determined from just the gluon propagator correction. On the other hand,
in other gauges a three-point vertex must also be calculated: It can be shown that
the gauge-fixed classical action, including counterterms, is BRST invariant only up
to wave-function renormalizations; i.e, the most general counterterms needed (with
a BRST preserving regularization) are BRST-invariant terms with additional mul-
tiplicative renormalizations of the quantum fields. Thus, BRST invariance, unlike
gauge invariance, is not strong enough to relate the gluon coupling and wave-function
renormalizations. Not only does this mean evaluating many more graphs, but graphs
which make the propagator correction look easy by comparison. (This is not so dif-
ficult for just the one-loop divergences we have considered, but the difficulty grows
exponentially with the number of loops.)
However, in the background-field gauge the L-loop propagator correction has
(L − 1)-loop vertex subdivergences, similar to those in other gauges. The net re-
sult is: (1) We still have a BRST-invariant “classical” action, containing the same
(counter)terms that appear in other gauges (including quantum ghosts), but covari-
antized with respect to background gauge fields (and including coupling to other
422 VIII. GAUGE LOOPS

background fields). However, the coefficients need be calculated only to order L − 1

for the L-loop effective action, one loop less than in other gauges. (2) In addition,
we have background-field-only terms in the classical action whose L-loop coefficients
do need to be calculated, but with a relatively small amount of additional effort,
due to gauge invariance. Thus renormalization consists of two steps: (1) adding
BRST-invariant counterterms for the quantum fields (background covariantized) to
cancel subdivergences, and (2) adding gauge-invariant counterterms for the back-
ground fields (which can be interpreted as vacuum renormalization for the quantum
fields) to cancel superficial divergences. Consequently, background-field gauges save
about one loop of difficulty as far as renormalization is concerned. Furthermore,
similar simplifications occur for calculations of finite parts (e.g., effective potentials),
because of simplifications from gauge invariance.

4. Grand Unified Theories

The best result of GUTs is their prediction that the gauge couplings of the Stan-
dard Model coincide at some high energy, as a consequence of the running of the
couplings with energy. (Mixed results have been obtained for masses, arguably be-
cause renormalization group arguments are accurate only for high energies, and thus
leptons with large masses. A “failed” prediction is proton decay, which has already
eliminated the nonsupersymmetric SU(5) model with minimal Higgs.) The numerical
details of this prediction are model dependent (and thus easy to fudge, given enough
freedom in choice of nonminimal fields), but the fact that all three couplings come
close together at high energies is already strong evidence in favor of unification.
Thus we make only the crudest form of this calculation, using only the one-loop
results of the previous subsection. The main assumption is that there is a “desert”
between the Standard Model unification scale (around the masses of the intermediate
vector bosons W and Z) and the Grand Unification scale MGU T , with no fundamental
particles with masses in that range (although, of course, a huge number of hadrons
appear there). This allows us to crudely approximate all fundamental particles below
that region (i.e., those of the Standard Model) as massless, and all above as infinitely
massive. In particular, in the framework of the minimal SU(5) GUT, this means all
the fermions are treated as massless.
Therefore the calculation is to use the one-loop results to calculate the running
of the couplings in the Standard Model, and use the relation of the gauge couplings
in the SU(5) GUT to identify those of the Standard Model in terms of that of this
 A. PROPAGATORS 423

GUT. From the previous subsection, the running of the couplings is given by
 2  X
1 1 MGU T
≈ − β 1 ln , β 1 = 2 cR (−1) (4s − 3 )
1 2s 2 1
2 2
g (−p ) g02
−p 2
R,s

1
for two helicities of spin s (with an extra factor of 2 for only 1 helicity of spin 0),
where g0 ≡ g(MGU 2
T ).

If we use g1 , g2 , g3 to label the couplings of U(1), SU(2), and SU(3) that are
identified with the single SU(5) gauge coupling at the unification scale, then their
relation to those of the Standard Model (as normalized in excercise IVB2.1) is
2
1 6 1 6 cos θW 1 1 sin2 θW 1 1
= 5 02 = 5 , = 2 =2 , = 2
g12 g e2 g22 g e2 2
g3 gs

where gs and g the usual SU(3) and SU(2) couplings, and the factor of 65 is because
the U(1) generator (see subsection IVB4) satisfies trD (G2 ) = 56 in terms of SU(5)
matrices. (We generally normalize to trD (G2 ) = 1 for each generator. Physical
couplings are preserved if changes in normalization of generators are accompanied by
changes in coupling normalization so as to preserve gi Gi .)
Then the values of the β1’s for the Standard Model are

β1,1 = 0 − 4 − 1
10
= − 41
10
, β1,2 = 22
3
−4− 1
6
= 19
6
, β1,3 = 11 − 4 + 0 = 7

where we have listed the contributions from spins 1, 12 (for 3 families), 0, respec-
tively. (Note that the spinors contribute the same to each because they are all
effectively massless: They don’t notice the SU(5) breaking. Also, we can ignore
SU(2)⊗U(1) breaking when calculating these β’s, since we have neglected the corre-
sponding masses.)
Excercise VIIIA4.1
Calculate the contribution of the spinors to the β1’s, in terms of both SU(5)
and SU(3)⊗SU(2)⊗U(1) multiplets. (Note the chiral couplings for spinors,
so for cR a complex representation and its complex conjugate might not both
contribute.)
The experimental values of the couplings (in the MS prescription) at µ = MZ ≈
91 GeV are
1 1
2
≈ 804, sin2 θW ≈ .231, ≈ 106
e gs2
Unfortunately, taking any two of the equations for 1/gi2 gives widely varying answers:
e.g.,
MGU T ≈ 1015±2 GeV
424 VIII. GAUGE LOOPS

Alternatively, since we have used only two parameters to fit three experimental num-
bers, we can try to predict the value of any one of e, θW , or gs from the rest: e.g.,
from e and gs we can find
sin2θW ≈ .207

which shows the same disagreement (but looks better than the exponentiated error
for MGU T ).
The result is not very accurate, since we have made many approximations, which
can be improved with some effort: Two-loop corrections add ln ln terms to the one-
loop ln terms; including the mass dependence of the effective couplings also adds
significant corrections. But the most important approximation assumption we made
was the desert: Undiscovered particles, such as new fermions, nonminimal Higgs, or
supersymmetric partners, change even the one-loop expressions β1 . Specifically, since
by definition the unification scale is where the masses of all unobserved vectors reside,
these new particles will all have spins 0 or 12 , and thus make the β’s more negative.
In particular, supersymmetrization yields a result consistent with experiment, with

MGU T ≈ 2.2 × 1016 GeV

(This has been interpreted as the only experimental verification of supersymmetry.)

Excercise VIIIA4.2
Let’s examine the effects of supersymmetry:
a Supersymmetrize the Standard Model contributions to β1 by adding the su-
persymmetric partners to each spin: 1 → 1 ⊕ 12 , 12 → 12 ⊕ 0 ⊕ 0, 0 → 12 ⊕ 0 ⊕ 0
(where the Higgs scalars have doubled because chiral scalar superfields can’t
satisfy reality conditions) to find the result

β1,1 = 0 − 6 − 3
5
= − 33
5
, β1,2 = 6 − 6 − 1 = −1, β1,3 = 9 − 6 + 0 = 3

b Solve for 1/g02 , ln(MGU

2 2 2
T /MZ ) (and thus MGU T ), and sin θW in terms of 1/e
2

and 1/gs2 . Then plug in to find the numerical values.

c Show the consistency condition relating the 3 couplings is
β1,2 − β1,3 β1,3 − β1,1 β1,1 − β1,2
+ + =0
g12 g22 g32
and that the closest integer values for the couplings from the above data,
1 1 1
= 742, = 371, = 106
g12 g22 g32
 A. PROPAGATORS 425

satisfies it exactly. (OK, so this is just a numerical coincidence, considering

experimental inaccuracies and theoretical approximations, but isn’t it still
nice?) Also, note that
1 1
2
=2 2 ⇒ sin2 θW = 3
13
g1 g2

d Drop the contributions of the Higgs (and its superpartners) to the β’s in both
the supersymmetric and nonsupersymmetric cases, and reevaluate sin2θW ,
showing both give the same (poor) value. (Thus, Higgs can make a difference.)

5. Supermatter
Although the problem with infrared renormalons may be only technical, the ap-
pearance of this same problem in several different approaches (including a nonpertur-
bative one; see later) strongly suggests that the “correct” approach to quantum field
theory, in the sense of a practical method for unambiguously (i.e., with predictive
power) calculating perturbative and nonperturbative effects, might be to consider
only theories that are perturbatively finite. In this subsection we will analyze gen-
eral properties of supersymmetric field theory using superspace, and in particular
improved UV behavior, concentrating on finite theories.
Finite supersymmetric theories must be in particular one-loop finite. This turns
out to be enough to guarantee finiteness to all loops: Two-loop finiteness is automatic,
while an appropriate renormalization prescription is required to guarantee finiteness
is preserved order by order in perturbation theory. (No constraints on the coupling
constants are needed beyond those found at one loop, but without the renormalization
prescription infinities cancel between different loop orders.) Of course, wave-function
renormalizations are gauge dependent: N=1 supersymmetric gauges eliminate some
of these unphysical divergences (and gauges with higher supersymmetry more), as
do background-field gauges even in nonsupersymmetric theories. So, “finite theory”
in general gauges refers only to the “physical” divergences — those that affect the
high-energy behavior of the theory, namely those that appear in couplings and masses.
Because of the nonrenormalization of chiral terms in the action (see subsection
VIC5), it might seem that the corresponding couplings and masses are always un-
renormalized. However, the kinetic terms of chiral superfields can receive quantum
corrections, and the true couplings are defined by field redefinitions that eliminate
these rescalings. This means that all such renormalizations are related, and given
by the wave function renormalizations. The only other couplings are the Yang-Mills
426 VIII. GAUGE LOOPS

ones, whose renormalization is also given by kinetic terms in background-field gauges.

Thus, all “physical” renormalizations in supersymmetric theories can be found from
just propagator corrections. In particular, this means that if the effective action is
calculated with background-field supergraphs, then it is completely finite in a finite
theory.
A possible exception to our statement of all physical renormalizations coming from
R
propagator corrections would seem to be the Fayet-Iliopoulos tadpole term d4 θ V .
However, massless tadpoles vanish in dimensional regularization, and massive ones
require real representations, which cannot generate explicit-prepotential terms. (In
particular, at more than one loop such terms never appear in the background-field
gauge for any representation.)

The simplest one-loop propagator correction is to φ̄φ. (The φφ correction van-

R
ishes, since d4 θ φ2 = 0: See subsection VIC5.) There are two graphs to consider,
one with two internal φ̄φ propagators, and one with internal φ̄φ and V V propagators.
The d algebra for the two graphs is identical: Both get a d2 and a d¯2 inside the loop,
exactly enough to give a nonvanishing graph (using [d̄2d2 δ 4(θ − θ0 )]|θ0 =θ = 1). There
is also a symmetry factor of 12 for the two φ̄φ propagators, and a −1 for the mixed
graph because the two different types of internal propagator have opposite sign (and,
as usual, an overall −1 to get Γ from the T-matrix). Thus, the supersymmetry (spin)
part of the algebra is almost trivial in this case.
On the other hand, the internal group theory is slightly messy, so we treat the
general case immediately: We take vector multiplets V i for an arbitrary group (though
we will need a semisimple group for finiteness, since Abelian groups are not even
P
asymptotically free). Sums G are over each simple subgroup (or each Abelian
factor), since they can have independent coupling constants gRG for representation R
(especially gAG for the adjoint, which we use for the pure super Yang-Mills term for
definiteness; except for the Abelian factors, where a nontrivial representation should
P
be substituted). Similarly, sums R are over irreducible representations of the group;
ΠRI J is the corresponding projection operator. For the simple (or single-component
Abelian) factors of the group
ηGij = cAG ΠGij
is used (but again, with a different normalization for the Abelian factors). We also
use the group theory identities (normalizations) from subsection IB2, now generalized
 A. PROPAGATORS 427

to these nonsimple groups and reducible representations:

X X cRG
GiI K GjK J ΠRJ I = cRG ΠGij = ηGij
G G
cAG

ij
X X cRG dAG
GiI K GjK J ηG = kRG ΠRI J = ΠRI J
cAG dR
R R

Then from the Lagrangian

Z Z 
L = − d θ φ̄ (e )I φJ +
4 I V J 2 1 IJK
d θ 6 λ φI φJ φK + h.c.

X 1 Z
− 2
d2 θ 12 W iαWαj ηGij
gAG
G

(ignoring mass terms) the result is simply

Z X
2 cRG dAG
Γ1,φφ̄ = h̄ d4 θ φ̄I MI J Â2φJ , MI J = gAG ΠRI J − 12 λ̄IKL λJKL
c d
AG R
R,G

where again Â2 is the operator representing the one-loop propagator correction (T-
matrix) for self-interacting scalars. (Of course, this operator may vary depending on
the internal masses; here we are concerned mostly with the divergences and leading
high-energy behavior, which is mass-independent. As usual, we can rescale the gauge
fields by their couplings in the Lagrangian; this moves these couplings from the prop-
agators into the vertices, giving the same result for this term in Γ, since it has no V ’s.)
Of course, this is the identical group theory that appears in the nonsupersymmetric
case; we have been more general here because we want to consider exact cancelation,
while in the nonsupersymmetric case simplicity is usually more important.

6. Supergluon
The supergluon self-energy calculation is similar to the nonsupersymmetric cases
considered in subsections VIIIA2-3. Examining the Feynman rules, we see that those
for the vector multiplets are similar to the nonsupersymmetric ones for vectors (as
expected), while those for the scalar multiplets are similar to those for spinors: d2
and d̄2 are analogous to (in 2×2 matrix notation) ∂ and ∂*, etc.
There are now only two kinds of loops to consider, vector and scalar multiplets:
As for the nonsupersymmetric case, ghosts in background-field gauges couple the
same as matter, since at one loop the only coupling is to background fields and thus
covariant, even for ghosts. For the real scalar superfield describing the quantum
428 VIII. GAUGE LOOPS

vector multiplet, looking at the terms in the action quadratic in V (from subsection
VIB10) Z .
α
S2V = dx d4 θ 14 V ( + 2iW α Dα + 2iW D̄α.)V

we see that vertices have only 1 spinor derivative at most. However, we need at
least 4 spinor derivatives (2 d’s and 2 d’s) ¯ per loop (see subsection VIC5), since the
result of reducing any loop to a point in θ space always leaves the tadpole θ-integral
[d...dδ 4(θ − θ0)]|θ0 =θ , which vanishes for fewer than 4 derivatives. Thus, a V loop in
a super Yang-Mills background vanishes for fewer than 4 external lines. This means
the entire contribution of quantum super Yang-Mills to the supergluon propagator
correction (or 3-point correction from real representations) in the background-field
gauge comes from the 3 ghosts (including the Nielsen-Kallosh ghost), which couple
the same as −3 scalar multiplets in the adjoint representation. Thus, for example,
we see without evaluating a single graph that this correction vanishes for N=4 super
Yang-Mills, which has also 3 physical adjoint scalar multiplets. (See subsection IVC7.)
For the scalar multiplets, we can find the analog of the squared-propagator trick:
The easiest way is by the method of subsection IIIC4, which automatically takes
care of factors of 12 , and can be applied classically, without worrying about functional
determinants. This method requires we consider the massive theory at intermediate
stages of the calculation, although the mass can be dropped at the end. The only
resulting limitation is that we must restrict to real representations of the gauge group.
(In other words, the couplings must preserve parity: For these terms, CP invariance is
automatic, and reality means C invariance, so P invariance is implied.) However, this
is a restriction of the usefulness of the squared-propagator trick anyway: Otherwise we
get expressions like (∂/ + iA/ )(∂/ − iA
/ *) which do not yield useful simplifications. (They
require as much work as without the trick.) In such cases we are stuck with doing the
calculations the hard way. This is not just a technical difficulty, it is a consequence
of the final result being messier in such cases: For example, for real representations
there is no possibility of anomalies. However, we can separate the generators into
the real (scalar) and imaginary (pseudoscalar) ones: Then this trick simplifies the
real (polar vector) couplings but not the imaginary (axial vector) couplings. (As for
Pauli-Villars in subsection VIIIB2 below, but also for the physical fields before taking
the mass to vanish after the trick has been applied, the mass term can be chosen to
preserve the polar symmetries and thus violate the axial ones.)
However, by comparison of the propagator correction for complex and real rep-
resentations without (the supersymmetric version of) the squared-propagator trick,
we see that the only difference between the two is in the (Yang-Mills) group theory.
 A. PROPAGATORS 429

Thus, we can calculate for real representations first, using the trick, and then for
complex representations by simply replacing the group-theory factor in the result for
the real ones.
Repeating the procedure of subsection IIIC4 with spinors replaced with chiral
R
superfields, we begin with the Lagrangian (S = dx L)
Z Z Z 
L = − d θ φ̄φ + √2
4 m 2 1 2 2 1
d θ 2 φ + d θ̄ 2 φ̄ 2

where the chiral superfields are covariantly chiral (or background-covariantly chiral)

∇α.φ = ∇αφ̄ = 0

Treating φ̄ as auxiliary (the φ̄2 term has no Yang-Mills coupling, as can be seen, e.g.,
in an “antichiral” representation), we eliminate it by its algebraic field equation
√
φ̄ = m
∇φ
2 2

After a trivial rescaling

√
φ → 2−1/4 mφ
R R 2
(and using d4 θ = d2 θ ∇ ) we obtain the action
Z Z
2 2
Lφ = − d θ 2 φ(∇ ∇ − 2 m )φ = − d2 θ 14 φ(
2 1 1 2
− m2 + i[W α, ∇α])φ

( = ∇a ∇a) using an identity from subsection VIC5.

In the chiral vacuum-bubble loop, we no longer have an explicit chiral superfield
2 2
to convert ∇ ∇2 to + .... However, using the chiral representation ∇ = d̄2 , we can
write the kinetic operator as

d̄2 ∇2 = d̄2 d2 + d¯2 (∇2 − d2 )

to separate the truly free part from the background interactions. Then quantization
can be preformed as usual (see subsection VIC5): Essentially, we can now use the
free − 0 + m2 = p2 + m2 as kinetic operator, since at each vertex there is a d̄2 to
project back to chiral superfields. Of course, in general we need only one projector in
any trace over a subspace: In this case that result is obtained by integrating the d¯2 ’s
by parts in the loop back and forth across the free propagators, since sandwiching
any ∇2 − d2 between them produces

d̄2 (∇2 − d2 )d̄2 = 12 ( − 0 + i[W α, ∇α])d̄2

430 VIII. GAUGE LOOPS

Repeating the procedure till only one d¯2 is left, the Feynman rules for this loop become
1
propagator : 1 2 δ 4(θ − θ0 )
2 (p +m2 )
one vertex : d̄ (∇ − d2 )
2 2

other vertices : 1
2( − 0 + i[W α, ∇α])

We thus see that one vertex has at most 3 derivatives (d¯2 d) while the other has
at most 1 (d):
d̄2 (∇2 − d2 ) = d¯2 [iAαdα + 12 i(dαAα ) − 12 AαAα ]
1
2( − 0 +i[W
α
, ∇α]) = iW α dα + 12 i(dαWα )− 12 [W α, Aα] +iAa∂a + 12 i(∂ aAa)− 12 Aa Aa

exactly the minimum needed. (Thus, there are insufficient derivatives for a tadpole
contribution to the propagator.) The result for this diagram is then the same as the
corresponding diagram in bosonic ϕ3 theory, with a group theory factor tr(Gi Gj ),
and replacing ϕ(−p)ϕ(p) with
Z
d4 θ d4 θ0 [iW iα(−p, θ0 )d0α δ 4(θ − θ0 )][iAjβ (p, θ)d¯2 dβ δ 4(θ − θ0 )]
Z Z
4 iα j
= dθ 1
2 W (−p, θ)Aα (p, θ) = d2 θ 12 W iα (−p, θ)Wαj (p, θ)

using dδ = −d0 δ, integration by parts, [d̄2d2 δ 4(θ −θ0)]|θ0 =θ = 1, and Wα = d¯2 Aα (chiral
representation). Written in the notation of subsection VIIIA3, the 2-supergluon part
of the unrenormalized 1-loop effective action is then
Z
Γ1,2sg = −h̄ tr dx d2 θ 12 W α ( 12 cR )Â2 Wα

for a scalar multiplet, and exactly −3 times that for a vector multiplet, including the
massive case. Thus, cancelations again survive the introduction of masses. Also, if
the masses of the various scalar multiplets are equal the entire propagator correction
is canceled in such theories, while for unequal masses only the divergence, and the
corresponding leading (logarithmic) high-energy term, is canceled.
Excercise VIIIA6.1
Show this result agrees with the restriction to N=1 supersymmetric theories
of the component result of subsection VIIIA3.
Excercise VIIIA6.2
Take the result of subsection VIIIA3 literally for all spins s (arbitrarily large).
Using the fact that multiplets with N+1 supersymmetries can be written as
2 multiplets with N supersymmetries, differing in maximum helicity by 1/2,
 A. PROPAGATORS 431

recursively find the result for general s (now labeling maximum helicity) for
all values of N≥ 1, and show it vanishes for N≥ 3.
Excercise VIIIA6.3
Calculate the chiral scalar contribution to the one-loop supergluon propagator
correction without the squared-propagator trick. (Hint: There are 8 spinor
derivatives in the loop. Integrating them by parts off one propagator produces
3 terms, since the number of d’s and d̄’s inside must be equal, because what’s
left is always spacetime derivatives on [d̄2 d2 δ 4(θ − θ0 )]|θ0 =θ .)
Generalizing the group theory as in the previous subsection, we have the total
result
Z X X cRG
Γ1,V V = −h̄ dx d2 θ 2 W ( 2 MG )ηGij Â2 Wα ,
1 iα 1 j
MG = −3
G R
cAG

(For Abelian factors, irrelevant for finiteness, we should take the cAG factor out of
P
ηGij and put it into MG ; then cAG = 0 for Abelian groups, so MG → R cRG > 0.)
Therefore, combining with the results of the previous subsection, the conditions for
finiteness are
X cRG X cRGdAG
2
= 3, gAG ΠRI J = 12 λ̄IKL λJKL
R
cAG R,G
cAG dR

In particular, for the case of N=4 super Yang-Mills written in terms of N=1 superfields
(see subsection IVC7), we have 3 adjoint chiral scalars φI with I = iI 0, where i is
the adjoint label and I 0 = 1, 2, 3 (which appeared as the label I in subsection IVC7,
where the adjoint label was implicit in matrix notation). Then
0 0 0
λIJK = gA f ijk I J K

and the above two finiteness conditions reduce to

0 0 0 0 0
δII0 = 3, δIJ0 = 12 I 0 K 0 L0 J K L (δij = fikl f jkl )

As explained in the previous subsection, in the general case the finiteness conditions
may receive quantum corrections at 3 loops and beyond, depending on the model and
renormalization prescription, but no new conditions are added.
Presently there is no deep understanding for the finiteness of these models (at
least, not deep enough to always avoid the quantum corrections to the finiteness con-
ditions). Note that they are finite for arbitrary values of the couplings, up to the
two above restrictions: For example, we can scale all the couplings by a common
432 VIII. GAUGE LOOPS

factor. Thus, they are finite order-by-order in perturbation theory (loops). Non-
supersymmetric theories can also be finite, but only for specific numerical values of
the coupling, i.e., not for arbitrarily small values of the coupling, and thus not order-
by-order in the loop expansion; they therefore suffer from the renormalon problem.
(The renormalon-like behavior of instantons is not a problem in the framework of the
1/Nc expansion.) The finiteness of theories with extended supersymmetry has been
explained by various arguments (in particular, for N=2 there are no divergences be-
yond 1 loop even for theories that are just renormalizable), but none of these applies
to the general case of simple supersymmetry.
To obtain more realistic models, we may want to consider adding “soft” super-
symmetry breaking terms (those which have little effect on high-energy behavior), as
introduced in subsection IVC6, to these finite theories. Finiteness can be maintained,
but the conditions become considerably more complicated in the general case. Note
that spontaneous breaking of supersymmetry is not allowed, because the first condi-
R
tion prohibits U(1) factors (with cAG = 0; thus no d4 θ V terms), while the second
R
prohibits gauge-singlet matter (with cRG = 0; thus no d2 θ φ terms).

7. Bosonization
A common method in field theory is to consider simpler models where calculations
are easier, and see if they are analogous enough to give some insight. In particular,
two-dimensional models sometimes have perturbative features that are expected only
nonperturbatively in four dimensions: For example, we saw in subsection VIIB3
the generation of bound states at one loop in the 2D CP(n) model. Of course,
some of the features may be misleadingly simple, and may have no analog in D=4.
Two-dimensional theories, especially free, massless ones, are also useful to describe
the quantum mechanics of the worldsheet in string theory (see chapter XI). In this
subsection we consider free, massless 2D theories: Essentially, this means just the
scalar and the spinor, since there are no transverse dimensions to give gauge fields
nontrivial components.
Spinor notation is very simple in D=2, since the Lorentz group is SO(1,1)=GL(1).
For that purpose it’s convenient to use lightcone notation. 2D γ matrices can be
chosen as

0 −i



√
γ+ = 00
, γ− = , γ−1 = √1 −i 0 ; Υ= 0 −i
= 2γ0
i0 0 0 2 0 i i 0
 
ψ⊕
Ψ= , Ψ̄ = ( iψ̄ −iψ̄⊕ )
ψ
 A. PROPAGATORS 433

In general, even-D γ matrices can be constructed as direct products of D/2 sets of

2D γ matrices, so tr(I) = 2D/2 . (For details, see subsection XC1.)
The Lagrangian for a massless, complex spinor can be written this way as

L = Ψ̄i∂/ Ψ = ψ̄⊕(−i∂ )ψ⊕ + ψ̄ (−i∂⊕⊕)ψ

(This also follows from truncation of 4D spinor notation.) Note that ψ⊕ and ψ
transform independently under proper Lorentz transformations, as do their real and
imaginary parts. Thus, we can not only impose a reality condition, but also a chirality
condition, dropping ψ⊕ or ψ : A single real component is enough to not only define
a spinor Lorentz representation, but also construct an action.
In position space, the propagator for a massless scalar is (see excercise VIIB1.1)
1
2π(−i)δ 2(x − x0) = −ln[(x − x0 )2]
− 12

up to a real, dimensionful constant: We use units µ = 1. The “−i” for the δ comes
from Wick rotation back to Minkowski space. The propagator for a massless spinor is
then ( = −2∂+∂− , shortening the double ⊕⊕’s and ’s from spinor to lightcone
vector notation)
1 −i
2π(−i)δ 2(x − x0) = i∂±{−ln[(x − x0 )2]} =
−i∂∓ (x − x0)±

which has the unusual consequence (its wave equation)

1
∂± = 2πiδ 2(x − x0)
(x − x0)∓

with an appropriate -prescription implicit:

Excercise VIIIA7.1
Show (e.g., by an infinitesimal Wick rotation) that the correct i prescription
for the spinor propagator is
−i −i −i
= θ(t − t0 ) + θ(t0 − t)
(x − x0)± − i(t − t )
0 (x − x ) − i
0 ± (x − x0)± + i

and that it satisfies the wave equation. (t − t0 can be replaced with (x − x0 )∓

in the above.)
In D=2, the theory of a massless spinor field is equivalent to that of a massless
scalar field (“bosonization”). The explicit correspondence between the spinor and
scalar is shown by separating the scalar into its left-propagating and right-propagating
434 VIII. GAUGE LOOPS

parts: This can be accomplished by differentiating with respect to x± and then inte-
grating back. We thus write
φ = φ(+) + φ(−)

where φ(±) is a function of only x± and not x∓ on-shell, and has propagator −ln x±.
The product of two fermion propagators, for ψ̄(x)ψ(x0) and ψ̄(x0)ψ(x), as would
follow from multiplying ψ̄(x)ψ(x) with ψ̄(x0)ψ(x0) in the functional integral (including
the −1 for the fermion “loop”, from reordering the ψ’s) is
  
−i −i
− = ∂±∂±0 {−ln[(x − x0)2 ]}
(x − x0)± (x0 − x)±

This shows that ∂±φ(±) has the same propagator as ψ̄α ψα, with α = (⊕, ) not
summed, so we can equate
∂±φ(±) = ψ̄αψα

which gives an explicit expression for the scalar in terms of the spinor after integration
over x± .

The inverse relation is (quantum mechanically, not classically)

ψα = e−iφ(±)

which can be checked by a sum of multiloop diagrams: The equivalent of connecting

0
ψα(x) to ψ̄α(x0) by a single propagator is to expand e−iφ(±) (x) and eiφ(±) (x ) each to nth
order and connect x and x0 with n φ(±) propagators, and sum over n (with appropriate
combinatoric factors). The φ(±) propagators are

−ln[i(x − x0)± ]

with the “i” determined by Wick rotation. The propagator of the composite fermions
is then:
X∞
−i
1
{−ln[i(x − x0 )±]}n =
n=0
n!
(x − x0 )±

Although this gives the appearance of a scalar being the bound state of spinors,
and vice versa, even in the free theory, there is a simpler interpretation, even clas-
sically: Massless particles in D=2 travel at the speed of light in one of two possible
 A. PROPAGATORS 435

directions. Thus, a collection of free “left-(or right-)handed” massless particles travels

along together, not separating, and thus acting like a bound state. (As shown in sub-
section VC8, singularities in perturbative quantum field theory directly correspond
to configurations in classical mechanics.)
Bosonization extends to massive fermions: The “massive Thirring model”
m
√
L = ψ̄⊕(−i∂ )ψ⊕ + ψ̄ (−i∂⊕⊕)ψ + 2
(ψ̄⊕ψ + ψ̄ ψ⊕) + g ψ̄⊕ψ̄ ψ⊕ψ

is equivalent to the “sine-Gordon model”

L= [ (∂φ)2
1 1
β2 4
+ 12 µ2 (1 − cos φ)]

with the above relation between the spinor and scalar fields, and
1 µ2
= 1 + 2g, ∼m
β2 β2
(Note in particular the free massive fermion for β = 1.) In this case the bound states
are dynamical. Note that the relation is between strong coupling in one theory and
weak in the other (“duality”).

8. Schwinger model
The simplest interacting model in D=2 is the “Schwinger model”, massless QED.
This theory is even simpler than scalar theories because its interactions occur only
through a massless gauge vector, which has no physical polarizations in two dimen-
sions (D−2=0).
The most interesting feature of the Schwinger model is that all amplitudes with
external vectors can be calculated exactly. In fact, the only nonvanishing 1PI vector
amplitude is the one-loop propagator correction, which gives just a mass term. In
that sense the theory is trivial, and describes just a massive vector. However, the
methods of calculation are instructive. We first consider some simple methods of
calculation of just the propagator correction, and then show that it is the only 1PI
vector graph. One method we have already considered is dimensional regularization;
from subsection VIIIA2-3 we have the contribution to the effective action (correcting
for the 2D normalization tr(I) = 2)
Z
1
Γ1 = dx F F

where we write Fab = ab F in D=2. Although this calculation needs no renormal-
ization, regularization is still necessary to allow naive manipulation of the integrand:
436 VIII. GAUGE LOOPS

Using dimensional regularization, we see from the result of subsection VIIIA3 that
1
we get a factor of D−1 − 4s2 ∼  in D = 2 + 2, canceling the 1/ pole from the scalar
integral.
It can also be calculated in position space, using the methods of the previous
subsection. The Lagrangian in lightcone notation is

L = − 4e12 F 2 + [ψ̄⊕(−i∂ + A )ψ⊕ + ψ̄ (−i∂⊕⊕ + A⊕⊕)ψ ]

F = ∂⊕⊕ A − ∂ A⊕⊕

We can calculate separately the contributions of ψ⊕ and ψ to fermion loops. The

“photon” propagator correction consists of the product of two fermion propagators,
as given in the previous subsection. We then find for the effective action (including
another −1 for T → Γ and a 12 for identical external lines), after including a finite
counterterm to restore gauge invariance,
1 1 1
1
2 A− (∂+ )
2
A− + 12 A+(∂−)2 A+ − A+ A− = F F
− 12 − 12

(after integration by parts).

This same calculation also gives the “axial anomaly”: Consider an axial vector
gauge field B that couples to the current ±ψ̄αψα (not summed), in addition to A’s
coupling to ψ̄α ψα. (In D=2, Wa = a b Vb ⇒ W± = ±V± .) The contribution to
the 1-loop effective action with one of each vector externally is, after including a
counterterm to preserve A gauge invariance (and therefore break B gauge invariance),
1 1 1
B− (∂+)2 A− − B+ (∂−)2 A+ − B−A+ + B+A− = (∂ · B) F
− 12 − 12 − 12

The anomaly is the breaking of B gauge invariance,

Z Z
δΓ
δB = ∂λ ⇒ δΓ = − λ∂ · = −2 λF
δB

An anomaly is by definition a quantum effect: As we have seen from the 2D

axial anomaly, it is related to a divergence that violates naive classical arguments,
since the regulator itself violates the symmetry. In the axial case there is no actual
divergent term in the effective action, but a finite term results from a / type of
cancelation. Dimensional analysis immediately reveals that the propagator correction
is the only graph in D=2 that can contribute such a term from the fermion loop.
(Fermion propagators go as 1/p, while the vertex is a constant: The electric charge
has dimension in D6=4.)
 A. PROPAGATORS 437

The complete one-loop effective action for the vectors then follows directly from
the complete anomaly for the axial current, and the vanishing of the anomaly for the
polar current: By separating out the anomalous term in the effective action,
Z
1 δΓ δ(∆Γ)
Γ = F F + ∆Γ; J= , ∆J =
δA δA
∂·J = 0, ∂×J = −2F ⇒ ∂·(∆J) = ∂×(∆J) = 0 ⇒ ∆J = 0 ⇒ ∆Γ = 0
(up to an irrelevant constant), where ∂ × J = ab ∂aJb is the curl of the polar current,
but also the divergence of the axial current. (There are some questions of boundary
conditions in solving the divergence- and curl-free conditions as ∆J = 0, but these
are resolved by working in Euclidean momentum space.)
Similar remarks apply to external gravity: From a similar calculation, replacing
the vector current with the energy-momentum tensor, we find
1
∂m T mn = 0, ∂m m n T np ∼ pm ∂m R ⇒ T m m ∼ R, Γ ∼ R R

where R is the 2D curvature (which is just a scalar, as the vector field strength is
a pseudoscalar). While in the vector case the finite local counterterm was chosen to
preserve polar gauge invariance and thus violate axial, for the tensor case a term is
chosen to preserve local conservation of energy-momentum and thus violate conformal
invariance T m m = 0. (The above expressions are linearized, but the results can be
generalized to fully nonlinear gravity.)
Excercise VIIIA8.1
Calculate the gravitational anomalies from a massless spinor loop in D=2,
using the classical expressions (as follow from dimensional and Lorentz anal-
ysis)
↔
T±± ∼ 12 ψ̄α i ∂ ±ψα , T+− = 0
(If you work in terms of Γ you can define the perturbative field hab such that
δΓ/δhab = T ab .)
The simple form of the effective actions in the Schwinger model is a consequence
of bosonization: Thus, including coupling to electromagnetism and gravity, the action
for the massless spinor is equivalent to

L = − 14 φ φ + (F + R)φ

Integrating out the scalar generates the above effective actions classically.
Excercise VIIIA8.2
The above action is dual to the mass term of the Stückelberg action:
438 VIII. GAUGE LOOPS

a Consider the first-order Lagrangian

L = −G2 + Ga (mAa + ∂aφ)

Eliminating the auxiliary field Ga by its field equation yields the usual mass
term for the Stückelberg model. Show that if we vary φ instead and solve
the resulting constraint on G, we obtain (the nongravitational part of) the
previous action.
b Generalize this construction to D=4, where the field dual to the Stückelberg
scalar is now an antisymmetric tensor gauge field. (See excercise IIB2.1.)

REFERENCES
1 A.A. Slavnov, Theor. Math. Phys. 10 (1972) 99;
J.C. Taylor, Nucl. Phys. B33 (1971) 436:
graphical form of unitarity for nonabelian gauge theories (later simplified as BRST).
2 G. ’t Hooft, Nucl. Phys. B33 (1971) 173, G. ’t Hooft and M. Veltman, Nucl. Phys. B44
(1972) 189, 50 (1972) 318;
B.W. Lee and J. Zinn-Justin, Phys. Rev. D5 (1972) 3121, 3137, 3135, 7 (1973) 1049:
renormalization proof for nonabelian gauge theories.
3 V.S. Vanyashin and M.V. Terent’ev, Sov. Phys. JETP 21 (1965) 375:
first 1-loop gluon coupling renormalization, but neglecting ghosts.
4 I.B. Khriplovich, Sov. J. Nucl. Phys. 10 (1970) 235:
first complete 1-loop gluon coupling renormalization.
5 G. ’t Hooft, unpublished comment after K. Symanzik’s talk at Colloquium on renor-
malization of Yang-Mills fields and applications to particle physics, Marseille, June 19
-23, 1972;
H.D. Politzer, Phys. Rev. Lett. 30 (1973) 1346;
D.J. Gross and F. Wilczek, Phys. Rev. Lett. 30 (1973) 1343:
asymptotic freedom.
6 R.J. Hughes, Phys. Lett. 97B (1980) 246:
analyzed 1-loop gluon propagator corrections in terms of spinless + spin contributions.
7 T.L. Curtright, Phys. Lett. 102B (1981) 17:
extended this analysis to higher spins, supersymmetry, and massless sectors of reduced
strings.
8 Z. Bern, G. Chalmers, L. Dixon, and D.A. Kosower, hep-ph/9312333, Phys. Rev. Lett.
72 (1994) 2134;
G.D. Mahlon, hep-ph/9312276, Phys. Rev. D49 (1994) 4438:
generalization of Parke-Taylor amplitudes to one loop.
9 Z. Bern, L. Dixon, D.A. Kosower, hep-ph/9602280, Ann. Rev. Nucl. Part. Sci. 46 (1996)
109:
review of modern methods for one-loop graphs.
10 H. Georgi, H. Quinn, and S. Weinberg, Phys. Rev. Lett. 33 (1974) 451:
test of running couplings in SU(5) GUT.
11 S. Dimopoulos, S. Raby, and F. Wilczek, Phys. Rev. D24 (1981) 1681:
test of running couplings in supersymmetric SU(5) GUT.
 A. PROPAGATORS 439

12 P. Langacker and N. Polonsky, Phys. Rev. D47 (1993) 4028:

a review of accurate tests of supersymmetric and nonsupersymmetric GUTs, empha-
sizing the improvement from supersymmetry.
13 A. Parkes and P. West, Phys. Lett. 138B (1984) 99;
D.R.T. Jones and L. Mezincescu, Phys. Lett. 138B (1984) 293:
one-loop finiteness implies two-loop finiteness.
14 D.I. Kazakov, Phys. Lett. 179B (1986) 352;
A.V. Ermushev, D.I. Kazakov, and O.V. Tarasov, Nucl. Phys. B281 (1987) 72;
O. Piguet and K. Sibold, Phys. Lett. 177B (1986) 373:
one-loop finiteness implies all-loop finiteness.
15 D.R.T. Jones, L. Mezincescu, and Y.-P. Yao, Phys. Lett. 148B (1984) 317:
one-loop finite theories with soft breaking.
16 I. Jack and D.R.T Jones, hep-ph/9405233, Phys. Lett. 333B (1994) 372:
one-loop finiteness of softly-broken theories implies two-loop finiteness.
17 Avdeev, Kazakov, and Kondrashuk; Arkani-Hamed, Giudice, Luty, and Rattazzi; loc.
cit. (IVC):
one-loop finiteness of softly-broken theories implies all-loop finiteness.
18 S. Hamidi, J. Patera, and J.H. Schwarz, Phys. Lett. 141B (1984) 349, S. Hamidi and
J.H. Schwarz, Phys. Lett. 147B (1984) 301;
D.R.T. Jones and S. Raby, Phys. Lett. 143B (1984) 137, J.E. Bjorkman, D.R.T. Jones,
and S. Raby, Nucl. Phys. B259 (1985) 503;
J. León, J. Perez-Mercader, M. Quiros, and J. Ramirez-Mittelbrunn, Phys. Lett. 156B
(1985) 66;
X.-D. Jiang and X.-J. Zhou, Phys. Lett. 197B (1987) 156;
D. Kapetanakis, M. Mondragón, and G. Zoupanos, hep-ph/9210218, Z. Phys. C 60
(1993) 181;
D.I.Kazakov, M.Yu.Kalmykov, I.N.Kondrashuk, and A.V.Gladyshev, hep-ph/9511419,
Nucl. Phys. B471 (1996) 389:
realistic finite models.
19 Mandelstam; Brink, Lindgren, and Nilsson; loc. cit. (VIB):
finiteness proof for N=4 using lightcone superfields.
20 Grisaru and Siegel, loc. cit. (VIC, ref. 12):
nonrenormalization theorems for extended supersymmetry.
21 P. Howe, K.S. Stelle and P. West, Phys. Lett. 124B (1983) 55:
finiteness proof for N=2 (+ matter) and N=4 using nonrenormalization theorems.
22 V.A. Novikov, M.A. Shifman, A.I. Vainshtein, and V.I. Zakharov, Nucl. Phys. B229
(1983) 381:
finiteness proof for N≥2 based on instantons.
23 P. Jordan, Z. Phys. 93 (1935) 464;
M. Born and N.S. Nagendra Nath, Proc. Ind. Acad. Sci. 3 (1936) 318;
A. Sokolow, Phys. Z. Sowj. 12 (1937) 148;
S. Tomonaga, Prog. Theo. Phys. 5 (1950) 544:
early attempts at bosonization.
24 W. Thirring, Ann. Phys. 3 (1958) 91.
25 T.H.R. Skyrme, Proc. Roy. Soc. A262 (1961) 237;
D. Mattis and E. Lieb, J. Math. Phys. 6 (1965) 304;
B. Klaiber, The Thirring model, in Lectures in theoretical physics, eds. A.O. Barut and
W.E. Brittin (Gordon and Breach, 1968) v. X-A, p. 141;
440 VIII. GAUGE LOOPS

R.F. Streater and I.F. Wilde, Nucl. Phys. B24 (1970) 561;
J. Lowenstein and J. Swieca, Ann. Phys. 68 (1971) 172;
K. Bardakçi and M.B. Halpern, Phys. Rev. D3 (1971) 2493;
G. Dell’Antonio, Y. Frishman, and D. Zwanziger, Phys. Rev. D6 (1972) 988;
A. Casher, J. Kogut, and L. Susskind, Phys. Rev. Lett. 31 (1973) 792, Phys. Rev. D10
(1974) 732;
A. Luther and I. Peschel, Phys. Rev. B9 (1974) 2911;
A. Luther and V. Emery, Phys. Rev. Lett. 33 (1974) 598;
S. Coleman, Phys. Rev. D11 (1975) 2088;
B. Schroer, Phys. Rep. 23 (1976) 314;
S. Mandelstam, Phys. Rev. D11 (1975) 3026;
J. Kogut and L. Susskind, Phys. Rev. D11 (1975) 3594:
bosonization (mostly for sine-Gordon ↔ Thirring).
26 J. Schwinger, Phys. Rev. 128 (1962) 2425.
 B. LOW ENERGY 441

........................ ........................
........................ B. LOW ENERGY ........................

In general, the only loop corrections that can be evaluated exactly in terms of el-
ementary functions are the one-loop propagator corrections. However, limiting forms
of vertex corrections, for various low- or high-energy limits, explicitly yield the most
important pieces for certain applications.

1. JWKB

Some low-energy contributions to the effective action can be obtained by vari-

ous quantum mechanical JWKB approximations. This involves an expansion of the
external field about its vacuum value in spacetime derivatives (momenta). Such an
expansion makes sense if this field is massless, since then small spatial momentum
means also small energy, in the relativistic sense. (Otherwise one needs to expand
nonrelativistically, about ~p = 0 but E = m. Such treatments were considered in sub-
section IIB5, and will be applied to loops in subsection VIIIB6.) It also can be useful
when the mass of the external field is small compared to the mass scale relevant to
the interactions, such as for chiral symmetry breaking in the low-energy description
of light mesons (subsection IVA4).
On the other hand, the fields we are integrating out must be massive, with a mass
greater than the energy we want to investigate: Otherwise, the internal particles
would show up as poles (and cuts) in the amplitudes, where Taylor expansion in
momenta would be a poor approximation. The basic principle for analyzing the
behavior of such a theory in a certain energy range is thus to first find contributions
to the effective action where: (1) only particles with masses of lower energy appear
on external (background) lines, and (2) only particles with masses of higher energy
appear on internal (quantum) lines. These contributions are approximated by Taylor
expansion to finite order in external momenta, yielding a local effective action. We
could then consider finishing the functional integration by integrating out the lighter
particles on internal lines: However, in this approximation it would be inaccurate to
consider such particles in loops, since there they would include energies above the
approximation scale. Thus, the effective action obtained by integrating out just the
heavier fields is useful only when the lighter fields are treated classically. We apply
the same approximation scheme to the classical action: Eliminate the heavier fields by
their classical equations of motion, and Taylor expand their propagators in momenta
to the desired order to get a local result.
442 VIII. GAUGE LOOPS

In subsection VIIB2 we saw the simplest example, the effective potential: In that
case the constant background scalar field acted as just a correction to the mass. We
now consider more complicated cases, where spin and gauge invariance play roles for
the internal or external fields. In particular, adding coordinate dependence to the
background fields means we need to consider more general propagators for quadratic
kinetic operators, such as harmonic oscillators.
We saw in subsection VIB1 the most general relativistic particle action for a scalar
.
in external fields that was quadratic in x and x. We now consider such actions in more
detail: They are the most general ones for which we can derive one-loop results to all
orders in the external fields (i.e., without performing the JWKB expansion beyond the
first quantum correction, which requires Taylor expanding the exponential in terms
that are beyond quadratic, thus expanding in the number of external fields).
Without loss of generality, we can consider Lagrangians that are homogeneous
. .
of second order in x and x: Terms linear in x are boundary (in τ ) terms, and were
already eliminated by a gauge transformation (radial gauge). Terms linear in x can
be removed by a translation, in the presence of an x2 term (which is needed to bound
an x term in the potential). (Both these kinds of terms can be restored trivially at
the end.) A constant term is also trivial, giving a contribution to the classical action
RT
that is just that times T (after integration 0 dτ ), and can be treated separately. (It
doesn’t contribute to the equations of motion.) The remaining contribution to the
mechanics action is then of the form (as usual, in the gauge v = 1)
Z T
. . .. .
S= dτ 12 [−x2 + xAx + xBx] ⇒ x + Ax + Bx = 0
0
Z T
. .. .
⇒ S= dτ 12 (−x2 − xx) = − 12 (xx)|T0
0
where A is an antisymmetric matrix and B symmetric. The steps to this contribution
to the one-loop effective field action are then: (1) Solve the equations of motion, which
are homogeneous second-order differential equations. (2) Change variables from the
two parameters used for each x to x(0) and x(T ). (Second-order differential equations
require two initial conditions, or one initial and one final.) (3) Find S(x(0), x(T )),
including separately the contribution from the constant term in the Lagrangian. (4)
Find the propagator for “time” T , including the e−iS and the van Vleck determinant.
(See excercise VA2.4.) (5) Integrate the propagator over T to find Γ. (See subsection
VIIB2.)
For example, consider in QED the contribution to Γ from a fermion loop. If we
are interested in only the properties of photons, then this gives the entire contribu-
tion to the functional integral from integrating out the fermions: This contribution,
 B. LOW ENERGY 443

plus the classical (free) Maxwell action, gives a nonlocal “classical” action of self-
interacting photons, which can itself be quantized to give the exact QED result for
external photons. Although this one-loop effective action is too difficult to calculate
exactly, the first-quantized JWKB approximation can give an accurate description
at energies small compared to the electron mass. Note that we are simultaneously
approximating to the first quantum correction in JWKB expansions of both the field
(second-quantized) type (one-loop) and the mechanics (first-quantized) type.
The mechanics action for a massive particle in a constant external electromagnetic
field strength (the lowest nontrivial order, but also the highest that keeps the action
quadratic), in the radial gauge for the background field and affine parametrization of
the worldline, is (see subsection VIB1)
Z
. .
S = dτ 12 (−x2 + xaFab xb + M 2 )

To include spin, we identify (see subsection VIIIA3)

M 2 = m2 − iS abFab

Since the only appearance of spin operators in the calculation of the propagator
(denominator) is this constant matrix, it commutes with everything, so we can treat
it as a number till the last step. The equation of motion
.. .
x + Fx = 0
.
is easily solved in matrix notation. (Hint: Solve for x first.) Finding xi = x(0) and
.
xf = x(T ) in terms of our integration parameters and inverting, then expressing x(0)
.
and x(T ) in terms of xi and xf (and T and F ), and making use of the antisymmetry
of F , the result is

S = − 14 (xf − xi )F coth( F2T )(xf − xi ) − 12 xf F xi + 12 M 2 T

The propagator is then given by (see subsections VA2 and VIIB2)

s
∂ 2 (−iS) −iS
hxf |e−iT H |xi i = det e
∂xf ∂xi

Plugging in, and then Wick rotating T → −iT , we find for the propagator with ends
tied together r
−T H iF
e−M T /2
2
hx|e |xi = det
1−e −iF T
444 VIII. GAUGE LOOPS

Finally, the contribution to the effective action is (see subsection VIIB2)

Z Z ∞ r r !
dT iF −M 2 T /2 I −m2 T /2
Γ = −c dx tr det e − det e
0 T 1 − e−iF T T

Z Z "r #
∞
iF T
= −c dx dT T −D/2−1 e −m2 T /2
det tr(eiS·F T /2) − tr(I)
0 1 − e−iF T

where c = − 12 for fermions, for statistics and squaring the propagator. (The “det” is
for the vector indices on Fab , the “tr” is for the spin indices from powers of S ab in
“S · F ”.)
Excercise VIIIB1.1
Explicitly evaluate the determinant and trace for D=2.
Excercise VIIIB1.2
Expand Γ in F and show the resulting F 2 terms agree with those obtained in
subsection VIIIA2-3.
Excercise VIIIB1.3
Consider the quadratic action
Z
. .
S = dτ 12 [−x2 + x(a − aT )x − xaaT x]

where the matrix a commutes with its transpose ([a, aT ] = 0). Solve the field
equations for S(xi , xf ; T ). Find hx|e−T H |xi.

2. Axial anomaly
The axial anomaly comes from a finite graph, as we have already seen in subsection
VIIIA8 for the case D=2. However, the naive manipulations that would show the
graph to preserve gauge invariance involve evaluating the finite difference between
divergent graphs, each of which needs regularization. Although in some cases the
graph can be evaluated explicitly, and then shown to be anomalous, it is generally
easier, and more instructive, to analyze the anomaly by itself.
The axial anomaly is associated with the use of  tensors. In renormalizable
theories in D=4, these occur only through γ−1 ’s for spinors. (In nonrenormalizable
theories, or in D=2,  tensors can occur in scalar theories. There is also the term
abcd Fab Fcd , which is a total divergence, and has no effect in perturbation theory.)
In general even dimensions, the massless kinetic term for a spinor is invariant under
transformations generated by γ−1 , but the mass term is not. Chiral symmetry is thus
 B. LOW ENERGY 445

related to masslessness; this is also true for conformal invariance, so it’s not surpris-
ing that quantum corrections can break both. (In fact, in supersymmetric theories
conformal symmetry is related to a particular chiral symmetry by supersymmetry, so
breaking of one requires breaking of the other if supersymmetry is to be preserved.)
Dimensional regularization manifestly preserves neither conformal nor chiral in-
variance; no regularization does. The existence of these anomalies proves the impos-
sibility of such a regularization. Furthermore, dimensional reduction has difficulty
dealing with γ−1 ; it even has inconsistencies in the presence of axial anomalies. On
the other hand, Pauli-Villars regularization is especially convenient for dealing with
axial anomalies because it regularizes by introducing masses. Thus, it breaks chiral
symmetry explicitly but softly, conveniently parametrizing the breaking by mass pa-
rameters. We therefore will use Pauli-Villars regularization for the single purpose of
evaluating the axial anomaly.
The basic idea of Pauli-Villars regularization is to include massive “ghost” fields
which would cancel graphs from physical fields if they had the same mass. But the
masses of the ghosts are used as regulators; after subtracting local divergences, the
regulator mass is taken to infinity. In our case, as we’ll see by explicit evaluation, the
anomaly itself is finite, so no subtraction is necessary.

The graph whose anomaly we want to evaluate is a one-loop 1PI graph with
external vectors and a massless internal spinor. Of the vectors, all but one is a
“polar” vector, coupling to ψ̄γa ψ, while the last is an “axial” vector, coupling to
ψ̄γ−1 γa ψ. These are the currents associated with the symmetries ψ 0 = eiθ ψ and
ψ 0 = eθγ−1 ψ (γ−1
2
= − 12 ). We add to this graph a similar one, but with a massive
spinor, and give the second graph an overall relative minus sign. Since the mass
breaks chiral invariance, we have explicitly broken the gauge invariance of the axial
vector, while preserving those of the polar vectors. Note that this is a feature of the
regularization: If a regularization existed that preserved chiral symmetry, then we
could freely move the γ−1 around the graph from one vertex to the next using the
usual naive anticommutation relations, thus moving also the anomaly from one vertex
to the next (i.e., violating gauge invariance in any vector we choose).
446 VIII. GAUGE LOOPS

Gauge invariance is represented by vanishing divergence of the corresponding

R
current: At each vertex we have the coupling A · J, with gauge invariance δA = ∂λ,
implying ∂ ·J = 0 by integration by parts, where J may be polar or axial depending on
the vertex. These currents are conserved classically. We know they are also conserved
quantum mechanically in the absence of γ−1 ’s, since dimensional regularization and
renormalization preserve the gauge invariance of the effective action. In graphical
terms, taking the divergence at a vertex kills a propagator (since ∂ · J is proportional
to the field equations of the internal field), and this can be shown to lead to vanishing
of the graph.
However, with the Pauli-Villars regulator, the classical conservation of the axial
current is explicitly broken. The result is that the complete axial anomaly can be
found by looking at just the contribution coming from this explicit classical violation
of current conservation (inserted into the one-loop graph). (The classically vanish-
ing contributions are actually nonvanishing because of the anomaly, but they cancel
between the physical and regulator fields, precisely because the regularization allows
the naive manipulations that justify dropping them.)
We therefore want to evaluate the anomaly
δΓ
∂a J a (x) ≡ ∂a
δAa(x)
R
where we start with a term in the classical action A · J, so classically J = δS/δA,
and then evaluate its quantum correction by looking at J ≡ δΓ/δA in terms of the
one-loop part of the effective action Γ. Classically, we find a contribution from only
the regulator,
√
∂ · J ≡ ∂ · (− 2iψ̄γ−1 γψ) = 2mψ̄γ−1 ψ
So, all we need to evaluate is a one-loop diagram with the axial vector coupling to the
R
regulator replaced with a pseudoscalar coupling φψ̄γ−1 ψ, and look at the graphs
with one external pseudoscalar and the rest polar vectors. Clearly this is the same
as coupling the pseudoscalar to the propagator of a bosonic spinor regulator in an
external vector field:
!  
1 1
∂ · J = 2m tr γ−1 = 2m tr γ−1 (−i∇ m
/ + √2 ) 2 1 2
i∇/ + √m2 ∇
/ + 2m
 
√ 2 1
= 2m tr γ−1 2 1 2
∇
/ + 2m
where the trace is in the γ-matrix space. In the limit m → ∞, graphs with more
external lines vanish more rapidly. On the other hand, we need at least D/2 fac-
tors of S ab (D γ-matrices) to give a nonvanishing γ-matrix trace. Thus, the leading
 B. LOW ENERGY 447

contribution will be, using −2∇/ 2 = + iS ab Fab from subsection IIIC4,

 
√ 2 1 1 1
∂ · J = 2m tr γ−1 1 2 1 ab
2 iS Fab 1 · · · 12 iS abFab 1 2
2 (m − 2 (m − 2 (m −
2
0) 0) 0)

with D/2+1 propagators, where 0 = (∂a)2 .

Thus, the only Feynman diagram we actually need to evaluate is the one-loop 1PI
diagram with external and internal scalars. The limit internal m → ∞ is the same
as the limit external p → 0. (The result does not depend on the internal momentum,
which is integrated over, nor the external mass, which would appear only in external
propagators.) Thus, this is just an effective potential calculation. We therefore have
the integral (see subsection VIIB1)
Z
1 1
dk 1 2 = D
2
[ 2 (k + m )] D/2+1 Γ( 2 + 1) 12 m2
√
2 2
⇒ ∂ · J = D tr[γ−1 ( 12 iS abFab )D/2]
( 2 )!
Excercise VIIIB2.1
Check this result by using the expression from subsection VIIIB1 for the
propagator in a constant external electromagnetic field (strength).
To evaluate in arbitrary even D, we note that the normalization of γ−1 is such
that we can choose

(γ−1 )2 = − 12 ⇒ γ−1 = (−i)D/22(D−1)/2γ 0 γ 1 · · · γ D−1

tr(I) = 2D/2 , 01···D = −1 ⇒ tr[γ−1( 12 iS abFab )D/2] = √1 ( 1 )D/2 ab···cd Fab...Fcd

2 2
1
⇒ ∂·J =2 D/2 D
ab···cd Fab ...Fcd
2 ( 2 )!
Thus, for example, for the Schwinger model (D=2) we have

∂ · J = −2F

in agreement with subsection VIIIA8, while for D=4

∂ · J = 14 abcd FabFcd

3. Anomaly cancelation
When the anomaly occurs in a current that couples to a gauge field, unitarity
is destroyed, since gauge invariance implies current conservation. This is a potential
448 VIII. GAUGE LOOPS

problem, since axial vector couplings occur in the Standard Model. (Actually, they
are “V−A”: (vector)−(axial vector).) The only way to avoid this problem is to have
an anomaly cancelation between the different spinors: The coefficient of the anomaly
is given purely by group theory, as tr(A, {B, C}), where A, B, C are the matrices
representing the couplings of the three vectors to all spinors, and the anticommuta-
tor comes from Bose symmetrization (from the crossed and uncrossed graph in the
S-matrix, or the single contribution multiplying commuting fields in the effective ac-
tion). We therefore require this trace (which represents the sum over all spinors)
to vanish. (See excercises IB5.3 and VC9.2d for an example of the calculation of
this trace.) The representations in the Standard Model have been chosen so this
cancelation occurs in each family.
We already know in terms of Dirac notation that axial anomalies appear only
in the presence of γ−1 ’s. An absence of γ−1 ’s is equivalent in terms of Weyl no-
tation to the use of a (pseudo)real representation for undotted Weyl spinors. For
example, consider a real representation that is reducible to a smaller (by half) rep-
resentation “R” and its complex conjugate “R̄”: Then we can complex conjugate
the complex-conjugate representation to produce a dotted Weyl spinor that is the
same representation as the undotted spinor. The undotted and dotted spinor can
then be combined, as usual, to form a Dirac spinor, which transforms as the complex
representation, without γ−1 ’s, and thus the same goes for the coupling of the gauge
vector:
ψRα ⊕ ψR̄α → ψRα ⊕ ψ̄Rα. → ΨR

So, in Dirac notation we can see that such representations do not contribute to anoma-
lies because of the absence of γ−1 ’s. Similar remarks apply to general real or pseudo-
real representations: We can take an arbitrary (pseudo)real representation and make
a Majorana spinor, as
ψRα → ψRα ⊕ ψ̄Rα. → ΨR

where now ψ̄Rα. is simply the complex conjugate of ψRα since R = R̄.
This cancelation also can be seen directly in terms of Weyl spinors: The (pseudo)-
reality of the representation is charge conjugation invariance (which is equivalent to
parity invariance for spin-1 couplings to spinors, since such couplings are always CP
invariant). Anomaly cancelation is then a generalization of Furry’s theorem (see
subsection VIIA5). Real and pseudoreal representations of the generators (including
complex + complex conjugate) are antisymmetric, up to a unitary transformation,
since they are hermitian:
GT = G* = −UGU −1
 B. LOW ENERGY 449

(so δψ = iGψ preserves reality or pseudoreality). Thus

tr(A, {B, C}) = tr(−AT , {−B T , −C T }) = −tr(A, {B, C}) ⇒ tr(A, {B, C}) = 0

In particular, any mass term (without Higgs) ψ T αψα requires a real representation (so
its variation yields G + GT = 0); a pseudoreal representation won’t work because it
uses an antisymmetric metric which, when combined with Cαβ , makes ψ T αψα vanish
by symmetry (since ψ is anticommuting). A related way to see in Weyl (or Dirac)
notation that real representations are nonanomalous is to use the same squared-
propagator trick we used for the propagator correction in subsection VIIIA3 (or re-
lated complex action from subsection IIIC4), which resulted in simplified Feynman
rules only for real representations: With those rules, there are no potentially diver-
gent 3-point graphs other than those that already occur for scalars (as part of the
covariantization of the propagator divergence).
The absence of γ−1 ’s is a special case of parity invariance. However, even par-
ity invariance is not enough to enforce cancelation of anomalies, since some parity
invariant theories have axial gauge vectors, which couple to axial currents Ψ̄γ−1 γa Ψ,
and the appearance of these γ−1 ’s can be sufficient to introduce anomalies. In these
anomalous cases, even if there is a C, the charge conjugation argument above does
not apply because the C following from the usual CP and the obvious P does not
simply replace A → −AT , but is some other permutation of similar representations.
Thus, in general P (and C) invariance is unrelated to anomaly cancelation: We can
have one without the other. Having real representations (i.e., no γ−1 ’s) is a special
case of both.
Excercise VIIIB3.1
Consider chiral symmetry (as in subsection IVA4 or IVB1) for a single flavor
— U(1)L ⊗U(1)R . Now gauge that symmetry:
a In Weyl spinor notation, write the action for massless Weyl spinors ψLα , ψRα
each coupled to their own gauge vector. Clearly there is one anomaly for 3
external ALa’s, due to ψL, and another for ARa, due to ψR, and no mixing.
Now assume the left and right coupling constants are equal (so the anomalies
are equal). Write the resulting symmetry transformations on all fields under
CP, C, and P.
b Rewrite this theory in Dirac notation. Using P, find the combinations of AL
and AR that are (polar) vector and axial vector. Relate the anomaly cal-
culations in the two notations. Show that dropping the axial vector gives
450 VIII. GAUGE LOOPS

(massless) QED. Find the theory that results from dropping the vector in-
stead: Give the gauge symmetry, and show it is anomalous, and explain the
anomaly (vs. the cancelation of the anomaly in QED) in both Weyl and Dirac
language.
c Generalize all the above results to U(n)L⊗U(n)R. (Note that C will now
include complex conjugation on the hermitian matrices for the vectors, so
that P won’t.)
The simplest way to prove anomalies cancel in the Standard Model is to use our
previous results for GUTs (subsection IVB4): (1) One way is to consider the GUT
gauge group SU(4)⊗SU(2)⊗SU(2). First, we note that tr(Gi ) = 0 for all of the
representations used, so there are no mixed anomalies. Then we see that the SU(4)
couplings are the usual “color”-type couplings, without γ−1 ’s (i.e., 4 ⊕ 4̄), so it has no
anomalies. On the other hand, SU(2) has only (pseudo)real representations, so neither
SU(2) has anomalies. Thus, anomalies cancel in the SU(4)⊗SU(2)⊗SU(2) GUT.
Finally, breaking to SU(3)⊗SU(2)⊗U(1) (which also spontaneously breaks parity)
leaves an extra singlet per family, which decouples, showing the cancelation for the
Standard Model.
(2) Another way is to start with SO(10), which is anomaly free for any represen-
tation of fermions:
tr(Gab , {Gcd , Gef }) = 0

simply because there is no combination of Kronecker δ’s with the appropriate sym-
metry (and similarly for SO(N), except for N=2 or 6, where such a term can be
produced with the  tensor). Breaking to the Standard Model again drops just a
singlet (as does breaking to SU(5), showing its anomaly cancelation; breaking to
SU(4)⊗SU(2)⊗SU(2) drops nothing, again showing its cancelation). In general, prov-
ing anomaly cancelation requires (a) using such arguments about real representations,
or (b) the absence of anomalies for certain groups (namely, only SU(N) for N>2, or
U(1), can have anomalies), or (c) explicitly calculating the relevant traces.

4. π0 → 2γ
When an anomalous axial symmetry appears only as a global symmetry classi-
cally, unitarity is preserved, since no gauge field couples to that current. This can be
a useful way to explain approximate global symmetries. The fact that the anomaly
is always a total derivative (because of the  tensor and the Bianchi identity for F )
means that the global symmetry is not broken perturbatively. (However, when the
 B. LOW ENERGY 451

external vectors are nonabelian, there can be contributions from field configurations
like instantons: See subsection IIIC6.) In subsection IVA4, we saw that the neutral
pion (π 0), the lightest hadron, could be considered as the pseudogoldstone boson of
an axial U(1) symmetry. We also want to consider the pion as a bound state of a
quark and antiquark: If we knew the wave function, we could write the coupling,
and calculate directly the decay of the neutral pion into two photons (π 0 → 2γ) via
quark-antiquark annihilation, or at least find the leading low-quark-energy contribu-
tion from the δ-function part of the wave function (in the relative coordinates of the
quark and antiquark), corresponding to the coupling to ψ̄γ−1 ψ. (An expansion of
the wave function in derivatives of the δ function would give coupling to currents
containing derivatives.)
Lacking such detailed information, the best we can do is extend the nonlinear σ
model approach, which is to look for the terms in the phenomenological Lagrangian
(expressed in terms of composite meson fields, not fundamental quark fields) with
fewest derivatives (i.e., those most important at low energy), applying the condition
of (approximate) chiral symmetry. Specifically, the global axial symmetry π 0 = π −2θ,
where A0 = A for the photon field, along with the electromagnetic gauge invariance
for A, under which the neutral pion field is invariant, would suggest couplings of pion
to photon involving only ∂π and F .
However, the anomaly allows the existence of another term: Since by definition
(from considering coupling to an unphysical axial gauge field) the anomaly is given
from a local axial transformation, while the pion field transforms in a trivial way
under this transformation, we can attribute the anomaly to the pion coupling as
Z
δπ = −2θ, δΓ = − θ∂ · J , Γ = Γ0 + ∆Γ, δ(∆Γ) = 0
Z Z
1
⇒ Γ0 = 1
2 π∂ ·J = π ab···cd Fab ...Fcd
2D/2( D2 )!
Thus, in four dimensions we find the contribution
Z
Γ0 = π 81 abcd FabFcd

Using the abelian form of the Chern-Simons form (subsection IIIC6), we also can
write this as Z
Γ0 = − 16 abcd (∂a π)Bbcd

(In the nonabelian case, we can neglect the surface term only if the vacuum value of
π has already been subtracted.) Adding this term to those found previously (the π
and A kinetic terms, as well as the quark terms that define the normalization of the
452 VIII. GAUGE LOOPS

π field through its coupling), the decay rate for π 0 → 2γ can be calculated (including
the 2 relevant flavors of quarks, and 3 colors, using the values of their electromagnetic
charges), and is found to agree closely with the experimental value.
The global anomaly in the nonperturbative case can be applied to the strong
interactions (QCD), although not as straightforwardly: Considering the external vec-
tors to be gluons (so there is an implict trace above over the group indices), Γ0 gives
a coupling of a neutral meson to a pseudoscalar glueball, as discussed in subsection
IC4. If the vacuum gives a nontrivial value to tr(abcd Fab Fcd ) (as for instantons), this
also leads to anomalous CP violation in the strong interactions.

5. Vertex
One-loop triangle graphs can’t be evaluated in terms of elementary functions.
However, in QED the most important effects are at low energy. We therefore will
evaluate the effective action in the quantum mechanical version of the JWKB ex-
pansion, as an expansion in derivatives. The resulting approximation to the effective
action thus will be local, but include terms of higher dimension than the classical
action, whose coefficients are therefore finite and unrenormalized: By dimensional
analysis, this means their coefficients will have powers of the inverse electron mass,
which can be considered as the expansion parameter. (See also subsection VIIB7,
where a scalar 1-loop vertex divergence was evaluated.)

The propagator corrections have been found already in subsection VIIIA1; now
we calculate the vertex correction. The integral is
Z
Na
Aa,3,QED = dk
D

Na = γ b (k/ + p/0 + √m2 )γa (k/ + p/ + m

√ )γ ,
2 b
D = 12 k 2 12 [(k + p0 )2 + m2] 12 [(k + p)2 + m2]

Without loss of generality, we can drop terms that vanish by the free fermion field
equations; this corresponds to canceling them by fermion field redefinitions. We then
evaluate the numerator by applying the identities

p2 = p02 = −m2, q = p0 − p ⇒ (p + p0 )2 = −4m2 − q 2

 B. LOW ENERGY 453

v/γv/ = 12 v 2γ − vv/
m
as well as the identities of subsection VIC4 for γ b ...γb, and the field equations p/ = √
2
on the far right and p/0 = √m2 on the far left, to obtain

N = (k/ + p/)γ(k/ + p/0 ) + m2

2
γ + m
√
2
(2k + p + p0 )

(k/ + p/)γ(k/ + p/0 ) = (k/ + p/ + p/0 )γ(k/ + p/ + p/0 ) − p/γp/ − p/0 γp/0 − p/0 γp/ − k/ γp/ − p/0 γk/

= [ 12 k 2 +k · (p+p0)− 2m2 − 12 q 2]γ − (k +p+p0)(k/ +2 √m2 )+m2γ + √m2 (p+p0 )− m2 γ + √m2 k

2

⇒ N = ( 12 k 2 γ − kk/ ) + [k · (p + p0 )γ − (p + p0 )k/ + m
√
2
k] − (m2 + 12 q 2 )γ

For the momentum integral we evaluate

Z Z
eik·x
A3 (x, m , q ) = dk
2 2
= d3 τ λ−D/2 e−E
D

E= 11 2
2 λx + ix · 12 [(α1 + α2 )(p + p0 ) + (α1 − α2 )q] + 12 λ[(α1 + α2 )2 m2 + α1 α2 q 2]

again on the fermion mass shell. We also have

Z Z ∞ Z 1  X  Z ∞ Z 1 Z 1−α1
3
d τ= dλ λ2
d α δ 1−
3
α = dλ λ 2
dα1 dα2
0 0 0 0 0

using, e.g., the definitions

Z b Z ∞ Z b
dx δ(x)f(x) = θ(−a)θ(b)f(0), dx θ(x − a)θ(b − x)f(x) = dx f(x)
a −∞ a

As for the fermion propagator, we clearly separate UV and IR divergent integrals by

the changes of variables
Z Z ∞ Z 1 Z α
α = α1 + α2 , β = α1 − α2 ⇒ 3
dτ= dλ λ2 1
dα 2 dβ
0 0 −α

followed by
Z Z ∞ Z 1 Z 1
λ
λ → 2, β → αβ ⇒ dτ→ 3
dλ λ 2
dα α−5 12 dβ
α 0 0 −1

which modifies the integral to

Z ∞ Z 1 Z 1
A3 = dλ λ
dα α−1−2 12 dβ e−E
0 0 −1

E = 12 { λ1 α2 x2 + iαx · (p + p0 + βq) + λ[m2 + 14 (1 − β 2)q 2]}

454 VIII. GAUGE LOOPS

We now expand to O(x2) and O(q 2). The β integral is then trivial (the integrand
becomes quadratic in β), the λ integral gives the usual, and the α integral is similar
to the case of the fermion propagator. The result is
 2
 2

1 2 − 1 q 1 q 1
A3 ≈ − 2 Γ(1 + )( 2 m )
1
−6 2 + 1 − 6 2 + ix · (p + p0 )
m m IR
 
0 2 1 1 2 2
+ 8 (x · (p + p )) + 4 m x + 24 (x · q) +
1 1 2 2 1 2
mx
U V 4
This leads to the expression for the vertex correction
 2
 2

1 2 − 1 q 1 1 1 1 q
A3,QED ≈ Γ(1 + )( 2 m ) 1+ 3 2 + 7
+ 2 + 12 2 γ
m IR 2 U V m
  
q 2 p + p0
+ 14 1 − 16 2 √
m m/ 2
where we have used
δaa = D
in evaluating the contribution from the k 2 term. (Remember that all algebra from
indices on the fields is done in 4 dimensions, while all algebra from indices on momenta
is done in D dimensions. Since the two parts of the calculation are usually done
separately, this should cause no confusion; however, the difference in evaluating δaa is
the main thing to watch.) Using the on-shell identity

4 √m2 γ = {p/ + p/0 , γ} + [q/, γ] = −(p + p0 ) + [q/, γ]

we can rewrite this as (again keeping only O(q 2))

    2 
1 2 − 1 1 1 1 1 q 1 [q
/, γ]
A3,QED ≈ Γ(1 + )( 2 m ) + 5
+2 + 3 + 1
γ+4 √
IR 2 U V IR 4 m2 m/ 2

The next step is to cancel the UV divergence by adding the counterterm for
electron wave-function renormalization from subsection VIIIA1:
    2
1 m2 q [q/, γ]
A3,QED,r = A3,QED + δA3,QED ≈ 3 1
− ln 2 + 4 1
2
γ + 14 √
IR µ m m/ 2
Equivalently, we can take the q = 0 piece of A3,QED, and note that it combines with
A2e of subsection VIIIA1 to gauge-covariantize the term proportional to p/ → p/ + A /.
(The unrenormalized effective action is thus automatically gauge invariant, as is the
counterterm.) At this point we can see the anomalous magnetic moment: Combining
the tree and 1-loop result (including coupling), and writing as spinless + magnetic
moment contributions, we have
     2 0
1 m2 q 1 p+p [q/, γ]
γ + e A3,QED,r ≈ 1 + e 3
2 2 1
− ln 2 + 4 1
2
(− 4
) √ + (1 + e2) 14 √
IR µ m m/ 2 m/ 2
 B. LOW ENERGY 455

We can translate these 1-loop corrections into a contribution to the effective action
as (with the usual −1 for the effective action)

Γ1,3,QED,r = −Ψ̄(−p0 )A3,QED,rΨ(p) · A(q)

We then note, again using the spinor (free) field equations to imply (p + p0 ) · q = 0,
to O(q 2),
−4 √m2 q a γ b q[aAb] ≈ q a(p + p0 )b q[aAb] = q 2 (p + p0 ) · A

The low-energy part of the renormalized effective action exhibiting up to order q 2/m2
corrections to the coupling is then, in gauge invariant form,
Z 
e2
Γ0+1,2e,r ≈ dx Ψ i∂/ − A / + √m2 − √ iS abFab
2 2m
    
e2 1 1 m2
− 2 3 − ln 2 + 4 γ (∂ Fab ) Ψ
1 a b
m IR µ
6. Nonrelativistic JWKB
As for other processes, the application of quantum field theory to bound states has
two steps: (1) Calculate the (gauge-invariant) effective action; (2) find solutions to
the field equations following from the effective action (“on-shell” states). For bound
states such solutions are nonperturbative; however, their determination is easier for
nonrelativistic systems, since we can ignore production and annihilation of additional
nonrelativistic (massive) particles in the second step because their effect already has
been included as small corrections to the effective action.
The Lamb shift is the (field theoretic) quantum contribution to the energy levels
of the hydrogen atom, which is described accurately even at one loop. The relativistic
solution is found by perturbing the relativistic effective action in derivatives about
the nonrelativistic one, whose solutions are the usual exact ones of the nonrelativistic
Schrödinger equation. For atoms the electron speed p/m is of the order of α (= 2πe2),
so the loop and derivative expansions are in the same small parameter.
The effective action is more conveniently calculated with manifestly relativistic
methods, since the internal (“virtual”) particles can be relativistic (especially those
that contribute to the UV divergences). On the other hand, the solutions to the field
equations are more conveniently calculated in a representation that takes better ad-
vantage of the nonrelativistic expansion, since the external particles are nonrelativis-
tic. Therefore, the second step begins by performing a field redefinition that converts
the manifestly Lorentz invariant effective action to a form recognizable as nonrel-
ativistic field theory with low-energy relativistic and loop corrections. (Originally
456 VIII. GAUGE LOOPS

the Lamb shift was calculated without this transformation. Higher-order calculations
then required use of the relativistic Bethe-Salpeter equation, which made collection
of terms of a given order more difficult.) In subsection IIB5 we considered the gen-
eralized Foldy-Wouthuysen transformation and its application to minimal coupling;
we now apply it to the nonminimal coupling introduced by loop corrections. (In the
literature this step has been performed on the Feynman diagrams themselves; how-
ever, as usual we can save some effort by working directly with the effective action.)
Here the nonminimal correction to the transformation is easy, since the nonminimal
terms are already near the order to which we work.
We first perform some dimensional analysis, using the fact that the leading be-
havior is given by the usual nonrelativistic Schrödinger equation. Then the only
parameters in units h̄ = 1 (but there is no c in the nonrelativistic theory with just
Coulomb interaction) are the mass m and speed e2, so (in the notation of subsection
IIB5)
π i ∼ me2, π 0 ∼ me4

(neglecting the rest mass contribution). It is then convenient to reorganize the ex-
pansion in 1/m to relate to the expansion in e2: For example, we can identify the two
by choice of units
m
1
∼ e2 ⇒ π i ∼ 1, π0 ∼ m
1

along with c = 1 (since we will include relativistic corrections).

The relativistic form of the Schrödinger equation is obtained by multiplying 2γ 0
in front of the kinetic operator of the electron in a background electromagnetic field,
as obtained from the effective action. Approximating the proton as infinitely massive
(for which we can partially correct by using the reduced mass for the electron), we
take the electric field as described by the usual static “scalar” potential, and drop the
magnetic field along with the “vector” potential.
We therefore modify the expansion of subsection IIB5 by (1) reorganizing the
1/m expansion according to our dimensional analysis, (2) using only a static electric
background, and (3) working directly in terms of γ matrices: We can either plug in
the Dirac case of the spin operators into the expressions of subsection IIB5, including
the reality-restoring transformation of subsection IIB4,

S ab → − 12 γ [a γ b] , S −1a → − √12 γ a

or just expand the Dirac operator directly,

√
2γ 0 (−π/ + m
√
2
) = π 0 − 2γ 0 γ i π i + 2mγ 0
 B. LOW ENERGY 457

(and similarly for the loop correction terms). From the reuslts of the previous sub-
section, we thus choose to order 1/m4
√ 0
E−1 = 2γ , O0 = 2γ i pi γ 0 , E1 = mπ 0
   
1 m2
O3 = −m2e2 √12 iγ iF 0i , E4 = m e 3
2 2 1
− ln 2 + 14 ∂ i F 0i
IR µ
(others vanishing), where we have included explicit m dependence so that the coeffi-
cients En and On are of order m0 according to our above dimensional analysis (so our
expansion in m makes sense). Using

tanh x ≈ x − 13 x3

the relevant commutators from IIB5 are then, for the nonvanishing generators to this
order
mG = − 12 {[G, ∆E] + LG coth(LG )O}E−1

⇒ G1 = − 12 O0 E−1
G3 = − 12 ([G1 , E1 ] + 13 [G1 , [G1 , O0]])E−1
G4 = − 12 O3 E−1

and for the transformed kinetic operator

F 0 = E + tanh( 12 LG )O

⇒ F10 = E1 + 12 [G1 , O0]

F30 = 12 [G3 , O0] − 1
24
[G1, [G1 , [G1, O0 ]]]
F40 = E4 + 2 [G1 , O3 ] + 2 [G4 , O0 ]
1 1

Remembering that E−1 commutes with even and anticommutes with odd, we have
identities like

(LG1 )nO0 = (−1)n(n−1)/2 (O0)n+1 (E−1 )n , (E−1 )2 = 1

Substituting for G into F 0:

F10 = E1 + 12 (O0 )2E−1

F30 = − 18 [O0, [O0, E1 ]] − 18 (O0)4 E−1
F40 = E4 + 12 {O0, O3 }E−1
458 VIII. GAUGE LOOPS

The final result is, using

(O0 )2 = (pi )2, me2[O0, E1] = −2O3E−1

and setting E−1 = −1 on the right for positive energy,

F10 = mπ 0 − 12 (pi )2
F30 = − 14 m[ 12 (∂ iF 0i) − iS ij {F 0i, pj }] + 18 (pi )4
   
0 1 m2
F4 = m e 3
2 2 1
− ln 2 ∂ F + i 2 S {F , p }
i 0i 1 ij 0i j
IR µ
As expected from dimensional analysis, F10 is the nonrelativistic result, F30 is the
lowest-order relativistic correction, and F40 is the lowest-order part of the one-loop
correction. Putting it all together, to this order we have
 i 2    
0 (p ) (pi )4 1 m2 i 0i 1 + 2e
2
F ≈π − 0
− − 1
8m 2 1 − 8 2
3
e − ln ∂ F + iS ij {F 0i, pj }
2m 8m3 IR µ2 4m2

Excercise VIIIB6.1
Find the additional terms in F 0 to this order when the electromagnetic field
is arbitrary (magnetic field, time derivatives of background), assuming the
same dimensional analyis for the background.

REFERENCES
1 W. Pauli and F. Villars, Rev. Mod. Phys. 21 (1949) 434.
2 J. Steinberger, Phys. Rev. 76 (1949) 1180;
Schwinger, loc. cit. (VB, ref. 2, second ref.):
triangle graph, for π 0 → 2γ .
3 J.S. Bell and R. Jackiw, Nuo. Cim. 60A (1969) 47:
identified axial anomaly.
4 S. Adler, Phys. Rev. 177 (1969) 2426:
explained axial anomaly.
5 S.L. Adler and W.A. Bardeen, Phys. Rev. 182 (1969) 1517:
showed axial anomaly is only at 1 loop.
6 H. Georgi and S.L. Glashow, Phys. Rev. D6 (1973) 429:
anomaly cancelation for general GUTs.
7 H. Bethe, Phys. Rev. 72 (1947) 339:
calculation of Lamb shift.
8 H. Fukuda, Y. Miyamoto, and S. Tomonaga, Prog. Theo. Phys. 4 (1949) 47, 121;
N.M. Kroll and W.E. Lamb, Jr., Phys. Rev. 75 (1949) 388;
Y. Nambu, Prog. Theo. Phys. 4 (1949) 82;
J.B. French and V.F. Weisskopf, Phys. Rev. 75 (1949) 1240:
relativistic calculation of Lamb shift.
9 W. E. Caswell and G. P. Lepage, Phys. Lett. 167B (1986) 437:
nonrelativistic effective actions.
 B. LOW ENERGY 459

10 L.S. Brown, loc. cit. (VA), p. 511;

A. Pineda and J. Soto, hep-ph/9711292, Phys. Lett. 420B (1998) 391:
Lamb shift with only dimensional regularization.
11 Quantum electrodynamics, ed. T. Kinoshita (World Scientific, 1990):
higher-loop QED.
460 VIII. GAUGE LOOPS

....................... .......................
....................... C. HIGH ENERGY .......................

The sign of the one-loop correction to the gauge coupling is opposite in QCD to
that of QED: The photon coupling is weak at “low” energies (actually, any observable
energy, since the coupling runs so slowly), while the gluon coupling is weak at high
energies (with respect to the hadronic mass scale). Thus, typically perturbation in
loops is used to study high-energy behavior of QCD, while the low-energy behavior
awaits the discovery of a general nonperturbative approach.

1. Conformal anomaly
Symmetries of the classical action that are violated at the quantum level are called
“anomalous”. There are two major sources for such “anomalies” in renormalizable
quantum field theory: (1) There are anomalies associated with the totally antisym-
metric matrix a1...aD , called “axial” (see subsections VIIIA8 and VIIIB2-4). When
they occur, they are found in graphs with at least (D+2)/2 external lines. They are
associated with graphs that have no divergences, yet require regularization. (2) The
existence of divergences requires the introduction of a mass scale even in theories
that are classically conformal. If anywhere, these show up at least in the most diver-
gent graphs, the propagator corrections. Normally, both kinds of anomalies will first
appear at one loop.
When anomalies are associated with global symmetries, they provide a natural
way to explain approximate symmetries, in the sense of the perturbative approx-
imation. However, when they occur in local symmetries, they destroy the gauge
invariance needed to prove unitarity. The latter type of theory therefore must be
avoided by applying the condition of anomaly cancelation in local symmetries.
We have already seen the appearance of the conformal anomaly in our renormal-
ization of divergent loop graphs: The introduction of a renormalization mass scale
breaks the scale invariance of a classically scale-invariant theory. The simplest ex-
ample, and generally the most important, is the one-loop propagator correction. If
we examine only high-energy behavior, then we can neglect masses from the classical
action.
Using dimensional regularization, the generic effect on the effective action of the
complete one-loop propagator correction is to modify the kinetic term of an arbitrary
massless theory to  
1 1
2 φK + β1 ln 2 φ
g2 µ
 C. HIGH ENERGY 461

where φ is an arbitrary-spin field that we have normalized φ → φ/g for some appro-
priate coupling g (like the Yang-Mills coupling if φ is the Yang-Mills vector), K is the
classical kinetic operator, µ is the renormalization mass scale, and β1 is a constant
determined by the one-loop calculation. As long as β1 is nonvanishing (i.e., the theory
is not finite) we can rewrite this as

1
2 φKβ1 ln φ
M2
where
M 2 = µ2 e−1/β1 g
2

is a renormalization-independent mass scale: Any physical measurement will observe

g and µ in only this combination. A choice of different renormalization mass scale
is equivalent to a finite renormalization of g 2 , such that M is unchanged. In the
case where g is dimensionless (the relevant one, since we are studying the conformal
anomaly), the coupling constant has undergone dimensional transmutation, being
replaced with a dimensionful constant.
If there is more than one coupling constant, things are more complicated, but
the same phenomenon occurs: One dimensionless coupling is replaced with a mass.
A particularly interesting case is pure Yang-Mills theory: Then we can write an
important contribution to the effective action as

F β1 ln F
M2
where F is now the complete nonabelian field strength, and is the square of the
covariant derivative. Since this contribution by itself gives the complete 1-loop con-
formal anomaly, the rest of the 1-loop effective action is conformally invariant. (All
its M dependence cancels.)
Note that the (one-loop) anomaly itself is local: If we perform an infinitesimal
conformal transformation on the one-loop part of the effective action, this variation
gives a local quantity. This is clear from the way this anomaly arose in dimensional
regularization: If there were no infinities, there would be no anomaly, since the naive
conformal invariance of the classical theory would be preserved at each step. However,
to regularize the divergence we needed to continue the theory to arbitrary dimensions,
and the theory is not conformal away from 4 dimensions. The scale variation of a
4D conformal action in 4-2 dimensions is proportional to  times that action, as
follows from dimensional analysis; this scaling can be associated with the nonvanishing
(engineering) dimension of the coupling away from D=4. (Usually, we write the
coupling as gµ , where g is dimensionless. The fields have engineering dimension
462 VIII. GAUGE LOOPS

independent of D, defined by the value in D=4: E.g., in ∇ = ∂ + A, A has the same

dimension as ∂.) However, the one-loop effective action is coupling independent;
thus, when dimensionally regularized but unrenormalized, it’s scale invariant. For
example, in the propagator correction discussed above, we get a regularized term
− 1 β1 φK( 12 )− φ, which is scale invariant but divergent. On the other hand, the
counterterm added to make it finite is from the 4D conformal action, and thus is not
scale invariant in D6=4; so the breaking of scale invariance can be associated entirely
with the counterterm. (I.e., the anomaly coming from the renormalized, nonlocal
effective action is equal to that coming from the infinite, local counterterm.) Since
the counterterm is local, the anomaly is local. It’s also finite, since it’s proportional
to  (from the variation) times 1/ (from the divergent coefficient of the counterterm).
In our propagator example, we have

− 1 β1 φK( 12 )− φ + 1 ( 12 µ2 )− β1 φKφ ≈ β1φK ln φ

µ2
A similar situation occurs for the axial anomaly with Pauli-Villars regularization:
After regularization, the anomaly comes entirely from the regulator graph, which is
not only finite by power counting, but local in the infinite-mass limit because that is
the zero-momentum (effective potential or JWKB) limit.
There is a physical significance to the sign of the constant β1. (We saw some
evidence of this already in our analysis of renormalons in section VIIC.) Instead of
thinking in terms of the renormalization-independent mass scale M, we can treat g
as an effective energy-dependent (“running”) coupling,

1 1 p2
= + β 1 ln
g 2(p2 ) g 2 µ2

In the case β1 > 0 the coupling gets weaker at high energy (“asymptotic freedom”),
while for β1 < 0 the coupling gets stronger at high energy (until it reaches the Lan-
dau ghost). (For low energy the situation is generally more subtle, since we usually
have complications from physical masses.) For QCD, this weakening of the cou-
pling at high energy allows the separation of an amplitude into a nonperturbative
low-energy piece (describing the observed particles, the bound-state hadrons), which
is determined experimentally, and a perturbative high-energy piece (describing the
non-asymptotic, fundamental “partons”, gluons and quarks), which can be calculated.
(This sometimes goes under the somewhat misleading name of “perturbative QCD”.)
This strongly contrasts with QED, where the weakening of the coupling at low en-
ergy means both fundamental particles (photons, electrons, etc.) and bound states
(positronium, atoms, etc.) can be treated perturbatively, and the only experimentally
 C. HIGH ENERGY 463

determined quantities are the values of masses and the electron charge (coupling at
low-energy). Thus, in QED one in principle can calculate anything, while in QCD
one is restricted to parts of certain amplitudes. (Various nonperturbative methods
also have been developed for QCD, but so far they have successfully calculated only a
few low-energy constants, as used in σ models, i.e., masses and low-energy couplings.)
Although experimental verification of these results is sufficient to confirm the QCD
description of hadrons, a practical description of hadronic cross sections at all ener-
gies would seem to require a string model that can incorporate behavior attributed
to both strings and partons.

2. e e → hadrons

If quarks and gluons are confined, how can QCD be useful? QED is useful be-
cause the coupling is small: e2 ≈ 1/861 is the perturbation parameter in relativistic
(quantum field theory, or 4D) calculations, α = 2πe2 ≈ 1/137 in nonrelativistic
(quantum mechanics, or 3D). Energy levels of the hydrogen atom can be calculated
quite accurately, without the question of freely existing electrons and protons coming
up. The speed of the bound electron is also α, another way to understand why pair
creation/annihilation and other relativistic or multiparticle effects are small, and can
be treated perturbatively.
Therefore, the real usefulness of a field theory depends not on how “physical” the
choice of fields is, but how accurate the perturbation expansion is. “Nonperturba-
tive” results may give some nice qualitative features, but they are ultimately useless
unless they can be used as the basis of a new perturbation expansion. (Attempts
at nonperturbative approaches to 4D quantum field theory continue, but so far the
results are meager compared to perturbative results, or to nonperturbative results in
quantum mechanics or 2D quantum field theory.)
The simplest application of QCD is to the production of hadrons by a photon
created by the annihilation of an electron and a positron. The total cross section for
such an event is given (according to the optical theorem) by the imaginary part of
quark contributions to the photon propagator: Since hadrons are made up of partons
(quarks and gluons), we assume a sum over hadrons can be written as a sum over
partons. This assumption, that hadrons can be described by a resummation of the
perturbation expansion, should be good at least at high energies, where the partons’
asymptotic freedom takes effect (and perhaps at lower energies by an appropriate
extrapolation). To lowest order for the process under consideration this is a 1-loop
464 VIII. GAUGE LOOPS

graph, with a quark in the loop. If we compare this to the production of, e.g., muon-
antimuon pairs (but not back to electron-positron pairs, because that includes the
crossed diagram) by the same procedure, and we neglect masses (at high enough
energies), then the only difference should be in the group theory: Hadron production
should be greater by a factor of the number of colors times the sum over flavors of
the square of the quark’s electric charge:
P (e+e− → h) X
R≡ ≈ Nc qf2
P (e+ e− → µ+ µ− ) f

Experimentally this relation is confirmed for Nc = 3, if the only flavors included in

the sum are those with masses below the photon energy ((2mf )2 < s).
This result can be extended to the case where the momenta of hadrons are ob-
served (not summed over): Although individual partons are not observed as asymp-
totic states, the dominant contribution to the cross section at high energies is given
by the conversion of the quarks into hadrons by the creation from the vacuum of
parton pairs with energies, and angular deviation from the partons created by the
photon, smaller than experimental accuracy. We treat all partons as approximately
massless, with respect to the energy scale of the photon. Thus, each parton created
by the photon starts out initially as free, is then accompanied by parallel partons of
small energy to form hadrons, and then these hadrons may further decay, but with a
small angular spread with respect to the directions of each of the initial partons. Such
collections of final-state hadrons are called “jets”. For high-energy electron-positron
annihilation, the dominant hadronic decay mode of this off-shell photon is thus into
two jets. This experimental result is further verification of QCD, and in particular a
jet is the most direct observation of a parton. Of course, even for asymptotic states
the directness of experimental observations varies widely: For example, compare a
photon or electron to a neutrino. A closer analogy is unstable particles: For example,
the neutron is observed as a constituent of the nucleus (as quarks are constituents of
hadrons), but eventually decays outside (as quarks “decay” into jets of hadrons).
 C. HIGH ENERGY 465

A similar analysis can be applied to the creation of any electroweak boson by

annihilation of a lepton with an antilepton.

3. Parton model
We have already seen that in quantum field theory coupling constants are usually
energy-dependent. However, the dependence is only logarithmic, and thus can be
treated as perturbative unless the relevant energy scale is within a few orders of
magnitude of the mass scale that appears by dimensional transmutation. In QED,
the value quoted for the electron charge is at the scale of the electron mass m. Using
the result of subsection VIIIA2 (or VIIIA3) for the 1-loop propagator correction, we
find (neglecting higher-loop corrections)
MQED 2
= e3/4e = 2.837890(82) × 10280 ⇒ MQED = 1.450159(42) × 10277 GeV
m
(where for fun we have included the 1-standard-deviation uncertainties for this 1-loop
result as the figures in parentheses; the e in the exponent is the electron charge).
Since the mass of the observable universe is of the order of 1080 GeV, and the Planck
mass (beyond which a particle will gravitationally collapse from its Compton radius
falling within its Schwarzschild radius) is of the order of “only” 1020 GeV, there is
little worry of observing the QED Landau ghost, even if QED were correct to that
scale.
On the other hand, the mass scale for QCD is (in the MS scheme)

MQCD ≈ .2 GeV

(This result depends on renormalization scheme, and is also an effective mass in the
sense that the usual experimental energy scale is among the quark masses, so the
high-energy approximation of the renormalization group is inaccurate, and the full
propagator correction with quark mass dependence should be used.) This indicates
that perturbative QCD is inadequate to describe properties for which the energy of
the quarks is low, such as hadron masses (although nonrelativistic quark models have
had partial successes).
However, in certain processes a single “parton” (quark or gluon) in a hadron is
given a high energy with respect to the other partons, usually a quark by electroweak
interaction. In those cases, the “strong” (chromodynamic) interaction of that par-
ton with the others in its original hadron is negligible: It has been liberated. The
approach is then to factor the amplitude into a piece with the electroweak and high-
energy (“hard”) chromodynamic interactions of this parton, which can be calculated
466 VIII. GAUGE LOOPS

perturbatively, and the low-energy (“soft”) chromodynamic part of the remaining

partons, which is left as an unknown, to be experimentally determined. (Thus, the
hard part is the easy part, while the soft part is the difficult part.) The predictive
power is thus limited to the dependence of the amplitude on the energy of this parton,
and on the particulars of the electroweak particles involved.
Another possible complication would be the effect of exciting many partons within
a hadron, indirectly through the first parton’s interactions with the rest: Then one
would have several terms to sum in an amplitude, each with a different unknown
soft factor, making the approach useless. Originally, it was thought that the high
energy alone was enough to explain the parton acting as free once liberated from the
hadron (based on “intuitive” arguments), but soon it was realized that this possibility
depended totally on the high-energy behavior of the theory: It requires the decrease of
the coupling with increasing energy, asymptotic freedom (or superrenormalizability,
or finiteness with effective asymptotic freedom through the Higgs effect). Based on
this property, one can show from the usual perturbation expansion that one soft factor
(per each hadron with an excited parton) is sufficient as a leading approximation, a
property known as “factorization”. This feature is a consequence of the fact that the
dominant contributions to Feynman graphs in this high-energy limit are those where
the values of the momenta of some of the partons are those corresponding to their
classical mechanics, as described in subsection VC8 and VIIA6.
This new approximation scheme is effectively a perturbation expansion in the
inverse of the energy being channeled into this parton. One neglects terms that are
smaller by such powers (including those from masses and renormalons), but incorpo-
rates logarithms through the renormalization group and other loop corrections to the
hard factor. Since available energy scales are much nearer to MQCD than to MQED in
QED, such an approximation scheme tends to break down around two loops, where
the corrections compete with the neglected terms, ambiguities in renormalization
schemes, and the relative size (convergence) of successive terms in the expansion. Al-
though the accuracy of the predictions of this approach cannot compare numerically
with those of QED, it is the only method to describe such processes that can lay
claim to being a theory, and provides direct experimental evidence of the validity of
QCD, both as a qualitative description of nature and as a valid perturbation scheme.
(As in the previous subsection, we also have processes where all the partons appear
only in intermediate states, or effectively so for final states in total cross sections via
the optical theorem, so factorization is unnecessary.)
 C. HIGH ENERGY 467

The most effective application of factorization is to “Deep(ly) Inelastic Scattering

(DIS)”. (An equivalent method for this process is the “operator product expansion”,
but unlike factorization there is no useful generalization of it to general processes.)
In this process a high-energy photon (or intermediate vector boson) is exchanged
between a lepton (usually an electron) and a quark. (This is the leading-electroweak-
order interaction of a lepton with a hadron.) The quark and rest of the hadron
do not interact again: Color singlets are obtained by the creation of soft partons
from the vacuum, which split from their own singlets and eventually combine with
the separated quark and hadron. For this process one calculates only the total cross
section, at least as far as all the strongly interacting particles are concerned (“inclusive
scattering”) but again this can be generalized to the observation of jets (“exclusive
scattering”). Applying the optical theorem, and ignoring the leptons, the leading
contribution to this process is given by the tree graph for scattering of a vector
boson off a quark, where the intermediate quark has a cut propagator. This is the
perturbatively calculated hard part, which is later attached to the soft factor. Thus
the hard part is the lepton-parton cross section, while the soft part is the “parton
distribution”, giving the probability of finding a parton in the hadron with a particular
fraction ξ (≥ 0, ≤ 1) of its momentum p. To leading order this fraction is determined
by kinematics: Since the hadron and scattered parton are treated as on-shell and
massless,
q2
p2 = (q + ξp)2 = 0 ⇒ ξ = x ≡ −
2q · p
so the “(Bjorken) scaling variable” x is a useful dimensionless parameter even when
(at higher orders) ξ 6= x. The energy scale is set by the square of the momentum q
of the vector boson.

k
hard

≈
k
ξp q
ξp
soft
p

There are several approximations used in this analysis, all of which can be treated
as the beginnings of distinct perturbation expansions: (1) The hard part is expanded
468 VIII. GAUGE LOOPS

in the usual (loop/coupling) perturbation expansion of field theory. The leading con-
tribution is that of the naive (pre-QCD) parton model (“leading order”), where the
quark that scatters off the photon is treated as free with respect to the strong inter-
actions. One-loop corrections (“next-to-leading-order”) introduce the running of the
coupling associated with asymptotic freedom, which justifies the validity of the par-
ton picture. (Two-loop corrections lead to various ambiguities, and have not proven
as useful yet.) This is usually the only perturbation expansion considered, because
such corrections are logarithmic in the energy of the exchanged parton (rather than
powers), and thus more important and easier to isolate from the data. Furthermore,
by the usual renormalization group methods such logarithmic corrections can be re-
duced by careful choice of renormalization scale (µ2 close to q 2 in ln(q 2/µ2 )). (2) In
calculating the hard part “light” quarks are approximated as massless. One can rec-
tify this by also perturbing in the masses, as a Taylor expansion in the square of
each mass divided by the square of the vector boson’s energy (m2 /q 2). (3) In the
explicit calculation the momentum ξp of the excited parton is assumed to be pro-
portional to the momentum p of the initial hadron. In the rest frame of the initial
hadron (which is massive in real life), this corresponds to the nonrelativistic approx-
imation of motionless quarks; one quark is then set into relativistic motion by the
photon, liberating it from the hadron. This approximation can be corrected by a
JWKB expansion (expressed in operator language, the operator product expansion),
also known as an expansion in “twist” (effectively, the power of momentum trans-
verse to p). However, this means a separate soft part for each term in the expansion:
Since these are determined experimentally, such an expansion would lead to a loss
of predictability. Thus generally (with few exceptions), parton model predictions are
restricted to high enough energies (q 2) that such corrections can be neglected. (4) Ex-
pansions in renormalons (see subsections VIIC2-3) introduce new coupling constants,
effectively nonperturbative corrections to the otherwise perturbative hard part. Like
all but the first of these expansions, this leads to correction terms that are down by
powers of 1/q 2 . (It may be possible to absorb this type of correction into the previous
one, since in principle the hard parts should contain all nonperturbative corrections
by definition.)
The other common application of the parton model is to “Drell-Yan scattering”:
In this case two hadrons scatter producing, in addition to hadrons, a photon (or other
electroweak boson) that decays into a lepton-antilepton pair. To lowest order, the
relevant diagram is the same as for DIS (crossing some of the lines). Because both of
the initial particles are hadrons, 2 soft parts are required; however, each of these is the
same as that used in DIS (“universality”) so they do not need to be redetermined. In
 C. HIGH ENERGY 469

fact, there is a direct progression from e+e− to DIS to Drell-Yan: The above diagrams
are similar except for the number (0 → 1 → 2) of soft parts (corresponding to the
number of initial hadrons); the leading contribution comes from the same diagram,
rotated to various positions (crossing).

B
p' soft
fBb
ξ'p' b ξ'p'

hard dσab ≈ q
ξp a ξp
fAa
soft
p A

More generally, we can consider not only hard parts involving identified quarks
in the initial state of the hard part, but also in the final state, by examining jets.
Thus, for soft parts we have not only the “parton distribution functions”, which
are probabilities found from amplitudes for an initial hadron → parton + anything
(summing over anything), we have “fragmentation functions”, which are probabilities
from amplitudes for parton → final hadron + anything. In principle these are related
by crossing symmetry: The diagrams are similar to the previous, with the partons
connecting to the hard part, but the external hadron lines may be either initial or final
(and the opposite for the corresponding parton with respect to the hard subgraph).
As for the parton distributions, the fragmentation function for any particular parton
and hadron is measured in one particular experiment, then used universally. (The
simplest is deep inelastic scattering for the parton distribution, and e+e− annihilation
with one of the two jets → hadron + anything for fragmentation.) Then the cross
section is generally of the form
XZ
dσA...B = dξa · · · dξb fAa (ξ1 ) · · · fBb (ξn )dσa...b (ξi )
a...b

where dσA...B is the observed (differential) cross section, a...b label the different par-
tons and A...B their hadrons (we leave off the labels for non-strongly interacting
470 VIII. GAUGE LOOPS

particles), the sum is over different kinds (and flavors) of partons (and perhaps over
different hard parts, if corrections down by powers are desired), ξa is the momentum
fraction for parton a of hadron A’s momentum, fAa is either the parton distribu-
tion function for A → a + X (X = “anything”) or the fragmentation function for
a → A + X, and dσa...b is the hard cross section (calculated perturbatively), which
is just the original with all the hadrons replaced by partons. For the parton distri-
R1 R∞
butions we integrate 0 dξ, while for fragmentation we integrate 1 dξ, or change
variables to the hadron’s fraction of the parton’s momentum ζ = 1/ξ and integrate
R1
0
dζ.
Note that, while physical cross sections are independent of the renormalization
mass scale µ, the same is not true of the hard cross sections calculated perturbatively
in the above factorized expressions, since they are expressed in terms of unphysical
quark “states”. However, these hard parts satisfy renormalization group equations, as
calculated in the usual perturbative way. (Of course, nontrivial contributions require
calculating beyond leading order.) This implies corresponding renormalization group
equations (see subsection VIIC1), the “evolution” or “Gribov-Lipatov-Dokshitzer-
Altarelli-Parisi (GLDAP) equations”, to be satisfied by the parton distributions, so
that µ dependence cancels in the complete cross sections. This determines the energy
dependence of the parton distributions. The equations take the form
X Z 1 dζ  
2 d 2 2 ξ 2 2
µ fAa (ξ, µ ) = fAb (ζ, µ )Pba , g (µ )
dµ2 b ξ ζ ζ

where fAa describes A → a+X, fAb describes A → b+X 0 , the “splitting functions” Pba
describe b → a+X 00 (X = X 0 +X 00 ), and the sum is over the intermediate parton b. For
hadron A with momentum p, the intermediate parton b has momentum ζp, and parton
a has momentum ξp, so ξ/ζ is a’s fraction of b’s momentum. The kinematics are such
that 0 ≤ x ≤ ξ ≤ ζ ≤ 1 (momentum is lost to X’s as A → b → a). The splitting
functions are calculated perturbatively from the corresponding renormalization group
equation for the hard part, since the combined µ dependence must cancel in the
physical cross section.
For similar reasons, the hard cross sections are infrared divergent; the soft parts
of the complete cross sections deal with low energies. This leads to complications
beyond next-to-leading order, due to the fact that the renormalization group scale µ,
which relates to ultraviolet divergences (high-energy behavior), and the “factorization
scale”, which relates to infrared divergences (it determines the division between hard
and soft energies), are in principle independent scales. This allows an ambiguity in
factorization prescriptions, in addition to the usual ambiguity in UV renormalization
 C. HIGH ENERGY 471

prescriptions. (In more general processes there can be other energy scales than just
q 2, each with its own factorization scale, further complicating matters.)

REFERENCES
1 R.P. Feynman, Phys. Rev. Lett. 23 (1969) 1415;
J.D. Bjorken and E.A. Paschos, Phys. Rev. 185 (1969) 1975:
parton model.
2 J.C. Collins, D.E. Soper, and G. Sterman, Factorization of hard processes in QCD, in
Perturbative quantum chromodynamics, ed. A.H. Mueller (World-Scientific, 1989) p. 1;
G. Sterman, Introduction to perturbative QCD, in Perspectives in the standard model ,
proc. TASI ’91, Boulder, Colorado, June 2-28, eds. R.K. Ellis, C.T. Hill, and J.D.
Lykken (World-Scientific, 1992) p. 475;
Sterman, loc. cit.;
G. Sterman, hep-ph/9606312, Partons, factorization and resummation, in QCD & be-
yond , proc. TASI ’95, Boulder, Colorado, June 4-30, ed. D.E. Soper (World-Scientific,
1996) p. 327:
reviews of factorization.
3 G. Sterman and S. Weinberg, Phys. Rev. Lett. 39 (1977) 1436:
justification for jets from perturbative QCD.
4 V.N. Gribov and L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 438, 675;
Yu.L. Dokshitzer, Sov. Phys. JETP 46 (1977) 641;
G. Altarelli and G. Parisi, Nucl. Phys. B126 (1977) 298.
472 IX. GENERAL RELATIVITY

PART THREE: HIGHER SPIN

Higher-spin (unstable) particles have been observed experimentally. Whether
they are considered elementary depends on how their theory is formulated. In partic-
ular, a description of hadrons in terms of strings would have many advantages, such
as unification of all hadrons, manifestation of duality symmetry, and calculability
through an accurate perturbation scheme.
Gravity and supergravity also include higher-spin particles. String theory might
also yield some solutions to some of their problems, especially renormalizability and
unification of all particles. Such gravitational strings would differ from hadronic
strings in their mass scale and in the appearance of massless particles, including the
graviton itself (in contrast to the massive “pomeron”, the analog of the graviton in
hadronic strings). Gravitational strings might require supergravity.
For these and other reasons supergravity and strings are two of the major areas of
research in theoretical high energy physics today (although not the only ones). Most
of the discussion of this part is introductory, and can be covered earlier, but it is not
essential to the course; however, its inclusion in a field theory text is essential at least
for reference.

IX. GENERAL RELATIVITY

Before discussing supergravity we need to study ordinary gravity. Both can be
treated as generalizations of Yang-Mills theory. We use this approach rather than
the traditional one, based on the metric, which is insufficient for describing spinors or
supersymmetry: There is no useful definition of distance in anticommuting directions
in curved (super)space.
Gravity is the only observed long-range (massless) force mediated by a higher-
spin (2) field. It is relevant for astrophysics, cosmology, and unification, all of which
have applications to the particles of lower spin.

............................ ............................
............................ A. ACTIONS ............................

We begin with the general principles that define pure gravity as a nonabelian
gauge theory, and use them to derive actions and couple to matter.
 A. ACTIONS 473

1. Gauge invariance

General relativity can be described by a simple extension of the methods used to

describe Yang-Mills theory. The first thing to understand is the gauge group. We start
with coordinate transformations, which are the local generalization of translations,
since gravity is defined to be the force that couples to energy-momentum in the same
way that electromagnetism couples to charge. However, these are not enough to define
spinors. This is easy to see already from the linear part of coordinate transformations:
Whereas SO(3,1) is the same Lie group as SL(2,C), GL(4) (a Wick rotation of U(4))
does not have a corresponding covering group; there is no way to take the square root
of a vector under coordinate transformations. So we include Lorentz transformations
as an additional local group. We therefore have a coordinate transformation group,
which includes translations and the orbital part of Lorentz transformations, and a
local Lorentz group, which includes the spin part of Lorentz transformations.
Clearly the coordinates xm themselves, and therefore their partial derivatives ∂m ,
are not affected by the (spin) Lorentz generators. We indicate this by use of “curved”
vector indices m, n, .... On the other hand, all spinors should be acted on by the
Lorentz generators, so we give them “flat” indices α, β, .., and we also have flat vector
indices a, b, ... for vectors that appear by squaring spinors. Flat indices can be treated
the same way as in flat space, with metrics Cαβ and ηab to raise, lower, and contract
them.
Some gravity texts, particularly the more mathematical ones, emphasize the use
of “index-free notation”. An example of such notation is matrix notation: Matrix
notation is useful only for objects with two indices or fewer, as we saw in our treatment
of spinor indices in chapter II. Such mathematical texts consider the use of indices
as tantamount to specifying a choice of basis; on the contrary, as we have seen in
previous chapters, indices in covariant equations usually act only (1) as place holders,
indicating where contractions are made and how to associate tensors on either side of
equations, and (2) as mnemonics, reminding us of representations and transformation
properties. Thus, the full content of the equation can be seen at a glance. In contrast,
many mathematical-style equations (when indeed equal signs are actually used) say
little more than “A = B”, with the real content of the equation buried in the text of
preceding paragraphs.
We therefore define the elements of the group as

g = eλ , λ = λm ∂m + 12 λab Mba
474 IX. GENERAL RELATIVITY

where ∂m acts on all coordinates, including the arguments of the real gauge parameters
λm and λab and any fields. Mab = −Mba are the second-quantized Lorentz generators:
They act on all flat indices, including those on λab and any fields that carry flat
indices. As a shorthand notation, sometimes we will also write
1 ab
2 λ Mba = λI MI

(and similarly for other appearances of the antisymmetric index pair ab).
In general, the relation between first- and second-quantized group generators is
the same as the relation between active and passive transformations, and the relation
between a matrix representation and the corresponding coordinate representation,
as discussed in subsection IC1, where in this case the fields are the coordinates. In
particular, the second-quantized Lorentz operators Mab have the same action as the
first-quantized Lorentz operators Sab introduced in subsection IIB1: For any field ψ,
Mhψ| = hψ|S, etc. The action of the Lorentz generators on vector indices is thus
given by

[Mab , Vc ] = V[a ηb]c ⇒ λI [MI , Va ] = 12 λbc [Mcb , Va ] = λa b Vb

This implies the commutation relations

[c
[Mab, M cd ] = −δ[aMb] d]

In explicit calculations, only two indices in the commutator will match, and they
reduce to simple expressions such as

[M12, V2 ] = η22V1 , [M12, M23] = η22 M13

As for derivatives, when acting on functions instead of operators we can write the
action of the Lorentz generator as simply Mab Vc without the commutator.
When spinors are involved in four dimensions, it’s simpler to convert all flat
indices to spinor indices. In that case, we can write
..
λ = λm ∂m + λI MI , λI MI = 12 λab Mba = 12 λαβ Mβα + 12 λαβ M . .
βα

[Mαβ , ψγ ] = ψ(αCβ)γ ⇒ λI [MI , ψα] = 12 λβγ [Mγβ , ψα ] = λα β ψβ

(γ
[Mαβ , M γδ ] = δ(αMβ) δ)
in terms of the SL(2,C) generators Mαβ = Mβα . Note that (Mαβ )† = +M . . because
αβ
(Mab )† = −Mab . We have used conventions consistent with OSp generators
1 BC
2λ [MCB , ψA} = λA B ψB , ηAB = (ηab , Cαβ , C . .)
αβ
 A. ACTIONS 475

Relating vector to spinor indices as usual as Va = Vαα., etc., then fixes the Lorentz
subgroup of the OSp group as (see excercise IIB7.2a)

M . .Vγ γ. = Vαα.Cβγ C . . − V .Cαγ Cα.γ.

ααβ β βγ ββ

= − 12 (C . .V(αγ.Cβ)γ + Cαβ Vγ(α.C . .) = − 12 (C . .Mαβ + Cαβ M . .)Vγ γ.

αβ β)γ αβ αβ

⇒ M . . = − 12 (C . .Mαβ + Cαβ M . .)
ααβ β αβ αβ

⇒ λ . . = C . .λαβ + Cαβ λ . .
ααβ β αβ αβ

For most of the remaining discussion of gravity, we’ll limit ourselves to bosonic fields
in vector notation, which is easy to generalize to arbitrary dimensions. For spinors,
we must either choose a dimension and use its corresponding spinor notation (for
D ≤ 6), or work in mixed spinor-vector notation (which is much messier).
Matter representations of the group work similarly to Yang-Mills. We define such
fields to have only flat indices. Then their transformation law is

ψ 0 = eλ ψ

where the transformation of a general Lorentz representation follows from that for a
vector (or spinor, if we include them), as defined above. Alternatively, the transforma-
tion of a vector could be defined with curved indices, being the adjoint representation
of the coordinate group:

V 0 = V 0m ∂m = eλ V e−λ
m∂ m∂
V = V m ∂m ⇒ m m

However, as in Yang-Mills theory, it is more convenient to identify only the gauge

field as an operator in the group. In any case, only the adjoint representation (and
direct products of it) has such a nice operator interpretation.
As an example of this algebra, we now work out the commutator of two transfor-
mations in gory detail: We first recall that the coordinate transformation commutator
was already worked out in subsection IC2, using the usual quantum mechanical rela-
tions (see also subsection IA1)

[f, f] = [∂, ∂] = 0, [∂, f] = (∂f)

for any function f. For the Lorentz algebra we will use the additional identities

[Mab, ∂m ] = [Mab , λm ] = [Mab, λcd Mdc ] = 0

476 IX. GENERAL RELATIVITY

all expressing the fact the Lorentz generators commute with anything lacking free flat
indices (i.e., Lorentz scalars). The commutator algebra is then

[λm 1 ab n 1 cd
1 ∂m + 2 λ1 Mba ,λ2 ∂n + 2 λ2 Mdc ]

= λm n m 1 ab 1 ab 1 cd
[1 [∂m , λ2] ]∂n + λ[1 [∂m , 2 λ2] ]Mba + 2 λ1 [Mba , 2 λ2 ]Mdc

= (λn[1 ∂n λm 1 m ab ac b
2] )∂m + 2 (λ[1 ∂m λ2] + λ[1 λ2]c )Mba

One fine point to worry about: We may consider spaces with nontrivial topologies,
where it is not possible to choose a single coordinate system for the entire space. For
example, on a sphere spherical coordinates have singularities at the two poles, where
varying the longitude gives the same point and not a line. (However, the sphere can
be described by coordinates with only one singular point.) We then either treat such
points by a limiting procedure, or choose different sets of nonsingular coordinates on
different regions (“patches”) and join them to cover the space.

2. Covariant derivatives
We can also define covariant derivatives in a manner similar to Yang-Mills theory;
however, since ∂m is now one of the generators, the “∂” term can be absorbed into
the “A” term of ∇ = ∂ + A:

∇a = eam ∂m + 12 ωa bc Mcb

in terms of the“vierbein (tetrad)” eam and “Lorentz connection” ωa bc . Now the action
of the covariant derivative on matter fields looks even more similar to the gauge
transformations: e.g.,

δφ = λm ∂m φ, ∇ a φ = e a m ∂m φ

δVa = λm ∂m Va + λa b Vb , ∇aVb = ea m ∂m Vb + ωab c Vc

δψα = λm ∂m ψα + λα β ψβ , ∇aψβ = eam ∂m ψβ + ωaβ γ ψγ
I.e., the covariant derivative ∇a is essentially D elements (labeled by “a”) of the
gauge algebra.
Excercise IXA2.1
Write the transformation law and covariant derivative of an antisymmetric
tensor in spinor notation (fαβ ), and compare to vector notation as above.
Note that the free index on the covariant derivative is flat so that it transforms
nontrivially under
∇0a = eλ∇ae−λ
 A. ACTIONS 477

Explicitly, for an infinitesimal transformation δ∇ = [λ, ∇] we have

δeam = (λn ∂n eam − ean ∂n λm ) + λa b eb m

δωa bc = λm ∂m ωa bc + (−eam ∂m λbc + ωa d[b λd c] + λa d ωd bc )

This commutator is the same as for [λ1 , λ2 ] in the previous subsection, except for the
two additional terms coming from the Lorentz generators acting on the free index on
∇a. In particular, the vierbein eam transforms on its flat index as the vector (defining)
representation of the local Lorentz group, and on its curved index (and argument)
as the vector (adjoint) representation of the coordinate group. Also, it should be
invertible, since originally we had ∇ = ∂ + A: We want to be able to separate out
the flat space part as eam = δam + ha m for perturbation theory or weak gravitational
fields. That means we can use it to convert between curved and flat indices:

V m = V a ea m ⇔ V a = V m em a

where em a is the inverse of eam . Furthermore, if we want to define the covariant

derivative of an object with curved indices, we can simply flatten its indices, take the
covariant derivative with ∇, and then unflatten its indices.
Flat indices are the natural way to describe tensors: (1) They are the only way
to describe half-integral spin. (2) Even for integer spin, they correspond to the way
components are actually measured. For example, consider nonrelativistic momentum
in two flat spatial dimensions, but in polar coordinates: Now using xm to represent
just the nonrelativistic spatial coordinates,
dxm . .
xm = (r, θ), pm = m = (mr, mθ) = pa eam
dt
 
m 1 0 a . .
ea = −1
, p = (m r, mr θ)
0 r
Then the two components of pa (with the simplest choice of eam ) are the usual com-
ponents of momentum in the radial and angular directions. On the other hand, one
component of pm is still the radial component of the momentum, while the other
component of pm is the angular momentum — a useful quantity, but not normally
considered as a component along with the radial momentum, which doesn’t even have
the same engineering dimensions.
Excercise IXA2.2
Show that the above choice of eam actually describes flat space: Use the fact
that pa transforms as a scalar under the coordinate transformations that ex-
press r and θ in terms of Cartesian coordinates x and y, and as a vector under
478 IX. GENERAL RELATIVITY

local “Lorentz” transformations, which are in this case just 2D rotations, to

. .
transform it to the usual Cartesian p0a = (mx, my).
This direct conversion between curved and flat indices also leads directly to the
covariant generalization of length: In terms of momentum (as would appear in the
action for the classical mechanics of the particle),
dxm
pm = m , −m2 = p2 = pa pb ηab ⇒ −ds2 = dxm dxn em a enb ηab ≡ dxm dxngmn
ds
Equivalently, the metric tensor gmn is just the conversion of the flat-space metric ηab
to curved indices. Also, in terms of differential forms,

Ωa = dxm em a ⇒ −ds2 = Ωa Ωb ηab

The field strengths are also defined as in Yang-Mills:

[∇a, ∇b ] = Tabc ∇c + 12 Rab cd Mdc

where we have expanded the field strengths over ∇ and M rather than ∂ and M so
that the “torsion” T and “curvature” R are manifestly covariant:

M 0 = eλMe−λ = M ⇒ T 0 = eλT e−λ, R0 = eλ Re−λ

The commutator can be evaluated as before, with the same change as for going from
[λ1 , λ2 ] to [λ, ∇a] (i.e., now there are two free indices on which the Lorentz generators
can act), except that now we rearrange terms to convert ∂m → ea m ∂m → ∇a . Making
the further definitions

e a = e a m ∂m , [ea, eb ] = cab c ec ⇒ cab c = (e[aeb] m )em c = −eam ebn ∂[m en] c

for the “structure functions” cab c , we find the explicit expressions

Tabc = cab c + ω[ab]c = −ea m eb n (∂[m en] c + e[m d ωn]d c )

Rab cd = e[aωb] cd − cab e ωe cd + ω[a ce ωb]e d = eam ebn (∂[m ωn] cd + ω[m ce ωn]e d )

If we ignore the action of ∇ on curved indices (it doesn’t act on them, but alternatively
we could flatten them, act, then curve them back), we can also write

Tmna = −∇[m en] a , [∇m , ∇n ] = 12 Rmn ab Mba

where
∇m = em a ∇a = ∂m + 12 ωm ab Mba
 A. ACTIONS 479

is essentially a covariant derivative for the Lorentz group only.

From this expression for the torsion we find the following expressions for the curl
and divergence of a vector in terms of curved indices: Defining

e ≡ det eam

⇒ cba b = (e[b m ∂m ea]n )en b = (ebm ∂m ean )en b − eam [(∂meb n )en b ] = ∂m eam − eam ∂m ln e

we have
ea m eb n ∂[m Vn] = e[a (eb]m Vm ) − cab c ec m Vm = ∇[aVb] − Tab c Vc

e ∂m e−1 V m = eam ∂m V a + cba b V a = ∇aV a + Tba b V a

Excercise IXA2.3
Relate the two above identities by comparing (in D=4) ∇[aVbcd] (generalizing
∇[aVb] ) and ∇aV a for Vabc = abcd V d .
In practice, a useful way to evaluate the commutator is to first evaluate the com-
mutators of the Lorentz generators with the whole covariant derivative, and then
subtract out the double-counted [M, M] term. This is particularly convenient when
considering some explicit solution to the field equations with a reduced set of com-
ponents (e.g., spherically symmetric), so that explicit indices may be lost except on
the Lorentz generators. Schematically, we then calculate

[∇1, ∇2] = [e1 + ω1 , e2 + ω2 ]

= {[e1, e2 ] + (e1ω2 )M2 − (e2 ω1 )M1 } + {ω1 [M1, ∇2] − ω2 [M2, ∇1] − ω1 ω2 [M1, M2 ]}

This method turns out to be the simplest way to calculate explicit solutions (as
opposed to discussing general properties). (For examples, see subsection IXC5 below.)
The covariant derivative satisfies the Bianchi (Jacobi) identities

0 = [∇[a, [∇b, ∇c]]] = [∇[a, Tbc]d ∇d + 12 Rbc] de Med ]

= (∇[aTbc] d )∇d + 12 (∇[a Rbc] de )Med − T[ab|e (Te|c] f ∇f + 12 Re|c] f g Mgf ) − R[abc] d ∇d

⇒ R[abc] d = ∇[a Tbc]d − T[ab|e Te|c] d , ∇[aRbc] de − T[ab|f Rf |c] de = 0

To make the transformation laws manifestly covariant we can define instead

λ = λa ∇a + 12 λab Mba
480 IX. GENERAL RELATIVITY

which is just a redefinition of the gauge parameters. The infinitesimal transformation

law of the covariant derivative is then

δ∇a = [(δeam )em b ]∇b + 12 (ea m δωm bc )Mcb = [λb ∇b + 12 λbc Mcb , ∇a]
= (−∇aλb + λc Tca b + λa b )∇b + 12 (−∇aλbc + λd Rda bc )Mcb

⇒ (δeam )emb = −∇aλb + λc Tcab + λa b , eam δωm bc = −∇aλbc + λd Rda bc

A “Killing vector” is a transformation that leaves the covariant derivative invari-

ant. (The term is usually used to refer to just the general coordinate part λm of the
transformation, but we’ll use it in a generalized sense to refer to the complete λ.)
It represents a symmetry; the existence of Killing vectors depends on the particular
space described by the covariant derivative. It then follows from the Jacobi identity
for [λ1, [λ2 , ∇]] that the Killing vectors form a group, the symmetry group of that
space. Invariance of the covariant derivative requires:

−∇aλb + λc Tca b + λa b = 0 ⇒ ∇(aλb) = λc Tc(ab), λab = 12 ∇[aλb] − 12 λc Tc[ab]

−∇aλbc + λd Rda bc = 0 ⇒ ∇aλbc = λd Rda bc

These equations are referred to as the “Killing equations”. (Again, usually it is just
the first equation, on λa , that is called by this name, but we’ll use it to refer also
to the equations for λab , which are needed to describe the symmetry when acting on
spinors, etc.)
Excercise IXA2.4
Express the Hamiltonian of the classical relativistic particle in terms of the
vierbein:
ea = eam pm , H = 12 (η ab eaeb + m2 )

Doing the same for general coordinate transformations λ = λa ea, examine

the condition for invariance [λ, H] = 0 using the Poisson bracket. Using the
commutation relations for the ea’s, show that this implies the Killing equation
∇(aλb) = λc Tc(ab).
Excercise IXA2.5
Solve the Killing equations explicitly in the case of flat space ∇a = ∂a. Show
this gives the Poincaré group, including both orbital and spin pieces.
 A. ACTIONS 481

3. Conditions
There are two kinds of conditions we can impose to eliminate some degrees of
freedom: gauge choices and constraints. Gauge choices explicitly determine degrees
of freedom that drop out of the action anyway. If the gauge is not completely fixed,
the form of the residual gauge transformations may change, since using particular
gauge parameters to fix the gauge, rather than eliminating those parameters, may
just determine them in terms of the remaining parameters: We require that the
residual transformations do not violate the gauge conditions that have already been
applied. Similar remarks apply to global symmetries: If they do not commute with
the gauge transformations for the gauge that was fixed, then they may aquire extra
gauge-transformation terms to preserve the gauge choiuce. On the other hand, con-
straints are chosen to be covariant under the transformation laws, and thus do not
alter them, while eliminating degrees of freedom that might otherwise appear in the
action (although not in all possible terms). Furthermore, the simplest explicit solu-
tion to constraints can itself introduce new gauge invariances. (An example of this
situation is supersymmetric Yang-Mills: see subsections IVC3-4.) In this subsection
this analysis will be applied to Lorentz invariance: We already saw that global Lorentz
transformations are included in coordinate transformations, and that local Lorentz
invariance is unnecessary when only integer spin (and in particular, pure gravity) is
treated. We now examine the consequences of eliminating this useful but redundant
invariance and the gauge field associated with it.
Of course, we can eliminate local Lorentz transformations by hiding flat indices:
For the vierbein itself, we have the local Lorentz invariant

g mn = η ab eam eb n

which is the inverse “metric tensor”. However, we have seen that tensors with flat
indices have simpler coordinate transformations, and there is no way to get rid of flat
indices when spinors are involved. Furthermore, the metric has the constraint that it
have Minkowski signature: This constraint is solved by expressing the metric in terms
of the flat-space Minkowsi metric η and the vierbein. Thus, solving the constraint
introduces local Lorentz invariance. (However, in this case the constraint does not
eliminate degrees of freedom, but only limits their range.)
The Lorentz transformations in λab are redundant to those in λm . The extra
gauge parameters also can be fixed by an appropriate gauge choice: For example,
consider the gauge

δeam = λa b eb m ⇒ Lorentz gauge ηm[a eb] m = 0

482 IX. GENERAL RELATIVITY

A coordinate transformation takes us to a different Lorentz gauge, since the Lorentz

gauge condition is not a scalar. This means that any coordinate transformation λm
must be accompanied by a Lorentz transformation λab to preserve this gauge, where
this λab is completely determined in terms of λm . This is easy to see perturbing ea m
about δam: To lowest order we have simply

0 = δ(ηm[a eb]m ) ≈ −λab + ∂[a λb] ⇒ λab ≈ ∂[aλb]

Excercise IXA3.1
Let’s further analyze this gauge condition:
a By looking at the transformation of a vector, identify the specific terms in
the Taylor expansions of λm and λab whose coefficients can be identified with
global Lorentz transformations, in the approximation used above.
b Using the same methods as excercise IVC4.3, and writing in matrix notation
eam = (eh )a m for some matrix h, solve explicitly for λab in terms of λm and
eam to all orders.
Similarly, the Lorentz connection ωa bc that gauges the λab transformations is
redundant to the vierbein that gauges λm : ω can be completely determined in terms
of e by constraining the torsion to vanish. To see this, we first notice that in the
general case the expression for the torsion in terms of the structure functions and
connection can be inverted to give the connection in terms of the other two. One
way to do this is to use the definition and permute the indices a → b → c (odd
permutations are redundant because of the antisymmetry of the equation in the first
two indices):

Tabc = cabc + ωabc − ωbac , Tbca = cbca + ωbca − ωcba , Tcab = ccab + ωcab − ωacb

Using the antisymmetry of the connection in its last two indices, we add the first and
last equation and subtract the second to obtain

ωabc = 12 (c̃bca − c̃a[bc] ), c̃abc = cabc − Tabc

Since the torsion is a covariant tensor, we can freely set it to vanish without affecting
the transformation laws of the remaining objects (it’s a covariant constraint, not a
gauge condition):
Tabc = 0 ⇒ ωabc = 12 (cbca − ca[bc] )

From now on we assume this constraint is satisfied. This simplifies the form of curls
and divergences, which implies that ∇ can be integrated by parts in covariant actions
 A. ACTIONS 483

(see below). However, we have already seen that the torsion is nonvanishing in su-
perspace (subsection IVC3): In that case the symmetry on flat indices is constrained,
so the connection has fewer components than the torsion, and can be determined by
setting only part of the torsion to vanish. (See subsection XA1 below.)
Excercise IXA3.2
Show explicitly that when the torsion vanishes the Killing equations from
δeam = 0 imply those from δωm ab = 0:

∇(aλb) = 0, λab = 12 ∇[aλb] ⇒ ∇a λbc = λd Rda bc

Excercise IXA3.3
Consider using the group GL(D) on the flat indices instead of SO(D−1,1).
(This construction is not useful for fermions.) Compensate for the extra gauge
invariance by replacing the Minkowski metric ηab with a “flat”-index metric
gab (and its inverse g ab ) that is coordinate dependent, but covariantly constant:

λ = λm ∂m + λa b Gb a ; λa b [Gb a , Vc ] = λc aVa , λa b [Gb a , V c ] = −V a λa c

∇a = ea + ωab c Gc b , [∇a, ∇b ] = Tabc ∇c + Rabc d Gd c

g mn = g ab eam eb n , ∇ag bc = 0

where now there is no (anti)symmetry associated with the indices on Ga b (or

λa b , etc.). As a result, gab transforms nontrivially under both coordinate and
GL(D) transformations. Use it (in place of η) to raise and lower flat indices.
a Find the explicit expressions for the torsion and curvature in terms of the
vierbein and connection. Solve these, and ∇g = 0, for the connection in
terms of the torsion and vierbein as

ωabc = 12 (c̃bca − c̃a[bc] ) + 12 (ec gab − e(agb)c ), c̃abc = cabc − Tabc

Show that there exists a GL(D) gauge

gab = ηab

(assuming gab has the right signature), that gauge has as a residual flat-index
invariance SO(D−1,1), and the resulting covariant derivative is identical to
that used earlier in this subsection.
b Show that one can instead choose a GL(D) gauge

eam = δam ⇒ g mn = g ab
484 IX. GENERAL RELATIVITY

and that this completely fixes the GL(D) invariance. Since the vierbein has
a curved index, the covariant derivatives are no longer covariant: Unlike the
previous gauge, to maintain this gauge any coordinate transformation must
be accompanied by a GL(D) transformation whose parameter is determined
by the coordinate transformation parameter. Find the solution for λa b in
terms of λm in the infinitesimal case. Compare with the transformation law
for curved indices (see subsection IC2). In this gauge the connection is known
as the “Christoffel symbols”.
The vanishing of the torsion simplifies the Bianchi identities on the curvature:

Tabc = 0 ⇒ R[abc]d = ∇[a Rbc]de = 0 ⇒ Rabcd = Rcdab

In terms of SU(N)-like Young tableaux, this means the curvature is of the form .
For SO(N) Young tableaux, we subtract out the trace pieces:

Rabcd → ⊕ ⊕·

where the first term is the “Weyl tensor” Wabcd (traceless), the last two terms combine
to give the “Ricci tensor” Rab ≡ Ra c bc , and the last (singlet) term is the “Ricci scalar”
R ≡ Ra a = Rab ab . They’re simpler in spinor notation in D = 4: Since [ab] → (αβ)
..
and (αβ),
Rabcd → R . . ⊕ R(αβ)(γδ) = W(αβγδ) + C(α(γ Cδ)β)R
(αβ)(αβ)

in terms of Weyl W(αβγδ), the traceless part of Ricci R . . , and the Ricci scalar
(αβ)(αβ)
R. Later we’ll see that the Ricci tensor is fixed exactly by the equations of motion.
That leaves the Weyl tensor as the on-shell field strength. As explained in subsection
IIB7, it describes helicity ±2.
Excercise IXA3.4
Prove that Rabcd = Rcdab follows from the Bianchi identity R[abc]d = 0 and the
antisymmetry of Rabcd in both ab and cd.

4. Integration
The antihermitian form of the group generators was a convenient choice because
partial derivatives are antihermitian, and the generators of the Lorentz group (which
is real and orthogonal) are antisymmetric in the vector representation. Thus, the
generators are real. However, the group elements are not unitary, since hermitian
conjugation reorders λm with respect to ∂m . The fix comes from noticing that

e = det eam ⇒ δ ln e = ema δeam = λm ∂m ln e − ∂m λm

 A. ACTIONS 485

← ← ←

⇒ δ(e−1) = e−1 λ = e−1 λm ∂ m ⇒ (e−1 )0 = e−1 e λ

←
where the derivatives ∂ act on everything to the left, λ now includes just coordinate
transformations, and we have exponentiated by the same method as for Lie groups
in subsection IA3. (Note that if we expand the exponential in a Taylor series such
derivatives in all but the first factor will hit λ’s, just as for those in eλ acting to the
right.) Any function that transforms in this way is known as a “density” (see subsec-
tion IIIB1 for the 1D case). We can easily see from the infinitesimal transformation
that a density times any scalar is also a density. This allows invariant actions to be
constructed as Z
S = dx e−1 L

for any scalar L. For cases without spinors we can also use
√
g ≡ det gmn = −e−2 ⇒ e−1 = −g

where gmn is the inverse of g mn. This can also be understood in terms of differential
forms, since

Ωa = dxm em a ⇒ Ω4 = dxm dxn dxp dxq em 0 en1ep 2 eq 3 = d4 x e−1

Z 0 Z
0a 0
Ω (x ) = Ω (x) ⇒
a
Ω L = Ω4 L
4

under coordinate transformations.

Excercise IXA4.1
Let’s look at some properties of transformations acting backwards:
a Show that for any function f
← ← ←

λf = [λ, f] = [f, λ] ⇒ eλf = eλfe−λ = e− λ fe λ

and use it to show that the product of e−1 with any scalar transforms the
same way as e−1 (i.e., is a density) under a finite coordinate transformation.
b Derive ←
 ←

λ
fe = 1·e λ
(eλ f)
←
(where the derivatives in each factor of λ act on everything to the left, but
vanish on “1”).
Excercise IXA4.2
We now examine finite transformations in terms of transformed coordinates
(see subsection IC2):
486 IX. GENERAL RELATIVITY

b Show that  
∂x0 ←

det = 1 · e− λ
∂x
by evaluating
Z Z  
−1 0 0−1 0 0 ∂x0
dx e (x) = dx e (x ), dx = dx det
∂x

c Show that   ←
∂ x̃
det = 1 · eλ
∂x
by similarly evaluating
Z Z Z
dx e (x) = dx̃ e (x̃) = dx e0−1(x)
−1 −1

From the results of subsection IXA2, we then have that covariant derivatives can
be integrated by parts in such actions, since
Z Z
Tba = 0 ⇒
b
dx e ∇aV = dx ∂m e−1 V m
−1 a

Excercise IXA4.3
Derive the expression for the covariant divergence in terms of e and the partial
divergence by assuming integration by parts:
Z Z
e φ∇a V = − e−1 V a ∇a φ
−1 a

Use this to find a simple form for the covariant d’Alembertian on a scalar:
1 √
φ ≡ ∇2φ = √ ∂m −gg mn ∂nφ
−g

Actions for matter are constructed in a similar way to Yang-Mills: Starting with
the flat-space action, replace ordinary derivatives with covariant derivatives. The new
ingredient is the extra factor of e−1. This prescription, as for Yang-Mills, is unam-
biguous up to only field-strength (curvature) terms, which can usually be eliminated
by symmetry requirements and dimensional analysis. (At least for low energies, we
want terms of the lowest mass dimension.) This uniqueness (at low energies or long
distances) is known as the “equivalence principle”: Inertial “mass” (really energy,
but also momentum), as determined by the kinetic term, is the same as gravitational
mass, as determined by the coupling of the gravitational field.
A simple example of matter is a real scalar field:
Z
S = e−1 14 [(∇χ)2 + m2 χ2 + aRχ2]
 A. ACTIONS 487

The constant a can sometimes be fixed by symmetry: In the massless case, to preserve
the global symmetry δχ = , we must have a = 0. (For self-interacting scalars, this
generalizes to a global nonabelian symmetry.) To preserve conformal symmetry (see
subsection IXA7), also for the massless case, we need a = 14 D−2
D−1
.
This form of actions in terms of scalar Lagrangians also suggests we modify the
definition of functional variation for convenience and covariance:
Z
δS
δS = dx e−1 (δφ)
δφ
or, equivalently, we use the covariant form of the δ function,
δφ(x)
= e(x)δ(x − x0)
δφ(x0)

As in flat space, the action for electromagnetism follows from gauge invariance:
Z Z
−1 2
√
1
S = 8e2 e Fab = 8e2 1
−gg mn g pq Fmp Fnq

where Fmn = ∂[m An] . Integration by parts then gives a simple form for Maxwell’s
equations. Such simple covariant equations of motion that don’t require explicit
expressions for the Lorentz connection appear only for antisymmetric tensors (which
in practice means just spin 0 and 1 in 4D).
Excercise IXA4.4
Methods related to differential forms can be applied to these special cases:
a Rewrite the above action for electromagnetism in terms of Aa and covariant
derivatives. Find the field equations following from both forms of the action,
and use this to find a simple expression for the covariant divergence of an
antisymmetric tensor with curved indices using just the metric. Compare the
results of the previous excercise.
b By converting flat indices on the covariant tensor abcd to curved, show that
√ 1
−gmnpq and √ mnpq
−g

are also covariant tensors. Use these, and the covariance of the curl (see
subsection IC2), to arrive at the same expression for the covariant divergence
of an antisymmetric tensor.
Another example is a Dirac spinor:
Z
S = e−1Ψ̄(γ a i∇a + m
√
2
)Ψ
488 IX. GENERAL RELATIVITY

where γ a are the usual constant Dirac matrices, in terms of which the spin operator
appearing in ∇ is the usual Mab → Sab = − 12 γ[a γb] . In 4D, we can rewrite this in
spinor notation by simply replacing ∂ . → ∇ . in the flat-space expressions given in
αβ αβ
subsection IIIA4, and replacing Mab → Mαβ as described in subsection IXA1, as well
R R
as d4 x → d4 x e−1 .

5. Gravity
The Einstein-Hilbert action for gravity follows from choosing the only available
scalar second-order in derivatives, the Ricci scalar:

LG = − 14 R = − 14 Rab ab

This action normally has a coefficient of 1/κ2 (compare Yang-Mills), but we’ll gener-
ally use (natural/Planck) units κ = 1; then κ is used only to parametrize expansion
about the vacuum and define the weak-field limit. (Actually, Planck units normally
use G = 1, whereas in our conventions κ = 1 → G = π.) In any case, the κ’s can
always be absorbed (unlike Yang-Mills) by a field redefinition of eam , and then appear
only in the definition of the “vacuum” (perturbative ground state, or solution that
defines the boundary conditions at infinity):

heam i = κ2/(D−2)δam

This makes eam ∂m , and thus dxm em a and ds2 , dimensionless. In this sense, gravity is
a theory with “spontaneous breakdown” of conformal invariance: Coordinate trans-
formations include conformal transformations, but this invariance is broken by the
vacuum, which introduces a length scale (κ).
Excercise IXA5.1
Consider the covariant derivative for nonvanishing torsion. By solving for the
Lorentz connection in terms of the structure functions and torsion, express the
◦
covariant derivative in terms of the torsion-free covariant derivative ∇ and the
◦
torsion. Thus, any action in terms of ∇ can be rewritten in terms of ∇ and T ,
so any theory with a nonvanishing torsion is equivalent to a similar one with
vanishing torsion (assuming the action is only second-order in derivatives of
the vierbein, and thus algebraic in the torsion). Take the commutator of two
◦
∇’s to find the curvature in terms of the torsion-free curvature Rabcd . Write
◦ ◦
the Einstein-Hilbert action with nonvanishing torsion in terms of R, ∇, and
T , to find:
◦ ◦
R = R − (Tab b )2 − 12 T abcTbca + 14 T abcTabc − 2∇a Tabb
 A. ACTIONS 489

Since the last term vanishes upon integration, T appears as an auxiliary field,
◦
so R is equivalent to just R.
Excercise IXA5.2
For some general applications, where the form of the vierbein is not specified,
it is useful to have a more explicit expression for the action in terms of the
vierbein. We found in subsection IXA2 that for vanishing torsion

∇aV a = e ∂m (e−1ea m V a ) = eaV a − cab b V a

Use this to show

R = (cab b )2 + 12 cabc cbca − 14 cabc cabc − 2e ∂m [eam∂n(ea n e−1 )]

We can drop the last term in the action integral under appropriate boundary
conditions. (Hint: Use the result of the previous excercise for ω = 0.)
Excercise IXA5.3
In two dimensions there is a single Lorentz generator,

Mab = ab M ⇒ ∇a = ea + ωa M, [∇a, ∇b] = − 12 ab RM

a Show that the connection and the only surviving part of the curvature then
take the simple forms
← ←
ωa = −ab e∂me−1 eam , R = −2e∂m [eam∂n(e−1 ean )] = −2e−1 e a e a e

b Derive, for the sphere in spherical coordinates,

 
m 1 0
ea =
0 sin1 θ
(Hint: First use ds2 = dxm dxn gmn in 3D flat space.)
R
b Use these results to calculate dx e−1R for the sphere in two ways: (1) by
showing R is a constant and pulling it out of the integral, and (2) by converting
it into a boundary term, where the “boundary” consists of infinitesimal circles
around the coordinate singularities at the poles. (In general, even for spaces
without true boundaries, one has to treat the boundaries of patches as such.)
It’s also possible to add a “cosmological term” to the gravitational action:
Z
Scos = Λ dx e−1

with the “cosmological constant” Λ. This term has no derivatives, and is thus anal-
ogous to a mass term. However, it only contributes to the nonpropagating spin-0
490 IX. GENERAL RELATIVITY

mode of the vierbein (see later), so it doesn’t give a physical mass, but does modify
the vacuum.
Excercise IXA5.4
Show that the action for gravity can be made polynomial in ea m by a field
redefinition (rescaling) of the form

ea m → ek ea m

when k takes the values

n+1
k=− , n = 2, 3, 4, ...
D−2
and that the resulting action is order Dn + 2 in the field. In what cases (of n
and D) is the cosmological term also polynomial?
The variation of the curvature can be obtained directly by varying its definition
in terms of [∇, ∇]. We start with the definition

δeam ≡ ζa b eb m ⇔ ζa b ≡ (δeam )em b

and work in terms of the flattened object ζab . Then we drop its Lorentz piece, choosing
ζab = ζba . We find:
δ∇a = ζa b ∇b + 12 ζa bc Mcb
⇒ 12 (δRab cd )Mdc = [∇[a, δ∇b]] = (∇[aζb] c )∇c − ζ[ac 12 Rb]c de Med + 12 ∇[aζb] cd Mdc + ζ[ab] c ∇c

⇒ ∇[aζb] c + ζ[ab] c = 0, δRabcd = ∇[aζb] cd − ζ[a e Rb]e cd

⇒ ζabc = ∇[b ζc]a

⇒ δRab cd = 12 {∇[a, ∇[c}ζb] d] − 12 (ζ[a e Rb]e cd + ab ↔ cd )

δe−1 = −e−1 δ ln e = −e−1em a δeam = −e−1 ζ a a

⇒ δ(e−1R) = 2e−1 [(η ab − ∇a∇b ) + (Rab − 12 η ab R)]ζab

where ≡ ∇a∇a . Thus for pure gravity we have the field equations

δSG = 0 ⇒ Rab − 12 ηab R = 0 ⇒ Rab = R = 0

while with a cosmological constant we have

δSG +δScos = 0 ⇒ Rab − 21 ηab (R−4Λ) = 0 ⇒ Rab − D1 ηabR = 0, D

R = 4 D−2 Λ

Note that calculating a variation is the same as performing a perturbation to

lowest order: We will use this result in subsection IXB1.
 A. ACTIONS 491

Excercise IXA5.5
For gravity, a first-order formalism follows from not imposing the torsion con-
straint (see excercise IXA5.1), so either the torsion or the Lorentz connection
can be treated as the auxiliary variable.
a Find a first-order action for gravity (in all D) by treating em a and ωm ab as the
independent variables. In D=4, using mnpq , write this action as polynomial
in these variables, eliminating the explicit e, to obtain
Z
SG = d4 x 16 1 mnpq
 abcd em a enb Rpq cd

with Rpq cd in terms of just ω.

b Vary this action with respect to both e and ω (independently) to find the
field equations, expressed in terms of torsion and curvature, using δ[∇m, ∇n]
to find the variation of Rpq cd (see subsection IXA2).
Excercise IXA5.6
As discussed in subsection IIIC4 for Yang-Mills, in four dimensions we can
write a complex first-order action for gravity that yields the usual gravity
action up to a surface term. For Yang-Mills, the complex action was obtained
by starting with a normal first-order formalism and replacing the auxiliary
field with its self-dual part.
a Starting with the first-order action of the previous problem, find the analog for
gravity by keeping just the part of ωm ab self-dual in ab, using spinor notation.
b Associate the coupling κ with the term quadratic in ω (analogously to the
Yang-Mills case). As for Yang-Mills, associate the self-dual theory with the
limit κ → 0. Find the equation for em a that follows from varying ω in this
◦
case, and show that it is equivalent to setting the self-dual part of ω m ab to zero,
◦
where ω is the usual torsion-free connection. Show this is equivalent to setting
the self-dual part of the curvature Rmn ab to vanish, in an appropriate gauge.
(Technically, this means we must either complexify the fields, or Wick rotate
to 4+0 or 2+2 space+time dimensions, where the Lorentz group factorizes.)

6. Energy-momentum
In subsection IIIB4 we saw that in the same way as a current in electrodynam-
ics or Yang-Mills is defined as the matter contribution to the gauge field’s equation
of motion, δSM /δAa = J a (in that case SM excludes only the pure Yang-Mills ac-
tion), the “energy-momentum tensor” is defined as the matter contribution to the
492 IX. GENERAL RELATIVITY

gravitational field equation (in this case SM excludes only the pure gravity action):
Z Z Z
−1 ab
√ √
δSM = e ζ Tab = 2 1
−g(δg )Tmn = − 2
mn 1
−g(δgmn )T mn

The case where ζab represents the invariances of the action implies restrictions on this
tensor: Using the separate gauge invariance of the matter action δgauge SM = 0 and
the matter field equations δSM /δ(matter) = 0 (as for the Yang-Mills case), gauge
variation of the gravity fields in SM implies
(
λab = −λba ⇒ T [ab] = 0 : Lorentz
ζab =
− 12 ∇(aλb) ⇒ ∇aT ab = 0 : coordinate

so coordinate invariance of the action implies local conservation of energy-momentum.

For example, for a real scalar field:
Z
S = e−1 14 [(∇χ)2 + m2 χ2 + aRχ2]

⇒ 2Tab = (∇aχ)(∇bχ) − 12 ηab[(∇χ)2 + m2χ2 ] + a[(ηab − ∇a∇b ) + (Rab − 12 ηab R)]χ2

Notice that for a 6= 0, the energy-momentum tensor gets extra total-derivative terms
which are separately conserved in flat space (since they come from the Rχ2 term,
which is separately covariant).
Excercise IXA6.1
Using the action given in subsection IXA4 and the variation of the covariant
derivative from subsection IXA5, find the energy-momentum tensor for the
Dirac spinor, and use its field equations to show this tensor is conserved.
R√
Note that this is not the same as ordinary conservation ∂m T mn = 0: −gT 0n
does not define a conserved total energy-momentum. This is in contrast with the
conserved current in electrodynamics, since we then can derive the usual global con-
servation law
Z Z Z
−1 −1 m d
0 = d x e ∇ a J = d x ∂m e J ∼
D a D
dD−1 x e−1 J 0
dt
On the other hand, it’s closely related to Yang-Mills, where δAa = −∇aλ leads to
∇aJ a = 0 in terms of the derivative ∇ covariantized with respect to the Yang-Mills
field (as well as gravity, if in curved space), so ∂m e−1J m = −e−1 [iAm, J m] 6= 0 (see
subsection IIIC1).
However, if there is a Killing vector Ka , then the component of momentum in
that direction is conserved:

J a ≡ Kb T ba ⇒ ∇a J a = (∇aKb )T ba + Kb (∇aT ba) = 0

 A. ACTIONS 493

(Remember ∇(aKb) = 0.) Some simple examples of this in flat space are (Ka )b = δab
(translational invariance), for which the corresponding “charge” is the total momen-
[b
tum, and (Ka )bc = δa xc] (Lorentz invariance), for which the charge is the total angular
momentum.
Including the variation of the gravitational action, we get the gravitational field
equations
Rab − 12 ηab R = 2Tab

Coordinate invariance of SG implies ∇a (Rab − 12 η ab R) = 0, which also follows from

the Bianchi identities: In that sense gauge invariance is said to be “dual” to Bianchi
identities, one implying the other through variation of the action: In general, for any
gauge field φ with gauge parameter λ
Z
δS δS
δφ = Oλ, 0 = δS = dx (Oλ) ⇔ OT =0
δφ δφ

where the “transpose” OT is defined by integration by parts. Positivity of the energy

(contained in any infinitesimal volume) is the condition T 00 ≥ 0. The addition of
the cosmological term modifies the left-hand side of the above equation of motion by
adding a term 2ηab Λ.
Although there is no covariant definition of total energy-momentum, in the case
where spacetime is asymptotically flat (the metric falls off to the flat metric sufficiently
fast at infinity), one can define a noncovariant energy-momentum tensor tab for gravity
itself which is covariant with respect to coordinate transformations that themselves
fall off at infinity. (See excercise IIIC1.2 for the analogous Yang-Mills case.) This
tensor satisifes ∂m (T mn + tmn) = 0 (where T mn is the usual tensor for matter), so the
usual conservation laws can be derived for the total energy-momentum coming from
integrating T + t. Many equivalent expressions exist for t. One way to derive it is to
expand the field equations order-by-order in h as

1
2 (Rab − 12 ηab R) ≡ Lab − tab

where Lab is the linearized part of the field equations (see subsection IXB1) and −tab
is the quadratic and higher-order parts. By the linearized Bianchi identities, we know

0 = ∂a Lab ≡ ∂a ( 12 Rab − 14 η ab R + tab) = ∂a(T ab + tab)

where we used the field equations in the last step. Note that there is a great deal
of ambiguity here: We could have linearized by expanding the metric around its flat
space value instead of the vierbein, or by expanding Rmn or Rmn instead of Rab , etc.
494 IX. GENERAL RELATIVITY

Because of the expression in terms of Lab ∼ ∂∂h, the integral of T + t, which gives the
total energy-momentum vector, can be expressed as a surface term, just as Gauss’
law in electrodynamics. Since space was assumed to be asymptotically flat, only the
quadratic part of t contributes in the surface integral, which is why there is so much
freedom in the definition of t. Since t is not covariant, the energy-momentum of
the gravitational field is not localized (coordinate transformations shift it around).
However, since the total energy-momentum is invariant, one can ask questions about
how much energy is radiated to infinity, etc.

7. Weyl scale
The simplest way to describe conformal transformations in field theory is as a
local scale transformation. If the theory is not coupled to gravity, we couple it to
gravity as in Yang-Mills theory by replacing a Poincaré invariant Lagrangian L(∂, ψ)
with L(∇, ψ) (where all fields ψ have flat indices), but also including the e−1 factor
in the action. We then transform the fields as

eam → Φeam , ψ → Φw+(D−2)/2 ψ

where Φ is the gauge parameter and w + D−2 2

is the engineering dimension (scale
weight) of the field ψ. (See subsection IIB1.) Effectively, eam has dimension 1, since
it’s the only field with curved indices, and thus any derivative must appear in the
combination eam ∂m , while the measure appears as dx e−1 . Of course, the action
won’t be locally scale invariant unless it is globally scale invariant, i.e., has only
dimensionless coupling constants (and thus no masses).
If the gravity-coupled theory is invariant under this local scale transformation,
then the theory will be conformally invariant after decoupling gravity. This follows
from the fact that the most general combined coordinate and local scale and Lorentz
transformation that preserves the flat-space vierbein ea m = δam is exactly a conformal
transformation. This is equivalent to our previous definition in terms of the scal-
ing of the flat-space ds2 under conformal transformations, since dx0m dx0ngmn 0
(x0) =
dxm dxn gmn (x) under coordinate transformations.
Excercise IXA7.1
Derive the usual conformal transformations by finding the most general local
scale + Lorentz + coordinate transformation that preserves the flat-space
vierbein.
A simple example is Yang-Mills theory. We look at the Yang-Mills field with
curved index, since its gauge transformation does not depend on the vierbein. (δAm =
 A. ACTIONS 495

∂m λ + ... vs. δAa = ea m ∂m λ + ....) To avoid interference with the Yang-Mills gauge
transformation, the Yang-Mills field with curved index must be scale invariant. Then
the action Z
S = 8e2 e−1 eam ebneap eb q FmnFpq
1

transforms with a factor Φ4−D , and so is invariant in D = 4 only.

Excercise IXA7.2
Consider a more general gauge field A and field strength F defined by
1 1
δAm1 ···mN = ∂ λ
(N −1)! [m1 m2 ···mN ]
, Fm1 ···mN +1 = ∂ A
N ! [m1 m2 ···mN +1 ]

where A is totally antisymmetric in its N indices. (Such theories were en-

countered in excercise IIB2.1b.)
a Define an action in terms of F 2. In what dimension D(N) is it conformally
invariant?
b Show that this theory is related by a “duality transformation” (switching
Bianchi identities and field equations) to the theory with N 0 indices on a new
A, where N 0 = D − 2 − N, and D(N 0 ) = D(N).
c Examine the cases N = D, D − 1, D − 2. Note that the scalar obtained by
duality does not have an Rφ2 term in its action, and thus is conformal only
in D=2.
Gravity is not scale invariant, but it will prove useful to examine its scale breaking
explicitly. To preserve gauge covariance and dimensional analysis, the scale transfor-
mation law of the covariant derivative must take the form

∇0a = Φ∇a + k(∇bΦ)Mab

where the Φ scaling of eam was defined above, and the linearity of δω in Φ follows
from the homogeneity of ∇ in e. (Alternatively, we could put in something more
arbitrary, but it would be eliminated by the rest of the procedure anyway.) From the
variation of commutation relations we then find
1 0 cd
2 Rab Mdc = [∇0a, ∇0b ]

= Φ2 [∇a, ∇b ] + (1 − k)Φ(∇[aΦ)∇b] + kΦ(∇[a∇c Φ)Mb]c + k 2 (∇Φ)2Mab

R0ab cd = Φ2 Rab cd + Φδ[a
[c
⇒ k = 1, ∇b] ∇d]Φ − δ[a
c d
δb](∇Φ)2
If we make the redefinition (at least for Φ positive)

Φ = φ−2/(D−2)
496 IX. GENERAL RELATIVITY

then we find the very simple scaling law for the integrand of the Einstein-Hilbert
action:
(e−1 R)0 = e−1(φ2 R − 4 D−1
D−2
φ φ)

Consider a field theory without gravity that has a conformally invariant action.
Spontaneous breakdown of scale invariance produces a Goldstone boson for that sym-
metry, the “dilaton”. Any theory can be made globally conformally invariant trivially
by performing a local scale transformation and making the parameter the dilaton field.
The dilaton can also act as a Higgs field: If we couple the dilaton to conformal
gravity (gravity with local Weyl scale invariance), the Higgs effect reduces conformal
gravity to ordinary (Einstein) gravity. For example, if we introduce the dilaton into
pure gravity by the local scale transformation above (in analogy to the Stückelberg
model), Z
SG = 4 D−1
D−2
dx e−1 14 φ( − 1 D−2
4 D−1
R)φ

Up to an (important) overall negative factor, this is the action for a conformal scalar.
The dilaton field φ is a compensator for local scale transformations, and acts as a Higgs
field for this gauge symmetry: By gauging it to its vacuum value hφi = κ1 , we regain
the usual form of the gravity action. (Alternatively, we can set hφi = 1, and introduce
κ through the proportionality constant in hea m i ∼ δam .) In this formalism, where
we require the action to be locally scale invariant, the terms which were conformally
invariant before coupling to gravity are easy to recognize: They’re just the ones which
have no φ-dependence. (This may require some field redefinition: typically rescaling
the matter fields according to their weight as above.) The cosmological term becomes
R
Scos = e−1 Λφ2D/(D−2), which is a conformal self-interaction term for a scalar.
Because what was the vierbein now appears only in the combination eam →
φ−2/(D−2)ea m , there is now the local scale invariance

ea m → Φeam , φ → Φ(D−2)/2φ

since this transformation leaves the combination invariant. Gauge invariance of the
matter action is then (using the infinitesimal parameter Φ = 1 + ζ):
δSM D−2 δSM
0 = δSM ∼ eam + 2
φ
δeam δφ
δSM
⇒ T aa = − D−2
2
φ
δφ
Thus, conformal matter has vanishing T a a, since it decouples from φ. (Actually, we
also need to scale the matter as above to achieve this decoupling, and there is a
 A. ACTIONS 497

corresponding δSM /δψ term in the above derivation, so the trace may vanish only
after applying the matter field equations, as in the derivation of ∇aT ab = 0 from
coordinate invariance in the previous subsection.) In particular, this is easy to check
. .
for the massless point particle, where T aa ∼ X m X n gmn = 0.
An interesting effect is obtained by eliminating the compensator by its field equa-
tion. (We’ll consider just the classical theory here: In the quantum case, integrating
out this field produces an additional 1-loop contribution to the effective action.) Be-
cause this manipulation involves integration by parts, we first expand the compensator
about its vacuum (asymptotic) value:

φ = 1 + 12 χ ⇒ L = 14 [χ( D−1
D−2
− 14 R)χ − Rχ − R]

Then eliminating χ by its field equation,

!
1
L→ 1
R R−R
4
R − 4 D−1
D−2

This action still describes Einstein gravity, but is locally scale invariant (though
not globally, because of the extraction of the vacuum value, and the way boundary
terms were neglected). Of course, it is nonlocal, and the nonlocality becomes more
complicated if nonconformal matter is included. Such terms also appear quantum
mechanically: In two dimensions, dimensionally regularizing D=2+2, in a Weyl scale
invariant theory we can get a divergent, yet still Weyl scale invariant, contribution to
the effective action proportional to
!
1 1 1 1
R R − R ≈ − R − 12 R R
 R − 4 D−2
D−1

After renormalizing the divergent term, which is topological and thus locally scale
invariant in exactly D=2, but not in D=2+2, the remaining finite term contributes
a conformal anomaly (see subsections VIIIA8 and C1).
Excercise IXA7.3
The statement that the R term is topological in D=2 neglects boundaries. In
general the topological invariant (the “Euler number”) is (the “Gauss-Bonet
theorem”) Z 2 I
d x 1 −1 1 ab taDtb
χ= e R +
2π 2 2π ηab tatb
where ta is a tangent vector to the boundary X m (τ ), as for the worldline of
the particle, and D is the covariant differential (as for the particle equation
of motion and the radial gauge; see subsections IXB2 and 4 below):
. .
tm = v −1X m , Dta = dX m em b ∇b ta = dτ vt · ∇ta = dτ (ta − vtbtc ωbc a )
498 IX. GENERAL RELATIVITY

(We have used the usual counterclockwise contour, and our convention 01 =
1, or xy = 1 in Euclidean space.) The additional term in χ is the angle
subtended by the boundary with respect to the surface (/2π), as obtained
from the cross product of t and t + Dt. We have written it in a form that
is manifestly invariant under the reparametrization of τ , and the v’s cancel.
(Of course, it is also manifestly coordinate invariant.)
a Prove that it is also scale invariant by showing that the connection part of
the D exactly cancels the contribution of R to the boundaries, leaving
Z 2  I
d x 1 −1  1 abta dtb
χ= e R   +
2π 2 patch boundaries 2π ηab ta tb

where we have turned the R term into a boundary term, and its remaining
contribution is from the fake boundaries at the borders of patches (or sur-
rounding singularities; R = ∂ω because the 2D Lorentz group is Abelian: see
excercise IXA5.3).
b Note that the dt term doesn’t contribute if we choose a gauge where

ta = δ1a

(i.e., tm = e1 m ). Demonstrate this by evaluating χ in polar coordinates for a

disk, and in spherical coordinates for the half-sphere. Show the result is half
that for a whole sphere (excercise IXA5.3). Repeat the calculation for the
disk in Cartesian coordinates (so then only the dt term contributes).

REFERENCES
1 A. Einstein, Sitz. Preuss. Akad. Wiss. Berlin, Math.-phys. Kl. (1914) 1030, (1915) 778,
799, 831, 844, Ann. der Phys. 49 (1916) 769:
general relativity.
2 É. Cartan, Leçons sur la géométrie des espaces de Riemann (Gauthier-Villars, 1928):
general relativity in terms of vierbein and (GL(D) or Lorentz) connection.
3 Weyl, loc. cit. (IC):
covariant derivatives on spinors.
4 L. Infeld and B.L. van der Waerden, Sitz. Preuss. Akad. Wiss. Berlin, Math.-phys. Kl.
(1933) 380:
general relativity in two-component spinor notation.
5 H. Weyl, Mat. Z. 2 (1918) 384:
Weyl tensor.
6 D. Hilbert, Nachrichten Königl. Ges. Wiss. Göttingen, Math.-phys. Kl. (1915) 395:
found the action for Einstein’s equations slightly earlier than Einstein.
7 A. Palatini, Rend. Circ. Mat. Palermo 43 (1919) 203:
first first-order action for gravity (in terms of metric and Christoffel symbols).
 A. ACTIONS 499

8 H. Weyl, Sitz. Preuss. Akad. Wiss. Berlin, Math.-phys. Kl. (1918) 465,
Raum·zeit·materie (Springer, 1919) p. 246 [English: Space-time-matter (Dover, 1952)]:
Weyl scale.
9 B. Zumino, Effective Lagrangians and broken symmetries, in Lectures on elementary
particles and quantum field theory, proc. 1970 Brandeis University Summer Institute in
Theoretical Physics, eds. S. Deser, M. Grisaru, and H. Pendleton (MIT, 1970) v. 2, p.
437:
dilaton.
10 E.S. Fradkin and V.I. Vilkovisky, Phys. Lett. 73B (1978) 209:
nonlocal action.
11 C.W. Misner, K.S. Thorne, and J.A. Wheeler, Gravitation (Freeman, 1970):
introductory, long-winded.
12 S.W. Hawking and G.F.R. Ellis, The large-scale structure of spacetime (Cambridge
University, 1973):
mathematical.
13 R.M. Wald, General relativity (University of Chicago, 1984):
intermediate between the above two.
14 S. Weinberg, Gravitation and cosmology (Wiley & Sons, 1972):
old-fashioned.
500 IX. GENERAL RELATIVITY

............................. .............................
............................. B. GAUGES .............................

We now consider various gauge choices for coordinate, Lorentz, and scale trans-
formations.

1. Lorentz
We begin with gauges that preserve global Lorentz invariance, which are useful
for perturbation theory. Therefore, we look first at perturbation by finding the kinetic
term, which is sufficent for finding linear gauge conditions. (It can also be derived from
general principles, as will be shown in subsection XIIA5.) We expand the vierbein
about its flat value,
eam = δa m + ha m

At the linearized level, local Lorentz invariance implies that only the symmetric part
of the field, 12 h(ab) , appears in the curvature and the action; we will denote this by hab
to simplify notation. (In other words, the linearized curvature is invariant under the
linearized local Lorentz transformations, which gauge away the antisymmetric part
of the field. This is equivalent to working directly with the metric.) We then can find
the linearized curvature, e.g., from the results of subsection IXA5 for the variation of
the curvature, by considering variation about flat space: i.e., replacing ζab → hab and
∇a → ∂a. The result is
Rab cd ≈ ∂[a∂ [c hb] d]

⇒ Rab − 12 ηab R ≈ hab + ∂a∂b hc c − ∂(a∂ c hb)c − ηab ( hc c − ∂ c∂ d hcd )

Since this comes from varying the action, the quadratic part of the gauge-invariant
action must be
Z
SG ≈ − 14 [hab hab + 2(∂ b hab )2 − haa hb b + 2ha a∂ b ∂ c hbc ]

This part of the action, and the linearized curvature, are invariant under the linearized
gauge transformations δhab = −∂(aλb) .
Excercise IXB1.1
Take the Newtonian (weak-field, nonrelativistic) limit of gravity: (1) Linearize
the action by perturbing about flat space (ea m = δam +ha m ). Keep just the part
of the pure gravity action quadratic in the perturbation, the part of the matter
coupling linear in it, and the complete flat-space matter action. (2) Assume
small velocities. Now consider the problem of a massive point particle in the
 B. GAUGES 501

field of a much more massive point particle (or spherical body); in the above
approximations:
a Show the effect of the gravitational field generated by the heavier particle on
the lighter particle is given by the action for the lighter particle (in the gauge
x0 ≡ t = τ ) Z
.
S = ms ≈ dt (m − 12 mx2i + mh00)

b Show this field is given by solving Laplace’s equation

R00 ≈ ∆h00 = T00

c Show that, with our conventions for normalizing functional differentiation, a

point mass M in D=4 generates
Mπ
T00 = M(2π)2 δ 3(x) ⇒ h00 = −
r
using the usual solution to Laplace’s equation for a point source. Combining
these results, we see that the potential energy for the particle is
Mmπ
U = mh00 = −
r
which agrees with Newtonian gravity if we identify G = π. (If we restore
units, this becomes G = κ2 π.)
The BRST transformations (see subsection VIA4) for gravity again follow from
the gauge transformations:

Qeam = C n ∂n ea m − ean ∂nC m + Ca b eb m

QC m = C n ∂n C m, QCab = C n∂nCab + Ca c Ccb

QC̃ m = −iB m, QC̃ab = −iBab
(Other forms follow from different parametrizations of the gauge transformations,
and are equivalent to field redefinitions. For theories without spinors, we can work in
terms of the metric, and avoid Lorentz gauge fixing.)
Lorentz gauges for coordinate invariance are similar to Yang-Mills. For gravity,
the gauge-fixing function is
fa = ∂ bhab − 12 ∂ahb b
The BRST procedure works similarly to Yang-Mills. Looking at just the graviton
kinetic term, the gauge-fixed quadratic Lagrangian for gravity is then, in the Fermi-
Feynman gauge,

LG → LG,F F = LG + 12 (∂ b hab − 12 ∂a hb b )2 = − 14 hab hab + 18 ha a hb b

502 IX. GENERAL RELATIVITY

plus ghost terms. Note that the trace part of h appears with opposite sign to the
traceless part. This prevents any redefinition which would allow rewriting the La-
grangian in the simple form − 14 hab hab . However, all derivatives have been absorbed
into ’s, which makes the linearized field equation a simple Klein-Gordon equation.
There are various generalizations of this gauge condition to include nonlinear
terms, such as the “de Donder (harmonic) gauge”, which uses the gauge-fixing func-
tion
√
f n = 12 ∂m ( −gg mn )

For example, this allows the field equation for a scalar to be written with only terms
with both partial derivatives acting on the scalar.

2. Geodesics
Consider the field equations for coupling gravity and electromagnetism to a scalar
particle: From subsection IIIB3, the action for a particle in external fields, rewritten
in Hamiltonian form, is
Z
.
SH = dτ {−xm em a (x)[πa − Aa(x)] + vH}, H = 12 π 2 + φ(x)

where we have pulled the v out of H for convenience, and use the “covariant momen-
tum”
πa = ea m pm + Aa(x) = ea m (pm + Am )

in place of pm (the canonical conjugate to xm ) for covariance. All the equations of

motion except the Lagrange-multiplier constraint
1 2
2π +φ=0

follow from the usual Poisson-bracket relation

.
v −1 O = i[H, O]

which can be evaluated by using the canonical commutation relations (following from
the simpler ones for pm )

i[πa, xm ] = eam , i[πa, πb ] = cab c πc + Fab, [x, x] = 0

Thus, πa acts effectively like −iea + Aa, which is the covariant derivative for gravity
.
and electromagnetism, less the Lorentz term. The x equation is the obvious
.
vπ a = xm em a
 B. GAUGES 503

that follows from varying SH with respect to πa , while the equation of motion for π
is
.
v −1π a = −ca bc πb πc − Fab πb − ∇aφ
Using the relation

0 = Tabc = cabc + ω[ab]c ⇒ ca(bc) + ω(bc)a = 0

we find
.
v −1 πa − π b π c ωbc a + F ab π b + ∇a φ = 0
This is the coordinate-covariant form of the Lorentz force law (plus scalar field).
With only the gravitational effects we have the covariantization of the free particle
equation,
.
Dpa ≡ pa − vpb pc ωbc a = 0
where “D” is understood as a covariantized τ derivative (along a worldline with metric
v).
It’s useful to consider a continuum of particles (“dust”) moving under the influ-
ence of these fields, such that any two infinitesimally close particles have infinitesi-
mally different velocities, and only one particle passes through any particular point
in spacetime (at least within some small region of spacetime). We then can treat πa
(or pm ) as a field defined for all x: Choosing a point x also chooses a curve X(τ ) for
which x = X(τ ) for some τ , so we can write π(x) in place of π(τ ). Specifying the
field π also determines this family of curves, since the tangent to any curve is given
by the X equation of motion
.
X m = vπ aea m
(To determine the τ parametrization, we also specify v, and the hypersurface given by
the collection of points X(0) from each curve.) Then we can express the τ derivative
in terms of x derivatives:
d .
= X m ∂m = vπ aea
dτ
which gives the manifestly covariant form of the equation of motion

π b ∇ b πa + Fa b πb + ∇ a φ = 0

For vanishing F (and thus A) and constant φ (= 12 m2), this equation

pb ∇b pa = 0

describes “geodesics”, which are curves of extremal length, since the action is

S = ms, −ds2 = dxm dxngmn

504 IX. GENERAL RELATIVITY

for massive particles. These are the analogs of straight lines in flat space. (For
positive-definite metric, they are shortest lines. Because of the indefinite signature of
the Minkowski metric, the worldlines of massive particles are actually longest lines,
while massless particles travel along lines with no length.)
For some purposes we need a weaker (but equivalent) form of the geodesic equa-
tion: If for some scalar f and vector na

(n · ∇)na = fna ⇒ pa = gna , (p · ∇)pa = 0, f = −(n · ∇)ln g

and we can integrate f to find g. In particular, we can identify

.
g = v −1 ⇒ nm = X m

Thus, the more general geodesic equation allows arbitrary parametrization of the
geodesics, while the stricter version (f = 0) corresponds to affine parametrization
.
(v = 1) if we still want to identify p with X. (Remember, as with all constrained
systems, the equations of motion p · ∇p = 0 imply (d/dτ )p2 = 0, so any geodesic
satisfying the stricter equation will have some fixed mass along that particular curve.)
Excercise IXB2.1
Show that in D=2 (one space dimension, one time) any lightlike curve is a
geodesic, using the weaker form of the geodesic equation. (Find f.) This is
a consequence of the fact that it is impossible to change direction in D=2
without slowing down.
The particle (geodesic) version of the conservation of momentum in the direction
of a Killing vector is
d
p · ∇ pa = 0 ⇒ p·∇ K·p =0 ⇒ K·p=0
dτ
where covariant conservation p · ∇ has become ordinary conservation d/dτ (no con-
nection term) because K · p is a scalar. (See also excercise IXA2.4.) This is the same
as for the conserved current J a = Kb T ba (subsection IXA6).
Excercise IXB2.2
Check explicitly that the current and energy-momentum tensor are covari-
antly conserved for the scalar particle coupled to gravity and electromag-
netism (see subsection IIIB4), using the geodesic equation of motion for the
latter. (Note: In subsection IIIB4 the δ’s need e’s to be covariant in curved
space.)
 B. GAUGES 505

3. Axial
The definition of axial gauges in terms of the covariant derivative is the same as
for Yang-Mills (n · ∇ = n · ∂). In terms of the explicit gravity fields,

n·∇=n·∂ ⇒ nm ≡ na ea m = na δam , na ωa bc = 0

In the case of gravity, this implies that lines in the na direction are geodesics (see
previous subsection):

∂m na = 0 ⇒ (n · ∇)n = (n · ∂)n = 0

To analyze the consequences of axial gauge conditions for the metric, we need a
further identity: For any vector field na , consider the action of n · ∇ on nm = em a na,
treating it as a scalar; in this calculation we ignore any indirect action of ∇ on curved
indices. Then

(n · ∂)ema na = (n · ∇)em a na = em a (n · ∇)na + na (n · ∇)em a

The last term simplifies for vanishing torsion, since:

nn ∇n em a = nn ∇[n em]a + nn ∇m ena = −nn Tnm a + nn∇m ena = ∇mna − ena ∇m nn

= ∇m na − en a ∂m nn

We thus have

(n · ∂)ema na = em a (n · ∇)na + ∂m ( 12 n2a ) − (ena na )∂m nn

Applying this identity to the axial gauge condition, we find

n · ∇ = n · ∂, ∂m na = 0 ⇒ (n · ∂)ema na = 0 ⇒ nm ≡ em a na = δm
a
na

by choosing the appropriate constants of integration. (This amounts to fixing a

residual gauge invariance.) We can now determine the form of the gauge condition
on the metric:

nm = na δam, a
nm = δm na ⇒ nm ≡ nngnm = nn ηnm

The lightcone gauge is again useful for eliminating unphysical degrees of freedom.
The linearized lightcone gauge conditions with na = δ−a
imply, for the unsymmetrized
h,
na eam = na δam ⇒ h+a = 0
506 IX. GENERAL RELATIVITY

ema na = δm
a
na ⇒ ea m δm
b
nb = na ⇒ ha+ = 0
and thus h(+a) = 0, so we can again work with just the symmetrized h. Separating
out the trace part as hij = hTij + δij h, where hTij is traceless, we find for the linearized
gauge-fixed action

L0G ≈ − 14 hab hab + 18 ha a hb b − 12 f 2

= − 14 hT ij hT ij − 12 (h0−i )2 + D−2
2
hh0−−

h0−i ≡ f i = −∂ +h−i + ∂ j hT ij − D−4 i

2
∂h
h0−− ≡ ∂ +f − + D−4
4
h = −∂ h +2 −− + i −i
+∂ ∂ h − D−2 + −
2
∂ ∂ h + D−4
4
h

where we have simplified some algebra by writing the gauge-invariant action as the
Lorentz gauge one minus its gauge-fixing terms. (There is some ambiguity in that
we can shift h0−i by a ∂ i h term, and absorb the generated terms into h0−− .) We see
that all but hTij are auxiliary fields (we redefined h−i and h−− by just shifting and
applying ∂ +), and can be eliminated (but watch out if there are matter couplings,
when eliminating them gives Coulomb-like interactions).
The temporal gauge (known also as “Gaussian normal coordinates”) is used when
treating time and space separately: In this case we have for the metric

nm = δ0m ⇒ g0m = η0m

An alternate way of defining the temporal gauge is to start with a spatial hypersurface,
and determine the geodesics normal to this hypersurface (g0i = 0), where the positions
on this hypersurface define xi, constant along the geodesics, and the proper times
along the geodesics define x0 (g00 = −1), with x0 = 0 at the hypersurface. The fact
that these are geodesics guarantees that the hypersurfaces of fixed, but nonvanishing,
(proper) time are still orthogonal to the geodesics (g0i stays zero):

n · V = 0, (n · ∇)n = 0 ⇒ n · [(n · ∇)V ] = 0

Equivalently, we can consider a dust of massive particles and choose an initial hyper-
surface orthogonal to their (timelike) geodesics to define x0 = s = 0. This coordinate
system is thus the “rest frame” of the dust; all the information about the geometry of
the space is contained in the time dependence of the spatial separation of the particles
(gij ). There is still the residual coordinate ambiguity of how to assign xi on the initial
hypersurface.
Gaussian normal coordinates thus can be useful for studying the dynamics of
particles: For example, we can study a gravitational field of distant, unknown (or
 B. GAUGES 507

ignored) origin (i.e., the curvature of spacetime) by watching the relative motion of
two nearby particles of such a dust, neglecting the gravitational force/curvature effect
acting between the two particles themselves. If the two particles start out relatively
at rest at some initial time (which is well-defined only if they are close and relatively
slow), then in the temporal gauge the paths of both particles are described by fixed
xi , independent of x0, since their geodesics are simply lines in the time (na = δ0a)
direction, and the proper times of both particles are the same as the time x0 . Then
the distance between the particles at any given time is given by the magnitude of
dxm em a , with dx0 = 0 and dxi their infinitesimal separation. Thus, since the xi ’s,
and thus dxi , are fixed, we want to study the change in em a (really just eia ; e0a = δ0a)
with time. Using our evaluation of (n · ∇)ema from above, we find

(n · ∇)2em a = (n · ∇)∇m na = [n · ∇, ∇m]na = nn [∇n, ∇m ]na = −nb nc Rbdc a em d

using the axial gauge condition n · ∇ = n · ∂. For the Gaussian case na = δ0a , we then
have
..
e m a = −R0b0 aem b
(Of course, vanishing curvature implies geodesics that start parallel remain that way,
because the space is then flat.) By observing different sets of particles initially at
rest with respect to each other, we can choose different timelike directions n, and
determine all the curvature components from their linear combinations.
Excercise IXB3.1
Let’s examine some 2D examples of axial gauges in spaces with positive-
definite metric:
a Gaussian normal coordinates need not be Cartesian in flat space. Show that
polar coordinates for the plane define an axial gauge. What is the coordinate
in the “na ” direction? Give the geodesic interpretation.
b Repeat the above for a curved space — the (2D) sphere in spherical coordi-
nates.
..
c Apply the above “equation of motion” ( e = −Re) to the sphere. (See excercise
IXA5.3.) Show its solution agrees with the obvious.

4. Radial
Another useful gauge similar to the axial gauge is the radial gauge (“Riemann
normal coordinates”), discussed for Yang-Mills in subsection VIB1. In this case we
have
nm = xm ⇒ (n · ∇)na = xm ∂m xnδna = na
508 IX. GENERAL RELATIVITY

a case of the more general form of the geodesic equation. Applying the same identity
as for the axial, we again have

n · ∇ = n · ∂, ∂m na = δm
a
⇒ (n · ∂)ema na = (n · ∂)δma na ⇒ em a na = δm
a
na

but now the boundary condition is already implied by the gauge condition near the
origin: For any infinitesimal xm = m ,

m em a (0) = m δm
a
⇒ em a (0) = δm
a

m ωm ab (0) = 0 ⇒ ωm ab (0) = 0

Thus, there is no residual gauge invariance, unlike axial gauges (where the coordi-
nates of the initial hypersurface need additional determination). Any reference frame
satisfying these conditions at the origin is called a “local inertial frame”, and is the
most natural for an observer at that point in spacetime. (In flat space, this yields
Cartesian coordinates.)
Excercise IXB4.1
We can think of Gaussian normal coordinates as defined by a dust of particles
with affine parametrization v = 1 and unit mass m = 1, with τ = s = x0 and
.
xi constant for any particle (X = p = n). For Riemann normal coordinates
we can think of particles radiating out from the origin xm = 0 in all possible
directions in space and time, but then some must be antiparticles (traveling
backward in time), some must be massless (for the lightlike geodesics), and
some must be tachyons, with m2 < 0 (for the spacelike geodesics). However,
as for the Gaussian case, we can still identify
.
nm = X m

Using the radial gauge condition, show that these can be chosen as geodesics
with
v = eτ , X(τ ) = eτ X(0), p = X(0)
so all particles start at the origin at τ = −∞, and their position at τ = 0
is determined by their initial (constant) momentum. (Thus particles with
proportional momenta travel the same path, but arrive at different points at
τ = 0; however, in this case τ is neither the time x0 nor the proper time s,
but just an arbitrary parameter.)
As we saw in subsection VIB1, the radial gauge is related to gauge-covariant
translation (in general relativity, “parallel transport”) as, for any tensor ψ,
a (y)D a (y)E m (y)D
ψ̃(ỹ) = ex a
ψ(y) = eΛψ(ỹ), ỹ m = ex a m
ym
 B. GAUGES 509

where y is the “origin”, Λ = ΛI MI is just a Lorentz transformation, and D is the

covariant derivative acting at y:
∂
Da = Ea m (y)Dm + $a I (y)MI , Dm = ; [Da , Db ] = Tab c Dc + Rab I MI
∂y m
As in general for coordination transformation parameters λa , xa now transforms under
local Lorentz transformations. (In background field language, this “quantum field”
transforms under the “background” Lorentz transformations.) Thus, xa is now a
function of y; it cannot be made even covariantly constant in general:

Da xb = 0 ⇒ 0 = [Da , Db ]xc = −xdRabd c

(For more practical reasons, if we defined it to be invariant or constant, the manip-

ulations that follow would break down.) At this point we have only made a Lorentz
transformation on ψ, since it and ψ̃ are evaluated at the same point ỹ. However, as
for Yang-Mills in subsection VIB1, for the next step we want to identify xa as the
new coordinate:
ψ 0(xa) = ψ̃(ỹ m (y m , xa)) = ex Da ψ(y)
a

where ψ 0 has implicit dependence on y, since in radial gauges the choice of origin
is gauge parameters that define the gauge. (The coordinates are defined as radial
with respect to the origin y.) Thus, we have made a Lorentz transformation ψ → ψ̃
followed by a coordinate transformation ψ̃ → ψ 0.
We also want to define a covariant derivative for x by

∇ψ 0 = (Dψ)0 = ex
aD
a
Da ψ(y)

where ∇ (as for Yang-Mills) contains only ∂a = ∂/∂xa and not Dm :

∇a = ea b (x)∂b + ωa I (x)MI

At this point we no longer distinguish flat and curved indices, since the Lorentz gauge
has been fixed. We have then transformed

y, ψ, D → x, ψ 0, ∇

Note that the y-coordinate tensors are the x-coordinate tensors evaluated at the
origin:
ψ 0(0) = ψ(y), (∇ψ 0)(0) = (Dψ)(y)
We can identify this as the radial gauge when x(y) satisfies the geodesic condition,
since then
(x · D)x = 0 ⇒ x0 ≡ ex·D x = x
510 IX. GENERAL RELATIVITY

⇒ x · ∇ψ 0 = x · (Dψ)0 = (x · Dψ)0 = x · Dψ 0 = x · ∂ψ 0

making use of ψ 0(x) = ex·D ψ(y).

Unfortunately, it is somewhat difficult to continue this construction in terms of
the covariant derivative, but simpler in terms of the “dual” differential forms. We
therefore define the (Lorentz-covariantized) Lie derivative as

Lx·D ψ̂ = x · Dψ̂, Lx·D D̂ = [x · D, D̂]

for any “tensor” (object carrying only flat indices) ψ̂ and any “covariant derivative”
(object with a flat vector index free, but multiplying partial derivatives and Lorentz
generators) D̂. We generalize to evaluate on not only ψ and D, but to apply the
Lie derivative also as part of the transformation exp(Lx·D ). For that reason, for the
remainder of this section we will abbreviate Lx·D as just L. We then have

[x · D, Da ] = −(Da xb )Db + xb (Tba c Dc + Rba I MI )

Defining Lie derivatives to satisfy the usual Leibniz and distributive rules like any
derivative (since we use them as infinitesimal transformations), we then find

(LEa m )Em b = −Da xb + xc Tca b , Ea m L$m I = xb Rba I

In terms of differential forms, defined as

E a = dy m Em a , $I = dy m $m I

D = E a Da = d + $I MI ⇒ Dxa = dxa − xb $b a

we then have
Lxa = 0, L(T , R) = x · D(T , R)

LE a = Dxa − E b xc Tcb a

L$I = E a xb Rba I ⇒ L(Dxa ) = −xb E c xd Rdcb a

which covers all the quantities that appear in evaluating eL on E a and $I . The
geodesic condition prevents higher derivatives of xa from appearing in the transfor-
mation law, and allows us to freely reorder all the x’s to the left at the end of the
calculation for identifying the coefficients of the Taylor expansion, at which point
we can forget that x depends on y. Thus, these few equations for the action of L
allow any transformed quantity to be evaluated straightforwardly by iteration, Taylor
expanding eL in powers of L.
 B. GAUGES 511

The important distinction between the transformation laws for Da and E a is

that for E a the derivatives of x appear only in the combination dx, which makes
changing coordinates from y (or ỹ) to x easier. Specifically, by iterating the above
Lie derivatives, we find a solution of the form

E 0a = eL E a = E b Ab a + (Dxb )Bb a , $0I = eL $I = E a AaI + (Dxa )BaI

where Ab a, Bb a , Aa I , BaI are functions of x and of tensors evaluated at the “origin”

((D · · · DT )(y), (D · · · DR)(y)). For Riemann normal coordinates, we want to fix y
(e.g., y = 0), so we evaluate the above at dy = 0. Furthermore, we can choose the
gauge ω(0) = 0 (at least for vanishing torsion), so also Dx → dx. Then the solution
is
E a = dxb Bb a , ω I = dxa BaI

Thus, Bb a and BaI are the inverse vierbein em a and Lorentz connection ωm I for the
new coordinate system,
∇a = (B −1 )ab (∂b + Bb I MI )

written explicitly as a Taylor expansion in x by the above method, all of whose

coefficients are tensors (torsions and curvatures and their derivatives) evaluated at
the origin.
However, we can also use these results for first-quantization (where actions are
expressed in terms of, e.g., E 0a /dτ for the particle) in background field gauges, by
choosing y as the background and x as the quantum coordinate (see subsection VIB1);
then we keep both the dy and Dx terms.
Excercise IXB4.2
Find the first few orders of this expansion.
a Using the above method, show that for vanishing torsion

E a = dxb (δba − 16 xc xd Rcbd a + ...), ωI = 12 dxa xb Rba I + ...

b Check the validity of this result by evaluating [∇, ∇] to this order from the
∇ given by this E and ω.
c Use instead the covariant-derivative method of subsection VIB1. In this case,
we find
∇a = ex·D̃ D̃a e−x·D̃ + ha b ex·D̃ ∂b e−x·D̃

where ha b is chosen to cancel all Dm terms in ∇, and we have defined

fI
D̃a = Ea m (y)Dm + $a I (y)M
512 IX. GENERAL RELATIVITY

where now xa is “constant”, so Dm and M f do not act on it. (Otherwise, in

this approach, we would be stuck with tons of D · · · Dx terms.) In terms of
the previously defined Lorentz generators,

fab + x[a ∂b]

Mab = M

Find h to this order, and use it to obtain

∇a = (δab + 16 xc xd Rcad b + ...)∂b + 12 ( 12 xb Rba cd )Mdc

f to M.
restoring M

5. Weyl scale
The gauge-fixed kinetic term can be simplified by including the conformal com-
pensator (see subsection IXA7). The quadratic part of the gauge-invariant Lagrangian
is then (φ = 1 + 12 χ)
L0 = − 14 e−1φ(R − 4 D−1
D−2
)φ
≈ − 14 [hab hab +2(∂ b hab )2 − ha a hb b +2ha a ∂ b∂ c hbc ] − 12 χ( ha a − ∂ a∂ b hab )+ 14 D−1
D−2
χ χ
The nicest (globally Lorentz) covariant gauge comes from choosing the coordinate
and scale gauge-fixing functions

fa = ∂ bhab − 12 ∂ahb b + 12 ∂a χ, f = χ − ha a

We use these to obtain the gauge-fixed Lagrangian (see subsection VIB9)

L = L0 + 12 (∂ bhab − 12 ∂ahb b + 12 ∂a χ)2 − 18 (χ − ha a ) (χ − hb b )

= − 14 hab hab + 1 1
4 D−2
χ χ

plus ghost terms. Now the h kinetic term is simpler. Also, remember that χ de-
couples from conformal matter. These features of gauge fixing make this formalism
closely analogous to the Stückelberg formalism for the massive vector. We can also
define nonlinear versions of these gauge-fixing functions, such as ∂m (φ e−1/2ea m ) or
√ √
∂m (φ2 −gg mn ) for the coordinate gauge, and φ e−1/2 or φ2 −g for the scale.
Excercise IXB5.1
Find the ghost terms for linearized gravity in the Fermi-Feynman gauge, and
its simplification with the compensator.
The scale gauge can also be fixed in terms of the vierbein/metric alone: For
example, we can fix the gauge
e=1
 B. GAUGES 513

in which case φ acts simply as a renaming of e. A more unusual gauge is

R=0

This is not a restriction on the geometry, since the physical Ricci scalar is effectively
replaced by its scale transform

R0 = φ−(D+2)/(D−2)(R − 4 D−1
D−2
)φ

which is scale invariant. In the gauge φ = 1, R0 = R, but in the gauge R = 0 it is

proportional to φ.
Excercise IXB5.2
Show that the ghosts for scale transformations propagate in the gauge R = 0:
Find their contribution to the action.
More general gauges are possible when matter fields appear. For example, con-
sider coupling gravity, with compensator, to a physical conformal scalar ψ. With
appropriate normalization of the compensator and physical scalar, the kinetic terms
for the two fields are identical except for sign: There is a manifest O(1,1) symmetry.
We can take advantage of this by using a “lightcone” basis for these fields: Defining
R
φ± = φ ± ψ, the full nonlinear (in gravity) Lagrangian L becomes (S = dx e−1L)

L = φ+( D−1
D−2
− 14 R)φ−

The overall normalization is arbitrary, including sign, since we can rescale either field
by a constant. Many Weyl scale gauges are possible, and somewhat more transparent
than making field redefinitions on the corresponding action without compensator.
Effectively, we can redefine the fields φ± arbitrarily as long as we don’t fix φ+ /φ− to
a constant, since that combination is scale invariant. (I.e., φ+ /φ− can be redefined,
but not fixed.)
Some of the more interesting choices are:

φ± = 1 ± ϕ ⇒ L = − 14 R − ϕ( D−1
D−2
− 14 R)ϕ
φ± = e±ϕ ⇒ L = − 14 R − D−1
D−2
ϕ ϕ
φ± = ϕ1±a ⇒ L = ϕ[(1 − a2) D−1
D−2
− 14 R]ϕ
φ+ = ϕ, φ− = 1 ⇒ L = − 14 Rϕ

We can also have any of these gauge-fixed Lagrangians with opposite overall sign,
simply by changing the choice of either φ+ or φ− by a sign. The first two choices
are useful because they put the action in standard form, as the usual gravity action
514 IX. GENERAL RELATIVITY

plus a physical scalar kinetic term. (Thus, coupling a massless scalar to gravity either
conformally or minimally is equivalent, and the two cases are distinguished only by
interactions.) In fact, the first choice, or “temporal gauge” φ+ + φ− = constant, just
returns us to the form without compensator, φ = 1. On the other hand, changing the
sign of φ− yields the “axial gauge” φ+ − φ− = constant, which is fixing the physical
scalar as ψ = 1. The overall sign of the action changes because the physical scalar
is traded for the compensator, or the corresponding part of the metric. This gauge
is closely related to the “string gauge”: In our third choice above the gravity action
is invisible until the surviving scalar has been expanded about its vacuum value.
The constant a is arbitrary except that it must not vanish (so that φ+/φ− is not a
constant). In particular, this action appears in string theory, with the choice
1
a= √ ⇒ L = ϕ( − 14 R)ϕ
D−1
which eliminates explicit D-dependence. Again the scalar appears with the wrong-
sign kinetic term, but R appears with the right sign (or vice versa), because of more
complicated redefinitions. The sign of the changes back to the usual for |a| > 1.
However, for |a| = 1, it disappears completely. A similar result occurs for the last
choice, or “lightcone gauge” φ− = 1.
Excercise IXB5.3
The property that distinguishes this kinetic term for a scalar coupled to grav-
ity is the O(1,1) symmetry:
a Before fixing the Weyl scale gauge, the continuous SO(1,1) subgroup of this
symmetry is just the scaling φ± → Λ±1 φ± . After gauge fixing, this trans-
formation may change the gauge, and thus may need to be combined with
a constant Weyl scale transformation to preserve the gauge. In that case
the vierbein will also transform under the resulting modified SO(1,1) trans-
formation. Find the SO(1,1) transformations for ϕ and eam in the above 4
gauges.
b There is also the “parity” transformation of this O(1,1), φ+ ↔ φ− . Find the
modified form of this transformation for ϕ and eam .
Although all these choices are equivalent in perturbation theory (though the
physical scalar may require a nonvanishing vacuum value), they aren’t necessarily
so nonperturbatively, depending on the ranges of the various scalars. Unfortunately,
nonperturbative gravity is not understood well enough (even classically) to make such
distinctions, even though they may be important physically. The above considera-
tions generalize straightforwardly to the case with many physical scalars, where we
 B. GAUGES 515

may consider symmetry groups such as O(n,1). If the physical scalars form a nonlin-
ear σ model, the compensator may join in to make the σ-model groups noncompact:
Examples of this appear in supergravity and strings (see below).
The appearance of a physical scalar can also affect the way scale gauges are chosen
in conjunction with coordinate gauges. For example, a result similar to the one found
at the beginning of this subsection can be obtained from the (linearized) action with
both compensator and physical scalar ψ (where hψi = 0),

L ≈ L0 − 14 ψ ψ

choosing the same χ-dependent coordinate-fixing term (fa )2 , but imposing the scale
gauge
ψ = √12 (χ − ha a )

The result is identical to the one given at the beginning of this subsection, except
that now no scale ghosts appear: The scalar that appears as ha a is now physical, and
no longer needs a ghost to cancel it. This is the perturbative “string gauge” for scale
invariance, which appears automatically in covariantly gauge-fixed string theory.
Excercise IXB5.4
Let’s investigate such gauge choices further:
a Starting with the Fermi-Feynman-gauge-fixed linearized gravity action of sub-
section IXB1, add the physical-scalar kinetic term − 14 ψ ψ. Separate the
traceless and trace pieces of hab . Show that the string-gauge action (i.e, the
one given at the beginning of this subsection if we ignore ghosts) follows from
simply switching
ψ ↔ √1D ha a
√
and then identifiying the new ψ with 2χ/(D − 2).
b The way the physical scalar of string theory appears in the gauge-invariant
and gauge-fixed action is slightly more clever than as described above. (See
subsections XIB5-6 below.) The kinetic term (already in the string gauge for
scale invariance) is Z
S= dx Φ( − 14 R)Φ

where the missing e has been absorbed into Φ by a field redefinition. (Since
Φ is thus not a scalar, we define Φ by e−1/2 e1/2Φ, since e1/2Φ is a scalar.)
Expanding Φ = 1 + χ, the linearized gauge fixing is now simply

L → L + 12 (∂ bhab + ∂a χ)2
516 IX. GENERAL RELATIVITY

(or we can use the nonlinear gauge-fixing function ∂m (Φea m )). Show the result
is the same as above.

REFERENCES
1 T. de Donder, La gravifique einsteinienne (Paris, 1921).
2 J. Scherk and J.H. Schwarz, Gen. Rel. and Grav. 6 (1975) 517;
M. Kaku, Nucl. Phys. B91 (1975) 99;
M. Goroff and J.H. Schwarz, Phys. Lett. 127B (1983) 61:
lightcone gauge for gravity.
3 Cartan, loc. cit. (IXA), ch. X;
M. Spivak, A comprehensive introduction to differential geometry (Publish or Perish
Inc., 1979) v. II, p. 299:
Riemann normal coordinate expansion.
4 U. Müller, C. Schubert, and A.E.M. van de Ven, A closed formula for the Riemann
normal coordinate expansion, gr-qc/9712092.
5 W. Siegel, Phys. Lett. 149B (1984) 162, 151B (1985) 396, 211B (1988) 55:
string gauge.
 C. CURVED SPACES 517

..................... .....................
..................... C. CURVED SPACES .....................

There are some important solutions of general relativity that have no close analog
in Yang-Mills. Here we consider the ones relevant to the only experimental verifica-
tions of this theory: Solutions outside approximately spherical matter distributions
(like the Sun and Earth), and those describing the Universe itself.

1. Self-duality
Plane wave solutions can be constructed for gravity in the same way as for Yang-
Mills (see subsection IIIC3): A little more work (solving the torsion constraint, or
using the result of the free theory) gives

∇+ = ∂ + − 12 xi xj R+i+j (x−)∂ − − xiR+i+j (x−)M −j (∇− = ∂ −, ∇i = ∂ i )

where R+i+j is an arbitrary function of x−, but symmetric in ij, and the empty-space
field equations imply it is also traceless:

R+i+i = 0

If we want to couple Yang-Mills to gravity, then we can still write exact solutions as
long as both waves are parallel; then

R+i+i = 2T ++ = 1
g2
tr(F +iF +i )

where here g 2 refers to the Yang-Mills coupling. (Similarly, we can add in other fields,
such as massless, neutral scalars or particles.)
Excercise IXC1.1
Check that the gravitational plane wave solution satisfies the field equations
and torsion constraint. Show that we can also find more-special solutions of
this form satisfying

1 1
gmn = ηmn + g2
tr(Am An), Rmnpq = g2
tr(FmnFpq )

This has the interpretation that the “graviton” is the bound-state of two
“gluons”. However, it is only a kinematic effect, since the two gluons happen
to be traveling in the same direction at the same speed. (We saw in subsection
VIIIA7 that a similar effect always occurs in D=2, since there only two spatial
directions exist.)
518 IX. GENERAL RELATIVITY

Self-duality for Yang-Mills was discussed in subsection IIIC4. Similar remarks

apply to gravity: We again impose

[∇a, ∇b] = ± 12 abcd [∇c, ∇d ]

Self-duality again implies the field equations, by dualizing the Bianchi identities: For
gravity
R[abc]d = 0 ⇒ 0 = ± 12 abcd Rbcde = 14 abcd bcf g Rf g de = Ra e

While it might appear that the self-duality condition is still second-order because
solving the torsion constraint makes the Lorentz connection the derivative of the
vierbein, the self-duality allows the gauge where the connection is also self-dual, and
this condition effectively becomes a first-order field equation:

Rabcd = Rcdab ⇒ Rabcd = ± 12 cdef Rab ef ⇒ ωabc = ± 12 bcde ωa de

In four dimensions (2 space + 2 time), lightcone methods can again be applied

(see subsection IIIC5): Now
0 0 0 0 0 0
[∇αβ , ∇γδ ] = C αγ 12 Rβ δ  ζ M0 ζ 0 (ωαβ 0 γδ = 0)
0 0
The fact that [∇(⊕α , ∇ )β ] has only an Mα0 β 0 term poses an additional constraint;
the full solution is then
0 0 0 0 0 0 0 0 0
∇⊕α = ∂ ⊕α , ∇ α = ∂ α + (∂ ⊕α ∂ ⊕β φ)∂ ⊕β 0 + 12 (∂ ⊕α ∂ ⊕β ∂ ⊕γ φ)Mβ 0 γ 0
0 0 0 0 0 0 0 0
Rα β γ δ = −i∂ ⊕α ∂ ⊕β ∂ ⊕γ ∂ ⊕δ φ
(In this case, the existence of covariantly constant spinors is a consequence of self-
duality.) The equation of motion that follows from the final condition is now
0 0
φ − i(∂ ⊕α ∂ ⊕β φ)(∂ ⊕α0 ∂ ⊕β 0 φ) = 0

2. De Sitter
The simplest spaces are those where the Ricci scalar is constant, and the other
parts of the curvature vanish:
Rab cd = kδ[a
c d
δb]

These are special solutions of the field equations without matter, but with a cosmo-
logical term, where there are no physical gravitons (the Weyl tensor vanishes), and
thus represent the vacuum. Since there are no physical degrees of freedom, we can
represent this space by just the conformal compensator: i.e. the vierbein (metric)
 C. CURVED SPACES 519

is just the flat one up to a local Weyl scale transformation. We thus have (from
subsection IXA7)
[c
Rab cd → Φδ[a∂b] ∂ d] Φ − δ[a
c d
δb](∂Φ)2 = kδ[a
c d
δb]

where we have written the curvature as a scale transformation of flat space Rab cd =
0, ∇ = ∂: The space is “conformally flat”. Separating this equation into its ir-
reducible parts with respect to the Lorentz group, the Weyl tensor part vanishes
identically, leaving

2Φ Φ − D(∂Φ)2 = Dk, D∂a ∂b Φ = ηab Φ

(The latter equation isn’t implied in D = 2, where the global conformal group is
larger, and more general coordinate choices are possible for this solution. However,
we can still use it consistently.) The latter equation can be solved easily: Looking at
a 6= b, we see that Φ is a sum of functions of one variable. Then looking at a = b tells
us that these functions are quadratic and have the same quadratic coefficient, while
the former equation gives k:

Φ = A + B axa + C 12 xaxa , k = 2AC − B 2

We can choose any A, B a , and C that give the desired value of k: For example, we
can choose the solution Φ = 1 + 14 kx2 (giving the usual flat-space coordinates for
k = 0), or Φ = B a xa (choosing the direction of B a as appropriate to k = −B 2 —
spacelike, lightlike, or timelike).
Excercise IXC2.1
Show that after a Weyl scale transformation the action for gravity including a
cosmological term is (up to a sign) that of a conformal self-interacting scalar
coupled to gravity. Use this to show that the de Sitter space solution (in the
R=0 gauge) yields an “instanton” for this scalar theory, and compare with the
Yang-Mills instanton of subsection IIIC6. Show that similar solutions exist
for massless scalars in arbitrary dimensions with potentials ∼ φn for arbitrary
n (but then k = 0).
The geometry of this space can be understood most easily as that of a D-
dimensional hyperboloid embedded in a flat (D+2)-dimensional space, where we add
one space and one time dimension: Again using the methods of subsections IA6 and
IVA2, we now supplement the constraint

y2 = 0 ⇒ y A = ewA, (w+ , w−, wa ) = (1, 12 xa xa, xa )

520 IX. GENERAL RELATIVITY

with the additional constraint

1 1
nA yA = 1 ⇒ e= =
nA w A −n− − n+ 12 xa xa + na xa

where nA is a (D+2)-vector, yielding the intersection of a cone and plane. In partic-

ular, for n2 6= 0 we can write the metric on the space whose coordinates are all but
n · y:
y A = (|n2 |−1/2, z A ) ⇒ z 2 + n−2 = 0, −ds2 = dz 2

which is the definition of a hyperboloid. Comparing the metric, we find the previous
result:
−ds2 = dy 2 = e2dxa dxa , e = Φ−1 ⇒ k = −n2

This gives the most general coordinate system for de Sitter space as a local scale
of flat space, since conformal transformations are the most general coordinate trans-
formations that will just replace this scale factor with another, and they just rotate
nA . The symmetry group of the D-dimensional subspace that satisfies these two con-
straints is as big as the Poincaré group, namely SO(D,1), ISO(D−1,1), or SO(D−1, 2),
depending on whether n2 <, =, or >0: The former constraint preserves the conformal
group, while the latter kills a timelike, lightlike, or spacelike coordinate.
Excercise IXC2.2
We can also start instead with a D+1-dimensional space, which is a natural
choice for the symmetry group of de Sitter space: Consider the metric and
constraint

−k ds2 = dz 2 = k dz a dz b ηab + dzD+1

2
, 1 = z 2 = k z az b ηab + zD+1
2

Both equations have the same global symmetry group, determined by the sign
of k; k = 0, flat space, can be considered as a limiting case of the others.
a Solve the constraint y 2 = 1 → z(x) as in subsection IVA2 for φ2 = m2 → φ(χ),
and substitute to find the metric in terms of x.
b Find the conformal transformation on xa that relates this coordinate sys-
tem to the more general one above. (Hint: Use z of the D+2-dimensional
construction.)
 C. CURVED SPACES 521

3. Cosmology
To a good approximation the universe can be described by a spacetime which is
(spatially) rotationally invariant (“isotropic”) with respect to a preferred time direc-
tion. Furthermore, it should be (spatially) translationally invariant (“homogeneous”),
so the metric should depend only on that time coordinate. This means that the 3D
subspace at any fixed time should be 3D spherical, flat, or de Sitter space, up to an
overall time-dependent scale factor:

−ds2 = −dτ 2 + φ2(τ )Υ

where Υ is the de Sitter metric for the 3 other dimensions for k = 1, 0, −1 (given,
e.g., by the coordinates in the previous subsection.) By a simple redefinition of the
time coordinate, this can be put in a form which is conformal to a static space:
b 2,
ds2 = φ2(t)ds b 2 = −dt2 + Υ
ds

where by “φ(t)” we really mean “φ(τ (t))”, and the two time coordinates are related
by Z Z
1
dτ = dt φ ⇒ t = dτ or τ = dt φ(τ (t))
φ(τ )
b 2 has curvature
Using previous results for 3D de Sitter space, we find ds

R̂ij kl = kδ[ik δj]l , rest = 0

where k = 1, 0, −1.
To a good approximation the matter in the universe can be approximated as a
“dust”, a collection of noninteracting particles. It should also be rotationally invariant
with respect to the preferred time direction, so the momenta of the particles should
be aligned in that time direction. (Really it is this matter that defines the time
direction, since it generates the curvature of spacetime.) Furthermore, the dust should
be translationally invariant, so the energy-momentum tensor should depend only on
that time coordinate. We then can write (compare excercise IIIB4.1)
ab
TM = ρM (t)uaub , ua = δ0a

where ρM is just the spatial density of particles in the “rest” frame. One way to
derive the φ dependence of ρM that generalizes straightforwardly to other cases is by
using conservation laws: By considering particles all of the same mass in units m = 1,
or by considering J and T for each individual particle (since in this case we neglect
interactions), we have from current conservation

J a = ρM ua ⇒ 0 = ∇aJ a = e∂m e−1J a eam = (φ−4 )∂0(φ4 )(ρM φ−1 )

522 IX. GENERAL RELATIVITY

⇒ ρM = 3aφ−3

for some nonnegative constant 3a, and using (covariant) energy-momentum conser-
vation as a check,

ua ∇aub = ∇0ub = φ−1 ∂0 ub = 0 (geodesic) ⇒ ∇a TM

ab
= ub ∇ a J a + J a ∇ a ub = 0

where we have used (from the result of subsection IXA7 for scaling covariant deriva-
tives)
b 0 = ∂0 ⇒ ∇0 = φ−1 ∂0
∇

(Note, however, that ∇i has Lorentz pieces, and M0i Ji ∼ J0 6= 0 even though Ji = 0.)
For radiation, the momenta of the photons can’t be timelike (they’re lightlike, of
course), but we can still use rotational and translational invariance, together with the
fact that the trace of the energy-momentum tensor vanishes (from scale invariance:
see subsection IXA7). Then

TRab = ρR (t) 13 (4ua ub + η ab )

There is no conserved current, but energy-momentum conservation alone determines

ρR ρR
0 = ∇a TRab = 43 ρM uaub ∇a + ∇b 13 ρR = φ−1 δ0b ( 43 ρM ∂0 − 13 ∂0 ρR )
ρM ρM
−4/3
= φ−1 δ0b ρM ∂0ρM ρR ρR = 32 bφ−4
4/3
⇒

for some nonnegative constant 32 b.

Writing the vierbein as a scale-transformation of the constant curvature space
discussed above (de Sitter in spatial directions, flat in other directions), the gravita-
tional field equations with matter and radiation become (using results from subsection
IXA6 or 7):

b aφ)(∇
6[(∇ b b φ) − 1 ηab (∇φ)
b 2 ] + [(ηab b − ∇
b a∇
b b) + (R̂ab − 1 ηab R̂)]φ2 = 2(TM ab + TRab)φ4
2 2

The only independent components of this equation are the 00-component and trace,
which are, after multiplying by an appropriate power of φ:
. ..
1 2
2φ + 12 kφ2 = aφ + 12 b, φ + kφ = a

For k = 1, these are just energy conservation and the equation of motion for a
harmonic oscillator (centered at φ = a). The 00 equation gave energy conservation
because T00 is the energy density. The trace equation gave the field equation for φ
due to the relation of T aa to δS/δφ given earlier. These equations are easily solved:
 C. CURVED SPACES 523

Imposing the initial condition φ(0) = 0 (i.e., we set the “Big Bang”, when curvatures
.
and energy density were infinite, to be t = 0) and φ(0) > 0 (so φ ≥ 0),
     
1   1 − cos t  √  sin t 
k= 0 : φ = a 12 t2 + b t
     
−1 cosh t − 1 sinh t
R
The “physical” time coordinate is then τ = 0 dt φ. In general φ can’t be expressed
directly in terms of τ , so we use the expressions for both in terms of t. For example,
for k = 1 and b = 0 (just matter), we get a cycloid, which has only such a parametric
expression. Explicit expressions can be found for a = 0 (just radiation): φ(τ ) is then
a circle, parabola, or hyperbola for k=1, 0, −1. Also, for k = 0 and b = 0, φ ∼ τ 2/3
√
(vs. τ for a = 0).
For the case of pure matter (b = 0), the energy conservation equation written in
terms of the τ coordinate becomes, using dτ = φ dt,
 2
dφ a k
1
2 − =−
dτ φ 2
This is the same as the Newtonian equation for the radial motion of a particle under
the influence of a fixed point mass (or the relative motion of 2 point particles).

4. Red shift
The most obvious effect of the cosmological expansion is the cosmological “red
shift”. The expansion of the universe causes photons to lose energy, including those
of the black-body radiation of the universe as well as those emitted long ago from
distant sources. The easiest way to see this is to consider Killing vectors. Since
the cosmological solutions are related to static, isotropic, homogeneous spaces by a
time dependent (but space independent) scale transformation, the symmetries of this
space are just in the spatial directions, and are basically the same as before the scale
transformation. Specifically, the Killing vectors that survive the scale transformation
∇a = Φ∇ b a + (∇b bΦ)Mab satisfy

b (a(ΦKb) ) − 2ηab K · ∇Φ
0 = ∇(aKb) = ∇ b

⇒ ba
Ka = Φ−1 K b · ∇Φ
for K b =0

We then find for conserved momenta K a pa ∼ Φ−1 pa . Since the K’s which survive are
just the spatial ones, we at first find only the spatial components of Φ−1 pa conserved,
but the conservation of the time component follows from pa pa = 0 for photons. Thus,
524 IX. GENERAL RELATIVITY

pa ∼ Φ ∼ φ−1 . Since pa is what an observer measures as the components of momen-

tum (in his “local inertial frame”, a gauge where at his location the metric is flat
and its first derivative vanishes), observers measure the photon’s energy, frequency,
and corresponding black-body radiation (whose distribution depends only on E/T )
as having time dependence ∼ φ−1 (and wavelength as φ). The spectrum of radiation
emitted by a distant object is then shifted by this energy loss, so the amount of shift
determines how long ago it was emitted, and thus the distance of the emitter.
Excercise IXC4.1
Using this result for the φ dependence of the momenta of individual particles,
we can now rederive the φ dependence of ρ’s of the previous subsection directly
from the explicit expressions for J and T of the point particle.
a Rederive J and T in curved space as in subsection IIIB4 and show that

J m δm
0
= (p0 )e(2π)2δ 3(x − X), T mn = J m pn = J npm

b From Killing vectors we just saw that

 0
φ (m 6= 0)
p δa ∼
a 0
φ−1 (m = 0)
where the massive particles are at rest (pa = mδ0a). Combine these results to
find  −3
−3 φ (m 6= 0)
J δa ∼ φ ,
a 0
ρ ∼ T δa δb ∼
ab 0 0
−4
φ (m = 0)
c Find the factors multiplying the ρ’s in T ab for the two cases of dust and
radiation from the explicit expression for T mn. For the massive case (dust)
all particles can be taken at rest, but for the massless case the particles travel
at the speeed of light, so average over particles traveling in the three spatial
directions. (ρ is a continuous function obtained by summing the δ functions
of all the particles. However, for the above results it is sufficient to consider
each individual particle for the massive case, and 3 particles at the same point
going in orthogonal directions for the massless case.)
Astronomers use 3 parameters which are more directly observable:
dφ/dτ φ d2 φ/dτ 2 ρ
H≡ , q≡− , σ≡
φ (dφ/dτ )2 3H 2
where ρ = T 00 = ρM +ρR . Note that in our conventions spatial integrals are weighted
R
as dD−1 x/(2π)D/2; thus the relation of our density to the more standard one is (see,
e.g., excercise IXC4.1a)
4π
Gρusual
ρ = (2π)2ρusual ⇒ σ= 3
H2
 C. CURVED SPACES 525

where G = π in our conventions (but sometimes G = 1 is useful, especially for

solutions describing stars and planets. See excercise IXB1.1.) The “Hubble constant”
H measures the expansion rate, and gives a length (time) scale. The “deceleration
parameter” q is a dimensionless parameter which tells how fast the expansion rate is
slowing down. The “density parameter” σ is another dimensionless parameter which
measures matter density with respect to the amount needed to close the universe. In
the case of pure matter, and with vanishing cosmological constant, σ = q. Then the
“critical” value is q = 12 , for which k = 0: For q > 12 , k = 1, while for q < 12 , k = −1.
In this case we also see that for a given value of H the critical value of the matter
density is ρc = 32 H 2 . If the matter of the universe has this density, we have k = 0,
and spacetime is conformally flat. If it has greater density, we have k > 1, and space
is closed.
Excercise IXC4.2
Solve for σ and q in terms of just a, b, k, and φ (but no time derivatives). In
particular, show

b=0 ⇒ σ = q = (2 − ka φ)−1
a=0 ⇒ 2σ = q = (1 − kb φ2 )−1

These quantities are difficult to measure. Some recent estimates for their present
values are:
H −1 = 14(2) × 109 yrs., .02 < q < 2

Even the gravitational constant is not so easy to measure: Its presently accepted
value is
G = 6.67259(85) × 10−11 m3kg −1 s−2

which is accurate to only a few parts per 10,000, compared to the standard atomic
and nuclear constants, which are known to better than 1 part per million. In “natural
(Planck) units,” c = G/π = h̄ = 1, H −1 = 1.5(2) × 1061 .

5. Schwarzschild
All gravitational experiments outside of cosmology are based on the “Schwarz-
schild solution”, which describes spherical symmetry outside the region with matter.
Assuming also time independence, which is a consequence of spherical symmetry
(Birkhoff’s theorem), we look for a metric of the form

−ds2 = −A−2 (r)dt2 + B −2(r)dr2 + r2 (dθ2 + sin2 θ dφ2 )

526 IX. GENERAL RELATIVITY

(Other coordinate choices are possible, e.g. −A−2(r)dt2 + B −2 (r)[dr2 + r2 (dθ2 +

sin2θ dφ2 )].) The first step is the choice of a vierbein: The simplest choice following
from this metric is
1 1
et = A∂t, er = B∂r , e θ = ∂θ , eφ = ∂φ
r r sin θ
(This can also be used as a starting point in place of the metric.) The next step is to
find the commutators of the e’s, which tells us what ω terms the ∇’s must have to
cancel these cab c ’s (vanishing torsion):

[eθ , eφ] = −r−1 cot θ eφ ∇φ has Mθφ ∇r = B∂r

[er , et] = B(ln A)0et ∇t has Mtr ∇t = A∂t + αMtr
⇒ ⇒
[er , eθ ] = −r−1 Beθ ∇θ has Mrθ ∇θ = r−1 ∂θ + βMrθ
[er , eφ] = −r−1 Beφ ∇φ has Mrφ ∇φ = (r sin θ)−1 ∂φ + γMrφ + δMθφ

where α, β, and γ depend only on r, while δ depends also on θ. (Their explicit forms
are already clear at this point, but we’ll collect the results below.)
We can now determine these Lorentz connections and compute the curvatures by
calculating the ∇ commutators. Since we now use explicit functions for the vierbein
and connections, we use the method described in subsection IXA2 for this situation:
Using the identities

[M12, V2 ] = η22 V1 , [M12, M23 ] = η22M13

[∇1, ∇2] = [e1 + ω1 , e2 + ω2 ]

= {[e1, e2] + (e1ω2 )M2 − (e2ω1 )M1 } + {ω1 [M1, ∇2] − ω2 [M2, ∇1 ] − ω1 ω2 [M1 , M2]}

we then find:

[∇t, ∇θ ] = −αβMtθ ⇒ Rtθtθ = −αβ

[∇t, ∇φ ] = −αγMtφ ⇒ Rtφtφ = −αγ
[∇t, ∇r ] = −B(ln A)0et − Bα0Mtr + α∇t = [α − B(ln A)0]et + (α2 − Bα0)Mtr
⇒ α = B(ln A)0, Rtrtr = α2 − Bα0
B B
[∇r , ∇θ ] = − eθ + Bβ 0Mrθ + β∇θ = (β − )eθ + (β 2 + Bβ 0)Mrθ
r r
B 0
⇒ β= , Rrθrθ = −(β + Bβ )
2
r
B
[∇r , ∇φ ] = − eφ + Bγ 0Mrφ + Bδ 0Mθφ + γ∇φ
r
B
= (γ − )eφ + (γ 2 + Bγ 0)Mrφ + (γδ + Bδ 0)Mθφ
r
 C. CURVED SPACES 527

B
⇒ γ= , Rrφrφ = −(γ 2 + Bγ 0), Rrφθφ = −(γδ + Bδ 0)
r
cot θ 1
[∇θ, ∇φ ] = − eφ + (∂θ δ)Mθφ + δ∇φ + βγMθφ − βδMrφ
r r
cot θ 1
= (δ − )eφ + (γ − β)δMrφ + (δ 2 + βγ + ∂θ δ)Mθφ
r r
cot θ 1
⇒ δ= , Rθφθφ = −(δ 2 + βγ + ∂θ δ), Rrφθφ = 0
r r
Collecting the results:

∇t = A∂t + B(ln A)0Mtr Rtrtr = BA[B(A−1)0 ]0

B2
∇r = B∂r Rtθtθ = Rtφtφ = − (ln A)0
r
1 B BB 0
∇θ = ∂θ + Mrθ Rrθrθ = Rrφrφ = −
r r r
1 cot θ B 1 − B2
∇φ = ∂φ + Mθφ + Mrφ Rθφθφ =
r sin θ r r r2
Excercise IXC5.1
Find the covariant derivative for the 2-sphere in spherical coordinates

ds2 = dθ2 + sin2 θ dφ2

in terms of the single SO(2) generator Mab = ab M by the above methods
(and not that of excercise IXA4.5). Calculate the curvature. Find the three
Killing vectors. (Hint: What is the symmetry of the sphere?)
Having all the curvatures, we can now calculate the Ricci tensor, which appears
in the field equations. The nonvanishing components are:

Rtt = Rtrtr + 2Rtθtθ , Rrr = −Rtrtr + 2Rrθrθ , Rθθ = Rφφ = −Rtθtθ + Rrθrθ + Rθφθφ

Vanishing of Rab − 12 ηab R is equivalent to vanishing of Rab . In terms of these curvatures,

we see it also implies

−Rtrtr = 2Rtθtθ = −2Rrθrθ = Rθφθφ

These are easy to solve: First,

Rtθtθ = −Rrθrθ ⇒ (ln A)0 = −(ln B)0 ⇒ A = B −1

where we have fixed the proportionality constant by requiring A, B → 1 as r → ∞

(redefining t by a constant scale transformation). Also,
1 k
−2Rrθrθ = Rθφθφ ⇒ (1 − B 2 )0 = − (1 − B 2) ⇒ 1 − B2 =
r r
528 IX. GENERAL RELATIVITY

r
k
⇒ B= 1−
r
for some constant k. The last field equation is then redundant. (As usual, the field
equations are related by the Bianchi identity.) The constant k can be related to the
nonrelativistic result by comparing at large distances. (See excercise IXB1.1.) We
then find k = 2GM, so the final result is:
   −1
2GM 2GM
−ds = − 1 −
2
dt + 1 −
2
dr2 + r2 (dθ2 + sin2θ dφ2 )
r r

More generally, if we have some spherically symmetric, static matter distribu-

tion, then the only nonvanishing components of the energy-momentum tensor will be
Ttt, Trr , and Tθθ = Tφφ (representing energy denisty, radial pressure, and isotropic
pressure), all functions of just r. Repeating the above procedure, we integrate
r
[r(1 − B 2)]0 = 2r2 Ttt, [ln(AB)]0 = − 2 (Ttt + Trr )
B
while the remaining equation is redundant.
For example, for a spherically symmetric, static electromagnetic field the only
nonvanishing components of the field strength are Ftr and Fθφ , corresponding to
electric and magnetic charges, respectively. Then the invariance of T under a duality
transformation (see subsections IIA7, IIIA4) implies

Tθθ = Tφφ ⇒ Ttt = −Trr ⇒ A = B −1

again, since on this Fab duality effectively replaces (θ, φ) ↔ (it, r), with the i from
Wick rotation. Local scale invariance (see subsection IXA7) then tells us

Ta a = 0 ⇒ Ttt = −Trr = Tθθ = Tφφ

Excercise IXC5.2
Let’s rederive these results by brute force:
a Derive Tab for a general electromagnetic field by varying its action with respect
to eam or g mn .
b Find each of T ’s components explicitly in terms of Ftr and Fθφ in the case
where those are the only nonvanishing components, and show they appear
only in the combination Ftr2 + Fθφ
2
.
As usual, these field strengths can be found easily from the integral form of Gauss’
law by integrating over a sphere: For example, for the magnetic field
Z
magnetic charge ∼ 2 dxm dxn Fmn = 4πr2 Fθφ
1
 C. CURVED SPACES 529

for the Fθφ component of Fab (integrating over θ and φ), since the metric (and vierbein)
for θ and φ is the same as for flat space. By duality, the solution for Ftr in terms of
the electric charge is the same. The result is
Q
Ttt = , Q = π 2(e2 + g 2)
r4
for electric charge e and magnetic charge g. The 1/r4 dependence also follows from
scale invariance, since the charges are dimensionless (and the matter field equations
decouple from A and B). (Again, since the solution does not extend to r = 0, we
normalize by comparing Fab or Tab at r = ∞ to the flat-space solution.) The net
effect on the Schwarzschild metric is
2GM 2GM 2Q
1− →1− + 2
r r r
Our solution relates to the usual mechanics normalization of the charges (see subsec-
tion VIIA3), restoring G, as

2Q = G2π(e2 + g 2 ) = G(e2m + gm
2
)

Excercise IXC5.3
Let’s also apply brute force to solving Maxwell’s equations ∇aF ab = ∇[aFbc] =
0 (outside the matter).
a As a warm-up, using directly the above covariant derivatives, show that in
flat space
V a = δra Vr ⇒ ∇aJ a = r−2 ∂r r2 Jr

Note that the covariant derivative of a vanishing component doesn’t necessar-

ily vanish (just as the ordinary derivative of a function that vanishes at some
point doesn’t necessarily vanish at that point): Components of ∇ other than
∇r contain Lorentz generators that rotate other components of V to Vr .
b Solve Maxwell’s equations in differential form for Fab in the above case. Use
the empty-space solution to define the normalization at infinity. (Actually,
in this case, the charge is well-defined in terms of the flux of the fields, as
described above, but gives the same result here because the space is asymp-
totically flat.)
Excercise IXC5.4
Consider the plane wave in the coordinates
 − −

−ds2 = −2dx+ dx− + L2 (x−) e2β(x ) dy 2 + e−2β(x ) dz 2
530 IX. GENERAL RELATIVITY

Calculate the covariant derivatives and curvature tensor by the method de-
scribed above (double-counting and subtracting). Show that the field equa-
tions reduce to
L00 + (β 02 )L = 0

Excercise IXC5.5
Use the method applied to the Schwarzschild metric to calculate the covariant
derivative and curvature tensor for the metric

−ds2 = −dt2 + 2ex dt dy − 12 e2xdy 2 + dx2 + dz 2

Show that this metric satisfies the field equations with a cosmological term
for a dust at rest with respect to this time coordinate; i.e.

Rmn − 12 g mn (R − 4Λ) = ρδ0m δ0n

where Λ and ρ are both constants.

Excercise IXC5.6
Use this method to calculate the covariant derivative and curvature tensor for
the cylindrically symmetric metric

−ds2 = −A−2(r)dt2 + B −2 (r)dr2 + r2 dθ2 + dz 2

Assume the matter in this problem is a “perfect fluid”,

T ab = ρua ub + P (η ab + ua ub ) (u2 = −1)

Solve the equations of motion for the gravitational field to find A and B, as
well as the pressure P and particle density ρ. What is the implied relation
between P and ρ?
Excercise IXC5.7
Use this method to calculate the covariant derivative and curvature tensor for
the following metric, corresponding to that outside a planar mass distribution:

−ds2 = −A−2(z)dt2 + B −2(z)(dx2 + dy 2 ) + dz 2

Solve Einstein’s equations in empty space to find A and B (up to some con-
stants of integration).
 C. CURVED SPACES 531

6. Experiments
All experiments (excluding cosmology) are based on the Schwarzschild metric.
The first type of experiment involves gravitational redshift, but unlike the cosmolog-
ical case, the relevant reference frames of observation are not local inertial frames
but the static reference frame in which the Schwarzschild metric is defined. (There
are also measurements of redshift from airplanes, whose reference frame is defined
with respect to the Schwarzschild one.) In this reference frame the relevant Killing
vector is the one which expresses the fact that the space is static, K m ∂m = ∂/∂t. The
momentum which is measured by the observer is pa , not pm or pm , since the observer
still uses a reference frame for which the metric at his position is flat (but not its first
derivative, since he is not in free fall). (In fact, this is one of the purposes for using
a vierbein, as a frame of reference.) The conserved quantity is then
r
2GM b
E = −Ka p = 1 −
a
E
r

where the energy of a particle Eb is the time component of pa as measured in this

frame. Thus, conservation of E for a photon gives the r-dependence of the observed
energy E b (and thus the frequency, which in turn determines the wavelength, since
p2 = 0).
To compare with nonrelativistic mechanics, we instead evaluate E for a massive
particle in the Newtonian limit:
 
GM GMm
E ≈ 1− (m + K) ≈ m + K −
r r

b (rest mass m +
giving the “conserved energy” E in terms of the “particle energy” E
kinetic K), including the potential energy.
The other type of experiment involves properties of geodesics, so we need to
solve the geodesic equations of motion. Without loss of generality, we can choose the
angular coordinates such that the initial position and direction of the particle is in the
equatorial plane θ = π/2, where it remains because of the symmetry θ ↔ π − θ, as in
the nonrelativistic case. Also as in the nonrelativistic case, we can find constants of
the motion corresponding to the energy E and (z-component of) angular momentum
L by using the Killing vectors K m ∂m = ∂/∂t and ∂/∂φ to find the conserved quantities
.
K a pa = K m gmn xn (in the parametrization v = 1):
 
.m 2GM . . .
E ≡ −gtm x = 1 − t, L ≡ gφm xm = r2 φ
r
532 IX. GENERAL RELATIVITY

In the case where the particles come from infinity, these are the initial kinetic energy
and angular momentum. We have chosen an affine parametrization, which requires
   
.m .n 2GM .2 2GM −1 .2 .
−m = gmn x x = − 1 −
2
t + 1− r + r2 φ2
r r
. .
Solving the previous equations for t and φ, this reduces to the radial equation
  2 
.2 2GM L
0 = −E + r + 1 −
2
+m 2
r r2
 
1 .2 GMm2 L2 GML2
⇒ 2r + − + 2− = 12 (E 2 − m2 )
r 2r r3
This looks like a typical nonrelativistic Hamiltonian for “energy” 12 (E 2 − m2)
with the same terms as in the Newtonian case but with an extra r−3 term. (To
take the nonrelativistic limit for the massive case, first scale the affine parameter
τ → s/m.) Since there are good coordinate systems for a “black hole” using r as a
coordinate (e.g., see the following subsection: r and r00 + t00, as seen from the figure
for Kruskal-Szkeres), this equation can even be used to descibe a fall into a black
hole. (For example, for L = 0 we get the same cycloid solution as in cosmology and
in Newtonian gravity, reaching the singularity at r = 0 in finite proper time.)
Because of the r−3 term in the potential, noncircular orbits are no longer closed.
In particular, let’s consider orbits which are close to circular. Circular orbits are
found by minimizing the potential for the r-equation:
dV GMm2 L2 3GML2
0= = − 3 +
dr r2 r r4
d2 V 2GMm2 3L2 12GML2
0< 2 =− + 4 −
dr r3 r r5
The near-circular orbits are described by small (harmonic) oscillations about this
minimum, with angular frequency given by
d2 V GMm2 (r − 6GM)
ωr2 = =
dr2 r3 (r − 3GM)
from solving for L2 = GMm2 r2 /(r − 3GM). On the other hand, the frequency of the
.
circular orbit itself in terms of its angular dependence is just φ = L/r2 , giving
GMm2
ωφ2 =
r2 (r − 3GM)
This means that the perihelion (closest approach to the Sun) of an orbit, which occurs
every period 2π/ωr of the radial motion, results in the change of angle
Z 2π/ωr  −1/2
dφ 2π GM
2π + δφ = dτ = ωφ = 2π 1 − 6
0 dτ ωr r
 C. CURVED SPACES 533

GM
⇒ δφ ≈ 6π
r
in the weak-field approximation. This effect contributes to the measurement of the
precession of the perihelion of the (elliptical) orbit of Mercury, but so do the precession
of Earth’s axis, the oblateness of the Sun, and gravitational interaction with other
planets. As a result, this relativistic effect contributes less than 1% to the observed
precession. In particular, the solar oblateness is difficult to measure.
The effects on geodesics of photons are much easier to measure, since there are
no Newtonian effects. As a result, the weak field approximation is sufficient. We first
consider bending of light by the Sun: A photon comes in from infinity and goes back
out to infinity (actually to the Earth, which we assume is much farther from the Sun
than the photon’s closest approach to it), and we measure what angle its trajectory
was bent by. (For example, we look at the apprarent change of position in stars when
the Sun passes in their direction during an eclipse.) Starting with the exact solution
. .
for a photon’s geodesic (case m2 = 0 above), we use the equations for r and φ to find
r
dr E2 4
= r − r2 + 2GMr
dφ L2
Changing variables,
b L GM du
u≡ , b≡ , a≡ ⇒ dφ = √
r E b 1 − u2 + 2au3
The impact parameter b ≡ L/E would be the closest approach to the Sun neglecting
gravitational effects (L = rp = bE). We now make the weak field approximation: For
a small,
 
du u3
dφ ≈ √ 1−a
1− u2 1 − u2
3

sin χ
= dχ 1 − a 2 (u ≡ sin χ)
cos χ
  
1
= d χ − a cos χ +
cos χ
Defining φ = 0 at r = ∞, the integral is
(1 − cos χ)2 (1 − cos φ)2
φ ≈ χ−a ⇒ χ≈φ+a
cos χ cos φ
⇒ u = sin χ ≈ sin φ + a(1 − cos φ)2
The change in φ from incoming photon to outgoing photon follows from the two
solutions for r = ∞:
u = 0 ⇒ φ = 0, π + 4a
534 IX. GENERAL RELATIVITY

Therefore the deviation of φ from a straight line is 4GME/L. (Mathematical note: All
variable changes were those suggested by the flat space case a = 0: E.g., b/r = sin χ,
where χ is what φ would be in flat space.)
A similar experiment involves measuring the round-trip travel time for radio waves
from Earth to some reflector (on another planet or an artificial solar satellite), with
and without the Sun near the path of the waves. Now, instead of dr/dφ we want, in
units b = 1  r
dr 2a 1 2a
= 1− 1− 2 + 3
dt r r r
rdr dr
⇒ dt ≈ √ 2 + 2a √ 2
hr√ − 1 r −1 i
= d r2 − 1 + 2a cosh−1 r
h√ i
−1 r
= d r − b + 2GM cosh
2 2
b
.
putting the b’s back. (We have neglected the gravitational effect on r, which is
.
negligible compared to that on t, since for most of the path r  b.) We then integrate
from r = rmin ≈ b to r = rEarth , add the integral from r = rmin to r = rref lector ,
multiply by 2 for the round trip, and throw in a factor to convert to the proper time
s of the observer (which turns out to have a negligible effect to this order in a). This
result is then compared to the same measurement when both observer and reflector
have revolved further about the Sun, so b changes significantly (but not rEarth nor
rref lector ). The biggest contribution (as seen from ∆t/∆b), for b  rEarth and rref lector ,
comes from the cosh term: For x  1, cosh−1 x ≈ ln(2x), so

∆s ≈ −8GM∆(ln b)

7. Black holes
For physical massive bodies the Schwarzschild solution applies only outside the
body, where Tab = 0. The form of the solution inside the body depends on the
distribution of matter, which is determined by its dynamics. Generally the surface of
the body is at r  GM, but we can try to find a solution corresponding to a point
mass by extending the coordinates as far as possible, till the curvature components
Rab cd blow up. The Schwarzschild metric is singular at r = 2GM. In fact, r and
t switch their roles as space and time coordinates there. There is no corresponding
singularity there in the curvatures, which are ∼ r−3 . This unphysical singularity can
be eliminated by first making the coordinate transformation, for r > 2GM,
Z  −1  r 
0 2GM
r = dr 1 − = r + 2GM ln −1
r 2GM
 C. CURVED SPACES 535

and then making a second coordinate transformation by rescaling the “lightcone”

coordinates as
r
00 00 (r 0 ±t)/4GM r
r ± t = 4GMe = 4GM − 1 e(r±t)/4GM
2GM
The result is the “Kruskal-Szekeres coordinates”
2GM −r/2GM
−ds2 = e (−dt002 + dr002 ) + r2 (dθ2 + sin2 θ dφ2 )
r
where r(r00 , t00) is defined by
 r 
002 002
r −t = (4GM) 2
− 1 er/2GM
2GM
This can now be extended past r = 2GM down to the physical singularity at r = 0.
The complete space now looks like (plotting just r00 and t00):

t"

M
2G
r=

r=0

r"

r=0
r=
2G
M

In this diagram lines at 45◦ to the axes represent radial lightlike geodesics. Since
nothing travels faster than light, this indicates the allowed paths of physical objects.
Curves of fixed r are hyperbolas: In particular, the physical singularity is the curve
t002 − r002 = (4GM)2 (r = 0), while t002 − r002 = 0 (r = 2GM) is the “event hori-
zon” which allows things to go only one way (out from the bottom half or into the
top half), and r = ∞ is both r00 = ±∞. Nothing can communicate between the 2
“outside worlds” of the left and right 90◦ wedges. In particular, a star which col-
lapses (“gravitational collapse”) inside its “gravitational radius” 2GM is crushed to
a singularity, and the spherically symmetric approximation to this collapse must be
536 IX. GENERAL RELATIVITY

represented by part of the Kruskal-Szekeres solution (outside the star) by Birkhoff’s

theorem, patched to another solution inside the star representing the contribution
of the matter (energy) there to the field equations. This means using just the top
and right 90◦ wedges, with parts near the left edge of this modified appropriately.
The top wedge is called a “black hole”. (If a situation should exist described by just
the bottom and right wedges, the bottom wedge would be called a “white hole”.)
Similarly, stable stars are described by just the right wedge, patched to some interior
solution. This right wedge represents the original Schwarzschild solution in the region
r > 2GM where its coordinates are nonsingular. In that region lines of constant t
are just “straight” radial lines in the Kruskal-Szekeres coordinate system (r00 ∼ t00).
Besides the fact that nothing can get out, another interesting feature of the black
hole is that an outside observer never sees something falling in actually reach the event
horizon: Consider an observer at fixed r > 2GM using Schwarzschild coordinates,
so his proper time s ∼ t. Then light radiating radially from an in-falling object is
received later and later, up till t = ∞, by the observer as the object approaches the
event horizon, although it takes the object a finite amount of proper time to reach
the event horizon and the physical singularity.
There are also more complicated black-hole solutions with spin and electric charge.
Another interesting effect of the event horizon is the eventual decay of the black
hole (“Hawking radiation”): Pair creation can result in a similar way to that in
an electrostatic potential of sufficient strength (see excercise IIIB5.1). Particles are
emitted near the event horizon (the edge of the gravitational barrier), carrying energy
off to infinity, while their antiparticles fall into the singularity.
There are two features of the black hole that are less than desirable: the existence
of singularities indicates a breakdown in the field equations, and the existence of event
horizons results in an “information loss”. Both these properties might be avoidable
quantum mechanically: For example, quantum effects can generate curvature-squared
terms in the effective action, which modify the short-distance behavior of the theory.
One might think that such short-distance effects would have an effect only at short
distances away from regions of high curvature such as the singularity, and thus remove
the singularities but not the event horizons. However, it is possible (and examples of
such solutions have been given) that the prevention of the creation of the singularity
in stellar collapse would eventually result in a reversal of the collapse (“gravitational
bounce”): The would-be black hole solution is patched to a would-have-been white
hole by short-distance modifications, resulting in an exploding star that initially re-
sembled a black hole but has no true event horizon.
 C. CURVED SPACES 537

REFERENCES
1 J. Ehlers and W. Kundt, Exact solutions of gravitational field equations, in Gravitation:
An introduction to current research, ed. L. Witten (Wiley, 1962) p. 49:
gravity plane waves.
2 J.F. Plebański, J. Math. Phys. 16 (1975) 2395:
reduction of self-dual metric to single component.
3 Siegel, loc. cit. (IVC, 2nd ref. 17):
lightcone gauge for self-dual (super)gravity.
4 W. de Sitter, Proc. Kon. Ned. Akad. Wet. 19 (1917) 1217, 20 (1917) 229.
5 A. Friedmann, Z. Phys. 10 (1922) 377;
H.P. Robertson, Astrophys. J. 82 (1935) 284, 83 (1936) 187, 257;
A.G. Walker, Proc. Lond. Math. Soc. 42 (1936) 90:
cosmology in general relativity.
6 Particle Data Group, loc. cit. (IC):
astrophysical “constants”.
7 K. Schwarzschild, Sitz. Preuss. Akad. Wiss. Berlin, Math.-phys. Kl. (1916) 189.
8 M.D. Kruskal, Phys. Rev. 119 (1960) 1743;
G. Szekeres, Publ. Mat. Debrecen 7 (1960) 285.
9 S.W. Hawking, Comm. Math. Phys. 43 (1975) 199.
10 V.P. Frolov and G.A. Vilkoviskii, Phys. Lett. 106B (1981) 307:
possibility that quantum corrections to gravity eliminate black holes.
538 X. SUPERGRAVITY

X. SUPERGRAVITY
In the previous chapter we studied the symmetry principles behind general rel-
ativity; now we add supersymmetry to the picture. Supergravity is a fundamental
part of many of the applications of supersymmetry.
......................... .........................
......................... A. SUPERSPACE .........................

We first need to understand the “geometry” associated with local supersymmetry.

1. Covariant derivatives
In subsection IVC3 we discussed superspace covariant derivatives for super Yang-
Mills. Similar methods can be applied to supergravity, the theory of the graviton (spin
2) and gravitino (spin 3/2). In that case we want to gauge the complete (unbroken)
global symmetry of the theory: Besides the obvious Poincaré and supersymmetry,
there is also the axial U(1) (“R”) symmetry that transforms the spin-3/2 field. (The
best we might have expected is superconformal symmetry, which also has conformal
boosts and scale, but which are broken by the vacuum just as in ordinary gravity,
and S-supersymmetry, which is also broken because it’s the square root of conformal
boosts.) It is introduced in the same way as local Lorentz invariance in ordinary
gravity, and acts on flat spinor indices (but cancels on vector indices). We therefore
want to gauge the translations ∂M (which have been generalized naturally to super-
space from ∂m appearing in ordinary gravity to include supersymmetry), the Lorentz
generators Mαβ , M . . of ordinary gravity, and the (second-quantized) hermitian U(1)
αβ
generator Y , defined to act on the covariant derivatives as

[Y, ∇α] = − 12 ∇α, [Y, ∇α.] = 12 ∇α., [Y, ∇a] = 0

We now use the “ ” to refer to hermitian conjugation without reordering, i.e., keep-
ing the partial derivatives and other generators on the right. (As in ordinary gravity,
transformations are not truly unitary, and covariant derivatives truly hermitian, be-
cause of ordering.)
Then the gauge parameter, covariant derivative, and field strengths are expanded
over these generators, as in ordinary gravity:
..
K = K M ∂M + 12 K αβ Mβα + 12 K αβ M . . + iK−1 Y
βα
..
∇A = EAM ∂M + 12 ΩAβγ Mγβ + 12 ΩAβ γ M . . + iAAY
γβ
 A. SUPERSPACE 539

..
[∇A, ∇B } = TAB C ∇C + 12 RAB γδ Mδγ + 12 RAB γ δ M . . + iFAB Y
δγ
M
(EA is known as the “supervierbein” or “vielbein”.) Alternatively, we can write the
M and Y terms collectively as 12 K AB MBA (and similarly for the covariant derivative
and field strengths), where MAB are the generators of OSp(3,1|4), by algebraically
constraining K AB to contain just the appropriate pieces (and relating K ab to K αβ in
the usual way). Also, the shorthand K I MI now includes Lorentz and U(1) terms.
Excercise XA1.1
Use the definition in the above commutation relations to express the torsion
TAB C directly in terms of the structure functions CAB C , Lorentz connection
ΩA βγ , and U(1) connection AA.
The constraints in supergravity are a combination of the kinds used in ordinary
gravity and super Yang-Mills: those that (1) define the vector derivative in terms of
the spinor ones
−i∇ . = {∇α, ∇ .}
αβ β

(2) define the spinor (Lorentz and R) connections

.
Tαβ γ = T .β γ) = Tαbb = 0
α,β(β

and (3) allow the existence of chiral (scalar) superfields

∇αΦ̄ = 0 ⇒ {∇α, ∇β }Φ̄ = 0 (Y Φ = 0)

(The first two constraints imply the generalization of this chirality condition to Y 6= 0
and chiral superfields with undotted indices, like the Yang-Mills field strength.)
Excercise XA1.2
Rewrite the first and last set of constraints directly in terms of field strengths.
The explicit solution of all these constraints is a bit messy, but we will need
only a certain subset of them to find the prepotentials and supergravity action. The
form of the solution is a generalization of super Yang-Mills in a way similar to how
general relativity generalizes ordinary Yang-Mills. In particular, just as the vierbein
ea = eam ∂m is a generalization of the Yang-Mills vector Aa to describe gauging of the
translations, the generalization of the super Yang-Mills prepotential V to supergravity
is H = H m (−i)∂m, which appears in an exponential eH just as V appears as eV : The
chirality-preserving constraints, expressed explicitly in terms of the vielbein, is

{Eα , Eβ } = Cαβ γ Eγ
540 X. SUPERGRAVITY

If the commutator vanished like the Yang-Mills case, Eα would be partial derivatives
on some complex two-dimensional subspace, the usual ∂α up to some complex (su-
per)coordinate transformation, as for dα in flat superspace. However, the fact that
their algebra still closes means they still generate translations within such a subspace,
and are thus linear combinations of such partial derivatives:

Eα = ψNαµ Êµ , ʵ = e−Ω ∂µ eΩ, Ω = ΩM (−i)∂M

where we have separated the matrix coefficient into a local complex scale (scale ⊕
U(1)) ψ and a local Lorentz transformation Nα µ . For most purposes we will find it
convenient to fix all these invariances by choosing the gauge

ψNα µ = δαµ ⇒ Eα = Êα

As for Yang-Mills, solution of the chirality condition introduces a new, chiral

gauge invariance:
0
eΩ = eiΛ̄eΩe−iK ; Λ = ΛM (−i)∂M , K = K M (−i)∂M

[∂¯µ., Λ] ∼ ∂¯ν. ⇒ ∂¯µ.Λm = ∂¯µ.Λµ = 0

.
where Λµ is not chiral, since it generates terms in the transformation law of Eµ. that
. .
can be canceled by including ∂¯µ.Λν terms in the transformation law of ψ̄N̄α.µ . This
.
means we can use Λµ and K µ to gauge
.
Ωµ = Ωµ = 0 ⇒ Ω = Ωm (−i)∂m ⇒ ʵ = ∂µ + ʵ m ∂m

where ʵ m = −i∂µ Ωm +... from expanding the exponentials as multiple commutators.
We can again transform to a chiral representation, and work in terms of
0
eU = eΩeΩ̄ , eU = eiΛ̄eU e−iΛ ; ʵ. = ∂µ., ʵ = e−U ∂µ eU
.
where U now generalizes the constant hUi = θµ θ̄µ (−i)∂µµ. used in flat superspace. Also
as for Yang-Mills, the usual local component transformations (now for coordinate,
supersymmetry, scale, U(1), and S-supersymmetry) reappear in the chiral parameters
Λm and Λµ .
For a component analysis, we look at the linearized transformation (see excercise
IVC4.3)
δU m ≈ i(Λ̄ − Λ)m − i 12 [hUi, Λ̄ + Λ]m
. . . . . .
= i(Λ̄ − Λ)m − 12 θν θ̄ν ∂ν ν.(Λ̄ + Λ)m + (θµ Λ̄µ − θ̄µ Λµ ) + i 12 (θ̄2 θν ∂ν µ Λµ − θ2 θ̄ν ∂ µν.Λ̄µ )
 A. SUPERSPACE 541

where for Λ we use as independent just the chiral parameters Λm and Λµ ; the nonchiral
.
Λµ , having already been used to gauge away U µ , is now fixed in terms of the others
as
. .
Λµ = e−U Λ̄µ eU

to maintain U µ = 0. The first term in the transformation tells us that the surviving
component fields are the same as for super Yang-Mills, with “m” as the group index:

U m = ea m (θθ̄), ψαm (θ̄2θ), ψ̄α.m (θ2 θ̄), Am (θ2θ̄2 )

The second term in the transformation law gives δeam ≈ −∂a λm from Λ̄m | = Λm | =
λm . Then Λµ contains the rest of the gauge parameters:

Λµ = µ , a + ib(θ), λν µ (θ), ζ µ (θ2 )

= supersymmetry, scale +iU(1), Lorentz, S-supersymmetry

The third term in the transformation law then shows scale and Lorentz gauge way
pieces of the vierbein, as usual, while S-supersymmetry gauges away the trace of ψαm .
.
It also forces Λm to include ¯µ at order θ to maintain the gauge; we then see that
ψαm is the gauge field for supersymmetry, with contributions from the second and
fourth terms. Finally, the fourth term also shows that Am is the gauge field for U(1).
The resulting component content is that of “conformal supergravity”, which will be
transformed later to ordinary supergravity through a compensator superfield.
For perturbation theory, or comparison with global supersymmetry, we should
expand about “flat” superspace (which is nontrivial because of nonvanishing torsion
T .c in empty superspace). We then modify the chiral representation:
αβ

eΩ eΩ̄ = eU = ehU i/2eH ehU i/2 ⇒ ʵ. = dµ., ʵ = e−H dµ eH

We now expand all derivatives over the covariant derivatives dM of global supersym-
metry (constructed from hUi as before)

H = H M (−i)dM , EA = EA M dM

instead of over partial derivatives ∂M , which is just a change of basis. This also
modifies the description of the Λ gauge parameters:
0
eH = eiΛ̄ eU e−iΛ; Λ = ΛM (−i)dM , K = K M (−i)dM
.
[d̄µ., Λ] ∼ d̄ν. ⇒ Λµ dµ + Λm ∂m = 12 {d̄ν , [d̄ν., Lµ dµ ]}
. .
⇒ Λµ = d¯2 Lµ , Λµµ = id̄µ Lµ
542 X. SUPERGRAVITY

in terms of a new parameter Lµ . From this we find the linearized transformation law
. . .
δH µµ ≈ dµ L̄µ − d̄µ Lµ

Excercise XA1.3
Expand this transformation law in components, and compare with the previ-
ous analysis.
Besides chirality, we also need a certain combination of the other constraints:
.
0 = Tαb b − T .β = (−1)B CαB B + Cαβ β − iAα
αβ

. .
where the A term comes from the contribution 12 iAαδ γ. to T .γ . (See excercise XA1.1.)
β αβ
Using the gauge Eα = Êα without loss of generality, we then find (comparing similar
manipulations in subsection IXA2)
←
−iAα = E∂M E −1 Eα M = E −1 E α E
← ←
where the backwards arrow on E α = Eα M ∂ M means all derivatives act on everything
to the left (see subsection IA2), and

E ≡ sdet EA M

(The superdeterminant was defined in subsection IIC3.)

We now need the general identity, for any function A and first-order differential
operator B,
← ← ← ← ← ← ← ←

AeB = (1 · eB e−B )AeB = (1 · eB )(e−B AeB ) = (1 · eB )(eB Ae−B )

←

= (1 · eB )(eB A)
← ← ← ←

⇒ 1 = (1 · e−B )eB = (1 · eB )[eB (1 · e−B )]

The final result in the gauge ψ = 1 (Nα µ is trivially restored) is then

←

iAα = EαT, eT = E(1 · e−Ω )

This can be used to solve chirality conditions on matter fields: In this gauge, we
have
Y Φ = yΦ ⇒ 0 = ∇α.Φ = (Ēα. + iy Āα.)Φ

⇒ Φ = eyT̄ eΩ̄φ, ∂¯µ.φ = 0

 A. SUPERSPACE 543

Again as for Yang-Mills, the chiral-representation field φ transforms under only the
Λ transformations:
←

φ0 = (1 · ei Λ )y eiΛφ, Λ = −i(Λm∂m + Λµ ∂µ )

Thus, scalars in the real representation become densities in the chiral representation
(except for y = 0). In particular, we have for the special case
←
Z
0
y = 1 ⇒ φ = φe iΛ
⇒ δ dx d2 θ φ = 0

which will prove useful later for chiral integration. For now, we note that such a
chiral scalar, with y 6= 0, can be seen to compensate from the ∂µ Λµ term: This term
allows U m to eat the complex “physical” scalar and spinor, fixing scale, U(1), and
S-supersymmetry, while the complex auxiliary scalar survives, along with coordinate,
Lorentz, and supersymmetry invariance. We also note that for perturbation about
flat superspace we have
←
i(1 · Λ) = ∂m Λm − dµ Λµ = −d̄2dµ Lµ

Excercise XA1.4
Show that preservation of the chirality of φ implies the previous chirality con-
ditions on Λ. Thus, as for nonsupersymmetric or nongravitational theories,
the gauge group follows more simply from starting with matter representa-
tions.
.
We also note that in the gauge ψ = 1, and also Ωµ = Ωµ = 0, the superdetermi-
nant is simply (see subsection IIC3)

E −1 = det(Em a )

where Em a is a component of EM A (not the inverse of Ea m ).

2. Field strengths
These constraints can be completely solved for all the field strengths. Alterna-
tively, we can impose them, together with the Bianchi identities (Jacobi identities
of the covariant derivatives), to find a smaller set of algebraically independent field
strengths, and the differential equations that relate them. The method is analogous
to the case of super Yang-Mills treated in subsection IVC3. We begin with the con-
straints analogous to the Yang-Mills ones:

{∇α, ∇ .} = −i∇ ., {∇α, ∇β } = Rαβ I MI

β αβ
544 X. SUPERGRAVITY

(where the latter will simplify from later results). From just the latter, we find

[∇(α, {∇β , ∇γ)}] = 0 ⇒ R(αβγ)δ − i 12 δ(αδ Fβγ) = ∇(αRβγ) I = 0

Using both constraints, we also have

[∇(α, {∇β), ∇γ.}] + [∇γ., {∇α, ∇β }] = 0

.
⇒ [∇(α, ∇β)γ ] = iRαβ γ ∇ . − 12 Fαβ ∇γ. − i(∇γ.Rαβ I )MI
. .δ
δ
.
⇒ [∇α, ∇β γ.] = −iCαβ W γ. + 12 iRαβ γ.δ ∇ . − 14 Fαβ ∇γ. − 12 i(∇γ.Rαβ I )MI
δ

for some operator W γ. = W γ.A∇A + W γ.I MI .

So far the excercise has been analogous
to the super Yang-Mills case (where the extra “i” in the definition of W is due to
our use of antihermitian generators, except for Y ). Now we impose the remaining
constraints, which can be combined conveniently as
. . .
0 = Tα,β γ.β δ = −iCαβ W γ.β δ ⇒ W α. = W α.β ∇β + W α.β ∇ . + W α.I MI
β

Following again the steps for Yang-Mills, we analyze the next-higher-dimension

Jacobis, beginning with

0 = {∇(α, [∇β), ∇ .]} + [∇ ., {∇α, ∇β }] = iCγ(α{∇β), W .} + ∆ .

γδ γδ δ γαβ δ

.
∆ . = 12 i(∇(αR . )∇. − 14 (∇(αFβ)γ )∇ . − 12 i(∇(α∇ .Rβ)γ I )MI
γαβ δ β)γ δ δ δ
.
+ 12 R . ∇β). + 14 iFγ(α∇ . − 12 i(∇ .Rγ(αβ)δ )∇δ − 14 (∇ .Fγ(α)∇β)
γ(αδ β)δ δ δ
.
+ (∇ .Rαβ I )MI − Rαβγ δ ∇δγ. − R . ∇γ .
γδ αβ δ

By inspection, or applying the previous Jacobis, we see

∆ .= 0 ⇒ ∆ . = Cγ(α ∆ ., ∆ . = − 13 ∆γ .
(γαβ)δ γαβ δ β)δ αβ γαβ

automatically, so the only new information comes from the trace of this Jacobi,

{∇α, W .} = i∆ .
β αβ

Evaluating {∇, W} in terms of its pieces, we find

.. . .
Rαβ γ δ = Fαβ = 0, γ δ
Rαβ γδ = δ(α δβ)B̄, W α.β = −B̄δ β.
α

W α. = ∇α.B̄ + ∇β W αβ
. , W α.βγ = − 12 ∇(β W α.γ)
 A. SUPERSPACE 545

..
∇αW . = ∇αW .γ δ = 0, ∇α W .γδ = −δα(γ (W .δ) + 12 i∇δ) .)B̄
β β β β β

where Wα is the Y part of Wα (= WαiY + ...).

Excercise XA2.1
Show that

{∇α, ∇β } = B̄Mαβ , ∇[α∇β ∇γ] = 0 ⇒ ∇α(∇2 + B̄) = − 12 B̄∇β Mβα

2
and thus ∇ +B gives a chiral superfield when acting on any superfield without
dotted indices.
For the other Jacobi of this dimension, we have

0 = [∇αα., {∇β , ∇γ.}] + {∇β , [∇γ., ∇αα.]} + {∇γ., [∇β , ∇αα.]}

= −i[∇αα., ∇β γ.] − i{∇β , Cγ.α.Wα + 12 (∇αB)M γ.α.} − i{∇γ., CβαW α. + 12 (∇α.B̄)Mβα}

2
= −iCα.γ.[fαβ − {∇β , Wα } + 12 (∇ B̄)Mαβ ] − h.c.
2
.
α
⇒ fαβ = 12 {∇(α, Wβ)} − 12 (∇ B̄)Mαβ , {∇α, Wα } + {∇ , W α.} = 0
(Here “h.c.” means “hermitian conjugate” without the reordering, which would gen-
erate non-operator terms.)
Evaluating {∇, W} in terms of its pieces, and combining with the results of the
previous Jacobi, we obtain the final result:
{∇α., ∇ .} = BM . ., {∇α, ∇ .} = −i∇ .
β αβ β αβ

[∇α., −i∇ .] = C . .Wβ − 12 (∇β B)M . ., [−i∇αα., −i∇ .] = C . .fαβ − h.c.

ββ βα αβ ββ βα

. . .
Wα = − B∇α − Gα ∇ + 2 (∇ Gα γ )M . . + 12 Wαβγ Mγβ + iWα Y + i 16 W β Mβα
β. 1 β
β βγ
. .
fαβ = i 12 G(α γ ∇β)γ. − 12 (∇(αB + i 13 W(α)∇β) + Wαβ γ ∇γ − 12 (∇(αGβ) γ )∇γ.
.
− ( 12 ∇2 B̄ + B B̄ + 12 1
i∇γ Wγ )Mαβ − i 18 [(∇(αδ G .)Mβ γ + α ↔ β]
γ)δ
. .
+ 12 Wαβ γδ Mδγ + 14 (∇(α∇γ Gβ) δ )M . . + i 12 (∇(αWβ) )Y
δγ
Wαβγδ = 1
∇ W
4! (α βγδ)

The “reduced tensors” B, Ga , Wα , Wαβγ satisfy the “reduced Bianchi identities”

.
Ga = Ḡa , ∇α.B = ∇α.Wα = ∇α.Wαβγ = 0, ∇αGαα. = ∇αB − iWα
. .
∇α Wαβγ − i 13 ∇(β Wγ) = −i 12 ∇(β α Gγ)α., ∇αWα + ∇αW α. = 0
546 X. SUPERGRAVITY

Note that B or Wα may vanish in certain gauges, for reasons to be explained in

subsection XA4.
Excercise XA2.2
In IXA4 we saw that integrals of total covariant derivatives vanished in curved
space by virtue of the identity Tab b = 0. Show that these torsions satisfy the
superpace generalization
(−1)B TAB B = 0

Excercise XA2.3
Using the expression for abcd in terms of spinor indices from subsection IIA5,
show
Tbcd = Ga abcd

Thus Ga is an axial vector.

Excercise XA2.4
By hermitian conjugation, find the commutators not written explicitly above,
and show the result is essentially the same as switching dotted and undotted
.α
indices (and similarly for bars), except that Ga , Y , and W α (and W ) get
extra minus signs. This illustrates CP invariance, and the fact that Ga is an
axial vector, while Y is a pseudoscalar (and similarly for W α ).
Excercise XA2.5
Derive the Bianchi identities in the absence of constraints, in terms of the
torsions and curvatures (as follow from the Jacobi identity):

∇[ATBC)D − T[AB|E TE|C)D = R[ABC) D

∇[ARBC) I − T[AB|E RE|C) I = 0

Excercise XA2.6
Show that in 4-component notation we can write

Tαβ c = −iγαβ
c
; Taβ γ = γaβδ Gδγ , Gαβ = −Gβα , ∇β Gβα = W α

This gives another way to see the result of excercise XA2.2. Show that this
expression for Gαβ = (Ga , B, B̄) gives it an interpretation as an SO(3,3) 6-
vector in SL(4) notation (see subsection IC5).
 A. SUPERSPACE 547

3. Compensators
Just as in ordinary gravity, compensators for scale transformations can be intro-
duced, but for supergravity the compensator should be a supersymmetric multiplet.
The simplest choice is the chiral scalar superfield Φ considered earlier: Its complex
“physical” scalar (scalar +i pseudoscalar) compensates local scale (the real part) and
U(1) (the imaginary part), its spinor compensates local S-supersymmetry, and its
auxiliary complex scalar appears as one of the auxiliary fields of supergravity.
Compensators are much more important in supergravity than in ordinary gravity:
Almost any flat space action can be coupled to gravity by the minimal coupling
prescription — replacing derivatives with covariant ones, and throwing a factor of
e−1 in for the measure. In supergravity this is not the case: As we’ll see in the next
section, we have both integrals over all superspace, which use E −1 , but also integrals
over chiral superspace (for integrating chiral superfields), which instead use Φ for the
measure. The minimal coupling procedure is then: (1) Use Φ (and Φ̄) to make a
flat-superspace action superconformally invariant, (2) replace the flat derivatives dA
with the curved ones ∇A, and (3) throw in the measure factors appropriate for the
integrals. (The last two steps couple conformal supergravity to a globally conformally
invariant theory.)
Another compensator that is commonly used is the “tensor multiplet”. (This
is sometimes confused in the literature with the “(complex) linear multiplet”, an-
other version of the scalar multiplet with no gauge fields whatsoever.) Treated as a
matter multiplet, it has the same physical content as the scalar multiplet, but the
pseudoscalar is replaced with a second-rank antisymmetric tensor gauge field,

δBmn = ∂[m λn]

To make things simpler, let’s look at flat space. We first note that this tensor is
“dual” to a pseudoscalar in the sense of switching field equations and constraints of
the field strength (see excercises IIB2.1 and VIIIA8.2): For the free fields,

F a = ∂a ϕ ⇒ ∂[aFb] = 0, Ga = 12 abcd ∂ bB cd ⇒ ∂ a Ga = 0

with the field equations following from “self-duality” under F ↔ G:

Fa = Ga ⇒ ∂ aFa = ∂[aGb] = 0

Since the theory of Bab must be described in terms of Ga alone (because of gauge
invariance), no renormalizable self-interactions are allowed; thus, this field is of little
548 X. SUPERGRAVITY

interest in quantum field theory outside of supergravity. In terms of the scalar, the
fact that only the field strength Fa appears in the field equations means there is the
global symmetry
δϕ = ζ
for constant parameter ζ. This generalizes to the nonabelian symmetries of nonlin-
ear σ models, resulting in derivative interactions (again nonrenormalizable) but no
potentials.
Excercise XA3.1
Consider coupling the tensor field to Yang-Mills: To preserve the tensor’s own
gauge symmetry, this coupling must be nonminimal. To produce such a cou-
pling, we start with the scalar and duality transform. The coupling we choose
is another 4D analog to the 2D model we considered in excercise VIIIA8.2,
replacing the pseudoscalar and total derivative 12 ab Fab with tr( 18 abcd FabFcd ).
(In general dimensions, the dual to a scalar is a rank-D−2 antisymmetric
tensor.) We start with the Lagrangian

L = − 14 φ φ + λφ 16
1
tr(abcd Fab Fcd )

for some coupling constant λ. Making use of the Chern-Simons form Babc of
subsection IIIC6 to write φ in this action only as ∂aφ, write a first-order form
of this action and perform a duality transformation to obtain
1 e2 eabc = 1 ∂[aBbc] + λBabc
L0 = 24
H , H 2

e is gauge invari-
Find the Yang-Mills gauge transformation of Bab . (Hint: H
ant.)
The tensor multiplet is described by a chiral spinor gauge field

δφα = id¯2 dα K (K = K̄)

Duality is then described in terms of the real scalar superfield strength (in the free
case)
.
F = φ + φ̄ ⇒ d¯2 dα F = 0, G = 12 (dα φα + d¯α. φ̄α) ⇒ d¯2 G = 0

with the field equation

F =G
(Fa appears at order θθ̄ in G, and Bab at order θ in φα.) Again the pseudoscalar has
a global symmetry: In terms of the superfield,

δφ = iζ
 A. SUPERSPACE 549

Now we return to curved space, covariantizing the above with respect to confor-
mal supergravity. We now identify the above global symmetry with the local axial
U(1) (R-)symmetry of supergravity. Thus, the superfield G does not compensate for
this symmetry; it remains as a symmetry in actions that use this compensator. In
R
particular, there are no d2 θ terms in such theories, except those that are locally
superscale invariant (so the compensator decouples). The matter tensor multiplet
also differs from the scalar multiplet in that it has no auxiliary fields (except for the
auxiliary components of the gauge field).

4. Scale gauges
Since all the covariant derivatives are built up from the spinor part of the vielbein,
we define the local superscale transformations for the covariant derivatives by first
defining
Eα0 = LEα
where L is a real, unconstrained superfield. The constraints then imply
.
0 0 β
∇α = L∇α + 2(∇ L)Mβα + 6(∇α L)Y,
β
∇α. = L∇α. + 2(∇ L)M . . − 6(∇α.L)Y
βα

From the anticommutator we find

−i∇0αα. =L2 (−i)∇αα. + 4L(∇α L)∇α. + 4L(∇α.L)∇α

.
β 4
+ 2 L (∇α∇ L )M . . + 12 L−2 (∇α.∇β L4 )Mβα − 32 L−2 ([∇α, ∇α.]L4)Y
1 −2
βα

Using the commutation relations, we then can show

B 0 = L6 (∇2 + B)L−4 , Wα0 = L3 [Wα − 12i(∇2 + B)∇α ln L]

G0αα. = (2[∇α, ∇α.] + Gαα.)L2 , 0

Wαβγ = L3 Wαβγ
From the way they appear in the commutators we also have that

Y Ga = 0, Y Wα = 12 Wα , Y Wαβγ = 12 Wαβγ , YB = B

From linearization, we see that B and W α pick out exactly the two irreducible halves
of the real scalar superfield L: The “vector multiplet” in Wα0 and the “scalar multiplet”
in B 0. (Compare the vector multiplet field strength and chiral scalar gauge fixing for
the prepotential V as described in subsections IVC4 and VIB9.) This means we can
completely fix the superscale gauge by the choice

B = Wα = 0
550 X. SUPERGRAVITY

as the generalization of the scale gauge in ordinary gravity that fixes the Ricci scalar
to vanish.
Excercise XA4.1
Derive the superscale transformations by use of the Bianchi identities:
a Use the commutation relations of the covariant derivatives (and the solution
to the Jacobi identities) to find all the transformations above. Show they
imply E 0 = L4 E.
b An easier way is to use the reduced Bianchi identities: Determine the trans-
formations of the reduced field strengths, up to constants, using chirality,
dimensional analysis, etc., and then solve for the constants by plugging into
the reduced identities.
We then define the scale (and U(1)) transformations of the compensators:

Φ0 = L2 Φ, Y Φ = 13 Φ; ∇α.Φ = 0

G0 = L4 G, Y G = 0; (∇2 + B)G = 0, G = Ḡ
where the scale weights follow from the U(1) weights (vanishing for G by reality) by
consistency with the constraints they satisfy.
Excercise XA4.2
Show that a superfield can be chiral only if it has no dotted indices. Then
show the relation that any such superfield has between scale and U(1) weights.
All these transformations can be derived either by consistency with the con-
straints, or by using the solution of the constraints: In terms of the unconstrained
superfields that solve the constraints, the superscale transformation is trivial:

ψ 0 = Lψ, Nα0 µ = Nαµ , Ω0 = Ω

The net result, as for super Yang-Mills, is that all the superficial transformations of
the constrained covariant derivatives are completely replaced with the new invariances
that appear upon solving the constraints: K−1 and L eliminate ψ, K αβ kills Nαµ , and
K M reduces ΩM to its real part U M , which transforms only under ΛM .
Excercise XA4.3
Rederive Aα as in subsection XA1, but in a general gauge, to find
←

iAα = Eα T, eT = ψ 2E(1 · e−Ω )

Show this result gives a superscale transformation for Aα that agrees with the
result above. Show the explicit solution for Φ in terms of φ and T also gives
it a superscale transformation that agrees with the above.
 A. SUPERSPACE 551

Excercise XA4.4
Often it is easier to use the solution to the constraints than the Jacobi iden-
tities:
a Solve for FAB in terms of Aα, and use the solution for Aα from subsection
XA1, to derive
2
Wα = −i(∇ + B)∇α(T + T̄ )
and use this to rederive the superscale transformation above. (Hint: Define
and use the chiral representation.)
b Find an explicit expression for B, and use it to rederive its superscale trans-
formation. (Hint: You will need to find Ωα βγ first. Since B is a scalar, you
can choose the Lorentz gauge Nαµ = δαµ .)
In subsection XA1 we found that a convenient way to simultaneously fix Lorentz,
U(1), and scale gauges was to choose Eα = Êα . (However, the corresponding com-
ponent invariances reappeared in the chiral gauge invariances.) Compensators allow
more freedom for gauge fixing: For example, we can fix the gauge B = Wα = 0 as
described above, or we can fix to 1 the compensator or a physical matter multiplet
(string gauge, as for gravity in subsection IXB5): The possibility of gauges such as
Φ = 1 or G = 1 depends on the existence in the action of such fields, and not on the
details of how they appear (as long as the gauge choice is consistent with the allowed
vacuum values). In particular, it does not depend on the signs of their kinetic terms,
which is the only thing that determines what is physical and what is a compensator.
Note that either Φ = 1 or G = 1 completely fixes the superscale gauge, in spite
of the constraints on these superfields. (E.g, Φ = Φ0 = 1 ⇒ L = 1.) This is due to
the appearance of the U(1) connection: For example, before fixing the scale and U(1)
gauges the chirality condition on Φ, rather than constraining Φ, actually determines
the spinor U(1) connection Aα:

∇αΦ̄ = (Eα − i 13 Aα)Φ̄ = 0 ⇒ Aα = −3iEα ln Φ̄

(But the chirality of the ratio of two chiral superfields with the same weights really
fixes it to be chiral; in other words, chirality of scalars makes all but one truly
chiral, since the U(1) connection can be determined only once.) As a result, the
scale (ΦΦ̄ = 1) and U(1) (Φ/Φ̄ = 1) gauge choice Φ = 1 determines Wα :

Φ=1 ⇒ Aα = 0 ⇒ Wα = 0

Similarly,
G=1 ⇒ 0 = (∇2 + B)G = B
552 X. SUPERGRAVITY

Conversely, we see that whenever one of the two field strengths B and W α is elim-
inated by a superscale(/U(1)) gauge choice in terms of one of the two compensators
Φ and G, the other field strength can be made superscale invariant: If we introduce
the compensator by a superscale transformation (as for gravity in subsection IXA7),
substituting either
L−4 → Φ̄Φ or G
in the above transformation laws, we find
2 2
B̃ = (Φ̄Φ)−3/2 (∇ + B)Φ̄Φ = Φ−1/2Φ̄−3/2 (∇ + B)Φ̄
fα = G−3/4 [Wα + 3i(∇2 + B)∇αln G]
W
fα or G for B̃ yields zero. We
as locally superscale invariant, where using Φ̄Φ for W
can therefore interpret gauging away the compensators as gauging them into the field
strengths: We have a choice of either

Φ=1 ⇒ Wα = 0, B̃ = B

G=1 ⇒ B = 0, fα = Wα
W
This is analogous to Stückelberg gauges (and their nonlinear generalizations): One of
these two tensors (gauge fields with respect to superscale) “eats” the compensator.
However, it differs from Stückelberg in that a second “gauge field” is completely
gauged away.
In fact, we’ll see in the next section that the pure supergravity actions constructed
using either of these compensators gives the corresponding field strength as its field
equation:
δ
⇒ B̃ = 0
δΦ
δ fα = 0
⇒ W
δΦα
Thus, either compensator can be used to eliminate both B and Wα , one as a field
equation and the other as a gauge choice. This result is already clear at this point
from dimensional analysis and chirality; similarly, we must have
δ
⇒ G̃a = 0
δU m
where G̃a is the result of applying a superscale transformation to Ga with whichever
of the two compensators is being used in the action. This leaves Wαβγ as the on-shell
field strength. The analogy to ordinary gravity is

(R, Rab − 12 ηab R, Wabcd ) ↔ (B/Wα, Ga , Wαβγ )

 A. SUPERSPACE 553

Although the Ricci tensor must appear in Ga , superscale invariance allows the
choice of gauges where the θ = 0 component is arbitrary: From the above we find the
linearized transformations

δGαα. ≈ 4[∇α, ∇α.]L, δAαα. ≈ −6[∇α, ∇α.]L

(For purposes of evaluating at θ = 0 we can neglect Aα | in δAa.) Thus, this axial vec-
tor component field can be moved around as convenient for component expansions. In
G̃a , they appear only in their invariant combination, Ga + 23 Aa in this approximation.
In nonsupersymmetric gauges, we can even gauge Ga | = 0.
Excercise XA4.5
Use the Bianchi identities instead of explicit superscale to track down the
axial vector:
a Use the relation of Wαβγ (which is scale covariant) to Wα (the field strength
for AA) and Ga to show that it is just this combination that appears in Wαβγ
(as its curl, for U(1) invariance).
b Show that
G=1 ⇒ B=0 ⇒ ∇a Ga = 0

Thus, in this gauge the axial vector gauge field Aa has been gauged out of
Ga | (although its field strength may appear at higher order: the gauge G = 1
doesn’t fix U(1)). What replaces it? (Hint: What’s in G?)

REFERENCES
1 D.Z. Freedman, P. van Nieuwenhuizen, and S. Ferrara, Phys. Rev. D13 (1976) 3214:
supergravity.
2 V.P. Akulov, D.V. Volkov, and V.A. Soroka, JETP Lett. 22 (1975) 187:
covariant derivatives for supergravity.
3 V. Ogievetsky and E. Sokatchev, Nucl. Phys. B124 (1977) 309;
V.P. Akulov, D.V. Volkov, and V.A. Soroka, Theor. Math. Phys. 31 (1977) 285;
S. Ferrara and B. Zumino, Nucl. Phys. B134 (1978) 301:
linearized off-shell supergravity (including auxiliary fields).
4 P. Breitenlohner, Phys. Lett. 67B (1977) 49; Nucl. Phys. B124 (1977) 500:
off-shell supergravity in components.
5 Wess and Zumino; Wess; loc. cit. (IVC, ref. 5):
superspace field equations.
6 W. Siegel, Supergravity superfields without a supermetric, Harvard preprint HUTP-
77/A068 (November 1977);
The superfield supergravity action, Harvard preprint HUTP-77/A080 (December 1977);
A polynomial action for a massive, self-interacting chiral superfield coupled to super-
gravity, Harvard preprint HUTP-77/A077 (December 1977);
554 X. SUPERGRAVITY

A derivation of the supercurrent superfield, Harvard preprint HUTP-77/A089 (Decem-

ber 1977):
off-shell supergravity in superspace, compensator superfield.
7 M. Kaku, P.K. Townsend, and P. van Nieuwenhuizen, Phys. Rev. D17 (1978) 3179:
conformal supergravity in components.
8 M.F. Sohnius and P.C. West, Phys. Lett. 105B (1981) 353:
R-symmetry generator in superspace covariant derivative.
9 V.I. Ogievetsky and I.V. Polubarinov, Sov. J. Nucl. Phys. 4 (1967) 156;
K. Hayashi, Phys. Lett. 44B (1973) 497;
M. Kalb and P. Ramond, Phys. Rev. D9 (1974) 2273;
E. Cremmer and J. Scherk, Nucl. Phys. B72 (1974) 117:
antisymmetric tensor gauge field.
10 J. Wess, Acta Phys. Austriaca 41 (1975) 409:
N=2 supersymmetric tensor multiplet.
11 W. Siegel, Phys. Lett. 85B (1979) 333:
N=1 supersymmetric tensor multiplet.
12 H. Nicolai and P.K. Townsend, Phys. Lett. 98B (1981) 257;
E. Bergshoeff, M. de Roo, B. de Wit, and P. van Nieuwenhuizen, Nucl. Phys. B195
(1982) 97;
G.F. Chapline and N.S. Manton, Phys. Lett. 120B (1983) 105:
Yang-Mills appearing as Chern-Simons contribution to tensor field strength.
13 P.S. Howe and R.W. Tucker, Phys. Lett. 80B (1978) 138:
superscale transformations.
14 Gates, Grisaru, Roček, and Siegel, loc. cit.:
complete superspace treatment of N=1 supergravity.
 B. ACTIONS 555

............................. .............................
............................. B. ACTIONS .............................

Now that we understand the structure of superfields in curved superspace, we an-

alyze various supergravity theories through their actions. We will use several methods
for finding and evaluating supergravity actions. These actions are significantly more
complicated than those we have encountered previously, and it is difficult to see all
their features simultaneously, so for any particular application we use the method
which best simplifies the property we most need: (1) Superspace methods are the
best for finding general actions and their symmetry properties, manifesting super-
symmetry, using globally supersymmetric gauges, and performing quantum calcula-
tions. (2) Component methods are useful for comparing actions and other properties
to nonsupersymmetric theories. Such approaches sometimes make some use of super-
space, but not superspace integration. (3) Compensators are useful in conjunction
with either of these methods, and can extract many important features and terms
in the action with little more than the results of global supersymmetry. They reveal
useful broken symmetries, and are the simplest way to analyze the “superhiggs effect”
(Higgs for local supersymmetry).

1. Integration
The action for supergravity follows from dimensional analysis: Since the usual
R
Einstein-Hilbert Lagrangian has dimension +2, as does d4 θ, the superspace La-
grangian must be dimensionless. The only covariant possibility in terms of the po-
tentials (EA M and ΩAI ) is thus
Z
SSG = 3 dx d4 θ E −1

including a normalization factor that will prove convenient later. Introducing the
compensator and local scale invariance must also make the usual action for super-
gravity look like the kinetic term for the compensator multiplet, i.e.,
Z
SSG,c = 3 dx d4 θ E −1 Φ̄Φ

(For simplicity we will restrict ourselves for the most part to the simplest compensator,
the chiral scalar.) The previous form then corresponds to the scale gauge Φ̄Φ = 1;
often the scale + U(1) gauge ψ = 1 is more convenient.
There should also be a supersymmetrization of the cosmological term. This might
seem difficult, requiring explicit prepotentials. However, we know from our study
of de Sitter space that the cosmological term is basically a statement about the
556 X. SUPERGRAVITY

conformal compensator. Therefore, the cosmological term for supergravity, in terms

of the superconformal compensator Φ, should be the supersymmetrization of the
corresponding term in ordinary gravity, a dimensionless self-interaction for a scalar.
The solution of the chirality condition can be written as
←

Y Φ3 = Φ3 ⇒ Φ3 = φ3e Ω̄ E

in the gauge ψ = 1. The result for the cosmological term is then

Z Z
−1 3
Sscosmo = dx d θ E Φ + h.c. = dx d2 θ φ3 + h.c.
2

independent of scale or U(1) gauge: As we saw in subsection XA1, this expression is

invariant under Λ transformations, and the integrand itself is invariant under K and
L transformations.
Excercise XB1.1
Although this final result for the cosmological term in terms of φ is locally
superscale invariant, the derivation started in the gauge ψ = 1. Generalize
the derivation, and the result in terms of Φ, to arbitrary gauges (see excercise
XA4.3).
A chiral expression for SSG can be found by similar methods:
Z Z
SSG = 3 dx d θ E B = 3 dx d2 θ̄ E −1 B̄
2 −1

Thus, as for super Yang-Mills, the action can be expressed as a real, chiral, or an-
tichiral integral. With the compensator,
Z Z
2
SSG = 3 dx d θ E Φ(∇ + B)Φ̄ = 3 dx d2 θ̄ E −1 Φ̄(∇2 + B̄)Φ
2 −1

or, more generally,

Z Z Z
2
dx d θ E L = dx d θ E (∇ + B)L = dx d2 θ̄ E −1 (∇2 + B̄)L
4 −1 2 −1

which is just the naive covariantization of the flat-space result (at least in the gauge
ψ = 1: see excercise XB1.1). Clearly this method generalizes to coupling to other
R R
multiplets, and allows both d2 θ and d4 θ integrals to be generalized to curved su-
perspace. In fact, the analysis of the compensator is much simpler than that of the
conformal supergravity that couples to it to produce ordinary supergravity: Just as
in ordinary gravity, where use of just the compensator allowed us to study certain in-
teresting solutions in gravity, namely de Sitter space and cosomology, some properties
of supergravity can be analyzed in terms of just the compensator.
 B. ACTIONS 557

A simple expression (as simple as the super Yang-Mills case) can be written for
the supergravity action in terms of unconstrained superfields. We first give a first-
order action, analogous to the one for Yang-Mills (subsection IVC5): In that case the
action was in terms of V and Aa; here it is in terms of U m and Em a. Using just the
constraints solved in subsection XA1, we write the action as
Z .
SSG,1 = 3 dx d4 θ E −1 Φ̄Φ(1 − 14 Tα,α.αα)

in terms of the torsion T .c . (Note the similarity to the Yang-Mills case, replacing the
αβ
Chern-Simons form with the same component of the torsion.) We already evaluated
everything except this torsion, which is also easily found in the gauge ψ = 1, Nα µ = δαµ,
.
Ωµ = Ωµ = 0: Z
SSG,1 = 3 dx d4 θ det(Em a )Φ̄Φ(1 − 14 Êa m Em a)
←

Φ = [det(Ema )]−1/3(1 · e Ω̄ )1/3eΩ̄ φ, b̄ .} = −iÊ .m ∂

{Êα , E α αα m

In the chiral representation this simplifies to

Z ← .
SSG,1 = 3 dx d4 θ [det(Ema )]1/3(1 · e−U )1/3φ(e−U φ̄)(1 − 14 Em ααi∂¯α.Êαm )

Excercise XB1.2
Find the algebraic field equation for Em a . Use this to eliminate it from the
action, yielding expressions for E m a , E, and the (second-order) action in
terms of U m only.
The expansion of the superspace action in terms of unconstrained superfields is
needed for supergraphs, the most efficient way to do quantum calculations. We will
not consider quantization here; the methods are similar to those described in subsec-
tions VIB5, 9-10, and C5 for super Yang-Mills. In particular, one uses background
field methods: For example, as for super Yang-Mills,
eΩ → eΩB eΩQ ⇒ Êα → e−H Eˆα eH , eH = eΩQ eΩ̄Q
The end result is that the expansion is about background-covariant derivatives DA,
e.g.,
b α = e−H Dα eH ,
∇ ∇α = e−H (ψDα + Ωα I MI )eH
etc. However, DA satisfy the same constraints as the full
. covariant derivatives: For
example, they have nonvanishing torsion T . = −iδαδ .. This differs from the ex-
c γ γ
αβ β
pansion implied above in U m about partial derivatives, which anticommute without
torsion: For a perturbation expansion useful for quantum calculations, one must ex-
pand in hab about hea m i, rather than in ea m itself; thus (at least) the vacuum value
hUi must be separated from U.
558 X. SUPERGRAVITY

2. Ectoplasm
Although all supersymmetric theories can be analyzed directly in superspace (in-
cluding classical solutions, effective potentials, Feynman graphs, etc.), for comparison
to nonsupersymmetric theories it is necessary to expand superfields in components.
Since all fundamental theories are described by actions, it is sufficient to give a pre-
scription for evaluating any action in terms of component fields, as in subsection IVC2
for global supersymmetry. In locally supersymmetric theories, the vielbein needs to
be expanded in terms of the prepotentials for supergraphs. We can also find compo-
nent actions by a straightforward Taylor expansion in θ of the prepotentials in the
superspace action. However, for component expansions of classical actions, one can
get by more simply by applying Bianchi identities to the covariant derivatives and
differential forms (antisymmetric tensors). It is unnecessary to know even the explicit
form of the measure in terms of the vielbein or prepotentials.
The fundamental idea is to think of the Lagrangian not as a scalar times a mea-
sure, but more “geometrically” as an antisymmetric tensor. Although this approach
does not work for the usual superspace Lagrangians because of the peculiarities of
fermionic integration, it can be applied to component Lagrangians integrated over
4D spacetime, treated as the bosonic subspace of superspace. We thus write the
component action as
Z
S = 4! dxm dxn dxp dxq Lmnpq (x, θ)
1

where LM N P Q is a graded antisymmetric superfield. Of course, the action should be

independent of θ, even though we have integrated over only x. This is equivalent to
requiring that the integral should be independent of the choice of 4D hypersurface in
superspace. We are familiar with a similar requirement for conserved charges, which
are defined as integrals over 3D hypersurfaces: Treating the conserved current in
terms of the 3-form dual to the vector,
Z
√
Jmnp = −gmnpq J q
⇒ Q = 3! dxm dxn dxp Jmnp
1

the dual to the usual conservation law as vanishing (covariant) divergence of the
vector is vanishing curl of the 3-form:
d
Q = 0 ⇒ ∂[m Jnpq] = 0
dt
An important point is that neither the definition of the charge nor the conservation
law requires a metric, since integration in general does not. We thus require for our
supersymmetric action
∂
S = 0 ⇒ ∂[M LN P QR) = 0, δLM N P Q = 3!1 ∂[M λN P Q)
∂θ
 B. ACTIONS 559

where the gauge invariance allows us to drop terms in the Lagrangian that are total
derivatives (surface terms). Note that both L and λ are assumed to be local func-
tions of the fields and their (finite-order) derivatives. (As a result, this is not the
usual “cohomology”, where both would be allowed to be arbitrary functions of the
coordinates.)
Converting the curl-free condition to flat indices (see subsection IVC5 for the
Chern-Simons superform),

1
∇ L
4! [A BCDE)
− 2!3!
1
T[AB|F LF |CDE) = 0, δLABCD = 1
∇ λ
3! [A BCD)
− (2!)
1 E
2 T[AB| λE|CD)

The plan is then to find LABCD , in terms of which the action can be written as
Z
S = dx (− 4!1 )mnpq Eq D Ep C En B Em ALABCD

where Em A is exactly the nontrivial part of the inverse vielbein EM A :

Em A = (ema , ψm α )

namely the inverse vierbein and the gravitino. (If we also write ψm α = em a ψaα , we
can collect all em a factors into a factor of e−1 using the  tensor.)
The next step is to explicitly solve the curl-free condition on the 4-form in terms
of the usual scalar superspace Lagrangian. (An alternative is to solve the Bianchis for
the field strength of the 3-form gauge field, which is also a 4-form.) The procedure
is the same as that used to solve the Bianchi identities for covariant derivatives (in
subsection XA2): We start with the lowest-dimension equations and work up. The
equations that include the constant (vacuum/flat-space) part of the torsion can be
solved algebraically, the rest give differential constraints. Of course, we will need to
use the results of subsection XA2 for the torsions and their constraints. The result is
. .
α 2
Lαβcd = α ,β α,cd L̄,
α .
Lαbcd = iαα,bcd ∇ L̄,
. Labcd = abcd [(∇ + 3B)L̄ + h.c.]

and their complex conjugates, the rest vanishing, where L is the usual chiral super-
space Lagrangian (superpotential):

∇α.L = 0
.
(We use the shorthand notation a = (αα), etc.)
The result is thus the component expansion of the usual curved superspace action
Z
S = dx d2 θ E −1 L + h.c.
560 X. SUPERGRAVITY

Using the Bianchi identities of the covariant derivatives one can also covariantize the
usual solution to the chirality condition:
2
L = (∇ + B)L

which allows us to identify the action as

Z
S = dx d4 θ E −1 (L + L̄)

so L can be taken real without loss of generality for general d4 θ integrals. On the
other hand, for supergravity we can take

LSG = 3, L̄SG = 0 ⇒ LSG = 3B, L̄SG = 0

..
or vice versa, and the curvature appears in terms of Rαβ αβ and not R̄ . .αβ (like the
αβ
corresponding fαβ without f¯. . for Yang-Mills), with half as many terms to collect
αβ
(for the same final result). In general, we thus have an expression for S in terms of
Em A and the components of LABCD , and for the latter in terms of curvatures and
covariant derivatives of L, which can be evaluated by the same methods as in flat
space (except that the commutation relations of the covariant derivatives are more
complicated).
Components of a superfield are again defined by evaluating its covariant deriva-
tives at θ = 0. However, as in the case of global supersymmetry, the value of θ is
arbitrary, since the result for the action is θ independent: We therefore will gener-
ally drop the “ | ” in component expansions of actions; any superfield then implicitly
refers to the corresponding component. This θ independence also means that it is
not necessary to make any gauge choices: These methods automatically express the
action in terms of just the component fields that cannot be completely gauged away.
(For example, Eα m | never appears.)
Excercise XB2.1
Collect all the above results:
R
a Find the complete component expression for the most general chiral ( d2 θ)
action in terms of covariant derivatives of the superpotential, and the super-
gravity fields.
R
b Do the same for a real ( d4 θ) action.
Excercise XB2.2
Evaluate the above actions for massive φ3 theory (for a chiral matter field φ)
in terms of the components of φ.
 B. ACTIONS 561

Excercise XB2.3
Do the same for super Yang-Mills:
a Solve the Bianchi identities for super Yang-Mills in curved superspace.
b Use this result to evaluate the component expansion of its action.

3. Component transformations

We saw in the previous subsection that in component expansions the supergravity

gauge fields naturally appear as Em A, since by definition we restrict to the bosonic
submanifold. Similar remarks apply to the component form of their supercoordinate
transformations (i.e., local supersymmetry), and the related component expansion
of their field strengths: In subsection IXB4, we saw that coordinate transformations
(as applied to solving the radial gauge condition) were simpler for EM A because the
derivative term on the parameter was just ∂M K A .
We therefore begin by rewriting the gauge (coordinate) transformations in terms
of Em A . As for gravity (see subsections IXA2 and IXB4), we can choose to make the
transformation laws more manifestly covariant by writing the generators in terms of
covariant derivatives. Then, as for gravity,

K = K A ∇A + K I MI , δ∇A = [K, ∇A] ⇒

δEM A = ∇M K A −EM B K C TCB A +K I MI EM A, δΩM I = −∇M K I +EM B K C RCB I

δ(TAB C , RAB I ) = K(TAB C , RAB I )

using (δEAM )EM B = −EA M δEM B . Here ∇M = EM A ∇A = ∂M + ΩM I MI , so these

transformations on Em A and Ωm A contain only bosonic derivatives ∂m , other than
those implicit in the torsions and curvatures. (Similar remarks apply to only Em A
and Ωm I appearing on the right.) Since the component expansion is an expansion in
θ, it is really only supersymmetry K α = α which is no longer manifest; specializing
to that, the transformations on the physical gauge fields become, using the results of
subsection XA2 for the torsions,
. . . .
δem αα
= −i( ψ̄m + ¯ ψm ),
α α α α
δψm = ∇m  + iem (β Gα . − ¯ .δβαB)
α α ββ
β β

We have not yet determined the solution for the Lorentz connection ωm ab and the
transformation laws for the auxiliary fields. We will also need the relation between
562 X. SUPERGRAVITY

the usual curvature and the components of the superfield strengths. All of these can
be found by use of the identities (see subsection IXA2):

−En C Em B TBC A = −TmnA = ∂[m En] A + E[m B Ωn]B A

En D Em C RCD ab = Rmn ab = ∂[m Ωn] ab + Ω[m ac Ωn]c b

The application of these identities is similar to that for expansion of the action in
the previous section: The separation of the factors of Em A into bosonic and fermionic
parts yields an expansion in powers of the gravitino field. For the torsion case where
the index A = a we solve for ωmab in terms of the torsions (auxiliary fields and
constants) and vielbein; for the torsion case A = α we solve for Tab γ , used for the
transformation law of the auxiliary fields, and for the curvature case we solve for
Rab cd , used in the component expansion of the action, in terms of these auxiliaries,
the vielbein, and the just-determined connection. The U(1) connection Am needs no
solution: It is pure (superscale) gauge, and will cancel in actions (after perhaps an
appropriate redefinition of Ga ). For these manipulations we use the relations that
TAB C and RAB cd have to B, Ga , Wα , Wαβγ , and their derivatives (as expressed by the
solution to the Bianchi identities given in subsection XA2). The solution is
◦ .
ωmbc = em a [ωabc − 12 (T̂bca − T̂a[bc])], T̂ab c ≡ ea m eb n Tmn c = abcd Gd + iψ[aγ ψ̄b]γ

Tab γ = −eam eb n∇[m ψn] γ + i(ψaβ Gγ . − ψ̄ .δβγ B − a ↔ b) ≡ Cαβ t . .γ + C . .tαβ γ

β aβ αβ αβ
.
δB = − 23 α tαβ β , δGαα. = −β (tβαα. + 13 Cβαt . .β ) + h.c.
γβ

Rab γδ = eam eb n Rmnγδ + B̄ψa (γ ψbδ)

. .
− 2i[ψa (Cβ tγδ . + 13 δ(β
γ β
δ)t . . ) − ψ̄a β 16 t(β γδ) − a ↔ b]
β β

◦
where “ ” refers to the usual expression for pure gravity, ψaβ ≡ ea m ψm β , and we
have chosen the superscale gauge W α = 0 for simplicity.

4. Component approach
We can now take the superspace action of subsection XB1, as expanded in com-
ponents by the ectoplasm method of subsection XB2, and substitute the component
expansions of the field strengths found in subsection XB3, to find the component
action
LSG = LG + Lψ + e−1La
.
LG = − 14 e−1 R, Lψ = mnpq ψ̄mα. 12 {enαα, ∇p }ψqα, La = − 38 (Ga )2 + 3B̄B
 B. ACTIONS 563

(Note the signs are again consistent with Ga and B, B̄ forming a 6-vector of SO(3,3),
though not in the same way as in excercise XA2.6.) Here ∇ and LG are the usual
covariant derivative and Einstein-Hilbert action of general relativity in terms of e and
ω, but ω is slightly different from any of the connections used previously (see excercise
XB4.1 below). It also differs from the ω given above in that we have explicitly
extracted the Ga piece (which is the sole source of ω in the ectoplasm approach).
An alternative to ectoplasm to determine ω is to use a first-order formalism: Rather
than imposing the usual torsion constraint, we can leave the Lorentz connection as
an independent field in R and in the ∇ in Lψ . Eliminating the Lorentz connection
by its field equation yields a modified torsion constraint, and produces ψ 4 terms in
the action. We have written Lψ in a form manifestly symmetric with respect to
integration by parts. (Alternatively, we can write ψ̄e∇ψ − ψe∇ψ̄.)
As an alternative to deriving the component action from the simpler superspace
expression, we can postulate the component action directly. In the component ap-
proach writing the action in components is more direct than the superspace approach
by definition, but proving supersymmetry invariance is less so. This is not so true
when coupling to matter, where writing component actions can also be as compli-
cated as deriving them from superspace, so here we consider the simplest case, pure
supergravity. We thus begin by postulating LSG = LG + Lψ ; the first term is obvious,
while the second follows from minimal coupling for the free gravitino action, which can
be derived easily by many methods (see, e.g., subsection XIIA5 below). We ignore
the auxiliary fields, which are necessary for off-shell closure of the supersymmetry
algebra, but not for supersymmetry invariance of the action.
We write the action for gravity in a form that more resembles the gravitino action
(see excercise IXA5.5):

LG = − 14 e−1 R = 1 mnpq
16
 abcd em a enb Rpq cd = 18 mnpq em a enb R̃pqab
. .
= i 18 mnpq em αα(R̄npα.β e . − Rnpα β eqβ α.)
qαβ
where we have switched to spinor notation for the curvature (see subsection IXA1) and
used duality in both vector and spinor notation (see subsection IIA7). (Alternatively,
we can regard this as the definition of the gravity action.) We then have for the
variation of this part of the action (after some integration by parts)
. . .
1 mnpq
δLG = i 4  em {[R̄npα δe − Rnpα δeqβ α] − [(δ ω̄nα )T . − (δωnα β )Tpqβ α.]}
αα . β . β . . β
qαβ pqαβ

where we have used (see excercise IIIC1.2)

Z Z
δRmn = ∇[m δωn] ,
I I
dx ∇mV = dx ∂m V m = 0
m
564 X. SUPERGRAVITY

Next, we pick the obvious transformation law for the gravitino field as the gauge
field of supersymmetry:
δψmα = ∇m α

The transformation laws for e and ω will be derived as a by-product of the invariance
proof, as will the explicit expression for ω in terms of e and ψ. Substituting this
expression for δψ into Lψ ,
. .
δLψ = mnpq [α∇m 12 {enαα , ∇p}ψ̄qα. − ¯α.∇m 12 {enαα, ∇p }ψqα
. .
+ 12 ψ̄mα.(δenαα )∇pψqα − 12 ψmα (δenαα )∇pψ̄qα.
. . .
+ 12 ψ̄mα.enαα(δωpα β )ψqβ − 12 ψmα enαα(δ ω̄pα.β )ψ̄ .]
qβ

where we have integrated by parts to free the supersymmetry parameters of deriva-

tives. We then use the antisymmetrization on all curved indices to collect the resulting
terms into torsion and curvature as

∇{e, ∇} = {e, ∇∇} + (∇e)∇ = 12 {e, R} − 12 T ∇

The curvature terms then cancel those from δLG, if we choose for δe in δLG the
transformation law
. . .
δemαα = −i(αψ̄m α + ¯αψm α )

Then we also substitute this expression for δe in δLψ , and note that half those terms
immediately drop out, since
ψ[m α ψn]α = 0

by antisymmetry. The remaining terms from both LG and Lψ then can be collected
as
.
δLSG = − 18 mnpq Temnαα∆pqαα.
. . .
Temnαα ≡ Tmnαα − iψ̄[m α ψn] α
.
∆pqαα ≡ i(δ ω̄[pα )e . − i(δω[pαβ )eq]β α. + α∇[pψ̄q]α. − ¯α.∇[pψq]α
. . β
q]βα

We now note that the former factor in δLSG vanishes by virtue of the equation of
motion from varying the connection: Rather than vanishing, the torsion now satisfies
.
second − order : Temn αα = 0

We can regard Te as the “supersymmetrized torsion”; this is equivalent on shell to the

result we found in the previous subsection from superspace. We can therefore quit
now, since in a second-order formalism the torsion (and thus the Lorentz connection)
 B. ACTIONS 565

would satisfy this equation even off shell. (This approach, using the second-order
formalism but not bothering to substitute the supersymmetry variation of the con-
nection, is called the “1.5-order formalism”.) On the other hand, we can just as easily
recognize that in the first-order formalism cancelation of δLSG is also guaranteed by
allowing vanishing of the latter factor to define the supersymmetry variation of the
(independent) connection:

first − order : ∆pqαα. = 0

Thus, use of the first-order formalism requires no more work than 1.5-order (contrary
to remarks in the literature), which is really the same as second-order, and provides
the bonus of yielding the transformation law for ω. However, it is useful to note that
not all quantities should have their longer forms substituted at the beginning of a
calculation (just as we learned in high-school algebra not to plug in numbers till the
end).
Excercise XB4.1
Let’s complete this calculation to the bitter end, finding all the properties of
the connection:
a Solve the torsion constraint for ω (see subsection IXA3).
b Find the transformation law for ω that follows from cancelation of the above
terms off shell (i.e., without imposing the torsion constraint).
c Show the above two results are consistent (modulo terms with ψ field equa-
tions, which can be canceled by contributions from auxiliary fields) by plug-
ging the expressions for δe and δψ into the variation of the result for part a
and comparing with the result for part b.
d Compare these results with the connection found in the previous subsection.
How does the appearance of Ga affect the transformation law?

5. Duality
Although antisymmetric tensor gauge fields can be avoided in general, they tend
to turn up in string theory, so we now look at them a little more generally, examining
their actions and how they relate to those for scalars. In particular, we note that
a sensible action for such a tensor alone cannot be constructed that is conformally
invariant: From the same analysis as for electromagnetism or Yang-Mills (subsection
IXA7), we see that (Fa)2 does not give a scale-invariant action in four dimensions.
Thus, such a field is not suitable as a compensator for pure gravity. However, in
566 X. SUPERGRAVITY

supergravity the tensor multiplet (see subsection XA3) also has an ordinary scalar,
and an appropriate power of it can make the tensor’s action conformal. Therefore,
we now examine general duality transformations for the supersymmetric case, which
is more relevant for understanding its use in gravity. We will consider explicitly a
flat superspace background for simplicity, but generalization to curved superspace
by covariantization is straightforward, replacing flat superspace derivatives with (su-
perconformal) covariant derivatives, introducing supergravity field strengths where
2
necessary (in this case, just d¯2 → ∇ + B for chirality), and using the covariant
integration measures.
Duality transformations can be performed directly in the action by use of first-
order formulations. Starting with the general tensor multiplet action
Z
Stm = dx d4 θ K(G)

where K is some function and G = dα φα + h.c., we write this in first-order form as

Z
0 e ) + V G]
Stm = dx d4 θ [K(V

where V is an unconstrained real superfield and K e is the Legendre transform of K:

For this action to reduce to the previous upon applying the algebraic field equation
of V , we must have
e ) + V G]| ∂ K̃(V )
[K(V = K(G)
∂V
=−G

The duality transformation is then performed by varying φα instead of V in SG0 :

Remembering that φα is chiral, so
Z Z
δ d x d θ V G = 2 d4 x d2 θ (δφα)d¯2 dα V + h.c.
4 4 1

we solve the condition on V as

d¯2 dα V = 0 ⇒ V = φ + φ̄

since thinking of V as the prepotential for a vector multiplet says that it is pure
gauge. The dualized action is then
Z
Sφ = dx d4 θ K(φe + φ̄)

We can also reverse the procedure through another first-order action

Z
0
Sφ = dx d4 θ [K(V ) − V (φ + φ̄)]
 B. ACTIONS 567

where in this case varying with respect to φ implies

d̄2 V = 0 ⇒ V =G

while varying with respect to V gives the inverse Legendre transform

e + φ̄)
[K(V ) − V (φ + φ̄)]| ∂K(V ) =φ+φ̄ = K(φ
∂V

The simplest case is the Lagrangian 12 G2 : We then find

K(V ) = 12 V 2 ⇔ e ) = −1V 2
K(V 2

so the duality is
Ltm = 12 G2 ⇔ Lφ = − 12 (φ + φ̄)2

In flat space, this gives the usual free result Lφ = −φ̄φ, but in curved space the
− 12 φ2 + h.c. part does not vanish because E −1 is not chiral. Consequently, this action
is not the conformal one (as we already knew from the component argument above).
However, the conformal one is easy to find by starting with −φ̄φ: Making the field
redefinition φ → eφ expresses the action in terms of φ + φ̄. Legendre transforming,

e ) = −eV
K(V ⇔ K(V ) = V (ln V − 1)

so the duality is
Lφ = −eφ+φ̄ ⇔ Ltm = G(ln G − 1)

These two conformal actions for matter, when coupled to conformal supergravity,
become the two “minimal” actions for supergravity, when the overall sign is changed
to make the matter fields into compensators: The version with φ as the compensator
is called “old minimal”, while that with φα is called “new minimal”. They differ only
off-shell, in their choice of auxiliary fields. Note that the field equations for the two
conformal multiplets,

d¯2 Φ̄ = 0 (Φ ≡ eφ ), d¯2 dα ln G = 0

fα = 0
reproduce the compensator part of the supergravity field equations B̃ = 0 and W
described in subsection XA4. Again, the full expressions follow from the usual su-
pergravitational and superscale invariances, which were used to find B̃ and Wfα ; the
compensator dependence is enough to identify them as the appropriate covariantiza-
tions.
Excercise XB5.1
We saw in subsection IVC5 for the Chern-Simons form, or XB2 for ectoplasmic
568 X. SUPERGRAVITY

integrals, that differential forms can be defined in superspace. Do the same

for the tensor multiplet:
a By generalizing the bosonic case to superspace with curved indices, and then
“flattening” the indices (as for the Chern-Simons superform), show that the
super 2-form BAB with field strength HABC and gauge parameter λA is de-
scribed by
δBAB = ∇[AλB) − TAB C λC , HABC = 12 ∇[ABBC) − 12 T[AB|D BD|C)
(Hint: Show that replacing EA → ∇A and CAB C → TAB C yields only cancel-
ing connection terms.)
b Show that the torsions given in subsection XA2 satisfy
Z .
dx d4 θ E −1 Hα,α,. αα = 0

Note that this also implies the gauge invariance of the Chern-Simons form of
the super Yang-Mills action in curved superspace.
c Show that the constraints
Hαβγ = Hαβ γ. = Hαβc = 0, H . . = −iCαγ C . .G
α,β,γ γ βγ

(and complex conjugates) can be solved by

Bαβ = B . = 0, B . . = −iC . .φβ ; ∇α.φβ = 0
αβ α,β β αβ
.
Bab = C . .bαβ + Cαβ b̄ . ., bαβ = 12 ∇(αφβ); G = 12 (∇αφα + ∇α.φ̄α )
αβ αβ
Relate the results for the gauge fields BAB to those for the Yang-Mills field
strengths FAB (subsection IVC3).
d Supersymmetrize the construction of excercise XA3.1: Show one can define a
field strength
eABC = HABC + BABC
H
using the Chern-Simons superform BABC .
The supergravity component action with the tensor multiplet compensator differs
from the one of the previous subsection in that B and the longitudinal part of G have
been replaced by the gauge field Bmn :
La → 12 mnpq Gm ∂n Bpq
which is the only possibility that preserves the gauge invariances of both G and B
while leaving them both auxiliary. (Their field equations are that their field strengths
vanish.)
 B. ACTIONS 569

6. Superhiggs
Supergravity affects spontaneous supersymmetry breaking in a simple way: From
the discussion of the immediately preceding subsections, we know that supergravity
can be described more simply as conformal supergravity coupled to a compensator.
Simple (N=1) conformal supergravity contains no scalars: It consists of only confor-
mal gravity (the traceless part of the metric), the conformal (traceless) part of the
gravitino field, and an auxiliary gauge vector. Since symmetry breaking involves giv-
ing vacuum values to only scalars, we can replace supergravity by just its compensator
for these purposes.
For a general analysis, consider a kinetic term
Z
SK = dx d4 θ 3φ̄φe−K(χ ,χ̄i )/3
i

This is the most general kinetic term with the usual number of spacetime derivatives:
Any term of the form f(φ, φ̄, χi , χ̄i) can be rewritten in this form after appropriate field
redefinitions. In particular, if we start with fields with arbitrary Weyl scale weight,
then this form follows after rescaling fields so only φ carries scale weight, since all
terms in the Lagrangian must have the same scale weight, fixed by (super)conformal
invariance. φ is then the only field to carry U(1) weight, which is proportional to
scale weight by superconformal invariance. Then φ appears only as φ̄φ, while K is an
arbitrary function of χi and χ̄i . The first step in evaluating this action in components
is to simply ignore conformal supergravity altogether, and evaluate this action as is,
in terms of matter and compensator multiplets, by the methods we have considered
previously for evaluating θ integration. The next step is to add back in some parts of
conformal supergravity: (1) the conformal graviton, which can be put back in easily
and uniquely using coordinate and local scale invariance; (2) the U(1) axial gauge
vector, whose coupling is minimal, and thus follows directly from U(1) covariantizing
the spacetime derivatives; and (3) the conformal gravitino, whose quartic couplings
can be quite complicated, but as a practical matter we are interested in only the
mass term, which is determined from the mass of the Goldstone fermion it eats,
which appears in the compensator (and the kinetic term, which is the usual one).
(The exponential form will prove convenient for later component analysis.)
Before considering the general case, we look at the pure supergravity case under
this analysis: Looking at just the bosons, we find
Z
SSG,b = dx e−1 12 {−3[(∇ − i 13 A)φ̄] · [(∇ + i 13 A)φ] − 12 φ̄φR + 6B̄B}
570 X. SUPERGRAVITY

where ∇ is the usual covariant derivative of general relativity, A is the U(1) gauge
vector, and the relative coefficent of the R term was fixed by local scale invariance
(see subsection IXA7). Note that here B is the usual auxiliary field from φ, and is
not associated with conformal supergravity. Choosing the component U(1) and scale
gauges φ| = 1, this reduces to
Z
SSG,b → dx e−1 (− 14 R − 16 A2 + 3B̄B)

Relating to Ga , we recall that if we had included it from conformal supergravity,

for this compensator G̃a = Ga + 23 Aa, so we can identify Aa with 32 G̃a . Thus, the
compensator method immediately yields the bosonic action, including auxiliary fields.
Returning to the general case, the part of the action for the “physical” scalars (φ|
and χ|) then starts out as
Z
SK,ps = dx e−1 e−K/3 12 {−3[(∇ − i 13 A)φ̄] · [(∇ + i 13 A)φ] + φ̄[(∇ + i 13 A)φ] · (∂¯i K)∇χ̄i

+φ[(∇ − i 13 A)φ̄] · (∂i K)∇χi + φ̄φ[(∂i∂¯j K) − 13 (∂iK)(∂¯j K)](∇χ̄j ) · (∇χi) − 12 φ̄φR}

ignoring until the following subsection the auxiliary scalars, which are irrelevant for
the kinetic term. We use the notation ∂i = ∂/∂χi, ∂¯i = ∂/∂ χ̄i. We then choose the
U(1) and scale gauges
φ| = eK(χ |,χ̄i |)/6
i

where we have explicitly written the |’s to emphasize that this is a nonsupersymmetric
gauge choice for the component φ|. Finally, we eliminate A by its algebraic field
equation. We thus obtain
Z
SK,ps → dx e−1 21 [(∂i ∂¯j K)(∇χ̄j ) · (∇χi) − 12 R]

Except for the R term and covariant derivatives, this is what would follow in flat
R
superspace from the action − dx d4 θ K. For supersymmetry breaking, we also need
the super cosmological term
Z
Sc = dx d2 θ λφ3 + h.c.

i
for some constant λ. We could consider more general potentials φ3ef (χ ) (again the
power of φ is fixed by scale and U(1)), but then the field redefinition φ → φe−f /3
would remove it while replacing K → K + f + f. ¯ (This invariance, and the form of
the “metric” on the space of fields χ and χ̄ appearing in the action, identify K as a
“Kähler potential”.)
 B. ACTIONS 571

The analysis for SK can also be made by performing a duality transformation on

the compensator. Following the same steps as described in the previous subsection for
the case without matter (factoring the overall −3 out of the process for convenience),
we find Z
SK → − dx d4 θ [3G ln G + GK(χ, χ̄)]

Since in this form A decouples, the result is obvious from the flat-space result.
Normally any kind of symmetry breaking will generate a cosmological term, since
a scalar getting a vacuum value implies the potential itself getting one, giving a term
R
dx e−1 constant. This would require adding a cosmological term to the action by
hand to cancel the generated one, since within observational limits no cosmological
constant is observed in nature. (In any case, the constant generated would corre-
spond to a subatomic length scale, whereas a realistic cosmological constant requires
a cosmological length scale, which means a constant, going as 1/length2 , of the order
of 10−80 in subatomic units.) An exception is when the potential is flat in some direc-
tion: In supersymmetry energy is always positive, and the supersymmetric vacuum
has zero energy, but some potentials allow other, perhaps nonsupersymmetric, vacuua
that also have zero energy, and thus generate no cosmological constant. This avoids
the ad hoc procedure of “fine tuning” the cosmological constant of an added term for
exact cancelation (or at least to order 10−80 ).

7. No-scale
A useful example of the superhiggs effect with a flat potential is “no-scale super-
gravity”. This theory has an explicit super-cosmological term, but the kinetic term
is such that this term does not generate a component cosmological term, but does
spontaneously break supersymmetry. The simplest example describes supergravity
coupled to a single chiral scalar multiplet. The kinetic term has an SU(1,1) symme-
try, and also appears in N=4 supergravity (see subsection XC6 below). Written in
terms of just the compensator part of supergravity, it is
Z
SK = dx d4 θ 3(φ̄χ + χ̄φ)

where φ is the compensator and χ is the matter. We have written it in a manifestly

U(1,1) covariant form, where the U(1,1) metric is off-diagonal ( 01 10 instead of the


usual diagonalized 10 −1
0
). For the above component analysis we redefine
Z
χ → φχ ⇒ SK → dx d4 θ 3φ̄φ(χ̄ + χ) ⇒ K = −3 ln(χ̄ + χ)
572 X. SUPERGRAVITY

(Many other superfield redefinitions are possible to put this in more conventional
forms, such as (3φ̄φ − χ̄χ), φ̄φ(3 − χ̄χ), etc.) The kinetic term for the physical scalars
follows from the same analysis we applied to the CP(1) model in subsection IVA2.
The only differences here are: (1) the symmetry is U(1,1), not U(2), and (2) the
constraint on the norm of the complex 2-vector follows not from a Lagrange multipler
(or a low-energy limit), but as a local scale gauge chosen to give the Einstein-Hilbert
curvature term the usual normalization. Alternatively, we can use the analysis given
in the previous subsection for the general case to find
Z  
−1 1 |∇χ|2
SK,ps → dx e 2 3 − R1
(χ̄ + χ)2 2

However, to study just the supersymmetry breaking, we want to look at the “po-
tential” terms: terms that involve the auxiliary scalars instead of spacetime deriva-
tives. We thus now need to include the super cosmological term, which breaks the
SU(1,1) invariance. Again evaluating at first without conformal supergravity, then
putting some (all but the conformal gravitino) back in, we find the contributions from
SK and Sc
Z
Saux = dx e−1 3[B̄B(χ̄ + χ) + (B̄φb + B φ̄b̄) + λ(Bφ2 + B̄ φ̄2)]

where B = d2 φ and b = d2 χ. We then see that eliminating the auxiliaries gives

nothing, so there is no potential to generate a cosmological term. However, there is
still a mass term for the gravitino: As always, Sc also contains the spinor term

6λ(φζ 2 + h.c.)

where ζα = dα φ is the trace of the gravitino. The gravitino in this model therefore
has a mass proportional to λhχ̄ + χi−1/2.
SU(1,1) invariant kinetic terms also appear in superstring theory, but unlike N=4
and no-scale supergravity, the kinetic term is (φ̄χ + χ̄φ)1/3 instead of just φ̄χ + χ̄φ.
(See subsection XIB6.) When applying no-scale supergravity to nature, more matter
multiplets are added,
Z Z 
3 f (χi/φ)
S = dx d θ 3(φ̄χ + χ̄φ − χ̄i χ ) +
4 i 2
dx d θ λφ e + h.c.

generalizing SU(1,1) to SU(n,1) in the first term. (N=5 supergravity has such an
SU(5,1) symmetry; see below.) Then χ acts as the “hidden” matter sector that doesn’t
directly couple to the observed matter χi, but serves only to break supersymmetry.
 B. ACTIONS 573

REFERENCES
1 W. Rarita and J. Schwinger, Phys. Rev. 60 (1941) 61:
action for spin 3/2.
2 Freedman, van Nieuwenhuizen, and Ferrara, loc. cit. (XA):
complete component formalism for supergravity, except for auxiliary fields.
3 S. Deser and B. Zumino, Phys. Lett. 62B (1976) 335:
first-order formalism for supergravity.
4 A.H. Chamseddine and P.C. West, Nucl. Phys. B129 (1977) 39;
P.K. Townsend and P. van Nieuwenhuizen, Phys. Lett. 67B (1977) 439:
1.5-order formalism.
5 P. van Nieuwenhuizen, Phys. Rep. 68 (1981) 189:
general review of supergravity.
6 S.J. Gates, Jr., hep-th/9709104, Ectoplasm has no topology: the prelude, in Supersym-
metries and quantum symmetries, proc., July 22-26, 1997, Dubna (Lecture notes in
physics, v. 524), eds. J. Wess and E.A. Ivanov (Springer, 1999) p. 46;
S.J. Gates, Jr., M.T. Grisaru, M.E. Knutt-Wehlau, and W. Siegel, hep-th/9711151,
Phys. Lett. 421 (1998) 203:
ectoplasm.
7 Wess and Zumino; Wess; loc. cit. (IVC, ref. 5):
component expansions for supergravity.
8 W. Siegel, loc. cit. (XA);
K.S. Stelle and P.C. West, Phys. Lett. 74B (1978) 330;
S. Ferrara and P. van Nieuwenhuizen, Phys. Lett. 74B (1978) 333:
old-minimal supergravity.
9 Sohnius and West, loc. cit. (XA):
new-minimal supergravity.
10 D.V. Volkov and V.A. Soroka, JETP Lett. 18 (1973) 312;
S. Deser and B. Zumino, Phys. Rev. Lett. 38 (1977) 1433:
superhiggs.
11 B. Zumino, Phys. Lett. 87B (1979) 203:
Kähler in N=1 nonlinear σ models.
12 E. Cremmer, S. Ferrara, C. Kounnas, and D.V. Nanopoulos, Phys. Lett. 133B (1983)
61:
no-scale supergravity.
574 X. SUPERGRAVITY

................. .................
................. C. HIGHER DIMENSIONS .................

A convenient method for describing extended supersymmetry in D=4 is to ap-

ply dimensional reduction to supersymmetry in D>4, since (1) spinors are bigger in
D>4, so even simple supersymmetry reduces to extended supersymmetry, and (2) the
Lorentz group is bigger in D>4, so some 4D scalars arise as parts of higher-D vectors,
etc., meaning fewer Lorentz representations in the multiplet in D>4.

1. Dirac spinors
We saw in subsection IC1 that coordinate representations of orthogonal groups
SO(D) could be defined in terms of self-conjugate fermions,

Gab = 12 [γa, γb ], {γa , γb } = δab

We now will construct explicit matrix representations of the Dirac matrices for arbi-
trary D, and examine their properties. This is useful for understanding: (1) repre-
sentations of internal symmetries, such as in Grand Unified Theories; (2) theories in
higher dimensions, which give simpler formulations of certain four-dimensional the-
ories when the extra dimensions are eliminated, and appear in string theory; and
(3) properties of spinors that are independent of D, or their dependence on D, which
is useful for comparison and for perturbation in quantum field theory.
An explicit solution can be found easily by first looking at even dimensions, and
breaking up the problem into D/2=n two-dimensional problems. Furthermore, we can
look first at the Euclidean case (SO(D)), and solve for the other cases (SO(D+,D− ))
by Wick rotation. The solution for SO(2) is just two of the Pauli σ matrices. The
general solution then comes from the direct product of the two-dimensional cases,
using the third σ matrix to introduce appropriate “Klein factors” (see excercise IA2.3)
to insure that the γ matrices from one two-dimensional subspace anticommute with
those from another. The resulting γ matrices are then:
1 √ √ √ 1 √ √
√ ( 2σ3 ⊗ · · · ⊗ 2σ3 ⊗ 2σi ⊗ I ⊗ · · · ⊗ I), √ ( 2σ3 ⊗ · · · ⊗ 2σ3)
2 2
√
where i = 1, 2, there are a total of n factors, and the number of 2σ3 and I factors
in the first expression ranges from 0 to n−1. The last matrix can always be included
to extend SO(2n) to SO(2n+1); in fact, up to normalization, it’s simply the product
of all the other γ’s. (In other words, the product of all the γ matrices is proportional
to the identity.)
 C. HIGHER DIMENSIONS 575

The next step is to notice that this construction generally gives a reducible rep-
resentation. Reducibility comes from two properties: (1) For SO(2n) we really have
SO(2n+1); and (2) the representation may be real. In fact, most of the interest-
ing cases involve SO(2n) (in particular, SO(3,1) for Lorentz and SO(4,2) for confor-
mal in four dimensions). In that case we can call the first (or any other) γ matrix
(σ1 ⊗ I ⊗ · · · ⊗ I) for SO(2n+1) “γ−1 ”, and take the rest as those for SO(2n). Then
the projection operators
√
Π± = 12 (1 ± 2γ−1 ) ⇒ Π±2 = Π± , Π+Π− = Π−Π+ = 0, Π+ + Π− = 1

commute with the SO(2n) generators Gab ∼ γγ, so they can be used to project the
representation of the γ’s into two representations of SO(2n). These two halves of
a Dirac spinor are known as “Weyl spinors”. A convenient representation of the γ
matrices for this purpose is the one given in subsection IIA6, with the representation
of the Pauli matrices used in our SU(2)/SL(2,C) discussion of subsections IIA1 and
5,




σ1 = √12 10 −1
0
, σ2 = √12 01 10 , σ3 = √12 0i −i
0

We then can write the spinor, which has 2n components (since it represents the direct
product of n representations of σ matrices, each of which has two components) as
two 2n−1 -component spinors projected by



Π± = σ± ⊗ I ⊗ ... ⊗ I, σ+ = 10 00 , σ− = 00 01

The γ matrices then take the block-diagonal form

γ−1 = √12 I0 −I
0
, other γ = 0 σ̃
σ0

We will refer to these reduced matrices σ (and σ̃), and the γ matrices themselves for
SO(2n+1), as generalized Pauli (σ) matrices.
The reality properties of the representation depend on the existence of a metric
η .B (or Ω . B for psuedoreality, which doesn’t reduce the representation), as in our
A A
discussion of classical groups of subsection IB5. In fact, all the spinor representations
of any orthogonal group are also defining representations of another group: For less
than seven dimensions, this leads to the identification of covering groups discussed in
subsection IC5; for more than six dimensions, it only identifies the orthogonal group
as a subgroup of this new group. (An interesting exception is SO(8), where the spinor
representations are also 8-dimensional, and are the two other defining representations
of SO(8).) In matrix notation, we look for a matrix C = η or Ω such that we can
define the operation of charge conjugation as

Ψ → C −1Ψ*, GΨ → C −1 (GΨ)* ⇒ G = C −1 G*C

576 X. SUPERGRAVITY

If we like, we can also choose

C = C † = C −1
without loss of generality. For a representation to be invariant under charge conju-
gation (i.e., real)
Ψ = C −1 Ψ* ⇒ C* = C −1

For our γ matrix representation, the matrix to look at is

√ √
C = ... ⊗ C2 2σ3 ⊗ C2 ⊗ C2 2σ3 ⊗ C2

where
0 −i


C2 = i 0

independent of the representation used for the Pauli matrices. (Our representation is
√
simplest, since then C2 2σ3 = I, and C = C †. In other representations, C may also
need an n-dependent factor of i if we want C = C † .) Using properties of σ matrices
we found in our discussion of SO(3) in subsection IIA2, such as C2σ*C2 = −σ, we
find

C −1 γ*C = (−1)nγ, C* = (−1)n(n+1)/2 C −1 , C T = (−1)n(n+1)/2 C, C † = C −1

This distinguishes 8 cases, where the irreducible spinors are:

SO(8m): Weyl and real
SO(8m+1): real
SO(8m+2): Weyl
SO(8m+3): pseudoreal
SO(8m+4): Weyl and pseudoreal
SO(8m+5): pseudoreal
SO(8m+6): Weyl
SO(8m+7): real

For SO(4m+2), charge conjugation does not preserve γ−1 , and thus Π±. Therefore,
in those cases there is no metric η . B or Ω . B on the irreducible spinor: The Dirac
A A
spinor consists of two irreducible spinors that are complex conjugate representations
of each other. In general, a Dirac spinor has 2n complex components for SO(2n) and
SO(2n+1); the Weyl condition reduces this a factor of two for SO(2n), as does reality
where applicable. (Pseudoreality does nothing.)
It is useful to know .the other group metrics, if they exist. For unitarity properties
we look for a metric ΥAB such that

G = −Υ−1 G† Υ ⇒ Υ = Υ†
 C. HIGHER DIMENSIONS 577

(Thus g = eG satisfies Υg −1 = g † Υ.) We can also choose

Υ = Υ−1

without loss of generality. We therefore look for a metric satisfying

Υ−1 γ † Υ = γ

so G ∼ [γ, γ] is antihermitian with respect to Υ. For SO(D), we have simply

Υ=I

since the hermiticity of the σ matrices implies that of the γ matrices. We then can also
define a metric to raise and lower indices in terms of these two metrics, by contracting
the dotted (or undotted) indices: In matrix notation, we then have

(C T Υ)−1 γ T (C T Υ) = (−1)n γ ⇒ G = −(C T Υ)−1 GT (C T Υ)

For all cases except SO(4m), this also defines the symmetry properties of the gener-
alized σ matrices: They can be defined as C T Υγ for SO(2n+1), and as its diagonal
blocks with respect to Π± for SO(4m+2); but for SO(4m) it’s off-diagonal, so the
generalized σ matrices appearing there carry one each of the two different kinds of
spinor indices, and thus have no symmetry. Then we rewrite the above result as

(C T Υγ)T = (−1)n(n−1)/2(C T Υγ)

2. Wick rotation
The indefinite-metric groups SO(D+ ,D−) (D+ 6= 0 6= D− ) can be treated by
Wick rotation: giving i’s to D− of the γa ’s, so the corresponding components of ηab
get minus signs. Since this affects γ T in the same way as γ, the metric C T Υ is
unchanged. In other words,

C T Υ = CET (ΥE = I)

in terms of the Euclidean C of the previous subsection. However, γ* and γ † are

affected in the opposite way to γ and γ T (−i’s instead of i’s). Υ then becomes (up
to normalization) the product of the timelike γ’s,
Y √
Υ∼ 2γa
ηaa <0

which also determines the modification of C. In the above equations for γ † and γ T ,
we then find a factor of −1 for each rotated dimension, coming from anticommutation
578 X. SUPERGRAVITY

with each timelike γ in Υ, so we redefine all γ’s by an overall factor of iD− to preserve
their pseudohermiticity. This changes the normalization to

Gab = (−1)D− 12 [γa , γb ], {γa , γb } = (−1)D− ηab

In odd dimensions the σ matrices are the γ matrices (up to multiplication by one
of the metrics), while in even dimensions the γ matrices consist of two off-diagonal
blocks of the σ matrices. To write actions we also need the “dual” Dirac spinor, in
the sense of a Hilbert-space inner product,

Ψ̄ = Ψ† Υ
√
with Υ as defined above. In particular, it is just 2γ0 in D− = 1.
We then find (e.g., using the explicit representation given above) that C has the
same properties with regard to symmetry and γ−1 for SO(D+ ,D−) as for SO(D+ − 1,
D− − 1). Thus, the properties of these metrics on the irreducible (as opposed to
Dirac) spinors follow easily from the Euclidean case by using

CT Υ : SO(D+ , D− ) ← SO(D+ + D− )

C: SO(D+ , D−) ← SO(D+ − D− )

Then the properties of Υ follow from the above two in the cases where all 3 exist;
the few cases where only Υ exists, which have D even, follow from the next higher D
(increasing D+ by 1).
Excercise XC2.1
Find the explicit γ matrices for D = 2 and 4 from the construction of the
previous section, and apply this Wick rotation. Compare the results with the
conventions of subsections VIIIA7 and IIA6.
We can use these properties to determine that the number of real components D0
of an irreducible spinor is

x mod 8 0 1 2 3 4 5 6 7
D0 = 2[D−2+f (D+ −D− )]/2, f(x) 0 1 2 3 2 3 2 1

The complete results can be summarized by the following table, showing for each
case of SO(D − D− ,D− ), for D mod 8 and D− mod 4, the types of irreducible spinors
ψ, the types of metrics η (symmetric) and Ω (antisymmetric) for these irreducible
spinors, and the type of generalized σ matrices (and its symmetry, where relevant):
 C. HIGHER DIMENSIONS 579

D− 0 1 2 3
D Euclidean Lorentz conformal
ψα ψα 0 ψα ψα. ψα ψα 0 ψα ψα.
. .
0 η αβ ηα.β η αβ η αβ η αβ Ωα.β Ωαβ η αβ
σαβ 0 σ . σαβ 0 σ .
αβ αβ

ψα ψα ψα ψα
. . . .
1 η ηα.β η αβ
αβ
η ηα.β η αβ
αβ
η αβ
Ωα.β Ωαβ η αβ
Ωα.β Ωαβ
σ(αβ) σ(αβ) σ(αβ) σ(αβ)
ψα ψ α ψα ψ α ψα ψ α ψα ψ α
. .
2 η αβ ηα.β Ωαβ Ωα.β
σ(αβ) σ (αβ) σ(αβ) σ (αβ) σ(αβ) σ (αβ) σ(αβ) σ (αβ)
ψα ψα ψα ψα
. . . .
3 Ω Ωα.β η αβ
αβ
Ω ηα.β Ωαβ
αβ
Ω ηα.β Ωαβ
αβ
Ω Ωα.β η αβ
αβ

σ(αβ) σ(αβ) σ(αβ) σ(αβ)

ψα ψα 0 ψα ψα. ψα ψα 0 ψα ψα.
. .
4 Ω Ωα.β η αβ
αβ
Ωαβ
Ω ηα.β Ωαβ
αβ
Ωαβ
σαβ 0 σ . σαβ 0 σ .
αβ αβ

ψα ψα ψα ψα
αβ
. . . .
5 Ω Ωα.β η αβ Ω αβ
Ωα.β η αβ Ω αβ
ηα.β Ωαβ Ω αβ
ηα.β Ωαβ
σ[αβ] σ[αβ] σ[αβ] σ[αβ]
ψα ψ α ψα ψ α ψα ψ α ψα ψ α
. .
6 η αβ Ωα.β Ωαβ ηα.β
σ[αβ] σ [αβ] σ[αβ] σ [αβ] σ[αβ] σ [αβ] σ[αβ] σ [αβ]
ψα ψα ψα ψα
αβ
. . . .
7 η ηα.β η αβ η αβ
Ωα.β Ωαβ η αβ
Ωα.β Ωαβ η αβ
ηα.β η αβ
σ[αβ] σ[αβ] σ[αβ] σ[αβ]

Excercise XC2.2
The vectors of SO(D+ ,D− ) for 3≤ D+ + D− ≤6 were expressed as tensors
with two spinor indices, with the appropriate symmetry and tracelessness
conditions, in subsection IC5.
a Show that the spinor metrics found from the Dirac analysis are sufficient to
identify each of these covering groups.
b Show the equivalence of each orthogonal group to its covering group by show-
ing that (1) the Lie algebras have the same dimension, and (2) the deter-
minant (or its square root) of this tensor gives the appropriate orthogonal
580 X. SUPERGRAVITY

metric. (Hint: Do the cases D+ + D− = 4 and 6 first, and specialize to 3 and

5.)
Excercise XC2.3
Consider the groups SO(n,n) and SO(n,n+1). Explicitly construct a real
representation of the γ matrices, demonstrating that all such spinors are real.
Besides the Dirac spinors and Dirac matrices γ, and the irreducible spinors and
Pauli matrices σ, it is also useful to introduce irreducible real (“Majorana”) spinors
and corresponding matrices Γ. When the irreducible spinors are already real these
are the same, but when the irreducible spinors are complex this real spinor is just the
direct sum of the irreducible spinor and its complex conjugate, a spinor with twice as
many components. In general, these generalized Majorana spinors and matrices have
many properties that are independent of the number of dimensions, but depend on
the number of time dimensions:

D− 0 1 2 3
Euclidean Lorentz conformal
ψα ψα 0 ψα ψ α ψα ψα 0 ψα ψ α
η αβ Ωαβ
Γαβ 0 Γ(αβ) Γ(αβ) Γαβ 0 Γ[αβ] Γ[αβ]

0
For D odd, there is only one irreducible spinor, so there is a metric M αβ or M αβ
to relate the two spinors listed. For D − 2D− twice odd (2 mod 4), the original
irreducible spinor was complex, so there is a metric representing a U(1) generator
that rotates the complex spinor and its complex conjugate oppositely. (I.e., it’s the
identity on the complex spinor and minus the identity for the complex conjugate.)
For D − 2D− = 3, 4, 5 mod 8, the original spinor was pseudoreal, and this U(1)
can be extended to an SU(2): Since the complex spinor and its complex conjugate
transform the same way under the orthogonal group, they can be paired as a doublet
of SU(2). This doubled representation is a real representation of the orthogonal
group ⊗ SU(2), since the direct product of the two antisymmetric charge-conjugation
matrices is symmetric (two −’s under transposition).

3. Other spins

Before considering supersymmetry in higher dimensions, we first study represen-

tations of the Poincaré group there. From the general analysis of section IIB, we
know that general on-shell representations follow from the massless ones, which can
 C. HIGHER DIMENSIONS 581

be classified by their representation of the lightcone little group SO(D−2). Specifi-

cally, the bosons can be described as traceless tensors of a certain symmetry (labeled
by a Young tableau), while the fermions can be labeled as the direct product of such
tensors with an irreducible spinor, with a tracelessness condition imposed between
any vector index and the spinor index using a γ or σ matrix. Similar methods can be
used to find the off-shell representations in terms of representations of SO(D−1,1),
but without subtracting traces. (For full details, see chapter XII.) The gauge degrees
of freedom can be subtracted from these Lorentz representations by dropping all lower
vector indices with the value “−”, by the usual lightcone gauge condition; this tells
us the number of total physical + auxiliary degrees of freedom.
In practice, the only interesting massless fields in higher dimensions are: (1) the
metric (graviton), (2) totally antisymmetric tensors (including scalars and vectors),
(3) spin-3/2 (gravitino), desribed by vector⊗spinor, and (4) spinors. By the methods
described above, the counting of physical, auxiliary, and gauge degrees of freedom
for these fields is (where D0 is the number of components of an irreducible spinor of
SO(D−1,1) — see the previous subsection):

field physical auxiliary gauge

2 D(D − 3)
1
h(ab) D D
D−2

D−2

D−1


A[a1 ...an] n n−1 n−1
1 0 1 0 0
ψaα 2 D (D − 3) 2 D (D + 1) D
1 0 1 0
χα 2D 2D 0

Excercise XC3.1
Derive all the entries in the table. For each type of field, find the minimum
D for which physical degrees exist.
We next consider exactly how many higher dimensions are relevant. From the
previous subsection, we see that an irreducible spinor (which we use for the super-
symmetry generators) has 1 component in D=2, 2 in D=3, 4 in D=4, 8 in D= 5 or
6, 16 in D=7, 8, 9, or 10, 32 in D=11, etc. Since the maximal Lorentz symmetry can
be obtained by looking at the maximum D for which a certain size spinor exists, we
see that the appropriate D for which an irreducible spinor reduces to N irreducible
spinors (for N-extended supersymmetry) in D=4 is
582 X. SUPERGRAVITY

N D
1 4
2 6
4 10
8 11

etc. From the discussion of subsection IIC5, we know that supergravity exists only
for N≤8, and super Yang-Mills only for N≤4. This means that simple supergravity
(i.e., any supergravity) exists only for D≤11, and simple super Yang-Mills for D≤10.
Since theories with massless states of spin>2 are not of physical interest (in fact, no
interacting examples have been constructed), we can restrict ourselves to looking at
just D=4, 6, 10, and 11. In general, an irreducible multiplet in some D can become
reducible in lower D. However, since irreducible multiplets of supersymmetry are con-
structed as the direct product of the smallest representation of supersymmetry with
an arbitrary representation of the Poincaré group, this reducibility corresponds di-
rectly to the reducibility of that Poincaré representation, which occurs simply because
the Lorentz group gets smaller upon reduction. In particular, the smallest represen-
tation of supersymmetry is itself irreducible. For the case of simple supersymmetry,
this is the scalar multiplet (scalars and spinors) in D=6, the vector multiplet (super
Yang-Mills: vectors, spinors, and scalars) in D=10, and supergravity in D=11. The
statement that it is the smallest multiplet in that number of dimensions is directly
related to the fact that it does not exist in higher dimensions.

4. Supersymmetry

We first generalize to arbitrary dimensions some definitions used earlier: To dis-

cuss the properties of supersymmetry that are common to all dimensions (but one
time), it’s most convenient to use the Majorana form

{qα, qβ } = Γaαβ pa

which is consistent with the general symmetry of these matrices. The supersymmetry
generators are then
∂ ∂
qα = −i α + 12 Γaαβ θβ a
∂θ ∂x
and αqα generates the infinitesimal transformations

δθα = α, δxa = i 12 Γaαβ α θβ

 C. HIGHER DIMENSIONS 583

where (qα)† = −qα. The covariant derivatives are

∂
dα = α
+ 12 Γaαβ θβ pa
∂θ
and they satisfy the same algebra as supersymmetry

{dα , dβ } = Γaαβ pa

but with the opposite hermiticity condition (dα )† = +dα . The invariant infinitesimals
are
dθα , dxa + i 12 (dθα )Γaαβ θβ

Superfields can be expanded as either

Φ(x, θ) = φ(x) + θα ψα(x) + ...

or
ψα = dα Φ, ...

giving the transformations

δφ = α ψα, δψα = −iβ 12 Γaαβ ∂a φ + ..., ...

Representations can be found as for D=4; we don’t have twistors in general, but
we can always use a lightcone frame. We first need to define Γaαβ , which in general
is independent of Γaαβ (only the latter was needed to define supersymmetry above):
The analog of the Dirac anticommutation relations (which can be reconstructed if we
combine the two Γ’s, as generalized σ’s, to form a generalized γ) is

Γ(a
αγ Γ
b)γβ
= η abδαβ

In the lightcone frame the momentum is just pa = δ+ p with p+ = ±1 being the sign
a +

of the (canonical) energy. In this frame we have the constraint Γ− q = 0. This projects
away half the q’s, since −Γ± Γ∓ are projection operators: Using the anticommutation
relations of Γ± ,

Π± = −Γ± Γ∓ ⇒ Π±2 = Π±, Π+Π− = Π−Π+ = 0, Π+ + Π− = 1

The equality of the sizes of the two subspaces follows from parity symmetry, Γ± ↔ Γ∓ .
We thus need to consider only half of the q’s, namely Γ+ q. We therefore switch
to a notation where we consider the truncated spinor qµ with just that half of the
584 X. SUPERGRAVITY

components. This “lightcone spinor” is an irreducible spinor of SO(D−2). In a

Majorana basis it satisfies the same commutation relations as Dirac matrices,

{qµ , qν } = δµν

Since qµ has an even number of components in D > 3, the states that represent
this algebra form a Dirac spinor of SO(2n ) (n > 0) that is reducible to two Weyl
spinors. (These spinors should not be confused with those of SO(D−2), such as qµ,
which is a vector of this SO(2n ).) Since supersymmetry takes each of these “spinors”
into the other, one spinor contains all the bosons, while the other contains all the
fermions. There are an equal number of physical boson and fermion states because
the two Weyl spinors are equal in size. Since SO(D−2)⊂SO(2n ), each Weyl spinor
of SO(2n ) is reducible with respect to SO(D−2). The only exception is D=4, where
SO(D−2)=SO(2n )=SO(2), and there is one bosonic state and one fermionic one.
This “Dirac spinor” of SO(2n ) is the smallest representation of supersymmetry. It
can also be represented in terms of anticommuting coordinates, by dividing up qµ into
two halves, one of which is complex coordinates, the other half being both the complex
and canonical conjugate (as for the fermionic harmonic oscillators of excercise IA2.3).
The most general representation of supersymmetry is then the direct product of this
one with an arbitrary representation of the Poincaré group.
All the results of this section can be extended to “extended supersymmetry”,
with supersymmetry generators qiα for an N-valued “internal” index i, as expected
from our discussion of supergroups in subsection IIC4: For example, in D=4 the
supergroup describing extended conformal supersymmetry, SU(2,2|N), includes con-
formal symmetry SU(2,2), internal symmetry U(N), N supersymmetries, and N S-
supersymmetries. In general, the supersymmetries then satisfy the algbera

{qiα, qjβ } = δij Γaαβ pa

The smallest representation of an extended supersymmetry follows as before, where

now the complete lightcone q acts as Dirac matrices for SO(N2n). Other representa-
tions are again found by direct product, now between this smallest supersymmetry
representation and an arbitrary representation of both Poincaré and the internal sym-
metry. For the more interesting cases, where N itself is a power of 2, the smallest
representation can also be derived by dimensional reduction from higher dimensions
of N=1 (“simple”) supersymmetry, changing the higher-dimensional algebra only by
setting some components of the momentum to vanish, and noting that a spinor of
higher dimensions reduces to many spinors, as clear from our explicit construction
 C. HIGHER DIMENSIONS 585

earlier. (Other representations tend to be reducible, since the Poincaré representation

in the direct product is reducible upon dimensional reduction.) Dimensional reduc-
tion can also be defined for an action (for supersymmetric or nonsupersymmetric
theories), by again setting the derivatives with respect to the “extra” coordinates to
vanish, and also restricting the integration to the reduced set of coordinates. An-
other interpretation is that we expand the fields over all momentum modes in the
extra coordinates, and then drop all but the zero (constant) modes.
We also recall from subsection XC2 the index structure of spinors in D=6, 10,
and 11, which we need to write supersymmetry covariant derivatives. We thus have,
for simple supersymmetry,

D = 6 : {diα , djβ } = −Cji i∂αβ

D = 10 : {dα , dβ } = −σαβ
a
i∂a
D = 11 : {dα , dβ } = −σαβ
a
i∂a

where in the case of D=6 we have taken advantage of the fact that SO(5,1)=SU*(4)
to eliminate vector indices, and introduced the SU(2) index i for spinors to make
them Majorana.

5. Theories
We first consider the scalar multiplet in D=6. The constraints and field equations
are given by the statement, in terms of supersymmetry covariant derivatives, that
there are only scalars and spinors on shell, and by supersymmetry their physical
polarizations must be equal in number. Since a spinor has 4 polarizations in D=6,
we must have 4 real scalars, and thus

diα φjk0 = Cji ψk0 α

The second SU(2) index k 0 is introduced again to make a spinor (this time the field)
Majorana, and performs a similar service for the scalars. This one equation is suf-
ficient to completely describe this multiplet on shell in the free case; interactions
require derivatives, so we won’t consider them here. This multiplet reduces to N=2
in D=4 in a very simple way: The SU(2) index on d labels the 2 supersymmetries,
and the 4-component spinor index reduces in the obvious way to SL(2,C) indices,
.
α → (α, α), with appropriate 6D spinor conventions.
Excercise XC5.1
Show the equations given for the 6D scalar multiplet give the complete field
586 X. SUPERGRAVITY

equations for all the components, and that only the scalars and spinors shown
explicitly in that equation survive on shell.
This six-dimensional theory gives a simple example of nontrivial dimensional re-
duction: Assume we have a 5-dimensional theory with a nontrivial U(1) symmetry.
Then we can dimensionally reduce by choosing the fields to depend on the fifth coor-
dinate in such a way that the fifth component of the momentum of each field is equal
to a constant m (with dimensions of mass) times its U(1) charge Q:

p4 = Z = mQ

This is consistent at the interacting level because each term in the action satsifies con-
servation of the U(1) charge as well as conservation of momentum. (This is equivalent
to how we introduced masses by dimensional reduction earlier for free fields, since any
free field can be “complexified”.) This has an interesting effect on the supersymmetry
algebra: It introduces a U(1) charge Z (called “central” because it commutes with
the rest of the algebra). For example, if we start with the 6D supersymmetry algebra
(like the above algebra for the supersymmetry covariant derivatives), introduce the
central charge in reducing to 5, and then do an ordinary reduction to 4 (or vice versa),
the supersymmetry algebra becomes (see subsection IVC7)

{qiα, q̄ j.} = δij p ., {qiα, qjβ } = Cαβ Cij Z, {q̄αi., q̄ j.} = C . .C ij Z

β αβ β αβ

If the higher-dimensional theory was massless, then p2 + Z 2 = 0 for the 4D theory.

More generally, if the higher-dimensional theory already had masses before the central
charge was introduced, then by supersymmetry it satisfied p2 + M02 = 0, M02 ≥ 0
(since supersymmetry always has positive potentials), while afterwards the 4D theory
satisfies
p2 + Z 2 + M02 = 0 ⇒ M 2 = M02 + Z 2 ≥ Z 2

where M is the 4D mass, in terms of the higher-D mass M0 . However, in general, in

the absence of central charges, massive representations of supersymmetry are bigger
than massless ones (because there are twice as many independent supersymmetry
generators on shell, since q is a spinor with 1 helicity for the massless case, but an
SU(2) doublet for the massive). So, M 2 = Z 2 > 0 has the advantage of allowing
smaller massive representations than when M 2 > Z 2 = 0 or when M 2 > Z 2 > 0.
Note that when M 2 = Z 2 , so all masses arise from the central charge, (total) mass is
conserved, just as in nonrelativistic physics, although in the relativistic case the mass
Z can be negative. (Of course, its square is always positive, as is physical energy. The
relation between the relativistic and nonrelativistic cases can be understood through
 C. HIGHER DIMENSIONS 587

dimensional reduction: See excercise IA4.5. The mass is also a central charge for the
Galilean group, but there the reduction is for a lightlike dimension.)
In the present case, we can choose our U(1) symmetry to be a subgroup of the
extra SU(2) internal symmetry (k 0 index) of the 6D scalar multiplet. Note that the
algebra of the d’s is modified in the same way as that of the q’s.
Super Yang-Mills is a bit more interesting, because interactions are easier to
introduce. From the counting arguments given in subsection XC3, we see that a
supersymmetric theory consisting of 1 vector and 1 spinor can exist in D=3, 4, 6,
or 10. This corresponds directly with our analysis of the largest dimensions for
simple supersymmetries: Dimensional reduction of a vector gives also scalars, so the
condition of no scalars gives maximum dimensions. We now make an analysis similar
to that of the previous subsection: By dimensional analysis for physical fields, and
using single-Majorana-spinor-index notation,

{∇α , ∇β } = −Γaαβ i∇a

[∇α, ∇a] = Γaαβ W β

[∇a, ∇b ] = iFab

Applying the Jacobi (Bianchi) identities, we find

Γa(αβ Γaγ)δ = 0

This identity can be satisfied only in D=3, 4, 6, or 10. The Bianchi identities imply
the field equations for D=10.
Excercise XC5.2
Multiply the identity Γa(αβ Γaγ)δ = 0 by Γbαβ , and use the Γ matrix anticom-
(a
mutation relation Γαβ Γb)βγ = η ab δαγ to show that D=3, 4, 6, or 10.
Similar methods can be applied to D=11 supergravity. Our component counting
for general dimensions, and our helicity analysis for general extended supersymmet-
ric theories in D=4 (applied to the dimensionally reduced theory), can be satisfied
by adding to the metric (44 physical components) and gravitino (128) a third-rank
antisymmetric tensor gauge field (84) Amnp (with field strength Fmnpq = 16 ∂[m Anpq]).
The action for the graviton and gravitino are like those in 4D N=1, while A has not
only the obvious quadratic term but also a “Chern-Simons term”:

L = e−1 [− 14 R + ψ̄m γ mnp ∇nψp + 1

96
(Fabcd )2 + ψ 2 F + ψ 4]
1 mnpqrstuvwx
+ 4·3!(4!) 2 Amnp Fqrst Fuvwx
588 X. SUPERGRAVITY

(There are also more-complicated fermion interaction terms than in 4D N=1.) The
necessity of the last term can be shown by finding the component form of the super-
symmetry transformations, or by finding the field equations implied by the superspace
formulation.

6. Reduction to D=4
We now look instead at the component formulation of higher-dimensional super
Yang-Mills. This formulation is off shell except for the lack of auxiliary fields. Since
the fields are just a vector and a spinor, the Lagrangian consists of just that of super
Yang-Mills coupled to a spinor in the adjoint representation of the Yang-Mills group.
Upon dimensional reduction, the vector produces some scalars. For example, the
D=10 theory has an SO(9,1) symmetry, which reduces in D=4 to the SO(3,1)⊗SO(6)
subgroup. The SO(6) symmetry of the 6 flattened dimensions is the SU(4) symmetry
of the N=4 supersymmetries. Under this reduction, the vector becomes 10 → (4, 1) ⊕
(1, 6), namely a 4-vector and scalars that form a 6 of SU(4), while the spinor becomes
16 → (4, 4), a 4D spinor that is also a 4 of SU(4) (like the supersymmetry generators).
Although γ (or σ, or Γ) matrices are necessary in D=10, in D=4 we can convert
to spinor notation for both SO(3,1) (=SL(2,C)) and SO(6) (=SU(4)). Thus vectors
and the Minkowski metric reduce as
. ij . . .
V → (V , V );
a αα
(V )* ≡ V̄ = V ,
βα αβ αβ
(V ij )* ≡ V̄ij = 12 ijkl V kl
.
ηab → (Cαβ C . ., 14 ijkl ) : V · W → V ααWαα. + 12 V ij W ij
αβ

while spinors and Pauli matrices reduce as

 iα   
ψ iCβαV̄ij δij V .
ψ → √2
α 1 . , σαβ Va →
a αβ
ψ̄i α δji Vβ α. iC . .V ij
βα
. . .
ψV/ χ → Vαα. 12 (ψ χ̄i − χ ψ̄i ) + 12 i(V̄ij ψ χ α + V ij ψ̄i α χ̄j α.)
iα α iα α iα j

The two terms in the 10D Lagrangian then reduce as

1 2
8
F → 8
F + 18 [∇, φ̄ij ] · [∇, φij ] − 32
1 2 1
[φ̄ij , φ̄kl ][φij , φkl ]
.α .
ψ α σαβ
a β
[−i∇a, ψ ] → ψ̄i [−i∇αα., ψ iα] + 12 i(ψ iα[φ̄ij , ψαj ] + ψ̄i α [φij , ψ̄j α.])

Excercise XC6.1
Looking at the SU(3) subgroup of SU(4), decompose the states of N=4 super
Yang-Mills into those of N=3. (Use the analysis of subsection IIC5 to count
states, in SU(N) representations.) Do the same to decompose N=4 into N=2
 C. HIGHER DIMENSIONS 589

super Yang-Mills plus scalar multiplet, this time using the SU(2)⊗SU(2) sub-
group for which 4 → ( 12 , 0) ⊕ (0, 12 ) (i.e., i → (i, i0)). This is another way of
understanding where the second SU(2) of the scalar multiplet comes from.
Excercise XC6.2
Derive the commutation relations of the N=4 Yang-Mills covariant derivatives
of subsection IVC7 by dimensional reduction of those for 10D N=1 given in
the previous subsection. (Don’t forget the scalars come from the components
of the vector covariant derivative in the extra dimensions.)
Dimensional reduction of (super)gravity is an example of the comparative sim-
plicity of the vierbein (covariant derivative) formalism vs. the metric or even inverse
vierbein (differential form) formalisms. The reason in this case is that gravity is
treated like Yang-Mills theory, and gauge vectors result from reducing the graviton.
This is seen most easily from comparison of the coordinate transformation laws:

δeam = λn ∂n eam − ea n∂nλm

δem a = λn ∂n em a + ena ∂m λn

δgmn = λp ∂p gmn + gp(m ∂n) λp

Fixing the index m = −1 on λm to get the gauge transformations of an Abelian vector

resulting from reduction from one extra dimension, and setting ∂−1 = 0 when acting
on any field as the definition of reduction, we see the identification (in an appropriate
gauge for the SO(D,1)/SO(D−1,1) generators M−1a )

m −1 !
a ea m Aa
ea m →
−1 0 ψ

where A transforms in the usual way for a gauge vector, and ψ is an additional scalar.
A more transparent way to write this is as

λ ≡ λm ∂m , e a ≡ e a m ∂m ; δea = [λ, ea]

λ → λ + λ−1 ∂−1 , ea → (ea + Aa ∂−1, ψ∂−1 )

which makes it clear that reduction has simply U(1)-covariantized the gauge param-
eter, transformation, and field, where ∂−1 is the U(1) generator. (Under reduction
all fields are U(1) neutral, but nonabelian or charged fields can be obtained by more
complicated compactifications.) On the other hand, the reduction of em a, being the
inverse of ea m , and gmn , being the square of that, yields nonlinear reductions, and
590 X. SUPERGRAVITY

the U(1) covariantization is not manifest. (In particular, in the metric formalism the
metric, and thus the U(1) vector, does not even appear in the covariant derivative,
except through its derivatives.)
Excercise XC6.3
Let’s work out the details of this simple example, reduction of pure gravity
from one extra dimension:
a Find the reduction of cab c by examining the commutators of the reduced
ea. (Fab comes out directly.) Using the expression of the Lagrangian in
terms of the c’s from excercise IXA5.2, find the reduced action, including a
R
cosmological term. (Drop the dx−1 . We can think of this as compactification
on a circle, independence from x−1 yielding a constant factor upon integration,
which can be absorbed.)
b The scalar appears in a funny way, seen previously in subsection IXB5. Rather
than field redefinitions, it is more convenient to reintroduce local scale invari-
ance (after the reduction), as in subsection IXB5, introducing the dilaton φ.
Then make a simple redefinition that replaces ψ and φ with the “canonical”
fields φ±. (The F 2 and cosmological terms then appear with powers of φ± .)
Similar results can be obtained for supergravity, but the results are more com-
plicated, because the scalars (which appear for N>3) appear in nonlinear σ models.
Furthermore, although these models can be constructed by the coset method dis-
cussed in subsection IVA3, the coset space G/H is noncompact, because the group G
is noncompact, although the subgroup H is compact. This is a consequence of the fact
that the “compensating” scalars of the group H=U(N) (or SU(8) for N=8) appear
with the wrong-sign kinetic term (as the dilaton even in ordinary gravity). Thus, con-
formal supergravity is coupled to “matter” with scalars in the adjoint representation
of the noncompact group G, while gauging away the compensating scalars leaves the
physical scalars of the coset space G/H. A simpler analog is N=1 supergravity coupled
to a scalar multiplet (see subsection XB7). This is the same as conformal supergravity
coupled to the matter action φ̄φ − χ̄χ, which has a symmetry G=U(1,1), while N=1
supergravity has a gauge group U(1). Including Weyl scale invariance GL(1), the
physical scalars then inhabit the coset space U(1,1)/U(1)⊗GL(1)=SU(1,1)/GL(1).
In the case of extended supergravity, the group G can be found by noting that
the physical scalars parametrizing G/H form the representation φijkl (totally anti-
symmetric, and complex conjugate) of the group H. We then look for the group G
whose adjoint representation transforms under the H subgroup as these scalars + ad-
joint of H. We can also determine G by defining group generators for G as Mi j for H,
 C. HIGHER DIMENSIONS 591

ijkl
and Mijkl (and hermitian conjugate M ) for G/H, and write commutation relations
ijkl
consistent with covariance under H. For N=8 we also have M = 4!1 ijklmnpq Mmnpq
(and the same for the corresponding physical scalars). The result for the coset space
G/H is

N = 4 : SU(4) ⊗ SU(1, 1)/U(4) = SU(1, 1)/U(1)

5 : SU(5, 1)/U(5)
6 : SO*(12)/U(6)
8 : E7(+7)/SU(8)

where E7(+7) is a noncompact form (Wick rotation) of the exceptional group E7 .

An additional complication is that the vectors represent the full H symmetry
only on shell. For example, for N=2 we have a single vector, as in electromagnetism.
Maxwell’s equations without sources have a U(1) symmetry, “S-duality”, that trans-
forms fαβ by a phase (and f¯. . by the opposite), that mixes the field equations with
αβ
the Bianchi identities. With sources, it mixes electric and magnetic charge, since it
mixes electric and magnetic fields. So, in general we must introduce both electric
and magnetic potentials for each vector. Furthermore, for N=6 the vectors appear
ij
as both fαβ and fijklmnαβ (one extra vector). (For N=8, the two are related by an
 tensor, just as φ and φ̄.) In this version of extended supergravity, all the vectors
are Abelian. There is also a version where they gauge SO(N), but that theory has a
cosmological constant.

REFERENCES
1 Georgi, loc. cit. (IB):
Dirac and irreducible spinors for SO(n).
2 Kugo and Townsend, loc. cit. (IIC):
Dirac and irreducible spinors for SO(n+ ,n− ).
3 Siegel, loc. cit.:
spinors and γ, σ, Γ matrices in SO(n+ ,n− ).
4 L. Brink, J.H. Schwarz, and J. Scherk, Nucl. Phys. B121 (1977) 77:
D=10 super Yang-Mills.
5 E. Cremmer, B. Julia, and J. Scherk, Phys. Lett. 76B (1978) 409:
D=11 supergravity.
6 W. Nahm, Nucl. Phys. B135 (1978) 149:
arbitrary free representations of supersymmetry in arbitrary dimensions.
7 W. Siegel, Phys. Lett. 80B (1979) 220:
higher-dimensional superspace.
8 E. Cremmer and B. Julia, Phys. Lett. 80B (1978) 48, Nucl. Phys. B159 (1979) 141:
N=8 supergravity.
592 X. SUPERGRAVITY

9 L. Brink and P. Howe, Phys. Lett. 88B (1979) 268;

W. Siegel, Nucl. Phys. B177 (1981) 325:
N=8 superspace formulation of supergravity.
10 E. Cremmer and S. Ferrara, Phys. Lett. 91B (1980) 61;
L. Brink and P. Howe, Phys. Lett. 91B (1980) 384:
superspace formulation of D=11 supergravity.
11 Th. Kaluza, Sitz. Preuss. Akad. Wiss. Berlin, Math.-phys. Kl. (1921) 966;
Klein, loc. cit. (IIB);
H. Mandel, Z. Phys. 39 (1926) 136:
dimensional reduction/compactification of general relativity.
 A. SCATTERING 593

XI. STRINGS
There are three areas of application of QCD, as defined by the region of momen-
tum space they address: (1) One is perturbative QCD, which applies to large relative,
“transverse” velocity of (some of) the constituents of the hadrons. In this approach
such an amplitude is divided into a half consisting of high-energy, asymptotically-free
partons, which is calculated perturbatively in the gauge coupling, and a half con-
sisting of low-energy, confined partons, which is nonperturbative, and therefore not
calculated.
(2) Another area deals with the low-energy behavior of QCD — with properties
of the vacuum (e.g., broken chiral symmetry), or the lowest-mass hadrons, scattering
at small relative velocities. This approach is nonperturbative with respect to the
gauge coupling, and instead perturbs in derivatives, as in first-quantized JWKB. The
methods used include instantons, lattice QCD, current algebra, dispersion relations,
nonlinear σ models, and duality. This low-energy behavior really says nothing about
confinement, just as the low-energy states of the hydrogen atom tell us nothing about
ionization.
A closely related problem is that the nonperturbative information about QCD
that comes from (electromagnetic-type) duality considerations, which relates “weak”
coupling to “strong” coupling as g ↔ 1/g, is not really relating quark-gluon physics
to hadronic physics, but is relating quark-gluon physics to monopole physics; i.e., it
relates a description of weakly coupled “electric” color charges to a similar looking
theory of weakly coupled “magnetic” color charges. Thus, the dual theory, being
formally of the same type as the original, except for a relabeling of what is called
“electric” and what is called “magnetic”, does not give anything that looks any more
like hadrons, or make it any easier to calculate.
(3) The one nonperturbative approach that does deal with high (hadron) energies
is string theory: It incorporates hadrons of arbitrarily high mass, and studies their
scattering at high energies. It also shows that stringy (hadron-like) behavior is a
characteristic of QCD coupling g ≈ 1, while g ≈ 0 or ∞ have non-stringy (parton-
like) behavior: String perturbation expands in G = ln g, not g nor 1/g; duality is the
symmetry G ↔ −G. Furthermore, this G is the coupling that defines the free string,
i.e., how partons bind to form strings. The coupling that determines how hadrons
couple to each other is 1/Nc , as described topologically in subsection VC9. (However,
the relation of duality to the 1/Nc expansion is unclear, since duality has been studied
so far only in relation to theories where the group is spontaneously broken to U(1),
594 XI. STRINGS

so effectively Nc = 1, or with respect to instantons, which are always defined for

SU(2) subgroups, so effectively Nc = 2. Thus, it has not been possible to apply such
duality arguments simultaneously with a 1/Nc analysis. Similarly, since both these
duality approaches deal only with low-energy behavior, they are difficult to relate to
confinement.)
Most of the research effort on string theory has been directed toward models with
“critical dimension” D4 (10, 11, or 26): Attempts to relate to the real world involve
choosing a solution of minimal energy with all but 4 dimensions “compactified”, and
considering perturbations about that solution. (The extra dimensions cannot be com-
pletely eliminated without losing renormalizability.) Unfortunately, the ambiguity in
choice of such a solution results in the loss of predictability, which is what unification
and renormalizability are all about.
However, historically the true usefulness of such string theories has been for the
concepts and features of field theory they have revealed: For example, supersymmetry
(sections IIC and IVC, and chapter X), the Gervais-Neveu gauge (subsection VIB4),
topological (1/N) expansion (subsection VIIC4), first-quantized BRST approach to
gauge theory (chapter XII), and certain simplifications in one-loop amplitudes were
all discovered through studies of 10- and 26-dimensional string theory, even though
they all are now understood more simply through ordinary field theory. This is due
to the fact that string theories are so complex and restrictive that they require the
most powerful techniques available. Clearly such strings are useful toy models for
learning about particle field theories, and about general properties of string theory
that might lead to generalizations to include realistic 4-dimensional string theories.
(In fact, the first paper on string theory was written in 1747 by d’Alembert, and was
the first appearance of the wave equation and the d’Alembertian. Thus, field theory,
quantum mechanics, and special relativity can trace their origins to string theory.)
In subsections IVB1 and VIIC4 we briefly discussed how hadrons are expected to
arise as strings from QCD. In this chapter we analyze the dynamics of this mecha-
nism. We begin by formulating the theory in terms of strings directly. Perturbative
calculations are performed using first-quantized path integrals. The only experimen-
tal evidence for strings is as a description of hadrons; to some extent the way that
QCD leads to strings can be understood with similar first-quantized methods, based
on random lattices.
 A. SCATTERING 595

......................... .........................
......................... A. SCATTERING .........................

String theories are the only known theories that exhibit (S-matrix) duality. Unlike
other S-matrix approaches, they provide an explicit perturbative calculational scheme,
like field theory, and string theory can be formulated as a field theory. Also like field
theory, string theory has consistency conditions at the classical and quantum levels,
related to gauge invariance and renormalizability.
The defining concept of the string is that it is a two-dimensional object: Just as
the particle is defined as a point object whose trajectory through spacetime is one-
dimensional (a worldline), the string has as its trajectory a two-dimensional surface,
the “worldsheet”. This leads to a much simpler picture of interactions: For interact-
ing particles, the geometric picture of a worldline becomes a graph, whose geometry
is singular at the interaction points. For interacting strings, we have instead a world-
sheet with nontrivial topology: sphere, disk, torus (doughnut), etc. For example, a
tree graph now looks more like a real tree, in that the branches now have thickness,
and they join smoothly to the rest of the tree. There are two types of free strings:
open (two ends) and closed (no boundary). Their worldsheets are a rectangle and a
tube (cylinder).

τ τ

σ
σ

1. Regge theory

In principle there is no difference between a fundamental state and a bound

state: We can always write an action with every state represented by an independent
field. Of course, such an action might not be renormalizable, but that seems more
of a formal distinction. A more physical one is based on the qualitative property
that bound states have radial and other excitations with related properties, while
fundamental states are more unique.
Regge theory is an approach to bound states that treats them as fundamental. A
family of states that are different excitations of the same ground state is treated as a
single entity. Although basically an approach based on fundamental properties of the
S-matrix, when combined with perturbation theory it leads directly to string theory.
596 XI. STRINGS

A quantitative definition of this concept follows from a generalization of a concept

seen in perturbative field theory. In amplitudes following from Feynman diagrams
the nature of intermediate states can be seen from the momentum-space behavior:
Single-particle states appear as poles (in the sense of complex analysis) in some mo-
mentum invariants, 1/(p2 + m2 ), where this p is the sum of some of the external
momenta, representing the momentum of the internal state. (Any tree graph is a
simple example.) Two-particle states appear as cuts in these invariants, where the
branch point represents the state where the two particles are at rest with respect to
one another, and the rest of the cut corresponds to arbitrary relative velocities. (For
example, a one-loop propagator correction has a branch point at −p2 = (m1 + m2)2
for intermediate particles of masses m1 and m2.) Similar remarks apply to other
multi-particle states. “Analytic S-matrix theory” was an attempt to formulate par-
ticle physics in terms of the S-matrix by replacing the property of locality of the
action with “maximal” analyticity of the S-matrix in momentum space. (Of course,
unitarity and Poincaré invariance can be described easily in terms of the S-matrix;
even analogs of renormalizability can be formulated in terms of certain properties of
the high-energy behavior.) Unfortunately, the most general form of such nonanalytic
behavior (poles, cuts, etc.), as discovered from analyzing Feynman diagrams, proved
to be too complicated to provide a practical method for defining a theory.
Since Poincaré invariance means that not only momentum is conserved but also
angular momentum, a natural next step was to consider the analytic behavior in that
variable as well. This behavior is seen already in nonrelativistic theories; here we will
approach the concept in a language most relevant to relativistic physics. The simplest
example of “Regge behavior” is the 4-point S-matrix; this is the relativistic analog
of a nonrelativistic particle in a potential. (We can think of an infinitely massive
second particle as producing the potential, or separate center-of-mass and relative
coordinates for two finite-mass particles.) Also, the Feynman diagrams that appear
in the nonrelativistic problem are “ladder diagrams”: The sides of the ladder represent
the two scattering particles, while the rungs represent a perturbation expansion for
the potential. It can be shown that such diagrams give the leading behavior of this
amplitude at high energies. Here the appropriate high energy limit is defined in
terms of the Mandelstam variables (see subsection IA4); by high energy we mean,
e.g., s → −∞ for fixed t.
The high-energy behavior of ladder diagrams can be shown to be of the form (in
units of an appropriate mass)

A(s, t) = kg 2 Γ[−α(t)]sα(t), α(t) = a + g 2b(t)

 A. SCATTERING 597

where a is a constant that describes the behavior of the tree graph, and b(t) is de-
termined by the one-loop graph. (Both, and the constant k, are independent of g.)
This amplitude takes a simple form under a modified type of “Sommerfeld-Watson
transform”:
I X∞
dJ
A(s, t) = J
Γ(−J)s f(J, t) = 1
J!
(−s)J f(J, t)
2πi J=0

The contour integral is taken as counterclockwise about the positive real axis to obtain
the last form, where it picks up the poles of Γ(−J) (see excercise VIIA2.3b), but can
be deformed to surround the singularities of f. In this case, a pole at J = α(t) in f
will reproduce our original ladder amplitude, while integrating it around the positive
real axis gives
g2 X
∞
g2
f(J, t) = −k ⇒ A(s, t) = k 1
(−s)J
J − α(t) J=0
J!
J − α(t)

which shows that particles of spin J contribute simple poles in t to the amplitude
at α(t) = J when α(t) can be approximated as linear near that value. The spin of
the intermediate particle follows from the sJ factor. (This is clear from examining a
4-point tree graph where the external lines are scalars and the internal line carries J
indices, and must contract the momenta of its two ends. There are also contributions
of lower spins from traces.) Thus the “Regge trajectory” α(t) determines not only
the high-energy behavior of the amplitude (for negative t), but also the spins and
masses of the bound states (for positive t): Looking at the graph for J = α(t), there
√
is a bound state of spin J and mass t whenever the curve crosses an integer value
of J. The contribution at n loops in perturbation theory to f(J, t) is a multiple pole
(J − a)−(n+1) , which contributes to A(s, t) a term proportional to sa(ln s)n .
Unfortunately, for field theories with a finite number of fundamental particles, the
trajectories are rather boring, containing only a finite number of bound states. In cer-
tain cases a trajectory may include one of the fundamental particles itself (“Reggeiza-
tion”). Because of the usual infrared divergences, such calculations can be applied
directly to S-matrix elements only for fundamental massive particles; for fundamental
massless particles, as in confining theories (like QCD), these results require external
lines to be off-shell, and some knowledge of the parton wave functions is needed.
Regge behavior thus gives a measurable definition of confinement: If the scattering
amplitudes of color-singlet states (or color-singlet channels of off-shell amplitudes of
color-nonsinglet states) have linear trajectories, the constitutent color-nonsinglet par-
ticles can be said to be “confined”. On the other hand, if the Regge trajectory rises
only to finite spin and then falls, as with the Higgs effect, then there is only “color
598 XI. STRINGS

screening”; color-singlet states might not be observable, but we do not see the infinite
number of radial excitations characteristic of confinement. Another possibility is that
arbitrarily high spin is reached at finite energy: This is characteristic of Coloumb
binding, and indicates that a new, “ionized” phase is reached above that energy.
Experimentally, hadrons are observed to have Regge behavior with respect to
both high-energy behavior and spectrum. However, those Regge trajectories are
approximately linear, thus indicating an (near) infinite number of bound states. The
linearity of the trajectories can be shown to be related to the relative stability of these
unstable particles (as compared to what is found in ladder approximations). This
suggests a formulation of the theory of hadrons where the whole Regge trajectory
is treated as fundamental. It can be shown that in any such “Regge theory” based
on a perturbation expansion where the “tree” graphs have only poles in the angular
momentum J (whose accuracy is implied by the linearity of the observed trajectories),
that the theory has a further property called (s-t) “duality”: This property states
that the amplitude can be expressed as a sum of poles in either the s or t “channel”,
rather than as a sum over both:
X Cn (t) X C̃n(s)
A(s, t) =
s − sn
=
t − tn
Σ =Σ
n n

This holds even when the sets of particles exchanged in the two channels are dif-
ferent, due to quantum numbers of the external states. This relation has also been
experimentally verified (approximately).
Explicit realizations of such “dual models” of the S-matrix in terms of first-
quantized systems are called “string theories”. They explain the linearity of the
Regge trajectories by the harmonic-oscillator structure of the string Hamiltonian,
and the duality of the amplitudes by the conformal invariance (“stretchiness”) of the
string worldsheet.

2. Classical mechanics
We now consider string theory as derived by first quantization. As for particles,
the first step is to study the classical mechanics, which determines the appropriate
set of variables, the kinetic term of the field theoretic action, some properties of
the interactions, and some techniques useful for perturbation. Just as the simplest
such action for the particle produces only the relatively uninteresting case of the
scalar, the most obvious action for the string yields a model that is not only too
simple, but quantum mechanically consistent only in 26 dimensions. However, this
 A. SCATTERING 599

toy model exhibits many relevant qualitative features, such as Regge behavior and
duality. Later we’ll consider the source of its problems by relating to four-dimensional
particle theories.
The simplest classical mechanics action for the string is a direct generalization of
that for the massless scalar particle: For the Lagrangian form of this action we write
Z 2
1 d σ√
SL = 0 −gg mn 12 (∂m X a ) · (∂nX b )ηab
α 2π
where X a (σ m ) is the position in spacetime of a point at worldsheet coordinates σ m =
(σ 0, σ 1) = (τ, σ), g mn (σ m ) is the (inverse) worldsheet metric, and α0 is a normalization
constant, the string tension. It can also be associated with the flat-space spacetime
metric ηab ; if we couple a spacetime metric, then its vacuum value can be taken as
ηab /α0 , where α0 is the gravitational coupling, as discussed in subsection IXA5.
Excercise XIA2.1
Analyze the classical mechanics of the string by approximating σ by a set of
discrete points, so X 0 (σ) → Xn+1 − Xn , etc. Choose the gauge X 0 = τ . Show
that the string then acts as a bunch of particles connected by springs, and
find all the usual spring properties: tension, speed of wave propagation, etc.
A new feature of this action (compared to the particle’s) is that it is (2D) Weyl
scale invariant. This gauge invariance can be used to gauge away one component of the
metric, in addition to the two that can be gauged away using 2D general coordinate
invariance. The net result is that the worldsheet metric can be completely gauged
away (except for some bits at boundaries), just as for the particle. However, this same
invariance prevents the addition of a worldsheet cosmological term: In the particle
case, such a term was needed to introduce mass. Here, mass is introduced through
the coefficient 1/α0 of the (∂X)2 term: The same scale invariance that prevents use
of a cosmological term also prevents this coefficient from being absorbed into the
definition of the worldsheet metric.
Just as for the particle, the metric can be eliminated by its equation of motion,
resulting in a more geometrical, but less useful, form of the action: In this case the
equation of motion (“Virasoro constraints”)

(∂m X) · (∂nX) = 12 gmn g pq (∂pX) · (∂q X)

after taking the determinant of both sides, gives

Z 2 p
1 dσ
S= 0 −g̃, g̃mn = (∂m X) · (∂n X)
α 2π
600 XI. STRINGS

This is the area of the string in terms of the “induced” metric g̃mn , analogously to the
particle case. The induced metric measures length as usually measured in spacetime:

dσ m dσ n g̃mn = (dσ m ∂m X) · (dσ n ∂nX) = (dX)2

Equivalently, this action can be written in terms of the area element dX a ∧ dX b :

Z q
1
S= − 12 (dX a ∧ dX b )2 , dX a ∧ dX b = (dσ 0∂0 X [a )(dσ 1∂1 X b] )
2πα0

For purposes of quantization, it’s also useful to have the Hamiltonian form of
the action. This also allows us to see how the Virasoro constraints generalize the
Klein-Gordon equation, and then find the BRST operator. By the usual methods of
converting from Lagrangian to Hamiltonian, we find
Z 2 √
dσ . −g 1 0 2 0−1 02 g01 0
SH = (−X · P + H), H= 2 (α P + α X )+ X ·P
2π g11 g11
where . = ∂0 and 0 = ∂1. Various combinations of components of the worldsheet
metric now appear explicitly as Lagrange multipliers. If we define

P̂(±) = √1 (α01/2 P ± α0−1/2X 0 ) ⇒ [P̂(+) , P̂(−) ] = 0

2

2
the constraints can be written as two independent sets P̂(±) .
Excercise XIA2.2
2
Show that if we call g± the Lagrange multipliers for P̂(±) , then in convenient
local Lorentz and Weyl scale (but not coordinate) gauges we can write in a
lightcone basis
e± = e±m ∂m = √12 (∂0 ± g∓ ∂1 )
while in another scale gauge we can write

dxm em± = √1 (dx0 g±

2
± dx1 )

Excercise XIA2.3
Find the (equal-τ ) commutation relations [P̂(±) , P̂(±) ]. Show that the (semi-
2
classical) commutation relations of the constraints P̂(±) close. (Hint: Use the
0 0 0
identity f(a)δ (a − b) = f(b)δ (a − b) − f (b)δ(a − b).)
Since 2D general coordinate (and even just Lorentz) invariance is no longer man-
ifest, for some purposes we need to generalize this to a form that is first-order with
respect to both τ and σ derivatives:
Z 2
1 dσ
S1 = − 0 [(∂m X) · P m + (−g)−1/2gmn 12 P m · P n ]
α 2π
 A. SCATTERING 601

obviously reproduces SL after eliminating P m . Eliminating just P 1 gives a simpler

way of deriving SH (with P 0 = α0 P ).
Since open strings have boundaries, the action implies boundary conditions, orig-
inating from integration by parts when deriving the field equations. In the last form
R
of the action variation of the first term gives, in addition to the d2 σ terms (δP ) · ∂X
R
and −(δX) · ∂P for the field equations, a boundary term dσ m mn (δX) · P n , where
dσ m is a line integral along the boundary, and the mn picks the component of P m
normal to the boundary. We thus have

nm P m = 0 at boundaries

where nm is a vector normal to the boundary. From the constraint imposed by varying
gmn , it then follows that

(tm P m )2 = 0 at boundaries

where tm is a vector tangent to the boundary (or any vector, for that matter). Since
by the field equations P m ∼ g mn ∂nX, this means that the boundary is lightlike in
spacetime: The ends of the string travel at the speed of light.

3. Gauges

In direct analogy to the particle (subsection IIIB2), the two most useful gauges
are the “conformal gauge”, defined by completely fixing the worldsheet metric, and
the lightcone gauge, which is not manifestly globally covariant but is a complete fixing
of the residual gauge invariance of the conformal gauge. In the conformal gauge we
set
gmn = ηmn

by using the 2 coordinate invariances and the 1 scale invariance to fix the 3 compo-
nents of the symmetric tensor gmn . The coordinate part of this gauge is essentially the
temporal gauge g0m = η0m , just as for the particle (−g00 = v 2 = 1). Also as for the
particle, this gauge can’t be fixed everywhere (see also subsections IIIA5 and IIIC2),
but the equation of motion from the metric is implied everywhere by imposing it at
the just the boundaries in τ . In this gauge the equations of motion for X are just the
2D Klein-Gordon equation, which is easy to solve in 2D lightcone coordinates:

∂+ ∂− X = 0 ⇒ X = X(+) (τ + σ) + X(−) (τ − σ)
602 XI. STRINGS

(We have used τ ± σ in place of σ ± for later convenience.) The constraints are then
0
2
P̂(±) ∼ (X(±) )2 = 0. This directly relates to the form of 2D conformal transformations,
which are infinite-dimensional in D=2:

ds2 = 2dσ + dσ − ⇒ σ 0+ = f(+) (σ +), σ 0− = f(−) (σ −)

The constraints are the generators of these conformal transformations. (As described
in subsection IIIA5, the constraints generate the gauge transformations; the global
transformations are those that preserve the temporal gauge.)
For the lightcone gauge, we again fix the (spacetime) +-components of the vari-
ables, and solve the +-components of the equations of motion (found by varying the
−-components). Looking at the equations of motion first, using the first-order form
of the action,
δS
0 = −m ∼ ∂m X + + (−g)−1/2 gmn P +n
P
−1/2 −1
⇒ (−g) gmn = (A · B) (mp Apnq Aq − Bm Bn ); Am = P +m , Bm = ∂m X +

(as seen, e.g., by using mnAn, Bm as a basis), and

Z
δS d
0 = − ∼ ∂m P +m
⇒ dσ P +0 = 0
X dτ
R
which identifies dσ P +0 as the conserved momentum p+ , up to a factor of 2πα0
.
(since p is really the coefficient of x in the action, where X(σ) = x + ...). Similarly,
δS/δg mn determines P −m , and thus X − .
We then choose as our main set of gauge conditions

X + = kτ, P +0 = k

for some constant k, which explicitly determines τ , and determines σ up to a function

of τ : An equivalent way to define the lightcone σ in terms of an arbitrary spacelike
coordinate σ 0 is Z σ
−1
σ=k dσ 0 P +0 (σ 0)
0

which identifies σ as the amount of momentum p+ between that value of σ and σ = 0

(at fixed τ ). We thus have that the length of the string (the range of σ, not the
physical length) is Z
l = k −1 dσ P +0 = 2πα0 p+ k −1

We then need to fix the location of σ = 0 as some function σ 0(τ ): Since in this gauge

∂1P +1 = 0
 A. SCATTERING 603

so P +1 is also a function of just τ , we further fix the gauge for σ by choosing

P +1 = 0 ⇒ (−g)−1/2gmn = ηmn

Thus the lightcone gauge is a special case of the conformal gauge, after also fixing
scale gauge g = −1. For the open string, this almost fixes σ 0 (τ ) at σ = 0, which we
can take as one boundary: The boundary condition for X + is now

0 = n · ∂X + ∼ n0

since in this (and any conformal) gauge ∂m X ∼ ηmnP n . Thus the normal to the
boundary must be in the σ direction, so the boundary is at constant σ. This means
we have one constant left to fix:

σ = 0 at one boundary (open string)

This invariance was left because all our previous gauge conditions preserved global σ
translation. Unfortunately, there is no corresponding convenient gauge choice for the
closed string, so there we leave just this one invariance. In summary, our complete
set of ligtcone gauge conditions is now:

gauge : X + = kτ, P +m = kδ0m , σ = 0 at one boundary (open string)

The lightcone action is now, in Hamiltonian form,

Z  Z i
.− + dσ h . 0 2 0−1 02
Slc = − dτ x p + X i Pi − 2 (α Pi + α Xi )
1
2π

The only distinction between open and closed strings is the boundary condition
(since closed strings by definition have no boundary). For closed strings we have only
periodicity in σ (by definition of “closed”), while for open strings we have

X 0 (τ, 0) = X 0 (τ, l) = 0

One consequence, as we just saw, is that closed strings have one residual gauge invari-
ance in the lightcone gauge. These two strings can be made to resemble each other
more closely by extending the open string to twice its length, defining X for negative
σ by
X(τ, −σ) = X(τ, σ)
This is the known as the “method of images”: X(τ, −σ) is identified with its mirror
image in the τ axis, X(τ, −σ). Then the two strings both satisfy periodic boundary
conditions, while the open string has this one additional condition. We also choose

0 + 1 (open) π
k = 2κα p , κ= 1 ⇒ l=
2 (closed) κ
604 XI. STRINGS

so the length of the closed string is 2π, while the open string has original length π that
has now been doubled to match the closed string. Our choice of “phase” in relating X
for positive and negative σ for the open string automatically enforces the boundary
condition X 0 (τ, 0) = 0 at one end of the string, while the condition X 0 (τ, π) = 0 at
the other end is implied in the same way by the closed string “boundary condition”
of periodicity, which can be written as X(τ, π) = X(τ, −π). The picture is then that
the open string is a closed string that has collapsed on itself, so that for half of the
range of σ X doubles back over the path it covered for the other half.
Because σ has a finite range, X can always be expanded in Fourier modes in that
variable; the boundary conditions slightly restrict the form of this expansion. We saw
that the equations of motion, being second-order in τ -derivatives, gave two modes
for each initial state: a left-handed one and a right-handed one. We need to be a bit
more precise about the zero-modes (independent of σ): We can separate them out as
q Z
2πα0 α0
X(τ, σ) = x + pτ + 2 [Y(+) (τ + σ) + Y(−) (τ − σ)], dσ Y(±) = 0
l
where Y contains only nonzero-modes. (The normalization of p, conjugate to x, comes
.
from the −x·p term in the Lagrangian.) Then x represents the “center of mass” of the
string, and p its total momentum. Note that this implies X(±) aren’t quite periodic:

X(±) (σ + 2π) = X(±) (σ) + 2πκα0 p

Now the periodicity boundary conditions shared by open and closed strings imply

Y(±) (σ + 2π) = Y(±) (σ)

while the extra boundary condition for the open string implies

Y(+) (σ) = Y(−) (σ) = Y (σ)

allowing us to drop the subscript in that case. Thus, the closed string has twice as
many modes as the open, except for the nonperiodic part, corresponding to the total
momentum and average position. This is related to the interpretation that the open
string is a closed string with its two halves occupying the same path. This doubling
2
also shows up in the constraints: For the closed string we have P̂(±) , while for the
2 2 2
open string we can consider just P̂(+) , since P̂(−) (σ) = P̂(+) (−σ). In the lightcone
gauge we solve these constraints for X − , by integrating
 2
.2 .i .− . √
0 = P̂(±) ∼ X (±) = X (±) − k X (±) ∼ (Y (±) + κ 2α0 p)2
2
 A. SCATTERING 605

Excercise XIA3.1
Rederive the solution to the boundary conditions for the open string without
using X(τ, −σ) = X(τ, σ) (and periodicity): The string, as originally, extends
between boundaries at 0 and π.
This separation of zero-modes from nonzero-modes also allows us to find the spin
and mass of the string: In any conformal gauge,
Z π/κ Z
1 dσ 1 . 2 02 1 X dσ . 2
2
0 =p +M = 2
(X + X ) ⇒ M = 2
Y
κα02 0 2π 2 2κα0 ± 2π (±)
Z X Z dσ
1 π/κ dσ [a . b] .
ab [a b]
J =x p +S = 0 ab
X X ⇒ S =ab
Y a(±) Y b(±)
α 0 2π ±
2π
(using the Hermitian form of the Lorentz generators, for classical purposes), where
for the open string we can replace
X Z π dσ Z 2π
dσ
→ , Y(±) → Y
± 0 2π 0 2π
For the lightcone gauge we then have the gauge condition to determine X + and
Virasoro constraints to determine X − :
q  q 2
+ 0 − 1 0
.− 1 0 i
.
1 0 i
Y(±) = 0, κα p + 2 α Y (±) = κα p + 2 α Y (±)
2κα0 p+
Excercise XIA3.2
Consider gauge fixing in the temporal gauge, replacing X + with X 0 . The
classical interpretation is now simpler, since τ and X 0 can now be identified
with the usual time. Everything is similar except that the Virasoro constraints
can’t be solved (e.g., for X 1 ) in general without square roots.
a Show that some 3D solutions (2 space, 1 time) for the open string are given
by
. .
√1 (Y 1 − iY 2 )(τ ) = ce−inτ ,
2
for nonzero integer n. (Without loss of generality, we can choose c real and
positive.) Find the mass (energy) and spin as
c c2 α0
M=√ , S 12 = = M2
α0 n n
Find X explicitly, and show it describes an “n-fold spinning rod”.
b Show that the above solution can be generalized to closed strings by using
two such Y ’s, and fixing the relative magnitude of the two c’s. Consider the
special cases where n− = ±n+. Find the explicit masses, spins, and X’s, and
show that one describes another n-fold spinning rod, while the other is an
“n-fold oscillating ring”.
606 XI. STRINGS

4. Quantum mechanics
The more interesting features of the string don’t appear until quantization. In
particular, we can already see at the free level the discrete mass spectrum character-
istic of Regge theory, or of bound states in general.
Canonical quantization is simplest in the lightcone gauge. As for particles, canon-
ical quantization is convenient only in mechanics (first quantization), not field theory
(second quantization). As can be seen from the lightcone action, the Hamiltonian
is part of the constraints: For the spinless particle, we had only the constraint
p2 + m2 = 0, which became E = H in the lightcone gauge X + = p+ τ after iden-
tifying the lightcone “energy” E = p+ p− and its Hamiltonian H = 12 (p2i + m2 ). (See
subsection IIIB2.) The string Hamiltonian can be rewritten conveniently in terms of
P̂ . Since the closed string is effectively just a doubling of the open string, we treat
the open string first. The Hamiltonian is simply
Z π
dσ 1 2
H= 2 P̂i
−π 2π

where P̂ = P̂(+) . Since we have chosen X + = 2α0 p+ τ , we have E = 2α0 p+ p− .

To identify the individual particle states, we Fourier expand the worldsheet vari-
ables in σ. As for the particle, we can work at τ = 0, since all the dynamics is
contained in the constraints. Equivalently, from the nonrelativistic view of the light-
cone formalism, we can work in the Schrödinger picture where the τ dependence is in
the wave function instead of the operators. We expand as
X
∞
√
P̂ (σ) = ãne−inσ , ã0 = 2α0 p, ã−n = ãn†
n=−∞

From the canonical commutation relations for P and X,

[P̂i (σ1), P̂j (σ2)] = −2πiδ 0(σ2 − σ1)δij

we have
[ãim, ãjn] = mδm+n,0δij
as well as the usual [pi , xj ] = −iδij , and thus can relate the modes to the usual
harmonic oscillator creation and annihilation operators:
√ √
ãn = nan , ã−n = nan† ⇒ [am, an† ] = δmn

for positive n. After normal ordering, we find for the Hamiltonian

X
∞
H = α0 p2i + N − α0 , N= nain† ain
n=1
 A. SCATTERING 607

⇒ H − E = α0 (p2a + M 2 ), M 2 = α0−1 (N − α0 )
for some constant α0 .
From the expression for the mass in terms of the number operator N, we see
that the nth oscillator ain† raises the mass-squared of the ground state |0i by n
(and similarly for multiple applications of these oscillators). For any given mass, the
highest-spin state is the symmetric, traceless tensor part of multiple ai1† ’s acting on
|0i: This describes the leading Regge trajectory, with spins

j = α0 M 2 + α0

Let’s look first at the first excited level, obtained by acting on the scalar ground
state |0i with the lowest-mass oscillators ai1† . Clearly this describes a (lightcone)
transverse vector, with no Stückelberg scalar for describing a massive vector. (I.e., it
has only D−2 components, not the D−1 necessary for a massive vector.) Thus this
state describes a massless vector, so

α0 = 1

The ground state is then a scalar tachyon with M 2 = −α0−1. For any given level
past the first excited level, one can check explicitly that the states coming from the
various oscillators include the necessary Stückelberg fields. For example, at the second
excited level, ai1† aj1 † contains a traceless, symmetric tensor and a scalar (coming from
the trace), while ai2 † is a vector; they combine to describe a massive tensor. The
proof that this works to all mass levels is closure of the Poincaré algebra quantum
mechanically: The only nontrivial commutator is [J −i, J −j ] = 0, since only J −i is
higher than quadratic (cubic) in oscillators (from the form of X − and P − after solving
constraints), so normal-ordering ambiguities lead to more than just constant terms.
For reasons to be explained in chapter XII, the algebra(ic computations) in calculating
this commutator is the same in any first-quantized theory as the first-quantized BRST
algebra. Of course, the proof of closure is already the same in principle because both
algebras are a consequence of the constraints, the conformal algebra. Thus, any
anomaly must show up in the conformal algebra itself, which will be considered in
the following subsection.
The closed string works similarly to the open, but with two sets of harmonic
oscillators, but with
p(+) = p(−) = 12 p
In that case we find
M 2 = 2α0−1 (N(+) + N(−) − 2)
608 XI. STRINGS

where N(+) and N(−) are the number operators for the two independent sets of os-
cillators. In the lightcone gauge the closed string has the residual gauge invariance
R
generated by dσ X 0 · δ/δX; this gives the residual constraint

N(+) = N(−)

The closed-string states are thus the direct product of two open-string states of the
same mass: For example, the ground state is a scalar tachyon with M 2 = −2α0−1 ,
while the first excited states are massless ones from the product of two vectors — a
scalar, an antisymmetric tensor, and a symmetric, traceless tensor. The leading Regge
trajectory consists of states created with equal numbers of ai1(+)† ’s and ai1(−) †’s, with

j = 12 α0 M 2 + 2

In summary, the leading trajectory for open or closed string is given by

1
j = κα0 M 2 +
κ

Covariant quantization of the string can be performed in several ways: One is to

use the OSp methods of chapter XII, as applied to the Lorentz generators derived
from the lightcone analysis. Another is to use the usual BRST of subsection VIA, as
applied to gravity in subsection IXB1, treating the mechanics of the string as a 2D
field theory. For the case of the conformal gauge, introducing ghosts Cm corresponding
to the gauge parameters, and antighosts B mn paired with the Lagrange multipliers
of the gauge conditions (Nakanishi-Lautrup fields), we find the ghost action
√ √
−ggmn = ηmn , δ −ggmn = ∇(mλn) − gmn g pq ∇pλq

⇒ Lg = B++ ∇−C− + B−− ∇+C+

where in the last step we have introduced a background “zweibein” for applications
such as the background field gauge, or geometries that do not admit the conformal
gauge globally, and flattened the indices on the ghosts so the tracelessness of B (which
√
follows from that of δ −ggmn ) can be solved explicitly. (The conformal-gauge Weyl
and Lorentz gauge conditions do not involve derivatives in the transformation laws
of the gauge-fixing conditions, so the ghosts are just algebraic.)
 A. SCATTERING 609

5. Anomaly

Closure of the BRST algebra, or lightcone Poincaré algebra, is nontrivial because

of the infinite summations over oscillators, or equivalently because of integration over
the two-dimensional “momentum” (which is quantized as mode number in the σ di-
rection because of the finite extent of σ). As usual, BRST invariance is equivalent
to gauge invariance, and we can check for anomalies in the usual way, now applied
to the 2D “field theory” corresponding to the mechanics of the string. Because of
conformal invariance the boundaries of the worldsheet are irrelevant; we can confor-
mally transform them to infinity; then the anomaly calculations are similar to those
applied to the Schwinger model in subsection VIIIA8. (In particular, see excercise VI-
IIA8.1.) The gauge invariances in question are coordinate invariance and local scale
invariance, whose preservation is the vanishing of the divergence and trace of the
energy-momentum tensor. As seen from our analysis for the Schwinger model, this
implies that the quantum corrections to the energy-momentum tensor must them-
selves vanish when the external fields are restricted to gravity only. The calculation
again involves one-loop propagator corrections; performed in position space, we get
the product of two propagators between the same two points, with various numbers
of derivatives acting on either end of either propagator.
The classical energy-momentum tensors can be found by various methods: A
straightforward way is to use the methods of chapter IX. A simpler method is to note
that the local Lorentz group in D=2 is Abelian (SO(1,1)=GL(1)), and the coupling of
the Lorentz connection is proportional to the “spin” (number of + minus − indices,
as used in the 2D spinor notation of subsections VIIIA7-8). This immediately tells
us the coupling of the “zweibein” from the action for terms linear in derivatives:
↔
iL = ψw ∇−ψ1−w ⇒ −iT++ = 12 ψw ∂ + ψ1−w + (w − 12 )∂+(ψw ψ1−w )

where ψw is any field with w the number of + vector indices minus − (half the number
of spinor indices). We have written the most general terms with one derivative. (It
is ∂+ to get the right Lorentz weight for T++ , but also because ∂− vanishes by the
free field equation.) The first term is the “zweibein” term, the second is the Lorentz
connection term; the coefficient of the second term is fixed by considering the cases
w = 0 (scalar, so all derivatives act on it) and 1 − w = 0. (As a check, the second
term vanishes by antisymmetry for w = 1 − w = 1/2.) For convenience we rewrite
this result as
−iT++ = wψw ∂+ ψ1−w + (w − 1)(∂+ψw )ψ1−w
610 XI. STRINGS

The same result holds for T−− from a ∇+ action. This includes the classical string
action
P+ · P− + P+ · ∇−X + P− · ∇+X

as well as the conformal-gauge ghost action

B++ ∇−C− + B−− ∇+C+

(The factor of e−1 can be absorbed into the fields by a local scale transformation; it’s
irrelevant for defining T±±.)
The calculation for the matrix element for two T++ ’s (i.e., for two h−− background
fields) is now simple: The propagators go as 1/(x − x0), and the various directions
of derivatives from the vertices give either both hitting the same propagator or one
hitting each, giving either (∂x−1)2 = x−4 or x−1 ∂ 2x−1 = 2x−4 . The matrix element
is then proportional to

[w2 + (w − 1)2 ] + 2[2w(w − 1)] = 6(w − 12 )2 − 1

2

with the usual extra minus sign in the fermionic case. (As a check, this result is
consistent with bosonization: A real, bosonic scalar gives the same result as a complex,
fermionic spinor.) Taking (spacetime dimension) D w = 0 X’s and −2 (fermionic)
w = 2 ghosts (actually one with w = 2 and the conjugate with w = −1), this adds
up to
D − 26 = 0 ⇒ D = 26

for anomaly cancelation. Thus, we have two conditions on this string that make
it unsuitable for describing mesons: unphysical intercept α0 for the leading Regge
trajectory (massless particles) and unphysical spacetime dimension D.
Excercise XIA5.1
In the derivation of the anomaly for X used for anomaly cancelation, the effect
of the P 2 term was neglected. Using the result for the h−− T++ term from
P+ · ∇−X, eliminate P± from the action, and redo the anomaly calculation
in terms of this form of SL . (Hint: Show each step in the above calculation
is still correct, even though P and X aren’t independent.)
 A. SCATTERING 611

6. Tree amplitudes

Unlike particles, Feynman diagrams for strings can be treated by first-quantized

methods for arbitrary loops. The basic idea is that interacting strings are just strings
with nontrivial geometries: For example, while an open-string propagator can be
described by a rectangle, an open-string tree graph can be described by a rectangle
that has parallel slits cut from two opposite ends of the rectangle part-way into the
interior; this describes initial strings that join and split at their ends (interactions).
This is the lightcone picture of interactions, where conservation of total p+ means
conservation of the sum of the lengths of the strings. This corresponds to the choice
k = 1 in the language of subsection XIA3, since the worldsheet coordinates must be
chosen consistently over the whole worldsheet:

X+ = τ ⇒ l = 2πα0 p+

More general conformal gauges are defined by conformal transformations of this con-
figuration: For example, the boundary of this slit rectangle can be transformed to a
single straight line by the usual methods of complex analysis, so the worldsheet be-
comes simply a half-plane. (For the infinite rectangle, relevant for asymptotic states,
P
the transformation is ρ = p+ r ln(z − Zr ).) Then even the geometry is irrelevant; all
that matters is the topology, which tells how many loops the diagram has. (See the
discussion of subsection VIIC4.)
First-quantized path integrals are then the easiest way to calculate arbitrary
S-matrix elements in string theory. However, the calculations still can be quite com-
plicated (as expected from a theory with an infinite number of one-particle states),
so here we will consider just the tree-level scattering of ground states, which is suf-
ficient to illustrate the qualitative features. We can start from a gauge where the
string is an infinite strip (a rectangle of infinite length but finite width), with all
but two of the external states associated with points on one side of the strip, the
remaining two states being at the ends at infinity. This is equivalent to a picture of
a propagator in an external field, with all but two of the external states associated
with the external field. Similar calculations are possible for particles, but give only
a single graph; for strings this gives the only graph, since different cyclic orderings
612 XI. STRINGS

are related by conformal transformations. (We are restricted to cyclic orderings by

group theory, as for the 1/N expansion for particles.) This method can be used in
either the lightcone gauge or Lorentz covariant conformal gauge. In the latter case,
we integrate over the relative position of the vertices on the end of the string; this
R∞
is equivalent to integrating 0 dτ e−τ H = 1/H for the propagators between vertices.
The result is the usual particle type expression V H −1 V...H −1 V . The ground-state
vertex of external momentum k is simply eik·X(T ,0), where the vertex is positioned at
the end of the string at σ = 0.
Since the functional integral is Gaussian, all we need to evaluate it is the two-
dimensional propagator, subject to the boundary condition X 0 = 0. As for 2D elec-
trostatic problems, we simplify the problem by a conformal transformation from the
infinite strip to the half-plane:
z = e−ρ

where ρ = τ − iσ is the complex (Wick-rotated worldsheet) coordinate for the strip.

The bottom of the strip (σ = 0, the real axis) is mapped to the positive real axis, while
the top of the strip (σ = π) is mapped to the negative real axis; the interior is mapped
to the upper-half complex plane. The 2D propagator (in worldsheet coordinate space,
needed for the explicit τ = Tr dependence of the vertices) now follows from that for
the whole plane by the method of images:

G(z, z 0 ) = πδ 2(z − z 0) ⇒ G(z, z 0 ) = 12 (ln|z − z 0 | + ln|z − z 0 *|)

(Note that we have changed normalization from our earlier discussions of general
dimensions.)
Thus, the N-point amplitude is given by the path integral

Y
N −1 Z zr−1 Z
N −2
AN = g lim
z1 →∞
dzr z12 DX e−S
zN →0 r=3 zr+1

X
N Z
d2 σ
S = −i kr · X(zr ) + 1
2 (∂X)
2

r=1
π
where we have chosen units 2α0 = 1. (So H = 12 p2 + N − 1.) Since the zr are the
values of z for the vertices, we integrate over zr from 0 to 1, but

zr > zr+1 , z1 = ∞, z2 = 1, zN = 0

While only z3, ..., zN −1 are integrated over, and vertices are inserted originally at only
z2, ..., zN −1, we have inserted two extra vertices at the ends of the string at z1 and zN
 A. SCATTERING 613

(i.e., τ = ±∞), to excite the true ground states from k = 0 to nonzero momentum
(as in ordinary quantum mechanics, where eikx |0i = |ki for P |ki = k|ki.) The extra
factor of z12 was inserted to amputate the external lines of these two vertices:
1 2 +N −1
|ki = −zeik·X(z)z 2 p |0i = −eik·X(z)|0i
1
hk| = −h0|z −( 2 p
2 +N −1)
eik·X(z) z = −h0|eik·X(z) z 2

The extra factors of −z come from the conformal transformation of this “vertex
operator”: Conformal invariance requires that dρ V (ρ) = dz V (z), and it is actually
V (ρ) = −zV (z) that creates an appropriately normalized state acting on |0i.
As usual, we perform the functional integral by completing the square,
Z Z Z Z
−1
−S = 2 X X + iJX →
1 1
2J J=21
JGJ

Here J is a density in terms of kr and a δ function; we can thus use its integration to
define it (rather then bothering with its explicit form in terms of δ functions in ρ or
z): Z Z Z
X X
J ·X = kr · X(zr ) ⇒ J · GJ = kr · ks G(zr , zs )
r r,s

(using also the fact that X and G are scalars). We drop the infinite terms coming
from connecting a vertex to itself with a propagator: This is a “renormalization” of
the vertex, called “normal ordering” in the operator approach. (It can also be treated
in a more careful way by taking the vertices to correspond to finite-width strings, as
they would in the lightcone approach, and taking the limit where their widths vanish.)
The result of the functional integral is then

Y
N −1 Z zr−1 Y
N −2
AN = g lim
z →∞
dzr z12 (zr − zs )kr ·ks
1
zN →0 r=3 zr+1 1≤r<s≤N

Y
N −1 Z zr−1 Y
N −2
=g dzr (zr − zs )kr ·ks
r=3 zr+1 2≤r<s≤N

where we have used the ground-state mass-shell condition k2 = 2 for k1 .

The simplest case is the four-point function,
Z 1
A4 = g 2
dz z −α(s)−1(1 − z)−α(t)−1
0

α(s) = 12 s + 1, s = −(k3 + k4 )2 , t = −(k2 + k3 )2

614 XI. STRINGS

which we recognize as the Beta function (see subsection VIIA2)

Γ[−α(s)]Γ[−α(t)]
A4 = g 2 B[−α(s), −α(t)] = g 2
Γ[−α(s) − α(t)]
Duality of this amplitude follows from the fact that it can be written as a sum of
poles in either the s or t channel (see excercise XIA6.1 below):
X∞  
[α(t) + j][α(t) + j − 1] · · · [α(t) + 1] 1
A4 =
j=0
j! j − α(s)

where the pole at j = α(s) has maximum spin j, as indicated by the residue of order
tj . Regge behavior is also found in the appropriate limit (see excercise XIA6.2 below):

lim A4 = g 2 Γ[−α(t)][−α(s)]α(t)
s→−∞
t f ixed

Excercise XIA6.1
Derive the pole structure of the 4-point string amplitude by Taylor expanding
the integrand of the Beta function in z.
Excercise XIA6.2
Use the Stirling approximation (see excercise VIIC2.1) to derive the Regge
limit of the 4-point string amplitude. Show that the same result can be
obtained directly from the integral (Beta function) representation of the am-
plitude, where the main contribution comes from z near 1.
Excercise XIA6.3
Show the Regge limit can be obtained from the Sommerfeld-Watson transform
of subsection XIA1. (Hint: The Beta function is a sum of Regge trajectories.)
On the other hand, in the correspsonding limit with fixed angle (i.e., fixed t/s;
again using the Stirling approximation),

lim A4 ∼ e−f (cos θ)α(s)

s→−∞
θ f ixed
   
t t −s u −s
cos θ ≈ 1 + 2 , f≈ ln + ln
s −s t −s u
This limit corresponds to that treated in perturbative QCD, with large transverse en-
ergies. However, this Gaussian behavior in the momentum differs from the power-law
behavior (with logarithmic corrections) found in QCD (see subsection XIC3 below).
Similar methods can be used for calculating closed string diagrams: The vertices
are then located anywhere on the worldshseet instead of just on the boundary, so there
 A. SCATTERING 615

are integrals over both z and z̄. Since the closed-string Hilbert space is the direct
product of two open-string Hilbert spaces (except for momentum), for left- and right-
handed modes, the z-z̄ integrands are products of two open-string integrands. There
are also generalizations to strings with worldsheet fermions; the main differences are
supersymmetry and D=10 (instead of 26).

= =

Because of the local scale invariance of string theory, different string graphs are
distinguished by their topology rather than their geometry: Any two graphs that can
be “stretched” into one another are equivalent. In particular, in any loop graph any
hole or window can be pulled out so that it appears as a tadpole. The result is that
any graph is equivalent to a tree graph with insertions of some one-loop open- or
closed-string tadpoles. However, this does not mean that any graph constructed with
only open-string propagators and interactions can be expressed as an open-string tree
graph with tadpole insertions: The one-loop open-string graph with two “half-twists”
on the open-string propagators in the loop is equivalent to a tree graph with a closed-
string intermediate state, as can be seen by stretching the surface, or by tracing the
routes of the boundaries. (For example, drawing this graph in a psuedo-planar way, as
a flat ring with external states connected to both the inner and outer edges, pulling
the inner edge out of the plane reveals a closed string connecting the two edges.)
This phenomenon is similar to 2D bosonization: A closed string can be represented
as the “bound state” of two free open strings just as a massless scalar in D=2 can be
represented as the bound state of two free massless spinors.

REFERENCES
1 T. Regge, Nuo. Cim. 14 (1959) 951, 18 (1960) 947.
2 R. Dolen, D. Horn, and C. Schmid, Phys. Rev. Lett. 19 (1967) 402:
duality.
616 XI. STRINGS

3 G. Veneziano, Nuo. Cim. 57A (1968) 190;

M. Suzuki, unpublished:
birth of string theory (as dual models).
4 K. Bardakçi and H. Ruegg, Phys. Rev. 181 (1969) 1884;
C.J. Goebel and B. Sakita, Phys. Rev. Lett. 22 (1969) 257;
Chan H.-M. and T.S. Tsun, Phys. Lett. 28B (1969) 485;
Z. Koba and H.B. Nielsen, Nucl. Phys. B10 (1969) 633, 12 (1969) 517:
generalization of 4-point amplitude.
5 M. Virasoro, Phys. Rev. D1 (1970) 2933;
I.M. Gelfand and D.B. Fuchs, Functs. Anal. Prilozhen 2 (1968) 92:
Virasoro constraints.
6 D.J. Gross, A. Neveu, J. Scherk, and J.H. Schwarz, Phys. Lett. 31B (1970) 592;
C. Lovelace, Phys. Lett. 34B (1971) 500;
E. Cremmer and J. Scherk, Nucl. Phys. B50 (1972) 222:
string loops as 1-loop tadpole insertions.
7 Y. Nambu, Quark model and the factorization of the Veneziano amplitude, in Proc. of
International conference on Symmetries and quark model , ed. R. Chaud, Wayne State
U., June, 1969 (Gordon and Breach, 1970) p. 269;
L. Susskind, Phys. Rev. D1 (1970) 1182, Nuo. Cim. 69A (1970) 457;
H.B. Nielsen, An almost physical interpretation of the integrand of the n-point
Veneziano model, 15th international conference on high energy physics, Kiev, 1970:
dual model as string, through mechanics.
8 Y. Nambu, lectures at Copenhagen Symposium, 1970, unpublished;
O. Hara, Prog. Theo. Phys. 46 (1971) 1549;
T. Goto, Prog. Theo. Phys. 46 (1971) 1560;
H. Noskowitz, unpublished:
area action for string.
9 P. Goddard, J. Goldstone, C. Rebbi, and C.B. Thorn, Nucl. Phys. B56 (1973) 109:
lightcone quantization for string.
10 P.A. Collins and R.W. Tucker, Phys. Lett. 64B (1976) 207;
L. Brink, P. Di Vecchia, and P. Howe, Phys. Lett. 65B (1976) 471;
S. Deser and B. Zumino, Phys. Lett. 65B (1976) 369:
mechanics of string with worldsheet metric.
11 A.M. Polyakov, Phys. Lett. 103B (1981) 207:
worldsheet ghosts.
12 C. Lovelace, Phys. Lett. 34B (1971) 500:
discovery of D=26 requirement.
13 D. Friedan, Introduction to Polyakov’s string theory, in Recent advances in field theory
and statistical mechanics, eds. J.B. Zuber and R. Stora, proc. 1982 Les Houches summer
school (Elsevier, 1984) p. 839:
covariant vertex operators with ghosts.
14 S. Mandelstam, Nucl. Phys. B64 (1973) 205:
lightcone path integrals for interacting strings.
15 O. Alvarez, Nucl. Phys. B216 (1983) 125:
covariant path integrals for interacting strings.
16 M.B. Green, J.H. Schwarz, and E. Witten, Superstring theory, 2 v. (Cambridge Uni-
versity, 1987);
 A. SCATTERING 617

J. Polchinski, String theory, 2 v. (Cambridge University, 1998):

comprehensive texts on strings.
17 P.H. Frampton, Dual resonance models and string theories (World Scientific, 1986);
Siegel, loc. cit.;
M. Kaku, Introduction to superstrings and M-theory, 2nd ed. (Springer-Verlag, 1999):
other string texts.
18 S. Fubini; G. Veneziano; V. Alessandrini, D. Amati, M. Le Bellac, and D. Olive; J.H.
Schwarz; C. Rebbi; S. Mandelstam; Dual theory, ed. M. Jacob (North-Holland, 1974);
J. Scherk, Rev. Mod. Phys. 47 (1975) 123:
old string reviews.
618 XI. STRINGS

........................ ........................
........................ B. SYMMETRIES ........................

In the previous section we looked in detail at the simplest of the string models.
In this section we examine some of the general properties of string theory, shared by
all known models, but expected to apply also to more realistic strings. These features
can be used for phenomenological applications of string theory, but also may help
point to new generalizations.
When string theory is used as a unified theory of gravity and other forces, the
most interesting predictions are those for the “low-energy” (with respect to the Planck
mass) part of the theory. Although the possible low-energy limits of known string
theories have not all been explored, the indications are that there are only a few
restrictions beyond the usual field theoretic ones: (1) The dilaton counts the number
of loops. (2) The spectrum of the closed string is given by the direct product of
two open strings. (3) String theory has noncompact symmetries, “S-duality” and
“T-duality”, resulting from the amplitudes also having this direct-product structure.
All these properties survive the low-energy limit. In the supersymmetric case, the
last property follows from the first two. (However, the D=10 superstring is actually
a D=11 supermembrane nonperturbatively. Since the observed features of hadrons,
as well as qualitative arguments from QCD, indicate stringy but not membrane-
like behavior, we will abstract only the perturbative features of higher-dimensional
strings.)

1. Massless spectrum
The known string models all have massless particles. A string model with massless
particles can be applied to hadrons only if masses are given to all these states through
the Higgs mechanism or some other change in the vacuum. An alternative is to use
such a model to describe fundamental massless particles (graviton, photon, gluons,
neutrinos), although this would also require the usual Higgs of the Standard Model
for generating masses for some particles (W, Z, quarks, charged leptons, Higgs). In
particular, all known string models have a graviton, and there is no known method
whereby this graviton would gain mass, so these models seem suited only for unified
theories of gravity plus matter. For this purpose, the massive fields have little phe-
nomenological interest. They might improve high-energy behavior, but only near the
Planck scale, which is effectively unobservable. Therefore, it is necessary to analyze
the massless subsector of such string theories to find signs of fundamental strings in
nature.
 B. SYMMETRIES 619

The low-energy limits of the known string models follow simply from gauge in-
variance once the spectrum is known, and the spectrum itself follows from simple
considerations. In particular, we saw in subsection XIA4 for the simplest model that
the massless spectrum of the closed string was simply the direct product of that of
two open strings. This is true for all known models, and is expected to be a general
feature: It is a consequence of the fact that all free massless fields in D=2 propagate
to either the left or right; the left-handed modes are associated with one open-string
factor, the right-handed with the other. In general, the two factors may be from
different open-string theories.
The massless sector of the open string includes spin 1 and no higher. This is
true for the known string models, and also is expected to be a general result, since
otherwise the closed string would include massless states with spin higher than 2, for
which no consistent interacting theory is known. Spin 1/2 leads to supersymmetry,
as described in subsection XIB3 below; for now we limit ourselves to bosonic strings.
For the spin-1 particles to obtain nontrivial self-interactions (Yang-Mills), we need
to introduce internal symmetry. This is naturally introduced by associating indices
with the ends of the open string (as if they were quarks), as applied to particle
field theory in subsection VC9 (but which originated in string theory). These “Chan-
Paton factors” can also be associated with worldsheet variables that live only on string
boundaries. As in the field theory case, these indices are associated with orientation
of the boundaries (arrows) only for U groups, not for SO or USp. (Exceptional groups
cannot be described by Chan-Paton factors.)
The bosonic string contains at least the graviton, a scalar (usually going by the
misnomer of “dilaton”), and a pseudoscalar (the “axion”, described by an antisym-
metric-tensor gauge field), and can contain additional vectors and scalars (if the
open strings had scalars in addition to the vector). This analysis can be performed
covariantly, but it is simpler to use a helicity or lightcone analysis. Then the helicities
of the closed-string states are just the sums of those of the open strings: For the
product of two vectors (the minimal case),

(+1 ⊕ −1) ⊗ (+1 ⊕ −1) = +2 ⊕ 0 ⊕ 0 ⊕ −2

giving the graviton, scalar, and axion. Similarly, additional scalars for one open string
give additional vectors for the closed, while additional scalars for both open strings
give also additional scalars.
Excercise XIB1.1
Make the same analysis in terms of covariant fields, both for the fields them-
620 XI. STRINGS

selves and their gauge transformations. Note that the trace of the gravita-
tional field hab (determinant of the metric gmn ) is missing. (It’s unphysical,
and can be found from the ghost sector, as explained in chapter XII.)

2. Reality and orientation

In the case that the two open strings are the same, it is possible to restrict this
direct product to its symmetric part. This eliminates the axion, but not the scalar.
Thus, a massless scalar appears even in the simplest case.
To understand this restriction better, we examine the discrete symmetries of the
worldsheet. As in D=4, local, unitary, Poincaré invariant 2D field theories are always
CPT invariant. (In particular, CPT doesn’t switch left- and right-handed modes,
which differ in some string theories.) Thus, we can always impose invariance of the
wave function/field under the worldsheet CPT transformation

worldsheet CP T : Φ[X(σ)] = Φ† [X(π − σ)]

where parity transforms σ → π − σ to preserve σ[0, π] for the open string; for
the closed string the π is irrelevant because of periodicity and invariance under σ
translation. (As for the particle, we drop τ dependence. We have written only the
X coordinate explicitly for simplicity; similar remarks apply to other coordinates,
such as ghosts, with possible extra signs due to 2D Lorentz indices.) Hermitian
conjugation for the open string (for the closed string the field is not a matrix), instead
of just complex conjugation (for C), simply switches the internal symmetry (Chan-
Paton) factors associated with the left and right ends of the open string (matrix
transposition), as also required by parity. In particular, this implies that the matrices
associated with the Yang-Mills fields are hermitian, so the Yang-Mills group is unitary.
In addition to this reality condition, if the 2D theory is also invariant under CP
and T, it is also possible, though not necessarily required, to impose such a quantum
mechanical invariance under CP, and thus T:

worldsheet T : Φ[X(σ)] = MΦ*[X(σ)]M −1 ⇔

worldsheet CP : Φ[X(σ)] = MΦT [X(π − σ)]M −1

where the matrix “M” is the Yang-Mills group metric for the open string (we drop the
M for the closed string), either symmetric or antisymmetric depending on whether
the group is orthogonal or symplectic; without imposing T and CP the group is
just unitary. (The T condition is the usual reality condition for the particle, as
 B. SYMMETRIES 621

discussed in subsection IA5.) Thus, all the classical groups are allowed (at least
in the classical field theory). Since imposing invariance of the states (not just the
action) under CP and T makes it impossible to observe the left/right handedness of
the worldsheet, such strings are “unoriented”, as opposed to the “oriented” strings
that satisfy just the CPT condition. Thus, orientability of the surface is directly
related to orientability of its boundaries (oriented for U, unoriented for SO or USp).
Also, as in the particle field theory, unorientability allows “twisted” worldsheets that
are prohibited in the oriented case (because we can distinguish the “front” of the
worldsheet from the “back”): This allows such exotic geometries as Möbius strips
and Klein bottles. Open strings produce closed ones as bound states (open and
closed strings are parts of the same worldsheet with different boundaries); in theories
of open and closed strings, they must be both oriented or both unoriented. Since the
worldsheet-CP and -T switch lefty and righty modes, this invariance on the closed
string results in the restriction introduced earlier, keeping only the symmetric part
of the direct product.

3. Supergravity
Again focusing on just the massless spectrum, we now look at the restrictions
imposed by supersymmetry. The open string can also contain massless spin 1/2, but
only if it is related by supersymmetry to its massless spin 1, since it leads to spin
3/2 in the closed string, and massless spin 3/2 is known to be inconsistent in an
interacting theory unless related by supersymmetry to the graviton. Thus there are
two possibilities for the massless sector of each open string: (1) vectors and scalars
for an open bosonic string, or (2) vector multiplets (vectors, spinors, and scalars,
all related by some number of supersymmetries) for an open superstring. From our
analysis of subsection IIC5, there are furthermore 3 types of vector multiplets in D=4,
corresponding to N=1,2, or 4 supersymmetries.
This leads to four types of closed strings: (1) The bosonic string, from bosonic
⊗ bosonic, was discussed in subsection XIB1. (2) The “heterotic” string, comes from
bosonic ⊗ super. It thus can have N=1,2, or 4 supersymmetries. (3) The “(Type II)
superstring” comes from super ⊗ super. The total number of its supersymmetries is
the sum of those from the open strings: Depending on the type of supersymmetric
open strings used, the superstring can have N=2,3,4,5,6, or 8 (in other words, anything
greater than 1, since N=7 supersymmetry is equivalent to N=8, and 8 is the maximum
for supergravity). (4) In the super case, if the left and right open strings are the same,
we can impose symmetry as in the bosonic case (“Type I”). Then we may also include
622 XI. STRINGS

the open strings in the spectrum: This symmetrization also identifies the left and right
supersymmetries, so then N=1,2 or 4, the same for open and closed states (so they
can be consistently coupled).
The spectrum again can be analyzed by helicity: For example, for the N=1 het-
erotic string, we have

(1 ⊕ 12 ⊕ − 12 ⊕ −1) ⊗ (1 ⊕ −1) = (2 ⊕ 32 ⊕ − 32 ⊕ −2) ⊕ ( 12 ⊕ 0 ⊕ 0 ⊕ − 12 )

which is supergravity plus a scalar multiplet. As for the bosonic string, all supersym-
metric closed strings include the scalar, again coming from vector ⊗ vector.
Excercise XIB3.1
Make the same analysis for the superstring.

4. T-duality
Another symmetry of all known string models is “T-duality”. It is closely related
to the open ⊗ open structure of closed string states, and thus expected to be a general
property of string theory. We consider the simple bosonic model as an example.
Including constant background fields, working in the conformal gauge for convenience,
the Lagrangian is

L = −(∂+X m )(∂−X n )Mmn , Mmn = Gmn + Bmn

where the curved indices now refer to spacetime, Gmn is the spacetime metric, and
Bmn is an antisymmetric tensor gauge field (“axion”). Writing the action in first-order
form
L0 = −P+m ∂− X m − P−m ∂+ X m + P+m P−n M mn

where M mn is the inverse of Mmn , we vary X instead of P to solve the field equation

∂+P−m + ∂−P+m = 0 ⇒ P+m = ∂+ X̃m , P−m = −∂−X̃m

and substitute to find the “dual” Lagrangian

L00 = −(∂+X̃m )(∂−X̃n )M mn

Thus the “duality transformation” from X to X̃ is an invariance of the theory,

as long as we also transform the background:

X m → X̃m , Mmn → M mn
 B. SYMMETRIES 623

Note that in flat space (M mn = η mn), using the P equation of motion in L0 , we have

P+m = ηmn ∂+X n , P−m = ηmn ∂− X n

so duality just changes the sign of the right-handed modes (∂−X = −∂−X̃) while
leaving invariant the left-handed ones (∂+X = ∂+X̃). (The treatment of the zero-
modes is more tricky: We have ignored them by taking the background constant.)
We can see this to lowest order in the background, since M → M −1 , to lowest
order in perturbation about hMi = η, changes the sign of the field, corresponding to
the fact that their vertex operators are linear in both left- and right-handed modes
((∂+X)(∂− X)). However, in full nonlinearity, duality mixes the spacetime metric
Gmn (X) and axion Bmn (X).
This invariance can be generalized to a continuous O(D,D) symmetry by com-
bining it with (global) Lorentz transformations. The above discrete symmetry is a
kind of “parity” for this larger group: There are also “reflections” from performing
the duality on just one component of X m . The easiest way to see the full symmetry
is in the Hamiltonian formalism, where it can be made manifest: We first combine
X 0m and the canonical momentum Pm into an O(D,D) vector:

ZM = (Pm , X 0m )
 n 
0 0 δm
⇒ [ZM (1), ZN (2)] = −iδ (2 − 1)ηM N , ηM N =
δnm 0
where the O(D,D) metric ηM N is constant even in curved space. (We have abbreviated
“1” for “σ1 ”, etc.) The Virasoro constraints are then

1 MN
2η ZM ZN = 12 M M N ZM ZN = 0

where M is not only symmetric but also an element of the O(D,D) group:
 
Gmn Gmp Bpn
M MN
= = M N M = η M P (M −1 )P Q η QN
−Bmp Gpn Gmn − Bmp Gpq Bqn

If the fields are constant in only d of the D dimensions, than the symmetry is re-
duced to O(d,d); thus O(d,d) is a symmetry of the dimensionally reduced theory
with arbitrary fields.
624 XI. STRINGS

5. Dilaton
We can extend the spectrum analysis off shell: The procedure (to be justified in
chapter XII) includes the ghost and antighost (multiplets) for the vector (multiplet)
as a doublet of the ghostly Sp(2) symmetry. The direct product of vector ⊗ vector
now clearly gives a traceless symmetric tensor (graviton), the corresponding trace
(physical scalar), and an antisymmetric tensor (axion). In the direct product of the
ghosts, the Sp(2) singlet gives the trace part of the metric tensor, which is the true
dilaton. This dilaton (the determinant of the metric tensor in the nonlinear case)
is required in gravity for constructing local actions (see subsection IXA7), but does
not contain a physical degree of freedom. The physical polarizations of the graviton
are contained in the traceless (actually det = −1) part of the metric, which describes
the conformal part of gravity. The direct products involving ghosts also give Sp(2)
nonsinglets, which are the ghosts of the massless sector of the closed string. BRST
transformations (and thus gauge transformations) can also be obtained by this direct-
product procedure.
The natural coupling of background fields in the classical mechanics of the string
reflects this direct-product structure, as seen in the previous subsection. This means
that the background metric as we have defined it has as its determinant not the usual
one, but that times a power of the physical scalar: It is a physical degree of freedom.
T-duality mixes physical degrees of freedom with each other.
If we try to construct a low-energy action for the massless fields of the bosonic
string, it is not too difficult to find a scalar invariant under T-duality to act as the
R √
Lagrangian. However, it is impossible to use the usual measure dx −g because g
is not invariant under T-duality. This problem is solved by including the spacetime
dilaton field Φ(X): It couples to the string as
Z 2
d σ√
Sdil = − −g 12 r ln Φ(X)
2π

where we denote the worldsheet curvature by r(σ) (only in this subsection) to distin-
guish it from the spacetime curvature R(x). (There are also boundary contributions:
see excercise IXA7.3.) This term can also be expressed as a coupling to the world-
sheet ghosts (according to the above arguments), allowing the worldsheet metric to
be completely fixed by gauge transformations, as usual.
Since there is no X dependence of Sdil for constant dilaton field (no ∂X factors,
unlike G and B), the constant dilaton is invariant under T-duality. Furthermore, since
it couples to the worldsheet curvature, which counts the number of loops, the dilaton
 B. SYMMETRIES 625

must appear homogeneously in the classical action. The dilaton that appears as above
in the string action transforms as a density under general coordinate transformations,
allowing the construction of actions invariant under both T-duality and coordinate
transformations. The resulting spacetime action is
Z
Smassless = dx Φ( − 14 R + 24 1
H abc Habc + Λ)Φ

where Habc = 12 ∇[aBbc] is the field strength for the axion. T-duality determines
the only arbitrary coefficient, the relative weight of the and R terms. Note the
absence of the factor e−1 , which has been absorbed into the definition of Φ: The
covariant derivative acting on Φ, since it is a density that transforms as e−1/2, acts
as ∇aΦ = e−1/2eae1/2Φ.
Excercise XIB5.1
Find the field equations following from this action. Then make the field
redefinition Φ = e−1/2eφ , to find the result:
δ
⇒ (∇φ)2 + φ − 14 R + 1
24
H2 +Λ=0
δΦ
δ
⇒ ∇aHabc + 2Habc ∇aφ = 0
δBab
δ
⇒ Rab = 2∇a ∇b φ
δeam

Excercise XIB5.2
This action is in the string gauge (see subsection IXB5).
a Make the physical scalar explicit in the action by the field redefinition (Weyl
scaling: see subsection IXA7)

ea m → χeam

leaving Φ and Bmn unchanged.

b The resulting scalar action can be (off-)diagonalized by further redefinitions:
√
Noting that the known string theories are defined for D − 1 an (odd) integer
(5 or 3), write the dimension in general as (for any D > 1, n not necessarily
integer)
D = n2 + 1

Restoring the e−1 to the action, redefine

Φ = e−1/2φ(n−1)/2(n+1)
+ φ(n+1)/2(n−1)
− , χ = φ1/(n+1)
+ φ−1/(n−1)
−
626 XI. STRINGS

which also gives the scalars φ± the canonical Weyl scale weights, to obtain
R
the final result for the Lagrangian L (where S = dx e−1L)
2 1 (n+5)/(n+1) (n−5)/(n−1) 2
L = φ+ ( n2n−1 − 14 R)φ− + φ
24 +
φ− H

+Λφ(n−1)/(n+1)
+ φ(n+1)/(n−1)
−

c This redefinition is singular for D = 2 (n = 1). Fix this by making the

additional redefinition
(n±1)/2
φ± → φ±

and then taking the limit n → 1. (We can also use redefinitions equivalent in
(D−2)/4
the limit, such as φ− → φ− .) Show the result is then
1 3 −2 2
L → 14 φ+ ( ln φ− − R) + φ φ H
24 + −
+ Λφ−

We can no longer choose the gauge φ+ = 1, since it is now scale invariant,

but we can still choose φ− = 1.

6. Superdilaton
When applied to the supersymmetric cases (superstring or heterotic string), the
inclusion of ghosts in the direct-product procedure also gives the auxiliary fields.
(The dilaton itself is an auxiliary field.) For example, in the heterotic case, the
direct product of the physical parts of the vector and vector multiplet give conformal
supergravity (the supersymmetrization of the traceless part of the metric) and a
physical tensor multiplet (the supersymmetrization of the axion and scalar). On the
other hand, the ghosts of the vector multiplet form a chiral scalar superfield; its
procuct with the scalar ghost of the vector gives another chiral scalar superfield, the
compensator, containing the dilaton. (See chapter X.)
The two conditions of supersymmetry and that the dilaton must appear homoge-
neously (quadratically after an appropriate field redefinition) are now enough to fix
the form of the action (except for the nonminimal heterotic cases, where the open
string’s scalars introduce extra vector multiplets). For convenience we redefine the
chiral scalar compensator as φ → φ2/3 so that it appears quadratically in the cos-
R
mological term d4 x d2 θ φ2. Thus, by dimensional analysis φ now has scale weight
3
2
. The axial-vector field strength of the axion appears as [∇α, ∇α.]G, so G has scale
weight 2. (Gauge fields are Weyl scale invariant with curved indices for consistency
with gauge transformations; thus Hmnp has weight 0 while Habc has weight 3.) Of
course, these weights also follow from local superscale transformations, the global part
 B. SYMMETRIES 627

of which transforms fields as L2w (see subsection XA4). The only action quadratic
in the dilaton consistent with global scale and U(1) (R) invariance is then (with
implicit covariantization with respect to conformal supergravity, which makes these
invariances local)
Z  Z 
4 −1/2 2 2
S = dx d θ φ̄φG + Λ dx d θ φ + h.c.

Excercise XIB6.1
Use the methods of subsection XB6 to find all of the terms in this action
involving only bosonic fields. Compare to the bosonic string action of the
previous subsection.
Besides T-duality, string theories also have “S-duality” symmetries that are re-
alized only on the field equations, or after performing electromagnetic-type duality
transformations on the fields: If we convert G into a second, physical chiral multiplet
χ by such a duality as described in subsection XB5, the above action is converted to
Z  Z 
4 2/3 1/3 2 2
S = dx d θ (φ̄φ) (χ + χ̄) + Λ dx d θ φ + h.c.

After the redefinitions

φ2 → φ, φ2χ → χ
the first term becomes manifestly SU(1,1) invariant (see subsection XB7):
Z  Z 
4 1/3 2
S = dx d θ (φ̄χ + χ̄φ) + Λ dx d θ φ + h.c.

It is now the original T-duality that can be realized only on shell. Also, in this form
the condition that the dilaton should appear homogeneously is obscured. (Such S-
dualities were first seen in extended supergravity theories, especially when obtained by
reduction from higher dimensions, where antisymmetric tensors are often required.)
Excercise XIB6.2
Apply the results of excercise XB5.1 to include vector multiplets in the above
actions by replacing G → G̃ in the first action and performing duality trans-
formations. (The super Yang-Mills appears in the spectrum from the product
(vector ⊕ scalars) ⊗ vector multiplet in the heterotic string.) This substi-
tution is dictated by homogeneity in the dilaton, which prevents the usual
R
conformal d2 θ W 2 term. Such terms occur naturally in higher-dimensional
couplings of supergravity to super Yang-Mills.
Classical and quantum symmetries of mechanics formulations of particle and
string theories in background fields are often used to derive equations for those back-
grounds. These features are not peculiar to these theories or their formulations: They
628 XI. STRINGS

are a general feature of describing a particle/field of some (super)spin in a gauge back-

ground. These equations fall into two distinct types: (1) A supersymmetric system
in a gauge background of higher superspin generates constraints on the background,
necessary for consistently defining the coupling (see subsections IVC4 and XA1). (2)
Any gauge system in a background of the same gauge field generates field equations
for the background (see excercise VIB8.2).
For example, the classical symmetries of the superparticle always generate con-
straints on its background, but give field equations for it only if the number of super-
symmetries is enough to insure its superspin is as high as that of its background (e.g.,
10D N=1 in background super Yang-Mills or 11D N=1 in background supergravity).
Similarly, the bosonic string generates field equations for background gravity at the
quantum mechanical level because quantization is required to reveal the massless
graviton excited state contained in the string itself. On the other hand, the 10D
superstring already generates field equations for background supergravity classically,
since the ground state of the superstring (closed if boundary conditions are ignored),
the only part that is evident (semi)classically, already contains supergravity.

7. Conformal field theory

We saw in subsections VIIIA7-8 some unusual features of massless theories in

D=2. Since the mechanics of the string is mathematically equivalent to 2D field
theory (as the mechanics of the particle is to 1D field theory), we now examine such
field theories in a little more detail. In particular, since the string we studied in
section XIA possessed local Weyl scale invariance on the worldsheet, we are directed
to 2D conformal field theories coupled to 2D gravity.
Since for the most part we will be interested in free fields, quantization will be
described most easily by the path-integral method. Although 2D field theory already
looks quite different from the 1D field theory of particle mechanics, free 2D massless
fields depend on only one of the two lightcone coordinates σ ± (or are the sum of
two such terms), and hence 2D conformal field theory is similar to 1D massive field
theory. Consequently some of the features of particle mechanics or nonrelativistic
field theory, such as the commutator, can still be useful and 2D Lorentz covariant.
In particle mechanics, the (equal-time) commutator is evaluated by path-integral
methods as
h[A, B](t)i ≡ limhA(t + )B(t) − B(t + )A(t)i
→0
 B. SYMMETRIES 629

(and similarly for the anticommutator), since A and B are treated as classical func-
tions when evaluating the path integral
Z
hfi ≡ Dφ fe−iS

where now “h i” refers not to just the vacuum expectation value, but incorporates
arbitrary initial and final states through the boundary conditions, or explicit wave
functions in the path integral (see subsections VA1, XIA6). In general, this definition
of h i actually gives the time-ordered expectation value, as follows from the derivation
of subsection VA1: The ’s were introduced to enforce the appropriate ordering. For
the rest of this subsection time ordering will be implicit in expectation values.
For simplicity, we assume the worldsheet boundary is at infinity; this can be
achieved by conformal (coordinate) transformation (z = e−ρ on the Euclidean world-
sheet: see subsection XIA6), or we can look at just short-distance effects. The upper-
half plane, for the open string, can be completed into the full plane, as describes the
closed string, by extending the open string to have closed-string boundary conditions,
but with half as many fields, as described in subsection XIA3: In the following we
consider only fields that are left- or right-handed (functions on-shell of only σ + or
σ −, as X(±) or P̂(±) ), which are always defined on the whole worldsheet (or, before
conformal transformation, periodic on the cylinder).
The simplest example is the 2D “spinor”. From subsection VIIIA7, we have
−i
hψ̄α (σ m )ψα(σ 0m)i = h1i + ...
(σ − σ 0)± − i(τ − τ 0)
(not summed over α) or in operator notation simply
−i
ψ̄α(σ m )ψα (σ 0m) = + ...
(σ − σ 0)± − i(τ − τ 0)
where “...” means terms that are finite in the limit τ → τ 0, resulting from propagators
that extend not between these two fermions, but between either one and the initial
or final wave function. Using the identity (see excercise VA3.1)
i i
− = 2πδ(x)
x + i x − i
we then have the usual

{ψ̄α (σ ±), ψα(σ 0±)} = 2πδ(σ ± − σ 0±)

Normally, this would taken at equal τ , but in 2D conformal field theory, since the
fields’ time-dependence is given by their depending on just σ + or just σ −, and since
630 XI. STRINGS

even interacting string calculations in terms of these “free” (with respect to first-
quantization) 2D fields factorizes into two separate calculations for the left-handed
fields and for the right-handed fields, we generally treat just σ + or just σ − as the only
argument. (After Wick rotation, this becomes just the complex variable z or just z̄.)
This limiting procedure can be avoided when evaluating commutators of σ-inte-
grated quantities, such as conserved charges: Since such “surface” integrals in D=2
are basically contour integrals (see excercise IIA1.2c), a commutator of the form
Z 
dσ
λ(σ)ψ̄(σ), ψ(σ ) = λ(σ 0)
0
2π

(now dropping the α index) for some anticommuting function λ, which by the above
definition is the integral over the difference of two contours, becomes the integral over
a closed contour surrounding σ 0, directly picking up the contribution from the pole
in σ − σ 0 (see excercise VA3.1b).
One unusual consequence of 2D massless field theory is bosonization. In subsec-
tion VIIIA7 we saw from propagators that the usual massless scalar φ = φ(+) + φ(−)
can be used to construct fermions as ψα = e−iφ(±) , ψ̄α = eiφ(±) . From the above argu-
ments this implies the usual anticommutation relations. (Note that such commutators
are completely quantum mechanical and not merely semiclassical: The calculation of
subsection VIIIA7 involved multiloop diagrams with arbitrary numbers of propaga-
tors of scalars, whereas Poisson brackets effectively use only a single propagator of a
fundamental field.)
As a further check of this equivalence we now examine their conformal transfor-
mations. The commutator of the scalar conformal generators

T±± = 12 (∂±φ)2 = 12 (∂±φ(±) )2

with that scalar follows from a single propagator (see subsection VIIIA7):

φ(±)(σ m )φ(±) (σ 0m) = −ln[i(σ ± − σ 0±)] + ...

1
⇒ [∂±φ(±) (σ m)]φ(±) (σ 0m) = − + ...
σ± − σ 0±
⇒ [∂±φ(±) (1), φ(±) (2)] = −i2πδ(1 − 2)

(again dropping the ± on σ, using “1” for “σ1 ”, etc.), confirming that ∂φ is canonically
conjugate to φ. We then find

[T±±(1), φ(±) (2)] = −i2πδ(1 − 2)φ0(±) (2), [T∓∓, φ(±) ] = 0

 B. SYMMETRIES 631

Similarly, for the composite fermion we find

0m ) ia 0m
[∂±φ(±) (σ m)]eiaφ(±)(σ =− eiaφ(±)(σ ) + ...
σ± −σ 0±

⇒ [∂±φ(±) (1), ψ̄α(2)] = 2πδ(1 − 2)ψ̄α(2),

[∂±φ(±) (1), ψα (2)] = −2πδ(1 − 2)ψα (2)

in agreement with the identification ∂±φ(±) = ψ̄α ψα. Then the product with T gives
two factors of 1/(σ − σ 0) for the most singular term, as well as a less-singular term
from connecting a propagator to either ∂φ (but not both):
 2  0
m iaφ(σ0m ) a iaφ(σ0m ) 1 iaφ(σ0m )
1
(∂φ)2
(σ )e = − 1
e − e + ...
2 2
σ ± − σ 0± σ ± − σ 0±
where we have used
0m )
 0m
0
[ia∂φ(σ m)]eiaφ(σ = eiaφ(σ ) + O(σ ± − σ 0±)

The final result is thus:

[T±± (1), ψα(2)] = −i2πδ(2 − 1)ψα0 (2) − iπδ 0(2 − 1)ψα (2), [T∓∓, ψα] = 0

These results generalize easily to general linear exponentials for multiple scalars:
Including an indefinite metric

h0|φi(±) (σ m )φj(±) (σ 0m)|0i = −ηij [i ln(σ − σ 0)± ]

we have
0m ) 0m )
= [i(σ ± − σ 0±)]a·beia·φ(±)(σ
m m )+ib·φ
eia·φ(±)(σ ) eib·φ(±)(σ (±) (σ
+ ...

as the most singular contribution to the path integral, using the same method as in
subsection VIIIA7, where indices are raised and lowered with ηij , as usual. Further-
more, we can generalize the conformal generators with a linear term, as results from
a 12 Rµi φi term in the action:

L = 14 η ij (∇φi ) · (∇φj ) + 12 Rµi φi ⇒ T±± = 12 η ij (∂±φi(±) )(∂±φj(±) ) − µi ∂±2 φi(±)

Its commutators then generalize, for any “covariant” fields χ, as

 Z 
dσ 0
i λ(σ )T±± (σ ), χ(σ) = λ(σ)χ0(σ) + w(±) λ0 (σ)χ(σ)
0 0
2π
1
i [T±±(1), χ(2)] = δ(2 − 1)χ0 (2) + w(±) δ 0(2 − 1)χ(2)
2π
632 XI. STRINGS

where w(±) are the conformal weights: w(+) + w(−) is the usual conformal weight,
while w(+) − w(−) is the “spin”. For the fields eia·φ(±) we then find
w(±)(a) = 12 a2 + iµ · a, w(∓) = 0
In particular, for µ = 0 we see that ψα has w(±) = 12 , while φ(±) is covariant with
weight w(±) = 0.
Excercise XIB7.1
Use the result for eiaφ eibφ to derive ψ̄α ψα = ∂±φ(±) (up to an infinite constant),
defining the equal-time product as a limit, as above.
Excercise XIB7.2
The iµ · a term in w is classical, since it comes from a single propagator:
a Derive the above Lagrangian for a single scalar by starting with the nonlocal
term R(1/ )R and applying a local Weyl scale transformation, introducing
the scalar as the compensator.
b Find the classical scale weight of eiaφ from its local scale transformation.
(Hint: In deriving the local scale transformation in part a, an exponential
will be needed, so that eφ transforms homogeneously.)
The most important use of such exponentials, outside of bosonization, is in the
use of external fields (or “vertex operators”): The conformal generators (energy-
momentum tensor) have conformal weight 2; requiring that background fields preserve
conformal invariance implies that such vertex operators must have conformal weight
1 and be local on the worldsheet:
1
i [T̂±±(1), T̂±± (2)] = δ 0(2 − 1)[T̂±±(1) + T̂±±(2)]
2π
T̂±± (σ) = T±±(σ) + Wf (σ) ⇒ W f has w = 1, f(σ), W
[W f(σ 0)] = 0

⇒ f(σ) = 2πδ(σ − σ0 )W (σ0)

W
where we have assumed the conformal anomaly cancels (or ignored its contribution),
and solved for closure of the algebra perturbatively in the background.
If we write a background spacetime field Φ(X(σ m )) as a Fourier transform, then
we see that its conformal weight is proportional to k2 , the square of the external
momentum. Hence a vertex operator consisting of just a scalar field produces the
tachyonic ground state. Excited states are created by products of derivatives of X
times fields (with spacetime Lorentz indices contracted); the derivatives add to the
conformal weight, forcing k 2 to decrease in compensation, resulting in massless and
massive (m2 > 0) states.
 B. SYMMETRIES 633

8. Triality
Bosonization can also be applied to representations of groups (Lorentz or inter-
nal). In particular, to obtain the correct anticommutation relations for a fermion eia·φ
and its conjugate e−ia·φ we require a2 = 1. Their weights are then 12 ± µ · a. The
simplest choice to obtain 2n fermions from n scalars is, for ηij = δij , in a (complex)
null basis for SO(2n),

SO(2n) vector : aiV = (±1, 0, 0, ...), (0, ±1, 0, ...), ...

Klein factors should be included to make fermions using different scalars anticom-
mute (see subsection IA2). This construction follows that for the Dirac matrices in
subsection XC1: Each scalar corresponds to a 2D subspace of SO(2n), and each com-
ponent of a D-vector ai is the corresponding eigenvalue of the 2D spin. From that
construction we see that spinors should be

SO(2n) spinors : aiS,S0 = (± 12 , ± 12 , ...)

with independent ±’s, with the product of all of them equal to 1 for one Weyl spinor
and −1 for the other. (The conventions are slightly different from subsection XC1:
Now we use a representation where σ3 is diagonal, and γ−1 is chosen as the last γ.)
However, the resulting operators have the correct anticommutation relations and
conformal weights only for SO(8) (n = 4), corresponding to the lightcone symmetry
for the 10D superstring. This follows from the “triality” symmetry between the vector
and two spinors: By simply changing the scalar-field basis so that the a’s for one of
the spinors are ± the basis elements, the a’s for the vector and the other spinor take
the same form as above for the two spinors (i.e., the above expressions for the vector
and spinor a’s are permuted).
Similar triality constructions apply to lower dimensions by taking into account
supersymmetry, which also relates a vector to a spinor. In D=6, simple supersym-
metry has an internal SU(2) (R) symmetry: Thus, there is a triality relating this
SU(2) to the two SU(2)’s of the lightcone’s SO(4). In terms of these, the vector is
the ( 12 , 12 , 0) representation, while the spinors are ( 12 , 0, 12 ) and (0, 12 , 12 ). The resulting
operators are given by

aV = (± √12 , ± √12 , 0), aS = (± √12 , 0, ± √12 ), aS0 = (0, ± √12 , ± √12 )

To relate to the SO(8) results we use the vector to define the basis, yielding

aV = (±1, 0, 0), (0, ±1, 0)

634 XI. STRINGS

⇒ aS = ( 12 , 12 , ± √12 ), (− 12 , − 12 , ± √12 ); aS0 = ( 12 , − 12 , ± √12 ), (− 12 , 12 , ± √12 )

This also follows directly from the SO(8) result by dropping the third and fourth
√
scalars for the vector, and using only (1/ 2×) their sum for the spinors. (I.e., it
represents only internal symmetry.) For D=4 the construction is even simpler: Besides
the SO(2)=U(1) of the lightcone, there is a second U(1) for R symmetry. In terms of
the complex plane defined by these two quantum numbers, there is an obvious triality
for the three cube roots of 1; thus
√ √
aV = ±(1, 0), aS = ±(− 12 , 2
3
), aS0 = ±(− 12 , − 2
3
)

which again also follows from SO(8), now combining its last 3 scalars.
Excercise XIB8.1
We now extend the analogy to the construction of subsection XC1:
a Show for general SO(2n), in analogy to the Dirac γ’s, that the (integral of)
products of two vector fermions, antisymmetrized in the vector indices, act in
the same way as the group generators, by examining their commutators with
each other and with the vector and spinor operators.
b Show for the triality cases that the (anti)commutator of two representations
yields the third (supersymmetry).
These constructions can be generalized from the lightcone to manifest Lorentz
covariance by adding equal numbers of scalars of positive and negative metric (at
least one of each): Their contributions to the spinors’ operator product (power of
σ) then cancel, preserving the anticommutation relations. One of the extra scalars
of positive metric yields the two “longitudinal” spacetime directions to complete the
SO(D−2) vector and spinor representations to SO(D−1,1). The rest of the scalars
come from “ghosts”. Note that the spacetime metric is unrelated to the metric for the
scalars: The Minkowski spacetime metric comes from Wick rotation of the scalars, as
applied in subsection XC2 to the construction of subsection XC1 for Dirac spinors.
For either Euclidean or Minkowski spacetime the basis is null; the only difference is
in reality.

REFERENCES
1 Ramond; Neveu and Schwarz; loc. cit. (IIC):
almost superstrings (fermions included).
2 M.B. Green and J.H. Schwarz, Phys. Lett. 109B (1982) 444, 149B (1984) 117:
superstrings.
3 D.J. Gross, J.A. Harvey, E. Martinec, and R. Rohm, Phys. Rev. Lett. 54 (1985) 502,
Nucl. Phys. B256 (1985) 253, 267 (1986) 75:
heterotic string.
 B. SYMMETRIES 635

4 W. Siegel, Phys. Lett. 134B (1984) 318;

T.H. Buscher, Phys. Lett. 194B (1987) 59, 201B (1988) 466:
T-duality as transformation on worldsheet.
5 K. Kikkawa and M. Yamasaki, Phys. Lett. 149B (1984) 357;
N. Sakai and I. Senda, Prog. Theor. Phys. 75 (1986) 692;
V.P. Nair, A. Shapere, A. Strominger, and F. Wilczek, Nucl. Phys. B287 (1987) 402;
B. Sathiapalan, Phys. Rev. Lett. 58 (1987) 1597;
R. Dijkgraaf, E. Verlinde, and H. Verlinde, Comm. Math. Phys. 115 (1988) 649;
K.S. Narain, M.H. Sarmadi, and E. Witten, Nucl. Phys. B279 (1987) 369;
P. Ginsparg, Phys. Rev. D35 (1987) 648;
P. Ginsparg and C. Vafa, Nucl. Phys. B289 (1987) 414;
S. Cecotti, S. Ferrara, and L. Girardello, Nucl. Phys. B308 (1988) 436;
R. Brandenberger and C. Vafa, Nucl. Phys. B316 (1988) 391;
A. Giveon, E. Rabinovici, and G. Veneziano, Nucl. Phys. B322 (1989) 167;
A. Shapere and F. Wilczek, Nucl. Phys. B320 (1989) 669;
M. Dine, P. Huet, and N. Seiberg, Nucl. Phys. B322 (1989) 301;
J. Molera and B. Ovrut, Phys. Rev. D40 (1989) 1146;
K.A. Meissner and G. Veneziano, Phys. Lett. 267B (1991) 33;
A.A. Tseytlin and C. Vafa, hep-th/9109048, Nucl. Phys. B372 (1992) 443;
M. Roček and E. Verlinde, hep-th/9110053, Nucl. Phys. B373 (1992) 630;
J.H. Horne, G.T. Horowitz, and A.R. Steif, hep-th/9110065, Phys. Rev. Lett. 68 (1992)
568;
A. Sen, Phys. Lett. 271B (1991) 295;
A. Giveon and M. Roček, hep-th/9112070, Nucl. Phys. B380 (1992) 128:
T-duality as symmetry of spacetime fields.
6 Siegel, loc. cit. (IXB, 1st and 2nd papers of ref. 5):
coupling of (unphysical) dilaton to string (through ghosts).
7 E.S. Fradkin and A.A. Tseytlin, Phys. Lett. 158B (1985) 316, Nucl. Phys. B261 (1985)
1:
coupling of dilaton through worldsheet curvature.
8 T. Banks, D. Nemeschansky, and A. Sen, Nucl. Phys. B277 (1986) 67:
relation between above two couplings.
9 Fradkin and Tseytlin, loc. cit.;
C.G. Callan, D. Friedan, E.J. Martinec, and M.J. Perry, Nucl. Phys. B262 (1985) 593:
low-energy closed-string actions for massless fields.
10 S. Cecotti, S. Ferrara, and M. Villasante, Int. J. Mod. Phys. A 2 (1987) 1839:
superspace action for 4D massless part of heterotic string.
11 W. Siegel, Phys. Lett. 211B (1988) 55:
massless part of heterotic string is old-minimal supergravity coupled to tensor multiplet,
as follows from direct product of open strings.
12 W. Siegel, hep-th/9510150, Phys. Rev. D53 (1996) 3324:
massless part of string actions from direct product and dilaton homogeneity.
13 Virasoro; Gelfand and Fuchs; loc. cit. (XIA);
S. Fubini and G. Veneziano, Nuo. Com. 67A (1970) 29, Ann. Phys. 63 (1971) 12;
A. Galli, Nuo. Cim. 69A (1970) 275;
J.L. Gervais, Nucl. Phys. B21 (1970) 192;
J.-L. Gervais and B. Sakita, Nucl. Phys. B34 (1971) 477;
636 XI. STRINGS

A. Chodos and C.B. Thorn, Nucl. Phys. B72 (1974) 509:

theory of free conformal fields.
14 R. Marnelius, Nucl. Phys. B211 (1983) 14:
effect of Rφ term on Virasoro operators.
15 E. Witten, D=10 superstring theory, Fourth workshop on grand unification, University
of Pennsylvania, Philadelphia, April 21-23, 1983, eds. H.A. Weldon, P. Langacker, and
P.J. Steinhardt (Birkhäuser, 1983) p. 395:
vectors and spinors of SO(8) as different exponentials of bosons.
16 W. Siegel and B. Zwiebach, Nucl. Phys. B263 (1986) 105:
bosonization of ghosts (bosonic string).
17 D. Friedan, E. Martinec, and S. Shenker, Phys. Lett. 160B (1985) 55, Nucl. Phys. B271
(1986) 93;
V.G. Knizhnik, Phys. Lett. 160B (1985) 403:
vectors and spinors of SO(9,1) as different exponentials of bosons.
18 M. Kaku, Strings, conformal fields, and topology: an introduction (Springer-Verlag,
1991);
P. Ginsparg and G. Moore, hep-th/9304011, Lectures on 2D gravity and 2D string
theory, in Recent directions in particle theory: from superstrings and black holes to the
standard model , Proceedings of the Theoretical Advanced Study Institute in Elementary
Particle Physics (TASI 1992), Boulder, CO, June 1-26, 1992, eds. J. Harvey and J.
Polchinski (World Scientific, 1993) p. 277;
S.V. Ketov, Conformal field theory (World Scientific, 1995);
P. Di Francesco, P. Mathieu, and D. Sénéchal, Conformal field theory (Springer, 1997).
 C. LATTICES 637

............................ ............................
............................ C. LATTICES ............................

Integrals are defined as limits of sums. For some cases it can be convenient to
define quantum theories on discrete spacetimes (“lattices”), perform all calculations
there, and then take the limit of continuous spacetime. Two types of such lattices
will be considered here: (1) Physical four-dimensional spacetime can be treated as a
regular hypercubic lattice. Then the existence and uniqueness of a continuum limit
where Lorentz invariance is restored must be proven. (2) In first-quantization of
particles or strings, the worldline or worldsheet can be approximated as a random
lattice. Integration over the metric of the worldline or worldsheet is then replaced
with summation over lattices with different geometries. The continuum limit is not
required by physical criteria, but only for purposes of comparison to the theory as
defined in the continuum.

1. Spacetime lattice
The use of a regular 4D lattice for quantizing QCD has three main advantages:
(1) The lattice acts as a gauge invariant regulator for UV divergences (and, if the
lattice is finite, also IR ones). (2) Gauge fixing is no longer necessary, since the path
integral can be performed without it. (3) Nonperturbative calculations are possible,
some analytically and some numerically (if the lattice is small enough).
Gauge fields are associated with translations through the covariant derivative.
However, on a lattice, even a regular one, infinitesimal translations are no longer
possible: For example, scalar fields are defined only at vertices of the lattice. We
therefore consider covariantizing finite translations, as in subsections IIIA5 and IIIC2,
  Z x 
−km ∇m
e = P exp −i dx · A e−k·∂ = Ux,x−k e−k·∂
0
x−k

Without loss of generality, we can restrict ourselves to translations along links, from
one vertex straight to an adjacent one (keeping all coordinates but one constant),
and successive combinations of these. Then the gauge field is replaced with the group
element Ux,x−k associated with each link, where k is now any of the 4 orthonormal
basis vectors (in Euclidean space). The gauge transformation of this representation
of the gauge field follows from either the path-ordered definition or the covariant-
translation definition:
0
e−k·∇(x) = g(x)e−k·∇ g−1 (x) ⇒ 0
Ux,x−k = g(x)Ux,x−k g−1 (x − k)

Note that, while the gauge field is a group element associated with a link, the gauge
transformation is a group element associated with a vertex. Furthermore, the field
638 XI. STRINGS

strength can be associated with the product of these group elements of the links
bounding a “plaquet”:
 H  0 0
−i dx·A
Ux,x−k Ux−k,x−k−k0 Ux−k−k0 ,x−k0 Ux−k0 ,x = P e = e−k·∇ e−k ·∇ ek·∇ ek ·∇
0
≈ e[k·∇,k ·∇] ≈ 1 + ik ak 0b Fab

where we have used

1
eB eC = eB+C+ 2 [B,C]+...

(In general, there is a geometric prescription associating a scalar with a point, a vector
with a line, a second-rank antisymmetric tensor with a surface, etc.)
We now define a gauge-invariant action by looking for an expression in terms of
these group elements that approximates the usual Yang-Mills action to lowest order in
the lattice spacing, while involving the least number of factors of the group elements.
The result is:
X
S = − g12 tr (Ux,x−k Ux−k,x−k−k0 Ux−k−k0 ,x−k0 Ux−k0 ,x − 1)
plaquets
X X
a 0b
≈ − g12 tr 1
2 (ik k Fab )
2
∼ 1
g2
tr F 2(x)
plaquets x

(expanding the exponential as above to quadratic order, and noting that total commu-
tators vanish when traced). Since our fields are now represented by group elements,
we no longer need to fix the gauge to make the functional path integral well defined:
In contrast to the continuum case, where integrating a gauge-invariant action over
gauge transformations would produce an infinite factor, here such an integral at any
one point is just an integral over the group space, which is finite (for compact groups,
which have finite volume). The functional integration is now integration over U for
each link, where the range of U is the group space (which is finite, since the group is
compact).
Matter can also be introduced: Scalars are naturally associated with vertices,
just as vectors are with links, and second-rank antisymmetric tensors with plaquets.
However, fermions do not have such a natural geometric interpretation. In particular,
it has been proven (the “Nielsen-Ninomiya theorem”) that massless fermions can’t be
defined in a useful way on the lattice without “fermion doubling”: There must be a
multiple of 2D massless fermion fields for D lattice dimensions. This is closely related
to the existence of axial anomalies: The absence of an anomaly is implied by the
existence of a regularization that manifestly preserves a symmetry (in this case, chiral
symmetry as a consequence of the existence of lattice-regularized fermions). However,
 C. LATTICES 639

massless fermions can be defined as limits of massive ones (so chiral invariance is not
manifest). Alternatively, nonlocal spinor kinetic operators can be found that preserve
masslessness and chirality without doubling. (The nonlocality can be controlled, but
at the cost of a significantly more complicated action.)
In general, fermions are more difficult to integrate over, particularly when using
“numerical methods” (computers), since fermions are not numbers. In principle one
can integrate out the fermions analytically to produce functional determinants in
terms of bosonic fields, but nonlocality makes them hard to evaluate by iterative
schemes. In practice fermion loops are usually ignored (“quenched approximation”),
which corresponds to leading order in an expansion in the inverse of the number
of flavors, or the approximation of heavy quarks. The resulting accuracy of QCD
calculations for low-energy parameters (masses of light hadrons, decay constants, etc.)
is of the order of 5-10%. (Getting good numbers in nonperturbative calculations is
significantly harder than in perturbative ones. The situation is expected to improve
somewhat with the advent of faster computers.) Finding scattering amplitudes, or
other properties that involve high-mass hadrons, is presently beyond the scope of
lattice methods. However, lattice QCD is the only method so far to obtain numbers
for comparison with experiment from nonperturbative calculations with the QCD
action.
The spacetime lattice allows a direct nonperturbative analysis of confinement.
For example, consider the potential between a heavy quark-antiquark pair. The
heaviness again allows us to ignore pair creation, and to treat the quarks as static.
For simplicity, consider scalar quarks, as described by first-quantization. Since we
approximate the quarks as static, the only relevant term in the quark mechanics action
R . R
is the interaction term dτ x · A = dx · A. Taking into account the nonabelian
nature of the group, and ignoring the first-quantized path integration Dx (since x is
−S
R fixed), the factor e for the quark becomes just the path-ordered expression
assumed
P(e−i dx·A) we have been considering, while for the antiquark we get the inverse
expression. To get a gauge-invariant expression, we connect the paths at top and
bottom, since the fields will be fixed at the boundaries at t = ±∞. (Functional
integration over any gauge-field link picks out just the singlet part of the integrand,
since the integral is over the group, and nonsinglet representations can be rotated to
minus themselves by an appropriate group element, canceling the contribution.) The
result is a “Wilson loop”
 H 
tr P e−i dx·A
640 XI. STRINGS

The strong-coupling expansion is applied by expanding the functional integrand

−S
e in powers of S, which is an expansion in powers of 1/g2 , and which is also an ex-
pansion in the number of plaquets. Clearly the dominant term in this expansion is the
one with the fewest factors of S. To be nonvanishing, each link variable must appear
in a singlet combination: The function of that link, when expanded in irreducible
group representations, must include a term that is proportional to the identity. For
example, for any unitary group, this is true for the product Ui j (U −1 )k l , where the two
U’s are for the same link, and the indices are the group indices; this has the constant
piece Ui j (U −1 )j i . For the case of the Wilson loop, if we assume the simplest case
where the path is a rectangle, then we need at least a factor of S for each plaquet
enclosed by the loop, so there will a UU −1 for each link on the boundary (one factor
from the loop, one from S), as well as for each link enclosed by it (both factors from
the contribution to S from either side). The result for the path integral is then
D  H E  rt
−i dx·A −V t 1
A = tr P e ∼e ∼
g2
R
where V is the (potential) energy (S = dt(V + T ) in Euclidean space), t is the
time separation between the top and bottom of the rectangle, and r is the spatial
separation between the two sides. We thus have a linear quark-antiquark potential

V (r) ∼ (ln g)r

so the quark-antiquark pair is confined.

Unfortunately, we can get a similar result from QED, by defining the U(1) group
in terms of a phase factor (so effectively the range of group integration is 2π, defining
a “compact” group). The reason is that for this U(1) theory this strong coupling
expansion is not accurate. The approximation is better for nonabelian theories, but
the persistence of confinement has not been proven in the continuum limit (small
coupling). In fact, while the transition to deconfinement in Abelian theories has been
found at finite coupling, it has been proven that such a phenomenon can occur in the
nonabelian theory only near zero coupling. However, the perturbative properties of
the continuum theory show that this is exactly where one expects the appearance of
ambiguities in the theory (known in lattice terminology as “nonuniversality”); this
is directly related to the ambiguities in the Borel transform discussed in subsection
VIIC3.
 C. LATTICES 641

2. Worldsheet lattice
In string theory there are two spaces, the two-dimensional space of the world-
sheet, and physical spacetime. In the previous subsection we considered approximat-
ing spacetime by a lattice; in this subsection we instead approximate the worldsheet
by lattices. For the spacetime of QCD we used a regular lattice, representing the
fixed geometry of flat spacetime. In string theory we considered worldsheets of ar-
bitrary geometry, described by a worldsheet metric, so our lattices should be more
arbitrary; in fact, functional integration over the worldsheet metric must be replaced
by summation over different lattices. We saw that the topological expansion of QCD
in 1/N generated polyhedra analogous to the worldsheet, with 1/N acting as the
string coupling. We therefore identify the Feynman diagrams themselves, with faces
chosen by the 1/N expansion, as these lattices, to give a more precise correlation
between the second-quantized path integral of QCD (and other field theories) and
the first-quantized path integral of string theory.
Presently the relation between such field theories and string theory is not well
understood, and has been described only for the bosonic string. Since the bosonic
string has only the worldsheet metric and spacetime coordinates as degrees of free-
dom, it corresponds to a (N×N-matrix) scalar field theory. Since a lattice requires a
scale, while conformal invariance includes scale invariance, we must break the confor-
mal invariance of the worldsheet. The simplest coordinate-invariant yet scale-variant
property of a space is its volume, so we add a volume (area) term to the string action.
Furthermore, to describe interactions we need to include a term containing the string
coupling constant. In string theory the coupling is topological, in the sense that the
power of the coupling constant is counted by the “genus” of the worldsheet (the num-
ber of “windows” plus twice the number of “handles”). This is the “Euler number”
χ, counted by the integral of the worldsheet curvature (see excercise IXA7.3). Our
worldsheet action thus consists of the three terms
Z 2  
d σ√ 1 mn 1
S= −g 0 g 2 (∂m X) · (∂nX) + µ + (ln κ) 2 R
1
2π α
A lattice version of this action is (with χ given in subsection VIIC4)
 
1 X X X X X
S1 = 0 2 (Xj − Xk ) + µ
1 2
1 + (ln κ)  1− 1+ 1
α
hjki j j hjki J

where j are vertices of the lattice, hjki are the links, and J are the plaquets.
Excercise XIC2.1
Put the particle on a random lattice “Minkowski” worldline. (See excercise
642 XI. STRINGS

VB1.2.) Show the propagator for a massless particle, written in momentum

space, before taking the limit lattice spacing  → 0, is
2
∆=
1 − e−ip2
Show this has unphysical poles at p2 = 2πn/ for arbitrary integer n. How
do these results differ if the propagator is defined for Wick-rotated τ ?
The corresponding field theory is easily found, according to our earlier discussions,
by (1) identifying the worldsheet lattice with a position-space Feynman diagram (the
vertices of the lattice being those of the diagram, the links of the lattice being the
Feynman propagators; see subsection VC8), and (2) using the 1/N expansion to
associate the faces of the worldsheet polyhedra with the U(N) indices of the scalar
field (see subsection VIIC4).
We then can immediately identify the three terms in the string action with their
counterparts in the scalar field theory: (1) The X term gives the propagators, (2)
the area term (which counts the vertices) gives the vertex factor (coupling constant),
and (3) the curvature term (which classifies the topology) gives the 1/N factors of the
topological expansion. Thus, the three constants in the string action can be identified
with the mass, coupling, and number of colors of the scalar field theory. Explicitly,
the field theory action is
Z
dD x 0
S2 = N tr 0 D/2
( 12 φe−α /2 φ − G n1 φn )
(2πα )
where we have identified

G = e−µ , 1
N
= κ, m2 = 2
α0

and we have put an overall factor of N (associated with 1/h̄) so that G (and m2) is
fixed (rather than G times some power of N) when the 1/N expansion is performed.
(The reverse can be made true by rescaling φ.) The unusual kinetic operator
0
e−α /2
= 1
m2
(m2 − + ...)

comes from identifying the second-quantized particle propagator as it appears in the

first-quantized path integral for the string:
Z Y
dD Xj 0
e−S1 ⇒ ∆(x, y) = e−(x−y) /2α
2
A= 0 D/2
j
(2πα )

Unlike the spacetime lattice, the worldsheet lattice preserves spacetime Poincaré
symmetry, so it’s not necessary to take any limits to define a physical theory (or at
 C. LATTICES 643

least taking limits won’t improve the physical relevance of this model). This model
thus describes a stiff or lumpy string. The usual continuum-worldsheet string then can
be identified with a particular limit of this more general string. Explicit calculations
have demonstrated that this lattice regularization of the worldsheet reproduces the
results of the continuum approach. These results have been limited to spacetime
dimension ≤ 1 because of the inconsistencies introduced by the tachyon, which is
the ground state in higher dimensions. Unfortunately, this prevents study of the
more interesting properties, such as scattering amplitudes and the precise form of
the potential (we have left n arbitrary in S2 ), since it’s superrenormalizable in D≤2
regardless of its form. However, these limitations probably would not appear in a
corresponding formulation of the superstring, which has no tachyons.
An interesting feature of this model is the use of Gaussian propagators to get
rid of the usual perturbative divergences of momentum integration. Naively, one
might suspect that such field theories were completely finite. However, we know in
this case that the bosonic string does have divergences perturbatively in the string
coupling, and that there are further problems unless D=26. This demonstrates that
modifying a theory to fix problems seen in perturbation theory does not preclude the
reappearance of such difficulties nonperturbatively.

3. QCD strings
The main problem with known string theories is that they describe spacetimes
of dimension D=10. To describe physics in D=4, it is usually assumed that 6 of
the dimensions choose to “compactify” to submicroscopic dimensions, corresponding
to length scales well below the range of present experiments. Such a solution to
the classical field equations with minimal energy is chosen as the “vacuum”, about
which perturbations are performed, but nothing is known to preclude contributions to
the functional integral from other vacuua, whether 4-dimensional, 10-dimensional, or
elsewhere. Although strings have been chosen to describe quantum gravity because
of their renormalizability (finiteness), this advantage is lost after compactification,
since the arbitrariness in choice of compactification is tantamount to the loss of pre-
dictability in nonrenormalizable theories. However, there are some general features
of string theories, such as the appearance of a dilaton-like physical scalar, that have
suggested certain 4D models (such as no-scale supergravity, which is also suggested
by extended supergravity: see subsection XB7).
Furthermore, D=10 string theories have recently been discovered to be (compacti-
fications of) D=11 membrane theories in disguise, where the eleventh dimension shows
644 XI. STRINGS

up only nonperturbatively. Not only is using a formalism where not all of the dimen-
sions are manifest a technical obstacle, but the quantum mechanics of membranes
suffers from several problems, including nonrenormalizability, which is what string
theories were chosen to avoid in the first place. This suggests that D=10 strings are
nonrenormalizable at the nonperturbative level. Furthermore, even the few proper-
ties of 4D theories suggested by strings are no longer required, since compactification
from D=11 need not proceed by way of D=10.
On the other hand, renormalizabilty of theories with a finite number of fields
predicts D=4, since theories in higher dimensions are all nonrenormalizable (or have
unbounded potentials: φ3 theory). Furthermore, both experiments with hadrons
and theoretical arguments in QCD suggest the existence of an inherently 4D string
theory. (For example, the existence of a continuum limit for confining spacetime-
lattice theories requires asymptotic freedom.)
The reason 10D strings avoid the D=4 limit is clear from the worldsheet lattice
approach of the previous subsection: The partons that make up the string have Gaus-
sian propagators, and Gaussian integrals always converge. This leads to Gaussian be-
havior of fixed-angle scattering (subsection XIA6), in conflict with hadronic physics,
where power-law behavior is observed for partons with large transverse momenta, and
is a theoretical consequence of asymptotic freedom with the usual propagators. (In
fact, it is the main empirical verification of QCD.)
Since nonrelativistic first-quantization gives Gaussian propagators e−x /t, it is not
2

surprising that the simplest strings should result in partons with Gaussian propaga-
tors e−x . However, the fact that first-quantization for particles leads instead to,
2

e.g., 1/x2 propagators for massless particles in 4D position space suggests that an
analogous treatment for strings should be possible. We thus attempt to follow the
derivation of the previous subsection from parton to string, but starting with realis-
tic parton propagators. The first step is to exponentiate the propagator so that the
exponent can be identified with a first-quantized action. The easiest way, and that
most analogous to the nonrelativistic case, is to use the Schwinger parametrization
of the propagator, which follows from the appearance of the worldline metric in the
action: Z ∞
1
dτ e−τ p /2
2
1 2 =
2p 0

As we saw in subsection VC8, a Feynman diagram in a scalar field theory with

nonderivative self-interactions is then written as
Z P
0 − [τij p2ij /2−i(xi −xj )·pij ]
dxi dpij dτij e hiji
 C. LATTICES 645

In the (worldsheet) continuum limit of this expression, p becomes a worldsheet vector,

so τ must become a symmetric worldsheet tensor. Since on a regular square lattice
(“flat” worldsheet) there are two propagators per vertex (for the two independent
directions), τ must be a traceless tensor. (This also explains why τ can’t be just a
scalar.) Imposing this tracelessness through a Lagrange multiplier λ, we can write
the (Wicked rotated) continuum action as
Z 2
dσ √
S= {−iP m · ∂m X + 12 τmn (P m · P n + λgg mn ) + −g[µ + (ln κ) 12 R]}
2π
Thus τ acts as a kind of second worldsheet metric. However, since Schwinger param-
eters are positive, τmn must be positive definite, and thus a Euclidean metric. This
also implies that gmn must be Minkowskian, to be consistent with the tracelessness
condition. Note that if we set λ equal to a constant, and ignore the positivity condi-
tion on τ , then eliminating τ by the equation of motion from varying gmn reproduces
the usual string action, where we can identify α0 = µ/hλi. This indicates a possible
approximation scheme.
The two components of τ that survive this tracelessness condition correspond to
the two lightlike directions defined by gmn : If we use a “zweibein”, defined as usual by
gmn = −e(m+ en)− , to flatten the indices on τ , then the Lagrange multiplier constraint
can be solved by simply setting τ+− = 0. The action is then
Z 2
d σ√
S= −g[iP± · e∓m ∂m X + 12 τ±± P∓ · P∓ + µ + (ln κ) 12 R]
2π
Back on the lattice, this implies that the directions chosen by the propagators (links)
on which P is defined are lightlike. Thus, the matrix model defined by this theory
should have only 4-point vertices, with the four propagators coming from any vertex
forming the worldsheet lightcone at that point on the worldsheet. The field theory
action is thus Z
dD x
S2 = N tr (− 1 φ φ − G 14 φ4 )
(2π)D/2 4
For D = 4, this action describes an asymptotically free theory, “wrong-sign” φ4
theory.
Unlike conventional strings, the QCD string has critical dimension D=4 for renor-
malizability. (In conventional strings all momentum integrals are Gaussian and thus
converge.) Another reason for D=4 is T-duality: T-duality interchanges the positions
of the vertices with the momenta of the loops. This is clear from our discussion of
the classical mechanics of Feynman diagrams in subsection VC8, if we note that the
procedure we used there to translate from coordinates to loop momenta is exactly the
646 XI. STRINGS

random lattice version of the T-duality transformation performed in subsection XIB4

(with X̃ as the loop momenta). Thus, invariance of a string theory under T-duality
must include invariance of the propagators of the underlying field theory under Fourier
transformation. This is trivial for conventional strings, since the Fourier transform
of a Gaussian is a Gaussian. However, by dimensional analysis (or explicit evalua-
tion: see excercise VIIB4.2), we see that the Fourier transform of 1/p2 is 1/x2 only
in D=4: T-duality implies both D=4 and masslessness. Furthermore, we can look at
interactions by considering the simplest case: The flat worldsheet is represented by a
regular, flat lattice. For φ4 theory we have the usual square lattice, which is self-dual
under switching vertices with loops (T-duality). On the other hand, triangular and
hexagonal lattices, corresponding to φ6 and φ3 theory, are dual to each other (i.e., φn
is dual to φ2n/(n−2), as follows from geometry). Thus T-duality also implies the φ4
interaction.
Excercise XIC3.1
Let’s examine T-duality for the random lattice more carefully:
a Repeat the T-duality transformation of subsection XIB4, but for the QCD
string (see subsection VC8), without a background (M mn = η mn ). Show that
invariance under X m → X̃m requires that the matrix τ also be replaced by
its inverse, with some factors of the 2D  tensor. (λ also transforms; you can
avoid this complication by using the zweibein form of the action.)
b Write the massless scalar propagator in momentum space of arbitrary dimen-
sion D as an exponential using a Schwinger parameter τ . Show that after
T-duality — Fourier transformation combined with τ → 1/τ (which leaves
the exponent invariant) — a τ -dependent “measure” factor is introduced,
except for D=4.

REFERENCES
1 K.G. Wilson, Phys. Rev. D10 (1974) 2445:
QCD lattice.
2 H.B. Nielsen and M. Ninomiya, Nucl. Phys. B185 (1981) 20, 195 (1982) 541, 193
(1981) 173, Phys. Lett. 105B (1981) 219.
3 P.H. Ginsparg and K.G. Wilson, Phys. Rev. D25 (1982) 2649;
P. Hasenfratz, hep-lat/9709110, Nucl. Phys. Proc. Suppl. 63A-C (1998) 53,
hep-lat/9802007, Nucl. Phys. B525 (1998) 401, P. Hasenfratz, V. Laliena, and F. Nie-
dermayer, hep-lat/9801021, Phys. Lett. 427B (1998) 125;
H. Neuberger, hep-lat/9707022, Phys. Lett. 417B (1998) 141, hep-lat/9801031, Phys.
Lett. 427B (1998) 353;
M. Lüscher, hep-lat/9802011, Phys. Lett. 428B (1998) 342:
avoiding Nielsen-Ninomiya, while preserving chiral invariance.
 C. LATTICES 647

4 M. Creutz, Phys. Rev. D21 (1980) 2308;

E. Marinari, G. Parisi, and C. Rebbi, Phys. Rev. Lett. 47 (1981) 1795:
computer calculations with lattice QCD.
5 E.T. Tomboulis, Phys. Rev. D25 (1982) 606, Phys. Rev. Lett. 50 (1983) 885:
deconfinement can occur only at zero coupling.
6 ’t Hooft, loc. cit. (VIIC, ref. 3):
convergence of singularities in the complex coupling plane at zero coupling.
7 H.J. Rothe, Lattice gauge theories, an introduction (World Scientific, 1992).
8 H.B. Nielsen and P. Olesen, Phys. Lett. 32B (1970) 203;
D.B. Fairlie and H.B. Nielsen, Nucl. Phys. B20 (1970) 637;
B. Sakita and M.A. Virasoro, Phys. Rev. Lett. 24 (1970) 1146:
worldsheet lattice as Feynman diagrams.
9 F. David, Nucl. Phys. B257 [FS14] (1985) 543;
V.A. Kazakov, I.K. Kostov and A.A. Migdal, Phys. Lett. 157B (1985) 295:
integration over worldsheet metric as sum over Feynman diagrams.
10 M.R. Douglas and S.H. Shenker, Nucl. Phys. B335 (1990) 635;
D.J. Gross and A.A. Migdal, Phys. Rev. Lett. 64 (1990) 127;
E. Brézin and V.A. Kazakov, Phys. Lett. 236B (1990) 144:
continuum limit for worldsheet lattice with metric and 1/N expansion.
11 Veneziano, loc. cit. (XIA);
V. Alessandrini, D. Amati, and B. Morel, Nuo. Cim. 7A (1971) 797;
D.J. Gross and P.F. Mende, Phys. Lett. 197B (1987) 129, Nucl. Phys. B303 (1988)
407;
D.J. Gross and J.L. Mañes, Nucl. Phys. B326 (1989) 73:
Gaussian behavior of string amplitudes.
12 P.A. Collins and R.W. Tucker, Nucl. Phys. B112 (1976) 150;
B. de Wit, M. Lüscher, and H. Nicolai, Nucl. Phys. B305 (1988) 545:
problems quantizing membranes.
13 C.M. Hull and P.K. Townsend, hep-th/9410167, Nucl. Phys. B438 (1995) 109:
the 11th dimension arises nonperturbatively in string theory.
14 Siegel, loc. cit. (VC):
QCD string with Schwinger parameters as second worldsheet metric.
15 A.M. Polyakov, Nucl. Phys. B268 (1986) 406:
two-metric formulation of usual string.
648 XII. MECHANICS

XII. MECHANICS
String theories describe particles of arbitrarily large spins: So far in this text we
have concentrated on lower spins, but we can describe (at least) free gauge-invariant
actions for arbitrary spins based on quantum mechanical BRST.
Gauge invariance is required in field theory to manifest Lorentz invariance. The
basic problem is that a four-vector wave function cannot have the obvious Minkowski
inner product, since the time component would have a minus sign in its normalization,
resulting in negative probability. In the classical action there is a gauge invariance
that allows the time component to be dropped from the action. However, such gauges
destroy manifest Lorentz invariance, since a three-vector cannot represent Lorentz
transformations in a local way. More useful gauges keep all components of the four-
vector, while also introducing scalar fermionic “ghosts” to cancel the effects of the
bad part of the four-vector. A certain symmetry between the bosonic and fermionic
unphysical degrees of freedom is needed to enforce this cancelation: It is the field
theoretic version of the BRST symmetry discussed in section VIA.
Another complication is that gauge transformations do not allow the elimination
of traces in a simple way: Although it is Lorentz covariant to constrain a tensor
to vanish when a pair of its vector indices is contracted, this interferes with gauge
invariance in interacting theories, such as gravity. A related complication is massive
theories, which can’t always be described simply by adding mass terms to massless
theories.
There is a simple solution to all these problems, which determines the free part
of the action for any theory. (Interactions are a separate problem.) This method
automatically introduces all the correct fields, including ghosts, for any massless or
massive theory. It also gives a simple universal expression for the BRST symmetry
that cancels unphysical modes, as well as providing a simple proof that these modes
disappear in the lightcone gauge. The method is based on the idea of introducing
extra fermionic dimensions to spacetime that are unphysical (unlike superspace for
supersymmetry), which cancel unphysical degrees of freedom associated with the time
dimension.
Although for most purposes the only spins of fundamental particles relevant in
field theory for are 0, 1/2, 1, 3/2 (maybe), and 2, and these few cases can be studied
separately, in this chapter we’ll analyze all free theories because: (1) The ultimate
theory of particles may require them; (2) some of the theories presently under most ac-
tive investigation (such as strings and membranes) require them; (3) many observed,
 A. OSp(1,1|2) 649

though perhaps not fundamental, particles have higher spin; and (4) a better under-
standing of field theory can be obtained by determining exactly which properties all
fields have in common as well as how they differ.

............................. .............................
............................. A. OSp(1,1 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
This construction involves the introduction of spacetime symmetries that are not
manifest on the physical coordinates. An important analog is the conformal group in
D dimensions, which acts nonlinearly on the usual D spacetime coordinates, but can
be represented linearly on D+2 coordinates, since the group is SO(D,2). As described
in subsection IA6 for spin 0, 1 space and 1 time coordinate can be eliminated, so that
SO(D,2) is still represented, but SO(D−1,1) is the largest orthogonal group that is
still manifest. We have also seen that in the lightcone gauge this manifest symmetry
is reduced again in the same way, leaving SO(D−2). In our case the relevant group is
OSp(D,2|2), the natural generalization of the orthogonal group to D space, 2 time, and
2 anticommuting dimensions. This allows rotations between timelike and fermionic
directions, eventually resulting in their cancelation.

1. Lightcone

We saw in subsection IIB3 how the single equation of motion Sa b ∂b + w∂a = 0,

applied to field strengths, universally described all spins in all dimensions, for free,
massless particles. (A possible exception is the spinless case, where we need = 0,
which is redundant otherwise. However, we can use the universal field equation even
in that case if we use the vector field strength formulation of spin 0.) One way we
solved this equation was to perform a unitary transformation. We can use the same
unitary transformation, plus the constraints, to simplify the Lorentz generators. To
further simplify matters, we can use the constraint = 0, solved for ∂ −, to choose
the gauge x+ = 0, which is equivalent to working in the Schrödinger picture (no
time dependence for operators). The procedure is thus: (1) Start with the manifest
(antihermitian) representation of the Lorentz generators,

J ab = x[a ∂ b] + S ab

(2) Apply the transformation

∂i
J → UJU −1 , ln U = S +i
∂+
650 XII. MECHANICS

which eliminates the only S +i term, in J +i (while complicating J −i ). (3) Finally, apply
the constraints, which have already been transformed by this same transformation,

= 0 (⇒ gauge x+ = 0), S ab ∂b + w∂ a = 0 → S +− − w = S i− = 0

Our Lorentz generators are then

J +i = −xi ∂ +, J +− = −x−∂ + + w, J ij = x[i ∂ j] + S ij

1 1 i j 2
J i− = −x−∂ i + +
[ 2 x (∂ ) + S ij ∂j + w∂ i ]
∂
These generators satisfy the pseudo(anti)hermiticity condition

J ab† (∂ +)1−2w = −(∂ +)1−2w J ab

This means that the Hilbert-space metric needs a factor of (∂ +)1−2w . This is related
to the fact that the w terms can be eliminated by a nonunitary transformation with
the appropriate power of ∂ +: As part of step 2, we could have applied a second
transformation
−+
U2 = (∂ +)S

with the result of eliminating all S +− terms, so the redundant constraint S +− = w

would not have been needed, so w would not appear. In any case, we generally choose
w = 0 for bosons.
We previously applied dimensional reduction to the field equations for the field
strengths, to obtain the equations for the massive, free theories from the massless
ones. The same methods can be applied to the Lorentz (or Poincaré) generators.
(The Lorentz generators will be used later to find the BRST operator, to obtain the
field equations in terms of the gauge fields, and the action. For that purpose the
dimensional reduction can be performed at any stage in the derivation.) We thus find
the general result

J +i = −xi ∂ +, J +− = −x−∂ + + w, J ij = x[i ∂ j] + S ij

1 1 i j 2
J i− = −x−∂ i + +
{ 2 x [(∂ ) − m2] + S ij ∂j + S i−1 im + w∂ i}
∂
(The −1 is still an index, and should not be confused with an inverse.)
Note that S −i and S −+ were eliminated (after the unitary transformation) by the
constraints, and that S +i just dropped out. (In other words, S +i = 0 was the gauge
choice for the constraint S −i = 0.) This leaves only S ij (and S i−1 in the massive case),
whose representation is that of the highest-weight part of the original field strength.
 A. OSp(1,1|2) 651

However, we can more simply choose the representation of S ij as our starting point,
since it is just the transverse part of the gauge field; it defines the representation of the
Poincaré group. We therefore have an explicit construction of the generators of the
Poincaré group, for arbitrary representations, defined on just the physical degrees of
freedom, given directly by the little group SO(D−2) spin generators S ij (or SO(D−1)
generators S ij and S i−1 in the massive case) that identify the representation. For
example, in D=4, SO(2) has just one generator, the helicity, so for any state of a
given helicity we know the action of the Poincaré generators.
Excercise XIIA1.1
Check that this lightcone representation of the Lorentz generators satisfies
the correct commutation relations.
Classical free field theory is easy to define in the lightcone, since solving the
constraints in the lightcone formalism has picked out just the physical components,
so the only remaining constraint is the Klein-Gordon equation. Thus, the kinetic term
for any massless bosonic field is simply − 12 φ( 12 )φ, where 12 = −∂ +∂ − + 12 (∂ i )2,
and ∂ − is considered the time derivative. (In general, the kinetic operator for a
massless boson is some second-order differential operator, which reduces to on the
physical components.) For fermions we have instead /∂ +, since we must then have
an odd number of derivatives to avoid getting a trivial result after integration by
parts. (For a boson, φ∂φ = ∂( 12 φ2 ), for a fermion ψ∂∂ψ = ∂(ψ∂ψ) − (∂ψ)2 = ∂(ψ∂ψ)
by anticommutativity. In general, the kinetic operator for a massless fermion is some
first-order differential operator, which reduces to /∂ + after eliminating auxiliary
fields.)
This quantum mechanical representation of the Lorentz generators has a simple
translation into classical field theory, in terms of field theory Poisson brackets. The
definition of Poisson brackets in lightcone quantum field theory follows directly from
the action: Defining as usual the canonical momentum π as (minus, in our conven-
tions) the variation of the Lagrangian with respect to the time derivative ∂+ (= −∂ −)
of the variable φ, we find the fundamental bracket for bosons

1
π = −∂ +φ ⇒ [φ(x−, xi), φ(x0−, x0i)] = i +
(2π)D/2δ(x− − x0−)δ D−2 (xi − x0i )
∂

(Note that the ∂ + was essential for the antisymmetry of the bracket. We can also
evaluate its inverse as an integral, so ∂1+ δ(x− − x0−) = 12 (x− − x0−). For fermions
we have instead an anticommutator and no 1/∂ + .) We then find that any quantum
mechanical group generator J (including internal symmetries) can be represented in
652 XII. MECHANICS

field theoretic form as

Z
dx− dD−2 xi 1 +
J = D/2 2 φ∂ Jφ = i 2 hφ|Jφi
1
(2π)
where we have used the relativistic inner product of subsection VB2, but for a lightlike
hypersurface: For positive-energy solutions
Z
dx−dD−2 xi ↔
h1|2i = D/2
ψ 1 * 1
2 (−i) ∂ +
ψ2
(2π)
Note that the free Poincaré generators are local in this form, from cancelation of
∂ +’s. In interacting theories, the generator J −i , as well as the translation generator
P −, which is also the Hamiltonian, gets additional terms higher-order in the fields.
In this manner, relativistic quantum field theory can be quantized in a way that
more resembles nonrelativistic field theory than in non-lightcone methods, since
is quadratic in the usual time derivative ∂0 . We won’t consider lightcone quantum
field theory further; however, in the following sections we’ll use this construction to
derive free gauge theory and its covariant quantization, in a way that we’ll generalize
straightforwardly to interactions. Thus, the same construction directly gives the
formulation of free representations of the Poincaré group, from field strengths to
transverse fields to covariant gauge fields.

2. Algebra
From the definition of the graded determinant in terms of Gaussian integrals (see
subsection IIC3), we see that anticommuting coordinates act like negative dimen-
sions: For example, sdet(kI) = k a−b for a commuting and b anticommuting dimen-
sions. Thus, if we add equal numbers of commuting and anticommuting dimensions,
they effectively cancel. Here we’ll do the same for theories with spin, which allows
the restoration of manifest Lorentz covariance to lightcone theories: Adding 2 com-
muting and 2 anticommuting dimensions to SO(D−2) gives OSp(D−1,1|2) (see also
subsection IIC3), which has an SO(D−1,1) subgroup.
We have seen that quantum field theory requires unphysical anticommuting fields
to cancel the commuting unphysical fields introduced by using gauges that do not
eliminate longitudinal polarizations. For example, the gauge field for electromag-
netism has only D−2 components in the lightcone gauge, but needs to keep all D
components to maintain manifest Lorentz covariance; this requires 2 “ghosts” to can-
cel the 2 extra components of the gauge field. The general result, at least for bosonic
gauge fields, is to produce fields that form representations of OSp(D−1,1|2), includ-
ing gauge fields and ghosts. Furthermore, by adding 2 anticommuting dimensions the
 A. OSp(1,1|2) 653

BRST transformations that relate the ghosts to the longitudinal degrees of freedom
can be introduced in a natural way, as translations in the new coordinates. The result
is that OSp(D−1,1|2) multiplets are automatic, and gauge fixing receives a geometric
interpretation. In this section we’ll see that an even more natural interpretation of
these BRST transformations is as rotations of the anticommuting coordinates, and
that they not only make gauge fixing to the simplest Lorentz covariant gauge triv-
ial, but also give a simple derivation of the gauge invariant action itself. (The two
points of view are related in that translations can be considered as part of “conformal
rotations”, as we saw in subsection IA6.)
The basic idea is very simple: Take the lightcone representation of the Poincaré
generators, found in the previous subsection, and extend the SO(D−2) indices and
representations to OSp(D−1,1|2) ones (including appropriate signs for the grading).
Conversely, we can begin the construction with the “conformal” group OSp(D+1,3|2),
find the equations of motion for the “Poincaré” group OSp(D,2|2), and solve them
for the “lightcone group” OSp(D−1,1|2). So we can use the same expressions for the
generators, but now the “transverse” OSp(D−1,1|2) index is

i = (a, α)

where a is a D-component index of SO(D−1,1) and α is a 2-component index of Sp(2).

The OSp(D−1,1|2) metric is
η ij = (η ab, C αβ )

Furthermore, we divide up the full OSp(D,2|2) index as

(±, i) = (A, a), A = (±, α)

We now interpret the SO(D−1,1) subgroup that acts on the a index as the usual
physical one, since the generators take the usual covariant form (because all the
transverse generators are linear). The orthogonal subgroup OSp(1,1|2) that acts on
the A index, and leaves the a index alone, is then interpreted as a symmetry group
of the unphysical degrees of freedom, an extension of BRST. However, the generators
with − indices are nonlinear, since the ± indices are no longer independent from the
rest. (The + were gauged away, the − were fixed by equations of motion.) As a
result, they act on transverse indices in a nontrivial way.


longitudinal, nonlinear SO(1,1): ± = A : ghost OSp(1,1|2)
α
transverse, manifest OSp(D−1,1|2): i =
a : Lorentz SO(D−1,1)
654 XII. MECHANICS

Since the OSp(1,1|2) generators act only in the unphysical directions, all phys-
ical states should be singlets under this symmetry. This is clear from the original
construction: We started with linear generators for OSp(D,2|2) (and translations and
dilatations for (D,2|2) dimensions), applied the equations of motion in terms of them,
and now we apply the OSp(1,1|2) singlet condition last. If we had instead applied
the OSp(1,1|2) singlet condition first to the (D,2|2) dimensional space, we would have
gotten the usual (D−1,1|0) dimensional space, and finally applying the equations of
motion would have given us the lightcone results of subsection IIB3.

OSp(D,2|2)
manifest symmetry: %
. &
SO(D−1,1) =⇒ OSp(D−1,1|2)
& field equations (fix ±) field strengths gauge fields/BRST
. BRST singlets (fix A) & .
%
% add 2+2 (extend i → (a, α)) SO(D−2)
lightcone

Explicitly, the OSp(1,1|2) generators are (choosing w = 0)

J +α = −xα∂ +, J −+ = x− ∂ +, J αβ = x(α ∂ β) + S αβ
1 1 α
J α− = −x−∂ α + [ x ( − m2 + ∂ β ∂β ) + S αb ∂b + S α−1 im + S αβ ∂β ]
∂+ 2
while the SO(D−1,1) generators take their usual manifest form

J ab = x[a ∂ b] + S ab

Excercise XIIA2.1
Write the general commutation relations of OSp(D−1,1|2). Specialize to the
case OSp(1,1|2), in lightcone notation. Show that this representation satisfies
them, paying special attention to signs. (Use the OSp(D−1,1|2) commutators
for the S’s.)
We can also add a “spin” part to the general “orbital” part of the OSp(1,1|2) alge-
bra we have already derived, in the sense that the two parts commute and separately
satisfy the commutation relations:

J AB → J AB + S̃ AB

The simplest choices are to choose this spin part to be just quadratic in new, “non-
minimal” coordinates and momenta. It will prove convenient to perform some trans-
formations J → UJU −1 that make the OSp(1,1|2) generators more similar to what
 A. OSp(1,1|2) 655

they were before adding the spin parts. We therefore make two consecutive transfor-
mations:
−+
J → U2 U1 JU1−1 U2−1 :
+α ∂ +
U1 = eS̃ α /∂
, U2 = (∂+)S̃

to return J −+ and J +α to their previous forms. In fact, these are just the OSp(1,1|2)
version of the same transformations we used in the previous subsection to remove S +i
and S +− . The result is

J +α = −xα∂ +, J −+ = x− ∂ +, J αβ = x(α ∂ β) + Ŝ αβ
1 α
J α− = −x−∂ α + [x (−K + 12 ∂ β ∂β ) + Qα + Ŝ αβ ∂β ]
∂+
Here we have

K = − 12 ( − m2), Ŝ αβ = S αβ + S̃ αβ , Qα = S̃ α− + S αb ∂b + S α−1 im − S̃ +α K

but more generally we can satisfy the commutation relations by requiring only that
K, Qα, and Ŝαβ are independent of the unphysical coordinates x− and xα and their
momenta, and satisfy that their only nontrivial commutators are
(γ
{Qα, Qβ } = 2K Ŝ αβ , [Ŝ αβ , Qγ ] = Q(αC β)γ , [Ŝαβ , Ŝ γδ ] = δ(α Ŝβ) δ)

3. Action
We saw in the previous subsection that physical states are singlets under the
OSp(1,1|2) BRST symmetry. It was introduced in a trivial way, but became nontrivial
after solving the equations of motion; on the other hand, applying the singlet condition
first reproduced the usual lightcone analysis. Reversing the order of applying the two
conditions has the advantage of allowing the physical state condition to be expressed
as a single equation, which can be derived from an action.
There are two ways of doing this: One is to use this algebra to generalize to gauge
fields the first-quantized BRST as applied to field theory in subsection VIA3. Because
of their quantum mechanical origin, the gauge-invariant ΦQΦ actions directly give a
form suitable for choosing the Fermi-Feynman gauge, where the kinetic operator is
simply − m2. However, this is somewhat unusual for fermions, whose simplest field
equation is first-order. (But it is useful for supersymmetry, where bosons and fermions
are treated symmetrically.) As a result, this approach gives actions for fermions with
an infinite number of auxiliary and ghost fields. The most convenient way to discover
the usual finite-component gauge-invariant first-order actions hidden there is by per-
forming an appropriate unitary transformation, after which this action (for bosons or
656 XII. MECHANICS

fermions) appears as the sum of three terms: the usual gauge-invariant action, a term
giving the usual second-quantized BRST transformations, and a nonderivative term
that would be considered nonminimal under second-quantized BRST. This approach
will be described in detail in the following sections.
The other way is to define a δ function in the generators of the group OSp(1,1|2),
and use it as the kinetic operator for the action:
Z
S = − dx dx−d2 xα 14 Φ∂ +δ(JAB )Φ

where the integration is over all the coordinates appearing in the OSp(1,1|2) gener-
ators. (The dx part is the usual dD x/(2π)D/2.) ∂ + comes from the usual relativistic
inner product; it is also a “measure” factor, which is a consequence of our using
generators satisfying the pseudohermiticity condition

J †AB ∂ + = −∂ +JAB

Equivalently, we could redefine Φ → (∂ +)−1/2Φ, J → (∂ +)−1/2 J(∂ +)1/2 (assuming

∂ + 6= 0, as usual in lightcone formalisms) to eliminate it and restore hermiticity.
(This would only affect J −+ → J −+ − 12 , J α− → J α− + ∂ α/2∂ +, making hermitian
the terms 12 {x−, ∂ +} and 14 {xα, ∂ β ∂β }/∂ +.) Because of the δ function, this action has
the gauge invariance
δΦ = 12 J BAΛAB

Thus, the field equations and gauge invariance reduce Φ to states in the OSp(1,1|2)
cohomology.
More explicitly, the δ function can be written as

∂ +δ(JAB ) = ∂ +δ(Jαβ 2 )δ(J −+)δ 2(J +α )δ 2(J α−)

= δ(x−)δ 2(xα )δ(Ŝαβ 2 )∂ +2 J −α2

where we have used

1
J −+ δ(J −+) = δ(J −+)J −+ = 0 ⇒ δ(J −+) = δ(x−)
∂+
(There is freedom in ordering of the original δ functions: Reordering of any two δ’s
produces terms that are killed by the other δ’s.) The δ(Ŝαβ 2 ) can be interpreted as a
Kronecker δs0 in the Sp(2) “spin” s:

− 12 Ŝ αβ Ŝαβ = 4s(s + 1)
 A. OSp(1,1|2) 657

(remember Ŝ αβ is antihermitian, and iŜ ⊕ is always integer while s can be half-

integer). The rest of the explicit δ’s are Dirac δ’s in the unphysical coordinates,
which can therefore be trivially integrated out, leaving:
Z
S = dx Lgi , Lgi = 12 φKgi φ, Kgi = 12 (− + m2 + 12 QαQα )

where φ is Φ evaluated at xα = x− = s = 0. Furthermore, the remaining gauge

invariance is
δφ = δs0 12 QαΛα

from J α−, since J αβ , J +α , and J −+ have been used to gauge to s = xα = x− = 0,

respectively.
Excercise XIIA3.1
Show explicitly that this action is invariant under the OSp(1,1|2) gauge trans-
formations. (Hint: Use the same method as excercise XA2.1.)

4. Spinors
As we saw in subsection VIA3, the BRST algebra for the (Dirac) spinor requires
nonminimal terms. For the general case of fermions we add these terms in the general
way described in subsection XIIA2, choosing them in terms of a (second) set of
OSp(1,1|2) γ matrices:

S̃ AB = − 12 [γ̃ A , γ̃ B }, {γ̃ A, γ̃ B ] = −η AB

where γ̃ A = (−κ, −µ; ξ̃, iζ̃) in the notation of subsection VIA3. In particular, we find

Ŝ αβ = S αβ − 12 γ̃ (α γ̃ β)

Qα = S αb ∂b + S α−1im + γ̃ α[γ̃ − + γ̃ + 21 ( − m2 )]

The next step for general massless fermions (and similarly for the massive case)
is to apply
Q2 = 12 S αaSα b ∂a ∂b − (γ̃ − − γ̃ +K)γ̃ α Sα a ∂a − 12 K
where we have used
γ̃ αγ̃α = C αβ 12 [γ̃β , γ̃α ] = 1
At this point we note that the gauge invariance generated by Qα, for gauge parameter
Λα = γ̃α Λ, includes a term γ̃ − Λ that allows us to choose the gauge

γ̃ +φ = 0
658 XII. MECHANICS

One way to think of this is to treat γ̃ + as an anticommuting coordinate and γ̃ − as

its derivative; another way is to treat them as 2×2 matrices. Alternatively, we can
unitarily transform the action to contain just the γ̃ − term of the operator: From the
discussion of 2D γ matrices of subsection VIIIA7 we find, including that part of the
spinor metric,


Υ̃γ̃ − = − 10 00
then acts as a projection operator. Either way, the net result is to reduce the action
to, now restoring the mass,
Z
Sf = dx Lgi,f , Lgi,f = 12 φ̂Kgi,f φ̂, Kgi,f = 12 γ̃ α(Sα a ∂a + Sα−1im)

where φ̂ is φ with the γ̃ ± -dependence eliminated (the top component in the above
matrix representation). Thus, φ̂ differs from the bosonic case in that it not only
depends on xa and is a representation of S ij , but is also a representation of γ̃ α, which
appears in Ŝ αβ to define s = 0.
The only type of representation we have missed in this analysis is self-dual anti-
symmetric tensors. In terms of field strengths, these satisfy

Fa1...aD/2 = ± (D/2)!
1
a1 ...aD/2b1 ...bD/2 F b1 ...bD/2

which is consistent, with Lorentz metric, if D/2 is odd (as seen from applying the
 tensor twice). A similar condition holds for the gauge field in the lightcone gauge
(with a (D−2)-dimensional  tensor). Because of the  tensor, this condition can’t be
described by adding extra dimensions to the lightcone. However, the direct product of
two spinors contains all antisymmetric tensors, and the rank D/2 one can be picked
out by an appropriate OSp invariant constraint. The self-dual part of this tensor
comes from the direct product of chiral spinors.
Excercise XIIA4.1
We now consider this construction in more detail:
a Derive the generalization of γ−1 to OSp γ matrices, anticommuting with both
the fermionic and bosonic γ’s, {γ−1 , γA } = 0. We can use the usual product
for the fermionic γ’s, but obviously the bosonic ones will need something
different. (Hint: For each pair of fermionic or bosonic γ’s there is a Klein
factor, as in excercise IA2.3e; for the fermionic γ’s the exponential is equal to
the usual product.)
b In twice-odd dimensions, consider the direct product of two spinors by rep-
resenting the OSp spin operators as a sum in terms of the two different sets
 A. OSp(1,1|2) 659

of OSp γ matrices acting on the two different spinor indices. Define the U(1)
(O(2)) symmetry that mixes the two γ matrices by taking linear complex
combinations of the γ matrices to form fermionic creation and annihilation
operators, so the OSp-invariant U(1) generator is a†A aA. Show that the eigen-
values of this generator pick out the different Lorentz representations. These
can be made irreducible by including γ−1 projections. Using explicit U(1)
and (both) γ−1 projectors in the action, show that self-dual tensors can be
described. (Note: This description contains an infinite number of auxiliary
fields.)

5. Examples

The OSp(1,1|2) method is thus an efficient method for finding gauge-invariant

actions (though not so useful for gauge fixing). We begin with examples of massless
bosons, for which the gauge-invariant kinetic operator is

Kgi = 12 (− + 12 Qα Qα), Qα = S αa∂a

The scalar is a trivial example; the simplest nontrivial example is the massless vec-
tor: In terms of the basis |i i for an OSp(D−1,1|2) vector (D-vector plus 2 ghosts),
normalized to
hi |j i = η ij ⇒ ha |b i = η ab , hα |β i = C αβ

we can write the OSp(D−1,1|2) generators as

ij
S(1) = |[i ihj) |

The Sp(2)-singlet field is then (dropping the |α i term)

φ = |aiAa (x)

We then have

Qα = (|αiha | − |aihα |)∂a ⇒ Q2 ≡ 12 QαQα = |a ihb |∂a∂b − 12 |αihα |

⇒ Lgi(1) = 18 (Fab )2

and for the gauge invariance

δφ = δs0 12 QαΛα ⇒ Λα = |α iλ(x) ⇒ δAa = ∂a λ

660 XII. MECHANICS

A more complicated example is the graviton (massless spin 2): We write the field,
a graded symmetric, traceless OSp(D−1,1|2) tensor, in terms of the direct product
of two vectors, with basis |i i|j i. The spin operators are thus
ij
S ij = S(1) ⊗ I(1) + I(1) ⊗ S(1)
ij

where the first factor in each term acts on the first factor in |i i|j i, etc.; I(1) is the
spin-1 identity. The s = 0 part of the field is then

φ = |i i|j ihji , hi i = ha a + hα α = 0 ⇒ φ = (|a i|b i + 12 |α i|α iη ab )hab

where hab includes its trace. The rest is straightforward algebra; we use identities
such as:
Q2 = Q2(1) ⊗ I(1) + I(1) ⊗ Q2(1) + Qα(1) ⊗ Q(1)α

Qα(1) ⊗ Q(1)α = (|α i|α iha |hb | + |ai|b ihα |hα |)∂a ∂b

(hα |hα |)(|β i|β i) = −hα |β ihα |β i = −2

where Q2(1) was evaluated above, and in the last identity we used the fact that |αi is
anticommuting. The final result is then

Lgi(2) = − 14 [hab hab + 2(∂ b hab )2 − ha a hb b + 2ha a ∂ b ∂ c hbc ]

in agreement with subsection IXB1. The original OSp(1,1|2) gauge invariance reduces
to
Λα = (|αi|a i + |ai|α i)λa ⇒ δhab = ∂(aλb)

Excercise XIIA5.1
Consider a (D−2)-rank antisymmetric tensor (i.e., totally antisymmetric in
D−2 indices in D dimensions; see excercises IIB2.1 and VIIIA8.2, and sub-
section XA3).
a Show from a lightcone analysis that it is equivalent to a scalar. Derive the
gauge-invariant action using OSp methods. Find the gauge transformations
and field strength.
b Find a first-order form for the action, (auxiliary field)2 + (auxiliary field) ×
(field strength). Show that eliminating the gauge field as a Lagrange multipler
results in the action for a scalar. Show that switching between scalar and
antisymmetric tensor is equivalent to switching field equation and constraint
for the field strength.
c Find the description for the massive case by dimensional reduction.
 A. OSp(1,1|2) 661

Excercise XIIA5.2
Consider a tensor totally symmetric in its vector indices. In the lightcone
gauge, the irreducible tensor is traceless. Show that, upon covariantiza-
tion, the field appearing in the gauge-invariant action satisfies a double-
tracelessness condition (or equivalently the fields appearing there are the
totally traceless tensor and another totally tracelsss tensor with two less in-
dices).
For massless fermions we saw

Kgi,f = 12 γ̃ α Sα a∂a

The next step is to use the fact that arbitrary fermionic representations are con-
structed by taking the γ-traceless piece of the direct product of a (Dirac) spinor with
an irreducible bosonic representation. (Just as an irreducible bosonic representation
of an orthogonal group is found by taking the direct product of vectors, choosing an
appropriate symmetry, as described by the Young tableau, and requiring the trace
in any two vector indices to vanish; here we also require that using a γ matrix to
contract the spinor index with any vector index also vanishes. Of course, simpler
methods can be used for SO(3,1), but we need methods that apply to all dimensions,
so they can be applied to orthosymplectic groups.) We then can write

S ij = Š ij − 12 [γ i, γ j } ⇒ Ŝ αβ = Š αβ − a†(α aβ)

where Š ij is the part of the spin acting on just the vector indices, and we have
combined γ α and γ̃ α into creation and annihilation operators, as in subsection VIA3:

aα = √1 (γ α
2
+ iγ̃ α), a†α = √1 (γ α
2
− iγ̃ α); [aα, a†β ] = −δαβ

For spin 1/2 Š = 0, δs0 projects to the ground-state of the oscillators, and we
immediately find
Sα a = −γα γ a ⇒ Lgi(1/2) = 14 φ̂γ a i∂aφ̂

where we have used

γ̃ α γα = i 12 (a†α aα − aαa† α) = i(a†α aα − 1) = −i(N + 1)

and this “N” counts the a†α excitation level. (Note that the Hilbert-space inner
product between the spinors includes the usual factor of γ0 .) A less trivial case is
spin 3/2: Now
φ = |i iφi , S αa = −γ α γ a + |[α iha] |
662 XII. MECHANICS

where φi has an explicit vector index, and an implicit spinor index. Then from γ-
tracelessness (for irreducibility) we have for the Sp(2) singlets

γ i φi = 0 ⇒ φα = −γα γ a φa ⇒ aαφa = 0, φα = − √12 a† αγ a φa

After a little algebra, using identities such as

1 [a b c]
6
γ γ γ = γ a γ b γ c + 12 (η b(a γ c) − η ac γ b )

we find
Lgi(3/2) = − 12
1
φ̂a γ [a γ b γ c] i∂b φ̂c

From inspection, or from δφ = δs0 12 Qα Λα , we find the gauge invariance

δ φ̂a = ∂aλ

Excercise XIIA5.3
Let’s now examine some massive examples:
a Find the gauge-invariant actions for massive spin 2 and spin 3/2 by dimen-
sional reduction of the massless cases.
b Note that for the spin-2 case the part of the mass term quadratic in h is
proportional to −h[aa hb] b . More generally, we might have expected (hab )2 +
k(ha a )2 for arbitrary constant k, since the first part gives mass to the physical
(transverse, traceless) part of h, while the second term affects only the unphys-
ical pieces. Find the Stückelberg terms generated from this generalized mass
term by the linearized gauge invariance. Looking at just the terms quadratic
in the Stückelberg vector, what is special about k = −1, and why do other
values of k give ghosts? (Hint: Compare gauge-fixed electromagnetism.)

REFERENCES
1 K. Bardakçi and M.B. Halpern, Phys. Rev. 176 (1968) 1686:
lightcone representations of Poincaré algebra.
2 M. Fierz and W. Pauli, Proc. Roy. Soc. A173 (1939) 211;
S.J. Chang, Phys. Rev. 161 (1967) 1308;
L.P.S. Singh and C.R. Hagen, Phys. Rev. D9 (1974) 898;
C. Fronsdal, Phys. Rev. D18 (1978) 3624;
T. Curtright, Phys. Lett. 85B (1979) 219;
B. deWit and D.Z. Freedman, Phys. Rev. D21 (1980) 358;
T. Curtright and P.G.O. Freund, Nucl. Phys. B172 (1980) 413;
T. Curtright, Phys. Lett. 165B (1985) 304:
covariant actions for totally symmetric representations of all (4D) spin.
 A. OSp(1,1|2) 663

3 W. Siegel and B. Zwiebach, Nucl. Phys. B282 (1987) 125:

lightcone representations of Poincaré algebra, and covariant actions, for arbitrary spin
and dimension.
4 G. Parisi and N. Sourlas, Phys. Rev. Lett. 43 (1979) 744:
cancelation of extra dimensions.
5 R. Delbourgo and P.D. Jarvis, J. Phys. A15 (1982) 611;
J. Thierry-Mieg, Nucl. Phys. B261 (1985) 55;
J.A. Henderson and P.D. Jarvis, Class. and Quant. Grav. 3 (1986) L61:
ghosts from extra dimensions for indices.
6 W. Siegel, Nucl. Phys. B284 (1987) 632:
1st-quantized BRST for fermions.
7 N. Berkovits, hep-th/9607070, Phys. Lett. 388B (1996) 743:
1st-quantized BRST for self-dual tensors.
8 Siegel, loc. cit.
9 Fierz and Pauli, loc. cit. (ref. 2 above):
mass term for spin 2.
664 XII. MECHANICS

............................... ...............................
............................... B. IGL(1) ...............................

Although the OSp(1,1|2) method is the simplest way to derive general free gauge-
invariant actions, it does not yield a simple method for gauge fixing, even though
the Hilbert space contains exactly the right set of ghosts. We now describe a related
method that is slightly less useful for finding gauge-invariant actions (it includes
redundant auxiliary fields), but allows gauges to be fixed easily.

1. Algebra
For this method we use a subset of the OSp(1,1|2) constraints, and show they are
sufficient. A simple analog is SU(2): To find SU(2) singlets, it’s sufficient to look for
states that are killed by both T3 and the raising operator T1 + iT2. This approach
gives a formalism that turns out to be easier to generalize to interacting theories, as
well as allowing a simple gauge-fixing procedure. We first divide up the Sp(2) indices
as α = (⊕, ) (not to be confused with ±). We then make a similarity transformation
that simplifies some of the generators (while making others more complicated):
⊕
J → UJU −1 : U = (∂ +)iJ

which changes the Hilbert-space metric (and corresponding hermiticity conditions) to

Z
† iJ ⊕
Υ = U U = (−1) , hΨ|Ξi = Ψ†ΥΞ

This simplifies (looking at the massless case without S̃ AB for simplicity)

1 +
J + → J = −x
∂+
J −+ → J −+ − iJ ⊕ , J ⊕ → J ⊕

J ⊕− → J ⊕−∂ + + iJ ⊕ ∂ ⊕ = (−x−∂ +)∂ ⊕ + 12 x⊕ + S ⊕a ∂a + S ⊕⊕ ∂⊕

(We use the same conventions for raising and lowering Sp(2) indices as for SU(2) in
subsection IIA4 and SL(2,C) in subsection IIA5.) These four generators form the
subgroup GL(1|1) of OSp(1,1|2) (=SL(1|2)): We can write the generators as JI J ,
where I = (+, ⊕). In subsection XIIB4 we’ll see that the singlets of this GL(1|1) are
the same as those of OSp(1,1|2).
On the other hand, because of the simplified form of these generators, it’s easy
to see how to reduce the group even further:

J + = −x = 0, J −+ + iJ ⊕ = x−∂ + = 0
 B. IGL(1) 665

⇒ J ⊕ = −ix⊕∂⊕ + i + S ⊕ , J ⊕− = 12 x⊕ + S ⊕a ∂a + S ⊕⊕ ∂⊕

(Of course, this further reduction could have been performed even without the trans-
formation.) We are now left with the group IGL(1), with just J ⊕ and J ⊕− as gener-
ators. (J ⊕− acts as translations for the GL(1) generator J ⊕ .) Also, we have reduced
the unphysical coordinates to just x⊕. We now simplify notation by relabeling

c = x⊕ , b = ∂⊕ , S 3 = iS ⊕ , J = iJ ⊕ + 1, Q = J ⊕−

⇒ J = cb + S 3 , Q = 12 c + S ⊕a ∂a + S ⊕⊕ b

J and Q are versions of the ghost-number and BRST operators introduced in sub-
section VIA1. The net result for obtaining these IGL(1) generators from the original
OSp(1,1|2) generators can also be stated as

J = iJ ⊕ |∂ ⊕=0,∂ + =1 , Q = J ⊕−|∂ ⊕ =0,∂ + =1

where ∂ ⊕ = 0 can be regarded as a gauge condition for the constraint x = 0, and

∂ + = 1 for x−∂ + (∼ ∂/∂(ln ∂ +)) = 0. The IGL(1) algebra is

[J, Q] = Q, Q2 = 12 {Q, Q} = 0

Excercise XIIB1.1
Show that any IGL(1) subgroup of OSp(1,1|2) (J = iJ ⊕ , Q = J ⊕− ) satis-
fies these commutation relations. Check that this final representation of the
IGL(1) algebra satisfies them.

2. Inner product
The new inner product can be derived by the same steps: Starting with the
lightcone inner product of subsection XIIA1, we add extra dimensions to get the
OSp(D,2|2) inner product. We next drop dependence on x , which will be eliminated
⊕
in the IGL(1) formalism. Then we perform the transformation with (∂ +)iJ =
(∂ +)J−1 used to simplify the BRST operator. This acts on both fields in the inner
product; applying integration by parts turns one such factor into (−∂ +)−J . The net
effect is that it cancels the ∂ + in the Hilbert-space metric, which allows us to drop
the x− integration, and it introduces a factor of (−1)J .
Rather than defining a Hilbert-space inner product, which is sesquilinear, it is
slightly more convenient to define a symplectic inner product, replacing the Hermitian
666 XII. MECHANICS

conjugate of the wave function/state on the left with the transpose, in analogy to an
ordinary vector inner product. The inner product is then
Z
hΨ|Ξi = −i(−1) Ψ
dx dc ΨT (x, c)(−1)J Ξ(x, c)

A Hilbert-space inner product can then be defined simply as hΨ*|Ξi (where (Ψ*)T =
Ψ† ). We have included a sign factor corresponding to what would be obtained if the
dc integration were moved to the symmetric position between the two wave functions:
By (−1)Ψ we mean take Ψ = 0 in the exponent if Ψ is bosonic and 1 if Ψ is fermionic.
We can make this manifest by defining
Z
Ψ(x, c) = hx, c|Ψi ⇒ I = −i dx dc |x, ci(−1)J hx, c|

which allows the inner product to be evaluated between hΨ| and |Ξi by inserting this
form for the “identity” I.
As a result, any commuting or anticommuting constant factor “a” can be moved
out of the inner product from the left or right in the usual way:

hΨ|Ξai = hΨ|Ξia, haΨ|Ξi = ahΨ|Ξi

As a consequence of the anticommutativity of the integration measure, we have

(−1)hΨ|Ξi = (−1)Ψ+Ξ+1

meaning that the statistics of the inner product is the opposite of an ordinary product;
we can think of “ | ” as a fermion.
Because of the change in metric from the lightcone, the IGL(1) generators now
satisfy
J T = 1 − J, QT = Q

where the constant comes from dropping the extra coordinates, and the transpose
“ T ” indicates integration by parts (the usual transpose in the infinite-matrix repre-
sentation of operators):
Z Z
ΨOΞ = (−1)ΨO (OT Ψ)Ξ

Before our transformations the generators all satisfied GT = −G; now they are pseu-
doantisymmetric with respect to the metric (−1)J , up to the constant:

hΨ|OΞi = hÕΨ|Ξi ⇒ Õ = (−1)J OT (−1)J

 B. IGL(1) 667

J˜ = 1 − J, Q̃ = −Q
From J T = 1 − J also follows the symmetry property of the inner product:

hΨ|Ξi = (−1)(Ψ+1)(Ξ+1)hΞ|Ψi

This can be interpreted as antisymmetry once the anticommutativity of the “ | ”

(metric) is taken into account.
The hermiticity conditions that follow from the change from the lightcone are

J † = 1 − J, Q† = Q

Before all generators were antihermitian; now they are pseudoantihermitian, up to a

constant:
Ô = (−1)J O† (−1)J
⇒ Jˆ = 1 − J, Q̂ = −Q
The factor of i in the inner product compensates for the funny hermiticity of (−1)J .
We then find the usual hermiticity condition for a vector inner product,

hΨ|Ξi* = hΞ† |Ψ†i

3. Action
As explained in subsection VIA1, we are interested in states in the cohomology
of the BRST operator Q, which means states satisfying

QΦ = 0, δΦ = iQΛ

In particular, the physical states are states in the cohomology of Q at ghost number
J = 0. However, now QΦ = 0 is the wave equation (as in subsection VIA3), and QΛ
contains the usual gauge transformations (as in the OSp(1,1|2) action of the previous
section).
The free gauge-invariant action for an arbitrary field theory is then
Z Z
S0 = −(−1) Φ
dx dc 2 Φ δJ1 QΦ = −(−1)
1 T Φ
dx dc 12 ΦT QδJ0Φ

for a real column-vector field Φ. (Complex fields can be decomposed into their real
and imaginary parts. For relating to quantum mechanics, we will usually consider
the column-vector in our OSp Hilbert-space notation.) This action gives QΦ = 0 as
the equations of motion, and has δΦ = QΛ as a gauge invariance, so the solutions are
the cohomology of Q. The projector δJ0 is a Kronecker δ restricting Φ to vanishing
668 XII. MECHANICS

ghost-number (and thus QΦ to ghost number 1, since [J, Q] = Q). As we’ll see in
the next section, this projector is redundant: The states in the cohomology with
nonvanishing momentum automatically have vanishing ghost-number, and the states
with nonvanishing ghost-number are needed for gauge fixing. The projection is useful
only for eliminating states that are redundant for discussing gauge invariance; we’ll
drop it for the remainder of this section. The “complete” free action is then
Z
S0 = −(−1) Φ
dx dc 12 ΦT (−1)J−1 QΦ = 12 hΦ|iQΦi = S0 †

where we have included the inner-product metric. This is just the translation of
the free BRST operator from first- to second-quantized form, as for the lightcone in
subsection XIIA1.
We now consider some simple examples, to see how this method reproduces the
usual results. The simplest example is the scalar: As shown in subsection VIA3,
Z
Q = c 2 ( − m ) ⇒ − dc 12 Φ(−1)J−1 QΦ = − 12 φ 12 ( − m2)φ
1 2

without restrictions to vanishing ghost number, unitary transformations, gauge fixing,

etc. Thus the scalar is in no way a gauge field: The kinetic operator follows from
simple kinematic considerations.
The fundamental example of a gauge theory is a vector: It is the defining repre-
sentation of the Lorentz group, and of the extended Lorentz group we used to define
the BRST operator. In the rest of this chapter we will see from its action most of the
general properties of gauge theories: ghosts, gauge invariance, BRST transformations
of the fields, the gauge-invariant action, gauge fixing, backgrounds, mass, etc. In
addition to the equations given for this case in subsection XIIA5, we will use

h⊕ | i = −h |⊕i = i

The BRST and ghost-number operators are

Q = 12 c + (|⊕iha | − |a ih⊕ |)∂a + 2|⊕ih⊕ |b, J = cb + i(|⊕ih | + | ih⊕ |)

The field is real; it can’t be called hermitian, since it is a column vector, but each
component of that vector is hermitian. (This is the same as reality, but for anti-
commuting objects reality includes extra signs that are defined to be exactly those
coming from hermitian conjugation.) Thus,
ˇ − |⊕iČ)
Φ = Φ* = (|a iAa − i| iC + i|⊕iC̃) − ic(|aiǍa − | iC̃
 B. IGL(1) 669

ˇ | − Čh⊕ |)ic
ΦT = Φ† = (Aaha | + iCh | − iC̃h⊕ |) + (Ǎaha | − C̃h
where we denote the “antifields” (those at order c) by a “ ˇ ”.
The BRST transformations of the fields can be found by comparing terms in Φ
and QΦ: If we define a second-quantized BRST operator Q̂ such that QΦ = Q̂Φ, but
Q̂ acts only on the fields while Q (as usual) acts on |ii and the coordinates (x and c),
then

Φ = |i i(φi − icψi ) ⇒ QΦ = Q̂Φ = |i i[(−1)i Q̂φi − ic(−1)i+1 Q̂ψi ]

In other words, we compare terms in Φ and QΦ, and throw in a minus sign for
transformations of fermions. (So, e.g., “Q̂Aa” is the coefficient of |a i in QΦ.) Dropping
the “ ˆ ” on Q, the result is

QAa = −∂aC, QC = 0, ˇ − 1 ∂ · A)
QC̃ = −2i(C̃ 2

ˇ
QǍa = −i( 12 Aa − ∂aC̃), QČ = 1
C̃ + ∂ · Ǎ, ˇ = −1 C
QC̃
2 2

Note that although Q is hermitian, it is antihermitian with respect to the inner-

product metric (−1)J , as expected from our convention of using antihermitian gener-
ators for spacetime symmetries. (The same extra sign for hermiticity vs. pseudoher-
miticity, also because of ghosts introduced by relativistic quantum mechanics, occurs
for the spatial Dirac matrices γ i : see subsection XC2.) As a result, our transforma-
tions agree with those of subsection VIA4. However, while the first-quantized Abelian
transformations also agree with those of subsections VIA1-3, the second-quantized
nonabelian transformations will have the extra i demonstrated in subsection VIA4,
following from the i introduced in the inner-product metric in the previous subsec-
tion. (This minor yet annoying factor will be further discussed in section XIIC when
we relate first- and second-quantized BRST.)
The Lagrangian then can be expanded as (after some integration by parts)
Z
ˇ + Ǎ · QA − ČQC − C̃Q
L0 = − dc 12 ΦT (−1)J−1 QΦ = −i 12 (A · QǍ − CQČ − C̃QC̃ ˇ C̃)

ˇ − 1 ∂ · A)2 − iC̃ 1 C + iǍ · ∂C

= 18 (Fab )2 + (C̃ 2 2

where we have used the transpose of the field on the left of ΦQΦ. To find the gauge-
invariant action, we can evaluate it by keeping just the (anti)fields with vanishing
ˇ and then eliminate the remaining antifields by their equa-
ghost number (Aa and C̃),
tions of motion:
ˇ − 1 ∂ · A)2
L → 18 (Fab )2 + (C̃ ⇒ Lgi = 18 (Fab)2
2
670 XII. MECHANICS

Excercise XIIB3.1
Consider the example of the second-rank antisymmetric tensor (see excercises
IIB2.1, VIIIA8.2, and XIIA5.1, and subsection XA3):
a Construct the states by direct product of two vectors. Decompose into fields
plus antifields, physical plus ghost: In particular, note the Sp(2) representa-
tion of each SO(D−1,1) representation.
b Find the BRST transformations for all the (anti)fields. In particular, note
that the tensor transforms into vector ghosts (as expected from the gauge in-
variance), which themselves transform into scalar ghosts (“ghosts for ghosts”).
c Graph all the states for s (of the Sp(2) S αβ ) vs. J, and indicate there how
BRST relates them.
d Find the gauge-invariant action from ΦQΦ.
e Generalize to arbitrary-rank antisymmetric tensors. Compare the results of
excercise XIIA5.1a.

4. Solution
The identity of the cohomology and the physical states can be proven most easily
by making a unitary (“gauge”) transformation to the “lightcone gauge”:

(Q, J) → U(Q, J)U −1 :

+i ∂ +⊕ b)/∂ +
U = e(S i +S

which simplifies Q while leaving J unchanged:

Q → 12 c − S ⊕− ∂ +, J → cb + S 3

These are the usual lightcone indices ± of any D-vector, not to be confused with the
± used earlier when reducing from D+2 bosonic dimensions. Except for the extension
to include the ⊕ index, this is the same transformation used in subsection IIB3 (and
XIIA1).
This makes the generators separable, allowing us to treat the two terms in Q and
J independently. Specifically, if we integrate the action over c,
Z
Φ = φ−icψ ⇒ − dc 12 ΦT (−1)J−1 QΦ = − 14 φT (−1)S φ−iψ T (−1)S −1 S ⊕− ∂ +φ
3 3

Then ψ is just a Lagrange multiplier enforcing the algebraic constraint S ⊕− φ = 0

(ignoring ∂ +, which we always assume is invertible in the lightcone approach), leaving
 B. IGL(1) 671

just the Klein-Gordon term for the part of φ that satisfies the constraint. We also
have the gauge invariance

Λ = λ + cχ, δΦ = iQΛ ⇒ δφ = −iS ⊕−∂ +λ, δψ = −S ⊕−∂ +χ − 1

2 λ

so we can shoose the gauge where φ is restricted (algebraically) to be in the cohomol-

ogy of S ⊕− .
To solve for the cohomology of S ⊕− it is sufficient to consider the reducible repre-
sentations formed by direct products of vectors (for bosons), or the direct products of
these with a single Dirac spinor (for fermions), since by definition the OSp(D−1,1|2)
generators S ij don’t mix different irreducible OSp(D−1,1|2) representations. We’ll
show that this cohomology restricts any reducible OSp(D−1,1|2) representation to
the corresponding reducible SO(D−2) lightcone representation, and therefore restricts
any irreducible OSp(D−1,1|2) representation to the irreducible SO(D−2) represen-
tation from which it was derived. (Also, the irreducible representations in arbitrary
dimensions are most conveniently found by such a construction, where reduction is
performed by symmetrization and antisymmetrization and subtracting traces of vec-
tor indices, and in the fermionic case also subtracting gamma-matrix traces and using
Majorana/Weyl projection.)
For bosons, we first consider the representation from which all the rest are con-
structed, the vector. Writing the basis for the vector states as |i i, where S ij |k i =
|[i iη j)k , we find
S ⊕− = 0 ⇒ not |+i, | i
δ = S ⊕− ⇒ not |−i, |⊕i
This leaves only the transverse lightcone states, as advertised. For the direct product
of an arbitrary number of vectors, we find the same result: The unphysical directions
are eliminated from each vector in the product.
We might worry that extra states in the cohomology would arise from a can-
celation of two terms, resulting from the action of the “⊕” and “−” parts of S ⊕−.
Specifically, this could happen if we could separate out the supertraceless part of
(the graded symmetric part of) the product of two OSp(1,1|2) vectors. (For exam-
ple, for an SO(n) vector we can separate the traceless part of a symmetric tensor as
Tij − n1 δij Tkk .) However, this is not possible, since for OSp(1,1|2)

str(δAB ) = 2 − 2 = 0

Explicitly, we can look at the two likely candidates for extra states in the cohomology,

|+i|− i ± i|⊕i| i
672 XII. MECHANICS

(and their transposes). But using

S ⊕− |+ i = −|⊕i, S ⊕− | i = −i|−i

for these states we find

S ⊕− (|+i|− i + i|⊕i| i) = −2|⊕ i|−i

S ⊕− i|+i| i = |+i|− i − i|⊕i| i

so neither state is in the cohomology. Note that we take S ⊕− to anticommute with

|α i; the states in the Hilbert space are assigned statistics. (This is the simplest way
to allow a direct relation between wave functions and fields.)
Note that in the Lagrangian − 14 φT φ − iψ T S ⊕− ∂ +φ the fields in φ that are
nonzero when acted upon by S ⊕− are auxiliary, killed by the Lagrange multiplier ψ.
On the other hand, the fields that are S ⊕− on something are pure gauge, and do not
appear in the ψS ⊕−φ term because S ⊕− is nilpotent, while they drop out of the φ φ
term because the fields multiplying them there are exactly the auxiliary ones that
were killed by varying ψ. This follows from the fact that a field that is pure gauge
with respect to S ⊕− has a nonvanishing inner product only with an auxiliary field,
since φ1 = S ⊕−λ ⇒ φ2 φ1 = φ2 S ⊕− λ. Equivalently, a field redefinition ψ → ψ + A φ
can cancel any terms in φ φ where one φ is S ⊕− on something.
For the example of the vector, we have explicitly for the transformation to the
lightcone
1  + i 
ln U = +
(| ih | − |i ih+ |)∂i + (|+ih⊕ | − |⊕ih+ |)b
∂
under which the Lagrangian becomes

L → L0 = − 14 A · ˇ +A− − iǍ− ∂ +C
A − iC̃ 12 C + C̃∂

The lightcone gauge transformations are

δA+ = −∂ +λ, ˇ = −1 λ
δ C̃ 2

δ C̃ = −∂ +ζ, δ Ǎ− = − 12 ζ

δ Č = 1
2 ξ , δ Ǎ+ = 1
2 ξ+ , δ Ǎi = 1
2 ξi

where “i” here refers to the transverse (D−2) components.

 B. IGL(1) 673

5. Spinors
In general we can add nonminimal terms S̃ AB of subsection XIIA2: The easiest
way is to add them as the last step, remembering that Q comes from J ⊕− and J from
J 3; this yields

Q → 12 c + S ⊕a ∂a + S ⊕⊕ b + S̃ ⊕− , J → cb + S 3 + S̃ 3

This result can also be seen from first-quantization of spin 1/2 (subsection VIA3).
Alternatively, if we add S̃ at the beginning as in subsection XIIA2, performing the
transformations given there, followed by the transformation
⊕
U = (∂ +)iJ

of subsection XIIB1, where J ⊕ itself now contains S̃ terms, we again find

1 +
J + → J = −x , J −+ → J −+ − iJ ⊕
∂+
J ⊕ → J ⊕ , J ⊕− → J ⊕− ∂ + + iJ ⊕ ∂ ⊕

Adding a final transformation

U = e−S̃
+⊕ b

(which actually undoes part of an earlier one), we again obtain the above result.
As in the previous subsection, we can also transform to the lightcone gauge, to
find
Q → 12 c − S ⊕− ∂ + + S̃ ⊕−

When analyzing the BRST cohomology in the lightcone gauge, the effect of this
nonminimal term is to replace (again ignoring the factor of −∂ +)

S AB → Ŝ AB = S AB + S̃ AB

(although actually only the S ⊕− and S 3 parts are used here).

As in subsection XIIA4, when treating fermions we choose the Dirac spinor rep-
resentation of OSp(1,1|2) for S̃ AB . Thus, for the case of spin 1/2, where S AB also
is a Dirac spinor representation, Ŝ AB is represented by the direct product of two
OSp(1,1|2) spinors. We now use the harmonic oscillator interpretation of the ghost
coordinates used in subsection XIIA5 (extending it trivially to the fermionic ones),
which can be applied to arbitrary OSp groups:

S AB = − 12 [γ A, γ B }, S̃ AB = − 12 [γ̃ A, γ̃ B }
674 XII. MECHANICS

aA = √1 (γ A
2
+ iγ̃ A), a†A = √1 (γ A
2
− iγ̃ A)
⇒ {aA, a†B ] = −η AB , Ŝ AB = −a†[AaB)
By expanding about the oscillator vacuum, we find this representation of OSp(1,1|2)
consists of the direct sum of totally (graded) antisymmetrized tensors: |0i, |A i =
a†A |0i, |[AB)i = a†Aa†B |0i,... . But we have already treated this case for the bosons,
the result being that only the singlet (vacuum) survives. Of course, the complete spin
representation is given by the direct product of the representation of the unphysical
variables (γ A, γ̃ A ) and the physical ones, namely the transverse lightcone gamma
matrices. Thus, the states in the cohomology are given by the direct product of all
the lightcone states with the vacuum of the unphysical variables. To treat arbitrary
fermions, generalization to direct products of the Dirac spinor with arbitrary numbers
of vectors works the same way, since the spinor looks like the direct sum of parts of
direct products of vectors as far as Ŝ AB is concerned.

6. Masses
As usual masses can be added by dimensional reduction: Our complete result for
application to massless and massive, bosons and fermions is then

Q = 12 c( − m2) + S ⊕a ∂a + S ⊕−1 im + S ⊕⊕ b (+S̃ ⊕−), J = cb + S 3 (+S̃ 3)

with extra i’s introduced implicitly by the procedure given in subsection IIB4.
For example, for the vector we have (the “Stückelberg formalism”)

Q = 12 c( − m2 ) + (|⊕iha | − |a ih⊕ |)∂a + (|⊕ih−1 | + |−1 ih⊕ |)m + 2|⊕ih⊕ |b

J = cb + i(|⊕ih | + | ih⊕ |)
Compared to the massless case treated in subsections XIIB3-4, the corresponding
field now has the extra terms

Φ → Φ + |−1 iφ − ic|−1iφ̌

giving the action

ˇ
L0 = Lgi + [C̃ − 12 (∂ · A + mφ)]2 − iC̃ 12 ( − m2)C + iC(∂ · Ǎ + mφ̌)

Lgi = 18 (Fab)2 + 14 (mA + ∂φ)2

For the spinor

Q = 12 c( − m2 ) − γ ⊕(∂/ − i √m2 ) − (γ ⊕ )2b − γ̃ ⊕γ̃ −

 B. IGL(1) 675

J = cb − iγ ⊕γ − iγ̃ ⊕ γ̃

where in the notation of subsection VIA3,

γ ⊕ = ξ, γ = iζ, γ̃ ⊕ = ξ,
˜ γ̃ = iζ̃, γ̃ − = −µ

7. Background fields
The coupling of external fields can be treated by suitable modification of the
BRST operator. In terms of self-interacting field theories, this corresponds to writing
the field as the sum of quantum and background fields, and keeping in the action only
the terms quadratic in the quantum fields, as discussed for semiclassical expansions
in subsection VA2 and for the background field method in subsection VIB8.
One interesting case is the coupling of an external vector gauge field. Clearly the
spacetime derivatives in Q must be modified by the minimal coupling prescription
∂ → ∇ = ∂ + iA, but dimensional analysis and Lorentz covariance also allow the
addition of a nonminimal term proportional to F abSab to . With the appropriate
coefficient, the general result is (see subsection VIIIA3)

QI = 12 c( − iF abSba ) + S ⊕a ∇a + S ⊕⊕b

where is now the covariant ∇2 . (J is unchanged.)

In the case of spin 0, this modification is trivial. For spin 1/2, we substitute
the graded generalization of the Dirac matrices, S ij = − 12 [γ i, γ j }, as discussed in
subsection XIIA5. We then find Q2I = 0 fixes the above coefficient of the nonminimal
term, the same as from squaring γ a ∇a. This follows from the simple factorization of
S ij in QI :

QI = −c(γ a ∇a)2 − γ ⊕(γ a ∇a) − (γ ⊕ )2b = −(γ a∇a + γ ⊕b)c(γ b ∇b + γ ⊕b)

where we have neglected the S̃ ⊕− term of subsection XIIB5, and used

[γ a, c] = {γ ⊕ , c} = 0

In the spin-1 case, we find the interesting result that Q2 = 0 requires not only
the above coefficient for the nonminimal term (as expected from supersymmetry),
but also that the background terms satisfy the free field equation ∂b F ab = 0. On
the other hand, for spins >1, Q2 = 0 implies Fab = 0, so these spins can’t couple
minimally (at least in flat spaces). Similar remarks apply to coupling gravity (spin
2) to spins >2.
676 XII. MECHANICS

Excercise XIIB7.1
Check these statements for spin 1. Compare the analogous result for back-
ground fields in the field theoretic approach from excercise VIB8.2 for Yang-
Mills for both the gauge transformations of the gauge-invariant action and
the field-theoretic BRST transformations of the gauge-fixed action.
Excercise XIIB7.2
Show that electromagnetism can’t couple minimally to spin 2 (i.e., gravitons
can’t have charge) by considering Q2 = 0 for spin 2 (symmetric traceless OSp
tensor) in an external vector field.
Another interesting feature of the spin-1 case is that we can define a “vacuum”
state
|0i = | i
which is in the BRST cohomology only at zero momentum (constant field), where
Q simplifies to S ⊕⊕b without background. However, this state has ghost number
J = −1. In fact, it corresponds to the global part of the gauge invariance of the theory:
Gauge parameters satisfying QΛ = 0 have no effect in the free theory (where δΦ =
iQΛ), but can act in the interacting theory: They do not contribute an inhomogeneous
term to gauge transformations. However, gauge parameters of the form Λ + QΞ have
the same effect as Λ, up to trivial transformations proportional to the field equations.
Thus, while the BRST cohomology at J = 0 gives the physical states, that at J = −1
gives the global invariances associated with the gauge field.
Now the physical states can be derived by operating on the vacuum with appro-
priate vertex operators: If we expand QI in the number of external fields,

QI = Q + V + ...

Q2I = 0 ⇒ {Q, V } = 0
δQI = i[QI , λ] ⇒ δV = i[Q, λ]
where λ is the gauge parameter for the background field. We therefore see that V
is in the “operator cohomology” of Q. The states in the cohomology of Q are then
given by
Φ = V |0i, Λ = λ|0i ⇒ QΦ = 0, δΦ = iQΛ
We can check this explicitly, as

V = i 12 c({Aa, ∂ a} − F abSba ) + iAaS ⊕a ⇒ V |0i = Aa|a i + ic 12 (∂ a Aa)|0i

ˇ = 1 ∂ · A, in agreement with the field equations.
The second term gives C̃ 2
 B. IGL(1) 677

8. Strings
Another interesting example is strings. Since first-quantization is essential in
string S-matrix calculations, it’s natural to associate string field theory with quantum
mechanical BRST. As usual, for massive fields this formalism automatically includes
the Stückelberg fields that would have been found by dimensional reduction, as well
as all the ghosts. However, the explicit expression for the BRST operator does not
explicitly correspond to that obtained by dimensional reduction: Although the spin
operators S ⊕a , S ⊕⊕ , and S 3 are quadratic in oscillators, S ⊕−1 is cubic (because P̂ −,
and thus X − , is quadratic in the lightcone gauge). Nevertheless, the representation
on any particular irreducible Poincaré representation contained among all the string
states is the same as obtained by dimensional reduction, as follows from the generality
of our analysis.
As for any Poincaré representation, reducible or not, all we need is the lightcone
spin operators, given for the general case in subsection XIIA1. In subsection XIIA2
we saw that the OSp(1,1|2) generators followed immediately from just a change in
notation. The IGL(1) generators were then found in subsection XIIB1 by a uni-
tary transformation and solving half the contraints of GL(1|1); the net result was
equivalent to applying “gauge conditions” to the original OSp(1,1|2) generators:

J = iJ ⊕ |∂ ⊕=0,∂ + =1 , Q = J ⊕−|∂ ⊕ =0,∂ + =1

For the string, the gauge condition ∂ + = 1 simply removes the last vestige of X +,
whose oscillator modes were already eliminated by the string lightcone gauge. On the
⊕ ⊕
other hand, the condition ∂ ⊕ = 0 makes X ⊕ = X(+) +X(−) the sum of two conformally
covariant objects: With the elimination of the linear τ term in the expansion of X ⊕,
⊕
X(±) are periodic in their arguments, and have the usual mode expansion in terms of
exponentials only (no linear term).
As a result, the decomposition of Q as obtained from the lightcone becomes
trivial (J was easy anyway, since it’s quadratic): Relabeling the result for the string’s
lightcone Lorentz generators from subsection XIA3,
Z . .
⊕− dσ ⊕ − − ⊕
J =i (X X − X X )
2πα0
(where now the “i” comes from using the antihermitian form of the Lorentz spin),
. .⊕ 0⊕
separating X ⊕ into its (±) pieces, using their “chirality” X (±) = ±X(±) to convert
the τ derivative into a σ derivative, integrating by parts, and applying the definition
.
1
P̂(±) = √2α 0
(X ± X 0 )
678 XII. MECHANICS

to X − , we obtain r Z
2 dσ X ⊕ −
Q=i X P̂
α0 2π ± (±) (±)
In this form we can easily apply the Virasoro constraint, as solved in the lightcone,
√ 1
P̂ 2 = 0, P̂ + = κ 2α0 p+ ⇒ P̂ − = i √ (P̂ i )2
2κ 2α0
where we have applied ∂ + = 1. Finally, we relabel

⊕
X(±) = C(±)

for purposes of identification with the usual BRST procedure in terms of ghosts C and
antighosts B (see subsections XIA4-5); comparison of the (equal-time) commutation
relations then gives the further identification

[P̂ i(±) (1), P̂ j(±) (2)} = ∓iη ij 2πδ 0(2 − 1)

q
α0 ij
⇒ [P̂ i(±) (1), X j(±) (2)} = −i 2
η 2πδ(2 − 1)

{B(±) (1), C(±) (2)} = 2πδ(2 − 1)

q

⇒ B(±) = − α20 P̂(±)

The final result is then

Z
1X dσ 0
Q= 2
C(±) (− 12 P̂(±) ± iC(±) B(±) )
κ ± 2πα0

in agreement with direct first-quantization of this string in the conformal gauge, in

2
terms of the constraints P̂(±) . The ghosts can be separated into zero- and nonzero-
modes as q q
α0 ⊕ 0
1
C(±) = 2 c + 2 Y(±) , B(±) = 2κb ∓ α20 Y(±)

(For the closed string there is also an extra zero-mode in C and B, enforcing the
constraint that the “+” contributions to M 2 equal the “−”.)
Excercise XIIB8.1
By separating zero-modes in the string’s Q, and comparing with its generic
expression
Q = 12 c( − M 2 ) + S ⊕a ∂a + iS ⊕−1 M + S ⊕⊕ b

show that
X Z
dσ i 0j
ij
S =i ± Y Y
±
2π (±) (±)
 B. IGL(1) 679

1 X Z dσ 0j 2
S i−1
M= √ i i
Y(±) (Y(±) )
2κ 2α ±0 2π
Z
2 1 X dσ 0i 2
M = 0
(Y(±) )
2κα ± 2π

Of particular interest is the massless level: As mentioned in subsection XIB5,

using the fact that the Hilbert space of the closed string is the direct product of the
Hilbert spaces of open strings gives a simple analysis of the massless states of any
closed string, since the massless states of any open string are given by a vector (multi-
plet) plus perhaps some scalars in the nonsupersymmetric case. To find the complete
off-shell structure, including auxiliary fields and ghosts, we can either take the direct
product of the two lightcone representations and then add 2+2 dimensions, or first
add the 2+2 dimensions and then take the direct product of the two OSp(D−1,1|2)
representations. In the supersymmetric case the latter is more convenient, since the
procedure of adding dimensions to superspace is not yet understood, but quantization
of the vector multiplet is (at least for N=1, and probably for N=2, in D=4).
For example, for the bosonic closed string we just multiply two OSp vectors,
producing a tensor tij , which we can decompose into its symmetric traceless part
hij (graviton plus ghosts), antisymmetric part Bij (axion plus ghosts), and trace χ
(physical scalar):

tij |j i ⊗ |i i → hij = t(ij] − 2

η tk ,
D−2 ij k
Bij = t[ij), χ = ti i

(See subsection XIIA5.) However, we know that string theory prefers to treat fields
in the string gauge for Weyl invariance, where the action (kinetic term) is not diag-
onalized in the fields (until coordinate invariance is fixed also). This is understood
from this direct product structure: The “natural” string fields are

 t(ab) graviton
string gauge : t[ab] axion
 α
t α dilaton (T-duality invariant)
since duality affects only the X modes, while the diagonal fields (representations of
OSp(D−1,1|2)) are

 hab = t(ab) − D−2 ηab (t c + t γ )
2 c γ
graviton
normal gauge : B =t axion
 ab a [ab] α
χ =t a+t α physical scalar
Excercise XIIB8.2
Show that the above OSp analysis is consistent with the diagonalizing field re-
definitions of the low-energy string action found in excercise XIB5.2. Explain
680 XII. MECHANICS

the result in terms of the redefinitions

1 √
Φ2 √ → e2φ, Φ2g mn → −gg mn
−g

In the heterotic case, as mentioned in subsection XIB6, we take the product of

the real prepotential plus two chiral ghosts of super Yang-Mills with the usual vector
plus two scalar ghosts of bosonic Yang-Mills:

(V ⊕ φα ) ⊗ (Aa ⊕ Cα ) = Ha ⊕ (Vα ⊕ φaα ) ⊕ φ ⊕ φ(αβ)

The result is a vector prepotential Ha describing the physical supergravity and tensor
multiplets (in a string gauge, G = 1), a chiral scalar compensator (“superdilaton”) φ
appropriate for “old minimal” supergravity, first-generation ghosts (Sp(2) doublets)
Vα and φaα, and second-generation ghosts (an Sp(2) triplet, for the tensor multiplet)
φ(αβ). If the vector is accompanied by scalars, the closed string also has additional
vector multiplets:
(V ⊕ φα) ⊗ ϕI = VI ⊕ φIα

Background fields in string theory can be treated similarly to the previous sub-
section: Since any open string includes Yang-Mills, we again choose the Yang-Mills
ghost as the vacuum state. (In the bosonic case, even the tachyon can be obtained
from it, by what would have been considered an annihilation operator with respect to
the tachyonic vacuum.) Again vertex operators can be considered as additions to the
BRST operator, are in the BRST operator cohomology, create states (in the state co-
homology) from the vacuum, etc. For strings, such operators are local functions of the
string variables: They are associated with coupling an external field at a particular
value of σ (and τ ). These local, fermionic vertex operators with ghost number 1 (so
they can be added to Q) can be derived from related, integrated, bosonic operators
with vanishing ghost number:
Z
[Q, W] = 0, W = W ⇒ [Q, W ] = V 0

⇒ {Q, V }0 = 0 ⇒ {Q, V } = 0
(Usually it is convenient to use conformal field theory methods for such methods, so
R
we use the notation of subsection XIB7, which mainly means dropping ±’s.) Thus V
is the only part of V in the cohomology, the rest being a BRST variation: Moving
V to a different value of σ is a gauge transformation. Thus V (σ) has explicit σ
dependence, while Q does not:

QI = Q + V (σ)
 B. IGL(1) 681

corresponding to a vertex local in σ (as found in subsection XIB7).

The operators W are those that appear with the gauge-fixed Hamiltonian:

HI = H + W (σ) = {QI , Λ} = {Q, Λ} + {V, Λ}

where the gauge-fixing operator is

Z
Λ = (−B + f), Λ2 = 0

the first term in Λ giving the usual 12 (p2 + M 2 ) in H, while f is left arbitrary (as long
as Λ2 = 0 is preserved) to allow more general gauges. Thus, either of V or W can be
derived from the other:

V 0 = [Q, W ], W = {Λ, V } (⇒ {Q, V } = [Λ, W ] = 0)

In general we have

1
i 2π [Ĝ(1), χ(2)] = δ(2 − 1)χ0 (2) + wδ 0(2 − 1)χ(2)
Z
0
⇒ i[H, χ] = χ , H = Ĝ

for any conformally covariant operator, so assuming V is one,

Ĝ(σ) = {Q, (−B + f)(σ)}, ĜI (σ) = {QI , (−B + f)(σ)}

⇒ [Q, W ] = [Q, {Λ, V }] = [{Q, Λ}, V ] = [H, V ] = V 0

in agreement with the previous. (Similarly, we can plug V 0 = [Q, W ] into {Λ, V 0 } to
show the consistency of W = {Λ, V }.) In the notation of subsection XIB7, Ĝ are the
conformal generators T±± , and ĜI are T̂±± , while

f (σ) = 2πδ(σ − σ0 )W (σ0)

{V (σ0), (−B + f)(σ)} = W

The simplest examples come from adding a background directly to the Virasoro
operators (as for gravity in subsection XIB4): Then for the bosonic string
Z Z
0
V = CW (X, P ), 1 2
Q = C(− 2 P̂ + iC B), Λ = −B,

If we assume that W has conformal weight 1 with respect to the gauge-fixed conformal
(Virasoro) generators, so that W is conformally invariant, then V has conformal
weight 0.
682 XII. MECHANICS

9. Relation to OSp(1,1|2)
For comparison to OSp(1,1|2), we perform a unitarity (gauge) transformation on
the IGL(1) action. We first define an almost-inverse of S ⊕⊕ : Since S ⊕⊕ annihilates
states with s3 = s, where we define

− 12 S αβ Sαβ = 4s(s + 1), S 3 |si = 2s3 |si

(so that s and s3 take their usual integer or half-integer values), we can define S ⊕⊕−1
such that
S ⊕⊕−1 S ⊕⊕ = 1 − δs3 ,s , S ⊕⊕ S ⊕⊕−1 = 1 − δs3 ,−s

S ⊕⊕ S ⊕⊕−1 S ⊕⊕ = S ⊕⊕ , S ⊕⊕−1 S ⊕⊕S ⊕⊕−1 = S ⊕⊕−1

We then apply the transformation

Q → Qdiag = UQU −1

Q = −cK + Q⊕ + S ⊕⊕ b, ln U = c[{S ⊕⊕−1 , Q⊕} − S ⊕⊕−1 S ⊕⊕ Q⊕ S ⊕⊕−1 ]

The exponent of U is nilpotent from the c, so it generates only a linear term, U =

1 + ln U. Using the commutation relations from subsection XIIA2

[S ⊕⊕, Q⊕] = 0, (Q⊕)2 = KS ⊕⊕

we find

Qdiag = cδs3 ,s (−K + Q⊕S ⊕⊕−1 Q⊕ )δs3 ,−s + (cbQ⊕δs3 ,s + bcδs3 ,−s Q⊕ ) + S ⊕⊕ b

Now we apply the identity

[S 3, A] = 0 ⇒ δs3 ,s Aδs3,−s = Aδs0

since any matrix element between hs, s3 |...|s0, s03i gives s = s3 = s03 = −s0 → s =
s0 = 0, as well as the facts

S ⊕⊕ Q δs0 = [S ⊕⊕, Q ]δs0 = 2iQ⊕δs0 ⇒ S ⊕⊕−1 Q⊕δs0 = − 12 iQ δs0

{Qα, Qβ }δs0 = 0

This yields the final result

Qdiag = −c(K + 14 Qα Qα)δs0 + (cbQ⊕δs3 ,s + bcδs3 ,−s Q⊕) + S ⊕⊕ b

 B. IGL(1) 683

Excercise XIIB9.1
Use the commutation relations of S αβ , as well as − 12 S αβ Sαβ = 4s(s + 1), to
derive
S S ⊕⊕ = 4[s(s + 1) − s3 (s3 − 1)]

from which follows the explicit expression

1 1 − δs3 ,s
S ⊕⊕−1 = S = S
S S ⊕⊕ 4(s − s3)(s + s3 + 1)
R
Integrating over c in the action Sdiag = − 1 T
2 Φ (−1)
J−1
Qdiag Φ as in subsections
XIIB3-4, we find the Lagrangian
3 −1
δs3 ,−s Q⊕φ − 12 ψ T (−1)S S ⊕⊕ ψ
3
Ldiag = 12 φT (K + 14 QαQα )δs0φ + iψ T (−1)S

(Note that in a product of the form ψ T χ a state of eigenvalue s3 multiplies one of

eigenvalue −s3 , since (S 3 )† = (S 3 )T = −S 3.) We now see that in this action only the
s = 0 (physical) part of φ appears in the φφ term, while only the s3 = −s (“minimal”)
part (including physical) appears in the φψ term. The only part of ψ that appears in
the φψ term is the s3 = s (minimal) part of ψ (the “antifields” to the corresponding
ones in φ), while all, but only, the remaining (“nonminimal”) part of ψ appears in
the ψψ term. In particular, the φφ term is recognized as the OSp(1,1|2) action of
subsection XIIA3. The terms involving ψ can be eliminated by ψ’s gauge invariance
and field equation, and ψ contains no propagating degrees of freedom (fields with
equations of motion), as can be seen by the methods used to analyze the cohomology
in subsection XIIB4. However, the auxiliary fields ψ, and the ghosts in (the s 6= 0
part of) φ to which they couple, are useful in gauge fixing, as we’ll see in the next
section.
Again looking at the example of the vector:

S ⊕⊕ = 2|⊕ ih⊕ | ⇒ S ⊕⊕−1 = 14 S = 12 | ih |

Q⊕ = (|⊕ iha | − |a ih⊕ |)∂a ⇒ ln U = −i 12 c(| iha | + |a ih |)∂a

The transformation is then (cf. subsection XIIB3)

Qdiag = 12 c( − |aihb |∂a ∂b)δs0 + cb|⊕iha |∂a − bc|aih⊕ |∂a + 2|⊕ ih⊕ |b
ˇ2
Ldiag = 18 (Fab)2 + iǍ · ∂C + C̃

The BRST transformations now simplify to

QAa = −∂aC, QC = 0, ˇ
QC̃ = −2iC̃
684 XII. MECHANICS

ˇ =0
QǍa = −i 12 ∂ b Fba, QČ = ∂ · Ǎ, QC̃

Excercise XIIB9.2
Find the ghosts and simplified BRST transformations for massless spin 2.

REFERENCES
1 M. Kato and K. Ogawa, Nucl. Phys. B212 (1983) 443;
S. Hwang, Phys. Rev. D28 (1983) 2614;
K. Fujikawa, Phys. Rev. D25 (1982) 2584:
1st-quantized BRST (for strings).
2 W. Siegel, Phys. Lett. 149B (1984) 157, 151B (1985) 391:
gauge-invariant actions from 1st-quantized BRST.
3 E. Witten, Nucl. Phys. B268 (1986) 253;
A. Neveu, H. Nicolai, and P.C. West, Phys. Lett. 167B (1986) 307:
BRST operator as kinetic operator.
 C. GAUGE FIXING 685

....................... .......................
....................... C. GAUGE FIXING .......................

Although the quantum mechanical BRST operator is clearly useful for gauge fix-
ing, its relation to the second-quantized BRST we applied in chapter VI is not obvious,
since the latter BRST operator does not include the gauge-invariant action. Here we
relate the two, and extend the former to interacting field theories. In particular, we
show how the ΦQΦ action leads directly to the gauge-fixed kinetic term as simply as
it led to the gauge-invariant one, without applying any transformations.

1. Antibracket
In the usual Hamiltonian formalism we work in a phase space (q, p) on which is
defined a Poisson bracket, useful for studying symmetry properties and equations of
motion in the classical theory, and for relating to the commutator of the quantum
theory (see subsections IA1-2). We want to interpret the present case of interest as
an analogous phase space, for which the fields φ (in Φ = φ − icψ) correspond to q
and the antifields ψ to p. This automatically follows from the lightcone commutator
of subsection XIIA1, by the same steps used to derive the IGL(1) algebra and inner
product Z
hΨ|Ξi = −i(−1)Ψ dx dc ΨT (x, c)(−1)J(c) Ξ(x, c)

in subsections XIIB1-2: We thus define this generalization of the Poisson bracket (see
subsection IA2) in terms of the inner product as
* ←  
δ  δ
(f[Φ], g[Φ]) = f ◦ g, ◦=η IJ
δΦI  δΦJ

where we have expanded the column vector Φ over a basis in the usual way (see
subsections IB1,5),
Φ = |I iΦI , ΦT = ΦI hI |

hI |J i = η IJ = (−1)I η JI , (η IJ )* = η JI , η IK ηJK = δIJ

etc., and the indices I, J (not to be confused with the ghost-number operator J(c)
appearing in the definition of the inner product above) run over all indices on the
field, which determine the statistics of the corresponding component as (−1)I . (Thus
Φ is always bosonic, the statistics of the fields coming always from expansion over |I i.)
This generalized commutator “( , )” is called an “antibracket” because of the unusual
statistics associated with it, following from the same unusual statistics of the inner
product. (The ordering of indices on η IJ in the antibracket is the opposite of usual
686 XII. MECHANICS

to take into account the extra sign factor from the intervention of the anticommuting
“ | ” between the two Φ’s.) Plugging in the definition of the inner product, we have
more explicitly
Z ←
δ δ
◦ = i dx dc I
η JI (−1)J(c) J
δΦ (x, c) δΦ (x, c)
Note that while the inner product is defined between two functions of the coordinates,
the antibracket is defined between two functionals of Φ, f and g, which don’t depend
explicitly on (x, c) (although we can specialize to cases where they depend on other
values of the coordinates (x0, c0)). Thus the orbital term cb in J(c) acts only on the
argument c of ΦJ (x, c) (and not on g), while the spin term S 3 acts only on the index
J of ΦJ (x, c).
We have used the fact that δ/δΦ is antihermitian, as follows from the fact that
the graded commutator between it and Φ is always a commutator, since they always
have opposite statistics:
 †
δ J 0 0 δ J 0 0
Φ (x , c ) = Φ (x , c )
δΦI (x, c) δΦI (x, c)
 
δ
= I
, Φ (x , c ) = δIJ δ(x0 − x)δ(c0 − c)
J 0 0
δΦ (x, c)
 †  T
δ δ δ
⇒ = =−
δΦ δΦ δΦ
Thus,
 †  †   ←
δ † δ δ δ
f = , f† = f, − =f
δΦ δΦ δΦ δΦ
Other properties of the bracket follow directly from those of the inner product:

(−1)(f,g) = (−1)f +g+1

(f, ga) = (f, g)a, (af, g) = a(f, g)

(f, g) = −(−1)(f +1)(g+1)(g, f)

(f, gh) = (f, g)h + (−1)(f +1)g g(f, h)

(−1)(f +1)(h+1)(f, (g, h)) + cyc. = 0

(f, g)† = (g †, f † )

(for some commuting or anticommuting constant a). Thus, the bracket has the exact
opposite symmetry as the inner product (as is the case with the usual brackets):
R
It would be symmetric in its two arguments if not for the dc that sits effectively
 C. GAUGE FIXING 687

between the two arguments. Most of the properties follow from this fact, and that the
signs obtained from pushing things around are determined by moving things naively
while treating the “ , ” in the middle as anticommuting. Furthermore, the existence
of a bracket with these properties allows the definition of a Lie derivative,

LAB ≡ (A, B)

Excercise XIIC1.1
Find all the usual properties of this derivative (statistics, linearity, distribu-
tivity, hermiticity, algebra, etc.), and relate to the usual Lie derivative.
We also have functional identities such as
δ
hΦ|Ψi = iδIJ (−1)J(c) ΨI (x, c)
δΦI (x, c)
δ
(ΦI (x, c), f[Φ]) = −i(−1)J(c)η IJ f[Φ]
δΦJ (x, c)
from which follow

(ΦI (x, c), ΦJ (x0 , c0)) = −i(−1)J(c)η IJ δ(c0 − c)δ(x0 − x) = −i(−1)S η IJ (c + c0 )δ(x − x0)
3

as well as
(Φ(x, c), hΦ|Ψi) = Ψ(x, c), (hΨ|Φi, hΦ|Ξi) = hΨ|Ξi

Here Ψ and Ξ are wave functions in the same space as Φ, but need not be taken as
bosonic (or real): We can even take them as functionals of Φ when applying the chain
rule, using the above expressions for the terms where the δ/δΦ’s don’t act on them.
Expressions quadratic in Φ will be used to perform the second-quantized (or just
classical field theoretic) version of linear first-quantized transformations:

OA ≡ 12 hΦ|AΦi ⇒ (OA, OB ) = O[A,B}, AΦ = (Φ, OA)

Z Z
I δ δ
df = − dx dc (−1) dΦI
I
f ⇒ (OA, f) = dx dc (−1)I (AΦI ) I f
δΦ δΦ
where A and B must satisfy

[(−1)J(c) A]T = (−1)J(c) A ⇒ A = −(−1)J(c) AT (−1)J(c)

to give nontrivial contributions when appearing symmetrically between the two fac-
tors of Φ. Corresponding group elements come from exponentiating bosonic first-
quantized generators, yielding fermionic second-quantized generators:

δf = (OA, f) = LOA f ⇒ f 0 = eLOA f

688 XII. MECHANICS

(−1)A = 1 ⇒ (−1)OA = −1 ⇒ (OA, f) = −(f, OA)

Clearly, the latter relations must hold when replacing OA with their nonlinear second-
quantized generalizations. We then find (also for bosonic A)

AΦ = (Φ, OA ) = −(OA, Φ)

where the minus sign is the usual from translating first-quantized to second-quantized
language (see subsection IC1).
Expanding (Φ, Φ) in c as
3
ΦI = φI − icη IJ (−1)S φ̌J

we find
(φ̌I (x), φJ (x0)) = −(φJ (x), φ̌I (x0 )) = δIJ δ(x − x0 )

This allows us to reexpress the antibracket as

 ← 
Z ←
δ δ δ δ 
◦ = − dx (−1)I  +
δ φ̌I δφI δφI δ φ̌I

For example, the antibrackets of the component fields for the vector are

ˇ −|⊕iČ)
Φ = |i iΦi = |i i[ηji Aj −ic(−1)S Ǎi ] = (|a iAa −i| iC +i|⊕iC̃)−ic(|aiǍa −| iC̃
3

(Ǎi (x), Aj (x0 )) = −(Aj (x), Ǎi(x0)) = δij δ(x − x0 )

⇒ (Ǎa, Ab) = ηab δ, (Č, C) = δ, ˇ C̃) = δ

(C̃,

where now “i” refers to the OSp(D−1,1|2) index (and we use C = A⊕, C̃ = A ,
ˇ = Ǎ ).
Č = Ǎ⊕, C̃

2. ZJBV
To prove gauge independence of the path integral, it’s useful to draw an anal-
ogy of relativistic quantum mechanical BRST to second-quantized BRST. (We’ll see
below that this is not just an analogy, but an equivalence.) Translating the BRST
quantization of subsection VIA2 into path integral language, the general Lagrangian
path integral for BRST quantization in quantum physics is
Z
0
A = Dq e−iS , S 0 = S + {Q, Λ}
 C. GAUGE FIXING 689

where q is all coordinates, including ghosts. While S and Λ depend only on q, the
BRST operator is linear in the conjugate momenta p — It generates a coordinate
transformation:
[Q, q m} = −iδQq m ⇒ Q = (δQ q m)pm

Here the index “m” includes all dependence of q, including time. (We saw a more
explicit expression of this result in subsection VIA2, assuming the constraints Gi are
themselves linear in the physical momenta.) We also have

[Q, S] = 0

where S can include not only the gauge-invariant action, but arbitrary additional
gauge-invariant pieces. Such pieces can be used to construct states in the BRST
cohomology from the vacuum. (This is the path-integral translation of the operator
construction given in subsection VIA1.) It can also include pure BRST variations,
{Q, Λ0}. Thus, to prove gauge independence of A, we need only prove the vanishing
of its variation under infinitesimal change of Λ,
Z
Dq e−iS {Q, δΛ} = 0

But this is trivial, since Q acts as a total derivative, and [Q, S] = 0. More generally,
we require only
← ←
0 = 1 · Q − i[Q, S] = −i(δQq m) ∂ m − (δQq m )∂m S
←
where ∂m = ∂/∂q m, and “1 · Q” means the derivatives in Q act backwards onto the
1 (and itself), as found by integration by parts. In cases we have considered (and
←
almost always), 1· Q and [Q, S] separately vanish. More generally, since there is a 1/h̄
multiplying S implicitly, nonvanishing values would require a “quantum correction”
to S. We can also write this condition as
←
e−iS Q = 0

Furthermore, we can write

← ∂ 2Q
1 · Q = −i
∂pm ∂q m
These manipulations can be applied to the field theory expressions J for group
generators, as found in subsection XIIA1 for the lightcone: The BRST operator as
found from these generators is

S = 12 hΦ|iQΦi, Q2 = 0 ⇔ (S, S) = 0
690 XII. MECHANICS

Then a unitary transformation can be implemented by exponentiating the infinitesi-

mal transformation

δQ = [G, Q] ⇔ δS = 12 hΦ|i(δQ)Φi = (G, S), G = 12 hΦ|GΦi

Q0 = eGQe−G ⇔ S 0 = 12 hΦ|iQ0Φi = eLG S

We want to implement gauge fixing by performing a unitary transformation on

Q and then evaluating S at the antifields ψ = 0:
Z
0
A = Dφ e−iS |ψ=0 , S 0 = eLΛ S ⇒ Sgf = (eLΛ S)|ψ=0

By similar manipulations to the BRST case, we see that gauge independence means
Z Z
0 = Dφ e (S, δΛ) = i Dφ (e−iS , δΛ)
−iS

where we evaluate this expression at ψ = 0, and we have again included arbitrary

gauge-invariant pieces in S. We thus obtain gauge independence from (S, S) = 0, or
more generally (again using integration by parts)
Z Z
I δ2 −iS I δ S
2
0 = dx (−1) e = dx (−1) + i 12 (S, S)
δ φ̌I δφI δ φ̌I δφI
This is the approach to BRST of Zinn-Justin, Batalin, and Vilkovisky (ZJBV).
Excercise XIIC2.1
Find the unitary transformation, in ZJBV language, that transforms the
“untransformed” action for the massive vector (that which gives the gauge-
invariant action upon dropping antifields) into the action that has only the
vector field (and not the scalar) upon dropping antifields.
Writing the BRST transformations in this second-quantized ZJBV notation will
allow us to gauge fix interacting theories (found by adding interaction terms to the
free S) in a gauge-independent way. In particular, it proves the equivalence of the
manifestly unitary lightcone gauge (which has no ghosts, only physical degres of free-
dom) to the manifestly Lorentz covariant Fermi-Feynman gauges (where the kinetic
operator is simply − m2 ). Specifically, the ΦQΦ action is already unitarily trans-
formed to the Fermi-Feynman gauge: Keeping just the φ terms, we have for a bosonic
theory
Z Z
3
SF F = S|ψ=0 = − dx dc 2 Φ (−1) c 2 ( −m )Φ = − dx 12 φT (−1)S 12 ( −m2)φ
1 T J−1 1 2

In other words, the Fermi-Feynman kinetic term is just the sum over all fields (but
not antifields) of a − m2 term (using the OSp(D−1,1|2)-invariant inner product:
 C. GAUGE FIXING 691

3
the (−1)S is just a sign, and can be absorbed by a field redefinition). This can
also be seen from the result for the complete ΦQΦ action after dropping the antifield
terms: For example, for a vector the gauge-fixed (free) action is simply (see subsection
XIIB3)
LF F = − 14 Aa Aa − iC̃ 12 C

(The Fermi-Feynman gauge for fermions also gives a − m2 kinetic term, but with
an infinite number of ghosts; this may be useful for supersymmetry.)
Excercise XIIC2.2
Let’s again consider arbitrary-rank antisymmetric tensors (see excercises
XIIA5.1 and XIIB3.1):
a Find the Fermi-Feynman actions.
b Do the same for the massive case. (Note: There are more fields.)
On the other hand, we saw in subsection XIIB9 that a unitary transformation,
and evaluation at ψ = 0, gave the gauge-invariant OSp(1,1|2) action in terms of just
the physical fields:

Λ0 = 12 hΦ|c[{S ⊕⊕−1 , Q⊕} − S ⊕⊕−1 S ⊕⊕Q⊕ S ⊕⊕−1 ]Φi ⇒ S 0 = Sdiag

Z
= dx [ 12 φT (K + 14 Qα Qα)δs0 φ + iψ T (−1)S −1 δs3 ,−s Q⊕φ − 12 ψ T (−1)S S ⊕⊕ ψ]
3 3
Sdiag
Z
Sgi = Sdiag |ψ=0 = dx 12 φT (K + 14 Qα Qα)δs0 φ

In the usual gauge-fixing approach, we would start with Sdiag and do the inverse
transformation to obtain the Fermi-Feynman gauge:

Λ = −Λ0 ⇒ SF F = (e−LΛ0 Sdiag )|ψ=0

The same Λ can be used in the interacting case, since the effect on the quadratic
piece of the action will be the same. Thus, to apply the usual ZJBV procedure we
can either start with Sdiag and apply some Λ 6= 0 sufficient to fix the gauge, or we can
start with S (the one we found from quantum mechanical BRST) and apply some
equivalent Λ, or no Λ at all (for Fermi-Feynman gauge).
To compare with the lightcone gauge, we start with
Z
Slc = dx [− 14 φ(−1)S φ − iψ(−1)S −1 S ⊕− ∂ +φ]
3 3

which was itself obtained by unitary transformation from S (see subsection XIIB4),
R
and make a further unitary transformation, of the form Λ = dx 12 φA φ, that has
692 XII. MECHANICS

the effect ψ → ψ + A φ for A such that all terms in φ φ containing auxiliary fields,
and thus also pure-gauge fields, are canceled. For this transformed Slc we then have
Z
Slc,diag |ψ=0 = − dx 14 φ δ(S AB )φ

where the projection operator δ(S AB ) picks out the singlets of S AB (A = (±, α)),
i.e., the transverse (physical) degrees of the light cone. A similar procedure can be
applied in the interacting case.
For the example of the vector:

ln U = −i 12 c(| ih+ | + |+ih |)(∂ +)−1

Qlc = 12 c − S ⊕− ∂ + → Qlc,diag = 12 c |i ihi | − S ⊕− ∂ +

Llc → Llc,diag = − 14 Ai ˇ +A− − iǍ− ∂ +C
A + C̃∂
i

which has just the transverse (lightcone) degrees of freedom when the antifields are
dropped:
Llc,gf = − 14 Ai Ai

3. BRST
In subsection VIA2, we saw that BRST could be used to gauge fix by adding a
BRST variation to the gauge-invariant Lagrangian. In that case, physical states are
those that are not only in the BRST cohomology, but also satisfy the equations of
motion.
On the other hand, for relativistic mechanics we saw that the equations of motion
are rather redundant, since τ is unphysical, and so p2 (+m2) = 0 is already included
as a constraint, and contained in the quantum mechanical BRST operator. In fact,
we have seen how in the most general case of a free field the correct spectrum is
specified by just the cohomology of the quantum mechanical BRST operator.
We therefore want to identify the quantum mechanical BRST cohomology condi-
tion with the combination of the second-quantized BRST cohomology condition and
the wave equation. This essentially has been accomplished in subsection XIIC2 by
decomposing Qdiag with respect to c, as we’ll now see by some further analysis.
From subsection XIIB9 we have

Qdiag = −c(K + 14 Qα Qα)δs0 + (cbQ⊕δs3 ,s + bcδs3 ,−s Q⊕) + S ⊕⊕ b

where the first term gives Lgi in terms of the physical part (s = 0) of φ, the second
term gives the minimal BRST transformations in terms of the minimal (anti)fields,
 C. GAUGE FIXING 693

and the last term adds the nonminimal stuff needed for fixing to general gauges. In
particular, we see that the BRST transformation of the physical fields are

δ(δs0φ) ∼ δs0Q⊕φ = δs0 Q⊕(δs,1/2δs3 ,−1/2φ)

(Note that this differs somewhat from the expression found from Q, since the trans-
formation to Qdiag is effectively a redefinition of ψ, adding to it a piece proportional
to an operator on φ.)
Thus the only occurrences of the physical fields in Qdiag are in the term that
gives the gauge-invariant action and the term that gives their BRST transformation;
the remaining terms introduce nonminimal fields, as well as account for the BRST
transformations of the ghosts. But this is the definition of BRST: Take the classical
action in terms of physical fields, construct the BRST transformation from the gauge
transformation that leaves the classical action invariant, add terms to the BRST
operator that insure its nilpotency on these ghosts, and add nonminimal terms to
allow gauge fixing. We have just seen that Qdiag is exactly of this structure, where
gauge fixing gives the desired Fermi-Feynman gauge by unitary transformation to Q
(which becomes a canonical transformation in second-quantized language, using the
antibracket).
All that is left to see is how the ZJBV combination of the gauge-invariant action
with the BRST operator is equivalent to ordinary BRST. Expanding in antifields,
ZJBV gives the gauge fixed action as

S = S0 +ψm(Qφm )+ψnmψnm ⇒ Sgf = S0 +(δΛ/δφm)(Qφm )+(δΛ/δφnm)(δΛ/δφnm )

where we have used the fact that the three terms in Qdiag contain only the physical,
minimal (“m”; including physical), and nonminimal (“nm”) fields, respectively. In
the usual ZJBV and BRST formalisms, derived from BRST without antifields, there
is no ψ 2 term, since this generates a BRST transformation

iQφnm = (S, φnm ) ∼ ψnm

One instead introduces further nonminimal fields, the “Nakanishi-Lautrup fields”,

such that
Qφnm = φN L

and use an extended gauge-fixing function

Λ̂ = Λ + φnm φN L
694 XII. MECHANICS

Then the gauge fixed action is

Sgf = S0 + {Q, Λ̂} = S0 + (δΛ/δφm)(Qφm ) + (δΛ/δφnm )φN L + φ2N L

After eliminating the NL fields by their (algebraic) equations of motion, we obtain

the same result as found from ZJBV. (Of course, the NL fields can also be introduced
directly into the ZJBV formalism, but are redundant for purposes of finding Fermi-
Feynman gauges.)
Consider the special case of Yang-Mills: Generalizing our results for free Yang-
Mills to the interacting case, making use of the BRST transformations of subsection
VIA4, we have
ˇ 2 + iǍ · [∇, C] − ČC 2
LZJBV = 18 (Fab )2 + C̃
The basic antibrackets are

(Ǎa, Ab) = ηab δ, (Č, C) = δ, ˇ C̃) = δ

(C̃,

From the general relations we saw earlier, and the definition of S in terms of Q for
the free case, we have
iQΦ = (Φ, S) = (S, Φ)
Since in the above we have pulled out factors to the left of the fields, as
3
Φ = |i iΦi = |i i[ηji Aj − ic(−1)S Ǎi]

we pull them out of the left of the antibracket, to obtain

i(Qφ)I = (S, φI ), i(Qφ̌)I = (S, φ̌I )

where as before (Qφ)I , etc., means to evaluate the corresponding component of QΦ

and introduce the corresponding signs for effectively pulling those factors to the left.
We then find the previous results for the BRST transformations of the fields (by
construction), but also those of the antifields:
ˇ
QAa = −[∇a, C], QC = iC 2, QC̃ = −2iC̃

QǍa = −i 12 [∇b , Fba] + i{C, Ǎa}, QČ = [∇·, Ǎ] + i[C, Č], ˇ =0
QC̃
(Remember that the funny signs of the antibracket come from its symmetry, plus
treating the comma in “( , )” as anticommuting. Note the generic terms i[C, }.)
Excercise XIIC3.1
Generalize the above results for the action and BRST transformations with
antifields when Yang-Mills is coupled to matter.
 C. GAUGE FIXING 695

BRST was described in a different way in subsection VIA4: Here we apply quan-
tum mechanical BRST, and find it equivalent to applying the ZJBV form of BRST
to second-quantization. The ZJBV action consists of the gauge-invariant action, plus
the antifields times the BRST transformations of the fields, plus (antifield)2 terms.
The difference between the BRST transformations obtained by the general methods
of subsection VIA1 as applied to second-quantization, and those found in this chap-
ter by applying OSp methods to first-quantization of relativistic systems, is that the
Nakanishi-Lautrup field is treated as a field in the former approach and as an an-
tifield in the latter. The two give equivalent results: The latter uses fewer fields,
but is slightly more restricted in choices of gauge; however, this restriction is avoided
in practice. (More “nonminimal” fields can be added to allow more general gauge
choices in either case.) For example, the ZJBV action for the former treatment of
Yang-Mills can be obtained from that for the latter by the replacement
ˇ 2 → C̃B
C̃ ˇ

The gauge-fixed action is the canonically transformed action (with respect to the
antibracket) evaluated at vanishing antifields:
Sgf = eLΛ SZJBV |
Consider Yang-Mills in the most common type of gauge, where some function of A is
fixed. From the usual BRST approach (see subsection VIA4), or the ZJBV approach
with B, we find
Z
∂f
Λ = tr 12 C̃[f(A) + 12 αB] ⇒ Lgf = Lgi − 12 B[f(A) + 12 αB] − 12 iC̃ · [∇, C]
∂A
while in the ZJBV approach without B we have
Z
∂f
Λ = tr 12 C̃f(A) ⇒ Lgf = Lgi + 14 f(A)2 − 12 iC̃ · [∇, C]
∂A
which is equivalent to the previous for positive α (after elimination of B).

REFERENCES
1 Siegel and Zwiebach, loc. cit. (XIIA).
2 Siegel, loc. cit.
3 J. Zinn-Justin, in Trends in elementary particle theory, eds. H. Rollnik and K. Dietz
(Springer-Verlag, 1975) p. 2:
introduced antifields, antibracket, etc.
4 I.A. Batalin and G.A. Vilkovisky, Phys. Lett. 102B (1983) 27, 120B (1983) 166; Phys.
Rev. D28 (1983) 2567, D30 (1984) 508; Nucl. Phys. B234 (1984) 106; J. Math. Phys.
26 (1985) 172:
generalized Zinn-Justin’s approach.
5 W. Siegel and B. Zwiebach, Nucl. Phys. B299 (1988) 206:
2nd-quantized ZJBV from 1st-quantization.
696

............................... ...............................
............................... AfterMath ...............................

Conversions to other common conventions

ηab → −ηab, γa → √1 γa ,
2
S → −S

g2 dD x g2
g2 → [via 1
2 (m
2
− ) → m2 − , → dD x] or nonabelian
8π 2 (2π)D/2 16π 2

Natural (Planck) units

c = h̄ = k = κ = 1 (G = π)
1 8π 2 2π 2π
2
= 2 = 2 = = 861.022516(39), H −1 = 1.5(2) × 1061
e ef t em α
1 kg = 2.59194(17) × 107 , 1 m = 1.096782(70) × 1035 , 1 s = 3.28807(21) × 1043

1 K = 3.98170(25) × 10−33 , 1 GeV = 4.62054(30) × 10−20

Indices
a, b, c, ... — (flat) vector
i, j, k, ... — transverse (D−1 or D−2) vector or internal
m, n, p, ... — (curved) vector or large summation
A, B, C, ... — (flat) super or conformal vector
I, J, K, ... — internal
M, N, P, ... — (curved) super
A, B, C, ... — conformal spinor
α, β, γ, ...; µ, ν, π, ... — spinor (usually 2-valued) or fermionic
ι, κ — internal
0 — time
−1 — mass (dimensional reduction)
±; t, t̄ — lightcone (longitudinal; transverse)
⊕, — spinor or spacecone reference line
Integration Z Z Z Z
dD x dD p
dx ≡ , dp ≡
(2π)D/2 (2π)D/2
δ(x − x0) ≡ (2π)D/2δ D (x − x0 ), δ(p − p0 ) ≡ (2π)D/2 δ D (p − p0 )

hx|x0 i = δ(x − x0), hp|p0 i = δ(p − p0 ); hx|pi = eip·x, hp|xi = e−ip·x; pa = −i∂a
0
δ(p − p )
on − shell : hp|p0 i =
2πδ[ 12 (p2 + m2 )]
 697

LORENTZ (I,II)
−m2 = p2 = pa pb ηab = −(p0 )2 + (p1 )2 + (p2 )2 + (p3 )2 = −2p+ p− + 2pt p̄t

E = p0 = −p0 ; p± = √1 (p0
2
± p1 ), pt = √1 (p2
2
− ip3 )
−ds2 = dx2 = dxa dxb ηab , pa ds = m dxa , pa dτ = dxa

2-spinor .  + t


Cαβ = −C αβ = C . . = 0i −i
0
; pαβ
= ppt pp̄−
αβ
.
β . β .
ψα = ψ Cβα , ψα = ψ C .; ψ 2 = 12 ψ α ψα = iψ ⊕ψ = −iψ⊕ψ
βα
. . . .
ψ̄ α = (ψ α)† ⇒ (ψα)† = −ψ̄α., (ψ 2)† = ψ̄ 2 = 12 ψ̄ αψ̄α. = iψ̄ ⊕ψ̄
.
A[αβ] = Aαβ − Aβα = Cαβ C Aγδ , γδ
A[αβγ] = 0; V = −2 det V = V
2 αβ
V .
αβ

η . . = Cαβ C . .,  . . . . = i(Cαβ Cγδ C . .C . . − Cαδ Cβγ C . .C . .)

αα,β β αβ αα,β β,γ γ,δ δ αδ β γ αβ γ δ
.
V = Vα β ⇒ V * = −V β α., V W * + W V * = (V · W )I
0123 = −0123 = 1, (V, W, X, Y ) = i tr(V W *XY * − Y *XW *V )

4-spinor
 
ψα .
Ψ= , Ψ̄ = Ψ† Υ = (χα ψ̄ α); −γ a γ b = 12 η ab + S ab
χ̄α.
 .  ..  !
0 ∇α β √ 0 C̄ αβ −iδαβ 0
∇
/ = , Υ= 2γ0 = , γ−1 = √1 .
∇β α. 0 C αβ 0 2
0 iδ . β
α
γ γa = −2,
a a
γ a/γa = a/, γ a//bγa = a · b,
a a
γ a//b/cγa = c//ba/
tr(I) = 4, tr(a//b) = −2a · b, tr(a//b/c/d) = a · b c · d + a · d b · c − a · c b · d
. .
Ψ = |αiψα + |α ]χ̄α., Ψ̄ = Ψ† = χαhα | + ψ̄ α [α.|
.
γ . = −|αi[ .| − | .]hα|; Π+ = |α ihα |, Π− = |α ihα.|
αβ β β

Superspace
 .   
∂ ∂ .
qα = −i − θ̄ p . ,
1 β
q̄α. = −i .−
1 β .
2 θ pβ α ; (q α)† = q̄ α
∂θα 2 αβ
∂ θ̄ α

∂ . ∂ .
dα = α + 2 θ̄ p .,1 β
d¯α. = 1 β
.α + 2 θ pβ α.; (dα )† = −d̄α
∂θ αβ
∂ θ̄
. . . Z
{q α, q̄ β } = {dα , d¯β } = pαβ , d2 θ = d2 = 12 dα dα , d¯2 d2 d¯2 = 1
2 d¯2
698

ACTIONS (III,IV,XII)
Z
. .
L= − 12 q 2 g(q) + qA(q) + U(q), S= dt L
Z
.
LH = −qp + H(q, p), SH = dt H − dq p
Z Z
−iS
A= Dφ e Ψ; W ick : A= Dφ e−S Ψ, S≥0

Mechanics
OSp(1,1|2):

K = − 12 ( − m2), Qα = S αa∂a + S α−1 im [+γ̃ α(γ̃ − − γ̃ + K)]

Ŝ αβ = S αβ (− 12 γ̃ (αγ̃ β) ), − 12 Ŝ αβ Ŝαβ = 4s(s + 1)

Z
S= dx Lgi , Lgi = 12 φT Kgi φ, Kgi = 12 (− + m2 + 12 QαQα ) (Ŝ αβ φ = 0)

IGL(1):

Q = 12 c( − m2 ) + S ⊕a ∂a + S ⊕−1 im + S ⊕⊕b (+S̃ ⊕− ), J = cb + S 3 (+S̃ 3 )

Z
S = −(−1) Φ
dx dc 12 ΦT (−1)J−1 QΦ = 12 hΦ|iQΦi, Sgi = S|JΦ=0
Z
3
SF F = S|bΦ=0 = − dx 12 φT (−1)S 12 ( − m2)φ

Quantum ChromoDynamics

Gi † = Gi , [Gi , Gj ] = −ifij k Gk , (Gi ψ)A = (Gi )AB ψB

∇a = ∂a + iAa = ∂a + iAai Gi , −i[∇a, ∇b ] = Fab = Fab i Gi = ∂[aAb] + i[Aa, Ab]

L= 1
8g 2
tr F abFab + L(∇, ψ), trD (Gi Gj ) = δij , trA(Gi Gj ) = 2Nδij

QAa = −[∇a, C], QC = iC 2, QC̃ = −iB, QB = 0, Qφ = iCφ (for δφ = iλφ)

Z
Sgf = Sgi −iQΛ, Λ = tr 12 C̃(f + 12 αB) ⇒ Lgf = Lgi − 12 B(f + 12 αB)+ 12 iC̃(δf)|λ=C
. .
LM ajorana = ψ αi∇αα ψ̄α. + √ (ψ α ψα
m
2 2
→ − 14 ψ α( − m2)ψα − 12 ψ αfα β ψβ
+ ψ̄ αψ̄α.)
. . .
/ + √m2 )Ψ = (ψ̄ αi∇αα.ψα + χ̄αi∇αα.χα) + √m2 (ψ α χα + ψ̄ αχ̄α.)
LDirac = Ψ(i∇
Z  Z Z Z 
SsuperQCD = − dx g2 tr d θ W + d θ φ̄φ +
1 2 2 4 2 √ m 1 2
d θ 2 2 φ + h.c.
 699

FEYNMAN (V)
Z N Z
Y 
δ
S = S0 + SI , S0 = 1
2 φKφ; AN = ψN i Z[φ]
i=1
δφ
Z Z 
−W [ϕ] 1 δ 1 δ
Z[ϕ] = e = Dφ e−(S0 [φ]+SI [φ+ϕ]) = exp 2 e−SI [ϕ]
δϕ K δϕ
Effective action Γ[φ] (unrenormalized): (E.g., L = − 14 φ( − m2 )φ + 16 gφ3 .)
(A1) 1PI graphs only (plus S0 ). (For W [φ], connected graphs only.)
R
(A2) Momenta: label consistently with conservation, with dp for each loop.
(A3) Propagators: 1/K for each internal line. (E.g., 1/ 12 (p2 + m2 ).)
(A4) Vertices: read off of −SI . (E.g., −g)
R P
(A5) External lines: attach the appropriate (off-shell) fields and dp, with δ( p).
(A6) Statistics: 1/n! for n-fold symmetry of internal/external lines
(or keep just 1 of n! related graphs); −1 for fermionic loop; overall −1.
Vacuum: (Renormalize before for minimal subtraction/after for MOM.)
(B1) Find the minimum of the effective potential (for scalars).
(B2) Shift (scalar) fields to perturb about minimum; drop constant in potential.
(B3) Find resulting masses; find wave function normalizations.
T-matrix:
(C1) Connected trees of (shifted, renormalized) Γ: (A2-4) for L=0 with S → Γ.
(C2) Amputate external Γ0 -propagators.
(C3) External lines: appropriate to Γ0 wave equation K̃ψ = 0. (E.g., 1.)
(C4) External-line statistics: No symmetry factors; −1 for fermion permutation.

Probabilities X 
Sconnected = iδ p T
X  Y (2π)D/2 Y dD−1 p Y (2π)D/2
−D/2
dP = |Tf i|2δ D p , P = 2(Im T ii )(2π)
ω out
(2π)D−1 in
ω
all
dP 2 Im Tii dP 2 Im Tii dP
= , = = −2 Im M, = 2 Im Tii = −Im M 2
dt ω ds m dτ
dP X  (2π)D Y dD−1 p (2π)D/2
dσ = = |Tf i|2δ D p , σ = 2(Im T ii )
v12 λ12 out (2π)D/2−1ω λ12
λ212 = (p1 · p2 )2 − m21 m22 = 14 [s − (m1 + m2 )2][s − (m1 − m2 )2]
dσ 1
= 12 (2π)3|Tf i|2 2 ; s = −(p1 + p2 )2 , t = −(p1 + p3 )2, u = −(p1 + p4 )2
dt λ12
dσ |~p3 |3 λ34
= (2π)2|Tf i|2 = (2π)2|Tf i|2 (CoM)
dΩ λ12[ 2 (s − m3 − m4)ω3 − m3ω4 ]
1 2 2 2 λ12 s
700

GAUGES (VI): Gervais-Neveu

LA = − 14 A · A − iAaAb ∂b Aa − 14 AaAb AaAb
Y ang − Mills : L = LA + LC , LC = − 12 iC̃(∂ + iA)2C − 12 C̃C(∂ · A + iA2)
Gervais − Neveu : L = LA + 14 m2 A2

Twistors .
hp| = pα , |pi = pα ; [p| = pα , |p] = pα.
hpqi* = [qp] = −[pq], hpqihrsi + hqrihpsi + hrpihqsi = 0
p = p+ |+i[+| + p− |−i[−| + pt |−i[+| + p̄t |+i[−|, h+−i = [−+] = 1
P = |pi[p| ⇒ p+ = hp−i[−p], p− = h+pi[p+], pt = h+pi[−p], p̄t = hp−i[p+]

Spacecone
n · A = 0, n = |+i[−|
L = L2 + L3 + L4
+−2P
1 2
−
+ 1 2
L2 = A (− 2 P )A + ψ ψ−
p
 ∓   ∓ 
p ± ± ∓ − p ±
L3 = A ([A , pA ] + {ψ , ψ }) +
+
ψ [A±, ψ ∓]
p p
1 1
L4 = ([A+, pA−] + {ψ +, ψ −}) 2 ([A−, pA+ ] + {ψ +, ψ −}) − [A+, ψ −] [A−, ψ +]
p p
[−p] h+pi
A+ = , A− = ; ψ + = [−p], ψ − = h+pi
h+pi [−p]
p− + p+ − p− + p+ −
ref. lines : A = A = ψ = ψ =1
p p p p
Background-field
φ → ϕ + φ; ∇ → D + iA, Fab → Fab + D[a Ab] + i[Aa, Ab ]
∂ · A → D · A, C̃∂ · ∇C → C̃D2 C + C̃D · i[A, C]
1 − loop : K = − 12 ( − iF abSba )

Supergraphs
R
(A2 12 ) θ’s: one for each vertex, with an d4 θ.
(A30 ) Propagators:
 
d2 m d¯2 1
(V V, φ̄φ, φφ, φ̄φ̄) : 1, −1, √2 1 2 , √2 1 2 1 2
m
δ 4(θ − θ0 )
−2p −2p 2
2 (p + m )

(A4 1 ) Chiral vertex factors: d¯2 on the φ end(s) of every chiral propagator,
2
d2 on the φ̄ end(s), but drop any one such factor at a superpotential vertex.
 701

LOOPS (VII,VIII): Gamma function

Z " #
∞ Y
∞
1 X∞
Γ(z) = dλ λz−1 e−λ = 1z e−γz ze
z/n
= 1z exp −γz + 1
n
ζ(n)(−z)n
0 n=1
1 + n n=2
!
X
n X∞
1
γ = lim −ln n + 1
m
= 0.5772156649..., ζ(z) =
n→∞
m=1 n=1
nz
√
Γ(z) 21−2z π
Γ(z + 1) = zΓ(z), Γ(z)Γ(1 − z) = π csc(πz), =
Γ(2z) Γ(z + 12 )
√ (2n)! √
Γ(n + 1) = n!, Γ(n + 12 ) = (n − 12 )(n − 32 )... 12 π = π
n!22n
r  
2π z z √
lim Γ(z) ≈ , lim Γ(az + b) ≈ 2π(az)az+b−1/2 e−az
z→∞ z e z→∞
Z 1 Z ∞
Γ(x)Γ(y)
B(x, y) = dz z (1 − z)
x−1 y−1
= dτ τ x−1 (1 + τ )−x−y =
0 0 Γ(x + y)

Regularization
Z ∞
Γ(a) a−1 −τ (p2 +m2 )/2
1 2 = dτ τ e
[ 2 (p + m2)]a 0
Z ∞  X 
1= dλ δ λ − τi , τi = λαi
0
Z Z
−k2 /2 ka ...kb
dk e = 1, dk 1 2 a = 0
(2k )
Z
Γ(a) Γ(b) Γ(a + b − D2 ) D
dk 1 2 a 1 = 1 2 a+b−D/2 B( 2 − a, D2 − b)
( 2 k ) [ 2 (k + p)2 ]b (2p )

Schemes
MS : h̄

MS : Γ( D2 )h̄, Γ(1 − )h̄, etc.

(−1)D0 /2 Γ(1 − 2)
G: h̄, h̄, etc.
Γ(2 − D2 )B( D2 − 1, D2 − 1) Γ(1 + )[Γ(1 − )]2

Running coupling
Z
Γ1,2g ≈ tr dx 18 F abβ1 (ln − 1 )Fab ; β1 = 12 cR (−1)2s (4s2 − 13 ); cF +F̄ = 2, cA = 2N
Z   
β2 µ2 1/β1g 2 2 β2 /β12
ΓM ≈ tr dx 1
8
F β1ln + ln β1 ln 2 F, = e (g )
M 2 β1 M M2
702

GRAVITY (IX,X)
[c
[Mab , Vc ] = V[a ηb]c ⇒ 1 ab
2 λ [Mba , Vc ] = λc a Va , [Mab , M cd ] = −δ[aMb] d]
[M12, V2 ] = η22 V1 ,[M12, M23 ] = η22M13
..
1 ab 1 αβ 1 αβ .
2 λ Mba = 2 λ Mβα + 2 λ Mβ α.
(γ
[Mαβ , ψγ ] = ψ(αCβ)γ ⇒ 1 αβ
2 λ [Mβα , ψγ ] = λγ αψα , [Mαβ , M γδ ] = δ(αMβ)δ)
∇a = ea + 12 ωa bc Mcb , e a = e a m ∂m ; [∇a, ∇b] = Tabc ∇c + 12 Rab cd Mdc
[∇1, ∇2] = [e1 + ω1 , e2 + ω2 ]
= {[e1, e2] + (e1ω2 )M2 − (e2ω1 )M1 } + {ω1 [M1, ∇2] − ω2 [M2, ∇1 ] − ω1 ω2 [M1 , M2]}
−ds2 = dxm dxngmn , gmn = em a enb ηab
Z
S= dx e−1L, e = det eam L = − 14 R + L(∇, ψ), R = Rab ab
Z
Rab − 12 ηab R = 2Tab , c
Rab = R acb , δSM = dx e−1 (em a δebm)Tab

Supergravity
..
∇A = EAM ∂M + 12 ΩA βγ Mγβ + 12 ΩA βγ M . . + iAAY, [Y, ∇α] = − 12 ∇α
γβ
..
[∇A, ∇B } = TAB C ∇C + 12 RAB γδ Mδγ + 12 RAB γ δ M . . + iFAB Y
δγ
{∇α., ∇ .} = BM . ., {∇α, ∇ .} = −i∇ .
β αβ β αβ

[∇α., −i∇ .] = C . .Wβ − 12 (∇β B)M . ., [−i∇αα., −i∇ .] = C . .fαβ − h.c.

ββ βα αβ ββ βα
. . .
Wα = − B∇α − Gα β ∇ . + 12 (∇β Gα γ )M . . + 12 Wα βγ Mγβ + iWαY + i 16 W β Mβα
β βγ
. .
fαβ = i 12 G(α γ ∇β)γ. − 12 (∇(αB + i 13 W(α )∇β) + Wαβ γ ∇γ − 12 (∇(αGβ) γ )∇γ.
.
− ( 12 ∇2B̄ + B B̄ + 12 1
i∇γ Wγ )Mαβ − i 18 [(∇(αδ G .)Mβ γ + α ↔ β]
γ)δ
. .
+ 12 Wαβ γδ Mδγ + 14 (∇(α∇γ Gβ) δ )M . . + i 12 (∇(αWβ))Y
δγ
Wαβγδ = 1
∇ W
4! (α βγδ)
.
Ga = Ḡa , ∇α.B = ∇α.Wα = ∇α.Wαβγ = 0, ∇α Gαα. = ∇αB − iWα
. .
∇αWαβγ − i 13 ∇(β Wγ) = −i 12 ∇(β α Gγ)α., ∇αWα + ∇α W α. = 0

Ectoplasm Z
S= dx (− 4!1 )mnpq eq D ep C enB em ALABCD
. .
α 2
. L̄,
Lαβcd = αα ,β α,cd . ∇ L̄,
Lαbcd = iαα,bcd Labcd = abcd [(∇ + 3B)L̄ + h.c.]
2
∇α.L = 0 ⇒ L = (∇ + B)L
 703

................................. .................................
................................. INDEX .................................
Topics are listed by subsection (or section, etc.).

Abelian . . . . . . . . . . . . . . . . . . . . . . . . . . . . IB2 Bosonization . . . . . . . . . VIIIA7,XIA5-6,B7-8

Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . III-IV Bracket . . . . . . . . . . . . . . . . . . . . . IA1-2,XIIA1
Adjoint representation . . . . . . . . . . . IB2,5,C2 BRST . . . . . VIA-B,VIIA1,IXB1,XIA4-5,XII
Advanced propagator . . . . . . . . . . . VA3,B1-2
Affine parametrization . . . . . . . . . . . . . . IIIB2 Cabibbo-Kobayashi-Maskawa matrix IVB3
Affine transformation . . . . . . . . . . . . . . . . IB1 Cartan metric . . . . . . . . . . . . . . . . . . . . . . . IB2
Analytic S-matrix theory . . . . . . . . . . . . XIA1 Cartan subalgebra . . . . . . . . . . . . . . . . . . . IB5
Anni’l’n op.IC1,IIB3,C5,IIIA2,VIA1,3,XIA4 Casimir operator . . . . . . . . . . . . . . . . . . . . . IB2
Anomaly . . . . . . . . . . . VIIIA8,B2-4,C1,XIA5 Causality . . . . . . . . . . . . . . . . . VA4,C6,VIIA1
Antibracket . . . . . . . . . . . . . . . . . . . . . . . XIIC1 Center-of-mass frame . . . . . . . IA4,VC7,VIC4
Anticommutation . . . . . . . . . . . . . . . . . . . . IA2 Central charge . . . . . . . . . . . . . . . . . IVC7,XC5
Anticommuting number/variable . . . . . . IA2 Channels . . . . . . . . . . . . . . . . . . . . . . . . . VIIB6
Antifield . . . . . . . . . . . . . . . . . . . . VIA3,XIIB3 Chan-Paton factors . . . . . . . . . . . . . . . . . XIB1
Anti-Gervais-Neveu gauge . . . . . . . . . VIB4-5 C. conj. IA5,IIC5,IIIA4,IVA4,B1,XC1,XIB2
Antighost . . . . . . . . . . . . . . . . . . . . . . . . . VIA2 Chern-Simons form IIIC6,IVC5,VIIIB2,XC5
Antiparticle . . . . . . . . . . . . . . . . IA4-5,IIIB4-5 Chirality . . . . . . . . . . . . . . . . . . . . . . IIA7,IVC1
Antisym.tensor IIB2,VIIIA8,XA3,XIIA5,B3 Chiral projector . . . . . . . . . . . . . . . . . . . IIA6-7
Arnowitt-Fickler gauge . . . . . . . . . . . . . IIIC2 Chiral representation . . . . . . . . . . . . . . IVC4-5
Asymptotic freedom . . . . . VIIIA3-4,C,XIC3 Chiral spinor . . . . . . . . . . . . . . . . . . IIA7,IIIC4
Atiyah-Drinfel’d-Hitchin-Manin . . . . . . IIIC7 Chiral superfield . . . . . . . . IIC5,IVC1-2,VIB7
Auxiliary variable . . . IIIA1,5,B2,C2,4,IVC2 Chiral superspace . . . . . . . . . . . . . . . . . . IVC1
Axial anomaly . . . . . . . . . . . . VIIIA8,B2-4,C1 Chiral symmetry . . . . . IVA4,VIIIB2-4,XIC1
Axial current . . . . . . . . . . . IVA4,VIIIA8,B2-4 Chiral theory . . . . . . . . . . . . . . . . . . . . . . . IIA7
Axial gauge . . . . . . . . . . . IIIC2,VIB6-8,IXB3 Chromodynamics . . . . . . . . . IC4,IVB1,VIIC4
Axial vector . . . . . . . . . . . . . . . . . . . . . . . . . IA5 Classical group . . . . . . . . . . . . . . . . . . IB4,IIC3
Axion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XIB Classical limit . . . . . . . . . . . . . IA1-2,IIA2-3,V
Clebsch-Gordan-Wigner coefficients . . . IIA4
Background field . . . . . VC1,3,VIIB2,XIIB7 Closed string . . . . . . . . . . . . . IVB1,VIIC4,XI
Background-field gauge . . . VIB8,10,VIIIA3 Coherent state . . . . . . . . . . . . . . . . . . . . . . . IA2
Bag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVB1 Cohomology . . . . . . . . . . . . . VIA1,3,XB2,XII
Baryon number . . . . . . . . . . . . . . . . . IVA4,B1 Coleman-Weinberg mechanism . . . . . . VIIB2
Baryon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IC4 Colinear divergence/particle . . . . . . . . VIIA6
Bhabha scattering . . . . . . . . . . . . . . . . . . VIC4 Color . . . . . . . . . . . . . . . . . . . . . . . . . . IC4,IVA4
Bianchi id. . IIA7,IIIC1,IVC3,IXA2,XB2,C5 Color decomposition/expansion VC9,VIIC4
Big Bang . . . . . . . . . . . . . . . . . . . . . . . . . . IXC3 Color ordering . . . . . . . . . . . . . . . VC9,VIC1-3
Birkhoff’s theorem . . . . . . . . . . . . . . . . IXC5,7 Compact group . . . . . . . . . . . . . . . . . . . . IB2,4
Bjorken scaling variable . . . . . . . . . . . VIIIC3 Compactification . . . . . . . . . . . . . . IIB4,XIC3
Black hole . . . . . . . . . . . . . . . . . . . . . . . . . IXC7 Compensator . . . . . . . . . . . . IVA5,IXA7,XA3
Borel summation/transform . . . . . . . . VIIC3 Complex projective . . . . . . . . . . . IVA2,VIIB3
Bnd.cond. IC1,IIIA5,C,VA,VIA1,IXA5,XIA Complex representation . . . . . . . . . . . . . IB1-2
Bnd.state VC1,VIIB3,C2,VIIIA7,B4,XIA1,6 Component approach . . . . . . . . . . . . . . . . XB4
Bohr-Sommerfeld quantization . . . . . . . VA2 Component transformation . . . . . . IIC2,XB3
 704

Compton scattering . . . . . . . . . . . . . . . . VIC4 Degree of freedom . . . IVC3-4,VB3,XC3,XII

Confinement . . . . . . . . . . . . . . . . . . . IVB1,XIC Density . . . . . . . . . . . . . . . . . IIIB1,IXA4,XIB5
Conformal . . . . . . . . . . . . . IA6,IIB-C,IIIC5-7 Density parameter . . . . . . . . . . . . . . . . . . IXC3
Conformal anomaly . . . . . . . . . VIIIC1,XIA5 De Sitter space . . . . . . . . . . . . . . . . . . . IXC2-3
Conformal boost . . . . . . . . . . . . . IA6,IIB1,6-7 Determinant . . . . . . . . . . . . . IB3,C2,5,IIA1,5
Conformal field theory . . . . . . . . . . . . . . XIB7 Differential cross section . . . . . . . . . . . . . VC7
Conformal gauge . . . . . . . . . . . . . . . . . . . XIA3 Differential form . . . . . . . . . . . IC2,IXA4,XB2
Conformally flat . . . . . . . . . . . . . . . . . . . . IXC2 Dilatation . . . . . . . . . . . . . . . . . . . . IA6,IIB1,6
Connected graph . . . . . . . . . . . . . . . . . . . . VC2 Dilaton . . . . . . . . . . . . . . . . . . . . . . . IXA7,XIB
Conservation . . . . . . . . . . . . . IA1,IIIB4,IXA6 Dim.red IIB4,IIIC8,VIB6,VIIA1,VIIIA3,XC
Constituent quark . . . . . . . . . . . . . . . . . . . IC4 Dimensional regularization . . . . . . . . . . . . VII
Constraint . . . . . . . . . . . . . . . . . . . . . . . . . IIIA5 Dimensional transmutation . . . . . . . . . VIIB3
Constructive quantum field th. IIIC4,VIIC3 Dirac delta . . . . . . . . . . . . . . . . . . . . IIIA1,VA2
Continuum limit . . . . . . . . . . . . . . . . VA1,XIC Dirac equation . . . . . . . . . . . . . . . . IIB2,IIIA4
Contour integral . . . . . . . . . . . . . . . . . . . . IIA1 Dirac matrices/spinor . . . . . . . IC1,IIA6,XC1
Contracted indices . . . . . . . . . . . . . . . . . . . IA1 Direct product/sum . . . . . . . . . . . . . . . . . . IB2
Contraction . . . . . . . . . . . . . . . . . . . . . . . . IIC3 Discrete symmetry . . . . . . . . . . . . . . . . . . . IA5
Coordinate representation . . . . . . . . . . . . IC1 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . VII
Coordinate transform. . IA,C2,IIIB1,3,IXA1 Division algebra . . . . . . . . . . . . . . . . . . . . IIC4
Coset space . . . . . . . . . . . . . . . . . . . . . . . . IVA3 Double covering . . . . . . . . . . . . . . . . . . . . . IIA3
Cosmological red shift . . . . . . . . . . . . . . IXC4 Drell-Yan scattering . . . . . . . . . . . . . . VIIIC3
Cosmological term . . . . . IXA5-7,C2,XB1,6-7 DualityIIA7,IIIA4,C4,IXC1,XB5,C6,XIB4,6
Cosmology . . . . . . . . . . . . . . . . . . . . . . . . . IXC3 Dual model . . . . . . . . . . . . . . . . . . . . . . . . XIA1
Counterterm . . . . . . . . . . . . . . . . . . . . . . VIIA1 Dual representation/space . . . . . . . . . . . . IB1
Covar.deriv. IIC2,IIIA4,C1,IVC3,IXA2,XA1 Duffin-Kemmer matrices . . . . . . . . . . . . . IIB4
Covariant momentum . . . . . . . . . . . IA5,IIIC1 Dust . . . . . . . . . . . . . . . . . . . . . . . . . . . IXB2,C3
Covering group . . . . . . . . . . . . . . . . . . . . . . IC5 Dynkin index . . . . . . . . . . . . . . . . . . . . . . . . IB2
CP(n) model . . . . . . . . . . . . . . . . . IVA2,VIIB3
CPT theorem . . . . . . . . . . . . . . . . . IVB1,VC8 Ectoplasm . . . . . . . . . . . . . . . . . . . . . . . . . XB2
Creat’n op. IC1,IIB3,C5,IIIA2,VIA1,3,XIA4 Effective action . . . . . . . . . . . . . . . . . . . . . VC2
Critical dimension . . . . . . . . . . . . . . . . . . . . XI Effective potential . . . . . . . . . . . . . . . . . VIIB2
Crossing symmetry . . . . . . . . . . . . VC8,VIC4 Einstein-Hilbert action . . . . . . . . . . . . . . IXA5
Cross product . . . . . . . . . . . . . . . . . . . . . . IIA1 Einstein summation . . . . . . . . . . . . . . . . . . IA1
Cross section . . . . . . . . . . . . . . . . . . VC7,VIC4 Elastic scattering . . . . . . . . . . . . . . . . . . . VC7
Current . . . . . . . . . . . . . . . IIA7,IIIA4,B4,VC3 Electromagnetism . . . IA5,C4,IIA7,B5,IVB2
Current quark . . . . . . . . . . . . . . . . . . . . . . . IC4 Energy . . . . . . . . . . . . . . . . . IA1,5,IIA5,B6,C1
Curvature . . . . . . . . . . . . . . . . . . . . . . . . . IXA2 Energy-momentum tensor . . IIIA4,B4,IXA6
Curved indices . . . . . . . . . . . . . . . . . . . . . IXA1 Equivalence principle . . . . . . . . . . . . . . . IXA4
Cut propagator . . . . . . . . . . . . . . . . . . . . . VC7 Euclidean space . . . . . . . . . . . IA4,IIIC4,VB4
Cutting rule . . . . . . . . . . . . . . . . . . . . . . . . VC6 Euler number . . . . . . . . . . VIIC4,IXA7,XIC2
Euler-Mascheroni constant . . . . . . . . . VIIA2
Deceleration parameter . . . . . . . . . . . . IXC3 Euler’s theorem . . . . . . . . . . . . . . . . . . . VIIC4
De Donder gauge . . . . . . . . . . . . . . . . . . . IXB1 Event horizon . . . . . . . . . . . . . . . . . . . . . . IXC7
Deep inelastic scattering . . . . . . . . . . . VIIIC3 Evolution equations . . . . . . . . . . . . . . . VIIIC3
Defining representation . . . . . . . . . . . . . . . IB4 Exceptional group . . . . . . . . . . . . . . . IB4,XC6
 705

Exclusive scattering . . . . . . . . . . . . . . . VIIIC3 General linear group . . . . . . . . . . . . . . . . . IB4

Explicit supersymmetry breaking . . . . IVC6 General relativity . . . . . . . . . . . . . . . . . . . . . IX
Explicit symmetry breaking . . . . . . . . . . IVA General superfield . . . . . . . . . . . . . . . . . . IVC1
Extended supersymmetry . . . . . IVC7,XC3-6 Generating functional . . . . . . . . . . . . . . VC1-2
External field . . . . . . . . . . . . . . IIIA1,B3-5,VC Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . IA1
External line . . . . . . . . . . . . . . . . . . . . . . . . VC2 Genus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XIC2
Extrinsic length . . . . . . . . . . . . . . . . . . . . IIIB1 Geodesic . . . . . . . . . . . . . . . . . . . . . . . . . IXB2-4
Gervais-Neveu gauge . . . . . . . . . . . . . . VIB4-5
Face . . . . . . . . . . . . . . . . . . . . . . . VIIC4,XIC2 Gervais-Neveu model . . . . . . . . . . . . . . . IVA6
Factorization . . . . . . . . . . . . . . . . . . . . . VIIIC3 Ghost . . . . . . . . . . . . . . . . . . . . . . . . . . VIA,XII
Family . . . . . . . . . . . . . . . . . . . . . . . . IC4,IVB3 Ghost number . . . . . . . . . . . . . . . VIA1,XIIB1
Fayet-Iliopoulos term . . . . . . . . . . . . . IVC5-7 Glashow-Iliopoulos-Maiani mechanismIVB3
Fermi-Feynman gauge . . VIB2-3,9,IXB1,XII Glashow-Salam-Weinberg model . . . . . IVB2
Fermion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IA2 GLDAP equations . . . . . . . . . . . . . . . . VIIIC3
Fermion doubling . . . . . . . . . . . . . . . . . . XIC1 Global symmetry/transformation . . . . . . . . I
Feynman diagram/graph . . . . . . . . . . . . . VC2 Glueball . . . . . . . . . . IC4,IVB1,VIIC4,VIIIB4
Feynman parameter . . . . . . . . . . . . . . . VIIA2 Gluino . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5
Feynman tree theorem . . . . . . . . . . . . . . . VC6 Gluon . . . . . . . . . . . . IC4,IIIC,IVB1,VIC,VIII
Field strength . . . . . . . IIA7,B3,5,IIIC1,XA2 Goldstone boson/theorem . . . . . . . . . . . . IVA
Fine tuning . . . . . . . . . . . . . . . . . . . . . . . . . XB6 Goldstone fermion . . . . . . . . . . . . . IVC6,XB6
Finite theory . . . . . . . . . . . . . . . . . . . VIIIA5-6 Graded commutation . . . . . . . . . . . . . . . . . IA2
Finite transformation . . . . . . . . . . . . . . . . IA3 Graded group/symmetry . . . . . IA2,IIC3,XII
1st-order IIIA5,C4,IVC5,IXA5,XB4-5,XIA2 Grand Unified Theory . . . . . . . IVB4,VIIIA4
Flat indices . . . . . . . . . . . . . . . . . . . . . . . . IXA1 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VC2
Flat superspace . . . . . . . . . . . . . . . . . . . . IVC3 Gravitational collapse/radius . . . . . . . . IXC7
Flavor . . . . . . . . . . . . . . . . . . . . . . . . IC4,IVA-B Gravitational red shift . . . . . . . . . . . . . . IXC6
Flavor-changing neutral currents . . . . . IVB3 Graviton . . . IIIA4,C1,IVA1,IXB1,C1-2,XIB
Foldy-Wouthuysen transform. . IIB5,VIIIB6 Gravity . . . . . . . . . . . . . . . . . . . . . . . . . IIIA4,IX
Four-vector . . . . . . . . . . . . . . . . . . . . . . . . . . IA4 Green function . . . . . . . . . . . . . . . . . . . . . VA-B
Fragmentation function . . . . . . . . . . . VIIIC3 Gribov ambiguity . . . . . . . . . . . . . . . . . . VIB2
Functional . . . . . . . . . . . . . . . . . . . . . IC1,IIIA1 Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . IA3-II
Functional integral . . . . . . . . . . . . . . . . . . VA1 Group contraction . . . . . . . . . . . . . . . . . . IIC3
Functional derivative . . . . . . . . . . . . . . . IIIA1 Group metric . . . . . . . . . . . . . . . . . . . . . . . . IB4
Fundamental representation . . . . . . . . . . . IB4 G-scheme . . . . . . . . . . . . . . . . . . . . . . . . . VIIA3
Furry’s theorem . . . . . . . . . . . . VIIA5,VIIIB3
Hadron . . . . . IC4,IVA4,B1,VIIC4,VIIIC,XI
Galilean boost/group . . . . . . . . . . . . . . . . IA1 Hamiltonian . . . . . . . . . . IA1,IIIA,VA1,VIA1
Gamma matrix . . . . . . . . . . . . . IC1,IIA6,XC1 Hamiltonian density . . . . . . . . . . . . . . . . IIIA3
Gauge . . . . . . . IIIB2,C1,VIB,IXB,XA4,XIA3 Hamilton-Jacobi equations . . . . . . . . . . . VA2
Gauge field . . . . . . . . . . . . . . . . . . . . . . IIA7,III Hamilton’s equations . . . . . . . . . . . . . . . . . IA1
Gauge invar. . IIA7,IIIA5,B1,C1,IXA1,XIA2 Handle . . . . . . . . . . . . . . . . . . . . . . VIIC4,XIC2
Gauge parameter . . . . . . . . . . . . . . . . IIA7,III Hard particle . . . . . . . . . . . . . . . . . . . . . VIIIC3
Gaugino . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5 Harmonic gauge . . . . . . . . . . . . . . . . . . . . IXB1
Gauss-Bonet theorem . . . . . . . . . . . . . . . IXA7 Helicity . . . . . . . . . . . . . . . . . . . . . . . . . IIB7,C5
Gaussian . . IB3,IIC3,VA2,5,B1,VIIA2,XIC3 Heterotic string . . . . . . . . . . . . . . . . . . . XIB3,6
 706

Hidden matter . . . . . . . . . . . . . . . . . . . . . . XB7 Lab frame . . . . . . . . . . . . . . . . . . . . . . . . . VC7

Higgsino . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5 Ladder diagram . . . . . . . . . . . . . . . . . . . . XIA1
Higgs mechanism . . . . . . . . . . . . . . . . . . . IVA6 Lagrange multiplier . . . . . . . . . . . . . . . IIIA1,5
Higgs scalar . . . . . . . . . . . . . . . . . . . . IC4,IVA6 Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . IIIA1
Higher dimensions . . . . . . . . . . . . . . . . . . . . XC Lagrangian density . . . . . . . . . . . . . . . . . IIIA3
Hilbert-space metric . . . . . . . IB4,VC5,XIIA1 Lamb shift . . . . . . . . . . . . . . . . . . . . . . . VIIIB6
Hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VIIC4 Landau equations/singularity . . . . . . . . VC8
Homogeneous space . . . . . . . . . . . . . . . IXC3-4 Landau gauge . . . . . . . . . . . . . . . . . . . . VIB2-3
Hubble constant . . . . . . . . . . . . . . . . . . . . IXC3 Landau ghost . . . . . . . . . . . . . . . . . . . . . VIIC2
Hypermultiplet/hypersymmetry . . . . . IVC7 Laplace transform . . . . . . . . . . . . . . . . . VIIC3
Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XIC
Ideal mixing . . . . . . . . . . . . . . . . . . . . . . . . IC4 Legendre transform . . . . . . . . . . . . . VC3,XB5
i prescription . . . . . . . . . . . . . . . . . . . VA3,B1 Length . . . . . . . . . . . . . IA4,IIIB1,IXB2,XIA2
Improved perturbation . . . . . . . . . . . . . VIIC1 Lepton . . . . . . . . . . . . . . . . . . . . . . . . . IC4,IVB
Inclusive scattering . . . . . . . . . . . . . . . VIIIC3 Levi-Civita tensor . . . . . . . . . . . . . . . . IA5,B3
Index notation . . . . . . . . . . . . . . . . . IB,IIA4-5 Lie algebra/bracket/derivative/group . . IA3
Induced metric . . . . . . . . . . . . . . . . . . . . . XIA2 Light bending . . . . . . . . . . . . . . . . . . . . . . IXC6
Inelastic scattering . . . . . . . . . . . . . . . . . . VC7 Lightcone . . . . . . . . . . . . . . . . . . . . . . . . . . . IA6
Inhomogeneous group . . . . . . . . . . . . . . XIIB1 Lightcone basis . . . . . . . . . . . . . . . . . . . . . . IA4
Infinitesimal transformation . . . . . . . . . . . IA1 Lightcone coordinates . . . . . . . . IA4,6,XIIA1
Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII Lightcone frame . . . . . . . . . . . . . . . . . . . IIB3,6
Infrared divergence . . . . . . . . . . VIIA6,VIIIB Lightcone gauge . . . . . . . . . . . . IIB3,IIIB2,C2
Infrared slavery . . . . . . . . . . . . . . . . . . . . IVB1 Lightcone spinor . . . . . . . . . . . . . . . . . . . . XC4
Inhomogeneous transformation . . . . . IA4,B1 Linear multiplet . . . . . . . . . . . . . . . . . . . . XA3
Inner product . . . . . . . . . . . . . . . . . IB1,XIIB2 Link . . . . . . . . . . . . . . . . . . . . . . . . . VC2,8,XIC
Instanton . . . . . . . . . . . . . . . . . . IIIC6-7,VIIC2 Little group . . . . . . . . . . . . . . . . . . . . . . . IIB3-4
Interaction picture . . . . . . . . . . . . . . . . . . VA4 Local inertial frame . . . . . . . . . . . . . . . . . IXC4
Internal line . . . . . . . . . . . . . . . . . . . . . . . . VC2 Locality . . . . . . . . . . . . . . . . . . . . . . . . . IIIA1,3
Intrinsic length . . . . . . . . . . . . . . . . . . . . . IIIB1 Local symmetry/transformation . . . . . . . . III
Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . IA6 Loop graph . . . . . . . . . . . . IIIB5,VC,VII-VIII
Irreducible representation . . . . . . . . . . . . . IB2 Lorentz connection . . . . . . . . . . . . . . . . . IXA2
Isospin . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVB2 Lorentz force law . . . . . . . . . . . . . . . . . . . IXB2
Isotropic space . . . . . . . . . . . . . . . . . . . IXC3-4 Lorentz gauge . . . . . . . . . . . . . . . . VIB2,IXB1
Lorentz transf. . . . . . . . . . . . . . . . . IA4-5,IIA5
Jacobi identity . . . . . . . . . . . . . . . . . . . . IA1,3 Lowering operator . . . . . . . . . . . IA6,B5,VIA1
Jet . . . . . . . . . . . . . . . . . . . . . . . . . . . . VIIIC2-3
JWKB . . . . . . . . IA1,VA2,C3,VIB1,VIIIB1,6 Magnetic charge . . . . . . . . . . . . . . . . . . IIIC8
Magnetic moment . . . . . IIIC4,VIIA3,VIIIB5
Kähler manifold/potential . . . . . . . . . . XB6 Magnetic monopole . . . . . . . . . . . . . . . . . IIIC8
Killing equation/vector . . . . IXA2,6,B2,C4,6 Majorana spinor . . . . . . . . . . IIA6,IIIA4,XC2
Kinoshita-Lee-Nauenberg . . . . . . . . . . VIIA6 Mandelstam variables . . . . . . IA4,VC4,VIC4
Klein factor/transformation . IA2,VC5,XC1 Manifest symmetry . . . . . . . . . . . . . . . . . . IB1
Klein-Gordon equation . . . . . . . . . . . . . . IIB1 Maximal helicity violation . . . . . . . . . . . VIC1
Kronecker delta . . . . . . . . . . . . . . . . . . . . . . IA1 Maxwell’s equations . . . . . . . . IIA7,B2,IIIA4
Kruskal-Szekeres coordinates . . . . . . . . IXC7 Mechanics . . . . . . . . . . . . . . . . IA,IIIB,VB,XII
 707

Mellin transform . . . . . . . . . . . . . . . . . . VIIC3 Pair creation . . IIIB5,VIIB4,5,VIIIA2-3,C2

Meson . . . . . . . . . IC4,IVA4,B1,VIIC4,VIIIB4 Parallel transport . . . . . . . . . . . . . . . . . . IXB4
Method of images . . . . . . . . . . . . . . . . . XIA3,6 Parity . . . . . . IA5,IIA5,IVA4,B1,VIIA5,XIB2
Metric . . . . . . . . . . . . . . IA4,B2,4,IIIA1,IXA2 Parke-Taylor . . . . . . . . . . . . . . . . . . . . . VIC1-3
Minimal couplingIA5,IIB5,IIIA2,4,C1,IXA4 Partially conserved axial current . . . . . IVA4
Minimal subtraction . . . . . . . . . . . . . . . VIIA3 Particles . . . . . . . . . . . . . . . . . IA,IIIB,VB,XII
Minkowski metric/space . . . . . . . . . . . . . . IA4 Parton . . . . . . . . . . . . . . . . . . . . . . . . . . . . VIIIC
Modified minimal subtraction . . . . . . . VIIA3 Parton distribution/model . . . . . . . . . VIIIC3
Møller scattering . . . . . . . . . . . . . . . . . . . VIC4 Path integral . . . . . . . . . . . . . . . . . VA1-2,C1-4
Momentum integral . . . . . . . . . . . . . . VII-VIII Path ordering . . . . . . . . . . . . . . . . IIIC2,XIC1
Momentum subtraction . . . . . . . . . . . . VIIA3 Pati-Salam model . . . . . . . . . . . . . . . . . . IVB4
Monopole . . . . . . . . . . . . . . . . . . . . . . . . . . IIIC8 Pauli-Lubański vector . . . . . . . . . . . . . . . IIB7
Pauli matrices . . . . . . . . . . . . IIA1,IIIA2,XC1
Nakanishi-Lautrup . . . . . . . . VIA2,4,XIIC3 Pauli-Villars regularization . . . . . . . . VIIIB2
Natural units . . . . . . . . . . . . . . . . . IA1,4,IXC4 Penrose transform . . . . . . . . . . . . . . IIB6-7,C5
Negative energy . . . . . . . . . . IA4-5,IIIB4,VB1 Perfect fluid . . . . . . . . . . . . . . . . . . . . . . . IXC5
Neutrino . . . . . . . . . . . . . . . . . . . . . IC4,IVB2-4 Perihelion precession . . . . . . . . . . . . . . . . IXC6
New minimal supergravity . . . . . . . . . . . XB5 Perturbation . . . . . . . . . . . . . . . . . . . . . V-VIII
Newtonian limit . . . . . . . . . . . . . . . . . IXB1,C6 Pfaffian . . . . . . . . . . . . . . . . . . . . . . . . . . IB3,C5
Nielsen-Ninomiya theorem . . . . . . . . . . XIC1 Phase space . . . IA5,VC7,VIIA5,B4-5,XIIC1
Nonperturbative . . . . . . . . . . . . . . . . . VIIC,XI Photino . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5
Nonrelativity . . . . . . . . . . . . . . . . . . . . . . . . IA1 Physical region . . . . . . . . . . . . . . . . . . . . . VC7
Nonrenormalization theorem . . . . . . . . VIC5 Pion . . . . . . . . . . . . . . . . . . . . IC4,IVA4,VIIIB4
Normal ordering . . . . . . . . . . . . VIA1,XIA4,6 Pion decay constant . . . . . . . . . . . . . . . . IVA4
No-scale supergravity . . . . . . . . . . . . . . . . XB7 Planar diagram . . . . . . . . . . . . . VC8-9,VIIC4
Null basis . . . . . . . . . . . . . . . . . . . . . . . . . . . IA4 Planck units . . . . . . . . . . . . . . . . . . IA1,4,IXC4
Plane wave . . . . . . . . . . . IIIC3,VB2,C7,IXC1
Octonion . . . . . . . . . . . . . . . . . . . . . . IB4,IIC4 Plaquet . . . . . . . . . . . . . . . . . . . . . . . . . . . XIC1
Off shell . . . . . . . . . . . . . . IIB1,IVC1,3,7,XC3 Poincaré group . . . . . . . . . . . . . . . . . . . IA4,IIB
Okubo-Zweig-Iizuka rule . . . . . . . . . . . VIIC4 Poisson bracket . . . . . . . . . . . . . . . . . . . . . . IA1
Old minimal supergravity . . . . . . . . . . . . XB5 Polar vector . . . . . . . . . . . . . . . . . . . . . . . . . IA5
1/N expansion . . . . . . . . . . . . . . . VIIC4,XIC2 Positive action . . . . . . . . . . . . . . . . . IIIB1,VA5
One-particle-irreducible graph . . . . . . . . VC2 Positive energy . . . . . . . IA4,IIC1,IIIA4,IVC2
1.5-order . . . . . . . . . . . . . . . . . . . . . . . . . . . XB4 Power counting . . . . . . . . . . . . . . . . . . . . VIIA5
On shell . . . . . . . . . . . . . . . . . . . . . . . IIB1,XC3 Prepotential . . . . . . . . . . . . . . . . . IVC1,4,XA1
Open string . . . . . . . IVB1,VIIC4,VIIIA3,XI Projective lightcone . . . . . . . . . . . . . . . . . . IA6
Operator cohomology . . . . . . . . . . . . . . . VIA1 Projective transformation . . . . . . . IA6,IVA2
Operator product expansion . . . . . . . VIIIC3 Propagator . . . . . . . . . . . . . . . . . . . . . . . . VA-B
Optical theorem . . . . . . . . . . VIIA4,6,VIIIC3 Propagator correction . . . . . . VIIB4-7,VIIIA
O’Raifeartaigh model . . . . . . . . . . . . . . . IVC6 Proper Lorentz transformation . . . . . . . . IA5
Orthonormal basis . . . . . . . . . . . . . . . . . . . IA4 Proper time . . . . . . . . . . . . . . . . . . . . . . . . . IA4
Orthosymplectic group . . . . . . . . . . . IIC3,XII Pseudogoldstone boson . . . . . . . IVA4,VIIIB4
Orthogonal group . . . . . . . . . . . . . . . . . . . . IB4 Pseudohermiticity . . . . . . . . . . . . . . . . . . . . IB4
Outer product . . . . . . . . . . . . . . . . . . . . . . IIA1 Pseudoreal representation . . . . . . . . . . . . . IB2
Overlapping divergence . . . . . . . . . . . . VIIB7 Pseudoscalar/pseudotensor . . . . . . . . . . . IA5
 708

QCD . . . . . . . . . . . . . . IVB1,VIC3,VIIIA3,C Schwarzschild solution . . . . . . . . . . . . IXC5-7

Quantum-chromodynamic string . . . . . XIC3 Schwinger-Dyson equations . . . . . . . . . . VC3
Quark . . . . . . . . . . . . . . . . . . . . . . . IC4,IVA4,B Schwinger model . . . . . . . . . . . . . . . . . VIIIA8
Quark-gluon plasma . . . . . . . . . . . . . . . . IVB1 Schwinger parameter . . . . . . . . . . VB1,VIIA2
Quaternion . . . . . . . . . . . . . . . . . . IB4,IIA1,C4 S-duality . . . . . . . . . . . IIA7,IIIC4,IXC1,XC6
Quenched approximation . . . . . . . . . . . . XIC1 Self-duality . . . . . . . . . . . . . . . . . . . IIIC4,IXC1
Semiclassical expansion . . . . . IIIA4,VA2,C3
Radial gauge . . . . . . . . . . . . . . . . VIB1,IXB4 Semisimple group . . . . . . . . . . . . . . . . . . . . IB2
Radiation . . . . . . . . . . . . . . . . . . . . . . . . IXC3-4 Sesquilinear . . . . . . . . . . . . . . . . . . . . . . . . . IB4
Raising operator . . . . . . . . . . . . . . . IB5,VIA1 Sigma matrices . . . . . . . . . . . IIA1,IIIA2,XC1
Rapidity . . . . . . . . . . . . . . . . . . . . . . . . . . . IIA5 Sigma model . . . . . . . . . . . . . . . . . . . . . IVA2-4
Real representation . . . . . . . . . . . . . . . . . . IB2 Simple group . . . . . . . . . . . . . . . . . . . . . IA6,B2
Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . VIC2 Simple supersymmetry . . . . . . . . . . . . . . IVC7
Red shift . . . . . . . . . . . . . . . . . . . . . . . . . . IXC4 Singularity . . . . . . . . . . . . . . VC8,VIIC3,IXC7
Reducible representation . . . . . . . . . . . . . . IB2 Slepton . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5
Reference momentum/vector . . . . . . . . VIB6 SL(2,C) . . . . . . . . . . . . . . . . . . . . . . . . IC5,IIA5
Reflection . . . . . . . . . . . . . . . . . . . . . IA5,IIA3,5 S-matrix . . . . . . . . . . . . . . . . . . . . . . . . . VA4,C
Regge behavior/theory/trajectory . . . XIA1 Soft divergence/particle . . . . . . . . . . . . VIIA6
Regularization . . . . . . . . . . . . VIIA-B,VIIIB2 Sommerfeld-Watson transform . . . . . . . XIA1
Relativistic Schrödinger equation . . . . . IIB1 SO(3) . . . . . . . . . . . . . . . . . . . . . . . . . . IA3,IIA2
Relativity . . . . . . . . . . . . IA4,IIA5-B,IIIA4,IX Source . . . . . . . . . . . . . . . . . . . . . . . . IIIA4,VC3
Renormalizability . . . . . . . . . . . . . . . . VIIA1,5 Spacecone . . . . . . . . . . . . . . . . . . . VIB6-7,C1-3
Renormalizable gauge . . . . . . . . . . . . . . . VIB3 Spacetime lattice . . . . . . . . . . . . . . . . . . . XIC1
Renormalization . . . . . . . . . . . . . . . . . . . . VIIA Special groups . . . . . . . . . . . . . . . . . . . . . . . IB4
Renormalization group . . . . . . . . VIIB3,6,C1 Special relativity . . . . . . . . IA4,IIA5-B,IIIA4
Renormalization mass scale . . . . . VIIA3,B3 Spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II
Renormalon . . . . . . . . . . . . . . . . . . . . . . . VIIC2 Spin operators . . . . . . . . . . . . . . . . . . . IIB,XII
Representation . . . . . . . . . . . . . . . . . . . . . IB-C Spinor helicity . . . . . . . . . . . . . . . . . . . . VIB2,6
Representation space . . . . . . . . . . . . . . . . . IB1 Spinor index . . . . . . . . . . . . . . . . . . . . . . . . . . II
Residual gauge invar.VIA1,B2,IXB3,XIA3-4 Spinor notation . . . . . . . . . . . . . . . IB5,C5,IIA
Resummation . . . . . . . . . . . . . . . . . . . . . . VIIC Spinor representation . . . . . . . . . . IC5,XC1-2
Retarded propagator . . . . . . . . . . . . VA3,B1-2 Spin-statistics theorem . . . . . . . . . . . . . . VC5
Ricci scalar/tensor . . . . . . . . . . . . . . . . . IXA2 Splitting function . . . . . . . . . . . . . . . . . VIIIC3
Riemann normal coordinates . . . . . . . . IXB4 Spontaneous symmetry breakdown . . . IVA1
Riemann zeta function . . . . . . . . . . . . . VIIA2 Squark . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5
Rotation . . . . . . . . . . . . . . . . . . . . . . . IA1,IIA2 S-supersymmetry . . . . . . . . . . . . . . . . . . . IIC4
R symmetry . . . . . . . . . . . . . . . . . . . . . . . . XA1 Standard Model . . . . . . . . . . . . . . . . . IC4,IVB
Running coupling . . . . . . . . . . . . . . . . . . VIIB6 Stationary phase/steepest descent . . . . VA5
Stirling approximation . . . . . . . . . . . . . VIIC2
Scalar field . . . . . . . . . . . . . . . . . . . . . . . . . IA1 Stokes’ theorem . . . . . . . . . . . . . . . . . . . . . . IC2
Scalar multiplet . . . . . . . . . . . . . . . . . . . . IVC1 String . . . . . . . . . . . . . IVB1,VIIC4,VIIIA3,XI
Scalar potential . . . . . . . . . . . . . . . . . . . . IIIA1 String gauge . . . . . . . IXB5,XA4,XIB5,XIIB8
Scale weight . . . . . . . . . . . . . . . . . . . . IA6,IIB1 String tension . . . . . . . . . . . . . . . . . . . . . . XIA2
Scaling variable . . . . . . . . . . . . . . . . . . . VIIIC3 Strong interaction . . . . . . . . . . . . . . IC4,IVB1
Scattering matrix . . . . . . . . . . . . . . . . . VA4,C Structure constants . . . . . . . . . . . . . . . . . . IA3
 709

Structure function . . . . . . . . . . . . . . . . . . IXA2 Tree graph . . . . . . . . . . . . . . . . . . . . VC2,XIA6

Stückelberg formalism . . . . . . . . . . IIB4,IVA5 Triality . . . . . . . . . . . . . . . . . . . . . . . . . . . . XIB8
Subtraction scheme . . . . . . . . . . . . . . VIIA1,3 Triangle diagram . . . . . . . . . . . . . . . . VIIIB2-5
Summation convention . . . . . . . . . . . . . . . IA1 Twist . . . . . . . . . . . . . . . . . . . . . . . . . . . . VIIIC3
Superconformal group . . . . . . . . . . . . . . . IIC4 Twistor . . . . IIB6-7,C5,IIIC5-7,VIB6-7,C1-3
Supercoordinate . . . . . . . . . . . . . . . IIC2,IIIB1 Two-by-two matrix . . . . . . . . . . . . . . . . IIA-B
Superdeterminant . . . . . . . . . . . . . . . . . . . IIC3
Superficial divergence . . . . . . . . . . . . VIIA1,5 Ultraviolet divergence . . . . . . . . . . . VII-VIII
Superfield . . . . . . . . . . . . . . . . . . . . . . . . . . IIC2 Unitarity . . . . . . . . . . . . IB4,VA4,C5-6,VIIA4
Supergraph . . . . . . . . . . . . . . . . . . . . . . . . VIC5 Unitary gauge . IIIA5,B2,C2,IVA6,VIA4,B3
Supergravity . . . . . . . . . . . . . . . . . . . . . X,XIB3 Unitary group . . . . . . . . . . . . . . . . . . . . . . . IB4
Supergroup . . . . . . . . . . . . . . . . . . . . . . . . . IIC3 U(1) problem . . . . . . . . . . . . . . . . . . . . . . . . IC4
Superhelicity . . . . . . . . . . . . . . . . . . . . . . . IIC5
Superhiggs . . . . . . . . . . . . . . . . . . . . . . . . . XB6 Vacuum . . . . . . . . . . . . IC1,IIIB4,IVA,IXA5
Superpotential . . . . . . . . . . . . . . . . . . . . . IVC2 Vacuum bubble . . . . . . . . . . . . . . . . . . . VIIB1
Superrenormalizable . . . . . . . . . . VIC5,VIIA5 Vacuum value . . . . . IIC5,IVA1,VIIC3,IXA5
Superspace . . . . . . . . . . . . . . . . . . . . . . . . . IIC2 Van Vleck determinant . . . . . . . . . . . . . . VA2
Superspacecone . . . . . . . . . . . . . . . . . VIB7,C3 Vector . . . . . . . . . . . . . . . . . . . . . . . . IA3-4,IIA1
Superstring . . . . . . . . . . . . . . . . . . . . . . XIB3,6 Vector field . . . . . . . . . . . . . . . . . . . . . . . . . . IC2
SupersymmetryIIC,IVC,VIC3,5,VIIIA5-6,X Vector multiplet . . . . . . . . . . . . . . . . . . IVC3-4
Supersymmetry breaking . . . . . . . . . . . . IVC6 Vector potential . . . . . . . . . . . . . . . . . . . . IIA7
Supertrace . . . . . . . . . . . . . . . . . . . . . . IA2,IIC3 Vector representation . . . . . . . . . . . . . . . . . IC5
Supertwistor . . . . . . . . . . . . . . . . . . . . . . . . IIC5 Velocity . . . . . . . . . . . . . . . . . . . . . . . IA1,IIIB1
Super Yang-Mills . . . . . . . . . . . . . . . . . IVC3-5 Vertex . . . . . . . . . . . . . . . . . . . . . . . VC2,VIIIB
SU(2) . . . . . . . . . . . . . . . . . . . . . . . . . . IA3,IIA2 Vertex operator . . . . . . . XIA6,B4,7,XIIB7-8
Symplectic . . . . . . . . . . . . . . . . . . . . . . . . . . IB4 Vielbein . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC3
Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . I-IV Vierbein . . . . . . . . . . . . . . . . . . . . . . . . . . . IXA2
Symmetry breaking . . . . . . . . . . . . . . . IVA,C6 Virasoro constraints . . . . . . . . . . . . . XIA2,B4
Volume element . . . . . . . . . . . . . . . . . . . . IIIB1
Tadpole . . . . . . . . . . . . . . . . . VIB8,C5,VIIB1
T-duality . . . . . . . . . . . . . . . . . . . . . . . . . . XIB4 Ward-Takahashi identity . . . . . . . . . . . VIB8
Temporal gauge . . . . . . . . . . . . . . . . . . . . IIIC2 Wave equation . . . . . . . . . . . . . . . . . . . . . . . IIB
Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . XIA2 Wave-function renorm. . . VIB8,VIIIA3,5,B5
Tensor multiplet . . . . . XA3,B5,XIB6,XIIB8 W boson . . . . . . . . . . . . . . . . . . . . . . . IC4,IVB2
Tensor notation . . . . . . . . . . . . . . . . . . . . . . IB5 Weak interaction . . . . . . . . . . . . . IC4,IVB2-4
Tetrad . . . . . . . . . . . . . . . . . . . . . . . . . . . . IXA2 Weak mixing (Weinberg) angle . . . . . . IVB2
’T Hooft ansatz . . . . . . . . . . . . . . . . . . . . IIIC6 Wess-Zumino gauge . . . . . . . . . . . . . . . . IVC4
Three-vector . . . . . . . . . . . . . . . . . . . . . . . . IIA1 Weyl scale . . . . . . . . . . . . . . . IXA7,B5,XA3-4
Time development . . . . . . . . . . . . . . . . . . . IA1 Weyl spinor . . . . . . . . . . . . . . . . . . . . IC1,IIA5
Time ordering . . . . . . . . . . . . . . . . . IIIA5,VA1 Weyl tensor . . . . . . . . . . . . . . . . . . . . . . . . IXA3
Time reversal . . . . . . . . . . . . . IA5,IIA5,XIB2 White hole . . . . . . . . . . . . . . . . . . . . . . . . . IXC7
Time translation . . . . . . . . . . . . . . . . . . . . . IA1 Wick rotation . . . . . . . . . . . . . . . . IC5,VA5,B4
T-matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . VC4 Wilson loop . . . . . . . . . . . . . . . . . . . . . . . . XIC1
Topological expansion . . . . . . . . . . . . . . VIIC4 Window . . . . . . . . . . . . . . . . . . . . . . . . XIA6,C2
Torsion . . . . . . . . . . . . . . . . . . IVC3,IXA2,XA1 Wino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5
 710

Worldline . . . . . . . . . . . . . . . . . . . . . . . . . . IIIB1 Young tableau . . . . . . . . . . . . . . . . . . . . . . . IC3

Worldsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . XI Yukawa coupling . . . . . . . . . . . . . . . . . IVA4,B
Worldsheet lattice . . . . . . . . . . . . . . . . . . XIC2
Zinn-Justin-Batalin-Vilkovisky . . . . . XIIC2
X-ray . . . . . . . . . . . . . . . . . . . . . . just kidding Zino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVC5
Zweibein . . . . . . . . . . . . . . . . . . . . . . . XIA5,C3
Yang-Mills . . . . . . . . . . . . . . . . . . . . . . . . . IIIC
.......... ..........
.......... Comments on Warren Siegel’s Fields: . . . . . . . . . .
“The price is right.”
“Oh, Warren, what have you done to your students now?”
“I can see you put a lot of work into it.”
“That’s nice, honey.”
“You might want to add a reference to my paper...”
“It’s different.”
“Is this going to be on the exam?”
“I’ll have a look at it when I get the time.”
“Aren’t there enough field theory books already?”
“So this is why you haven’t written any papers lately.”
“Where are the jokes?”