a

r
X
i
v
:
h
e
p
-
p
h
/
0
5
0
6
3
3
0
v
1


3
0

J
u
n

2
0
0
5
What is Renormalization?
G.Peter Lepage
Newman Laboratory of Nuclear Studies
Cornell University, Ithaca, NY 14853
Talk presented at TASI89, June 1989
1 Introduction
As everyone knows the quantized theory of electrodynamics was created in the
late 1920s and early 1930s. The theory was analyzed in perturbation theory,
and was quite successful to leading order in the ne-structure constant . How-
ever all sorts of innities started to appear in calculations beyond that order,
and it was almost twenty years before the technique of renormalization was de-
veloped to deal with these innities. What resulted is quantum electrodynamics
(QED), one of the most accurate physical theories ever created: the g-factor of
the electron is predicted (correctly) by the theory to at least 12 signicant dig-
its! At rst sight, renormalization appears to be a rather dubious procedure
for hiding embarrassing innities, and the success of QED seems nothing less
than miraculous. Nevertheless, persuaded by success, most physicists decided
that renormalizability was an essential ingredient in any physically relevant eld
theory. Such thinking played a crucial role rst in the development of a funda-
mental theory of weak interactions and then in the discovery of the underlying
theory of strong interactions. In this lecture I argue that renormalizability is
not an essential characteristic of useful eld theories. Indeed it is possible, some
would say likely, that none of known interactions is described completely by
a renormalizable eld theory. Modern developments in renormalization theory
have given meaning to nonrenormalizable eld theories, thereby generalizing
and greatly clarifying our understanding of quantum eld theories. As a result
nonrenormalizable interactions are possible and seem likely in most theories
constructed to deal with the real world.
In the rst part of this lecture we will examine the technique of renormaliza-
tion, rst illustrating the conventional ideas and then extending these to deal
with nonrenormalizable interactions. Central to this discussion is the notion of
a cut-o eld theory as a low-energy approximation to some more general (and
possibly unknown) theory. Much of this material warrants a more detailed dis-
cussion than we have time for, and so a number of exercises have been included
to suggest topics for further thought.
In the second part of the lecture we will examine the implications of our new
perspective on renormalization for theories of electromagnetic, strong, and weak
1
interactions. Here we will address such issues as the origins and signicance of
renormalizability, the importance of naturalness in physical theories, and the
experimental limits on nonrenormalizable interactions in electromagnetic and
weak interactions. We will also show how to use renormalization ideas to create
rigorous nonrelativistic eld theories that greatly simplify the analysis of such
nonrelativistic systems as positronium or the meson.
Most of the ideas presented in this lecture are well known to many people.
However, since little of the modern attitude towards renormalization has made
it into standard texts yet, it seems appropriate to devote a lecture to the subject
at this Summer School. The discussion presented here is largely self-contained;
the annotated Bibliography at the end lists a few references that lead into the
large literature on this diverse subject.
2 Renormalization Theory
2.1 The Problem with Quantum Fields
The innities in quantum electrodynamics, for example, originate in the fact
that the electric eld E(x, t) becomes a quantum-mechanical operator in the
quantum theory. Thus measurements of E(x, t) in identically prepared systems
tend to dier: the electric eld has quantum uctuations. By causality, adjacent
measurements of the eld, say at points x and x +a, are independent and thus
uctuate relative to one another. As a result the quantum electric eld is rough
at all length scales, becoming innitely rough at vanishingly small length scales.
Exercise: Dene E(x, t) to be the electric eld averaged over a spherical region of
radius a centered on point x. This might be roughly the eld measured by a
probe of size a. Show that
0|
_
E(x + a, t) E(x, t)
_
2
|0
1
a
4
(1)
as a 0i.e., the uctuations in the eld from point to point diverge as the
probe size goes to zero, even for the vacuum state! Physically, the uctua-
tions arise because it is impossible to probe the electric eld at a point without
creating photons.
This structure at all length scales is characteristic of quantum elds, and
is quite dierent from the behavior of classical elds which typically become
smooth at some scale. The roughness of the eld is at the root of the problem
with dening the quantum eld theory. For example, how does one dene
derivatives of a eld E(x) when the dierence E(x +a) E(x) diverges as the
separation a vanishes? In perturbation theory the roughness at short distances
results in divergent integrations over loop momenta, divergences associated with
intermediate states carrying arbitrarily large momenta (and having arbitrarily
short wavelengths). The innities that result demonstrate rather dramatically
that the short-distance structure of the quantum elds plays an important role
in determining the long-distance (low-momentum) behavior of the theory; the
short-distance structure cannot be ignored.
2
To give any meaning at all to a quantum eld theory one must rst regulate
it, by in eect removing from the theory all states having energies much larger
than some cuto . With a cuto in place one is no longer plagued by innities
in calculations of the scattering amplitudes and other properties of the theory.
For example, integrals over loop momenta in perturbation theory are cut o
around and thus are well dened. However the cuto seems very articial.
The use of a cuto apparently contradicts the notion, developed above, that the
short-distance structure of the theory is important to the long-distance behavior;
with the cuto one is throwing away the short-distance structure. Furthermore
is a new and articial parameter in the theory. Thus it is traditional to remove
the cuto by taking to innity at the end of any calculation. This last step is
the source of much of the mystery in the renormalization procedure, and it now
appears likely that this last step is also a wrong step in the nonperturbative
analysis of many theories, including QED. Rather than follow this route we now
will examine what it means to keep the cuto nite.
2.2 Cut-o Field Theories
The basic idea behind renormalization is that all eects of the very high-energy
states in the Hilbert space on the low-energy behavior of the theory can be
simulated by a set of new local interactions. So we can discard the states with
energy greater than some cuto provided we modify the theorys Lagrangian to
account for the eects that result from the discarded states. In this section we
will see how this idea provides the basis for the conventional renormalization
procedure, using QED as an example of a renormalizable eld theory. This will
lay the groundwork for our discussion, in the next section, of the origins and
signicance of nonrenormalizable interactions.
According to conventional renormalization theory QED is dened by a La-
grangian,
L
0
= (i e
0
A m
0
)
1
2
(F

)
2
, (2)
together with a regulator that truncates the theorys state space at some very
large
0
.
(a)
The cut-o theory is correct up to errors of (1/
2
0
). It is worth
emphasizing that e
0
and m
0
are well-dened numbers so long as
0
is kept nite;
in QED each can be specied to several digits (for any particular value of
0
).
Given these bare parameters one need know nothing else about renormaliza-
tion in order to do calculations. One simply computes scattering amplitudes,
cutting all loop momenta o at
0
and using the bare parameters in propaga-
tors and vertices. The renormalization takes care of itself automatically. To
compute e
0
and m
0
for a particular
0
one chooses two convenient processes
or quantities, computes them in terms of the bare parameters using L
0
, and
(a)
The simplest way to regulate perturbation theory is to simply cut o the integrals over loop
momenta at
0
. Although such a regulator can (and has been) used, it complicates practical
calculations because it is inconsistent with Lorentz invariance and gauge invariance. However
the details of the regulator are largely irrelevant to our discussion, and so for simplicity we will
speak of the regulator as though it is a simple cuto. One of the more conventional regulators,
such as Pauli-Villars or lattice regulators, is recommended for real calculations.
3
k
p p

Figure 1: The one-loop vertex correction to the amplitude for an electron scat-
tering o an external eld.
adjusts the bare parameters until theory and experiment agree. Then all other
predictions of the theory will be correct, up to errors of (1/
2
0
).
(b)
To understand the role of the cuto in dening the theory, we start with
QED as dened by L
0
and cuto
0
, and we remove from this theory all states
having energies or momenta larger than some new cuto (
0
). Then we
examine how L
0
must be changed to compensate for this further truncation
of the state space. Of course the new theory that results can only be useful
for processes at energies much less than , and so we restrict our attention to
such processes. Furthermore we will analyze the eects of the new cuto using
perturbation theory, although our results are valid nonperturbatively as well.
We now want to discard all contributions to the theory coming from loop
momenta greater than the new cuto . Consider rst, the one-loop radiative
corrections to the amplitude for an electron to scatter o an external electro-
magnetic eld.
Working in the L
0
theory with the original cuto, the part of the vertex
correction (Fig. 1) that is being discarded is
T
(a)
(k > ) = e
3
0
_
0

d
4
k
(2)
4
1
k
2

u(p

1
(p

k) m
0
A
ext
(p

p)
1
(p k) m
0

u(p). (3)
Since the masses and external momenta are assumed to be much less than ,
we can neglect m
0
, p and p

in the integrand as a rst approximation, thereby

(b)
Newcomers to eld theory sometimes nd it hard to believe that this is all there is to
conventional renormalization, given the length and complexity of the treatment generally
accorded the subject in texts. What happened to counterterms, subtractions points, and nor-
malization conditions? These are all related to the detailed implementation of the general
concepts we are discussing. Such implementations tend to be highly optimized for particular
sorts of calculationse.g., for high-order calculations in perturbation theory, or for lattice
simulationsand as such can be fairly complex. Such details are important in actual calcu-
lations. Here however our focus is on conceptual issues and so we can dispense with much of
the detail. Anyone planning to do real calculations is advised to consult the standard texts.
4
greatly simplifying the integral:
T
(a)
(k > ) e
3
0
_
0

d
4
k
(2)
4
1
k
2
u(p

k
k
2
A
ext
(p

p)
k
k
2

u(p)
e
3
0
u(p

) A
ext
(p

p) u(p)
_
0

d
4
k
(2)
4
1
(k
2
)
2
. (4)
Applying a similar analysis to the other one-loop corrections, we nd that the
part of the electrons scattering amplitude that is omitted as a result of the new
cuto has the form
T(k > ) ie
0
c
0
(/
0
) u(p

) A
ext
(p

p) u(p) (5)
where c
0
is dimensionless and thus can depend only upon the ratio /
0
, these
being the only scales left in the loop integration.
Exercise: Show that
c0(/0) =
0
6
log(/0). (6)
Clearly T(k > ) is an important contribution to the electrons scattering
amplitude; it cannot be dropped. However such a contribution can be reincor-
porated into the theory by adding the following new interaction to L
0
:
L
0
= e
0
c
0
(/
0
) A . (7)
Thus we can modify the Lagrangian to compensate for the removal of the states
above the cuto, at least for the purpose of computing the electrons scattering
amplitude.
It is important that the new interaction L
0
is completely specied by a
single number, the coupling constant c
0
. In analyzing the vertex correction to
electron scattering (Eq. (3)), we can neglect the external momenta relative to
the internal momentum k with the result that the coupling c
0
is independent
of the external momenta. Thus the interaction is characterized by a number,
rather than by some complicated function of the external momenta. Momentum
independence, or more generally polynomial dependence on external momenta,
indicates that these eective interactions are local in coordinate space; that is,
they are polynomial in elds or derivatives of the elds all evaluated at the same
point x. This important result actually follows from the uncertainty principle
and is quite general: interactions involving intermediate states with momenta
greater than the cuto are local as far as the (low-energy) external particles
are concerned. Since by assumption the external particles have momenta far
smaller than , intermediate states above the cuto must be highly virtual. In
quantum mechanics a state can be highly virtual provided it is short-lived, and
so these highly-virtual intermediate states can exist for times and propagate
over distances of only (1/). Such distances are tiny compared with the
wavelengths of the external particles, 1/p 1/, and thus the interactions
are eectively local.
5
a)
k
p
b) k
p
Figure 2: Examples of one-loop radiative corrections to electron-electron scat-
tering.
Although electron scattering on external elds has been xed, we must worry
about the eect of the new cuto on other processes. Consider for example the
one-loop corrections to electron-electron scattering (Fig. 2). The k > contri-
butions due to vertex and self-energy corrections to the one-photon exchange
process (e.g., Fig. 2a) are correctly simulated by L
0
, just as they are in the
case of an electron scattering on an external eld. That leaves only the k >
contribution from the two-photon exchange diagrams (e.g., Fig. 2b). Again one
can neglect the external momenta and masses in the internal propagators. The
resulting amplitude must involve spinors for each of the external electrons, in
combinations like u

u u

u or uu uu. These all have the dimension of [energy]

2
,
while in general a four-particle amplitude must be dimensionless. Thus the part
of the amplitude involving loop momenta larger than contributes something
like
d(/
0
)
u

u u

2
, (8)
where d is dimensionless and where the factor in the denominator is
2
, since
is the only important scale left in the loop integral. Clearly such a contribution
is suppressed by (p/)
2
and can be ignored (for the moment) since we are
assuming p.
A similar analysis for, say, electron-electron scattering into four electrons
and two positrons (e.g., Fig. 3) shows that intermediate states above the cuto
contribute something of order
(uu)
2
(uv)
2

8
, (9)
which is even less important. Evidently the more external particles involved in
a loop, the larger the number of hard internal propagators, and the smaller the
eect of the cuto.
Exercise: Show that an amplitude with n external particles has dimension [energy]
4n
when relativistic normalization (i.e., p . . .|p

. . .
_
p
2
+ m
2

3
(pp

)) is used
for all states. Use this fact to show that the addition of an extra pair of external
fermions to a k > loop results in an extra factor of 1/
3
, while adding an
extra external photon leads to an extra factor of 1/. For some processes addi-
tional factors of external momenta or of the electron mass may also be required
6
k
p
Figure 3: A one-loop radiative correction to electron-electron scattering i into
four electrons and two positrons.
respectively by gauge invariance or chiral symmetry (i.e., electron-helicity con-
servation when m=0). Such factors result in additional factors of 1/. (Note
that the relativistic normalization condition for Dirac spinors is uu =2m, and
that photon polarization vectors are normalized by = 1.)
Using simple dimensional and power-counting arguments of this sort, one
can show that the only scattering amplitudes that are strongly aected by the
cuto are those involving the electron-photon vertex, and for these the loss of
the k> states is correctly compensated by adding the single correction L
0
to
the Lagrangian. The only other physical quantity that is strongly aected by
the cuto is the mass of the electron: the physical mass of the electron is m
0
plus a self-energy correction that involves the k > states. The eect of these
states on the mass is easily simulated by adding a term of the form
m
0
c
0
(/
0
) (10)
to the Lagrangian.
(c)
Thus the theory with Lagrangian
L

= (i e

A m

)
1
2
(F

)
2
, (11)
cuto , and coupling parameters
e

= e
0
(1 +c
0
(/
0
)) (12)
m

= m
0
(1 + c
0
(/
0
)) (13)
gives the same results as the original theory with cuto
0
(up to corrections
of (1/
2
)).
(c)
It is not obvious at rst glance that this mass correction is proportional to m
0
. On strictly
dimensional grounds one might expect a term proportional to
0
. However such a term is
ruled out by the chiral symmetry of the massless theory. If m
0
is set equal to zero, then
the original theory is symmetric under chiral transformations of the form exp(i
5
).
Changing the cuto does not aect this symmetry and therefore any new interaction that
violates chiral symmetry must vanish if m
0
vanishes. Thus upon calculating the coecient of
the new interaction one nds that it is proportional to m
0
rather than
0
; that is, the
mass renormalization is logarithmically divergent rather than linearly divergent. Note by way
of contrast that chiral symmetry is explicitly broken in Wilsons implementation of fermions
on a lattice, and consequently the mass renormalization in such a theory is proportional to

0
rather than m
0
.
7
With this result we see that a change in the cuto can be compensated by
changing the bare coupling and mass in the Lagrangian in such a way that the
low-energy physics of the theory is unaected. This is the classical result of
renormalization theory.
Exercise: Sketch out arguments for the validity of this result to two-loop order by
examining some simple process like electron-electron scattering. Divide each
loop in a diagram into high-energy and low-energy parts, with as the dividing
line. This means that each two-loop diagram will be divided into four contribu-
tions depending upon loop energies: high-high, high-low, low-high, and low-low.
The low-low contribution is still present with the new cuto. The low-high and
high-low contributions are removed by the cuto, but these are either local in
character, down by 1/, or are automatically simulated in the new theory by
diagrams in which the high-energy loop is replaced by one of the new interaction
introduced above to correct one-loop results (Eqs. (7) and (10)). The part of
the low-high and high-low contributions that is local can be treated with the
high-high contribution. The high-high contribution must be completely local
and can be simulated by a local interaction in the Lagrangian. Again by power-
counting, only the electron-photon vertex and the electron mass are appreciably
changed by the cuto, and thus these new local interactions are taken care of
by introducing corrections of relative order e
4
0
into e and m.
Exercise: Parameters e and m vary as the cuto is varied. The couplings are
said to run as more or less of the state space is included in the cut-o theory.
Show that these couplings satisfy evolution equations of the form:

de
d
= (e) (14)

dm
d
= mm(e). (15)
(In these equations we assume that m is negligible compared with ; more
generally and m depend also on the ratio m/.) Compute and m to
leading order in e for QED.
The bare parameters e and m can be thought of as the eective charge and
mass of an electron at energy-momentum scales of (). This is particularly
relevant in analyzing a process that is characterized by only a single scale, say
Q. To compute the amplitude for such a process in a cut-o theory one must
take much larger than Q. However the main eect of vertex and self-energy
corrections to the amplitude is to replace e and m by eQ and mQ everywhere
in the amplitude. Thus such amplitudes are most naturally expressed in terms
of the bare parameters for the theory with cuto = Q. This result is not
surprising insofar as physical amplitudes are independent of the actual value of
the cuto used. The natural way to express this independence is to calculate
with Q but then to reexpress the result in terms of the running parameters
at scale Q. In this way one removes all explicit reference to the actual cuto, and
in particular one removes large logarithms of /Q that otherwise tend to spoil
the convergence of perturbation theory. This procedure, while useful in QED,
has proven essential in perturbative QCD where the convergence of perturbation
theory is marginal at best.
8
2.3 Beyond Renormalizability
It is clear from our analysis in the last section that the cut-o theory is accurate
only up to corrections of (p
2
/
2
) where p is typical of the external momenta.
In practice such errors may be negligible, but if one wishes to remove them there
are two options. One is the traditional option of taking to innity. The other
is to keep nite but to add further corrections to the Lagrangian. This second
choice is far more informative and useful.
To illustrate the procedure consider again the k> part of the amplitude for
an electron scattering on an external eld as in Eq. (3). In our earlier analysis
we neglected external momenta and masses relative to the loop momentum. We
can correct this approximation by making a Taylor expansion of the amplitude
in powers of p/, p

/, and m
0
/ to obtain terms of the form
T(k > ) = ie
0
c
0
uA
ext
u

ie
0
m
0
c
1

2
uA

ext

(p p

ie
0
c
2

2
(p p

)
2
u A
ext
u +. . . , (16)
where coecients c
0
, c
1
. . . are all dimensionless, and where the structure of the
amplitude is constrained by the need for current conservation and for chiral
invariance in the limit m
0
=0. The eects of all of these terms can be simulated
by adding new local interactions to the Lagrangian. The rst term was handled
in the previous section by simply replacing the bare charge e
0
by e

; the only
dierence now is that contributions to c
0
of (m
2
0
/
2
) must be retained. The
remaining terms in T(k>) require the introduction of new types of interaction:
L
a
2
=
e
0
m
0
c
1

2
F

+
e
0
c
2

2
i

. (17)
These interactions are designed by including a eld for each external particle
in T(k > ), and a derivative for each power of an external momentum. By
augmenting the Lagrangian with such terms we can systematically remove all
(p
2
/
2
) errors from the cut-o theory. Of course such errors arise in processes
other than electron scattering o a eld, and further terms must be added to
the Lagrangian to compensate for these. For example, the p
2
/
2
contribution
coming from k > in electron-electron scattering (Eq. (8)) is compensated by
interactions like
L
b
2
=
d

2
(

)
2
. (18)
Luckily power-counting and dimensional analysis tell us that only a few pro-
cesses are aected by the new cuto to this order, and therefore only a -
nite number of terms need be added to the Lagrangian to remove all errors of
(p
2
/
2
) in all processes.
It seems remarkable that the p
2
/
2
errors for an innity of processes can
removed from the cut-o theory by adding a nite (even small) number of new
interactions to the Lagrangian. However, one can show, even without examining
9
particular processes, that there is only a nite number of possible new interac-
tions that might have been relevant to this order. Interactions that simulate
k > physics must be locali.e., polynomial in the elds, and derivatives of
the eldsand they must have the same symmetries as the underlying theory.
In QED, these symmetries include Lorentz invariance, gauge invariance, parity
conservation, and so on.
(d)
In addition the interaction terms must have (energy)
dimension four; thus an interaction operator of dimension n + 4 must have a
coecient of (1/
n
) or smaller, being the smallest scale in the k> part of
the theory. Dimension-six operators are obviously important in (1/
2
), but
operators of higher dimension are suppressed by additional powers of 1/ and
so are irrelevant to this order. There are very few operators of dimension six
or less that are Lorentz and gauge invariant, and that can be constructed from
polynomials of (dimension 3/2), A

(dimension 1), and

(dimension 1).
And these few are the only ones needed to correct the Lagrangian of the cut-o
theory through order p
2
/
2
.
Exercise: It is critical to our discussion that an operator of dimension n + 4 in the
Lagrangian only aect results in order (p/)
n
(or less). So introducing, for ex-
ample, a dimension-eight operator should not aect the predictions of the theory
through order (p/)
2
. This is obviously the case at tree level in perturbation
theory, since the coecient of the dimension-eight operator has a factor 1/
4
that must appear in the nal result for any tree diagram involving the interac-
tion. Such a contribution will then be suppressed by a factor (p/)
4
. However
in one-loop order (and beyond) loop integrations can supply powers of that
cancel the powers of 1/ explicit in the interaction, resulting in contributions
from the new interaction that are not negligible in second order. Show that these
new contributions can be cancelled by appropriate shifts in the coecients of
the lower dimension operators in the Lagrangian. Thus the dimension-eight
operator has no net eect on the results of the theory through order p
2
/
2
.
This procedure can obviously be extended so as to correct the theory to any
order in p/, but the price of this improved accuracy is a more complicated
Lagrangian. So why bother with cut-o eld theory? Certainly in perturbation
theory it seems desirable that the number of interactions be kept as few as
possible so as to avoid an explosion in the number of diagrams. On the other
hand the theory with no cuto or with is ill-dened. Furthermore in
practice it is essential to reduce the number of quantum degrees of freedom
before one is able to solve a quantum eld theory. A cuto does just this.
While cutos tend to appear only in intermediate steps of perturbative analyses,
they are of central importance in most nonperturbative analyses. In numerical
treatments, using lattices for example, the number of degrees of freedom must
be nite since computers are nite.
The practical consequences of our analysis are important, but the most strik-
ing implication is that nonrenormalizable interactions make sense in the context
(d)
If the regulator breaks one of the symmetries of the theory then interactions that break
the symmetry will also arise. These interactions serve to cancel the symmetry-breaking eects
of the regulator. In lattice gauge theory, for example, the lattice regulator breaks the Lorentz
symmetry and as a consequence interactions in such a theory can be Lorentz non-invariant.
10
of a cut-o eld theory. Insofar as experiments can probe only a limited range
of energies it is natural to analyze experimental results using cut-o eld theo-
ries. Since nonrenormalizable interactions occur naturally in such theories it is
probably wrong to ignore them in constructing theoretical models for currently
accessible physics.
Exercise: The idea behind renormalization is rather generally applicable in quantum
mechanics. Suppose we have a problem in which the states of most interest
all have low energies, but couple through the Hamiltonian to states with much
higher energies. To simplify the problem we want to truncate the state space,
excluding all states but the (low-energy) ones of interest. Suppose we dene
a projection operator P that projects onto this low-energy subspace; its com-
plement is Q 1 P. Then show that the full Hamiltonian H for the theory
can be replaced by an eective Hamiltonian that acts only on the low-energy
subspace:
H
e
(E) = HPP + HPQ
1
E HQQ
HQP (19)
where HPQ PHQ. . . . In particular if |E is an eigenstate of the full Hamil-
tonian then its projection onto the low-energy subspace |EP P|E satises
H
e
(E)|EP = E|EP. (20)
Note that since the energies E in the P-space are much smaller than the energies
in the Q-space, the second term in H
e
can be expanded in a powers of E/HQQ.
Thus the new interactions are polynomial in E and therefore local in time. In
practice only a few terms might be needed in this expansion. Such truncations
are made all the time in atomic physics. For example, if one is doing radio-
frequency studies of the hyperne structure of the ground state of hydrogen one
usually wants to forget about all the radial excitations of the atom (i.e., optical
frequencies).
2.4 Structure and Interpretation of Cut-o Field Theories
In the preceding sections we illustrated the nature of a cut-o eld theory using
QED perturbation theory. It should be emphasized that most of what we dis-
covered is not tied to QED or to perturbation theory. The central notion, that
the eects of high-energy states on low-energy processes can be accounted for
through the introduction of local interactions, follows from the uncertainty prin-
ciple and so is quite general. We can summarize the general ideas underlying
cut-o eld theories as follows:
A nite cuto can be introduced into any eld theory for the purposes
of discarding high-energy states from the theory. The cut-o theory can
then be used for processes involving momenta p much less than .
The eects of the discarded states can be retained in the theory by adjust-
ing the existing couplings in the Lagrangian, and by adding new, local,
nonrenormalizable interactions. These interactions are polynomial in the
elds and derivatives of the elds. The nonrenormalizable couplings do
not result in intractable innities since the theory has a nite cuto.
11
Only a nite number of interactions is needed when working to a partic-
ular order in p/, where p is a typical momentum in whatever process
is under study. The cuto usually sets the scale of the coecient of an
interaction operator: the coecient of an operator with (energy) dimen-
sion n + 4 is (1/
n
), unless symmetries or approximate symmetries of
the theory further suppress the interaction. Therefore one must include
all interactions involving operators with dimension n+4 or less to achieve
accuracy through order (p/)
n
.
Conversely an operator of dimension n+4 can only aect results at order
(p/)
n
or higher, and so may be dropped from the theory if such accuracy
is unnecessary.
Note that the interactions in a cut-o eld theory are completely specied
by the requirement of locality, by the symmetries of the theory and regulator,
and by the accuracy desired. The structure of the operators introduced into
the Lagrangian does not depend upon the detailed dynamics of the high-energy
states being discarded. It is only the numerical values of the coecients of these
operators that depend upon the high-energy dynamics. Thus while the high-
energy states do have a strong eect on low-energy processes we need know very
little about the high-energy sector in order to compute low-energy properties of
the theory. The coupling constants of the cut-o theorye

, m

, c
1
, c
2
, d. . . in
the previous sectionscompletely characterize the behavior of the discarded
high-energy states for the purposes of low-energy analyses. In cases where we
understand the dynamics of the discarded states, as in our analysis of QED, we
can compute these coupling constants. In other situations we are compelled to
measure them experimentally.
The expansion of a cut-o Lagrangian in powers of 1/ is somewhat anal-
ogous to multipole expansions used in classical eld theory. For example, the
detailed charge distribution of a nucleus is of little importance to an electron in
an atomic orbital. To compute the long range electrostatic eld of the nucleus
one need only know the charge of the nucleus, and, depending upon the level
of accuracy desired, perhaps its dipole and quadrupole moments. Again, for all
practical purposes the eects of short-range structure on long-range behavior
can be expressed in terms of a nite number of numbers, the multipole moments,
characterizing the short-range structure.
Armed with this new understanding of eld theory, it is time to reexamine
traditional theories in an eort to better understand why they are the way they
are and how they might be changed to accommodate future experiments.
3 Applications
3.1 Why is QED renormalizable?
Having argued that nonrenormalizable interactions are admissible one has to
wonder why QED is renormalizable after all. What do we learn from the fact
12
of its renormalizability? Our new attitude towards renormalization suggests
that the key issue in addressing a theory like QED is not whether or not it
is renormalizable, but rather how renormalizable it isi.e., how large are the
nonrenormalizable interactions in the theory.
It is quite likely that QED is a low-energy approximation to some compli-
cated high-energy supertheory (a string theory?). Consequently there will be
some large energy scale beyond which QED dynamics are insucient to describe
nature, where the supertheory will become necessary. Since we have yet to g-
ure out what the supertheory is, it is natural in this scenario that we introduce
a cuto equal to this energy scale so as to exclude the unknown physics. The
supertheory still aects low-energy phenomena, but it does so only through the
values of the coupling constants that appear in the cut-o Lagrangian:
L

= (i e

A m

)
1
2
(F

)
2
+
+
e

c
1

2
F

+
e

c
2

2
i

+
d

2
(

)
2
+. . . . (21)
We cannot calculate the coupling constants in this Lagrangian until we discover
and solve the supertheory; the couplings must be measured. The nonrenormal-
izable interactions are certainly present, but if is large their aect on current
physics will be very small, down by (p/)
2
or more. The fact that we havent
needed such terms to account for the data tells us that is indeed large. This is
the key to the signicance of renormalizability and its origins: very low-energy
approximations to arbitrary high-energy dynamics can be formulated in terms
of renormalizable eld theories since nonrenormalizable interactions would be
suppressed in their eects by powers of p/ and therefore would be irrelevant
for p . This analysis also tells us how to look for low-energy evidence of
new high-energy dynamics: look for (small) eects caused by the leading non-
renormalizable interactions in the theory. Indeed we can estimate the energy
scale at which new physics must appear simply by measuring the strength
of the nonrenormalizable interactions in a theory. The low-energy theory must
fail and be replaced at energies of order . Processes with p start to probe
the detailed structure of the supertheory, and can no longer be described by the
low-energy theory; the expansion of the Lagrangian in powers of 1/ no longer
converges.
As we noted, the successes of renormalizable QED imply that the energy
threshold for new physics in electrodynamics is rather high. For example, the
F interaction in L

would shift the g-factor in the electrons magnetic

moment by an amount of order (m
e
/)
2
. So the fact that renormalizable QED
accounts for the g-factor to almost 12 digits implies that
m
2
e

2
< 10
12
(22)
from which we can conclude that is probably larger than a TeV.
13
3.2 Strong Interactions: Pions or Quarks?
QED is spectacularly successful in explaining the magnetic moment of the elec-
tron. Its failure to explain the magnetic moment of the proton is equally spec-
tacular: g
p
is roughly three times larger than predicted by QED when the proton
is treated as an elementary, point-like particle. Nonrenormalizable terms in L

make contributions of order unity,

m
2
p

2
1, (23)
indicating that the cuto in proton QED must be of order the protons mass
m
p
. Thus QED with an elementary proton must fail around 1 GeV, to be
replaced by some other more fundamental theory. That new theory is quantum
chromodynamics, of course, and the large shift in the protons magnetic moment
is a consequence of its being a composite state built of quarks. This example
illustrates how the measured strength of nonrenormalizable interactions can be
used to set the energy scale for the onset of new physics. Also it strongly suggests
that one ought to use QCD in analyzing strong interaction physics above a GeV.
This result also suggests that one can and ought to treat the proton as a
point-like particle in analyses of sub-GeV physics. In the hydrogen atom, for
example, typical energies are of order a few eV. At such energies the protons
electromagnetic interactions are most eciently described by a cut-o QED (or
nonrelativistic QED) for an elementary proton. The cuto should be set
below a GeV, and the anomalous magnetic moment, charge radius and other
properties of the extended proton should be simulated by nonrenormalizable
interactions of the sort we have discussed. This theory can be made arbitrarily
precise by adding a sucient number of interactionsan expansion in 1/and
by working to suciently high order in perturbation theoryan expansion in .
With the cuto in place, the nonrenormalizable interactions cause no particular
problems in perturbation theory.
In the same spirit one expects very low-energy strong interactions to be ex-
plicable in terms of a theory of point-like mesons and baryons. Indeed these
ideas account for the great success of PCAC and current algebra in describing
pion-pion and pion-nucleon scattering at threshold. A cut-o theory of elemen-
tary pions, protons and other hadrons must work above threshold as well, but
will ultimately fail somewhere around a GeV. It is still an open question as to
just where the cuto lies. In particular it is unclear whether or not nuclear
physics can still be eciently described by pion-nucleon theories at energies of a
few hundred MeV, or whether the quark structure is already important at such
energies. To evaluate the utility of the pion-nucleon theory one must treat it as
a cut-o eld theory, taking care to make consistent use of the cuto through-
out, and systematically enumerating the possible interactions, determining their
couplings from experimental data. It also seems likely that some sort of (sys-
tematic) nonperturbative approach to the problem is essential. This is clearly a
crucial issue for nuclear physics since a theory of point-like hadrons, if it works,
should be far simpler to use than a theory of quarks and gluons.
14
3.3 Weak Interactions: Renormalization, Naturalness and
the SSC
The Standard Model of strong, electromagnetic and weak interactions has been
enormously successful in accounting for experimental data up to energies of order
100GeV or more. Still there remain several unanswered questions, and the most
pressing of these concerns the origins of the masses for the W and Z bosons. A
whole range of possible mechanisms has been suggestedthe Higgs mechanism,
technicolor, a supersymmetric Higgs particle, composite vector bosonsand
extensive experimental searches have been conducted. Yet we still do not have
a denitive answer. What will it take to unravel this puzzle?
In fact new physics, connected with the Z mass, must appear at energies of
order a few TeV or lower. Current data indicates that the vector bosons are
massive and that their interactions are those of nonabelian gauge bosons, at least
insofar as the quarks are concerned. This suggests that a minimal Lagrangian
for the vector bosons that accounts for current data would be the standard
Yang-Mills Lagrangian for nonabelian gauge elds plus simple mass terms for
the Ws and Z. Traditionally such a theory is rejected immediately since the
mass terms spoil gauge invariance and ruin renormalizability. However from
our new perspective, the appearance of nonrenormalizable terms in the minimal
theory poses no problems; it indicates that this theory is necessarily a cut-o
eld theory with a nite cuto. Thus the theory with minimal particle content
becomes inadequate at a nite energy of order the cuto, and new physics is
inevitable. By examining the nonrenormalizable interactions one nds that the
value of the cuto is set by the vector-boson mass M:
(e)

4M

(24)
This cuto is of order a few TeV, and it represents the threshold energy for new
physics.
This analysis is the basis for one of the most important scientic arguments
for building an accelerator, like the SSC, that can probe few-TeV physics; the
secret behind the W and Z masses almost certainly lies in this energy range.
One potential problem with this argument is the possibility that the Higgs par-
ticle exists and has a mass well below a few TeV, in which case the interesting
physics is at too low an energy for an SSC. In fact the argument survives be-
cause a theory with a low-mass Higgs particle is unnatural unless there is new
(e)
It is the longitudinal degrees-of-freedom in a massive Yang-Mills theory that spoil the
renormalization. The longitudinal part of the action can be isolated through gauge transfor-
mations and shown to be equivalent to a nonlinear sigma model, which is nonrenormalizable.
To see this simply, one can examine the standard theory with a Higgs particle. One arranges
the scalar couplings of this theory in such a way that the mass of the Higgs becomes innite,
while keeping the vector-boson masses constant. What remains, once the Higgs particle is
removed in this fashion, is just a theory of massive gauge bosons. The part of the Lagrangian
dealing with the Higgs particle becomes a nonlinear sigma model in this limit, with the mag-
nitude of the complex scalar eld frozen at its vacuum expectation value v 200 GeV. The
sigma model is nonrenormalizable but makes sense as a cut-o theory provided the cuto is
less than (4v), which is equivalent to the limit given in the text.
15
structure, again at energies of order a few TeV. If we regard the standard model
with a Higgs eld as a cut-o eld theory, approximating some more complex
high-energy theory, then we expect that the scale of dimensionful couplings in
the cut-o Lagrangian is set by the cuto . In particular the bare mass of
the scalar boson is of order the cuto, and thus the renormalized mass of the
Higgs particle ought to be of this order as well. Since this theory is technically
renormalizable, one can make the Higgs mass much smaller than the cuto by
ne tuning the bare mass to cancel almost exactly the large mass generated
by quantum uctuations; but such tuning is highly contrived, and as such is
unlikely to occur in the low-energy approximation to any more complex theory.
Thus the Higgs theory is sensible only when the cuto is nite. To estimate
how large a natural cuto might be, we compare the bare mass with the mass
renormalization due to quantum uctuations. Barring (unlikely) accidental can-
cellations, one expects the physical and bare masses to have roughly the same
order of magnitude; that is, we expect
m
2
H

2
+m
2
|
2
|, (25)
where m
2
is the mass renormalization, and m
2
H
and
2
are the squares of the
physical and bare masses respectively. The mass renormalization is quadrati-
cally sensitive to the cuto,
m
2

2
(26)
where is the
4
coupling constant. Therefore the bare and physical masses
can be comparable only if the cuto is less than

|
2
|

, (27)
which turns out to be the same limit we obtained above for the nonrenormaliz-
able massive Yang-Mills theory (Eq. (24)).
(f)
Therefore new physics is expected
in the few-TeV region whether or not there is a Higgs particle at lower energies.
The Yang-Mills theory coupled to a Higgs doublet gained widespread accep-
tance because it was renormalizable. However, our new understanding of cut-o
eld theories suggests that naturalness, rather than renormalizability, is the key
property of a physical theory. And from this perspective, the renormalizable
theory with a Higgs doublet is neither better nor worse than the nonrenormal-
izable massive Yang-Mills theory. One might even argue that the latter model
is the more attractive given that it is minimal. Both theories are predictive in
the 100 GeV region and below; neither theory can survive without modication
much beyond a few TeV.
(f)
Recall that the complex scalar eld generates a mass for the gauge bosons by acquiring
a vacuum expectation value. Replacing i by i gA in the Lagrangian for a free scalar
boson leads to an interaction term g
2
A
2

2
/2. This becomes a mass term for the A eld if the
eld has nonzero vacuum expectation value v, the mass being gv. Thus v is of order M/

.
This v arises from the competition between the bare scalar mass term and the
4
interaction,
with the result that v
2
is also of order |
2
|/, which in turn is of order the cuto squared in
a natural theory. Combining these two expressions for v gives the limit in Eq. (24).
16
Our analysis of the Higgs sector of the standard weak interaction theory il-
lustrates a general feature of physically relevant scalar eld theories: the renor-
malized mass of the eld is most likely as large as the bare mass, and both are
at least as large as the mass renormalization due to quantum uctuations. In
strongly interacting theories this means that the renormalized mass is of order
the cuto; scalars with masses small compared to are unnatural. This re-
sult should be contrasted with the situation for spin-1/2 particles and for gauge
bosons. In the case of vector bosons like the photon, the bare mass in the La-
grangians mass term, M
2

A
2
/2, would indeed be of order the cuto were it not
for gauge invariance, which requires M

=0. Similarly for fermions like the elec-

tron, chiral symmetry implies that all mass renormalizations must vanish if the
bare mass does. Consequently one has m
e
m
e
even though the cuto is much
larger. In general one expects low-mass particles in a theory only when there is a
symmetry, like gauge invariance or chiral symmetry, that protects the low mass.
No such symmetry exists for scalar bosons, unless they are tied to fermionic
partners via a supersymmetry. This may explain why all existing experimental
data in high-energy physics can be understood in terms of elementary particles
that are either spin-1/2 fermions or spin-1 gauge bosons.
3.4 Nonrelativistic Field Theories
Nonrelativistic systems, such as positronium or the and mesons, play an
important role in several areas of elementary particle physics. Being nonrel-
ativistic these systems are generally weakly coupled, and as a result typically
involve only a single channel. Thus, for example, the is predominantly a
bound state of a b quark and antiquark. It has some probability for being in a
state comprised of a bb pair and a gluon, or a bb pair and a uu pair, etc., but the
probabilities are small and therefore these channels have only a small eect on
the gross physics of the meson. This is an enormous simplication relative to
relativistic systems where many channels may be important, and it is this that
accounts for the prominence of such systems in fundamental studies of electro-
magnetic and strong interactions. Nevertheless there are signicant technical
problems connected with the study of nonrelativistic systems. Central among
these is the problem of too many energy scales. Typically a nonrelativistic sys-
tem has three important energy scales: the masses m of the particles involved,
their three-momenta pmv, and their kinetic energies KEmv
2
. These scales
are widely dierent in a nonrelativistic system since v 1 (where the speed of
light c = 1), and this greatly complicates any analysis of such a system. In this
section we will see how renormalization can be used as a tool to dramatically
simplify studies of this sort.
One can appreciate the problems that arise when analyzing these systems
by considering a lattice simulation of the meson. The space-time grid used in
such a simulation must accommodate wavelengths covering all of the of scales in
the meson, ranging from 1/mv
2
down to 1/m. Given that v
2
0.1 in the , one
might easily need a lattice as large as 100 sites on side to do a good job. Such
a lattice could be three times larger than the largest wavelength, with a grid
17
Figure 4: A two-loop kernel contributing to the Bethe-Salpeter potential for
positronium.
spacing three times smaller than the smallest wavelength, thereby limiting the
errors caused by the grid. This is a fantastically large lattice by contemporary
standards and is quite impractical.
The range of length scales also causes problems in analytic analyses. As an
example, consider a traditional Bethe-Salpeter calculation of the energy levels
of positronium. The potential in the Bethe-Salpeter equation is given by a
sum of two-particle irreducible Feynman amplitudes. One generally solves the
problem for some approximate potential and then incorporates corrections using
time-independent perturbation theory. Unfortunately, perturbation theory for
a bound state is far more complicated than perturbation theory for, say, the
electrons g-factor. In the latter case a diagram with three photons contributes
only in order
3
. In positronium a kernel involving the exchange of three photons
(e.g., Fig. 4) can also contribute to order
3
, but the same kernel will contribute
to all higher orders as well:
(g)
K
3
=
3
m
_
a
0
+a
1
+a
2

2
+. . .
_
. (28)
So in the bound state calculation there is no simple correlation between the
importance of an amplitude and the number of photons in it. Such behavior is
at the root of the complexities in high-precision analyses of positronium or other
QED bound states, and it is a direct consequence of the multiple scales in the
problem. Any expectation value like that in Eq. (28) will be some complicated
function of ratios of the three scales in the atom:
K
3
=
3
m f(p/m, KE/m) . (29)
(g)
The situation is actually even worse. The contribution from a particular kernel is highly
dependent upon gauge. For example, the coecients a
0
and a
1
vanish in Coulomb gauge
but not in Feynman gauge. The Feynman gauge result is roughly 10
4
times larger, and it
is spurious: the large contribution comes from unphysical retardation eects in the Coulomb
interaction that cancel when an innite number of other diagrams is included. The Coulomb
interaction is instantaneous in Coulomb gauge and so this gauge generally does a better
job describing the elds created by slowly moving charged particles. On the other hand
contributions coming from relativistic momenta are more naturally handled in a covariant
gauge like Feynman gauge; in particular renormalization is far simpler in Feynman gauge
than it is in Coulomb gauge. Unfortunately most Bethe-Salpeter kernels have contributions
coming from both nonrelativistic and relativistic momenta, and so there is no optimal choice
of gauge. This is again a problem due to the multiple scales in the system.
18
Since p/m and KE/m
2
, a Taylor expansion of f in powers of these
ratios generates an innite series of contributions just as in Eq. (28). Similar
series do not occur in the g-factor calculation because there is but one scale in
that problem, the mass of the electron.
Traditional methods for analyzing either of these systems fail to take advan-
tage of the nonrelativistic character of the systems. One way to capitalize on
this feature is to introduce a cuto of order the constituents mass or smaller
into the eld theory. The cuto here can be thought of as the boundary be-
tween relativistic and nonrelativistic physics. Since the gross dynamics in the
problems of interest is nonrelativistic, relativistic physics is well simulated by
local interactions in the cut-o Lagrangian.
The utility of such a cuto is greatly enhanced if one also transforms the
Dirac eld so as to decouple its upper components from its lower components.
This is called the Foldy-Wouthuysen transformation, and it transforms the Dirac
Lagrangian into a nonrelativistic Lagrangian. In QED one obtains
(iD m)

iD
0
+

D
2
2m

D
4
8m
3

e
2m

B
e
8m
2

E + (30)
where D

+ieA

is the gauge-covariant derivative,

E and

B are the elec-
tric and magnetic elds, and is a two-component Pauli spinor representing
the electron part, or upper components, of the original Dirac eld. The lower
components of the Dirac eld lead to analogous terms that specify the elec-
tromagnetic interactions of positrons. The Foldy-Wouthuysen transformation
generates an innite expansion of the action in powers of 1/m. As an ordinary
eld theory this expansion is a disaster; the renormalizability of the
theory is completely disguised, requiring a delicate conspiracy involving terms
of all orders in 1/m. However, setting m implies that the Foldy-Wouthuysen
expansion is an expansion in 1/, and from our general rules for cut-o theories
we know that only a nite number of terms need be retained in the expansion
if we want to work to some nite order in p/ p/m v. Thus, to study
positronium through order v
2

2
, we can replace QED by a nonrelativistic
QED (NRQED) with the Lagrangian
L
NRQED
=
(F

)
2
2
+

_
i
t
e
0
A
0
+

D
2
2m
0
c
1
e
0
2m
0


B c
2
e
0
8m
2
0

E
c
3
_
ie
0
8m
2
0


E
ie
0
4m
2
0


E
__

+
d
1
m
2
0
(

)
2
+
d
2
m
2
0
(

)
2
+positron and positron-electron terms. (31)
19
The coupling constants e
0
, m
0
, c
1
. . . are all specied for a cuto of m.
Renormalization theory tells us that there exists a choice for the coupling con-
stants in this theory such that NRQED reproduces all of the results of QED up
to corrections of order (p/m)
3
.
NRQED is far simpler to use than the original theory when studying non-
relativistic atoms like positronium. The analysis falls into two parts. First
one must determine the coupling constants in the NRQED Lagrangian. This
is easily done by computing simple scattering amplitudes in both QED and
NRQED, and then by adjusting the NRQED coupling constants so that the
two theories agree through some order in and v. The coupling constants are
functions of and the mass of the electron; to leading order one nds that the
d
i
s vanish while all the c
i
s equal one. As the couplings contain the relativis-
tic physics, this part of the calculation involves only scales of order m; it is
similar in character to a calculation of the g-factor. Furthermore there is no
need to deal with complicated bound states at this stage. Having solved the
high-energy part of QED by computing L
NRQED
, one goes on to solve NRQED
for any nonrelativistic process or system. To study positronium one uses the
Bethe-Salpeter equation for this theory, which is just the Schrodinger equation,
and ordinary time-independent perturbation theory. One of the main virtues
of this approach is that it builds directly on the simple results of nonrelativis-
tic quantum mechanics, leaving our intuition intact. Even more important for
high-precision calculations is that only two dynamical scales remain in the prob-
lem, the momentum and the kinetic energy, and these are easily separated on
a diagram-by-diagram basis. As a result innite series in can be avoided in
calculating the contributions due to individual diagrams, and thus it is trivial
to separate, say, (
6
) contributions from (
5
) contributions.
Obviously NRQED is useful in analyzing electromagnetic interactions in any
nonrelativistic system. In particular it provides an elegant framework for incor-
porating relativistic eects into analyses of many-electron atoms and solids in
general.
For heavy-quark mesons one can replace QCD by a nonrelativistic theory
(NRQCD) whose Lagrangian has basically the same structure as L
NRQED
.
Since the coupling constants relate to physics at scales of order m and above,
they can be computed perturbatively if the quark mass is large enough. The
lattice used in simulating NRQCD can be much coarser than that required for
ordinary QCD since structure at wavelengths of order 1/m has been removed
from the theory. For example, v 1/3 for the and thus the lattice spacing
can be roughly three times larger in NRQCD, leading to a lattice with 3
4
fewer
sites. Furthermore, by decoupling the quark from the antiquark degrees of free-
dom we convert the numerical problem of computing quark propagators from
a boundary-value problem for a four-spinor into an initial-value problem for a
two-spinor, resulting in very signicant gains in eciency. The move from QCD
to NRQCD makes the one of the easiest mesons to simulate numerically.
20
4 Conclusion
In this lecture we have seen that renormalizability is not miraculous; on the
contrary, approximate renormalizability is quite natural in a theory that is a
low-energy approximation to some more complex high-energy supertheory. Fur-
thermore one expects symmetries like gauge invariance or chiral symmetry in the
low-energy theory; low-mass particles are unnatural without such symmetries.
In such a picture the ultraviolet cuto, originally an artice required to give
meaning to divergent integrals, acquires physical signicance as the threshold
energy for the appearance of new physics associated with the supertheory; it is
the boundary between what we do and do not know. With the cuto in place it
is quite natural to have small nonrenormalizable interactions in the low-energy
theory, the size of these interactions being intimately related to the energy range
over which the low-energy approximation is valid: the smaller the coupling con-
stants for nonrenormalizable interactions, the larger the range of validity for
the theory. Thus in studying weak, electromagnetic and strong processes we
should be on the lookout for evidence indicating such interactions since this will
give us clues as to where new physics might be found. Finally, we saw that
renormalization and cut-o Lagrangians are powerful tools that can be used to
separate energy scales in a problem, allowing us to deal with one scale at a time.
In light of these developments we need no longer apologize for renormalization.
Acknowledgements
I thank Kent Hornbostel for his comments and suggestions concerning this
manuscript. These were most useful. This work was supported by a grant
from the National Science Foundation.
Bibliography
Cut-o eld theories have played an important role in the modern development
of the renormalization group as a tool for studying both quantum and statistical
eld theories. An elementary discussion of this subject, with lots of references
to the original literature, can be found in Wilsons Nobel lecture:
K.G. Wilson, Rev. Mod. Phys. 55, 583(1983); see also K.G. Wilson, Sci.
Am. 241, 140 (August, 1979).
The eects of a nite cuto are naturally important in numerical simulations
of lattice QCD, where it is hard enough to even get the lattice spacing small, let
alone vanishingly small. The use of nonrenormalizable interactions to correct
for a nite lattice spacing has been explored extensively; see, for example,
K. Symanzik, Nucl. Phys. B226, 187(1983).
The use of Monte Carlo techniques in renormalization-group analyses was dis-
cussed by R. Gupta at this Summer School.
Another area in which cut-o Lagrangians have come to the fore is in ap-
plications of current algebra to low-energy strong interactions. Nonlinear chiral
21
models provide the natural starting point for any attempt to model such physics
in terms of elementary meson and baryon elds. A good introduction to the
general ideas underlying these theories is given in Georgis book:
H. Georgi, Weak Interactions and Modern Particle Theory, Benjamin/Cummings,
Menlo Park, 1984.
Note that Georgi refers to cut-o theories as eective theories. I have avoided
this usage to prevent confusion between cut-o Lagrangians and eective clas-
sical Lagrangians, in which all loops are incorporated into the action. Serious
attempts have been made to carry chiral theories beyond tree level using per-
turbation theory; see, for example,
J. Gasser and H. Leutwyler, Ann. Phys. (N.Y.) 158, 142(1984).
Unfortunately, the convergence of perturbation theory is not always particularly
impressive here. Pion-nucleon theories are analyzed extensively in the literature
of nuclear physics. However most of these analyses are inconsistent in the ap-
plication of cutos and the design of the cut-o Lagrangian.
Experimental limits on possible deviations from the standard model have
been studied extensively in preparation for the SSC and other multi-TeV accel-
erators. For example, a study of the limits on nonrenormalizable interactions in
QED, QCD, etc. can be found in the proceedings of the rst Snowmass workshop
on SSC physics:
M. Abolins et al., in Proceedings of the 1982 DPF Summer Study on
Elementary Particle Physics and Future Facilities, edited by R. Donaldson
et al..
The signicance of an extra anomaly in the electrons magnetic moment, beyond
what is predicted by QED, was rst discussed in
R. Barbieri, L. Maiani and R. Petronzio, Phys. Lett. 96B, 63(1980); and
S.J. Brodsky and S.D. Drell, Phys. Rev. D22, 2236(1980).
The proceedings of the Snowmass workshops and similar workshops contain
extensive discussions of the various scenarios for the weak interactions at TeV
energies. See also
M.S. Chanowitz, Ann. Rev. Nucl. Part. Sci. 38, 323(1988)
where among other things, the phenomenology of the (nonrenormalizable) mas-
sive Yang-Mills version of weak interactions is discussed. The naturalness of
eld theories became an issue with the advent of early theories of the grand
unication of strong, electromagnetic and weak interactions, these GUTs be-
ing renormalizable but decidedly unnatural. For a brief introduction to such
theories see Quiggs book:
C. Quigg, Gauge Theories of Strong, Weak, and Electromagnetic Interac-
tions, Benjamin/Cummings, Reading, 1983.
Nonrelativistic QED is practically as old as quantum mechanics, but the
idea of using this theory as a cut-o eld theory, thereby permitting rigorous
calculations beyond tree level, originates with the (too) short paper
W.E. Caswell and G.P. Lepage, Phys. Lett. 167B, 437(1986).
Here the theory is applied in state-of-the-art bound-state calculations for positro-
nium and muonium. The application of these ideas to heavy-quark physics is
discussed in
22
G.P. Lepage and B.A. Thacker, in Field Theory on the Lattice, edited
by A. Billoire et al., Nucl. Phys. (Proc. Suppl.) 4 (1988); B.A. Thacker,
Ph.D. Thesis, Cornell University (September 1989); B.A. Thacker and
G.P. Lepage, Cornell preprint (December, 1989).
23