Advanced Quantum Mechanics

Peter S. Riseborough
February 19, 2015

Contents
1 Introduction 5

2 Quantum Mechanics of a Single Photon 6

2.1 Rotations and Intrinsic Spin . . . . . . . . . . . . . . . . . . . . . 7
2.2 Massless Particles with Spin Zero . . . . . . . . . . . . . . . . . . 11
2.3 Massless Particles with Spin One . . . . . . . . . . . . . . . . . . 12

3 Maxwell’s Equations 14
3.1 Vector and Scalar Potentials . . . . . . . . . . . . . . . . . . . . . 15
3.2 Gauge Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

4 Relativistic Formulation of Electrodynamics 20

4.1 Lorentz Scalars and Vectors . . . . . . . . . . . . . . . . . . . . . 20
4.2 Covariant and Contravariant Derivatives . . . . . . . . . . . . . . 22
4.3 Lorentz Transformations . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Invariant Form of Maxwell’s Equations . . . . . . . . . . . . . . . 27

5 The Simplest Classical Field Theory 30

5.1 The Continuum Limit . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Normal Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.3 Rules of Canonical Quantization . . . . . . . . . . . . . . . . . . 39
5.4 The Algebra of Boson Operators . . . . . . . . . . . . . . . . . . 41
5.5 The Classical Limit . . . . . . . . . . . . . . . . . . . . . . . . . . 44

6 Classical Field Theory 44

6.1 The Hamiltonian Formulation . . . . . . . . . . . . . . . . . . . . 46
6.2 Symmetry and Conservation Laws . . . . . . . . . . . . . . . . . 47
6.2.1 Conservation Laws . . . . . . . . . . . . . . . . . . . . . . 47
6.2.2 Noether Charges . . . . . . . . . . . . . . . . . . . . . . . 49
6.2.3 Noether’s Theorem . . . . . . . . . . . . . . . . . . . . . . 50
6.3 The Energy-Momentum Tensor . . . . . . . . . . . . . . . . . . . 51

1
7 The Electromagnetic Lagrangian 54
7.1 Conservation Laws for Electromagnetic Fields . . . . . . . . . . . 58
7.2 Massive Spin-One Particles . . . . . . . . . . . . . . . . . . . . . 64
7.3 Polarizations of Massive Spin-One Particles . . . . . . . . . . . . 65
7.4 The Propagator for Massive Photons. . . . . . . . . . . . . . . . 68

8 Symmetry Breaking and Mass Generation 70

8.1 Symmetry Breaking and Goldstone Bosons . . . . . . . . . . . . 70
8.2 The Kibble-Higgs Mechanism . . . . . . . . . . . . . . . . . . . . 72

9 Gravitational Interactions 74
9.1 Mathematical Structure of General Relativity . . . . . . . . . . . 78
9.2 Linearized Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.3 The Massive S = 2 Graviton . . . . . . . . . . . . . . . . . . . . 81
9.4 The Sourced Equations . . . . . . . . . . . . . . . . . . . . . . . 84
9.5 The Modes and Energies of Massive Gravitons . . . . . . . . . . 86

10 Quantization of the Electromagnetic Field 89

10.1 The Lagrangian and Hamiltonian Density . . . . . . . . . . . . . 91
10.2 Quantizing the Normal Modes . . . . . . . . . . . . . . . . . . . . 94
10.2.1 The Energy of the Field . . . . . . . . . . . . . . . . . . . 96
10.2.2 The Electromagnetic Field . . . . . . . . . . . . . . . . . 97
10.2.3 The Momentum of the Field . . . . . . . . . . . . . . . . 98
10.2.4 The Angular Momentum of the Field . . . . . . . . . . . . 100
10.3 Uncertainty Relations . . . . . . . . . . . . . . . . . . . . . . . . 106
10.4 Coherent States . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
10.4.1 The Phase-Number Uncertainty Relation . . . . . . . . . 113
10.4.2 Argand Representation of Coherent States . . . . . . . . . 114

11 Non-Relativistic Quantum Electrodynamics 115

11.1 Emission and Absorption of Photons . . . . . . . . . . . . . . . . 117
11.1.1 The Emission of Radiation . . . . . . . . . . . . . . . . . 118
11.1.2 The Dipole Approximation . . . . . . . . . . . . . . . . . 120
11.1.3 Electric Dipole Radiation Selection Rules . . . . . . . . . 124
11.1.4 Angular Distribution of Dipole Radiation . . . . . . . . . 132
11.1.5 The Decay Rate from Dipole Transitions. . . . . . . . . . 138
11.1.6 The 2p → 1s Electric Dipole Transition Rate. . . . . . . . 141
11.1.7 Electric Quadrupole and Magnetic Dipole Transitions. . . 144
11.1.8 The 3d → 1s Electric Quadrupole Transition Rate . . . . 148
11.1.9 Two-photon decay of the 2s state of Hydrogen. . . . . . . 151
11.1.10 The Absorption of Radiation . . . . . . . . . . . . . . . . 158
11.1.11 The Photoelectric Effect . . . . . . . . . . . . . . . . . . . 163
11.1.12 Impossibility of absorption of photons by free-electrons. . 166
11.2 Scattering of Light . . . . . . . . . . . . . . . . . . . . . . . . . . 168
11.2.1 Rayleigh Scattering . . . . . . . . . . . . . . . . . . . . . 172
11.2.2 Thomson Scattering . . . . . . . . . . . . . . . . . . . . . 175

2
 11.2.3 Raman Scattering . . . . . . . . . . . . . . . . . . . . . . 179
11.2.4 Radiation Damping and Resonance Fluorescence . . . . . 180
11.2.5 Natural Line-Widths . . . . . . . . . . . . . . . . . . . . . 183
11.3 Renormalization and Regularization . . . . . . . . . . . . . . . . 185
11.3.1 The Casimir Effect . . . . . . . . . . . . . . . . . . . . . . 186
11.3.2 The Lamb Shift . . . . . . . . . . . . . . . . . . . . . . . . 198
11.3.3 The Self-Energy of a Free Electron . . . . . . . . . . . . . 201
11.3.4 The Self-Energy of a Bound Electron . . . . . . . . . . . . 205
11.3.5 Brehmstrahlung . . . . . . . . . . . . . . . . . . . . . . . 210

12 The Dirac Equation 217

12.1 Conservation of Probability . . . . . . . . . . . . . . . . . . . . . 221
12.2 Covariant Form of the Dirac Equation . . . . . . . . . . . . . . . 223
12.3 The Field Free Solution . . . . . . . . . . . . . . . . . . . . . . . 225
12.4 Coupling to Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 231
12.4.1 Mott Scattering . . . . . . . . . . . . . . . . . . . . . . . . 232
12.4.2 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . 235
12.4.3 The Gordon Decomposition . . . . . . . . . . . . . . . . . 237
12.5 Lorentz Covariance of the Dirac Equation . . . . . . . . . . . . . 241
12.5.1 The Space of the Anti-commuting γ µ -Matrices. . . . . . . 252
12.5.2 Polarization in Mott Scattering . . . . . . . . . . . . . . . 259
12.6 The Non-Relativistic Limit . . . . . . . . . . . . . . . . . . . . . 262
12.7 Conservation of Angular Momentum . . . . . . . . . . . . . . . . 266
12.8 Conservation of Parity . . . . . . . . . . . . . . . . . . . . . . . . 268
12.9 Bi-linear Covariants . . . . . . . . . . . . . . . . . . . . . . . . . 272
12.10The Spherically Symmetric Dirac Equation . . . . . . . . . . . . 276
12.10.1 The Hydrogen Atom . . . . . . . . . . . . . . . . . . . . . 287
12.10.2 Lowest-Order Radial Wavefunctions . . . . . . . . . . . . 300
12.10.3 The Relativistic Corrections for Hydrogen . . . . . . . . . 302
12.10.4 The Kinematic Correction . . . . . . . . . . . . . . . . . . 307
12.10.5 Spin-Orbit Coupling . . . . . . . . . . . . . . . . . . . . . 308
12.10.6 The Darwin Term . . . . . . . . . . . . . . . . . . . . . . 314
12.10.7 The Fine Structure of Hydrogen . . . . . . . . . . . . . . 315
12.10.8 A Particle in a Spherical Square Well . . . . . . . . . . . 319
12.10.9 The MIT Bag Model . . . . . . . . . . . . . . . . . . . . . 325
12.10.10The Temple Meson Model . . . . . . . . . . . . . . . . . . 329
12.11Scattering by a Spherically Symmetric Potential . . . . . . . . . 331
12.11.1 Polarization in Coulomb Scattering. . . . . . . . . . . . . 331
12.11.2 Partial Wave Analysis . . . . . . . . . . . . . . . . . . . . 335
12.12An Electron in a Uniform Magnetic Field . . . . . . . . . . . . . 338
12.13Motion of an Electron in a Classical Electromagnetic Field . . . . 341
12.14The Limit of Zero Mass . . . . . . . . . . . . . . . . . . . . . . . 345
12.15Classical Dirac Field Theory . . . . . . . . . . . . . . . . . . . . . 353
12.15.1 Chiral Gauge Symmetry . . . . . . . . . . . . . . . . . . . 356
12.16Hole Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
12.16.1 Compton Scattering . . . . . . . . . . . . . . . . . . . . . 364

3
 12.16.2 Charge Conjugation . . . . . . . . . . . . . . . . . . . . . 369

13 Local and Global Gauge Symmetries 373

14 The Many-Particle Dirac Field 375

14.1 The Algebra of Fermion Operators . . . . . . . . . . . . . . . . . 375
14.2 Quantizing the Dirac Field . . . . . . . . . . . . . . . . . . . . . 377
14.3 Parity, Charge and Time Reversal Invariance . . . . . . . . . . . 382
14.3.1 Parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
14.3.2 Charge Conjugation . . . . . . . . . . . . . . . . . . . . . 386
14.3.3 Time Reversal . . . . . . . . . . . . . . . . . . . . . . . . 387
14.3.4 The CPT Theorem . . . . . . . . . . . . . . . . . . . . . . 390
14.4 The Connection between Spin and Statics . . . . . . . . . . . . . 391

15 Massive Gauge Field Theory 392

15.1 The Gauge Symmetry . . . . . . . . . . . . . . . . . . . . . . . . 393
15.2 The Coupling to the Gauge Field . . . . . . . . . . . . . . . . . . 394
15.3 The Free Gauge Fields . . . . . . . . . . . . . . . . . . . . . . . . 395
15.4 Breaking the Symmetry . . . . . . . . . . . . . . . . . . . . . . . 398

4
1 Introduction
Non-relativistic mechanics yields a reasonable approximate description of phys-
ical phenomena in the range where the particles’ kinetic energies are small com-
pared with their rest mass energies. However, it should be noted that when the
relativistic invariant mass of a particle is expressed in terms of its energy E and
momentum p via
E 2 − p2 c2 = m2 c4 (1)
it implies that its dispersion relation has two branches
q
E = ±c p2 + m2 c2 (2)

In relativistic classical mechanics, it is assumed expedient to neglect the negative-

energy solutions. This assumption based on the expectation that, since energy
E(p)

+mc2

2mc2
pc

-mc2

Figure 1: The positive and negative energy branches for a relativistic particle
with rest mass m. The minimum separation between the positive-energy branch
and the negative-energy branch is 2mc2 .

can only change continuously, it is impossible that a particle with positive energy
can make a transition from the positive to negative-energy states. However, in
quantum mechanics, particles can make discontinuous transitions. Therefore, it
is necessary to consider both the positive and negative-energy branches. These
considerations naturally lead one to the concept of particles and anti-particles,
and also to the realization that one must consider multi-particle quantum me-
chanics or field theory.

We shall take a look at the quantum mechanical description of the elec-

tromagnetic field, Dirac’s relativistic theory of spin one-half fermions (such as
leptons and quarks), and then look at the interaction between these particles
and the electromagnetic field. The interaction between charged fermions and
the electromagnetic field is known as Quantum Electrodynamics. Quantum
Electrodynamics contains some surprises, namely that although the interaction

5
appears to be governed by a small coupling strength
 2
e 1
∼ (3)
h̄ c 137.0359979
perturbation theory does not converge. In fact, straightforward perturbation
theory is plagued by infinities. However, physics is a discipline which is aimed
at uncovering the relationships between measured quantities. The quantities
e and m which occur in quantum electrodynamics are theoretical constructs
which, respectively, describe the bare charge of the electron and bare mass of
the electron. This means one is assuming that e and m would be the results
of measurements on a (fictional) electron which does not interact electromag-
netically. That is, e and m are not physically measurable and their values are
therefore unknown. What can be measured experimentally are the renormal-
ized mass and the renormalized charge of the electron. The divergences found in
quantum electrodynamics can be shown to cancel or drop out, when one relates
different physically measurable quantities, as only the renormalized masses and
energies enter the theory. Despite the existence of infinities, quantum electrody-
namics is an extremely accurate theory. Experimentally determined quantities
can be predicted to an extremely high degree of precision.

The framework of quantum electrodynamics can be extended to describe

the unification of electrodynamics and the weak interaction via electro-weak
theory, which is also well tested. The scalar (U (1)) gauge symmetry of the elec-
tromagnetic field is replaced by a matrix (SU (2)) symmetry of the combined
electro-weak theory, in which the gauge field couples to the two components of
the spinor wave functions of the fermions. The generalization of the gauge field
necessitates the inclusion of additional components. Through symmetry break-
ing, some components of the field which mediates the electro-weak interaction
become massive, i.e. have finite masses. The finite masses are responsible for
the short range of the weak interaction. More tentatively, the gauge theory
framework of quantum electrodynamics has also been extended to describe the
interactions between quarks which is mediated by the gluon field. The gauge
symmetry of the interaction is enlarged to an SU (3) symmetry. However, un-
like quantum electrodynamics where photons are uncharged and do not interact
with themselves, the gluons do interact amongst themselves.

2 Quantum Mechanics of a Single Photon

Maxwell’s equations were formulated to describe classical electromagnetism. In
the quantum description, the classical electromagnetic field is described as being
composed of a very large number of photons. Before one describes multi-photon
quantum mechanics of the electromagnetic field, one should ascertain the form
of the Schrödinger equation for a single photon. The photon is a massless, un-
charged particle of spin-one.

6
 A spin-one particle is described by a vector wave function. This can be
heuristically motivated as follows:

A spin-zero particle has just one state and is uniquely described by a one-
component field ψ.

A spin one-half particle has two independent states corresponding to the two
allowed values of the z-component of the intrinsic angular momentum S z =
± h̄2 . The wave function ψ of a spin one-half particle is a spinor which has two
independent components
 (1) 
ψ (r, t)
ψ(r, t) = (4)
ψ (2) (r, t)
These two components can be used to represent two independent basis states.

We conjecture that since a particle with intrinsic spin S has (2S + 1) inde-
pendent basis states, then the wave function should have (2S + 1) independent
components.

A (non-relativistic) spin-one particle should have three independent states

corresponding to the three possible values of the z-component of the intrinsic
angular momentum. From the conjecture, one expects that the wave function
ψ of a spin-one particle should have three components.
 (1) 
ψ (r, t)
ψ(r, t) =  ψ (3) (r, t)  (5)
ψ (3) (r, t)
This conjecture can be verified by examining the transformational properties of
a vector field under rotations. Under a rotation of the field, the components
of the field are transported in space, and also the direction of the vector field
is rotated. This implies that the components of the transported field have to
be rotated. The rotation of the direction of the field is generated by operators
which turn out to be the intrinsic angular momentum operators. Specifically,
the generators satisfy the commutation relations defining angular momentum,
but also correspond to the subspace with angular momentum one.

2.1 Rotations and Intrinsic Spin

Under the transformation which takes r → r0 = R r, the magnitude of the
scalar field ψ at r is transferred to the point r0 . This defines the transformation
ψ → ψ 0 . The transformed scalar field ψ 0 is defined so that its value at r0 has
the same value as ψ(r). That is
ψ 0 (r0 ) = ψ(r) (6)

7
or equivalently

ψ 0 (R̂ r) = ψ(r) (7)

The above equation can be used to determine ψ 0 (r) by using the substitution
r → R̂−1 r so
ψ 0 (r) = ψ(R̂−1 r) (8)
If ê is a unit vector along the axis of rotation, the rotation of r through an

e
exr

ϕ
ex(exr)
r(e.r)
r

Figure 2: A rotation of the vector r through an angle ϕ about an axis ê.

infinitesimal angle δϕ is expressed as

R̂ r = r + δϕ ê ∧ r + . . . (9)

where terms of order δϕ2 have been neglected. Hence, under an infinitesimal

ψ ψ(R-1r)

ψ(r) y-axis

r'=R-1r

x-axis

Figure 3: The effect of a rotation R̂ on a scalar field ψ(r).

rotation, the transformation of a scalar wave function can be found from the
Taylor expansion

ψ 0 (r) = ψ(r − δϕ ê ∧ r)
= ψ(r) − δϕ ( ê ∧ r ) . ∇ ψ(r) + . . .
= ψ(r) − δϕ ( r ∧ ∇ ) . ê ψ(r) + . . .
i δϕ
= ψ(r) − ( ê . L̂ ) ψ(r) + . . .
h̄

8
  
i δϕ
= exp − ( ê . L̂ ) ψ(r) (10)
h̄

where the operator L̂ has been defined as

L̂ = − i h̄ r ∧ ∇ (11)

Therefore, locally, rotations of the scalar field are generated by the orbital an-
gular momentum operator L̂.

Since the operation R̂ is a rotation, it also rotates a vector field ψ(r). Not
only does the rotation transfer the magnitude of ψ(r) to the new point r0 but it
must also rotate the direction of the transferred vector ψ(r). The rotated vector
ψ 0 (r0 ) is denoted by R̂ ψ(r). Mathematically, the transformation is expressed
0.5
0
-0.5
-1
-1
-0.5
-0.5
0
0
0.5
0.5
Figure 4: The effect of a rotation R̂ on a vector field ψ(r). The rotation affects
both the magnitude and direction of the vector. 1
1 -2
as 0 2 1 0 -1 -2
2 ψ 0 (r0 ) = R̂ ψ(r) (12)
or equivalently

ψ 0 (R̂ r) = R̂ ψ(r) (13)

The above equation can be used to determine ψ 0 (r) as

ψ 0 (r) = R̂ ψ(R̂−1 r) (14)

The part of the rotational operator designated by R̂ does not affect the posi-
tional coordinates (r) of the vector field, and so can be found by considering
the rotation of the vector field ψ at the origin
 
R̂ ψ = Iˆ + δϕ ê ∧ ψ (15)

9
That is, the operator R̂ only produces a mixing of the components of ψ. Hence,
the complete rotational transformation of the vector field can be represented as

ψ 0 (r) = R̂ ψ(r − δϕ ê ∧ r)
= ψ(r − δϕ ê ∧ r) + δϕ ê ∧ ψ(r − δϕ ê ∧ r)
= ψ(r − δϕ ê ∧ r) + δϕ ê ∧ ψ(r) + . . .
i δϕ
= ψ(r) − ( ê . L̂ ) ψ(r) + δϕ ê ∧ ψ(r) + . . .
h̄
i δϕ i δϕ
= ψ(r) − ( ê . L̂ ) ψ(r) − ( ê . Ŝ ) ψ(r) (16)
h̄ h̄
where the terms of order (δϕ)2 have been neglected and a vector operator Ŝ has
been introduced. The operator Ŝ only admixes the components of ψ µ , unlike
L̂ which only acts on the r dependence of the components. The components of
the three-dimensional vector operator S are expressed as 3 × 3 matrices1 , with
matrix elements
(Ŝ (i) )j,k = − i h̄ ξ i,j,k (18)
where ξ i,j,k is the antisymmetric Levi-Civita symbol. The Levi-Civita symbol
is defined by ξ i,j,k = 1 if the ordered set (i, j, k) is obtained by an even number
of permutations of (1, 2, 3) and is −1 if it is obtained by an odd number of
permutations, and is zero if two or more indices are repeated. Specifically, the
antisymmetric matrices are given by
 
0 0 0
Ŝ (1) = h̄  0 0 −i  (19)
0 i 0

and by  
0 0 i
Ŝ (2) = h̄  0 0 0  (20)
−i 0 0
and finally by  
0 −i 0
Ŝ (3) = h̄  i 0 0  (21)
0 0 0
By using a unitary transform, the above three operators can be transformed
into the standard representation of spin-one operators where S (3) is chosen to
be diagonal. It is easily shown that the components of the matrix operators L̂
and Ŝ satisfy the same type of commutation relations

[ L̂(i) , L̂(j) ] = i h̄ ξ i,j,k L̂(k) (22)

1 The component of the matrix denoted by
(Ŝ)j,k (17)

denotes the element of Ŝ in the j-th row and k-th column.

10
and
[ Ŝ (i) , Ŝ (j) ] = i h̄ ξ i,j,k Ŝ (k) (23)
where the repeated index (k) is summed over. The above set of operators
form a Lie algebra associated with the corresponding Lie group of continuous
rotations. Thus, it is natural to identify these operators which arise in the
analysis of transformations in classical physics with the angular momentum
operators of quantum mechanics. In terms of these operators, the infinitesimal
transformation has the form
i δϕ
ψ 0 (r) ≈ ψ(r) − ê . ( L̂ + Ŝ ) ψ(r) + . . . (24)
h̄
or  
i δϕ
ψ 0 (r) = exp − ê . ( L̂ + Ŝ ) ψ(r) (25)
h̄
Thus, the transformation is locally accomplished by
 
i δϕ
ψ 0 (r) = exp − ( ê . Ĵ ) ψ(r) (26)
h̄
where
Ĵ = L̂ + Ŝ (27)
is the total angular momentum. The operator Ŝ is the intrinsic angular mo-
mentum of the vector field ψ. The magnitude of S is found from

Sˆ2 = (Ŝ (1) )2 + (Ŝ (2) )2 + (Ŝ (3) )2 (28)

which is evaluated as
 
1 0 0
Sˆ2 = 2 h̄2  0 1 0  (29)
0 0 1
and is the Casimir operator. It is seen that a vector field has intrinsic angular
momentum, with a magnitude given by the eigenvalue of Sˆ2 which is

S ( S + 1 ) h̄2 = 2 h̄2 (30)

hence S = 1. Thus, it is seen that a vector field is associated with an intrinsic

angular momentum of spin one.

2.2 Massless Particles with Spin Zero

First, we shall try and construct the Schrödinger equation describing a massless
uncharged spinless particle. A spinless particle is described by a scalar wave
function, and an uncharged particle is described by a real wave function. The
derivation is based on the energy-momentum relation for a massless particle

E 2 − p2 c2 = 0 (31)

11
which is quantized by using the substitutions
∂
E → i h̄
∂t
p → p̂ = − i h̄ ∇ (32)

One finds that the real scalar wave function ψ(r, t) satisfies the wave equation

1 ∂2
 
2
− ∇ ψ = 0 (33)
c2 ∂t2

since h̄ drops out. This is not a very useful result, since it is a second-order differ-
ential equation in time, and the solution of a second-order differential equation
can only be determined if two initial conditions are given. Usually, the initial
conditions are given by

ψ(r, 0) = f (r)

∂ψ 
= g(r) (34)
∂t t=0

In quantum mechanics, measurements disturb the state of the system and

so it becomes impossible to design two independent measurements which can
uniquely specify two initial conditions for one state. Hence, one has reached
an impasse. Due to this difficulty and since there are no known examples of
massless spinless particles found in nature, this theory is not very useful.

2.3 Massless Particles with Spin One

The wave function of an uncharged spin-one particle is expected to be repre-
sented by a real vector function.

We shall try and factorize the wave equation for the vector E into two first-
order differential equations, each of which requires one boundary condition.
This requires one to specify six quantities. Therefore, one needs to postulate
the existence of two independently measurable fields, E and B. Each of these
fields should satisfy the two wave equations

1 ∂2
 
2
− ∇ E = 0 (35)
c2 ∂t2

and
1 ∂2
 
− ∇2 B = 0 (36)
c2 ∂t2

12
 The first-order equations must have the form
 
∂E
i h̄ = c a p̂ ∧ E + b p̂ ∧ B
∂t
 
∂B
i h̄ = c d p̂ ∧ B + e p̂ ∧ E (37)
∂t
since the left-hand side is a vector, the right-hand side must also be a vector
composed of the operator p̂ and the wave functions. Like Newton’s laws, these
equations must be invariant under time-reversal invariance, t → − t. The
transformation leads to the identification

a = d = 0 (38)

and

b = −e (39)

if one also requires that one of the two fields changes sign under time reversal2 .
We shall adopt the convention that the field E retains its sign, so

E → E
B → −B (40)

under time-reversal invariance. On taking the time derivative of the first equa-
tion, one obtains
2
 
2 ∂ E 2 2
− h̄ = − c b p̂ ∧ p̂ ∧ E (41)
∂t2
Likewise, the B field is found to satisfy
∂2B
 
2 2 2
− h̄ = − c b p̂ ∧ p̂ ∧ B (42)
∂t2
Thus, one has found the two equations
2
  
2 ∂ E 2 2 2
− h̄ = −c b − p̂ E + p̂ p̂ . E (43)
∂t2
and
∂2B
  
2 2 2 2
− h̄ = −c b − p̂ B + p̂ p̂ . B (44)
∂t2
On substituting the operator p̂ = − i h̄ ∇ , one obtains

∂2E
  
2 2 2
= − c b − ∇ E + ∇ ∇ . E (45)
∂t2
2 For the non-relativistic Schrödinger equation, time-reversal invariance implies that t →

t0 = −t and ψ → ψ 0 = ψ ∗ .

13
and
∂2B
  
2 2 2
= −c b − ∇ B + ∇ ∇.B (46)
∂t2

so h̄ drops out. To reduce these equations to the form of wave equations, one
needs to impose the conditions

∇.E = 0 (47)

and
∇.B = 0 (48)
On identifying the coefficients with those of the wave equation, one requires
that
b2 = 1 (49)
Thus, one has arrived at the set of the source-free Maxwell’s equations
1 ∂E
= ∇ ∧ B
c ∂t
1 ∂B
− = ∇ ∧ E
c ∂t
∇.E = 0
∇.B = 0 (50)

which describe the one-particle Schrödinger equation for a massless spin-one

particle, with the wave function (E, B). These have a form which appears to be
completely classical, since h̄ has dropped out. Furthermore, in the absence of
sources, Maxwell’s equations are invariant under the symmetry transformation
(E, B) → (−B, E).

3 Maxwell’s Equations
Classical Field Theories describe systems in which a very large number of par-
ticles are present. Measurements on systems containing very large numbers of
particles are expected to result in average values, with only very small devia-
tions. Hence, we expect that the subtleties of quantum measurements should be
completely absent in systems that can be described as quantum fields. Classical
Electromagnetism is an example of such a quantum field, in which an infinitely
large number of photons are present.

In the presence of a current density j and a charge density ρ, Maxwell’s

equations assume the forms
1 ∂E 4π
∇ ∧ B − = j
c ∂t c

14
 1 ∂B
∇ ∧ E + = 0
c ∂t
∇.E = 4πρ
∇.B = 0 (51)

The field equations ensure that the sources j and ρ satisfy a continuity equation.

Taking the divergence of the first equation and combining it with the time
derivative of the third, one obtains
   
1 ∂ 4π
∇. ∇ ∧ B − ∇.E = ∇.j
c ∂t c
 
1 ∂ 4π
− ∇.E = ∇.j
c ∂t c
4 π ∂ρ 4π
− = ∇.j (52)
c ∂t c
Hence, one has derived the continuity equation
∂ρ
+ ∇.j = 0 (53)
∂t
which shows that charge is conserved.

3.1 Vector and Scalar Potentials

Counting each component of Maxwell’s equations separately, one arrives at eight
equations for the six components of the unknown fields E and B. As the equa-
tions are linear, this would over-determine the fields. Two of the eight equations
must be regarded as self-consistency equations for the initial conditions on the
fields.

One can solve the two source-free Maxwell equations, by expressing the
electric E and magnetic fields B in terms of the vector A and scalar φ potentials,
via
1 ∂A
E = − − ∇φ (54)
c ∂t
and
B = ∇ ∧ A (55)
The expressions for B and E automatically satisfy the two source-free Maxwell’s
equations. This can be seen by examining
1 ∂B
∇ ∧ E + = 0 (56)
c ∂t

15
which, on substituting the expressions for the electromagnetic fields in terms of
the vector and scalar potentials, becomes
   
1 ∂A 1 ∂
∇ ∧ − − ∇φ + ∇ ∧ A = 0 (57)
c ∂t c ∂t

which is automatically satisfied since

 
∇ ∧ ∇φ = 0 (58)

and the terms involving A cancel since A is analytic. The remaining source-free
Maxwell equation is satisfied, since it has the form

∇.B = 0 (59)

which reduces to
 
∇. ∇ ∧ A = 0 (60)

which is identically zero.

Therefore, the six components of E and B have been replaced by the four
quantities A and φ. These four quantities are determined by the Maxwell equa-
tions which involve the sources, which are four in number.

The fields are governed by the set of non-trivial equations which relate A and
φ to the sources j and ρ. When expressed in terms of A and φ, the remaining
non-trivial Maxwell equations become
   
1 ∂ 1 ∂A 4π
∇ ∧ ∇ ∧ A + + ∇φ = j
c ∂t c ∂t c
 
1 ∂A
−∇. + ∇φ = 4πρ (61)
c ∂t

but since    
∇ ∧ ∇ ∧ A = ∇ ∇.A − ∇2 A (62)

the pair of equations can be written as

1 ∂2A
   
2 1 ∂φ 4π
− ∇ A + 2 +∇ ∇ . A + = j
c ∂t2 c ∂t c
 
2 1 ∂
−∇ φ − ∇.A = 4πρ (63)
c ∂t

We shall make use of gauge invariance to simplify these equations.

16
3.2 Gauge Invariance
The vector and scalar potentials are defined as the solutions of the coupled
partial differential equations describing the electric and magnetic fields
1 ∂A
E = − − ∇φ (64)
c ∂t
and
B = ∇ ∧ A (65)
Hence, one expects that the solutions are only determined up to functions of
integration. That is the vector and scalar potentials are not completely deter-
mined, even if the electric and magnetic fields are known precisely. It is possible
to transform the vector and scalar potentials, in a way such that the E and B
fields remain invariant. These transformations are known as gauge transforma-
tions of the second kind3 .

In particular, one can perform the transform

A → A0 = A − ∇ Λ
1 ∂Λ
φ → φ0 = φ + (66)
c ∂t
where Λ is an arbitrary analytic function and this transformation leaves the E
and B fields invariant. The magnetic field is seen to be invariant since

B0 = ∇ ∧ A0
 
= ∇ ∧ A − ∇Λ

= ∇ ∧ A
= B (67)

where the identity  

∇ ∧ ∇Λ = 0 (68)

valid for any scalar function Λ has been used. The electric field is invariant,
since the transformed electric field is given by

1 ∂A0
E0 = − − ∇ φ0
c ∂t
3 The transformation
 
0 iχ
ψ → ψ = ψ exp
h̄
p → p̂0 = − i h̄ ∇ − ∇ χ
used in quantum mechanics is known as a gauge transformation of the first kind.

17
    
1 ∂ 1 ∂Λ
= − A − ∇Λ − ∇ φ +
c ∂t c ∂t
1 ∂A
= − − ∇φ
c ∂t
= E (69)

In the above derivation, it has been noted that the order of the derivatives can
be interchanged,
∂Λ ∂
∇ = ∇Λ (70)
∂t ∂t
since Λ is an analytic scalar function.

The gauge invariance allows us the freedom to impose a gauge condition

which fixes the gauge. Two gauge conditions which are commonly used are the
Lorenz gauge
1 ∂φ
∇.A + = 0 (71)
c ∂t
and the Coulomb or radiation gauge

∇.A = 0 (72)

The Lorenz gauge is manifestly Lorentz invariant, whereas the Coulomb gauge
is frequently used in cases where the electrostatic interactions are important.

It is always possible to impose one or the other of these gauge conditions. If

the vector and scalar potentials (φ, A) do not satisfy the gauge transformation,
then one can perform a gauge transformation so that the transformed fields
(φ0 , A0 ) satisfy the gauge condition.

For example, if the fields (φ, A) do not satisfy the Lorenz gauge condition,
since
1 ∂φ
∇.A + = χ(r, t) (73)
c ∂t
where χ is non-zero, then one can perform the gauge transformation to the new
fields (φ0 , A0 )

1 ∂φ0 1 ∂φ 1 ∂2Λ
∇ . A0 + = ∇ . A − ∇2 Λ + + 2
c ∂t c ∂t c ∂t2
2
 
1 ∂
= χ − ∇2 − 2 Λ (74)
c ∂t2

The new fields satisfy the Lorentz condition if one chooses Λ to be the solution
of the wave equation

1 ∂2
 
2
∇ − 2 Λ = χ(r, t) (75)
c ∂t2

18
This can always be done, since the driven wave equation always has a solution.
Hence, one can always insist that the fields satisfy the gauge condition
1 ∂φ0
∇ . A0 + = 0 (76)
c ∂t
Alternatively, if one is to impose the Coulomb gauge condition

∇ . A0 = 0 (77)

one can use Poisson’s equations to show that one can always find a Λ such that
the Coulomb gauge condition is satisfied4 .

In the Lorenz gauge, the equations of motion for the electromagnetic field
are given by
1 ∂2
 
2 4π
− ∇ + 2 2
A = j
c ∂t c
1 ∂2
 
− ∇2 + 2 φ = 4πρ (78)
c ∂t2
Hence, A and φ both satisfy the wave equation, where j and ρ are the sources.
The solutions are waves which travel with velocity c.

In the Coulomb gauge, the fields satisfy the equations

1 ∂2
 
2 4π 1 ∂
− ∇ + 2 2
A = j − ∇φ
c ∂t c c ∂t
− ∇2 φ = 4 π ρ (79)

The second equation is Poisson’s equation and has solutions given by

ρ(r0 , t)
Z
φ(r, t) = d3 r 0 (80)
| r − r0 |
which is an “instantaneous” Coulomb interaction. However, the force from the
electric field E is not transmitted instantaneously from r0 to r, since there is a
term in the equation for A which compensates for the “instantaneous” interac-
tion described by φ.

Exercise:

Consider the case of a uniform magnetic field of magnitude B which is ori-

ented along the z-axis. Using the Coulomb gauge, find a general solution for
the vector potential.

4 Imposing a gauge condition is insufficient to uniquely determine the vector potential A,

since in the case of the Coulomb gauge, the vector potential is only known up to the gradient
of any harmonic function Λ.

19
4 Relativistic Formulation of Electrodynamics
Physical quantities can be classified as either being scalars, vectors or tensors
according to how they behave under transformations. Scalars are invariant un-
der Lorentz transformations, and all vectors transform in the same way.

4.1 Lorentz Scalars and Vectors

The space-time four-vector has components given by the time t and the three
space coordinates (x(1) , x(2) , x(3) ) which label an event. The zeroth-component
of the four-vector x(0) (the time component) is defined to be ct, where c is the
velocity of light, in order that all the components have the dimensions of length.
In Minkowski space, the four-vector is defined as having contravariant compo-
nents xµ = (ct, x(1) , x(2) , x(3) ), while the covariant components are denoted
by xµ = (ct, −x(1) , −x(2) , −x(3) ). The invariant length is given by the scalar
product

xµ xµ = ( c t )2 − ( x(1) )2 − ( x(2) )2 − ( x(3) )2 (81)

where repeated indices are summed over. The invariant length xµ xµ is related
to the proper time τ . This definition can be generalized to the scalar product
of two arbitrary four-vectors Aµ and B µ as

Aµ Bµ = A(0) B (0) − A(1) B (1) − A(2) B (2) − A(3) B (3) (82)

In special relativity, the four-vector scalar product can be written in terms of

the product of the time-index components and the scalar product of the usual
three-vectors as
Aµ Bµ = A(0) B (0) − A . B (83)
The Lorentz invariant four-vector scalar product can also be written as

Aµ Bµ = gµ,ν Aµ B ν (84)

where gµ,ν is the Minkowski metric. These equations imply that

Aµ = gµ,ν Aν (85)

That is, the metric tensor transforms contravariant components to covariant

components. The Minkowski metric can be expressed as a four by four matrix
 
1 0 0 0
 0 −1 0 0 
gµ,ν =   0 0 −1
 (86)
0 
0 0 0 −1

20
where µ labels the rows and ν labels the columns. If the four-vectors are ex-
pressed as column-vectors
 (0) 
A
A(1) 
Aν = 

 A(2) 
 (87)
A(3)
and  
A(0)
 A(1) 
Aν =  
 A(2)  (88)
A(3)
then the transformation from contravariant to covariant components can be
expressed as
     (0) 
A(0) 1 0 0 0 A
 A(1)   0 −1 0 0   A(1) 
 A(2)  =  0 0 −1 (89)
     
0   A(2) 
A(3) 0 0 0 −1 A(3)
The inverse transform is expressed as

Aµ = g µ,ν Aν (90)

where, most generally, g µ,ν is the inverse metric

g µ,ν = ( gµ,ν )−1 (91)

In our particular case of Cartesian coordinates in (flat) Minkowski space, the

inverse metric coincides with the metric.

Example: Covariant and Contravariant Components of a Euclidean

Vector

Consider a two-dimensional Euclidean space. However, instead of simply

considering Cartesian axes, consider two unit vectors ê1 and ê2 which are not
co-linear
ê1 . ê2 = cos θ (92)
A general vector X can be expressed in terms of its contravariant components
(X (1) , X (2) ) via
X = X (1) ê1 + X (2) ê2 (93)
The scalar product of two vectors X and Y can be evaluated by standard means
in terms of their contravariant components. The result can be expressed in terms
of a matrix product as
   (1) 

1 cos θ Y
X .Y = X (1) X (2) (94)
cos θ 1 Y (2)

21
involving the metric tensor gµ,ν
 
1 cos θ
(95)
cos θ 1
Likewise, the length |X| of the vector can be expressed as
   (1) 
2 (1) (2)

1 cos θ X
|X| = X X (96)
cos θ 1 X (2)
The covariant components of the vector (X(1) , X(2) ) are found from the con-
travariant components by the action of the metric tensor
     (1) 
X(1) 1 cos θ X
= (97)
X(2) cos θ 1 X (2)
The covariant components of any vector can be found geometrically by dropping
normals from the tip of the vector to the axes. The intersection of a normal with
its corresponding axis determines the covariant component. For a Cartesian co-
ordinate system, the covariant components are identical to the contravariant
components.

A familiar example of the Lorentz invariant scalar product involves the mo-
mentum four-vector with contravariant components pµ ≡ ( Ec , p(1) , p(2) , p(3) )
where E is the energy. The covariant components of the momentum four-vector
are given by pµ ≡ ( Ec , −p(1) , −p(2) , −p(3) ) and the scalar product defines the
invariant mass m via
 2
E
p µ pµ = − p2 = m2 c2 (98)
c
Another scalar product which is frequently encountered is pµ xµ which is given
by
pµ xµ = E t − p . x (99)
This scalar product is frequently seen in the description of planes of constant
phase of waves.

4.2 Covariant and Contravariant Derivatives

We shall now generalize the idea of the differential operator ∇ to Minkowski
space. The generalization we consider, will have to be modified when the metric
varies in space, i.e. when gµ,ν depends on the coordinates xµ of the points in
space.

Consider a scalar function φ(xµ ) defined in terms of the contravariant coor-

dinates xµ . Under an infinitesimal translation aµ
xµ → xµ0 = xµ + aµ (100)

22
the scalar function φ(xµ0 ) is still a scalar. Therefore, on performing a Taylor
expansion, one has
∂
φ(xµ + aµ ) = φ(xµ ) + aµ φ(xµ ) + . . . (101)
∂xµ
which is also a scalar. Therefore, the quantity
∂
aµ φ(xµ ) (102)
∂xµ
is a scalar and can be interpreted as a scalar product between the contravariant
vector displacement aµ and the covariant gradient
∂
φ(xµ ) (103)
∂xµ
The covariant gradient can be interpreted in terms of a covariant derivative
∂
∂µ = µ
∂x
 
1 ∂ ∂ ∂ ∂
= , , ,
c ∂t ∂x(1) ∂x(2) ∂x(3)
 
1 ∂
= , ∇ (104)
c ∂t

Likewise, one can introduce the contravariant derivative as

∂
∂µ =
∂xµ
 
1 ∂ ∂ ∂ ∂
= , − (1)
, − (2)
, −
c ∂t ∂x ∂x ∂x(3)
 
1 ∂
= , −∇ (105)
c ∂t

These covariant and contravariant derivative operators are useful in making

relativistic transformational properties explicit. For example, if one defines the
four-vector potential Aµ via
 
µ (1) (2) (3)
A = φ, A , A , A
 
= φ, A (106)

then the Lorenz gauge condition can be expressed as

1 ∂φ
∂µ Aµ = + ∇.A = 0 (107)
c ∂t

23
which is of the form of a Lorentz scalar. Likewise, if one introduces the current
density four-vector j µ with contravariant components
 
jµ = c ρ , j (1) , j (2) , j (3)
 
= cρ, j (108)

then the condition for conservation of charge can be written as

∂ρ
+ ∇.j = 0
∂t
∂µ j µ = 0 (109)

which is a Lorentz scalar. Also, the gauge transformation can also be compactly
expressed in terms of a transformation of the contravariant vector potential

Aµ → Aµ0 = Aµ + ∂ µ Λ (110)

where Λ is an arbitrary scalar function. The gauge transformation reduces to

1 ∂Λ
φ → φ0 = φ +
c ∂t
A → A0 = A − ∇Λ (111)

Similarly, one can use the contravariant notation to express the quantization
conditions
∂
E → i h̄
∂t
p → − i h̄ ∇ (112)

in the form
pµ → i h̄ ∂ µ (113)
One can also express the wave equation operator in terms of the scalar product
of the contravariant and covariant derivative operators

1 ∂2
∂ µ ∂µ = − ∇2 (114)
c2 ∂t2
Hence, in the Lorenz gauge, the equations of motion for the four-vector potential
Aµ can be expressed concisely as
4π µ
∂ ν ∂ν Aµ = j (115)
c
However, these equations are not gauge invariant.

24
4.3 Lorentz Transformations
A Lorentz transform can be defined as any transformation which leaves the
scalar product of two four-vectors invariant. Under a Lorentz transformation,
an arbitrary four-vector Aµ is transformed to Aµ0 , via
Aµ0 = Λµ ν Aν (116)
where the repeated index ν is summed over. The inverse transformation is
represented by
Aµ = ( Λ−1 )µ ν Aν 0 (117)
Since the scalar product is to be relativistically invariant, one requires that
Aµ0 Bµ 0 = Aµ Bµ (118)
The left hand-side is evaluated as
Aµ0 Bµ 0 = Λµ ν Aν gµ,σ Λσ τ B τ (119)
If the scalar product is to be invariant, the transform must satisfy the condition
gν,τ = Λµ ν gµ,σ Λσ τ (120)
If this condition is satisfied, then Λµ ν is a Lorentz transformation.

Like the metric tensor, the Lorentz transformation can be expressed as a

four by four matrix
 0
Λ 0 Λ0 1 Λ0 2 Λ0 3

 Λ1 0 Λ1 1 Λ1 2 Λ1 3 
Λµ ν =   Λ2 0 Λ2 1 Λ2 2 Λ2 3 
 (121)
3 3 3 3
Λ 0 Λ 1 Λ 2 Λ 3
where µ labels the rows and ν labels the columns. In terms of the matrices, the
condition that Λ is a Lorentz transformation can be written as
g = ΛT g Λ (122)
where ΛT is the transpose of the matrix Λ, i.e.
( ΛT )ν µ = Λµ ν (123)

A specific transformation, which is the transformation from a stationary

frame to a reference frame moving along the x(1) axis with velocity v, is repre-
sented by the matrix
− vc
 
1 0 0
 − v 1 q 0 0 
1 c
µ
 
Λ ν = q 
0 0 1 − v 2
0
 (124)
v 2  c2 
1 − c2  q
2

0 0 0 1 − vc2

25
 x(3)
x'(3)

v
x(2)
x'(2)

O
O'

x(1) x'
(1)

Figure 5: Two inertial frames of reference moving with a constant relative ve-
locity with respect to each other.

which can be seen to satisfy the condition

gν,τ = Λµ ν gµ,σ Λσ τ (125)

which has to be satisfied if Λµ ν is to represent a Lorentz transformation.

x(3)

x'(2)
x(2)

x'(1)
φ
O x(1)

Figure 6: Two inertial frames of reference rotated with respect to each other.

Likewise, the rotation through an angle ϕ about the x(3) -axis represented by
 
1 0 0 0
 0 cos ϕ sin ϕ 0 
Λµ ν =  0 − sin ϕ cos ϕ 0 
 (126)
0 0 0 1

is a Lorentz transformation, since it also satisfies the condition eqn(120).

Since the boost velocity v and the angles of rotation ϕ are continuous, one
could consider transformations where these quantities are infinitesimal. Such
infinitesimal transformations can be expanded as

Λµ ν = δ µ ν + µ ν + . . . (127)

26
where δ µ ν is the Kronecker delta function representing the identity transforma-
tion5 and µ ν is a matrix which is first-order in the infinitesimal parameter. The
condition on µ ν required for Λµ ν to be a Lorentz transform is given by

0 = µ ν gµ,τ + gν,σ σ τ (129)

or, on using the metric tensor to lower the indices, one has

τ,ν = − ν,τ (130)

Hence, an arbitrary infinitesimal Lorentz transformation is represented by an

arbitrary antisymmetric 4 × 4 matrix τ,ν . This matrix occurs in the expression
for the transformation matrix

Λτ,ν = gτ,ν + τ,ν + . . . (131)

which transforms the contravariant components of a vector into covariant com-

ponents. It follows that, if ν and τ are either both space-indices or are both
time-indices, the components of the finite Lorentz transformation matrix Λτ ν
are antisymmetric on interchanging τ and ν. Whereas if the pair of indices ν
and τ are mixed space and time-indices, the components of the transformation
matrix Λτ ν are symmetric.

Exercise:

Show that a Lorentz transformation from the unprimed rest frame to the
primed reference frame moving along the x(3) -axis with constant velocity v, can
be considered as a rotation through an imaginary angle θ = i χ in space-time,
where i c t plays the role of a spatial coordinate. Find the equation that deter-
mines χ.

4.4 Invariant Form of Maxwell’s Equations

In physics, one strives to write the fundamental equations in forms which are
independent of arbitrary choices, such as the coordinate system or the choice of
gauge condition. However, in particular applications it is expedient to choose
the coordinate system and gauge condition in ways that highlight the symme-
tries and simplify the mathematics.

We shall introduce an antisymmetric field tensor F µ,ν which is gauge in-

variant. That is, the form of F µ,ν is independent of the choice of gauge. We
5 The student more adept in index-gymnastics may consider the advantages and disadvan-

tages of replacing the Kronecker delta function δ µ ν by g µ ν , since

δ µ ν ≡ g µ ν = g µ,ρ gρ,ν = δµ,ν
(128)

27
shall express the six ( (16−4)
2 = 6) independent components of the antisymmetric
tensor in terms of the four-vector potential Aµ and the contravariant derivative
as
F µ,ν = ∂ µ Aν − ∂ ν Aµ (132)
so the tensor is antisymmetric

F µ,ν = − F ν,µ (133)

It is immediately obvious that F µ,ν is invariant under gauge transformations

Aµ → Aµ0 = Aµ + ∂ µ Λ (134)

since
∂µ∂ν Λ − ∂ν ∂µΛ ≡ 0 (135)
µ,ν
Alternatively, explicit evaluation of F shows that the six independent com-
ponents can be expressed in terms of the electric and magnetic fields, which are
gauge invariant. Components of the field tensor are explicitly evaluated from
the definition as
1 ∂ (1) ∂
F 0,1 = A − φ
c ∂t ∂x1
1 ∂ (1) ∂
= A + φ
c ∂t ∂x(1)
= − E (1) (136)

and
∂ ∂
F 1,2 = A(2) − A(1)
∂x1 ∂x2
∂ ∂
= − A(2) + A(1)
∂x(1) ∂x(2)
= − B (3) (137)

The non-zero components of the field tensor are related to the spatial compo-
nents (i, j, k) of the electromagnetic field by

F i,0 = E (i) (138)

and
F i,j = − ξ i,j,k B (k) (139)
where ξ i,j,k is the Levi-Civita symbol. Therefore, the field tensor can be ex-
pressed as the matrix

−E (1) −E (2) −E (3)

 
0
 E (1) 0 −B (3) B (2) 
F µ,ν =  E (2) B (3)
 (140)
0 −B (1) 
E (3) −B (2) B (1) 0

28
Maxwell’s equations can be written in terms of the field tensor as
4π µ
∂ν F ν,µ = j (141)
c
For µ = i, the field equations become
1 ∂ 0,i ∂ 4 π (i)
F + F j,i = j
c ∂t ∂xj c
1 ∂ (i) ∂ 4 π (i)
− E + ξ i,j,k B (k) = j
c ∂t ∂xj c
 (i)
1 ∂ (i) 4 π (i)
− E + ∇ ∧ B = j (142)
c ∂t c

while for µ = 0 the equations reduce to

∂ 4 π (0)
F j,0 = j
∂xj c
∂
j
E (j) = 4πρ
 ∂x 
∇.E = 4πρ (143)

since F 0,0 vanishes. The above field equations are the two Maxwell’s equations
which involve the sources of the fields. The remaining two sourceless Maxwell
equations are expressed in terms of the antisymmetric field tensor as

∂µ Fν,ρ + ∂ρ Fµ,ν + ∂ν Fρ,µ = 0 (144)

where the indices are permuted cyclically. These internal equations reduce to

∇.B = 0 (145)

when µ, ν and ρ are the space indices (1, 2, 3). When one index taken from the
set (µ, ν, ρ) is the time index, and the other two are different space indices, the
field equations reduce to
1 ∂B
+ ∇ ∧ E = 0 (146)
c ∂t
If two indices are repeated, the above equations are satisfied identically, due to
the antisymmetry of the field tensor.

Alternatively, when expressed in terms of the vector potential, the field equa-
tions of motion are equivalent to the wave equations
 
4π µ
∂ν ∂ ν Aµ − ∂ µ ∂ν Aν = j (147)
c

29
Since four-vectors Aµ and j µ transform as
Aµ0 = Λµ ν Aν
j µ0 = Λµ ν j ν (148)
and likewise for the contravariant derivative
∂ µ0 = Λµ ν ∂ ν (149)
then one can conclude that the field tensor transforms as
F µ,ν 0 = Λµ σ Λν τ F σ,τ (150)
This shows that, under a Lorentz transform, the electric and magnetic fields
(E, B) transform into themselves.

Exercise:

Show explicitly, how the components of the electric and magnetic fields
change, when the coordinate system is transformed from the unprimed refer-
ence frame to a primed reference frame which is moving along the x(3) -axis with
constant velocity v.

5 The Simplest Classical Field Theory

Consider a string stretched along the x-axis, which can support motion in the
y-direction. We shall consider the string to be composed of mass elements
mi = ρ a, that have fixed x-coordinates denoted by xi and are separated by
a distance a. The mass elements can be displaced along the y-axis. The y-
coordinate of the i-th mass element is denoted by yi . We shall assume that the
string satisfies the spatial boundary conditions at each end. We shall assume
that the string satisfies periodic boundary conditions, so that y0 = yN +1 .

The Lagrangian for the string is a function of the coordinates yi and the
velocities dy
dt . The Lagrangian is given by
i

i=N   2  2 
X mi dyi κi
L = − yi − yi−1 (151)
i=1
2 dt 2

The first term represents the kinetic energy of the mass elements, and the second
term represents the increase in the elastic potential energy of the section of the
string between the i-th and (i − 1)-th element as the string is stretched from its
equilibrium position. This follows since, ∆si the length of the section of string
between mass element i and i − 1 in a non-equilibrium position is given by
∆s2i = ( xi − xi−1 )2 + ( yi − yi−1 )2
= a2 + ( yi − yi−1 )2 (152)

30
 y
yi+1

yi yi+1-yi

yi-1 a

xi-1 xi xi+1 x

Figure 7: A string composed of a discrete set of particles of masses mi sepa-

rated by a distance a along the x-axis. The particles can be moved from their
equilibrium positions by displacements yi transverse to the x-axis.

since the x-coordinates are fixed. Thus, if one assumes that the spring constant
for the stretched string segment is κi , then the potential energy of the segment
is given by
κi
Vi = ( yi − yi−1 )2 (153)
2
We shall consider the case of a uniform string for which κi = κ for all i.

The equations of motion are obtained by minimizing the action S which is

defined as the integral
Z T
S = dt L (154)
0
between an initial configuration at time 0 and a final configuration at time T .
The action is a functional of the coordinates yi and the velocities dydt , which are
i

to be evaluated for arbitrary functions yi (t). The string follows the trajectory
yiex (t) which minimizes the action, which travels between the fixed initial value
yi (0) and the final value yi (T ). We shall represent the deviation of an arbitrary
trajectory yi (t) from the extremal trajectory by δyi (t), then
δyi (t) = yi (t) − yiex (t) (155)
The action can be expanded in powers of the deviations δyi as
S = S0 + δ 1 S + δ 2 S + . . . (156)
where S0 is the action evaluated for the extremal trajectories. The first-order
deviation found by varying δyi is given by
Z T i=N
X     ex     
dδyi dyi
δ1 S = dt mi − κ δyi yiex − yi−1
ex ex
+ κ δyi yi+1 − yiex
0 i=1
dt dt
(157)

31
in which yi (T ) and dy
dt are to be evaluated for the extremal trajectory. Since
i

the trajectory which the string follows minimizes the action, the term δ 1 S must
vanish for an arbitrary variation δyi . We can eliminate the time derivative of
the deviation by integrating by parts with respect to t. This yields
T i=N
d dyiex
Z X       
1 ex ex ex ex
δ S = dt − mi δyi − κ δyi yi − yi−1 + κ δyi yi+1 − yi
0 i=1
dt dt
 ex T
X dyi 
+ mi δyi (t) (158)
i
dt 0

The boundary term vanishes since the initial and final configurations are fixed,
so
δyi (T ) = δyi (0) = 0 (159)
Hence the first-order variation of the action reduces to
Z T i=N
d dyiex
X     
1 ex ex ex
δ S = dt δyi − mi − κ 2 yi − yi−1 − yi+1
0 i=1
dt dt
(160)
The linear variation of the action vanishes for an arbitrary δyi (t), if the term in
the square brackets vanishes
d dyiex
   
ex ex ex
mi + κ 2 yi − yi−1 − yi+1 = 0 (161)
dt dt

Thus, out of all possible trajectories, the physical trajectory yiex (t) is determined
by the equation of motion
   
d dyi
mi = − κ 2 yi − yi−1 − yi+1 (162)
dt dt

The momentum pi which is canonically conjugate to yi is determined by

 
∂L
pi = (163)
∂( dy
dt )
i

which yields the momentum as

dyi
pi = mi (164)
dt

The Hamiltonian is defined as the Legendre transform of L, so

X dyi
H = pi − L (165)
i
dt

32
The Hamiltonian is only a function of the pairs of canonically conjugate mo-
menta pi and coordinates yi . This can be seen, considering infinitesimal changes
in yi , dy
dt and pi . The resulting infinitesimal change in the Hamiltonian dH is
i

expressed as
i=N
X      
dyi dyi ∂L dyi ∂L
dH = dpi + pi d( ) − d( ) − dy i
i=1
dt dt ∂( dy
dt )
i dt ∂yi
i=N    
X dyi ∂L
= dpi − dyi (166)
i=1
dt ∂yi

since, the terms proportional to the infinitesimal change d( dy

dt ) vanish identi-
i

cally, due to the definition of pi . From this, one finds

∂H dyi
= (167)
∂pi dt
and  
∂H ∂L
= − (168)
∂yi ∂yi

Therefore, the Hamiltonian is only a function of the pairs of canonically

conjugate variables pi and yi . The Hamiltonian is given by
i=N
X  p2  2 
i κi
H = + yi − yi−1 (169)
i=1
2 mi 2

When expressed in terms of the Hamiltonian, the equations of motion have the
form
dyi ∂H
=
dt ∂pi
dpi ∂H
= − (170)
dt ∂yi
The Hamilton equations of motion reduce to
dyi pi
=
dt mi
   
dpi
= − κi yi − yi−1 + κi+1 yi+1 − yi (171)
dt
for each i value N ≥ i ≥ 1.

One can define the Poisson brackets of two arbitrary quantities A and B in
terms of derivatives with respect to the canonically conjugate variables
  i=N
X  ∂A ∂B 
∂B ∂A
A, B = − (172)
i=1
∂yi ∂pi ∂yi ∂pi

33
The Poisson bracket is antisymmetric in A and B
   
A, B = − B, A (173)

The Poisson brackets of the canonically conjugate variables are given by

 
pi , yj = − δi,j (174)

and    
pi , pj = yi , yj = 0 (175)

We shall show how energy is conserved by considering a finite segment of

the string. For example, we shall consider the segment of the string consisting
of the i-th mass element and the string which connects the i-th and (i − 1)-th
mass element. The energy of this segment will be described by Hi , where
 2  2
ρa dyi κ
Hi = + yi − yi−1 (176)
2 dt 2

The rate of increase energy in this segment is given by

 2        
dHi d yi dyi dyi dyi−1
= ρa +κ yi − yi−1 − κ yi − yi−1
dt dt2 dt dt dt
(177)
One can use the equation of motion to eliminate the acceleration term, leading
to
     
dHi dyi dyi−1
= κ yi+1 − yi − κ yi − yi−1 (178)
dt dt dt

The increase in energy of this segment, per unit time, is clearly given by the
difference of the quantity
  
dyi
Pi = − κ yi+1 − yi (179)
dt

at the front end of the segment and Pi−1 at the back end of the segment. Since,
from continuity of energy, the rate of increase in the energy of the segment must
equal the net inflow of energy into the segment, one can identify Pi as the flux
of energy flowing out of the i-th into the (i + 1)-th segment.

34
5.1 The Continuum Limit
The displacement of each element of the string can be expressed as a function
of its position, via
yi = y(xi ) (180)
where each segment has length a, so that xi+1 = xi + a. The displacement
y(xi+1 ) can be Taylor expanded about xi as

a2 ∂ 2 y 
 
∂y 
y(xi+1 ) = y(xi ) + a + + ... (181)
∂x xi 2! ∂x2 xi

We intend to take the limit a → 0, so that only the first few terms of the series
need to be retained. The summations over i are to be replaced by integrations
N Z L
X 1
→ dx (182)
i=1
a 0

The tension in the string T is given by

T = κa (183)

and this has to be kept constant when the limit a → 0 is taken.

In the continuum limit, the Lagrangian L can be expressed as an integral of

the Lagrangian density L as
Z L
L = dx L (184)
0

where   2  2 
1 dy ∂y
L = ρ − κa (185)
2 dt ∂x

The equations of motion are found from the extrema of the action
Z T Z L
S = dt dx L (186)
0 0

It should be noted, that in S time and space are treated on the same footing
and that L is a scalar quantity.

In the continuum limit, the Hamiltonian is given by

Z L
H = dx H (187)
0

35
where the Hamiltonian density H is given by
  2  2 
1 dy ∂y
H = ρ + κa (188)
2 dt ∂x

and the energy flux P is given by

  
dy ∂y
P = −κa (189)
dt ∂x

The condition of conservation of energy is expressed as the continuity equation

dH ∂P
+ = 0 (190)
dt ∂x

5.2 Normal Modes

The solutions of the equations of motion are of the form of a (real) superposition
of plane waves  
1
ψk (x) = √ exp i ( k x − ω t ) (191)
L
The above expression satisfies the wave equation if the frequency ω satisfies the
dispersion relation
ωk2 = v 2 k 2 (192)
The above dispersion relation yields both positive and negative frequency solu-
tions. If the plane-waves are to satisfy periodic boundary conditions, k must be
quantized so that
2π
kn = n (193)
L
for integer n. The positive-frequency solutions shall be written as
 
1
ψk (x) = √ exp i ( kn x − ωn t ) (194)
L
and the negative frequency solutions as
 
∗ 1
ψ−k (x) = √ exp i ( kn x + ωn t ) (195)
L
These solutions form an orthonormal set since
Z L
dx ψk∗0 (x) ψk (x) = δk0 ,k (196)
0

Hence, a general solution can be written as

X  
∗ ∗
y(x) = ck (0) ψk (x) + c−k (0) ψ−k (x) (197)
k

36
where the ck are arbitrary complex numbers that depend on k. If the time
dependence of the ψk (x) is absorbed into the complex functions ck via
 
ck (t) = ck (0) exp − i ωk t (198)

then one has

   
1 X ∗
y(x) = √ ck (t) + c−k (t) exp i k x (199)
L k

which is purely real. Thus, the field y(x) is determined by the amplitudes of
the normal modes, i.e. by ck (t). The time-dependent amplitude ck (t) satisfies
the equation of motion
d2 ck
= − ωk2 ck (200)
dt2
and, therefore, behaves like a classical harmonic oscillator. To quantize this
classical field theory, one needs to quantize these harmonic oscillators.

The Hamiltonian is expressed as

Z L   2 
1 1 2 2 ∂y(x)
H = dx p(x) + κ a (201)
2a 0 ρa ∂x

On substituting y(x) in the form

   
1 X ∗
y(x) = √ ck (t) + c−k (t) exp i k x (202)
L k

and
dc∗−k
   
ρa X dck
p(x) = √ + exp i k x (203)
L k dt dt
then after integrating over x, one finds that the energy has the form
dc∗k dc∗−k
  
ρ X dc−k dck
H = + +
2 dt dt dt dt
k
   
κa X 2
+ k c−k (t) + c∗k (t) ck (t) + c∗−k (t) (204)
2
k

Furthermore, on using the time-dependence of the Fourier coefficients ck (t), one

has
  
ρ X 2 ∗ ∗
H = − ωk c−k (t) − ck (t) ck (t) − c−k (t)
2
k
  
κa X 2 ∗ ∗
+ k c−k (t) + ck (t) ck (t) + c−k (t) (205)
2
k

37
but the frequency is given by the dispersion relation
 
κa
ωk2 = v 2 k 2 = k2 (206)
ρ
Therefore, the expression for the Hamiltonian simplifies to
X  
H = ρ ωk2 c∗k (t) ck (t) + c−k (t) c∗−k (t)
k
X  
= ρ ωk2 c∗k (0) ck (0) + c−k (0) c∗−k (0) (207)
k

which is time-independent, since the time-dependent phase factors cancel out.

Thus, one can think of the energy as a function of the variables ck and c∗−k .
Since the Hamiltonian is strictly expressed in terms of canonically conjugate co-
ordinates and momenta, one should examine the Poisson brackets of ck and c∗−k .

The variables y(xi ) and p(xj ) have the Poisson brackets

 
p(xi ) , y(xj ) = − δi,j
   
p(xi ) , p(xj ) = y(xi ) , y(xj ) = 0 (208)

Due to the orthogonality properties of the plane-waves, one has

   
a X
ck + c∗−k ≈ √ y(xi ) exp − i k xi (209)
L i
and also
   
a X
− i ωk 0 ρ a ck0 − c∗−k0 ≈ √ p(xj ) exp − i k xj0
(210)
L j

These relations are simply the results of applying the inverse Fourier transform
to y(x) and p(x). One can find the Poisson brackets relations between ck and
c∗k from
   
− i ωk 0 ρ ck0 − c∗−k0 , ck + c∗−k
   
a X
= − p(xi ) , y(xj ) exp − i ( k xi + k 0 xj )
L i,j
 
a X 0
= + δi,j exp − i ( k xi + k xj )
L i,j
 
a X 0
= + exp − i ( k + k ) xi
L i
= + δk+k0 (211)

38
Likewise, one can obtain similar expressions for the other commutation relations.
This set of equations can be satisfied by setting
 
i
c∗k0 , ck = δk,k0 (212)
2 ωk ρ

and    
c∗k0 , c∗k = ck0 , ck = 0 (213)

The above set of Poisson brackets can be recast in a simpler form by defining
1
ck = √ ak (214)
2 ωk ρ

etc., so that the Poisson brackets reduce to

 
∗
ak0 , ak = i δk,k0 (215)

and    
a∗k0 , a∗k = ak0 , ak = 0 (216)

where the non-universal factors have cancelled out.

5.3 Rules of Canonical Quantization

The first rule of Canonical Quantization states, “Physical quantities should be
represented by operators”. Hence ak and a∗k should be replaced by the operators
âk and â†k . The second rule of Canonical Quantization states, “Poisson Brackets
should be replaced by Commutators”. Hence,
 
i h̄ A , B → [ Â , B̂ ] (217)

So one has
[ â†k0 , âk ] = − h̄ δk,k0 (218)
and
[ â†k0 , â†k ] = [ âk0 , âk ] = 0 (219)
To get rid of the annoying h̄ in the commutator, one can set
√
âk = h̄ b̂
†
√ k†
âk = h̄ b̂k (220)

Whether it was noted or not, b̂†k is the Hermitean conjugate of b̂k . The Her-
mitean relation can proved by taking the Hermitean conjugate of ŷ(xi ), and

39
noting that the third rule of quantization states, “Measurable quantities are to
replaced by Hermitean operators”. Therefore, the operator
s    
1 X h̄
ŷ(x) = √ b̂k (t) + b̂†−k (t) exp i k x (221)
L 2 ρ ωk
k

must be Hermitean. What this means is, the Hermitean conjugate

s    
† 1 X h̄ † † †
ŷ (x) = √ b̂k (t) + ( b̂−k (t)) exp − i k x (222)
L k 2 ρ ωk

has to be the same as ŷ(x). On setting k = −k 0 in the above equation, one has
s    
† 1 X h̄ † † † 0
ŷ (x) = √ b̂−k0 (t) + ( b̂k0 (t)) exp + i k x (223)
L k0 2 ρ ωk 0

For ŷ † (x) to be equal to ŷ(x), it is necessary that the Hermitean conjugate of

the operator b̂†k0 is equal to b̂k0 . This shows that the pair of operators are indeed
Hermitean conjugates. The quantum field is represented by the operator
s    
1 X h̄ † 0
ŷ(x) = √ b̂−k0 (t) + b̂k0 (t) exp + i k x (224)
L 0 2 ρ ωk 0
k

where the time-dependent creation and annihilation operators are given by

 
b̂k (t) = b̂k exp − i ωk t
 
b̂†k (t) = b̂†k exp + i ωk t (225)

The quantized Hamiltonian becomes

 
† †
X
2
Ĥ = ρ ωk ĉk ĉk + ĉ−k c−k
k
X h̄ ωk  † †

= b̂k b̂k + b̂−k b̂−k (226)
2
k

On transforming k → −k in the second term of the summation, one obtains the

standard form
X h̄ ωk  † †

Ĥ = b̂k b̂k + b̂k b̂k (227)
2
k

40
where the b̂k and b̂†k are to be identified as annihilation and creation operators
for the quanta.

The quantum operator P̂ corresponding to the classical quantity P

Z L
P = dx P (228)
0

is evaluated as
  X   
κa
P̂ = − h̄ k b̂†−k − b̂k b̂†k + b̂−k (229)
2ρ
k

where the plane-wave orthogonality properties have been used. This quantity
can be expressed as the sum of two terms
   
κa X † †
P̂ = − h̄ k b̂−k b̂k − b̂k b̂−k
2ρ
k
   
κa X
+ h̄ k b̂k b̂†k − b̂†−k b̂−k (230)
2ρ
k

The operator P̂ can be shown to be equivalent to

h̄ k b̂†k b̂k
X
P̂ = v 2 (231)
k

which obviously is proportional to the sum of the momenta of the quanta. The
quantity κρa is just the square of the wave velocity v 2 . On noting that the
quanta travel with velocities given by v sign(k) and have energies given by
h̄ ωk = h̄ v | k |, one sees that P is the expressed as the total energy flux
associated with the quanta.

5.4 The Algebra of Boson Operators

The number operator n̂k can be defined6 as

n̂k = b̂†k b̂k (232)

which has eigenstates | nk > with eigenvalues nk

n̂k | nk > = nk | nk > (233)

The eigenvalues nk are positive integers, including zero. This can be inferred
from the commutation relations

[ b̂†k , b̂k0 ] = − δk,k0 (234)

6 P. Jordan and O. Klein, Zeit. fūr Physik, 45, 751 (1927).

41
which has the consequence that

[ n̂k , b̂†k0 ] = + δk,k0 b̂†k0

[ n̂k , b̂k0 ] = − δk,k0 b̂k0 (235)

Hence, when b̂k acts on an eigenstate of n̂k with eigenvalue nk it produces

another eigenstate of n̂k but with an eigenvalue of nk − 1, as can be seen since

n̂k b̂k | nk > = b̂k ( n̂k − 1 ) | nk >

= ( nk − 1 ) b̂k | nk > (236)

Therefore, since b̂k lowers the eigenvalue of the number operator by one unit,
one can write
b̂k | nk > = C(nk ) | nk − 1 > (237)
where the complex number C(nk ) has to be determined. The normalization
coefficient C(nk ) can be determined by noting that

< nk | b̂†k (238)

is the Hermitean conjugate of the state

b̂k | nk > (239)

On taking the norm of the state and its conjugate, one finds the normalization

< nk | b̂†k b̂k | nk > = C ∗ (nk ) C(nk ) < nk − 1 | nk − 1 >

= | C(nk ) |2 (240)

However, on using the definition of the number operator and the normalization
condition, one finds that
| C(nk ) |2 = nk (241)
so, on choosing the phase factor, one can define
√
b̂k | nk > = nk | nk − 1 > (242)

as the annihilation operator.

Likewise, one can see that Hermitean conjugate operator b̂†k when acting on
an eigenstate of the number operator increases its eigenvalue by one unit

n̂k b̂†k | nk > = b̂†k ( n̂k + 1 ) | nk >

= ( nk + 1 ) b̂†k | nk > (243)

Therefore, one has 0

b̂†k | nk > = C (nk ) | nk + 1 > (244)

42
 0
The coefficient C (nk ) is found from the normalization condition
0
< nk | b̂k b̂†k | nk > = C ∗ (nk ) C 0 (nk ) < nk + 1 | nk + 1 >
0
= | C (nk ) |2 (245)
which with
b̂k b̂†k = n̂k + 1 (246)
yields 0
| C (nk ) |2 = nk + 1 (247)
Since, the phase factor has already been determined by the Hermitean conjugate
equation, one has √
b̂†k | nk > = nk + 1 | nk + 1 > (248)
which raises the eigenvalue of the number operator.

Hence, one sees that the eigenvalues of the number operator are separated
by integers. Furthermore, the smallest eigenvalue corresponds to nk = 0, since
for nk = 0 the equation
√
b̂k | nk > = nk | nk − 1 > (249)
reduces to
b̂k | 0 > = 0 (250)
Hence, the hierarchy of states produced by the annihilation operator acting on
a number operator eigenstate terminates at nk = 0. Thus, the eigenvalues of
the number operator nk can have integer values 0 , 1 , 2 , 3 , . . . , ∞.

Therefore, any arbitrary number operator eigenstate | {nk } > , in which

the number of excitations (nk ) in the each normal mode has been specified, can
be written in terms of the vacuum state | 0 > and the creation operators as
Y  ( b̂† )nk 
| {nk } > = √k |0 > (251)
k
nk !

The repeated operation of the creation operator b̂†k creates a state with nk
bosonic excitations present in mode k and the denominator provides the correct
normalization for this state.

Any arbitrary state | Ψ > can be expressed as a linear superposition of

number operator eigenstates
X
|Ψ >= C({nk }) | {nk } > (252)
{nk }

where the sum runs over all possible number eigenstates, and the complex coef-
ficients C({nk }) are arbitrary except that they must satisfy the normalization
condition X
| C({nk }) |2 = 1 (253)
{nk }

43
5.5 The Classical Limit
The classical limit of the quantum field theory can be characterized by the limit
in which the field operator can be replaced by a function. This requires that the
“classical” states are not only described as states with large numbers of quanta
in the excited normal modes, but also that the state is a linear superposition
of states with different number of quanta, with a reasonable well defined phase
of the complex coefficients. For a quantum state to ideally represent a given
classical state, one needs the quantum state to be composed of a coherent su-
perposition of states with different numbers of quanta.

That states which are eigenstates of the number operators ( | {nk } > ) can
not represent classical states, can be seen by noting that the expectation value
of the field operator is zero

< {nk } | ŷ(x) | {nk } > = 0 (254)

follows from the expectation value of the creation and annihilation operators

< {nk } | ak | {nk } > = 0 (255)

Despite the fact that the average value of the field is zero, the fluctuation in the
field amplitude is infinite since
  
2 1 X h̄ † †
< {nk } | ŷ(x) | {nk } > = < {nk } | b̂k0 + b̂−k 0 b̂k + b̂−k0
0 | {nk } >
L 0 2 ρ ωk 0
k
1 X h̄
= ( 1 + 2 nk 0 ) (256)
L 0 2 ρ ωk 0
k

and the zero-point contribution diverges logarithmically at the upper and lower
limits of integration.

Hence, the eigenstates of the number operator. or equivalently Ĥ, do not

describe the classical states of the string. Classical states must be expressed as
a linear superposition of energy eigenstates.

6 Classical Field Theory

The dynamics of a multi-component classical field φα is governed by a Lagrange
density L, which is a scalar quantity that is a function of the fields φα and their
derivatives ∂µ φα . The equations of motion for the classical field are determined
by the principle of extremal action. That is, the classical fields are those for
which the action S
Z Z  
0 3 α α
S = dt d x L φ , ∂µ φ (257)

44
is extremal. An arbitrary field φα can be expressed in terms of the extremal
value φα α
ex and the deviation δφ as

φα = φα
ex + δφ
α
(258)

The space and time derivatives of the arbitrary field can also be expressed as
the derivatives of the sum of the extremal field and the deviation

∂ν φα = ∂ν φα
ex + ∂ν δφ
α
(259)

The first-order change in the action δS is given by

Z t Z     
∂ ∂
δS = dt0 d3 x δφα L φα
ex , ∂ φα
µ ex + (∂ν δφ α
) L φα
ex , ∂ φα
µ ex
0 ∂φα ∂(∂ν φα )
(260)
On integrating by parts with respect to xν in the last term, and on assuming
appropriate boundary conditions, one finds
Z t Z      
0 3 α ∂ α α ∂ α α
δS = dt d x δφ L φex , ∂µ φex − ∂ν L φex , ∂µ φex
0 ∂φα ∂(∂ν φα )
(261)
which has to vanish for an arbitrary choice of δφα . Hence, one obtains the
Euler-Lagrange equations
    
∂ α α ∂ α α
L φex , ∂ µ φ ex = ∂ν L φex , ∂ µ φ ex (262)
∂φα ∂(∂ν φα )

This set of equations determine the time dependence of the classical fields
φα
ex (x). That is, out of all possible fields with components φα , the equations
of motion determine the physical field which has the components φα ex . It is
convenient to define the field momentum density πα0 (x) conjugate to φα as
 
0 ν 1 ∂ β β
πα (x ) = L φ , ∂µ φ (263)
c ∂(∂0 φα )

The Hamiltonian density H is then defined as the Legendre transform

X
H = c πα0 (∂0 φα ) − L (264)
α

which eliminates the time-derivative of the fields in terms of the momentum

density of the fields.

Exercise:

Starting from the Lorentz scalar Lagrangian

  2 
1 µ mc 2
L = ( ∂µ φ ) ( ∂ φ ) − φ (265)
2 h̄

45
for a real scalar field φ, determine the Euler-Lagrange equation and the Hamil-
tonian density H.

Exercise:

Consider the Lagrangian density

  2 
1 mc
L = ( ∂µ ψ ∗ ) ( ∂ µ ψ ) − | ψ |2 (266)
2 h̄

for a complex scalar field ψ. Treat ψ and ψ ∗ as independent fields.

(i) Determine the Euler-Lagrange equation and the Hamiltonian density H.
(ii) By Fourier transforming with respect to space and time, determine the form
of the general solution for ψ.

Exercise:

The Lagrangian density for the complex field ψ representing a charged par-
ticle is given by

h̄2
        ∗ 
h̄ ∂ψ ∂ψ
L = − ∇ ψ∗ . ∇ ψ − ψ∗ − ψ − ψ ∗ V (x) ψ
2m 2i ∂t ∂t
(267)
(i) Determine the equation of motion, and the Hamiltonian density H.
(ii) Consider the case V (x) ≡ 0, then by Fourier transforming with respect to
space and time, determine the form of the general solution for ψ.

6.1 The Hamiltonian Formulation

The Hamiltonian formulation reserves a special role for time, and so is not
Lorentz covariant. However, the Hamiltonian formulation is the most conve-
nient formulation for quantizing fields. The Hamilton equations of motion are
determined from the Hamiltonian
Z
H = d3 x H (268)

by noting that H is only a functional of πα0 and φα . This can be seen, since as
Z  X 
H = d3 x c πα0 (∂0 φα ) − L (269)
α

then, the first-order variation of the Hamiltonian δH is given by

Z  X   
3 0 α 0 α
δH = d x c δπα (∂0 φ ) + πα (∂0 δφ ) − δL (270)
α

46
but, from the Lagrangian formulation of field theory, one has
1
δL = δφα (∂0 πα0 ) + (∂0 δφα ) πα0 (271)
c
where the Euler-Lagrange equations were substituted into the first term. There-
fore, the variation in the Hamiltonian is given by
Z X  
3 0 α α 0
δH = d xc δπα (∂0 φ ) − δφ (∂0 πα ) (272)
α

which does not involve the time derivative of the fields. This implies that
the Hamiltonian is a function of the fields πα0 , φα and their derivatives. On
calculating the variation of H using the independent variables πα0 and φα , and
integrating by parts, one finds that the Hamiltonian equations of motion are
given by
   
∂H ∂H
c ∂0 φα = − ∇
∂π 0 ∂(∇πα0 )
 α  
∂H ∂H
− c ∂0 πα0 = − ∇ (273)
∂φα ∂(∇φα )

The structure of these equations are similar to those of the classical mechanics of
point particles. Similar to classical mechanics of point particles, one can define
Poisson Brackets with fields. When quantizing the fields, the Poisson Bracket
relations between the fields can be replaced by commutation relations.

6.2 Symmetry and Conservation Laws

Emmy Noether produced a theorem linking continuous symmetries of a La-
grangian to conservation laws7 .

6.2.1 Conservation Laws

Consider a Lagrangian density L which is a function of a set of fields φα (x) and
their derivatives defined in a Minkowski space x. Consider how the Lagrangian
density changes for a particular choice of a combination of infinitesimal trans-
formations of the field components

φα (x) → φα0 (x) = φα (x) + δφα (x) (274)

and, as a consequence, the derivatives of the field components also transform as

∂µ φα (x) → ∂µ φα0 (x) = ∂µ φα (x) + ∂µ δφα (x) (275)

7 E. Noether, Nachr. d. Kgl. Gessch. d. Wiss. Gottingen, K1. Math. Phys. (1918) 235.

47
Under this combined transformation, the Lagrangian density changes by an
infinitesimal amount δL, given by
 
∂L ∂L
δL = α
∂µ δφα + δφα (276)
∂(∂µ φ ) ∂φα

where the field index α is to be summed over. However, the generalized mo-
mentum density παµ (x) is defined by
 
µ ∂L
πα (x) = (277)
∂(∂µ φα )
so
∂L
δL = παµ (∂µ δφα ) + δφα (278)
∂φα
The Euler-Lagrange equation for each field φα is given by
∂L
∂µ παµ − = 0 (279)
∂φα
where φα satisfies the appropriate boundary conditions. Thus, on adding and
subtracting a term
(∂µ παµ ) δφα (280)
to δL, one finds
   
∂L
δL = παµ ∂µ δφα + (∂µ παµ ) δφα + − ∂ π
µ α
µ
δφα
∂φα
   
∂L
= ∂µ παµ δφα + − ∂ π
µ α
µ
δφα
∂φα
 
= ∂µ παµ δφα (281)

since the last term in the second line vanishes if the fields φα satisfy the Euler-
Lagrange equations. If the Lagrangian is invariant under the transformation,
then δL = 0, so  
∂µ παµ δφα = 0 (282)

where the field index α is to be summed over. The above equation can be
re-written as a continuity equation

∂µ j µ = 0 (283)

where the conserved current j µ (x) is given by

j µ (x) ∝ παµ (x) δφα (x)

 
∂L
∝ δφα (x) (284)
∂(∂µ φα )

48
up to a constant of proportionality. The normalization of the conserved current
is arbitrary and can be chosen at will. Since it is recognized that δφα is in-
finitesimal, the normalization is chosen by introducing an infinitesimal constant
 via

 j µ (x) = παµ (x) δφα (x)

 
∂L
= δφα (x) (285)
∂(∂µ φα )

The conserved charge Q is defined as the integral over all space of the time
component of the current density j (0) . That is, the conserved charge is given by
Z
Q = d3 x j (0) (x) (286)

or, more specifically

Z
Q = d3 x πα(0) (x) δφα (x)
Z  
∂L
= d3 x δφα (x) (287)
∂(∂0 φα )

Since  is a constant, the total charge Q is constant. Therefore, the total time
derivative of Q vanishes
dQ
= 0 (288)
dt
The spatial components of j µ form the current density vector.

6.2.2 Noether Charges

Consider the infinitesimal variation of a complex field φα (x) defined by
α
X
φα (x) → φ0 (x) = φα (x) + i  λα β φβ (x) (289)
β

If this infinitesimal variation leads to L being invariant, one has a conserved

current
X  ∂L 
jµ = i λα β φβ (x) (290)
∂(∂µ φα )
α,β

An important example is given by the infinitesimal transformation

ψ0 = ψ + iψ
ψ ∗0 = ψ∗ − i  ψ∗ (291)

49
where ψ and its complex conjugate ψ ∗ are regarded as independent fields. The
transformation represents a an infinitesimal constant shift of the phase of the
field8 . The conserved current is
    
µ ∂L ∂L ∗
j = −i ψ(x) − ψ (x) (292)
∂(∂µ ψ) ∂(∂µ ψ ∗ )

which is the electromagnetic current density four-vector.

Exercise:

The Lagrangian density for the complex Schrödinger field representing a

charged particle is given by

h̄2
        ∗ 
∗ h̄ ∗ ∂ψ ∂ψ
L = − ∇ψ . ∇ψ − ψ − ψ − ψ ∗ V (x) ψ
2m 2i ∂t ∂t
(293)
(i) Determine the conserved Noether charges.

Exercise:

Determine the Noether charges for a complex Klein-Gordon field theory,

governed by the Lagrangian density
  2 
1 mc
L = ( ∂µ ψ ∗ ) ( ∂ µ ψ ) − | ψ |2 (294)
2 h̄

6.2.3 Noether’s Theorem

The basic theorem can be generalized to the case where the Lagrangian density
is not invariant under the infinitesimal transformation, but instead changes by
a combination of total derivatives. That is,

δL =  ∂µ Λµ (295)
8 This particular transformation is a specific example of a gauge transformations of the first

kind, in which
 
q
ψ 0 (x) = exp − i Λ(x) ψ(x)
h̄ c

A gauge transformation of the second kind is one in which the field changes according to
Aµ0 = Aµ + (∂ µ Λ)
q
Since p̂µ = i h̄ ∂ µ , the combination of these transformations keep the quantity (p̂µ − c
Aµ )ψ
invariant

50
for some analytic vector function with components Λµ . This type of transforma-
tion does not change the total action. If the Lagrangian changes by the above
amount for the combined transformation δφα
φα (x) → φα0 (x) = φα (x) + δφα (x) (296)
then as has been previously shown
   
∂L
δL = παµ (∂µ δφα ) + (∂µ παµ ) δφα + − ∂µ π µ
α δφα
∂φα
   
∂L
= ∂µ παµ δφα + − ∂µ π µ
α δφα
∂φα
 
= ∂µ παµ δφα (297)

one has
 
 ∂µ Λ µ = ∂µ παµ δφα (298)

If the conserved currents are identified as

 
µ µ α
 j = πα δφ −  Λµ (299)

then the continuity condition

∂µ j µ = 0 (300)
holds.

6.3 The Energy-Momentum Tensor

An example of Noether’s theorem is given by the transformation
φα (x) → φα (x + ) = φα (x) + µ (∂µ φα ) (301)
which represents an infinitesimal space-time translation. This is a symmetry
appropriate to a Lagrangian density L which has no explicit x dependence. We
shall assume that the Lagrangian density only depends on the field φα and its
derivatives ∂ν φα  
L = L φα , (∂ν φα ) (302)

In this case, the change in the Lagrangian density is given by the total derivative
   
∂L α ∂L
δL = (∂ ν δφ ) + δφα
∂(∂ν φα ) ∂φα
     
∂L ∂L
= µ ∂ ∂
ν µ φα
+ ∂µ φα
∂(∂ν φα ) ∂φα
= µ ∂µ L (303)

51
where the last line follows since the Lagrangian only depends implicitly on xµ
through the fields. Hence, the change in the Lagrangian is a total derivative

δL = µ ∂µ Λ (304)

where Λ = L. Therefore, for transformations of the type

φα → φα + µ (∂µ φα ) (305)

Noether’s theorem takes the form

        
∂L ∂L
µ ∂µ L = µ ∂ν ∂µ φα
+ ∂ µ φ α
∂(∂ν φα ) ∂φα
      
∂L ∂L
= µ ∂ν ∂µ φα
+ ∂ν ∂µ φα
∂(∂ν φα ) ∂(∂ν φα )
  
∂L
= µ ∂ν ∂µ φα (306)
∂(∂ν φα )

where the Euler-Lagrange equation has been used in the second line. Thus, the
fields satisfy the continuity conditions
   
∂L
0 = µ ∂ν ∂µ φα
− δ ν
µ L (307)
∂(∂ν φα )

where

δµ ν = 1 if µ = ν
δµ ν = 0 otherwise (308)

The conserved current density is identified as

   
∂L
T νµ = ∂µ φα
− δ ν
µ L (309)
∂(∂ν φα )

which is the energy-momentum density Tµν . The energy momentum tensor sat-
isfies the conservation law
∂ν T ν µ = 0 (310)
The second-rank tensor can be written in contravariant form as
   
∂L
T ν,µ = ∂ µ α
φ − g ν,µ
L (311)
∂(∂ν φα )

where the metric tensor has been used to raise the index µ. The component
with µ = ν = 0 is the Hamiltonian density H for the fields
   
0,0 ∂L α
H = T = ∂0 φ − L (312)
∂(∂0 φα )

52
so the total energy of the field is given by
Z Z
3
E = d xH = d3 x T 0,0 (313)

The energy is conserved since

1 ∂H X ∂
+ (j)
T j,0 = 0 (314)
c ∂t j
∂x

where the components c T j,0 represents the components of the energy-density

flux. Likewise, the components
 
T 0,j = c πα(0) ∂ (j) φα (315)

are related to the momentum density since the total momentum of the field is
given by Z
1
P (j) = d3 x T 0,j (316)
c

Since T 0,j is the momentum density, one expects that the components of the
orbital angular momentum density are proportional to

M 0,j,k = T 0,j x(k) − T 0,k x(j) (317)

One can define a third-rank tensor via

M µ,ν,ρ = T µ,ν xρ − T µ,ρ xν (318)

The divergence of the third-rank tensor is evaluated as

   
µ,ν,ρ µ,ν ρ µ,ν ρ µ,ρ
∂µ M = ∂µ T x + T δ µ − ∂µ T xν − T µ,ρ δ ν µ

= T ρ,ν − T ν,ρ (319)

where the conservation law for T µ,ν and the condition

∂µ xρ = δµ ρ (320)

expressing the independence of the variables xρ and xµ have been used. The
divergence of the third-rank tensor vanishes if T µ,ν is symmetric. Thus, the
angular momentum tensor M µ,ν,ρ is conserved if the energy-momentum tensor
is symmetric.

It should be noted that the tensor T µ,ν is only symmetric for scalar fields.
This is related to the fact that a vector or tensor field carries a non-zero intrin-
sic angular momentum. It is possible to incorporate an additional term in the

53
momentum-energy tensor of a vector field to make it symmetric.

Exercise:

(i) Determine the momentum-energy tensor for a complex scalar field ψ governed
by the Lagrangian density
  2 
1 ∗ µ mc 2
L = ( ∂µ ψ ) ( ∂ ψ ) − |ψ| (321)
2 h̄

(ii) Find the forms of the energy and momentum density of the field.
(iii) Using the form of the general solution, find expressions for the total energy
and momentum of the field in terms of the Fourier components of the field.

Exercise:

(i) Determine the energy-momentum tensor for the Lagrangian density for the
complex Schrödinger field representing a charged particle given by

h̄2
        ∗ 
∗ h̄ ∗ ∂ψ ∂ψ
L = − ∇ψ . ∇ψ − ψ − ψ − ψ ∗ V (x) ψ
2m 2i ∂t ∂t
(322)
(ii) Find the forms of the energy and momentum density of the field.
(iii) Find the forms of the generalized orbital angular momentum density of the
field.
(iv) Consider the case where V (x) ≡ 0. Using the form of the general solution,
find expressions for the total energy and momentum of the field in terms of the
Fourier components of the field.

7 The Electromagnetic Lagrangian

The Lagrangian for a source-free electromagnetic field must be gauge invariant
and must be a Lorentz scalar. An appropriate scalar Lagrange density can be
constructed as
1
L = − F µ,ν Fµ,ν (323)
16 π
where Aµ are the fields. The constant of proportionality is merely a matter
of convention. The Euler-Lagrange equations are found by expressing the La-
grangian density in the symmetrical form
1
L = − F µ,ν gµ,σ gν,τ F σ,τ (324)
16 π
From the above expression, it is seen that the two factors of the antisymmetri-
cal second-rank field tensors produce identical variations of the action. Further-
more, in this form the Lagrangian only depends on the contravariant derivatives

54
of the contravariant components of the field. This form allows variations to be
made directly without using the properties of the metric tensor. The first-order
variation of the action can be expressed as
Z  
2 4 µ,ν µ,ν
δS = − d x ( ∂µ δAν ) F − ( ∂ν δAµ ) F
16 π c
Z   
2 4 µ,ν ν,µ
= d x δAν ∂µ F − F
16 π c
Z
1
= d4 x δAν ( ∂µ F µ,ν ) (325)
4πc
where the second line has been obtained by integrating by parts and the last
line was obtained by using the antisymmetric nature of the field tensor. The
vanishing of the first-order variation of the action δS, for arbitrary δAν , yields
the Euler-Lagrange equation

∂µ F µ,ν = 0 (326)

which is the same as Maxwell’s equations in the absence of any sources.

In the absence of the source, the Lagrangian density is gauge invariant.

This can be seen by noting that the contravariant field tensor F µ,ν is gauge
invariant, and the covariant tensor is obtained from the contravariant tensor by
lowering both indices with the metric tensor. The contravariant field tensor can
be expressed as the matrix

−E (1) −E (2) −E (3)

 
0
 E (1) 0 −B (3) B (2) 
F µ,ν ≡   E (2) B (3)
 (327)
0 −B (1) 
E (3) −B (2) B (1) 0

and the co-variant field tensor can be expressed as the matrix

E (1) E (2) E (3)

 
0
 −E (1) 0 −B (3) B (2) 
Fµ,ν ≡   −E (2) B (3)
 (328)
0 −B (1) 
−E (3) −B (2) B (1) 0

in which the sign of the terms with mixed time and space indices have changed.
Therefore, the Lagrangian density can be expressed in terms of the electromag-
netic fields as
1
L = ( E2 − B2 ) (329)
8π
Since the Lagrangian density is completely expressed in terms of the electro-
magnetic field, it is gauge invariant.

55
 In the presence of source densities, the Lagrangian density is extended to
include the interaction to become
1 1
L = − F µ,ν Fµ,ν − Aµ j µ (330)
16 π c
This interaction term is the only Lorentz scalar that one can form with the
four-vector current and the field. It should be noted that the last term is not
gauge invariant. This action yields the equation of motion
4π ν
∂µ F µ,ν = j (331)
c
as expected.

The lack of gauge invariance in the interaction Lagrangian

1 µ
Lint = − A jµ (332)
c
does not affect the equations of motion. On performing the gauge transforma-
tion
Aµ → Aµ0 = Aµ + ∂ µ Λ (333)
one finds that the interaction part of the Lagrangian density is transformed to
 
1
Lint 0 = − Aµ + ∂ µ Λ jµ (334)
c

Since charge is conserved, the current density must satisfy the continuity equa-
tion
∂ µ jµ = 0 (335)
The continuity condition can be used to express the interaction as the untrans-
formed Lagrangian density and a perfect derivative
1 µ 1 µ
Lint 0 = − A jµ − ∂ ( Λ jµ ) (336)
c c
The perfect derivative term only adds a constant term to the action which does
not affect the equations of motion9 . Hence, although the Lagrangian density
is not gauge invariant in the presence of sources, the Lagrangian equations of
motion are gauge invariant.

The momentum density conjugate to Aµ is calculated as

c
π 0,µ = − F 0,µ (337)
4π
9 The change in the form of the interaction Lagrangian density produced by a gauge trans-

formation should be taken as a warning against considering quantities in a field theory as

being localized.

56
which vanishes for µ = 0, indicating that the scalar potential A0 is not a
dynamic variable. This suggests that it may be appropriate to completely fix
the scalar potential by a choice of gauge, such as the Coulomb gauge which leads
to the scalar potential φ being fixed by Poisson’s equation. In the presence of
sources, the Hamiltonian density is expressed as
1
H = − ( ∂0 Aν ) F 0,ν − L
4π
1
= − ( F0,ν + ∂ν A0 ) F 0,ν − L
4π
1 1 1 µ
= − ( F0,ν + ∂ν A0 ) F 0,ν + ( B2 − E2 ) + j Aµ
4π 8π c
1 1 1 µ
= + ( E2 + B2 ) − ( ∇ . E ) A0 + j Aµ
8π 4π c
1
+ ∇ . ( A(0) E ) (338)
4π
The fourth line has been derived by noting that the non-zero components of
F 0,µ are only non-zero for space-like µ and are given by

F 0,i = − E (i) (339)

Thus, the first term in the third line is given by

1 1
− F0,ν F 0,ν = + E2 (340)
4π 4π
which can be combined with the term
1
− ( E2 − B2 ) (341)
8π
originating from the Lagrangian density. This combination results in the term
 
1
E2 + B2 (342)
8π
which is recognized as the usual expression for the energy density of a free
electromagnetic field. On substituting eqn(339) into the second term in the
third line, one finds
1
+ ( ∇A0 ) . E (343)
4π
which can be expressed as
1 1 1
( ∇A0 ) . E = ∇ . ( A0 E ) − A0 ( ∇ . E ) (344)
4π 4π 4π
This relation has been used in arriving at the fourth line of eqn(338). Since the
divergence of the electric field satisfies Gauss’s law

∇.E = 4πρ (345)

57
the expression given in eqn(344) simplifies to
1 1
( ∇A0 ) . E = ∇ . ( A0 E ) − A0 ρ (346)
4π 4π
Therefore, the Hamiltonian density can be expressed as
1 1 µ
H = + ( E 2 + B 2 ) − ρ A0 + j Aµ
8π c
1
+ ∇ . ( A(0) E ) (347)
4π
On combining the term ρ A0 with the last term
1 µ 1
j Aµ = ρ A0 − j.A (348)
c c
which originates from the Lagrangian interaction (−Lint ), one finds that the
terms proportional to A0 ρ in the Hamiltonian density cancel. On neglecting
the total derivative term [ + 41π ∇ . ( φ E ) ], one finds that the Hamiltonian
density reduces to
1 1
H = ( E2 + B2 ) − j.A (349)
8π c
The first term is the energy density of the free electromagnetic field and the
second term represents the energy of the interaction between the electromag-
netic field and “charged particles”. It should be noted that the interaction
Hamiltonian is expressed entirely in terms of an interaction between the current
density and the vector potential, which demonstrates that the Hamiltonian is
not invariant under a Lorentz transformation
1
Hint = − j.A (350)
c
but is invariant under rotations in space. This situation is to be contrasted
with the interaction term in the Lagrangian which was Lorentz invariant as it
explicitly included an interaction between the scalar potential and the charge
density.

7.1 Conservation Laws for Electromagnetic Fields

The Lagrangian density L of an electromagnetic field is given by the Lorentz
scalar
1 1 µ
L = − F µ,ν Fµ,ν − j Aµ (351)
16 π c
or
      
1 1
L = − ∂ µ Aν − ∂ ν Aµ ∂µ Aν − ∂ν Aµ − jµ Aµ
16 π c
(352)

58
The Noetherian energy-momentum tensor T ν,µ is found from
  
∂L
T νµ = ∂µ Aρ
− δν µ L
∂(∂ν Aρ )
  
∂L
= ∂µ Aρ − δ ν µ L (353)
∂(∂ν Aρ )
The derivative of the Lagrangian density is evaluated as
   
∂L 1
= − F ν,ρ − F ρ,ν
∂(∂ν Aρ ) 8π
1
= − F ν,ρ (354)
4π
Therefore, the energy-momentum density is found as
 
ν 1 ν,ρ
T µ = − F ∂µ Aρ − δ ν µ L (355)
4π
On raising the index µ with the metric tensor, one has the contravariant second-
rank tensor
 
ν,µ 1 ν,ρ
T = − F ∂ Aρ − g ν,µ L
µ
(356)
4π
The energy-momentum tensor is not gauge invariant, as it explicitly involves
the fields Aµ . On using the expression for the source-free Lagrangian density
 
1
L = E2 − B2 (357)
8π
one finds that the time components of T µ,ν are given by
   
0,0 1 2 2 1
T = E + B + ∇. φE (358)
8π 4π
The expression T 0,0 is the Hamiltonian density H, in the absence of sources,
which represents the energy density of the free field. The momentum density is
given by the mixed time and space components, and is given by
1
T 0,j = − F 0,ρ ( ∂ (j) Aρ ) (359)
4π
but since F µ,ν is antisymmetric, only the terms where ρ is a spatial index are
non-zero. Hence, one has
1 X 0,i
T 0,j = − F ( ∂ (j) Ai )
4π i
 
1 X 0,i (j) (i) (i) (j) 1 X 0,i
= + F ∂ A − ∂ A + F ( ∂ (i) A(j) )
4π i 4π i
(360)

59
where the relation between the space-like components of the covariant and con-
travariant four-vector Ai = − A(i) has been used. Since the time component
of the field tensor is given by

F 0,i = − E (i) (361)

and10   X
(i) (j) (j) (i)
∂ A − ∂ A = − ξ i,j,k B (k) (362)
k

one finds that the momentum density is given by

1 X i,j,k (i) (k) 1 X (i)
T 0,j = − ξ E B − E ( ∂ (i) A(j) )
4π 4π i
i,k
 (j)
1 1
= E ∧ B + E . ( ∇ A(j) ) (363)
4π 4π
On noting that in the absence of sources, one has

∇.E = 0 (364)

and by adding a term proportional to A(j) ( ∇ . E ) to the expression for T 0,j

in eqn(363), one arrives at the result
 (j)  
0,j 1 1 (j)
T = E ∧ B + ∇. A E (365)
4π 4π

The components T 0,ν , apart from the terms involving total derivatives which
integrate out to zero, are related to the total energy and the components of the
total momentum of the electromagnetic field. The components of T µ,ν satisfy
the continuity equations
∂µ T µ,ν = 0 (366)
which represent the conservation of energy and momentum. The other mixed
time and spatial components of the energy-momentum tensor are evaluated as
 (j)    (j)  
1 1 1 ∂
T j,0 = E ∧B + ∇∧ φB − φ E (j)
4π 4π c ∂t
(367)
The components T j,0 represent the components of the energy flux.

It should be noted that the energy-momentum tensor T µ,ν is not symmet-

ric. This has the consequence that the covariant generalization of the angular
momentum to the third-rank tensor

M µ,ν,ρ = T µ,ν xρ − T µ,ρ xν (368)

10 Since the vector relationship B = ∇ ∧ A involves the covariant derivative, there is a

negative sign in the analogous expression involving the contravariant derivative.

60
is not conserved as the energy-momentum tensor is not symmetric. Additional
terms can be added to the energy-momentum tensor11 , to create a symmetric
tensor Θµ,ν . These extra terms account for the intrinsic angular momentum of
the photon.

The symmetric energy-momentum tensor Θµ,ν can be found by substituting

(∂ ν Aλ ) = − F λ,ν + (∂ λ Aν ) (369)

into the expression for T µ,ν , to yield

 
µ,ν 1 µ,ρ λ,ν 1 µ,ν ρ,λ 1 µ,ρ
T = g Fρ,λ F + g Fρ,λ F − g Fρ,λ (∂ λ Aν )
4π 4 4π
(370)
The first two terms are symmetric w.r.t. µ and ν and are gauge invariant. These
two terms will form the basis for Θµ,ν , which will be expressed as
 
µ,ν 1 µ,ρ λ,ν 1 µ,ν ρ,λ
Θ = g Fρ,λ F + g Fρ,λ F (371)
4π 4

The expression Θµ,ν is symmetric under the interchange of µ and ν, as can be

seen by writing
 
1 1 µ,ν
Θµ,ν = F µ λ F λ,ν + g Fρ,λ F ρ,λ
4π 4
 
1 1
= F µ,λ Fλ ν + g µ,ν Fρ,λ F ρ,λ (372)
4π 4

If Θµ,ν and T µ,ν are to represent the same set of conserved quantities, the last
term in eqn(370) must be expressible as a total derivative. That this is true can
be seen by examining the asymmetric term
1 µ,ρ 1
− g Fρ,λ (∂ λ Aν ) = − F µ,λ (∂λ Aν ) (373)
4π 4π
where the index ρ was raised by using the metric tensor. On combining the
above expression with the source free Maxwell equation

(∂λ F µ,λ ) = 0 (374)

11 J. Belinfante, Physica 6, 887 (1939) has shown that the modified tensor Θµ,ν defined by
 
Θµ,ν = T µ,ν + ∂ρ Λρ;µ,ν

where Λρ;µ,ν is an arbitrary tensor that is antisymmetric under the interchange of the first
pair of indices
Λρ;µ,ν = − Λµ;ρ,ν
will automatically satisfy the same continuity conditions as T µ,ν and leave the total energy
and momentum unaltered.

61
one obtains
 
1 µ,ρ 1
− g Fρ,λ (∂ λ Aν ) = − F µ,λ ν ν µ,λ
(∂λ A ) + A (∂λ F )
4π 4π
 
1 µ,λ ν
= − ∂λ F A (375)
4π

which is a total covariant derivative of a third-rank tensor which is antisymmet-

ric under the interchange of µ and λ. Furthermore, adding this term does to the
energy-momentum tensor does not change the energy-momentum conservation
law. This can be seen by observing that the difference between the two forms
of energy-momentum conservation law involves the double derivative
 
1
∂µ ( Θµ,ν − T µ,ν ) = − ∂µ ∂λ F λ,µ Aν (376)
4π

and F λ,µ is antisymmetric. On interchanging the order of the derivatives in the

right hand side, switching the summation labels, and using the antisymmetric
property of F λ,µ , one has
 
1
∂µ ( Θµ,ν − T µ,ν ) = − ∂µ ∂λ F λ,µ Aν
4π
 
1
= − ∂λ ∂µ F λ,µ Aν
4π
 
1
= − ∂µ ∂λ F µ,λ Aν
4π
 
1
= + ∂µ ∂λ F λ,µ Aν (377)
4π

On comparing the right hand sides of the first and last line, one finds that they
have opposite signs and, therefore, they are zero. Thus, the difference between
continuity relations vanish

∂µ ( Θµ,ν − T µ,ν ) = 0 (378)

Hence, since T µ,ν is conserved, then the symmetrized energy-momentum tensor

Θµ,ν is also conserved.

Thus, the symmetric energy-momentum tensor Θµ,ν expressed by

 
µ,ν 1 µ,ρ λ,ν 1 µ,ν ρ,λ
Θ = g Fρ,λ F + g Fρ,λ F (379)
4π 4

is a conserved quantity. The purely temporal component is given by

 
1
Θ0,0 = E2 + B2 (380)
8π

62
and the mixed temporal and spatial components are given by
 (j)
0,j 1
Θ = E ∧ B (381)
4π

The temporal and spatial components of Θ0,µ are, respectively, recognized as

being the energy-density and the momentum-density vector of the field. The
components Θj,0 are recognized as forming the Poynting vector which represents
the energy flux of the electromagnetic field. The spatial components are given
by
 
i,j 1 (i) (j) (i) (j) 1 i,j 2 2
Θ = − E E + B B − δ (E + B ) (382)
4π 2

Noether’s theorem is purely classical, but there are generalizations for quan-
tum fields. Quantum generalizations includes the Ward-Takahashi and Taylor-
Slavnov identities.

Exercise:

Evaluate the components T j,0 and T i,j of the (asymmetric) energy-momentum

tensor for a source-free electromagnetic field.

Exercise:

Show that in the presence of sources, the symmetric energy-momentum ten-

sor has components with the form
 
0,0 1 2 2 1
Θ = E + B − j.A
8π c
 (j)
1
Θ0,j = E ∧ B − ρ A(j) (383)
4π

Verify the form of the conservation laws for energy and momentum.

Exercise:

Show that the extra term added to the tensor T i,j in order that Θi,j will
be symmetric produces a contribution to the angular momentum density of the
form  (j)
0,j 1
S = E ∧ A (384)
4π
which is the intrinsic spin density of the electromagnetic field.

63
7.2 Massive Spin-One Particles
The electromagnetic theory has been unified with the theory of weak interac-
tions. This generalization requires the existence of two new types of spin-one
particles in addition to the photon, which together mediate the electro-weak
interaction. These new particles have non-zero mass. The massive spin-one
particle has to satisfy the equation12

pµ pµ = m2 c2 (385)

and with the quantization condition,

∂
pµ → p̂µ = i h̄ (386)
∂xµ

the four-vector field Aµ must satisfy the Klein-Gordon equation

2 
1 ∂2
 
mc 4π µ
− ∇2 + Aµ = j (387)
c2 ∂t2 h̄ c

where h̄ no longer drops out. This equation can be derived from the Lagrangian
 2
1 1 mc 1 µ
L = − F µ,ν Fµ,ν + Aµ Aµ − j Aµ (388)
16 π 8π h̄ c

For example, on varying Aµ , one obtains the equation of motion

 2
ν,µ mc 4π µ
∂ν F + Aµ = j (389)
h̄ c

Neither the Lagrangian, nor the equation of motion are gauge invariant. The ap-
propriate gauge condition can be enforced by imposing conservation of charge13

∂µ j µ = 0 (390)

On taking the four-divergence of the equation of motion, one finds

 2
ν,µ mc 4π
∂µ ∂ν F + ∂µ Aµ = ∂µ j µ (391)
h̄ c

The first term on the left-hand side vanishes due to the definition of F µ,ν , since

F ν,µ = ∂ ν Aµ − ∂ µ Aν (392)

one finds
∂ν F ν,µ = ∂ν ∂ ν Aµ − ∂ µ ∂ν Aν (393)
12 A. Proca, J. Phys. et Radium 7, 147 (1936).
13 Note that, unlike the massless photon, charge conservation has to be imposed as an
additional assumption.

64
therefore

∂µ ∂ν F ν,µ = ∂ν ∂ ν ∂µ Aµ − ∂µ ∂ µ ∂ν Aν
= 0 (394)

The term on the right-hand side of eqn(391) also vanishes, because it was chosen
to impose charge conservation. Hence, one finds that Aµ for a massive spin-one
particle must satisfy the Lorenz gauge condition

∂µ Aµ = 0 (395)

Exercise:

Starting from the expression eqn(379), determine the symmetrized energy-

momentum tensor for the massive vector field. Hence, find the energy and
momentum densities.

7.3 Polarizations of Massive Spin-One Particles

The four-vector potential with contravariant components Aµ (x) can be Fourier
transformed as
1 X  
µ µ ν
A (x) = √ A (k) exp[ i k xν ] + c.c. (396)
2V k

which results in four components associated with four polarization vectors which
are denoted by êµ (k). For massive photons, if one assumes charge conservation,
the gauge fields must satisfy the Lorentz gauge condition. The Lorentz gauge
condition
∂µ Aµ (x) = 0 (397)
results in the Fourier components satisfying the condition

kµ Aµ (k) = 0 (398)

which involving the photon’s four-momentum. Apparently, the gauge condition

reduces the number of independent components of the four-vector potential from
four to three. However, we are used to thinking that massless photons only have
two independent polarizations. In the following, we shall see how this comes
about.

For a photon with mass m, one can apply the Lorentz gauge condition in
the laboratory frame. We shall choose the z-axis as the direction of the pho-
ton’s three-momentum k. With this choice, the photon’s four-momentum has

65
 q
2
components (k (0) , 0, 0, k (3) ) where k (0) = ( mc 2
h̄ ) + k . The Lorentz condition
becomes
k (0) A(0) (k) − k (3) A(3) (k) = 0 (399)
(3)
Hence, A(0) (k) = kk(0) A(3) (k). This implies that in the photon’s rest frame, the
four-vector potential only has three spatial components and the scalar potential
is zero. In any case, there are only three-independent components of the four-
vector potential. The four-vector potential can be expressed in terms of the
three independent components as

1 X  k (3)
A(x) = √ A(3) (k) ( (0) ê(0) (k) + ê(3) (k) )
2V k k
 
(1) (1) (2) (2) ν
+ A (k) ê (k) + A (k) ê (k) exp[ i k xν ] + c.c.

(400)

Thus, the four-vector associated with the (longitudinal) contravariant compo-

(3)
nent A(3) (k) has a polarization which is proportional to kk(0) ê(0) (k) + ê(3) (k).
The vector with components A(3) (k) is physical since it lead to an electromag-
netic field. The electric and magnetic fields are defined by
1 ∂
E = − A − ∇φ
c ∂t
B = ∇ ∧ A (401)

The longitudinal component of the electric field has a Fourier amplitude given
by
 
(3) (0) (3) (3) (0)
E (k) = − i k A (k) − k A (k)

k (3)2
 
= −i k (0) − A(3) (k)
k (0)
 2
m c
h̄
= −i A(3) (k) (402)
k (0)
and the longitudinal component of the magnetic field B (3) (k) is zero since
 
(3) (3)
0 = − i k ê ∧ A(k) . ê(3) (k) (403)

Thus, the component A(3) (k) contributes to the E field but not the B field.
For massive photons, the unsymmmetrized energy-momentum tensor T µ,ν is
expressed in terms of contributions from, not only the E and B fields, but also

66
from the gauge field Aµ . The tensor is evaluated as
2
g µ,ν
j ρ Aρ
  
µ,ν 1 µ,ν ρ,σ µ,ρ ν mc
T = g Fρ,σ F −4F ∂ Aρ − Aρ Aρ + g µ,ν
16 π 8π c h̄
(404)
The energy-momentum tensor can be “symmetrized” by adding and subtracting
a term
4
+ F µ,ρ ∂ρ Aν (405)
16 π
to express the second term in a gauge-invariant form. We then note that that
remaining term (with the negative sign) could be put into the form of a diver-
gence of an antisymmetric third rank-tensor, if we could combine it with a term
of the form
1 1
− ( ∂ρ F µ,ρ ) Aν = + ( ∂ρ F ρ,µ ) Aν (406)
4π 4π
This can be accomplished by adding zero in the form of Aν times the Euler-
Lagrange equation
 2
mc 4π µ
+ ∂ρ F ρ,µ + Aµ − j = 0 (407)
h̄ c

therefore, completing the divergence at the expense of adding extra mass and
source terms. The divergence of third-rank antisymmetric tensor can be dropped,
leading to the “symmetrized” energy-momentum tensor given by
 
µ,ν 1 µ,ν ρ,σ µ,ρ ν
Θ = g Fρ,σ F + 4F Fρ
16 π
2 2
g µ,ν m c
 
1 mc
− Aρ Aρ + Aµ Aν
8π h̄ 4π h̄
j ρ Aρ j µ Aν
+ g µ,ν − (408)
c c
It is seen that the coupling of the electromagnetic fields to the current densities
spoils the symmetry of the energy-momentum tensor. Hence, the charge cur-
rents act as sources of angular momentum and results in the electromagnetic
field’s angular momentum not being conserved. The energy-density H is given
by Θ0,0 so
2 
j.A
   
1 2 2 1 mc (0)2 2
H = E + B + A + A − (409)
8π 8π h̄ c

and the momentum density P is given by

2
j (0) A
  
1 1 mc
P = E ∧ B + A(0) A − (410)
4πc 4πc h̄ c2

67
Thus, the longitudinal photon with components A(3) (k) and A(0) (k) does have
physical effects, as do the two transverse photons A(1) (k) and A(2) (k). There-
fore, the four-vector potential of a massive electromagnetic field has three phys-
ical components.

Exercise:

Determine the contribution to the energy and momentum of the electromag-

netic field which originates from the coupled time and space-like (longitudinal)
components of the four-vector potential.

For a massless photon m = 0, so one sees that the longitudinal component

of the electric field E (3) (k) is zero. Also since for m = 0 one has k (0) = k (3) ,
so the Lorentz gauge condition requires that the component A(3) (k) is equal to
A(0) (k) and, therefore, the longitudinal component is associated with a single
polarization vector ( ê(0) (k) + ê(3) (k) ). Since A(0) (k) = A(3) (k), the norm of
the vector (A(0) (k), 0, 0, A(3) (k)) is identical zero. For m = 0, the longitudinal
component A(3) (k) neither contributes to the E field nor to the B field. Since
the energy and momentum densities are expressed in terms of the E and B
fields via  
1 2 2
H = E + B (411)
8π
and  
1
P = E ∧ B (412)
4πc
the component A(3) (k) or equivalently A(0) (k) has no physical effect. Hence,
it can be deemed unphysical. Therefore, the four-vector potential of a mass-
less electromagnetic field only has two physical components A(1) (k) and A(2) (k)
which are the components transverse to the direction of propagation.

7.4 The Propagator for Massive Photons.

The four-vector potential can be expressed entirely in terms of the currents,
via the introduction of a photon propagator. The equation of motion for the
massive spin-one particle is
 2
µ,ν mc 4π ν
∂µ F + Aν = j (413)
h̄ c

On Fourier transforming the linear equation with respect to space and time, one
obtains
  2 
mc 4π ν
− kµ k µ + Aν (k) + kµ k ν Aµ (k) = j (k) (414)
h̄ c

68
which can be re-written as
  2  
ρ mc ν ν 4π ν
− kρ k + δµ + kµ k Aµ (k) = j (k) (415)
h̄ c
The photon propagator is defined via

Aµ (k) = Dµ,λ (k) jλ (k) (416)

which leads to the phonon propagator being determined as the solution of a

matrix equation
  2  
mc 4π ν
− kρ k ρ + g ν,µ + k ν k µ Dµ,λ (k) = δ (417)
h̄ c λ
The above matrix equation can be solved by first writing the propagator as

Dµ,ν (k) = g µ,ν A(k 2 ) + k µ k ν B(k 2 ) (418)

and then solving for the unknown quantities B(k 2 ) and A(k 2 ). This leads to
the following expression for the photon propagator
 
µ ν
  g µ,ν − (km ck)2
4π h̄
Dµ,ν (k) = 2 (419)
c
 
m c ρ k
h̄ − k ρ

Thus, the Fourier component of the four-vector potential is given by

 
µ,ν kµ kν
g − ( m c )2  
h̄ 4π
Aµ (k) =   2 jν (k) (420)
c

m c ρ
h̄ − k kρ

Hence, Aµ (x) can be found from the inverse Fourier transform. It should be
noted that, if one imposes current conservation

k ν jν (k) = 0 (421)

the second term in the numerator of the photon propagator has no physical
effect. We also note that
  2
4π h̄
kµ Aµ (k) = k ν jν (k) (422)
c mc
So, once again we see that assumption of charge conservation enforces the
Lorentz gauge condition, even if the massive photon is involved in a virtual
process.

69
8 Symmetry Breaking and Mass Generation
We shall first look at an example of Goldstone’s theorem which states that,
if a system described by a Lagrangian which has a continuous symmetry (and
only short-ranged interactions) has a broken symmetry state then the system
supports a branch of small amplitude excitations with a dispersion relation ωk
that vanishes at k = 0. We shall then examine the situation in which the system
is coupled by long-ranged interactions, as modelled by an electromagnetic field.
As was first pointed out by Anderson, the long-ranged interactions alter the
excitation spectrum of the symmetry broken state by removing the Goldstone
modes and generating a branch of massive excitations.

8.1 Symmetry Breaking and Goldstone Bosons

Consider a Lagrangian density for a complex scalar field of the form
 2  2
mc
L = ( ∂µ ψ ∗ ) ( ∂ µ ψ ) − ψ ∗ ψ − φ20 (423)
2 h̄ φ0

The Lagrangian density is invariant under the continuous global symmetry

 
ψ → ψ 0 = exp − i α ψ (424)

for any real constant α. The static or minimum energy solution corresponds to

| ψ | = φ0 (425)

which leaves the phase of ψ undetermined. Since the phase of ψ is continuous,

the ground state is infinitely degenerate. If one writes

ψ = φ1 + i φ2
ψ∗ = φ1 − i φ2 (426)

then the Lagrangian can be written as a Lagrangian density involving the two
real scalar fields φ1 and φ2 . The Lagrangian density has a U (1) symmetry which
corresponds to the rotation of ψ around a circle about the origin in the (φ1 , φ2 )
plane.

We shall assume the field ψ representing the physical ground state corre-
sponds to only one of the infinite number of possible candidates. The physical
state must have a phase, which shall be defined as zero. That is, one starts with
a ground state ψ = φ0 , and then consider the small amplitude excitations. A
low-energy excited state corresponds to the complex field

ψ = φ0 + δψ (427)

70
 v(Ψ)
Re [Ψ]

Im [Ψ] φ0

Figure 8: The potential V [ψ] described by the Lagrangian is invariant under

global rotations of the phase of ψ. The minima occurs at values of ψ which have
magnitudes φ0 , therefore, the uniform static field is infinitely degenerate.

where δψ is static and uniform and can be considered to be very small. The
small amplitude complex field δψ can be expressed in terms of its real and
imaginary parts
δψ = χ1 + i χ2 (428)
The Lagrangian density takes the form
 2  2
µ µ mc
L = ( ∂µ χ1 ) ( ∂ χ1 ) + ( ∂µ χ2 ) ( ∂ χ2 ) − 2 φ0 χ1 + χ21 + χ22
2 h̄ φ0
(429)
If one only considers infinitesimally small amplitude oscillations, one only needs
consider terms quadratic in the fields. The quadratic Lagrangian density LF ree
describes non-interacting fields. The quadratic Lagrangian density is given by
 2
mc
LF ree = ( ∂µ χ1 ) ( ∂ µ χ1 ) − χ21 + ( ∂µ χ2 ) ( ∂ µ χ2 ) (430)
h̄
The symmetry breaking has resulted in the complex field breaking up into two
fields: The first field χ1 describes massive excitations m and the second field χ2
describes massless excitations. The first field χ1 has plane-wave solutions if the
energy and momentum are related via the dispersion relation
2
m c2

ω 2 = c2 k 2 + (431)
h̄
and represents excitations which corresponds to a “stretching” of φ0 . It is
massive since this excitation moves the field away from the minimum of the
potential. The second excitation χ2 represents δψ which is transverse to φ0 in

71
the (φ1 , φ2 ) plane. This last excitation is known as a Goldstone boson14 . The
Goldstone boson has a dispersion relation

ω 2 = c2 k 2 (432)

which vanishes at k = 0. The Goldstone boson dynamically restores the sponta-

neously broken U (1) symmetry since, at k = 0, it just corresponds to a change
of the value of the (static and uniform) broken symmetry field from (φ0 , 0)
to the new direction (φ0 , χ2 ). Therefore, if infinitely many zero-energy Gold-
stone bosons are excited in the system, the resulting state should correspond
to a new ground state with a different value of the phase. As noted by An-
derson15 prior to Goldstone’s work, the Goldstone theorem breaks down when
long-ranged interactions are present. Anderson was responsible for the concept
of mass generation through symmetry breaking due to the coupling with gauge
fields16 . This concept was subsequently developed by Peter Higgs17 and Tom
Kibble and coworkers18 .

8.2 The Kibble-Higgs Mechanism

We shall now consider the coupling of a scalar field ψ with charge q to a gauge
field Aµ . The Lagrangian density is related to the sum of the Lagrangian density
for the electromagnetic field and the Lagrangian density for the charged scalar
particle. The coupling between the fields is found from the minimum coupling
assumption
q µ
p̂µ → p̂µ0 = p̂µ − A (433)
c
which becomes
q µ
i h̄ ∂ µ → i h̄ ∂ µ − A (434)
c
Therefore, the Lagrangian density for the coupled fields has the form
q q 1
L = ( ∂µ − i Aµ ) ψ ∗ ( ∂ µ + i Aµ ) ψ − F µ,ν Fµ,ν
h̄ c h̄ c 16 π
 2  2
mc ∗ 2
− ψ ψ − φ0 (435)
2 h̄ φ0
14 J. Goldstone, Il Nuovo Cimento, 19, 154 (1961).
15 P. W. Anderson, Phys. Rev., 112 1900 (1958).
16 P.W. Anderson, “Plasmons, Gauge Invariance, and Mass”. Physical Review 130, 439

(1963).
17 P.W. Higgs, “Broken Symmetries, Massless Particles and Gauge Fields”, Physics Letters,

12, 132 (1964): P.W. Higgs, “Broken Symmetries and the Masses of Gauge Bosons”. Physical
Review Letters 13, 508 (1964):P.W. Higgs, “Spontaneous Symmetry Breaking without mass-
less Bosons”, Physical Review 145, 1156 (1966)

18 G.S. Guralnik, C.R. Hagen, and T.W.B. Kibble, “Global Conservation Laws and Massless

Particles”. Physical Review Letters 13, 585-587 (1964): T.W.B. Kibble, “Symmetry Breaking
in non-Abelian Gauge Theories”, Physical Review, 155, 1554 (1967).

72
The Lagrangian density is invariant under the local gauge transformation19
 
0 q
ψ → ψ = exp − i Λ ψ
h̄ c
Aµ → Aµ0 = Aµ + ∂ µ Λ (436)

where Λ = Λ(xµ ) is a function of space-time. The system has minimum energy

when ψ has a constant value with a magnitude given by

| ψ | = φ0 (437)

and the Aµ vanish. Any local gauge transformation leads to a state with the
same energy, therefore, the ground state is infinitely degenerate.

We shall assume that a physical system spontaneously breaks the symmetry

in that it corresponds to a specific constant value of Λ. We shall choose the
local gauge Λ(x) such that the field ψ representing the excited states is purely
real. However, once the gauge has been fixed, no further gauge transformations
can be made.

The small amplitude excitations can be expressed as

ψ = φ0 + δψ (438)

The fluctuations can be expressed as

δψ = χ1 (439)

and on substituting in the Lagrangian and collecting the quadratic terms, one
obtains
 2
mc
LF ree = ( ∂µ χ1 ) ( ∂ µ χ1 ) − χ21
h̄
 2
1 q φ0
− F µ,ν Fµ,ν + Aµ Aµ (440)
16 π h̄ c

Therefore, one finds that the charged boson field has a mass m and the gauge
field has acquired a mass mA given by
 2
q φ0
m2A = 8π (441)
c2

Hence, by coupling an electromagnetic field with two components to a scalar

charged boson field, one has found a massive vector boson gauge-field with three
independent components. The massless spin-less component of the charged bo-
son field which described the Goldstone mode has become the longitudinal mode
19 V. Fock, Z. Phys., 39, 226, (1927).

73
of the gauge field. More specifically, the field χ2 which initially corresponded
to the Goldstone mode became unphysical when the massless vector field was
introduced as it could be gauged away. This is seen by writing
   
χ2
ψ = φ0 + χ1 + i χ2 ≈ φ0 + χ1 exp i (442)
φ0

so χ2 could be removed by a gauge transformation involving the particular

choice of Λ
χ2
Λ =   (443)
q φ0
h̄ c

9 Gravitational Interactions
We introduce the field theory of the Gravitational Interaction, in the weak field
limit, by analogy with electromagnetism. More specifically, we shall develop the
classical field theory of the massive graviton in parallel with the Proca’s theory
of a massive spin-one particle that we discussed previously. This approach has
the disadvantage that it does not have the same beautiful geometric basis as
Einstein’s field equations.

Electromagnetic fields are mediated by spin-one bosons and gravitational

fields are mediated by spin-two bosons. Electromagnetic forces between two-
particles with like charges is repulsive but Gravitational forces are always at-
tractive. These facts are quantified by Coulomb’s law
q1 q2
V (r1 − r2 ) = (444)
| r1 − r2 |

and Newton’s Force law

G m1 m2
V (r1 − r2 ) = − (445)
| r1 − r2 |

where G is the gravitational constant. The value of G was first determined as

G = 6.754 × 10−11 m3 sec−2 kgm−1 by Henry Cavendish in 1798. However,
G remains the most poorly known physical constant. Recent high precision
measurements of G yield values with error bars that are mutually exclusive.
The high precision values differ in the third significant digit. The above two
potentials show that the electric and gravitational fields have the same type
of classical Green’s functions associated with massless bosons. Therefore, we
expect that the difference in the signs of the interaction is caused by the spin
values of the bosons involved in the exchanges. We shall examine the difference
of the signs of the electromagnetic and gravitational forces by examining the

74
quantum propagators.

Consider (massive) electromagnetic radiation in the presence of a source.

The Lagrangian density formulated by Proca20 is given by
 2
1 1 mc 1 µ
L = − F µ,ν Fµ,ν + Aµ Aµ − A jµ (446)
16 π 8π h̄ c

where
F µ,ν = ∂ µ Aν − ∂ ν Aµ (447)
The mass term spoils the gauge invariance of the source-free Lagrangian. The
action S is given by Z
S = d4 x L (448)

which, on integrating by parts, can be expressed as

Z    2  
4 1 ρ mc µ,ν 1
S = d x Aµ ∂ρ ∂ + g − ∂ ∂ − jν
µ ν
Aν (449)
8π h̄ c

The equation of motion for the four-vector potential is

  2  
mc 4π µ
∂ρ ∂ ρ + g µ,ν − ∂ µ ∂ ν Aν = j (450)
h̄ c

If charge is conserved
∂µ j µ = 0 (451)
On operating on the equation of motion with ∂µ , one finds that because of the
finite mass the field Aν must satisfy the condition

∂ ν Aν = 0 (452)

Thus, conservation of charge removes the choice of gauge. In the rest frame
of a photon, k = ( mh̄ c )(1, 0, 0, 0), so one has A0 (k) = 0 and so one finds that
the massive photon has three components. This is expected for a particle with
spin S = 1. The electromagnetic propagator can be defined as the kernel of the
integral relation
Z
4π
Aν (x) = d4 x0 Dν,λ (x, x0 ) j λ (x0 ) (453)
c
Thus, the propagator satisfies the partial differential equation
  2  
mc
∂ρ ∂ ρ
+ g µ,ν
− ∂ ∂µ ν
Dν,λ (x, x0 ) = δλµ δ 4 (x − x0 ) (454)
h̄
20 A. Proca, C.R. Acad. Sci., Paris 203, 709-711 (1936), J. de Physique et Radium, 8, 23-28

(1937).

75
which can be solved by Fourier transforming. On performing the Fourier trans-
formation, one finds the algebraic equation
  2  
mc
− kρ k ρ + g µ,ν + k µ k ν Dν,λ (k) = δλµ (455)
h̄
which can be solved and has the solution
− g kν k λ 
ν,λ + ( m c )2
h̄
Dν,λ (k) = (456)
k ρ kρ − ( mh̄ c )2

We shall determine the interaction mediated by the photon, by considering

the contribution to the energy density from two charges. For a static charge
density, the interaction energy can be expressed as minus half the interaction
Lagrangian. For static charges, the kinetic energy terms of the free-field La-
grangian do not contribute to the energy. Furthermore, since the static field
free Lagrangian is quadratic in the field, the equation of motion can be used to
show that it cancels half the interaction term which is linear in the field. The
interaction Lagrangian can be written as
Z
1
Lint = − d3 r jµ Aµ (457)
c
which, for t = 0, becomes
dk00 d4 k
Z Z
1
Lint = − j ∗ (k 0 , k) Aµ (k0 , k) (458)
c (2π) ( 2 π )4 µ 0
On eliminating the four-vector potential, the interaction portion of the La-
grangian can be expressed as
− g k ν kλ
dk00 d4 k ν,λ +
Z Z 
1 4π ( mh̄ c )2
Lint = − j ν (k00 , k)∗ j λ (k0 , k)
c (2π) ( 2 π )4 c k ρ kρ − ( mh̄ c )2
(459)
Using the condition of continuity of charge

kν j ν (k) = 0 (460)

the interaction can be written as

dk00 d4 k
 
− gν,λ
Z Z
1 ν 0 ∗ 4 π
Lint = − j (k0 , k) j λ (k0 , k)
c (2π) ( 2 π )4 c k ρ kρ − ( mh̄ c )2
dk00
 ν 0
d4 k j (k0 , k)∗ jν (k0 , k)
Z Z 
4π
= + 2 (461)
c (2π) ( 2 π )4 k ρ kρ − ( mh̄ c )2
For two static charges, for which

ρ(k) = 2 π δ(k0 ) ρ(k) (462)

76
one has
d3 k ρ(k)∗ ρ(k)
Z  
4π
Lint = (463)
c2 ( 2 π )3 − k 2 − ( mh̄ c )2
Thus, the interaction Lagrangian between two like charges is negative and this
corresponds to a positive (replusive) interaction potential V .

The massive photon corresponds to fluctuations of the four-vector potential

Aν , which has four components. If we assume the continuity of charge, then the
four-vector potential must satisfy the gauge condition

k ν Aν (k) = 0 (464)

Therefore, the massive photon has three independent components corresponding

to the three different S z eigenvalues of an S = 1 particle. Consider the form of
the photon propagator
− g kν k λ
ν,λ +

( mh̄ c )2
Dν,λ (k) = (465)
k ρ kρ − ( mh̄ c )2

The numerator originates from the product of polarization vectors α (k) for the
massive spin-one particle. The polarization vectors can can be chosen as

(1) = (0, 1, 0, 0)
(2) = (0, 0, 1, 0)
(3) = (0, 0, 0, 1) (466)

and for a photon at rest, one has

mc
k ν = (( ), 0, 0, 0) (467)
h̄
The polarization four-vectors and the energy-momentum four-vector satisfy the
gauge condition
k ν α
ν (k) = 0 (468)
which is Lorentz invariant. The propagator involves a polarization vector αν (k)
at the source where the photon is emitted. The photon then propagates at the
sink, where it is absorbed. The absorption involves the factor α
λ (k). Since the
polarization is not measured, it is summed over. Thus, the numerator of the
propagator is given by factor
X
α α
ν (k) λ (k) (469)
α

Since this quantity transforms as a Lorentz tensor, it can only be expressed in

terms of the metric gν,λ and the product kν kλ . The combination can be fixed,
up to a normalization, by noting that since one must use the gauge condition

k ν α
ν (k) = 0 (470)

77
the propagator must satisfy the condition

k ν Dν,λ (k) = 0 (471)

Thus, if the propagator has the form

 
A gν,λ + B kν kλ
Dν,λ (k) = (472)
k ρ kρ − ( mh̄ c )2

the gauge condition requires that

A kλ + B k ν kν kλ = 0 (473)

or  2
mc
A + B = 0 (474)
h̄
Therefore, the numerator of the propagator can be expressed as
 
kν kλ
A gν,λ − m c 2 (475)
( h̄ )

The constant of proportionality A can be fixed by considering the massive pho-

ton in its rest frame, and choosing one spatial index for ν and λ in the sum
over polarization components (eg. ν = λ = 1). This fixes A = −1. Thus, the
numerator of the photon propagator is given by
 
kν kλ
Gν,λ (k) = − gν,λ + m c 2 (476)
( h̄ )

We have shown that the photon propagator is given by

 
Gν,λ (k)
Dν,λ (k) = (477)
k ρ kρ − ( mh̄ c )2

9.1 Mathematical Structure of General Relativity

The fully non-linear Einstein field equation is given by
8πG
Gµ,ν + Λ gµ,ν = Tµ,ν (478)
c3
where Λ is the cosmological constant, Gµ,ν is the Einstein tensor and Tµ,ν is the
energy-momentum tensor. The Einstein tensor is given in terms of the Ricci
tensor Rµ,ν via
1
Gµ,ν = Rµ,ν − gµ,ν R (479)
2

78
where the Ricci scalar R is defined as

R = g µ,ν Rµ,ν (480)

The Ricci tensor is obtained from the Riemann tensor via the contraction
λ
Rµ,ν = Rµ,λ,ν (481)

and the Riemann tensor is given in terms of the Christoffel symbols via
ρ
Rσ,µ,ν = ∂µ Γρν,σ − ∂ν Γρµ,σ + Γρµ,λ Γλν,σ − Γρν,λ Γλµ,σ (482)

where, in turn, the Christoffel symbols Γσρ,µ are given in terms of the metric
tensors by  
1
Γσρ,µ = g σ,ρ ∂µ gν,ρ + ∂ν gρ,µ − ∂ρ gµ,ν (483)
2
In the absence of a source, the Lagrangian density of the source-free gravitational
field is given by √
L = −g R (484)
where g = det gµ,ν is negative.

9.2 Linearized Gravity

The graviton is supposed to correspond to a particle with spin S = 2. The field
of the graviton corresponds to the fluctuating part of the metric tensor hµ,ν .
Therefore, hµ,ν is the linearized part of a symmetric second-rank tensor, where
the lowest-order piece corresponds to the metric tensor ηµν for flat Minkowski
space. Thus
gµ,ν = ηµ,ν + hµ,ν (485)
where
hµ,ν  1 (486)
Since the metric tensor is symmetric, it has 10 independent components. How-
ever, under a general coordinate transformation

xµ → x0µ = xµ − Λµ (487)

the full metric changes according to

0 ∂xσ ∂xτ
gµ,ν → gµ,ν = gσ,τ (488)
∂x0 µ ∂x0 ν
so the linearized metric transforms as

hµ,ν → h0µ,ν = hµ,ν + ∂µ Λν + ∂ν Λµ (489)

79
Thus, both metrics describe the same physical gravitational field and so gravi-
tational theory has a gauge invariance. The gravitational gauge transformation
is very similar to the electromagnetic gauge transformation

Aµ → A0µ = Aµ + ∂µ Λ (490)

We can use this similarity to motivate the expectation that a graviton has S = 2.
On introducing a mass for the photon, we found that charge is not automatically
conserved. Furthermore, we found that if one enforces charge conservation, then
the electromagnetic field must satisfy the gauge condition

∂ µ Aµ = 0 (491)

Therefore, the four-vector potential has only three independent components as

expected for a massive spin-one particle. Likewise, on giving the graviton a
mass, one expects that the massive graviton’s field should satisfy four different
Lorentz gauge conditions
∂ µ hµ,ν = 0 (492)
If hµ,ν is to represent a particle with S = 2, we anticipate that it must also be
traceless
η µ,ν hµ,ν = 0 (493)
That is, a massive S = 2 graviton’s field must have 5 independent components
which correspond to the five different eigenvalues of S z . Therefore, since the
field hµ,ν has 10 components, it must satisfy five conditions.

In the Lorentz gauge, the linearized Einstein equations reduce to the form
 
ρ 1 ρ σ,τ 16 π G
∂ρ ∂ hµ,ν − ηµ,ν ∂ρ ∂ η hσ,τ = − Tµ,ν (494)
2 c3

where the energy-momentum tensor Tµ,ν is assumed to be symmetric. On mul-

tiplying by η µ,ν and summing over the indices µ and ν, one finds
 
d 16 π G
(1 − ) ∂ρ ∂ ρ η σ,τ hσ,τ = − η µ,ν Tµ,ν (495)
2 c3

where d = 4. This equation can be used to eliminate the second term in the
linearized equation of motion, leading to
  
16 π G 1
∂ρ ∂ ρ hµ,ν = − T µ,ν − ηµ,ν η σ,τ
T σ,τ (496)
c3 2

The second term in the source projects out unwanted components of the energy-
momentum tensor. For the vacuum, where Tµ,ν = 0, the linearized equations
take the form
∂ρ ∂ ρ hµ,ν = 0 (497)

80
which is just the relativistic wave equation for a massless spin-two particle. For
a graviton of mass m, the wave equation could be expected to have the form
  2 
ρ mc
∂ ∂ρ + hµ,ν = 0 (498)
h̄

9.3 The Massive S = 2 Graviton

The source-free equations of motion
  2 
ρ mc
∂ ∂ρ + hµ,ν = 0 (499)
h̄

describing a massive S = 2 particle can be found by considering Fierz-Pauli

Lagrangian density for a symmetric second-rank tensor field21 . The Lagrangian
density is given by
1 1
L = + ∂λ hµ,ν ∂ λ hµ,ν − ∂λ ηµ,ν hµ,ν ∂ λ ησ,τ hσ,τ
2 2
− ∂µ hν,λ ∂ ν hµ,λ + ∂µ hµ,ν ∂ν η σ,τ hσ,τ
 2  
1 mc µ,ν µ,ν σ,τ
− hµ,ν h − η hµ,ν η hσ,τ (500)
2 h̄

up to a constant factor. The form of the mass term is not enforced by any known
symmetry. In fact, like the Proca Lagrangian for the massive photon, the mass
term in the Fierz-Pauli Lagrangian density violates the gauge symmetry. The
principle of extremal action leads to the equations of motion

0 = − ∂λ ∂ λ hµ,ν + η µ,ν ∂λ ∂ λ ησ,τ hσ,τ

+ ∂λ ∂ µ hλ,ν + ∂λ ∂ ν hλ,µ
− ∂ µ ∂ ν η σ,τ hσ,τ − η µ,ν ∂σ ∂τ hσ,τ
 2  
mc µ,ν µ,ν σ,τ
− h − η ησ,τ h (501)
h̄

On operating on the equation with ∂µ and on noting that only the mass term
remains, one finds the equation

∂ µ hµ,ν − ∂ν η σ,τ hσ,τ = 0 (502)

This is a mandatory “Lorentz gauge” condition for linearized (massive) gen-

eral relativity. Here this is not a choice since the condition is enforced by the
21 M. Fierz and W. Pauli, Proc. Roy. Soc. London, 187, 211-232 (1937).

81
existence of the mass. Substituting this mandatory condition back into the
equations of motion yields
 2  
ρ σ,τ mc σ,τ
∂ ∂ρ hµ,ν − ∂µ ∂ν η hσ,τ + hµ,ν − ηµ,ν η hσ,τ = 0 (503)
h̄

On taking the trace of the above equation, if m 6= 0, one recovers the condition
that hµ,ν must be traceless
η µ,ν hµ,ν = 0 (504)
as anticipated. Furthermore, this also implies the four mandatory “gauge con-
ditions”
∂ µ hµ,ν = 0 (505)
must be satisfied. On applying the above five constraints to the equation of
motion, one finds
  2 
mc
∂ ρ ∂ρ + hµ,ν = 0 (506)
h̄
Thus, the Fierz-Pauli Lagrangian for hµ,ν does indeed describe a massive particle
with S = 2, and we have confirmed the form of the five conditions on the field.
In the presence of a source, the equation could be expected to have the form
  2   
mc 16 π G
∂ ρ ∂ρ + hµ,ν = − T̃µ,ν (507)
h̄ c3

where T̃µ,ν is an appropriate expression linear in the energy-momentum tensor

which radiates a tensor field with five independent components. The unwanted
components of T̃µ,ν must be projected out, similar to what we have done for
the linearized theory of general relativity. The propagator takes care of this
projection. The equation of motion is a linear equation with a source, so it can
be solved by introducing a propagator which satisfies the equation
  2   
mc 0 1
ρ
∂ ∂ρ + Dµ,ν;σ,τ (x, x ) = δµσ δντ + δµτ δνσ δ 4 (x−x0 ) (508)
h̄ 2

so that the solution of the inhomogeneous equation can be written as

  Z
16 π G
hµ,ν (x) = d4 x0 Dµ,ν;σ,τ (x, x0 ) Tσ,τ (x0 ) (509)
c3

which involves a convolution of the propagator with the source.

The above analysis implies that the Fourier transformed propagator for a
graviton with mass m must have the form
 
Bµ,ν;σ,τ (k)
Dµ,ν;σ,τ (k) = (510)
k ρ kρ − ( mh̄ c )2

82
The graviton’s field is normalized according to
X
α α ˆ
µ,ν (k) µ,ν (k) = I (511)
α

The numerator of the graviton propagator must involve a combination which

transforms as a tensor under a Lorentz transformation. The propagator’s form
must be such that the symmetric tensor hµ,ν must satisfy the four gauge condi-
tions and be traceless. Using the five conditions repeatedly, one can show that
the numerator is proportional to
 
X
α α 2
µ,ν (k) σ,τ (k) ∝ Gµ,σ (k) Gν,τ (k) + Gµ,τ (k) Gν,σ (k) − Gµ,ν (k) Gσ,τ (k)
α
3
(512)
where  
kµ kν
Gµ,ν (k) = − ηµ,ν + m c 2 (513)
( h̄ )
The sign of the propagator is fixed by considering a massive graviton in its rest
frame and setting µ = σ = 1 and ν = τ = 2. This leads to
 
 Gµ,σ (k) Gν,τ (k) + Gµ,τ (k) Gν,σ (k) − 23 Gµ,ν (k) Gσ,τ (k) 
1
Dµ,ν;σ,τ (k) =
2 k ρ kρ − ( mh̄ c )2
(514)
Einstein’s equation links the Riemann curvature, or equivalently hµ,ν , to the
metric tensor T µ,ν . Since energy and momentum are conserved, in flat space
the energy-momentum tensor satisfies

∂µ T µ,ν = 0 (515)

or, on Fourier transforming

kµ T µ,ν (k) = 0 (516)

The above condition eliminates the the k-dependent terms in the propagator
when coupled to the energy-momentum tensor, so one can replace the Gµ,ν (k)
by − ηµ,ν . It should be noted that the energy density is positive

T 0,0 (k) > 0 (517)

and for time-like indices η0,0 = 1. The interaction Lagrangian corresponding

to two static masses is given by

d3 k 1 + 1 − 23
  Z
8πG 0,0∗
Lint = − T (0, k) T 0,0 (0, k)
c2 ( 2 π )3 − k 2 − ( mh̄ c )2 + i η
(518)
so for static mass densities, the interaction Lagrangian is positive. Thus, the
gravitational interaction between two masses is attractive. However, after we

83
have introduced the factor of − 12 to obtain the static interaction potential, we
that the result differs from the result expected from Newton’s law of gravity
by a factor of 43 . That is, one expects that the factor of 32 in the last term of
the propagator should be replaced by unity. The propagator corresponding to
the linearized theory of gravity (for which m = 0) does have the 23 replaced
by 1. Due to this discrepancy, we shall examine the connection between the
propagator and the source in more detail below.

9.4 The Sourced Equations

The Fierz-Pauli Lagrangian in the presence of a source can be written as

c4

L = + ∂λ hµ,ν ∂ λ hµ,ν − ∂λ ηµ,ν hµ,ν ∂ λ ησ,τ hσ,τ
32 π G
− 2 ∂µ hν,λ ∂ ν hµ,λ + 2 ∂µ hµ,ν ∂ν η σ,τ hσ,τ
 2  
mc
− hµ,ν hµ,ν − η µ,ν hµ,ν η σ,τ hσ,τ
h̄
− c hµ,ν T µ,ν (519)

In the presence of a source, the Fierz-Pauli equation of motion becomes

 
16 π G
− T µ,ν = ∂λ ∂ λ hµ,ν − η µ,ν ∂λ ∂ λ ησ,τ hσ,τ
c3
− ∂λ ∂ µ hλ,ν − ∂λ ∂ ν hλ,µ
+ ∂ µ ∂ ν η σ,τ hσ,τ + η µ,ν ∂σ ∂τ hσ,τ
 2  
mc µ,ν µ,ν σ,τ
+ h − η ησ,τ h (520)
h̄

On operating on the equation with ∂µ and on noting that only the mass term
remains, one finds the equation
 2    
mc µ σ,τ 16 π G
∂ hµ,ν − ∂ν η hσ,τ = − ∂µ T µ,ν (521)
h̄ c3

We see that in the absence of a graviton mass, the Lagrangian theory of the
graviton requires that the energy-momentum tensor must be conserved. For the
massive graviton, the conservation of energy and momentum is not automat-
ically ensured, but is an independent assumption. We shall assume that the
energy-momentum tensor remains a conserved quantity.

On assuming conservation of energy and momentum, one can substitute the

condition
∂ µ hµ,ν − ∂ν η σ,τ hσ,τ = 0 (522)

84
into the equation of motion, yielding
 2    
mc 16 π G
∂ ρ ∂ρ hµ,ν − ∂µ ∂ν η σ,τ hσ,τ + hµ,ν − ηµ,ν η σ,τ hσ,τ = − Tµ,ν
h̄ c3
(523)
Taking the trace, we find that
 2  
mc 16 π G
( d − 1 ) η σ,τ hσ,τ = η σ,τ Tσ,τ (524)
h̄ c3

so that hµ,ν is not traceless in the presence of a source. On substituting the

trace in the condition, one finds that the four mandatory gauge conditions are
modified to become
16 π h̄2 G
 
1
∂ µ hµ,ν = ∂ν η σ,τ Tσ,τ (525)
d − 1 m2 c5

Thus, automatic conservation of energy and momentum and gauge invariance

appear to be linked. That is, when energy and momentum conservation has
to be inferred from other considerations, a specific gauge condition must be
enforced. Substituting the expression for the trace of hµ,ν into the equations of
motion leads to
 2     
ρ mc 16 π G 1 ∂µ ∂ν σ,τ
∂ ∂ρ hµ,ν + hµ,ν = − Tµ,ν − ηµ,ν + m c 2 η Tσ,τ
h̄ c3 d − 1 ( h̄ )
(526)
On Fourier transforming, one obtains
 
16 π G
c3    
1 kµ kν σ,τ
hµ,ν (k) = ρ T µ,ν (k) − ηµ,ν − η T σ,τ (k)
k kρ − ( mh̄ c )2 d − 1 ( mh̄ c )2
(527)
or
 
16 π G
c3
   
1 kµ kν
hµ,ν (k) = ρ ηµ,σ ην,τ − ηµ,ν − m c 2 ησ,τ T σ,τ (k)
k kρ − ( mh̄ c )2 d − 1 ( h̄ )
(528)
As we are assuming that energy and momentum are conserved

k µ Tµ,ν (k) = 0 (529)

the above equation can be re-written as

 
16 π G
c3
 
1
hµ,ν (k) = ρ Gµ,σ (k) Gν,τ (k) − Gµ,ν (k) Gσ,τ (k) T σ,τ (k)
k kρ − ( mh̄ c )2 d − 1
(530)

85
Since hµ,ν is symmetric in the indices, one should also symmetrize the propa-
gator. The symmetrized propagator is given by
 
8 π G
c3  
2
hµ,ν (k) = ρ G µ,σ (k) G ν,τ (k) + G µ,τ (k) G ν,σ (k) − G µ,ν (k) G σ,τ (k) T σ,τ (k)
k kρ − ( mh̄ c )2 d − 1
(531)
2
in agreement with the result of our previous calculation. The factor of d−1 is a
2
real discrepancy with the m = 0 result which contains d−2 . This discrepancy
is due to the fact that in the limit m → 0, the 5 excitations of the Fierz-Pauli
theory decouple into helicity ± 2 excitations, helicity ± 1 vector excitations and
a scalar particle. The vector particles do not mediate an interaction, but the
scalar particle does. The scalar particle is the so-called longitudinal graviton.
However, the theory with m exactly equal to zero only has two helicity ± 2
excitations and does not include the longitudinal graviton. The discontinuity
between the m = 0 theory and the limit m → 0 was investigated by Vainshtein22 .

9.5 The Modes and Energies of Massive Gravitons

The Fierz-Pauli Lagrangian density
c4

L = + ∂λ hµ,ν ∂ λ hµ,ν − ∂λ ηµ,ν hµ,ν ∂ λ ησ,τ hσ,τ
32 π G
− 2 ∂µ hν,λ ∂ ν hµ,λ + 2 ∂µ hµ,ν ∂ν η σ,τ hσ,τ
 2  
mc
− hµ,ν hµ,ν − η µ,ν hµ,ν η σ,τ hσ,τ
h̄
(532)
can be re-written by integrating the third term by parts twice, leading to
c4

L = + ∂λ hµ,ν ∂ λ hµ,ν − ∂λ ηµ,ν hµ,ν ∂ λ ησ,τ hσ,τ
32 π G
− 2 ∂ ν hν,λ ∂µ hµ,λ + 2 ∂µ hµ,ν ∂ν η σ,τ hσ,τ
 2  
mc
− hµ,ν hµ,ν − η µ,ν hµ,ν η σ,τ hσ,τ
h̄
(533)
The vanishing of the trace and the mandatory gauge conditions obtained from
the equation of motion can be used to reduce the Lagrangian density to an
effective Lagrangian density
2
c4
  
mc
Lef f = + ∂λ hµ,ν ∂ λ hµ,ν − hµ,ν hµ,ν
32 π G h̄
(534)
22 A.I. Vainshtein, Phys. Lett. B, 39, 393-394 (1972).

86
Hence, the energy of the field can be expressed as
2
c3
Z   
3 µ,ν µ,ν mc µ,ν
P0 = d r ∂0 hµ,ν ∂0 h + ∇ hµ,ν . ∇ h + hµ,ν h
32 π G h̄
(535)
or, on Fourier transforming,
2 
c3 π 2
Z  
mc
P0 = d3 k k02 + k 2 + hµ,ν (k) hµ,ν (k) (536)
4G h̄

so the energy is positive.

The fields are restricted by the four constraints

k µ hµ,ν (k) = 0 (537)

and the trace condition

ηµ,ν hµ,ν (k) = 0 (538)
The four constraints lead to the elimination of the time-like components of the
tensor as independent variables since
X kj
h0,i (k) = − hi,j (k) (539)
j
k0

and
X ki kj
h0,0 (k) = hi,j (k) (540)
i,j
k0 k0

The trace condition is

h0,0 (k) − h1,1 (k) − h2,2 (k) − h3,3 (k) = 0 (541)

Therefore, we have three independent off-diagonal amplitudes h1,2 , h1,3 , and

h2,3 . The independent diagonal amplitudes can be chosen as h3,3 and the dif-
ference h1,1 − h2,2 , since the sum h1,1 + h2,2 is determined in terms of h0,0 − h3,3
by the trace condition. The energy-density was found to be proportional to

hµ,ν (k) hµ,ν (k) (542)

which can be re-written in terms of the independent amplitudes as

X i j 2 X  X k i 2
µ,ν k k X
|hi,j (k)|2
  
hµ,ν (k) h (k) =  
0 k0
h i,j (k) − 2 
0
h i,j (k) +
i,j
k 
j

i
k 
i,j
(543)

87
For k directed along the z-direction, one has
 2 2  2  
µ,ν k 2 k 2 2
hµ,ν (k) h (k) = −1 |h3,3 (k)| − 2 −1 |h3,1 (k)| + |h3,2 (k)|
k02 k02
+ 2 |h1,2 (k)|2 + |h1,1 (k)|2 + |h2,2 (k)|2
2
3 k2
  2  
2 k 2 2
= − 1 |h 3,3 (k)| − 2 − 1 |h 3,1 (k)| + |h 3,2 (k)|
2 k02 k02
1
+ 2 |h1,2 (k)|2 + | h1,1 (k) − h2,2 (k) |2 (544)
2
Thus, the massive graviton field has five independent components.

If one assumes that  2

2 mc
k02 = k + (545)
h̄
then one sees that the two off-diagonal modes h3,1 (k) and h3,2 (k) have an energy
proportional to
 2  
mc
4 |h3,1 (k)|2 + |h3,2 (k)|2 (546)
h̄
It can be shown that the amplitudes do not diverge in the limit m → 0, so
these modes do not carry energy in this limit. The two modes have helicities of
S z = ± 1. This can be seen by noting that the linear combinations
h3,1 (k) ± i h3,2 (k) (547)
transform as
 
0 0
h3,1 (k) ± i h3,2 (k) = exp[ ∓ i ϕ ] h3,1 (k) ± i h3,2 (k) (548)

under a rotation of ϕ around the z-axis. Thus, these combinations have angular
momenta of ± 1 around the z-axis. The physical effects of the modes h3,1 (k)
and h3,2 (k) vanish when m → 0.

The two modes h1,1 (k) − h2,2 (k) and h1,2 (k) carry energy in the limit of zero
mass, have the helicity S z = ± 2 and correspond to the graviton. That is, the
linear combination
 
h1,1 (k) − h22 (k)
± i h1,2 (k) (549)
2
transform as
  
h1,1 (k) − h22 (k)
exp[ ∓ i 2 ϕ ] ± i h1,2 (k) (550)
2
under a rotation of ϕ around the z-axis and, thus, correspond to a helicity
S z = ± 2.

88
 The remaining mode h3,3 (k) is the longitudinal graviton which corresponds
to S z = 0. The amplitude of the longitudinal graviton is expected to diverge
in the limit m → 0. This can be seen by examining the equation for the source
of the gravitational radiation
      
ρ mc 2 16 π G 1 kµ kν
k kρ − ( ) hµ,ν (k) = ηµ,σ ην,τ − ηµ,ν − m c 2 ησ,τ T σ,τ (k)
h̄ c3 d − 1 ( h̄ )
(551)
and noting that, for k along the z-direction, the term in the source which di-
verges when m → 0 only couples to h3,3 (k). Thus, we have

1
h3,3 (k) ∼ (552)
( mh̄ c )2

and, therefore, this mode yields a finite contribution to the energy. It is this
longitudinal mode which gives rise to the discontinuity at m = 0.

10 Quantization of the Electromagnetic Field

Following the work of Dirac23 , the energy, momentum and angular momentum
of the electromagnetic field shall be reduced into contributions from a set of
normal modes. A particular normal mode will correspond to a particular wave
vector and a particular polarization of the field. The normal modes can be
described in terms of a set of harmonic oscillators and, when quantized, the
normal modes will be described by quantum mechanical harmonic oscillators.

In the absence of sources, the (classical) wave equation for the vector poten-
tial has the form
1 ∂2
 
2
− ∇ + 2 A = 0 (553)
c ∂t2
when the Coulomb gauge condition is imposed

∇.A = 0 (554)

The Fourier transformation, with respect to space is defined as

Z  
1 3
A(k, t) = √ d r exp i k . r A(r, t) (555)
V
23 P. A. M. Dirac, Proc. Roy. Soc. A 114, 243 (1927).

In this paper Dirac uses two different approaches to quantizing electromagnetism. In one
approach he treated a single photon as satisfying a single-particle Schrödinger equation, that
has a similar form to Maxwell’s equations. The other approach treated the fields as dynamical
variables and then quantized them. Dirac then showed that these two methods produce
equivalent results. By doing this, Dirac created second quantization.

89
where V is the volume of the system. The inverse Fourier Transform is given
by  
1 X
A(r, t) = √ exp − i k . r A(k, t) (556)
V k

On Fourier transforming the wave equation with respect to space and time, one
finds the equation of motion
  2
2 1 ∂
k + 2 A(k, t) = 0 (557)
c ∂t2

and the Coulomb gauge condition becomes

k . A(k, t) = 0 (558)

We shall look for solutions for A(k, t) that have a time dependence given by
linear superpositions of the terms proportional to
 
exp ∓ i ωk t (559)

By substituting the above terms into the wave equation, it is found that linear
superpositions of plane-waves are solutions of Maxwell’s equation but only if
the frequency ωk and wave vector k are related via the dispersion relation

ωk2 = c2 k 2 (560)

The gauge condition also requires that the vector potential is oriented perpen-
dicular to the direction of propagation. Therefore, an arbitrary plane-wave
solution can be represented as a linear superposition of two polarized waves
with polarizations described by two mutually orthogonal unit vectors denoted
by ˆα (k). The polarization vectors satisfy

k . ˆα (k) = 0
ˆα (k) . ˆβ (k) = δα,β (561)
 
We shall assume that three vectors k, ˆ1 (k), ˆ2 (k) form a mutually orthogonal
coordinate system. We shall define

ˆ1 (−k) = ˆ1 (k)

ˆ2 (−k) = ˆ2 (k) (562)

The algebraic equations for A(k) can be solved trivially. One can express the
vector potential as a linear superposition
 
1 X
A(r, t) = √ ˆα (k) exp − i k . r Φα (k, t) (563)
V k,α

90
 E
B
k

Figure 9: The normal modes of the classical electromagnetic field are plane-
polarized waves, in which E and B are transverse to the direction of propagation
k, and oscillate in phase.

However, since the vector potential is real

A(r, t) = A∗ (r, t) (564)
one must have
Φα (k, t) = Φ∗α (−k, t) (565)
Therefore, if Φα (k) and Φ∗α (k) are to be considered as being independent fields,
then one must restrict k to have values in a volume of k-space that does not
contain both k and −k for any fixed value of k. This curiosity is associated with
the fact that, for purely real fields, particles are identical to their anti-particles.

10.1 The Lagrangian and Hamiltonian Density

The Lagrangian density L for the electromagnetic field can be expressed as
 
1 2 2
L = E − B (566)
8π
in the Coulomb gauge, the electromagnetic field is given by
1 ∂A
E = −
c ∂t
B = ∇ ∧ A (567)
Hence, the Lagrangian density is expressed as
  2  2 
1 1 ∂A
L = − ∇ ∧ A (568)
8 π c2 ∂t

91
The Lagrangian is given by the space integral of the Lagrangian density
Z
L = d3 r L (569)

On substituting A(r, t) in the form of eqn(563) and integrating over r and using
the identity Z  
1 3 0
d r exp i ( k + k ) . r = δk+k0 (570)
V
one finds the Lagrangian is given by
1 X X
L = δk+k0
8π
k,k0 α,β

∂Φβ (k 0 )
   
0 1 ∂Φα (k)
× ˆα (k) . ˆβ (k ) 2
c ∂t ∂t

0 0 0
+ ( k ∧ α (k) ) . ( k ∧ β (k ) ) Φα (k) Φβ (k )

1 ∂Φ∗α (k)
    
1 X ∂Φα (k) 2 ∗
= − k Φα (k) Φα (k)
8π c2 ∂t ∂t
k,α

(571)
In the above expression, the summation over k is unrestricted. If the Lagrangian
is to be expressed in terms of the independent components, then the summation
over k must be restricted to half the set of allowed values. With this restriction,

-k

Figure 10: A possible partition of k-space, which does not contain both k and
its inverse −k.

one obtains
  ∗   
2 X0 1 ∂Φα (k) ∂Φα (k) 2 ∗
L = − k Φ α (k) Φ α (k)
8π c2 ∂t ∂t
k,α

(572)

92
where the prime over the summation denotes the restriction of k to values
in the “positive” half volume of k-space. Since there are half the number of
independent normal modes, their contributions are twice as big. The Lagrangian
is a function of the six generalized variables Φα (k) and Φ∗α (k) for the independent
k values. The generalized momenta variables are found as
 ∗ 
2 ∂Φα (k)
Πα (k) =
8 π c2 ∂t
 
∗ 2 ∂Φ α (k)
Πα (k) = (573)
8 π c2 ∂t

The Lagrangian equations of motion of the field are given by

k2
  
∂ 1 ∂Φα (k)
= − Φα (k) (574)
∂t 8 π c2 ∂t 8π
or
∂ 2 Φα (k)
 
= − ωk2 Φα (k) (575)
∂t2
where ωk = c k. Thus, the classical field Φα (k) has a time-dependent ampli-
tude which resembles that of a harmonic oscillator with frequency ωk = c k.
The Hamiltonian can be obtained from the Lagrangian, via the Legendre Trans-
formation
X  ∂Φ∗α (k)

0 ∂Φα (k)
H = Π∗α (k) + Πα (k) − L (576)
∂t ∂t
k,α

which leads to the explicit expression for the Hamiltonian

X  8 π c2 2 2 ∗

0
H = Π∗α (k) Πα (k) + k Φα (k) Φα (k) (577)
2 8π
k,α

where the summation over (k, α) runs over the independent normal modes.
Hence, the k summation only runs over the set of points in k space which are
not related via the inversion operator. The Hamiltonian is related to the energy
of the electromagnetic field, as shall be seen below.

The energy density H for the electromagnetic field can be expressed as

 
1
H = E2 + B2 (578)
8π

in the Coulomb gauge. The energy density can be written in terms of the vector
potential as
  2  2 
1 1 ∂A
H = + ∇ ∧ A (579)
8 π c2 ∂t

93
The energy is the integral of the energy density over all space
Z
H = d3 r H (580)

When expressed in terms of the generalized coordinates and the generalized

momenta, the energy reduces to the expression
X  8 π c2 1 2 ∗

∗
H = Πα (k) Πα (k) + k Φα (k) Φα (k) (581)
4 8π
k,α

in which the summation over k is unrestricted. Thus, the above expression for
the energy is identical to the Hamiltonian for the electromagnetic field. Fur-
thermore, the Hamiltonian has been expressed in terms of a set of the normal
modes labeled by (k, α).

10.2 Quantizing the Normal Modes

The quantized Hamiltonian is obtained from the classical Hamiltonian by re-
placing the field components and their canonically conjugate momenta
Φα (k) , Πα (k) (582)
by the operators
Φ̂α (k) , Π̂α (k) (583)
and their complex conjugates are replaced by the Hermitean conjugate opera-
tors. The canonically conjugate coordinates and momenta operators satisfy the
commutation relations
[ Φ̂α (k) , Π̂β (k 0 ) ] = i h̄ δα,β δk,k0
[ Π̂α (k) , Π̂β (k 0 ) ] = 0
0
[ Φ̂α (k) , Φ̂β (k ) ] = 0 (584)
The quantized Hamiltonian for the electromagnetic field is given by
X  8 π c2 1 2 †

†
Ĥ = Π̂α (k) Π̂α (k) + k Φ̂α (k) Φ̂α (k) (585)
4 8π
k,α

The Hamiltonian can be factorized by introducing the annihilation operators

 s s
8 π c2 2 k2

1 †
âk,α = √ i Π̂α (k) + Φ̂ (k) (586)
2 2 h̄ ωk 8 π h̄ ωk α

and the Hermitean conjugate operators

s s
8 π c2 † 2 k2
 
† 1
âk,α = √ − i Π̂α (k) + Φ̂α (k) (587)
2 2 h̄ ωk 8 π h̄ ωk

94
known as creation operators. The commutation relations for the creation and
annihilation operators can be obtained directly from the commutation relations
of the field operators Φ̂α (k) and Π̂α (k) which are shown in eqn(584). It can
be shown that the creation and annihilation operators satisfy the commutation
relations

[ âk,α , â†k0 ,β ] = δα,β δk,k0

[ â†k,α , â†k0 ,β ] = 0
[ âk,α , âk0 ,β ] = 0 (588)

The field operators can be expressed in terms of the creation and annihilation
operators. Starting with
 s s
8 π c2 2 k2

1 †
âk,α = √ i Π̂α (k) + Φ̂ (k) (589)
2 2 h̄ ωk 8 π h̄ ωk α

transforming k → −k and then by noting that Π̂α (−k) = Π̂†α (k) and Φ̂†α (−k) =
Φ̂α (k), one finds
 s s
8 π c2 † 2 k2

1
â−k,α = √ i Π̂α (k) + Φ̂α (k) (590)
2 2 h̄ ωk 8 π h̄ ωk

One can eliminate Π̂†α (k) by adding the expression for the creation operator
given by eqn(587) and the expression for the annihilation operator with mo-
mentum −k given by eqn(590). This process yields the expression for the field
component operators Φ̂α (k) in the form
r  
2 π h̄ ωk †
Φ̂α (k) = âk,α + â−k,α (591)
k2
and, by an analogous procedure, the Hermitean conjugate operator is found to
be given by r  
† 2 π h̄ ωk †
Φ̂α (k) = âk,α + â−k,α (592)
k2
which is identical to Φ̂α (−k). Likewise, the canonically conjugate momenta
operators are given by
r  
h̄ ωk †
Π̂α (k) = i â−k,α − âk,α (593)
8 π c2
and their Hermitean conjugates are
r  
h̄ ωk
Π̂†α (k) = − i â−k,α − â†k,α (594)
8 π c2
as was anticipated.

95
10.2.1 The Energy of the Field
The Hamiltonian of the electromagnetic field
X  8 π c2 1 2 †

Ĥ = Π̂α (k) Π̂†α (k) + k Φ̂α (k) Φ̂α (k) (595)
4 8π
k,α

can be expressed in terms of the creation and annihilation operators as

X h̄ ωk   †  
Ĥ = â−k,α − âk,α â−k,α − â†k,α
4
k,α
  
+ â†k,α + â−k,α âk,α + â†−k,α
X h̄ ωk  † 
= âk,α âk,α + âk,α â†k,α + â†−k,α â−k,α + â−k,α â†−k,α
4
k,α

(596)

If one sets k → −k in the second set of terms, then one finds the Hamiltonian
becomes the sum over independent harmonic oscillators for each k value and
polarization
X h̄ ωk  † †

Ĥ = âk,α âk,α + âk,α âk,α (597)
2
k,α

The number operator for each normal mode is given by

n̂k,α = â†k,α âk,α (598)

and has integer eigenvalues denoted by nk,α . Hence, the energy eigenvalues E
are given by  
X 1
E = h̄ ωk nk,α + (599)
2
k,α

The energy of the electromagnetic field is quantized in units of h̄ ωk = h̄ c k.

The quanta are known as photons.

It should be noted that the contributions to the total energy from the zero-
point energy terms h̄ω2 k diverge. However, in most circumstances, only the
excitation energy of the field is measurable, hence the divergence is mainly ir-
relevant. The zero-point energy does have physical consequences, and can be
observed if the volume or boundary conditions of the field are changed. The
change in the zero-point energy of the field due to change in volume or boundary
conditions is known as the Casimir effect24 .

24 H. B. G. Casimir, Proc. Neth. Aka. Wetenschapen, 51, 793 (1948), M. J. Sparnaay,

Physica 24, 761 (1959)

96
10.2.2 The Electromagnetic Field
The quantized vector potential is given by the operator Â(r), given by
s
2 π h̄ c2
   
†
X
Â(r) = ˆα (k) âk,α + â−k,α exp − i k . r (600)
ωk V
k,α

In the Heisenberg representation, the time dependence of the vector potential

is found from
∂A(r, t)
i h̄ = [ A(r, t) , Ĥ ] (601)
∂t
which has the solution
   
t t
Â(r, t) = exp + i Ĥ Â(r, 0) exp − i Ĥ (602)
h̄ h̄
or
s
2 π h̄ c2
      
â†k,α exp
X
Â(r, t) = ˆα (k) i ωk t + â−k,α exp − i ωk t exp − ik.r
ωk V
k,α

(603)

The above equation was obtained by noting that, in the basis composed of
eigenstates of the number operators |nk,α >, one has
   
âk,α (t) |nk,α > = exp + iωk t (â†k,α âk,α + 1/2) âk,α (0) |nk,α > exp − iωk t (nk,α + 1/2)
   
√
= exp + iωk t (nk,α − 1/2) nk,α |nk,α − 1 > exp − iωk t (nk,α + 1/2)
 
= exp − i ωk t âk,α |nk,α > (604)

and that the time-dependent creation operator is given by the Hermitean con-
jugate expression. Thus, the explicit form of time dependence of the vector
potential is a consequence of the explicit time dependence of the creation and
annihilation operators in the Heisenberg representation. Alternatively, one can
find the time dependence of the creation and annihilation operators directly
from the Heisenberg equations of motion without invoking a privileged set of
basis states. The equation of motion for the creation operator is given by

∂â†k,α
i h̄ = [ â†k,α , Ĥ ] (605)
∂t
and the commutator is evaluated as

[ â†k,α , â†k0 ,β âk0 ,β ] = − â†k,α δα,β δk,k0 (606)

97
so the equation of motion simplifies to

∂â†k,α
i h̄ = − h̄ ωk â†k,α (607)
∂t
Therefore, one finds the result
 
â†k,α (t) = â†k,α exp i ωk t (608)

Likewise, the annihilation operator satisfies the equation of motion

∂âk,α
i h̄ = [ âk,α , Ĥ ] (609)
∂t
and as
[ âk,α , â†k0 ,β âk0 ,β ] = + âk,α δα,β δk,k0 (610)
so the equation of motion simplifies to
∂âk,α
i h̄ = + h̄ ωk âk,α (611)
∂t
Hence, one finds that the time-dependent annihilation operator is given by
 
âk,α (t) = âk,α exp − i ωk t (612)

which is just the Hermitean conjugate of the â†k,α (t) that was found previously.
Therefore, the time-dependence of the vector potential is entirely due to the
time-dependence of the Heisenberg representation of the creation and annihila-
tion operators.

10.2.3 The Momentum of the Field

The total momentum operator for the electromagnetic field is given by the
integral over all space of the Poynting vector
Z  
1 3
P̂ = d r Ê ∧ B̂ (613)
4πc

This will be evaluated by expressing the Ê and B̂ field operators in terms of the
vector potential A operator via

1 ∂ Â
Ê = −
c ∂t
B̂ = ∇ ∧ Â (614)

98
The vector potential operator can be written in terms of the creation and anni-
hilation operators for the normal modes as
s
2 π h̄ c2
   
†
X
Â(r, t) = ˆα (k) âk,α (t) + â−k,α (t) exp − i k . r (615)
ωk V
k,α

then the E and B field operators are found as

r    
2 π h̄ ωk
â†k,α − â−k,α
X
Ê(r) = − i ˆα (k) exp − i k . r (616)
V
k,α

and
s
2 π h̄ c2
   
â†k,α + â−k,α
X
B̂(r) = − i ( k ∧ ˆα (k) ) exp − ik.r
ωk V
k,α
(617)
For a fixed k, the polarization vectors ˆα (k) and k are mutually orthogonal.
Therefore, one has
ˆα (k) ∧ ( k ∧ ˆβ (k) ) = k ( ˆα (k) . ˆβ (k) ) − ˆβ (k) ( k . ˆα (k) )
= k δα,β (618)
Hence, the total momentum of the electromagnetic field is determined from
  
h̄ X
P̂ = ˆα (k) ∧ ( k ∧ ˆα (k) ) â†k,α − â−k,α â†−k,α + âk,α
2
k,α
  
h̄
k â†k,α − â−k,α â†−k,α + âk,α
X
= (619)
2
k,α

It should be noted that the momentum from each normal mode of the field is
parallel to its direction of propagation. Since the creation operators commute
â†k,α â†−k,α = â†−k,α â†k,α (620)
and that the annihilation operators also commute
â−k,α âk,α = âk,α â−k,α (621)
one finds that the part of the momentum represented by the summation over k
given by
X  † †

h̄ k âk,α â−k,α − â−k,α âk,α = 0 (622)
k,α
vanishes since the summand is odd under inversion symmetry. Thus, the mo-
mentum of the electromagnetic field is given by
 
h̄ X † †
P̂ = k âk,α âk,α − â−k,α â−k,α
2
k,α
 
1 X
= h̄ k â†k,α âk,α − h̄ k â†−k,α â−k,α − h̄ k (623)
2
k,α

99
where the commutation relations for the creation and annihilation operators
were used to obtain the last line. The last term vanishes when summed over k,
due to inversion symmetry. Hence, the momentum of the field is given by the
operator
 
1 X † †
P̂ = h̄ k âk,α âk,α − h̄ k â−k,α â−k,α (624)
2
k,α

Finally, on transforming −k to k in the last term of the summand, one finds the
total momentum of the field is carried by the excitations since

h̄ k â†k,α âk,α
X
P̂ = (625)
k,α

Thus, each quantum excitation of wave vector k has momentum h̄ k.

Since a photon has an energy h̄ c k and momentum h̄ k, these quanta are

massless because the mass of the quanta are defined as the relativistic invariant
length of the momentum four-vector
 2
E
− p2 = m2 c2 (626)
c
which yields m = 0. The energy-momentum dispersion relation of the quanta
of the electromagnetic field was conclusively demonstrated by A. H. Compton25 .
Compton showed that when quanta are scattered by charged particles, the pho-
ton’s dispersion relation follows directly by application of conservation laws to
the recoiling particle.

10.2.4 The Angular Momentum of the Field

The total angular momentum operator of the electromagnetic field JˆEM is given
by Z  
ˆ 1 3
J EM = d r r ∧ ( Ê ∧ B̂ ) (627)
4πc
The i-th component is given by
Z  
ˆ(i) 1 3 i,j,k (j) (k)
JEM = d rξ x ( Ê ∧ B̂ )
4πc
Z  
1 3 i,j,k (j) k,l,m (l) (m)
= d rξ x ξ Ê B̂
4πc
∂ Â(p)
Z  
1
= d3 r ξ i,j,k x(j) ξ k,l,m Ê (l) ξ m,n,p
4πc ∂x(n)
(628)
25 A. H. Compton, Phys. Rev. Second Series, 21, 483 (1923).

100
However, due to the identity
 
ξ k,l,m ξ m,n,p = δ k,n δ l,p − δ k,p δ l,n (629)

one finds
∂ Â(l) (k)
Z  
(i) 1 (l) ∂ Â
JˆEM = d3 r ξ i,j,k x(j) Ê (l) − x(j)
Ê (630)
4πc ∂x(k) ∂x(l)
On integrating by parts in the last term, one has
(l)
Z    
ˆ(i) 1 3 i,j,k (j) (l) ∂ Â ∂ (j) (l) (k)
JEM = d rξ x Ê + x Ê Â
4πc ∂x(k) ∂x(l)
(631)
Since the divergence of the electric field vanishes26 ,
∂ Ê (l)
= 0 (634)
∂x(l)
and since
∂x(j)
= δ j,l (635)
∂x(l)
the total angular momentum can be re-written as
∂ Â(l)
Z  
(i) 1
JˆEM = d3 r ξ i,j,k Ê (l) x(j) + Ê (j)
Â(k)
(636)
4πc ∂x(k)
The first term can be recognized as the orbital angular momentum of the field.
The orbital angular momentum operator L̂(i) is given by
∂
L̂(i) = − i h̄ ξ i,j,k x(j) (637)
∂x(k)
so the total angular momentum of the field is given by
Z  
ˆ (i) i 3 (l) (i) (l) i,j,k (j) (k)
JEM = d r Ê L̂ Â − i h̄ ξ Ê Â
4 π h̄ c
Z  
i
= d3 r Ê (l) L̂(i) Â(l) + Ê (j) ( Ŝ (i) )j,k Â(k) (638)
4 π h̄ c
26 In the presence of a charge density q |ψ(r)|2 , the angular momentum of the EM field will

contain a term given by Z

q
− d3 r ψ ∗ (r) r ∧ A(r) ψ(r) (632)
c
which is gauge-dependent. This term combines with a corresponding term in the expression
for the orbital angular momentum of the charged particles to yield the gauge-invariant term
Z
q
d3 r ψ ∗ (r) r ∧ ( p̂ − A(r) ) ψ(r) (633)
c
Hence, although the total angular momentum of a system of charged particles in an electro-
magnetic field is well-defined and gauge-invariant, there could be some confusion as to how
the angular momentum is distributed between the particles and the field.

101
where the definition
( Ŝ (i) )j,k = − i h̄ ξ i,j,k (639)
for Ŝ, the intrinsic spin operator for the photon, has been used in obtaining the
second line. The total vector angular momentum operator can be expressed as
Z  
i 3 (j) j,k j,k
Ĵ EM = d r Ê L̂ δ + ( Ŝ ) Â(k) (640)
4 π h̄ c
which shows that the orbital angular momentum is diagonal with respect to
the field components and the spin angular momentum mixes the different field
components.

The total spin component of the angular momentum operator for the elec-
tromagnetic field is given by
Z  
(i) i 3 (j) (i) j,k (k)
ŜEM = d r Ê ( Ŝ ) Â
4 π h̄ c
Z
1
= d3 r ξ i,j,k Ê (j) Â(k)
4πc
Z  (i)
1
= d3 r Ê ∧ Â (641)
4πc
This can be expressed in terms of the photon creation and annihilation operators
as
 
(i) h̄ X (j) i,j,k (k)
ŜEM = − i ˆβ (k) ξ ˆα (k)
2
k,α,β
  
† †
× â−k,β − âk,β âk,α + â−k,α (642)

The first term in parenthesis is recognized as the i-th component of the vector
product
ˆβ (k) ∧ ˆα (k) (643)
and, therefore, it is antisymmetric in the polarization indices α and β and the
non-zero contributions are restricted to the case α 6= β. Since the creation
and annihilation operators corresponding to different polarizations commute,
the product of the two remaining parenthesis can be re-arranged as the sum of
two terms
 (i)
(i) h̄ X
ŜEM = − i ˆβ (k) ∧ ˆα (k)
2
k,α,β
 
† †
× â−k,β âk,α − âk,β â−k,α
 
† †
+ â−k,β â−k,α − âk,α âk,β (644)

102
On transforming the summation variable k → −k and commuting the operators,
one finds that the first term in the square brackets is symmetric under the
interchange of α and β whereas the second term is antisymmetric. Hence, on
summing over the polarization indices, the contribution from the first term
vanishes, as it is the product of a symmetric and the antisymmetric (vector
product) term. Therefore, the total spin operator of the electromagnetic field is
expressed as
 (i)  
(i) h̄ X † †
ŜEM = i ˆβ (k) ∧ ˆα (k) âk,α âk,β − â−k,β â−k,α
2
k,α,β

(645)

On defining the sense of the polarization vectors relative to k̂ (≡ ê3 (k) the unit
vector in the direction of propagation) via
 
ˆ1 (k) ∧ ˆ2 (k) = k̂ (646)

so that k̂ corresponds to the z-direction, one finds that

  
h̄ X
Ŝ EM = i k̂ â†k,2 âk,1 − â†k,1 âk,2
2
k
 
− k̂ â†−k,2 â−k,1 − â†−k,1 â−k,2 (647)

On setting −k → k in the second part of the summation, the spin of the elec-
tromagnetic field is found as
X  † 
Ŝ EM = i h̄ k̂ âk,2 âk,1 − â†k,1 âk,2 (648)
k

It should be noted that in this expression, the indices (1, 2) refer to directions
in three-dimensional space and do not refer to the z-component of the spin an-
gular momentum. Therefore, the above equation shows that a plane-polarized
photon is not an eigenstate of the single-particle spin operator quantized along
the k-axis27 .

In our Cartesian component basis, the eigenstates of the component of the

spin operator parallel to the direction of propagation Ŝ (3) , where
 
0 −i 0
Ŝ (3) = h̄  i 0 0  (649)
0 0 0
27 Strictly speaking, this quantum number corresponds to the helicity as it is the spin eigen-

value which is quantized along the direction of propagation.

103
 e (k), where
are given by Φ m
 
1
1
e (k)
Φ +1 = − √  i 
2 0
 
0
e (k)
Φ 0 =  0 
1
 
1
1 
e (k)
Φ −1 = √ −i  (650)
2 0

and where the subscript m refers to the eigenvalue of Ŝ (3) , in units of h̄. From
this, it follows that an arbitrary transverse vector wave function Φ(k) can only
be expressed as a linear superposition of states involving m = ±1, and that the
m = 0 component is absent. On expressing an arbitrary (non-transverse) vector
wave function Φ(k) with components Φ(1) (k), Φ(2) (k) and Φ(3) (k) in terms of
its components referred to the helicity eigenstates Φm (k) one has
     (1) 
Φ+1 (k) − 1 i √0 Φ (k)
1
 Φ0 (k)  = √  0 0 2   Φ(2) (k)  (651)
Φ−1 (k) 2 1 i 0 Φ(3) (k)
This relation between the two bases can be expressed in the alternate form
m=1
X
Φ(k) = êm Φm (k) (652)
m=−1

where the circularly-polarized unit vectors are introduced via

1
ê+1 = − √ ( ˆ1 (k) + i ê2 (k) )
2
ê0 = ˆ3 (k)
1
ê−1 = √ ( ˆ1 (k) − i ˆ2 (k) ) (653)
2
The circularly-polarized unit vectors are associated with photons which have
definite helicity eigenvalues. It should be noted that these complex unit vectors
are orthogonal, and satisfy

ê∗m0 . êm = δm,m0 (654)

The above relations allow one to define the circularly-polarized creation and
annihilation operators via their relation to the quantum fields. This procedure
yields
i=3
X m=1
X
ˆi (k) âk,i = êm (k) âk,m (655)
i=1 m=−1

104
Hence, the photon annihilation operators corresponding to a definite helicity
are related to the annihilation operators for plane-polarized photons via
1
âk,m=+1 = − √ ( âk,1 − i âk,2 )
2
âk,m=0 = âk,3
1
âk,m=−1 = √ ( âk,1 + i âk,2 ) (656)
2
and the inverse relations are given by
1
âk,1 = − √ ( âk,m=1 − âk,m=−1 )
2
i
âk,2 = − √ ( âk,m=1 + âk,m=−1 )
2
âk,3 = âk,m=0 (657)
When expressed in terms of the circularly-polarized unit vectors, the spin oper-
ator for the electromagnetic field becomes
X  † †

Ŝ EM = h̄ k̂ âk,m=1 âk,m=1 − âk,m=−1 âk,m=−1 (658)
k

which is expressed in terms of photons with definite helicity. Within the man-
ifold of single-photon states with momentum h̄ k, the spin operator has eigen-
values of ±h̄ when measured along the direction k̂. It is seen that the photon
has helicity m = ±1 but does not involve the helicity state with m = 0 since
the electromagnetic field is transverse. The transverse nature of the field is due
to the photon being massless. In general, a massive particle with spin S should
have (2S + 1) helicity states. However, a massless particle can only have the
two helicity states corresponding to m = ±S.

The angular momentum of the elementary excitation of the electromagnetic

field was inferred from experiments in which beams of circularly-polarized light
were absorbed by a sensitive torsional pendulum28 . Quantum electromagnetic
theory shows that the angular momentum density of left circularly-polarized
light is just h̄ times the photon density or, equivalently, is just ω −1 times the
energy density which is also the case for classical electromagnetism29 . Hence, the
net increase of angular momentum per unit time can easily be calculated from
the excess of the angular momentum flux flowing into the pendulum over that
flowing out. Beth’s experiments verified that the net torque on the pendulum
was consistent with the theoretical prediction. Thus, the quantized electromag-
netic field has been shown to be related to a massless particle with spin h̄ and
energy-momentum given by the four-vector (h̄ωk /c, h̄k). This particle is the
photon. Every quantized field is to be associated with a type of particle.

28 R. A. Beth, Phys. Rev. 50, 115 (1936).

29 J. H. Poynting, Proc. Roy. Soc. A82, 560 (1909).

105
 e1

e2 k

Figure 11: The circularly-polarized normal modes of a classical electromag-

netic field are composed of two plane-polarized waves which are out of phase,
are mutually orthogonal, and are transverse to the direction of propagation k.
The resulting electric field spirals along the direction of propagation. The left
circularly-polarized wave shown in the diagram corresponds to a helicity of +h̄.

10.3 Uncertainty Relations

The eigenstates of the field operators such as Â(r, t) do not correspond to eigen-
states of the photon number operators.

Consider the electric field

1 ∂ Â
Ê = − (659)
c ∂t
Although the expectation value of Ê vanishes for any eigenstate of the set of
occupation numbers | {nk0 ,β } >

< {nk0 ,β } | Ê | {nk0 ,β } > = 0 (660)

since
< {nk0 ,β } | ak,α | {nk0 ,β } > = 0 (661)
the fluctuation in the field is given by
< {nk0 ,β } | Ê . Ê | {nk0 ,β } > − | < {nk0 ,β } | Ê | {nk0 ,β } > |2
= < {nk0 ,β } | Ê . Ê | {nk0 ,β } >
4π X 1
= h̄ωk ( nk,α + )
V 2
k,α

→ ∞ (662)
The fluctuations in the field diverge because the zero-point energy fluctuations
diverge.

106
 The commutation relations between the x-component of the E field and the
B field at the same instant of time are non-zero30 . That is,
 
2π X
[ Êx (r) , B̂y (r0 ) ] = h̄ ωk ˆα (k)x ( k̂ ∧ ˆα (k) )y exp i k . ( r0 − r )
V
k,α
 
2π X
− h̄ ωk ˆα (k)x ( k̂ ∧ ˆα (k) )y exp − i k . ( r0 − r )
V
k,α
 
4 π h̄ c X 0
= − kz exp − i k . ( r − r )
V
k
 
4 π c h̄ ∂ X
= i exp − i k . ( r0 − r )
V ∂z
k
c h̄ ∂ 3 0
= i δ (r − r) (663)
2 π 2 ∂z

The fact that the two polarizations are transverse to the unit vector k̂ has been
used to obtain the third line. Since Ê and B̂ do not commute, it follows that E
and B obey an uncertainty relation in that the values of E and B cannot both
be specified to arbitrary accuracy at the same point.

However, if two points in space time x and x0 are not causally related, i.e.

| r0 − r | =
6 c | t0 − t | (664)

then the operators commute

[ Êx (r, t) , B̂y (r0 , t0 ) ] = 0 (665)

Thus, if the two points in space-time are not connected by the propagation of
light, then the Ex and By fields can both be determined to arbitrary accuracy.

10.4 Coherent States

We shall focus our attention on one normal mode of the electromagnetic field,
and shall drop the indices (k, α) labelling the normal mode. A coherent state
| aϕ > is defined as an eigenstate of the annihilation operator

â | aϕ > = aϕ | aϕ > (666)

For example, the vacuum state or ground state is an eigenstate of the annihila-
tion operator, in which case aϕ = 0.

30 P. Jordan and W. Pauli Jr. 47, 151 (1927).

107
 The coherent state31 can be found as a linear superposition of eigenstates of
the number operator with eigenvalues n
∞
X
| aϕ > = Cn | n > (667)
n=0

On substituting this form in the definition of the coherent state

X
â | aϕ > = Cn â | n >
n
X
= aϕ Cn | n > (668)
n

and using the property of the annihilation operator, one has

X √ X
Cn n | n − 1 > = aϕ Cn | n > (669)
n n

On taking the matrix elements of this equation with the state < m |, and using
the orthonormality of the eigenstates of the number operator, one finds
√
Cm+1 m + 1 = aϕ Cm (670)

Hence, on iterating downwards, one finds

 m 
aϕ
Cm = √ C0 (671)
m!
and the coherent state can be expressed as
∞  n 
a
√ϕ
X
| aϕ > = C0 |n > (672)
n=0 n!

The normalization constant C0 can be found from

∞  n∗ n 
∗
X aϕ aϕ
1 = C0 C0 (673)
n=0
n!

by noting that the sum exponentiates to yield

 
1 = C0∗ C0 exp a∗ϕ aϕ (674)

so, on choosing the phase of C0 , one has

 
1 ∗
C0 = exp − a aϕ (675)
2 ϕ

108
 0.2

0.15

Pn
0.1

0.05

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Figure 12: The probability of finding n photons P (n) in a normal mode repre-
sented by a coherent state.

From this, it can be shown that if the number of photons in a coherent state
are measured, the result n will occur with a probability given by
( a∗ϕ aϕ )n
 
∗
P (n) = exp − aϕ aϕ (676)
n!
Thus, the photon statistics are governed by a Poisson distribution. Furthermore,
the quantity a∗ϕ aϕ is the average number of photons n present in the coherent
state.

The coherent states can be written in a more compact form. Since the state
with occupation number n can be written as

( ↠)n
|n > = √ |0 > (677)
n!
the coherent state can also be expressed as
∞
( aϕ ↠)n
  X
1 ∗
| aϕ > = exp − aϕ aϕ |0 > (678)
2 n=0
n!

or on summing the series as an exponential

   
1 ∗
| aϕ > = exp − aϕ aϕ exp aϕ ↠| 0 > (679)
2
Thus the coherent state is an infinite linear superposition of states with different
occupation numbers, each coefficient in the linear superposition has a specific
phase relation with every other coefficient.

31 R. J. Glauber, Phys. Rev. Lett. 10, 84 (1963).

109
 The above equation represents a transformation between number operator
states and the coherent states. The inverse transformation can be found by
expressing aϕ as a magnitude a and a phase ϕ
 
aϕ = a exp i ϕ (680)

The number states can be expressed in terms of the coherent states via the
inverse transformation
√   Z 2π  
n! 1 2 dϕ
|n > = exp + a exp − i n ϕ | aϕ >
an 2 0 2π
(681)
by integrating over the phase ϕ of the coherent state. Since the set of occupa-
tion number states is complete, the set of coherent states must also span Hilbert
space. In fact, the set of coherent states is over-complete.

The coherent state | aϕ > can be represented by the point aϕ in the Argand
plane. The overlap matrix elements between two coherent states is calculated
1
Im z

aφ
0.5
a

φ Re z
0
-1 -0.5 0 0.5 1

-0.5

-1

Figure 13: Since a coherent state | aφ > is completely determined by a complex

number aϕ , it can be represented by a point in the complex plane.

as  
| < a0ϕ0 | aϕ > | 2
= exp − | aϕ − a0ϕ0 2
| (682)

Hence, coherent states corresponding to different points are not orthogonal. The
coherent states form an over complete basis set. The over completeness relation
can be expressed as
d <e aϕ d =m aϕ
Z
| aϕ > < aϕ | = Iˆ (683)
π

110
This relation can be proved by taking the matrix elements between the occupa-
tion number states < n0 | and | n >, which leads to
d <e aϕ d =m aϕ
Z
< n0 | aϕ > < aϕ | n > = δn0 ,n (684)
π
which can be evaluated as
d <e aϕ d =m aϕ
Z
= < n0 | aϕ > < aϕ | n >
π
0
dϕ a∗ϕ n anϕ
Z ∞ Z 2π  
= da a √ exp − | aϕ |2
0 0 π n0 ! n!
Z ∞ n+n0
  Z 2π  
a 2 dϕ 0
= da a √ exp − | aϕ | exp i ( n − n ) ϕ
0 n0 ! n! 0 π
Z ∞ 0
an+n
 
= da a √ exp − a2 2 δn,n0 (685)
0 n0 ! n!
On changing variable to s = a2 , one proves the completeness relation by noting
that Z ∞  
n
ds s exp − s = n! (686)
0
Hence, the coherent states form a complete basis set.

The effect of the creation operator on the coherent state can be expressed
as
   
† † 1 ∗ †
â | aϕ > = â exp − a aϕ exp aϕ â |0 >
2 ϕ
   
1 ∗
= exp − a aϕ ↠exp aϕ ↠| 0 >
2 ϕ
   
1 ∗ ∂ †
= exp − a aϕ exp aϕ â |0 >
2 ϕ ∂aϕ
   
1 ∗ ∂ 1 ∗
= exp − a aϕ exp + a aϕ | aϕ >
2 ϕ ∂aϕ 2 ϕ
(687)
The coherent state is not an eigenstate of the creation operator, since the re-
sulting state does not include the zero-photon state.

The expectation value of the field operators between the coherent states
yields the classical value, since
< aϕ | ( ↠+ â ) | aϕ > = ( a∗ϕ + aϕ ) (688)
In deriving the above equation, the definition
â | aϕ > = aϕ | aϕ > (689)

111
has been used in the term involving the annihilation operator and the term orig-
inating from the creation operator is evaluated using the Hermitean conjugate
equation
< aϕ | ↠= < aϕ | a∗ϕ (690)
One also finds that that the expectation value of the number operator is given
by
< aϕ | ↠â | aϕ > = a∗ϕ aϕ (691)
so the magnitude of aϕ is related to the average number of photons in the
coherent state n. This identification is consistent with the Poisson distribution
of eqn(676) which governs the probability of finding n photons in the coherent
state. The coherent state is not an eigenstate of the number operator since
there are fluctuations in any measurement of the number of photons. The rms
fluctuation ∆n can be evaluated by noting that
< aϕ | n̂2 | aϕ > = < aϕ | ↠â ↠â | aϕ >
= < aϕ | ↠↠â â | aϕ > + < aϕ | ↠â | aϕ >
= ( a∗ϕ )2 ( aϕ )2 + a∗ϕ aϕ (692)
where the boson commutation relations have been used in the second line. Thus,
the mean squared fluctuation in the number operator is given by
< aϕ | ∆n̂2 | aϕ > = a∗ϕ aϕ (693)
The rms fluctuation of the photon number is only negligible when compared to
the average value if aϕ has a large magnitude
a∗ϕ aϕ  1 (694)

The expectation values of coherent states almost behave completely clas-

sically. The deviation from the classical expectation values can be seen by
examining
< aϕ | â ↠| aϕ > = a∗ϕ aϕ + 1 (695)
which is evaluated by using the commutation relations. It is seen that the ex-
pectation values can be approximated by the classical values, if the magnitude
of aϕ is much greater than unity.

Exercise:

Determine the expectation values for the electric and magnetic field opera-
tors in a coherent state which represents a plane-polarized electromagnetic wave.

Exercise:

Determine the expectation values for the electric and magnetic field opera-
tors in a coherent state which represents a left circularly-polarized electromag-
netic wave composed of photons with a helicity of +1.

112
10.4.1 The Phase-Number Uncertainty Relation
From the discussion of coherent states, it is seen that the coherent state has a
definite phase, but does not have a definite number of quanta. In general, it is
impossible to know both the phase of a state and the number of a state. This
is formalized as a phase - number uncertainty relation.

The phase and amplitude of a state is related to the annihilation operator.

Since the annihilation operator is non-Hermitean, one can construct the annihi-
lation operator as a function of Hermitean operators. Formally, the amplitude
can be related to the square root operator, and the phase to a phase operator32 .
Hence, one can write
√ˆ
 
âk,α = exp + i (ϕ̂k,α − ωk t) nk,α (696)

and the Hermitean conjugate operator, the creation operator can be expressed
as
√
 
â†k,α = nˆk,α exp − i (ϕ̂k,α − ωk t) (697)
√
since it has been required that ˆn and ϕ̂ are Hermitean. Furthermore, the
√
operator ˆn must have the property
√ˆ √ˆ
nk,α nk,α = n̂k,α (698)

On substituting the expressions for the creation and annihilation operators,

in terms of the phase and amplitude, into boson commutation relations

[ âk,α , â†k0 ,β ] = δk,k0 δα,β (699)

etc, one finds

√ˆ √ ˆ
   
δk,k0 δα,β = exp + i (ϕ̂k,α − ωk t) nk,α nk0 ,β exp − i (ϕ̂k0 ,β − ωk0 t)

√ √ˆ
   
− nˆk0 ,β exp − i (ϕ̂k0 ,β − ωk0 t) exp + i (ϕ̂k,α − ωk t) nk,α

(700)

Thus, for k = k 0 and α = β, one has

     
exp + i ϕk,α n̂k,α − n̂k,α exp + i ϕk,α = exp + i ϕk,α (701)

32 P. A. M. Dirac, Proc. Roy. Soc. A 114, 243 (1927).

113
This relationship is satisfied, if the phase and number operators satisfy the
commutation relation
[ n̂k,α , ϕk,α ] = i (702)
If one can construct the Hermitean operators that satisfy this commutation
relation, then one can show that the rms uncertainties phase and number must
satisfy the inequality

(∆ϕk,α )rms (∆nk,α )rms ≥ 1 (703)

It should be noted that only the relative phase can be measured33 . Thus, if the
phase difference of any two components (k, α) and (k 0 , α0 ) is specified precisely,
then the occupation number of either component can not be specified.

Exercise:

Express the vector potential and the electric and magnetic field operators in
terms of the amplitude and phase operators.

10.4.2 Argand Representation of Coherent States

The coherent state | aϕ > can be represented by the point aϕ in the Argand
plane. The overlap matrix elements between two coherent states is calculated
as  
| < a0ϕ0 | aϕ > |2 = exp − | aϕ − a0ϕ0 |2 (704)

Hence, coherent states are not orthogonal. In fact, their overlap decreases expo-
nentially with large “separations” between the points aϕ and a0ϕ0 in the Argand
plane. We shall denote | aϕ | by a. Two states separated by distances a ∆ϕ or
∆a such that a ∆ϕ ≥ 1 and ∆a ≥ 1 are effectively orthogonal or independent.
However, states within an area given by ∆a × a ∆ϕ ≈ 1 have significant overlap
and so can represent the same state. Therefore, the minimum uncertainty state
occupies an area ∆a × a ∆ϕ ≈ 1. We note that 2 a ∆a can be interpreted
as a measure of the uncertainty ∆nϕ in the particle number for the state, and
∆ϕ is the uncertainty in the phase of the state. Hence, the phase - number
uncertainty relation sets the area of the Argand diagram that can be associated
with a single state as
a ∆a ∆ϕ ∼ 1 (705)

33 L. Susskind and J. Glogower, Physics 1, 49 (1964).

114
 Im z

a ∆φ

∆a
∆φ

Re z

Figure 14: Due to the phase-number uncertainty principle, the minimum area
of the Argand diagram needed to represent a minimum uncertainty state has
dimensions such that a ∆a ∆ϕ ∼ 1.

11 Non-Relativistic Quantum Electrodynamics

The non-relativistic Hamiltonian for a particle with charge q and mass m inter-
acting with a quantized electromagnetic field can be expressed as

p̂2 q2
 
q 2
Ĥ = + q φ(r) − p̂ . Â(r) + Â(r) . p̂ + Â (r)
2m 2mc 2 m c2
 2 0 2
Ê (r ) + B̂ (r0 )
Z 
+ d3 r 0 (706)
8π

when the vector potential is chosen to satisfy the Coulomb gauge. The second,
third and fourth terms are to be evaluated at the location of the charged point
particle, r, and the last term is evaluated at all points in space. The Hamiltonian
can be expressed as
Ĥ = Ĥ0 + Ĥrad + Ĥint (707)
where Ĥ0 is the Hamiltonian for the charged particle in the electrostatic poten-
tial φ
p̂2
Ĥ0 = + q φ(r) (708)
2m
and Ĥrad is the Hamiltonian for the electromagnetic radiation and Hint is the
interaction
q2
 
q 2
Ĥint = − p̂ . Â + Â . p̂ + Â (709)
2mc 2 m c2

The interaction term is composed of a paramagnetic interaction which is linearly

proportional to the vector potential and the diamagnetic interaction which is
proportional to the square of the vector potential. When the electromagnetic

115
field is quantized, the radiation Hamiltonian has the form
X h̄ ωk  † 
Ĥrad = âk,α âk,α + âk,α â†k,α (710)
2
k,α

Since the quantized vector potential is given by

s
2 π h̄ c2
   
1 X †
Â(r, t) = √ êα (k) âk,α + â−k,α exp − i k . r (711)
V k,α ωk

the paramagnetic interaction can be expressed as

s
2 π h̄ c2
   
q X †
Ĥpara = − p̂ . ˆα (k) âk,α + â−k,α exp − i k . r
mc V ωk
k,α
(712)
in which the transverse gauge condition ∇ . A = 0 has also been used. The

(k,α) (k,α)

p p
p'
p'

Figure 15: The paramagnetic interaction leads to scattering of an electron from

p to p0 by either (a) absorbing a photon, or (b) by emitting a photon.

diamagnetic interaction is expressed as

q2 X  2 π h̄ c2   
0 0
Ĥdia = √ ˆβ (k ) . ˆα (k) exp − i ( k + k ) . r
2 m c2 ωk ωk 0 V
k,k0 ,α,β
 
× â†k0 ,β â†k,α + â†k0 ,β â−k,α + â−k0 ,β â†k,α + â−k0 ,β â−k,α (713)

For charged particles with spin one-half, analysis of the non-relativistic Pauli
equation shows that there is another interaction term involving the particles’
spins. This interaction can be described by the anomalous Zeeman interaction
 
q h̄
ĤZeeman = − σ.B (714)
2mc
where
B = ∇ ∧ A(r) (715)

116
and σi are the three Pauli matrices.

Generally, the paramagnetic interaction has a greater strength than the Zee-
man interaction. This can be seen by examining the magnitudes of the interac-
tions. The paramagnetic interaction has a magnitude given by
e
p.A (716)
mc
and for an atom of size a , the uncertainty principle yields
h̄
p ∼ (717)
a
The Zeeman interaction has a magnitude given by
e h̄
σ.(k ∧ A) (718)
mc
but since k is the wavelength of light
1
k ∼ (719)
λ
Hence, since the wave length of light is larger than the linear dimension of an
atom, λ > a, one finds the inequality between the magnitude of the paramag-
netic interaction and the Zeeman interaction
e h̄ 1 e h̄ 1
A > A (720)
mc a mc λ
Both the paramagnetic and Zeeman coupling strengths are proportional to the
magnitude of the vector potential A, hence the ratio of the strengths of the
interactions are independent of A. Therefore, there magnitudes satisfy the in-
equality
1 1
> (721)
a λ
so the Zeeman interaction can frequently be neglected in comparison with the
paramagnetic interaction.

11.1 Emission and Absorption of Photons

An atom in an electromagnetic field has its constituent charges perturbed by
the oscillating field, and those perturbations may lead to either to the absorp-
tion of radiation or emission of further radiation. However, thermal equilibrium
between matter and radiation can only be reached if, in addition to these in-
duced processes, there exists also a spontaneous process in which an excited
atom emits radiation even in the absence of any measurable radiation. This
spontaneous emission process may be considered as being induced by the zero-
point fluctuations of the electromagnetic field.

117
11.1.1 The Emission of Radiation
We shall consider a state | (nlm) {nk0 ,β } > which is an energy eigenstate
of the unperturbed Hamiltonian Ĥ0 and the radiation Hamiltonian Ĥrad . The
interaction Ĥint causes the system to make a transition from the initial state to
a final state. In the initial state, the electron is in an energy state designated
by the quantum numbers (n, l, m) and the electromagnetic field is in a state
specified by the number of photons in each normal mode. That is, the photon
field is in an initial state which is specified by the set of photon quantum num-
bers, {nk0 ,β 0 }. We shall consider the transition in which the electron makes a
transition from the initial state to a final state denoted by (n0 , l0 , m0 ). Since the

(hω'/c,hk')
0

Enlm
-4
E

-8
e-

-12
En'l'm'

-16

Figure 16: An electron in the initial atomic state with energy En,l,m makes a
transition to the final atomic state with energy En0 ,l0 ,m0 , by emitting a photon
with energy h̄ωk0 .

photon is emitted, the final state of the photon field described by the set {n0k0 ,β }
where
n0k0 ,β = nk0 ,β for (k 0 , β) 6= (k, α) (722)
and the number of photons in a normal mode (k, α) is increased by one
n0k,α = nk,α + 1 (723)
The transition rate for the electron to make a transition from (n, l, m) to
(n0 , l0 , m0 ) can be calculated34 from the Fermi-Golden rule expression
   
1 2π X
= | < n0 l0 m0 {n0k0 ,β } | Ĥint | nlm {nk0 ,β } > |2 δ( Enlm − En0 l0 m0 − h̄ ωk,α )
τ h̄
k,α
(724)
34 P. A. M. Dirac, Proc. Roy. Soc. A 112, 661 (1926), A 114, 243 (1927).

118
The delta function expresses the conservation of energy. The energy of the
initial state is given by
X 1
Enlm + h̄ ωk0 ,β ( nk0 ,β + ) (725)
0
2
k ,β

and the final state has energy

X 1
En0 l0 m0 + h̄ ωk0 ,β ( n0k0 ,β + ) (726)
2
k0 ,β

The difference in the energy of the initial state and final state is evaluated as
Enlm − En0 l0 m0 − h̄ ωk,α (727)
which is the argument of the delta function and must vanish if energy is con-
served. The sum over k can be evaluated by assuming that the radiation field is
confined to a volume V . The allowed k values for the normal modes are deter-
mined by the boundary conditions. In this case, the sum over k is transformed
to an integral over k-space via
Z
X V
→ d3 k (728)
( 2 π )3
k

The matrix elements of the interaction Hamiltonian between photon energy

eigenstates is evaluated as
s
q X 2 π h̄ c2
< {n0k0 ,β } | Ĥint | {nk0 ,β } > = −
mc V ωk
k,α
 
× < {n0k0 ,β } | ˆα (k) . p̂ ( â†k,α + â−k,α ) exp − ik.r | {nk0 ,β } >

(729)
since only the paramagnetic part of the interaction has non-zero matrix ele-
ments. For the photon emission process, the matrix elements of the creation
operator between the initial and final states of the electromagnetic cavity is
evaluated as
< {n0k0 ,β } | â†k,α | {nk0 ,β } > =
p
nk,α + 1 (730)
hence, the matrix elements of the interaction are given by
s
q 2 π h̄ c2 p
< n0 l0 m0 {n0k0 ,β } | Ĥint | nlm {nk0 ,β } > = − nk,α + 1
mc V ωk
 
× < n0 l0 m0 | ˆα (k) . p̂ exp − i k . r | nlm >

(731)

119
Therefore, the transition rate for photon emission can be expressed as
2
2 π h̄ c2 X
 Z  
1 2π q V 3
= d k ( nk,α + 1 )
τ h̄ mc ( 2 π )3 V ωk α
 
× | < n0 l0 m0 | ˆα (k) . p̂ exp − i k . r | nlm > |2 δ( Enlm − En0 l0 m0 − h̄ ωk )

(732)

The above expression shows that the rate for emitting a photon into state (k, α)
is proportional to a factor of nk,α + 1, which depends on the state of occupation
of the normal mode. The term proportional to the photon occupation number
describes stimulated emission. However, if there are no photons initially present
in this normal mode, one still has a non-zero transition rate corresponding to
spontaneous emission. These factors are the result of the rigorous calculations35
based on Dirac’s quantization of the electromagnetic field, but were previously
derived by Einstein36 using a different argument. From the above expression, it
is seen that the number of photons emitted into state (k, α) increases in propor-
tional to the number of photons present in that normal mode. This stimulated
emission increases the number of photons and can lead to amplification of the
number of quanta in the normal mode, and leads to the phenomenon of Light
Amplification by Stimulated Emission of Radiation (LASER).

11.1.2 The Dipole Approximation

The dipole approximation is justified by noting that in an emission process, the
typical energy of the photon is of the order of 10 eV. Hence, a typical wave
length of the photon is given by
2 π c h̄
λ = ∼ 3000 Å (733)
h̄ ω
whereas the typical length scale r for the electronic state is of the order of an
Angstrom. Therefore the product k r ∼ 10−3 , so the exponential factor in the
vector potential can be Taylor expanded as
 
exp − i k . r ∼ 1 − i k . r + ... (734)

The first term in the expression produces results that are equivalent to the ra-
diation from an oscillating classical electric dipole. If only the first term in the
expansion is retained, the resulting approximation is known as the dipole ap-
proximation. The second term in the expansion yields results equivalent to the
radiation from an electric quadrupole. The dipole approximation, where only
the first term in the expansion is retained, is justified for transitions where the
35 P. A. M. Dirac, Proc. Roy. Soc. A 114, 243 (1927).
36 A. Einstein, Verh. Deutsche Phys. Ges. 18, 318 (1916), Phys. Z. 18, 121 (1917).

120
 A(r)

Ψ(r)

V(r)

Figure 17: A cartoon depicting the relative length-scales assumed in the dipole
approximation.

successive terms in the expansion are successively smaller by factors of the order
of 10−3 . The dipole approximation crudely restricts consideration to the case
where the emitted photon can only have zero orbital angular momentum. This
follows from the dipole approximation’s requirement that the size of the atom is
negligible compared with the scale over which the vector potential varies. Then
the vector potential in the spatial region where the electron is located only de-
scribes photons with zero orbital angular momentum.

In the dipole approximation, the transition rate for single photon emission
is given by
2
2 π h̄ c2 X
 Z  
1 2π q V 3
≈ d k ( nk,α + 1 )
τ h̄ mc ( 2 π )3 V ωk α
× | ˆα (k) . < n0 l0 m0 | p̂ | nlm > |2 δ( Enlm − En0 l0 m0 − h̄ ωk )
(735)
The matrix elements of the momentum can be evaluated by noting that the
states | nlm > are eigenstates of the unperturbed electronic Hamiltonian so
Ĥ0 | nlm > = Enlm | nlm > (736)
where the unperturbed Hamiltonian is given by
p̂2
Ĥ0 = + V (r) (737)
2m
The electronic momentum operator p̂ can be expressed in terms of the commu-
tator of the Hamiltonian Ĥ0 and r through the relation
h̄
[ r , Ĥ0 ] = i p̂ (738)
m
On using this relation, the matrix elements of the momentum operator can be
written in terms of the matrix elements of the electron’s position operator r by
m
< n0 l0 m0 | p̂ | nlm > = − i < n0 l0 m0 | [ r , Ĥ0 ] | nlm >
h̄

121
 m
= i ( Enlm − En0 l0 m0 ) < n0 l0 m0 | r | nlm >
h̄
(739)

Therefore, in the dipole approximation, the transition rate is given by

q2
Z
1 X
≈ d3 k ωk ( nk,α + 1 )
τ (2π) α
× | ˆα (k) . < n0 l0 m0 | r | nlm > |2 δ( Enlm − En0 l0 m0 − h̄ ωk )
(740)

where the property of the delta function has been used to set
( Enlm − En0 l0 m0 )2
δ( Enlm − En0 l0 m0 − h̄ ωk ) = ωk2 δ( Enlm − En0 l0 m0 − h̄ ωk )
h̄2
(741)
It is seen that the volume of the electromagnetic cavity has dropped out of the
expression of eqn(740) for the transition rate. We shall assume that the number
of photons nk,α in the initial state is zero. The (complex) factor

dnlm,n0 l0 m0 = q < n0 l0 m0 | r | nlm > (742)

is defined as the electric dipole moment, and the electronic energy difference is
denoted by the frequency

Enlm − En0 l0 m0 = h̄ ωnl,n0 l0 (743)

With this notation, the transition rate can be expressed as

 2
ωnl,n0 l0 X
Z 
1 1
≈ d3 k | ˆα (k) . dnlm,n0 l0 m0 |2 δ( h̄ ωnl,n0 l0 − h̄ ωk )
τ 2π ωk α
(744)

The integration over d3 k can be performed by separating the integration over

the direction dΩk of the outgoing photon and an integration over the magnitude
of k. The integration over the magnitude of k can be performed by noting that
the integrand is proportional to a Dirac delta function, so the transition rate
can be evaluated as
2 Z ∞  2 X
ωnl,n
Z
1 0 l0 k
= dΩk dk | ˆα (k) . dnlm,n0 l0 m0 |2 δ( ωnl,n0 l0 − c k )
τ 2 π h̄ 0 ω k α
3
ωnl,n0 l0
Z X
= dΩk | ˆα (k) . dnlm,n0 l0 m0 |2 (745)
2 π h̄ c3 α

The above expression yields the rate at which an electron makes a transition
between the initial and final electronic state, in which one photon of any polar-
ization is emitted in any direction.

122
 If one is only interested in the decay rate of the electronic state via the
emission of a photon, one should sum over all polarizations and integrate over
all directions of the emitted photon. The direction of the emitted photon k̂ is
expressed in terms of polar coordinates defined with respect to an arbitrarily
chosen polar axis. The direction of the photon’s wave vector k̂ is defined as

dΩk

z e2(k)
k
e1(k)
θk
y

φk

Figure 18: A photon is emitted with wave vector k with a direction denoted
by the polar coordinates (θk , ϕk ). The polarization vector ê1 (k) is chosen to be
in the plane containing the polar-axis and k, therefore, ê2 (k) is parallel to the
x − y plane.

k̂ = (sin θk cos ϕk , sin θk sin ϕk , cos θk ). The directions of the two transverse
polarizations α are defined as

ˆ1 (k) = (cos θk cos ϕk , cos θk sin ϕk , − sin θk )

ˆ2 (k) = (− sin ϕk , cos ϕk , 0) (746)

The scalar product between the polarization vectors and the dipole moment can
be expressed in terms of the Cartesian components via
(i)
X
ˆα (k) . dnlm,n0 l0 m0 = ˆ(i)
α (k) . ( dnlm,n0 l0 m0 ) (747)
i

As neither the polarization nor the direction of the outgoing photon are mea-
sured, the transition rates is determined as an integral over all directions
3
ωnl,n
  X X Z
1 0 l0 (j) ∗ (i)
= dΩk ˆ(j) ˆ(i)
α (k) α (k) ( dnlm,n0 l0 m0 ) ( dnlm,n0 l0 m0 )
τ 2 π h̄ c3 i,j α
(748)

123
On using the identity
Z
1 X 2
dΩk ˆ(j) ˆ(i)
α (k) α (k) = δi,j (749)
4π α 3

one finds that the transition rate is given by the scalar product of complex
vectors ∗
3
4 ωnl,n0 l0 dnlm,n0 l0 m0 . dnlm,n0 l0 m0
 
1
= (750)
τ 3 h̄ c3
The electric dipole matrix elements can be shown to vanish between most pairs
of states. The selection rules determine which matrix elements are non-zero
and, therefore, which electric dipole transitions are allowed.

11.1.3 Electric Dipole Radiation Selection Rules

Electric dipole induced transitions obey the selection rules ∆l = ± 1 and either
∆m = ± 1 or 0, where l is the quantum number for electron’s orbital angular
momentum and m is the z-component. The dipole selection rules can be derived
by writing the wave functions for the one-electron states as

ψn,l,m (r) = Rnl (r) Yml (θ, ϕ) (751)

where Rn,l (r) is the radial wave function, and Yml (θ, ϕ) is the spherical harmonic
function quantized along the z-direction. The components of an arbitrarily
oriented electric dipole matrix elements involve matrix elements of the quantities

x = r sin θ cos ϕ
y = r sin θ sin ϕ
z = r cos θ (752)

Since the above expressions are the components of a vector, they can be re-
written as combinations of the spherical harmonics with angular momentum
l = 1, via
r  
1 8π 1 1
x = r Y−1 (θ, ϕ) − Y1 (θ, ϕ)
2 3
r  
i 8π 1
y = r Y−1 (θ, ϕ) + Y11 (θ, ϕ)
2 3
r
4π 1
z = r Y0 (θ, ϕ) (753)
3
Hence, the components of the vector r can be written as
r  
4π êx + i êy 1 1 êx − i êy 1
r = r √ Y−1 + êz Y0 − √ Y1 (754)
3 2 2

124
The circular polarization vectors are given by
êx − i êy
êm=−1 = √
2
êm=0 = êz
êx + i êy
êm=+1 = − √ (755)
2
which are orthogonal
ê∗m0 . êm = δm,m0 (756)
Hence, the vector r can be written in the alternate forms
r
4π X ∗ 1
r = r êm Ym (θ, ϕ)
3 m
r
4π X
= r êm Ym1 (θ, ϕ)∗ (757)
3 m

This illustrates that through the dipole approximation coupling term

 
r . êm âk,m + ê∗m â†k,m (758)

(where k ≈ 0), an electron with angular momentum quantized along the z-

direction most naturally couples to circularly-polarized light with the same
quantization axis. The electric dipole matrix elements involve the three fac-
tors
Z 2π Z π
0
dϕ dθ sin θ Yml 0 (θ, ϕ)∗ Y±1
1
(θ, ϕ) Yml (θ, ϕ)
0 0
Z 2π Z π
0
dϕ dθ sin θ Yml 0 (θ, ϕ)∗ Y01 (θ, ϕ) Yml (θ, ϕ) (759)
0 0

which come from the angular integrations. Conservation of angular momentum

leads to the dipole-transition selection rules
l0 = l ± 1 (760)
and
m0 = m ± 1
m0 = m (761)
because one unit of angular momentum is carried away by the photon in the
form of its spin37 .

37 In the dipole approximation, the photon is restricted to have zero orbital angular mo-

mentum. Therefore, the angular momentum is completely transformed to the photon’s spin.
More generally, the spatial (plane-wave) part of the vector potential should be expanded in
terms of spherical harmonics to exhibit the photon’s orbital angular momentum components.

125
The m-selection rules for electric dipole transitions.

The selection rules on the z-component of the angular momentum, m, follow

directly from the ϕ-dependence of the spherical harmonics
 
l l 1
Ym (θ, ϕ) = Θm (θ) √ exp i m ϕ (762)
2π
so, the integral over the Cartesian components of the dipole matrix elements
involve
Z 2π     
1 sin ϕ 1 − i δm+1,m0 + i δm−1,m0
dϕ exp i ( m − m0 ) ϕ =
2π 0 cos ϕ 2 δm+1,m0 + δm−1,m0
(763)
and Z 2π  
1
dϕ exp i ( m − m0 ) ϕ = δm,m0 (764)
2π 0
The above results lead to the selection rules for the z-component of the electron’s
orbital angular momentum

m0 = m ± 1
m0 = m (765)

An alternate derivation of the selection rules for the z-component of the

electron’s orbital angular momentum can be found from considerations of the
commutation relations

[ L̂z , x ] = i h̄ y
[ L̂z , y ] = − i h̄ x
[ L̂z , z ] = 0 (766)

On taking the matrix elements between states with definite z-components of the
angular momenta, one finds

< n0 l0 m0 | [ L̂z , x ] | nlm > = i h̄ < n0 l0 m0 | y | nlm >

< n0 l0 m0 | [ L̂z , y ] | nlm > = − i h̄ < n0 l0 m0 | x | nlm >
< n0 l0 m0 | [ L̂z , z ] | nlm > = 0 (767)

which reduce to

( m0 − m ) < n0 l0 m0 | x | nlm > = i < n0 l0 m0 | y | nlm >

( m0 − m ) < n0 l0 m0 | y | nlm > = − i < n0 l0 m0 | x | nlm >
( m0 − m ) < n0 l0 m0 | z | nlm > = 0 (768)

126
From the last equation, it follows that either m0 = m or that

< n0 l0 m0 | z | nlm > = 0 (769)

On combining the first two equations, one finds that

( m0 − m )2 < n0 l0 m0 | x | nlm > = i ( m0 − m ) < n0 l0 m0 | y | nlm >

= < n0 l0 m0 | x | nlm > (770)

The above equation is solved by requiring that either

( m0 − m )2 = 1 (771)

or
< n0 l0 m0 | x | nlm > = 0 (772)
Hence, the m-selection rules for the electric dipole transitions are ∆m = ± 1, 0.

The l-selection rules for electric dipole transitions.

The selection rules for the magnitude of the electron’s orbital angular mo-
mentum can be found by considering the double commutator
 
[ L̂2 , [ L̂2 , r ] ] = 2 h̄2 r L̂2 + L̂2 r (773)

On taking the matrix elements of this equation between different eigenstates of

the magnitude of the orbital angular momentum, one finds
 2  
l0 ( l0 + 1 ) − l ( l + 1 ) < n0 l0 m0 | r | nlm > = 2 l0 ( l0 + 1 ) + l ( l + 1 ) < n0 l0 m0 | r | nlm >
(774)
Since
 2
0 0
l (l + 1) − l(l + 1) = ( l0 + l + 1 )2 ( l0 − l )2 (775)

and
   
0 0 0 2 0 2
2 l (l + 1) + l(l + 1) = (l + l + 1) + (l − l) − 1 (776)

the above equation is satisfied if, either

< n0 l0 m0 | r | nlm > = 0 (777)

or   
( l0 + l + 1 )2 − 1 ( l0 − l )2 − 1 = 0 (778)

127
The first factor in eqn(778) is always positive when l0 6= l, therefore, the electric
dipole selection rule becomes ∆l = ± 1.

The actual values of the matrix elements can be found from explicit calcu-
lations. The θ-dependence of the matrix elements is governed by the associated
Legendre functions through
s
( 2 l + 1 ) ( l − m )! l
Θlm (θ) = P (cos θ) (779)
2 ( l + m )! m
which obey the recursion relations
l+1 l−1
l Pm (cos θ) − Pm (cos θ)
sin θ Pm−1 (cos θ) = (780)
2l + 1
and
l−1 l+1
l ( l + m ) ( l + m + 1 ) Pm (cos θ) − ( l − m ) ( l − m + 1 ) Pm (cos θ)
sin θ Pm+1 (cos θ) =
2l + 1
(781)
appropriate for the ∆m = ± 1 transitions and
l+1 l−1
l ( l − m + 1 ) Pm (cos θ) + ( l + m ) Pm (cos θ)
cos θ Pm (cos θ) =
2l + 1
(782)
for the constant m transition, ∆m = 0. Using the recursion relations, one
finds that
s
l ( l + m + 2 ) ( l + m + 1 ) l+1
sin θ Θm (θ) = Θm+1
(2l + 1)(2l + 3)
s
( l − m ) ( l − m − 1 ) l−1
− Θm+1
(2l − 1)(2l + 1)
(783)
for ∆m = 1, while for ∆m = − 1 one finds
s
l ( l + m ) ( l + m − 1 ) l−1
sin θ Θm (θ) = Θm−1
(2l + 1)(2l − 1)
s
( l + 2 − m ) ( l + 1 − m ) l+1
− Θm−1
(2l + 1)(2l + 3)
(784)
and for constant m
s
(l + 1 + m)( l + 1 − m ) l+1
cos θ Θlm (θ) = Θm
(2l + 1)( 2l + 3)
s
(l + m)(l − m)
+ Θl−1 (785)
(2l − 1)(2 l + 1) m

128
The coefficients in the above equation have a similar form to the Clebsch-Gordon
coefficients. The dipole matrix elements can be evaluated by taking the matrix
0
elements of the above set of relations with Θlm0 (θ)∗ and then using the orthog-
onality properties. The above three relations give rise to the selection rules for
the magnitude of the orbital angular momentum l

l0 = l ± 1 (786)

Hence, not only have the selection rules on l been re-derived but the angular
integrations have also been evaluated.

What the above mathematics describes is how the spin angular momentum
of the emitted photon is combined with the orbital angular momentum of the
electron in the final state, so that total angular momentum is conserved. This
implies the selection rules which leads to the magnitude of the initial and final
electronic angular momentum l having to satisfy the triangular inequality

l0 + 1 ≥ l ≥ | l0 − 1 | (787)

as required by the rules of combination of angular momentum. The evaluation of

the dipole matrix elements is an explicit example of the Wigner-Eckart theorem.
For this example, the irreducible tensor is the vector V with components V µ
given by
r
± (x ± iy) 4π 1
V = ∓ √ = r Y
2 3 ±1
r
0 4π 1
V = z = r Y (788)
3 0
Then, since the electric dipole carries angular momentum (1, µ), the Wigner-
Eckart theorem reduces to
1
< n0 l0 m0 | V µ | nlm > = √ < l, m; 1, µ | l0 m0 > < n0 l0 | |V | | nl >
2 l0 + 1
(789)
where the first term which represents the angular integration is a Clebsch-
Gordon coefficient and the second factor is the reduced matrix element which
does depend on the form of the particular vector, but is independent of any
choice of coordinate system. Furthermore, the Wigner-Eckart theorem yields
the selection rules for the electric dipole transition

l + l0 ≥ 1 ≥ | l − l0 | (790)

Exercise:

129
 Table 1: Matrix Elements of the Components of the Dipole Moment

l0 m0 x y z

q q
(l+2+m)(l+1+m) (l+2+m)(l+1+m)
m0 = m + 1 1
2 (2l+1)(2l+3) − 2i (2l+1)(2l+3) -
q
(l+1+m)(l+1−m)
l0 = l + 1 m0 = m - - (2l+1)(2l+3)
q q
(l+2−m)(l+1−m) (l+2−m)(l+1−m)
m0 = m − 1 − 12 (2l+1)(2l+3) − 2i (2l+1)(2l+3) -

q q
(l−m)(l−1−m) (l−m)(l−1−m)
m0 = m + 1 − 12 (2l−1)(2l+1)
i
2 (2l−1)(2l+1) -
q
0 0 (l+m)(l−m)
l =l−1 m =m - - (2l−1)(2l+1)
q q
(l+m)(l−1+m) (l+m)(l−1+m)
m0 = m − 1 1
2 (2l−1)(2l+1)
i
2 (2l−1)(2l+1) -

Using the commutation relations for the j-th component of a vector V̂ j with
the i-th component of the orbital angular momentum L̂i ,
X
[ L̂i , V̂ j ] = i h̄ ξ i,j.k V̂ k (791)
k

where ξ i,j,k is the antisymmetric Levi-Civita symbol, show that

 
[ L̂2 , V̂ ] = − i h̄ L̂ ∧ V̂ − V̂ ∧ L̂
 
= − 2 i h̄ L̂ ∧ V̂ − i h̄ V̂ (792)

From the above equation, derive the double commutation relation

   
2 2 2 2 2 2
[ L̂ , [ L̂ , V̂ ] ] = 2 h̄ V̂ L̂ + L̂ V̂ − 4 h̄ L̂ L̂ . V̂

(793)
and that the last term of the above expression is zero if V̂ = r.

The parity selection rule.

In addition to the electronic orbital angular momentum selection rules, there

is a parity selection rule. The parity operation is an inversion through the origin
given by r → − r. The parity operator P̂ has the effect
P̂ ψ(r) = ψ(−r) (794)

130
The parity operator is its own inverse since for any state ψ(r)

P̂ 2 ψ(r) = P̂ ψ(−r)
= ψ(r) (795)

Therefore, the parity operator has eigenvalues p = ±1 for the eigenstates which
are defined by
P̂ φp (r) = p φp (r) (796)
so
P̂ 2 φp (r) = p2 φp (r) = φp (r) (797)
2
which yields p = 1 or p = ± 1. In polar coordinates, the parity operation is
equivalent to a reflection
θ → π − θ (798)
followed by a rotation
ϕ → ϕ + π (799)
In electromagnetic processes, parity is conserved since the Coulomb potential is

z r

r
θ
y
π−θ
π+ϕ ϕ
x

-r

Figure 19: The effect of the parity operator on a displacement vector r, in

spherical polar coordinates.

symmetric under reflection38 . Therefore, the parity operator P̂ commutes with

the Hamiltonian
[ P̂ , Ĥ ] = 0 (800)
and so one can find states | φn > that are simultaneous eigenstates of Ĥ and
P̂.

Ĥ | φn > = En | φn >
P̂ | φn > = pn | φn > (801)
38 The weak interaction does not conserve parity.

131
Inversion transforms vector operators according to

P̂ r P̂ −1 = − r (802)

Hence, for any matrix elements of r between any eigenstates of the parity oper-
ator, one has

< φn0 | r | φn > = − < φn0 | P̂ r P̂ −1 | φn >

= − pn0 pn < φn0 | r | φn > (803)

Therefore, the parity must change in an electric dipole transition

pn 0 pn = − 1 (804)

This is known as the Laporte selection rule for electric dipole transitions39 . The
validity of this selection follows from the fact that inversion commutes with the
orbital angular momentum operator. The spherical harmonics are eigenstates
of the parity operator since

P̂ Yml (θ, ϕ) = ( − 1 )l Yml (θ, ϕ) (805)

This is proved by examining

 
l
Yll (θ, ϕ) ∝ sin θ exp ilϕ (806)

which is seen to be an eigenfunction of the parity operator with eigenvalue

( − 1 )l . All other spherical harmonics with the same value of l have the
same eigenvalue since the lowering operator (like any component of the angular
momentum) commutes with the parity operator. Therefore, one can use the
angular momentum selection rule to show that parity does change in an electric
dipole transition since 0
( − 1 )l+l = − 1 (807)
The Laporte selection rule is satisfied since ∆l = 1.

11.1.4 Angular Distribution of Dipole Radiation

We shall assume that the initial state is polarized so that the electron is in an
electronic state labelled by m, where the axis of quantization is fixed in space.
The decay rate in which a photon of polarization α is emitted into the solid
angle dΩk is given by
3
1 ωnl,n0 l0 X
= 3
dΩk | ˆα (k) . dnlm,n0 l0 m0 |2 (808)
τdΩk 2 π h̄ c α
39 O. Laporte, Z. Physik, 23, 135 (1924).

132
For a photon emitted in the direction k̂
k̂ = (sin θk cos ϕk , sin θk sin ϕk , cos θk ) (809)
the polar polarization vectors are given by
ˆ1 (k) = (cos θk cos ϕk , cos θk sin ϕk , − sin θk )
ˆ2 (k) = (− sin ϕk , cos ϕk , 0) (810)
Therefore, the scalar products of the transition matrix elements of r with the
polarizations are given by
1
ˆ1 (k) . < n0 l0 m0 | r | nlm > = cos θk exp[ − i ϕk ] < n0 l0 m0 | (x + iy) | nlm >
2
1
+ cos θk exp[ + i ϕk ] < n0 l0 m0 | (x − iy) | nlm >
2
− sin θk < n0 l0 m0 | z | nlm > (811)
and
i
ˆ2 (k) . < n0 l0 m0 | r | nlm > = − exp[ − i ϕk ] < n0 l0 m0 | (x + iy) | nlm >
2
i
+ exp[ + i ϕk ] < n0 l0 m0 | (x − iy) | nlm >
2
(812)
Due to the m-selection rules
< n0 l0 m0 | (x + iy) | nlm > ∝ δm0 −m−1
< n0 l0 m0 | (x − iy) | nlm > ∝ δm0 −m+1 (813)
and
< n0 l0 m0 | z | nlm > ∝ δm0 −m (814)
the cross-terms in the square of the matrix elements are zero. Hence, on sum-
ming over the polarizations, one finds that the (θk , ϕk ) dependence of the decay
is governed by the dipole matrix elements through
 
X 1
| ˆα (k) . rnlm,n0 l0 m0 |2 = 1 + cos2 θk | < n0 l0 m0 | (x + iy) | nlm > |2
α
4
 
1
+ 1 + cos θk | < n0 l0 m0 | (x − iy) | nlm > |2
2
4
+ sin2 θk | < n0 l0 m0 | z | nlm > |2 (815)
0
For l = l + 1, the above sum is found to depend on the angular factors
 
m0 2 1 (l + 2 + m)(l + 1 + m)
Il0 =l+1 (θk , ϕk ) = 1 + cos θk δm0 −m−1
4 (2l + 1)(2l + 3)
 
1 (l + 2 − m)(l + 1 − m)
+ 1 + cos2 θk δm0 −m+1
4 (2l + 1)(2l + 3)
(l + 1 + m)(l + 1 − m)
+ sin2 θk δm0 −m (816)
(2l + 1)(2l + 3)

133
Since the z-component of the final electron’s orbital angular momentum is not
measured, m0 should be summed over. The angular distribution of the emitted
radiation for the l0 = l + 1 transition when neither the polarization nor the final
state m0 value are measured is given by
X 0 1 (l + 2)(l + 1) + m2 (l + 1)2 − m2
Ilm
0 =l+1 (θk , ϕk ) =( 1 + cos2 θk ) + sin2 θk
2 (2l + 1)(2l + 3) (2l + 1)(2l + 3)
m0
(817)
This factor determines the angular dependence of the emitted electromagnetic
radiation, which clearly depends on the value of m specifying the initial elec-
tronic state. On rearranging the expression, one finds that the anisotropy is
governed by the factor
X 0 (l + 2)(l + 1) + m2 1 l(l + 1) − 3 m2
Ilm
0 =l+1 (θk , ϕk ) = + sin2 θk
(2l + 1)(2l + 3) 2 (2l + 1)(2l + 3)
m0
(818)
which shows that for m = 0 the photons are preferentially emitted perpendicular
to the direction of quantization axis since this maximizes the overlap between
the polarization and the dipole matrix element. In the opposite case of large
values of m2 [ 3 m2 > l (l + 1) ], one finds that the photons are preferentially
emitted parallel (or anti-parallel) to the axis of quantization. On integrating
over all directions of the emitted photon, one obtains
Z
1 X 0 2 (l + 1)
dΩk Ilm
0 =l+1 (θk , ϕk ) = (819)
4π 0
3 (2l + 1)
m

The independence of the result on m follows since, in this case, there are no an-
gular correlations and the choice of direction of quantization of m is completely
arbitrary. The total decay rate for an electron in a state with fixed m due to
an l0 = l + 1 transition is given by
2
4 e2 ωnl,n
3  Z ∞
1 0 l0 (l + 1)  2 ∗

= dr r Rn0 l+1 (r) r Rnl (r)  (820)
τl0 =l+1 3 h̄ c3 (2l + 1)  0

for l0 = l + 1. This decay rate would be measured in experiments in which

neither the final state of the electron nor the final photon state is measured.

However, if the initial electronic state is unpolarized, then one should sta-
tistically average over the initial m. In this case, the emitted radiation becomes
isotropic
l
1 X X 0 2 (l + 1)
Ilm
0 =l+1 (θk , ϕk ) = (821)
2l + 1 0
3 (2l + 1)
m=−l m

since
l
1 X 1
m2 = l(l + 1) (822)
2l + 1 3
m=−l

134
Hence, if the initial electronic state is unpolarized, the electromagnetic radiation
is isotropic. The decay rate for the l0 = l +1 transition starting with a statistical
distribution of m values is given by
2
4 e2 ωnl,n
3  Z ∞
1 0 l0 (l + 1)  2 ∗

= dr r Rn0 l+1 (r) r Rnl (r)  (823)
τl0 =l+1 3 h̄ c3 (2l + 1)  0
for l0 = l + 1. This is the same result that was previously obtained for the decay
rate of a level with a specific m value, when the m0 value of the final state and
the polarization or direction of the emitted photon are not measured.

For the case where l0 = l − 1, one finds that the decay rate involves the
angular factor
 
m0 2 1 (l − m)(l − 1 − m)
Il0 =l−1 (θk , ϕk ) = 1 + cos θk δm0 −m−1
4 (2l − 1)(2l + 1)
 
2 1 (l − 1 + m)(l + m)
+ 1 + cos θk δm0 −m+1
4 (2l − 1)(2l + 1)
(l + m)(l − m)
+ sin2 θk δm0 −m (824)
(2l − 1)(2l + 1)
which on summing over the final values of m0 yields the angular dependence of
the radiation field
X 0 1 l(l − 1) + m2 l2 − m2
Ilm
0 =l−1 (θk , ϕk ) = ( 1 + cos2 θk ) + sin2 θk
2 (2l − 1)(2l + 1) (2l − 1)(2l + 1)
m0
(825)
The anisotropy of the emitted radiation is determined by the factor
X 0 l(l − 1) + m2 1 l(l + 1) − 3 m2
Ilm0 =l−1 (θk , ϕk ) = + sin2 θk (826)
0
(2l − 1)(2l + 1) 2 (2l − 1)(2l + 1)
m

which shows that for m = 0 the photons are preferentially emitted perpendicular
to the direction of quantization axis since this maximizes the overlap between
the polarization and the dipole matrix element. In the opposite case of larger
m2 [ 3 m2 > l (l + 1) ], one finds that the photons are preferentially emitted
parallel (or anti-parallel) to the axis of quantization.

Again it is noted that if the initial state is unpolarized, so that m has to be

averaged over, then the radiation field is isotropic since
l
1 X X 0 2 l
Ilm
0 =l−1 (θk , ϕk ) = (827)
2l + 1 3 (2l + 1)
m=−l m0

Therefore, the decay rate in which the photon is emitted in any direction is
given by the expression
2
4 e2 ωnl,n
3  Z ∞
1 0 l0 l  2 ∗

=  dr r R n 0 l−1 (r) r Rnl (r)
 (828)
τl0 =l−1 3 h̄ c3 (2l + 1) 
0


135
for l0 = l − 1.

Classical Interpretation.

The quantum mechanical results for the angular distribution of the radiation
can be understood in terms of a simple classical model of the atom. In Bohr’s
model, a single electron orbits a central nucleus to which it is bound by the
attractive Coulomb potential. We shall assume that the radius of the orbit is a
and that the electron is performing a circular orbit in the x − y plane. Since the
direction of the electron’s orbital angular momentum is aligned with the z-axis,
it corresponds to the case where m ≈ l and l  1. In this case, the electron
has an oscillating dipole moment given by
 
d(t) = q a cos ω t êx + sin ω t êy
   
= q a <e êx − i êy exp i ω t (829)

This rotating dipole moment can be decomposed into two orthogonal linear
dipole moments which oscillate out of phase with each other. It should be

m=l
-
e
d(t) ωt
x

Figure 20: A classical electron orbiting in the x-y plane (m = l) can be consid-
ered as producing two perpendicular linearly-oscillating electric-dipole moments.

recalled that a classical oscillating (linear) electric dipole moment radiates power
P (ω) into a solid angle dΩk with a distribution given by
   4
dP c ω
= | d |2 sin2 Θkd (830)
dΩk linear 8π c
where Θkd is the angle between the detector and the direction of the electric
dipole. On considering the radiation from the atom to be generated from two
orthogonal linear oscillating dipoles, one finds
   4  
dP c ω 2 2 2
= |d| sin Θkx + sin Θky (831)
dΩk dipole 8π c

136
 dΩk

ez k

Θky
θk
ey
Θkx

ex

e- m=l

Figure 21: The polarization of the radiated electromagnetic field for an electron
orbiting in the x-y plane (m = l) can be comprehended in terms of the classical
radiation emanating from two linearly oscillating electric-dipole moments. The
angles Θkx and Θky , respectively, are the angles between the emitted radiation
and the x-axis and the angle subtended by the emitted radiation and the y-axis.

which on using

cos Θkx = sin θk cos ϕk

cos Θky = sin θk sin ϕk (832)

becomes
   4  
dP c ω 2 2
= |d| 1 + cos θk (833)
dΩk dipole 8π c

Since the energy of the emitted photon is given by h̄ ω, one finds the angular
dependence of the semi-classical prediction of the decay rate is given by
3 
e2
 
1 ωa
= 1 + cos2 θk dΩk (834)
τdΩk 8 π h̄ a c
The polarization vector is parallel to the direction of the electric field, which in
turn is given by the direction of the oscillating dipole that produced it. Hence,
a detector which is arranged to accept radiation travelling in the direction k̂ will
detect polarizations that are found by projecting the electron’s orbit onto the
plane perpendicular to k̂. For example, in this case where the electron’s orbit
is in the x − y plane, so radiation along the z-axis will be circularly-polarized,
whereas radiation in the x − y plane will be linearly-polarized.

137
 Circular

e-
k

Linear

Figure 22: The polarization at a field point can be determined by considering

the projection of the electrons orbit on a plane perpendicular to the direction
of emission k. The polarization vector of the classical EM wave follows the
projected orbit of the dipole moment.

The angular dependence of the decay rate follows directly from the expres-
sions of eqn(817) and eqn(825) by setting m ≈ l  1, replacing the radial matrix
elements of r by a, adding the expressions and inserting them into eqn(808). The
analysis shows that quantum mechanics reproduces the classical limit correctly,
as is expected from the correspondence principle.

11.1.5 The Decay Rate from Dipole Transitions.

The decay rate due to dipole transitions includes processes in which photons
of all polarizations are emitted in all directions. Accordingly, the decay rate
is found by summing over all polarizations and integrating over the directions
of the emitted photon. For a spherically symmetric system, the energy will
be independent of the z-component of the orbital angular momentum. In this
case, one should sum over all values of m0 corresponding to the degenerate final
states. On summing over all final states corresponding to a specific l0 value, that
is on summing over m0 where m0 = m, m ± 1, one finds that the transition
rate can be expressed as
(l+1) 2
4 e2 ωnl,n
3 ∞
    Z
1 0 l0
(2l+1)
 2

= l
 dr r Rn0 l0 (r) r Rnl (r)  (835)
τ 3 h̄ c3 (2l+1)

0

for 
l+1
l0 = (836)
l−1
It should be noted that, for a fixed l0 , the lifetime of the state | nlm > is inde-
pendent of the value of m. This is expected since the choice of the quantization

138
Table 2: Radial wave functions Rnl (ρ) forRa Hydrogenic-like atom, where ρ =
Zr ∞
a . The functions are normalized so that 0 dρ ρ2 Rnl Rnl = 1.
 
n = 1 l = 0 2 exp − ρ

   
√1 ρ ρ
n = 2 l = 0 2
1 − 2 exp − 2

 
1 ρ
l = 1 2
√
6
ρ exp − 2

   
2 2 2 2 ρ
n = 3 l = 0 3 1 − 3 ρ + 27 ρ exp − 3
32

5
   
22 ρ ρ
l = 1 7 1 − 6 ρ exp − 3
32

3
 
22 2 ρ
l = 2 9 √ ρ exp − 3
32 5

139
 R∞
Table 3: Values of | 0
dr r2 Rnl r Rn0 l−1 |2 in atomic units.

n, l n0 , l − 1

np 1s 28 n7 (n − 1)2n−5 (n + 1)−2n−5

2s 217 n7 (n2 − 1)(n − 2)2n−6 (n + 2)−2n−6

3s 28 37 n7 (n2 − 1)(n − 3)2n−8 (7n2 − 27)2 (n + 3)−2n−8

nd 2p 219 3−1 n9 (n2 − 1)(n − 2)2n−7 (n + 2)−2n−7

3p 211 39 n9 (n2 − 1)(n2 − 4)(n − 3)2n−8 (n + 3)−2n−8

nf 3d 213 39 5−1 n11 (n2 − 1)(n2 − 4)(n − 3)2n−9 (n + 3)−2n−9

direction is completely arbitrary.

There are no selection rules associated with the radial integration in the
dipole matrix elements Z ∞
dr r2 Rn0 l−1 r Rnl (837)
0
The radial part of the dipole matrix element can be expressed in terms of the
hypergeometric function F (a, b, c) via
Z ∞
dr r2 Rn0 l−1 r Rnl
0
0
s 0
a (−1)n −l (n + l)!(n0 + l − 1)! (4n0 n)l+1 (n − n0 )n +n−2l−2
 
=
4(2l − 1) (n − l − 1)!(n0 − l)! (n0 + n)n0 +n
2
4n0 n
 0
4n0 n
 
0 n −n 0
× F (l + 1 − n, l − n , 2l, − 0 )− F (l + 1 − n, l − n , − 0 )
(n − n)2 n0 + n (n − n)2
(838)

Simple analytic expressions for the squares of the matrix elements for small val-
ues of (n0 , l) are shown in Table(3).

The radial integrations were evaluated by Schrödinger40 using the generating

40 E. Schrödinger, Ann. der Phys. 79, 362 (1926).

140
function expansion for the Laguerre polynomials. Eckart41 and Gordon42 have
calculated these dipole matrix elements by other means. In general, the lifetime
of the hydrogenic states increases with increasing n, varying roughly as n3 for
a fixed value of l. The decrease in the dipole matrix elements with increasing n
is simply due to the increasing numbers of nodes in the radial wave functions.

The magnitude of the decay rate is estimated as

 3  2 
1 c ωa e
∼ (839)
τ a c h̄ c

where the magnitude of the dipole matrix element is estimated as e a where a

is the Bohr radius. On setting h̄ω equal to the electrostatic energy of hydrogen,
the remaining factor is estimated to have the magnitude
   2
ωa e
∼ (840)
c h̄ c

where the length scale a has dropped out. Hence, as

 2
e 1
≈ (841)
h̄ c 137.0359979

one finds that the decay rate is given by

 2 4
1 c e
∼ (842)
τ a h̄ c

so the decay time is approximately eight orders of magnitude larger than the
time taken for the photon to cross the atom. When averaged over l, the electric
dipole decay rate is given by
1 X (2l + 1) 9
∝ ∼ n− 2 (843)
τn n5
l

so, as seen in Table(4), the decay is slower for the higher energy levels.

11.1.6 The 2p → 1s Electric Dipole Transition Rate.

Consider the decay of the 2p state (with m = 0) to the 1s state in the hydrogen
atom. As can be seen from Table(2), the initial state is described by an electronic
wave function
   r
1 r 1 r 3
ψ2p (r) = √ exp − cos θ (844)
2 6a 3 a 2 a 4π
41 C. Eckart, Phys. Rev. 28, 927 (1926).
42 W. Gordon, Ann. der Phys. 2, 1031 (1929).

141
Table 4: Electric Dipole Transition Rates for Hydrogen, in units of 108 sec−1 .

Initial Final n=1 n=2 n=3

2p ns 6.25 - -

3s np - 0.063 -
3p ns 1.64 0.22 -
3d np - 0.64 -

4s np - 0.025 0.018
4p ns 0.68 0.095 0.030
4p nd - - 0.003
4d np - 0.204 0.070
4f nd - - 0.137

and the final state electronic is given by

 
2 r 1
ψ1s (r) = √ exp − √ (845)
a3 a 4π
where the length scale a is the Bohr radius

h̄2
a = (846)
m e2
The decay rate in the Fermi-Golden rule, evaluated in the dipole approximation,
is given by
1 3
4 ω1,2 d∗1s,2p . d1s,2p
= (847)
τ 3 h̄ c3
The frequency is evaluated from

h̄ ω12 = E2p − E1s

m e4
 
1
= 1 −
2 h̄2 4
3 e2
= (848)
8 a
Hence,
 3  2 3 2 2
 d1s,2p 2
 
1 4 3 e e a
=  
τ 3 8 h̄ a h̄ c3  ea 

142
 4 2
e2

c  d1s,2p 

9
= (849)
128 h̄ c a  ea 
Therefore, the scattering rate is determined by the ratio ac but also is modified
by the fourth power of the dimensionless electromagnetic coupling strength
 2
e 1
≈ (850)
h̄ c 137.0359979
The smallness of this factor allows us to only consider the Fermi-Golden rule
expression for the decay rate. The dimensionless dipole matrix elements are
expected to be non-zero, since they obey the selection rules. They are non-zero,
as can be directly verified by performing an integration. The only non-zero
dipole matrix element originates from the z-component of the dipole
Z
∗
d1s,2p = e d3 r ψ1s (r) r ψ2p (r) (851)

since only the z-component satisfies the ∆m = 0 selection rule. The angular
integration is evaluated as
√ Z 2π Z π √
3 3 2
dϕ dθ sin θ cos2 θ = 2π
4π 0 0 4 π 3
1
= √ (852)
3
and the radial integration yields
Z ∞ ∞
r4
Z  
∗ 2 3 r
dr r2 R1s (r) r R2p (r) = √ dr exp −
0 2
6 0 a3 2 a
 5 Z ∞  
a 2 4
= √ dx x exp − x
6 3 0
 5  5
4! 2 √ 2
= a √ = 4a 6 (853)
6 3 3
Hence, the magnitude of the dipole matrix element is evaluated as
 5
d1s,2p √ 2
= 4 2 (854)
ea 3
Therefore, the dipole allowed decay rate is given by
 8  2 4
1 2 e c
= (855)
τ 3 h̄ c a

Hence, the time scale τ is of the order of 10−10 seconds. The exact value of the
decay time is calculated to be 1.6 × 10−9 seconds.

143
11.1.7 Electric Quadrupole and Magnetic Dipole Transitions.
Consider decays such as the 3d state (with m = 0) to the 1s state in the hydrogen
atom. Since, in this transition, the change in the electron’s angular momentum
is two units, the transition is forbidden in the dipole approximation. Therefore,
the transition rate is evaluated by keeping the next order term in the expansion
 
exp − i k . r ≈ 1 − i k . r + ... (856)

The second term in the expansion describes electric quadrupole and magnetic
dipole transitions.

The matrix elements that have to be evaluated are of the form

< n0 l0 m0 | ( k . r ) ( ˆα (k) . p̂ ) | nlm > (857)

This shall be written as the sum of two terms, with different symmetries with re-
spect to interchange of r and p. These two terms will describe electric quadrupole
and magnetic dipole transitions. The matrix elements are written as the sum
of a term symmetric under the interchange of r and p̂ and a term that is anti-
symmetric
 
1
( k . r ) ( ˆα (k) . p̂ ) = ( k . r ) ( ˆα (k) . p̂ ) + ( k . p̂ ) ( ˆα (k) . r )
2
 
1
+ ( k . r ) ( ˆα (k) . p̂ ) − ( k . p̂ ) ( ˆα (k) . r )
2
(858)

The first term represents the matrix elements for the electric quadrupole tran-
sitions43 , and the second term represents the matrix elements for the magnetic
dipole transitions. The first term can be written as the scalar products of a
symmetric dyadic  
k. r p̂ + p̂ r . ˆα (k) (859)

The scalar products are organized such that the left most vector outside the
parenthesis forms a scalar product with the left most vectors within the paren-
thesis, and likewise with the right most vectors. The electronic matrix elements
only involve the dyadic operator, as the wave vector and polarization vectors
are properties of the photon. The matrix elements
 
0 0 0
< nlm | r p̂ + p̂ r | nlm > (860)

are evaluated by first noting that

[ r , p̂2 ] = 2 i h̄ p̂ (861)
43 J. A. Gaunt and W. H. McCrea, Proc. Camb. Phil. Soc. 23, 930 (1927).

144
which allows the momentum operator to be written as
im
p̂ = [ Ĥ , r ] (862)
h̄
Therefore, the matrix elements of the dyadic can be expressed in the form of
the matrix elements of the commutator with the dyadic
 
0 0 0 im
< nlm | r p̂ + p̂ r | nlm > = < n0 l0 m0 | r [ Ĥ , r ] + [ Ĥ , r ] r | nlm >
h̄
im
= < n0 l0 m0 | [ Ĥ , r r ] | nlm >
h̄
( En0 l0 m0 − Enlm )
= im < n0 l0 m0 | r r | nlm >
h̄
= i m ωn0 ,n < n0 l0 m0 | r r | nlm > (863)

The decay rate in the Fermi-Golden rule, evaluated in the electric quadrupole
approximation, is given by
2 Z
m2 c2 ωk2 X

1 e
= d3 k | k . < n0 l0 m0 | r r | nlm > . ˆα (k) |2
τ mc 8 π ωk α
× δ( Enlm − En0 l0 m0 − h̄ ωk )
5 Z
e2

ωnl,n0 l0 X
= dΩk | k̂ . < n0 l0 m0 | r r | nlm > . ˆα (k) |2
8 π h̄ c α
(864)

where, in the second line k is restricted to have the magnitude

ωnl,n0 l0
k = (865)
c
The frequency is evaluated from

h̄ ωnl,n0 l0 = Enl − En0 l0

2
m e4 e2

∼ = m c2
h̄2 h̄ c
e2
∼ (866)
a
Hence,
5
e2 e2
 Z
1 X
∼ dΩk | k̂ . < n0 l0 m0 | r r | nlm > . ˆα (k) |2
τ h̄ a c 8 π h̄ α
 2 6
e c
∼ (867)
h̄ c a

145
Therefore, the scattering rate is determined by the ratio ac but also is modified
by the sixth power of the dimensionless electromagnetic coupling strength
 2
e 1
≈ (868)
h̄ c 137.0359979

The smallness of this factor allows us to only consider the Fermi-Golden rule
expression for the decay rate. The dimensionless quadrupole matrix elements
are expected to be non-zero, since they obey the selection rules which involve
the exchange of two units of angular momentum. They are non-zero, as can
be directly verified by performing an integration. Therefore, the quadrupole
allowed decay rate is given by
6
e2

1 c
∼ (869)
τ h̄ c a

Hence, the time scale τ is expected to be of the order of 10−6 seconds44 .

This type of transition is known as an electric quadrupole transition. Because

of the transversality condition

k̂ . ˆα (k) = 0 (870)

one can add a diagonal term to the dyadic without affecting the result. A
diagonal term with a magnitude that makes the resulting dyadic traceless is
added to the dyadic, leading to the expression
 
1
Qi,j = e xi xj − δi,j | r |2 (871)
3

Therefore, the transition rate is governed by the electric quadrupole tensor

< n0 l0 m0 | Qi,j | nlm > (872)

The symmetric dyadic Qi,j has six inequivalent components, which because of
the restriction that the dyadic is traceless, can be reduced to five independent
components. Due to the transformational properties of the dyadic under ro-
tation, it can be expressed as a linear combination of the spherical harmonics
Ym2 (θ, ϕ) and nothing else45 . This can be seen from rewriting the quadrupole
tensor Q̃  
2
xx − r3 xy xz
Q̃ 2
=  yx yy − r3 yz (873)
 
e

2
r
zx zy zz − 3
44 This estimate will be modified upwards by several orders of magnitude, due to the presence

of a large dimensionless factor that was not accounted for.

45 The transformational properties of the dyadic follow immediately from the transforma-

tional properties of the vector r

146
in terms of spherical polar coordinates
2 2
 1 1 2 1

Q̃ 2 sin θ cos 2ϕ − 6 (3 cos θ − 1) 2 sin θ sin 2ϕ sin θ cos θ cos ϕ
1 2 2
=  2 sin θ sin 2ϕ − 2 sin θ cos 2ϕ − 16 (3 cos2 θ − 1)
1
sin θ cos θ sin ϕ 
e r2 1 2
sin θ cos θ cos ϕ sin θ cos θ sin ϕ 3 (3 cos θ − 1)
(874)
The presence of states with orbital angular momentum of only two makes the
dyadic an irreducible second rank tensor. Application of the Wigner-Eckart
theorem to an irreducible second rank tensor results in the electric quadrupole
selection rules
l + l0 ≥ 2 ≥ | l − l0 | (875)
The angular momentum carried away by the photon consists of the spin-one
carried away by the photon in addition to the component of the photon’s wave
function described by the spherical Bessel function j1 (kr) ∼ k r which carries
off one unit of orbital angular momentum. In addition to the angular momen-
tum selection rules, there are parity selection rules for the electric quadrupole
transitions. Since the parity operator satisfies

P̂ r P̂ −1 = − r (876)

then the electric quadrupole matrix elements satisfy

< n0 | r r | n > = < n0 | P̂ r r P̂ −1 | n >

= pn 0 pn < n0 | r r | n > (877)

Therefore, the parity does not change in an electric quadrupole transition as

pn0 pn = 1.

The magnetic dipole matrix elements are given by

 
1
( k . r ) ( ˆα (k) . p̂ ) − ( k . p̂ ) ( ˆα (k) . r ) (878)
2
which can be re-written as
 
1
( k ∧ ˆα (k) ) . ( r ∧ p̂ ) (879)
2
The first term is of the form

B = ∇ ∧ A → k ∧ A → ∇ ∧ ˆα (k) (880)

and the second term

r ∧ p (881)
is the orbital angular momentum. The orbital angular momentum produces the
orbital magnetic moment given by
e
µ = (r ∧ p) (882)
2mc

147
The magnetic dipole transition should be extended from orbital angular mo-
mentum to include the spin magnetic moment which is of the same order
 
e h̄
σ . ( k ∧ ˆα (k) )
2mc
 
e
( r ∧ p ) . ( k ∧ ˆα (k) ) (883)
2mc
since orbital angular momentum is quantized in units of h̄. The angular mo-
mentum selection rule for the magnetic dipole transition is given by

∆l = 0 (884)

and
1 ≥ | ∆m | (885)
also parity does not change.

Terms with higher-order orbital angular momentum that occur in the ex-
pansion of the photon’s wave function exp[ i k . r ] can be found by using the
Rayleigh expansion. The terms with orbital angular momentum l are propor-
tional to the spherical harmonics jl (kr) which vary as (kr)l when kr → 0, as is
found from the expansion of the exponential term. The presence of the extra
factors k l in the matrix element has the result that the electric 2s -th multi-pole
transition rates are found to vary as
 2s+1
1 ωn,n0
∝ a2s (886)
τ c
where s is the magnitude of the change in the electronic orbital angular mo-
mentum, which satisfies the inequality

( l + l0 ) ≥ s ≥ | l0 − l | (887)

The extra factors from the photon’s angular momentum results in an overall de-
2
crease in the electric multi-pole transition rate by a factor of ( h̄e c )2s . It should
also be noted that the relative strength of the higher-order electric multi-pole
transitions increase more rapidly with Z than the electric dipole transitions.
Therefore, it is frequently found that the quadrupole transitions cannot be ne-
glected for the heavy elements. Alternatively, higher-order multipole transitions
do become important in the x-ray region, since in this region the wavelength of
the radiation is comparable to the spatial extent of the charged particle’s wave
function.

11.1.8 The 3d → 1s Electric Quadrupole Transition Rate

The transition 3d → 1s is forbidden to occur via the dipole process, since it
involves a change of l by two units. It may occur as an electric quadrupole

148
transition. The electric quadrupole transition rate can be expressed as
 5 Z
1 1 ω X
= dΩk | k̂ . < 1s | Q̃ | 3d > . ˆα (k) |2 (888)
τ 8 π h̄ c α

where Q̃ represents the quadrupole tensor. The frequency factor can be evalu-
ated as    2
ω 4 e
= (889)
c 9 a h̄ c
hence, the rate can be expressed in the form
  5  2 6 Z
1 c 4 e X 1
e | 3d > . ˆα (k) |2
= dΩk 2 a4
| k̂ . < 1s | Q
τ 8πa 9 h̄ c α
e
(890)
We shall consider the transition from the m = 0 state of the 3d level to the 1s
state. As can be easily shown, the matrix elements of quadrupole tensor for this
transition are diagonal and are given by
− Q2zz
 
0 0
< 1s | Q̃ | 3d > = < 1s |  0 − Q2zz 0  | 3d > (891)
0 0 Qzz
Therefore, the transition matrix elements are of the form
Z X 1
dΩk e | 3d > . ˆα (k) |2
| k̂ . < 1s | Q
e2 a4
α
Z X | < 1s | Qzz | 3d > |2  2
1 1
= dΩk k̂z ˆα (k)z − k̂x ˆα (k)x − k̂y ˆα (k)y
α
e2 a4 2 2
(892)
The direction of the emitted photon k̂ is expressed as
k̂ = (sin θk cos ϕk , sin θk sin ϕk , cos θk ) (893)
and the polarization vectors are given by
ˆ1 (k) = (cos θk cos ϕk , cos θk sin ϕk , − sin θk )
ˆ2 (k) = (− sin ϕk , cos ϕk , 0) (894)
Thus for the m = 0 level, one finds that the integral over the angular distribution
is given by
Z X 1
dΩk e | 3d > . ˆα (k) |2
| k̂ . < 1s | Q
e2 a4
α
2
| < 1s | Qzz | 3d > |2
Z 
3
= dΩk − sin θ k cos θ k
e2 a4 2
3 | < 1s | Qzz | 3d > |2
= 4π (895)
10 e2 a4

149
The scattering rate becomes
  5  2 6
1 3 4 c e | < 1s | Qzz | 3d > |2
= (896)
τ 20 9 a h̄ c e2 a4

The dimensionless quadrupole matrix elements are evaluated as the product of

an angular integral and a radial integral

< 1s | Qzz | 3d >

Z
1
2
= dΩ Y00 (θ, ϕ)∗ ( 3 cos2 θ − 1 ) Y02 (θ, ϕ)
ea 3
Z ∞
∗ r2
× dr r2 R1s (r) 2 R3d (r)
0 a
r   92
1 4 3
= − 4 (897)
3 5 4

Finally, one finds the resulting expression for the quadrupole decay rate of the
3d state with m = 0  2 6
1 1 c e
= (898)
τ 3600 a h̄ c
which is evaluated as 228 sec−1 . From the above analysis, it is seen that angular
distribution for the emitted photon is governed by the factor

cos2 θk sin2 θk (899)

and the intensity is largest for the cone with θk ≈ 0.28 π or 0.72 π. This angular
dependence of the emitted radiation is the same as found by considering the
radiation from an oscillating classical quadrupole, for which the radiated power
is given by
   6
dP c ω
= Q2 cos2 θk sin2 θk (900)
dΩk quad 288 π c

However, the angular distribution of the emitted quadrupole radiation is to

be contrasted with the ∆m = 0 decay of a 2p electron, that has an angular
distribution of emitted photons given by

sin2 θk (901)
π
which is maximum for θk = 2.

The 3d → 1s quadrupole decay rate will be dwarfed by dipole allowed cas-

cade emission processes such as 3d → 2p followed by 2p → 1s. Therefore, the
intensity of the emitted light corresponding to the 3d → 1s process is expected
to be extremely weak. However, the quadrupole line is expected to be much
more readily observed in absorption spectra.

150
 m=0

Figure 23: The angular distribution of quadrupole radiation for the ∆m = 0

transition, as a function of polar angle θk .

11.1.9 Two-photon decay of the 2s state of Hydrogen.

The 2s state of the hydrogen atom can not decay via the paramagnetic interac-
tion since, as it can be shown that the matrix elements that govern the emission
intensity vanish
 
< 1s | ˆα (k) . p̂ exp − i k . r | 2s > = 0 (902)

First on integrating by parts, the matrix elements can be written as

 
i h̄ < 1s | ˆα (k) . ∇ exp − i k . r | 2s >
 
= + i h̄ < 2s | exp + i k . r ˆα (k) . ∇ | 1s >∗

(903)

On utilizing the expression for the 1s wave function

 
1 r
ψ1s (r) = √ exp − (904)
π a3 a

one finds that

r ψ1s (r)
∇ψ1s (r) = − (905)
r a

151
Hence, the transition matrix element is given by
Z ∞  
h̄ ˆα (k) . r ∗
−i d3 r ψ2s (r) exp − i k . r ψ1s (r) (906)
a 0 r

The matrix elements can be simply evaluated in spherical polar coordinates if

one chooses the direction of k as the polar axis. The plane-wave, therefore, only
depends on the z-component of r and since

ˆα (k) . k = 0 (907)

the factor ˆα (k) . r only depends on x and y and is antisymmetric with respect
to the transformations x → − x and y → − y. All other factors are even
functions of x and y. On integrating over the directions in the x − y plane, one
finds that the integral is identically zero.

The above result could have been (partially) anticipated by considering the
selection rules. The electric dipole transition is forbidden by parity. The mag-
netic dipole transition is zero in this non-relativistic treatment. All magnetic
and electric quadrupole and higher multipole transitions are forbidden by an-
gular momentum conservation.

The 2s state decays via two-photon emission which is described by the dia-
magnetic interaction and by the effect of the paramagnetic interaction taken to
second-order in time-dependent perturbation theory. Since only the part of the
paramagnetic interaction that creates a photon is involved, for our purposes the
paramagnetic interaction can be replaced by
1
q X 2 π h̄ c2 2
  
Ĥpara → − p̂ . ˆα (k) a†k,α exp − i k . r (908)
mc V ωk
k,α

Likewise, the diamagnetic interaction can be replaced by

hk

n
n'

Figure 24: The one-photon emission part of the paramagnetic interaction.

152
 q2 2 π h̄ c2 ˆα (k) . ˆα0 (k 0 ) †
   
†
X
0
Ĥdia → √ a a
k,α k0 ,α0 exp − i ( k + k ) . r
2 m c2 V ωk ωk 0
k,α;k0 ,α0
(909)
for two-photon emission. The system is assumed to be initially in an eigenstate

(k,α)

(k',α')

2s
1s

Figure 25: The two-photon emission part of the diamagnetic interaction.

of the unperturbed Hamiltonian | n > but, due to the interaction Ĥint makes
transitions to states | n0 >. These states are to be considered as products
of the electronic states and the states of the electromagnetic cavity. Following
the usual procedure of time-dependent perturbation theory, the above state
| ψn > can be decomposed in terms of a complete set of non-interacting energy
eigenstates | n > via
 
X i
| ψn > = Cn0 (t) exp − En0 t | n 0 > (910)
0
h̄
n

where Cn0 (t) are time-dependent coefficients. The probability of finding the
system in the final state | n0 > at time t is then given by |Cn0 (t)|2 . The rate
at which the transition n → n0 occurs is then given by the time-derivative of
|Cn0 (t)|2 .

It shall be assumed that the interaction is slowly turned on when t → − ∞.

The interaction can be reduced to zero at large negative times by introducing a
multiplicative factor of exp[ + η t ] in the interaction, where η is an infinitesi-
mally small positive constant. To first-order in the diamagnetic interaction, one
finds
q2 2 π h̄ c2
    
(1) −i 0 0
Cn0 (t) = < n | exp[ − i ( k + k ) . r ] | n > √
h̄ 2 m c2 ωk ωk 0 V
Z t  
i
× 2 ˆα (k) . ˆα0 (k 0 ) dt0 exp ( h̄ ω 0 + h̄ ω + En0 − En ) t0
−∞ h̄
(911)

where ω = c k and ω 0 = c k 0 are the energies of the two photons in the final
state. It should be noted that, since we have evaluated the photonic part of the

153
initial and final state, the labels n and n0 , now only describe the electronic part
of the inital and final states. The small quantity η has been absorbed as a small
imaginary part to the initial state energy

En → En + i η h̄ (912)

The paramagnetic interaction is of order of q and the diamagnetic interaction

is of order q 2 . Thus, to second-order in q, one must include the diamagnetic
interaction and the paramagnetic interaction to second-order. There are two
second-order terms which represent:

(a) emission of the photon (k, α) followed by the emission of a photon (k 0 , α0 ).

(b) emission of a photon (k 0 , α0 ) followed by the emission of the photon (k, α).

(k,α) (k',α') (k',α') (k,α)

n'' n''
n n' n n'

Figure 26: The two-photon emission processes to the paramagnetic interaction

to second-order.

The second-order contribution to the transition amplitude is given by

2  2   Z t Z t0
2 π h̄ c2

(2) −i q 0
Cn0 (t) = √ dt dt00
h̄ mc ωk ωk 0 V −∞ −∞
 X
i i
exp[ (En0 + h̄ω 0 − En00 ) t0 ] exp[ (En00 + h̄ω − En ) t00 ]
00
h̄ h̄
n
× < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ | nlm >
X i i
+ exp[ (En0 + h̄ω − En00 ) t0 ] exp[ (En00 + h̄ω 0 − En ) t00 ]
00
h̄ h̄
n

× < n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα0 (k 0 ) . p̂ | nlm >

(913)

The earliest time integration can be evaluated leading to

2   Z t
2 π h̄ c2
 
(2) −i q X i
Cn0 (t) = √ dt0 exp[ (h̄ω 0 + En0 − En − h̄ω) t0 ]
h̄ mc ωk ωk 0 V −∞ 00
h̄
n

154
 < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ | nlm >


( En − En00 − h̄ ω )
< n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα0 (k 0 ) . p̂ | nlm >

+ (914)
( En − En00 − h̄ ω 0 )

as long as the denominators are non-vanishing.

(1) (2)
The coefficients Cn0 (t) and Cn0 (t) have the same type of time-dependence.
The remaining integration over time yields
  Z t  
−i 0 i 0 0
dt exp (h̄ω + h̄ω + En0 − En − ih̄η) t
h̄ −∞ h̄
exp[ i
h̄ (h̄ω 0 + h̄ω + En0 − En − ih̄η) t ]
= − (915)
(h̄ω 0 + h̄ω + En0 − En − ih̄η)

The transition rate is given by

 2 
1 ∂  (1)
 C 0 (t) + C (2)

=  n n0 (t)
 (916)
τ ∂t 

but the time-dependence of the squared modulus is contained in the common

factor
 
 exp[ h̄i (h̄ω 0 + h̄ω + En0 − En − ih̄η) t ] 2
  exp 2 η t
  =
 (h̄ω 0 + h̄ω + En0 − En − ih̄η)  (h̄ω 0 + h̄ω + En0 − En )2 + h̄2 η 2
(917)
Since the momenta and polarizations of the emitted photons are not measured,
the rate is summed over (k, α) and (k 0 , α0 ). Therefore, the transition rate is
given by the expression
 
2 η exp 2 η t
1 X
= 0 + h̄ω + E 0 − E )2 + h̄2 η 2
M2 (918)
τ 0 (h̄ω n n
k,α:k ,α0

where the matrix elements M are due to the combined effect of the diamagnetic
interaction and the paramagnetic interaction taken to second-order. That is,

M = < 1s kα, k 0 α0 | Ĥdia | 2s >

X < 1s k α, k 0 α0 | Ĥpara | n00 l00 m00 k 0 α0 > < n00 l00 m00 k 0 α0 | Ĥpara | 2s >
+
00 00 00
E2s − En00 l00 m00 − h̄ωk0
n l m
X < 1s k α, k 0 α0 | Ĥpara | n00 l00 m00 kα > < n00 l00 m00 kα | Ĥpara | 2s >
+
E2s − En00 l00 m00 − h̄ωk
n00 l00 m00
(919)

155
These three terms add coherently, and it should be noted that the intermediate
state is only a virtual state and it can have a higher-energy than the 2s state46 .
In the limit η → 0 the first term in the expression for the transition rate
of eqn(918) reduces to a delta function which expresses conservation of energy
between the initial and final states.
 
η exp 2 η t
1
lim = δ( E2s − E1s − h̄ωk0 − h̄ωk )
η→0 π (h̄ω 0 + h̄ω + En0 − En )2 + h̄2 η 2
(920)
In the limit η → 0 the transition rate reduces to the Fermi-Golden rule
expression
1 2π X
= | M |2 δ( E2s − E1s − h̄ωk0 − h̄ωk ) (921)
τ h̄ 0 0 k,α:k ,α

The emitted photons have continuous spectra. In the expression for the matrix
elements M , the last two terms differ in the time-order that the two photons
are emitted. On inserting the expressions for the interactions into M , one can
pull out the common factors leaving a dimensionless matrix element M 0 . This
leads to the expression
q2 2 π h̄ c2
  
1
M = 2
√ M0 (922)
2mc V ωk ωk 0
where M 0 is the dimensionless factor given by
 
0 0 0
M = ˆα (k) . ˆα0 (k ) < 1s | exp − i ( k + k ) . r | 2s >

2 X < 1s | ˆα (k) . p̂ exp[ − i k . r ]| n00 l00 m00 > < n00 l00 m00 | ˆα0 (k 0 ) . p̂ exp[ − i k 0 . r ] | 2s >
+
m E2s − En00 l00 m00 − h̄ωk0
n00 l00 m00

2 X < 1s | ˆα0 (k 0 ) . p̂ exp[ − i k 0 . r ]| n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ exp[ − i k . r ] | 2s >
+
m E2s − En00 l00 m00 − h̄ωk
n00 l00 m00
(923
The first term is negligible, since 1  k . r and the electronic eigenstates
are orthogonal. The order of magnitude of the second term is given by the
electronic kinetic energy divided by the excitation energy. Hence, the reduced
matrix elements have a magnitude of the order of unity. The transition rate is
given by
 2 2
1 e ( 2 π )3 X h̄ c2
= 2 2 0
|M 0 |2 δ( E2s − E1s − h̄ωk − h̄ωk0 )(924)
τ 2mc V 0 0
k k
k,α;k ,α

46 Due to the Lamb shift, there is a 2p state with slightly lower energy than the 2s state.
However, due to the small magnitude of the energy difference, the part of the decay process
involving any real 2p transition is negligibly small.

156
One can assume that the dipole matrix elements of the intermediate states
should be randomly oriented in space, since the initial and final electronic states
are isotropic. After summing over the polarizations, the transition rate becomes
isotropic. On setting X
| M 0 |2 ≈ 1 (925)
α,α0

one finds
 2 2
h̄ c2 d3 k d3 k 0
Z Z
1 e
= δ( E2s − E1s − h̄ωk − h̄ωk0 )(926)
τ 2 m c2 ( 2 π )3 k k0
Since the integrand is independent of the direction of k and k 0 , the angular
integrations can be performed leaving
 2 2 Z ∞ Z ∞
1 e h̄ c2
= dk k dk 0 k 0 δ( E2s − E1s − h̄ωk − h̄ωk0 )(927)
τ m c2 2π 0 0

On integrating over k 0 , one obtains

 2 2 Z ωc
1 e c ω21
= dk k ( − k) (928)
τ m c2 2π 0 c
where ω12 is related to the energy difference of the 1s and 2s states. An ele-
mentary integration yields
 2 2  3
1 e c ω12
= (929)
τ m c2 12 π c
The first factor has dimensions of length squared and can be recognized as the
square of the classical radius of the electron. However, since
ω12 3 e2
= (930)
c 8 h̄ c a
and
h̄2
a = (931)
m e2
or
e2 a e4
2
= 2 2 (932)
mc h̄ c
one finds the decay rate is approximated by
  3  2 7
1 1 3 c e
= (933)
τ 12 π 8 a h̄ c

Thus, the estimated decay rate is 8.75 sec−1 . The exact value calculated by
Shapiro and Breit47 is 8.266 sec−1 .

47 J. Shapiro and G. Breit, Phys. Rev. 113, 179 (1959).

157
11.1.10 The Absorption of Radiation
If a process occurs in which only a photon with quantum numbers (k, α) is
absorbed, then the numbers of quanta in the initial and final state of the elec-
tromagnetic field are given by
n0k,α = nk,α − 1
n0k0 ,β = nk0 ,β (934)
The matrix elements of the paramagnetic interaction are given by
< n0 l0 m0 {n0k0 ,β } | Ĥpara | nlm {nk0 ,β } >
s
√ 2 π h̄ c2
  X  
q 0 0 0
= − nk,α < n l m | p̂ . ˆα (k) exp + i k . r | nlm >
mc V ωk
k,α

(935)
The photon absorption rate is found from the Fermi-Golden rule expression
2 
2 π h̄ c2
 
1 2π q X
= nk,α δ( Enlm + h̄ωk − En0 l0 m0 )
τ h̄ mc V ωk
n0 l0 m0
 
0 0 0
× | < n l m | p̂ . ˆα (k) exp + i k . r | nlm > |2 (936)

This is related to the lifetime due to stimulated emission, if the initial and final
states are interchanged.

The scattering cross-section for photon absorption σabsorb (ω) is found by

relating the number of photons absorbed (per second) to the product of the
incident flux and the cross-section. The photon flux is given by the photon
density times the velocity of light
nk,α
j = c k̂ (937)
V
Hence, the cross-section can be written as
2 
2 π h̄ c2
   
V 2π q X
σabsorb (ωk ) = nk,α δ( Enlm + h̄ωk − En0 l0 m0 )
nk,α c h̄ mc V ωk
n0 l0 m0
 
0 0 0
× | < n l m | p̂ . ˆα (k) exp + i k . r | nlm > |2 (938)

which simplifies to
4 π 2 e2
  X
σabsorb (ωk ) = δ( Enlm + h̄ωk − En0 l0 m0 )
m2 ωk c
n0 l0 m0
 
× | < n0 l0 m0 | p̂ . ˆα (k) exp + ik.r | nlm > |2

(939)

158
The absorption cross-section is independent of the volume of the electromag-
netic cavity and the number of photons in the incident beam. As a function of
frequency, the Born approximation for the cross-section for photon absorption
contains delta function lines corresponding to the atomic excitation energies.
Measured absorption lines do have natural widths ∆ωnl,n0 l0 and the absorbtion
spectra can be approximated by the sums of Lorentzian functions. The widths

70
60
σ(ω) [ h /m ]
50
40
2

30
20
10
0
0.7 0.8 0.9 1
hω [ Ryd ]
Figure 27: A sketch of the photon absorption cross-section σ(ω) (in units of
h̄2
m ) as a function of photon energy h̄ω (in units of Rydbergs). The plot over-
emphasizes the role of the photon lifetimes, since the ratio of the line-width to
e2 3
the photon frequency is of the order of ( h̄c ) .

of the lines are governed by half the sum of the decay rates of the initial and
final electronic levels.
 
1 1 1
∆ωnl,n0 l0 = + (940)
2 τnl τ n0 l 0
This formula implies that rapidly decaying levels will yield broad lines, but does
not imply the converse48 . The spectral widths can be described by the inclu-
sion of the effects of interaction to higher orders49 . The higher-order processes
produce small shifts of the atomic energy levels and also give the energies small
imaginary parts, resulting in a Lorentzian line shape. Since a typical atomic
transition rate is of the order of 108 sec−1 and a typical photon frequency is of
the order of 1015 sec−1 , the widths of the lines can usually be neglected.

The absorption cross-section can be evaluated in the dipole approximation

4 π 2 e2
  X
σabsorb (ωk ) = δ( Enlm + h̄ωk − En0 l0 m0 )
m2 ωk c 0 0 0 nlm
48 Lines in the absorption spectra with weak intensities can be broad if the final states are
rapidly decaying.
49 V. F. Weisskopf and E. Wigner, Z. Physik, 63, 54 (1930), Zeit. für Physik, 65, 18 (1930).

159
 × | < n0 l0 m0 | p̂ . ˆα (k) | nlm > |2 (941)

which can be re-written as

4 π 2 e2 ωk
  X
σabsorb (ωk ) = δ( Enlm + h̄ωk − En0 l0 m0 )
c 0 0 0 nlm
0 0 0
× | < n l m | r . ˆα (k) | nlm > |2 (942)

For an isotropic medium, the electronic states are degenerate with respect to
the z-components of the orbital angular momentum, so the initial state (n, l, m)
should be averaged over the different values of m
l
1 X
(943)
(2l + 1)
m=−l

and the values of m0 for the final states are summed over all possible values.
This averaging process results in an isotropic absorption rate, and is equivalent
to averaging the polarization vector over all directions in space. Therefore, in
the dipole approximation, the absorption cross-section for an isotropic medium
is given by the expression

4 π 2 e2
  X
σabsorb (ω) = ωn0 l0 ,nl | < n0 l0 m0 | r | nlm > |2 δ(ωn0 l0 ,nl − ω )
3 h̄ c 0 0 0
nlm
(944)
The strength of each absorption line can be found by integrating the cross-
section over a narrow frequency range centered on the frequency of the absorp-
tion line. (More specifically, the width of the interval of integration must be
greater than the natural line-width.) The integrated intensity of the transition
(nlm) → (n0 l0 m0 ) is given by
Z ωnl,n0 l0 +
4 π2
 2
e
dω σabsorb (ω) = ωn0 l0 ,nl | < n0 l0 m0 | r | nlm > |2
ωnl,n0 l0 − 3 h̄ c
(945)
The intensity of each line is proportional to the “oscillator strength” fnl→n0 l0
defined as
2 m ωn0 l0 ,nl
fnl→n0 l0 = | < n0 l0 m0 | r | nlm > |2 (946)
h̄
The intensities and the frequencies of all the transitions are related via sum
rules50 . These sum rules involve quantities of the form
X
ωnp 0 l0 m0 ,nlm | < n0 l0 m0 | r | nlm > |2 (947)
n0 l0 m0

160
 Table 5: Sum Rules for Dipole Transitions

ωnp 0 l0 m0 ,nlm | < n0 l0 m0 | r | nlm > |2

P
p n0 l0 m0

0 < nlm | r2 | nlm >

3 h̄
1 2 m

2
2 m ( Enlm − < nlm | V | nlm > )

h̄
3 2 m < nlm | ∇2 V | nlm >

and have values given in the Table(5). The sum rules can be used to provide
checks of experimental data.

Sum Rules for Dipole Radiation

There exists a systematic way of deriving sum rules for the weighted inten-
sities of the dipole allowed transitions. The sum rules are of the form
X p 0 0 0
ωnl,n0 l0 | < nlm | Â | n l m > |2 (948)
n0 l0 m0

where
h̄ ωnl,n0 l0 = Enl − En0 l0 (949)
and p is a positive integer.

Consider the expression

F (t) = < nlm | Â(t) † (0) | nlm > (950)

where the operator Â(t) is given in the Heisenberg representation

   
i i
Â(t) = exp + Ĥ0 t  exp − Ĥ0 t (951)
h̄ h̄
50 W. Thomas, Naturwiss. 11, 527 (1925).

F. Reiche and W. Thomas, Zeit. für Physik, 34, 510 (1925).

W. Kuhn, Zeit. für Physik, 33, 408 (1925).

161
Then, on taking successive derivatives of F (t) with respect to t, one finds
 
∂F i
= < nlm | [ Ĥ0 , Â(t) ] † (0) | nlm > (952)
∂t h̄

and
 2
∂2F
 
i
= < nlm | [ Ĥ0 , [ Ĥ0 , Â(t) ] ] † (0) | nlm > (953)
∂t2 h̄

etc. This process shows that the p-th derivative is expressed as p nested com-
mutators
 p   p
∂ F i
p
= < nlm | [ Ĥ0 , [ . . . [ Ĥ0 , [ Ĥ0 , Â(t) ] ] . . . ] ] † (0) | nlm >
∂t h̄
(954)
Alternatively, one can insert a complete set of states in the definition for F (t)
yielding
X
F (t) = < nlm | Â(t) | n0 l0 m0 > < n0 l0 m0 | † (0) | nlm > (955)
n0 l0 m0

but since the states | nlm > are eigenstates of Ĥ0 , one has

X   2
 < nlm | Â | n0 l0 m0 > 
 
F (t) = exp i ωnl,n0 l0 t   (956)
n0 l0 m0

On taking the p-th derivative of this form of F (t), one finds

2
∂pF
  X  
p p 0 0 0
 
= i ωnl,n 0 l0 exp i ωnl,n0 l0 t  < nlm | Â | n l m >  (957)

∂tp
n0 l0 m0

The sum rules are found by equating the two forms of the p-th time-derivative
and then setting t = 0
X  p  2
0 0 0
 
Enl − En0 l0  < nlm | Â | n l m > 
 
n0 l0 m0

= < nlm | [ Ĥ0 , [ . . . , [ Ĥ0 , [ Ĥ0 ,  ] ] . . . ] ] † | nlm >(958)

Hence, the p-th moment of the matrix elements of  is related to the expecta-
tion value of the product of the p-th nested commutator of Ĥ0 and  multiplied
by † .

The expectation value of p nested commutators of  can be expressed as the

expectation value of the product of p − q nested commutators of  with q nested
commutators of † . This can be demonstrated by noting that the expectation

162
value is homogeneous in time. The q-th nested commutator of the operator Â
can be defined by
B̂q = [ Ĥ0 , B̂q−1 ] (959)
where
B̂0 = Â (960)
†
Likewise, Ĉq can be defined as the q-th nested commutator of  . However, for
any pair of operators B̂p−q−1 and Ĉq , one has

< nlm | B̂p−q−1 (t) Ĉq (0) | nlm >

X  
= exp i ωnl,n l t < nlm | B̂p−q−1 | n0 l0 m0 > < n0 l0 m0 | Ĉq | nlm >
0 0

n0 l0 m0

= < nlm | B̂p−q−1 (0) Ĉq (−t) | nlm > (961)

This is an expression of the homogeneity of time. Hence, on taking a derivative

with respect to t and then setting t = 0, one finds

< nlm | [ Ĥ0 , B̂p−q−1 ] Ĉq | nlm > = ( − 1 ) < nlm | B̂p−q−1 [ Ĥ0 , Ĉq ] | nlm >
(962)

On using the definition of the operators B̂p and Cq , the above equation reduces
to

< nlm | B̂p−q Ĉq | nlm > = ( − 1 ) < nlm | B̂p−q−1 Ĉq+1 | nlm > (963)

By induction, this shows that the nested commutators can be distributed be-
tween the two sides of the expression.

< nlm | B̂p Ĉ0 | nlm > = ( − 1 )q < nlm | B̂p−q Ĉq | nlm > (964)

which was to be shown.

11.1.11 The Photoelectric Effect

The differential scattering cross-section for the absorption of a photon by a
hydrogen atom in the ground state accompanied by the emission of an electron
shall be derived. For emitted electrons with sufficiently high energies, the wave
function for the photo-emitted electron can be approximated by a plane-wave.
The transition rate is given by the Fermi-Golden rule expression involving the
paramagnetic interaction
2 
2 π h̄ c2 h̄2 k 02
   
1 2π e X
0
= nk,α | < k | p . ˆα (k) exp i k . r | 1s > |2 δ( E1s + h̄ ωk − )
τ h̄ mc V ωk 2m
k0
(965)

163
The cross-section is given by
X  4 π 2 e2  h̄2 k 02
 
0 2
σ = | < k | p . 
ˆα (k) exp i k . r | 1s > | δ( E 1s + h̄ ω k − )
0
m2 ωk c 2m
k
(966)
where the initial wave function is given by
 
1 r
ψ1s (r) = √ exp − (967)
π a3 a

As long as the emitted electron is not close to threshold, the final state wave
function can be approximated by a plane-wave
 
1
ψk0 (r) = √ exp i k 0 . r (968)
V
The sum over final states of the electron can be replaced by an integral over the
magnitude of its momentum and its direction
Z ∞ Z
X V 0 02
→ dk k dΩk0 (969)
0
( 2 π )3 0
k

It is seen that the factor of the volume in the density of final states cancels with
the factors from the normalization of the electron’s final state. The differential
cross-section corresponds to the part of the cross-section where the outgoing
electron is emitted into the solid angle dΩk0 . Hence,
Z ∞
e2 h̄2 k 02
  
dσ V 0 02 0 2
= dk k | < k | p . 
ˆα (k) exp i k . r | 1s > | δ( E 1s + h̄ ω k − )
dΩ0 2 π m2 ωk c 0 2m
(970)
The integration over the magnitude of electron’s final momentum k 0 can be
performed by using the properties of the energy conserving delta function. The
magnitude of electron’s final momentum is denoted by kf

2m
kf2 = ( h̄ωk + E1s ) (971)
h̄2
The result of the integration over k 0 is

e2
   
dσ V 0
= kf | < kf dΩ | p . ˆα (k) exp i k . r | 1s > |2
dΩ0 2 π h̄2 m ωk c
(972)
It is assumed that the initial photon is propagating along the x-axis and is
polarized along the z-direction. The matrix elements involving the momentum
operator only yield a finite result when p̂ acts on ψ1s (r), since k . ˆα (k) = 0.
However,
h̄ cos θ
ˆα (k) . p̂ ψ1s (r) = i ψ1s (r) (973)
a

164
 eα e- (Ek',k')
'
θ
(hωk,k)

Figure 28: The geometry for the photo-emission of an electron from an atom.
An electromagnetic wave, with polarization along the z-axis, is incident along
the x-axis. The photo-emitted electron propagates along the direction k 0 .

which results in the replacement p̂z → i h̄ cos

a
θ
. Thus, one finds

e2
   
dσ V 0
= kf | < kf dΩ | cos θ exp i k . r | 1s > |2 (974)
dΩ0 2 π m ωk c a2

where (θ, ϕ) are the polar coordinates of the vector r. The matrix elements are
evaluated using the dipole approximation for the photon wave function and set
 
exp i k . r ≈ 1 + i k . r + ... (975)

and only keep the first term of the expansion. The factor cos θ can be expressed
as a spherical harmonic through
r
4π 1
cos θ = Y0 (θ, ϕ) (976)
3
and the final state electronic wave function can be expressed in terms of the
Rayleigh expansion
  X
0
exp i kf k̂ . r = 4π il jl ( kf r ) Yml (θ, ϕ) Yml ∗ (θ0 , ϕ0 ) (977)
l,m

where (θ0 , ϕ0 ) are the polar coordinates of the electron’s final momentum. The
angular integration over the polar coordinates (θ, ϕ) can be performed by using
the orthogonality relations for the spherical harmonics. The end result is
Z ∞  
1 r
< kf dΩ0 | cos θ | 1s > = − 4 π i cos θ0 √ dr r2 j1 (kf r) exp −
π a3 V 0 a
(978)
where the cos θ0 dependence refers to the direction of the emitted electron’s
angular momentum. The radial integral is evaluated to yield
r
0 0 a3 2 kf a
< kf dΩ | cos θ | 1s > = − 4 π i cos θ (979)
π V ( 1 + kf2 a2 )2

165
Therefore, the differential cross-section is given by
    2   2
dσ kf e a 2 0 2 kf a
= 8 cos θ (980)
dΩ0 k m c2 ( 1 + kf2 a2 )2
Using
h̄2
a = (981)
m e2
the photo-emission cross-section can be re-written as
     2 2  2
dσ kf 2 e 2 0 2 kf a
= 8 a cos θ (982)
dΩ0 k h̄ c ( 1 + kf2 a2 )2
Thus, although the photon is propagating along the x-direction, the electron is
preferentially emitted along the direction of the polarization (θ0 ≈ 0). This can
be understood as being due to the effect of c being large, so that the photon’s
momentum is negligible compared with the energy, therefore, (in the dipole
approximation) only the direction of the polarization determines the angular
distribution of the emitted electron. It should be noted that in the relativistic
case, where the momentum of the photon is important, the electrons are pre-
dominantly ejected in the direction of the photon51 . This formula also breaks
down for emitted electrons with low energies. In this case, the correct electronic
wave function for the continuous spectrum of Ĥ0 should be used52 . The in-
clusion of the Coulomb attraction of the ion in the final state has the effect of
reducing the cross-section near the threshold.

11.1.12 Impossibility of absorption of photons by free-electrons.

Free electrons are described by the non-interacting Hamiltonian Ĥ0 where
p̂2
Ĥ0 = (983)
2m
which plane-waves as energy eigenstates
 
1
φk0 (r) = √ exp − i k0 . r (984)
V
corresponding to the energy eigenvalues
h̄2 k 02
Ek 0 = (985)
2m
The matrix elements for electromagnetic transitions in which a photon (k, α) is
absorbed is given by
r
√
  
q 2 π h̄ c
− < k 00 | p̂ . ˆα (k) exp + i k . r | k 0 > nk,α (986)
mc V ωk
51 F. Sauter, Ann. Phys. 9, 217 (1931), Ann. Phys. 11, 454 (1931).
52 M. Stobbe, Ann. Phys. 7, 661 (1930).

166
which is evaluated as
r
√

q 2 π h̄ c
− p̂ . ˆα (k) δk+k0 −k00 nk,α (987)
mc V ωk
This shows that momentum is conserved. Furthermore, for the transition rate

p'
p''

Figure 29: The absorption of a photon via the paramagnetic interaction.

to represent a real process, it is necessary that energy is conserved between the

initial and final states
h̄2 k 02 h̄2 k 002
h̄ ωk + = (988)
2m 2m
It is impossible for this process to satisfy the conditions for conservation of
energy and momentum. This can be seen by appealing to the relativistic for-
mulation where the four-vector momentum is conserved

pµ + p0µ = p00µ (989)

Hence,
( pµ + pµ0 ) ( pµ + p0µ ) = pµ00 p00µ (990)
but the electron’s momenta form a Lorentz scalar which is related to the rest
mass
pµ0 p0µ = pµ00 p00µ = m2 c2 (991)
and the photon has zero mass

pµ pµ = 0 (992)

Therefore, one finds that the cross-terms vanish

pµ p0µ = 0 (993)

In the rest frame of the electron one has pµ 0 = ( m c , 0 ), so the energy of

the photon is identically zero. Therefore, there is no photon and the absorption
process is impossible.

167
11.2 Scattering of Light
Kramers and Heisenberg evaluated the scattering cross-section for light incident
on atomic electrons53 . The incident photon is denoted by (k, α) and the scat-
tered photon by (k 0 , α0 ). The scattering cross-section involves the paramagnetic
interaction to second-order and the diamagnetic interaction to first-order. The
matrix elements of the diamagnetic interaction are given by
e2
 
0 0 0 0 0
< n l m k α | Ĥdia | nlmkα > = < n0 l0 m0 k 0 α0 | Â . Â | nlmkα >
2 m c2
e2
   
0 0 0 0 0 † † 0
= < n l m k α | ( ak,α ak0 ,α0 + ak0 ,α0 ak,α ) exp i ( k − k ) . r | nlmkα >
2 m c2
2 π h̄ c2
 
× √ ˆα (k) . ˆα0 (k 0 ) (994)
ωk ωk 0 V
where it has been assumed that only the initial and final photon are present. On
making use of the long-wavelength approximation λ  a, the matrix elements
simplify to
e2
 
< n0 l0 m0 k 0 α0 | Ĥdia | nlmkα > ≈ < n0 l0 m0 | nlm >
2 m c2
2 π h̄ c2
 
× √ ˆα (k) . ˆα0 (k 0 )
ωk ωk 0 V
(995)
The scattering cross-section will be expressed in terms of a transition rate and
the transition rate will be calculated using a similar procedure to that which
was used in describing two-photon decay. An arbitrary state | ψn > can be
expressed in terms of a complete set of non-interacting states | n >
 
X i
| ψn > = Cn0 (t) exp − En0 t | n 0 > (996)
0
h̄
n

where Cn (t) are time-dependent coefficients. Initially, the system is assumed to

be in an energy eigenstate | n > of the unperturbed Hamiltonian, and due to
the interaction makes a transition to a state | n0 >. The probability of finding
the system in the state | n0 > at time t is then given by |Cn0 (t)|2 . It shall be
assumed that the interaction is turned off when t → − ∞. The interaction
can be turned off at large negative times by introducing a multiplicative factor
of exp[ + η t ] in the interaction, where η is an infinitesimally small positive
constant. To first-order in the diamagnetic interaction, one finds
e2 2 π h̄ c2
    
(1) −i
Cn0 (t) = δn0 ,n √ 2 ˆα (k) . ˆα0 (k 0 )
h̄ 2 m c2 ωk ωk 0 V
Z t  
0 i 0 0
dt exp ( h̄ ω + En0 − h̄ ω − En ) t (997)
−∞ h̄
53 H. A. Kramers and W. Heisenberg, Z. Physik, 31, 681 (1925).

168
where ω = c k and ω 0 = c k 0 and the long-wavelength approximation has
been used. The small quantity η has been absorbed as a small imaginary part
to the initial state energy

En → En + i η h̄ (998)

The paramagnetic interaction is of order of e and the diamagnetic interaction

is of order e2 . Thus, to second-order in e, one must include the diamagnetic
interaction and the paramagnetic interaction to second-order. There are two

k'α'
k''α''

n n'

Figure 30: Photon scattering processes due to the diamagnetic interaction to

first-order.

terms which are second-order in the paramagnetic interactions that represent:

(a) absorption of the photon (k, α) followed by the emission of a photon (k 0 , α0 ).

(b) emission of a photon (k 0 , α0 ) followed by the absorption of the photon (k, α).

(k,α) (k',α')
(k',α')
(k,α)

n'' n''
n n' n n'

Figure 31: Photon scattering processes due to the paramagnetic interaction to

second-order.

The second-order contribution to the transition amplitude is given by

2  2   Z t Z t0
2 π h̄ c2

(2) −i e 0
Cn0 (t) = √ dt dt00
h̄ mc ωk ωk 0 V −∞ −∞
 X
i i
exp[ (h̄ω 0 + En0 − En00 ) t0 ] exp[ − (En − En00 + h̄ω) t00 ]
00
h̄ h̄
n

169
 × < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ | nlm >
X i i
+ exp[ (En0 − En00 − h̄ω) t0 ] exp[ − (En − En00 − h̄ω 0 ) t00 ]
00
h̄ h̄
n

× < n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα0 (k 0 ) . p̂ | nlm >

(999)

The earliest time integration can be evaluated leading to

2  t
2 π h̄ c2
   Z
(2) −i e X i
Cn0 (t) = √ dt0 exp[ (h̄ω 0 + h̄ω + En0 − En ) t0 ]
h̄ mc ωk ωk 0 V −∞ h̄
n00
0 0 0 0 00 00 00 00 00 00
< n l m | ˆα0 (k ) . p̂ | n l m > < n l m | ˆα (k) . p̂ | nlm >


( En − En00 + h̄ ω )
< n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα0 (k 0 ) . p̂ | nlm >

+ (1000)
( En − En00 − h̄ ω 0 )

as long as the denominators are non-vanishing.

(1) (2)
The coefficients Cn0 (t) and Cn0 (t) have the same type of time-dependence.
The remaining integration over time yields
  Z t  
−i 0 i 0 0
dt exp (h̄ω + En0 − En − h̄ω − ih̄η) t
h̄ −∞ h̄
exp[ i
h̄ (h̄ω 0 + En0 − En − h̄ω − ih̄η) t ]
= − (1001)
(h̄ω 0 + En0 − En − h̄ω − ih̄η)

The transition rate is given by

 2 
1 ∂  (1)
 C 0 (t) + C (2)

=  n n0 (t)
 (1002)
τ ∂t 

but the time-dependence of the squared modulus is contained in the common

factor
 
 exp[ h̄i (h̄ω 0 + En0 − En − h̄ω − ih̄η) t ] 2
  exp 2 η t
  =
 (h̄ω 0 + En0 − En − h̄ω − ih̄η)  (h̄ω 0 + En0 − En − h̄ω)2 + h̄2 η 2
(1003)
Therefore, one finds the transition rate is given by the expression
 
2 η exp 2 η t  2 2  2
1 e 2 π h̄ c2
= √ M 2 (1004)
τ (h̄ω 0 + En0 − En − h̄ω)2 + h̄2 η 2 m c2 ωk ωk 0 V

170
where the matrix elements are given by

M = ˆα (k) . ˆα0 (k 0 ) < n0 l0 m0 | nlm >
0
1 X < n0 l0 m0 | ˆα0 (k ) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ | nlm >
+
m 00 ( En − En00 + h̄ ω )
n
0
1 X < n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα0 (k ) . p̂ | nlm >

+
m 00 ( En − En00 − h̄ ω 0 )
n
(1005)

On taking the limit η → 0, the first factor in the decay rate reduces to an
energy conserving delta function. Therefore, one obtains the Fermi-Golden rule
expression
 2 2  2
1 2π e 2 π h̄ c2
= √ M 2 δ(h̄ωk0 + En0 − En − h̄ωk ) (1006)
τ h̄ m c2 ωk ωk 0 V

The magnitudes of the final state photon quantum numbers (k 0 ) must be inte-
grated over, since these are not measured. This integration imparts a physical
meaning to the expression for the rate which contains the Dirac delta function.
We shall assume that the direction of the scattered photon is to be measured
and that the photon is absorbed by a detector which subtends a solid angle dΩ0
to the atom. Therefore, the scattering rate is given by
 2 2 Z ∞ 2
2 π h̄ c2

1 2π e V 0 0 02
= dΩ dk k √ | M |2 δ(h̄ωk0 +En0 −En −h̄ωk )
τdΩ0 h̄ m c2 ( 2 π )3 0 ωk ωk 0 V
(1007)
Since h̄ ωk0 = h̄ c k 0 , the integration over the delta function can be performed,
yielding
 2 2 2
V dΩ0 ω 02 2 π h̄ c2

1 2π e
= √ | M |2 (1008)
τdΩ0 h̄ m c2 ( 2 π )3 h̄ c3 ω ω0 V
The scattering cross-section is defined as the transition rate divided by the
photon flux. The photon flux is found by noting that it has been assumed that
there is one photon per volume V so the photon density is V1 and the speed of
light is c. Hence, the photon flux is given by Vc . Therefore, the cross-section is
determined by the Kramers-Heisenberg formula
   2 2  0 
dσ e ω
= | M |2 (1009)
dΩ0 m c2 ω
The magnitude of the scattering rate is determined by the quantity re which
has the dimensions of length
 2 
e
re = (1010)
m c2

171
This quantity is often called the classical radius of the electron. The quantity
re can be expressed as
 2   2 
e e h̄
re = = ≈ 2.82 × 10−15 m (1011)
m c2 h̄ c mc

11.2.1 Rayleigh Scattering

Rayleigh scattering corresponds to the limit in which the light is elastically
scattered. Hence, one has
ω = ω0 (1012)
In the case of elastic scattering, all the terms in the Kramers-Heisenberg formula
are equally important. That all terms have a similar magnitude can be seen by
re-writing the first term ˆα (k) . ˆα0 (k 0 ) in a way which is similar to the second.
The scalar product of the polarization vectors can be expressed as
X
ˆα (k) . ˆα0 (k 0 ) = ˆα (k)i δi,j ˆα0 (k 0 )j (1013)
i,j

but one can re-write the Kronecker delta function in terms of the commutation
relation
[ xi , p̂j ] = i h̄ δi,j (1014)
Thus, one can express the scalar product as a commutator
1 X
ˆα (k) . ˆα0 (k 0 ) = ˆα (k)i [ xi , p̂j ] ˆα0 (k 0 )j
i h̄ i,j
1
= [ ˆα (k) . r , p̂ . ˆα0 (k 0 ) ] (1015)
i h̄
Since, in the dipole approximation, the diamagnetic contribution to the matrix
elements M is proportional to the overlap integral

< n0 l0 m0 | nlm > (1016)

the initial and final states must be identical if this is non-zero. Hence, the
result is equivalent to the expectation value in the state | nlm > . On replacing
the matrix elements by the expectation value and then insert a complete set of
electronic states, one finds

< n0 l0 m0 | nlm > ˆα (k) . ˆα0 (k 0 )

1 X 
= < n0 l0 m0 | r . ˆα (k) | n00 l00 m00 > < n00 l00 m00 | α0 (k 0 ) . p̂ | nlm >
i h̄ 00 00 00
n l m

− < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | r . ˆα (k) | nlm > (1017)

172
The matrix elements of r can be expressed in terms of the matrix elements of p̂
via
1
< n0 l0 m0 | p̂ | n00 l00 m00 > = < n0 l0 m0 | [ r , p̂2 ] | n00 l00 m00 >
2 i h̄
m
= < n0 l0 m0 | [ r , Ĥ0 ] | n00 l00 m00 >
i h̄
m
= ( En00 l00 m00 − En0 l0 m0 ) < n0 l0 m0 | r | n00 l00 m00 >
i h̄
(1018)

Therefore, one finds

i
< n0 l0 m0 | r | n00 l00 m00 > = < n0 l0 m0 | p̂ | n00 l00 m00 > (1019)
m ωn00 ,n0

where
En00 l00 m00 − En0 l0 m0 = h̄ ωn00 n0 (1020)
Thus, the elastic scattering term in the Kramers-Heisenberg formula is given by

δnlm,n0 l0 m0 ˆα (k) . ˆα0 (k 0 )

1 X < n0 l0 m0 | p̂ . ˆα (k) | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα0 (k 0 ) | nlm >
=
m 00 00 00 h̄ ωn00 n0
n l m

1 X < n0 l0 m0 | p̂ . ˆα0 (k 0 ) | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
−
m h̄ ωnn00
n00 l00 m00
(1021)

but since for elastic scattering Enlm = En0 l0 m0 , one has

δnlm,n0 l0 m0 ˆα (k) . ˆα0 (k 0 )

1 X < n0 l0 m0 | p̂ . ˆα (k) | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα0 (k 0 ) | nlm >
=
m 00 00 00 En00 − En
n l m

1 X < n0 l0 m0 | p̂ . ˆα0 (k 0 ) | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
+
m En00 − En
n00 l00 m00
(1022)

On substituting this back into the expression for the matrix elements M , one
obtains

1 X < n0 l0 m0 | p̂ . ˆα (k) | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα0 (k 0 ) | nlm >
M =
m 00 00 00 En00 − En
n l m

1 X < n0 l0 m0 | p̂ . ˆα0 (k 0 ) | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
+
m En00 − En
n00 l00 m00

173
 1 X < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
+
m ( En − En00 + h̄ ω )
n0 l00 m00
< n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα0 (k 0 ) | nlm >

1 X
+
m ( En − En00 − h̄ ω )
n00 l00 m00
(1023)

which simplifies to

< n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >

ω X
M =
m h̄ ωn00 n ( ωnn00 + ω )
n00 l00 m00
< n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα0 (k 0 ) | nlm >

−
ωn00 n ( ωnn00 − ω )
(1024)

In the limit of small photon frequencies compared with the electronic energies,
one can expand the denominators of the matrix element as
1 1 ω
= 2 ∓ 3 + ... (1025)
ωnn00 ( ωnn00 ± ω ) ωnn00 ωnn00

When this low-frequency expansion is substituted into the matrix elements, the
leading term vanishes. This can be seen since the leading term is proportional
to
X 1 
2 < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
ω nn00
n00

− < n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα0 (k 0 ) | nlm >

(1026)

which can be expressed as

X 
m 2
< n0 l0 m0 | ˆα0 (k 0 ) . r | n00 l00 m00 > < n00 l00 m00 | r . ˆα (k) | nlm >
n00

0 0 0 00 00 00 00 00 00 0
− < n l m | ˆα (k) . r | n l m > < n l m | r . ˆα0 (k ) | nlm >

(1027)

or, on using the completeness relation, one finds the expectation value of the
commutator is given by

m2 < n0 l0 m0 | [ ˆα0 (k 0 ) . r , r . ˆα (k) ] | nlm > = 0 (1028)

174
Thus, the leading term of the low-frequency expansion vanishes. Therefore, the
scattering rate is expressed as
   2   3
dσ re  X
4  1
= ω 
dΩ0 m h̄ ωnn00
n00 l00 m00

× < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
 2
0 0 0 00 00 00 00 00 00 0

+ < n l m | ˆα (k) . p̂ | n l m > < n l m | p̂ . ˆα0 (k ) | nlm > 

(1029)

Finally, the scattering rate can be expressed in terms of the dipole matrix ele-
ments as
   2  X  3
dσ re m 4 
 1
= ω
dΩ0 h̄  ωnn00
n00 l00 m00

× < n0 l0 m0 | ˆα0 (k 0 ) . r | n00 l00 m00 > < n00 l00 m00 | r . ˆα (k) | nlm >
 2
0 0 0 00 00 00 00 00 00 0

+ < n l m | ˆα (k) . r | n l m > < n l m | r . ˆα0 (k ) | nlm > 

(1030)

Hence, at long-wavelengths, the scattering cross-section varies as ω 4 as expected

from Rayleigh’s law. Since the typical electronic frequency ωnn00 is in the ultra-
violet spectrum, then
ωnn0  ω (1031)
for all frequencies in the visible optical spectrum. This leads to the phenomena
of blue skies in the day and red sunsets at dusk.

11.2.2 Thomson Scattering

Thomson scattering occurs for photons with sufficiently high energies

ω  ωnn00 (1032)

so that the photon energy is greater than the atomic binding-energy. In this
case, the second and third terms in the Kramers-Heisenberg formula can be
neglected. This is because
1
ω ∼ ω0  < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | p̂ . ˆα (k) | nlm >
m
(1033)

175
Therefore, the scattering predominantly occurs elastically and the scattering
cross-section is given by
   2
dσ 2 
 0 

= r e 
ˆα (k) . 
ˆα0 (k ) (1034)
dΩ0  

which is independent of ω. The above result is dependent on the scattering

angle via the polarization vectors.

In the investigation of the angular dependence of Thomson scattering, it is

convenient to introduce a coordinate system which is defined by the polarization
vectors and direction of propagation of the incident photon and its polarization
ˆ1 (k). The coordinate system is composed of the three orthogonal unit vec-
tors (ˆ 1 (k), ˆ2 (k), k̂). Thus the direction of the polarization vector ˆ1 (k) defines
the x-direction. In this coordinate system, the scattered photon (k 0 , α0 ) is in
the direction k 0 with polar coordinates (θk0 , ϕk0 ). The polarization of the final
photons ˆα (k 0 ) must be transverse to k 0 . Two polarization vectors are defined
according to

ˆ1 (k 0 ) = ( cos θk0 cos ϕk0 , cos θk0 sin ϕk0 , − sin θk0 ) (1035)

which lies in the plane of k and k 0 and

ˆ2 (k 0 ) = ( − sin ϕk0 , cos ϕk0 , 0 ) (1036)

which lies in the plane perpendicular to k. In terms of the chosen polarization

k
e2(k')
k'
e1(k')
θk'

e2(k)

φk'
e1(k)

Figure 32: The coordinate system and polarization vectors used to describe
Thomson scattering.

vectors, the scattering cross-section for incident radiation that is polarized along

176
the x-direction takes on the form
cos2 θk0 cos2 ϕk0 for α0 = 1
  
dσ 2
= r (1037)
0
dΩ x−pol e
sin2 ϕk0 for α0 = 2

if the polarizations of the final photon are measured.

If the incident beam has its polarization along the x-direction, and the de-
tector is not sensitive to the polarization, then the final polarization must be
summed over. In this case of a polarized beam and a polarization insensitive
detector, the cross-section is given by
   
dσ 2 2 2 2
= r e cos θ k 0 cos ϕk 0 + sin ϕk 0 (1038)
dΩ0 x−pol

where the polarizations of the final state photon have been summed over.

If the incident beam of photons is unpolarized, then ϕ is undefined since

the azimuthal direction of the scattered photon is defined with respect to the
assumed polarization ˆ1 (k). In the case of an unpolarized incident beam the
expression should be integrated over ϕ and divided by 2π. The scattering rate
is given by
re2 cos2 θk0 for α0 = 1
  
dσ
= (1039)
0
dΩ unpol 2 1 for α0 = 2

if the polarizations of the final state photons are measured. This result is iden-
tical to that obtained by assuming that the initial beam is composed of one half
of the number photons polarized along the x-direction and the other half of the
number of photons polarized along the y-direction. That is
2
for α0 = 1
   1 2 2
dσ 2 2 cos θk ( cos ϕk + sin ϕk )
0 0 0
= r 2 (1040)
0
dΩ unpol e 1 2
2 ( sin ϕk + cos ϕk )
0 0 for α0 = 2

The cross-section for unpolarized photons with a polarization insensitive detec-

tor is given by
re2
   
dσ 2
= 1 + cos θk0 (1041)
dΩ0 unpol 2
where the final polarizations have been summed over.

The total cross-section σ is obtained by integrating over all directions. The

total Thomson scattering cross-section is independent of whether the initial
beam was polarized or unpolarized. The final result is
8π 2
σ = r (1042)
3 e
which has a magnitude of 6.65 × 10−29 m2 . More massive charged particles,
such as protons, can also produce Thomson scattering but the cross-sections for

177
 m 2
these processes are smaller by factors of ( M ) . The derivation of the Thomson
scattering cross-section breaks down for photons which have energies of the
order of the electron’s rest energy

h̄ ω ∼ me c2 (1043)

For photons with these high-energies, one must describe the scattering process
relativistically. In this energy region, Compton scattering dominates.

Classical Interpretation

The classical counter-parts of Rayleigh and Thomson scattering can be de-

scribed by a two-step process. In the first step, the incident classical electro-
magnetic field causes an electron to undergo forced oscillations. In the second
step, the oscillating electrons emit electromagnetic radiation.

In the first process, an electron bound harmonically to the atom which re-
sponds to an electromagnetic field E 0 exp[ i ω t ] can be described by the
equation of motion
 
q
r̈ + ω02 r = E 0 <e exp i ω t (1044)
m
where ω0 is the frequency of the electron’s natural motion. In the steady state,
one finds q  
m E0
r = 2 <e exp i ω t (1045)
ω0 − ω 2
The acceleration of the charged particle can be described by
q
ω2 E 0
 
r̈ = − m2 <e exp i ω t (1046)
ω0 − ω 2
The accelerating charged particle radiates electromagnetic energy. The emitted
power is given by the Larmor formula

2 q2 r2 ω4
P (ω) =
3 c3
2 q E 20
4
ω4
= 2 (1047)
3 m c ( ω0 − ω 2 )2
2 3

while the incident energy flux is given by

c
E2 (1048)
4π 0
Hence, the scattering cross-section is described by

8π 2 ω4
σ = re 2 (1049)
3 ( ω0 − ω 2 )2

178
This formula has the correct frequency dependence in the limit ω  ω0 in which
case the classical cross-section varies as ω 4 , as expected for Rayleigh scattering.
On the other hand, in the limit ω  ω0 the cross-section becomes frequency
independent, as is expected for Thomson scattering.

11.2.3 Raman Scattering

For inelastic scattering, one has h̄ ω 6= h̄ ω 0 , therefore, the condition of con-
servation of energy requires that

Enlm + h̄ ω = En0 l0 m0 + h̄ ω 0 (1050)

Since it is most probable that the initial electron is in the ground state, one has

En0 l0 m0 > Enlm (1051)

which leads to the inequality

h̄ ω > h̄ ω 0 (1052)

Hence, the final photon has less energy than the initial photon. That is, the

Stokes
n → n'
I(ω')

anti-Stokes
n' → n

ω−ωn',n ω ω+ωn', ω'

Figure 33: The schematic frequency dependence of the observed intensity ex-
pected in a Raman scattering experiment. The ratio of intensities of the Stokes
and anti-Stokes lines provides a relative measure of the initial occupation of the
low-energy state n and the higher-energy excited state n0 .

electromagnetic field has lost energy and left the electron in an excited state.
This inelastic process describes the Stoke’s line. On the other hand, if the
electron is initially in an excited state, then it is possible that the electron loses
energy and makes a transition to the ground state. In this case,

En0 l0 m0 < Enlm (1053)

179
so the final photon is more energetic

h̄ ω < h̄ ω 0 (1054)

This process results in the anti-Stokes line.

11.2.4 Radiation Damping and Resonance Fluorescence

In the analysis of photon scattering, it has been assumed that the energy de-
nominators ( En − En00 + h̄ ω ) do not vanish. If the energy denominator
vanishes, the Kramers-Heisenberg formula becomes singular, however, the phys-
ically observed scattering cross-section may become large but does not diverge.
This is the phenomenon of resonance-fluorescence. Using the classical model,
one can describe the scattering cross-section, if damping is introduced to rep-
resent the lifetime of the electronic states. That is, the dynamics of the bound
electron is modelled by a damped harmonic oscillator
 
2 q
r̈ + γ ṙ + ω0 r = E <e exp i ω t (1055)
m 0
which has the solution
 
q 1
r = E <e 2 exp iωt (1056)
m 0 ω0 + i γ ω − ω 2

since γ is related to the decay rate and is of the order of 108 sec−1 , it is usually
negligible compared with the frequency of light which is estimated as ω ∼ 1015
sec−1 . Following our previous arguments, one finds that the scattering cross-
section is given by
 2 2
8π q ω4
σ(ω) = (1057)
3 m c2 ( ω 2 − ω02 )2 + γ 2 ω 2
which no longer diverges when the resonance condition is satisfied, because of
the damping of the electronic states.

The lifetime of a quantum mechanical state which at t = 0 is represented

by | ψn (0) > calculated to second-order in the interaction ĤI is given by the
Fermi-Golden rule expression. The rate can be expresses as the limit η → 0 by

1 2 X | < ψn0 | ĤI | ψn > |2

= − =m (1058)
τn h̄ 0
E n − En0 + i η
n

whereas the energy-shift found in second-order (Rayleigh-Schrödinger) pertur-

bation theory is also given by the limit η → 0 of
X | < ψn0 | ĤI | ψn > |2
∆En = <e (1059)
n
E n − En0 + i η

180
so
En = En(0) + ∆En (1060)
Hence, due to the form of the expressions for the shift and the lifetime as the
real and imaginary parts of a complex function, it is possible to consider an
unstable state as having a complex energy54 given by
h̄
En − i Γn ≈ En(0) + ∆En − i (1061)
2 τn
That is, the lifetime can be considered as giving the state an energy-width Γn .
This is the natural width of the electronic state. The factor of two in the width
can be understood by considering the time-dependence of the state | ψn (t) >
which is given by
 
i
| ψn (t) > = exp − ( En − i Γn ) t | ψn (0) > (1062)
h̄
Hence, the probability Pn (t) that the state has not decayed at time t is given
by
Pn (t) = | < ψn (0) | ψn (t) > |2
 
i
= | < ψn (0) | exp − ( En − i Γn ) t | ψn (0) > |2
h̄
 
2
= exp − Γn t (1063)
h̄
due to the normalization of the initial state. This time-dependence of Pn (t) is
interpreted in terms of the exponential decay of the probability for finding the
initial state  
t
Pn (t) = Pn (0) exp − (1064)
τn
This leads to the identification of the relation between the energy-width and
the lifetime
h̄
Γn = (1065)
2 τn
Hence, the lifetime τn of an unstable or metastable state can be incorporated
by introducing an imaginary part Γn to the energy.

Therefore, for the case of resonant scattering, one should replace the energies
by complex numbers such that the real part represents the state’s energy and
the imaginary part describes half the state’s decay rate. In the case of resonant
scattering, the Kramers-Heisenberg formula is modified55 to
   2 2  0 
dσ e ω
0
= 2
| M |2 (1066)
dΩ mc ω
54 That is, the perturbation produces a complex shift of the energy-shift which related to

the self-energy Σn (E) which is to be discussed later

55 P. A. M. Dirac, Proc. Roy. Soc. A 114, 710 (1927).

181
where the matrix elements are given by

M = ˆα (k) . ˆα0 (k 0 ) < n0 l0 m0 | nlm >

1 X < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ | nlm >
+
m ( En − En00 − i Γn00 + h̄ ω )
n00 l00 m00
< n0 l0 m0 | ˆα (k) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα0 (k 0 ) . p̂ | nlm >

1 X
+
m ( En − En00 − i Γn00 − h̄ ω 0 )
n00 l00 m00
(1067)
2
Since close to resonance, the resonant denominator is given by Γ ∼ h̄ac ( h̄e c )4
2
whereas the numerator is of the order of ea . Hence, on-resonance, the matrix
2
elements can be of the order ( h̄e c )−3 larger than the non-resonant matrix ele-
ments. Therefore, on resonance, the non-resonant terms may be neglected. In
the following, it shall be assumed that the resonant state is non-degenerate

| < n0 l0 m0 | ˆα0 (k 0 ) . p̂ | n00 l00 m00 > < n00 l00 m00 | ˆα (k) . p̂ | nlm > |2
   2 2  0 
dσ e ω
=
dΩ0 m2 c2 ω ( En − En00 + h̄ ω )2 + Γ2n00
(1068)
This expression can be re-expressed in terms of the product of two factors

e 2 π h̄ | < nlm | ˆα (k) . p̂ | n00 l00 m00 > |2 V

   2 
dσ
=
dΩ0 m2 ω V ( En − En00 + h̄ ω )2 + Γ2n00 c
V ω 02
 2 
2 π e 2 π h̄
× 2 0
| < n00 l00 m00 | ˆα0 (k 0 ) . p̂ | n0 l0 m0 > |2
h̄ m ω V ( 2 π )3 h̄ c3
(1069)

which is the probability for absorption from the ground state to the resonant
state | n00 l00 m00 > (divided by the incident flux) times the probability for its
decay via emission. On resonance, it appears that the process corresponds to
two sequential processes, first absorption and secondly emission.

For energies slightly off-resonance, the resonant scattering is expected to in-

terfere with the non-resonant scattering process. Likewise, if the resonant state
is degenerate, the sum over the degeneracy must be performed before the matrix
elements are squared leading to constructive interference.

The difference between a resonant process and two step process, is deter-
mined by the lifetime of the intermediate state | n0 l0 m0 > compared with the
frequency width of the photon beam. The frequency width of the photon beam
may be limited by the monochromator, or by the time-scale of the experiment
if it involves a pulsed light source. If the lifetime of the intermediate state is
sufficiently long compared with the time scale of experiment, it may be possi-
ble to observe the decay long after the incident light has been switched off. In

182
this case, the resonance can be considered to be composed of two independent
processes56 . Furthermore, it may be possible to perform further experiments
on the surviving intermediate state. In the opposite case, where the lifetime
of the intermediate state is shorter than the time-scale of the experiment, the
intermediate state will have decayed before the experiment has terminated.

11.2.5 Natural Line-Widths

The interaction representation will be used to calculate the natural width for
the absorption of light (k, α), by introducing the lifetimes of the initial and final
state. Strictly speaking, one should not take the exponential decay of a proba-
bility Pn (t) of finding an electron in state ψn too literally. If one considers the
approximate exponential decay as being rigorous, this implies that the Hamil-
tonian should be non-Hermitean which is strictly forbidden. One should think
of the decaying wave function as a wave packet or linear superposition of exact
energy eigenstates (with energies denoted by E). The Fourier transform of the
time-dependent wave function should provide the energy-distribution ρn (E) of
the exact energy eigenstates in the wave packet | ψn (t) >
Z ∞  
1 i
ρn (E) = dt exp + Et < ψn (0) | ψn (t) >(1070)
2 π h̄ −∞ h̄

On assuming the approximate form of a decaying wave packet

 
i |t|
< ψn (0) | ψn (t) > = exp − En t − (1071)
h̄ 2 τn

where the decay includes transitions to all possible final states, one finds
 
1 1 1
ρn (E) = −
2πi E − En − i 2 h̄τn E − En + i 2 h̄τn
h̄
1 2 τn
= (1072)
π ( E − En )2 + ( 2 h̄τ )2
n

This can only be an approximate form of the energy-distribution since the en-
ergy must be bounded from below. The existence of a lower-bound to en-
ergy distribution implies that the width of the electronic energy level has to
be energy-dependent 2 h̄τn = Γn as this must become zero below a threshold
energy. However, it should be noted that the width of the energy-distribution
will determine the approximate exponential decay. Since the perturbations in-
troduce an energy-dependent width to the wave packet, causality requires that
the energy-shift ∆En should also be energy-dependent. Hence, the effects of
the perturbation (such as the energy-shift and lifetime) should be described in
56 V. Weisskopf, Ann. der Physik, 9, 23 (1931).

183
terms of a self-energy Σn (E)

Σn (E) = <e Σn (E) + i =m Σn (E)

≈ ∆En − i Γn (1073)

The energy-dependent self-energy appears most naturally if one uses Brillouin-

Wigner perturbation theory to calculate the correction to an approximate energy
En . From second-order Brillouin-Wigner perturbation theory, one finds that the
energy-dependent self-energy, when evaluated just above the real E axis, is given
by
X | < n0 | ĤI | n > |2
Σn (E + iη) = (1074)
0
E + i η − E n0
n

This complex self-energy has a real and imaginary part. The imaginary part
can be thought of as occurring via amplification of the infinitesimal imaginary
term i η in the denominator, and can be seen to be non-zero when the energy
E of the component in the wave packet falls in the region when the spectral
density of the approximate En0 is finite. Hence, since the En0 are bounded from
below, then so is the energy-distribution ρn (E) since

1 =m Σn (E + iη)
ρn (E) = − (1075)
π ( E − En − <e Σn (E) )2 + ( =m Σn (E + iη) )2

The real part of the self-energy must also be energy-dependent, since it is related
to the imaginary part via the Kramer’s-Kronig relations
Z ∞
1 =m Σn (z + iη)
<e Σn (E) = − dz
π −∞ E − z
Z ∞
Pr <e Σn (z)
=m Σn (E + iη) = + dz (1076)
π −∞ E − z

where the Principal Part of an integral with a simple pole is defined as

Z ∞  Z ∞ Z − 
f (z) f (z) f (z)
Pr dz = lim dz + dz (1077)
−∞ z →0 + z −∞ z

Hence, the real part of the self-energy is also energy-dependent. The Kramers-
Kronig relation is an expression of causality.

Since the electronic states in the expression for the Fermi-Golden rule decay
rate
1 2π
= | < n0 l0 m0 | ĤI | nlm > |2 δ( En0 l0 − Enl − h̄ ω ) (1078)
τ nl→n0 l0 h̄
are to be interpreted as wave packets with a distribution of energies, the factor
expressing conservation of energy should be expressed in terms of the energy

184
conservation for the components of the wave packets. Hence, the decay rate
should be written as the convolution
Z ∞ Z ∞
1 2π
= | < n0 l0 m0 | ĤI | nlm > |2 dE 0 ρn0 l0 (E 0 ) dE ρnl (E) δ( E 0 − E − h̄ ω )
τ nl→n0 l0 h̄ −∞ −∞
Z ∞
2π 0 0 0 2
= | < n l m | ĤI | nlm > | dE ρn0 l0 ( E + h̄ ω ) ρnl (E) (1079)
h̄ −∞

We shall use the approximation for the energy distributions suggested by eqn(1072).
In this case, the convolution is evaluated by contour integration as
h̄ h̄
1 2 2 τn + 2 τn0 l
= | < n0 l0 m0 | ĤI | nlm > |2
τ nl→n0 l0 h̄ ( h̄ ω + Enl − En0 l0 )2 + ( 2 h̄τnl + h̄
2 τn0 l0 )
2

(1080)

since only the terms with poles on the opposite sides of the real-axis yield
non-zero contributions. From this, one can show that the optical absorption
cross-section is given by
1 1
4π

e2
 X ωn0 l0 ,nl ( 2 τn0 l0 + 2 τnl )
σabsorb (ω) = | < n0 l0 m0 | r | nlm > |2 1 1
3 h̄ c
n0 l0 m0
( ωn0 l0 ,nl − ω )2 + ( 2 τn0 l0 + 2 τnl )2
(1081)
which was first derived by Weisskopf and Wigner57 . Hence, the natural width
is given by the average of the decay rates for the initial and final electronic
states. This leads to the conclusion that even weak lines can be broad, if the
final electronic state has a short lifetime.

11.3 Renormalization and Regularization

Quantum Electrodynamics treats the interactions between charged particles and
the electromagnetic field, and often contains infinities. The zero-point energy of
the electromagnetic field is one such infinity. In most cases, these infinities can
be ignored since they are not measurable, since the infinities occur as modifica-
tions caused by the introduction of interactions between the charged particles
of a hypothetical system with an electromagnetic field. That is, the infinities
occur in the form of a renormalization of the quantities of the non-interacting
theory. These infinite renormalizations do not lead to the rejection of the theory
of Quantum Electrodynamics since the quantities of the non-interacting system
are not measurable. To be sure, the infinities occur in relations between hypo-
thetical quantities and physically measurable quantities, and so these infinities
can be ignored since the hypothetical quantities are undefined. However, it is
possible to use the theory to eliminate the unmeasurable quantities, thereby
yielding relations between physically measurable quantities to other physically
57 V. F. Weisskopf and E. Wigner, Z. Physik, 63, 54 (1930).

185
measurable quantities. In Quantum Electrodynamics, the infinities cancel in
equations which only contain physical measurable quantities. This fortunate
circumstance makes the theory of Quantum Electrodynamics renormalizable.

First, it shall be shown how the infinite zero-point energy of the electro-
magnetic field can lead to a (finite) physically measurable force between its
containing walls.

11.3.1 The Casimir Effect

The zero-point energy of the electromagnetic field can lead to measurable ef-
fects. In general relativity, the total energy including the zero-point energy
of the electromagnetic radiation is the source for the gravitational field. The
Casimir effect58 shows that the zero-point energy of the electromagnetic radi-
ation produces a force on the walls of the cavity. We shall consider a cubic
volume V = L3 which is enclosed by conducting walls that acts as a cavity
for the electromagnetic radiation. This volume is divided into two by a metallic
partition, which is located at a distance d from one side of the cavity. We shall

L-d d

Figure 34: The geometry of the partitioned electromagnetic cavity used to con-
sider the Casimir effect.

evaluate the total energy of this configuration and then deduce the form of the
interaction between the partition and the walls of the cavity.

We shall consider the total energy due to the zero-point fluctuations in the
container. Since the zero-point energy is divergent due to the presence of arbi-
trarily large frequencies, we shall introduce a convergence factor. The introduc-
tion of a convergence factor to remove infinities is the process of regularization.
The introduction of the convergence factor can be motivated by the observation
that, in matter, electromagnetic radiation becomes exponentially damped at
58 H. B. G. Casimir, Physica 19, 846 (1953).

186
large frequencies. Hence, one can write
 
1 X ωk,α
E = h̄ ωk,α exp − λ (1082)
2 c
k,α

and then take the limit λ → 0.

The presence of the conducting walls introduces boundary conditions such

that the EM field is zero at every boundary. The boundary conditions restrict
the allowed values of k so that the components satisfy
ki L = π ni (1083)
for i = x, y and ni are positive integers. The boundary condition for the re-
maining two boundaries leads to the restriction
kz d = π nz (1084)
The energy of the radiation in one part of the partition can be expressed as
s s
Z ∞ Z ∞ ∞ 2 2 
L2
  
X n z π nz π
Ed = h̄ c 2 dkx dky kx2 + ky2 + exp − λ kx2 + ky2 +
π 0 0 n =1
d d
z
(1085)
where the two polarizations have been summed over. The integration has cylin-
drical symmetry but only extends over the quadrant with positive kx and ky ,
therefore, it shall be re-written as
s s
2 Z ∞ ∞  2   2 
2L X nz π nz π
Ed = h̄ c dk k k2 + exp − λ k 2 +
4π 0 n =1
d d
z
(1086)
or, on changing variable to the dimensionless κ = k 2 ( nzd π )2
∞ 3 Z ∞
L2 X √ nz π λ √
  
nz π
Ed = h̄ c dκ κ + 1 exp − κ + 1
4 π n =1 d 0 d
z
(1087)
The factor of n3z can be expressed as a third-order derivative of the exponential
factor w.r.t. λ
∞ Z ∞
L2 X ∂3 nz π λ √
  
1
Ed = − h̄ c dκ exp − κ + 1
4 π n =1 0 κ + 1 ∂λ3 d
z
(1088)
The summation over nz can be performed, leading to
Z ∞ √
L2 ∂3 exp[ − πdλ κ + 1 ]
 
1
Ed = − h̄ c dκ √
4π 0 κ + 1 ∂λ3 1 − exp[ − πdλ κ + 1 ]
Z ∞
L2 ∂3
 
1 1
= − h̄ c dκ √ (1089)
4π 0 κ + 1 ∂λ3 exp[ πdλ κ + 1 ] − 1

187
 √
Let t = κ + 1 so
∞
2 L2 dt ∂ 3
Z  
1
Ed = − h̄ c π λ
(1090)
4π 1 t ∂λ3 exp[ d t] − 1

The factor of t−1 can be eliminated by performing one of the differentials with
respect to λ.
Z ∞
h̄ c L2 ∂2 exp[ πdλ t ]
 
Ed = dt (1091)
2d 1 ∂λ2 ( exp[ πdλ t ] − 1 )2

We shall set
πλt
s = exp[ ] − 1 (1092)
d
therefore
∞
h̄ c L2 ∂ 2
 Z
d ds
Ed = (1093)
2 d ∂λ2 πλ s0 s2

where the lower limit of integration depends on λ and is given by

πλ
s0 = exp[ ] − 1 (1094)
d
The integration can be performed trivially, yielding

h̄ c L2 ∂ 2
 
d
Ed =
2 d ∂λ2 π λ s0
d
h̄ c L2 ∂ 2
 
π λ
=
2 d ∂λ2 exp[ πdλ ] − 1
2  π λ
h̄ c L2 ∂ 2
 
d d
= (1095)
2 d ∂λ2 πλ exp[ πdλ ] − 1

The last factor in the above expression can be expanded as

∞
x X xn
= Bn (1096)
exp[x] − 1 n=0
n!

where Bn are the Bernoulli numbers, which are given by B0 = 1, B1 =

− 21 , B2 = 16 , B3 = 0, B4 = − 30 1
, etc. Therefore, the energy of the
electromagnetic cavity at zero temperature, is finite for a finite value of the
cut-off λ but diverges as λ−4 when λ → 0. The zero point energy of the cavity
can be expressed as
n−4 
h̄ c L2 π 2 X Bn (n − 2)(n − 3)
 
πλ
Ed = (1097)
2 d3 n=0 n! d

188
where the n = 0 term diverges as λ−4 in the limit as λ → 0 and is proportional
to the volume of the cavity d L2 . The term with n = 1 also diverges, but
diverges as λ−3 and has the form of a surface energy since it is proportional to
L2 . The terms with n = 2 and n = 3 are identically equal to zero. The term
with n = 4 remains finite in the limit λ → 0 and all the higher-order terms
vanish in this limit. Explicitly, one has

h̄ c L2 6 B0 d2 2 B4 π 2
 
2 B1 d
Ed = + + + O(λ) (1098)
2d π 2 λ4 π λ3 4! d2

The first term in the energy is proportional to L2 d, which is the volume of the
cavity and the second term is proportional to L2 the surface area of the walls.
The third term is independent of the cut-off and the higher order terms vanish
in the limit λ → 0.

The Casimir force is the force between two planes, which originates from
the energy of the field59 . This energy can be separated out into a volume
part and parts due to the creation of the surfaces and an interaction energy
between the surfaces. In order to eliminate both the volume dependence of the
energy and the surface energies, we are considering two configurations of the
partitions in the cavity. In one configuration the plane divides the volume into
two unequal volumes d L2 and (L − d) L2 , and the other configuration is a
reference configuration where the cavity is partitioned into two equal volumes
L3
2 . The difference of energies for these configurations is given by

∆E = Ed + EL−d − 2 E L (1099)
2

In the limit L → ∞ this is expected to reduce to the energy of interaction

between the planes separated by distance d. Since the volume and surface areas
of the two partitions are identical, one finds that the difference in energy of the
two configurations is finite and is given by

π 2 h̄ c L2
lim ∆E → − (1100)
Ld,λ→0 720 d3
The d-dependence of the energy difference leads to an attractive force between
the two plates separated by a distance d, which is the Casimir force

π2 L2
F = − h̄ c 4 (1101)
240 d
The force is proportional to L2 which is the area of the wall of the cavity. The
predicted force was measured by Sparnaay60 . A more recent experiment involv-
ing a similar force between a planar surface and a sphere has achieved greater
59 Our considerations only include the part of Fock space that corresponds to having zero

numbers of excited quanta. Hence, the Casimir force is due to the properties of the field, and
is not due to the transmission of real particles (photons) between the planes.
60 M. J. Sparnaay, Physica 24, 751 (1958).

189
 ter such a large perturbation, the feedback system required 6 3 1014 Hz, Eq. (5) gives a correction of order 20%
several minutes to reestablish equilibrium. at the closest spacings; our data does not support such
Assuming that the functional form for the Casimir force a deviation. However, the simple frequency dependence
is correct, its magnitude was determined by using linear of the electrical susceptibility used in the derivation of
least squares to determine a parameter d for each sweep Eq. (5) is not correct for Au, the index of refraction of
such that which has a large imaginary component above the plasma
Fcm sai d ­ s1 1 ddFcT sai d 1 b 0 . (9) frequency; a rough estimate using the tabulated complex
index [14] limits the conductivity correction as no larger
In this context, b 0 should be zero, and for the complete than 3%, which is consistent with our results [15].
data set, b 0 , 5 3 1027 dyn (95% confidence level). I thank Dev Sen (who was supported by the UW NASA
The average over the 216 sweeps gives d ­ 0.01 6 0.05, Space Grant Program) for contributions to the early stages
and this is taken as the degree of precision of the of this experiment, and Michael Eppard for assistance
measurement. There was no evidence for any variation with calculations.
of d depending on the region of the plates used for the
measurement.
The most striking demonstration of the Casimir force
is given in Fig. 4. The agreement with theory, with no
*Present address: Los Alamos National Laboratory,
Neutron Science and Technology Division P-23, M.S.
H803, Los Alamos, NM 87545.
[1] H. B. G. Casimir, Koninkl. Ned. Adak. Wetenschap. Proc.
51, 793 (1948).
[2] E. Elizalde and A. Romeo, Am. J. Phys. 59, 711 (1991).
[3] V. M. Mostepanenko and N. N. Trunov, Sov. Phys. Usp.
31, 965 (1988).
[4] M. J. Sparnaay, Physica (Utrecht) 24, 751 (1958).
[5] C. I. Sukenik, M. G. Boshier, D. Cho, V. Sangdohar, and
E. A. Hinds, Phys. Rev. Lett. 70, 560 (1993).
[6] E. M. Lifshitz, Sov. Phys. JETP 2, 73 (1956).
[7] T. H. Boyer, Phys. Rev. 174, 1764 (1968).
[8] J. Blocki, J. Randrup, W. J. Swiatecki, and C. F. Tsang,
Ann. Phys. (N.Y.) 105, 427 (1977).
[9] J. Schwinger, L. L. DeRaad, Jr., and K. A. Milton, Ann.
Phys. (N.Y.) 115, 1 (1978).
[10] J. Mehra, Physica (Utrecht) 37, 145 (1967).
[11] L. S. Brown and G. J. Maclay, Phys. Rev. 184, 1272
(1969).
[12] G. Ising, Philos. Mag. 1, 827 (1926).
[13] W. R. Smythe, Static and Dynamic Electricity (McGraw-
FIG. 4. Top: All data with electric force subtracted, averaged Hill, New York, 1950), pp. 121 –122.
into bins (of varying width), compared to the expected Casimir [14] CRC Handbook of Chemistry and Physics, 76th Ed. (CRC
Figure 35: The separation-dependent force between two closely spaced metallic
force for a 11.3 cm spherical plate. Bottom: Theoretical
Casimir force, without the thermal correction, subtracted from
Press, Boca Raton, 1995), pp. 12 –-130.
[15] S. Hacyan, R. Jauregui, F. Soto, and C. Villarreal, J. Phys.
surfaces due to the modification of the zero-point energy. The lower panel shows
top plot; the solid line shows the expected residuals. A 23, 2401 (1990).

the difference between 8the experimental results and the theoretical prediction
for the Casimir Force. [After S. K. Lamoreaux, Phys. Rev. Lett. 78, 5, (1997).]

accuracy61 .

To summarize, the physical quantity is the force or difference in energies

when one wall is moved. When the change in energy is calculated, the differ-
ence between the two divergent energies is finite and independent of the choice
of cut-off62 .

Cut-Off Independence

It is the boundary condition and not the cut-off that plays an important role
in the Casimir effect. For simplicity, one can choose zero boundary conditions.
The zero-point energy of a cylindrical electromagnetic cavity of radius R and
length d can be expressed as the sum
s 2  s
∞ Z 2 
h̄ c π R2 X
 
2
π n z 2
π nz
Ed = 2 dkρ kρ kρ + F kρ +
2 ( 2 π ) n =1 d d
z
(1102)
where F (z) is an arbitrary cut-off function (which may depend on an arbitrary
parameter λ which is ultimately going to be set to zero). The cut-off must not
effect the low energy-modes so one can choose F (0) = 1 and all the derivatives
of F (z) to be zero for finite values of z. These assumptions are all in accord with
61 S. K. Lamoreaux, Phys. Rev. Lett. 78, 5 (1997).
62 The independence of any cut-off procedure can be shown by evaluating the divergent sums
by using the Euler-Maclaurin summation formula.

190
 1.2

0.8

F(z)
0.4

0
0 0.2 0.4 0.6 0.8 1
z/N

Figure 36: The schematic form of the cut-off function F (z).

the ideal case of no cut-off function or F (z) = 1. The energy can be written as
∞
h̄ c R2 X
Ed = f (nz ) (1103)
2 n =1 z

where
s 2 s 2 
Z ∞  
π nz π nz
f (nz ) = dkρ kρ kρ2 + F kρ2 + (1104)
0 d d
The summation can be performed by changing it into an integral, however the
corrections due to smoothing will be kept. This is accomplished by the Euler-
Maclaurin formula. The integral between 0 and N of a function can be roughly
expressed as a summation
Z N N −1
1 X 1
dx f (x) ≈ f (0) + f (n) + f (N ) (1105)
0 2 n=1
2

by choosing to approximate the integral by the area under a histogram where

the x variable is binned into intervals of width unity centered around x = n.
The corrections at n = 0 and n = N are needed to account for the fact that the
range of integration excludes half the width of the rectangular blocks centered on
n = 0 and n = N . The Euler-Maclaurin formulae is equivalent to finding a good
smooth polynomial fit to the integrand, and then integrating the polynomial.
It generates corrections which are given by the derivatives at the end points
Z N N −1
1 X 1
dx f (x) = f (0) + f (n) + f (N )
0 2 n=1
2
B2 B4
+ ( f (1) (0) − f (1) (N ) ) + ( f (3) (0) − f (3) (N ) ) + . . .
2! 4!
(1106)

191
We shall assume that f (n) and all its derivatives vanishes in the limit of large
n, limN →∞ f (N ) → 0, due to the behavior of the cut-off function. The
corrections in the Euler-Maclaurin summation formulae can be evaluated by
noting that the first derivative of f (n) with respect to n is given by
Z ∞ s 2 
π2 n

(1) kρ 2 +
πn
f (n) = dk ρ F kρ (1107)
d2
q
0 k 2 + ( π n )2 d
ρ d

since the derivatives of F (z) all vanish for finite z. The integration over the
variable kρ is re-expressed in terms of an integration over the variable z, defined
by s  2
πn
z = kρ2 + (1108)
d
so
kρ
dz = dkρ s (1109)
  2
π n
kρ2 + d

The integration is evaluated through integration by parts

Z ∞
(1) π2 n
f (n) = dz F (z)
d2 πn
d
Z ∞
π2 n
 
∂z
= dz F (z)
d2 πn
d
∂z
∞
π2 n 
= 2
z F (z)
d πn
d

π 3 n2
= − (1110)
d3
In deriving the above expression, the condition that the first-order derivative of
F (z) vanishes for finite z has been used. It immediately follows that

2 π3 n
f (2) (n) = − (1111)
d3
and
2 π3
f (3) (n) = − (1112)
d3
and all higher order derivatives vanish. Hence, one finds that at z = 0 all the
m-th order derivatives f (m) (0) vanish, except for m = 3 which is given by

2 π3
f (3) (0) = − (1113)
d3

192
Hence, on evaluating the energy of the cylindrical cavity (and using the zero
boundary conditions), one finds that the energy is composed of the sum of an
integral and a finite number of other terms. The integral part of the expression
only depends on the volume of the cavity and is proportional to a divergent
integral, and hence drops out when the energy differences are taken. The only
terms that yield non-zero contributions to the energy difference originate with
f (3) (0) and depend on d. It is these terms that give rise to the Casimir force.
This approach also showed that any particular choice made for the cut-off is
irrelevant.

Mathematical Interlude:
The Euler-Maclaurin Summation Formula.

The Euler-Maclaurin formula allows one to accurately evaluate the difference

of finite summations and their approximate evaluations in the form of integrals.

The Euler-Maclaurin Formula

If N is an integer and f (x) is a smooth differentiable function defined for all

real values of x between 0 and N , then the summation
N
X −1
S = f (n) (1114)
n=1

can be approximated by an integral

Z N
I = dx f (x) (1115)
0

In particular, by utilizing the “trapedoizal rule”, one expects that

 
1
I ∼ S + f (0) + f (N ) (1116)
2
The Euler-Maclaurin formula provides expressions for the difference between the
sum and the integral in terms of the higher-derivatives f (n) at the end points
of the interval 0 and N . For any integer p, one has
  p  
1 X B2n
S + f (0) + f (N ) − I = f 2n−1 (N ) − f 2n−1 (0) + R
2 n=1
(2n)!
(1117)
where B1 = -1/2, B2 = 1/6, B3 = 0, B4 = -1/30, B5 = 0, B6 = 1/42, B7 =
0, B8 = -1/30, ... are the Bernoulli numbers, and R is an error term which is
normally small if the series on the right is truncated at a suitable value of p.

The Remainder Term

193
 The remainder R when the series is truncated after p terms is given by
Z N
Pp+1 (x)
R = (−1)p dx f (p+1) (x) (1118)
0 (p + 1)!

where Pn (x) = Bn (x − [x]) are the periodic Bernoulli polynomials. The remain-
der term can be estimated as
Z N
2
|R| ≤ dx | f 2p−1 (x) | (1119)
(2π)p 0

Derivation by Induction

First we shall examine the properties of the Bernoulli polynomials and the
Bernoulli numbers. Then we shall indicate how the Euler-Maclaurin formula
can be obtained by induction.

The Bernoulli polynomials Bn (x), for n = 0, 1, 2, ... are defined by the

generating function expansion
∞
z ezx X zn
G(z, x) = z
= Bn (x) (1120)
e −1 n=0
n!

Furthermore, when x = 0, one has

∞
z X zn
G(z, 0) = = B n (1121)
ez − 1 n=0
n!

where Bn are the Bernoulli constants. Hence, the Bernoulli constants are the
Bernoulli polynomials evaluated at x = 0, i.e. Bn (0) = Bn . Furthermore, on
differentiating the generating function w.r.t. x, one finds

∂G(z, x)
= z G(z, x) (1122)
∂x
which implies that
∞ ∞
X ∂Bn (x) z n X zn
= z Bn (x) (1123)
n=0
∂x n! n=0
n!

On equating the coefficients of z n in the above equation, one obtains the im-
portant relation
∂Bn (x)
= n Bn−1 (x) (1124)
∂x
Therefore, by integration it easy to show that Bn (x) are polynomials of degree
n. The first few Bernoulli polynomials can be explicitly constructed from the

194
generating function expansion. The first few polynomials are given by

B0 (x) = 1
1
B1 (x) = x−
2
1
B2 (x) = x2 − x +
6
3 1
B3 (x) = x3 − x2 + x
2 2
1
B4 (x) = x4 − 2x3 + x2 −
30
5 5 1
B5 (x) = x5 − x4 + x3 − x
2 3 6
... (1125)

From the generating function expansion, one can show that the Bernoulli poly-
nomials are either even or odd functions of x − 12 . The generating function can
be expressed
∞
zn
 
z(x− 12 ) z X
G(z, x) = e z z = Bn (x) (1126)
e − e− 2
2
n=0
n!

where the second factor is an even function of z, thus, the generating function is
invariant under the combined transformation z → −z and (x − 12 ) → −(x − 12 ).
Therefore, one has
∞ ∞
zn zn
   
X 1 1 X 1 1
Bn +x− = Bn + − x ( − 1 )n (1127)
n=0
2 2 n! n=0
2 2 n!

so the polynomials satisfy

Bn (x) = ( − 1 )n Bn (1 − x) (1128)

In particular for x = 1, one has

Bn (1) = ( − 1 )n Bn (0) (1129)

The generating function with x = 0 can be re-written as the sum of its even
and odd parts
z ∞
zn
 
2 z X
G(z, 0) = − = Bn (0) (1130)
tanh z2 2 n=0
n!

The even part has only even terms in its Taylor expansion, and there is only
one term in the odd part. Hence, the odd Bernoulli numbers vanish for n > 1,
i.e. B2n+1 (0) = 0 for n > 0. Therefore, for n ≥ 2, one has Bn (0) = Bn (1). This
equality can be used to evaluate the integrals of the Bernoulli polynomial over

195
the range from 0 to 1. On expressing the integral of Bn (x) in terms of Bn+1 (x),
one has
Z 1 Z 1
1 ∂Bn+1 (x)
dx Bn (x) = dx
0 (n + 1) 0 ∂x
Bn+1 (1) − Bn+1 (0)
=
( n+1 )
= 0 for n ≥ 1 (1131)

Hence, the Bernoulli polynomials may be defined recursively via the relation

∂Bn (x)
= n Bn−1 (x) (1132)
∂x
if the constant of integration is fixed by
Z 1
dx Bn (x) = 0 for n ≥ 1 (1133)
0

The periodic Bernoulli functions Pn (x) can be defined by

Pn (x) = Bn (x − [x]) (1134)

where [x] is the integral part of x. This definition of Pn (x) reproduces to the
Bernoulli polynomials on the interval (0, 1) since [x] = 0 in this interval. The
functions Pn (x) are periodic over an extended range of x with period 1.

The Euler-Maclaurin formula can be obtained by mathematical induction.

Consider the integral
Z n+1 Z n+1
∂v
dx f (x) = dx u (1135)
n n ∂x

with the identification of

u = f (x) (1136)
and
∂v
= 1 = P0 (x) (1137)
∂x
since P0 (x) = 1. Therefore, on using the recursion relation involving the
derivative of the Bernoulli polynomials, one finds that

v = P1 (x) (1138)

Integrating by parts, one obtains

Z n+1 Z n+1
n+1 ∂f (x)
dx f (x) = [ f (x) P1 (x) ]n − dx P1 (x) (1139)
n n ∂x

196
but since the periodic Bernoulli polynomial P1 (x) is given by
1
P1 (x) = (x − [x]) − (1140)
2
it has the value of 1/2 at the limits of integration. Hence, the integration reduces
to
Z n+1   Z n+1
f (n + 1) + f (n) ∂f (x)
dx f (x) = − dx P1 (x) (1141)
n 2 n ∂x

Summing the above expression from n = 1 to n = N − 1, yields

Z N   N −1 Z N
f (1) + f (N ) X ∂f (x)
dx f (x) = + f (n) − dx P1 (x) (1142)
1 2 n=2 1 ∂x
 
f (1) + f (N )
Adding 2 to both sides of the equation and rearranging, one finds

N Z N   Z N
X f (1) + f (N ) ∂f (x)
f (n) = dx f (x) + + dx P1 (x) (1143)
n=1 1 2 1 ∂x

The last two terms, therefore, give the error when the sum is approximated by
an integral. The first correction is simply the end point corrections from the
“trapezoidal rule”, and the second correction has to be evaluated to yield the
Euler-Maclaurin formula. The last correction is of the form of an integral which
can be expressed in terms of the sum of the integrals
Z n+1
dx f 0 (x) P1 (x) (1144)
n

where the prime refers to the derivative of f (x) w.r.t. x. The above expression
can be evaluated by integrating by parts. The integrand is re-written as
Z n+1 Z n+1
∂v
dx f 0 (x) P1 (x) = dx u (1145)
n n ∂x

where one identifies the two factors as

u = f 0 (x)
∂v
= P1 (x) (1146)
∂x
Since the indefinite integral is evaluated as
Z x
1
dx0 P1 (x0 ) = P2 (x) (1147)
2

197
the integration by parts yields
Z n+1 n+1 Z n+1
P2 (x) f 0 (x)

0 1
dx P1 (x) f (x) = − dx f 00 (x) P2 (x) (1148)
n 2 n 2 n
However, one has P2 (0) = P2 (1) = B2 , therefore the above expression simplifies
to
Z n+1  0 Z n+1
f (n + 1) − f 0 (n)

0 1
dx P1 (x) f (x) = B2 − dx f 00 (x) P2 (x)
n 2 2 n
(1149)
Then, on summing the above expression from n = 1 to n = N − 1, one finds
Z N  0 Z N
f (N ) − f 0 (1)

0 1
dx P1 (x) f (x) = B2 − dx f 00 (x) P2 (x) (1150)
1 2 2 1
This yields the first term in the series of end point corrections in the Euler-
Maclaurin formula, where the correction is the sum of the first derivatives at
the end points multiplied by B2 /2!. The above process can be iterated yielding
a complete proof of the Euler-Maclaurin summation formula.

In order to get bounds on the size of the error when the sum is approximated
by the integral, we note that the Bernoulli polynomials on the interval [0, 1] at-
tain their maximum absolute values at the endpoints and the value Bn (1) is the
n-th Bernoulli number.

References

T. M. Apostol, ”An Elementary View of Euler’s Summation Formula”,

American Mathematical Monthly, 106, 409-418 (1999).

D. H. Lehmer, ”On the Maxima and Minima of Bernoulli Polynomials”,

American Mathematical Monthly, 47, 533-538 (1940).

11.3.2 The Lamb Shift

The Lamb shift is a shift between the energy levels of the 2 2 S 12 and the 2 2 P 12
levels of hydrogen from the predictions of the Dirac equation as they have the
same n and j values (j = l ± s). The Dirac equation predicts that these two
levels should be degenerate. However, these levels were measured by Lamb and
Retherford63 who found that the 2 2 S 21 level is higher than the 2 2 P 12 level
by 1058 MHz or 0.033 cm−1 . Bethe explained this in terms of the interaction
between the bound electron and the quantized electromagnetic field64 . Simi-
lar shifts should also occur between the n 2 S 12 and the n 2 P 12 levels, but the
63 W. E. Lamb Jr. and R. C. Retherford, Phys. Rev.72, 241 (1947).
64 H. A. Bethe, Phys. Rev. 72, 339 (1947).

198
magnitude of the shifts should be much smaller, as the magnitude varies as n−3 .

Qualitatively, the electron interacts with the fluctuating electromagnetic

field and with the potential due to the nucleus. The zero-point fluctuations
cause the electron to deviate from its quantum orbit by an amount given by ∆r
and, therefore, experiences a potential given by
1
V (r + ∆r) = V (r) + ∆r . ∇V (r) + ( ∆r . ∇ )2 V (r) + . . . (1151)
2!
so one expects an energy-shift given by
1
∆E = < ∆r2 > < nlm | ∇2 V (r) | nlm > (1152)
3 . 2!
Due to the form of the Coulomb potential

e2
V (r) = − (1153)
r
the Laplacian is related to a point charge density at the nucleus

∇2 V (r) = 4 π e2 δ(r) (1154)

Hence, the shift due to the fluctuations in the electron’s potential energy occurs
primarily at the origin. The effect of the electromagnetic fluctuations on the
kinetic energy are not state specific, and can be considered as a uniform shift
of all the energy levels, like the electron’s rest mass energy m c2 . Thus, the
relative energy shift of the levels is solely determined by the potential at the
origin. Therefore, the states with non-zero angular momenta do not experience
the relative energy-shift since the electronic wave functions vanish at the origin.
Thus, only the 2s state experiences a shift but the 2p state is unshifted.

The magnitude of the Lamb shift can be ascertained by expressing ∆r in

terms of the zero-point fluctuations in the electromagnetic field65 . If it is as-
sumed that the electron is bound to the atom harmonically, ∆r is determined
from the equation of motion
q
∆r̈ + ω02 ∆r = E (1155)
m
where the electric field E has components that are fluctuating with wave vector k
or equivalently with frequency ω. This has the result that the position fluctuates
at the frequency ω with an amplitude given by
q 1
∆rω = Eω 2 (1156)
m ω0 − ω 2
65 T. A. Welton, Phys. Rev. 74, 1557 (1948).

199
 ∆r(t)

Figure 37: A cartoon depicting the modification of the classical orbit of an

electron due to the zero-point fluctuations of the electromagnetic field.

where Eω is the Fourier component of the fluctuating electric field. Hence, the
ω-component of the mean squared fluctuation66 in the particle’s position is given
by
 2
q Eω2
< | ∆r2ω | > = < | 2 | > (1157)
m ( ω0 − ω 2 )2
On approximating the electromagnetic energy associated with the fluctuating
electromagnetic field < | Eω2 | > by the half the sum of the zero-point energies
of the photon modes, one has
V 1
< | Eω2 | > = 2 h̄ ω (1158)
8π 4
where the factor 2 represents the two types of polarization of the normal modes.
Therefore, on summing over the normal modes, one finds that the mean squared
deviation of the electron’s trajectory from the classical orbit is proportional to
Z ∞
Eω2
Z
V
3
dΩ dω ω 2 < | 2 | >
(2πc) 0 ( ω0 − ω 2 )2
Z ∞
ω3
Z
4 π h̄ V
= 3
dΩ dω 2 (1159)
V (2πc) 0 ( ω0 − ω 2 )2
The integration over ω can be approximated as
mc2
m c2
Z h̄ dω
= ln (1160)
ω0 ω h̄ ω0
where an upper and lower cut-off have been introduced to prevent the integral
from diverging67 . The expectation value of the second derivative of the potential
66 The average squared fluctuation of the electromagnetic field should, in principle, be cal-

culated as an average over a volume in time and space which encompasses the electron’s
trajectory.
67 The upper limit can be considered as being determined by the spatial dimension of the

volume in which the electromagnetic fluctuations are being averaged over. The divergence at
the lower limit of integration is unphysical and is caused by the neglect of the lifetimes of
the electronic states. The inclusion of the lifetimes result in the integrand being finite at the
resonance frequency ω0

200
for the 2s state is given by
1
< | ∇2 V | > = 4 π e2 (1161)
π a3
where the second factor represents the 2s electron density at the origin. The
corresponding factor for an ns level is expected to vary proportionally to n−3 .
Combining the above expressions, one finds that the 2s level is shifted by an
energy given by
 2 3 
m e4 m c2

4 e
∆E2s = ln (1162)
2 π h̄ c h̄2 h̄ ω0
where the frequency of the electron’s orbit ω0 has been chosen as a lower cut-off
on the frequency of the electromagnetic fluctuations. Since the logarithmic fac-
e2
tor is approximately given by ∼ −2 ln h̄c , one can see that the above estimate
is consistent with the Lamb shift having a magnitude of approximately 4.372
×10−6 eV.

11.3.3 The Self-Energy of a Free Electron

The corrections to the energy of a free electron due to its coupling to the elec-
tromagnetic field are to be considered68 . It shall be assumed that the electro-
magnetic field is in the ground state | {0} > , and the energy of an electron in
a state with momentum q will be evaluated via perturbation theory.

The lowest-order correction to the electron’s energy comes from the diamag-
netic interaction. From first-order perturbation theory, one finds the correction

∆Eq(1) = < q {0} | Ĥdia | q {0} > (1163)

On using a plane-wave to represent the electronic wave function

 
1
ψq (r) = √ exp i q . r (1164)
V
then the first-order change in the electron’s energy due to the coupling to the
field is given by
e2 2 π h̄ c2 ˆα (k) . ˆα0 (k 0 )
  X  
∆Eq (1)
= √ < {0} | ak0 ,α0 a†k,α | {0} >
2 m c2 0 0
V ωk ωk 0
k,α,k ,α
Z      
1
× d3 r exp − i q . r exp i ( k 0 − k ) . r exp i q . r
V
68 W. Heisenberg and W. Pauli, Z. Physik, 56, 1 (1929),

W. Heisenberg and W. Pauli, Z. Physik, 59, 168 (1930).

I. Waller, Z. Physik, 59, 168 (1930).
I. Waller, Z. Physik, 61, 721 & 837 (1930).
I. Waller, Z. Physik, 62, 673 (1930).

201
 (k,α)

q q

Figure 38: The first-order correction to the rest mass of the electron due to the
diamagnetic interaction.

e2 2 π h̄ c2 ˆα (k) . ˆα0 (k 0 )

  X  
= √ δk,k0 δα,α0 (1165)
2 m c2 V ωk ωk 0
k,α,k,α0

since the electronic matrix elements give rise to the condition of conservation of
momentum. Hence, the correction to the energy is found as

e2
    Z
(1) V 2 π h̄ c 1
∆Eq = 2 3
2 d3 k
2mc (2π) V k
Z ∞
e2
   
V 2 π h̄ c
= 8π dk k
2 m c2 ( 2 π )3 V 0
 2  Z ∞
e h̄
= dk k (1166)
πmc 0

which diverges. This contribution is independent of the electron’s momentum

q, and since k = k 0 it can be seen that the contribution of the diamagnetic
interaction to first-order is independent of the quantum state of the electron.
This contribution to the electron’s energy can be lumped together with the elec-
tron’s rest-energy m c2 . However, since the corrections are being evaluated for
non-relativistic electrons for which the rest energy is not observable, it is cus-
tomary to ignore the rest-energy and, therefore, this correction shall no longer
be considered.

The paramagnetic interaction, when taken to second-order, also yields a

correction to the electron’s self-energy. This correction can be considered to be
due to a virtual process in which the electron emits a photon and then re-absorbs
it. The second-order correction to the energy is evaluated from

X < q {0} | Ĥpara | q 0 1k,α > < q 0 1k,α | Ĥpara | q {0} >
∆Eq(2) =
0
Eq0 + h̄ ωk − Eq
q ,k,α
(1167)

202
 (k,α)

q-k
q q

Figure 39: The second-order self-energy correction of a free electron due to

the paramagnetic interaction. The electron with momentum q emits a virtual
photon with momentum k and then reabsorbs it.

where | q 0 1k,α > is a one-photon intermediate state of the electron-photon

system. We assume that the process does not conserve energy, so that the
denominator is finite. The matrix elements are evaluated as
s
0 2 π h̄ c2
< q 1k,α | Ĥpara | q {0} > = h̄ ˆα (k) . q
V ωk
Z      
1 3 0
× d r exp − i q . r exp − i k . r exp i q . r
V
s
2 π h̄ c2
= h̄ ˆα (k) . q δq0 +k−q (1168)
V ωk

which leads to momentum conservation. The second-order correction to the

electron’s energy takes the form

| h̄ q . ˆα (k) |2
 2  X 
2 π h̄ c2

e
∆Eq(2) = 2 2 h̄2 (q−k)2
(1169)
m c V ωk 2
h̄ q 2
− − h̄ ω
k,α 2 m 2 m k

On summing over the polarizations by using the diadic completeness relation69

X
ˆα (k) ˆα (k) = Iˆ − k̂ k̂ (1170)
α
69 The
completeness relation merely expresses the fact that any vector in a three-dimensional
space can be expressed in terms of the components along three orthogonal directions êi
3
X
A = Ai êi
i=1

where the components are given by the scalar product

Ai = A . êi
Hence, the completeness relation follows as
X
I = êi êi
i
.

203
one finds that the numerator is given by
X
h̄2 | q . ˆα (k) |2 = h̄2 q 2 ( 1 − cos2 θ ) (1171)
α

where θ is the angle between q and k

q . k = q k cos θ (1172)

Hence, one has

h̄2 q 2 ( 1 − cos2 θ )
 2 
2 π h̄ c2
Z  
(2) e V 3
∆Eq = d k h̄2 (q−k)2
m2 c2 ( 2 π )3 V ωk h̄2 q 2
2 m − 2 m − h̄ ωk
∞ π 2
e2 h̄ h̄ q 2 ( 1 − cos2 θ )
  Z Z
= 2
dk k dθ sin θ h̄2 q k 2 2
2πm c 0 0 cos θ − h̄ k − h̄ c k
m 2 m
(1173)

This contribution can be written as being explicitly proportional to the kinetic

energy of the electron, and a factor of k can be cancelled from the numerator
and the denominator
Z ∞
h̄2 q 2
 2 Z π
e 2 ( 1 − cos2 θ )
∆Eq(2) = dk dθ sin θ
2m h̄ c π 0 0 2 q cos θ − k − 2 m h̄
c

(1174)

It should be evident that the integral diverges logarithmically at large k. The

divergent part of the integral can be written as
Z ∞
h̄2 q 2
 2 Z π
e 2 dk
∆Eq(2) ∼ − dθ sin θ ( 1 − cos2 θ )
2m h̄ c π 0 2mc
h̄
k
2 2  2  Z ∞
h̄ q e 8 dk
= − (1175)
2m h̄ c 3 π 2mc h̄
k

If an upper cut-off λ−1

+ is introduced, then the correction to the electron’s kinetic
energy can be estimated as

h̄2 q 2 8
 2  
(2) e h̄
∆Eq = − ln (1176)
2 m 3 π h̄ c 2 m c λ+

This shift can be interpreted as a (second-order) renormalization of the electron’s

mass from the un-renormalized mass to the physical mass m∗
  2   
1 1 8 e h̄
= 1 − ln + ... (1177)
m∗ m 3 π h̄ c 2 m c λ+

It is the renormalized mass m∗ which would be determined by an experiment.

204
11.3.4 The Self-Energy of a Bound Electron
The Lamb shift (a quantum electrodynamic shift of the 2s level of Hydrogen
upwards by 1058 MHz) is caused the self-energy of a bound electron. The
self-energy of the state nlm can be estimated from second-order perturbation
theory using the dipole approximation, as is appropriate for a completely non-
relativistic calculation. The second-order shift is given by
2 π h̄ c2 | < n0 l0 m0 | ˆα (k) . p̂ | nlm > |2
 2   
(2) e X
∆Enlm =
m2 c2 0 0 0
V ωk Enlm − En0 l0 m0 − h̄ ωk
k,α,n l m
(1178)
On summing over the polarizations using the completeness relation, one obtains

(2)
 2 
e h̄ c
Z ∞ Z π X | < n0 l0 m0 | p̂ | nlm > |2 ( 1 − cos2 θk )
∆Enlm = dk k dθ k sin θ k
m2 c2 ( 2 π ) 0 0 Enlm − En0 l0 m0 − h̄ ωk
n0 l0 m0
(1179)
where θk is the angle subtended between k and the matrix elements of p. The
angular integration can be performed, yielding

(2)
 2 
e 2 h̄ c
Z ∞ X | < n0 l0 m0 | p̂ | nlm > |2
∆Enlm = dk k
m2 c2 3π 0 Enlm − En0 l0 m0 − h̄ ωk
n0 l0 m0
(1180)
In the completely non-relativistic limit, the integration over k can be shown to
be linearly divergent at the upper limit of integration.

Hans Bethe argued70 that, within the same dipole approximation, the cor-
rection to the kinetic energy of the electron in the state | nlm > is given by an
expression analogous to that of an electron in a continuum state n
 2 2 Z ∞ X | < n0 | p̂ | n > |2
(2) 2 e h̄
∆Tn = dω ω (1181)
3 π h̄ c mc 0 0
En − En0 − h̄ ω
n

Since momentum is conserved for continuum states (on average), only the state
where n = n0 contribute so the denominator simplifies71 . The expression for the
mass renormalization is divergent and is given by
 2 2 Z ∞ X | < n0 | p̂ | n > |2
2 e h̄
∆Tn(2) = − dω ω
3 π h̄ c mc 0 h̄ ω
n0
 2  Z ∞
4 e h̄ dω ω < n | p̂2 | n >
= − 2
(1182)
3 π h̄ c mc 0 ω 2m
where the completeness relation has been used. This expression is valid if n
labels either a continuum or a discrete state, since only the mass of the electron
70 H. A. Bethe, Phys. Rev. 72, 339 (1947).
71 Since we are now using the dipole approximation, the recoil of the free electron which was
taken into account in our previous analysis is now being ignored. [See the denominator of the
first line of eqn(1173).]

205
is being altered and the expectation value of p̂ is unaltered. Thus, Bethe argued,
the kinetic energy of an electron in a bound state which has the physical mass
m∗ should be approximated as

p̂2 p̂2
< nlm | | nlm > = < nlm | | nlm >
2 m∗  22m  Z ∞
dω ω < nlm | p̂2 | nlm >

4 e h̄
−
3 π h̄ c m c2 0 ω 2m
(1183)

Now the bare Hamiltonian for an electron bound to a nucleus is given by

p̂2
Ĥ0 = + V (r) (1184)
2m
and the unperturbed energy of the state | nlm > of a hypothetical electron
with mass m is calculated in the non-relativistic Schrödinger theory as

(0) p̂2
Enlm = < nlm | | nlm > + < nlm | V (r) | nlm > (1185)
2m
However, in order to obtain a sensible numerical value for the approximate
(0)
energy, Enlm has to be expressed in terms of the observed physical mass m∗ .
Therefore, the bare Hamiltonian has to be expressed in terms of the physical
mass and compensating radiative corrections to the mass

(0) p̂2
Enlm = < nlm | ∗
| nlm > + < nlm | V (r) | nlm >
 22m  Z ∞
dω ω < nlm | p̂2 | nlm >

4 e h̄
+ 2
3 π h̄ c mc 0 ω 2m
2
p̂
= < nlm | ∗
| nlm > + < nlm | V (r) | nlm >
 2m
2  Z ∞
dω ω X | < n0 l0 m0 | p̂ | nlm > |2

4 e h̄
+
3 π h̄ c m c2 0 ω 0 0 0
2m
nlm
(1186)

The completeness relation was used in obtaining the last line. The second term
in the above expression for the unperturbed energy is the correction due to
the mass renormalization72 which should be combined with the second-order
radiative correction. The total energy (to second-order) is given by
(0) (2)
Enlm = En,l,m + ∆En,l,m
72 Renormalization is an idea which Bethe attributed to H. A. Kramers. Kramers had

proposed that physical quantities should be expressed in terms of observable quantities, with
all mention of bare quantities removed. Kramers was advocating a classical treatment from
which Bethe created a non-relativistic quantum treatment.

206
 p̂2
= < nlm | | nlm > + < nlm | V (r) | nlm >
2 m∗
 2 2 Z ∞ X | < n0 l0 m0 | p̂ | nlm > |2
2 e h̄
+ dω ω
3 π h̄ c mc 0 h̄ ω
n0 l0 m0
 2 2 Z ∞ X | < n0 l0 m0 | p̂ | nlm > |2
2 e h̄
+ dω ω
3 π h̄ c mc 0 0 0 0
Enlm − En0 l0 m0 − h̄ ω
nlm
(1187)
The overall (second-order) shift from the Schrödinger estimate of the energy for
the state | nlm > (as calculated with the physical mass) is given by the sum
of the last two terms, which is expressed as
 2 2 Z ∞ X | < n0 l0 m0 | p̂ | nlm > |2 ( Enlm − En0 l0 m0 )
shift = 2 e h̄
∆Enlm dω ω
3 π h̄ c mc 0 0 0 0
( Enlm − En0 l0 m0 − h̄ ω ) h̄ ω
nlm
(1188)
The integration over ω is logarithmically divergent, and can be made to con-
verge by introducing an upper cut-off ω+ = c λ−1 . Therefore, the difference
of the linearly divergent self-energy of the bound electron and the linearly di-
vergent self-energy of the free electron is only logarithmically divergent. After
introducing the cut-off, one finds the result
| < n0 l0 m0 | p̂ | nlm > |2
 2 X
shift 2 e
∆Enlm = −
3 π h̄ c m2 c2
n0 l0 m0
 h̄cλ−1 + En0 l0 m0 − Enlm 
 
× ( Enlm − En l m ) ln 
0 0 0  
En0 l0 m0 − Enlm 
(1189)
The square of the mass m can safely be replaced by the square of the physical
mass m∗ in the expression for the energy shift, since we are only working to
2
first order in h̄e c . All other quantities have been expressed in terms of the
physical mass. If the rest energy of the electron is used as the upper cut-off
energy m c2 ∼ 0.5 × 106 eV, and assuming that the averaged logarithm of the
electron excitation energy corresponds to an energy of the order of 17.8 Ryd,
then the logarithm has a value of about 7.63 and is not sensitive to the precise
value of Enlm − En0 l0 m0 and, therefore, can be taken outside the summation
 2 h̄2 c2 
 2  
shift = − 2
∆Enlm
e
ln  2 4 
3 π h̄ c Z e
X | < n0 l0 m0 | p̂ | nlm > |2
× ( Enlm − En0 l0 m0 )
0 0 0
m2 c2
nlm
(1190)
73 e2
As later shown by Dyson , that divergences found in any order in h̄ c can
73 F. J. Dyson, Phys. Rev. 75, 1736 (1949).

207
be removed by consistently using the ideas of mass and charge renormaliza-
tion74 . Hence, a completely consistent relativistic theory does yield a finite
shift, without the need to invoke any cut-off75 . The weighted sum over the
matrix elements can be evaluated by expressing it in terms of an expectation
value involving commutators of Ĥ0 with p̂. That is
X
| < n0 l0 m0 | p̂ | nlm > |2 ( Enlm − En0 l0 m0 )
n0 l0 m0
X
= < nlm | p̂ | n0 l0 m0 > < n0 l0 m0 | [ p̂ , Ĥ0 ] | nlm >(1191)
n0 l0 m0

and using the completeness relation, one obtains

= < nlm | p̂ [ p̂ , Ĥ0 ] | nlm > (1192)

Likewise,
X
| < n0 l0 m0 | p̂ | nlm > |2 ( Enlm − En0 l0 m0 )
n0 l0 m0
X
= − < nlm | [ p̂ , Ĥ0 ] | n0 l0 m0 > < n0 l0 m0 | p̂ | nlm >
n0 l0 m0
(1193)

which results in

= − < nlm | [ p̂ , Ĥ0 ] p̂ | nlm > (1194)

Thus, on adding the above two equations and dividing by two, one finds
X
| < n0 l0 m0 | p̂ | nlm > |2 ( Enlm − En0 l0 m0 )
n0 l0 m0
1
= < nlm | [ p̂ , [ p̂ , Ĥ0 ] ] | nlm > (1195)
2
On substituting
p̂ = − i h̄ ∇ (1196)
and
p̂2
Ĥ0 = + V (r) (1197)
2m
74 This statement does not imply that a properly renormalized perturbation theory is con-

vergent. In fact, one may argue that if the coupling constant changed sign then systems
containing electrons would be unstable to BCS pairing. Since the radius of convergence of
any expansion is limited by the closest singularity, perturbation theory may only have a zero
radius of convergence. In this case, the theory may be expected to contain non-analytic terms
of the form exp[ − h̄ c/ e2 ].
75 F. J. Dyson, Phys. Rev. 173, 617 (1948).

208
into the expression for the matrix elements, one obtains
X
| < n0 l0 m0 | p̂ | nlm > |2 ( Enlm − En0 l0 m0 )
n0 l0 m0
2 Z
h̄
= − d3 r ψnlm (r) ∇2 V (r) ψnlm (r) (1198)
2
Substituting the expressions for the matrix elements into the expression for the
Lamb-shift yields
 2 2  
 2 h̄2 c2 

shift = 2 e h̄ 2
∆Enlm ln  Z 2 e4  < nlm | ∇ V (r) | nlm >
 
3 π h̄ c m2 c2
(1199)

Thus, the energy-shift only occurs for bound electrons as the expectation value of
the Laplacian of the potential will vanish for extended states. For a hydrogenic-
like atom
∇2 V (r) = 4 π Z e2 δ 3 (r) (1200)
so
4 Z e2 e2 h̄2  2 h̄2 c2 
    
∆E shift
nlm = 2
| ψnlm (0) | ln  2 4  (1201)

3 h̄ c m2 c2 Z e
Therefore, the Lamb shift only occurs for electrons with l = 0, since electronic
wave functions with l 6= 0 vanish at the origin. The atomic wave function at
the position of the nucleus is given by
 3
1 Z
| ψn00 (0) |2 = (1202)
π na
This yields Bethe’s estimate for the Lamb shift as
 2 3  4 4 
 2 h̄2 c2 
 
shift = 4 e Z e m
∆En00 ln   (1203)
3 π n3 h̄ c h̄2  Z 2 e4 

The above formulae leads to the estimate of 1040 MHz which is in good agree-
ment with the experimentally determined value76 . The exact relativistic calcu-
lation77 yields the result
 2 5
 m c2 
   
shift 4 4 e 2 31
∆En00 = Z mc ln 
  + (1204)
3 π n3 h̄ c 2 h̄ ωn,n0  120

where the mc2 in the logarithm comes from the Dirac theory without invoking
any cut-off. The most recent experimentally measured value78 is 1057.851 MHz
which is in good agreement with the theoretical value of 1057.857 MHz.

76 W. E. Lamb Jr. and R. E. Retherford, Phys. Rev. 72, 241 (1947).

77 N. M. Kroll and W. E. Lamb Jr., Phys. Rev. 75, 388 (1949).
78 G. C. Bhatt and H. Grotch, Ann. Phys. 187, 1 (1987).

209
11.3.5 Brehmstrahlung
Accelerating (or decelerating) charged particles radiate. We shall consider the
radiation emitted by a charged particle (such as an electron) that scatters from
a massive charged particle via the Coulomb interaction. It is assumed that the
mass M of the massive charged particle (in most cases, this is a nucleus) is
significantly greater than the electron mass, so that the recoil of the nucleus can
be neglected. The (instantaneous) Coulomb interaction between the electron
and the nucleus is given by
Z e2
V (r) = − (1205)
r
The Hamiltonian of the unperturbed electron is simply the kinetic energy. The
incident electron is assumed to have a momentum q and the scattered electron
has momentum q 0 and the cross-section for the scattering process will be calcu-
lated via low-order perturbation theory.

Rutherford Scattering

To second-order, the scattering cross-section is expressed as Rutherford scat-

tering which is elastic and, therefore, involves no emission of photons. The

q q'

q-q'

Figure 40: The Rutherford scattering process.

Rutherford scattering cross-section is found from the Fermi-Golden rule decay

rate
 
1 2π
= | < q 0 | V (r) | q > |2 δ( Eq − Eq0 ) (1206)
τ Rutherford h̄
The matrix elements of the Coulomb potential is evaluated as
4 π Z e2
< q 0 | V (r) | q > = − (1207)
V | q − q 0 |2
On integrating over the magnitude of the scattered electron’s momentum, one
obtains
Z ∞ 2
4 π Z e2
  
1 2π V 0 0 02
= dΩ dq q δ( Eq − Eq0 )
τdΩ0 Rutherford h̄ ( 2 π )3 0 V | q − q 0 |2

210
 2
4 π Z e2

2π V m
= q dΩ0
h̄ ( 2 π ) h̄2
3 V | q − q 0 |2
(1208)

The denominator in the potential has to be evaluated on the energy shell. On

introducing the scattering angle θ0 and using the elastic scattering condition

| q − q 0 |2 = 2 q 2 ( 1 − cos θ0 )
θ0
= 4 q 2 sin2 (1209)
2
one finds

q'
2q sinθ'/2

θ'
q

Figure 41: The geometry for Rutherford scattering. For elastic scattering, the
magnitude of the initial momentum q is equal to the magnitude of the final
momentum q 0 and the scattering angle is θ0 .

2
4 π Z e2
  
1 2π V m
= q θ0
dΩ0
τdΩ0 Rutherford h̄ ( 2 π ) h̄2
3
V 4 q 2 sin2 2
(1210)

On diving the scattering rate by the incident flux F of electrons

h̄ q
F = (1211)
mV
the elastic scattering cross-section is found to be given by
2
V m2 4 π Z e2
  
dσ 2π V
= θ0
dΩ0 Rutherford h̄ ( 2 π ) 3
h̄3 V 4 q 2 sin2 2
2
m Z e2

= 0 (1212)
2 h̄2 q 2 sin2 θ2

which is the Rutherford scattering cross-section for electrons. The scattering

cross-section diverges at θ0 = 0 and is always finite at θ0 = π no matter how
large q is. The scattering at θ0 = π is known as back-scattering, and is caused

211
 dσ/dΩ'
0 0.25 0.5 0.75 1
θ'/π

Figure 42: The scattering angle dependence of the differential scattering cross-
section.

by the extremely high potential experienced by electrons with very small im-
pact parameters. It was the large cross-section for back-scattering of charged
α-particles from atoms, found by H. Geiger and E. Marsden in 191379 , that
was instrumental in verifying Rutherford’s 1911 conjecture80 that atoms have
nuclei which are of very small spatial extent. The divergence in the scattering
cross-section at θ0 = 0 is due to the long-ranged nature of the Coulomb inter-
action, which causes electrons to undergo scattering (no matter how slight the
scattering is) at arbitrarily large distances from the nucleus.

Brehmstrahlung

Elastic scattering of electrons by the Coulomb potential is highly unlikely,

since from classical electrodynamics it is known that accelerated particles ra-
diate. Hence, it is expected that photons should be emitted in this process.
This phenomenon is known as Brehmstrahlung. We shall calculate the Brehm-
strahlung scattering cross-section81 using low-order perturbation theory. The
electron is scattered between the free electron eigenstates due to a perturbation
which is a linear superposition of the Coulomb interaction with the nucleus and
the paramagnetic interaction.

The lowest-order probability amplitude describing Brehmstrahlung is a lin-

ear superposition of two processes. These are:

(a) Scattering of an electron from the nucleus followed by the emission of a pho-
ton. The initial state of the electron is assumed to have momentum q and the
final state of the electron is given by q 0 while the emitted photon has momentum
k. Therefore, from conservation of momentum, the momentum of the electron
79 H. Geiger and E. Marsden, Phil. Mag. 25, 1798 (1913).
80 E. Rutherford, Phil. Mag. 21, 669 (1911).
81 H. A. Bethe and W. Heitler, Proc. Roy. Soc. A 146, 82 (1934).

212
 (k,α)
(k,α)

q'+k q-k
q q' q q'

Figure 43: The two lowest-order processes contributing to Bremstrahlung.

in the intermediate state is given by q 0 + k.

(b) Emission of a photon followed by scattering from the nucleus. Conservation

of momentum indicates that the intermediate state has momentum given by
q − k.

The matrix elements for these second-order processes are given by

s
4 π Z e2 2 π h̄ c2
 
e h̄
Ma = 0 2
ˆα (k) . ( q 0 + k )
V |q − q − k| mc V ωk
 
1
× (1213)
Eq − Eq0 +k + i η

and
s
4 π Z e2 2 π h̄ c2
 
e h̄
Mb = 0 2
ˆα (k) . ( q − k )
V |q − q − k| mc V ωk
 
1
× (1214)
Eq − Eq−k − h̄ ωk + i η

It should be noted that the numerators of the matrix elements simplify because
the photons have transverse polarizations

α (k) . k = 0 (1215)

From the energy conserving delta function in the expression for the decay rate,
one finds
Eq = Eq0 + h̄ ωk (1216)
hence the first energy-denominator can be expressed in a similar form to the
second
Eq − Eq0 +k = Eq0 − Eq0 +k + h̄ ωk (1217)
For small k, the energy-denominators can be expanded, yielding

h̄ 0 h̄2 k 2
Eq0 − Eq0 +k + h̄ ωk = h̄ ωk − q .k − (1218)
m 2m

213
and
h̄ h̄2 k 2
Eq − Eq−k − h̄ ωk = − h̄ ωk + q.k − (1219)
m 2m
Since the energy of the photon cannot exceed the energy of the initial electron,
one must have q > k, so the third term is smaller than the second term. Due to
the large magnitude of c compared with the electron velocities h̄mq , the second
and third terms can be neglected. Therefore, the photon-energy dominates both
the energy-denominators. On substituting the above expressions in the sum of
the matrix elements, one finds
s
4 π Z e2 2 π h̄ c2
 
e
M a + Mb = 0 2
V |q − q − k| mc V ωk
0
ˆα (k) . q h̄ ˆα (k) . q h̄
 
× +
Eq0 − Eq0 +k + h̄ ωk + i η Eq − Eq−k − h̄ ωk + i η
s
4 π Z e2 2 π h̄ c2
   
e
≈
V | q − q 0 − k |2 mc V ωk
ˆα (k) . ( q 0 − q )
 
× (1220)
ωk

Using this approximation for the matrix elements, the transition rate is given
by
2  2 
4 π Z e2 2 π h̄ c2
 
1 2π X X e
=
τ h̄ 0
V | q − q 0 − k |2 mc V ωk
q k,α

 ˆα (k) . ( q 0 − q ) 2
 
×
  δ( Eq − Eq0 − h̄ ωk ) (1221)
ωk 

The terms proportional to k in the Coulomb scattering terms can be neglected,

for low k values. The inelastic scattering cross-section for Brehmstrahlung is
found by replacing the sums over q 0 and k 0 by integrals, and dividing by the
incident flux of electrons. This procedure results in the expression
2 X Z
d2 σ q0 2 m Z e2 e2
    
dΩk
=
dΩ0 dωk Brehmse q h̄2 | q − q 0 |2 α
4 π 2 ωk h̄ c
 h̄ ˆα (k) . ( q 0 − q ) 2
 
×
  (1222)
mc 

If the angular distributions of the emitted photon (dΩk ) and the scattered elec-
tron (dΩ0 ) are both measured, the scattering cross-section can be represented
as
d3 σ q0
   
dσ
=
dΩ0 dΩk dωk Brehmse q dΩ0 Rutherford

214
 e2  h̄ ˆα (k) . ( q 0 − q ) 2
  X  
1
×  
4 π 2 ωk h̄ c α
 m c 

(1223)

where the second factor is the probability of emitting a photon with energy h̄ ωk
into solid angle dΩk . On summing over the polarization α and integrating over
the directions of the emitted photon, one obtains
2
d2 σ q0 2 m Z e2
  
=
dΩ0 dωk Brehmse h̄2 | q − q 0 |2
q
 h̄ ( q 0 − q ) 2
 2 
2 e
×   (1224)
3 π ωk h̄ c  mc 

Hence, the scattering rate which includes the emission of a photon of energy
h̄ ωk is given by the product of the Rutherford scattering rate with a factor
0 2
q0 e2 2 q h̄ sin θ2
 
2
(1225)
q 3 π ωk h̄ c mc

This particular factorization of the cross-section involving the simultaneous

emission of a soft photon is common to many processes involving the emis-
sion of low-energy bosons. The soft-photon theorem82 shows that properties of
the emitted low-energy photon is insensitive to anything except the global prop-
erties (such as the total charge or total magnetic moment) of the scattered par-
ticle. The cross-section involving the emission of a low-energy photon diverges
as ωk → 0, due to the factor of ωk−1 in eqn(1224). This type of divergence
is an infrared divergence. What this implies is that, in Brehmstrahlung, arbi-
trary large numbers of low-energy photons are emitted. Furthermore, similar
singularities are also found in the ω = 0 limit when elastic scattering correc-
tions to the Rutherford scattering process are considered83 . In any experiment
with finite energy resolution, elastic scattering and very low-energy quasi-elastic
scattering processes cannot be distinguished, so it is might be expected that the
elastic scattering and quasi-elastic scattering divergences should be combined.

The divergences found in the problem of Brehmstrahlung were first con-

sidered by Bloch and Nordsieck84 who showed that the infra-red divergences
cancel. That is, the infra-red divergence does not exist85 . The cancelation
was achieved adding virtual emission processes for Rutherford scattering to the
Brehmstrahlung cross-section for the emission of photons of energy less than
ω0 , since these processes cannot be distinguished for sufficiently small photon
82 F. F. Low, Phys. Rev. 96, 1428 (1958).
83 R. H. Dalitz, Proc. Roy. Soc. A 206, 509 (1950).
84 F. Bloch and A. Nordsieck, Phys. Rev. 52, 54 (1937).
85 Since there are an infinite number of low-energy photons present in Brehmstrahlung, then

it is expected that the classical limit of quantum theory applies so that classical electromag-
netic theory should produce exact results.

215
frequencies ω0 . That is, on introducing an infra-red cut-off λ− , one finds that
the total inelastic scattering in which a photon with frequency less than ω0 is
emitted is given by
      2   
dσ dσ 1 e 2 ω 0 λ−
= A ln + ... + ...
dΩ Brehmse dΩ Rutherford 2 π h̄ c c
(1226)
where the factor A depends on the initial and final momentum of the electron.
This result is logarithmically divergent as λ− → 0. On the other hand, to the
same order, the elastic scattering cross-section is found as
      2   
dσ dσ 1 e h̄
= 1+ A ln + ... + ...
dΩ Elastic dΩ Rutherford 2 π h̄ c λ− m c
(1227)
Hence, on combining the results, one finds that the quasi-elastic scattering cross-
section is given by
      2   
dσ dσ 1 e 2 h̄ ω0
= 1+ A ln + ... + ...
dΩ Quasi-Elastic dΩ Rutherford 2 π h̄ c m c2
(1228)
so the cut-off λ− cancels and the scattering cross-section does not diverge log-
arithmically. With this reasoning, Bloch and Nordsieck found that the appro-
2 2
h̄ ω0
priate expansion parameter is not h̄e c but instead is given by h̄e c ln m c2 . The
higher-order perturbations may also describe processes involving larger numbers
of emitted soft photons and results in a multiplicative exponential factor to the
quasi-elastic scattering rate
      2 
dσ dσ 1 e 2 h̄ ω0
≈ exp B ln + . . .
dΩ Quasi-Elastic dΩ Rutherford 2 π h̄ c m c2
(1229)
Therefore, the scattering rate from soft photons vanishes in the limit ω0 → 0.
This occurs because perturbation theory causes the normalization of the starting
approximate wave function to change, and hence the probabilities of the vari-
ous processes are changed by including higher-order processes. In other words,
since the probability of emitting an arbitrarily large number of soft-photons is
finite, the probability of emitting either zero or any fixed number of soft photons
must be zero. Bloch and Nordsieck’s calculation was restricted to the case of
emission of sufficiently low-energy photons. Pauli and Fierz86 also considered
Brehmstrahlung in a non-relativistic approximation. Pauli and Fierz showed
that the infra-red divergences, discussed above, cancel. Pauli and Fierz went on
to examine the remaining ultra-violet divergences, and showed that portions of
the ultra-violet infinities that were found in the calculations of the scattering
processes could be associated with mass renormalization. Using a relativistic
theory Ito, Koba and Tomonaga87 showed that the remaining infinities could
86 W. Pauli and M. Fierz, Nuovo Cimento, 15, 167 (1938).
87 D. Ito, Z. Koba and S-I. Tomonaga, Prog. Theor. Phys. (Kyoto), 3, 276 (1948).

216
be absorbed into a renormalization of the electron charge. Similar conclusions
were arrived at by Lewis88 and by Epstein89 . Dyson90 showed that all infinities
that appear in Quantum Electrodynamics could be cured by renormalization to
arbitrarily high-orders in perturbation theory.

12 The Dirac Equation

In 1928, Dirac searched for a relativistically invariant form of the one-particle
Schrödinger equation for electrons
∂
i h̄ ψ = Ĥ ψ (1230)
∂t
Since this equation is only first-order in time, then the solution is uniquely
specified by the initial condition for ψ. It is essential to only require an evolution
equation which is first-order in time. Dirac91 searched for a set of coupled first-
order (in time) equations for a multi-component wave function ψ

ψ (0)
 
 ψ (1) 
 
 
ψ =  (1231)
 
.. 

 . 

 
ψ (N −1)

The wave function was assumed to satisfy an equation of the form

 
h̄ ∂
i − α . p̂ ψ = β m c ψ
c ∂t
 
h̄ ∂
i + i h̄ α . ∇ ψ = β m c ψ (1232)
c ∂t

The equations have to be of this form since, if the equation is a first-order partial
differential equation in time then it must also only involve the first-order partial
derivatives with respect to the spatial components for the resulting equation to
be relativistically covariant. The wave function ψ is a N -component (column)
wave function and the three as yet unknown components of α and β are three
N × N matrices. Since the Hamiltonian is the generator of time translations,
∂
then Ĥ should be equivalent to ih̄ ∂t . Hence, as the Hamiltonian operator Ĥ
must be Hermitean, then the operators α and β must be Hermitean matrices.
88 H. W. Lewis, Phys. Rev. 73, 173 (1948).
89 Saul T. Epstein, Phys. Rev. 73, 177 (1948).
90 F. J. Dyson, Phys. Rev. 75, 486 (1949).
91 P. A. M. Dirac, Proc. Roy. Soc. A 117, 610 (1928).

P. A. M. Dirac, Proc. Roy. Soc. A 118, 351 (1928).

217
This set of equations is required to yield the dispersion relation for a relativistic
particle
 2
E
− p2 = m2 c2 (1233)
c
which, following the ordinary rules of quantization, leads to the Klein-Gordon
equation
h̄2 ∂ 2
 
2 2
− 2 + h̄ ∇ ψ = m2 c2 ψ (1234)
c ∂t2
(which is a second-order partial differential equation in time). The requirement
that the Dirac equation is compatible with the Klein-Gordon equation imposes
conditions on the form of the matrices. On writing the Dirac equation as
 
h̄ ∂ψ
i = β m c − i h̄ α . ∇ ψ (1235)
c ∂t

and iterating, one has

 2 2  2
h̄ ∂ ψ
− = β m c − i h̄ α . ∇ ψ
c ∂t2

= β 2 m2 c2 − i h̄ m c ( β α + α β ) . ∇

− h̄2 ( α . ∇ )2 ψ (1236)

When expressed in terms of individual matrices α(j) , the above equation be-
comes
 2 2 
h̄ ∂ ψ 2 2 2
X
− = β m c − i h̄ m c ( β α(j) + α(j) β ) ∇j
c ∂t2 j

h̄2 X

− ( α(i) α(j) + α(j) α(i) ) ∇i ∇j ψ (1237)
2 i,j

since the derivatives commute. If the above equation is to be equivalent to the

Klein-Gordon equation, then the coefficients of the various derivatives must be
identical for both equations. Therefore, it is required that the constant terms
are equal
β 2 = Iˆ (1238)
It is also required that the first-order derivative terms vanish and that the
second-order derivative terms should be equal, hence the matrices must satisfy
the anti-commutation relations

α(i) β + β α(i) = 0
α (i)
α(j) + α(j) α(i) = 2 δ i,j Iˆ (1239)

218
On imposing the above conditions, Dirac’s form of the relativistic Schrödinger
equation is compatible with the Klein-Gordon equation.

From eqn(1238), one concludes that if the Hermitean matrices are brought to
diagonal form then the diagonal elements are given by ± 1. The possible dimen-
sions N of the matrix can be determined by considering the anti-commutation
relations. On taking the determinant of eqn(1239), one finds

det α(i) det β = ( −1 )N det β det α(i)

det α(i) det α(j) = ( −1 )N det α(j) det α(i) (1240)

Hence, on cancelling the common factors of determinants, one finds

( − 1 )N = 1 (1241)

so N must be even. Furthermore, the matrices must be traceless. This can be

seen by considering
α(i) α(j) = − α(j) α(i) (1242)
which on multiplying by α(i) , yields the relation

α(j) = − α(i) α(j) α(i)

α(j) = − ( α(i) )−1 α(j) α(i) (1243)

since α(i) is its own inverse. Apart from the negative sign, the form of the
left-hand side is of the form of an equivalence transformation. By using cyclic
invariance, it can be shown that the trace of a matrix is invariant under equiv-
alence transformations. Therefore, one has

Trace α(i) = − Trace α(i) (1244)

or
Trace α(i) = 0 (1245)
which proves that the matrices are traceless.

Since the Dirac matrices satisfy

β2 = Iˆ
(α(i) )2 = Iˆ (1246)

then their eigenvalues must all be ±1, as can be seen by operating on the
eigenvalue equation
β φβ = λβ φβ (1247)
with β. This process yields

β 2 φβ = λβ β φβ
= λ2β φβ (1248)

219
 ˆ requires that the eigenvalues must satisfy the equation
which with β 2 = I,

λ2β = 1 (1249)

This and the condition that the matrices are traceless implies that the set of
eigenvalues of each matrix are composed of equal numbers of +1 and −1, and
it also confirms the conclusion that dimension N of the matrices must be even.
The smallest value of the dimension for which there is a representation of the
matrices is N = 4. The smallest even value of N , N = 2 cannot be used since
one can only construct three linearly independent anti-commuting 2 × 2 ma-
trices92 . These three matrices are the Pauli spin matrices σ (j) . Hence, Dirac
constructed the relativistic theory with N = 4.

It is useful to find a representation in which the mass term is diagonal, since

this represents the largest energy which occurs in the non-relativistic limit.
When diagonalized, the β matrix has two eigenvalues of +1 and two eigenvalues
of −1 and so β can be expressed in 2 × 2 block-diagonal form. We shall express
the 4 × 4 matrices in the form of 2 × 2 block matrices. In this case, one can
represent the matrix in the block-diagonal form
 
I 0
β = (1250)
0 −I

If the three matrices α(i) are to anti-commute with β and be Hermitean, they
must have the off-diagonal form

A(i)
 
0
α(i) = (1251)
A(i)† 0

where A(i) is an arbitrary 2 × 2 matrix. We shall choose all three A(i) matrices
to be Hermitean. Since the three α(i) matrices must anti-commute with each
other, the A(i) must also anti-commute with each other. Since the three Pauli
matrices are mutually anti-commuting, one can set

σ (i)
 
0
α(i) = (1252)
σ (i) 0

where the σ (i) and I are, respectively, the 2 × 2 Pauli matrices and the 2 × 2
unit matrix. The Pauli matrices are given by
 
(1) 0 1
σ = (1253)
1 0
92 In d + 1 space-time dimensions, one can form 2d+1 matrices from products of the set of

d+1 linearly independent (anti-commuting) Dirac-matrices. We shall assume that the product
matrices are linearly independent. Since the number of linearly independent N × N matrices
is N 2 , the minimum dimension N which will yield a representation of the Dirac-matrices is
d+1
N =2 2 .

220
  
0 −i
σ (2) = (1254)
i 0
and  
(3) 1 0
σ = (1255)
0 −1
The matrix α(0) is defined as the 4 × 4 identity matrix
 
(0) I 0
α = (1256)
0 I

This set of matrices form a representation of the Dirac matrices. This can
be seen by directly showing that they satisfy the appropriate relations. Many
different representations of the Dirac matrices can be found, but they are all
related by equivalence transformations and the physical results are independent
of which choice is made.

Exercise:

By direct matrix multiplication, show that the above matrices satisfy the
relations
( α(j) )2 = β 2 = Iˆ (1257)
and the anti-commutation relations

α(i) β + β α(i) = 0
α (i)
α(j) + α(j) α(i) = 2 δ i,j Iˆ (1258)

and so form a representation of the Dirac matrices.

12.1 Conservation of Probability

One can find a conservation law for Dirac’s equation
 
∂ψ 2
i h̄ = − i h̄ c α . ∇ + β m c ψ (1259)
∂t

On pre-multiplying the Dirac equation by ψ † , which is the Hermitean conjugate

of the spinor wave function and is defined as the row vector formed by the
complex conjugate of the components
 
† (0)∗ (1)∗ (2)∗ (3)∗
ψ = ψ , ψ , ψ , ψ , (1260)

one obtains
 
∂ψ
† † † 2
i h̄ ψ = − i h̄ c ψ α . ∇ ψ + ψ β ψ m c (1261)
∂t

221
The Hermitean conjugate of the Dirac equation is given by
∂ψ †
 
† † † † 2
− i h̄ = + i h̄ c ∇ . ψ α + ψ β m c (1262)
∂t
Since α and β are Hermitean matrices, the Hermitean conjugate equation sim-
plifies to
∂ψ †
 
− i h̄ = + i h̄ c ∇ . ψ † α + ψ † β m c2 (1263)
∂t
Post-multiplying the Hermitean conjugate equation by the column-vector ψ,
yields
∂ψ †
 
− i h̄ ψ = + i h̄ c ∇ . ψ † α ψ + ψ † β ψ m c2 (1264)
∂t
On subtracting eqn(1264) from the eqn(1261) and combining terms, one obtains
∂
i h̄ ( ψ † ψ ) = − i h̄ c ∇ . ( ψ † α ψ ) (1265)
∂t
The above equation has the form of a continuity equation
∂ρ
+ ∇.j = 0 (1266)
∂t
in which the probability density is given by
ρ = ψ† ψ (1267)
Using the rules of matrix multiplication the probability density is a real scalar
quantity, which is given by the sum of squares
ρ = | ψ (0) |2 + | ψ (1) |2 + | ψ (2) |2 + | ψ (3) |2 (1268)
and so it is positive definite. Hence, unlike the Klein-Gordon equation, the
Dirac equation does not lead to negative probability densities. The probability
current density j is given by

j = c ψ† α ψ (1269)
In this case, the total probability
Z Z
3 †
Q = d xψ ψ = d3 x ρ (1270)

is conserved, since
Z
dQ ∂ρ
= d3 x
dt ∂t
Z
= − d3 x ∇ . j
Z
= − d2 S . j (1271)

222
where Gauss’s theorem has been used to represent the volume integral as surface
integral. For a sufficiently large volume, the current at the boundary vanishes,
hence the total probability is conserved
dQ
= 0 (1272)
dt

12.2 Covariant Form of the Dirac Equation

In the absence of an electromagnetic field, the Dirac equation can be expressed
in either of the two forms

αµ p̂µ ψ = βmcψ
i h̄ αµ ∂µ ψ = βmcψ (1273)

where it has been recalled that

α(0) = Iˆ (1274)

and the covariant momentum operator is given by

 
∂
p̂µ = i h̄ (1275)
∂xµ

Or equivalently, after multiplying the Dirac equation by β and then introducing

the four γ matrices via
γ µ = β αµ (1276)
one finds that the Dirac equation appears in the alternate forms

γ µ p̂µ ψ = mcψ
i h̄ γ µ ∂µ ψ = mcψ (1277)

The four gamma matrices satisfy the anti-commutation relations

γ µ γ ν + γ ν γ µ = 2 g µ,ν Iˆ (1278)

where Iˆ is the 4 × 4 identity matrix, and g µ,ν is the Minkowski metric. The
gamma matrices labelled by the spatial indices are Unitary and anti-Hermitean,
as shall be proved below.

It is easy to show that the matrix with the temporal index (0) is unitary
and Hermitean

( γ (0) )−1 = γ (0)

( γ (0) )† = γ (0) (1279)

since β is its own inverse and β is Hermitean.

223
 The gamma matrices with spatial indices are anti-Hermitean as

( γ (i) )† = ( β α(i) )†
= ( ( α(i) )† β † )
= ( α(i) β )
= ( − β α(i) )
= − γ (i) (1280)

since α(i) and β are Hermitean and, in the fourth line the operators have been
anti-commuted. Now, the gamma matrices with spatial indices can be shown
to be unitary since

γ (i) γ (i) = β α(i) β α(i)

= − β β α(i) α(i)
= − Iˆ (1281)

where, in obtaining the second line, the anti-commutation properties of α(i) and
β have been used, and the property

( α(i) )2 = β 2 = Iˆ (1282)

was used to obtain the last line. Since it has already been demonstrated that
the spatial matrices are anti-Hermitean

( γ (i) )† = − γ (i) (1283)

then it follows that γ (i) is unitary as

( γ (i) )† γ (i) = Iˆ (1284)

which completes the proof.

The continuity equation can also be expressed in a covariant form. The

†
covariant Dirac adjoint of ψ is defined as ψ where
†
ψ = ψ † γ (0) (1285)

Hence, since
( γ (0) )2 = Iˆ (1286)
†
the Hermitean conjugate wave function ψ can be expressed in terms of the
†
adjoint spinor ψ via
†
ψ † = ψ γ (0) (1287)
The continuity equation has the Lorentz covariant form
∂j µ
= 0 (1288)
∂xµ

224
where the four-vector conserved probability current j µ is given by

j µ = c ψ † αµ ψ (1289)

By using the definition of the Dirac adjoint, the current density can be re-
expressed as the four quantities
†
j (0) = c ψ γ (0) ψ
†
j (i) = c ψ γ (i) ψ (1290)

that, respectively, represent c times the probability density and the j (i) are the
contravariant components of the probability current density.

12.3 The Field Free Solution

In the absence of fields, the Dirac equation can be solved exactly by assuming
a solution in the form of plane-waves. This is because the momentum operator
p̂ commutes with the Hamiltonian Ĥ since in the absence of fields there is no
explicit dependence on position. The solution can be expressed as a momentum
eigenstate in the form
 (0) 
u
 u(1) 
 
µ
ψ =  (2)  exp − i kµ x
  (1291)
u
(3)
u

where the functions uµ (k) are to be determined. On substituting this form in

the Dirac equation, it becomes an algebraic equation of the form
 
(0) ˆ mc
k I − k.α − ( )β ψ = 0 (1292)
h̄

where k is a three-vector with components given by the contra-variant spa-

tial components of k µ . In order to write this equation in two by two block-
diagonal form, the four-component spinor ψ can be written in terms of two
two-components spinors  A 
φ
ψ = (1293)
φB
where the two two-component spinors are given by
 (0) 
A u
φ =
u(1)
 (2) 
B u
φ = (1294)
u(3)

225
Hence, the Dirac equation can be expressed as the block-diagonal matrix equa-
tion
   
(0) m c
 − k + h̄ I k.σ  A 
 φ
    B = 0 (1295)
φ
− k (0) − mh̄ c I
 
k.σ

where the three-vector scalar product involves the contra-variant components

of the momentum k (i) with the Pauli spin matrices σ (i) . The above equation
is an eigenvalue equation for k (0) . The eigenvalues are given by the solution of
the secular equation
 
 − k (0) + mh̄ c I k.σ
 

 = 0 (1296)
   

(0) m c
k.σ − k − h̄ I
 

which can be written as

  2   2
(0)2 mc
k − = σ.k (1297)
h̄

Using Pauli’s identity

      
σ.A σ.B = A.B I + iσ. A ∧ B (1298)

one finds the energy eigenvalues are given by the doubly-degenerate dispersion
relations s 2
mc
k (0)
= ± + k2 (1299)
h̄
Thus, the field free relativistic electron can have positive and negative-energy
eigenvalues given by q
E = ± m2 c4 + p2 c2 (1300)

Since the solutions are degenerate, solutions can be found that are simultaneous
eigenvalues of the Hamiltonian Ĥ given by

m c2 I
 
− i h̄ c σ . ∇
Ĥ = (1301)
− i h̄ c σ . ∇ − m c2 I

and another operator that commutes with Ĥ. It is convenient to choose the
second operator to be the helicity operator.

The helicity operator Σ̂ corresponds to the projection of the electron’s spin

along the direction of momentum. The (un-normalized) helicity operator is

226
 Σk = +1 Σk = −1

k k

Figure 44: A cartoon depicting the two helicity states of a spin one-half particle.

given by
 
σ.∇ 0
Σ̂ = − i h̄ (1302)
0 σ.∇
This is the appropriate relativistic generalization of spin valid only for free
particles93 , as the helicity is a conserved quantity since
[ Ĥ , Σ̂ ] = 0 (1303)
In the absence of electromagnetic fields, the Hamiltonian is evaluated as
m c2 I
 
h̄ c σ . k
Ĥ(k) = (1304)
h̄ c σ . k − m c2 I
Likewise, for the source free case, the properly normalized Helicity operator is
found as
 
σ . k̂ 0
Λ(k) = (1305)
0 σ . k̂
which has eigenvalues of ±1.

The axis of quantization of σ will be chosen to be along the direction of

propagation k̂. In this case, the helicity operator becomes
 (3) 
σ 0
Λ(k) = (1306)
0 σ (3)
and the eigenstates of helicity with eigenvalue +1 are composed of a linear
superposition of the spin-up eigenstates. We shall represent the two-component
spinors φA
+ via

φA
+ = u(0) χ+
 
(0) 1
= u (1307)
0
93 Helicity is not conserved for spherically symmetric potentials. However, if only a time-

independent vector potential is present, the generalized quantity

q
Σ̂ = σ . ( p̂ − A)
c
is conserved. This conservation law implies that the spin will always retain its alignment with
the velocity.

227
and φB
+ as

φB
+ = u(2) χ+
 
1
= u(2) (1308)
0

Therefore, one has

u(0) χ+
   
µ
ψ+ (x) = exp − i kµ x (1309)
u(2) χ+

Likewise, for the negative helicity states, φA

− can be represented via

φA
− = u(1) χ−
 
0
= u(1) (1310)
1

and φB
+ as

φB
− = u(3) χ−
 
0
= u(3) (1311)
1

Thus, the eigenstates with helicity −1 are the spin-down eigenstates

 (1)   
u χ− µ
ψ− (x) = exp − i kµ x (1312)
u(3) χ−

Clearly, states with different helicities are orthogonal since

χ†Λ0 χΛ = δΛ,Λ0 (1313)

which is as it should be since they are eigenstates of a Hermitean operator.

On substituting the helicity eigenstates ψΛ into the Dirac equation for the
free spin one-half particle
∂
i h̄ ψΛ = Ĥ ψΛ (1314)
∂t
one finds
φA m c2 σ (3) c h̄ k (3) φA
     
E Λ = Λ (1315)
φB
Λ σ c h̄ k (3)
(3)
− m c2 φB
Λ

Therefore, the complex amplitudes φA B

Λ and φΛ are found to be related by

σ (3) c h̄ k (3) A
φB
Λ = φ (1316)
E + m c2 Λ

228
This equation shows that the components φBΛ are small for the positive-energy
solutions, whereas the complementary expression

σ (3) c h̄ k (3) B
φA
Λ = − φ (1317)
m c2 − E Λ
shows that φA Λ is small for the negative-energy solutions. Hence, the two
positive-energy and two negative-energy (un-normalized) solutions of the Dirac
equation can be written as

χ+
 
 
ψ+ (x) = Ne   exp − i kµ xµ (1318)
c h̄ k(3)
E + m c2 χ+

for helicity +1 and

χ−
 
 
µ
ψ− (x) = Ne   exp − i kµ x (1319)
c h̄ k(3)
− E + m c2 χ−

for helicity -1. In this expression Ne is a normalization factor.

The normalization condition is

Z
d3 r ψ † ψ = 1 (1320)

which determines the magnitude of the normalization constant through

c2 h̄2 k 2
 
2
1 = V Ne 1 +
( E + m c2 )2
E + 2 E m c2 + m2 c4 + c2 h̄2 k 2
 2 
2
= V Ne
( E + m c2 )2
2 E + 2 E m c2
2
 
2
= V Ne
( E + m c2 )2
 
2 2E
= V Ne
E + m c2
Hence, the normalization constant can be set as
r
E + m c2
Ne = (1321)
2EV
for positive E.

For states with negative energies,

p
E = − m2 c4 + c2 h̄2 k 2 (1322)

229
the lower components are the large components. In this case, it is more conve-
nient to express the negative-energy solutions as

− m cc2h̄ −k E χ+
 
 
ψ+ (x) = Np   exp − i kµ xµ (1323)
χ+

for helicity +1 and

c h̄ k
 
m c2 − E χ−  
µ
ψ− (x) = Np   exp − i kµ x (1324)
χ−

for helicity -1. Furthermore, in this expression the normalization constant has
the form r
m c2 − E
Np = (1325)
−2EV
Hence, the positive and negative-energy solutions are symmetric under the in-
terchange E → − E, if Λ → − Λ and the upper and lower two-component
spinors (φA , φB ) are interchanged.

General Helicity Eigenstates

The Helicity operator for a particle with a momentum h̄ k is given by the

Hermitean operator

k (3) k (1) − i k (2)

 
1 
Λ(k) = 
k (1) (2) (3)
k + ik −k
 
cos θk sin θk exp[−iϕk ]
=   (1326)
sin θk exp[+iϕk ] − cos θk

which since
Λ(k) Λ(k) = I (1327)
94
has eigenvalues Λ of ±1. The helicity eigenstates are given by the two-
component spinors χΛ± . The positive helicity state is given by
 (1)
− i k (2)

k
1
χΛ+ = p  
2 k ( k − k (3) ) k − k (3)
cos θ2k exp[−i ϕ2k ]
 
ϕk 
= exp[ − i ]  (1328)
2 θk ϕk
sin 2 exp[+i 2 ]
94 C. G. Darwin, Proc. Roy. Soc. A 118, 654 (1928).

C. G. Darwin, Proc. Roy. Soc. A 120, 631 (1928).

230
in which (k, θk , ϕk ) are the polar coordinates of k. The negative helicity eigen-
state is given by the spinor χΛ−

− k + k (3)
 
1
χΛ− = p  
2 k ( k − k (3) ) k (1)
+ ik (2)

− sin 2 exp[−i ϕ2k ]

θk
 
ϕk 
= exp[ + i ]  (1329)
2 θk ϕk
cos 2 exp[+i 2 ]

Therefore, the general helicity eigenstate plane-wave solutions of the Dirac equa-
tion can be written in terms of two two-component spinors as
 
χΛ±  
µ
ψΛ± (x) = Ne   exp − i kµ x (1330)
c h̄ k Λ±
E + m c 2 χΛ±

In this expression Ne is a normalization factor

r
E + m c2
Ne = (1331)
2EV
These plane-wave solutions are useful in considerations of scattering processes.

12.4 Coupling to Fields

The Dirac equation describes relativistic spin one-half fermions, and their anti-
particles. It describes all massive leptons such as the electron, muon and tao
particle, and can be generalized to describe their interaction with the electro-
magnetic field, or its generalization the electro-weak field. In the limit m → 0,
the Dirac equation reduces to the Weyl equation95 which describes neutrinos.
The Dirac equation also describes massive quarks and the interaction can be
generalized to quantum chromodynamics.

In the absence of interactions, the Dirac equation can be expressed in either

of the two forms

αµ p̂µ ψ = βmcψ
i h̄ αµ ∂µ ψ = βmcψ (1332)

The interaction with electromagnetic field is introduced as follows. Using the

minimal coupling approximation, where
q
p̂µ → p̂µ 0 = p̂µ − Aµ (1333)
c
95 H. Weyl, Z. Physik, 56, 330 (1929).

231
and q is the charge of the particle, the Dirac equation in the presence of an
electromagnetic field becomes
 
q
αµ p̂µ − Aµ ψ = β m c ψ
c
 
q
i h̄ αµ ∂µ + i Aµ ψ = β m c ψ (1334)
h̄ c
This process has resulted in the inclusion of the interaction with the electro-
magnetic field in a gauge invariant, Lorentz covariant manner. The appearance
of the gauge field together with the derivative results in local gauge invariance.
Sometimes it is convenient to define a covariant derivative as the gauge-invariant
combination
q
Dµ ≡ ∂µ + i Aµ (1335)
h̄ c
The concept of the covariant derivative also appears in the context of other
gauge field theories. Using this definition we can express the Dirac equation in
the presence of an electromagnetic field in the compact covariant form
i h̄ γ µ Dµ ψ = m c ψ (1336)
The presence of an electromagnetic field does not alter the form of the conserved
four-vector current
†
jµ = c ψ γµ ψ (1337)
which is explicitly gauge invariant.

12.4.1 Mott Scattering

We shall consider the scattering of positive-energy electrons from a nucleus of
charge Z. The initial electron beam has momentum h̄ k which is scattered by
the target nucleus. The detector is positioned so that it detects all the scattered
electrons with momentum h̄ k 0 . The initial and final states of the positive-energy
electron can be represented by the Dirac spinors of the form ψσ
 
χσ  
µ
ψk,σ (x) = Nk   exp − i kµ x (1338)
c h̄ k . σ
Ek + m c 2 χσ

where the normalization constant is chosen as

s
Ek + m c2
Nk = (1339)
2 Ek V

The interaction Hamiltonian with the electrostatic field of the nucleus is given
by the diagonal matrix
Z e2
 
I 0
ĤInt = − (1340)
r 0 I

232
The flux of incident electrons is defined by
|v|
F = (1341)
V
where  
∂E
v = (1342)
∂p
Therefore, the electron flux is given by
h̄ k c2
 
F = (1343)
V Ek
The elastic scattering cross-section in which the final state polarization is un-
measured is given by
Ek V 2 X ∞
    Z
dσ 1
= dk 0 k 02 | < k 0 σ 0 | ĤInt | k, σ > |2 δ( Ek − Ek0 )
dΩ0 ( 2 π )2 h̄2 k c2 0 0
σ
(1344)
where the delta function ensures conservation of energy. Since the polarization
of the final state electron is unmeasured, the spin σ 0 is summed over. The
integration over k 0 can be performed, yielding
   2 X
dσ EV
= | < k 0 σ 0 | ĤInt | k, σ > |2 (1345)
dΩ0 2 π h̄2 c2 σ 0

where k and k 0 are restricted to be on the energy shell (E = Ek = Ek0 ). The

matrix elements can then be evaluated as
4 π Z e2 E + m c2
  
< k 0 , σ 0 | ĤInt | k, σ > = −
V | k − k 0 |2 2EV
c h̄ ( σ . k 0 ) ( σ . k )
2 2
 
T
× χσ0 I + χσ
( E + m c2 )2
(1346)
where the normalization constants have been combined, since energy is con-
served. Likewise, the complex conjugate matrix elements are given by
4 π Z e2 E + m c2
  
0 0
< k, σ | ĤInt | k , σ > = −
V | k − k 0 |2 2EV
c h̄ ( σ . k ) ( σ . k 0 )
2 2
 
T
× χσ I + χσ0
( E + m c2 )2
(1347)
These expressions for the matrix elements are inserted into the scattering cross-
section. Since the final state polarization is not detected, then σ 0 must be
summed over. The trace over σ 0 is evaluated by using the completeness relation
X
χσ0 χTσ0 = I (1348)
σ0

233
The resulting matrix elements involve the spin-dependent factor

c2 h̄2 ( σ . k ) ( σ . k 0 ) c2 h̄2 ( σ . k 0 ) ( σ . k )
  
T
χσ I + I + χσ
( E + m c2 )2 ( E + m c2 )2
(1349)

The products of matrix elements shown above can be evaluated with the aid of
the Pauli identity. The sum of the cross-terms can be evaluated directly using
the Pauli identity. We note that since the vector product are antisymmetric in
k and k 0 , the sum of the vector product terms cancel. That is

c2 h̄2
 
0 0
( σ . k ) ( σ . k ) + ( σ . k ) ( σ . k )
( E + m c2 )2
c2 h̄2
= 2 ( k . k 0 ) Iˆ (1350)
( E + m c2 )2
The remaining term is evaluated by using the Pauli identity for the inner two
scalar products, and then re-using the identity for the outer two scalar products.
Explicitly, this process yields

c4 h̄4
 
0 0
(σ.k)(σ.k )(σ.k )(σ.k)
( E + m c2 )4
c4 h̄4
= k 02 k 2 Iˆ (1351)
( E + m c2 )4
Hence, the cross-section is given by
2 
Z e2 c4 h̄4 k 2 k 02
   
dσ 2 2 2 2 0
= ( E + m c ) + 2 c h̄ k . k +
dΩ0 h̄2 c2 | k − k 0 |2 ( E + m c2 )2
(1352)
It should be noted that the last two terms originated from the combined action
of the Pauli spin operators and involved the lower two-component spinors. The
last term can be simplified by using the elastic scattering condition k = k 0 and
then using the identity

c4 h̄4 k 4 = ( E 2 − m2 c4 )2 (1353)

in the numerator. On canceling the factor of ( E + m c2 )2 in the denominator of

the last term with a similar factor in the numerator, the last term is recognized
as being just ( E − m c2 )2 . Hence, on combining the first and last terms, one
finds the result
2 
Z e2
   
dσ 2 2 4 2 2 0
= 2 ( E + m c ) + 2 c h̄ k . k (1354)
dΩ0 h̄2 c2 | k − k 0 |2
The scattering angle θ0 is introduced in the square parenthesis through

k . k 0 = k 2 cos θ0 (1355)

234
and also in the denominator of the Coulomb interaction by

θ0
| k − k 0 |2 = 4 k 2 sin2 (1356)
2
Furthermore, the factor of m2 c4 in the square parenthesis can be replaced by

m2 c4 = E 2 − c2 h̄2 k 2 (1357)

so that the cross-section takes the form

2  0
Z e2
   
dσ 2 2 2 2 2 θ
= 0 E − c h̄ k sin
dΩ0 2 h̄2 c2 k 2 sin2 θ2 2
2 2
2 Z e2 E θ0
     
v
= 2 2 2 2 θ0
1 − sin2 (1358)
4 h̄ c k sin 2 c 2

where the expression for the magnitude of the velocity

2
c2 h̄ k

2
v = (1359)
E

has been introduced. The above result is the Mott scattering cross-section96 ,
which describes the scattering of charged electrons. It differs from the Ruther-
ford scattering cross-section due to the multiplicative factor of relativistic origin,
which deviates from unity due to the electron’s internal degree of freedom. The
extra contribution to the scattering is interpreted in terms of scattering from the
magnetic moment associated with the electron’s spin interacting with the mag-
netic field of the nuclear charge that the electron experiences in its rest frame.
It should be noted that even if the initial beam of electrons is un-polarized, the
scattered beam will be partially spin-polarized (due to higher-order corrections).

12.4.2 Maxwell’s Equations

Maxwell’s equations can be written in the form of the Dirac equation. We
introduce a four-component wave function ψ given by
 
0
 B (1) − iE (1) 
ψ =   B (2) − iE (2) 
 (1360)
B (3) − iE (3)

Maxwell’s equations can be written in the form

4π
i αµ ∂µ ψ = − j (1361)
c
96 N. F. Mott, Proc. Roy. Soc. A 124, 425 (1929).

235
where j is the contravariant form of the current four-vector
 
cρ
 j (1) 
j =  j (2) 
 (1362)
j (3)

We shall require that the matrices αµ are Hermitean and that they satisfy the
equation
( αµ )2 = Iˆ (1363)
On comparing with the form of Maxwell’s equations97 , one finds that the Ma-
trices are given by  
1 0 0 0
 0 1 0 0 
α(0) =  0 0 1 0 
 (1364)
0 0 0 1
 
0 −1 0 0
 −1 0 0 0 
α(1) =   (1365)
 0 0 0 −i 
0 0 i 0
 
0 0 −1 0
 0 0 0 i 
α(2) =  −1 0
 (1366)
0 0 
0 −i 0 0
 
0 0 0 −1
 0 0 −i 0 
α(3) =   0 i 0
 (1367)
0 
−1 0 0 0
The matrices corresponding to the spatial indices are traceless and satisfy the
anti-commutation relations

α(i) α(j) + α(j) α(i) = 2 δi,j (1368)

and X
α(i) α(j) = i ξ i,j,k α(k) (1369)
k

On pre-multiplying Maxwell’s equations in the form

4π
i αµ ∂µ ψ = − j (1370)
c
97 Since the first element of ψ is zero, the first columns of the matrices are not determined

directly from the comparison. The first rows are determined by demanding that the matrices
are Hermitean.

236
with the operator
i αν ∂ν (1371)
one obtains
4π ν
− αν αµ ∂ν ∂µ ψ = − i α ∂ν j (1372)
c
Utilizing the anti-commutation of the spatial matrices, the left-hand side sim-
plifies to    
1 ∂
− − ∂µ ∂ µ + 2 αν ∂ν ψ (1373)
c ∂t
On substituting the new form of Maxwell’s equations in the second term, the
expression reduces to
 
8π ∂
− − ∂µ ∂ µ ψ + i 2 j (1374)
c ∂t
Thus, the equation becomes
 
4π 2 ∂
∂µ ∂ µ ψ = i − αν ∂ν j (1375)
c c ∂t
The zero-th component of the source term vanishes, due to conservation of
charge.

12.4.3 The Gordon Decomposition

The interaction of the Dirac particle with the electromagnetic field is described
by the interaction Hamiltonian which is described by the 4 × 4 matrix
 
q
ĤI = c γ (0) γ µ Aµ (1376)
c
The matrix interaction Hamiltonian operator yields an interaction Hamiltonian
density ĤI given by
 
q †
ĤI = c ψ γ µ ψ Aµ
c
 
q
= j µ Aµ (1377)
c
where j µ is the four-vector probability current density which satisfies the con-
dition for conservation of probability. Due to the prominence of the current
density operator in applications of the Dirac equation, since it naturally de-
scribes interactions with an electromagnetic field and the conservation laws, the
physical content of the current densities shall be examined next.

In the presence of an electromagnetic field, the four-vector current density

is given by the expression
†
jν = c ψ γν ψ (1378)

237
where
†
ψ = ψ † γ (0) (1379)
One can rewrite the current density by using the Dirac equation
 
q
i h̄ γ µ ∂µ + i Aµ ψ = m c ψ (1380)
h̄ c

and the Hermitean conjugate equation

 
q
− i h̄ ∂µ − i Aµ ψ † γ µ† = m c ψ † (1381)
h̄ c

On symmetrizing the current density and then substituting the Dirac equation
in one term and its Hermitean conjugate in the other term, one obtains
 
ν c † ν † ν
j = ψ γ ψ + ψ γ ψ
2
 
i h̄ q † µ† (0) ν † (0) ν µ q
= − ( ∂µ − i Aµ )ψ γ γ γ ψ + ψ γ γ γ ( ∂µ + i Aµ )ψ
2m h̄ c h̄ c
 
i h̄ q † (0) µ† (0) ν † ν µ q
= − ( ∂µ − i Aµ )ψ γ γ γ γ ψ + ψ γ γ ( ∂µ + i Aµ )ψ
2m h̄ c h̄ c
(1382)

where the partial derivatives only operate on the wave function immediately to
the right of it. The identity
γ (0) γ (0) = Iˆ (1383)
†
has been used to express ψ † in terms of ψ . However, since the γ matrices
satisfy
γ (0) γ µ† γ (0) = γ µ (1384)
the current can be further simplified to yield
 
ν i h̄ q † µ ν † ν µ q
j = − ( ∂µ − i Aµ )ψ γ γ ψ + ψ γ γ ( ∂µ + i Aµ )ψ
2m h̄ c h̄ c
(1385)

where, once again, the partial derivative only operates on the wave function
immediately to the right of it. Furthermore, if one sets
 
1
γµ γν + γν γµ = g µ,ν Iˆ
2
 
1
γµ γν − γν γµ = − i σ µ,ν (1386)
2

238
then the current density can be expressed as the sum of two contributions

jν = jcν + jsν
 
i h̄ µ,ν † † q µ,ν †
= − g ( ∂µ ψ ψ − ψ ∂µ ψ ) + 2 i g ψ Aµ ψ
2m h̄ c
 
h̄ ∂ †
− µ
ψ σ µ,ν ψ (1387)
2 m ∂x

where
 
i h̄ † † q †
jcν = − ( ∂ν ψ ψ − ψ ∂ν ψ ) + 2 i ψ Aν ψ
2m h̄ c
 
h̄ ∂ † µ,ν
jsν = − ψ σ ψ (1388)
2 m ∂xµ

This is the Gordon decomposition98 of the probability current density. A similar

expression can be derived for the matrix elements of the interaction operator
†
between states ψ β and ψα . As shall be shown, the first contribution in the Gor-
don decomposition is gauge invariant and dominates the current density in the
non-relativistic limit. The second contribution involves the matrix σ µ,ν which
is anti-symmetric in its indices and has the form of a spin contribution to the
current density.

Let us examine the first term in the probability current density. If ψ repre-
(0)
sents an energy eigenstate, then jc is given by
 
(0) E † q †
jc = ψ ψ − ψ A(0) ψ (1389)
mc mc

This contribution obviously yields the main contribution to (c times) the prob-
ability density
†
jc(0) ≈ c ψ ψ (1390)
in the non-relativistic limit since the rest mass energy dominates the energy
(i)
E ∼ m c2 . The spatial components of jc are given by
 
i h̄ † † q †
jc = (∇ψ )ψ − ψ (∇ψ) − ψ Aψ (1391)
2m mc

where the derivatives have been expressed as derivatives w.r.t. the contravariant
components x(i) of the position vector. This expression coincides with the full
non-relativistic expression for the current density j (i) .

We now examine the second term jsµ in the Gordon decomposition. For
future reference, the anti-symmetrized products of the Dirac matrices σ µ,ν will
98 W. Gordon, Zeit. für Physik, 50, 630 (1928).

239
be expressed in 2 × 2 block diagonal form. Therefore, since
 
(0) I 0
γ =
0 −I
σ (i)
 
(i) 0
γ = (1392)
−σ (i) 0

and  
i
σ µ,ν = γµ γν − γν γµ (1393)
2
the matrices are found as
σ (j)
 
0
σ 0,j = i (1394)
σ (j) 0

and
σ (k)
 
i,j
X
i,j,k 0
σ = ξ (1395)
0 σ (k)
k

The two by two block diagonal matrix of Pauli spin matrices will be denoted
(0)
by σ̂. For an energy eigenstate, the time component of js is identically zero.
(i)
Hence, the spatial components of js are given by
h̄ †
js = − ∇ ∧ ( ψ σ̂ ψ ) (1396)
2m
where σ̂ is the 2 × 2 block-diagonal Pauli spin matrix
 
σ 0
σ̂ = (1397)
0 σ

The additional term in the current density clearly involves the Pauli spin-
matrices. To elucidate its meaning, its contribution to the energy shall be
examined. On substituting this term in the interaction Hamiltonian density,
one finds a contribution
spin q
ĤI = − j .A
c s  
q h̄ †
= + A . ∇ ∧ ( ψ σ̂ ψ ) (1398)
2mc
On integrating over space, the interaction Hamiltonian density gives rise to the
interactions contribution to the total energy. By integrating by parts, it can
be shown that this energy contribution is equivalent to the energy contribution
caused by an equivalent form of the interaction Hamiltonian density

spin q h̄ †
ĤI ≡ − ( ψ σ̂ ψ ) . ( ∇ ∧ A )
2mc
q h̄ †
≡ − ( ψ σ̂ ψ ) . B (1399)
2mc

240
where B is the magnetic field. Hence, the interaction energy contains a term
which represents an interaction between the electron’s internal degree of free-
dom and the magnetic field.

12.5 Lorentz Covariance of the Dirac Equation

One goal of Physics is to write the laws in a manner which are independent
of any arbitrary choices that are made. Within special relativity, this implies
that the laws of Physics should be written in a way which is independent of
the choice of inertial reference frame. Dirac’s theory is Lorentz covariant if the
results are independent of the Lorentz frame used. To this end, it is required
that the Dirac equation in a Lorentz transformed frame of reference has the
same form as the Dirac equation in the original reference frame, and also that
the solutions of these two equations describe the same physical states. That is,
the two solutions must describe the same set of measurable properties in the
different reference frames, and therefore the results are simply related by the
Lorentz transformation.

The first step of the proof of the Lorentz covariance of the Dirac equation
requires that one should be able to show that under a Lorentz transformation
defined by
Aµ → Aµ0 = Λµ ν Aν (1400)
then the Dirac equation is transformed from
q
γ µ ( p̂µ − Aµ ) ψ = m c ψ (1401)
c
to an equation with an equivalent form
q 0
γ µ0 ( p̂0µ − A ) ψ0 = m c ψ0 (1402)
c µ
Furthermore, the four components of the spinor wave function ψ 0 are assumed
to be linearly related to the components of ψ by a four by four matrix R̂(Λ)
which is independent of xµ

ψ 0 (x0 ) = R̂(Λ) ψ(x) (1403)

Hence, the transformed Dirac equation can be re-written in terms of the un-
transformed spinor
q 0
γ µ0 ( p̂0µ − A ) ψ0 = m c ψ0
c µ
q 0
γ µ0 ( p̂0µ − A ) R̂(Λ) ψ = m c R̂(Λ) ψ (1404)
c µ

if such an R̂(Λ) exists. The γ µ0 matrices must satisfy the same anti-commutation
relations as the γ µ and, therefore, only differ from them by a similarity trans-

241
formation99 . The transformations of γ µ0 just results in the set of the four linear
equations that compose the Dirac equation being combined in different ways,
so this rearrangement can be absorbed in the definition of R̂(Λ). That is, one
can choose to impose the convention that γ µ0 = γ µ . The transformed Dirac
equation can be expressed as
q 0
γ µ0 ( p̂0µ − A ) R̂(Λ) ψ = m c R̂(Λ) ψ
c µ
q
γ µ0 Λµ ν ( p̂ν − Aν ) R̂(Λ) ψ = m c R̂(Λ) ψ (1405)
c
where the transformation properties of the momentum four-vector have been
used100 . On multiplying by the inverse of R̂(Λ), one has
q
R̂−1 (Λ) γ µ0 Λµ ν ( p̂ν − Aν ) R̂(Λ) ψ = mcψ (1406)
c
q
R̂−1 (Λ) γ µ0 R̂(Λ) Λµ ν ( p̂ν − Aν ) ψ = mcψ (1407)
c

where the four by four matrices R̂(Λ) have been commuted with the differential
operators and also with the components of the Lorentz transform. The condition
for covariance as
R̂−1 (Λ) γ µ0 R̂(Λ) Λµ ν = γ ν (1408)
The transformed Dirac equation has the same form as the original equation if
the transformed γ µ0 matrices satisfy the same anti-commutations and conditions
as the unprimed matrices. This can be achieved by choosing γ µ0 = γ µ . This
choice yields the condition for covariance as

R̂−1 (Λ) γ µ R̂(Λ) Λµ ν = γ ν (1409)

Since for a Lorentz transform one has

Λ µ ν Λ ρ ν = δµ ρ (1410)

then multiplying the above covariance condition by Λρ ν leads to

R̂−1 (Λ) γ µ R̂(Λ) = Λµ ν γ ν (1411)

The above equation determines the 4 × 4 matrix R̂(Λ). If R̂(Λ) exits, the Dirac
equation has the same form in the two frames of reference and the solutions
are linearly related. Pauli’s “fundamental theorem” guarantees that a matrix
R̂(Λ) exists which does satisfy the condition. Instead of following the general
theorem, the solution will be inferred from consideration of infinitesimal Lorentz
99 This is a statement of Pauli’s fundamental theorem [W. Pauli, Ann. Inst. Henri Poincaré

6, 109 (1936).]. For a general discussion, see R. H. Good Jr. Rev. Mod. Phys. 27, 187
(1955).
100 It should be noted that the matrices Λ ν and R̂ act on totally different spaces. The
µ
matrices Λµ ν act on the components of the four-vectors xν , whereas the R̂ matrices act on
the components of the four-component Dirac spinor ψ.

242
transformations.

The matrix R̂(Λ) will be determined by considering the effect of an infinites-

imal Lorentz transformation

Λµ ν = δ µ ν + µ ν + . . . (1412)

where δ µ ν is the Kronecker delta function. The matrix R̂(Λ) for the infinitesimal
transformation can also be expanded as
i µ
R̂ = Iˆ −  ν ωµ ν + . . . (1413)
4
where ωµ ν is a four by four matrix that has yet to be determined. The inverse
matrix can be written as
i µ
R̂−1 = Iˆ +  ν ωµ ν + . . . (1414)
4
to first-order in the infinitesimal quantity µ ν . On substituting the matrices for
the infinitesimal transform into the equation that determines R̂, one obtains
 
i ρ σ µ µ σ
 σ ωρ γ − γ ωρ = µ ν γ ν (1415)
4

or on raising and lowering indices

 
i ρσ µ µ ρσ
ρσ ω γ − γ ω = g µρ ρσ γ σ (1416)
4

Thus, since ρσ is anti-symmetric as it represents an infinitesimal Lorentz trans-

formation, the matrix ω ρσ can be restricted to be anti-symmetric in the indices,
because any symmetric part does not contribute to the matrix R̂. By making
specific choices for the anti-symmetric quantities ρσ , which are zero except for
a chosen pair of indices (say α and β), one finds that the anti-symmetric part
of ω αβ is determined from the equation
i
[ ω αβ , γ µ ] = g µα γ β − g µβ γ α (1417)
2
These sets of equations have to be satisfied even if arbitrary choices are made for
the infinitesimal Lorentz transformations ρσ . The infinitesimal unitary matrix
R̂ can be expressed in terms of six generators ω ρσ of the infinitesimal Lorentz
transformation
i
R̂ = Iˆ − ρσ ω ρσ + . . . (1418)
4
The set of matrices ω ρσ that define R̂ must satisfy the equation
i
[ ω ρσ , γ µ ] = g µρ γ σ − g µσ γ ρ (1419)
2

243
The set of (as yet unknown) matrices ω ρσ that solve the above set of equations
are given by
i
ω αβ = σ αβ = [ γα , γβ ] (1420)
2
which are the six generators of the general infinitesimal Lorentz transformation.
This solution, and hence, the existence of R̂(Λ) shows that the solutions of the
Dirac equation and the transformed equation are in a one to one correspondence.

——————————————————————————————————

Proof of Solution

It can be shown that the expression for σ α,β given in eqn(1420) satisfies the
requirement of eqn(1419), by evaluating the nested commutator through repeat-
edly using the anti-commutation properties of the γ matrices. The commutator
can be expressed as a nested commutator or as the sum of two commutators
i
[ σ αβ , γ µ ] = [ [ γα , γβ ] , γµ ]
2
i i
= [ γα γβ , γµ ] − [ γβ γα , γµ ] (1421)
2 2
On using the anti-commutation relation for the γ matrices
 
1
γα γβ + γβ γα = g α,β Iˆ (1422)
2
one can eliminate the second term leading to
[ σ αβ , γ µ ] = i [ γ α γ β , γ µ ] + i g α,β [ Iˆ , γ µ ]
= i [ γα γβ , γµ ] (1423)
where the second line follows since the identity matrix commutes with γ µ . One
notices that if the γ µ ’s are anti-commuted to the center of each product, some
terms will cancel and there may be some simplification. On using the anti-
commutation relation in the second term of the expression
 
αβ µ α β µ µ α β
[σ , γ ] = i γ γ γ − γ γ γ (1424)

one finds
 
[ σ αβ , γ µ ] = i γ α γ β γ µ + γ α γ µ γ β − 2 g µ,α γ β (1425)

Likewise, the γ matrices in the first term can also be anti-commuted, leading to
 
αβ µ µ,β α α µ β α µ β µ,α β
[σ , γ ] = i 2g γ − γ γ γ + γ γ γ − 2g γ
 
= 2 i g µ,β γ α − g µ,α γ β (1426)

244
since the middle pair of terms cancel. Hence, one has proved that
 
i αβ µ µ,α β µ,β α
[σ , γ ] = g γ − g γ (1427)
2

which completes the identification of the solution of the equation for ω α,β .
Therefore, since R̂(Λ) exists, it has been shown that the form of the Dirac
equation is maintained in the primed reference frame and that there is a one
to one correspondence between the solutions of the primed and unprimed frames.

——————————————————————————————————

Equivalence of Physical Properties

It remains to be shown that the ψ and ψ 0 describe the properties of the same
physical system, albeit in two different frames of reference. That is, the proper-
ties associated with ψ must be related to the properties of ψ 0 and the relation
can be obtained by considering the Lorentz transformation. The most complete
physical descriptions of a unique quantum mechanical state are related to the
probability density, which can only be inferred from an infinite set of position
measurements. The probability density, should behave similarly to the time
component of a four-vector as was seen from the consideration of the continuity
equation. Therefore, it follows that if the four-vector probability currents of ψ
and ψ 0 are related via a Lorentz transformation, then the two spinors describe
the same physical state of the system.

The probability current four-vector j µ in the unprimed frame is described

by
†
jµ = c ψ γµ ψ
= c ψ † γ (0) γ µ ψ (1428)

and in the primed frame, one has

j µ0 = c ψ †0 γ (0) γ µ ψ 0
= c ψ † R̂† γ (0) γ µ R̂ ψ (1429)

The identity
R̂−1 = γ (0) R̂† γ (0) (1430)
will be proved below, so on using this identity together with

γ (0) γ (0) = Iˆ (1431)

the probability current density can be re-written as

j µ0 = c ψ † γ (0) γ (0) R̂† γ (0) γ µ R̂ ψ

= c ψ † γ (0) R̂−1 γ µ R̂ ψ (1432)

245
However, because the covariant condition is given by

R̂−1 (Λ) γ µ R̂(Λ) = Λµ ν γ ν (1433)

the current density can be expressed as

j µ0 = c ψ † γ (0) Λµ ν γ ν ψ
†
= Λµ ν c ψ γ ν ψ
= Λµ ν j ν (1434)

Hence, the probability current densities j µ0 and j µ found in the two reference
frames are simply related via the Lorentz transformation. Therefore, the Dirac
equation gives consistent results, no matter what inertial frame of reference is
used.

——————————————————————————————————

Proof of Identity

The identity
R̂−1 (Λ) = γ (0) R̂† (Λ) γ (0) (1435)
can be proved by starting from the expression for the expression for R̂ appro-
priate for infinitesimal transformation given by
1
R̂ = Iˆ + µν [ γ µ , γ ν ] + . . . (1436)
8
Hence, the Hermitean conjugate is given by
1
R̂† = Iˆ + µν [ γ ν † , γ µ† ] + . . .
8
1
R̂† = Iˆ − µν [ γ µ† , γ ν † ] + . . . (1437)
8
since the Hermitean conjugate of a product is the product of the Hermitean
conjugate of the factors taken in opposite order. On forming the product
γ (0) R̂† γ (0) and inserting a factor of

γ (0) γ (0) = Iˆ (1438)

between the pairs of four by four γ matrices in the commutator and noting that

γ (0) γ µ† γ (0) = γ µ (1439)

one finds that

1
γ (0) R̂† γ (0) = Iˆ − µν [ γ µ , γ ν ] + . . .
8
= R̂−1 (1440)

246
The last line follows from the observation that on combining the expression for
R̂ with the expression for γ (0) R̂† γ (0) , the terms of order  cancel. Hence to
the order of 2 , the product γ (0) R̂† γ (0) coincides with R̂−1 . This concludes
the discussion of the desired identity.

Finite Rotations

Consider a finite rotation of the coordinate system specified by the transfor-

mation matrix Λ
xµ0 = Λµ ν xν (1441)
Specifically, a finite (passive) rotation through an angle ϕ about the ê3 direction
can be expressed in terms of the transformation matrix
 
1 0 0 0
 0 cos ϕ + sin ϕ 0 
Λ =   0 − sin ϕ
 (1442)
cos ϕ 0 
0 0 0 1
The above transformation represents a rotation of the coordinate system while
the physical system stays put. For an infinitesimal rotation through δϕ, the

x(2)
x(2)'

x(1)'

ϕ
x(1)

Figure 45: A passive rotation of the coordinate system through an angle ϕ about
the ê3 -axis.

transformation matrix reduces to

 
1 0 0 0
 0 1 + δϕ 0 
Λ =   0 − δϕ
 + ... (1443)
1 0 
0 0 0 1
to first-order in the infinitesimal quantity δϕ. Therefore, with the infinitesimal
form of the general Lorentz transformation

Λµ ν = δ µ ν + µ ν + . . . (1444)

247
on lowering the first index, one identifies

12 = − 21 = − δϕ (1445)

The infinitesimal transformation of a Dirac spinor was determined to be given

by
i
R̂(δϕ) = Iˆ − µν σ µν + . . . (1446)
4
Hence, for a infinitesimal rotation one has
i
R̂(δϕ) = Iˆ + ( δϕ σ 1,2 − δϕ σ 2,1 ) + . . .
4
i
= Iˆ + δϕ σ 1,2 + . . .
 2 
δϕ 1,2
= exp i σ (1447)
2
since σ µ,ν is anti-symmetric. On compounding N infinitesimal transformations
about the same axis R̂(δϕ) using their exponential form, and defining N δϕ =
ϕ, one obtains the finite rotation R̂(ϕ)
 N
R̂(ϕ) = R̂(δϕ)
 
δϕ 1,2
= exp i N σ
2
 
ϕ 1,2
= exp i σ (1448)
2
Therefore, for a finite rotation, the transformation matrix is given by
 
ϕ 1,2
R̂(ϕ) = exp i σ (1449)
2

which can be expressed in terms of even and odd-powers of σ 1,2 via

   
ϕ 1,2 ϕ 1,2
R̂(ϕ) = cos σ + i sin σ (1450)
2 2
but since
i
σ 1,2 = [ γ (1) , γ (2) ]
2
= σ̂ (3) (1451)

the transformation can be expressed as

   
ϕ (3) ϕ (3)
R̂(ϕ) = cos σ̂ + i sin σ̂ (1452)
2 2

248
The above expression can be simplified by expanding the trigonometric functions
in series of ϕ and then using the property of the σ̂ (j) matrices

( σ̂ (3) )2 = Iˆ (1453)

Since the repeated use of the above identity leads to

( σ̂ (3) )2n = Iˆ
( σ̂ (3) )2n+1 = σ̂ (3) (1454)

the series simplify and can be re-summed leading to

   
ϕ ˆ ϕ
R̂(ϕ) = cos I + i sin σ̂ (3) (1455)
2 2

Therefore, under a finite rotation through angle ϕ around the unit vector ê, a
spinor is rotated by the operator
ϕ ˆ ϕ
R̂(ϕ) = cos I + i sin ê . σ̂ (1456)
2 2
From the above equation, due to the presence of the half-angle, one notes that
a rotation ϕ and through ϕ + 2π are not equivalent, since

R̂(ϕ + 2π) = − R̂(ϕ) (1457)

which changes the sign of the spinor. For spin one-half electrons, it is necessary
to rotate through 4π to return to the same state

R̂(ϕ + 4π) = R̂(ϕ) (1458)

†
A quantity which is bi-linear in ψ and ψ will remain invariant under a rotation
of 2π.

Finite Lorentz Boosts

A finite Lorentz boost by velocity v along the ê1 direction can be expressed
in terms of the transformation
 
cosh χ − sinh χ 0 0
 − sinh χ cosh χ 0 0 
Λ =   (1459)
 0 0 1 0 
0 0 0 1

where the rapidity χ is defined by

v
tanh χ = (1460)
c

249
so
1
cosh χ =
1 − ( vc )2
p
v
c
sinh χ = (1461)
1 − ( vc )2
p

For an infinitesimal boost through δχ, the transformation matrix reduces to

 
1 − δχ 0 0
 − δχ 1 0 0 
Λ =   0
 + ... (1462)
0 1 0 
0 0 0 1
to first-order in the infinitesimal quantity δχ. Therefore, with the infinitesimal
form of the general Lorentz transformation
Λµ ν = δ µ ν + µ ν + . . . (1463)
on lowering the first index, one identifies
01 = − 10 = − δχ (1464)
The infinitesimal transformation of a Dirac spinor was determined to be given
by
i
R̂(δχ) = Iˆ − µν σ µν + . . . (1465)
4
Hence, for a infinitesimal Lorentz boost one has
i
R̂(δχ) = Iˆ + 2 δχ σ 0,1 + . . .
 4 
δχ 0,1
= exp i σ (1466)
2
On compounding N successive infinitesimal Lorentz boosts (with parallel veloc-
ities) given by R̂(δχ) and defining N δχ = χ, one obtains the finite Lorentz
boost R̂(χ)
 N
R̂(χ) = R̂(δχ)
 
δχ 0,1
= exp i N σ
2
 
χ 0,1
= exp i σ (1467)
2
Therefore, for a finite Lorentz boost, the transformation matrix is given by
 
χ 0,1
R̂(χ) = exp i σ (1468)
2

250
which can be expressed in terms of even and odd-powers of σ 0,1 via
   
χ 0,1 χ 0,1
R̂(χ) = cosh i σ + sinh i σ (1469)
2 2
but since
i
σ 0,1 = [ γ (0) , γ (1) ]
2
= i α(1) (1470)

the transformation can be expressed as

   
χ (1) χ (1)
R̂(χ) = cosh − α + sinh − α (1471)
2 2
The above expression can be simplified by expanding the hyperbolic functions
in series of χ and then using the property of the α matrices

( α(1) )2 = Iˆ (1472)

Since the repeated use of the above identity leads to

( α(1) )2n = Iˆ
( α(1) )2n+1 = α(1) (1473)

the series simplify and can be re-summed leading to

   
χ ˆ χ
R̂(χ) = cosh − I + sinh − α(1)
2 2
   
χ ˆ χ
= cosh I − sinh α(1) (1474)
2 2
Therefore, under a finite boost through velocity v, a spinor is rotated by the
operator
 
χ ˆ χ (1)
R̂(χ) = cosh I − tanh α
2 2
 
χ ˆ χ
= cosh I − tanh v̂ . α (1475)
2 2
where the rapidity χ is determined by
v
tanh χ = (1476)
c

Exercise:

251
 Determine the relationship between the rapidities for a combined Lorentz
transformation consisting of two successive Lorentz boosts with parallel veloci-
ties v0 and v1 .

Exercise:

Starting from a solution of a free stationary Dirac particle with spin σ,

perform a Lorentz boost to determine the solution for a Dirac electron with
momentum p0 .

v A

L'

Figure 46: A cartoon depicting a stationary free-electron confined in a volume

V with proper length L, viewed from a coordinate system moving with velocity
v anti-parallel to the ê1 -axis.

Exercise:

Show that the helicity eigenvalue of a free Dirac particle can be reversed by
going to a new reference frame which is “overtaking” the particle.

12.5.1 The Space of the Anti-commuting γ µ -Matrices.

One can form sixteen matrices Γi from the product of the four γ matrices. Since
the γ µ matrices obey the anti-commutation relations

{ γ µ , γ ν }+ = 2 g µ,ν Iˆ (1477)

all other products can be reduced to the above products. The order of the
matrices is irrelevant, since the different matrices anti-commute. Also, since
( γ µ )2 = ± I,ˆ one only needs to consider the products in which each matrix
enters at most one time. Hence, since each of the four matrices either appear as
a factor or do not, there are only 24 such matrices. These sixteen Γi matrices
can be constructed from I, ˆ γ µ , σ µ,ν = i γ µ γ ν , γ (4) and γ (4) γ µ , by choosing

252
Table 6: The Set of the Sixteen Matrices Γn with their Phase Factors (j > i)

Iˆ

γ (0) i γ (i)

γ (0) γ (i) i εi,j,k γ (i) γ (j)

γ (4) = i γ (0) γ (1) γ (2) γ (3)

− i γ (0) γ (4) γ (i) γ (4)

appropriate phase factors.

Closure under Multiplication

The set of matrices Γi formed from the set of γ µ are closed under multipli-
cation, so
Γi Γj = ai,j Γk (1478)
where a4i,j = 1. The sixteen Γi matrices can be chosen as the product of the
members of the above set multiplied by a phase factor taken from the set ± 1
and ± i, such that the condition

( Γi )2 = Iˆ (1479)

is satisfied. Furthermore, by counting the number of non-equivalent factors of

the γ µ in the products, one can show that

Γi Γj = Iˆ only if i = j (1480)

Also, by anti-commuting the factors of γ µ in the products, one can show that

Γi Γj = ± Γj Γi (1481)

Specifically, for a fixed Γi not equal to the identity, one can always find a
specific Γk such that
Γi Γk = − Γk Γi (1482)
which on multiplying by Γk results in

Γk Γi Γk = − Γi (1483)

253
Traceless Matrices

The above facts can be used to show that the Γi matrices, other than the
identity, are traceless. This can be proved by considering

− Trace Γi = Trace( − Γi ) = Trace( Γk Γi Γk )

= Trace( Γi Γk Γk ) = Trace Γi (1484)

in which the existence of a specific Γk which anti-commutes with Γi has been

used, and where the cyclic invariance of the trace has been used as has been
ˆ Hence, all the Γi matrices, other than the identity, are traceless
( Γk )2 = I.

Trace Γi = 0 (1485)

Linear Independence

The sixteen Γi matrices are linearly independent. The linear independence

can be expressed in terms of the absence of any non-trivial solution of the
equation X
Ci Γi = 0 (1486)
i

other than Ci ≡ 0 for all i. If the Γi are linearly independent, the only solution
of this equation is
Ci ≡ 0 for all i (1487)
This can be proved by multiplying eqn(1486) by any one Γj in the set which
leads to
X
Cj Iˆ + Ci Γi Γj = 0
i6=j
X
Cj Iˆ + Ci ai,j Γk = 0 (1488)
i6=j

On taking the trace one finds

X
0 = Cj Trace Iˆ + Ci ai,j Trace Γk
i6=j
= Cj 4 (1489)

since the matrices Γk are traceless. Hence, all the Cj are zero, so the matrices
are linearly independent.

254
Uniqueness of Expansions

The existence of sixteen linearly independent matrices require that the ma-
trices can be represented in a space of N × N matrices, where N ≥ 4. Any
matrix A in the space of 4 × 4 matrices can be uniquely expressed in terms of
the basis set of the Γi . For example, if
X
A = Ci Γi (1490)
i

then on multiplying by Γj and taking the trace, one has

X
Trace( A Γj ) = Ci Trace( Γi Γj )
i
X
= Cj Trace( Γj Γj ) + Ci Trace( Γi Γj )
i6=j
X
= Cj Trace( Iˆ ) + Ci Trace( ai,j Γk )
i6=j
= Cj 4 (1491)

Hence, the coefficients Cj in the expansion of A are uniquely determined as

1
Cj = Trace( A Γj ) (1492)
4

Schur’s Lemma

The uniqueness of the expansion can be used to show that the product of Γi
for fixed i with the set of Γj for leads to a different Γk for each j. This can be
shown by assuming that there exist two different (linearly independent) values
Γj and Γj 0 which lead to the same Γk

Γi Γj = ai,j Γk
Γi Γj 0 = ai,j 0 Γk
(1493)

On multiplying by Γi , one obtains

Γj = ai,j Γi Γk
Γj 0 = ai,j 0 Γi Γk
(1494)

Hence, one infers that

ai,j 0
Γj 0 = Γj (1495)
ai,j

255
which contradicts the assumption that Γj and Γj 0 are linearly independent.
Therefore for fixed i, the product of Γi Γj leads to a different result Γk for the
different Γj .

One can also prove Schur’s lemma. Schur’s Lemma states that if a matrix A
commutes with all the γ µ ’s, then A is a multiple of the identity. If A commutes
with the γ µ ’s, it also commutes with all the Γi ’s. Schur’s lemma follows from
the expansion of A as X
A = Ci Γi + Cj Γj (1496)
j6=i

ˆ Then, one notes that there exits a Γk such that

for any i such that Γi 6= I.

Γk Γi Γk = − Γi (1497)

Since it has been assumed that A commutes with all the Γi , for the specific Γk
one has

A = Γk A Γk
X
= Ci Γk Γi Γk + Cj Γk Γj Γk
j6=i
X
= − Ci Γi + Cj Γk Γj Γk (1498)
j6=i

Furthermore, since the Γi matrices either commute or anti-commute

Γk Γj Γk = ( ± 1 )j,k Γj (1499)

the above equation reduces to

X
A = − Ci Γi + Cj ( ± 1 )j,k Γj (1500)
j6=i

which should be compared with the assumed form of the expansion

X
A = Ci Γi + Cj Γj (1501)
j6=i

Since the expansion is unique, the coefficients of the Γj are unique and in par-
ticular
Ci = − Ci (1502)
ˆ Hence, if A commutes with all the Γi
so Ci = 0 for any i such that Γi 6= I.
then A must be proportional to the identity.

Pauli’s Fundamental Theorem

256
 Pauli’s fundamental theorem states that if there are two representations of
the algebra of anti-commuting γ-matrices, say γ µ0 and γ µ , then these represen-
tations are related via a similarity transformation

γ µ0 = Ŝ γ µ Ŝ −1 (1503)

where Ŝ is a non-singular matrix.

The theorem requires that one constructs a set of sixteen matrices Γi 0 from
the γ µ0 following the same rules with which the Γi were constructed from γ µ .
Then one can describe the non-singular matrix by
X
Ŝ = Γi 0 F Γi (1504)
i

where F is an arbitrary 4 × 4 matrix.

First one notes that

Γi Γj = ai,j Γk (1505)
so on iterating, one has

Γi Γj Γi Γj = a2ı,j Γ2k = a2i,j Iˆ (1506)

since
Γ2k = Iˆ (1507)
On pre-multiplying eqn(1506) by Γj Γi , one obtains

Γj Γi Γi Γj Γi Γj = a2i,j Γj Γi (1508)

but since

Γj Γi Γi Γj = Iˆ (1509)

eqn(1508) reduces to

Γi Γj = a2i,j Γj Γi (1510)

However, as
Γi Γj = ai,j Γk (1511)
the equation becomes

ai,j Γk = a2i,j Γj Γi (1512)

or since a4i,j = 1, the equation can be expressed as

Γj Γi = a3i,j Γk (1513)

257
 0
The Γi matrices are constructed so that they satisfy similar relations to the Γi .
0
In particular, the Γi matrices satisfy

Γi 0 Γj 0 = ai,j Γk 0 (1514)

with the same constants ai,j as the unprimed matrices.

Pauli’s theorem follows from the above relations by noting that

X
Γi 0 Ŝ Γi = Γi 0 ( Γj 0 F Γj ) Γ i (1515)
j

but on recalling that

Γj Γi = a3i,j Γk (1516)
and
Γi 0 Γj 0 = ai,j Γk 0 (1517)
one finds X
Γi 0 Ŝ Γi = a4i,j Γk 0 F Γk (1518)
j

Therefore, with a4i,j = 1, the above equation reduces to

X
Γi 0 Ŝ Γi = Γk 0 F Γk (1519)
j

However, since i is fixed and j is being summed over, every Γk appears once
and only once in the product. Therefore, the sum can be performed over k
X
Γi 0 Ŝ Γi = Γk 0 F Γk = Ŝ (1520)
k

If one can show that the matrix Ŝ has an inverse, then on post-multiplying by
Ŝ −1 , one finds
Γi 0 Ŝ Γi Ŝ −1 = Iˆ (1521)
Furthermore, since Γi 0 is its own inverse, then on pre-multiplying by Γi 0 the
equation reduces to
Ŝ Γi Ŝ −1 = Γi 0 (1522)
This is a generalization of the statement of the theorem. As a particular case,
one may choose Γi = γ µ in which case the theorem becomes

γ µ0 = Ŝ γ µ Ŝ −1 (1523)

which was the initial statement of Pauli’s fundamental theorem made above.

258
 The matrix Ŝ is non-singular and has an inverse. This can be shown by
using Schur’s Lemma. One can construct a matrix Ŝ 0 in a manner which is
symmetrical to the construction of Ŝ. That is
X
Ŝ 0 = Γi G Γi 0 (1524)
i

From symmetry it follows that since eqn(1520) is given by

Ŝ = Γi 0 Ŝ Γi (1525)

one also has

Ŝ 0 = Γi Ŝ 0 Γi 0 (1526)
Therefore, on taking the product, one obtains

Ŝ 0 Ŝ = Γi Ŝ 0 Γi 0 Γi 0 Ŝ Γi
= Γi Ŝ 0 Ŝ Γi (1527)

Hence, by Schur’s Lemma one sees that Ŝ 0 Ŝ commutes with all the matrices in
the space, therefore it must be a multiple of the identity

Ŝ 0 Ŝ = κ Iˆ (1528)

where κ is a constant. By a judicious choice of the magnitude of the elements

of F , the constant κ can be set to unity, yielding

Ŝ 0 Ŝ = Iˆ (1529)

Thus, Ŝ is non-singular so the inverse exists and is given by Ŝ −1 = Ŝ 0 .

12.5.2 Polarization in Mott Scattering

When evaluated in the Born Approximation, Mott scattering does not result
in the polarization of an unpolarized beam. However, when higher-order cor-
rections are included, Mott scattering produces a partially polarization of the
scattered electrons101 . If the incident beam is polarized by having a definite
helicity, it is expected that the helicity may change as a result of the scattering.

The probability of non-helicity flip scattering and helicity flip scattering can
be evaluated using the Born approximation. The initial beam will be considered
as having a momentum p parallel to the ê3 -axis and as having a helicity of +1.
The initial spinor is proportional to
s
Ep + m c2 p.r
   
χ+
ψp,+ (r) = c p exp i (1530)
2 Ep V Ep + m c2 χ+ h̄
101 N. F. Mott, Proc. Roy. Soc. A 124, 425 (1929).

259
 p'

p θ'

p'

Figure 47: Helcity non-flip and helicity flip Mott scattering of an electron with
helicity +1. The scattering angle is θp0 .

The electrons are assumed to be elastically scattered to a state with final mo-
mentum p0 . The scattering is defined to occur through an angle θp0 in the z − x
plane. The final state is composed of a linear-superposition of states with dif-
ferent helicities. Since the final state helicities are specified relative to the final
momentum, the final state helicity eigenstates can be obtained by rotating the
initial state helicity eigenstates through an angle θp0 around the ê2 -axis

ψp0 0 ,Λ0 (r) = R̂(θp0 ) ψp0 ,Λ0 (x)

s !
Ep + m c2 χΛ0 p0 . r
 
= R̂(θp0 ) 0
c p Λ exp i
2 Ep V Ep + m c2 χΛ0 h̄
(1531)

where the rotation operator is given by

 
θp0 ˆ θp0 (2)
R̂(θp0 ) = cos I − i sin σ̂ (1532)
2 2
which does not mix the upper and lower two-component spinors. Therefore, one
finds that the final state two-component spinors representing helicity eigenstates
are given by
 
0 θp0 θp0 (2)
χ+ = cos I − i sin σ χ+
2 2
!
θ 0
cos 2p
= θ 0 (1533)
sin 2p

and
 
θp0 θp0 (2)
χ0− = cos I − i sin σ χ−
2 2

260
 !
θ 0
− sin 2p
= θ 0 (1534)
cos 2p

Therefore, the final state basis states are given by

s !
Ep + m c2 χ0Λ0 p0 . r
 
0
ψp0 ,Λ0 (r) = c p Λ0
0 exp i (1535)
2 Ep V Ep + m c2 χΛ0 h̄

The Born approximation scattering cross-section can be expressed in terms of

the modulus squared matrix elements
2
Z e2
 Z
†
d3 r ψ 0 p0 ,Λ0 (r) γ (0)
 
 ψp,+ (r)  (1536)
 |r| 

which is evaluated as
2  2  2 2
4 π Z e2 c2 p2 E p + m c2
 
0
 0† 
1 + Λ  χ Λ0 χ+ 
V | p − p0 |2 ( Ep + m c2 )2 2 Ep  
2  2 2
2
( Ep + m c ) + Λ0 c2 p2
2 2
  
4πZ e  0† 
=  χ Λ0 χ+ (1537)
V | p − p0 |2 2 E p ( E p + m c2 )  

Therefore, the probability for non-helicity flips scattering is proportional to

2
4 π Z e2

θp0
∝ 0 2
cos2 (1538)
V |p − p | 2
whereas the probability for helicity flip scattering is given by
2  2
4 π Z e2 m c2

θp0
∝ sin2 (1539)
V | p − p0 |2 Ep 2
It is seen that the probability for helicity flip scattering vanishes in the ultra-
relativistic limit. Also, in the non-relativistic limit, a static charge cannot flip
the spin. Therefore, in the non-relativistic limit, if one expresses the spin eigen-
state as a linear superposition of the final helicity eigenstates
θp0 0 θp0 0
χ+ = cos χ − sin χ (1540)
2 + 2 −
one is lead to expect that the relative probability of helicity flip to non-helicity
flip will be governed by a factor of
θp0
tan2 (1541)
2
which agrees with the above matrix elements evaluated in the non-relativistic
limit. The cross-section for non-flip scattering is determined as
2
2 Z e2 E p
  
dσ θp0
= cos2 (1542)
dΩ0 +,+ 2 2
4 c p sin 22 θp0 2

261
whereas the cross-section for spin flip scattering is given by
2
2 Z e2 m c 2
  
dσ θp0
= sin2 (1543)
dΩ0 +,− 4 c2 p2
θ 0
sin2 2p 2

The Born approximation to the total cross-section for scattering of polarized

electrons, in which the final polarization is not measured, is given by
     
dσ dσ dσ
= +
dΩ0 + dΩ0 +,+ dΩ0 +,−
2  2
2 Z e2 Ep m c2
  
2 θp0 2 θp0
= θ 0
cos + sin
4 c2 p2 sin2 2p 2 Ep 2
2 2
2 Z e2 Ep
     
pc θp0
= 1 − sin2 (1544)
2 2 2 θp0 E 2
4 c p sin 2 p

which is the same as the cross-section as calculated for unpolarized electrons.

The degree of polarization of the scattered beam is given
θp0 θp0
Ep2 cos2 − m2 c4 sin2
 
2 2
P (θp0 ) = θ 0 θ 0
(1545)
Ep2 cos2 2p + m2 c4 sin2 2p

If the initial beam of electrons is unpolarized, the scattered electrons would be

observed to be partially polarized, where the net polarization is in the plane
perpendicular to the scattering plane. However, the polarization is due to pro-
cesses of higher-order than the Born approximation and is governed by the factor
2
( Zh̄ ec ).

12.6 The Non-Relativistic Limit

The non-relativistic limit of the Dirac equation should reduce to the Schrödinger
equation. As shall be seen, the appropriate Schrödinger equation for a particle
with positive-energy is modified due to the existence of spin. The non-relativistic
limit is described by the Pauli equation102 .

The Dirac equation can be written as

   
h̄ ∂ q q
i − A0 ψ = α . ( p̂ − A )+ β m c ψ (1546)
c ∂t c c

The equation can be written in 2 × 2 block diagonal form, if the wave function
is expressed in the form of two two-component spinors. We shall mainly focus
on the positive-energy solutions and recognize that, in the non-relativistic limit,
the largest component of the wave function is φA and the largest term in the
102 W. Pauli, Z, Phys. 44, 601 (1927).

262
energy is the rest mass energy m c2 . Therefore, the spinor wave function will
be expressed as
m c2
 A   
φ
ψ = exp − i t (1547)
φB h̄
The above form explicitly displays the rest-mass energy of the positive-energy
solution of the Dirac equation. Hence, the Dirac equation takes the form
c σ . ( p̂ − qc A ) φB
   A   
∂ φ
i h̄ − q A0 =
∂t φB c σ . ( p̂ − qc A ) φA
 
2 0
−2mc (1548)
φB
where the rest mass has been eliminated from the equation for the large com-
ponent φA of the positive-energy solution. Since the kinetic energy and the
potential energy are assumed to be smaller than the rest mass energy, the equa-
tion for the small component
   
∂ B q
i h̄ − q A0 φ = c σ . p̂ − A φA − 2 m c2 φB (1549)
∂t c
can be expressed as
 
1 q
φB = σ. p̂ − A φA (1550)
2mc c
Substituting the expression for the small component into the equation for the
large component, hence eliminating φB , one finds the equation
     2
∂ A 1 q
i h̄ − q A0 φ = σ . p̂ − A φA (1551)
∂t 2m c
which is the Pauli equation. The equation can be simplified by expanding the
terms involving the Pauli spin matrices. The Pauli identity can be used to
obtain
   2  2    
q q q q
σ . p̂ − A = I p̂ − A + iσ. p̂ − A ∧ p̂ − A
c c c c
 2  
q q h̄
= I p̂ − A − σ. ∇ ∧ A (1552)
c c
where the last term originates from the non-commutativity of the components
of p̂ and A. Since the magnetic field B is given by

B = ∇ ∧ A (1553)

the Pauli equation can be expressed as

 2
∂ A 1 q q h̄
i h̄ φ = p̂ − A φA + q A0 φA − σ . B φA (1554)
∂t 2m c 2mc

263
The Pauli equation103 is the non-relativistic limit of the Dirac equation. It rep-
resents the Schrödinger equation for a charged particle with spin one-half. The
two components of the spinor φA in the Pauli equation represent the internal
spin of the electron. The last term represents the anomalous Zeeman interaction
between the magnetic field and the electron’s spin.

The other contribution to the Zeeman interaction originates with the elec-
tron’s orbital angular momentum L. The ordinary Zeeman interaction occurs
between the constant magnetic field B and the orbital angular momentum and
originates from the gauge-invariant term in the Hamiltonian
 2  2
1 q 1 q
p̂ − A = p̂ − B ∧ r (1555)
2m c 2m 2c

where the vector potential has been expressed in terms of the uniform magnetic
field via
1
A = B ∧ r (1556)
2
The expression for the energy term can be further simplified to
2
p̂2
  
1 q q
p̂ − A = − p̂ . (B ∧ r) + (B ∧ r) . p̂
2m c 2m 4mc
2
q
+ A2
2 m c2
p̂2 q2
 
q
= − (B ∧ r) . p̂ + A2
2m 2mc 2 m c2
p̂2 q2
 
q
= − B . (r ∧ p̂) + A2
2m 2mc 2 m c2
p̂2 q2
 
q
= − B . L̂ + A2 (1557)
2m 2mc 2 m c2

In obtaining the second line, the i-th component of p̂ has been commuted with
the i-th component of ( B ∧ r ). In obtaining the third line, the (cyclic) vector
identity
(A ∧ B).C = (B ∧ C ).A (1558)
has been used. The first term in eqn(1557) represents the usual non-relativistic
expression for the kinetic energy of the electrons, the second term represents
the ordinary Zeeman interaction which originates from the paramagnetic inter-
action. The last term represents the diamagnetic interaction.

The total Zeeman interaction is the energy of the total magnetic moment M
in the field B
ĤZeeman = − M . B (1559)
103 W. Pauli, Z, Phys. 44, 601 (1927).

264
The Dirac equation results in the Zeeman interaction of the form
 
q
ĤZeeman = − B . L̂ + h̄ σ
2mc
 
q
= − B . L̂ + 2 S (1560)
2mc
where the spin angular momentum S has been identified as
h̄
S = σ (1561)
2
It is seen that both the spin angular momentum and the orbital angular momen-
tum of the charged particle interacts with the magnetic field, therefore, both
contribute to the magnetic moment. However, it is noted that the magnetic
moment can be written in the form
 
q
M = L̂ + g S (1562)
2mc

where the magnitude of the magnetic moment is determined by the factor 2 qmh̄ c
which is the Bohr magneton. The Dirac equation shows that the spin angular
momentum couples with a different strength to orbital angular momentum, and
the relative coupling strength g (the gyromagnetic ratio) is given by g = 2.
The existence of spin and the value of 2 for the gyromagnetic ratio were the
first successes of Dirac’s theory. Quantum Electrodynamics104 yields a small
correction to the gyromagnetic ratio of
  2 
1 e
g = 2 1 + + ... (1563)
2 π h̄ c
which has been experimentally verified to incredible precision105 . Using the fea-
tures associated with spin, Dirac’s theory correctly described the fine structure
of the Hydrogen atom. The second success of the Dirac equation followed Dirac’s
physical interpretation of the negative-energy states in terms of anti-particles106 .
The second round of success came with the discovery of the positron by Ander-
son107 .

Exercise:

The Dirac equation can be phenomenologically modified to describe particles

with anomalous magnetic moments. The Dirac equation is modified to
 
q q h̄ ˆ
i h̄ γ µ ( ∂µ + i Aµ ) + κ σ µ,ν
F µ,ν − m c I ψ = 0 (1564)
h̄ c 4 m c2
104 J. S. Schwinger, Phys. Rev. 73, 416 (1948).
105 H. M. Foley and P. Kusch, Phys. Rev. 73, 412 (1948).
R. S. Van Dyck Jr., P. B. Schwinberg and H. G. Dehmelt, Phys. Rev. Lett. 59, 26 (1987).
106 P. A. M. Dirac, Proc. Roy. Soc. A 126, 360 (1930).
107 C. D. Anderson, Phys. Rev. 43, 491 (1933).

265
Show that the modified equation is Lorentz covariant and that the Hamiltonian
is Hermitean. Also derive the corrections to the magnetic moment due to the
spin by examining the non-relativistic limit.

12.7 Conservation of Angular Momentum

The law of conservation of angular momentum will now be examined. For a
relativistic electron the orbital angular momentum and the spin angular mo-
mentum are not separately conserved. However, the total angular momentum
which is the sum of the orbital angular momentum and spin angular momentum
is conserved.

The orbital angular momentum L̂ defined by

L̂ = r ∧ p̂ (1565)

is not conserved for a spherically symmetric potential. The Dirac Hamiltonian

is given by
Ĥ = c α . p̂ + β m c2 + Iˆ V (r) (1566)
The matrices shall be expressed in a 2 × 2 block diagonal form. Therefore, the
identity matrix is written as
 
I 0
Iˆ = (1567)
0 I

and  
I 0
β = (1568)
0 −I
Finally, the α matrices are of off-diagonal form
 
0 σ
α =
σ 0
 
0 I
= σ̂ (1569)
I 0

where σ̂ is the 2 × 2 block-diagonal Pauli spin matrix. The rate of change of

orbital angular momentum is given by the Heisenberg equation of motion
∂
i h̄ L̂ = [ L̂ , Ĥ ] (1570)
∂t
The orbital angular momentum operator commutes with the mass term and
with the spherically symmetric potential V (r). The orbital angular momentum
does not commute with the momentum. Thus,
∂
i h̄ L̂ = c [ L̂ , α . p̂ ] (1571)
∂t

266
Hence, the Heisenberg equation of motion can be expressed in the form
 
∂ 0 I
i h̄ L̂ = c [ L̂ , σ̂ . p̂ ]
∂t I 0
 
0 I
= −c σ̂ . [ p̂ , L̂ ] (1572)
I 0

However, the components of the orbital angular momentum L̂(i) and momenta
p(j) satisfy the commutation relations
X
[ L̂(i) , p(j) ] = i h̄ ξ i,j,k p(k) (1573)
k

Therefore, one finds

 
∂ 0 I
i h̄ L̂ = i h̄ c ( σ̂ ∧ p̂ ) (1574)
∂t I 0

which shows that orbital angular momentum is not conserved for a relativistic
electron with a central potential.

The spin angular momentum is also not conserved. This can be seen by
examining the Heisenberg equation of motion for the Pauli spin operator
∂
i h̄ σ̂ = [ σ̂ , Ĥ ] (1575)
∂t

The spin operator commutes with Iˆ and β but does not commute with the α
matrices. Hence,
∂
i h̄ σ̂ = c [ σ̂ , α . p̂ ]
∂t  
0 I
= c [ σ̂ , σ̂ ] . p̂ (1576)
I 0

The components of the Pauli spin operators satisfy the commutation relations
X
[ σ (i) , σ (j) ] = 2 i ξ i,j,k σ (k) (1577)
k

which, clearly, have a similar form to the commutation relations for the orbital
angular momentum. Hence, spin angular momentum is not conserved since
 
∂ 0 I
i h̄ σ̂ = − 2 i c ( σ̂ ∧ p̂ ) (1578)
∂t I 0

267
 The total angular momentum Jˆ is defined via

Jˆ = L̂ + Ŝ
h̄
= L̂ + σ̂ (1579)
2
The total angular momentum is conserved since
∂ ˆ ∂ ∂
i h̄ J = i h̄ L̂ + i h̄ Ŝ
∂t ∂t  ∂t
   
0 I 0 I
= i h̄ c ( σ̂ ∧ p̂ ) − ( σ̂ ∧ p̂ )
I 0 I 0
= 0 (1580)

which follows from combining eqn(1574) and eqn(1578). This confirms the in-
terpretation of the quantity Ŝ defined by
h̄
Ŝ = σ̂ (1581)
2
as the spin angular momentum of the electron.

12.8 Conservation of Parity

Dirac was very conscious that his book “Principles of Quantum Mechanics”
never contained any mention of parity. It seems that he had questioned the
requirement of parity invariance108 since biological systems are not parity in-
variant. Dirac’s viewpoint was vindicated by the discovery that the weak inter-
action violates parity.

The parity transform P acting on the coordinates (t, r) has the effect

P (t, r) → (t0 , r0 ) = (t, −r) (1582)

which is an inversion of the spatial coordinates. Thus, the parity reverse the
space-like components of vectors, so the effects of the parity operation on the
position and momentum vectors are given by

P̂ r P̂ −1 = −r
P̂ p P̂ −1 = −p (1583)

However, the effect of the parity transform on pseudo-vectors such as orbital

angular momentum L = r ∧ p is such that

P̂ L P̂ −1 = L (1584)
108 The question of parity conservation in weak interactions was raised subsequently by T. D.

Lee and C. N. Yang [T. D. Lee and C. N. Yang, Phys. Rev. 104, 254 (1956).]

268
which is unchanged. This implies that spin angular momentum should also be
invariant under the parity transform
P̂ σ P̂ −1 = σ (1585)
If the Hamiltonian Ĥ is invariant under a parity transform, one requires that
Ĥ = P̂ Ĥ P̂ −1 (1586)
Imposing parity invariance of the Dirac Hamiltonian
Ĥ = c α . p̂ + β m c2 + Iˆ V (r) (1587)
yields a condition on the potential
V (r) = V (−r) (1588)
and also to conditions on the Dirac matrices
P̂ α P̂ −1 = −α
P̂ β P̂ −1 = β (1589)
The condition on the potential is the familiar condition for parity invariance
in classical mechanics. In the standard representation, in 2 × 2 block diagonal
form, the requirement of parity invariance on the Dirac matrices become the
matrix equations
   
0 σ 0 σ
P̂ P̂ −1 = −
σ 0 σ 0
   
I 0 I 0
P̂ P̂ −1 = (1590)
0 −I 0 −I
The above equation shows that, in the standard representation, the parity op-
erator can be uniquely factorized as
 
I 0
P̂ = P̂ (1591)
0 −I

where the operator P̂ only acts on the coordinates r. The presence of the matrix
in the parity operation on the Dirac spinor should be compared with the effect
of the parity operator on the four-vector potential of Electrodynamics Aµ (r)
which is given by the product of spatial inversion and a matrix operation
P̂ Aµ (r, t) = γ µ ν P̂ Aν (r, t)
= γ µ ν Aν (−r, t) (1592)
where the matrix γ µ ν given by
 
1 0 0 0
 0 −1 0 0 
γµν = 
 0 0 −1
 (1593)
0 
0 0 0 −1

269
reverses the direction of the spatial components of the vector field.

The effect of the parity operator on the Dirac four-component spinor wave
function can be computed from
 A 
φ (t, r)
P̂ ψ(t, r) = P̂
φB (t, r)
  A 
I 0 φ (t, −r)
=
0 −I φB (t, −r)
φA (t, −r)
 
= (1594)
− φB (t, −r)

Hence, in the standard representation, the parity operator changes the relative
sign of the two two-component spinors. Due to the presence of the term − I in
the lower diagonal block of the parity matrix, the lower two-component spinor
φB in the Dirac spinor is said to have a negative intrinsic parity.

The parity eigenstates satisfy the eigenvalue equation

P̂ ψ = ηp ψ (1595)

with eigenvalues ηp = ±1, since P̂ 2 = I. ˆ The application of the parity

operator on the Dirac spinor leads to the equation

P̂ φA ηp φA
   
= (1596)
−P̂ φB ηp φB

Hence, the two-component spinors φA (r) and φB (r) have opposite parities under
spatial inversion

P̂ φA (r) = ηp φA (r)
P̂ φB (r) = − ηp φB (r) (1597)

In polar coordinates, the spatial part of the parity operation P̂ is equivalent to

a reflection
θ → π − θ (1598)
followed by a rotation
ϕ → ϕ + π (1599)
which has the effect that

sin θ → sin θ
cos θ → − cos θ
   
m
exp i m ϕ → ( − 1) exp i m ϕ (1600)

270
Hence, the spherical harmonics with m = l
r
( − 1 )l
 
l 2l + 1 l
Yl (θ, ϕ) = sin θ exp i l ϕ (1601)
2l l! 4π

are eigenstates of the parity operator and have parity eigenvalues of (−1)l . The
lowering operator L̂− , defined via
  
− ∂ ∂
L̂ = − h̄ exp − i ϕ − i cot θ (1602)
∂θ ∂ϕ
is invariant under the parity transformation

P̂ L̂− P̂ −1 = L̂− (1603)

Therefore, on repeatedly operating on Yll (θ, ϕ) with the lowering operator L̂−
(l − m) times, one finds that under the parity transformation

Yml (θ, ϕ) → ( − 1 )l Yml (θ, ϕ) (1604)

which shows that all states with a definite magnitude of the orbital angular mo-
mentum l are eigenstates of the parity operator and have the same eigenvalue.

Exercise:

Show that under a parity transformation the positive-energy solution for the
+
free Dirac particle ψk,σ (x) transforms as

+ +
P̂ ψk,σ (x) = ψ−k,σ (x) (1605)
−
while the negative-energy solutions ψk,σ (x) transform as

− −
P̂ ψk,σ (x) = − ψ−k,σ (x) (1606)

Hence, the parity operation reverses the momentum and keeps the spin invari-
ant for the positive-energy and negative-energy solutions solution. The extra
negative sign implies that the negative-energy solution has opposite intrinsic
parity to the positive-energy solution.

Exercise:

Consider the parity transform as an example of an improper Lorentz trans-

formation Λ, for which det | Λ | = − 1. If the Lorentz transform is given
by
xµ0 = Λµ ν xν (1607)
the spinor wave function transforms via

ψ 0 (x0 ) = R̂(Λ) ψ(x) (1608)

271
where R̂(Λ) “rotates” the spinor. The covariant condition for the Dirac equation
is
R̂−1 (Λ) γ µ R̂(Λ) = Λµ ν γ ν (1609)
For a parity transformation, one has
xµ0 = xµ (1610)
µ
since the spatial components of x change sign. Hence, for a parity transforma-
tion, the transformation matrix is determined as
Λµ ν = gµ,ν (1611)
which is an improper Lorentz transformation since
det | g | = − 1 (1612)
Therefore, the covariant condition reduces to
R̂−1 (Λ) γ µ R̂(Λ) = gµ,ν γ ν (1613)
Solve for the matrix R̂(Λ) which shuffles the components of the Dirac spinor.

12.9 Bi-linear Covariants

Under an Lorentz transformation
xµ0 = Λµ ν xν (1614)
0
(where Λ 0 > 0 for an orthochronous transformation), the Dirac spinor ψ
transforms according to
ψ 0 (x0 ) = R̂(Λ) ψ(x) (1615)
and the condition that the Dirac equation is covariant under the orthochronous
Lorentz transformation is
R̂−1 (Λ) γ µ R̂(Λ) = Λµ ν γ ν (1616)
From the transformational properties of the Dirac spinors, together with the
identity
γ (0) R̂† (Λ) γ (0) = R̂−1 (Λ) (1617)
one can find the transformational properties of quantities that are bi-linear in
the Dirac spinors.
†
Thus, for example, the bi-linear quantity ψ ψ transforms according to
†0
ψ ψ0 = ψ †0 γ (0) ψ 0
= ψ † R̂† (Λ) γ (0) R̂(Λ) ψ
= ψ † ( γ (0) )2 R̂† (Λ) γ (0) R̂(Λ) ψ
= ψ † γ (0) R̂−1 (Λ) R̂(Λ) ψ
†
= ψ ψ (1618)

272
 †
Table 7: The sixteen bi-linear covariants ψ Q̂ ψ for the Dirac equation.

Quantity Bi-linear Transformed Number of

Covariant Matrix Matrices

†
ψ Q̂ ψ R̂−1 (Λ) Q̂ R̂(Λ)

†
Scalar ψ Iˆ ψ Iˆ 1

†
Vector ψ γµ ψ Λµ ν γ ν 4

†
Anti-symmetric Tensor ψ σ µ,ν ψ Λµ ρ Λν τ σ ρ,τ 6

†
Pseudo-scalar ψ γ (4) ψ det | Λ | γ (4) 1

†
Axial-Vector ψ γ (4) γ µ ψ det | Λ | Λµ ν γ (4) γ ν 4

where a factor of ( γ (0) )2 = Iˆ has been used in the third line and the identity
†
has been used in the fourth. Thus, one finds that ψ ψ transforms like a scalar.
†
Likewise, one can show that the bi-linear quantities ψ γ µ ψ transform like
the components of a four-vector. That is
†0
ψ γ µ ψ0 = ψ †0 γ (0) γ µ ψ 0
= ψ † R̂† (Λ) γ (0) γ µ R̂(Λ) ψ
= ψ † ( γ (0) )2 R̂† (Λ) γ (0) γ µ R̂(Λ) ψ
= ψ † γ (0) R̂−1 (Λ) γ µ R̂(Λ) ψ
†
= ψ R̂−1 (Λ) γ µ R̂(Λ) ψ
†
= Λµ ν ψ γ ν ψ (1619)
where the covariant condition has been used in obtaining the last line. Since
†
this relation holds for Lorentz boosts, rotations and spatial inversions, ψ γ µ ψ
is a four-vector.

The anti-symmetric quantity σ µ,ν defined as

i
σ µ,ν = [ γµ , γν ] (1620)
2
†
can be used to form a bi-linear quantity ψ σ µ,ν ψ. This bi-linear quantity

273
transforms like a second-rank anti-symmetric tensor, since
†0
ψ σ µ,ν ψ 0 = ψ †0 γ (0) σ µ,ν ψ 0
= ψ † R̂† (Λ) γ (0) σ µ,ν R̂(Λ) ψ
= ψ † ( γ (0) )2 R̂† (Λ) γ (0) σ µ,ν R̂(Λ) ψ
= ψ † γ (0) R̂−1 (Λ) σ µ,ν R̂(Λ) ψ
†
= ψ R̂−1 (Λ) σ µ,ν R̂(Λ) ψ (1621)

For µ 6= ν, the antisymmetric quantity σ µ,ν can be written as

σ µ,ν = i γ µ γ ν (1622)

Therefore, one may re-express the bi-linear quantity as

†0 †
ψ σ µ,ν ψ 0 = i ψ R̂−1 (Λ) γ µ γ ν R̂(Λ) ψ
†
= i ψ R̂−1 (Λ) γ µ R̂(Λ) R̂−1 (Λ) γ ν R̂(Λ) ψ
†
= i ψ Λµ ρ γ ρ Λν τ γ τ ψ
†
= Λµ ρ Λν τ ψ σ ρ,τ ψ (1623)

where we have inserted a factor of Iˆ = R̂(Λ) R̂−1 (Λ) in the second line, and
used the covariant condition (twice) in the third line. Hence, the bi-linear quan-
†
tity ψ σ µ,ν ψ transforms like an anti-symmetric second-rank tensor.

One can define a quantity γ (4) in terms of a product of all the γ-matrices

γ (4) = i γ (0) γ (1) γ (2) γ (3) (1624)

It is easily verified that γ (4) anti-commutes with all the γ µ ,

{ γ µ , γ (4) }+ = 0 (1625)

Furthermore, one has

( γ (4) )2 = Iˆ (1626)
(4)
In the standard representation of the Dirac matrices, the matrix γ has the
two by two block diagonal form
 
(4) 0 I
γ = (1627)
I 0
†
The quantity γ (4) can be used to construct a bi-linear covariant quantity ψ γ (4) ψ.
Under an orthochronous Lorentz transformation, the bi-linear quantity trans-
form according to
†0
ψ γ (4) ψ 0 = ψ †0 γ (0) γ (4) ψ 0

274
 = ψ † R̂† (Λ) γ (0) γ (4) R̂(Λ) ψ
= ψ † ( γ (0) )2 R̂† (Λ) γ (0) γ (4) R̂(Λ) ψ
†
= ψ γ (0) R̂† (Λ) γ (0) γ (4) R̂(Λ) ψ
†
= ψ R̂−1 (Λ) γ (4) R̂(Λ) ψ (1628)

A proper Lorentz transformation, such as a boost and a rotation, are generated

by the quantities σ µ,ν which involves the anti-symmetrized product of the two
Dirac matrices γ µ and γ ν . Since the matrices γ µ and γ ν individual anti-commute
with γ (4) , their product commutes with γ (4) . Hence, one can commute γ (4) and
R̂(Λ). Therefore, the bi-linear quantity transforms as
† †
ψ 0 (x0 ) γ (4) ψ 0 (x0 ) = ψ (x) γ (4) ψ(x) (1629)

which behaves like a scalar under a proper orthochronous Lorentz transforma-

tion for which det | Λ | = 1. However, for an inversion where det | Λ | = − 1,
one has R̂(P) = γ (0) which anti-commutes with γ (4) . Hence, for an inversion,
one has
† †
ψ 0 (x0 ) γ (4) ψ 0 (x0 ) = − ψ (x) γ (4) ψ(x) (1630)
so the quantity changes sign. In general, for an orthochronous transformation
one can show that

R̂−1 (Λ) γ (4) R̂(Λ) = det | Λ | γ (4) (1631)

so one has
† †
ψ 0 (x0 ) γ (4) ψ 0 (x0 ) = det | Λ | ψ (x) γ (4) ψ(x) (1632)
†
Therefore, the quantity ψ γ (4) ψ transforms as a pseudo-scalar.
†
One can also define the bi-linear axial-vector ψ γ (4) γ µ ψ. From consid-
erations similar to those used previously, one can show that these quantities
transform according to
† †
ψ 0 (x0 ) γ (4) γ µ ψ 0 (x0 ) = det | Λ | Λµ ν ψ (x) γ (4) γ ν ψ(x) (1633)
†
Hence, ψ γ (4) γ µ ψ transforms like a four-vector under proper orthochronous
Lorentz transformations. However, the spatial components do not change sign
†
under an inversion, but the time components do change sign. Therefore, ψ γ (4) γ µ ψ
transforms like an axial-vector.

Exercise:

Show that a modified Dirac equation described by

 
µ q κ q h̄ µ,ν (4) ˆ
i h̄ γ ( ∂µ + i Aµ ) − i σ γ Fµ,ν − m c I ψ = 0 (1634)
h̄ c 4 m c2

275
is covariant under proper Lorentz transformations, but is not covariant under
improper transformations.

Show, by considering the non-relativistic limit, that the above equation de-
scribes an electron with an electric dipole moment. Determine an expression for
the electric dipole moment.

12.10 The Spherically Symmetric Dirac Equation

The Dirac Hamiltonian for a (electrostatic) spherically symmetric potential is
given by
Ĥ = c α . p̂ + β m c2 + Iˆ V (r) (1635)

The angular momentum operator Jˆ and the parity operator P̂ commute with
the Hamiltonian Ĥ. Therefore, one can find simultaneous eigenstates of the
three operators Ĥ, Jˆ2 , Jˆz and P̂. The energy eigenstates satisfy the equation
 
c α . p̂ + β m c2 + Iˆ V (r) ψ = E ψ (1636)

On writing the four-component spinor in terms of the two two-component

spinors φA and φB the energy eigenvalue equation reduces to the set of cou-
pled equations

( E − V (r) − m c2 ) φA (r) = c ( σ . p̂ ) φB (r)

( E − V (r) + m c2 ) φB (r) = c ( σ . p̂ ) φA (r) (1637)

In spherical polar coordinates, the operator ( σ . p̂ ) can be expressed as

 
cos θ sin θ exp[−iϕ] ∂
( σ . p̂ ) = − i h̄
sin θ exp[+iϕ] − cos θ ∂r
 
i h̄ − sin θ cos θ exp[−iϕ] ∂
−
r cos θ exp[+iϕ] sin θ ∂θ
 
i h̄ 0 − i exp[−iϕ] ∂
−
r sin θ i exp[+iϕ] 0 ∂ϕ
(1638)

which has a quite complicated structure. For future reference, it shall be noted
that the matrix part of the coefficient of the partial derivative w.r.t. r is simply
equal to  
r.σ
(1639)
r
which is independent of the radial coordinate r. The operator ( σ . p̂ ) can be
cast in a more convenient form through the repeated use of the Pauli identity.

276
First, the 2 × 2 unit matrix can be written as
 2
r.σ
I = (1640)
r

since different Pauli spin matrices anti-commute

{ σ (i) , σ (j) }+ = 2 δ i,j Iˆ (1641)

and are their own inverses. Therefore, one can express the operator ( σ . p̂ ) as
 2
r.σ
( σ . p̂ ) = ( σ . p̂ )
r
 
r.σ
= ( r . σ ) ( σ . p̂ )
r2
  
r.σ
= r . p̂ + i σ . ( r ∧ p̂ )
r2
  
r.σ ∂
= − i h̄ r + i σ . L̂
r2 ∂r
  
r.σ ∂ 2i
= − i h̄ r + S . L̂ (1642)
r2 ∂r h̄

where the Pauli identity has been used in going between the second and third
lines. Therefore, the two-component spinors satisfy the set of coupled equations
  
2 A r.σ ∂ 2i
( E − V (r) − m c ) φ (r) = c − i h̄ r + S . L̂ φB (r)
r2 ∂r h̄
  
2 B r.σ ∂ 2i
( E − V (r) + m c ) φ (r) = c − i h̄ r + S . L̂ φA (r)
r2 ∂r h̄
(1643)

It is seen that, due to the effect of special relativity, the Dirac equation results
in the coupling of the spin and the orbital angular momentum.

Two-Component Spinor Spherical Harmonics

The angular dependence of the two-component wave functions φA (r) and

B
φ (r) are determined by the eigenvalue equations for the magnitude and the
z-components of the total angular momentum

J = L + S (1644)

Thus, the two-component spinor eigenstates of total angular momentum Ωlj,jz (θ, ϕ)
which describes the angular dependence, are formed by combining states of or-
bital angular momentum l, represented by Yml (θ, ϕ), and the spin eigenfunction

277
Table 8: The Clebsch-Gordon Coefficients for adding orbital angular momentum
(l, m) with spin quantum numbers ( 21 , sz ) to yield a state with total angular
momentum quantum numbers (j, jz ). The allowed values of m are given by
jz = m + sz .

sz = + 12 sz = − 12

q 1
q 1
1 l + jz + l − jz +
j =l+ 2 2 l + 1
2
2 l + 1
2

q 1
q
1 l − jz + l +jz + 12
j =l− 2 - 2 l + 1
2
2 l + 1

χ± . On combining states with orbital angular momentum l and spin s = 12 , one

finds states with total angular momentum which satisfy
1 1
l + ≥ j ≥ l − (1645)
2 2
Thus, it is found that the possible eigenstates correspond to j = l + 21 and
j = l − 12 . Furthermore, the corresponding eigenfunctions are expressed as
s s
1
l + 2 + jz l l + 12 − jz l
Ωll+ 1 ,jz (θ, ϕ) = Yjz − 1 (θ, ϕ) χ+ + Yjz + 1 (θ, ϕ) χ−
2 2l + 1 2 2l + 1 2
s s
l + 12 − jz l l + 12 + jz l
Ωll− 1 ,jz (θ, ϕ) = − Yjz − 1 (θ, ϕ) χ+ + Yjz + 1 (θ, ϕ) χ−
2 2l + 1 2 2l + 1 2

(1646)
where the coefficients are identified with the Clebsch-Gordon coefficients given
in Table(8). The functions Ωlj,jz (θ, ϕ) are the analogue of the spherical harmon-
ics Yml (θ, ϕ) in relativistic problems where spin and orbital angular momentum
are coupled.

However, since orbital angular momentum is not a good quantum number,

the angular dependence of the eigenstates of the Dirac Hamiltonian can be
expressed as a linear superposition of states with different values of the orbital
angular momentum l. For a fixed value of j, one finds that the possible values
of the orbital angular momentum l are determined by
1
j = l +
2
0 1
j = l − (1647)
2

278
where l0 = l + 1. The appropriate two-component spinor angular momentum
eigenstate with quantum numbers (j, jz ) found by combining a spin one-half
and orbital angular momentum l0 = (l + 1) is given by
s s
l+1 l + 32 − jz l+1 l + 32 + jz l+1
Ωl+ 1 ,j (θ, ϕ) = − Yj − 1 (θ, ϕ) χ+ + Yj + 1 (θ, ϕ) χ−
2 z 2l + 3 z 2 2l + 3 z 2

(1648)
0
As shall be seen later, the two-component spinors Ωll0 − 1 ,jz (θ, ϕ) and Ωll+ 1 ,jz (θ, ϕ)
2 2
have opposite parities. In fact, the two-component spinors generated by angular
momentum l and l0 = (l + 1) are related by the action of the pseudo-scalar
   
r.σ cos θ sin θ exp[−iϕ]
= (1649)
r sin θ exp[+iϕ] − cos θ

which changes sign under a parity transformation, (θ, ϕ) → (π − θ, ϕ + π). The

explicit relationship is given by
 
r.σ
Ωlj,jz (θ, ϕ) = − Ωl+1
j,jz (θ, ϕ) (1650)
r

as can be shown by examination of Table(1). Likewise, on using the identity

 2
r.σ
= I (1651)
r

one finds that the inverse relationship between the two-component spinors is
also given by  
r.σ
Ωl+1 l
j,jz (θ, ϕ) = − Ωj,jz (θ, ϕ) (1652)
r
Therefore, one concludes that the two angular momentum eigenstates have dif-
ferent properties under the spatial inversion transformation r → −r.

——————————————————————————————————

Mathematical Interlude:
The Action of the Operator ( r̂ . σ ) on the Spinor Spherical Harmon-
j± 1
ics Ωj,jz2 (θ, ϕ).

Here, it will be argued that the spinor spherical harmonics satisfy the equa-
tions
 
r.σ j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = − Ωj,jz2 (θ, ϕ)
r
 
r.σ j− 1 j+ 1
Ωj,jz2 (θ, ϕ) = − Ωj,jz2 (θ, ϕ) (1653)
r

279
 The components of the total angular momentum
Jˆ(i) = L̂(i) + Ŝ (i) (1654)
commute with ( r . Ŝ ). That is
[ Jˆ(i) , ( r . Ŝ ) ] = 0 (1655)
The complete proof of this statement immediately follows from the proof of the
relation for any one component Jˆ(i) , since ( r . Ŝ ) is spherically symmetric.
Thus, for i = 1, one has
[ Jˆ(1) , ( r . Ŝ ) ] = [ L̂(1) + Ŝ (1) , x(1) Ŝ (1) + x(2) Ŝ (2) + x(3) Ŝ (3) ]
= Ŝ (2) [ L̂(1) , x(2) ] + Ŝ (3) [ L̂(1) , x(3) ]
+ x(2) [ Ŝ (1) , Ŝ (2) ] + x(3) [ Ŝ (1) , Ŝ (3) ] (1656)
Using the commutation relations
[ Ŝ (i) , Ŝ (j) ] = i h̄ εi,j,k Ŝ (k) (1657)
and
[ L̂(i) , x(j) ] = i h̄ εi,j,k x(k) (1658)
one finds that
 
[ Jˆ(1) , ( r . Ŝ ) ] = i h̄ Ŝ (2) x(3) − Ŝ (3) x(2) + x(2) Ŝ (3) − x(3) Ŝ (2)

= 0 (1659)
which was to be shown. From repeated use of the above commutation relations
which involve the components Jˆ(i) , it immediately follows that
2
[ Ĵ , ( r . Ŝ ) ] = 0 (1660)
1 2
j±
Thus, since Ωj,jz2 is a simultaneous eigenstate of Ĵ and Jˆ(3) and because these
j± 1
operators commute with ( r . Ŝ ), then ( r . Ŝ ) Ωj,jz2 is also a simultaneous
eigenstate with eigenvalues (j, jz ).
1 2
j±
Since the states ( r . Ŝ ) Ωj,jz2 are simultaneous eigenstates of Ĵ and Jˆ(3)
with eigenvalues (j, jz ), and because this subspace is spanned by the basis com-
j± 1
posed of the two states Ωj,jz2 (θ, ϕ), the transformed states can be decomposed
as
 
r.σ j+ 1 j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = C++ (j, jz ) Ωj,jz2 (θ, ϕ) + C+− (j, jz ) Ωj,jz2 (θ, ϕ)
r
 
r.σ j− 1 j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = C−+ (j, jz ) Ωj,jz2 (θ, ϕ) + C−− (j, jz ) Ωj,jz2 (θ, ϕ)
r
(1661)

280
where the coefficients C±,± (j, jz ) will be determined below.

First, we shall show that the coefficients C±,± (j, jz ) are independent of jz .
This follows as Jˆ± commutes with ( r . Ŝ ) since all the components Jˆ(i) com-
mute with ( r . Ŝ ). Thus, one has
   
ˆ ± r.σ j+ 12 r . σ ˆ± j+ 12
J Ωj,jz (θ, ϕ) = J Ωj,jz (θ, ϕ) (1662)
r r
and
 
r.σ j+ 1 j+ 1 j− 1
Jˆ± Ωj,jz2 (θ, ϕ) = C++ (j, jz ) Jˆ± Ωj,jz2 (θ, ϕ) + C+− (j, jz ) Jˆ± Ωj,jz2 (θ, ϕ)
r
 
r . σ ˆ± j+ 12 j+ 1 j− 1
J Ωj,jz (θ, ϕ) = C++ (j, jz ± 1) Jˆ± Ωj,jz2 (θ, ϕ) + C+− (j, jz ± 1) Jˆ± Ωj,jz2 (θ, ϕ)
r
(1663)
Hence, on comparing the linearly-independent terms on the left-hand sides, one
concludes that
C++ (j, jz ± 1) = C++ (j, jz )
C+− (j, jz ± 1) = C+− (j, jz ) (1664)
etc. Therefore, the coefficients C±,± (j, jz ) are independent of the value of jz .
Henceforth, we shall omit the index jz in C±,± (j, jz ).

From considerations of parity, it can be determined that C++ (j) = C−− (j) =
0. Under the parity transformation r → − r, one has
j± 1 1 j± 1
Ωj,jz2 (θ, ϕ) → ( − 1 )j± 2 Ωj,jz2 (θ, ϕ) (1665)

which follows from the properties of the spherical harmonics Yml (θ, ϕ) under the
parity transformation. Also one has
   
r.σ r.σ
→ − (1666)
r r
under the parity transform. Thus, after the parity transform, one finds that the
transformed states have the decompositions
 
r.σ j+ 1 j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = − C++ (j) Ωj,jz2 (θ, ϕ) + C+− (j) Ωj,jz2 (θ, ϕ)
r
 
r.σ j− 1 j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = C−+ (j) Ωj,jz2 (θ, ϕ) − C−− (j) Ωj,jz2 (θ, ϕ)
r
(1667)
which by comparison with eqn(1661) leads to the identification
C++ (j) = C−− (j) = 0 (1668)

281
Therefore, recalling that the coefficients are independent of jz , one can express
the effect of the operator on the spinor spherical harmonics as
 
r.σ j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = C+− (j) Ωj,jz2 (θ, ϕ)
r
 
r.σ j− 1 j+ 1
Ωj,jz2 (θ, ϕ) = C−+ (j) Ωj,jz2 (θ, ϕ) (1669)
r
Furthermore, since
 2
r.σ
= I (1670)
r
one obtains the condition
C+− (j) C−+ (j) = 1 (1671)
 
r . σ
This condition can be made more restrictive as r is Hermitean, which
leads to
C+− (j) = C−+ (j)∗ (1672)
The above two equations suggest that C−+ (j) and C+− (j) are pure phase fac-
tors, such as
 
C+− (j) = exp + i φ(j)
 
C−+ (j) = exp − i φ(j) (1673)

The phase factor can be completely determined by considering the relations

(1669) with specific choices of the values of (θ, ϕ). As can be seen by examining
the case where ϕ = 0, the phase φ(j) is either zero or π. For the case ϕ = 0,
the operator simplifies to
   
r.σ cos θ sin θ
= (1674)
r sin θ − cos θ
The spinor spherical harmonics are given by
j+ 12
 q 
j+1−jz
j+ 1 − Y 1 (θ, ϕ)
Ωj,jz2 (θ, ϕ) =  q 2j+2 jz − 2
j+ 12

+ j+1+j2j+2
z
Y jz + 12
(θ, ϕ)
j− 12
 q 
j+jz
j− 1
2j Yjz − 21 (θ, ϕ)
Ωj,jz2 (θ, ϕ) =  q j− 12
 (1675)
j−jz
2j Yj + 1 (θ, ϕ)
z 2

which becomes real for ϕ = 0 since the spherical harmonics become real. Hence,
on inspecting eqn(1669) with ϕ = 0, one concludes that the phase factors are
equal and are purely real. That is
C+− (j) = C−+ (j) = ± 1 (1676)

282
Finally, by considering θ = 0, for which
   
r.σ 1 0
= (1677)
r 0 −1

and the spherical harmonics reduce

r
j± 1 2j + 1 ± 1
Yj ∓21 (0, ϕ) = δjz ∓ 12 ,0 (1678)
z 2 4π
j± 1
one finds that, for fixed j, only the four spinor spherical harmonics Ωj,jz2 (0, ϕ)
 q 
2j+1
j+ 1 − δjz − 12 ,0
Ωj,jz2 (0, ϕ) =  q 8π 
2j+1
+ 8 π δjz + 12 ,0
 q 
2j+1
j− 1
8 π δjz − 12 ,0
Ωj,jz2 (0, ϕ) =  q  (1679)
2j+1
8 π δ jz + 12 ,0

are non-zero. The spinor spherical harmonics with θ = 0 are connected via
 
1 0 j± 1 j∓ 1
Ωj,±21 (0, ϕ) = − Ωj,±21 (0, ϕ) (1680)
0 −1 2 2

Hence, one has determined that

C+− (j) = C−+ (j) = − 1 (1681)

which holds independent of the values of θ and j, so the effect of the operator
on the spinor spherical harmonics is completely specified by
 
r.σ j+ 1 j− 1
Ωj,jz2 (θ, ϕ) = − Ωj,jz2 (θ, ϕ)
r
 
r.σ j− 1 j+ 1
Ωj,jz2 (θ, ϕ) = − Ωj,jz2 (θ, ϕ) (1682)
r

as was to be shown.

——————————————————————————————————

The Ansatz

If one only considers the spatial part of the parity operator, P̂ , the two-
0 0
component spinor states Ωll0 ± 1 ,jz (θ, ϕ) have parities (−1)l
2

0 0 0
P̂ Ωll0 ± 1 ,jz (θ, ϕ) = (−1)l Ωll0 ± 1 ,jz (θ, ϕ) (1683)
2 2

283
Furthermore, as has been seen, the upper and lower two-component spinors of
the four-component Dirac spinor must have opposite intrinsic parity. Therefore,
the desired simultaneous eigenstates for the relativistic electron can be either
−
represented by the four-component Dirac spinor ψj,j z
(r) with parity (−1)l =
1 1
(−1)(l+ 2 − 2 ) of the form

f − (r)

l
− r Ω 1
l+ ,j
(θ, ϕ)
ψl+ 1 (r) =  g− (r) l+12 z  (1684)
2 ,jz i r Ωl+ 1 ,j (θ, ϕ)
2 z

+
or by ψj,j z
(r)
 
f + (r)
+ r Ωl+1
l+ 1 ,j
(θ, ϕ)
z
ψl+ 1
,j
(r) =  g + (r)
2  (1685)
2 z
i r Ωll+ 1 ,jz (θ, ϕ)
2

1 1
which has parity (−1)l+ 2 + 2 . In these expressions f ± (r) and g ± (r) are scalar
radial functions that have to be determined as solutions of the radial equation.
These states do not correspond to definite values of the orbital angular mo-
mentum since the upper and lower two-component spinors correspond to the
different values of either l or l0 = l + 1 for the orbital angular momentum.

To condense the notation, the energy eigenstates will be written in the com-
pact form
f ± (r)
!
lA
± r Ω j,jz (θ, ϕ)
ψj,jz (r) = ± (1686)
i g r(r) Ωlj,j
B
z
(θ, ϕ)
1 1
where lA = j ± 2 and lB = j ∓ 2.

The Radial Equation

±
We shall find the radial Dirac equation for the solution ψj,jz
(r). The Dirac
spinor wave functions in eqn(1685) and eqn(1684) are substituted into eqns(1643).
The spin-orbit interaction term can be evaluated by squaring the expression

Jˆ = L̂ + S (1687)

which leads to the identity

 
1 2 2
S . L̂ = Ĵ − L̂ − S2 (1688)
2

When this operator acts on the relativistic two-component spinor spherical har-
monic Ωlj,j
A
z
, one finds

h̄2
 
3
S . L̂ Ωlj,j
A
= j ( j + 1 ) − lA ( lA + 1) − Ωlj,j
A
(1689)
z
2 4 z

284
 1
which for j = lA + 2 yields

h̄2 1
S . L̂ Ωlj,j
A
= (j − ) Ωlj,j
A
(1690)
z
2 2 z

and for j = lA − 21 , one obtains

h̄2 3
S . L̂ Ωlj,j
A
= − (j + ) Ωlj,j
A
(1691)
z
2 2 z

The Dirac equation can be written in the general form

  
f (r) lA r.σ ∂ 2 g(r) lB
( E − V (r) − m c2 ) Ωj,jz = c h̄ r − S . L̂ Ωj,jz
r r2 ∂r h̄2 r
  
g(r) lB r.σ ∂ 2 f (r) lA
( E − V (r) + m c2 ) Ωj,jz = − c h̄ r − S . L̂ Ωj,jz
r r2 ∂r h̄2 r
(1692)
Following Dirac, it is customary to define an integer κ in terms of the eigenvalues
of S . L̂ via
h̄2
( S . L̂ ) Ωlj,j
A
= − ( 1 + κ ) Ωlj,j
A
z
2 z

2
h̄
( S . L̂ ) Ωlj,j
B
= − ( 1 − κ ) Ωlj,j
B
(1693)
z
2 z

j+ 1
Therefore, if Ωlj,j
A
z
= Ωj,jz2 , i.e. j = lA − 12 , then κ = (j + 12 ).
j− 1
Otherwise, if Ωlj,j
A
z
= Ωj,jz2 , i.e. j = lA + 12 , then κ = − (j + 12 ).

On substituting the above expressions in the Dirac energy eigenvalue equa-

tion for ψj,jz (r) one finds
  
f (r) lA r.σ ∂ g(r) lB
( E − V (r) − m c2 ) Ωj,jz = c h̄ r + 1 − κ Ωj,jz
r r2 ∂r r
  
g(r) lB r.σ ∂ f (r) lA
( E − V (r) + m c2 ) Ωj,jz = − c h̄ r + 1 + κ Ωj,jz
r r2 ∂r r
(1694)
Since the radial spin projection operator is independent of r, it can be com-
muted to the right of the differential operator in the large parenthesis. Then on
using either the relation given in eqn(1650) or in eqn(1651), one finds that the
relativistic spherical harmonics factor out of the equations, leading to
 
f (r) ∂ 1 − κ g(r)
( E − V (r) − m c2 ) = − c h̄ +
r ∂r r r
 
g(r) ∂ 1 + κ f (r)
( E − V (r) + m c2 ) = c h̄ +
r ∂r r r
(1695)

285
Table 9: The Relationship between j, lA , lB , κ and Parity. The parity eigenvalue
A
is given by ηp = (−1)l and κ = ±(j + 21 ).

κ lA lB Parity

κ = (j + 12 ) j+ 1
2 j− 1
2 (−1)κ

κ = −(j + 12 ) j− 1
2 j+ 1
2 (−1)1−κ

Therefore, the Dirac radial equation consists of the two coupled first-order dif-
ferential equations for f (r) and g(r). On multiplying by a factor of r and
simplifying the derivatives of f (r)/r, one finds the pair of more symmetrical
equations
 
∂ κ
( E − V (r) − m c2 ) f (r) + c h̄ − g(r) = 0
∂r r
 
∂ κ
( E − V (r) + m c2 ) g(r) − c h̄ + f (r) = 0
∂r r
(1696)
The above pair of equations are the central result of this lecture.

The Probability Density in Spherical Polar Coordinates.

The probability density P (r) that an electron, in an energy eigenstate of a

spherically symmetric potential, is found in the vicinity of the point (r, θ, ϕ) is
given by
P (r) = ψ † (r) ψ(r)
|f (r)|2 |g(r)|2
   
lA † lA †
= 2
Ω j,jz (θ, ϕ) Ω j,jz (θ, ϕ) + 2
Ωlj,j
B
(θ, ϕ) Ωlj,j
B
(θ, ϕ)
r r z z

(1697)
However, due to the identity
Ωlj,j
A
z
(θ, ϕ)† Ωlj,j
A
z
(θ, ϕ) = Ωlj,j
B
z
(θ, ϕ)† Ωlj,j
B
z
(θ, ϕ) = Aj,|jz | (θ) (1698)
the probability is independent of the azimuthal angle ϕ and the sign of jz (just
like in the non-relativistic case) and has a common angular factor of Aj,|jz | (θ).
Thus, the probability distribution factorizes into a radial and the angular factor
|f (r)|2 |g(r)|2
 
P (r) = + Aj,|jz | (θ) (1699)
r2 r2

286
 Table 10: Relativistic Angular Distribution Functions

j |jz | Aj,|jz | (θ)

1 1 1
2 2 4 π

3 1 1
2 2 8 π ( 1 + 3 cos2 θ )

3
2
3
2
3
8 π sin2 θ

5 1 3
2 2 16 π ( 1 − 2 cos2 θ + 5 cos4 θ )

5
2
3
2
3
32 π sin2 θ ( 1 + 15 cos2 θ )

5
2
5
2
15
32 π sin4 θ

The angular distribution function for a closed shell is given by the sum over the
angular distribution functions. Due to the identity,
j
X 2j + 1
Aj,|jz | (θ) = (1700)
jz =−j
4π

one finds that closed shells are spherically symmetric, as is expected. The first
few angular dependent factors Aj,|jz | (θ) are given in Table(10) and the corre-
sponding non-relativistic angular factors are given in Table(11). On compar-
ing the relativistic angular dependent factors with the non-relativistic factors
|Yml (θ, ϕ)|2 , one finds that they are identical for |jz | = j. Since the relativistic
distribution is the sum of two generally different positive definite forms origi-
nally associated with the two spinors χ+ and χ− , it generally does not go to
zero for non-zero values of θ.

12.10.1 The Hydrogen Atom

The radial energy eigenvalue equation for a hydrogenic-like atom is given by
Z e2
 
∂ κ
(E + − m c2 ) f (r) + c h̄ − g(r) = 0
r ∂r r
Z e2
 
∂ κ
(E + + m c2 ) g(r) − c h̄ + f (r) = 0
r ∂r r
(1701)

287
Figure 48: The relativistic (left) and non-relativistic (right) angular distribu-
tions Aj,|jz | (θ) for j = 21 and j = 32 .

288
Figure 49: The relativistic (left) and non-relativistic (right) angular distribu-
tions Aj,|jz | (θ) for j = 25 .

289
 Table 11: Non-Relativistic Angular Distribution Functions

l |m| Al,|m| (θ)

1
0 0 4 π

3
1 0 4 π cos2 θ

1 1 3
8 π sin2 θ

5
2 0 16 π ( 1 − 3 cos2 θ )2

2 1 15
8 π sin2 θ cos2 θ

2 2 15
32 π sin4 θ

The above equations will be written in dimensionless units, where the energy is
expressed in terms of the rest mass m c2 and lengths are expressed in terms of
the Compton wave length mh̄ c . A dimensionless energy  is defined as the ratio
of E to the rest mass energy
E
 = (1702)
m c2
For a bound state, m c2 > E > − m c2 so the value of the magnitude of
 is expected to be a little less than unity. A dimensionless radial variable ρ is
introduced which governs the asymptotic large r decay of the bound state wave
function. The variable is defined by
 
p
2
rmc
ρ = 1 −  (1703)
h̄
In terms of these dimensionless variables, the Dirac radial equations for the
hydrogen-like atom become
 r   
1 −  γ ∂ κ
− + f + − g = 0
1 +  ρ ∂ρ ρ
 r   
1 +  γ ∂ κ
+ g − + f = 0 (1704)
1 −  ρ ∂ρ ρ
where
Z e2
 
γ = (1705)
h̄ c

290
is a small number.

Boundary Conditions

The asymptotic ρ → ∞ form of the solution can be found from the asymp-
totic form of the equations
r
1 −  ∂
− f + g ∼ 0
1 +  ∂ρ
r
1 +  ∂
g − f ∼ 0 (1706)
1 −  ∂ρ
Hence, on combing these equations, one sees that the asymptotic form of the
equation is given by
∂2f
= f (1707)
∂ρ2
Therefore, one has
   
f ∼ A exp − ρ + B exp + ρ (1708)

and, likewise, g has a similar exponential form. If the solution is to be nor-

malizable, then the coefficient B in front of the increasing exponential must be
exactly zero (B ≡ 0).

The asymptotic ρ → 0 behavior of the solution can be found from

 
∂
γf + ρ − κ g = 0
∂ρ
 
∂
γg − ρ + κ f = 0 (1709)
∂ρ
where it has been noted that both the angular momentum term κ and the
Coulomb potential γ govern the small ρ variation, while the mass and en-
ergy terms are negligible. This is in contrast to the case of the non-relativistic
Schrödinger equation with the Coulomb potential, where for small r the Coulomb
potential term is negligible in comparison with the centrifugal potential. We
shall make the ansatz for the asymptotic small ρ variation

f ∼ A ρs
g ∼ B ρs (1710)

where the exponent s is an unknown constant and then substitute the ansatz in
the above equations. This procedure yields the coupled algebraic equations

γA + (s − κ)B = 0
γB − (s + κ)A = 0 (1711)

291
Hence, it is found that the exponent s is determined as solutions of the indicial
equation which is a quadratic equation. The solutions are given by
p
s = ± κ2 − γ 2 (1712)

Since, the wave function must be normalizable near ρ → 0,

Z  
lim dr | f |2 + | g |2 < ∞ (1713)
η→0 η

one must choose the positive solution for s. Normalizability near the origin
requires that 2 s > − 1. Hence, one may set
p
s = κ2 − γ 2 (1714)

This will be a good solution for κ = − 1 if Z does not exceed a critical value.
For values of Z greater than ≈ 172, the point charge can spark the vacuum
and spontaneously generate electron-positron pairs109 . The solution with the
negative value of s given by
p
s = − κ2 − γ 2 (1715)

could also possibly exist and be normalizable if γ is greater than a critical value
γc determined as
1 p
= 1 − γc2 (1716)
2
This critical value of γ is found from
√
3
γc = (1717)
2
which corresponds to Zc ∼ 118. The solutions corresponding to negative s
are, infact, un-physical and do not survive if the nucleus is considered to have
a finite spatial extent.

The Fröbenius Method

We shall use the Fröbenius method to find a solution. The solutions of the
radial equation shall be written in the form
 
f (r) = exp − ρ ρs F (ρ)
 
g(r) = exp − ρ ρs G(ρ) (1718)

109 H. Backe, L. Handschug, F. Hessberger, E. Kankeleit, L. Richter, F. Weik, R. Willwater,

H. Bokemeyer, P. Vincent, Y. Nakayama, and J. S. Greenberg, Phys. Rev. Lett. 40, 1443
(1978).

292
This form incorporates the appropriate boundary conditions at ρ → 0 and
ρ → ∞. The coupled radial equations are transformed to
 r   
1 −  ∂
− ρ + γ F + ρ + s − κ − ρ G = 0
1 +  ∂ρ
r   
1 +  ∂
ρ + γ G − ρ + s + κ − ρ F = 0
1 −  ∂ρ
(1719)

The functions F (ρ) and G(ρ) can be expressed as an infinite power series in ρ
∞
X
F (ρ) = an ρn
n=0
X∞
G(ρ) = bn ρn (1720)
n=0

where the coefficients an and bn are constants which have still to be determined.
The coefficients are determined by substituting the series in the differential
equation and then equating the coefficients of the same power in ρ. Equating
the coefficient of ρn yields the set of relations
 r 
1 − 
− an−1 + γ an + ( n + s − κ ) bn − bn−1 = 0
1 + 
r 
1 + 
bn−1 + γ bn − ( n + s + κ ) an + an−1 = 0
1 − 
(1721)

This equation is automatically satisfied for n = 0, since by definition a−1 =

b−1 ≡ 0 so the equation reduces to the indicial equation for s. These relations
yield recursion relations between the coefficients (an , bn ) with different values of
n. The form of the recursion relation can be made explicit by using a relation
between an and bn valid for any n. This relation is found by multiplying the
first relation of eqn(1721) by the factor
r
1 + 
(1722)
1 − 
and adding it to the second, one sees that the coefficients with index n − 1
vanish. This process results in the equation
r   r 
1 +  1 + 
γ − ( n + s + κ ) an + γ + ( n + s − κ ) bn = 0
1 −  1 − 
(1723)

valid for any n. The above equation can be used to eliminate the coefficients
bn and yield a recursion relation between an and an−1 . The ensuing recursion

293
relation will enable us to explicitly calculate the wave functions G(ρ) and hence
F (ρ).

Truncation of the Series

The behavior of the recursion relation for large values of n can be found by
noting that eqn(1723) yields
r
1 + 
n an ∼ n bn (1724)
1 − 
which when substituted back into the large n limit of the first relation of
eqn(1721) yields
n an ∼ 2 an−1 (1725)
Since the large ρ limit of the function is dominated by the highest powers of
ρ, it is seen that if the series does not terminate, the functions F (ρ) and G(ρ)
would be exponentially growing functions of ρ
 
F (ρ) ∼ exp + 2 ρ
 
G(ρ) ∼ exp + 2 ρ (1726)

Therefore, the set of recursion relations must terminate, since if the series does
not terminate, the large ρ behavior of the functions F (ρ) and G(ρ) would gov-
erned by the growing exponentials. Even when combined with the decaying
exponential term that appear in the relations
 
s
f (r) = ρ F (ρ) exp − ρ
 
s
g(r) = ρ G(ρ) exp − ρ (1727)

the resulting functions f (r) and g(r) would not satisfy the required boundary
conditions at ρ → ∞. We shall assume that the series truncate after the nr -th
terms. That is, it is possible to set

anr +1 = 0
bnr +1 = 0 (1728)

Thus, the components of the radial wave function may have nr nodes. Assuming
that the coefficients with indices nr + 1 vanish and using the first relation in
eqn(1721) with n = nr + 1, one obtains the condition
r
1 + 
bn = − anr (1729)
1 −  r

294
A second condition is given by the relation between an and bn
r   r 
1 +  1 + 
γ − ( n + s + κ ) an + γ + ( n + s − κ ) bn = 0
1 −  1 − 
(1730)

valid for any n. We shall set n = nr and then eliminate anr using the termination
condition expressed by eqn(1729). After some simplification, this leads to the
equation p
 γ = ( nr + s ) 1 − 2 (1731)
This equation determines the square of the dimensionless energy eigenvalue 2 .
On squaring this equation, simplifying and taking the square root, one finds
( nr + s )
 = ± p (1732)
( nr + s )2 + γ 2

or, equivalently, the energy of the hydrogen atom110 is given by

m c2
E = ± q (1733)
γ2
1 + ( nr + s ) 2

where r
1 2
s = (j + ) − γ2 (1734)
2
This expression for the energy eigenvalue is independent of the sign of κ and,
therefore, it holds for both cases
1
j = (l + 1) −
2
1
j = l + (1735)
2
Hence, the energy eigenstates are predicted to be doubly degenerate (in addition
to the (2j + 1) degeneracy associated with j3 ), since states with the same j but
have different values of l0 have the same energy. If the positive-energy eigenvalue
is expanded in powers of γ, one obtains
1 γ2
E ≈ m c2 − m c2 1 + ... (1736)
2 ( nr + j + 2 )2

which agrees with the energy eigenvalues found from the non-relativistic Schrödinger
equation. However, as has been seen, the exact energy eigenvalue depends on
nr and (j + 21 ) separately, as opposed to being a function of the principle quan-
tum number n which is defined as the sum n = nr + j + 21 . Hence, the Dirac
110 C. G. Darwin, Proc. Roy. Soc. A 118, 654 (1928).

C. G. Darwin, Proc. Roy. Soc. A 120, 621 (1928).

W. Gordon, Zeit. für Physik, 48, 11 (1928).

295
equation lifts the degeneracy between states with different values of the angular
momentum. The energy levels together with their quantum numbers are shown
in Table(12). The energy splitting between states with the same n and different
j values has a magnitude which is governed by the square of the fine structure
2
constant Z ( h̄e c ). That is
γ2

1
E ≈ m c2 1 −
2 ( nr + j + 12 )2
γ4
  
1 1 3
− − + . . .
2 ( nr + j + 12 )3 ( j + 12 ) 4 ( nr + j + 12 )
(1737)
The fine structure splittings for H-like atoms was first observed by Michelson111
and the theoretical prediction is in agreeement with the accurate measurements
of Paschen112 . The fine structure splitting is important for atoms with larger Z.
This observation has a classical interpretation which reflects the fact that for
large Z the electrons move in orbits with smaller radii and, therefore, the elec-
trons must move faster. Relativistic effects become more important for electrons
which move faster, and this occurs for atoms with larger values of Z. Although
the fine structure splitting does remove some degeneracy, the two states with
the same principle quantum number n and the same angular momentum j but
which have different values of l are still predicted to be degenerate. Thus, for
example, the 2Sj= 21 and the 2Pj= 12 states of Hydrogen are predicted to be de-
generate by the Dirac equation. It has been shown that this degeneracy is
removed by the Lamb shift, which is due to the interaction of an electron with
its own radiation field. The Lamb shift is smaller than the fine structure shifts
2
discussed above because it involves an extra factor of h̄e c .

The Ground State Wave Function

The ground state wave function of the hydrogen atom is slightly singular
at the origin. This can be seen by noting that it corresponds to nr = 0 and
κ = −1. Since the dimensionless energy is given by the expression
p
 = 1 − γ2 (1738)
one finds that the dimensionless radial distance ρ is simply given by
 
p
2
rmc
ρ = 1 − 
h̄
 
mc
= γ r
h̄
Z e2 m
= r (1739)
h̄2
111 A. A. Michelson, Phil. Mag. 31, 338 (1891).
112 F. Paschen, Ann. Phys. 50, 901 (1916).

296
Table 12: The Equivalence between Relativistic and Spectroscopic Quantum
Numbers.
 
n = nr + |κ| nr κ = ±(j + 12 ) nLj Degenerate E
m c2

Partner

p
1 0 -1 1S 12 1 − γ2

r
γ 2
2 1 -1 2S 12 2P 12 1− √
2+2 1−γ 2
2 1 +1 2P 12 2S 12 --
q
1
2 0 -2 2P 32 1− 4 γ2

r
γ 2
3 2 -1 3S 12 3P 12 1− √
5+4 1−γ 2
3 2 +1 3P 12 3S 12 --
r
γ 2
3 1 -2 3P 32 3D 32 1− √
5+2 4−γ 2
3 1 +2 3D 32 3P 32 --
q
1
3 0 -3 3D 52 1− 9 γ2

297
where the characteristic length scale is just the non-relativistic Bohr radius
divided by Z. The wave functions are written as
 
f (ρ) = exp − ρ ρs F (ρ)
 
g(ρ) = exp − ρ ρs G(ρ) (1740)

Since nr = 0, the recursion relations terminate immediately leading to the

functions F (ρ) and G(ρ) being given, respectively by constants a0 and b0 . Since
κ = − 1 for the ground state, the recursion relations are simply

γ a0 + ( s + 1 ) b0 = 0
γ b0 − ( s − 1 ) a0 = 0 (1741)

The solution of the equations results in the index s being given by

p
s = 1 − γ2 (1742)

and p
b0 s + κ 1 − γ2 − 1
= = (1743)
a0 γ γ
This shows that the lower component is smaller than the upper constant by
approximately γ, which has the magnitude of vc where v is the velocity in Bohr’s
theory. The ratio of b0 to a0 determines the radial functions as
 
f (r) s−1
= a0 ρ exp − ρ
r
p
1 − γ 2 − 1 s−1
 
g(r)
= a0 ρ exp − ρ (1744)
r γ

Since Y00 (θ, ϕ) = √1 , the angular spherical harmonics for the upper com-
4 π
ponents are just
1
ΩA (θ, ϕ) = √ χσ (1745)
4π
and the lower components are given by
r.σ 1
ΩB (θ, ϕ) = − √ χσ
r 4π
 
1 cos θ sin θ exp[−iϕ]
= − √ χσ (1746)
4π sin θ exp[+iϕ] − cos θ
Thus, apart from an over all normalization factor, the four-component spinor
Dirac wave function ψ is given by
 
√ 2   χσ 
N
ρ 1−γ −1 exp − ρ 

ψ = √ r . σ  (1747)
4π −i r χσ

298
 0.8
1S1/2
0.6

0.4

f(r)
0.2
g(r) x 100

0
0 0.5 1 1.5 2 2.5 3

-0.2

-0.4

mcγr/h

Figure 50: The large f (r) and small component g(r) radial wave functions for
the 1S 12 ground state of Hydrogen.

Hence, it is seen that as ρ approaches the origin, at first the wave function is
slowly varying since
√ 2 
γ2

γ2
ρ 1−γ −1 ∼ exp − ln ρ ∼ 1 − ln ρ (1748)
2 2

but for distances smaller than the characteristic length scale

 
h̄ 2
rc = exp − 2 (1749)
mcγ γ

the wave function exhibits a slight singularity. This length scale is much smaller
that the nuclear radius so, due to the spatial distribution of the nuclear charge,
the singularity is largely irrelevant. This singularity is not present in the non-
relativistic limit, since in this limit one assumes that the inequality | V (r) | 
m c2 always holds, although this assumption is invalid for r ∼ 0. Therefore, one
concludes that the relativistic theory differs from the non-relativistic theory at
small distances, which could have been discerned from the use of the Heisenberg
uncertainty principle.

299
 0.4 0.2

2P1/2
2S1/2
0.2 0
0 2 4 6 8 10

0 f(r)
0 2 4 6 8 10 -0.2
f(r) g(r) x 100
-0.2
g(r) x 100
-0.4

-0.4

-0.6
-0.6 mcγr/h
mcγr/h

Figure 51: The radial wave functions for the 2S 12 and 2P 12 states of Hydrogen.

12.10.2 Lowest-Order Radial Wavefunctions

The first few radial functions for the hydrogen atom can be expressed in the
form
s    s    
E r r r
f (r) = N 1 + exp − c0 − 2 c1
m c2 a a a
s    s    
E r r r
g(r) = − N 1 − 2
exp − d0 − 2 d1
mc a a a
(1750)
where the above form is restricted to the case where the radial quantum number
nr take on the values of 0 or 1. The index s is the same as that which occurs
in the Frobenius method and is given by the positive solution
p
s = κ2 − γ 2 (1751)
It is seen that the radial wavefunction depend on the dimensionless variable ρ
defined by
r
ρ = (1752)
a
where the length scale a is given in terms of the energy E and the Compton
wavelength by
  2 − 12  
E h̄
a = 1 − (1753)
m c2 mc
The values of the indices s, energy E, length scale a and normalization N
are given in Table(13). Since the two-component spinor spherical harmonics
Ωlj,jz (θ, ϕ) are normalized to unity, the normalization condition is determined
from the integral
Z ∞  
2 2 2
N dr | f | + | g | = 1 (1754)
0

300
involving the radial wave functions. The integral is evaluated with the aid of
the identity
Z ∞  
dρ ρa+b exp − 2 ρ = 2−(a+b+1) Γ(a + b + 1) (1755)
0

The coefficients cn and dn in the above expansion of the radial functions differ

Table 13: Parameters specifying the Radial Functions for the Hydrogen atom.
   
E γ m c
State s m c2 a h̄ N

p p s+ 1
κ = −1 1 − γ2 1 − γ2 1 √1 √2 2
a 2Γ(2s+1)
1S 12

q √   r
2 E
p 1 + 1 − γ2 ( −1) s+ 1
E 1 m c2 √1 √2 2
κ = −1 1 − γ2 2 2 m c2 2 ( 2 E2 ) a
m c Γ(2s+1)
2S 32

q √   r
2 E
p 1 + 1 − γ2 ( +1) s+ 1
E 1 m c2 √1 √2 2
κ=1 1 − γ2 2 2 m c2 2 ( 2 E2 ) a
m c Γ(2s+1)
2P 12

s+ 1
q
γ2
p
κ = −2 4 − γ2 1 − 2 √1 √2 2
4 a 2Γ(2s+1)
2P 32

from the coefficients an and bn that occur in the Frobenius expansion, since the
values of the ratio cnr /dnr has been chosen to simplify in the limit of large n. In
particular at the value of nr (at which the series terminates), the ratio is chosen
to satisfy
cnr
= 1 (1756)
d nr
instead of the condition
s
anr 1 + ( mEc2 )
= − (1757)
b nr 1 − ( mEc2 )

301
The relative negative sign and the square root factors in the coefficients have
been absorbed into the expressions for the upper and lower components f (r) and
g(r). The square root factors are responsible for converting the upper and lower
components, respectively, into the large and small components for positive E,
and vice versa for negative E. The expansion coefficients are given in Table(14).
Since the ratio of the magnitudes of the polynomial factors is generally of the

Table 14: Coefficients for the Polynomial in the Hydrogen atom Radial Wave-
functions.

State c0 c1 d0 d1

κ = −1 1 0 1 0
1S 12

   2 E
     2 E

E ( )+1 E ( )+1
m c2 m c2
κ = −1 2 m c2 2 s + 1 2 m c2 + 1 2 s + 1

2S 32

    2 E
    2 E

E ( )−1 E ( )−1
m c2 m c2
κ=1 2 m c2 − 1 2 s + 1 2 m c2 2 s + 1

2P 12

κ = −2 1 0 1 0
2P 32

order of unity, the ratio of the magnitudes of the small to large components is
found to be of the order of γ.

12.10.3 The Relativistic Corrections for Hydrogen

The Dirac equation for Hydrogen will be examined in the non-relativistic limit,
and the lowest-order relativistic corrections will be retained. The resulting equa-
tion will be recast in the form of a Schrödinger equation, in which the Hamil-
tonian contains additional interaction terms. The resulting interactions, when
treated by first-order perturbation theory, yield the fine structure. The physical
interpretation of the interactions will be examined. Historically, the following
type of analysis and the ensuing discussion of the Thomas precession played a

302
decisive role in compelling Pauli to reluctantly accept Dirac’s theory.

The Dirac equation can be expressed as the set of coupled equations

 
∂ 2
i h̄ − V − mc φA = − i h̄ c ( σ . ∇ ) φB
∂t
 
∂ 2
i h̄ − V + mc φB = − i h̄ c ( σ . ∇ ) φA (1758)
∂t

where φA and φB are, respectively, the upper and lower two component spinors
of the four-component Dirac spinor ψ. The energy eigenvalues of these equations
are sought, so to this end the explicit time-dependence of the energy eigenstates
will be separated out via
 A   
φ i
ψ = exp − Et (1759)
φB h̄

Also the non-relativistic energy  will be defined as the energy referenced with
respect to the rest-mass energy

E = m c2 +  (1760)

The coupled equations reduce to

 
 − V φA = − i h̄ c ( σ . ∇ ) φB
 
2
 − V + 2 mc φB = − i h̄ c ( σ . ∇ ) φA (1761)

The pair of equations will be expanded in powers of ( mp c )2 and only the first-
order relativistic corrections will be retained. One can express φB as

− i h̄ c ( σ . ∇ ) φA
φB =
 − V + 2 m c2
 −1
1  − V
= 1 + ( σ . p̂ ) φA
2mc 2 m c2
 
1  − V
≈ 1 − + ... ( σ . p̂ ) φA (1762)
2mc 2 m c2

to the required order of approximation. The above equation can be used to

obtain a Schrödinger-like equation for the two-component spinor φA . Since a
Schrödinger equation is sought for ψS , a correspondence must be established
between the pair of spinors (φA ,φB ) and ψS . The probability density is the
physical quantity which is directly associated with both types of wave functions.
The probability density associated with the Schrödinger equation should be
equivalent to the probability density associated with the Dirac equation. The

303
probability density associated with the four-component Dirac spinor depends
on both φA and φB ,
 
P (r) = φA† φA + φB † φB (1763)

The probability density associated with the two-component Schrödinger wave

function depends on ψS
P (r) = ψS† ψS (1764)
The probability density is normalized to unity. On equating the two expressions
for the normalization and substituting for φB , one obtains
Z Z  
1
d3 r ψS† ψS = d3 r φA† φA + ( σ . p̂ φA †
) ( σ . p̂ φ A
)
4 m2 c2
Z  
1
= d3 r φA† φA + φ A†
( σ . p̂ ) ( σ . p̂ ) φ A
4 m2 c2
p̂2
Z  
= d3 r φA† I + φA (1765)
4 m2 c2

Therefore, the two-component Schrödinger wave function can be identified as

p̂2
 
ψS = I + + ... φA (1766)
8 m2 c2

or, on inverting the expansion

p̂2
 
A
φ ≈ I − + ... ψS (1767)
8 m2 c2

Expressing φA in terms of ψS in the equation for φB yields the equation

p̂2
   
B 1  − V
φ ≈ 1 − ( σ . p̂ ) I − ψS
2mc 2 m c2 8 m2 c2
p̂2
     
1  − V
≈ ( σ . p̂ ) I − − ( σ . p̂ ) ψS
2mc 8 m2 c2 2 m c2
(1768)

On substituting φB and ψS into the equation for φA , one finds the (two-
component) energy eigenvalue equation

p̂2
  
 − V I − ψS
8 m2 c2
( σ . p̂ ) p̂2
     
 − V
= ( σ . p̂ ) I − − ( σ . p̂ ) ψS
2m 8 m2 c2 2 m c2
(1769)

304
or
p̂2 p̂2
 
 − V −  + V ψS
8 m2 c2 8 m2 c2
 2 
p̂2
   
p̂  − V
= I − − ( σ . p̂ ) ( σ . p̂ ) ψS
2m 8 m2 c2 4 m2 c2
(1770)
The above energy eigenvalue equation can be expressed as
p̂2 p̂2 p̂2
 
 − V − +  + V ψS
2m 8 m2 c2 8 m2 c2
p̂4
   
V
= − + ( σ . p̂ ) ( σ . p̂ ) ψS (1771)
16 m3 c2 4 m2 c2
The term proportional to the product of the energy eigenvalue  and the kinetic
energy can be re-written as
p̂2 p̂2
 ψS =  ψS
8 m2 c2 8 m2 c2
p̂2 p̂2
≈ ( V + ) ψS (1772)
8 m2 c2 2m
to the required order of approximation. On substituting the above expression
into the energy eigenvalue equation (1771), one finds
p̂2 p̂4 p̂2 p̂2
 
 − V − + + V + V ψS
2m 8 m3 c2 8 m2 c2 8 m2 c2
   
V
= ( σ . p̂ ) ( σ . p̂ ) ψS (1773)
4 m2 c2
The above equation will be interpreted as the non-relativistic energy eigenvalue
equation for the two-component wave function ψS , which contains relativistic
corrections of order ( vc )2 . The energy eigenvalue equation (1773) will be written
in the form
 2
p̂4

p̂
 ψS = + V ψS − ψS
2m 8 m3 c2
 2
p̂ V + V p̂2
    
V
− ψ S + ( σ . p̂ ) ( σ . p̂ ) ψS
8 m2 c2 4 m2 c2
(1774)
where the relativistic corrections are symmetric in p2 and V . This represents the
energy eigenvalue equation for a two-component wave function ψS , similar to
the Schrödinger wave function, but the above equation does include relativistic
corrections to the Hamiltonian. The first correction term is
p̂4
ĤKin = − (1775)
8 m3 c2

305
which is recognized as the relativistic kinematic energy correction, that origi-
nates from the expansion of the kinetic energy
p
 = m2 c2 + p2 c2 − m c2
p2 p4
≈ − + ... (1776)
2m 8 m3 c2
The remaining two correction terms
 2
p̂ V + V p̂2
    
V
( σ . p̂ ) ( σ . p̂ ) − (1777)
4 m2 c2 8 m2 c2

will be interpreted as the sum of the spin-orbit interaction and the Darwin term.
It should be noted that the sum of these two terms would identically cancel in
a purely classical theory. This cancellation can be shown to occur since, in the
classical limit, V and p commute, and then the Pauli-identity can be used to
show that the resulting pairs of terms cancel.

The factor
 
2 ( σ . p̂ ) V ( σ . p̂ ) − p̂2 V + V p̂2 (1778)

can be evaluated as
   
2 2
2 p̂ . V p̂ − p̂ V + V p̂ + 2iσ. p̂ ∧ V p̂ (1779)

The first two terms can be combined to form a double commutator, yielding
 
− [ p̂ , [ p̂ , V ] ] + 2 i σ . p̂ ∧ V p̂ (1780)

or  
+ h̄2 ∇2 V + 2 i σ . p̂ ∧ V p̂ (1781)

The last term can be evaluated, resulting in the expression

 
2 2
+ h̄ ∇ V + 2 h̄ σ . ∇ V ∧ p̂ (1782)

since
p̂ ∧ p̂ ≡ 0 (1783)
Using these substitutions, the remaining interactions can be expressed as the
sum of the spin-orbit interaction and the Darwin interaction

h̄2
 
h̄
ĤSO + ĤDarwin = + 2 2
σ . ∇ V ∧ p̂ + ∇2 V (1784)
4m c 8 m2 c2

306
The first term is the spin-orbit interaction term, and the second term is the
Darwin term. For central potentials, the Darwin term is only important for
electrons with l = 0. The evaluation and the physical interpretation of the
energy shifts due to the three fine-structure interactions will be discussed sepa-
rately.

12.10.4 The Kinematic Correction

The kinematic interaction
p̂4
ĤKin = − (1785)
8 m3 c2
originates from the expansion of the relativistic expression for the kinetic energy
of a classical particle
p
 = m2 c2 + p2 c2 − m c2
p2 p4
≈ − + ... (1786)
2m 8 m3 c2
The first-order energy shift due to the kinematic correction ĤKin can be eval-
uated by using the solution to the non-relativistic Schrödinger equation
2
p̂2 m c 2 Z e2 Z e2
  
ψS = − + ψS (1787)
2m 2 n2 h̄ c r
which leads to
p̂4
Z
∆EKin = − d3 r ψS† (r) ψS (r)
8 m3 c2
2 2
m c 2 Z e2 Z e2
Z  
1 3 †
= − d r ψ S (r) − + ψS (r)
2 m c2 2 n2 h̄ c r
4
Z e2 Z 2 e4
 Z
3 1
= mc 2
− d3 r ψS† (r) 2 ψS (r) (1788)
h̄ c 8 n4 2 m c2 r
Hence, the first-order energy shift due to the kinematic correction is evaluated
as
4 
Z e2
 
3 1
∆EKin = m c2 − (1789)
h̄ c 8 n4 n3 ( 2 l + 1 )
This term is found to lift the degeneracy between states with fixed principle
quantum numbers n and values of the angular momenta l. The relativistic kine-
matic correction to the energy is found to be smaller than the non-relativistic
energy by a factor of
2
Z e2

∼ Z 2 × 10−4 (1790)
h̄ c

307
which can be identified with a factor of ( vc )2 as can be inferred from an analysis
based on the Bohr model of the atom. One sees that the relativistic corrections
become more important for atoms with larger Z, since the correction varies as
Z 4 . This occurs because for larger Z the electrons are drawn closer to the nu-
cleus and, hence have higher kinetic energies, so the electron’s velocities draw
closer to the velocity of light.

12.10.5 Spin-Orbit Coupling

To elucidate the meaning of the spin-orbit interaction, the interaction will be
re-derived starting from quasi-classical considerations of the anomalous Zeeman
interaction of a spin with a magnetic field.

Consider a particle moving with a velocity v in a static electric field E. In

the particle’s rest frame, it will experience a magnetic field B 0 which is given by
1 v ∧ E 1
B0 = − q ≈ E ∧ v (1791)
c v2
1 − c c

for small velocities v. The magnetic field B 0 is a relativistic correction due to

the motion of the source of the electric field. If an electron is moving in a central
electrostatic potential φ(r) caused by a charged nucleus, the radial electric field
is given by  
r ∂φ
E = − (1792)
r ∂r
Hence, the magnetic field experienced by an electron in its rest frame is given
by
 
1 ∂φ
B0 = − r ∧ p
m c r ∂r
 
1 ∂φ
= − L (1793)
m c r ∂r
which is caused by the apparent rotation of the charged nucleus. In the electron’s
rest frame, the electron’s spin S should interact with the magnetic field through
the Zeeman interaction
rest = − q g B 0 . S
ĤInt (1794)
S
2mc
where gS is the gyromagnetic ratio for the electron’s spin. Dirac’s theory pre-
dicts that the spin is a relativistic phenomenon and also that gS = 2 for an
electron in its rest frame. This interaction with the magnetic field will cause
the spin of the electron to precess. The spin precession rate found in the elec-
trons rest frame is calculated as
e
ωrest = gS B 0 (1795)
2mc

308
However, the electron is bound to the nucleus and is orbiting with angular
momentum L. Therefore, one has to consider the corrections to the precession
rate (and the interaction) caused by the acceleration of the electron’s rest frame.

Thomas Precession

Electrons exhibit two different gyromagnetic ratios. The gyromagnetic ratio

of gS = 2 couples a spin to an external magnetic field and there is a gyro-
magnetic ratio of unity for the lab frame. This gyromagnetic ratio of unity (in
the lab frame) enters the coupling between the spin of an electron in a circular
orbit to the magnetic field B 0 experienced in the electron’s rest frame113 . We
shall find the gyromagnetic ratio in the lab frame, by calculating the rate of
precession that is observed in the lab frame and then inferring the (lab frame)
interaction which produces the same rate of precession.

In the electron’s rest frame, the gyromagnetic ratio due to the orbital mag-
netic field B 0 (caused by the charged nucleus) is given by gs = 2. This gyro-
magnetic ratio yields a spin precession rate in the electron’s rest frame of
e
ωrest = gS B 0 (1796)
2mc
The spin precession rate observed in the lab frame will be calculated later. The
rate of precession as observed in the electron’s rest frame has to be corrected
by taking into account the motion of the electron. The correction is due to
the non-additivity of velocities in successive Lorentz transformations. First,
the transformation properties of Dirac spinors under infinitesimal rotations and
boosts will be re-examined. Secondly, infinitesimal transformations will be suc-
cessively applied to describe the particle’s instantaneous rest frame and the
Thomas precession.

A Lorentz transform of a spinor field ψ is achieved by the rotation operator

R̂ via
ψ 0 (r) = R̂ ψ(R−1 r) (1797)
where R̂ shuffles the components of the spinor. For a passive rotation (of the
coordinate system) through the infinitesimal angle δϕ in the i - j plane, the
infinitesimal Lorentz transform has the non-zero elements

i,j = − j,i = − δϕ (1798)

Hence, the four-component spinor is transformed by a rotation operator of the

form
 
δϕ i,j
R̂(δϕ) = exp + i σ (1799)
2
113 L. H. Thomas, Nature 117, 514 (1926).

L. H. Thomas, Phil. Mag. 3, 1 (1927).

309
where
i
σ i,j = [ γ (i) , γ (j) ]
2  (k) 
X
i,j,k σ 0
= ξ (1800)
0 σ (k)
k

Hence, for a passive rotation through an infinitesimal angle δϕ, the four-component
Dirac spinor is rotated by
 (k) 
δϕ X i,j,k σ 0
R̂(δϕ) = Iˆ + i ξ + ... (1801)
2 0 σ (k)
k

which can be expressed in terms of the projection of the (block diagonal) spin
operator Ŝ = h̄2 σ̂ on the axis of rotation ê as

i δϕ
R̂(δϕ) = Iˆ + ( ê . σ̂ ) + . . .
2
i δϕ
= Iˆ + ( ê . Ŝ ) + . . . (1802)
h̄

which is in accord with the definition of spin Ŝ as the generator of rotations.

If the primed frame of reference has a velocity v along the k-axis relative
to the un-primed frame, the infinitesimal Lorentz transform has the non-zero
elements
v
0,k = − k,0 = − (1803)
c
A Lorentz boost along the k-axis corresponds to a rotation in the 0 - k plane
through an “angle” χ
 
χ 0,k
R̂(χ) = exp + i σ (1804)
2

where the “angle” χ is governed by the boost velocity v through

v
tanh χ = (1805)
c
However,
i
σ 0,k = [ γ (0) , γ (k) ] = i α(k) (1806)
2
so  
χ (k)
R̂(χ) = exp − α (1807)
2
Therefore, for a Lorentz boost with an infinitesimal velocity v along the k-th
direction, one finds
v
R̂(χ) = Iˆ − α(k) + . . . (1808)
2c

310
The infinitesimal transformation is guaranteed to be consistent with the source
free solution of the Dirac equation. For example, if the above transformation is
applied to the solution of the Dirac equation describing a positive-energy parti-
cle at rest, the transformed solution describes a particle moving with momentum
p = − m v when viewed from the moving frame of reference.

ωRest

a q ωΤ E

Figure 52: A cartoon depicting a rotating charged spin one-half particle, along
with the precession of the spin due to the external field in the particle’s rest
frame and the Thomas precession.

Consider the rotation R̂1 of a spinor due to an infinitesimal Lorentz trans-

formation with “small” velocity v, then
1
R̂1 = Iˆ − α . v + ... (1809)
2c
At a time δt later, the electron has changed its velocity since it is accelerating.
The new velocity of the electron’s rest frame is given by

v 0 = v + a δt (1810)

On performing a second Lorentz transform with the boost a δt, one finds the
rotation
1
R̂2 = Iˆ − α . a δt + . . . (1811)
2c
The combined Lorentz transform is given by
  
1 1
R̂ = R̂2 R̂1 = Iˆ − α . a δt Iˆ − α.v + ...
2c 2c
1 1
= Iˆ − α . ( v + a δt ) + ( α . a ) ( α . v ) δt + .(1812)
..
2c 4 c2
The Pauli identity can be used to evaluate the last term

(α.a)(α.v) = ( σ̂ . a ) ( σ̂ . v )
= a . v Iˆ + i σ̂ . ( a ∧ v ) (1813)

311
where, since the product of the two α’s yields a two by two block diagonal form
which involves the four by four matrices Iˆ and σ̂. Hence, the right-hand side acts
equally on both the upper and lower two-component spinors. Furthermore, since
the orbit is circular, the acceleration is perpendicular to the velocity, therefore

a.v = 0 (1814)

Thus, the combined boost corresponds to the transformation

1 i
R̂ = Iˆ − α . ( v + a δt ) + σ̂ . ( a ∧ v ) δt + . . . (1815)
2c 4 c2
The combined boost is identified as producing an infinitesimal Lorentz boost
through v + a δt and a rotation around an axis ê through the infinitesimal
angle δϕ given by
1
δϕ ê ≈ ( a ∧ v ) δt (1816)
2 c2
The rotation part acts on both the upper and lower two-component spinors in
the Dirac spinor. The rotation angle δϕ is linearly proportional to the time
interval δt. This class of rotations due to the combination of Lorentz boosts are
known as a Wigner rotations. Hence, it was shown that the spinor rotates with
the angular velocity given by
1
ωT = (a ∧ v)
2 c2
q
= (E ∧ v) (1817)
2 m c2
The magnitude of ωT is calculated as
e
ωT = B0 (1818)
2mc
and its direction is opposite to the precession of the spin in the electron’s rest
frame.

On combing the two precession frequencies, one finds that in the lab frame
the spin’s precession rate is given by

ωLab = ωrest − ωT
e
= ( gS − 1 ) B 0 (1819)
2mc
It is clear that the moving spin experiences an effective interaction which is
reduced by the factor  
gS − 1
(1820)
gS
when compared to the interaction in the electron’s rest frame. Hence, the gyro-
magnetic ratio that enters the spin-orbit coupling should not be gS but should

312
be given by ( gS − 1 ).

The Spin-Orbit Interaction

In the lab frame, the interaction between the moving electron’s spin S mag-
netic moment and its field is inferred to be
Lab = − q ( g − 1 ) B 0 . S
ĤInt (1821)
S
2mc
where gS is the gyromagnetic ratio. Since the magnetic induction field is given
by
 
0 1 ∂φ
B = − L (1822)
m c r ∂r
where the electrostatic potential is given by
q0
φ(r) = (1823)
r
the spin-orbit interaction can be expressed as
q q0
ĤSO = − ( gS − 1 ) L.S (1824)
2mc m c r3
Hence, the spin-orbit interaction is found to be given by
Z e2
ĤSO = ( gS − 1 ) L . S (1825)
2 m2 c2 r3
The spin-orbit coupling is a relativistic coupling which, apart from the Thomas
precession factor, indicates that the electron’s spin interacts with a magnetic
field in its rest frame via the gyromagnetic ratio of 2. The magnitude of the
interaction agrees precisely with the interaction found from the perturbative
treatment of the Dirac equation.

To first-order in perturbation theory, the spin-orbit coupling interaction

yields a shift of the energy levels. Since the total angular momentum J is a
good quantum number, one can write
 
1 3
L.S = j(j + 1) − l(l + 1) − (1826)
2 4
1
but j for a single electron can only take on the values j = l ± 2, so
 
1 1 1
L.S = ± (l + ) − (1827)
2 2 2
The expectation value of r−3 is evaluated as
Z  3
1 1 Z
d3 r ψS† (r) 3 ψS (r) = (1828)
r l ( l + 12 ) ( l + 1 ) n a

313
for l 6= 0. So the first-order energy-shift due to the spin-orbit coupling can be
expressed as
4  1
Z e2 ±(l + ) − 12
 
∆ESO = m c2 2
1 (1829)
h̄ c 4 n3 l ( l + 2 )(l + 1)

Therefore, the spin-orbit interaction lifts the degeneracy between states with
different j = l ± 12 values. For l = 0, the numerator vanishes since the total
angular momentum can only take the value
1
j = + (1830)
2
The energy shift produced by the spin-orbit coupling is about a factor of the
square of the fine structure constant
2 2
e2
 
1
∼ ∼ 10−4 (1831)
h̄ c 137

smaller than the energy levels of the hydrogen-like atom

2
m c2 Z e2

En ≈ − (1832)
2 n2 h̄ c

calculated using the non-relativistic Schrödinger equation. The spin-orbit split

levels are labeled by the angular momentum values and the j values, and are
denoted by nLj . Hence, for n = 2 and l = 1, one has the two levels 2P 12 and
2P 32 , while for n = 3 and l = 2 one has the levels 3D 32 and 3D 52 , and so on. It
is seen that the spin-orbit interaction is increasingly important for atoms with
large Z values, as it varies like Z 4 .

12.10.6 The Darwin Term

The Darwin term has no obvious classical interpretation. It only has physical
consequences for states with zero orbital angular momentum. However, it does
play an important role for the s electronic state of hydrogen, and is essential in
describing why the Dirac’s theory makes the 2S 21 and 2P 12 states of hydrogen
degenerate. This degeneracy was an essential ingredient in the discovery of the
Lamb shift and the subsequent development of Quantum Electrodynamics.

The Darwin interaction is given by

π Z e2 h̄2 3
ĤDarwin = δ (r) (1833)
2 m2 c2
which produces the first-order shift

π Z e2 h̄2 †
∆EDarwin = ψS (0) ψS (0) (1834)
2 m2 c2

314
Hence, the shift only occurs for electrons with l = 0. Furthermore, since the
probability density for finding the electron at the origin is given by
 3
1 Z
ψS† (0) ψS (0) = δl,0 (1835)
π na

to first order, the Darwin term produces a shift

4
m c2 Z 4 e2

∆EDarwin = δl,0 (1836)
2 n3 h̄ c

which shifts the energies of s states upwards. The Darwin term reflects the fact
that the relativistic corrections are important for small r since the inequality

Z e2
m c2  (1837)
r
required for the non-relativistic treatment to be reasonable is violated in this
region.

12.10.7 The Fine Structure of Hydrogen

E S P

Kinematic 2P3/2

Spin Orbit
2S1/2 2P1/2
Kinematic

Darwin

n=2

Figure 53: The Grotarian energy level diagram for the n = 2 shell of hydrogen
(blue). The diagram shows the magnitude and sign of the various relativistic
corrections. It should be noted that states with the same j are degenerate.

When the various relativistic corrections are combined, for l = 0, the Darwin
term exactly compensates for the absence of the spin-orbit interaction. There-
fore, the energy shifts combine to yield one formula in which l drops out. This
implies that the energy levels only depend on the principle quantum number n

315
and the total angular momentum j. States with different orbital angular mo-
menta are degenerate, even though the individual interactions appear to raise
the degeneracy. The relativistic corrections inherent in Dirac’s theory of hydro-
gen yields energy shifts and line-splittings which are described as fine structure.
The energy levels are described by

1 Z 2 α2 1 Z 4 α4
   
1 3
E ≈ m c2 1 − − − + . . .
2 n2 2 n3 ( j + 12 ) 4n
(1838)

where
e2
 
α = (1839)
h̄ c
is the fine structure constant. Generally, states with larger j values have higher
energies. The fine structure splittings decrease with increasing n like n−3 , but
increase with increasing Z like Z 4 . The splitting of the lower energy levels are
largest, for example

m c2 α 4
 
1 1
E2P 3 − E2P 1 = − − ≈ 4.533 × 10−5 eV (1840)
2 2 16 2 1

This splitting corresponds to a frequency of 10.96 GHz. The energy levels are
predicted to be doubly degenerate (in addition to the degeneracy associated
with j3 ), the degeneracy is just the number of states with different l values that
yield the same value of j. Since j is found by combining l with the electronic
spin s = 21 , there are two possible l values for each energy level which are given
by the solutions of either
1
j = l + (1841)
2
or
1
j = l − (1842)
2
The higher-order relativistic corrections do not alter the conclusion that the
states labeled by (n, j) are degenerate, as the energy levels found from the exact
solution of the Dirac equation only depend on n and j. For j = 12 the energy
levels, although predicted to be degenerate by Dirac’s theory, are experimentally
observed as being non-degenerate. The first experiments that revealed this split-
ting were performed by Lamb and Retherford114 . These scientists found that
the 2S 12 was shifted by about 1057 MHz to higher energies relative to the 2P 12 .
The relative shift of the nS 12 level of hydrogen with respect to the nP 12 level is
known as the Lamb shift.

Lamb and Retherford’s Experiment

114 W. E. Lamb Jr. and R. E. Retherford, Phys. Rev. 72, 241 (1947).

316
 E S P D
3D5/2
Kinematic
Kinematic Spin Orbit
3P3/2 3D3/2
Spin Orbit
3S1/2
3P1/2

Kinematic
Darwin

n=3

Figure 54: The Grotarian energy level diagram for the n = 3 shell of hydrogen
(blue). The diagram shows the magnitude and sign of the various relativistic
corrections. It should be noted that states with the same j are degenerate.

Lamb and Retherford designed an experiment to accurately measure the fine

structure of the hydrogen atom. In the experiment, the time scales were such
that the population of all excited states, other than the meta-stable 2S 12 state
of hydrogen, radiatively decayed to the ground state. Hence, the number of
induced transitions from the 2S 12 state could be monitored by simply observing
of the population of hydrogen atoms not in the ground state.

I
-
e
e-

Oven EM
H1 Cavity H1

Figure 55: A schematic of the apparatus used in the Lamb-Retherford exper-

iment. The beam of H molecules is produced in an oven, the beam is excited
by cross-bombardment with an electron beam. The population of the 2S 12 is al-
tered in the microwave resonator, and the population is observed via the current
emitted at the tungsten plate.

A beam of hydrogen atoms was produced by dissociating hydrogen molecules

317
in an oven. The thermal beam of hydrogen atoms was then cross-bombarded
with electrons, which excited some of the hydrogen atoms out of the ground
state. Since the electron-atom scattering doesn’t obey the radiation selection
rules, a finite population of atoms (about 1 in 108 ) were excited to the long-lived
2S 12 state. Subsequently, the other excited electronic states rapidly decayed to
the ground state by the emission of radiation. The beam of hydrogen atoms was
then passed through a tuneable (microwave) electromagnetic resonator, which
could cause the hydrogen atoms in the meta-stable level to make transitions
to selected nearby energy levels. Again, any non-2S excited state of hydrogen
produced by the action of the resonator rapidly decayed to the ground state.
The resulting beam of hydrogen atoms was incident on a Tungsten plate, and
the collision could result in electron emission if the atoms were in an excited
state, but no emission would take place if the hydrogen atom was in the ground
state. Therefore, the current due to the emitted electrons was proportional to
the number of meta-stable hydrogen atoms that survived the passage through
the resonator. Hence, analysis of the experiment yielded the number of transi-
tions undergone in the electromagnetic resonator.

Figure 56: The dependence of the current emitted from the tungsten plate on
the applied magnetic field. The resonance frequency was set to 9487 Megacycles.
[W. E. Lamb Jr. and R. C. Retherford, Phys. Rev. 72, 241 (1947).]

In the resonator, an applied magnetic field Zeeman split the excited levels of
hydrogen and, when the oscillating field was on-resonance with the splitting of
the energy levels, the hydrogen atom made transitions out from the meta-stable
2S 12 state. At resonance, the frequency of the oscillating electromagnetic field
is equal to the energy splitting. Therefore, for fixed frequency, knowledge of the
resonance magnetic field allowed the splitting of the energy levels to be accu-

318
rately determined. The field dependence of the resonance frequency indicated

Figure 57: The observed dependence of the resonance frequencies on the applied
magnetic field. The solid lines are the predictions of the Dirac theory and the
dashed lines are the result of Dirac’s theory if the energy of the 2S state is
simply shifted. [W. E. Lamb Jr. and R. C. Retherford, Phys. Rev. 72, 241
(1947).]

that at zero field the degeneracy between the 2S 21 and 2P 12 states were lifted,
with the 2S 21 state having the higher energy.

12.10.8 A Particle in a Spherical Square Well

The radial equation for a relativistic spin one-half particle in a spherically sym-
metric “square well” potential is given by
 
∂ κ
( E − V (r) − m c2 ) f (r) + c h̄ − g(r) = 0
∂r r
 
∂ κ
( E − V (r) + m c2 ) g(r) − c h̄ + f (r) = 0
∂r r
(1843)

319
We shall examine the case of an attractive central square well potential V (r)
which is defined by

− V0 for r < a
V (r) = (1844)
0 for r > a
In the region r < a where the potential is finite, the Dirac radial equation

0.5

0
V(r)/V0

-0.5

-1

-1.5
0 0.5 1 1.5 2

r/a

Figure 58: A spherically symmetric potential well, of depth V0 and radius a.

becomes
 
2 ∂ κ
( E + V0 − m c ) f (r) + c h̄ − g(r) = 0
∂r r
 
∂ κ
( E + V0 + m c2 ) g(r) − c h̄ + f (r) = 0
∂r r
(1845)

The function f (r) satisfies a second-order differential equation, which can be

found by pre-multiplying the second equation by the operator
 
∂ κ
c h̄ − (1846)
∂r r
and then eliminating g(r) by using the first equation. This process yields the
equation
 2   
∂ κ(κ + 1)
c2 h̄2 − f (r) = − ( E + V 0 )2
− m 2 4
c f (r)
∂r2 r2
(1847)

320
By using a similar procedure, starting from the second equation, one can find
the analogous equation for g(r)
 2   
2 2 ∂ κ(κ − 1) 2 2 4
c h̄ − g(r) = − ( E + V0 ) − m c g(r)
∂r2 r2
(1848)

It should be recognized that the term proportional to κ ( κ + 1 ) on the left-

hand side of the eqn(1847) for the large component, when divided by 2 m c2 ,
is equivalent to the centrifugal potential in the non-relativistic limit. The small
component experiences a different centrifugal potential. Furthermore, the quan-
tity
( E + V0 )2 − m2 c4 (1849)
plays a similar role to the kinetic energy in the non-relativistic Schrödinger
equation.

Real Momenta

If the quantity ( E + V0 )2 − m2 c4 is positive, it can be written as

( E + V0 )2 − m2 c4 = c2 h̄2 k02 > 0 (1850)

where k0 is real. These equations can be expressed in dimensionless form by

introducing the dimensionless variable variable ρ = k0 r. The radial equations
simplify to become
∂2f
 
ρ2 + ρ 2
− κ ( κ + 1 ) f = 0
∂ρ2
2
 
2 ∂ g 2
ρ + ρ − κ(κ − 1) g = 0 (1851)
∂ρ2
Since (apart from the sign) κ is identified with a form of angular momentum,
one sees that the upper and lower components experience different centrifugal
potentials. These equations have forms which are closely related to Bessel’s
equation. If one sets
1
f = ρ 2 X|κ+ 12 | (1852)
and
1
g = ρ 2 Y|κ− 12 | (1853)
the equations reduce to the pair of Bessel’s equations
∂ 2 X|κ+ 12 | ∂X|κ+ 12 |
 
2 2 1 2
ρ + ρ + ρ − (κ + ) X|κ+ 12 | = 0
∂ρ2 ∂ρ 2
∂ 2 Y|κ− 12 | ∂Y|κ− 12 |
 
1 2
ρ2 + ρ + ρ 2
− ( κ − ) Y|κ− 12 | = 0
∂ρ2 ∂ρ 2
(1854)

321
of half-integer order. The spherical Bessel functions and spherical Neumann
functions of order n are defined in terms of the Bessel functions via
r
π
jn (ρ) = J 1 (ρ)
2 ρ n+ 2
r
π
ηn (ρ) = N 1 (ρ) (1855)
2 ρ n+ 2

Therefore, the general solutions of each of the radial equations can be expressed
as
f (r)
= A0 j|κ+ 12 |− 12 (k0 r) + A1 η|κ+ 12 |− 12 (k0 r) (1856)
r
and
g(r)
= B0 j|κ− 12 |− 12 (k0 r) + B1 η|κ− 12 |− 12 (k0 r) (1857)
r
However, since the functions f (r) and g(r) in the upper and lower components
are related by the differential equations

( E + V0 + m c2 ) g(r)
    
∂ 1 + κ f (r)
+ = (1858)
∂ρ ρ r c h̄ k0 r

and
( E + V0 − m c2 )
    
∂ 1 − κ g(r) f (r)
+ = − (1859)
∂ρ ρ r c h̄ k0 r

the two sets of coefficients (A0 , A1 ) and (B0 , B1 ) must also be related. The
explicit relations can be found by using the recurrence relations for the spherical
Bessel functions jn (ρ)
 
∂ n+1
ρ jn (ρ) = ρn+1 jn−1 (ρ) (1860)
∂ρ

and  
∂
ρ−n jn (ρ) = − ρ−n jn+1 (ρ) (1861)
∂ρ
The spherical Neumann functions ηn (ρ) satisfy identical recurrence relations.
This yields the relations

E + V 0 + m c2
  
A0 = sign κ B0
c h̄ k0
E + V 0 + m c2
  
A1 = sign κ B1 (1862)
c h̄ k0

Hence, for positive-energy solutions. the upper components are the large com-
ponents and the lower components are the small components. In the inner
region, one must set A1 = B1 = 0, since the wave function are required to be

322
normalizable near the origin and the spherical Neumann functions ηn (ρ) diverge
as ρ−(n+1) as ρ → 0.

Imaginary Momenta

If the quantity ( E + V0 )2 − m2 c4 is negative, it can be written as

( E + V0 )2 − m2 c4 = − c2 h̄2 κ20 < 0 (1863)
where κ0 is real. This corresponds to the case of negative kinetic energies. In
this case, one can express the solution in terms of the modified spherical Bessel
functions
f (r)
= A00 i|κ+ 12 | (κ0 r) + A01 k|κ+ 12 | (κ0 r) (1864)
r
and
g(r)
= B00 i|κ− 12 | (κ0 r) + B10 k|κ− 12 | (κ0 r) (1865)
r
Because to the factors of i in the definitions of the modified spherical Bessel
functions, the amplitudes of the upper and lower components are related via
E + V 0 + m c2
 
0
A0 = − B00
c h̄ κ0
E + V0 + m c2
 
A01 = B10 (1866)
c h̄ κ0
where a minus sign has appeared in the first equation. Again, we see that for
positive energies, for r < a, the upper components are the larger components
and the lower components are the smaller components.

Bound States

The bound state energy E must occur in the energy interval

m c2 > E > − m c2 (1867)
so that the wave function in the region r < a where the potential is zero is
exponentially decaying. Since E 2 − m2 c4 < 0, the wave functions in the
outer region should also be expressed in terms of the modified spherical Bessel
functions. The quantity κ1 can be defined as
E 2 − m2 c4 = − h̄2 c2 κ21 (1868)
and the equations can be expressed in terms of the dimensionless variable
ρ = i κ1 r (1869)
In this case, it is more useful to express the solution of the radial Dirac equation
in terms of the spherical Hankel functions h± n (ρ). The spherical Hankel functions
are defined via
h±
n (ρ) = jn (ρ) ± i ηn (ρ) (1870)

323
For asymptotically large ρ, these functions are complex conjugates and represent
out-going or incoming spherical waves
  
1 1 π
lim h±n (ρ) → exp ± i ρ − ( n + ) (1871)
ρ→∞ ρ 2 2

The factor of ρ−1 reflects the fact that the intensity of an outgoing wave-packet
decreases in proportion to ρ−2 in order to conserve energy and probability. From
the asymptotic variation, it is seen that the spherical Hankel functions h± n (iρ)
with imaginary arguments, respectively, represent exponentially attenuating or
growing spherical waves. In the exterior region, the solutions are represented
by
f (r)
= C00 h+ |κ+ 12 |− 12
(iκ1 r) + C10 h− |κ+ 12 |− 12
(iκ1 r) (1872)
r
and
g(r)
= D00 h+ |κ− 12 |− 12
(iκ1 r) + D10 h− |κ− 12 |− 12
(iκ1 r) (1873)
r
The coefficients of the upper and lower components are related via

E + m c2
 
C00 = − D00
c h̄ κ1
E + m c2
 
0
C1 = D10 (1874)
c h̄ κ1

as can be seen by substituting the asymptotic form of the Hankel functions given
by eqn(1871) in the asymptotic form of the differential equations relating f (r)
and g(r) with V0 = 0. If this wave function is to be normalizable at ρ → ∞,
one must set C10 = D10 = 0.

The solutions for the wave functions have been found in the inner and outer
regions of the potential. The solution must also hold at r = a. This is achieved
by demanding that the upper and lower components of the wave function are
continuous at r = a. These conditions are demanded due to charge conservation
∂µ j µ = 0, since the current j µ only depends on the components of ψ and does
not (explicitly) depend on their derivatives.

Since the wave function at the origin must be normalizable, and since the
wave function must be exponentially decaying, when r → ∞, the matching
condition for the upper component becomes

A0 j|κ+ 21 |− 12 (k0 a) = C00 h+

|κ+ 1 |− 1
(iκ1 a) (1875)
2 2

and the matching condition for the lower components becomes

B0 j|κ− 12 |− 12 (k0 a) = D00 h+

|κ− 1 |− 1
(iκ1 a) (1876)
2 2

324
By eliminating the amplitudes from the two matching conditions by using
eqn(1874), one can arrive at the equation
  h+ (iκ1 a) 
E + V 0 + m c2 j|κ+ 21 |− 12 (k0 a) E + m c2
   
|κ+ 12 |− 12
sign(κ) = −
c h̄ k0 j|κ− 12 |− 12 (k0 a) c h̄ κ1 h+
|κ− 1 |− 1
(iκ1 a)
2 2
(1877)
In the above expression, the quantities k0 and κ1 are defined by
h̄2 c2 k02 = ( E + V0 )2 − m2 c4 (1878)
and
h̄2 c2 κ21 = m2 c4 − E 2 (1879)
These equations determine the allowed values for the energy. The above set of
equations have to be solved numerically to find the energy eigenvalues. We note
that for the Dirac particle, the spin effectively results in the formation of a cen-
trifugal barrier (either for the upper or the lower component) even for electrons
in s states. As a result, the potential V0 must exceed a critical strength if it is
to yield a bound state.

12.10.9 The MIT Bag Model

From the point of view of symmetry, a baryon, such as a neutron or proton, are
thought of as being composed of three (valence) quarks. For example, the proton
is considered to be made of two up quarks and a down quark (p = (uud)), while
the neutron is considered to be made of one up quark and two down quarks
(n = (udd)). These valence quarks are assumed to be surrounded by a sea of
gluons which bind the quarks together and a sea of virtual quark/anti-quark
pairs that are produced by the gluon field. Likewise, mesons are considered
to be made of a quark and an anti-quark, but these valence quarks are also
surrounded by a sea of gluons and quark/anti-quark pairs. The gluon force has
the property that the energy of interaction increases as the separation between
the quarks increases. It is this property of the gluon force that results in the
quarks being confined, so that no single quark can be found in nature.

The MIT bag model115 is a simple purely phenomenological model for the
structure of strongly interacting particles (hadrons). The model is based on
the spherically symmetric potential of radius a, but it will be assumed that the
quark mass can have one or the other of two values. The quark is assumed to
have a small mass (approximately zero) if it is located within a sphere of radius
a, and the mass is assumed to be very large (or infinite) if r > a. To be sure,
the quark mass is assumed to be a function of r such that
m = 0 if r < a
m → ∞ if r > a (1880)
115 A. Chodos, R. L. Jaffe, K. Johnson, C. B. Thorn, and V. F. Weisskopf, Phys. Rev. D 9,

3471, (1974).

325
It is the infinite mass of the quark for r > a that results in the confinement of
the quark to within the hadron. That is, in the exterior region, the infinite rest
mass energy exceeds the bound state energy so the exterior region is classically
forbidden, therefore, the particle is confined to the interior.

Inside the hadron, where both the potential energy and the mass m are zero,
the kinetic energy parameter k0 can be expressed entirely in terms of the energy
via E = h̄ c k0 since the potential is assumed to be zero. Therefore, the radial
components of the Dirac wave function can be expressed as
f (r)
= A0 j|κ+ 12 |− 12 (k0 r)
r
g(r)
= sign(κ) A0 j|κ− 12 |− 12 (k0 r) (1881)
r
where the amplitudes of the upper and lower components are the same, since
the potential and mass are zero for r < a.

Outside the hadron, where r > a, the energy E is assumed to be much less
than the rest mass energy, m c2  E, therefore, the momentum parameter is
imaginary and one can set h̄ c κ1 ≈ m c2 . In the exterior region, the radial
functions can be expressed as
f (r)
= C00 h+
|κ+ 1 |− 1
(iκ1 r)
r 2 2

g(r)
= − C00 h+
|κ− 1 |− 1
(iκ1 r) (1882)
r 2 2

since the imaginary momentum parameter has a magnitude which is governed

by the large mass m. Due to the large magnitude of κ1 , the wave function
decays very rapidly in the exterior region.

The bound state energy is determined from the matching condition

 h+ (iκ1 a) 
j|κ+ 12 |− 12 (k0 a)
 
|κ+ 12 |− 12
sign(κ) = − (1883)
j|κ− 12 |− 12 (k0 a) h+
|κ− 1 |− 1
(iκ1 a)
2 2

Due to the asymptotic properties of the spherical Hankel functions, their ratio
is unity for large κ1 . This leads to the energies of the quarks being governed by
the simplified matching condition

j|κ+ 21 |− 12 (k0 a) = − sign(κ) j|κ− 12 |− 12 (k0 a) (1884)

where
E = c h̄ k0 (1885)
The above equation governs the ground state and excited state energies of the
individual quarks inside the hadron. Since the spherical Bessel functions oscil-
late in sign, the above equations will result in a set of solutions for k0 with fixed

326
κ. From the structure of the equations, it is seen that the solutions k0 will only
depend on the integer number κ and the value of a. Since another boundary
condition should also be imposed at the bag’s surface, only states with angular
momentum j = 21 should be retained. This extra condition restricts the interest
to states with κ = − 1.

We shall examine the lowest-energy bound state which corresponds to the

case κ = − 1. The bound state energies are given by the matching condition

j0 (k0 a) = j1 (k0 a) (1886)

Since
sin ρ
j0 (ρ) = (1887)
ρ
and  
sin ρ − ρ cos ρ
j1 (ρ) = (1888)
ρ2
the energy eigenvalues are determined by the solutions of
1
ρ = (1889)
1 + cot ρ
which has an infinite number of solutions which, asymptotically, are spaced by
π. The smallest solution corresponds to k0 a = 2.04. Hence, the energy of the

1
P(ρ)/P(0)

0.5

0
0 1 2 3 4
ρ

Figure 59: The radial dependence of the quark-distribution in the ground state
of the MIT bag.

lowest-energy quark is give by the formula

2.04 c h̄
Eκ=−1,nr =0 = (1890)
a
The solutions with larger values of k0 , corresponding to excited states with
κ = − 1 are given by analogous expressions. Therefore, if one knows the value
of a, one could find the energies required to excite a single-quark between the

327
Table 15: The lowest single-particle energies (in units of Eκ,nr a/c h̄) of the MIT
Bag Model.

nr κ = −1 κ = +1 κ = −2 κ = +2

nr = 0 2.04 3.81 3.21 5.12

nr = 1 5.40 7.00 6.76 8.41

nr = 2 8.58 10.17 10.01 11.61

nr = 3 11.73 13.31 13.20 14.79

single particle levels. This could allow one to calculate the excitation energies
required to change the hadron’s internal structure.

In conclusion, the MIT bag model, when interpreted as being a strictly

single-particle picture, predicts that the set of excitation energies (for the in-
ternal structure) of each of the basic hadrons can be put into a one-to-one
correspondence with each other. That is, the family of excitation energies for
each hadron should fall on-top of each other, if one scales the energies by mul-
tiplying them with the hadron’s characteristic length scale a. The bag radius
is determined by the use of further phenomenological considerations. However,
although the model can be used to fit the right size for a nucleon (∼ 1 fm),
the model predicts that a meson (such as the pions which are composed of a
quark and anti-quark in the combinations of either (u, d), (d, u) or √12 (dd − uu)
) should have almost the same radii116
  14
an 3
= (1891)
aπ 2
Hence, the ratio of the nucleon mass Mn to the pion mass Mπ is expected to be
given by the formula
  14
Mn 3 × 2.04/an 3 2
= = (1892)
Mπ 2 × 2.04/aπ 2 3
116 It is assumed that the bag energy is given by the sum of a volume term B a3 and the sum

of the quark energies cah̄

P
n
αn . Minimizing the energy w.r.t a results in the bag radius a
being determined by
c h̄ X
a4 = αn
3B
n

328
Table 16: The Observed Energy Levels for the charmonium system (cc) in units
of MeV/c2 .

1 3 3 3 3
S0 S1 P0 P1 P2

2981 3097 3415 3510 3556

3686 3770 - - -
- 4040 - - -
- 4160 - - -

which yields a ratio of 1.36. This ratio is far too small for the triplet of π mesons
since Mπ ∼ 139 MeV/c2 , and Mn ∼ 938 MeV/c2 . Although it is in adequate
for the pseudo-scalar mesons, the MIT bag model is more appropriate for the ω
vector meson which is composed of √12 (uu + dd) and has a mass of Mω ∼ 783
MeV/c2 , or the ρ vector meson √1 2 (uu + dd) with a mass Mρ ∼ 776 MeV/c2 .
Hence, at best, the MIT Bag model produces mixed results. The MIT Bag
model is also quite unappealing, since the basic assumptions of the bag model
do not follow from Quantum Chromodynamics, and the model is neither re-
normalizable nor is it Lorentz invariant.

12.10.10 The Temple Meson Model

A quark and anti-quark pair form bound states. Thus, for example a charmed
quark/anti-quark pair (c, c) can form states with different internal quantum
numbers117 . The experimentally determined energies for the J/Ψ system118
are given in Table(16). Similarly, the Upsilon particle119 (bb) has a similar set
of energy levels. The energy levels of the Upsilon system120 are tabulated in
Table(17). For positronium121 , like the hydrogen atom, it is the electromagnetic
force mediated by vector photons which binds the electron and positron into a
bound state. For a quark/anti-quark bound state, it is the color force mediated
by massless vector gluons that bind the quark/anti-quark pair together. The
color force has the property that it increases with increasing separation of the
quark/anti-quark pair, which has the consequence that the quarks are confined.
Furthermore, high-energy inelastic scattering experiments on hadrons indicate
117 J. E. Augustin et al. Phys. Rev. Lett. 33, 1406 (1974).

J. J. Aubert Phys. Rev. Lett. 33, 1404 (1974).

118 The data are taken from the Particle Data Group: http://pdg.lbl.gov
119 S. W. Herb, et al. Phys. Rev. Lett. 39, 252 (1977).

W. R. Innes et al. Phys. Rev. Lett. 39, 1240 (1977).

120 The data are taken from the Particle Data Group: http://pdg.lbl.gov
121 M. Deutsch, Phys. Rev. 82, 455 (1951).

329
Table 17: The Observed Energy Levels for the Upsilon system (bb) in units of
MeV/c2 .

1 3 3 3
S0 P0 P1 P2

9460 9860 9893 9913

10025 10232 10255 10268
10355 - - -
10580 - - -

that at small separations the quarks only interact weakly. This property is
called asymptotic freedom. It was the realization by ’t Hooft122 , Gross and
Wilczek123 and Politzer124 that non-Abelian gauge theories possessed the prop-
erties of asymptotic freedom that led to the acceptance of the theory of Quantum
Chromodynamics. The screening of the color force between the quarks at large
distances (due to virtual quark/anti-quark pairs) is more than compensated by
an anti-screening due to virtual gluon pairs. However, at small distances the
color force vanishes.

The rest-mass energy of the quarks and anti-quarks will be modeled by

 
m(r) c = m0 c − i ω α . r (1893)

which describes an energy similar to that of an elastic string which couples to

the spin125 . The model has two undetermined parameters, the quark mass m0
and the string tension m0 c ω. The mass m(r), and the Dirac equation, can be
used to determine the energy levels of quarkonium.

Exercise:

Show that the positive energy eigenvalues of the Dirac equation with the
mass m(r) given by
 
m(r) c = m0 c − i ω α . r (1894)

are determined as √
En,j,l = m0 c2 tA + 1 (1895)
122 G. t’ Hooft, unpublished (1972).
123 D. J. Gross and F. A. Wilczek, Phys. Rev. Lett. 30, 1343 (1973).
124 H. D. Politzer, Phys. Rev. Lett. 30, 1346 (1973).
125 D. Ito, K. Mori and E. Carriere, Nuovo Cimento, 51 A, 1119, (1967).

P. A. Cook, Nuovo Cimento Lett. 1, 419 (1971).

330
where the dimensionless parameter t corresponding to the string tension is given
by  
h̄ ω
t = (1896)
m0 c2
and A is given in terms of the quantum numbers as
1
A = 2 (n − j) + 1 if j = l +
2
1
A = 2 (n + j) + 3 if j = l − (1897)
2
Hence, find the best fit to the excitation spectra of quarkonium.

12.11 Scattering by a Spherically Symmetric Potential

First, the polarization dependence of scattering of an electron from a Coulomb
potential will be examined in terms of the scattering amplitudes, and second,
by using a partial wave analysis, the scattering amplitudes will be expressed in
terms of phase shifts.

12.11.1 Polarization in Coulomb Scattering.

The scattering of a relativistic electron by a Coulomb force field results in spin-
flip scattering since the electron has a magnetic moment which interacts with
the magnetic field produced in the electron’s rest frame. Since the Coulomb
potential is spherically symmetric, the angular momentum Jˆ2 and Jˆ(3) com-
mute with the Hamiltonian, hence, (j, j3 ) are constants of motion. However,
the orbital angular momentum L̂ does not commute with Ĥ.

The Dirac wave function ψ(r) can be expressed in terms of two two-component
spinors  A 
φ (r)
ψ(r) = (1898)
φB (r)
One only need specify the upper component φA (r), since once φA (r) has been
specified φB (r) is completely determined. For example, for the in and out
asymptotes, the Dirac equation reduces to

E p − m c2
   A 
− c p̂ . σ φ (r)
= 0 (1899)
− c p̂ . σ E p + m c2 φB (r)

Hence, the lower two-component spinor is completely determined in terms of

the upper two-component spinor
c p̂ . σ
φB (r) = φA (r) (1900)
E p + m c2

331
 In the scattering experiment, a plane-wave with momentum p parallel to the
ê3 -axis falls incident on the target. The in-asymptote can be described by a
state which is in a superposition of eigenstates of Ŝ (3) given by
   
in χ± pr
ψ± (r) = NEp c p
± Ep + m c2 χ± exp i cos θ (1901)
h̄

From the Rayleigh expansion, one observes that the in-asymptotes are not eigen-
states of (Ĵ)2 = (L̂ + Ŝ)2 since they are formed of linear superpositions of many
2 2
states with differing eigenvalues of L̂ but have a fixed eigenvalue of Ŝ . How-
ever, the in-asymptote are eigenstates of Jˆ(3) = L̂(3) + Ŝ (3) with eigenvalues
± h̄2 .

Ψin
(θ,ϕ)
p

Ψout

Figure 60: The geometry of the asymptotic final state of Mott scattering. At
large r, the beam separated into an unscattered beam ψin and a spherical out-
going wave ψout .

The corresponding out-asymptotes can be described as spherical outgoing

waves. Even though the in-asymptote may have a definite eigenvalue of Ŝ (3) ,
the spherically symmetric out-asymptote waves may contain a component with
flipped spin, due to the action of the spin-orbit coupling
   
1
Ŝ . L̂ = Ŝ (3) L̂(3) + Ŝ + L̂− + Ŝ − L̂+ (1902)
2

active in the vicinity of the target. In spherical polar coordinates, the orbital

332
angular momentum raising and lowering operators are given by
 
∂ ∂
L̂± = ± h̄ exp[±iϕ] ± i cot θ (1903)
∂θ ∂ϕ

Hence, on noting that

Ŝ ± χ± ≡ 0 (1904)
one finds that the out-asymptotes can be expressed as
!
( f (θ) ± g(θ) exp[ ± iϕ ] Ŝ ∓ )
 
out χ± 1 pr
ψ± (r) = NEp c p ( êr . σ ) ∓ exp i
Ep + m c2 ( f (θ) ± g(θ) exp[ ± iϕ ] Ŝ ) χ± r h̄
(1905)
where êr is a unit vector in the radial direction. It should be noted that the
out-asymptote describes an outgoing spherical wave when r → ∞. Therefore,
the operator ( σ . p̂ ) appearing in the asymptote has simplified since
   
r.σ ∂ 2i Ŝ . L̂
lim ( σ . p̂ ) = lim − i h̄ +
r→∞ r→∞ r ∂r h̄ r
  
r.σ ∂
→ − i h̄ (1906)
r ∂r

which reflects that the spin-orbit coupling term is ineffective at r → ∞. Simi-

larly, the effect of the differential operator can be evaluated as
    
∂ 1 pr p pr
lim − i h̄ exp i → exp i (1907)
r→∞ ∂r r h̄ r h̄

In light of the comment about the upper two-component spinor, one sees that
the scattered wave is determined by
 
f (θ)
(1908)
g(θ) exp[+iϕ]

for an incident beam with positive helicity, and by

 
−g(θ) exp[−iϕ]
(1909)
f (θ)

if the initial beam has a negative helicity. The quantities f (θ) and g(θ) are gen-
eralized scattering amplitudes that have the dimensions of length, and depend
on θ but do not depend on ϕ as both the in and out asymptotes are eigenstates
of Jˆ(3) with eigenvalues ± h̄2 . A partial wave analysis can be performed on the
Dirac equation to yield expressions for the scattering amplitudes f (θ) and g(θ)
in terms of phase shifts. A detailed knowledge of the scattering amplitudes is
not required for the following analysis.

333
 If the in-asymptote has the spin quantized along the direction given by
(sin θs cos ϕs , sin θs sin ϕs , cos θs ), the upper component of the Dirac wave spinor
is determined by the two-component spinor

cos θ2s exp[−i ϕ2s ]

 
χs = (1910)
sin θ2s exp[+i ϕ2s ]

The out-asymptote is then determined by the two-component spinor

f (θ) cos θ2s exp[−i ϕ2s ] − g(θ) sin θ2s exp[+i ϕ2s ] exp[−iϕ]
   
A 1 pr
φ (r) = exp i
g(θ) cos θ2s exp[−i ϕ2s ] exp[+iϕ] + f (θ) sin θ2s exp[+i ϕ2s ] r h̄
(1911)
The probability for scattering is proportional to
 2
 θs ϕs θs ϕs 
I(θ, ϕ) ∝  f (θ) cos exp[−i ] − g(θ) sin exp[+i ] exp[−iϕ] 
2 2 2 2
 2
 θs ϕs θs ϕs 
+  g(θ) cos
 exp[−i ] exp[+iϕ] + f (θ) sin exp[+i ] 
2 2 2 2
   
2 2 ∗ ∗
= | f (θ) | + | g(θ) | + sin θs sin(ϕ − ϕs ) i f (θ) g(θ) − f (θ) g (θ)

(1912)

which clearly depends on the azimuthal angle ϕ.

If the initial beam is unpolarized, the direction of the initial spin (θs , ϕs )
should be averaged over by integrating over the solid angle dΩs = dϕs dθs sin θs .
This process yields the scattering probability for the unpolarized beam
Z  
dΩs 2 2
I(θ, ϕ) = | f (θ) | + | g(θ) | (1913)
4π
which is independent of the azimuthal angle ϕ. It should be noted that the
unpolarized cross-section differs from the polarized cross-section.

Even if the initial beam is unpolarized, the final beam will be partially
polarized. The direction of the net polarization is determined by evaluating the
matrix elements of Ŝ and averaging over the direction of the initial spin, θs and
ϕs . The result is proportional to
 ∗
f (θ) g(θ) − f (θ) g ∗ (θ)

h̄
Ŝ = i (sin ϕ, − cos ϕ, 0) (1914)
2 | f (θ) |2 + | g(θ) |2
Hence, the polarization is perpendicular to the scattering plane. It should also
be noted that the net polarization of the scattered wave is determined by the
relative deviation of the scattering cross-section for polarized electrons from the
unpolarized scattering cross-section.

334
12.11.2 Partial Wave Analysis
The Dirac equation with a spherically symmetric potential V (r) has solutions
of the form
j± 12
!
f (r)
r Ω j,jz (θ, ϕ)
ψ(r) = j∓ 12 (1915)
i g(r)
r Ω j,jz (θ, ϕ)
j± 1
where the two-component spinor spherical harmonics Ωj,jz2 (θ, ϕ) are given by
q 1 1
j± 12
 
j+ 2 ± 2 ∓jz
j± 1 ∓ 2j+1±1 Y jz − 12
(θ, ϕ)
Ωj,jz2 (θ, ϕ) =  q 1 1 j± 12
 (1916)
j+ 2 ± 2 ±jz
2j+1±1 Y j + 1 (θ, ϕ)
z 2

and the radial functions fκ (r) and gκ (r) satisfy

   
2 ∂ κ
E − V (r) − m c fκ (r) = − c h̄ − gκ (r)
∂r r
   
∂ κ
E − V (r) + m c2 gκ (r) = c h̄ + fκ (r) (1917)
∂r r
1
where κ = ± ( j + 2 ). If the momentum h̄ k is defined via

c2 h̄2 k 2 = E 2 − m2 c4 (1918)

the asymptotic r → ∞ form of the solutions of these coupled equations with

positive values of κ are of the form of a linear superposition
fκ (r)
= Aκ jκ (kr) + Bκ ηκ (kr) (1919)
r
where jκ (kr) and ηκ (kr) are the spherical Bessel and the spherical Neumann
functions. For negative values of κ, the solutions are given by
fκ (r)
= Aκ j−κ−1 (kr) + Bκ η−κ−1 (kr) (1920)
r
The spherical Bessel and spherical Neumann functions have the asymptotic
forms
cos(kr − (κ + 1) π2 )
jκ (kr) →
kr
sin(kr − (κ + 1) π2 )
ηκ (kr) → (1921)
kr
The solutions for a free particle do not involve the spherical Neumann functions,
since they are not normalizable at the origin. The amplitudes of the asymptotic
solution in the presence of a finite potential V (r) are usually written as
Bκ
= − tan δκ (k) (1922)
Aκ

335
where δκ (k) are the phase shifts that characterize the potential. The phase
shifts depend directly on κ (and the energy) and only depend indirectly on j
and l through κ. The phase shifts are defined so that the asymptotic variation
of the radial functions is given by
fκ (r) cos(kr − (κ + 1) π2 + δκ (k))
∼ eiδκ (k) (1923)
r r
and only differs from the asymptotic variation of the free particle solutions
through the phase shifts. Furthermore, if this is decomposed in terms of incom-
ing and outgoing spherical waves,
  
exp i k r − (κ + 1) π2 + 2 δκ (k)
fκ (r)
∼
r   2r 
π
exp − i k r − (κ + 1) 2
+ (1924)
2r
their fluxes are equal due to conservation of particles and, as written, the in-
coming spherical waves are not modified by the phase-shifts.

The general asymptotic r → ∞ form of the wave function for the scattering
is composed of the un-scattered wave and a spherical outgoing wave. The polar-
axis is chosen to be parallel to direction of the incident beam which is also chosen
to be the quantization axis for the spin. If the incident beam is polarized with
spin-up, the upper two-component spinor has the form
 
      exp i k r
1 f (θ)
φA
↑ (r) = exp i k r cos θ +
0 g(θ) exp[ i ϕ ] r
(1925)
whereas for a down-spin polarized incident beam
 
      exp i k r
0 − g(θ) exp[ − i ϕ ]
φA↓ (r) = exp i k r cos θ +
1 f (θ) r
(1926)
On recalling the Rayleigh expansion
  X
exp i k r cos θ = il ( 2 l + 1 ) jl (kr) Pl (cos θ) (1927)
l

one can find the scattered spherical outgoing wave by subtracting the un-
scattered beam from the total wave function. On using the asymptotic large r
variation, one obtains the asymptotic form
cos(kr − (l + 1) π2 )
  X
exp i k r cos θ → il ( 2 l + 1 ) Pl (cos θ) (1928)
kr
l

336
which has a similar form to the asymptotic form of the total wave function.
In particular, the spin and orbital angular momentum eigenstates can be de-
composed in terms of the spinor spherical harmonics. Thus, for the up-spin
polarized incident beam one has the upper two-component spinor
r
4π
Pl (cos θ) χ+ = Y l (θ, ϕ) χ+
2l + 1 0
√
√ √
 
4π
= l + 1 Ωll+ 1 , 1 − l Ωll− 1 , 1 (1929)
2l + 1 2 2 2 2

and for the down-spin beam

r
4π
Pl (cos θ) χ− = Y l (θ, ϕ) χ−
2l + 1 0
√
√ √
 
4π
= l + 1 Ωll+ 1 ,− 1 + l Ωll− 1 ,− 1 (1930)
2l + 1 2 2 2 2

Therefore, when expressed in terms of a superposition of continuum energy

eigenstates corresponding to different values of j and κ, the asymptotic form of
the Rayleigh expansion becomes
√ X cos(kr − (l + 1) π )  √ √
  
exp i k r cos θ χ+ → 4π il 2
l + 1 Ωll+ 1 , 1 − l Ωll− 1 , 1
kr 2 2 2 2
l
(1931)
and
√ X cos(kr − (l + 1) π )  √ √ l
  
l 2 l
exp i k r cos θ χ− → 4π i l + 1 Ωl+ 1 ,− 1 + l Ωl− 1 ,− 1
kr 2 2 2 2
l
(1932)
Although the coefficients Aκ of the exact wave function are as yet unknown, they
can be determined by requiring the scattered spherical wave does not contain
terms proportional to  
exp − ikr
(1933)
r
which would represent an incoming spherical wave. This requirement leads to
the outgoing spherical wave having a spin-up component given by
√   
4π X (l + 1)
√ exp[ 2 i δ−l−1 (k) ] − 1
2ik 2l + 1
l
 
   exp i k r
l
+ √ exp[ 2 i δl (k) ] − 1 Y0l (θ, ϕ)
2 l+ 1 r
(1934)

337
and the down-spin component is given by
√ r  
4π X l(l + 1)
exp[ 2 i δ−l−1 (k) ] − 1
2ik 2l + 1
l
 
   exp i k r
− exp[ 2 i δl (k) ] − 1 Y1l (θ, ϕ) (1935)
r

In the above expressions, the index on the phase-shifts δκ (k) refer to the value
of κ. Hence, for a spin-up polarized incident beam, the scattering amplitudes
are given in terms of the phase-shifts via
√   
4π X (l + 1)
f (θ) = √ exp[ 2 i δ−l−1 (k) ] − 1
2ik 2l + 1
l
 
l
+ √ exp[ 2 i δl (k) ] − 1 Y0l (θ, ϕ) (1936)
2l + 1
and
√ r  
4π X l(l + 1)
g(θ) exp[ i ϕ ] = exp[ 2 i δ−l−1 (k) ] − 1
2ik 2l + 1
l
 
− exp[ 2 i δl (k) ] − 1 Y1l (θ, ϕ) (1937)

A similar analysis can be applied to the scattering of an incident beam which is

down-spin polarized, giving similar results.

If the incident beam is un-polarized, the elastic scattering cross-section is

given in terms of the scattering amplitudes by
   
dσ
= | f (θ) |2 + | g(θ) |2 (1938)
dΩ
where the polar angle θ is the scattering angle.

12.12 An Electron in a Uniform Magnetic Field

We shall consider a Dirac electron in a constant magnetic field B = B (z) êz
aligned parallel to the z direction. The vector potential can be chosen such that

A = B x êy (1939)

We shall search for stationary states with energy E, where

 A   
φ i
ψ = exp − Et (1940)
φB h̄

338
In the standard representation, the energy eigenvalue equation is represented
by the set of coupled equations
q
( E − m c2 ) φA (r) = c σ . ( p̂ − A ) φB (r)
c
q
( E + m c2 ) φB (r) = c σ . ( p̂ − A ) φA (r)
c
(1941)

Substituting the expression for φB from the second equation into the first, one
obtains the second-order differential equation for φA
 2
2 2 4 A 2 q
(E − m c )φ = c σ . ( p̂ − A) φA (r)
c
 
2 q 2 q h̄
= c ( p̂ − A) − σ . B φA (r)
c c
 
= p̂ c + q B x − 2 q p̂y c B x − q c h̄ σ B φA (r)
2 2 2 2 2 (z)

(1942)

Since p̂y and p̂z commute with x, one can find simultaneous eigenstates of Ĥ,
p̂y and p̂z . Hence, the two-component spinor φA can be expressed as
 
A
φ (r) = exp i ky y + i kz z ΦA (x) (1943)

in which ΦA (x) is a two-component spinor which only depends on the variable

x. In this case, the exponential term can be factored out of the eigenvalue
equation. The resulting equation has the form
2
 
2 2 ∂
− h̄ c 2
+ ( c h̄ ky − q B x ) − q c h̄ B σ (z)
ΦA (x) = ( E 2 − m2 c4 − c2 h̄2 kz2 ) ΦA (x)
∂x2
(1944)
The equations decouple if the two-component spinor ΦA (x) can be taken to be
an eigenstate of the z-component of the spin operator

ΦA (x) = f (x) χσ (1945)

where
σ (z) χσ = σ χσ (1946)
in which the eigenvalues of σ (z) are denoted by σ. Therefore, the eigenvalue
equation can be reduced to
2 
∂2
 
c h̄ ky
− h̄2 c2 + ( q B )2 x− f (x) = ( E 2 − m2 c4 − c2 h̄2 kz2 + q c h̄ B σ ) f (x)
∂x2 qB
(1947)

339
which (apart from an overall scale factor) is formally equivalent126 to the (non-
relativistic) energy eigenvalue equation for a shifted harmonic oscillator, with
frequency 2 c | q | B. The modulus sign was inserted to ensure that the frequency
ωHO is positive. The energy eigenvalues are determined from

1
( E 2 − m2 c4 − c2 h̄2 kz2 + q c h̄ B σ ) = 2 | q | c h̄ B ( n + ) (1948)
2
Hence, for an electron with negative charge q = − e one finds that the positive-
energy eigenvalue is given by the solution
r
| e | h̄
E = c m2 c2 + h̄2 kz2 + ( 2 n + 1 + σ ) B (1949)
c
This expression has an infinite degeneracy as it is independent of the continuous
variable ky . It also has a discrete (two-fold) degeneracy between the levels with
quantum numbers (n, σ = 1) and (n + 1, σ = −1). The two-fold degeneracy
can be understood as a consequence of the generalized helicity σ . ( p̂ − qc A )
commuting with the Hamiltonian Ĥ. This results in the spin’s alignment with
the electron’s velocity being preserved, as the spin’s precession is precisely bal-
anced by the electron’s orbital precession. It should be noted that if the g factor
deviates from 2, and such an anomaly in the g factor is expected from Quantum
Electrodynamics and has been found in experiment, then this degeneracy will
be lifted. The calculated ( g − 2 ) anomaly for an electron is given by
     2  3  4
g − 2 1 α α α α
= − 0.3284986 + 1.17611 − 1.434 + ...
2 Theor 2 π π π π
(1950)
where  2
e
α = (1951)
h̄ c
is the fine structure constant. The experimentally determined value of the g
anomaly is found as
 
g − 2
= 0.0011659208 (1952)
2 Expt
126 The explicit (but dimensionally incorrect) analogy is obtained by setting the Harmonic

Oscillator mass, mHO , as

1
mHO =
2 c2
and then determine the frequency from
 2
qB
m2HO ωHO
2
=
c
.

340
and differs from the theoretical value in the last two decimal places127 . In
the non-relativistic limit, the expression for the relativistic energy eigenvalue
reproduces the expression for energies of the well-known Landau levels

h̄2 kz2
 
1 + σ | e | h̄ B
E ≈ m c2 + + (n + ) (1953)
2m 2 mc
which are doubly-degenerate.

12.13 Motion of an Electron in a Classical Electromag-

netic Field
Consider an electron in a classical electromagnetic field represented by the real
vector potential Aµ . For simplicity, electromagnetic field will be represented by
a plane wave defined over Minkowski space that depends on the phase φ defined
by
φ = kµ xµ (1954)
Hence, the vector potential is written as

Aµ = Aµ (φ) (1955)

The vector potential satisfies the Lorentz Gauge condition

∂µ Aµ = kµ Aµ (φ)0 = 0 (1956)

where the prime indicates differentiation with respect to φ. The classical vector
potential must satisfy the source-free wave equation

∂ν ∂ ν Aµ = kν k ν Aµ (φ)00 = 0 (1957)

which results in the condition

kν k ν = 0 (1958)

which is the dispersion relation for a free electromagnetic field.

The Dirac equation for a spin one-half particle with charge q can be used to
obtain the second-order differential equation
q2
 
2 µ q µ µ 2 2 q µ ν 0
− h̄ ∂µ ∂ − 2 i h̄ A ∂µ + 2 Aµ A − m c − i h̄ γ kµ γ Aν (φ) ψ = 0
c c c
(1959)
127 This discrepancy could indicate the importance of virtual processes in which heavy par-

ticle/antiparticle pairs are created. The (g − 2) anomalies for the muon and its anti-particle
have also been measured [G. W. Bennett et al., Phys. Rev. Lett. 92, 1618102 (2004).].
These experiments show that particles and anti-particles precess at the same rate. However,
the value of the (g − 2) anomaly is inconsistent with the theoretical prediction based on the
standard model of particle physics.

341
where ψ is the four-component Dirac spinor. In deriving this, the Lorenz gauge
condition has been used to re-write
     
µ ν µ ν µ,ν
γ γ ∂µ Aν ψ = γ γ ∂µ Aν ψ − g ∂µ Aν ψ

(1960)
in the diagonal terms.

Following Volkow128 , the solution of the second-order differential equation

can be found in the form
pµ xµ
 
ψ = exp − i F (φ) (1961)
h̄
where pµ is a four-vector and F (φ) is a four-component spinor. This form re-
duces to the form of a free particle solution when Aµ ≡ 0 in which case pµ
becomes the momentum of the free particle. The exponential form is unaltered
when the vector potential is non-zero since arbitrary multiples of the electro-
magnetic wave vector k can be added to the momentum of the free particle, in
which case pµ has a different interpretation. For a transverse polarized vector
potential describing an electromagnetic wave travelling in the ê3 direction, the
operators
p̂1 = i h̄ ∂1
p̂2 = i h̄ ∂2 (1962)
commute with the time-dependent Dirac Hamiltonian and are constants of mo-
tion. Although the particle’s energy and momentum operators do not commute
with the Hamiltonian, as these quantities are not conserved due to the interac-
tion with the field, the quantity
 
p̂3 − p̂0 = i h̄ ∂3 − ∂0 (1963)

does commutes with the Hamiltonian and, therefore, is conserved. The con-
servation of this quantity can be interpreted in terms of the energy absorbed
or emitted by the electron due to interaction with the classical electromagnetic
field being accompanied by the absorption or emission of similar amount of mo-
mentum129 . Despite the different interpretation of pµ in the presence of the
classical field, the four-vector pµ shall be chosen to satisfy the condition
pµ pµ = m2 c2 (1964)
which is the dispersion relation for a free electron130 .

128 D. M. Volkow, Zeit. für Physik, 94, 25 (1935).

129 For the quantized electromagnetic field, the absorption of a photon involves the absorption
of the energy and momentum given by the four-vector h̄ kµ , where kµ = (k, 0, 0, k).
130 If the condition on pµ is dropped, the function F (φ) will acquire an overall phase factor

that depends linearly on φ and on the constant value of pµ pµ − m2 c2 .

342
 The form of the wave function of eqn(1961) is to be substituted into the
second-order differential eqn(1959). It shall be noted that
Aµ ∂µ F (φ) = kµ Aµ F (φ)0 = 0
∂ µ ∂µ F (φ) = k µ kµ F (φ)00 = 0 (1965)
µ µ
since A satisfies the Lorenz gauge condition and k satisfies the dispersion
relation for electromagnetic waves in vacuum. On substituting the ansatz into
the second-order equation, using the above two equations and the choice of pµ
satisfying the free-electron dispersion relation, one finds that the second-order
equation reduces to a first-order differential equation for the spinor F (φ)
q2
 
q q
2 i h̄ pµ k µ F (φ)0 = 2 Aµ pµ − 2 Aµ Aµ + i h̄ γ µ kµ γ ν Aν (φ)0 F (φ)
c c c
(1966)
which only depends on φ since the exponential phase-factor which depends on
pµ xµ has been factored out. The first-order equation can be integrated w.r.t.
φ to yield
Z φ 
q γ µ kµ γ ν Aν
  
iq µ 0 1 q µ 0 0 0
F (φ) = exp − p Aµ (φ ) − A (φ ) A µ (φ ) dφ + F (0)
h̄ c pλ k λ 0 2 c c 2 pλ k λ
(1967)
where F (0) is an arbitrary constant four-component spinor. The exponential of
the matrix is defined in terms of its series expansion.
 Z φ   
iq µ 0 1 q µ 0 0 0
F (φ) = exp − p A µ (φ ) − A (φ ) A µ (φ ) dφ
h̄ c pλ k λ 0 2 c
q γ µ kµ γ ν Aν
 
× exp F (0) (1968)
c 2 pλ k λ
The above form can be simplified by expanding the last exponential factor due
to the identity  n
µ ν
γ kµ γ Aν = 0 (1969)

for all integers n such that n > 1. The identity can be proved by
γ µ kµ γ ν Aν γ τ kτ γ ρ Aρ = − γ µ kµ γ τ kτ γ ν Aν γ ρ Aρ + 2 g ν,τ Aν kτ γ µ kµ γ ρ Aρ
= − γ µ kµ γ τ kτ γ ν Aν γ ρ Aρ (1970)
where the first line follows by using the anti-commutation relations for the γ
matrices and the second line follows from applying the Lorenz gauge condition.
The expression can be further simplified by noting that on anticommuting the
first pair of γ matrices, one has
= − γ µ kµ γ τ kτ γ ν Aν γ ρ Aρ
= γ τ kτ γ µ kµ γ ν Aν γ ρ Aρ + 2 g µ,τ kµ kτ
= γ τ kτ γ µ kµ γ ν Aν γ ρ Aρ
= γ µ kµ γ τ kτ γ ν Aν γ ρ Aρ (1971)

343
the third line follows from the condition k µ kµ = 0 and the last line follows
from interchanging the first two pairs of summation indices. On comparing the
first and last lines, one notes that the right-hand side is zero. Therefore, one
has proved the identity

γ µ kµ γ ν Aν γ τ kτ γ ρ Aρ = 0 (1972)

Using, the above identity, the spinor F (φ) can be expanded as

 Z φ   
iq µ 0 1 q µ 0 0 0
F (φ) = exp − p Aµ (φ ) − A (φ ) Aµ (φ ) dφ
h̄ c pλ k λ 0 2 c
q γ µ kµ γ ν Aν
 
× Iˆ + F (0) (1973)
c 2 pλ k λ
Hence, the spinor solution of the second-order differential equation can be ex-
pressed as
q γ µ kµ γ ν Aν
  
S
ψ(x) = exp i Iˆ + F (0) (1974)
h̄ c 2 pλ k λ
where S given by
Z φ  
q 1 q µ 0
S = − pµ xµ − pµ Aµ (φ0 ) − A (φ ) Aµ (φ0 ) dφ0 (1975)
c pλ k λ 0 2 c
is the classical action of a particle moving in an electromagnetic field.

If the above equation is to be a solution of the Dirac equation, one needs to

exclude redundant solutions of the second-order equation. This can be achieved
by demanding that as r → ∞ one has Aµ → 0. In this limit, the above
solution reduces to
pµ xµ
 
ψ → exp − i F (0) (1976)
h̄
which satisfies the Dirac equation if
 
γ µ pµ − m c F (0) = 0 (1977)

Therefore, one demands that F (0) satisfies the above supplementary condition
which is the same as for a free particle. Hence, one can set
 
χσ
F (0) = NF   (1978)
p . σ
p(0) + m c
χσ

where the normalization constant is given by

s
p(0) + m c
NF = (1979)
2 p(0) V

344
The spectrum of eigenvalues of the electron’s energy can be found by Fourier
transforming the above solution with respect to time, which shows that the elec-
tron absorbs and emits radiation in multiples of h̄ ω. The Volkov solutions have
been used to describe the Compton scattering of electrons by intense coherent
laser beams, and is also the basis of the strong-field approximation sometimes
found useful in atomic physics131 .

The current density is derived from the expression

†
jµ = c ψ γµ ψ (1980)

Since the Dirac adjoint spinor is given by

q γ ν Aν γ µ kµ
   
† † S
ψ = F (0) Iˆ + exp − i (1981)
c 2 pλ k λ h̄

the current density is evaluated as

q pν Aν q 2 Aν Aν
  
c q µ
j µ = (0) pµ − A + kµ − (1982)
p V c c k λ pλ c2 2 k λ pλ

Hence, the current is composed of a constant component pµ and an oscillatory

component form the vector potential, and an oscillatory component which is
second order in the vector potential. This implies that the electromagnetic field
has measurable consequences. For a vector potential Aµ which is a periodic
function with a time-averaged value of zero, the time-averaged current density
is given by  2
q Aν Aν
 
µ
c µ µ
j = (0) p − k (1983)
p V c2 2 k λ pλ
which shows that the electromagnetic wave does not drop out from time-averaged
quantities.

12.14 The Limit of Zero Mass

The Dirac equation has the form

γ µ p̂µ ψ = m c ψ (1984)

where the γ matrices are any set of matrices which satisfy the anti-commutation
relations
γ µ γ ν + γ ν γ µ = 2 g µ,ν Iˆ (1985)
131 L. V. Keldysh, Zh. Eksp. Teor. Fiz. 47, 1945 (1964). [Sov. Phys. J.E.T.P. 20, 1307

(1965).]
F. H. M. Faisal, J. Phys. B 6, L89 (1973).
H. R. Reiss, Phys. Rev. A 22, 1786 (1980).

345
The Dirac equation is independent of the specific representation of the γ matri-
ces. We have chosen the representation
 
I 0
γ (0) = (1986)
0 −I

and
σ (i)
 
(i) 0
γ = (1987)
−σ (i) 0
where σ (i) are the Pauli-matrices. This is the standard representation.

We can find other representations which differ through unitary transforma-

tions
ψ 0 = Û ψ (1988)
where the explicit form of the γ matrices transform via

γ µ0 = Û γ µ Û † (1989)

and the Dirac adjoint is transformed via

†0 †
ψ = ψ 0 γ (0)0 (1990)

These unitary transformations of the gamma operators keep matrix elements of

the form Z
d3 r ψ † Â ψ (1991)

invariant.

The chiral representation is found by performing the unitary transform

 
1 I −I
Û = √ (1992)
2 I I

starting with the standard representation. In the chiral representation, the γ

matrices have the form  
(0)0 0 I
γ = (1993)
I 0
and
σ (i)
 
(i)0 0
γ = (1994)
−σ (i) 0
The components of the wave function in the chiral representation ψ 0 are denoted
as  L0 
0 φ
ψ = (1995)
φR0

346
The components φL and φR are related to the components of ψ in the standard
representation via
 L0   A
φ − φB

φ 1
= √ (1996)
φR0 2 φA + φB
The chiral representation is particularly useful for the description of massless
spin one-half particles, such as might be the case for the neutrino. The neutrino
masses are extremely small. The masses have evaded direct experimental mea-
surement. However, direct measurements have set upper limits on the masses
which decrease with time132 . In this case, with the limit m → 0, the Dirac
equation takes the form
   
∂
0 ∂t + c σ . ∇  L0 
  φ
   = 0 (1997)
φR0

∂
 
∂t − c σ . ∇ 0

Hence, the Dirac equation for a massless free particle reduces to two uncoupled
equations, each of which are equations proposed by Weyl133
 
∂
+ c σ . ∇ φR0 = 0 (1998)
∂t
and  
∂
− c σ . ∇ φL0 = 0 (1999)
∂t
The Weyl equation describes a spin one-half massless particle by a two com-
ponent spinor wave function. The Weyl equation violates parity invariance.
The Weyl equation was considered to be un-physical until the discovery of the
(anti-)neutrino134 and the associated violation of parity invariance135 . After the
parity violation of the weak interaction was established, the Weyl equation was
adopted to describe the neutrino136 .

Inexplicably nature seems to have selected the Weyl equation for φL , but not
φR to describing neutrinos. The solutions of the Weyl equation for free particles
 
∂
− c σ . ∇ φL = 0 (2000)
∂t
132 L. Langer and R. Moffat, Phys. Rev. 88, 689 (1952).

V. A. Lyubimov, F. G. Novikov, V. Z. Nozik, F. F. Tretyakov, and V. S. Kosik, Phys. Lett.

94B, 266 (1980).
A. I. Belesev et al., Phys. Lett. 350, 263 (1995).
133 H. Weyl, Zeit. für Physik, 56, 330 (1929).
134 C. L. Cowan Jr., F. Reines, F. B. Harrison, H. W. Kruse and A. D. McGuire, Science

124, 103 (1956).

F. Reines and C. L. Cowan Jr., Phys. Rev. 113, 273 (1959).
135 C. S. Wu, E. Ambler, R. W. Hayward, D. D. Hoppes and R. F. Hudson, Phys. Rev. 108,

1413 (1957).
136 T. D. Lee and C. N. Yang, Phys. Rev. 105, 1671 (1957).

A. Salam, Nuovo Cimento 5, 299 (1957).

L. Landau, Nuclear Phys. 3, 127 (1957).

347
can be written as
 (0)   
u 1 i
φL = √ exp − ( E t − p . r ) (2001)
u(1) V h̄

Since helicity is conserved, one can choose the direction of p as the axis of
quantization. The positive-energy solution is given by
   
L 0 1 i
φ− = √ exp − (Et − pz) (2002)
1 V h̄
which has negative helicity and has energy given by

E− = c p (2003)

The negative-energy solution is given by

   
L 1 1 i
φ+ = √ exp − (Et − pz) (2004)
0 V h̄
which has positive helicity and the energy is given by

E+ = − c p (2005)

This negative-energy solution will describe anti-particles. The Weyl equation for
φR has a positive-energy solution with positive helicity, and a negative-energy
solution with negative helicity. Since only neutrinos with negative helicity are
observed in nature, only φL is needed. The anti-neutrinos have positive helicity
and are represented by φR .

E Λ=+1 Λ=−1
R L Λ=−1
φ φ ν∗ ν

Elementary Excitations
Λ=+1

Figure 61: The dispersion relations for φL and φR . The elementary excitations
are the negative-helicity neutrino ν and a positive-helicity anti-neutrino ν.

The Neutrino

The neutrino was postulated by Pauli to balance energy and momentum

conservation in beta decay. In beta decay, it had been observed that neutron

348
decay products included a proton and an electron. However, it was observed
that the emitted electron had a continuous range of kinetic energies. Therefore,
another neutral particle must have been emitted in the decay. This particle was
termed the anti-neutrino, and the reaction can be written as
n → p + e− + ν e (2006)
Conservation of angular momentum requires that the neutrino has a spin of h̄2 .
Furthermore, since an energy of 1.2934 MeV is released in the transformation of
a neutron to a proton, and since sometime the decay processes produce electrons
which seem to take up all the released energy, the neutrino was suggested as
having zero mass. An upper limit on the neutrino’s mass of a few eV follows
from the Fermi-Kurie plot137 . The Fermi-Kurie plot of the electron energy

200
2 1/2

150
[N(p)/Fp ]

100

50

0
0 5 10 15 20
Energy [keV]

Figure 62: The Fermi-Kurie plot of the energy distribution of the electrons
emitted in the beta decay of tritium, 3 H → 3 He + e− + ν e . The decay releases
18.1 keV. It is seen that the electrons produced in the decay process have a
non-zero probability for carrying off most of the released energy. Hence, one
concludes that the anti-neutrinos are almost massless. The dashed blue curve
is the curve expected if the neutrino had a mass of 3 keV.

distribution is based on the phase space available for the emission of the electron
and anti-neutrino138 . The joint phase-space available for the electron of four-
momentum (Ee /c, p) and the anti-neutrino of four-momentum is (Eν /c, q) is
proportional to the factor
Z
dΓ = dp p2 dq q 2 δ(E − Ee (p) − Eν (q))
Z
p Ee (p) q Eν (q)
= dEe (p) 2
dEν (q) δ(E − Ee (p) − Eν (q))
c c2
137 L. Langer and R. Moffat, Phys. Rev. 88, 689 (1952).

V. A. Lyubimov, F. G. Novikov, V. Z. Nozik, F. F. Tretyakov, and V. S. Kosik, Phys. Lett.

94B, 266 (1980).
A. I. Belesev et al., Phys. Lett. 350, 263 (1995).
138 E. Fermi, Zeit. für Physik, 88, 161 (1934).

F. N. D. Kurie, J. R. Richardson and H. C. Paxton, Phys. Rev. 48, 167 (1935).

349
 
1 p 
2 − m2 c4 E (q) 
= 5
dE e (p) p E e (p) E ν (q) ν ν
c 
Eν (q)=E−Ee (p)
1 p
= dEe (p) p Ee (p) ( E − Ee (p) )2 − m2ν c4 ( E − Ee (p) )
c5
(2007)

where, since the anti-neutrino’s trajectory is unobservable, its momentum is in-

tegrated over. This phase-space factor partially governs the energy distribution
of the emitted electrons. The second to last factor in the accessible volume
of phase-space contains the dependence on the anti-neutrino’s mass mν and it
is this factor which is high-lighted by the Fermi-Kurie plot. The plot is de-
signed to exhibit a linear energy variation until the line cuts the E-axis, if the
anti-neutrino is massless. On the other hand, if the anti-neutrino has a finite
mass, the line should curve over and cut the E-axis vertically. In this case, the
anti-neutrino mass would be determined by the difference between the linearly
extrapolated intercept and the actual intercept.

The process of beta decay does not conserve parity. The non-conservation
of parity was discovered in the experiments of C. S. Wu et al.139 . In these
experiments, the spin of a 60 Co nucleus was aligned with a magnetic field. The
spin S = 5h̄ 60 Co nucleus decayed into a spin S = 4h̄ 60 N i nucleus by emitting
an electron and an anti-neutrino.
60
Co → 60
N i + e− + ν e (2008)

Since angular momentum is conserved, the spin of the electron and the anti-
neutrino initially must both be aligned with the field. In the experiment, the
angular distribution of the emitted electrons was observed. Because the helicity
of the electrons is conserved, the angular distribution of the electrons can be
used to prove that the electrons all have negative helicity, and hence it is inferred
that the anti-neutrinos should have positive helicity. Since helicity should be
reversed under the parity operation, and since only negative helicity electrons
are observed, the process is not invariant under parity. Hence, parity is not
conserved.

The electrons that are emitted in beta decay have negative helicities. If
the momentum of an emitted electron is given by (p, θp , ϕp ), then its helicity
operator is  
cos θp sin θp exp[−iϕp ]
Λp = (2009)
sin θp exp[+iϕp ] − cos θp
The helicity operator has eigenstates χ given by

Λp χ± ±
θp = ± χθp (2010)
139 C. S. Wu, E. Ambler, R. W. Hayward, D. D. Hoppes and R. F. Hudson, Phys. Rev. 108,

1413 (1957).

350
which are determined as
!
θ ϕ
cos 2p exp[−i 2p ]
χ+
θp = θ ϕ
sin 2p exp[+i 2p ]
!
θ ϕ
− sin 2p exp[−i 2p ]
χ−
θp = θ ϕ (2011)
cos 2p exp[+i 2p ]

Since angular momentum is conserved and the emitted electrons only have neg-
ative helicity, the angular distribution of the emitted electrons is proportional to
the square of the overlap of the initial electron spin-up spinor with the negative
helicity spinors

† − 2 θp
| χ+
θ=0 χθp | = sin2
2
1
= ( 1 − cos θp ) (2012)
2
which is in exact agreement with the experimentally observed distribution. From
the distribution of emitted electrons one is led to expect that the anti-neutrino
has positive helicity.

The helicity of the neutrino was measured in an experiment performed by

Maurice Goldhaber et al.140 . In the experiment, a 152 Eu nucleus with J = 0
140 M. Goldhaber, I. Grodzins, A. W. Sunyar, Phys. Rev. 109, 1015 (1958).

νe
S=h/2
e-
S=h/2

Co
S=5h Ni
S=4h

Figure 63: The spin S = 5h̄ of the Co nucleus is aligned with the magnetic field.
The Co undergoes beta decay to N i which has S = 4h̄ by emitting an electron
e− and an anti-neutrino ν e . The spin of the electron and the anti-neutrino
produced by the decay must initially be aligned with the magnetic field, due to
conservation of angular momentum.

351
 1.2

0.8

I(θ)
0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1

θ/π

Figure 64: The angular distribution of the emitted electron in the beta decay
experiment of Wu et al.

captures an electron from the K-shell and decays to the excited state of a 152 Sm
nucleus with angular momentum J = h̄ and emits a neutrino.
152
Eu + e− → 152
Sm∗ + νe (2013)

The J = h̄ excited state of Sm∗ subsequently decays into the J = 0 ground

state of Sm by emitting a photon.
152
Sm∗ → 152
Sm + γ (2014)

Goldhaber et al. measured the photons with the full Doppler shift, from which
they were able to infer the direction of the recoil of the nucleus. The photons

e-

Eu

Sm*
νe

J=1

γ Sm νe

Λγ=−1 J=0

Figure 65: A schematic depiction of the experiments of Goldhaber et al. which

determined the helicity of the neutrino.

were observed to be right-circularly polarized, which corresponds to having a

negative helicity. Therefore, the photon’s spin was parallel to the momentum of
the emitted neutrino. Since the ground state of Sm has zero angular momentum,
the excited state of the Sm∗ nucleus must have had its angular momentum

352
oriented along the direction of motion of the emitted neutrino. Since the sum of
the angular momentum of the excited state (J = h̄) and the emitted neutrino
must equal the spin of the captured electron h̄2 , the neutrino must have its spin
oriented anti-parallel to the angular momentum of the Sm∗ nucleus. Hence, the
neutrino has negative helicity.

12.15 Classical Dirac Field Theory

The Dirac Lagrangian density is given by
 
† q
L = ψ i h̄ c γ µ ( ∂µ + i Aµ ) − m c2 ψ (2015)
c h̄

which, since ψ † and ψ are independent, the momentum conjugate to ψ is

1 ∂L
Π =
c ∂(∂0 ψ)
†
= i h̄ ψ γ (0)
= i h̄ ψ † (2016)

The momentum conjugate to ψ † vanishes

1 ∂L
Π† =
c ∂(∂0 ψ † )
= 0 (2017)

The Lagrangian equation of motion is found from the variational principle which
states that the action is extremal with respect to ψ and ψ † . The condition that
the action is extremal with respect to variations in ψ † leads to the Dirac equation
q
i h̄ γ µ ( ∂µ + i Aµ ) ψ = m c ψ (2018)
c h̄
after the resulting equation has been multiplied by a factor of γ (0) . On making
a variation of the action with respect to ψ, one finds the Hermitean conjugate
equation
q † †
− i h̄ c ( ∂µ − i Aµ ) ψ γ µ − m c2 ψ = 0 (2019)
c h̄
That this is the Hermitean conjugate of the Dirac equation can be shown by
taking its Hermitean conjugate, which results in
q
i h̄ c γ µ† γ (0) ( ∂µ + i Aµ ) ψ − m c2 γ (0) ψ = 0 (2020)
c h̄
The above equation can be reduced to the conventional form by multiplying by
γ (0) and by using the identities

γ (0) γ (0) = Iˆ
γ (0) γ µ† γ (0) = γµ (2021)

353
Hence, the equation found by varying ψ is just the Hermitean conjugate of the
Dirac equation
q
i h̄ γ µ ( ∂µ + i Aµ ) ψ = m c ψ (2022)
c h̄
Furthermore, it is surmised that the starting Lagrangian is appropriate to de-
scribe the Dirac field theory.

The Hamiltonian density H is determined from the Lagrangian by the usual

Legendre transformation process
H = c Π ∂0 ψ + c Π† ∂0 ψ † − L
= i h̄ c ψ † ∂0 ψ − L
†
= i h̄ c ψ γ (0) ∂0 ψ − L
† q †
= − i h̄ c ψ γ . ( ∇ − i A ) ψ + ψ ( m c2 + q γ (0) A(0) ) ψ
c h̄
(2023)
where the relation between the covariant components of the vector potential to
the contravariant components A(i) = − A(i) has been used in the last line.
The result is identifiable with the Hamiltonian density that appears in the usual
expression for the quantum mechanical expectation value for the energy for the
Dirac electron.

The set of conserved quantities can be obtained from Noether’s theorem.

The momentum-energy tensor T µ ν is given by
∂L ∂L
T µν = ∂ν ψ + ∂ν ψ † − δ µ ν L (2024)
∂(∂µ ψ) ∂(∂µ ψ † )
which is evaluated as
†
T µν = i h̄ c ψ γ µ ∂ν ψ − δ µ ν L
 
† † q †
= i h̄ c ψ γ µ ∂ν ψ + δ µ ν − i h̄ c ψ γ ρ ( ∂ρ + i Aρ ) ψ + m c2 ψ ψ
c h̄
(2025)
Hence, one finds the energy density T 0 0 is given by
† q †
T 0 0 = − i h̄ c ψ γ . ( ∇ − i A ) ψ + ψ ( m c2 + q γ (0) A(0) ) ψ
c h̄
q
= − i h̄ c ψ † α . ( ∇ − i A ) ψ + ψ † ( β m c2 + q A(0) ) ψ
c h̄
(2026)
which is the Hamiltonian density H. On integrating over all space, one sees
that the energy of the Dirac Field is equal to the expectation value of the Dirac
Hamiltonian operator
Z Z
3 0
d rT 0 = d3 r ψ † Ĥ ψ (2027)

354
Likewise, (c times) the momentum density T 0 j is found from
†
T 0j = i h̄ c ψ γ (0) ∂j ψ
= i h̄ c ψ † ∂j ψ
= c ψ † p̂j ψ (2028)

where the partial derivative has been identified with the covariant momentum
operator. Hence, the contravariant component of the momentum is given by
∂
T 0,j = − i h̄ c ψ † ψ
∂xj
= c ψ † p̂(j) ψ (2029)

where the usual (contravariant) momentum operator is defined as

∂
p̂(j) = − i h̄ (2030)
∂xj
Therefore, the j-th component of the momentum is given by
Z Z
1
P (j) = d3 r T 0,j = d3 r ψ † p̂(j) ψ (2031)
c
which is equal to the expectation value of the momentum operator.

One can also determine the conserved Noether charges by noting that the
Lagrangian is invariant under a global gauge transformation
 
0
ψ → ψ = exp + i ϕ ψ
 
∗
∗ 0∗
ψ →ψ = exp − i ϕ ψ (2032)

where ϕ is a constant real number. The infinitesimal global gauge transforma-

tion produces a variation in the (independent) fields

δψ = + i δϕ ψ
δψ ∗ = − i δϕ ψ ∗ (2033)

Since the Lagrangian is invariant under the transformation, then

δL = 0 (2034)

so we have

0 = δ L       
∂L ∂L ∗ ∂L ∂L
= δψ + δψ + δ(∂µ ψ) + δ(∂µ ψ ∗ )
∂ψ ∂ψ ∗ ∂(∂µ ψ) ∂(∂µ ψ ∗ )
(2035)

355
After substituting the Euler-Lagrange equations for the derivatives w.r.t. the
fields ψ and ψ ∗ , the variation is expressed as
    
∂L ∂L ∗
0 = ∂µ δψ + δψ (2036)
∂(∂µ ψ) ∂(∂µ ψ ∗ )
For an arbitrary gauge transformation through the fixed infinitesimal angle δϕ,
this condition becomes
    
∂L ∂L ∗
0 = i δϕ ∂µ ψ − ψ (2037)
∂(∂µ ψ) ∂(∂µ ψ ∗ )
Hence, one finds that there is a current j µ which satisfies the continuity equation
∂µ j µ = 0 (2038)
where (apart from the infinitesimal constant of proportionality) the current is
given by     
∂L ∂L ∗
j µ ∝ i δϕ ψ − ψ (2039)
∂(∂µ ψ) ∂(∂µ ψ ∗ )
For the Dirac Lagrangian, the second term is identically zero and the first term
is non-zero. Hence, on adopting a conventional normalization, the conserved
current is identified as
†
jµ = c ψ γµ ψ (2040)
This is the same expression for the conserved current that was previously derived
for the one-electron Dirac equation. Hence, the one-particle Dirac equation
yields the same expectation values and obeys the same conservation laws as the
(classical) Dirac field theory.

12.15.1 Chiral Gauge Symmetry

In the limit of zero mass, the Dirac Lagrangian takes the form
†
L = i h̄ c ψ γ µ ∂µ ψ (2041)
Starting with the standard representation and making the unitary transform
 
1 I −I
Û = √ (2042)
2 I I
one finds that in the chiral representation the Dirac Lagrangian reduces to
 
µ µ
L = i h̄ c φ†L σL ∂µ φL + φ†R σR ∂µ φR (2043)

where φL and φR are two-component Dirac spinors and the two sets of quantities
σ µ and σ̃ µ are expressed in terms of the Pauli matrices as
µ
σL = ( σ0 , − σ )
µ
σR = ( σ0 , σ ) (2044)

356
 µ µ
The difference between σL and σR reflect the different chirality of φL and φR . In
the absence of the mass term, the Dirac Lagrangian possesses two independent
scalar gauge transformations. These transformations corresponds to the global
gauge transformations
 
L L0 L
φ → φ = φ exp i θL
 
R R0 R
φ → φ = φ exp i θR (2045)

where θL and θR are independent angles. The Lagrangian has a U (1) × U (1)
gauge symmetry. The presence of a mass term would couple the two fields and
reduce the gauge transformation to one in which θR = θL .

In the chiral representation, the Hermitean matrix defined by

γ (4) = i γ (0) γ (1) γ (2) γ (3) (2046)

takes the form  

(4) −I 0
γ = (2047)
0 I
The general gauge transformations for the massless fermion can be expressed as
the product of two independent transformations
       
θL + θR ˆ θR − θL
ψ → ψ 0 = exp i I exp i γ (4) ψ (2048)
2 2

where ψ is a four-components spinor

φL
 
ψ = (2049)
φR

The first factor represents the usual global gauge transformation for the Dirac
Lagrangian with finite mass. This transformation yields the usual conserved
four-vector current jVµ defined by
†
jVµ = c ψ γ µ ψ (2050)

The second factor is specific to the Dirac Lagrangian with zero mass. It is
called the chiral transformation or axial U (1) transformation. Using the anti-
commutation relation
{ γ (4) , γ µ }+ = 0 (2051)
one can show that the exponential factor in the chiral gauge transformation has
the property that
       
µ θR − θL (4) θR − θL (4)
γ exp i γ = exp − i γ γ µ (2052)
2 2

357
This property can be used to show that the Lagrangian is invariant under the
chiral transformation because
       
† θR − θL θR − θL
ψ 0 γ µ ∂µ ψ 0 = ψ † exp − i γ (4) γ (0) γ µ ∂µ exp i γ (4) ψ
2 2
= ψ † γ (0) γ µ ∂µ ψ
†
= ψ γ µ ∂µ ψ (2053)

which involves two commutations. Since the massless Dirac Lagrangian is invari-
ant under the chiral transformation, Noether’s theorem shows that the current
µ †
jA = c ψ γ µ γ (4) ψ (2054)

is conserved. This conserved current transforms like a vector under proper

orthochronous Lorentz transformations but does not transform as a vector under
improper orthochronous transformations. Therefore, the current is an axial
(0)
current. The conserved axial density jA is given by
(0) †
jA = ψ γ (0) γ (4) ψ
= ψ † γ (4) ψ
= − φ†L φL + φ†R φR (2055)

which is the difference between the number of particles with positive helicity
and the number of particles with negative helicity.

In the presence of a mass m, the Dirac Lagrangian in the chiral representation

is
   
µ µ
L = i h̄ c φ†L σL ∂µ φL + φ†R σR ∂µ φR − m c2 φ†L φR + φ†R φL
(2056)
and one finds that the axial current is not conserved because the mass term is
not invariant and acts like a current source
µ 2 m c † (4)
∂µ jA = i ψ γ ψ (2057)
h̄
To summarize, just like the Proca equation yields a zero mass for the photon
if one imposes U (1) gauge invariance on the electromagnetic field141 , the neu-
trino must have zero mass if one imposes a global U (1) chiral gauge invariance.
Furthermore, the existence of conservation of chirality for the massless neutrino
implies that the weak interaction must involve a coupling proportional to a fac-
tor of either (Iˆ + γ (4) ) or (Iˆ − γ (4) ).

141 Schwinger has noted that the condition of gauge invariance does not necessarily result

in the photon being massless. He argued that if the electromagnetic coupling strength were
larger, the photon could have finite mass. [J. Schwinger, “Gauge Invariance and Mass”,
Physical Review, 125, 397 (1962).]

358
Exercise:

By considering an infinitesimal chiral gauge transformation on the Lagrangian

for massive Dirac particles, determine δL and show that this leads to the axial
µ
current jA not being conserved.

12.16 Hole Theory

The negative-energy solutions of the Dirac equation lead to the conclusion that
one-particle quantum mechanics is an inadequate description of nature. In clas-
sical mechanics, the dispersion relation for a free particle is found to be given
by p
E = ± m2 c4 + p2 c2 (2058)
The negative-energy states found in classical mechanics can be safely ignored.
The rational for ignoring the negative-energy states in classical mechanics is
that, the dynamics is governed by a set of differential equations which result
in the classical variables changing in a continuous fashion. Since the particle’s
energy can only change in a continuous fashion, there is no mechanism which al-
lows it to connect with the negative branch of the dispersion relation. However,
in quantum mechanics, particles can make discontinuous transitions between
different energy levels, by emitting photons. Hence, if one has a single electron
in a positive-energy state where E > m c2 , this state would be unstable to
the electron making a transition to a negative-energy state which occurs with
the simultaneous emission of photons which carry away an energy greater than
2 m c2 . The transition rate for such process is quite large, therefore, one might
conclude that positive-energy particles should not exist in nature. Furthermore,
if one does have particles in the negative-energy branch, they might be able to
further lower their energies by multiple photon emission processes. Hence, the
states of negative-energy particles with finite momenta could be unstable to
states in which the momentum has an infinite value.

Dirac noted that if the negative-energy states were all filled, then the Pauli
exclusion principle would prevent the decay of positive-energy particles into
the negative-energy states. Furthermore, in the absence of any positive-energy
particles, the Pauli exclusion principle would cause the set of particles in the
negative-energy state to be completely inert. In this picture, the filled sea
of negative-energy states would represent the physical vacuum, and would be
unobservable in experiments. For example, if charge is measured, it is the
non-uniform part of the charge distribution that is measured, but the infinite
number of particles in the negative-energy states do produce a uniform charge
density. Likewise, when energies are measured, the energy is usually measured
with respect to some reference level. For the case of a vacuum in which all
the negative-energy states are filled with electrons, the measured energies cor-
respond to energy differences and so the infinite negative energy of the vacuum

359
 2
Unoccupied Positive Energy States

2 1
E/mc

-1

Occupied Negative Energy States

-2

Figure 66: A cartoon depicting the vacuum for Dirac’s Hole Theory, in which
the negative-energy states are filled and the positive-energy states are empty.

should cancel. Therefore, Dirac postulated that the vacuum consists of the
state in which all the negative-energy states are all filled with electrons142 . Fur-
thermore, physical states correspond to the states were a relatively few of the
positive-energy states are filled with electrons and a few negative states are un-
occupied. In this case, the electrons in the positive-energy states are identified
with observable electrons, and the unfilled states or holes in the distribution of
negative-energy states are also observable. These holes are known as positrons
and are the anti-particles of the electrons. The properties of a positron are
found by computing the difference between the property for a state with an
absent negative-energy electron and the property of the vacuum state.

We shall assume that the vacuum contains of N electrons which completely

fill all the N negative states and, for simplicity of discussion, the effect of cou-
pling to the electromagnetic field can be ignored. Then the charge of a positron
qp is the difference between the charge of the vacuum with one missing electron,
and the charge of the vacuum

q p = ( N − 1 ) q e − N qe (2059)

Therefore, one finds that the positron has the opposite charge to that of an
electron
qp = − qe (2060)
142 P. A. M. Dirac, Proc. Roy. Soc. A 126, 360 (1930).

360
Hence, the positron has a positive charge. Likewise, the energy of the vacuum
in which all the electrons occupy all the negative-energy states is denoted by
E0 . The positron energy will be denoted as Ep (pe ). The positron corresponds
all states with negative energy being filled except for the state with the energy
p
Ee (pe ) = − m2 c4 + p2e c4 (2061)

which is unfilled. The positron energy is defined as the energy difference

 
Ep (pp ) = E0 − Ee (pe ) − E0

= − Ee (pe )
p
= m2 c4 + p2e c4 (2062)

Therefore, the positron corresponds to a particle with a positive energy. From

this it is seen that the rest mass energy of the positron is identical to the
rest mass energy of the electron. If the vacuum corresponds to a state with
momentum P 0 and if the negative-energy state with momentum pe is unfilled,
then the momentum of the positron would be given by pp where
 
pp = P 0 − pe − P0

= − pe (2063)

Hence, the momentum of the positron is the negative of the momentum of the
missing electron
p p = − pe (2064)
Likewise, the spin of the positron is opposite to the spin of the missing electron,
etc. The velocity of an electron is defined as the group velocity of a wave packet
of momentum pe . Hence, one finds the velocity of the negative energy-electron
from
∂
ve = Ee (pe )
∂pe
pe c2
= − q (2065)
m2 c4 + p2e c2

while the velocity of the positron is given by

∂
vp = Ep (pp )
∂pp
pp c2
= q
m2 c4 + p2p c2

361
Table 18: The relation between properties of Negative Energy Electron and
Positron States.

Particle Charge Energy Momentum Spin Helicity Velocity

h̄
Electron −|e| −|E| +p + 2 σ σ.p v

h̄
Positron +|e| +|E| −p − 2 σ σ.p v

pe c2
= − q
m2 c4 + p2e c2
= ve (2066)

Therefore, the positron and the negative-energy electron states have the same
velocities.

Hole theory provides a simple description of the relation between a negative-

energy state and anti-particle states. Mathematically, this relation is expressed
in terms of the charge conjugation transformation. A unique signature of the
hole theory is that a positive-energy electron can make a transition to an unfilled
negative-energy state emitting radiation, which corresponds to the process in
which a electron-positron pair annihilates143

e + e → 2γ (2067)

In this process, it is necessary that the excess energy be carried off by two
photons if the energy-momentum conservation laws are to be satisfied. Likewise,
by supplying an energy greater than a threshold energy of 2 m c2 , it should be
possible to promote an electron from a negative-energy state, thereby creating
an electron-positron pair. Since it is unlikely that more than one photon can
be absorbed simultaneously, electron-positron pair creation only occurs in the
vicinity of a charged nucleus which can carry off any excess momentum.

γ → e + e (2068)

The positively charged electron, predicted by Dirac, was found experimentally

143 P. A. M. Dirac, Proc. Camb. Phil. Soc. 26, 361 (1930).

362
 2
Unoccupied Positive Energy States

1
2
E/mc

(k,α)

-1

Occupied Negative Energy States

-2

Figure 67: A cartoon depicting electron-positron production in Dirac’s Hole

Theory. In this case, an incident γ-ray produces an electron-hole pair. The
process is restricted to occur in the vicinity of heavy particles that can act as a
momentum sinks.

by Anderson144 and the electron-positron creation145 and annihilation pro-

cesses146 were observed shortly afterwards.

Dirac commented147 that in scattering processes involving low-energy elec-

trons, such as Thomson scattering, it is essential that negative-energy states
appear as virtual states, if one is to recover the correct scattering cross-section
in the non-relativistic limit. The involvement of negative-energy states in the
scattering of light is a consequence that, in the standard representation, the
lower two-component spinor in the Dirac wave function for a free (positive-
energy) electron vanishes in the low energy limit, and also because the coupling
144 C. D. Anderson, Phys. Rev. 43, 491 (1933). Anderson observed the curved trajectories

of the charged particles in a cloud chamber in the presence of a magnetic field. Anderson
inferred the charge of the particles from their direction of motion. The insertion of a lead
plate in a cloud chamber caused the particles to lose energy on one side of the plate which
was observed as a change in the radius of curvature of the particle’s track. Therefore, the
examination of the radius of curvature of the track on both sides of the plate allowed the
direction of motion to be established.
145 P. M. S. Blackett and G. P. S. Occhialini, Proc. Roy. Soc. A 139, 688 (1933). These

authors were the first who correctly identified the positively charged particle as the anti-
particle of the electron, in full accord with the predictions of Dirac’s hole theory.
146 J. Thibaud, Phys. Rev, 35, 78 (1934).
147 P. A. M. Dirac, Proc. Roy. Soc. A126, 360 (1930).

363
to the radiation field is produced by γ (0) γ . A. The interaction operator can
be expressed as
 
0 σ
ĤInt = − q α . A = − q .A (2069)
σ 0

which only connects the upper and lower two-component spinors of the initial
and final states ψn and ψn0 . Hence, as light scattering processes are at least of
second-order in A, the intermediate state ψn00 must involve a negative-energy
electron state. Since the Pauli exclusion principle forbids the occupation of
the filled negative-energy states, hole theory ascribes the intermediate states
as involving virtual electron-positron creation and annihilation processes. This
shows that, even for processes which appear to involve a single electron in the
initial and final states, one must abandon single-particle quantum mechanics
and adopt a multi-particle description. Therefore, a purely single-particle de-
scription is inadequate and one must consider a many-particle description such
as quantum field theory.

12.16.1 Compton Scattering

We shall consider Thomson scattering of light by free electrons. In this process,
light is scattered from the initial state (k, α) to the final state (k, α) and the
(positive-energy) electron makes a transition from the initial state (q, σ) to its
final (positive-energy) state (q 0 , σ 0 ). The Thomson scattering cross-section of
light is given by the expression
2
V ωk 0
  
dσ
= | M |2 (2070)
dΩk0 2 π h̄ c2

where the matrix element M are determined from

X  < q 0 , k 0 , α0 | ĤInt | q 00 > < q 00 | ĤInt | q, k, α >
M =
00
( Eq + h̄ ωk − Eq00 )
q

< q 0 , k 0 , α0 | ĤInt | q 00 , k, α, k 0 , α0 > < q 00 , k, α, k 0 , α0 | ĤInt | q, k, α >


+
( Eq − Eq00 − h̄ ωk0 )
(2071)

and where q indicates all the quantum numbers of a positive-energy free electron
state. The sum over q 00 represents a sum over all possible intermediate states
of the electron, no matter whether they are positive or negative-energy states.
The matrix element M is composed of a coherent superposition of matrix el-
ements for virtual processes which represent the absorption of a photon (k, α)
followed by the subsequent emission of a photon (k 0 , α0 ) and the process where
the emission of light precedes the absorption process.

364
 (k,α) (k',α')
(k,α) (k',α')

q''
q''
q' q q'
q

Figure 68: Processes involving negative electron states q 00 which contribute to

Compton scattering.

Since the basis set is composed of momentum eigenstates, the evaluation

of the spatial integration in the matrix elements of the interaction results in
the condition of conservation of momentum. Hence, for the process where the
photon (k, α) is absorbed before the emission of the photon (k 0 , α0 ), the momenta
are restricted by

k + q = q 00
q 00 = k 0 + q 0 (2072)

which leads to the identification of the momentum of the intermediate and final
states as

q 00 = q + k
q0 = q + k − k0 (2073)

In the second process, where the emission process precedes the absorption, con-
servation of momentum yields

k + q = k + k 0 + q 00
k + k 0 + q 00 = k 0 + q 0 (2074)

which yields

q 00 = q − k 0
q0 = q + k − k0 (2075)

The limit in which the initial electron is at rest q = 0 shall be considered. The
momenta of the incident and scattered photon will be assumed sufficiently low
so that the momentum of the electron in the intermediate state can be neglected
since q 00 ≈ 0. That is, the Compton scattering process will be consider in the
limit k → 0 and k 0 → 0.

365
 If the initial (positive-energy) electron is stationary and has spin σ, its wave
function can be represented by the Dirac spinor
 
1 χσ
ψσ,q (r) = √ (2076)
V 0
Because the interaction Hamiltonian has the form of an off-diagonal 2 × 2 block
matrix  
q 0 σ
ĤInt = − .A (2077)
c σ 0
the only non-zero matrix elements are those which connect the upper two-
component spinor to the lower two-component spinor of the virtual state. Also,
momentum conservation requires that the virtual state also be one of almost
zero momentum. Hence, the electron in the virtual state must have the form of
a negative-energy eigenstate
 
1 0
ψσ00 ,q00 (r) ≈ √ (2078)
V χσ00
since the contribution from a positive-energy state with small momentum is
negligibly small. Therefore, the electronic part of the matrix elements involving
the initial electron simply reduce to the expression
< ψσ00 ,q00 | ĤInt | ψσ,q > = | e | χ†σ00 σ χσ . A (2079)

Likewise, the matrix elements which involve the final (positive energy) electron
are evaluated as
< ψσ0 ,q0 | ĤInt | ψσ00 ,q00 > = | e | χ†σ0 σ χσ00 . A (2080)
From these one finds that, to second-order, the matrix elements that appear in
the transition rate are given by

2 π h̄ c2 ( χ†σ0 σ . α0 (k 0 ) χσ00 ) ( χ†σ00 σ . α (k) χσ )

  X 
M = e2 √
V ωk ωk 0 Eq − Eq00 + h̄ ωk
σ 00

( χ†σ0 σ . α (k) χσ00 ) ( χ†σ00 σ . α0 (k 0 ) χσ )


+
Eq − Eq00 − h̄ ωk0
e2 2 π h̄ c2
    X 
≈ √ ( χ†σ0 σ . α0 (k 0 ) χσ00 ) ( χ†σ00 σ . α (k) χσ )
2 m c2 V ωk ωk 0
σ 00

+ ( χ†σ0 σ . α (k) χσ00 ) ( χ†σ00 σ . α0 (k 0 ) χσ ) (2081)

where one has set

Eq − Eq00 ≈ 2 m c2 (2082)
On using the completeness relation for the two-component Dirac spinors

χσ00 χ†σ00 = I
X
(2083)
σ 00

366
the matrix elements are evaluated as
e2 2 π h̄ c2
    
† 0 0
M ≈ √ χσ0 ( σ . 
ˆα (k) ) ( σ . 
ˆα0 (k ) ) + ( σ . 
ˆα0 (k ) ) ( σ . 
ˆα (k) ) χσ
2 m c2 V ωk ωk 0
(2084)
The products in the above expression can be evaluated with the aid of the Pauli
identity. The result is

( σ . ˆα (k) ) ( σ . ˆα0 (k 0 ) ) = ( ˆα (k) . ˆα0 (k 0 ) ) + i σ . ( ˆα (k) ∧ ˆα0 (k 0 ) ) (2085)

Therefore, after combining both terms and noting that the pair of vector product
terms cancel since the vector product is antisymmetric under the interchange
k ↔ k 0 , one finds that the matrix elements reduce to

e2 2 π h̄ c2
    
† 0
M ≈ √ χσ0 2 ˆα (k) . ˆα (k ) χσ
0
2 m c2 V ωk ωk 0
e2 2 π h̄ c2
   
M ≈ √ δσ,σ0 2 ˆα (k) . ˆα0 (k 0 ) (2086)
2 m c2 V ωk ωk 0

Hence the matrix elements are diagonal in the spin indices. The above matrix
elements are identical to the matrix elements that occur in the non-relativistic
quantum theory of Thomson scattering. On substituting this result into eqn(2070),
one recovers the non-relativistic expression for the differential scattering cross-
section
  2 2
ωk 0
  
dσ e
≈ δσ,σ 0 | ˆα (k) . ˆα0 (k 0 ) |2
dΩk0 ωk m c2
  2 2
ωk 0

e
≈ δσ,σ0 cos2 Θ (2087)
ωk m c2

where
cos Θ = ˆα (k) . ˆα0 (k 0 ) (2088)
is the angle subtended by the initial and final polarization vectors. Hence, one
concludes that the negative-energy states do play an important role in light
scattering processes which involve low-energy electrons. The result, although
correct, does need re-interpretation, since the states of negative energy are as-
sumed to be filled with electrons in the vacuum and, therefore, the electron is
forbidden to occupy these levels in the intermediate states.

Electron-Positron Interpretation

The first contribution to the matrix elements, which was described above,
has to be re-interpreted as representing a process in which an electron that
initially occupies the negative-energy state q 00 makes a transition to the positive-
energy state q 0 while emitting the photon (k 0 , α0 ). This transition is subsequently
followed by the positive-energy electron q absorbing the photon (k, α) and falling

367
 q q
e- e-
(k',α') (k,α)

q''
e+ e+ q''

q' (k',α')
(k,α)
e-
-
e
q'

Figure 69: Processes involving positrons which contribute to Compton scatter-

ing.

into the empty negative-energy state. In this process, the negative-energy states
are completely occupied in the initial and final state, and the energy of the initial
and final states are conserved. By re-ordering the factors in the matrix elements
and noting that since

Eq + h̄ ωk = Eq0 + h̄ ωk0 (2089)

the contribution to the matrix element of these two descriptions are identical
(apart from an over all negative sign).

The second contribution to the matrix elements can be viewed as originating

from an electron which initially occupies a negative-energy state q 00 that absorbs
the photon (k, α) and makes a transition to the positive-energy state q 0 . This
is followed by the electron in the positive-energy state q emitting the photon
(k 0 , α0 ) and then falling into the empty negative-energy state q 00 . Again, on
re-ordering the matrix elements and noting that

Eq − h̄ ωk0 = Eq0 − h̄ ωk (2090)

one finds an identical expression (and the multiplicative factor of minus one).
Hence, Dirac hole-theory does lead to the correct classical result.

The above description is quite cumbersome, but can be made more concise by
adopting an anti-particle description of the unoccupied negative-energy states.
The first contribution to M first involves the creation of a virtual electron-
positron pair with the emission of the photon (k 0 , α0 ). The electron which has
just been created in the momentum eigenstate (q 0 , σ 0 ) remains unchanged in
the final state. Subsequently, the positron annihilates with the initial electron
(q, σ) while absorbing the photon (k, α). Since the intermediate state is a vir-
tual state, energy does not have to be conserved. The second contribution to M
involves the creation of a virtual electron-positron pair with the absorption of
the photon (k, α). The created electron (q 0 , σ 0 ) remains in the final state while

368
the positron subsequently annihilates with the initial electron (q, σ) and emits
the photon (k 0 , α0 ). This process is also a virtual process if the energy of the
incident light h̄ ωk is less than 2 m c2 .

The perturbative expression for the Compton scattering cross-section can

be evaluated exactly, without recourse to non-relativistic approximations. The
exact result is
 2 
1 2 ω0 ω0
  
dσ ω 2
= re + − 2 + 4 cos Θ (2091)
dΩ 4 ω ω0 ω
where Θ is the angle between the polarization vectors. This result was first
derived by Klein and Nishina148 in 1928.

12.16.2 Charge Conjugation

Charge conjugation is the operation of replacing matter by anti-matter, so that,
for example, electrons will be replaced by positrons and vice versa. The opera-
tion of charge conjugation consists of first taking the complex conjugate of the
Dirac equation
 
q
γ µ ( i h̄ ∂µ − Aµ ) − m c ψ = 0 (2092)
c
which describes a particle with charge q. We shall also assume that ψ describes
a positive-energy solution. Complex conjugation yields the equation
 
q ∗
γ µ∗ ( − i h̄ ∂µ − Aµ ) − m c ψ ∗ = 0 (2093)
c
The complex conjugate of a positive-energy solution ψ ∗ has a time-dependent
phase that identifies it with a negative-energy solution. The vector potential
Aµ is real. In the standard representation γ (0) , γ (1) and γ (3) are real, whereas
γ (2) is imaginary and, therefore, satisfies

γ (2)∗ = − γ (2) (2094)

We shall multiply the complex conjugate of the Dirac equation by γ (2) and anti-
commute γ (2) with the real γ µ∗ and commute γ (2) with the γ (2)∗ matrix. This
procedure changes the sign in front of the term originating from the differential
momentum operator w.r.t. the sign of the mass term. This procedure yields
 
q
γ (2) µ∗
γ ( − i h̄ ∂µ − Aµ ) − m c ψ ∗ = 0
c
 
q
µ
γ ( i h̄ ∂µ + Aµ ) − m c γ (2) ψ ∗ = 0 (2095)
c
148 O. Klein and Y. Nishina, Zeit. für Physik, 52, 843 (1928).

369
Hence, one sees that γ (2) ψ ∗ describes a Dirac particle with mass m and a charge
of − q moving in the presence of a vector potential Aµ . The fact that the opera-
tion of charge conjugation (in any representation) involves complex conjugation
is related to gauge invariance. Charge conjugation is a new type of symmetry for
particles that have complex wave functions which relates particles to particles
with opposite charges. The charge conjugate field ψ c is defined as
ψ c = Ĉ ψ ∗ (2096)
which is the result of the complex conjugation followed by the action of a linear
operator Ĉ. The joint operation can be represented as an anti-unitary operator.
The charge conjugation operator Ĉ is defined as the unitary and Hermitean
operator
Ĉ = − i γ (2) (2097)
The charge conjugation operator is Hermitean as
Ĉ † = + i γ (2)† = − i γ (2) = Ĉ (2098)
and it is unitary since
Ĉ † Ĉ = − γ (2) γ (2) = Iˆ (2099)
where the anti-commutation relations of the γ matrices have been used. It was
through this type of logic that Kramers149 discovered the form of the charge
conjugation transformation which turns a particle into an anti-particle.

The expectation values of an operator  in a general charge conjugated state

ψ c are related to the expectation values in a general state ψ via
 ∗
< ψ c | Â | ψ c > = − < ψ | γ (2) Â∗ γ (2) | ψ > (2100)

This can be shown in the position representation, by writing

Z Z
c†
3 c
d r ψ (r) Â ψ (r) = d3 r ψ ∗† (r) Ĉ † Â Ĉ ψ ∗ (r)
 Z ∗
3 † †∗ ∗ ∗
= d r ψ (r) Ĉ Â Ĉ ψ(r) (2101)

where we have used the identity z = (z ∗ )∗ in the second line. However, since Ĉ
is real, one finds
Z  Z ∗
3 c† c 3 † ∗
d r ψ (r) Â ψ (r) = d r ψ (r) Ĉ Â Ĉ ψ(r)
 Z ∗
= − d3 r ψ † (r) γ (2) Â∗ γ (2) ψ(r)

(2102)
149 H. A. Kramers, Proc. Amst. Akad. Sci. 40, 814 (1937).

370
This shows the relation between expectation values of a general operator  in
a state ψ(r) and its charge conjugated state ψ c (r).

We shall examine the effect of charge conjugation on the plane wave solutions
of the Dirac equation. The plane-wave solutions can be written as
r
( E + m c2 )
   
χσ µ
ψσ,k (x) = c h̄ σ . k exp − i k xµ (2103)
2EV E + m c2 χσ

The charge conjugate wave function is given by

c ∗
ψσ,k (x) = Ĉ ψσ,k (x)
r !
( E + m c2 ) χ∗σ
 
µ
= Ĉ ∗
c h̄ σ . k exp + i k xµ
2EV E + m c2 χ∗σ
(2104)

where

Ĉ = − i γ (2)
−iσ (2)
 
0
=
iσ (2) 0
 
0 0 0 −1
 0 0 1 0 
= 
 0 1 0 0 
 (2105)
−1 0 0 0

Therefore, the charge conjugate wave function is found to be given by

!
c h̄ σ ∗ . k
r
( E + m c2 )
 
c (2) − E + m c2 χ∗σ µ
ψσ,k (x) = i σ̂ exp + i k xµ
2EV χ∗σ
(2106)
which has the form of a plane-wave solution with negative energy E → − E,
and momentum h̄ k → − h̄ k. Furthermore, the spin of the charge conjugated
wave function has been reversed150 σ → − σ, since when i σ (2) acts on the
complex conjugated positive-eigenvalue eigenstate of the spin projected on an
arbitrary direction

cos θ2 exp[+i ϕ2 ]
 
χ+σ (θ, ϕ)∗ = (2107)
sin θ2 exp[−i ϕ2 ]
150 Note that the helicity is invariant under the joint transformation
σ → −σ
k → −k

371
it turns it into the negative-eigenvalue eigenstate

− sin θ2 exp[−i ϕ2 ]
 
χ−σ (θ, ϕ) = (2108)
cos θ2 exp[+i ϕ2 ]

That is, up to an arbitrary phase factor, the lower two-component spinor is

given by
i σ (2) χ+σ (θ, ϕ)∗ = χ−σ (θ, ϕ) (2109)
Likewise, it can be shown that the upper two-component spinor is proportional
to

i σ (2) ( σ ∗ . k ) χ+σ (θ, ϕ)∗ = − ( σ . k ) ( i σ (2) ) χ+σ (θ, ϕ)∗

= − ( σ . k ) χ−σ (θ, ϕ)
= ( σ . ( − k ) ) χ−σ (θ, ϕ) (2110)

The end result is that the charge conjugated single-particle wave function has
the form
r
c h̄ ( σ . (−k) )
( E + m c2 )
   
c − 2 χ−σ µ
ψσ,k (x) = E + m c exp + i k xµ
2EV χ−σ
(2111)
The properties described above are the properties of a state of a relativistic free
particle with a negative energy eigenvalue − E, momentum − h̄ k and spin − σ.
The absence of an electron in the charge conjugated state describes a positron,
with positive energy E, momentum h̄ k and spin σ.

More generally, even when an electromagnetic field is present, the charge

conjugated wave function of a positive-energy particle corresponds to the wave
function of a state with reversed energy E → − E, reversed spin σ → − σ
and reversed charge q → − q. Therefore, the charge conjugated state corre-
sponds to the (negative-energy) state which when unoccupied is described as an
anti-particle.

Exercise:

Consider massless Dirac particles, m → 0. (i) Show that the energy-helicity

eigenstates coincide with the eigenstates of γ (4) . (ii) Hence, show that the oper-
ators 12 ( Iˆ ± γ (4) ) project onto helicity eigenstates. These projection operators
relate the four-component Dirac spinors onto the independent two-component
Weyl spinors φL and φR . (iii) Show that charge conjugation transforms φL into
φR .

Exercise:

372
 Prove the completeness relation for the set of solutions for the Dirac equation
for a free particle
X  
φ†α (r)λ φα (r0 )ρ + φcα † (r)λ φcα (r0 )ρ = δ 3 (r − r0 ) δλ,ρ (2112)
α

where λ and ρ denote the components of the Dirac spinor151 .

13 Local and Global Gauge Symmetries

Global transformations are defined to be transformations which are the same
at every point in space-time, whereas local transformations vary from point to
point. Global symmetries, such as invariance under rotation, spatial translation,
and time translations, lead to laws of physics such as conservation of angular
momentum, linear momentum and energy. Local transformations do not lead
to new physical laws, but do lead to constraints on the form of the interactions.

A familiar example of a local transformation is the gauge transformation

in Quantum Electrodynamics. The Lagrangian density for Quantum Electro-
dynamics LQED given by the sum of the field-free electrodynamic Lagrangian
density and the Dirac Lagrangian density
 
1 † q
LQED = − F µ,ν Fµ,ν + c ψ γ µ ( p̂µ − Aµ ) − m c ψ (2113)
16 π c

The role of the source term for the electromagnetic field is played by the term
in the Dirac Lagrangian which represents the coupling to the vector potential.
The expression for the QED action
Z
SQED = dx4 LQED (2114)

is invariant under an infinite number of infinitesimal gauge transformations with

the form

Aµ → A0µ = Aµ + ∂µ Λ
 
iq
ψ → ψ0 = ψ − Λψ
h̄ c
 
iq
ψ† → ψ †0 = ψ † + Λ ψ† (2115)
h̄ c
151 Frequently, the relativistic free electron states are given a manifestly covariant normaliza-

tion, in order to facilitate covariant perturbation theory. The use of different normalization
conventions results in changes the form of the completeness relation.

373
Applying Noether’s theorem to the action yields an infinite number of continuous
currents, which, up to a multiplicative factor, are given by
    
ν ∂LQED ∂LQED iq
jΛ = ∂µ Λ − Λψ (2116)
∂ (∂ν Aµ ) ∂ (∂ν ψ) h̄ c
which satisfy the continuity conditions
∂ν jΛν = 0 (2117)
The currents are identified as
1 †
jΛν = F µ,ν ∂µ Λ + q ψ γ ν Λ ψ (2118)
4π
On substituting the Euler-Lagrange equation for the electromagnetic field
4π †
∂µ F µ,ν = q c ψ γν ψ (2119)
c
into the second term of the four-vector current density, one finds that the con-
tinuous current has the form
 
1
jΛν = ∂µ F µ,ν Λ (2120)
4π
The conserved charge Q can be defined as a volume integral of the temporal
component of the four-vector current density
Z
1 (0)
QΛ = d3 r jΛ (r)
c
Z  
1
= d3 r ∂µ F µ,0 Λ
4πc
Z
1
= d3 r ∇ . ( E Λ ) (2121)
4πc
For a global transformation for which Λ is constant over all space-time, one can
write Z
Λ
QΛ=const. = d3 r ( ∇ . E ) (2122)
4πc
On using Gauss’s law
∇.E = 4πρ (2123)
one finds that global gauge invariance ensures conservation of electrical charge
Z
Λ
QΛ=const. = d3 r ρ(r) (2124)
c
However, for any of the infinite number of spatially varying Λ which vanish on
the boundaries
Z
1
QΛ = d3 r ∇ . ( E Λ )
4πc
Z
1
= d2 S . E Λ
4πc
= 0 (2125)

374
Hence, although local transformations lead to an infinite number of symmetries
and an infinite number of continuous currents, all the conserved charges are
zero. Thus, local gauge invariance does not leads to new conserved quantities.
It does, however, constrain the form of the interactions.

14 The Many-Particle Dirac Field

14.1 The Algebra of Fermion Operators
Second quantization of fermions amounts to adopting an occupation number
representation. Therefore, we shall examine No. accounting for fermions152 .

Fermion operators satisfy anti-commutation relations. The anti-commutator

of two operators  and B̂ is defined as

{ Â , B̂ }+ ≡ Â B̂ + B̂ Â (2126)

The fermion creation and annihilation operators, ĉ†α and ĉα , satisfy the anti-
commutation relations

{ ĉ†α , ĉ†β }+ = 0
{ ĉα , ĉβ }+ = 0 (2127)

and
{ ĉ†α , ĉβ }+ = δα,β (2128)
where the quantum numbers α and β describe a complete set of single-particle
states.

The anti-commutation relation

ĉ†α ĉ†β = − ĉ†β ĉ†α (2129)

is merely a re-statement of the anti-symmetric nature of a fermionic many-

particle wave function under the permutation of a pair of particles, as is the
Hermitean conjugate relation

ĉα ĉβ = − ĉβ ĉα (2130)

The number operator n̂α is defined as

n̂α = ĉ†α ĉα (2131)

152 P. Jordan and E. Wigner, Zeit. für Physik, 47, 631 (1928).

375
The choice of anti-commutation relations results in the eigenvalues of the num-
ber operator to be restricted to either nα = 1 or nα = 0. This can be seen by
examining the identity
n̂α n̂α = n̂α (2132)
which follows from

n̂α n̂α = ĉ†α ĉα ĉ†α ĉα

= ĉ†α ĉα − ĉ†α ĉ†α ĉα ĉα
= ĉ†α ĉα + ĉ†α ĉ†α ĉα ĉα (2133)

where we have used the anti-commutation relation for the creation and annihila-
tion operator to obtain the second line and used the anti-commutation relation
for two annihilation operators to obtain the last line. On comparing the second
and third lines, one recognizes that

ĉ†α ĉ†α ĉα ĉα = 0 (2134)

Hence, we have

n̂α n̂α = ĉ†α ĉα

= n̂α (2135)

Thus, the eigenstates of the number operator satisfy the equation

n̂α n̂α | nα > = n̂α | nα >

n2α | nα > = nα | nα > (2136)

Therefore, for there to be non-trivial eigenstates the eigenvalues must satisfy

the equation
nα ( nα − 1 ) = 0 (2137)
which only has the solutions nα = 0 and nα = 1. Thus the choice of anti-
commutation relations for the creation and annihilation operators results in
the Pauli exclusion principle. The Pauli exclusion principle states that a non-
degenerate quantum state can not be occupied by more than one fermion.

The number operator satisfies the commutation relations

[ n̂α , ĉ†β ] = δα,β ĉ†β

[ n̂α , ĉβ ] = − δα,β ĉβ (2138)

as can be seen by using the fermion anti-commutation relations. The hierarchy

of eigenstates of the number operator can be found from the action of the
creation operator. In particular, one can define the eigenstate of the annihilation
operator with eigenvalue zero by

ĉα | 0 > = 0 (2139)

376
The state | 0 > is also an eigenstate of the number operator with eigenvalue
zero since
n̂α | 0 > = ĉ†α ĉα | 0 >
= 0 (2140)
A general eigenstate of the number operator with eigenvalue nα can be expressed
as
( ĉ† )nα
| nα > = √α |0 > (2141)
nα !
as can be seen by using the commutation relation
[ n̂α , ( ĉ†α )nα ] = nα ( ĉ†α )nα (2142)
If this operator equation acts on the state where the quantum state α is unoc-
cupied | 0 >, and using the condition
n̂α | 0 > = 0 (2143)
one finds the state of equation(2140) satisfies the eigenvalue equation
n̂α | nα > = nα | nα > (2144)
with eigenvalue of either unity or zero.

A general number operator eigenstate can be expressed in terms of the oc-

cupation numbers of all the single-particle states
∞ 
( ĉ†α )nα
Y 
| {nα } > = √ | 0 , 0 , 0 , ... , 0 > (2145)
α=1
nα !

where the allowed values of the set of occupation numbers nα are either unity
or zero. The sequencing or ordering of the creation operators in this expression
is crucial, since the interchange the positions of the operators may result in a
change in sign of the state. For example, the action of a creation operators on
a general number eigenstate has the effect
Pβ
ĉ†β | n1 n2 . . . nβ . . . > = ( − 1 )( i=1 ni ) | n1 n2 . . . nβ + 1 . . . > (2146)

where the sign occurs since this involves anti-commutating ĉ†β with
Pβ
i=1 ni
other creation operators to bring it into the β-th position.

14.2 Quantizing the Dirac Field

The quantization of the Dirac field proceeds exactly the same way as for non-
relativistic electrons153 . However, the negative-energy states will be described
153 W. Heisenberg and W. Pauli, Zeit. für Physik, 56, 1 (1929).

W. Heisenberg and W. Pauli, Zeit. für Physik, 59, 168 (1930).

377
with a different notation from the positive-energy states. The change of no-
tation is to reflect the intent of describing the (quasi-particle) excitations of
the system and not to describe the many-particle ground state which is unob-
servable. The wave functions φα (r) describing the positive-energy states of the
non-interacting electrons are indexed by the set of quantum numbers α ≡ (k, σ).
The negative-energy states are described as the charge conjugates of the positive-
energy states. Therefore, the negative-energy states are described by the same
set of indices α and the corresponding wave functions are denoted by φcα (r).
The annihilation operator for electrons in the positive-energy state α is denoted
by ĉα . However, the operator which removes an electron from the (negative-
energy) charge conjugated state φcα (r) is denoted by a creation operator b̂†α . The
change from annihilation operator to creation operator merely represents that
creating a positron with quantum numbers α is equivalent to creating a hole in
the negative-energy state154 . The effect of the annihilation operators on Dirac’s
vacuum | 0 >, in which all the negative-energy states are fully occupied are

ĉα | 0 > = 0
b̂α | 0 > = 0 (2147)

where the first expression follows from the assumed absence of electrons in the
positive-energy states, and the second expression follows from the assumption
that all the negative-energy states are completely filled, so adding an extra elec-
tron to the state φcα is forbidden by the Pauli-exclusion principle. More concisely,
the above relations state that the vacuum contains neither (positive-energy)
electrons nor positrons. It is seen that the form of the anti-commutation rela-
tions are unchanged by this simple change of notation. The anti-commutation
relations become

{ ĉ†α , ĉ†β }+ = { ĉα , ĉβ }+ = 0

{ ĉ†α , ĉβ }+ = δα,β (2148)

for the electron operators

{ b̂†α , b̂†β }+ = { b̂α , b̂β }+ = 0

{ b̂†α , b̂β }+ = δα,β (2149)

for the positron operators, and the mixed electron/positron anti-commutation

relations are given by

{ ĉ†α , b̂†β }+ = { ĉα , b̂β }+ = { ĉ†α , b̂β }+ = 0 (2150)

The mixed electron/positron anti-commutation relations are all zero, since the
operators describe electrons in different single-particle energy eigenstates. In
154 W. H. Furry and J. R. Oppenheimer, Phys. Rev. 45, 245 (1934).

378
this notation, the field operators are expressed as155
X  
c †
ψ̂(r) = φα (r) ĉα + φα (r) b̂α (2151)
α

and
X  
ψ̂ † (r) = φ∗α (r) ĉ†α + φcα ∗ (r) b̂α (2152)
α

The field operators ψ̂(r) and ψ̂ † (r) are expected to be canonically conjugate, as
we shall show below.

The Lagrangian density is given by

 
L = cψ ˆ † i h̄ γ µ ∂ − m c ψ̂ (2153)
µ

so the momentum field operator Π̂(r) canonically conjugate to ψ̂(r) is given by

Π̂(r) =
1 δL ˆ † (r) γ (0) = i h̄ ψ̂ † (r)
= i h̄ ψ (2154)
c δ(∂0 ψ̂)

Hence, one expects that the field operators ψ̂ † (r) and ψ̂(r) are canonically con-
jugate and, therefore, satisfy the equal-time anti-commutation relations

{ ψ̂ † (r)λ , ψ̂(r0 )ρ }+ = δ 3 (r − r0 ) δλ,ρ (2155)

where λ and ρ label the components of the Dirac spinor. The anti-commutation
relations for the field operators can be verified by noting that
X 
† 0
{ ψ̂ (r) , ψ̂(r ) }+ = { ĉ†α , ĉβ }+ φ∗α (r) φβ (r0 ) + { ĉ†α , b̂†β }+ φ∗α (r) φcβ (r0 )
α,β

+ { b̂α , ĉβ }+ φcα ∗ (r) φβ (r0 ) + { b̂α , b̂†β }+ φcα ∗ (r) φcβ (r0 )
X  
= δα,β φ∗α (r) φβ (r0 ) + δα,β φcα ∗ (r) φcβ (r0 )
α,β
X  
= φ∗α (r) φα (r0 ) + φcα ∗ (r) φcα (r0 )
α
= δ (r − r0 )
3
(2156)

where the fermion anti-commutation relations have been used in arriving at the
second line. The positive-energy states and their charge conjugated states form
a complete set of basis states for the single-particle Dirac equation, so their
155 W. Heisenberg and W. Pauli, Zeit. für Physik, 56, 1 (1929).

W. Heisenberg and W. Pauli, Zeit. für Physik, 59, 168 (1930).

379
completeness condition has been used in going from the third to the fourth line.
The equal-time field anti-commutation relations can be generalized to field anti-
commutators at space-time points with a general type of separation. In the case
where the two field points x and x0 have a space-like separation
µ
( xµ − x0 ) ( xµ − x0 µ ) < 0
causality dictates that the anti-commutators are zero
{ ψ̂ † (x) , ψ̂(x0 ) }+ = 0
That is, for space-like separations, there is no causal connection156 so a mea-
surement of a local field at x0 cannot affect a measurement at x. N. Bohr and

∆x2 > 0
ct

r
2
∆x < 0

Figure 70: Due to causality, the anti-commutator of the field operator should
vanish for space-like separations. The anti-commutators can be non-zero inside
or on the light cone.

L. Rosenfeld157 have put forward general arguments that the commutation rela-
tions also place limitations on the measurement of fields at time-like separations.

The Hamiltonian density for the (non-interacting) quantized Dirac field the-
ory can be expressed as the operator
 
Ĥ = ψ̂ † γ (0) c − i h̄ γ . ∇ + m c ψ̂ (2157)

and the Hamiltonian operator is given by

Z
Ĥ = d3 r Ĥ (2158)

When the expansion of the quantized field in terms of single-particle wave func-
tions is substituted into the Hamiltonian, one finds
X  
† c †
Ĥ = Eα ĉα ĉα + Eα b̂α b̂α
α
156 Outside the light-cone there is no way to distinguish between future and past.
157 N. Bohr and L. Rosenfeld, Kon. Dansk. Vid. Selskab., Mat.-Fys. Medd. XII, 8 (1933).

380
 X  
= Eα ĉ†α ĉα − Eα b̂α b̂†α (2159)
α

where the expression for the energy of the charge conjugated state

Eαc = − Eα (2160)

has been used. On anti-commuting the positron and annihilation operators, one
finds  
X
Ĥ = Eα ĉ†α ĉα + b̂†α b̂α − 1 (2161)
α

The last term, when summed over α, yields the infinitely negative energy of
Dirac’s vacuum in which all the negative-energy states are filled. The vacuum
energy shall be used as the reference energy, so the Hamiltonian becomes
X  
† †
Ĥ = Eα ĉα ĉα + b̂α b̂α (2162)
α

which describes the energy of the excited state as the sum of the energies of the
excited electrons and the excited positrons. The energies of the positrons and
electrons are given by positive numbers.

The momentum operator defined by Noether’s theorem is found as

 
† †
X
P̂ = h̄ k ĉk,σ ĉk,σ + b̂k,σ b̂k,σ (2163)
k,σ

which is just the sum of the momenta of the (positive-energy) electrons and the
positrons. The spin operator is defined as
Z
h̄
Ŝ = d3 r ψ̂ † σ̂ ψ̂ (2164)
2
This is evaluated by substituting the expression for the field operators in terms
of the single-particle wave functions and the particle creation and annihilation
operators. The expectation value of the spin operator in the charge conjugated
state φcα is given by
Z  Z ∗
d3 r φcα † (r) σ̂ φcα (r) = − d3 r φ†α (r) γ (2) σ ∗ γ (2) φα (r)
 Z ∗
= d3 r φ†α (r) σ (2) σ̂ ∗ σ (2) φα (r)
 Z ∗
3 †
= − d r φα (r) σ̂ φα (r)
 Z 
3 †
= − d r φα (r) σ̂ φα (r) (2165)

381
The third line follows from the identity

σ (2) σ̂ ∗ σ (2) = − σ̂ (2166)

The last line follows since σ is Hermitean. Hence, the spin operator is evaluated
as  
h̄ X † † †
Ŝ = χσ00 σ χσ0 ĉk,σ00 ĉk,σ0 + b̂k,σ00 b̂k,σ0 (2167)
2 0 00
k;σ ,σ

which is just the sums of the spins of the electrons and positrons.

Finally, the conserved Noether charge corresponding to the global gauge

invariance is given by
Z
Q̂ = d3 r ψ̂ † (r) ψ̂(r)
X  
† †
= ĉα ĉα + b̂α b̂α
α
X  
= ĉ†α ĉα − b̂†α b̂α + 1 (2168)
α

The last term in the parenthesis, when summed over all states α, yields the total
charge of the vacuum which is to be discarded. Hence, the observable charge is
defined as
X  
Q̂ = ĉ†α ĉα − b̂†α b̂α (2169)
α

which shows that the total electrical charge defined as the difference between
the number of electrons and the number of positrons is conserved.

14.3 Parity, Charge and Time Reversal Invariance

The Lagrangian density may posses continuous symmetries and it may also
posses discrete symmetries. Some of the discrete symmetries are examined be-
low.

14.3.1 Parity
The parity eigenvalue equation for a multi-particle state with parity ηψ can be
expressed as
P̂ | ψ > = ηψ | ψ > (2170)
Since the action of the parity operator on states is described by a unitary opera-
tor, operators transform under parity according to the general form of a unitary

382
transformation. In particular, the effect of the parity transformation on the
field operator is determined as

ψ̂(r) → ψ̂ 0 (r0 ) = P̂ ψ̂(r) P̂ (2171)

The parity transformation is going to be determined in analogy with the parity

transformation of a classical field, in which the creation and annihilation oper-
ators are replaced by complex numbers. The parity operation on the quantum
field can be interpreted as only acting on the wave functions and not the particle
creation and annihilation operators. Quantum mechanically, this corresponds
to viewing the parity operator as changing the properties of the states to the
properties associated with the parity reversed states. Since the field operator is
expressed as
X  
† c
ψ̂(r) = ĉα φα (r) + b̂α φα (r) (2172)
α

one has
X  
P̂ ψ̂(r) P̂ = ĉα P̂ φα (r) P̂ + b̂†α P̂ φcα (r) P̂ (2173)
α

However, under a parity transform a general Dirac spinor satisfies

P̂ φα (r) = ηαP φPα (r)

P̂ φcα (r) = ηαP c φcPα (r) (2174)

where ηαP is a phase factor which represents the intrinsic parity of the state.
Furthermore, since P̂ 2 = I,ˆ then the intrinsic parities ηαP and ηαP c have to
satisfy the conditions

( ηαP )2 = 1
( ηαP c )2 = 1 (2175)

So the intrinsic parities are ±1. The intrinsic parity of a state φα (r) and its
charge conjugated state φcα (r) are related by

ηαP c = − ηαP (2176)

This follows since charge conjugation flips the upper and lower two-component
spinors and these two-component spinors have opposite intrinsic parity. There-
fore, the state φα (r) and the charge conjugates state φcα (r) have opposite pari-
ties. Therefore, it follows that the field operator transforms as
X  
P̂ ψ̂(r) P̂ = ηαP ĉα φPα (r) − ηαP b̂†α φcPα (r) (2177)
α

so the quantum field operators transforms in a similar fashion to the classical

field.

383
 The relations between parity reversed states and parity reversed charge con-
jugated states can be verified by examining the free particle solutions of the
Dirac equation and noting that the parity operator consists of the product of
γ (0) and spatial inversion r → − r. This spatial inversion acting on a wave
function with momentum k and spin σ becomes a wave function with momentum
−k and spin σ, up to a constant of proportionality. A free particle momentum
eigenstate is given by
   
χσ (0)
φσ,k (x) = N c h̄ k . σ exp − i ( k0 x − k.r) (2178)
E + m c2 χσ

The application of the parity operator to the above wave function yields
   
(0) χσ (0)
P̂ φσ,k (x) = N γ c h̄ k . σ exp − i ( k0 x + k.r)
E + m c2 χσ
   
χσ
= N c h̄ k . σ exp − i ( k0 x(0) + k . r )
− E + m c2 χσ
= φσ,−k (x) (2179)

as anticipated. The charge conjugated state is given by

φcσ,k (x) = Ĉ φ∗σ,k (x)

!
χ∗σ
 
= N Ĉ c h̄ k . σ ∗ exp + i k µ xµ (2180)
E + m c2 χ∗σ

where

Ĉ = − i γ (2)
 
0 0 0 −1
 0 0 1 0 
= 
 0 1 0 0 
 (2181)
−1 0 0 0

Therefore, the charge conjugate wave function is given by

!
c h̄ k . σ ∗
 
∗
c (2) χσ µ
φσ,k (x) = − i σ̂ N E + m c 2
exp + i k xµ (2182)
− χ∗σ

The effect of the parity operator on this state leads to

!
c h̄ k . σ ∗
 
∗
c (2) χ σ (0)
P̂ φσ,k (x) = − i σ̂ N E + m c 2
exp + i ( k x0 + k . r )
+ χ∗σ
!
c h̄ ( − k . σ ∗ )
 
(2) − χ∗σ (0)
= − i σ̂ N E + m c2
exp + i ( k x0 + k . r )
+ χ∗σ
= − φcσ,−k (x) (2183)

384
where in the first line the parity operator has sent r → − r and the factor of
γ (0) has flipped the sign of the lower components. In the second line we have
re-written k as −(−k) in the two two-component spinor, in anticipation of the
comparison with eqn(2180) which allows us to identify the factor of φcσ,−k (x).
This example shows that a state and its charge conjugate have opposite intrinsic
parities.

From the general form of the parity transformation on Dirac spinors, one
infers that the parity transform of the field operator is given by
X  
P̂ ψ̂(r) P̂ = ĉα ηαP φPα (r) − b̂†α ηαP φcPα (r) (2184)
α

On setting α0 = P α and noting that α = P α0 , one finds

X  †

P P c
P̂ ψ̂(r) P̂ = ηPα0 ĉPα0 φα0 (r) − ηPα0 b̂Pα0 φα0 (r) (2185)
Pα0

and on transforming the summation index from α0 to α

X  
P̂ ψ̂(r) P̂ = P
ηPα P
ĉPα φα (r) − ηPα b̂†Pα φcα (r) (2186)
α

Thus, the parity operation can also be interpreted as only affecting the particle
creation and annihilation operators, and not the wave functions. Quantum
mechanically, this interpretation corresponds to viewing that the particles as
being transferred into their parity reversed states
X  
† c
P̂ ψ̂(r) P̂ = P̂ ĉα P̂ φα (r) + P̂ b̂α P̂ φα (r) (2187)
α

In this new interpretation, the effects of parity on the fermion operators are
found by identifying the operators multiplying the single-particle wave functions
in the previous two equations. The resulting operator equations are
P
P̂ ĉα P̂ = ηPα ĉPα (2188)
and
P̂ b̂†α P̂ = − ηPα
P
b̂†Pα (2189)
which shows that fermion particles and anti-particles have opposite intrinsic
parities. Therefore, we conclude that, irrespective of which interpretation is
used, the field operator transforms as
X  
P P † c
P̂ ψ̂(r) P̂ = ηα ĉα φPα (r) − ηα b̂α φPα (r) (2190)
α

which shows that the quantum field operators transforms in a similar fashion
to the classical field.

385
14.3.2 Charge Conjugation
Under charge conjugation, the classical Dirac field transforms as

ψ → ψ c = − i γ (2) ψ ∗ (2191)

(up to an arbitrary phase) since this is how the single-particle wave functions
transform. Classically, the (anti-linear) charge conjugation operator Cˆ is the
product of complex conjugation and the unitary matrix operator Ĉ = − i γ (2) .
If the classical field is expressed as a linear superposition of energy eigenfunc-
tions, the amplitudes of the eigenfunctions are represented by complex numbers.
In the charge conjugated state, these amplitudes are replaced by the complex
conjugates. In the quantum field, the amplitudes must be replaced by parti-
cle creation and annihilation operators. If an amplitude is associated with an
annihilation operator, then the complex conjugate of the amplitude is usually
associated with a creation operator. Hence, we should expect that charge con-
jugation will result in the creation and annihilation operators being switched.

Since the quantum field operator is expressed as

X  
ψ̂(r) = ĉα φα (r) + b̂†α φcα (r) (2192)
α

the charge conjugate operation Cˆ transforms the field operator via

X  
ψ̂ c (r) = Cˆ ψ̂(r) Cˆ = ĉ†α Cˆ φα (r) Cˆ + b̂α Cˆ φcα (r) Cˆ (2193)
α

where, in accord with the earlier comment about the relation between the quan-
tum and classical fields, the single-particle operators have been replaced by their
Hermitean conjugates. However, under charge conjugation general Dirac spinors
satisfy

Cˆ φα (r) = η c φcα (r)

Cˆ φcα (r) = η c φα (r) (2194)

therefore,
X  
ψ̂ (r) = Cˆ ψ̂(r) Cˆ =
c
η c
ĉ†α φcα (r) + b̂α φα (r) (2195)
α

However, if the charge conjugation operator is to be interpreted as only acting

on the single-particle operators, one has
X  
c
ψ̂ (r) = ˆ ˆ ˆ † ˆ c
C ĉα C φα (r) + C b̂α C φα (r) (2196)
α

386
For consistency, the two expressions for ψ̂ c (r) must be equivalent. Hence, the
operator coefficients of φα (r) and φcα (r) in the two expressions should be iden-
tical. Therefore, one requires that

Cˆ ĉα Cˆ = η c b̂α
Cˆ b̂†α Cˆ = η c ĉ†α (2197)

In other words, charge conjugation replaces particles by their anti-particles and

their quantum numbers α are unchanged. Furthermore, we identify the charge
conjugated field operator as

ψ̂ c = Cˆ ψ̂ Cˆ = − i η c γ (2) ψ̂ † (2198)

where ψ̂ † is the Hermitean conjugate (column) field operator. Apart from the
replacement of the complex amplitudes with the Hermitean conjugates of the
creation and annihilation operators, the above expression is identical to the ex-
pression for charge conjugation on the classical field.

The charge conjugation operator has the effect of reversing the current den-
sity operator
ˆ † γ µ ψ̂ Cˆ = − ψ
Cˆ ψ ˆ † γ µ ψ̂ (2199)
which is understood as the result in the change of the charge’s sign.

14.3.3 Time Reversal

The time-reversal operation interchanges the past with the future. Time reversal
transforms the space-time coordinates via

T̂ (ct, r) = (−ct, r) (2200)

Thus, under time reversal, the time and spatial components of the position
four-vector have different transformational properties. Furthermore, the energy-
momentum four-vector transforms as

T̂ (p(0) , p) = (p(0) , −p) (2201)

Hence, the position four-vector and momentum four-vector have different trans-
formational properties. Due to the above properties, angular momentum (in-
cluding spin) transforms as
T̂ J = − J (2202)
Therefore, we find that time reversal reverses momenta and flips spins.

According to the Wigner theorem158 , time reversal can only be implemented

by an anti-linear anti-unitary transformation. Since the time reversal operator
158 E. P. Wigner. Göttinger. Nachr. 31, 546 (1932).

387
T̂ interchanges the initial and final states, then

< T̂ ψf | T̂ ψi > = < ψi | ψf >

= < ψf | ψi >∗ (2203)

Thus, T̂ must be an anti-unitary operator. Furthermore, if the initial state is

given by a linear superposition
X
| ψi > = Cα | φα > (2204)
α

then the overlap is given by

X X
< T̂ ψf | T̂ Cα | φα > = Cα∗ < ψf | φα >∗ (2205)
α α

Hence, one infers that

X X
T̂ Cα | φα > = Cα∗ T̂ | φα > (2206)
α α

which is the definition of an anti-linear operator and so we identify T̂ as an

anti-linear operator.

It can be shown that the time-reversed Dirac wave function defined by

T̂ ψ(t, r) = − γ (1) γ (3) ψ ∗ (−t, r) (2207)

satisfies the Dirac equation with t → − t. For example, the plane wave
solutions of the Dirac equation can be shown to transform as

T̂ φσ,k (r, t) = − γ (1) γ (3) φ∗σ,k (r, −t)

= φ−σ,−k (r, t) (2208)

which flips the momentum and the spin angular momentum. It should be noted
that the matrix operator γ (1) γ (3) does not couple the upper and lower two-
component spinors, but nevertheless is closely related to the operator − i γ (2)
which occurs in the charge conjugation operator.

Also, if the Dirac field operator is required to satisfy

T̂ ψ̂(t, r) T̂ = − γ (1) γ (3) ψ̂ ∗ (−t, r) (2209)

then the single-particle operators must satisfy

T̂ cα T̂ = cT α
T̂ bα T̂ = bT α (2210)

which correspond to particles following time-reversed trajectories.

388
Table 19: Discrete Symmetries of Particles.
The charge conjugated of a state is a negative energy state with momentum −p
and spin −σ, that is interpreted as the state of antiparticle with momentum p
and spin σ.

Q p σ Λ

Charge Conjugation − + + +

Parity + − + −

Time Reversal + − − +

CPT − + − −

It is known that the weak interaction violates parity invariance. However,

there was a slight possibility that the weak interaction conserves the combined
operation of charge conjugation and spatial inversion159 . Christenson, Cronin,
Fitch and Turlay160 performed experiments which showed that the combined
operation C P is violated in the decay of K mesons. There is reason to believe
that the weak interaction is invariant under the combined symmetry operation
C P T , since this is related to Lorentz invariance.

The combined symmetry operation Cˆ P̂ T̂ transforms a Dirac spinor as

ψ 0 (x0 ) = Cˆ P̂ T̂ ψ(x)
 ∗
(2)
= −iγ P̂ T̂ ψ(x)
 ∗
= + i γ (2) γ (0) γ (1) γ (3) ψ ∗ (−x)

= i γ (2) γ (0) γ (1) γ (3) ψ(−x)

= i γ (0) γ (1) γ (2) γ (3) ψ(−x)
= γ (4) ψ(−x) (2211)
159 J. C. Wick, A. S. Wightman and E. P. Wigner, Phys. Rev. 88, 101 (1952).
160 J. Christenson, J. W. Cronin, V. I. Fitch and R. Turlay, Phys. Rev. Lett. 13, 138 (1964).

389
14.3.4 The CPT Theorem
The CPT theorem states that any local161 quantum field theory with a Her-
mitean Lorentz invariant Lagrangian which satisfies the spin-statistics theorem,
is invariant under the compound operation Cˆ P̂ T̂ , where the operators can be
placed in any order.

The proof of the theorem relies on the fact that any Lorentz invariant quan-
tity must be created out of contracting the indices of bi-linear covariants (quan-
tities such as the current density jµ which involve products of the γµ ) with the
indices of contravariant derivatives ∂ µ . Since the joint operation P̂ T̂ results in
each of the contravariant derivatives ∂ µ in the product changing sign, the theo-
rem ensures that the corresponding bi-linear covariants with which the deriva-
tives are contracted with must undergo an equivalent number of sign changes
under the compound operation Cˆ P̂ T̂ . The theorem only assumes invariance
under proper orthochronous Lorentz transformations and makes no assumptions
about reflection. The improper transformations are treated as analytic continu-
ation of the Lorentz transformation into complex space-time. The theorem was
first discussed by Lüders162 and Pauli163 , and then by Lee, Oehme and Yang164 .

The theorem has several consequences, such as the equality of the masses of
particles and their anti-particles. This follows since the mass mc is an eigenvalue
of p̂(0) in the particle’s rest frame and since one can find simultaneous eigenstates
of the commuting operators p̂µ and the product Cˆ P̂ T̂ . If one denotes the
compound operator as
Θ̂ = Cˆ P̂ T̂ (2212)
then

< Ψ | Ĥ | Ψ > = < Ψ | Θ̂−1 Θ̂ Ĥ Θ̂−1 Θ̂ | Ψ >

= < Ψ | Θ̂−1 Ĥ Θ̂ | Ψ > (2213)

since the CPT theorem ensures that Θ̂ commutes with the Hamiltonian

Θ̂ Ĥ Θ̂−1 = Ĥ (2214)

If | Ψ > represents a stable single-particle state, such as

| Ψ > = c†α | 0 > (2215)

161 A Local Field Theory is one expressible in terms of a local Lagrangian density in which

interactions can be expressed in terms of products of fields at the same point in space-time.
It would be truly remarkable if this concept were to continue to work at arbitrarily small
distances!
162 G. Lüders, Dan. Mat. Fys. Medd. 28, 5 (1954).

G. Lüders, Ann. Phys. 2, 1 (1957).

163 W. Pauli in Niels Bohr and the Development of Physics, Pergamon Press, London (1955).
164 T. D. Lee, R. Oehme and C. N. Yang, Phys. Rev. 106, 340 (1957).

390
then the state Θ̂ | Ψ > describes an anti-particle with flipped angular momen-
tum. This follows since the vacuum satisfies

Θ̂ | 0 > = | 0 > (2216)

Therefore, the single-particle state transforms as

Θ̂ | Ψ > = Θ̂ c†α Θ̂−1 Θ̂ | 0 >

= Θ̂ c†α Θ̂−1 | 0 > (2217)
ˆ P̂ and T̂ , one finds that the operator Θ̂ c†α Θ̂−1
By successive applications of C,
reduces to the creation operator for the anti-particle with reversed angular mo-
mentum. Therefore, the state Θ̂ | Ψ > describes an anti-particle with flipped
angular momentum. From the equality of the expectation values

< Ψ | Ĥ | Ψ > = < Ψ | Θ̂−1 Ĥ Θ̂ | Ψ > (2218)

one finds that the energy of a particle is equal to the energy of an anti-particle
with a reversed spin. However, as the rest mass cannot depend on the angular
momentum, the mass of a particle is equal to the mass of its anti-particle. For
unstable particles, the equality of the mass of the particle and anti-particle is
ensured by the invariance of the S-matrix under Cˆ P̂ T̂ .

Likewise, one can use the CPT theorem to show that the total decay rate
of a particle into products is equal to the total decay rate of the anti-particle
into its products165 . It should be noted that the partial decay rates into specific
final states are not equivalent, only the sums over all final states are equal.

14.4 The Connection between Spin and Statics

The above result for the energy operator of the Dirac field illustrates the “Spin
Statistics Theorem” proposed by Pauli166 . The theorem states that particles
with half odd-integer spins obey Fermi-Dirac Statistics and particles with integer
spins obey Bose-Einstein Statistics. The Dirac spinor describes spin one-half
particles, and if these particles are chosen to satisfy anti-commutation relations,
then the energy of the excited states is given by
X  
ĤDirac = Eα ĉ†α ĉα + b̂†α b̂α (2219)
α

which only has positive excitation energies. Hence, if the wave function changes
sign under the interchange of a pair of spin one-half particles the energy is
bounded from below. If the field operators had been chosen to obey commu-
tation relation, then the wave function would have been symmetric under the
165 T. D. Lee, R. Oehme and C. N. Yang, Phys. Rev. 106, 340 (1957).
166 W. Pauli, Phys. Rev. 58, 716 (1940).

391
interchange of particles. If this were the case, there would be a negative sign in
front of the positron energies so that the energy would have been unbounded
from below. This would have implied that the vacuum would not be stable, and
the theory is erroneous. This can be taken as implying that spin one-half par-
ticles must obey Fermi-Dirac Statistics. The other part of the theorem compels
integer spin particles to be bosons. Therefore, since photons have spin one, the
expression for the energy of the electromagnetic field is considered to be given
by
X h̄ ωk  † 
ĤPhoton = âk,α âk,α + âk,α â†k,α (2220)
2
k,α

This Hamiltonian represents the energy of a spin-one particle. The photon

creation and annihilation operators satisfy commutation relations, therefore,
the energy can be expressed as
X h̄ ωk  
ĤPhoton = 2 â†k,α âk,α + 1 (2221)
2
k,α

which is the sum of the vacuum energy (the zero-point energies) and the ener-
gies of each excited photon. The excitation energies are positive. If it had been
assumed that the photon wave functions were anti-symmetric under the inter-
change of particles, then one would have found that the photon energies would
have been identically equal to zero. Furthermore, the excited photons would
have carried zero momentum and, therefore, be completely void of any physi-
cal consequence. Hence, one concludes that spin-one photons must obey Bose-
Einstein Statistics. The generalized theorem167 is an assertion that a non-trivial
integer spin field cannot have a anti-commutator that vanishes for space-like sep-
arations and a non-trivial odd half-integer spin field cannot have a commutator
that vanishes for space-like separations.

15 Massive Gauge Field Theory

Following Yang and Mills168 , we shall consider a two-component complex scalar
field. The field can be expressed as a two-component field, representing states
with different isospin  
Φ1
Φ = (2222)
Φ2
where the Φi are complex scalars. That is
Φ1 = <e Φ1 + i =m Φ1
Φ2 = <e Φ2 + i =m Φ2 (2223)
167 R. Streater and A. S. Wightman, PCT, Spin and Statistics, and All That, Princeton
Univ. Press (2000).
168 C. N. Yang and R. L. Mills, Phys. Rev. 96, 191 (1964).

392
This is equivalent to assuming four independent real fields. The inner product
is defined as
Φ† Φ = Φ∗1 Φ1 + Φ∗2 Φ2 (2224)

15.1 The Gauge Symmetry

We shall assume that the Lagrangian is invariant under a generalized gauge
transformations of the form
 
0 (0)
Φ → Φ = exp − i α Û Φ (2225)

where α(0) is an arbitrary scalar. The invariance of the Lagrangian under mul-
tiplication of the wave function by the phase factor, is equivalent to the usual
U (1) gauge invariance which has been discussed in the context of the electro-
magnetic field. The operator Û must be a unitary operator, if the norm of Φ is
conserved by the generalized gauge transformation
Φ†0 Φ0 = Φ† Û † Û Φ
= Φ† Φ (2226)
Therefore, one requires
Û † Û = Iˆ (2227)
and so Û must be a unitary operator. The operator Û is assumed to be an
arbitrary unitary matrix that acts on isospin states, that is, it acts on the two
components of Φ. Furthermore, it shall be assumed that the unitary matrix
has determinant + 1. Hence, the Lagrangian is assumed to be invariant under
a set of SU (2) gauge transformations. A general transformation of SU (2) is
generated by the three operators
 
(1) 0 1
τ =
1 0
 
(2) 0 −i
τ =
i 0
 
(3) 1 0
τ = (2228)
0 −1
where these matrices generate a Lie algebra. That is, the algebra of the com-
mutation relations is closed, since
[ τ (i) , τ (j) ] = 2 i ξ i,j,k τ (k) (2229)
where ξ i,j,k is the antisymmetric Levi-Civita symbol. An arbitrary unitary
transformation can be expressed as
 X 
k (k)
Û = exp − i α τ (2230)
k

393
where the αk are three real quantities. This represents an arbitrary rotation in
isospin space169 . The U (1) gauge transformation can also be represented in the
same way. Namely, the U (1) transformation can be expressed as
 
Û0 = exp − i α(0) τ (0) (2231)

where τ (0) is the unit matrix

 
(0) 1 0
τ = (2232)
0 1
We should note that since τ (0) commutes with all isospin operators, the U (1)
symmetry is decoupled from the SU (2) symmetry, and hence, when a coupling
to gauge fields is introduced, the U (1) gauge field may have a coupling constant
which is different from the coupling constant for the three SU (2) gauge fields.

15.2 The Coupling to the Gauge Field

We shall start with a Lagrangian density Lscalar describing the field free two
component scalar field, given by
Lscalar = ∂µ Φ† ∂ µ Φ − V (Φ† Φ) (2233)
†
where V (Φ Φ) is an arbitrary scalar potential. For example, in a Klein-Gordon
field theory describing particles with mass m
 2
mc
V (Φ† Φ) = Φ† Φ (2234)
h̄
The Lagrangian is invariant under the combined gauge transformation if the
quantities αk are independent of x. In this case, the field is invariant under the
transformation which is identical at each point in space, so the Lagrangian is
said to have a global gauge invariance.

We shall alter the Lagrangian, such that it is invariant under a gauge trans-
formation which varies from point to point in space. These are local gauge
transformations, in which the αk (x) depend on x. If the Lagrangian is to be
invariant under local gauge transformations, then one must introduce a coupling
to gauge fields Aµ . This coupling compensates for the change of the derivatives
under the gauge transformation, so that
  †    
µ µ
∂µ − i g Aµ Φ ∂ − igA Φ
  †    
= ∂µ − i g A0µ Φ0 ∂ µ − i g Aµ0 Φ0 (2235)

169 We shall not stop and contemplate the question of what restricts our measurements have to

be quantized along the isospin z-direction, and shall not ponder why there is a super-selection
rule at work.

394
Since
Φ = Û † Φ0 (2236)
we require that
     
∂ µ − i g Aµ0 Φ0 = Û ∂ µ − i g Aµ Φ (2237)

so the fields Aµ must transform as

 
i
Aµ0 = Û Aµ Û † + Û ∂ µ Û † (2238)
g

where the derivative only acts on the unitary transformation. Since the Û are
generated by τ (k) , there must be four components of Aµ , i.e. the fields have
four components Aµ,k . The matrix form of Aµ is given by
3
X
Aµ = Aµ,k τ (k)
k=1
Aµ,(3) Aµ,(1) − i Aµ,(2)
 
= µ,(1) (2239)
A + i Aµ,(2) − Aµ,(3)

Under a gauge transformation Û , the vectors Aµ,k are transformed in isospin

space. For a global gauge transformation, the transformation is a rotation in
isospin space. The gauge field Aµ is also required to transform as a four-vector
under Lorentz transformations.

We shall identify the contravariant derivative for the massive scalar particles
as170 as
Dµ = ∂ µ − i g Aµ − i g 0 Aµ0 (2240)
and one recognizes that this has the same form as the coupling of charged
particles to the EM field. In that case, the coupling occurs solely via τ (0) , the
coupling constant is given by g 0 = h̄q c and the field Aµ(0) = Aµ is the
four-vector potential. Since τ (0) commutes with all isospin operators, it is not
necessary to consider g 0 to be identical with the g value for the SU (2) gauge
fields.

15.3 The Free Gauge Fields

We have four real four-vector fields Aµ,k . These are the gauge fields. The free
gauge fields exist in the absence of the particles, and has a free Lagrangian. The
field strength tensors F µ,ν are given by the SU (2) generalized form of the EM
field tensor
F µ,ν = Dµ Aν − Dν Aµ (2241)
170 This can be related to the contravariant derivative familiar in the context of general

relativity, if one follows the logic adopted by Weyl and considers GR as a gauge field theory.

395
where D is the covariant derivative only involving the SU (2) triplet of gauge
fields. It should be noted that since the gauge fields do not commute, this
involves terms which are second-order in the field amplitudes. That is
   
F µ,ν = ∂ µ Aν − ∂ ν Aµ − i g Aµ Aν − Aν Aµ (2242)

The quadratic terms can be evaluated by using the commutation relations of

the isospin operators τ (k) . The k-th component of the SU (2) triplet of gauge
fields is given by
3
X
Fkµ,ν = ∂ µ
Aνk − ∂ ν
Aµk + g ξ i,j,k ( Aµi Aνj − Aνi Aµj )
{i,j}=1
3
X
= ∂ µ Aνk − ∂ ν Aµk + 2 g ξ i,j,k ( Aµi Aνj − Aνi Aµj )
{i>j}=1

(2243)
where the indices i and j are summed over and ξ i,j,k is the Levi-Civita symbol.
In arriving at the above expression, we have used the identity
3
X
τ (i) τ (j) = δ i,j τ (0) + i ξ i,j,k τ (k) (2244)
k=1

found by combining the anti-commutation and commutation relations for the

Pauli spin matrices. There is no contribution to the last term in the field tensor
from the U (1) gauge field Aµ(0) since τ (0) commutes with all other matrices. The
zeroth-component of the field tensor is simply given by
µ,ν
F(0) = ∂ µ Aν(0) − ∂ ν Aµ(0) (2245)

as expected for an electromagnetic field. Since the SU (2) gauge fields don’t
commute, the field theory is a non-Abelian gauge field theory. Under an SU (2)
transformation, the field tensors transform according to
F µ,ν → F µ,ν 0 = Û F µ,ν Û † (2246)
which is just a local unitary transform in isospin space. The Lagrangian density
for all the free gauge fields can be expressed as
1
Lgauge = − Trace F µ,ν Fµ,ν (2247)
32 π
where the Trace is evaluated in isospin space and takes into account that there
are a total of four fields. The Lagrangian density can be expressed directly in
terms of the contributions from four components of the field. The result can be
expressed as
3
1 X k
Lgauge = − Fµ,ν F µ,ν,k (2248)
16 π
k=0

396
where we have decomposed the fields as
X
k
Fµ,ν = Fµ,ν τ (k) (2249)
k

evaluated the product of the Pauli spin matrices and used the fact that the
Pauli spin matrices τ (k) for k 6= 0 are traceless.

One can consider the k-components of the vector potential Aµ (i.e. the three
real components Aµk for fixed µ) as forming three-vectors Aµ in isospin space.
These quantities transform as three-vectors under transformations in isospin
space, and also the Aµ transform as four-vectors under Lorentz transformations
in Minkowsky space-time. The three-vector fields are spin-one bosons with
isospin one. Hence, we might expect that the isospin triplet should contain two
oppositely charged particles and one uncharged particle. These particles are
supplemented by the particle corresponding to the single uncharged field Aµ(0) .
In terms of this set of isospin vectors, the free gauge field Lagrangian density
can be written in the form of a sum of a scalar product in isospin space and an
isospin scalar
  
1 µ ν ν µ µ ν
Lgauge = − (∂ A − ∂ A ) + 2g A ∧ A . (∂µ Aν − ∂ν Aµ ) + 2g Aµ ∧ Aν
16 π
  
1 µ ν ν µ
− ∂ A(0) − ∂ A(0) ∂µ Aν,(0) − ∂ν Aµ,(0) (2250)
16 π
It should be noted that the Lagrangian reduces to the sum of four non-interacting
electromagnetic Lagrangians in the limit g → 0. However, at finite values of
g, the Lagrangian density contains cubic and quartic interactions with coupling
strengths that are fixed by gauge invariance in terms of the single gauge param-
eter g.

Figure 71: The interaction vertices representing the interaction of three and
four isospin triplet gauge field bosons.

Exercise:

397
 Determine the equations of motion for the vector gauge fields, in the presence
of a source term
1
Lint = − Trace ( Aµ . j µ ) (2251)
c
where the current source j µ has also been decomposed in terms of Pauli spin
matrices.

It is convenient to introduce the two combinations

 
1
A±µ = √ A(1)
µ ∓ i A(2)
µ (2252)
2
which appear in the isospin matrix form of Aµ . These combinations are mutually
complex conjugate. Likewise, one can introduce the combinations of the field
tensors
 
± 1 (1) (2)
Fµ,ν = √ Fµ,ν ∓ i Fµ,ν (2253)
2
which are evaluated as
± ± ±
Fµ,ν = ( ∂µ ∓ 2 i g A(3) (3)
µ ) Aν − ( ∂ν ∓ 2 i g Aν ) Aµ (2254)

The third component of the field tensor can be written as

(3) − −
Fµ,ν = ( ∂µ A(3)
ν − ∂ν A(3) + +
µ ) + 2 i g ( Aµ Aν − Aν Aµ ) (2255)

In terms of these new combinations, the free Lagrangian for the gauge fields
become
1 1 1
Lgauge = − (0)
Fµ,ν F µ,ν,(0) − (3)
Fµ,ν F µ,ν,(3) − F − F µ,ν,+ (2256)
16 π 16 π 8 π µ,ν
where the first two terms are recognized as being similar to the Lagrangian den-
sity for the electromagnetic field. It was first hypothesized by Sheldon Glashow
that the electro-weak interaction is produced by the massless vector bosons de-
scribed by the above Lagrangian171 . Masses for the gauge bosons should not
be added by hand, since the resulting theory would not be renormalizable. To
retain renormalizability of the theory, and to have massive vector bosons, we
need to break the symmetry.

15.4 Breaking the Symmetry

We shall assume that our massive charged scalar boson field has broken symme-
try172 . The small amplitude excitations of the field with broken symmetry will
171 S.
L. Glashow, Nuclear Physics 22, 579-588, (1961).
172 S.
Weinberg, Phys. Rev. Lett. 19, 1264 (1967), Abdus Salam, Proc. of the 8th Nobel
Symposium, Stockholm (1968).

398
be modified, as will be the excitations of the gauge fields. Due to the symmetry-
breaking of the scalar field, the U (1) vector gauge field will become coupled to
the triplet of SU (2) gauge fields. When the symmetry is broken, the elementary
excitations of the coupled system of fields change and these new excitations will
represent the observable particles.

The potential for the two-component scalar field is chosen to be given by

 2  2
mc
V (Φ) = Φ† Φ − φ20 (2257)
2 h̄ φ0
where φ0 is a fixed real constant. The lowest-energy state described by this
potential is given by
Φ† Φ = φ20 (2258)
This state is degenerate with respect to global rotations in the four-dimensional
space of <e Φ1 , =m Φ1 , <e Φ2 , =m Φ2 which keeps the magnitude of Φ
constant and uniform over space.

The symmetry is broken by assuming that the physical ground state corre-
sponds to one specific choice of the uniform field Φ. Given the specific ground
state which the system chooses spontaneously, one can make use of the global
gauge invariance to describe the ground state Φ0 as a field which has one non-
zero component which is real. That is, αk can be chosen so that
 
<e Φ1
Φ0 =
0
 
φ0
= (2259)
0
The excited states can be expressed as
 
φ0 + χ1
Φ = (2260)
0
where the local gauge degrees of freedom have been used to make χ1 real. This
excited field is invariant under the transformation
Φ → Φ0 = ÛEM Φ (2261)
where ÛEM is restricted to have the form
 
1  0 
ÛEM =   (2262)
0 exp − i Λ

This is a transformation in which the U (1) transformation is combined with a

specific SU (2) transformation
   
Λ
  exp + i 2 0
Λ  
ÛEM = exp − i     (2263)
2
exp − i Λ2
 
0

399
and will turn out to represent the residual U (1) gauge invariance of the electro-
magnetic field.

The Lagrangian density for the isospin doublet of scalar fields and their
couplings can be evaluated for the excited state as
 2
mc
Lscalar = (Dµ Φ)† Dµ Φ − χ21 (2264)
h̄

where the covariant derivative of the doublet of scalar fields is given by

Aµ,(3) ( φ0 + χ1 )
 µ   µ,(0)   
µ ∂ χ1 A ( φ0 + χ1 ) √
D Φ = − i g0 −ig
0 0 2 Aµ,− ( φ0 + χ1 )
(2265)
A new interaction strength λ can be defined as
q
λ = g02 + g 2 (2266)

and on defining an angle θ via

g
tan θ = (2267)
g0
the coupling constants can be represented as

g0 = λ cos θ
g = λ sin θ (2268)

Thus, the covariant derivative has the connection with the field

AµZ = cos θ Aµ,(0) + sin θ Aµ,(3) (2269)

The field AµZ will turn out to be the field that describes the neutral Z particle.
The field orthogonal to the Z field is defined as

AµEM = − sin θ Aµ,(0) + cos θ Aµ,(3) (2270)

When expressed in terms of the transformed fields and constants, the covariant
derivative terms become
 µ   µ 
Dµ Φ =
∂ χ1
− i ( φ0 + χ1 ) √λ AZµ,− (2271)
0 g 2A

The lowest-order terms in the Lagrangian density of the non-uniform scalar field
and all its couplings to the gauge fields are expressed as
 2
mc
Lscalar = ∂µ χ1 ∂ µ χ1 − χ21 + λ2 φ20 AµZ Aµ,Z + 2 g 2 φ20 Aµ+ A− µ
h̄
(2272)

400
The higher-order terms, which have been neglected, describe the self-interactions
between the scalar field and the residual interactions between the scalar field
and the gauge fields.

The Lagrangian density for the free gauge field is

1 1 1
Lgauge = − F (0) F µ,ν,(0) − F (3) F µ,ν,(3) − F − F µ,ν,+ (2273)
16 π µ,ν 16 π µ,ν 8 π µ,ν
has to be expressed in terms of the new fields AµZ and AµEM . The inverse
transform is given by

Aµ,(0) = cos θ AµZ − sin θ AµEM

Aµ,(3) = sin θ AµZ + cos θ AµEM (2274)

If one defines FZµ,ν and FEM

µ,ν
as the transformed field tensors evaluated to lowest-
order in the fields

FZµ,ν = ∂ µ AνZ − ∂ ν AµZ

µ,ν
FEM = ∂ µ AνEM − ∂ ν AµEM (2275)

then the original field tensors can be expressed as

µ,ν
F(0) = cos θ FZµ,ν − sin θ FEM
µ,ν

µ,ν
F(3) = sin θ FZµ,ν + cos θ FEM
µ,ν
+ 2 i g ( Aµ− Aν+ − Aν− Aµ+ )
(2276)

The Lagrangian density describing the small amplitude excitations of the scalar
field and the gauge fields can be written as
 2
µ mc
LF ree = ∂µ χ1 ∂ χ1 − χ21
h̄
1
− Fµ,ν,Z FZµ,ν + λ2 φ20 AµZ Aµ,Z
16 π
1 µ,ν
− Fµ,ν,EM FEM
16 π
1
− F − F µ,ν,+ + 2 g 2 φ20 Aµ+ A− (2277)
8 π µ,ν µ

In electro-weak theory, the first term represents the free uncharged scalar
boson. The second term describes an uncharged vector particle with mass MZ
proportional λ φ0 . The third term describes the uncharged massless vector
particle known as the photon. From the equations of motion for Aµ,± , the
remaining term can be shown to describe a pair of charged particles with masses
MW proportional to g φ0 = sin θ λ φ0 . These particles are known as the W +

401
and W − particles. The W + and W − particles are charged and the observed
charges are ± e. The interaction mediated by the massive vector bosons is found
to have a finite range (≈ 10−18 m), and is responsible for the weak interaction.
The experimentally determined masses173 are MW c2 ≈ 80.33 GeV and MZ c2 ≈
91.187 GeV. Nearly all the parameters of this theory have been determined
through experiment, the only exception is the mass m of the scalar particle
which remains to be discovered. The ratio of the masses determines the angle
θ via174
MW
= sin θ (2278)
MZ
which yields sin θ ≈ 0.8810. The W ± particles carry electrical charges ±e since
they couple to the electromagnetic field. This can be seen by examining the
W ± field tensor
F±µ,ν = ( ∂ µ ∓ 2 i g Aµ,(3) ) Aν± − ( ∂ ν ∓ 2 i g Aν,(3) ) Aµ± (2279)
where
Aµ,(3) = sin θ AµZ + cos θ AµEM (2280)
Therefore, the covariant derivatives of the fields Dµ Aν± couple them to the elec-
tromagnetic field AµEM with either a positive or negative coupling constant of
magnitude 2 g cos θ. Since only electrically charged particles couple to the
electromagnetic field, one can make the identification
 
e
= 2 g cos θ = λ sin 2θ (2281)
h̄ c
which determines the coupling strengths. Furthermore, because the coupling
strengths have been completely determined, the observed masses can be used
to determine φ0 . This leads to the identification
sin2 2θ ( MZ c2 )2
φ20 = e2
(2282)
8 π h̄ c ( h̄c )
√
which leads to φ0 ≈ 178 GeV / h̄c, where h̄c ≈ 197 MeV fm. Hence, the
only undetermined parameter is the mass of the Higgs particle m. Recent ex-
periments175 have found a narrow resonance with an energy of approximately
126 GeV. The resonance has properties consistent with those expected of the
Higgs particle. The resonance decays either into two photons or two vector
bosons. Yang Mills theories, even if symmetry is spontaneously broken as in the
Weinberg-Salam theory, were shown to be renormalizable by G. t’ Hooft176 .

173 G. Arnison, A. Astbury, B. Aubert, et al., Phys. Lett. B, 122 103-116 (1983).

G. Arnison, O. C. Allkofer, A. Astbury, et al., Phys. Lett. B 147, 493-508 (1984).

174 It is customary to define the Weinberg angle θ MW
W via cos θW = M .
Z
175 Physics Letters B, 716, 1 (2012), Physics Letters B, 716, 30 (2012).
176 G. t’ Hooft, Nuclear Physics, B 35, 167 (1971), G. ’t Hooft and M. Veltman Nuclear

Physics B 44, 189 (1972).

402