Nothing is mightier than an idea

whose time has come.

V. M. Hugo (1802-1885)

12
Quantum Electrodynamics

In Chapter 7 we have learned how to quantize relativistic free fields and in Chap-
ters 10 and 11 how to deal with interactions if the coupling is small. So far, this was
only done perturbatively. Fortunately, there is a large set of physical phenomena
for which perturbative techniques are sufficient to supply theoretical results that
agree with experiment. In particular, there exists one theory, where the agreement
is extremely good. This is the quantized theory of interacting electrons and photons
called quantum electrodynamics, or shortly QED.

12.1 Gauge Invariant Interacting Theory

The free Lagrangians of electrons and photons are known from Chapter 5 as
e γ
L(x) = L(x) + L(x), (12.1)

with
e
L(x) = ψ̄(x) (i/
∂ − m) ψ(x), (12.2)
γ 1 µν 1 µ ν ν µ
L(x) = − Fµν (x)F (x) = − [∂µ Aν (x) − ∂ν Aµ (x)] [∂ A (x) − ∂ A (x)] .(12.3)
4 4
When quantizing the photon field, there were subtleties due to the gauge freedom
in the choice of the gauge fields Aµ :

Aµ (x) → Aµ (x) + ∂µ Λ(x). (12.4)

For this reason, there were different ways of constructing a Hilbert space of free
particles. The first, described in Subsection 7.5.1, was based on the quantization
of only the two physical transverse degrees of freedom. The time component of
the gauge field A0 and the spatial divergence ∇ · A(x) had no canonically conjugate
field and were therefore classical fields, with no operator representation in the Hilbert
space. The two fields are related by Coulomb’s law which reads, in the absence of
charges:
∇2 A0 (x) = −∂ 0 ∇ · A(x). (12.5)

801
802 12 Quantum Electrodynamics

Only the transverse components, defined by

 
Ai⊥ (x) ≡ δ ij − ∇i ∇j /∇2 Aj (x, t), (12.6)
were operators. These components represent the proper dynamical variables of the
system. After fulfilling the canonical commutation rules, the positive- and negative-
frequency parts of these fields define creation and annihilation operators for the
electromagnetic quanta. These are the photons of right and left circular polarization.
This method had an esthetical disadvantage that two of the four components of
the vector field Aµ (x) require a different treatment. The components which become
operators change with the frame of reference in which the canonical quantization
procedure is performed.
To circumvent this, a covariant quantization procedure was developed by Gupta
and Bleuler in Subsection 7.5.3. In their quantization scheme, the propagator took a
pleasant covariant form. But this happened at the expense of another disadvantage,
that this Lagrangian describes the propagation of four particles of which only two
correspond to physical states. Accordingly, the Hilbert space contained two kinds
of unphysical particle states, those with negative and those with zero norm. Still,
a physical interpretation of this formalism was found with the help of a subsidiary
condition that selects the physical subspace in the Hilbert space of free particles.
The final and most satisfactory successful quantization was developed by Faddeev
and Popov [1] and was described in Subsection 7.5.2. It started out by modifying
the initial photon Lagrangian by a gauge-fixing term
LGF (x) = −D∂ µ Aµ (x) + αD 2 (x)/2. (12.7)
After that, the quantization can be performed in the usual canonical way.

12.1.1 Reminder of Classical Electrodynamics of Point Particles

In this chapter we want to couple electrons and photons with each other by an
appropriate interaction and study the resulting interacting field theory, the famous
quantum electrodynamics (QED). Since the coupling should not change the two
physical degrees of freedom described by the four-component photon field Aµ , it is
important to preserve the gauge invariance, which was so essential in assuring the
correct Hilbert space of free photons. The prescription how this can be done has been
known for a long time in the context of classical electrodynamics of point particles.
In that theory, a free relativistic particle moving along an arbitrarily parametrized
path xµ (τ ) in four-space is described by an action
s #1/2
v2 (t)
"
dxµ dxµ
Z Z
2
A = −mc dτ = −mc2 dt 1 − 2 , (12.8)
dτ dτ c
where x0 (τ ) = t is the time and dx/dt = v(t) the velocity along the path. If the
particle has a charge e and lies at rest at position x, its potential energy is
V (t) = eφ(x, t), (12.9)
12.1 Gauge Invariant Interacting Theory 803

where

φ(x, t) = A0 (x, t). (12.10)

In our convention, the charge of the electron e has a negative value to agree with
the sign in the historic form of the Maxwell equations

∇ · E(x) = −∇2 φ(x) = ρ(x),

∇ × B(x) − Ė(x) = ∇ × ∇ × A(x) − Ė(x)
h i 1
= − ∇2 A(x) − ∇ · ∇A(x) − Ė(x) = j(x). (12.11)
c
If the electron moves along a trajectory x(t), the potential energy becomes

V (t) = eφ (x(t), t) . (12.12)

In the Lagrangian L = T − V , this contributes with the opposite sign

Lint (t) = −eA0 (x(t), t) , (12.13)

adding a potential term to the interaction

Z
int
A |pot = −e dtA0 (x(t), t) . (12.14)

Since the time t coincides with x0 (t)/c of the trajectory, this can be expressed as
eZ
Aint |pot = − dx0 A0 (x). (12.15)
c
In this form it is now quite simple to write down the complete electromagnetic
interaction purely on the basis of relativistic invariance. The minimal Lorentz-
invariant extension of (12.15) is obviously
eZ
Aint = − dxµ Aµ (x). (12.16)
c
Thus, the full action of a point particle can be written, more explicitly, as
e
Z Z Z
A = dtL(t) = −mc ds − dxµ Aµ (x)
c
#1/2
v2
"
1
Z Z  
2 0
= −mc dt 1 − 2 −e dt A − v · A . (12.17)
c c

The canonical formalism supplies the canonically conjugate momentum

∂L v e e
P= = mq + A ≡ p + A. (12.18)
∂v 1 − v2 /c2 c c
804 12 Quantum Electrodynamics

Thus the velocity is related to the canonical momentum and external vector potential
via
v P − ec A
= r . (12.19)
c e
2
P− c
A + m2 c2

This can be used to calculate the Hamiltonian as the Legendre transform

∂L
H= v − L = P · v − L, (12.20)
∂v
with the result
s
2
e

H=c P− A + m2 c2 + eA0 . (12.21)
c
At nonrelativistic velocities, this has the expansion
2
1 e

2
H = mc + P− A + eA0 + . . . . (12.22)
2m c
The rest energy mc2 is usually omitted in this limit.

12.1.2 Electrodynamics and Quantum Mechanics

When going over from quantum mechanics to second-quantized field theory in Chap-
ter 2, we found the rule that a nonrelativistic Hamiltonian

p2
H= + V (x) (12.23)
2m
becomes an operator

∇2
Z " #
3 †
H= d x ψ (x, t) − + V (x) ψ(x, t). (12.24)
2m

For brevity, we have omitted a hat on top of H and the fields ψ † (x, t), ψ(x, t).
Following the rules of Chapter 2, we see that the second-quantized form of the
interacting nonrelativistic Hamiltonian in a static A(x) field with the Hamiltonian
(12.22) (minus mc2 ),

(P − eA)2
H= + eA0 , (12.25)
2m
is given by
" 2 #
1 e
Z 
3 † 0
H= d x ψ (x, t) − ∇−i A + eA ψ(x, t). (12.26)
2m c
12.1 Gauge Invariant Interacting Theory 805

The action of this theory reads

Z Z Z   
A= dtL = dt d x ψ † (x, t) i∂t + eA0 ψ(x, t)
3

2
1 † e
 
+ ψ (x, t) ∇ − i A ψ(x, t) . (12.27)
2m c
It is easy to verify that (12.26) reemerges from the Legendre transform
∂L
H= ψ̇(x, t) − L. (12.28)
∂ ψ̇(x, t)
The action (12.27) holds also for time-dependent Aµ (x)-fields.
These equations show that electromagnetism is introduced into a free quantum
theory of charged particles following the minimal substitution rule
e
∇ → ∇ − i A(x, t),
c
∂t → ∂t + ieA0 (x, t), (12.29)
or covariantly:
e
∂µ → ∂µ − i Aµ (x). (12.30)
c
The substituted action has the important property that the gauge invariance of
the free photon action is preserved by the interacting theory: If we perform the
gauge transformation
Aµ (x) → Aµ (x) + ∂ µ Λ(x), (12.31)
i.e.,
A0 (x, t) → A0 (x, t) + ∂t Λ(x, t),
A(x, t) → A(x, t) − ∇Λ(x, t), (12.32)
the action remains invariant provided thet we simultaneously change the fields
ψ(x, t) of the charged particles by a spacetime-dependent phase
ψ (x, t) → e−i(e/c)Λ(x,t) ψ(x, t). (12.33)
Under this transformation, the space and time derivatives of the field change like
e
 
−i(e/c)Λ(x,t)
∇ψ(x, t) → e ∇ − i ∇Λ(x, t) ψ,
c
−i(e/c)Λ(x,t)
∂t ψ → e (∂t − ie∂t Λ) ψ(x, t). (12.34)
The covariant derivatives in the action (12.27) have therefore the following simple
transformation law:
e e
   
−i(e/c)Λ(x,t)
∇ − i A ψ(x, t) → e ∇ − i A ψ(x, t),
c  c
e 0
  
∂t + i A ψ(x, t) → e−i(e/c)Λ(x,t) ∂t + ieA0 ψ(x, t). (12.35)
c
806 12 Quantum Electrodynamics

These combinations of derivatives and gauge fields are called covariant derivatives
and are written as
e
 
Dψ(x, t) ≡ ∇ − i A ψ(x, t),

c 
Dt ψ(x, t) ≡ ∂t + ieA0 ψ(x, t), (12.36)

or, in four-vector notation, as

e
 
Dµ ψ(x) = ∂µ + i Aµ ψ(x). (12.37)
c
Here the adjective “covariant” does not refer to the Lorentz group but to the gauge
group. It records the fact that Dµ ψ transforms under local gauge changes (12.29)
of ψ in the same way as ψ itself in (12.33):

Dµ ψ(x) → e−i(e/c)Λ(x) Dµ ψ(x). (12.38)

With the help of such covariant derivatives, any action which is invariant under
global phase changes by a constant phase angle [i.e., U(1)-invariant in the sense
discussed in Section 8.11.1]

ψ(x) → e−iα ψ(x), (12.39)

can easily be made invariant under local gauge transformations (12.31). We merely
have to replace all derivatives by covariant derivatives (12.37), and add to the field
Lagrangian the gauge-invariant photon expression (12.3).

12.1.3 Principle of Nonholonomic Gauge Invariance

The minimal substitution rule can be viewed as a consequence of a more general
principle of nonholonomic gauge invariance. The physics of the initial action (12.17)
is trivially invariant under the addition of a term
e
Z
∆A = − dt ẋµ (t)∂µ Λ(x). (12.40)
c
The integral runs over the particle path and contributes only a pure surface term
from the endpoints:
e
∆A = − [Λ(xb ) − Λ(xa )] . (12.41)
c
This does not change the particle trajectories. If we now postulate that the dynam-
ical laws of physics remain also valid when we admit multivalued gauge functions
Λ(x) for which the Schwarz integrability criterion is violated, i.e., which possess
noncommuting partial derivatives:

(∂µ ∂ν − ∂ν ∂µ )Λ(x) 6= 0. (12.42)

12.1 Gauge Invariant Interacting Theory 807

Then the derivatives

Aµ (x) = ∂µ Λ(x) (12.43)
have a nonzero curl Fµν = (∂µ Aν − ∂µ Aν ) = (∂µ ∂ν − ∂ν ∂µ )Λ(x) 6= 0, and the action
(12.41) coincides with the interaction (12.16).
Similarly we can derive the equations of motion of a wave function in an elec-
tromagnetic field from that in field-free space by noting the trivial invariance of
quantum mechanics without fields under gauge transformations (12.33), and by ex-
tending the set of permissible gauge functions Λ(x) to multivalued functions for
which the partial derivatives do not commute as in (12.42).
In either case, the nonholonomic gauge transformations convert the physical laws
obeyed by a particle in Euclidean spacetime without electromagnetism into those
with electromagnetic fields.
This principle is discussed in detail in the literature [11]. It can be generalized
to derive the equations of motion in a curved spacetime from those in flat spacetime
by nonholonomic coordinate transformations which introduce defects in spacetime.

12.1.4 Electrodynamics and Relativistic Quantum Mechanics

Let us follow this rule for relativistic electrons and replace, in the Lagrangian (12.2),
the differential operator ∂/ = γ µ ∂µ by
e e
   
γ µ ∂µ + i Aµ = ∂/ + i A/ ≡D
/. (12.44)
c c
In this way we arrive at the Lagrangian of quantum electrodynamics (QED)

1 2
L(x) = ψ̄(x) (i/
D − m) ψ(x) − Fµν . (12.45)
4
The classical field equations can easily be found by extremizing the action with
respect to all fields, which gives

δA
= (i/
D − m) ψ(x) = 0, (12.46)
δ ψ̄(x)
δA 1
= ∂ν F νµ (x) − j µ (x) = 0, (12.47)
δAµ (x) c

where j µ (x) is the current density:

j µ (x) ≡ ecψ̄(x)γ µ ψ(x). (12.48)

Equation (12.47) coincides with the Maxwell equation for the electromagnetic field
around a classical four-dimensional vector current j µ (x):

1
∂ν F νµ (x) = j µ (x). (12.49)
c
808 12 Quantum Electrodynamics

In the Lorenz gauge ∂µ Aµ (x) = 0, this equation reduces to

1
−∂ 2 Aµ (x) = j µ (x). (12.50)
c
The current density j µ combines the charge density ρ(x) and the spatial current
density j(x) of particles of charge e in a four-vector:

j µ = (cρ, j) . (12.51)

In terms of electric and magnetic fields E i = F i0 , B i = −F jk , the field equations

(12.49) turn into the Maxwell equations

∇ · E = ρ = eψ̄γ 0 ψ = eψ † ψ, (12.52)
1 e
∇ × B − Ė = j = ψ̄ ψ. (12.53)
c c
The first is Coulomb’s law, the second is Ampère’s law in the presence of charges
and currents.
Note that the physical units employed here differ from those used in many books
of classical electrodynamics [12], by the absence of a factor 1/4π on the right-hand
side. The Lagrangian used in those books is
1 2 1
L(x) = − Fµν (x) − j µ (x)Aµ (x) (12.54)
8π c
1 h 2 i  1

2
= E (x) − B (x) − ρ(x)φ(x) − j(x) · A(x) ,
4π c
which leads to Maxwell’s field equations

∇ · E = 4πρ,
4π
∇ × B − Ė = j. (12.55)
c
The form employed
√ conventionally
√ in quantum field theory arises from this by re-
placing A → 4πA and e → − 4πe2 . The charge of the electron in our units has
therefore the numerical value
√ q
e = − 4πα ≈ − 4π/137 (12.56)
√
rather than e = − α.

12.2 Noether’s Theorem and Gauge Fields

In electrodynamics, the conserved charge resulting from the U(1)-symmetry of the
matter Lagrangian by Noether’s theorem (recall Chapter 8) is the source of a mass-
less particle, the photon. This is described by a gauge field which is minimally
coupled to the conserved current. A similar structure will be seen in Chapters 27
12.2 Noether’s Theorem and Gauge Fields 809

and 28 to exist for many internal symmetries giving rise to nonabelian versions of
the photon, for instance the famous W - and Z-vector mesons, which mediate the
weak interactions, or the gluons which give rise to strong interactions. It is useful
to recall Noether’s derivation of conservation laws in such theories.
For a a locally gauge invariant theory, the conserved matter current can no longer
be found by the rule (8.118), which was so useful in the globally invariant theory.
Indeed, in quantum electrodynamics, the derivative with respect to the local field
transformation ǫ(x) would be simply given by

δL
jµ = , (12.57)
∂∂µ Λ

since this would be identically equal to zero, due to local gauge invariance. We may,
however, subject just the matter field to a local gauge transformation at fixed gauge
fields. Then we obtain the correct current

∂L 
jµ ≡  . (12.58)
∂∂µ Λ γ

Since the complete change under local gauge transformations δsx L vanishes identi-
cally, we can alternatively vary only the gauge fields and keep the electron field
fixed 
∂L 
jµ = −  . (12.59)
∂∂µ Λ e
This is done most simply by forming the functional derivative with respect to the
γ
gauge field, and by omitting the contribution of L:
e
∂L
jµ = − . (12.60)
∂∂µ Λ

An interesting consequence of local gauge invariance can be found for the gauge
field itself. If we form the variation of the pure gauge field action
 γ 
γ Aǫ 
Z
δs A = d4 x tr δsx Aµ , (12.61)
δAµ

and insert, for δsx A, an infinitesimal pure gauge field configuration

δsx Aµ = −i∂µ Λ(x), (12.62)

the variation must vanish for all Λ(x). After a partial integration, this implies the
local conservation law ∂µ j µ (x) = 0 for the current
γ
µ δA
j (x) = −i . (12.63)
δAµ
810 12 Quantum Electrodynamics

In contrast to the earlier conservation laws derived for matter fields which were valid
only if the matter fields obey the Euler-Lagrange equations, the current conservation
law for gauge fields is valid for all field configurations. It is an identity, often
called Bianchi identity due to its close analogy with certain identities in Riemannian
geometry.
To verify this, we insert the Lagrangian (12.3) into (12.63) and find j ν =
∂µ F µν /2. This current is trivially conserved for any field configuration due to the
antisymmetry of F µν .

12.3 Quantization
The canonical formalism can be used to identify canonical momenta of the fields
ψ(x) and Ai (x):

∂L
πψ (x) = = ψ † (x), (12.64)
∂ψ(x)

and

πAi (x) ≡ π i (x) = F 0i (x) = −E i (x), (12.65)

and to find the Hamiltonian density

∂L ∂L(x)
H(x) = ψ̇(x) + − L(x) (12.66)
∂ ψ̇(x) ∂ Ȧk (x)
1
= ψ̄(−i ∇ + m)ψ + (E2 + B2 ) + ∇A0 · E + eψ̄γµ ψAµ .
2
Here, and in all subsequent discussions, we use natural units in which the light
velocity is equal to unity.
The quantization procedure in the presence of interactions now goes as follows:
The Dirac field of the electron has the same equal-time anticommutation rules as in
the free case:

{ψ a (x, t), ψb† (x′ , t)} = δ (3) (x′ − x)δab ,

{ψ(x, t), ψ(x′ , t)} = 0, (12.67)
{ψ † (x, t)ψ † (x′ t)} = 0.

For the photon field we first write down the naive commutation rules of the spatial
components:

− [Ei (x, t), Aj (x′ , t)] = −iδ (3) (x − x′ )δij (12.68)

[Ai (x, t), Aj (x′ , t)] = 0, (12.69)
[Ei (x, t), Ej (x′ , t)] = 0. (12.70)
12.3 Quantization 811

As in Eq. (7.346), the first commutator cannot be true here, since by Coulomb’s
law (12.52):

∇ · E = −∇ · Ȧ − ∇2 A0
= eψ † ψ, (12.71)

and we want the canonical fields Ai to be independent of ψ, ψ † , and thus to commute

with them. The contradiction can be removed just as in the free case by postulating
(12.68) only for the transverse parts of Ei , and using δijT (x − x′ ) as in (7.347), while
letting the longitudinal part ∇ · A(x, t) be a c-number field, since it commutes with
all Ai (x, t). The correct commutation rules are the following:
h i
Ȧj (x, t), Aj (x′ , t) = −iδijT (x − x′ ),
[Aj (x, t), Aj (x′ , t)] = 0, (12.72)
h i
′
Ȧi (x, t), Ȧj (x , t) = 0.

To calculate the temporal behavior of an arbitrary observable composed of ψ, ψ † ,

and of Ai , Ȧi fields, only one more set of commutation rules has to be specified
which are those with A0 . This field occurs in the Hamiltonian density (12.66) and
is not one of the canonical variables. Moreover, in contrast to the free-field case in
Section 7.5, it is no longer a c-number. To see this, we express A0 in terms of the
dynamical fields using Coulomb’s law (12.71):
1 1
Z  
A0 (x, t) = d3 x′ eψ †
ψ + ∇ · Ȧ (x′ , t). (12.73)
4π |x′ − x|
This replaces Eq. (4.268) in the presence of charges. In an infinite volume with
asymptotically vanishing fields, there is no freedom of adding a solution of the
homogeneous Poisson equation (4.269). Hence, whereas ∇ · A is a c-number field,
the time component of the gauge field A0 is now a non-local operator involving the
fermion fields. Since these are independent of the electromagnetic field, A0 still
commutes with the canonical Ai , Ȧj fields:
h i h i
A0 (x, t), Ai (x′ ) = A0 (x, t), Ȧi (x′ , t) = 0. (12.74)

The commutator with the Fermi fields, on the other hand, is nonzero:
h i e
A0 (x, t), ψ(x′ , t) = − ψ(x, t). (12.75)
4π|x − x′ |

Note the peculiar property of A0 : It does not commute with the electron field, no
matter how large the distance between the space points is. This property is called
nonlocality. It is a typical property of the present transverse covariant quantization
procedure.
Certainly, the arbitrary c-number function ∇ · A(x, t) can be made zero by an
appropriate gauge transformation, as in (4.257).
812 12 Quantum Electrodynamics

In the Hamiltonian, the field A0 can be completely removed by a partial inte-

gration:
Z Z
3 0
d x ∇A · E = − d3 x A0 ∇ · E, (12.76)

if we set the surface term equal to zero. Using now the field equation

∇ · E = eψ † ψ, (12.77)

we derive
Z
H = d3 x H (12.78)
e 1 2
Z    
d3 x ψ̄ −i · (∇ − i A) + m ψ +

= E + B2 .
c 2
When looking at this expression, one may wonder where the electrostatic interaction
has gone. The answer is found by decomposing the electric field

E i = −∂ 0 Ai + ∂ i A0 (12.79)

into longitudinal and transverse parts ELi and ETi with EL · ET = 0:

∂i∂i
!
ELi i 0
= ∂ A − 2 Aj ,
∇
∂i∂j
!
ETi = −∂ δ − 2 Aj .
0 ij
(12.80)
∇
Then the field energy becomes
1Z 3  2 2
 1Z 3  2 2
 1Z 3 2
d x E +B = d x ET + B + d xEL . (12.81)
2 2 2
Using (12.73), we see that the longitudinal field is simply given by
1 1
Z
EL (x) = − ∇ d3 x′ eψ † (x′ , t)ψ(x′ , t)
4π |x − x′ |
1
 
†
= ∇ eψ (x)ψ(x) . (12.82)
∇2
It is the Coulomb field caused by the charge density of the electron eψ † (x)ψ(x). The
field energy carried by ELi (x) is
2
1 e2 1 †
Z Z   
d3 x E2L (x) = d3 x ∇ ψ (x)ψ(x)
2 2 ∇2
e2 1
Z
=− d3 x ψ † (x)ψ(x) 2 ψ † (x)ψ(x)
2 ∇
e2 1
Z
= d3 xd3 x′ ψ † (x, t)ψ(x, t) ψ † (x′ , t)ψ(x′ , t). (12.83)
8π |x − x′ |
12.4 Perturbation Theory 813

This coincides precisely with the classical Coulomb energy associated with the charge
1 R 3
density (12.52). The term 2 d x (E2T + B2 ) in Eq. (12.81), on the other hand, is
an operator and contains the energy of the field quanta.
In order to develop a perturbation theory for QED in this quantization, we
must specify the free and interacting parts of the action. Since A0 and ∇·A are
unquantized and appear only quadratically in the action, they may be eliminated
in the action in the same way as in the energy, so that the action becomes
1h 2
Z  i 
A= d4 x ψ̄(x)(i/
∂ − M)ψ(x) + ET (x) − B2 (x) + Aint . (12.84)
2
The first two terms are the actions of the Dirac field ψ and transverse electromagnetic
fields AT , and Aint denotes the interaction
e2 1 1
Z Z Z
int 3 3 ′ †
A =− dt d xd x ψ (x, t)ψ(x, t) ′
ψ(x′ , t)ψ(x′ , t) +
j · AT . d4 x
8π |x − x | c
(12.85)
The interaction contains two completely different terms: The first is an instanta-
neous Coulomb interaction at a distance, which takes place without retardation and
involves the charge density. It is a nontrivial field-theoretic exercise to show that
the absence of retardation in the first term is compensated by current-current inter-
action resulting from the second term, so that it does not cause any conflicts with
relativity. This will be done at the end of Section 14.16.
The special role of the Coulomb interaction is avoided from the beginning in the
Gupta-Bleuler quantization procedure that was discussed in Subsection 7.5.2. There
the free action was
1
Z  
4
A= ∂ − m) ψ − F µν Fµν − D∂ µ Aµ + D 2 /2 ,
d x ψ̄ (i/ (12.86)
4
and the interaction had the manifestly covariant form
Z
int
A =− d4 x j µ Aµ . (12.87)

12.4 Perturbation Theory

Let us now set up the rules for building the Feynman diagrams to calculate the
effect of the interaction. In this context, we shall from now on attach, to the free
propagator, a subscript 0. The propagator of the free photon depends on the gauge.
It is most simple in the Gupta-Bleuler quantization scheme, where [see (7.510)]
d4 q i
Z
Gµν ′ µν ′ µν ′
0 (x, x ) = −g G0 (x, x ) = −g 4 2
e−ik(x−x ) . (12.88)
(2π) q + iη
Since we want to calculate the effect of interactions, we shall from now on attach to
the free propagator a subscript 0.
814 12 Quantum Electrodynamics

The free-particle propagator of the electrons was given in (7.289):

d4 p i
Z
′ ′
S0 (x − x ) = 4
e−ip(x−x ) . (12.89)
(2π) p/ − M + iη
int
In a Wick expansion of eiA , each contraction is represented by one of these two
propagators:

Aµ (x)Aν (x′ ) = Gµν ′

0 (x − x ), (12.90)
ψ(x)ψ̄(x′ ) = S0 (x − x′ ). (12.91)

In the Feynman diagrams, they are pictured by the lines

i
= −g µν
q2 + iǫ , (12.92)
i
= . (12.93)
p/ − m
The interaction Lagrangian

Lint = −eψ̄γ µ Aµ (x) (12.94)

is pictured by the vertex

= −eγ µ .
(12.95)

With these graphical elements we must form all Feynman diagrams which can con-
tribute to a given physical process.
In the transverse quantization scheme, the Feynman diagrams are much more
complicated. Recalling the propagator (7.361), the photon line stands now for

µν i
= Pphys (q) , (12.96)
q2 + iǫ

with the physical off-shell polarization sum [compare (12.7)]

µν qµ qν qµ ην + qν ηµ q2
Pphys (q) = −g µν − + qη − η η
µ ν . (12.97)
(qη)2 − q 2 (qη)2 − q 2 (qη)2 − q 2

The photon propagator is very complicated due to the appearance of the frame-
dependent auxiliary vector η = (1, 0, 0, 0). As a further complication, there are
12.4 Perturbation Theory 815

diagrams from the four-fermion Coulomb interactions in (12.85). These can be

pictured by a photon exchange diagram

i 0
= 2
γ × γ 0. (12.98)
q

They may be derived from an auxiliary interaction

Z
Aint = − d4 x j 0 A0 , (12.99)

assuming the A0 -field to have the propagator

i (12.100)
0 0 = .
q2

If this propagator is added to the physical one, it cancels precisely the last term in
the off-shell polarization sum (12.97), which becomes effectively

µν qµ qν qµ ην + qν ηµ
Pphys,eff (k) = −g µν − 2 2
+ qη . (12.101)
(qη) − q (qη)2 − q 2
Of course, the final physical results cannot depend on the frame in which the
theory is quantized. Thus it must be possible to drop all η-dependent terms. We
shall now prove this in three steps:
First, a photon may be absorbed (or emitted) by an electron which is on their
mass shell before and after the process. The photon propagator is contracted with
an electron current as follows
µν
ū(p′ , s′3 )γµ u(p, s′3 )Pphys,eff (q). q = p′ − p. (12.102)

Since the spinors on the right and left-hand side satisfy the Dirac equation, the
current is conserved and satisfies

ū(p′ , s′3 )γµ u(p, s′3 )q µ = 0. (12.103)

This condition eliminates the terms containing the vector q µ in the polarization sum
(12.101). Only the reduced polarization sum
µν
Pred (q) = −g µν (12.104)

survives, which is the polarization tensor of the Gupta-Bleuler propagator (12.88).

The same cancellation occurs if a photon is absorbed by an internal line, although
due to a slightly more involved mechanism. An internal line may arise in two ways.
816 12 Quantum Electrodynamics

Figure 12.1 An electron on the mass shell absorbing several photons.

An electron may enter a Feynman diagram on the mass shell, absorb a number of
photons, say n of them, and leave again on the mass shell as shown in Fig. 12.1.
The associated off-shell amplitude is
1 1 1 1
a(p′ , p, qi ) = ′ q/ n q/ n−1 q/ n−1 · · · q/ 3 q/ 2 q/ 1 . (12.105)
p/ − M p/ n−1 − M p/ 2 − M p/ 1 − M
It has to be amputated and evaluated between the initial and final spinors, which
amounts to multiplying it from the left and right with ū(p′ , s′3 )(/
p ′ − M) and with
(/
p −M)u(p, s3 ), respectively. If an additional photon is absorbed, it must be inserted
as shown in Fig. 12.2. At each vertex, there is no current conservation since the

Figure 12.2 An electron on the mass shell absorbing several photons, plus one additional
photon.

photon lines are not on their mass shell. Nevertheless, the sum of all n + 2 diagrams
does have a conserved current.
To prove this we observe the following Ward-Takahashi identity for free particles
[2, 3]:
1 1 1 1
q/ = − . (12.106)
p/ r + q/ − M p/ r − M p/ r − M p/ r + q/ − M
More details on this important identity will be given in the next section.
The sum of all off-shell absorption diagrams can be written as
1 1
a(p′ , p, qi ; q) = ′ q/ n q/ n−1 · · ·
p/ + q/ − M p/ n−1 + q/ − M
12.4 Perturbation Theory 817

1 1 1
··· q/ · · · q/ 2 q/ 1 u(p, s3), (12.107)
p/ r + q/ − M p/ r − M p/ 1 − M

to be evaluated between ū(p′ + q, s′3 )(/ p ′ + q/ − M) and (/

p − M)u(p, s3 ). With
the help of the Ward-Takahashi identity we can now remove the q/ recursively from
a(p′ , p, qi ; q) and remain with the difference

a(p′ , p, qi ; q) = a(p′ , p, qi ) − a(p′ + q, p + q, qi ). (12.108)

When evaluating the right-hand side between the above spinors, we see that the
first term in the difference vanishes since the left-hand spinor satisfies the Dirac
equation. The same thing holds for the second term and the right-hand spinor.
Thus the polarization sum in the photon propagator can again be replaced by the
reduced expression (12.104).
Finally, the electron line can be closed to a loop as shown in Fig. 12.3. Here the

Figure 12.3 An internal electron loop absorbing several photons, plus an additional
photon, and leaving again on the mass shell.

amplitude (12.108) appears in a loop integral with an additional photon vertex:

d4 p d4 p
Z Z
q a(p′ , p, qi ; q)] =
tr[/ q [a(p′ , p, qi ) − a(p′ + q, p + q, qi )]}. (12.109)
tr{/
(2π)4 (2π)4

If the divergence of the integral is made finite by a dimensional regularization, the

loop integral is translationally invariant in momentum space and the amplitude
difference vanishes. Hence, also in this case, the polarization sum can be replaced
by the reduced expression (12.104).
Thus we have shown that due to current conservation and the Ward-Takahashi
identity, the photon propagator in all Feynman diagrams can be replaced by

(12.110)
818 12 Quantum Electrodynamics

µν −ig µν
= G (q) = 2 .
q + iη

As a matter of fact, for the same reason, any propagator

" #
µν i qµ qν
= G (q) = − 2 gµν − 2 (1 − α) (12.111)
q q
can be used just as well, and the parameter α is arbitrary. Indeed, this is the
propagator arising when adding to the gauge-invariant Lagrangian in (12.84) the
gauge-fixing expression
α 2
LGF = D − D∂ µ Aµ . (12.112)
2
For the value α = 1 favored by Feynman, the propagator (12.4) reduces to (12.110).

12.5 Ward-Takahashi Identity

From the application of Eq. (12.106), it is apparent that the Ward-Takahashi identity
plays an important role in ensuring the gauge invariance of loop diagrams. In fact,
the renormalizability of quantum electrodynamics was completed only after making
use of the diagonal part of it, which was the original Ward identity. For free particles,
we observe that the Ward-Takahashi identity (12.106) can be written in terms of
the electron propagator (12.90), and the free vertex function Γµ0 (p′ , p) = γ µ , as

S0−1 (p′ ) − S0−1 (p) = i(p − q)µ Γµ0 (p′ , p). (12.113)

The original Ward identity is obtained from this by forming the limit p′ → p:
∂ −1
S0 (p) = iΓµ0 (p, p). (12.114)
∂pµ
The important contribution of Ward and Takahashi was to prove that their
identity is valid for the interacting propagators and vertex functions, order by order
in perturbation theory. Thus we may drop the subscripts zero in Eq. (12.113) and
write
S −1 (p′ ) − S −1 (p) = i(p′ − p)µ Γµ (p′ , p). (12.115)
This identity is a general consequence of gauge invariance, as was first conjectured
by Rohrlich [4].
For the general proof of (12.115), the key observation is that the operator version
of the fully interacting electromagnetic current j µ (x) = eψ̄(x)γ µ ψ(x) satisfies, at
equal times, the commutation rules with the interacting electron and photon fields

[j 0 (x), ψ(y)]δ(x0 − y 0 ) = −eγ 0 ψ(x) δ(x0 − y 0 ), (12.116)

[j 0 (x), ψ̄(y)]δ(x0 − y 0 ) = eψ̄(x)γ 0 δ(x0 − y 0), (12.117)
[j 0 (x), Āµ (y)]δ(x0 − y 0 ) = 0. (12.118)
12.6 Magnetic Moment of Electron 819

This follows directly from the canonical equal-time anticommutation rules of the
electrons written in the form

{ψ(x), ψ † (y)}δ(x0 − y 0) = δ (4) (x − y). (12.119)

As a consequence of (12.116)–(12.118), we find for any local operator O(x):

   
∂µ T̂ j µ (x)O(y) = j 0 (x)O(yi ) δ(x0 − y 0) + T̂ (∂µ j µ (x)O(y)) . (12.120)

The first term on the right-hand side arises when the derivative is applied to the
Heaviside functions in the definition (2.232) of the time-ordered product. The gen-
eralization to many local operators reads:
 n
 X h i
∂µ T̂ j µ (x)O(y1 ) · · · O(yi ) · · · O(yn ) = T̂ O(y1 ) · · · j 0 (x), O(yi ) · · · O(yn )δ(x0−yi0 )
i=1

+ T̂ (∂µ j µ (x)O(y1) · · · O(yi ) · · · O(yn )). (12.121)

Since the electromagnetic current is conserved, the last term vanishes.

A particular case of (12.121) for a conserved current is the relation
   
∂µ T̂ j µ (x)ψ̄(y1 )ψ(y2 ) = eT̂ ψ̄(y1 )ψ(y2 ) [δ(x − y1 ) − δ(x − y2 )] . (12.122)

Taking this between single-particle states and going to momentum space yields an
identity that is valid to all orders in perturbation theory [5]

−i(p′ − p)µ S(p′ )Γµ (p′ , p)S(p) = S(p′ ) − S(p). (12.123)

This is precisely the Ward-Takahashi identity (12.113).

12.6 Magnetic Moment of Electron

For dimensional reasons, the magnetic moment of the electron is proportional to the
Bohr magnetic moment
eh̄
µB = . (12.124)
2Mc
Since it is caused by the spin of the particle, it is proportional to it and can be
written as
s
 = gµB . (12.125)
h̄
The proportionality factor g is called the gyromagnetic ratio. If the spin is polarized
in the z-direction, the z-component of  is
1 eh̄ 1
µ = gµB =g . (12.126)
2 2Mc 2
We have discussed in Subsec. 4.15 that, as a result of the Thomas precession, an
explanation of the experimental fine structure will make the g-factor of the electron
820 12 Quantum Electrodynamics

magnetic moment to have a value near 2. This is twice as large as that of a charged
rotating sphere of angular momentum L, whose magnetic moment is
eh̄ L
= , (12.127)
2Mc h̄
i.e., whose g-value is unity. The result g = 2 has been found also in Eq. (6.119) by
bringing the Dirac equation in an electromagnetic field to the second-order Pauli
form (6.110).
Let us convince ourselves that a Dirac particle possesses the correct gyromagnetic
ratio g = 2. Consider an electron of momentum p in a electromagnetic field which
changes the momentum to p′ (see Fig. 12.95). The interaction Hamiltonian is given
by the matrix element
Z
int
H = d3 xAµ (x)hp′ |j µ (x)|pi, (12.128)

where in Dirac’s theory:

hp′ , s′3 |j µ (x)|p, s3 i = eh0|a(p′ , s′3 ) ψ̄(x)γ µ ψ(x) a† (p, s3 )|0i. (12.129)
Inserting the field expansion (7.224) in terms of creation and annihilation operators
1 h i
e−ipx u(p, s3 )ap,s3 + eipx v(p, s3 )b† (p, s3 ) ,
X
ψ(x) = q (12.130)
p,s3 V Ep /M

and using the anticommutators (7.228) and (7.229), we obtain

′
ei(p −p)x
hp′ , s′3 |j µ (x)|p, s3 i = eū(p′ , s3 )γ µ u(p, s3 ) q q . (12.131)
V Ep′ /M V Ep /M

The difference between final and initial four-momenta

q ′ ≡ p′ − p (12.132)
is the momentum transfer caused by the incoming photon.
In order to find the size of the magnetic moment we set up a constant magnetic
field in the third space direction by assuming the second component of the vector
potential to be the linear function A2 (x) = x1 B3 . Then we put the final electron to
rest, i.e., p′µ = (M, 0), and let the initial electron move slowly in the 1-direction. We
1 0 1
create an associated spinor u(p, s3 ) by applying a small Lorentz-boost e−iζ (iγ γ )/2
to the rest spinors (4.676), and expanding the matrix element (12.131) up to the
first order in p. In zeroth order, we see that
ū(0, s′3 )γ µ u(0, s3 ) = χ† (s′3 )χ(s3 )δ µ 0 , (12.133)
showing that the charge is unity. The linear term in q 1 gives rise to a 2-component:
1 1
′ 2 −iζ 1 (iγ 0 γ 1 )/2 eiq x
hp , s′3 |j 2 (x)|p, s3 i = eū(0, s3 )γ e u(0, s3 ) . (12.134)
V
12.6 Magnetic Moment of Electron 821

The two normalization factors on the right-hand side of (12.131) differ only by
second-order terms in q 1 . Now, since ū(0, s3 )γ 2 u(0, s3 ) = 0 and iγ 1 γ 2 /2 = S3 , the
spinors on the right-hand side reduce to

−ieζ 1 u† (0, s3 )S3 u(0, s3 ) = −ieζ 1 s3 . (12.135)

Momentum conservation enforces ζ 1 = −q 1 /M, so that we find

1 1
′ q 1 eiq x
hp , s′3 |j 2 (x)|p, s3 i = ie s3 . (12.136)
M V
Inserting this into the interaction Hamiltonian (12.128), we obtain
e 1 e 1
Z Z
int 3 1 iq 1 x1 1 x1
E = lim d xA2 (x)iq e s3 = lim d3 x A2 (x)∂1 eiq s3
1
q →0 M V q 1 →0 M V
e 1
Z
= − d3 x ∂1 A2 (x) s3 . (12.137)
M V
Inserting here the above vector potential A2 (x) = x1 B3 , we obtain the magnetic
interaction Hamiltonian
e
H int = −B3 s3 . (12.138)
M
Since a magnetic moment µ interacts, in general, with a magnetic field via the energy
−B, we identify the magnetic moment as being (12.126), implying a gyromagnetic
ratio g = 2.
Note that the magnetic field caused by the orbital motion of an electron leads
to a coupling of the orbital angular momentum L = x × p with a g-factor g = 1. In
order to see this relative factor 2 most clearly, consider the interaction Hamiltonian
Z
int
H =− d3 xA(x)hp′ , s′3 |j(x)|p, s3 i, (12.139)

and insert the Dirac current (12.131). For slow electrons we may neglect quantities
of second order in the momenta, so that the normalization factors E/M are unity,
and we obtain
Z
d3 x A(x)ū(p′ , s3 ) u(p, s3 )e−i(p −p)x .
′
H int = −e (12.140)

At this place we make use of the so-called Gordon decomposition formula

1 i µν
 
ū(p′ , s′3 )γ µ u(p, s3) = ū(p′ , s′3 ) (p′µ + pµ ) + σ qν u(p, s3 ), (12.141)
2M 2M
where q ≡ p′ − p is the momentum transfer. This formula follows directly from the
anticommutation rules of the γ-matrices and the Dirac equation.
An alternative decomposition is
1 ′µ i µν
 
′ µ ′
hp |j |pi = eū(p ) (p + pµ )F1 (q 2 ) + σ qν F2 (q 2 ) u(p′ ), (12.142)
2M 2M
822 12 Quantum Electrodynamics

with the form factors F1 (q 2 ), F2 (q 2 ) related to F (q 2 ), G(q 2 ) via (12.141) by

F (q 2 ) = F1 (q 2 ), G(q 2 ) = F1 (q 2 ) + F2 (q 2 ). (12.143)
Then we rewrite the interaction Hamiltonian as
e Z 3
H int = − d xA(x) ū(p′ , s3 ) (p + q − iq × S) u(p, s3 )e−iqx , (12.144)
M
where we have used the relations (4.518) and (4.515). We now replace q by the
derivatives i∂x in front of the exponential e−iqx , and perform an integration by parts
to make the derivatives act on the vector potential A(x), with the opposite sign. In
the transverse gauge, the term A(x) · q gives zero while −iA(x) · (q × S) becomes
B · S. For equal incoming and outgoing momenta, this leads to the interaction
Hamiltonian
e
Z
int
H =− d3 x [A(x) · p + B(x) · S] . (12.145)
M
We now express the vector potential in terms of the magnetic field as
1
A(x) = B × x, (12.146)
2
and rewrite (12.145) in the final form
e
H int = − B · (L + 2S) , (12.147)
2M
where L = x × p is the orbital angular momentum. The relative factor 2 discovered
by Alfred Landé in 1921 between orbital and spin angular momentum gives rise to
a characteristic splitting of atomic energy levels in an external magnetic field. If
the field is weak, both orbital and spin angular momenta will precess around the
direction of the total angular momentum. Their averages will be, for example,
J·L J·S
L̄ = J , S̄ = J . (12.148)
J2 J2
By rewriting
1 2  1 2 
J·L= J + L2 − S2 , J·S= J − L2 + S2 , (12.149)
2 2
we see that
L̄ = fLJ J, S̄ = fSJ J (12.150)
with the factors
fLJ = [J(J + 1) + L(L + 1) − S(S + 1)] /2J(J + 1), (12.151)
fSJ = [J(J + 1) − L(L + 1) + S(S + 1)] /2J(J + 1). (12.152)
Inserting this into (12.147), we obtain the interaction energies of an atomic state
|JMi:
e
H int = −gLS BM, (12.153)
2M
12.7 Decay of Atomic State 823

where

gLS = fLJ + 2fSJ = 1 + [J(J + 1) − L(L + 1) + S(S + 1)] /2J(J + 1) (12.154)

is the gyromagnetic ratio of the coupled system. This has been measured in many
experiments as Zeeman effect, if the external field is small, and as anomalous Zeeman
effect or Paschen-Back effect, if the external field strength exceeds the typical field
strength caused by the electron orbit. Then orbital and spin angular momenta
decouple and precess independently around the direction of the external magnetic
field.

12.7 Decay of Atomic State

The first important result of quantum electrodynamics is the explanation of the
decay of an atom. In quantum mechanics, this decay can only be studied by means
of the correspondence principle.
Consider an electron in an atomic state undergoing a transition from a state n
with energy En to a lower state n′ with energy En′ , whereby a photon is emitted
with a frequency ω = (En′ −En )/h̄ (see Fig. 12.4). According to the correspondence

Figure 12.4 Transition of an atomic state from a state n with energy En to a lower state
n′ with energy En′ , thereby emitting a photon with a frequency ω = (En′ − En )/h̄.

principle, this is the frequency with which the center of charge of the electronic cloud
oscillates back and forth along the direction ˆ with an amplitude:

x0 = hn′ | · x|ni = ǫhn′ |ˆ · x|ni. (12.155)

The oscillating charge emits antenna radiation. The classical theory of this process
has been recapitulated in Section 5.1, where we have given in Eq. (5.37) the radiated
power per solid angle. Its directional integral led to the Larmor formula (5.38), and
reduced to (5.38) for a harmonic oscillator.
Quantum mechanically, the antenna radiation formula (5.37) can be applied to
an atom that decays from level n to n′ , if we replace |x0 |2 by the absolute square of
the quantum mechanical matrix element (12.155):

|x0 |2 → |hn′ | · x|ni| .

2
(12.156)
824 12 Quantum Electrodynamics

Then formula (5.37) yields the radiated power per unit solid angle
dĖn′ n e2 ω 4
= 2 3 |hn′ | · x|ni| sin2 θ.
2
(12.157)
dΩ 8π c
Integrating over all dΩ gives the total radiated power, and if we divide this by the
energy per photon h̄ω, we obtain the decay rate
4 e2 ω 3 4
Γ̇ n′ n = 3
|hn′
|  · x|ni|
2
= α ω|k|2 |hn′ | · x|ni| .
2
(12.158)
3 4π h̄c 3

Let us now confirm this result by a proper calculation within quantum electro-
dynamics. Consider a single nonrelativistic electron in a hydrogen-like atom with
central charge Ze. For an electron of mass M moving in a Coulomb potential
Zα
VC (x) ≡ − , (12.159)
|x|
the Hamiltonian reads in the transverse gauge with ∇A(x) = 0,
p2 1 e2 2
H= − A(x, t) · p + A (x, t) + VC (x) + eA0 (x, t) = HC + H int . (12.160)
2M M 2M
where
p2
HC =
+ VC (x) (12.161)
2M
is the Hamiltonian of the hydrogen-like atom by itself, and H int contains the in-
teraction of the electron with the vector potential Aµ (x, t) = 0. Its magnitude
is determined by the electronic charge distribution via the Coulomb law as shown
in Eq. (12.73). The radiation field A(x, t), has an expansion in terms of photon
creation and annihilation operators given in Eq. (7.350):
1
e−ikx (k, h)ak,h + h.c. .
X h i
A(x, t) = √ (12.162)
k,h 2V ωk

Let |ni be an excited initial state of an atom with an electron orbit having the
principal quantum number n, and suppose that it decays into lower state a† (k, h)|n′ i
with a principal quantum number n′ . In addition to the electron, the lower atomic
state contains a photon with wave vector k, energy ω = ck, and helicity h. According
to Eq. (9.235), the decay probability of the initial state per unit time is given by
Fermi’s golden rule (in the remainder of this section we use physical units):
dPn′n Z d3 kV 1 ′ 2
 
= 2πh̄δ(En + h̄ω − En ′)  hn |a(k, h)T |ni  , (12.163)
dt (2π)3 
h̄ 

where T is the TR -matrix which coincides, in lowest order perturbation theory, with
the matrix Hint d3 x Hint x) [see (9.132) and (9.288) ]. The matrix element is obvi-
ously
c e
hn′ |a(k, h)H int |ni = √ hn′ |e−ikx ∗ (k, h) · p|ni. (12.164)
2V ω Mc
12.7 Decay of Atomic State 825

Performing the integral over the photon momentum (neglecting recoil) we find from
(12.163) the differential decay rate [compare (9.338)]
dΓn′ n e2 ω
= 2 2 3
|hn′ |e−ikx ∗ (k, h) · p|ni|2 . (12.165)
dΩ 8π h̄ M c
Further calculations are simplified by the observation that the wavelength of the
emitted photons is the inverse of their energy (in massless units), and thus of the
order of h̄/Zα2 Mc, about 100 times larger than the atomic diameter which is of
the order of the Bohr radius aB = h̄/ZαMc for an atom of charge Z. The expo-
nential e−ikx is therefore almost unity and can be dropped. This yields the dipole
approximation to the atomic decay rate:
dΓn′ n e2 ω
≈ 2 |∗ (k, h) · hn′ |p|ni|2. (12.166)
dΩ 8π h̄ M 2 c3
Another way of writing this result is
dΓn′ n ω ω2 ∗
≈α | (k, h) · hn′ |x|ni|2 . (12.167)
dΩ 2πh̄ c2
The momentum operator p can be replaced by M ẋ = i[Ĥ, x]/h̄, and thus, in the
matrix element hn′ |p|ni|2 , by iM(En′ − En )x/h̄ = −iMωx. Multiplying the decay
rate by the energy of the photon h̄ω to get the rate of radiated energy, the result
(12.167) coincides with the classical result (12.158).
It is customary to introduce the so-called oscillator strength for an oscillator in
the direction ǫ:
 2  2
2Mω |k| X ′
fn′ n  hn′ | · x|ni  hn | · x|ni .
X  
≈ =2 (12.168)
  
h̄  ν  λ̄e ν 

This quantity fulfills the Thomas-Reiche-Kuhn sum rule:

fn′ n = 1.
X
(12.169)
n′

For an atom with Z electrons, the right-hand side is equal to Z.

To derive this sum rule (and a bit more) we define the operator
1 −i|k|·x
Ê  ≡ e , (12.170)
|k|
whose time derivative is
˙ i h̄  −i|k|·x
Ê  = [Ĥ, Ê  ] =  · ∇ +  · ∇e−i|k|·x .

e (12.171)
h̄ 2M
According to the canonical commutator [p̂i , x̂j ] = −ih̄δij , the Hermitian-conjugate
˙
of Ê  commutes with Ê  like
˙ h̄
[Ê  † , Ê  ] = −i . (12.172)
M
826 12 Quantum Electrodynamics

Taking this commutator between states hn| and |ni, and inserting a completeness
relation n′ |n′ ihn′ | = 1 in the middle, we may go to small k, to find indeed the sum
P

rule (12.169).
Let us calculate the angular properties of the radiation in more detail. The
decomposition of the hydrogen wave functions into radial and angular parts is
hx|nlmi = Rnl (r)Ylm (θ, ϕ). (12.173)
Then the matrix elements of x factorize:
hn′ l′ m′ |x|nlmi = rn′ l′ ;nl × hl′ m′ |x̂|lmi. (12.174)
The matrix elements of r,
Z ∞
rn′ l′ ;nl ≡ drr 2Rn′ l′ (r)rRnl (r), (12.175)
0

have been calculated by Gordon [25]:

v
′ ′
(−1)n −l u ′ ′ n+n −2l−2
t (n + l)!(n + l − 1)! (4nn′ )l+1 (n − n )
u
rn′ l′ ;nl = (12.176)
4(2l − 1)! (n′ − l)!(n − l − 1)! (n + n′ )n+n′
 !2 !
′ ′ ′
!
 4nn n−n 4nn 
× F −n′r , 2l, − − F −nr − 2, −n′r , 2l, − ,
 (n − n′ )2 n + n′ (n − n′ )2 

with F (a, b, c; z) being hypergeometric functions. The angular matrix elements of

the unit vector in (12.174),
Z
′ ′
hl m |x̂|lmi ≡ dx̂ Yl∗′ m′ (θ, ϕ)x̂Ylm (θ, ϕ), (12.177)

are easily calculated since x̂3 = cos θ, and the spherical harmonics satisfy the recur-
sion relation
v v
(l+1)2 − m2 l2 − m2
u u
u u
cos θ Ylm (θ, ϕ) = t Yl+1 m (θ, ϕ)+ t Yl−1 m (θ, ϕ). (12.178)
(2l+2)(2l+1) (2l+1)(2l−1)

On account of the orthonormality relation [recall (4F.3)]

Z
dx̂ Yl∗′m′ (θ, ϕ)Ylm (θ, ϕ) = δl′ l δm′ m , (12.179)

we obtain immediately the angular matrix elements of x̂3 :

v v
(l+1)2 −m2 l2 −m2
u u
, hl−1 m′ |x̂3 |lmi = t
u u
hl+1 m|x̂3 |lmi = t , (12.180)
(2l+2)(2l+1) (2l+1)(2l−1)

with all others vanishing. The matrix elements of x̂1 and x̂2 are found with the help
of the commutation rule
[L̂i , x̂j ] = iǫijk x̂k , (12.181)
12.8 Rutherford Scattering 827

which states that x̂i is a vector operator. As a consequence, the matrix elements
satisfy the Wigner-Eckart theorem,

hl′ m′ ± 1|x̂M |lmi = hl′ m′ |1M; lmix̂l′ l , (12.182)

′′
where hl′ m′ |1m′′ ; lmi are Clebsch-Gordan coefficients (see Appendix 4E) and x̂m
are the spherical components of x̂ [recall the definition in Eq. (4.893)]:
√ √
x̂3 = cos θ, x̂± = ∓(x̂2 ± x̂2 )/ 2 = sin θe±iϕ / 2. (12.183)

Explicitly:
v
u (l ± m + 2)(l ± m + 1)
u
hl + 1 m ± 1|x̂± |lmi = t ,
2(2l + 3)(2l + 1)
v
u (l ∓ m)(l ∓ m − 1)
u
hl − 1 m ± 1|x̂± |lmi = t .
2(2l + 1)(2l − 1)
(12.184)

12.8 Rutherford Scattering

The scattering of electrons on the Coulomb potential of nuclei of charge Ze,

ZE 2 Zα
VC (r) = − =− , (12.185)
4πr r
was the first atomic collision observed experimentally by Rutherford.
The associated scattering cross section can easily be calculated in an estimated
classical approximation.

12.8.1 Classical Cross Section

In a Coulomb potential the electronic orbits are hyperbola. If an incoming elec-
tron runs along the z-direction and is deflected by a scattering angle θ towards the
x direction (see Fig. 12.5), the nucleus has the coordinates

(xF , zF ) = (b, −a), (12.186)

where
a θ b θ
= tan , = cos . (12.187)
b 2 d 2
The parameter d is equal to aǫ, where ǫ > 1 is the excentricity of the hyperbola.
The distance of closest approach to the nucleus is

rc = d − a. (12.188)
828 12 Quantum Electrodynamics

Figure 12.5 Kinematics of Rutherford scattering.

It is determined by the conservation of the nonrelativistic energy:

p2 p′2 p2 Zα
E nr = = = c − , (12.189)
2M 2M 2M rc
where pc is the momentum at the point of closest approach. This momentum is
determined by the conservation of orbital angular momentum:
l = pb = p′ b = pc rc . (12.190)
From these equations we find
Zα
b= . (12.191)
2E tan(θ/2)
The quantity b is called the impact parameter of the scattering process. It is the
closest distance which an particle would have from the nucleus if it were not deflected
at all. Particles which would pass through a thin annular ring with the radii b and
b + db are, in fact, scattered into a solid angle dΩ given by
1 Zα 1
db = − 3 dΩ. (12.192)
4π 4E sin (θ/2) cos(θ/2)
The current density of a single randomly incoming electron is j = v/V . It would
pass through the annular ring, with a probability per unit time
dṖ = j 2πbdb. (12.193)
With this probability it winds up in the solid angle dΩ. Inserting Eq. (12.192), we
find the differential cross section [recall the definition Eq. (9.242)]
2
dσ dṖ db 1 Zα

= = 2πb = 4 . (12.194)
dΩ dΩ dΩ 4 sin (θ/2) 2E
12.8 Rutherford Scattering 829

12.8.2 Quantum-Mechanical Born Approximation

Somewhat surprisingly, the same result is obtained in quantum mechanics within the
Born approximation. According to Eqs. (9.147) and (9.248), the differential cross
section is
dσ M 2V 2
≈ |Vp′ p |2 , (12.195)
dΩ (2π)2
where
1 Zα 4π Zα
Z
′
Vp′ p = − d3 x ei(p −p)x =− . (12.196)
V r V |q2 |2
The quantity

q2 ≡ |p′ − p|2 = 2p2 (1 − cos θ) = 4p2 sin2 (θ/2) = 8EM sin2 (θ/2) (12.197)

is the momentum transfer of the process. Inserting this into (12.196), the differential
cross section (12.195) coincides indeed with the classical expression (12.194).

12.8.3 Relativistic Born Approximation: Mott Formula

Let us now see how the above cross section formula is modified in a relativistic
calculation involving Dirac electrons. The scattering amplitude is, according to
Eq. (10.142), Z
Sfi = −iehp′ , s′3 | d4 x ψ̄(x)γ µ ψ(x)|p, s3 iAµ (x), (12.198)
where Aµ (x) has only the time-like component
Ze Z
d3 q iqx Ze
A0 (x) = − =− e . (12.199)
4πr (2π)3 |q|2
The time-ordering operator has been dropped in (12.198) since there are no operators
at different times to be ordered in first-order perturbation theory. By evaluating the
matrix element of the current in (12.198), and performing the spacetime integral we
obtain s
M2 Ze2
Sfi = i2πδ(E ′ − E) ū(p′ ′
, s 3 )γ 0
u(p, s 3 ) , (12.200)
V 2E ′E |q|2
where E and E ′ are the initial and final energies of the electron, which are in fact
equal in this elastic scattering process.
Comparing this with (9.130) we identify the T -matrix elements
M Ze2
Tp′ p = ū(p′ , s′3 )γ 0 u(p, s3 ) 2 , (12.201)
VE |q|
In the nonrelativistic limit this is equal to
1 Ze2
Tp′ p = − . (12.202)
V |q|2
830 12 Quantum Electrodynamics

Its relativistic extension contains a correction factor

M
C= ū(p′ , s3 )γ 0 u(p, s3 ). (12.203)
E
Its absolute square multiplies the nonrelativistic differential cross section (12.194).
Apart from this, the relativistic cross section contains an extra kinematic factor
E 2 /M 2 accounting for the different phase space of a relativistic electron with respect
to the nonrelativistic one [the ratio between (9.247) and (9.244)]. The differential
cross section (12.194) receives therefore a total relativistic correction factor

E2
|C|2 = |ū(p′ , s′3 )γ 0 u(p, s3 )|2 . (12.204)
M2
If we consider the scattering of unpolarized electrons and do not observe the final
spin polarizations, this factor has to be summed over s′3 and averaged over s3 , and
the correction factor is
E2 1 X
2
|C|2 = ū(p′ , s′3 )γ 0 u(p, s3)ū(p, s3 )γ 0 ū(p′ , s′3 ). (12.205)
M 2 s′ ,s3
3

To write the absolute square in this form we have used the general identity in the
spinor space, valid for any 4 × 4 spinor matrix M:

(ū′ Mu)∗ = ūM̄ u′ , (12.206)

where the operation bar is defined for a spinor matrix in complete analogy to the
corresponding operation for a spinor:

M̄ ≡ γ 0 M † γ 0−1 . (12.207)

The Dirac matrices themselves satisfy

γ̄ µ = γ µ . (12.208)

The “bar” operation has the typical property of an “adjoining” operation. If it is

applied to a product of matrices, the order is reversed:

M1 · · · Mn = M n · · · M 1 . (12.209)

We use now the semi-completeness relation (4.702) for the u-spinors and rewrite
(12.205) as
E2
!
2 1 0p
/ + M 0 p/ + M
|C| = tr γ γ . (12.210)
M2 2 2M 2M
The trace over product of gamma matrices occurring in this expression is typical
for quantum electrodynamic calculations. Its evaluation is somewhat tedious, but
follows a few quite simple algebraic rules.
12.8 Rutherford Scattering 831

The first rule states that a trace containing an odd number of gamma matrices
vanishes. This is a simple consequence of the fact that γ µ and any product of an
odd number of gamma matrices change sign under the similarity transformation
γ5 γ µ γ5 = −γ µ , while the trace is invariant under any similarity transformation.
The second rule governs the evaluation of a trace containing an even number of
gamma matrices. It is a recursive rule which makes essential use of the invariance
of the trace under cyclic permutations
tr(γ µ1 γ µ2 γ µ3 · · · γ µn−1 γ µn ) = tr(γ µ2 γ µ3 · · · γ µn−1 γ µn γ µ1 ). (12.211)
This leads to an explicit formula that is a close analog of Wick’s expansion formula
for time-ordered products of fermion field operators. To find this, we define a pair
contraction between a/ and b/ as
1
a
/b/ ≡ tr(/
a b/ ) = ab. (12.212)
4
Then we consider a more general trace
1
tr(/
a 1 · · · a/ n ) (12.213)
4
and move the first gamma matrix step by step to the end, using the anticommutation
rules between gamma matrices (4.566), which imply that

a/1a/ i = −/ ai a/1 + 2 a
ai a/1 + 2a1 ai = −/ /1 a
/i . (12.214)
Having arrived at the end, it can be taken back to the front, using the cyclic invari-
ance of the trace. This produces once more the initial trace, except for a minus sign,
thus doubling the initial trace on the left-hand side of the equation if n is even. In
this way, we find the recursion relation
1 1 1
tr(/
a1 a/2 a/3 · · · a/n−1 a/n ) = tr(/a1a /2 ) + tr(/a1 a /2 a
/3 · · · a/n−1 a/n )
4 4 4
1 1
+ . . . + tr(/a1a /3 · · ·/an−1 a/n ) + . . . + tr(/a1 a
/2 a /2 a/3 · · · a
/n−1a
/n ).(12.215)
4 4
The contractions within the traces are defined as in (12.212), but with a minus sign
for each permutation necessary to bring the Dirac matrices to adjacent positions.
Performing these operations, the result of (12.215) is
1 1 1
tr(/
a1 a/2 a/3 · · · a/n−1 a/n )=(a1 a2 ) tr(/ a3 a/4 · · · a/n−1 a/n ) + (a1 a3 ) tr(/ a2 a/4 · · · a/n−1 a/n )
4 4 4
1 1
+ . . . + (a1 an−1 ) tr(/ a2 a/3 · · · a/n−2 a/n ) + (a1 an ) tr(/ a2 a/3 · · · a/n−1 a/n ).(12.216)
4 4
By applying this formula iteratively, we arrive at the expansion rule of the Wick
type:
1
(−)P (ap(1) ap(2) )(ap(3) ap(4) ) · · · (ap(n−1) ap(n) ), (12.217)
X
tr(/
a1 · · · a/n ) =
4 pair contractions
832 12 Quantum Electrodynamics

where P is the number of permutations to adjacent positions.

The derivation of the rule is completely parallel to that of the thermodynamic
version of Wick’s rule in Section 4.14, whose Eqs. (7.857) and (7.858) can directly
be translated into anticommutation rules between gamma matrices and the cyclic
invariance of their traces, respectively.
Another set of useful rules following from (12.214) and needed later is
γ µ a/ γ ν = a/ + 2aµ γ ν , (12.218)
γ µ γµ = 4, (12.219)
γ µ a/ γµ = −2/a, (12.220)
γ µ a/ b/ γµ = −2/a, (12.221)
γ µ a/ b/ b/ γµ = −2/c b/ a/ . (12.222)
Following the Wick rule (12.253) for γ-matrices, we now calculate the expression
(12.210) as
E2 1 1
 
2
|C|2 = 2
tr(γ 0 p/ γ 0 p/ ′ ) + M 2 tr(γ 02 )
M 2M 4
1  0 ′0 ′ 2

= 2p p − pp + M . (12.223)
2M 2
Inserting p0 = p′0 = E and, in the center-of-mass frame,
θ
pp′ = E 2 − |p|2 cos θ = M 2 + 2|p|2 sin2 , (12.224)
2
the total cross section in the center-of-mass frame becomes
dσ Z 2 α2 M 2 E2
= 4 4 × 2
|C|2 . (12.225)
dΩCM 4|p| sin (θ/2) M
This relativistic version of Rutherford’s formula is known as Mott’s formula.
In terms of the incident electron velocity, the total modification factor reads
E2
"  2 #
1 v θ
2
|C|2 = 2
1− sin2 . (12.226)
M 1 − (v/c) c 2
It is easy to verify that the same differential cross section is valid for positrons. In
the nonrelativistic case, this follows directly from the invariance of (12.195) under
e → −e. But also the relativistic correction factor remains the same under the
interchange of electrons and positrons, where (12.205) becomes
E2 2 1 X
|C| = v̄(p′ , s′3 )γ 0 v(p, s3 )v̄(p, s3 )γ 0 v̄(p′ , s′3 ). (12.227)
M2 2 s′ ,s3
3

Inserting the semi-completeness relation (4.703) for the spinors v(p, s3 ), this becomes
E2 / − M 0 p/ ′ − M
!
2 1 0p
|C| = tr γ γ , (12.228)
M2 2 2M 2M
which is the same as (12.205), since only traces of an even number of gamma matrices
contribute.
12.9 Compton Scattering 833

12.9 Compton Scattering

A simple scattering process, whose cross section can be calculated to a good accu-
racy by means of the above diagrammatic rules, is photon-electron scattering, also
referred to as Compton scattering. It gives an important contribution to the blue
color of the sky.1
Consider now a beam of photons with four-momentum ki and polarization λi
impinging upon an electron target of four-momentum pi and spin orientation σi .
The two particles leave the scattering regime with four-momenta kf and pf , and spin
indices λf , σf , respectively. Adapting formula (10.103) for the scattering amplitude
to this situation we have

Sfi = S (p′ , s′3 ; k′ , h′ |k, h; p, s3 )

R∞
−i dtVI (t) †
h0|a(p′ , s′3 )a(k′ , h′ )T̂ e −∞ a (k, h)a† (p, s3 )|0i
≡ R∞
−i dtV (t)
h0|e −∞ I |0i
≡ SN (p′ , s′3 ; k′ , h′ |k, h; p, s3 ) /Z[0], (12.229)

with R∞ R
−i dt VI (t) d4 x ψ̄(x)γ µ ψ(x)Aµ (x)
e −∞ = e−ie . (12.230)

Expanding the exponential in powers of e, we see that the lowest-order contribution

to the scattering amplitude comes from the second-order term which gives rise to the
two Feynman diagrams shown in Fig. 12.6. In the first, the electron s1 of momentum
p absorbs a photon of momentum k, and emits a second photon of momentum k ′ ,
to arrive in the final state of momentum p′ . In the second diagram, the acts of
emission and absorption have the reversed order. Before we calculate the scattering
cross section associated with these Feynman diagrams, let us first recall the classical
result.

Figure 12.6 Lowest-order Feynman diagrams contributing to Compton Scattering and

giving rise to the Klein-Nishina formula.

1
The blue color is usually attributed to Rayleigh scattering. This arises from generalizing Thom-
son’s formula (12.233) for the scattering of light of wavelength λ on electrons to that on droplets
of diameter d with refractive index n. That yields σRayleigh = 2π 5 d6 [(n2 − 1)/(n2 + 2)]2 /3λ4 .
834 12 Quantum Electrodynamics

12.9.1 Classical Result

Classically, the above process is described as follows. A target electron is shaken
by an incoming electromagnetic field. The acceleration of the electron causes an
emission of antenna radiation. For a weak and slowly oscillating electromagnetic
field of amplitude, the electron is shaken nonrelativistically and moves with an
instantaneous acceleration
e e
ẍ = E=  E0 e−iωt+ik·x , (12.231)
M M
where ω is the frequency and E0 the amplitude of the incoming field.
The acceleration of the charge gives rise to antenna radiation following Larmor’s
formula. Inserting (12.231) into (5.37), and averaging over the temporal oscillations,
the radiated power per unit solid angle is
!2
dĖ 1 e2
= E02 sin2 β, (12.232)
dΩ 2 4πM
where β is the angle between the direction of polarization of the incident light and
the direction of the emitted light.
For a later comparison with quantum electrodynamic calculations we associate
this emitted power with a differential cross section of the electron with respect to
light. According to the definition in Chapter 6, a cross section is obtained by dividing
the radiated power per unit solid angle by the incident power flux density cE20 /2.
This yields !2
dσ e2
= sin2 β = re2 sin2 β, (12.233)
dΩ 4πMc2
where
e2 h̄α
re = 2
= ≈ 2.82 × 10−13 cm (12.234)
4πMc Mc
is the classical electron radius. Formula (12.233) describes the so-called Thomson
scattering cross section. It is applicable to linearly polarized incident waves. For
unpolarized waves, we have to form the average between the cross section (12.233)
and another one in which the plane of polarization is rotated by 90%. Suppose
the incident light runs along the z-axis, and the emitted light along the direction
k̂ = (sin θ cos φ, sin θ sin φ, cos θ). For a polarization direction  = ẍ
ˆ along the x-axis,
the angle β is found from

sin2 β = (k̂ × )2 = cos2 θ + sin2 θ sin2 φ. (12.235)

For a polarization direction  = ẍ

ˆ along the y-axis, it is

sin2 β = (k̂ × )2 = cos2 θ + sin2 θ cos2 φ. (12.236)

The average is
1 2 1
sin2 β = cos2 θ + sin θ = (1 + cos2 θ). (12.237)
2 2
12.9 Compton Scattering 835

Integrating this over all solid angles yields the Thomson cross section for unpolarized
light
8π 2
σtot = r . (12.238)
3 e

12.9.2 Klein-Nishina Formula

The scattering amplitude corresponding to the two Feynman diagrams in Fig. 12.6
is obtained by expanding formula (10.103) up to second order in e:
Z
Sfi = −e2 ǫ ′ ψ(y)ψ̄(x)/
d4 xd4 yh0|ap′ ,s′3 T̂ ψ̄(y)/ ǫ ψ(x)a†p,s3 |0i, (12.239)

and by performing the Wick contractions of Section 7.8:

ǫ ′ ψ(y)ψ̄(x)/
h0|ap′ ,s′3 T̂ ψ̄(y)/ ǫ ψ(x)a†p,s3 |0i = h0|T̂ ap′ ,s′3 ψ̄(y) ǫ/′ ψ(y)ψ̄(x) ǫ/ ψ(x)a†p,s3 |0i

+h0|T̂ ap′ ,s′3 ψ̄(y)/ǫ′ ψ(y)ψ̄(x)/ǫψ(x)a†p,s3 |0i|0i. (12.240)

After Fourier-expanding the intermediate electron propagator,
d4 pi −ipi (y−x) i
Z
G0 (y, x) = 4
e i
, (12.241)
(2π) p/ − M
we find
Z
′ d 4 pi
Z
M
Sfi = −e2 d4 xd4 yek y−kx 4
√
(2π) V 2 EE ′ 2ω2ω ′
"
′ i i i
× ei(p −p )y−(p−p )x ū(p′ , s′3 )/
ǫ ′∗ i ǫ/u(p, s3 )
p/ − M
#
′ i i i
+ ei(p −p )x−(p−p )y ū(p′ , s′3 )/
ǫ i ǫ/′ u(p, s3 ) . (12.242)
p/ − M
One of the spatial integrals fixes the intermediate momentum in accordance with
energy-momentum conservation, the other yields a δ (4) -function for overall energy-
momentum conservation. The result is
M
Sfi = −i(2π)4 δ (4) (p′ + k ′ − p − k)e2 √ ū(p′ , s′3 )Hu(p, s3), (12.243)
2 ′
V EE 2ω2ω ′

where H is the 4 × 4-matrix in spinor space

(/
p + k/ ) + M p − k/ ′ ) + M ′∗
(/
H ≡ ǫ/′∗ ǫ
/ + ǫ
/ ǫ/ . (12.244)
(p + k)2 − M 2 (p − k ′ )2 − M 2
We have written ǫ(k, h), ωk as ǫ, ω, and ǫ(k′ , h′ ), ωk′ as ǫ′ , ω ′ , respectively, with a
similar simplification for E and E ′ . The second term of the matrix H arises from
the first by the crossing symmetry
ǫ ↔ ǫ′ , k ↔ −k ′ . (12.245)
836 12 Quantum Electrodynamics

Simplifications arise from the properties (12.246). It can, moreover, be simplified

by recalling that external electrons and photons are on their mass shell, so that

p2 = p′2 = M 2 , k 2 = k ′2 = 0, (12.246)

k ǫ = k ′ ǫ′ = 0. (12.247)
A further simplification arises by working in the laboratory frame in which the initial
electron is at rest, p = (M, 0, 0, 0). Also, we choose a gauge in which the polarization
vectors have only spatial components. Then

p ǫ = p ǫ′ = 0, (12.248)

since p has only a temporal component and ǫ only space components. We also
use the fact that H stands between spinors which satisfy the Dirac equation (/ p −
M)u(p, s3 ) = 0, ū(p, s3 )(/
p −M) = 0. Further we use the commutation rules (4.566)
for the gamma matrices to write [as in (12.214)]

ǫ/p/ = − p/ ǫ/ + 2pǫ. (12.249)

The second term vanishes by virtue of Eq. (12.248). Similarly, we see that p/ anti-
commutes with ǫ/′ . Using these results, we may eliminate the terms p/ + M occurring
in M. Finally, using Eq. (12.247), we obtain
k/ ′ ′∗
( )
k/
ū(p′ , s′3 )Hu(p, s3) = ū(p′ , s′3 ) ǫ′∗ ǫ/ + ǫ/ ǫ/ u(p, s3). (12.250)
2pk 2pk ′
To obtain the transition probability, we must take the absolute square of this. If
we do not observe initial and final spins, we may average over the initial spin and
sum over the final spin directions. This produces a factor 1/2 times the sum over
both spin directions, which is equal to

|ū(p′ , s′3 )Hu(p, s3 )|2 = ū(p′ , s′3 )Hu(p, s3)ū(p, s3 )Hu(p′, s′3 ). (12.251)
X X
F =
s′3 ,s3 s′3 ,s3

Here we apply the semi-completeness relation (4.702) for the spinors to find
1
F = p ′ + M)H(/
tr [(/ p + M)H] . (12.252)
4M 2
The trace over the product of gamma matrices can be evaluated according to the
Wick-type of rules explained on p. 830.
1
(−)P a/ 1 · · · a/ n .
X
tr(/
a 1 · · · a/ n ) = (12.253)
4 pair contractions

After some lengthy algebra (see Appendix 9A), we find

ω′
" #
1 ω
F = 2
+ ′ − 2 + 4|′∗ · |2 . (12.254)
2M ω ω
12.9 Compton Scattering 837

Figure 12.7 Illustration of the photon polarization sum h,h′ |′∗ |2 = 1 + cos2 θ in
P

Compton scattering in the laboratory frame. Incoming and outgoing photon momenta
with scattering angle θ are shown in the scattering plane, together with their transverse
polarization vectors.

This holds for specific polarizations of the incoming and outgoing photons. If we sum
over all final polarizations and average over all initial ones, we find (see Fig. 12.7)
1 X ′∗ 2 1
| | = (1 + cos2 θ), (12.255)
2 h,h′ 2

This can also be found more formally using the transverse completeness relation
(4.334) of the polarization vectors:
1 X ′∗ 2 1 X ′i∗ ′ ′ i
| | = ǫ (k , h )ǫ (k, h)ǫ∗j (k, h)ǫ′i (k′ , h′ ) (12.256)
2 h,h′ 2 h,h′
1 ij k i k j k ′j k ′i (k · k′ )2
! ! " #
ji 1 1
= δ − 2 δ − ′2 = 1 + 2 ′2 = (1 + cos2 θ).
2 k k 2 kk 2

The average value of F is therefore

ω′ 1 ω′
" # " #
1X 1 ω ω
F̄ = 2
+ ′
− 2 + 4| ′∗
· | 2
= 2
+ ′ − sin2 θ . (12.257)
2 h,h′ 2M ω ω M ω ω

We are now ready to calculate the transition rate, for which we obtain from
Eq. (9.298):
Z 3 ′ Z 3 ′
dP d pV d k V (4) ′
= V (2π)4 3
δ (p − p)|tfi |2 , (12.258)
dt (2π) (2π)3
with the squared t-matrix elements

1 M2 1
|tfi |2 = e4 4 ′ ′
F. (12.259)
V EE 2ω2ω 2
The spatial part of the δ-function removes the momentum integral over p′ . The
temporal part of the δ-function enforces energy conservation. This is incorporated
into the momentum integral over k′ as follows. We set Ef = p′0 + k0′ , and write

dω ′
d3 k ′ = k ′2 dk ′ dΩ = ω ′2 dΩdEf , (12.260)
dEf
838 12 Quantum Electrodynamics

where Ω is the solid angle into which the photon has been scattered. Then (12.258)
becomes
V2 ω′ dω ′  1 M2 1
! 
dP
= V e4 (2π)4 F. (12.261)
dtdΩ (2π)6 ω dEf Ef =Ei 4V 4 EE ′ 2


For an explicit derivative dω ′/dEf , we go to the laboratory frame and express the
final energy as
q q √
Ef = ω ′ + p′2 + M 2 = ω ′ + (k − p′ )2 + M 2 = ω ′ + ω 2 − 2ωω ′ cos θ + ω ′2 + M 2 ,
(12.262)
where θ is the scattering angle in the laboratory. This yields the derivative
dEf ω ′ − ω cos θ
= 1 + . (12.263)
dω ′ E′
By equating Ei = M + ω with Ef , we derive the Compton relation

ωω ′(1 − cos θ) = M(ω − ω ′ ), (12.264)

or
ω
ω′ = , (12.265)
1 + ω(1 − cos θ)/M
and therefore
dEf ω ′ − ω cos θ E ′ + ω ′ − ω cos θ M + ω − ω cos θ M ω
′
= 1 + ′
= ′
= ′
= ′ ′ . (12.266)
dω E E E E ω
Since E = M in the laboratory frame, Eq. (12.261) yields the differential probability
rate
!2
dP V2 ω′ 1 1
= V e4 (2π)4 V F. (12.267)
dtdΩ (2π)6 ω 4V 4 2

To find the differential cross section, this has to be divided by the incoming
particle current density. According to Eq. (9.315), this is given by
v
j= , (12.268)
V
where v is the velocity of the incoming particles. The incoming photons move with
light velocity, so that (in natural units with c = 1)
1
j= . (12.269)
V
This leaves us with the Klein-Nishina formula for the differential cross section
!2
dσ ω′ 1
= α2 F. (12.270)
dΩ ω 2
12.10 Electron-Positron Annihilation 839

In the nonrelativistic limit where ω ≪ M, the Compton relation (12.265) shows

that ω ′ ≈ ω, and Eq. (12.257) reduces to
1
F → (1 + cos2 θ). (12.271)
M2
Expressed in terms of the classical electron radius r0 = α/M, the differential scat-
tering cross section becomes, expressed in terms of the classical electron radius
r0 ≡ α/M,
dσ 1
≈ r0 2 (1 + cos2 θ). (12.272)
dΩ 2
This is the Thomson formula for the scattering of low energy radiation by a static
charge. To find the total Thomson cross section, we must integrate (12.272) over all
solid angles and obtain:
dσ 1 1
Z Z
σ= dΩ ≈ 2π d cos θ r02 (1 + cos2 θ). (12.273)
dΩ −1 2
In the low-energy limit we identify
8π
σThomson ≡ r0 2 . (12.274)
3
Let us also calculate the total cross section for relativistic scattering, integrating
(12.270) over all solid angles:
!2 "
α2 ω′ ω′
#
dσ 1 ω
Z Z
σ= dΩ = 2π d cos θ + ′ − sin2 θ . (12.275)
dΩ −1 2M 2 ω ω ω
Inserting
ω
ω′ = , (12.276)
1 + ω(1 − cos θ)/M
which follows from (12.265), the angular integral yields

σ = σThomson f (ω), (12.277)

where f (ω) contains the relativistic corrections to Thomson’s cross section:

( )
3 2ω[ω(ω + 1)(ω + 8) + 2]
f (ω) = − (2+2ω−ω 2) log(1 + 2ω) (12.278)
8ω 3 (2ω + 1)2
For small ω, f (ω) starts out like (see Fig. 12.8a):

f (ω) = 1 − 2ω + O(ω 2 ). (12.279)

For large ω, ωf (ω) increases like (see Fig. 12.8b):

3
ωf (ω) = log 2ω + O(1/ log ω). (12.280)
8ω
840 12 Quantum Electrodynamics

1.0 3.0

0.9 2.5

0.8 2.0

0.7 f (ω) 1.5 ωf (ω)

0.6 1.0

0.5 0.5

a) 0.0 0.2 0.4 0.6 0.8 1.0 b) -6 -4 -2 0 2 4 6

ω log(ω)

Figure 12.8 Ratio between total relativistic Compton cross section and nonrelativistic
Thomson cross section.

Figure 12.9 Lowest-order Feynman diagrams contributing to electron-positron annihi-

lation. It arises from the Compton diagram by the crossing operation p′ → −p′ , k → −k,
ǫ → ǫ∗ .

12.10 Electron-Positron Annihilation

The Feynman diagrams in Fig. 12.6 can also be read from bottom to top in which case
they describe annihilation processes. In order to see this, we use as much as possible
previous results and reinterpret the Feynman diagrams as shown in Fig. 12.9. Instead
of an outgoing electron with momentum p′ , we let a negative-energy electron go out
with momentum −p′ . Instead of an incoming photon with momentum k we let a
negative-energy photon go out with momentum −k. The former are represented
by spinors v̄(p′ , s3 ), which are negative energy solutions of the Dirac equation with
inverted momenta and spin directions. The S-matrix element to lowest order is
found from a slightly modified (12.239):
Z
Sfi = −e2 d4 xd4 yh0|T̂ ψ̄(y)/ ǫ ψ(x)b†p′ ,s′ ap,s3 |0i.
ǫ ′ ψ(y)ψ̄(x)/ (12.281)
3

Performing the Wick contractions as in (12.240) we obtain

d 4 pi M
Z Z
2 4 4 k ′ y+kx
Sfi = −e d xd ye 4
√
(2π) V EE ′ 2ω2ω ′
2
"
′ i i i
× ei(−p −p )y−(p−p )x v̄(p′ , s′3 )/
ǫ ′∗ i ǫ/∗ u(p, s3)
p/ − M
12.10 Electron-Positron Annihilation 841
#
i(−p′ −pi )x−(p−pi )y ′ i
+e v̄(p , s′3 )/
ǫ∗ i
ǫ/′ u(p, s3 ) . (12.282)
p/ − M

Note that this arises from the Compton expression (12.242) by the crossing opera-
tion

p′ → −p′ , ū(p, s3 ) → v̄(p, s3 ), k → −k, ǫ(k, h) → ǫ∗ (k, h). (12.283)

As before, one of the spatial integrals fixes the intermediate momentum in accor-
dance with energy-momentum conservation, the other yields a δ (4) -function for over-
all energy-momentum conservation. The result is

M
Sfi = −i(2π)4 δ (4) (k + k ′ − p − p′ )e2 √ v̄(p′ , s′3 )Hu(p, s3 ), (12.284)
V 2 EE ′ 2ω2ω ′

where H is the 4 × 4-matrix in spinor space

(/
p − k/ ) + M ∗ p − k/ ′ ) + M ′∗
∗ (/
H ≡ ǫ/′∗ ǫ
/ + ǫ
/ ǫ/ . (12.285)
(p − k)2 − M 2 (p − k ′ )2 − M 2

As before, we have written ǫ(k, h), ωk as ǫ, ω, and ǫ(k′ , h′ ), ωk′ as ǫ′ , ω ′ , respectively,

with a similar simplification for E and E ′ . The second term of the matrix H arises
from the first by the Bose symmetry

ǫ ↔ ǫ′ , k ↔ k′ . (12.286)

Simplifications arise from the mass shell properties (12.246), the gauge conditions
(12.247), and the other relations (12.248), (12.249). We also work again in the
laboratory frame in which the initial electron is at rest,
√ p = (M, 0, 0, 0) and the
positron comes in with momentum p′ and energy E ′ = p′2 + M 2 . This leads to

k/ ′ ′∗
( )
′ k/ ∗
v̄(p , s′3 )Hu(p, s3) = v̄(p ′
, s′3 ) ǫ ′∗
ǫ/ + ǫ/∗ ǫ/ u(p, s3 ). (12.287)
2pk 2pk ′

To obtain the transition probability, we must take the absolute square of this.
If we do not observe initial spins, we may average over the initial spin components
which gives a factor 1/4 times the spin sum

|v̄(p′ , s′3 )Hu(p, s3 )|2 = v̄(p′ , s′3 )Hu(p, s3)ū(p, s3 )Hu(p′, s′3 ). (12.288)
X X
F =
s′3 ,s3 s′3 ,s3

Now we use the semi-completeness relations (4.702) and (4.703) for the spinors to
rewrite (12.288) as

1
F = − p ′ + M)H(/
tr [(−/ p + M)H] . (12.289)
4M 2
842 12 Quantum Electrodynamics

The trace over a product of gamma matrices is now evaluated as before and we
obtain, for specific polarizations of the two outgoing photons, almost the same ex-
pression as before in Eq. (12.257) (see Appendix 12A):
ω′
" #
1 ω
F = 2
+ ′ + 2 − 4|′∗ · ∗ |2 . (12.290)
2M ω ω
This result can be deduced directly from the previous polarization sum (12.254) by
the crossing operation (12.283), apart from an overall minus sign, whose origin is
the negative sign in front of (12.289).
We are now ready to calculate the transition rate, for which we obtain from
Eq. (9.298)
dP d3 k ′ V d3 k ′ V (4) ′
Z Z
= V (2π)4 δ (p − p)|tfi |2 , (12.291)
dt (2π)3 (2π)3
with the squared t-matrix elements

2 1 M24 1
|tfi | = e 4 ′ ′
F. (12.292)
V EE 2ω2ω 4
The spatial part of the δ-function removes the momentum integral over k′ . The
temporal part of the δ-function enforces energy conservation. This is incorporated
into the momentum integral over k as follows. We set Ef = ω + ω ′ , and write
dω
d3 k = k 2 dkdΩ = ω 2 dΩdEf , (12.293)
dEf
where Ω is the solid angle into which the photon with momentum k emerges. Then
(12.291) becomes
V2 1 M2 1

dP ω dω 
 
= V e4 (2π)4 F. (12.294)
dtdΩ (2π)6 ω′ dEf Ef =Ei 4V 4 EE ′ 4


To calculate dω/dEf explicitly, we express the final energy in the laboratory frame
as
√ q q
Ef = ω + k′2 = ω + (p′ − k)2 = ω + p′2 − 2ω|p′| cos θ + ω 2 , (12.295)
where θ is the scattering angle in the laboratory. Hence:
dEf 1 d  ′2  1  ′ 
=1+ ′ p − 2ω|p′| cos θ + ω 2 = ωω − 2ω|p′
| cos θ + ω 2
dω 2ω dω ωω ′
1 k(p + p′ ) M(M + E ′ )
  q  
= ω M + p′2 +M 2 −2ω|p′ | cos θ = = . (12.296)
ωω ′ ωω ′ ωω ′
We must now divide (12.294) by the incoming positron current density j = |p′ |/E ′ V
[recall (9.315)], and find the differential cross section
α2
2
dσ ω 1 2

= ′ M F, (12.297)
dΩ |p |(M + E ′ ) M 4
12.10 Electron-Positron Annihilation 843

h,h′ | | in electron-
Figure 12.10 Illustration of the photon polarization sum
P ′∗ 2

positron annihilation in the laboratory frame. Incoming positron and outgoing photon
momenta with scattering angles θ and θ ′ are shown in the scattering plane, together with
their transverse polarization vectors.

where
M M + E ′ − |p′ | cos θ
= , (12.298)
ω M + E′
as follows from equating the right-hand sides of
1 1
kk ′ = (k + k ′ )2 = (p + p′ )2 = M(M + E ′ ), (12.299)
2 2
and
kk ′ = k(p + p′ − k) = ω(M + E ′ − |p′ | cos θ). (12.300)
If the incoming positron is very slow, then ω ≈ ω ′ ≈ M, and the two photons share
equally the rest energies of the electron and the positron. We can now substitute in
F of Eq. (12.290):
ω′ E ′ − |p′ |cos θ
= , (12.301)
ω M
and sum over all photon polarizations to obtain [see Fig. 12.10 and Eq. (12.257)]
2
(k · k′ )2 M M

|′∗ |2 = 1 + = 1 + cos2 (θ + θ′ ) = 1 −
X
2 ′2
− ′ . (12.302)
h,h′
k k ω ω

The last expression is found by observing that

k k ′ = ωω ′[1 − cos(θ + θ′ )], (12.303)

so that we can express

k k′ 1 (k + k ′ )2 1 (p + p′ )2 M(M + E ′ )
cos(θ + θ′ ) = 1 − = 1 − = 1 − = 1 − .
ωω ′ 2 ωω ′ 2 ωω ′ ωω ′
Energy conservation leads to
k k′ M(ω + ω ′)
cos(θ + θ′ ) = 1 − = 1 − , (12.304)
ωω ′ ωω ′
thus obtainig the right-hand side of (12.302). Note that from (12.303) and (12.304)
we find a relation
k k ′ = M(ω + ω ′ ). (12.305)
844 12 Quantum Electrodynamics

We now introduce the relativistic factor of the incoming positron γ ≡ E ′ /M and

write the relation (12.298) as
s
γ−1
s 1− cos θ
M γ−1 M γ+1
=1− cos θ, = √ . (12.306)
ω γ+1 ω′ γ − γ 2 − 1 cos θ

The result is integrated over all solid angles, and divided by 2 to account for Bose
statistics of the two final photons. This yields the cross section expressed in terms
of the classical electron radius r0 = α/M:

πr02 γ 2 + 4γ + 1
" #
1Z dσ γ+3
q  
σ(γ) = dΩ = 2
log γ + γ2 − 1 − √ 2 .
2 dΩ 1+γ γ −1 γ −1
(12.307)
For small incoming positron energy, the cross section diverges like

πr 2 γ πr 2 c
σ(γ) −−−→ σlow−energy (γ) ≡ √ 20 ≈ 0 , (12.308)
small p γ −1 v

whereas in the high-energy limit it behaves like

πr02
σ−
−−→ [log(2γ) − 1] . (12.309)
|p|→∞ γ

The detailed behavior is shown in Fig. 12.11.

The above result can be used to estimate the lifetime of a positron moving
through matter. We simply have to multiply the cross section σ by the incident
current density j = |p|/EV of a single positron and by the number N of target
electrons, which is Z per atom. For slow positrons we may use Eq. (12.308) to find
the decay rate
1.

0.8

0.6 σ(γ)/σ low−energy (γ)

0.4

0.2

γ
1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Figure 12.11 Electron-positron annihilation cross section divided by its low-energy lim-
p
iting expression (12.308) as a function of γ = 1/ 1 − v 2 /c2 of the incoming positron in
the laboratory frame.
12.11 Positronium Decay 845

N
Ṗ = πr02 v Z
. (12.310)
V
For lead, this yields a lifetime τ = 1/Ṗ ≈ 10−10 s.
The formula (12.313) is actually not very precise, since the incoming positron
wave is strongly distorted towards the electron by the Coulomb attraction. This
is the so-called initial-state interaction. It is closely related to the problem to be
discussed in the next section.

12.11 Positronium Decay

The previous result can be used to calculate the lifetime of positronium. Since the
momenta in positronium are nonrelativistic, the annihilation cross section (12.308)
is relevant. The wave function of positronium at rest in an s-wave is approximately
given by [recall (7.311)]
3
√
Z
d p † 8 πa3
|ψ S,S3 i = ψ̃(p)a†
b
p,s3 −p,s3
′ |0ihS, S 3 |s, s 3 ; s, s ′
3 i, ψ̃(p) = , (12.311)
(2π)3 (1 + a2 |p|2 )2
where S = 1 for ortho-positronium and S = 0 for para-positronium, and a is the
Bohr radius of positronium, which is twice the Bohr radius of hydrogen: a = 2 aH =
2/αM, the factor 2 being due to the reduced mass µ being half the electron mass
[recall the general formula (6.132)]. This amplitude can be pictured by the Feynman
diagrams shown in Fig. 12.12.

Figure 12.12 Lowest-order Feynman diagrams contributing to the decay of the spin
singlet para-positronium, i.e., the ground state.

In Eq. (7.310) we have calculated the charge conjugation parity ηC of these

states to be ∓1, respectively. Since a photon is odd under charge conjugation [recall
(7.550)], ortho-positronium can only decay into three photons (one is forbidden
by energy-momentum conservation). Thus only para-positronium decays into two
photons and the cross section calculated in the last section arises only from the spin
singlet contribution of the initial state. Since it was obtained from the average of a
total of four states, three of which do not decay at all, the decay rate of the singlet
state is four times as big as calculated previously from (12.308). The integral over all
momenta in the amplitude (12.281) can be factored out since the low-energy decay
rate is approximately independent of v and hence of p. Thus we obtain directly the
decay rate [15]
2
d3 p
Z
  1
Γpara→2γ ≡ Ṗpara = 4πr02 c  ψ̃(p)  = 4πr02 c |ψ(0)|2 = 4πr02 c
 
 (2π) 3  πa3
846 12 Quantum Electrodynamics

Mc2 α5
= . (12.312)
h̄ 2
Using the electron energy Mc2 ≈ 0.510 MeV and the Planck constant h̄ ≈ 6.682 ×
10−16 eV s, the ratio h̄/Mc2 is equal to 1.288 × 10−22 s. Together with the factor
2/α5 , this leads to a lifetime τ = 1/Γ ≈ 0.13 × 10−9 s.
The decay of the spin triplet ortho-positronium state proceeds at a roughly
thousand times slower rate
Mc2 α6 4(π 2 − 9)
Γorth→3γ ≡ Ṗorth = , (12.313)
h̄ 2 9π
with a lifetime of 140 × 10−9 s.

Figure 12.13 Lowest-order Feynman diagrams contributing to decay of the spin-triplet

ortho-positronium, the first excited state which lies 203.5 GHz above the ground state.

Decays into 4 and 5 photons have also been calculated and measured experimen-
tally. The theoretical rates are [14, 15]:

Mc2 7 α
 
Γpara→4γ ≈ 0.0138957 α 1−14.5 +O(α2 ) ≈ 1.43 × 10−6 × Γpara→2γ . (12.314)
h̄ π
Γorth→5γ ≈ 0.0189 α2 × Γorth→3γ ≈ 0.959 × 10−6 × Γorth→3γ . (12.315)

Experimentally, the branching ratios are 1.14(33) × 10−6 and 1.67(99) × 10−6 , in
reasonable agreement with the theoretical numbers. The validity of C-invariance
has also been tested by looking for the forbidden decays of para-positronium into
an odd and ortho-positronium into an even number of photons. So far, there is no
indication of C-violation.

12.12 Bremsstrahlung
If a charged particle is accelerated or slowed down, it emits an electromagnetic
radiation called Bremsstrahlung. This is a well-known process in classical electro-
dynamics and we would like to find the quantum field theoretic generalization of it.
First, however, we shall recapitulate the classical case.

12.12.1 Classical Bremsstrahlung

Consider a trajectory in which a particle changes its momentum abruptly from p to
p′ [see Fig. 12.14].
12.12 Bremsstrahlung 847

Figure 12.14 Trajectories in the simplest classical Bremsstrahlung process: An electron

changing abruptly its momentum.

The trajectory may be parametrized as:

(
µ τ p/M τ < 0,
x (τ ) = for (12.316)
τ p′ /M τ > 0,
where τ is the proper time. The electromagnetic current associated with this tra-
jectory is
dxµ (τ ) (4)
Z
µ
j (x) = e dτ δ (x − x(τ ))
dτ
e
Z h i
= dτ Θ(τ )pµ δ (4) (x − τ p/M) + θ(−τ )p′µ δ (4) (x − τ p′ /M) .(12.317)
M
After a Fourier decomposition of the δ-functions, this can be written as
d4 k −ikx µ
Z
j µ (x) = e j (k), (12.318)
(2π)4
with the Fourier components
pµ p′µ
!
µ
j (k) = −ie − ′ . (12.319)
pk p k
The vector potential associated with this current is found by solving the Maxwell
equation Eq. (12.50). Under the initial condition that at large negative time, the
vector potential describes the retarded Coulomb potential of the incident particle,
we obtain Z
Aµ (x) = i dx′ GR (x − x′ )j µ (x′ ), (12.320)
where GR (x − x′ ) is the retarded Green function defined in Eq. (7.162). At very
large times the particle has again a Coulomb field associated with it which can be
found by using the advanced Green function of Eq. (7.168):
Z
Aµout (x) = i dx′ GA (x − x′ )j µ (x′ ). (12.321)

As the acceleration takes place, the particle emits radiation which is found from the
difference between the two fields,
1
Z
Aµrad = dx′ [GR (x, x′ ) − GA (x, x′ )]j µ (x′ ). (12.322)
i
848 12 Quantum Electrodynamics

Remembering the list (7.212) of Fourier transforms of the various Green functions,
we see that the Fourier components of the radiation field are given by

Aµrad (k) = −i2πǫ(k 0 )δ(k 2 )j µ (k). (12.323)

The energy of the electromagnetic field is, by Eq. (7.427):

1
Z
H= d3 x (−Ȧµ Ȧµ − ∇Aµ ∇Aµ ). (12.324)
2
A classical field which solves the field equations can be Fourier decomposed into
positive- and negative-frequency components as in (7.390),
1  
Aµ (x) = e−ikx aµk + eikx aµ†
X
√ k , (12.325)
k 2V ωk
where
3
aµk ≡ ǫµ (k, λ)ak,λ
X
(12.326)
λ=0
are classical Fourier components. Then the combination of time and space deriva-
tives in (12.324) eliminates all terms of the form ak ak and a∗k a∗k , and we find
ωk
(−aµ∗ ν µ ν∗
X
H= k ak − ak ak ) gµν , (12.327)
k 2

just as in the calculation of the energy in (7.428). The radiation field (12.323)
corresponds to

aµk = −ij µ (k)|k0 =|k| . (12.328)

Inserting everything into (12.327) we derive, for large t, an emitted energy

d3 k
Z
k0
X ∗
E =− 0 3
jµ (k)j µ (k). (12.329)
2k (2π) h=1,2

Inserting (12.319) we obtain the energy emitted into a momentum space element
d3 k:
1 d3 k 2 2pp′ M2 M2
" #
dE = e − − , (12.330)
2 (2π)3 (pk)(p′ k) (pk)2 (p′ k)2
where polarization vectors have vanishing zeroth components. Dividing out the
energy per photon k 0 , this can be interpreted as the probability of omitting a photon
into d3 k:
d3 k 2pp′ M2 M2
" #
2
dP = 0 e − − . (12.331)
2k (2π)3 (pk)(p′ k) (pk)2 (p′ k)2
If we are interested in the polarization of the radiated electromagnetic field, we
make use of the local current conservation law ∂µ j µ which reads, in momentum
space,
k · j = k0j 0 . (12.332)
12.12 Bremsstrahlung 849

This allows us to rewrite

1
jµ∗ (k)j µ (k) = |j 0 (k)|2 − |j(k)|2 = 2
|k · j(k)|2 − |j(k)|2 = −|jT (k)|2 , (12.333)
k
where
kikj
!
jTi (k) ij
≡ δ − 2 j j (k) (12.334)
k
is the transverse part of the current. In the second-quantized description of the
energy, the transverse projection is associated with a sum over the two outgoing
photon polarization vectors [recall (4.334)]:

ki kj
ǫ∗i (k, h)ǫi (k, h).
X
δij − 2
= (12.335)
k h=1,2

The emitted energy (12.329) can therefore be resolved with respect to the polariza-
tion vectors as
d3 k
Z
k0 |(k, h) · j(k)|2 .
X
E= 0 3
(12.336)
2k (2π) h=1,2

This leads to an energy emitted into a momentum space element d3 k:

2
d3 k 2 X  pǫ p′ ǫ 

dE = e −  , (12.337)
2(2π)3 h=1,2  pk p′ k 


and a corresponding probability of omitting a photon into d3 k, by analogy with

(12.331).
Let us calculate the angular distribution of the emitted energy in Eq. (12.330).
Denote the direction of k by n:
k
n≡ . (12.338)
|k|
Then we can write pk as
pk = E|k|(1 − v · n) (12.339)
and find
d3 k e2 2(1−v · v′ ) M2 M2
" #
dE = − − . (12.340)
(2π)3 2 |k|2 (1−v · n)(1−v′ · n) E 2 (1−v · n)2 E ′2 (1−v′ · n)2

The radiation is peaked around the directions of the incoming and outgoing particles.

12.12.2 Bremsstrahlung in Mott Scattering

We now turn to the more realistic problem of an electron scattering on a nucleus.
Here the electron changes its momentum within a finite period of time rather than
abruptly. Still, the Bremsstrahlung will be very similar to the previous one. Let us
consider immediately a Dirac electron, i.e., we study the Bremsstrahlung emitted
850 12 Quantum Electrodynamics

Figure 12.15 Lowest-order Feynman diagrams contributing to Bremsstrahlung. The

vertical photon line indicates the nuclear Coulomb potential.

in Mott scattering. The lowest-order Feynman diagrams governing this process

are shown in Fig. 12.15. The vertical photon line indicates the nuclear Coulomb
potential
Zα
VC (x) = − . (12.341)
4πr
The scattering amplitude is found from the Compton amplitude by simply inter-
changing the incoming photon field
q
eAµ (x) = ǫµk,λe−ikx 2V k0
with the static vector potential
d4 q 1
Z
δ µ 0 VC (r) = −δ µ 0 Zα 4
2πδ(q 0)e−iqx 2 . (12.342)
(2π) |q|
The scattering amplitude is therefore
4πZαe M 1
Sfi = i 2
2πδ(p′0 + k ′ − p0 ) √ √
|q| V 2 E ′ E 2V k 0′
" #
′ ′ ′∗ 1 0 0 1 ′
× ū(p , s3 ) ǫ/ ′ γ −γ ′ ǫ/ u(p, s3 ), (12.343)
p/ + k/ ′ − M p/ − k/ ′ − M
where
q = p′ + k′ − p (12.344)
is the spatial momentum transfer. The amplitude conserves only energy, not spatial
momentum. The latter is transferred from the nucleus to the electron without any
restriction. The unpolarized cross section following from Sfi is
1 1 d 3 p′ d 3 k ′ F
Z
2 2 3
dσ = M Z (4πα) 0′ ′ 6
2πδ(p′0 + k ′0 − p0 ) 2 , (12.345)
2k E E v (2π) |q|
where we have used the incoming particle current density v/V = p/EMV and set
′
+ k/ ′ + M 0 / − k/ ′ + M ′
" !
1X ′∗ p
/ 0p p/ + M
F ≡ tr ǫ/ γ − γ ǫ/
2 h 2p′ k ′ 2pk 2M
p/ ′ + M
! #
0p
/ + k/ + M ′ p/ − k/ + M 0
γ ′
ǫ/ − ǫ/′ γ . (12.346)
2p k 2pk 2M
12.12 Bremsstrahlung 851

Let dΩe and dΩγ be the solid angles of outgoing electrons and photons dΩe and dΩγ ,
respectively. If we drop the prime on the emitted photon variables, so that ǫ′ and
k ′ are written as ǫ and k, and write for the energy k 0 of the outgoing photon the
variable ω, the differential cross section becomes
dσ Z 2 α3 M 2 |p′ |
= ωF. (12.347)
dΩe dΩγ dω π 2 |q|4 |p|
The calculation of F is as tedious as that of the trace (12.289) in the Klein-Nishina
cross section. It can be done again with the help of the formulas in Appendix 12A.
Let us introduce the angles θ and θ′ between the outgoing photon momentum and
the initial and final electron momenta p and p′ , respectively (see Fig. 12.16). Then

k
p′ p
θ′ θ

Figure 12.16 The angles θ ′ , θ, ϕ in the Bethe-Heitler cross section formula.

we calculate the polarization sums

( · p′ )2 = |p′2 | sin2 θ′ ,
X

( · p)2 = |p|2 sin2 θ,

X
(12.348)
h

(ǫ · p′ )( · p) = |p′ ||p| sin θ′ sin θ cos ϕ,

X
(12.349)
h
and obtain
1 p′2 sin2 θ′ p2 sin2 θ

2 2
F = (4E − q ) + (4E ′2 − q 2 )
4ω 2 (E ′ − p′ cos θ′ )2 (E − p cos θ)2
p2 sin2 θ + p′2 sin2 θ′
+ 2ω 2 ′
(E − p′ cos θ′ )(E − p cos θ)
p′ p sin θ′ sin θ cos ϕ

′ 2 2
−2 ′ (4E E − q + 2ω ) . (12.350)
(E − p′ cos θ′ )(E − p cos θ)
With this form of the function F , Eq. (12.347) is known as the Bethe-Heitler cross
section formula. For soft photon emission, ω → 0, the cross section becomes:
2
d3 k  ǫp′
 
dσ dσ  ǫp 
≈ × e2 −  . (12.351)
dΩe dΩe elastic 2ω(2π)3  kp′ kp 
 

It consists of the elastic cross section that is multiplied by the cross section of the
classical Bremsstrahlung.
852 12 Quantum Electrodynamics

12.13 Electron-Electron Scattering

The leading Feynman diagrams are shown in Fig. 12.17. The associated scattering

Figure 12.17 Lowest-order Feynman diagrams contributing to electron-electron scatter-

ing.

amplitude is given by

Sf i = (2ω)4δ 4 (p′1 + p′2 − p1 − p2 )(−ie)2 (12.352)

"
−igνρ
× −ū(p′1 , ε′1)γ ν u(p1, ε1 ) ′ 2
ū(p′2 , ε′2 )γ ρ u(p2 , ε2)
(p1 − p1 )
#
′ ′ ν −igνρ ′ ′ ρ
+ ū(p2 , ε2 )γ u(p1 , ε1) ū(p1 , ε1 )γ u(p2 , ε2 ) .
(p1 − p′2 )2

For the scattering amplitude tf i defined by

Sf i ≡ −ie2 (2π)4 δ 4 (p′1 + p′2 − p1 − p2 )tf i , (12.353)

we find
ū(p′1 , ε′1)u(p1 , ε1 )ū(p′2 , ε′2 )γν u(p2 , ε2 ) ū(p′2 , ε′2 )γ ν u(p1 , ε1 )ū(p′1 , ε′1 )γν u(p2 , ε2 )
tf i = − .
(p1 − p′1 )2 (p1 − p′2 )2
(12.354)

There is a manifest antisymmetry of the initial or final states accounting for the Pauli
principle. Due to the identity of the electrons, the total cross section is obtained by
integrating over only half of the final phase space.
Let us compute the differential cross section for unpolarized initial beams, when
the final polarizations are not observed. The kinematics of the reaction in the center
of mass frame is represented in Fig. 12.18, where θ is the scattering√ angle in this
frame. The energy E is conserved, and we denote |p| = |p′ | = p = E 2 − m2 . Using
the general formula (9.311) with the covariant fermion normalization V → 1/E, we
obtain
dσ M 2 e4
= |tf i |2 . (12.355)
dΩCM 4E 2 (2π)2

The bar on the right-hand side indicates an average over the initial polarizations
and a sum over the final polarizations. More explicitly, we must evaluate the traces1
12.13 Electron-Electron Scattering 853

Figure 12.18 Kinematics of electron-electron scattering in the center of mass frame.

1 X
|tf i |2 = |tf i |2
4 ε1 ε2 ε′ ε′
1 2

6 p1 + M 6 p′1 + M
( ! !
1 6 p + M ρ 6 p2 + M 1
= tr γν γp tr γ ν γ ′
4 2M 2M 2M 2M [(p1 − p1 )2 ]2
6 p + M 6 p2 + M ν 6 p + M ρ 6 p′1 + M
!
1
−tr γν γρ γ γ
2M 2M 2M 2M (p1 − p1 ) (p′2 − p1 )2
′ 2

+(p′1 ↔ p′2 )} . (12.356)

This is done using the formulas
tr [γν (6 p1 + M)γρ (6 p1 + M)] = 4(p1ν p′1 ρ − gν ρ p1 · p′1 + p1ρ p′1 ν + M 2 gνρ ),
tr [γν (6 p1 + M)γπ (6 p′1 + M)] × tr [γ ν (6 p2 + M)γ ρ (6 p2 + M)]
= 32[(p1 · p2 )2 + (p1 · p′2 )2 + 2M 2 (p1 · p′2 − p1 · p2 )], (12.357)
and further
γν (6 p1 + M)γρ (6 p2 + M)γ ν = −2 6 p′2 γρ 6 p1 + 4M(p′2 ρ + p1ρ ) − 2M 2 γρ , (12.358)
leading to
tr[γν (6 p1 + M)γρ (6 p′2 + M)γ ν (6 p2 + M)γ ρ (6 p′1 + M)] = −32(p1 p2 )2 − 2M 2 p1 p2 ],
and thus to
(p1 p2 )2 + (p1 p′2 )2 + 2M(p1 p′2 − p1 p2 )
(
1
|tf i |2 =
2M 4 [(p′1 − p1 )2 ]2
(p1 p2 )2 + (p1 p′1 )2 + 2M 2 (p1 p′1 − p1 p2 )
+
[(p′2 − p1 )2 ]2
(p1 p2 )2 − 2M 2 p1 p2
)
+2 ′ . (12.359)
(p1 − p1 )2 (p′2 − p1 )2
This can be expressed in terms of the Mandelstam variables s, t, u whose properties
were discussed in Eqs. (9.318)–(9.331), yielding
1 s2 + u 2 1 s2 + t2
( " # " #
1 2 2
|tf i |2 = + 2m (t − m ) + + 2m2 (u − m2 )
2M 4 t2 4 u2 4
1 s s
  
2 2
+ −m − 3m . (12.360)
tu 2 2
854 12 Quantum Electrodynamics

We may now easily express all invariants in terms of the center-of-mass energy ECM
and the scattering angle θ:
2
p1 p2 = 2ECM − M 2,
p1 p′1 = ECM
2
(1 − cos θ) + M 2 cos θ, (12.361)
p1 p′2 = ECM
2
(1 − cos θ) − M 2 cos θ.

This leads to the Møller formula (1932):

α2 (2E 2 −M 2 )2 2
−M 2 )2
" #
dσ 4 3 (ECM 4

= 2 CM 2 2 2 4 − 2 + 2 2 2
1+ 2 . (12.362)
dΩCM 4ECM (ECM −M ) sin θ sin θ (2ECM −M ) sin θ

In the ultrarelativistic limit of high incident energies M/ECM → 0, we have

α2 α2
!
dσ  4 2 1 1 1
  
≈ − + = + + 1 . (12.363)
2
dΩCM ur ECM sin4 θ sin2 θ 4 2
4ECM 4
sin θ/2 cos4 θ/2
2
For small energies where ECM ≃ M 2 , v 2 = (p2 − M 2 )/ECM
2
, we obtain the nonrela-
tivistic formula
dσ  α 2 1 4 3
    
= −
dΩCM nr M 4v 4 sin θ sin2 θ
4
 2 !
α 1 1 1 1
= + − , (12.364)
M 16v 4 sin4 θ/2 cos4 θ/2 sin2 θ/2 cos2 θ/2

that was first derived by Mott in 1930.

Comparing (12.364) with the classical Rutherford formula for Coulomb scattering
in Eq. (12.194), we see that the forward peak is the same for both if we set Z = 1
and replace M by the reduced mass M/2. The particle identity yields, in addition,
the backward peak.

12.14 Electron-Positron Scattering

Let us now consider electron-positron scattering. The kinematics and lowest-order
diagrams are depicted in Figs. 12.19 and 12.20. Polarization indices are omitted

Figure 12.19 General form of diagrams contributing to electron-positron scattering.

12.14 Electron-Positron Scattering 855

Figure 12.20 Lowest-order contributions to electron-positron scattering.

and in Fig. 12.20 four-momenta are oriented according to the charge flow. The
scattering amplitude may then be obtained from (12.354) by substituting

p2 → q1′ , u(p2 ) → v(q1′ ),

p′2 → −q1′ , u(p′2 ) → v(q1′ ),

and by changing the sign of the amplitude. The center of mass cross section is then
given by the formula

dσ M 4 e4
= 2
|tf i |2 (12.365)
dΩ 4ECM (2π)2

with
(p1 q1′ )2 + (p1 q1 )2 − 2M 2 (p1 q1 − p1 q1′ )
(
1
|tf i |2 =
2M 4 [(p′1 + p1 )2 ]2
(p1 q1′ )2 + (p1 p′1 )2 + 2M 2 (p1 p′1 + p1 q1′ )
+
[(p1 + q1 )2 ]2
(p1 q1′ )2 + 2M 2 p1 q1′
)
+2 . (12.366)
(p1 − p′1 )2 (p1 + q1 )2

This can be expressed in terms of the Mandelstam variables s, t, u whose properties

were discussed in Eqs. (9.318)–(9.331) as follows:

1 u 2 + s2 1 u2 + t2
( " # " #
1 2 2
|tf i |2 = + 2m (t − m ) + + 2m2 (s − m2 )
2M 4 t2 4 s2 4
1 u u
  
2 2
+ −m − 3m . (12.367)
st 2 2
It is then straightforward to derive the cross section formula first obtained by Bhabha
(1936):
− +
dσ e e 4
−M 4 2
−M 2 )2
"
α 5 8ECM (2ECM
= 2
− 2 2
+ 2
(12.368)
dΩ 2ECM 4 ECM (ECM −M 2 )(1−cos θ) 2(ECM −M 2 )2 (1−cos θ)2
4
(−1 + 2 cos θ + cos2 θ) + 4ECM
2
M 2 (1 − cos θ)(2 + cos θ) + 2M 4 cos2 θ
#
2ECM
+ 4
.
16ECM
856 12 Quantum Electrodynamics

In the ultrarelativistic limit, this becomes

− +
dσ e e α2 1 + cos4 θ/2 1 cos4 θ/2
" #
2
= + (1 + cos θ) − 2 . (12.369)
dΩ ur 8E 2
CM sin4 θ/2 2 sin2 θ/2
The nonrelativistic limit is simply
− +
dσ e e
2
α 1

= . (12.370)
dΩ nr M 16v 4 sin4 θ/2
This agrees again with the classical Rutherford cross section (12.194). The annihi-
lation diagram does not contribute in this limit,

Figure 12.21 Experimental data for electron-electron and electron-positron scattering at

θ = 900 as a function of the incident electron energy in the laboratory frame. (a) Electron-
electron scattering. The solid line represents the Møller formula, the broken one the Møller
formula when the spin terms are omitted. (b) Electron-positron scattering. The solid line
follows the Bhabha formula, the broken one the prediction when annihilation terms are
deleted. Data are from A. Ashkin, L.A. Page, and W.M. Woodward, Phys. Rev. 94, 357,
(1974).

The results of Eqs. (12.362) and (12.368) may be compared with experimental
data. At low energies we show in Fig. 12.21 some experimental data for electron-
electron scattering at 90 degrees [7]. Møller’s formula (12.362) is a good agreement
with the data. The agreement confirms the fact that the spin of the electron is really
1/2. If it was zero, the agreement would have been bad (see the dashed curve in
Fig. 12.21).
Electron-positron scattering data are fitted well by Bhabha’s cross section, and
the annihilation term is essential for the agreement. The energy of the incident
particle in the laboratory frame plotted on the abscissa is chosen in the intermedi-
ate range where neither the nonrelativistic nor the ultrarelativistic approximation
is valid. The numerical values show a significant departure from the ratio 2:1 be-
tween e− e− and e− e+ cross sections, expected on the basis of a naive argument of
indistinguishability of the two electrons.
12.15 Anomalous Magnetic Moment of Electron and Muon 857

12.15 Anomalous Magnetic Moment of Electron and Muon

The most directly observable effect of loop diagrams is a change in the magnetic
moment of the electron. Recall the precession equation (6.74).
As a consequence of loop diagrams in quantum electrodynamics, the gyromag-
netic ratio g in the relation (12.125) receives a correction and becomes g = 2(1 + a).
The number
a = (g − 2)/2 > 0 (12.371)
is called the anomalous magnetic moment of the electron. It has been measured
experimentally with great accuracy [16]:

a = 1 159 652 188.4(4.3) × 10−12 . (12.372)

The numbers in parentheses indicate the error estimate in the last two digits. For
the positron, the result is

ā = 1 159 652 187.9(4.3) × 10−12 . (12.373)

Quantum electrodynamics has been able to explain these numbers up to the last
digits – a triumph of quantum field theory.
To lowest order in α, the anomalous magnetic moment can easily be calculated.
Interestingly enough, it is found to be a finite quantity; no divergent integrals occur
in its calculation. The Feynman diagram responsible for it is the vertex correction
shown in Fig. 12.23. This diagram changes the electromagnetic current of an electron
from
hp′ , s′3 |j µ |p, s3 i = eū(p′ )γ µ u(p) (12.374)
to
hp′ , s′3 |j µ |p, s3 i = eū(p′ ) [γ µ + Λµ (p′ , p)] u(p), (12.375)

Figure 12.22 Cross section for Bhabha scattering at high energy, for scattering angle
450 < θ < 1350 as a function of total energy-momentum square s. The solid line is
calculated from quantum electrodynamics with first-order radiative corrections [8].
858 12 Quantum Electrodynamics

Figure 12.23 Vertex correction responsible for the anomalous magnetic moment.

where the vertex correction Λµ (p′ , p) is given by the Feynman integral

µ ′
Z
d4 k ν 1 1 1
Λ (p , p) = −i4πα 4
γ ′ γµ γν 2 . (12.376)
(2π) p/ + k/ − M p/ + k/ − M k

This integral is logarithmically divergent at large momenta kµ . It can be regularized

by cutting the integration off at some large but finite momentum Λ which is later
removed by a renormalization of the charge in the Lagrangian. There is also an
infrared divergence which is kept finite by cutting off the k-integration at a small
mass value k 2 = µ2 , much smaller than the electron mass, i.e., µ2 ≪ M 2 .
To do the integral, we rewrite the integrand as

p/′ + k/ + M µ p/ + k/ + M 1
γν ′ 2
γ 2 2
γν 2 , (12.377)
(p + k) − M (p + k) − M k − µ2

where we have introduced a small photon mass µ to avoid infrared divergencies at

small momenta. We now collect the product of denominators into a single denomi-
nator with the help of Feynman’s formula (11.158) for three denominators

1 Z 1 Z x
=2 dx dy[Ay + B(x − y) + C(1 − x)]−3 , (12.378)
ABC 0 0

so that
1 1 1
(12.379)
(p′ + k) − M (p + k) − M k 2
2 2 2
Z 1 Z x nh i h i o−3
=2 dx dy (p′ + k)2 −M 2 y + (p+k)2− M 2 (x−y) + (k 2 −µ2 )(1 − x) .
0 0

This can be simplified to

Z 1 Z x o−3
2
n
2 dx dy [k − p′ y − p(x − y)] − µ2 (1 − x) − M 2 x2 + q 2 y(x − y) .
0 0
(12.380)
After performing a shift of the integration variable

k → k + p′ y + p(x − y),
12.15 Anomalous Magnetic Moment of Electron and Muon 859

the vertex correction (12.376) takes the form

d4 k 1 x
Z Z Z
µ ′
Λ (p , p) = −i4πα 2 dx (12.381)
dy
0 0 (2π)4
p′ (1 − y) − k/ − p/ (x − y) + M]γ µ [/
[/ p (1 − x − y) − k/ − p/′ y + M]
×γ ν γν .
[k 2 − µ2 (1 − x) − M 2 x2 + q 2 y(x − y)]3
Instead of calculating this general expression, we shall restrict ourselves to matrix el-
ements of the current between electron states and evaluate ū(p′ , s′3 )Λµ (p′ , p)u(p, s3).
Then we can use the mass shell conditions p2 = p′2 = M 2 and the Dirac equa-
tions p/ u(p, s3) = Mu(p, s3 ) and ū(p′ , s′3 )/
p = M ū(p′ , s′3 ). We now employ ap-
propriately the anticommutation rules of the gamma matrices using the formulas
(12.220)–(12.222), perform a Wick rotation k 0 → ik 4 , and integrate over the Eu-
clidean four-momenta d4 kE = 2π 2 dk k 3 , setting p2 = −p2E , p′2 = −p′E 2 . Cutting off
the k-integral at kE = Λ, we arrive at the triple integral
1 x Λ kE3
Z Z Z
′
ū(p , s′3 )Λµ (p′ , p)u(p, s3 ) = α dx dy dkE kE3 3
0 0 0 [kE2 + µ2 (1−x)+M 2 x2 +qE2 y(x−y)]
n h i
×ū(p′ , s′3 ) γ µ kE2 − 2M 2 (x2 − 4x + 2) + 2q 2 (y(x − y) + 1 − x) (12.382)
o
− 4Mp′µ (y − x + xy) − 4Mpµ (x2 − xy − y) u(p, s3 ).
The denominator is symmetric under the exchange y → x−y. Under this operation,
the coefficients of pµ and p′µ are interchanged, showing that the vertex function is
symmetric in pµ and p′µ . We can therefore replace each of these coefficients by the
common average
( )
y − x + xy 1 1
→ [(y − x + xy) + (x2 − xy − y)] = − ,
x2 − xy − y 2 2x(1 − x)
and rewrite (12.382) as
1 x Λ kE3
Z Z Z
ū(p′ , s′3 )Λµ (p′ , p)u(p, s3) = α dx dy dkE kE3 3
0 0 µ [kE2 + µ2 (1−x)+M 2 x2 −q 2 y(x−y)]
n h i
′
×ū(p , s′3 ) γ µ
kE2 − 2M (x − 4x + 2) + 2q 2 (y(x − y) + 1 − x)
2 2
(12.383)
′µ µ
− 2M(p + p )x(1 − x)} u(p, s3).
This expression may be decomposed as follows:
1 ′µ
 
′
ū(p , s′3 )Λµ (p′ , p)u(p, s3) = ū(p ′
, s′3 ) µ
γ H(q ) − 2
(p + pµ )G(q 2 ) u(p, s3 ),
2M
(12.384)
with the invariant functions
α 1 x Λ kE2 − 2M 2 (x2 − 4x + 2) + 2q 2 [y(x − y) + 1 − x]
Z Z Z
2
H(q ) = dx dy dkE kE3 3
π 0 0 µ [kE2 + µ2 (1 − x) + M 2 x2 − q 2 y(x − y)]
(12.385)
860 12 Quantum Electrodynamics

and
α 2 1 x Λ 4M 2 x(1 − x)
Z Z Z
G(q 2 ) = M dx dy dkE kE3 3.
π 0 0 µ [kE2 + µ2 (1 − x) + M 2 x2 − q 2 y(x − y)]
(12.386)

The momentum integral in the second invariant function G(q 2 ) is convergent

at small and large momenta, such that we can set the photon mass µ to zero and
ultraviolet to infinity. Using the integral formula
∞ 1 1
Z
dkE2 kE2 = , (12.387)
0 (kE2 2 3
+ M1 ) 2M12

we obtain
α 2 1 x x(1 − x)
Z Z
G(q 2 ) = M dx dy . (12.388)
π 0 0 M 2 x2− q 2 y(x − y)
The integral over y yields
s
α 1 1 q2
Z
G(q ) = M 2
2
dx 4(1 − x) q arctan , (12.389)
π 0 q 2 (4M 2 − q 2 ) 4M 2 − q 2

leading to s
α 2M 2 q2
G(q 2 ) = q arctan , (12.390)
π q 2 (4M 2 − q 2 ) 4M 2 − q 2

which can be rewritten as

α 2θ q2
G(q 2 ) = , with sin2 θ ≡ . (12.391)
2π sin 2θ 4M 2
For small q 2 , it has the expansion

q2
!
α 2
G(q ) = 1+ + ... . (12.392)
2π 6M 2

In the first invariant function H(q 2), both the cutoff and the photon mass are
necessary to obtain a finite result. The divergence can be isolated by a subtraction
of the integrand, separating

H(q 2) = H(0) + ∆H(q 2 ), (12.393)

with a divergent integral

α 1 x ∞ kE2 − 2M 2 (x2 −4x+2)

Z Z Z
H(0) = dx dy dkE kE3 3
π 0 0 0 [kE2 + µ2 (1−x) + M 2 x2 ]
(12.394)
12.15 Anomalous Magnetic Moment of Electron and Muon 861

and a convergent one at large momenta

kE2 − 2M 2 (x2 −4x+2) + 2q 2 [y(x−y) + 1 − x]
(
2 αZ 1 Z x Z ∞
∆H(q ) = dx dy dkE kE3 3
π 0 0 0 [kE2 + µ2 (1−x) +M 2 x2 − q 2 y(x−y)]
kE2 − 2M 2 (x2 − 4x + 2)
)
− 2 3 . (12.395)
[kE + µ2 (1−x) + M 2 x2 ]
The divergent momentum integral in (12.394) must be performed with the help of
some regularization scheme, for which we choose the Pauli-Villars regularization,
replacing the photon propagator as follows:
1 1 1
2 2
→ 2 2
− 2 , (12.396)
k −µ k −µ k − Λ2
where Λ is a large cutoff mass. Then we may use the formula
kE2 + M22
(" # )
Z ∞
dkE2 kE2 2 2 2 3
− [µ2 → Λ2 ]
0 (kE + µ (1 − x) + M1 )
µ2 (1 − x) + M12 1 M22
= − log + . (12.397)
Λ2 (1 − x) 2 µ2 (1 − x) + M12
With this, the convergent momentum integral (12.395) yields
M 2 (x2 − 4x + 2) − q 2 [y(x − y) + 1 − x]
(
α 1 x Z Z
2
∆H(q ) = dx dy −
2π 0 0 µ2 (1 − x) + M 2 x2 − q 2 y(x − y)
M 2 (x2 − 4x + 2) µ2 (1 − x) + M 2 x2 − q 2 y(x − y)
" #)
+ 2 − log . (12.398)
µ (1 − x) + M 2 x2 µ2 (1 − x) + M 2 x2
If the photon mass is set equal to zero, the integral is divergent at x = 0. The
physical meaning of this infrared divergence will be explained later.
Let us first understand ∆H(q 2 ) for small q 2 by expanding ∆H(q 2 ) = ∆H ′ (0)q 2 +
O(q 4 ). Then ∆H ′ (0) is given by the integral
(
α 1 x y(x − y) + 1 − x
Z Z
′
∆H (0) = dx dy
2π 0 0 µ2 (1 − x) + M 2 x2
M 2 (x2 − 4x + 2)y(x − y)
)
y(x − y)
− + 2
[µ2 (1 − x) + M 2 x2 ]2 µ (1 − x) + M 2 x2
M2
!
α 1 1
= log 2 − , (12.399)
2π 3 µ 12
where we have discarded all terms which go to zero for µ → 0.
The full result is
M2
" ! !
2 α 2θ
H(q ) = H(0) + log 2 − 2 1 −
2π µ tan 2θ
#
4 2θ
Z θ
+ θ tan θ + dx x tan x + , (12.400)
tan 2θ 0 sin 2θ
which has precisely the first Taylor coefficient (12.399).
862 12 Quantum Electrodynamics

12.15.1 Form Factors

We now introduce the customary Lorentz-invariant decompositions of the matrix
elements of the current (12.375) of a spin-1/2 particle as follows:

i µν
 
′ µ ′ µ 2
hp |j |pi = eū(p ) γ F (q ) + σ qν G(q 2 ) u(p′ ). (12.401)
2M

The invariant functions F (q 2) and G(q 2 ) are the standard form factors of the elec-
tron. The relation between this and (12.384) follows directly from Gordon’s decom-
position formula (12.141), showing that G(q 2 ) in (12.386) coincides with G(q 2 ) in
(12.401), whereas

F (q 2 ) = 1 + H(q 2 ) − G(q 2 ). (12.402)

The charge form factor F (q 2 ) at q 2 = 0 specifies the charge of the electron.

Inserting (12.400), we obtain

M2
" ! !
2 α 2θ
F (q ) = F (0) + log 2 − 2 1 −
2π µ tan 2θ
#
4
Z θ
+ θ tan θ + dx x tan x , (12.403)
tan 2θ 0

where
M2
!
α Λ 9
F (0) ≡ 1 + log + − log 2 . (12.404)
2π M 4 µ
The value F (0) contains both the ultraviolet and the infrared cutoff. The subtracted
function ∆F (q 2 ) ≡ F (q 2) − F (0) has only an infrared divergence. In writing down
the expressions (12.403) and (12.404), we have ignored all contributions which vanish
for µ → 0 and Λ → ∞.
Due to the loop integral, the charge is changed to the new value
" !#
α Λ 9 M
e1 = eF (0) = e 1 + log + − 2 log . (12.405)
2π M 4 µ

The factor is commonly denoted as the renormalization constant Z1−1 . To order α,

it is
!
−1 α Λ 9 M
Z1 ≡ F (0) = 1 − log + − 2 log . (12.406)
2π M 4 µ

According to the theory of renormalization, this has to be equated with the exper-
imentally observed charge. After this we can replace, in Eq. (12.403), the number
F (0) by 1, and the bare fine-structure constant α by the physical one (keeping the
notation, for simplicity). The latter substitution is also done in G(q).
12.15 Anomalous Magnetic Moment of Electron and Muon 863

12.15.2 Charge Radius

For small momentum transfers q 2 , the renormalized charge form factor has the Taylor
series expansion
!
α M 3
FR (q 2 ) = 1 + q 2 2
log − + O(q 4 ). (12.407)
3πM µ 8

The form factor of the anomalous magnetic moment is, for small q,
α 1
GR (q 2 ) = q 2 2
+ O(q 4 ). (12.408)
3πM 4
Due to the emission and absorption of virtual photons, the electron is shaken
over a finite range. It is customary to define here a charge radius Re of the electron
by the first term in the expansion of the charge form factor2

Re2
F (q 2 ) = 1 − q 2 + ... . (12.409)
6
The factor 1/6 is due to the fact that for a uniformly charged shell of radius Re , the
Fourier transform of the charge density has the expansion
1
Z Z  
2
F (q ) = d x e ρ(x) = d x 1 + iqx − (qx)2 + . . . ρ(x)
3 iqx 3
2
2
1 R
Z
= 1 − q2 d3 x r 2 ρ(x) = 1 − q2 e + . . . . (12.410)
6 6
Setting q 0 = 0 in Eq. (12.407), we find that due to the vertex correction, the charge
radius Revc of the electron is given by
!
α 2 M 3
Re2 vc = 2
log − . (12.411)
πM µ 8

If the zeroth component of the current (12.401) couples to a static electric po-
tential A0 (x), the q 2 -term in (12.407) yields a correction factor

Re2 2
" #
1+ ∇ A0 (x). (12.412)
6

The fluctuations of the electron position leads to an extra term proportional to

the harmonic average of the potential. Since the potential obeys Gauss’ law, this
is nonzero only where there are charges. In an atom of nuclear charge Ze at the
origin, the potential is
Zα
eA0 (x) = VC (x) = − (12.413)
r
2
Note the difference of this quantity with respect to the classical electron radius (12.234).
864 12 Quantum Electrodynamics

and the harmonic average produces an additional δ-function at the origin:

e∇2 A0 (x) = 4πZα δ (3) (x). (12.414)

Thus the correction factor changes the Coulomb potential into an effective potential

Zα Re2 vc
VCeff = − + Zα 4πδ (3) (x). (12.415)
r 6
The prediction of this additional term is the origin of one of the early triumphs
of QED. The δ-function leads to an energy shift of s-wave orbits with respect to
those of nonzero orbital angular momentum, called the Lamb shift. It removes the
degeneracy between the 2S1/2 and the 2P1/2 predicted by Dirac’s equation in an
external Coulomb field. It will be discussed in detail below.

12.15.3 Anomalous Magnetic Moment

On the right-hand side of the current matrix element (12.401) we replace the γ µ -
term by a combination of vectors p′µ + pµ and σ µν qν , as before in (12.141), and see
that the magnetic moment of the free electron acquires a correction factor 1 + G(0).
Thus the gyromagnetic ratio g in Eq. (12.126) is changes from g = 2 to

g ≡ 2(1 + a) = 2[1 + G(0) + . . . ] (12.416)

[recall (12.371)]. The number G(0) yields therefore directly the anomalous magnetic
moment:
a = G(0). (12.417)
In contrast to the charge, this quantity is finite. Its value is [17]
α
a = G(0) = ∼ 1 161 409 74292 × 10−12 . (12.418)
2π
This result was first calculated by Schwinger [18]. It is about 1.5% larger than
the experimental values (12.372) and (12.373). The difference can be explained by
higher-order electrodynamic and strong-interaction corrections [19]. By including
all diagrams of sixth order in perturbation theory one finds, after a considerable
effort (there are 72 Feynman diagrams to sixth order), the expansion
 2  3
α α α
a= + c2 + c3 (12.419)
2π π π
with the coefficients [20]

197 1 3
 
c2 = + − 3 log 2 ζ(2) + ζ(4) = −0.328 478 965 . . . , r (12.420)
144 2 4
c3 = 1.176 11(42). (12.421)
12.15 Anomalous Magnetic Moment of Electron and Muon 865

Up to c2 , the theory is lower than the experimental number by only 1 part in 105 .
Adding the c3 -term, the theoretical value becomes
a = 1 159 652 140.4(27.1) × 10−12 , (12.422)
reducing the discrepancy to 1 part in 108 . The error is mainly due to the uncertainty
in the fine-structure constant [17].
The above calculation may be used to show that the anomalous magnetic moment
of an antiparticle is the same as for a particle. The matrix element of the initial
current between positron states is
hp′ , s′3 |j µ (x)|p, s3 i = eh0|b(p′ , s′3 )ψ̄(x)γ µ ψ(x)b† (p, s3 )|0i = −ev̄(p′ , s3 )γ µ u(p, s3 )eiqx ,
(12.423)
rather than (12.129). The minus sign is due to the odd number of exchanges of
field operators necessary to evaluate the anticommutators. If both momenta are
zero, only the zeroth component survives, showing that the charge of the positron
is −e. For small momentum transfers, we calculate the second spatial component
of (12.423) more explicitly as
1 (iγ 0 γ 1 )/2
hp′ , s′3 |j 2 (x)|p, s3 i = −ev̄(0, s3 )γ 2 eiζ v(0, s3 ), (12.424)
and compare this with (12.134). The linear term in ζ 1 contains the contribution
of the charge form factor to the magnetic moment. We see that (12.424) has the
opposite overall sign of (12.134) which is compensated by an opposite sign in the
exponent of the Lorentz transformation. Thus we obtain the matrix element of j 2
between positrons:
−ieζ 1 v̄(0, s3 )S3 v(0, s3 ) = ieζ 1 s3 , (12.425)
with an opposite sign in comparison with the electron in (12.135). The sign change at
the end is caused by the opposite spin orientation of the two-spinors ξ(s3 ) contained
in the spinors v(0, s3 ) [recall (4.684),(4.685) and (4.676)].
An opposite sign is also found for the contribution from the second form factor
where the matrix element of the electron is
i 21
eū(0, s3 ) σ q1 G(0)u(0, s3 ) = −ieū(0, s3 )S3 ζ 1G(0)u(0, s3) = −ieζ 1 s3 G(0).
2M
(12.426)
This produces the correction to the g-factor
g = 2[1 + G(0)]. (12.427)
For a positron, the matrix element of the current is
i 2ν
σ qν G(q 2 )]v(p, s3 ),
−ev̄(p′ , s′3 )[γ 2 F (q 2 ) + (12.428)
2M
and the second term becomes, for p′ = 0 and small p in the x-direction,
i 21
−ev̄(0, s3 ) σ q1 G(0)v(0, s3 ) = iev̄(0, s3 )S3 ζ 1G(0)v(0, s3 ) = ieζ 1 s3 G(0),
2M
(12.429)
866 12 Quantum Electrodynamics

which is exactly the opposite of (12.426). Thus a positron has the same correction
(12.427) to the g-factor as an electron. Note that the sign change found in (12.425)
is present also here, but it is compensated by a minus sign from an extra γ 0 -matrix
acting upon v † .
For a muon, the coefficients are [21]

c2 = 0.765 857 376(27), c3 = 24.050 508 98(44),

c4 = 24.050 508 98(44), c5 = +930(170), (12.430)

leading to

aµ = 1 165 847 057(29) × 10−12 . (12.431)

The strong interactions change this slightly via the composite Feynman diagrams

hadrons
γ

 

γ

Figure 12.24 Leading hadronic vacuum polarization corrections to aµ .

indicated in Fig. 12.24. Their contribution was estimated in [22] to amount to

astr
µ = 6 924(62) × 10
−11
to 6 988(111) × 10−11 . (12.432)

The amplitude is calculated from the formula

1 Z∞
astr
µ (vac. pol.) = 3 ds K (s) σ 0 (s)e+ e− →hadrons , (12.433)
4π 4mπ
2

where σ 0 (s)e+ e− →hadrons is the cross section for the process indicated in the subscript.
Some radiative corrections have been taken care of, such as initial state radiation
by a subtraction from the measured cross sections. The function K(s) is equal to

x2 x2
! " #
1 1+x 2

2 2
K(s) = x 1− + (1+x) 1+ 2 ln(1 + x)−x+ + x ln x, (12.434)
2 x 2 1− x
with
q
1− 1 − 4m2µ /s
x = q . (12.435)
1+ 1 − 4m2µ /s
12.15 Anomalous Magnetic Moment of Electron and Muon 867

√
s (GeV) astr
µ (vac. pol.) × 10
11

2mπ − 1.8 6343 ± 60

1.8 − 3.7 338.7 ± 4.6
3.7 − 5 + ψ(1S, 2S) 143.1 ± 5.4
5 − 9.3 68.7 ± 1.1
9.3 − 12 12.1 ± 0.5
12 − ∞ 18.0 ± 0.1
Total 6924 ± 62

Table 12.1 Different contributions to astr

µ (vac. pol.) in the integral 12.433.

It comes from the remaining part of the diagram. The contributions from the
different parts of the cross sections to astr
µ (vac. pol.) in the integral (12.433) are
separately listed in Table 12.15.3. The weak interactions in the standard model add
to this [23]

5 Gµ m2µ m2µ
" !#
1
aEW
µ (1 loop) = √ 1 + (1 − 4 sin2
θW ) 2
+ O ≈ 195 × 10−11 , (12.436)
3 8 2π 2 5 M2

where Gµ = 1.16637(1)×10−5 GeV−2 , sin2 θW ≡ 1−MW 2

/MZ2 ≃ 0.223. and M = MW
or MHiggs . See Fig. 12.25 for the Feynman diagrams.

 Z H

     
W W  

g g g

(a) (b) (c)

Figure 12.25 One-loop electroweak radiative corrections to aµ . The wiggly lines are
gluons.

Two-loop corrections change this slightly by

aweak
µ (2 loop) = −43(4) × 10−11 , (12.437)

if we assume a Higgs particle mass of mH ≃ 150 GeV (with little sensitivity to the
exact value). Altogether, we obtain

atheory
µ = 1 165 915 97(67) × 10−11 , (12.438)
868 12 Quantum Electrodynamics

in agreement with the experimental numbers [24]

aexp −9 −9 −9
µ = 11 659 204(7) × 10 , 11 659 202(22) × 10 , 11 659 204(7) × 10 . (12.439)

See Fig. 12.26 for a comparison of theory and experiment.

Figure 12.26 Measured values of aµ and prediction of the Standard Model (SM). The
small error bars of the theoretical value come from the left-hand estimate for the hadron
contribution in Eq. (12.432). For sources see Ref. [24].

12.16 Vacuum Polarization

Let us now turn to the vacuum polarization. The lowest-order Feynman diagram is
shown in Fig. 12.27.

Figure 12.27 Lowest-order Feynman diagram for the vacuum polarization.

12.16 Vacuum Polarization 869

In general, the propagator of the photon has the form

P µν (q)
Gµν (q) = i , (12.440)
q2
where the polarization tensor P µν (q) depends on the gauge [see (12.111)]. From the
lowest-order Feynman diagram in Fig. 12.27, the propagator receives a lowest-order
correction
Gµν κν
0 (q) [−iΠλκ (q)] G0 (q), (12.441)
where −Πµν (k) is given by the Feynman integral.

d4 p tr[γµ (/
p + q/ + M)γν (/
p + M)]
Z
2
−iΠµν (q) = e . (12.442)
(2π) [(p + q) − M ](p − M 2 )
4 2 2 2

Using Feynman’s formula (11.156)

1 1 1
Z
= dz , (12.443)
AB 0 [Az + B(1 − z)]2

we rewrite (12.442) as

2
Z 1 Z
d4 p tr[γµ (/
p + q/ + M)γν (/p + M)]
−iΠµν (q) = e dz . (12.444)
0 (2π) [(p + qz) + q (z − z ) − M 2 ]2
4 2 2 2

This expression is symmetric in µ and ν. A shift in the p-integration by an amount

qz brings it to the form

2
Z 1 Z
d4 p tr{γµ [/
p + q/ (1 − z) + M]γν (/ p − q/ z + M)}
−iΠµν (q) = e dz . (12.445)
0 (2π)4 [p2 + q 2 (z − z 2 ) − M 2 ]2
After evaluating the trace, dropping odd terms in p, and using the symmetry in µ
and ν, we obtain
Z 1 d4 p −(−gµν q 2 + 2qν qν )(z − z 2 ) −gµν (p2 −M 2 )+2pµ pν
Z
−iΠµν (q) = 4e2 dz .
0 (2π)4 [p2 + q 2 (z − z 2 ) − M 2 ]2
(12.446)
Because of the rotational symmetry of the integrand we can use the first replacement
rule in Eq. (11.136) to replace −gµν (p2 − M 2 ) + 2pµ pν by −gµν (p2 /2 − M 2 ).
The momentum integral is quadratically divergent, since there are two more
powers of integration variables in the numerator than in the denominator. The
situation is improved by imposing the requirement of gauge invariance, q µ Πµν (q) =
0. This makes the final results independent of the gauge choice in the photon
propagator. Thus we postulate, for the moment, the vanishing of the divergent
integral
1 d4 p −q 2 (z − z 2 ) − (p2 /2 − M 2 )
Z Z
−iq µ Πµν (q) = e2 qν 4 dz . (12.447)
0 (2π)4 [p2 + q 2 (z − z 2 ) − M 2 ]2
870 12 Quantum Electrodynamics

We shall verify in the next Subsection that this is guaranteed if we calculate the
Feynman diagrams by analytic regularization in D dimensions rather than D = 4.
Thus we are left with the logarithmically divergent integral
1d4 p 1
Z Z
2 2 2
−iΠµν (q) = −(−gµν q + qµ qν ) 8e dz(z − z ) .
0 (2π) [p + q (z − z 2 ) − M 2 ]2
4 2 2

(12.448)
2
Let us define the invariant function that accompanies the tensor (−gµν q + qµ qν ) =
q 2 Pµν (q) as −iΠ(q 2 ), i.e., we write

−iΠµν (q) ≡ −iPµν (q)q 2 Π(q 2 ). (12.449)

If we expand
Π(q 2 ) = Π(0) + Π′ (0)q 2 + . . . , (12.450)
we see that only Π(0) is logarithmically divergent:
Z 1 Z
d4 p 1
−iΠ(0) = −8e2 dz(z − z 2 ) , (12.451)
0 (2π)4 (p2 − M 2 )2

whereas all remaining terms in the expansion (12.450) are finite, for example the
first term:
1 d4 p 1
Z Z
−iΠ′ (0) = 16e2 dz(z − z 2 )2 . (12.452)
0 (2π) (p − M 2 )3
4 2

Since the mass M carries a small negative imaginary part, we now perform a Wick
rotation of the integration contour, setting p0 = ip4 and letting p4 run from −∞ to
∞. Thus we substitute p2 = −p2E and d4 p/(2π)4 = id4 pE /(2π)4 , and calculate from
(12.452):
′ e2 4 π 2 α 1
Π (0) = 4 2
= . (12.453)
(2π) 15 M 6πM 2 5
In the divergent quantity Π(0), we perform a Wick rotation so that the integral
d4 p/(2π)4 turns into an integral over all Euclidean four-momenta d4 pE /(2π)4 =
S4 dpE p3E = dp2E p2E /16π 2 . If we introduce an ultraviolet momentum cutoff at Λ2 , we
obtain
d 4 pE
1 1 1 1 ∞ p2E
Z Z
Π(0) = 8e2 2
= 8e2 dp2E
6 4
(2π) (pE + M ) 2 2 6 16π 2 0 (p2E + M 2 )2
Λ2
!
α1
= log 2 − 1 . (12.454)
π3 M

The complete invariant function of the polarization tensor is therefore

Λ2
" ! ! #
α 1 2 1 θ 1
 
2
Π(q ) = log 2 − 1 + + 1− − . (12.455)
π 3 M 3 3 sin2 θ tan θ 9
12.16 Vacuum Polarization 871

For small q 2 , we expand

!
2 1 θ 1 4
 
2
Π(q ) = + 1− = + θ2 + . . . , (12.456)
3 3 sin2 θ tan θ 9 15
and find the bracket to behave like
Λ2 1 q2
" ! #
′ 2 α 1
Π(0) + Π (0)q + . . . = log 2 − 1 + + O(q 4 ) . (12.457)
π 3 M 15 M 2
The last term agrees with the result (12.453) of a direct calculation of Π′ (0). It
is gratifying to find out that the condition (12.447) is fulfilled by the dimensional
regularization in d = 4 − ǫ dimensions.
Note that the expression (12.455) is real only for q 2 < M 2 . When q 2 > (2M)2 the
external field can produce electron-positron pairs and Π(q 2 ) acquires an imaginary
part. The imaginary part causes a decrease of the probability amplitude for the
occurrence of a pure scattering process below the threshold of pair production.
A string of vacuum polarization diagrams produces the geometric series

Gµν = Gµν µλ κν µλ κσ τν
0 + G0 [−iΠλκ ] G0 + . . . + G0 [−iΠλκ ]G0 [−iΠστ ]G0 + . . . , (12.458)

which can be summed up to

Gµν (q) = {[G−1 −1 −1

µν (q) + iΠµν (q)] } . (12.459)

Inserting the tensor decomposition (12.449), this can be written as

P µν (q) P µν (q)
Gµν (q) = i ≈ i (12.460)
q 2 [1 − Π(q 2 )] q 2 [1 + Π(0) + Π′ (0)q 2 − . . .]
For small q 2 , the photon propagator is, therefore,
−1
Pµν (q) α

µν
G (q) ≈ i 2
1 + Π(0) + 2
q2 . (12.461)
q 15πM
The divergent number 1+Π(0) can be absorbed into the field renormalization factor,
after defining
−1/2
AµR (x) = Z3 A(x) (12.462)
with
α Λ2
Z3 = [1 + Π(0)]−1 = 1 −
log 2 + . . . . (12.463)
3π m
This corresponds to renormalizing the charge to

αR = Z3 α. (12.464)

The propagator of the renormalized field AµR is then

!−1
Pµν (q) αR q 2
Gµν
R (q) =i 2 1+ . (12.465)
q 15π M 2
872 12 Quantum Electrodynamics

For a given static source of charge Ze at the origin, the Coulomb potential
Zα
VC (x) = − (12.466)
r
is obtained by applying the free propagator Gµν µ µ (3)
0 to the current j (x) = δ 0 Zeδ (x).
To lowest order in α, the vacuum polarization changes G0 µν (q) by a factor

αR q 2
!
1−
15π M 2
which is equivalent to multiplying the potential VC (x) by a factor

αR ∇2
!
1− .
15π M 2
The potential is therefore modified to
1 α
 
VCR (x) = −Zα + 4πδ (3) (x) . (12.467)
r 15πM 2
A comparison with (12.407)–(12.415) shows that the vacuum polarization decreases
the effective radius of the electron (12.411), derived from the vertex correction, by
α 1
Re2 vp = − , (12.468)
3πM 2 5
thus giving rise to a total effective radius determined to lowest order in α by:
!
α 1 M 3 1
Re2 vp =− log − − . (12.469)
3πM 2 5 µ 8 5
Since the finite radius of the electron gives rise to the Lamb shift to be derived
below, the vacuum polarization decreases the Lamb shift.

12.17 Dimensional Regularization

We still must show that the divergent integral (12.447) is really zero, to ensure the
gauge invariance of the self-energy (12.446). In D dimensions, the Dirac matrices
have the dimension 2D/2 , and (12.446) reads [26]

dD p −(−gµν q 2 +2qν qν )(z−z 2 )−gµν (p2 −M 2 )+2pµpν

1
Z Z
D/2 2
−iΠµν (q) = 2 e dz .
0 (2π)D [p2 + q 2 (z − z 2 ) − M 2 ]2
(12.470)
Then the previous replacement under the integral pµ pν → gµν /4 becomes pµ pν →
gµν /D [recall (11.136)]. The integral dD p over all four-momenta may now be Wick-
rotated to
Z Z ∞
i dD pE /(2π)D = iSD /2(2π)D dp2E (p2E )D/2−1 , (12.471)
0
12.18 Two-Dimensional QED 873

where SD is the surface of the unit sphere (11.126). Hence we can rewrite the integral
(12.447) in Euclidean spacetime as
1 dD pE −q 2 (z−z 2 )+ (1 − 2/D)p2E + M 2
Z Z
−iq µ Πµν (q) = e2 qν 2D/2 i dz , (12.472)
0 (2π)D [p2E + m2 ]2
where

m2 = m2 (q 2 , z) ≡ M 2 − q 2 z(1 − z), (12.473)

and further as
Z 1 Z
dD pE 2m2 /D + (1 − 2/D)(p2E + m2 )
−iq µ Πµν (q) = e2 qν 2D/2 i dz . (12.474)
0 (2π)D [p2E + m2 ]2
Now we use Formula (11.130) to replace

d D pE 1 Γ(D/2)Γ (ν − D/2)
Z
D 2 2 ν
= S̄ D Γ(ν)[m2 ]D/2−ν , (12.475)
(2π) (pE + m ) 2
so that
Γ(D/2)
−iq µ Πµν (q) = e2 qν 2D/2 iS̄D
2
2 2
   Z 1
× Γ (1−D/2) Γ(2) + 1− Γ (2−D/2) Γ(1) dz[m2 ]D/2−1 = 0, (12.476)
D D 0

thus guaranteeing the gauge invariance of the vacuum polarization tensor in any
dimension D [10].

12.18 Two-Dimensional QED

If the bare electron mass is zero, an interesting phenomenon arises in two dimensions:
The self-energy of the photon makes the photon massive. This happens in spite of
gauge invariance of the Lagrangian, as pointed out by Schwinger [6] in 1962. Since
this was a surprise, the two-dimensional massless QED is called the Schwinger model
[27].
Consider the self-energy (12.448) in D spacetime dimensions where the prefactor
4 in Eq. (12.446) is replaced by 2D/2 , so that (12.448) reads

dD p 1 1
Z Z
−iΠµν (q) = −(−gµν q 2 + qµ qν ) 2D/2+1 e2 dz(z−z 2 )
.
0 (2π) [p + q (z−z 2 )−M 2 ]2
D 2 2

(12.477)
D
Here the integral over d p in momentum spacetime is Wick-rotated, as in (12.471),
into the Euclidean momentum integral
SD ∞ 1 SD 2 D/2−2 Γ(D/2)Γ(2−D/2)
Z
i dp2E = i[m ] , (12.478)
2(2π)D 0 (p2E + m2 )2 2(2π)D Γ(2)
874 12 Quantum Electrodynamics

with m2 determined by Eq. (12.473). Inserting this into (12.477) yields, for −iΠµν (q)
of massless electrons, the tensor
1 SD Γ(D/2)Γ(2−D/2)
Z
2 D/2+1 2
−(−gµν q + qµ qν ) 2 ei dz[z(1 − z)]D/2−1 (−q 2 )D/2−2 ,
0 2(2π)D Γ(2)
(12.479)
which is equal to3

Γ2 (D/2) SD Γ(D/2)Γ(2−D/2)
−(−gµν q 2 + qµ qν )2D/2+1 e2 i (−q 2 )D/2−2 , (12.480)
Γ(D) 2(2π)D Γ(2)
so that we obtain, in D = 2 dimensions,

4e2 S2 e2 e2
−iΠµν (q) = −i(−gµν q 2 + qµ qν ) = −i(−g µν q 2
+ qµ qν ) = −iP µν ,
−q 2 2(2π)2 −πq 2 −πq 2
(12.481)
or, recalling (12.449),

e2
Π
! (q) = − . (12.482)
πq 2
Inserting this into (12.460), we obtain the renormalized photon propagator

P µν (q) P µν (q)
Gµν (q) = i = i . (12.483)
q 2 + q 2 Π(q) q 2 − e2 /π

This shows that the photon has acquired a nonzero mass m2γ = e2 /π. The effective
Lagrangian of the photon to this order is

Leff = − 14 F µν (1 + m2γ / )Fµν . (12.484)

The Schwinger model illustrates the remarkable fact that in spite of the gauge invari-
ance of the theory, a mass term can be generated for the photon by a loop diagram
of a massless fermion.

12.19 Self-Energy of Electron

The lowest-order Feynman diagram for the self-energy of the electron is shown in
Fig. 12.28. It adds to the electron propagator a term

G0 (p)[−iΣ(p)]G0 (p), (12.485)

where
i
G0 (p) = (12.486)
p/ − M
3
I.S. Gradshteyn and I.M. Ryzhik, Table of Integrals, Series, and Products, Academic Press,
New York, 1980, Formula 3.191.3.
12.19 Self-Energy of Electron 875

Figure 12.28 Lowest-order Feynman diagram for the self-energy of the electron.

is the free Dirac propagator and the self-energy −iΣ(p) of the electron is given by
the diagram in Fig. 12.28. Explicitly:

d4 k µ p/ − k/ + M 1
Z
−iΣ(p) = e2 4
γ 2 2
γµ 2 . (12.487)
(2π) (p − k) − M k

Using the anticommutation rule (12.214), we can simplify, in the numerator,

γ µ (/
p − k/ + M)γµ = −2(/
p − k/ ). (12.488)

The integral is logarithmically divergent at large k. It also has an infrared diver-

gence. To enforce convergence, we employ the Pauli-Villars regularization method
and modify the photon propagator as follows:
1 1 1
2
→ 2 2
− 2 , (12.489)
k k −µ k − Λ2
where Λ is a large cutoff mass and µ is a small photon mass. The self-energy becomes

d4 k −2(/
!
p − k/ ) + 4M 1 1
Z
−iΣ(p) = e2 − 2 . (12.490)
(2π) (p − k)2 − M 2
4 2
k −µ 2 k − Λ2

The mass parameter Λ cuts off the contribution of short-wave photons with k ≫ Λ2 .
At the end we shall take the cutoff to infinity.
By adding the same Feynman diagram repeatedly to an electron line, we obtain
the geometric series

G0 (p) + G0 (p)[−iΣ(p)]G0 (p) + G0 (p)[−iΣ(p)]G0 (p)[−iΣ(p)]G0 (p) + . . . , (12.491)

which can be summed up to

i
G(p) = . (12.492)
p/ − M − Σ(p)

Using Feynman’s formula (12.443), this can be rewritten as

d4 k
1
Z Z
2
Σ(p) = −ie dx
0 (2π)4
( )
−2(/ p − k/ ) + 4M
× − (Λ = µ) . (12.493)
[−(k − px)2 − p2 x(1 − x) + M 2 x + Λ2 (1 − x)]2
876 12 Quantum Electrodynamics

A simplification occurs by shifting the integration variable from k to k + px. Then

the terms proportional to k are off in k and can be dropped. After performing a
Wick rotation of the integration contour, we obtain
2π 2 e2 Z 1 Z 1
( )
−2/p (1 − x) + 4M
Σ(p) = ds dkE kE3 2 −(Λ = 0) .
(2π)4 0 0 [kE2 − p2 x(1 − x) + M 2 x + Λ2 (1 − x)]
(12.494)
The kE -integral is easily done, and yields
π 2 e2 Z 1
Σ(p) = dx [−2/
p (1 − x) + 4M]
(2π)4 0
−p2 x(1 − x) + M 2 x + Λ2 (1 − x)
( " # )
× log − (Λ = µ) . (12.495)
−p2 x(1 − x) + M 2 x
The right-hand side is a 4 × 4-matrix in spinor space which may be decomposed into
invariant functions
p − M)A(p2 ) + B(p2 ),
Σ(p) = (/ (12.496)
where
π 2 e2 Z 1 n o
A(p2 ) = dx[−2(1 − x)] log f (p 2
, x, Λ) − (Λ = µ) ,
(2π)4 0
π 2 e2 Z 1 n o
B(p2 ) = dx 2M(1 + x) log f (p 2
, x, Λ) − (Λ = µ) , (12.497)
(2π)4 0
with
−p2 x(1 − x) + M 2 x + Λ2 (1 − x)
f (p2 , x, Λ) = . (12.498)
−p2 x(1 − x) + M 2 x
The invariant functions are logarithmically divergent for large Λ.
We may expand them around the mass shell p2 = M 2 in powers of p2 − M 2 .
Then only the lowest expansion terms carry the logarithmic divergence:
π 2 e2 1 M 2 x2 + Λ2 (1 − x)
Z " #
2
A(M ) = dx [−2(1 − x)] log ,
(2π)4 0 M 2 x2
π 2 e2 Z 1 M 2 x2 + Λ2 (1 − x)
" #
2
B(M ) = dx 2M(1 + x) log ,
(2π)4 0 M 2 x2
π 2 e2 1
Z
B ′ (M 2 ) = dx 2M(1 + x)(1 − x)
(2π)4 0
Λ2 (1 − x) µ2 (1 − x)
" #
× − . (12.499)
M 2 x[M 2 x2 + Λ2 (1 − x)] M 2 x[M 2 x2 + µ2 (1 − x)]
All higher expansion terms are finite and can be evaluated with an infinite cutoff
Λ. We have also dropped all terms that vanish in the limit of zero photon mass µ.
Omitting the regular parts of the self-energy, the propagator becomes
i
G(p) =
p − M)[1 + A(M 2 )] + B(M 2 ) + B ′ (M 2 )(p2 − M 2 )
(/
12.20 Ward-Takahashi Identity 877

1 i
= , (12.500)
1 + A(M 2 ) + 2MB ′ (M 2 ) p/ − M − δM

where
B(M 2 )
δM ≡ − . (12.501)
1 + A(M 2 ) + 2MB ′ (M 2 )
The prefactor in the denominator is commonly denoted by

Z2 ≡ 1 + A(M 2 ) + 2MB ′ (M 2 ). (12.502)

1/2
It can be removed by renormalizing the field ψ(x) to ψR (x) = Z2 ψ(x). The
p − M). For large Λ2 ,
renormalized field has then a propagator with a pole term i/(/
we can simplify the results for Z2 and δM 2 to order α:

Λ2 1−x2 µ2 (1−x)
( " # " #)
α Z1 1−x
Z2 − 1 = − dx (1−x) log 2 +log 2 −2 1− 2 2 ,
2π 0 M x x M x + µ2 (1−x)
Λ2
" #
α 1 1−x
Z
2
δM = dx(1+x) log 2 + log 2 . (12.503)
2π 0 M x

Performing the integrals over x yields

M2
" #
α Λ 5
Z2 − 1 = − log + + 1 − log 2 ,
2π M 4 µ
α Z 1
Λ 3
 
δM 2 = dx 3 log + . (12.504)
2π 0 M 4

12.20 Ward-Takahashi Identity

It is important to realize that Z2 coincides with the renormalization constant Z1 =
F −1 (0) defined in Eq. (12.406) by the charge form factor to make the current matrix
element finite. This equality is a consequence of the Ward identity fulfilled by the
vertex correction Λµ (p′ , p) defined in Eq. (12.375):

∂
Λµ (p, p) = − Σ(p). (12.505)
∂pµ

For the total Dirac matrix Γµ (p′ , p) ≡ γ µ + Λµ (p′ , p) in the current (12.375), this
implies that
∂
Γµ (p, p) = [/
p − M − Σ(p)] . (12.506)
∂pµ

This relation holds to all orders in α. For off-diagonal matrix elements of the current,
there also exists a more general relation:

(p′ − p)µ Γµ (p′ , p) = [/

p ′ − M − Σ(p)] − [/
p − M − Σ(p)], (12.507)
878 12 Quantum Electrodynamics

from which (12.506) can be derived in the limit p′ → p. This is the famous Ward-
Takahashi identity. Its validity for a free theory is obvious. For the interacting
theory it will be proved at the end of this section..
A first important consequence of the Ward-Takahashi identity is the equality of
the renormalization constants Z1 = Z2 . To derive it, we use (12.506) to rewrite

∂
Γµ (p′ , p) = − Σ(p) + Γµ (p′ , p) − Γµ (p, p)
∂pµ
∂
= −γ µ + p − M − Σ(p)] + Γµ (p′ , p) − Γµ (p, p). (12.508)
[/
∂pµ

From the renormalization equation of the electron propagator (12.500), we see that

p/ − M − Σ(p) = Z2−1 [/
p − MR − ΣR (p)] , (12.509)

where MR = M + δM. This leads to

∂
Γµ (p′ , p) = γ µ (Z2−1 − 1) − Z2−2 ΣR (p) + Γµ (p′ , p) − Γµ (p, p). (12.510)
∂pµ

This must be compared with the definition of the renormalized vertex function

Γµ (p′ , p) = γ µ (Z1−1 − 1) + Z1−1 ΓµR (p′ , p), (12.511)

to conclude that

Z1 = Z2 ,
∂
ΓµR (p′ , p) = Z1 [Γµ (p′ , p) − Γµ (p, p)] − ΣR (p). (12.512)
∂pµ

The proof of the Ward-Takahashi identity (12.507) follows from the canonical
commutation relations of the current with the fields. These are certainly true in the
presence of interactions [28]:

[j 0 (x, x0 ), ψ(x′ , x0 )] = −eδ (3) (x − x′ )ψ(x, x0 ),

[j 0 (x, x0 ), ψ̄(x′ , x0 )] = eδ (3) (x − x′ )ψ̄(x, x0 ). (12.513)

These illustrate Noether’s theorem, which makes j 0 the generator of phase trans-
formations [recall Eq. (8.280)]. We now form the derivative of the time-ordered
expectation
h i
∂zµ h0|T̂ ψ(x′ )jµ (z)ψ̄(x)|0i = eh0|T̂ ψ(x′ )ψ̄(x)|0i δ (4) (x′ − z) − δ (4) (z − x) .
(12.514)
Expressed in terms of the full propagators, this equation becomes

iS(x′ − z) ∂zµ Γµ (z) iS(z − x) = iS(z − x) − iS(x′ − z). (12.515)

12.21 Lamb Shift 879

After a Fourier transformation, it reads

S(p′ )(p′ − p)µ Γµ (p′ , p)S(p) = S(p) − S(p′ ), (12.516)

or

(p′ − p)µ Γµ (p′ , p) = S −1 (p) − S −1 (p′ ), (12.517)

which is precisely the Ward-Takahashi identity (12.507).

As a consequence of the equality Z1 = Z2 , we find that the matrix elements of the
current between one-loop corrected electron states exhibit a unit charge, rather than
the divergent charge that is contained in the charge form factor F (0) = Z1−1 . Thus
we may conclude the important property of QED that the electric charge remains
unrenormalized to all orders in perturbation theory.

12.21 Lamb Shift

One of the most important early confirmations of the correctness of quantum elec-
trodynamics of electrons and photons came from the atomic Lamb shift. According
to Dirac’s theory, the energy spectrum of an electron in an external Coulomb field
is  1/2
2 2
 α Z 
Enl = Mc2 1 + q , (12.518)

n′ + (j + 1/2)2 − α2 Z 2 
where n′ = n − j − 1/2 = 0, 1, 2, . . . is the radial quantum number, and j =
1/2, 3/2, . . . the total angular momentum. Up to lowest order in the fine-structure
constant α = e2 /h̄c, this is approximately equal to

Z2 α2 Z 2
" ! #
1 1 3
Enl = Mc2 − α2 Mc2 2 1 + − + ... . (12.519)
2 n n j + 1/2 4n

These formulas show that the Schrödinger degeneracy, of all levels with the same
principal quantum number and different values of the orbital angular l, is modified in
Dirac’s theory, where levels with the same quantum numbers n and j are degenerate
for different l’s. The lowest states, where this degeneracy can be compared with
experimental data, are the n = 2 - states 2S1,2 and 2P1/2 . They are found to have
slightly different energies, thus contradicting the simple Dirac theory.
The energy difference is explained by quantum electrodynamics, and may be
attributed to three physical effects. First, the electron encircling the nucleus is
shaken by the vacuum fluctuations of the electromagnetic field over a range of the
order of the Compton wavelength. Thus it sees a harmonic average of the Coulomb
potential over this length scale. This lifts the level 2S1/2 against the level 2P1/2 by
roughly 27 MHz. Second, the anomalous magnetic moment of the electron changes
slightly the Coulomb attraction. Third, the photon running through the vacuum can
excite an electron-positron pair. These three effects together cause an upwards shift
880 12 Quantum Electrodynamics

of the level 2S1/2 with respect to the level 2P1/2 equal to [29] ∆E = 1 010 MHz, 68
MHz, and −27 MHz, respectively. The sum is roughly 1 052 MHz, a number which
was confirmed experimentally around 1950 [30]. The calculation of these effects will
now be reviewed.

12.21.1 Rough Estimate of the Effect of Vacuum Fluctuations

In order to estimate the first effect, consider a free nonrelativistic electron of mass
M in the vacuum. It is shaken by the zero-point oscillation of the electromagnetic
field, which causes an acceleration

M ẋ = eE. (12.520)

For a given frequency ω, the electron is shaken around its average position by a
displacement
e
∆x = − ω 2 E(ω). (12.521)
M
Its square-average is
e2 ∞ dω 2
Z
h(∆x)2 i = 2 hE (ω)i. (12.522)
M 0 ω4
The right-hand side can be estimated from the energy (7.341) of the free electro-
magnetic field in the vacuum, where it has the value [recall (7.434)]

1X
E = ωk . (12.523)
2 k,λ

The polarization sum runs over the two helicities. Hence, with the usual limiting
phase space integral (7.21) for the momentum sum, we have

d3 k 1 ∞ dω 3
Z Z
E= 3
ωk = ω . (12.524)
(2π) 2π 0 π

Since the vacuum energy is equally distributed between electric and magnetic fields,
we find
1 ∞ dω 3
Z
hE 2 i = ω , (12.525)
2π 0 π
and hence
2α ∞ dω 1
Z
2
h(∆x) i = 2 . (12.526)
M 0 π ω
The integral is divergent at small and large frequencies ω. A priori, it is unclear
which are the relevant frequencies that will contribute in a proper calculation. If we
consider only electromagnetic waves with wavelength shorter than the Bohr radius
aB = 1/αM, the integral starts at ωmin = Mα. Alternatively, we may expect the
energy of the atomic electrons to supply the relevant cutoff. Then ωmin = Mα2 . On
the high-frequency end, we omit wavelengths shorter than the Compton wavelength
12.21 Lamb Shift 881

of the electron, where classical considerations become invalid. Thus we cut off the
integral at ωmax = M. In this way we obtain a mean square somewhere in the range
2α
h∆xi2 = C, (12.527)
πM
with a constant C somewhere in the range
1 1
 
C ∈ log , 2 log ≈ 4.92 × (1, 2). (12.528)
ωmin ωmin
The electric interaction energy of an electron shaken over this region is modified as
follows. The Hamiltonian at the position x + ∆x(t) is
Z
H=e d3 xA0 (x + ∆x(t)). (12.529)

Averaging over ∆x gives

1
Z  
H = e d3 x A0 (x) + h∆xi ∆xj i∂i ∂j A0 (x)
2
1
Z  
3 2
= e d x 1 + h∆x · ∆xi∇ A0 (x)
6
1 α
Z  
3 2
= e d x 1+ C ∇ A0 (x). (12.530)
3π M 2
In an atom of nuclear charge Ze with a Coulomb potential (12.413), the Laplace
operator yields (12.414), and the potential is changed into an effective one:
Zα 1 α
VCeff = − + CZα4πδ (3) (x). (12.531)
r 3π M 2
For an atomic s-state with wave function ψn , we treat the extra potential VCeff
perturbatively. We evaluate its expectation value in a state of principal quantum
number n, and find the positive energy shift
4α
∆En = 2
ZαC|ψn (0)|2 . (12.532)
3M
Thus, the present rough estimate of the effect of vacuum fluctuations produces the
same term as before in (12.415), except for a different logarithmic factor (12.411).
For a hydrogen atom we insert
3/2
1 1

ψn (0) = √ , (12.533)
n3 π aB
where aB = 1/Mα is the Bohr radius. If the nuclear charge is Z, then aB is dimin-
ished by this factor. Thus, we obtain the energy shift
4α2 Z 1
∆En = 2
(MZα)3 3 C. (12.534)
3M nπ
882 12 Quantum Electrodynamics

For a hydrogen atom with n = 2, this becomes

α3 2
∆E2 = α MC. (12.535)
6π
The quantity Mα2 is the unit energy of atomic physics determining the hydrogen
spectrum to be En = −Mα2 /2n2 . Thus

Mα2 = 4.36 × 10−11 erg = 27.21eV = 2 Ry = 2 · 3.288 × 1015 Hz. (12.536)

Inserting this, together with α ≈ 1/137.036, into (12.542) yields4

∆E2 ≈ 135.6MHz × C, (12.537)

which amounts to the theoretical estimate

667.15 MHz < ∆E2 < 1334.3 MHz. (12.538)

The experimental Lamb shift5

∆ELamb shift ≈ 1 057 MHz, (12.539)

is indeed contained in this range.

12.21.2 Relativistic Estimate

The above simple estimate of the effect of vacuum fluctuations produces the same
type of correction to the Coulomb potential as the vertex correction in Eq. (12.415),
and the vacuum polarization in (12.467). Those two corrections yield an energy
shift in an s-state ψn (x) of principal quantum number n:

4α 2 4α3 Z 4 2
∆En = Zα|ψn (0)| C rel = α MCrel , (12.540)
3M 2 3πn3
with the constant Crel :
M 3 1
Crel ≡ log− − . (12.541)
µ 8 5
For a hydrogen atom in an s-state of principal quantum number n = 2 with n = 2,
this implies
α3
∆E2 = Mα2 Crel ≈ 135.641 MHz × Crel . (12.542)
6π
The result is not completely determined since it contains an infrared cutoff µ in the
constant Crel of Eq. (12.541). In a first approximation, we may imagine the atomic
energy Mα2 to provide the infrared cutoff for the photon energies. This leads to an
estimate for the energy shift of the 2S1/2 levels with respect to the 2P1/2 levels of
4
The precise value of the Lamb constant α4 M/6π is 135.641± 0.004 MHz.
5
See Notes and References.
12.21 Lamb Shift 883

the hydrogen atom that is about 6% smaller than the larger of the nonrelativistic
estimates (12.538):

∆E2 ≈ (1 334.3 − 51 − 27.3) MHz. ≈ 1 256 MHz. (12.543)

The intermediate expression exhibits the contribution of the terms −3/8 and −1/5 in
(12.541) (thus showing that vacuum polarization gives a negative shift −27.31 MHz).
This shift was first calculated by Uehling [31], who initially thought that vacuum
polarization was the main cause of the Lamb shift. He was disappointed to see it
contributing only with about 3%. In muonic atoms, however, vacuum polarization
does produce the dominant contribution to the Lamb shift for a simple reason:
While the above-calculated energy shifts contain a factor 1/M 2 in formulas such as
(12.593), where M is the mass of the muon, the leading vacuum polarization graph
still involves an electron loop containing the electron mass, thus being enhanced by
a factor (Mµ /Me )2 ≈ 2102 .

12.21.3 Effect of Wave Functions

In the above calculations the finite size of the electron was derived from a one-loop
Feynman-diagram in which the electron lines describe free particles. In an atom,
however, the electrons move in a Coulomb potential. The electron is bound to the
nucleus. A more accurate calculation should take into account the atomic wave func-
tions of the electron. This is most simply done in an approximation which treats
the electrons as nonrelativistic particles. Such an approximation carries an intrinsic
error caused by the fact that if a nonrelativistic electron emits a photon with energy
of the order Mc2 and larger, the recoil will necessarily make the electron relativistic.
Such an error can, however, be avoided by separating the relativistic from the non-
relativistic contributions. In the first contribution, the effect of the atomic binding
of the electrons is negligible, so that the electrons can be treated as free relativistic
particles. In the second contribution, the electrons remain approximately nonrela-
tivistic. There exists a natural energy scale K = Mα which is much larger than
the atomic energy Mα2 , but much smaller than the rest energy of the electron M.
The energy scale K serves to make the separation quantitative. For photons in the
upper energy regime, to be referred to as the hard-photon regime, we may equip the
photon with a mass α2 M ≫ µ ≪ αM, and deduce the Lamb shift from Eq. (12.540)
to be
4α 2 4α3 Z 4 2
∆En = Zα|ψ n (0)| C hard = α MChard , (12.544)
3M 2 3πn3
with the constant
M 3 1
Chard ≡ log − − . (12.545)
µ 8 5
The renormalization procedure has removed the ultraviolet divergences occuring in
the calculation of vertex corrections and vacuum polarizations. In the soft-photon
regime, this leaves us with the task of calculating the contribution from the photons
of low energy where the electrons stay nonrelativistic. In the transverse gauge with
884 12 Quantum Electrodynamics

∇A(x) = 0, the Hamiltonian has the form (12.160), with the Coulomb Hamiltonian
(12.161). The radiation field (12.162) contains, in the soft-photon regime under
consideration, only photon energies within the limited interval ω ∈ (0, K) where
K satisfies Mα2 ≪ K ≪ M. The radiation field is expanded in terms of photon
creation and annihilation operators as in Eq. (7.350):
1 h i
µ (x) = e−ikx ǫµ (k, λ)âk,λ + h.c. .
X
√ (12.546)
k,λ 2V ωk

The hats above the field operators are displayed, for clarity. To estimate the expected
effects we consider first the influence of the radiation field upon a free electron.

Effect of Vacuum Oscillations upon Free Electron

Let us first calculate the size of the energy shift for a free electron. This is the
quantum-mechanical version of the calculation in Subsection 9.10.1.
The Hamilton operator of a free electron of momentum p is Ĥ0 = p̂2 /2M, and
the wave functions is a plane wave
1
hx|pi = √ eipx (12.547)
V
of energy Ep = p2 /2M. If the electromagnetic field is quantized, there is a ground
state for the fluctuating vector potential, which is the vacuum state |0i of the pho-
ton field. The combined state will be denoted by |p; 0i. The calculation is done
perturbatively. Thus we expand the energy shift in powers of the coupling constant:
∆Ep = ∆Ep(1) + ∆Ep(2) + . . . , (12.548)

assuming the charge e to be sufficiently small. The first-order shift ∆Ep(1) is simply
the expectation value of the interaction operator
e e2
Ĥ int = − p̂ · Â(x) + 2 2 Â2 (x). (12.549)
Mc M c
Since this is odd in the field A(x) which has no expectation value, the first-order
energy shift vanishes. Thus we turn to the second-order shift ∆Ep(2) , in which we may
ignore the second term in the interaction (called the seagull term) that contributes
equally to all atomic levels. Then we have
e2 1
∆Ep = 2
hp; 0|p̂ · Â(x) p̂ · Â(x)|p; 0i. (12.550)
(Mc) Ep − Ĥ0
Inserting a complete set of intermediate electron-plus-single photon states |p; ki, we
find, in natural units with c = 1, h̄ = 1,

p · (k, λ)eikx p · ∗ (k′ , λ′ )e−ik x

h ih ′
i
e2 1 1
Z
d3 x
X
∆Ep =
M2 p2
q q
2V |k| 2V |k′ | 1
k,k′
λ,λ′ =1,2
− (p − k′ )2 − |k′ |
2M 2M
12.21 Lamb Shift 885

e2 X 1 |p · (k, λ)|2
= . (12.551)
M 2 k,λ=1,2 2V |k| p · k k2
− − |k|
M 2M
Performing the polarization sum
(p · k)2
|p · (k, λ)| = p −
2 2
X
, (12.552)
λ=1,2 k2

and replacing the sum over momenta by a phase space integral, V d3 k/(2π)3 , we
R

obtain
p2
∆Ep = JM (p2 ) (12.553)
2M
with
e2 Z d3 k 1 (p · k)2
" #
2 1
JM (p ) = 1 − . (12.554)
M (2π)3 |k| p2 k2 p·k k2
− − |k|
M 2M
Writing the last factor as
k2 p·k
1 1 2M − M
− + , (12.555)
k k k2 p·k
− +k
2M M
we obtain
JM (p2 ) = JM (0) + JM
′
(p2 ) (12.556)
with
e2 d3 k 1 (p · k)2 1
Z " #
JM (0) = − 1 − (12.557)
M (2π)3 k p2 k 2 k
and
k2 p·k
e2 Z 3
dk 1
"
(p · k) 2
# −
′
JM (p2 ) = 1− 2 2 2M M . (12.558)
M 3
(2π) k p k k2 p·k
− +k
2M M
The first term can easily be calculated:
e2 Z ∞ Z ∞
JM (0) = − 2 dk d cos θ(1 − cos2 θ)
4π M 0 0
4α
Z ∞
= − dk. (12.559)
3πM 0
Being in the soft-photon regime, a quadratic divergence at large k is avoided. The
integral is cut off at k = K << M. With the resulting finite JM (0), the kinetic
energy of the electron is changed from E(p) = p2 /2M to the renormalized energy
p2
ER (p) = [1 + JM (0)]. (12.560)
2M
886 12 Quantum Electrodynamics

Such a factor may be absorbed in the mass of the electron, by defining a renormalized
mass
M
MR = ∼ M − MJM (0). (12.561)
1 + JM (0)
′
In terms of this and the subtracted function JM (p2 ) of Eq. (12.558), we find the full
momentum dependence of the energy to order α:
p2 ′
ER (p) = [1 + JM (p2 )]. (12.562)
2MR R

Effect of Vacuum Oscillations upon Bound Electron

We now perform the same calculation once more in the presence of the Coulomb po-
tential VC (x), and study an electron in an orbit of principal quantum number n with
a wave function ψn (x), moving through a photon vacuum |0i. Then Eq. (30B.18)
becomes
e2 1
∆En = 2 hn; 0|p̂ · Â(x) p̂ · Â(x)|n; 0i. (12.563)
M En − ĤC
Inserting a complete set of states |n; 0i between the operators in Eq. (12.563) leads
to
e2 X X d3 k 1 (p̂eikx )nn′ (k, λ)(p̂e−ikx )n′ n ∗ (k, λ)
Z
∆En = 2 (12.564)
M n λ,λ′ (2π)3 2V |k| En − En′ − k

where (p̂eikx )nn′ denotes the matrix elements

Z
(p̂eikx )nn′ ≡ d3 xψn (x)(p̂eikx )nn′ ψn′ (x). (12.565)

A Schrödinger wave function corresponds graphically to an infinite set of static

photon exchanges. The energy shift to be calculated from Eq. (12.563) has therefore
the following graphical representation: The additional photon provides a radiative
nucleus
X
=
n
electron

Figure 12.29 Diagrammatic content in the calculation of the energy shift with the help
of Schrödinger wave functions. A hydrogen atom is represented by the fat line on the left
which results from an infinite sum of photon exchanges.

correction to the static photons which creates the bound state. The sum over n′
must, of course, include also an integral over the continuous wave functions. By
rewriting, as in (12.555),
1 1 En′ − En
=− + , (12.566)
En − En′ − k k k(En′ − En + k)
12.21 Lamb Shift 887

we obtain
e2 X ikx

−ikx

∆En = (p̂ i e ) nn ′ p̂ j e
M 2 n′ n′ n

d3 k 1
!
ki kj En′ − En
Z 
× 3 2
δij − 2 −1 + . (12.567)
(2π) 2k k En′ − En + k

The |k|-integration is again restricted to the soft-photon regime |k| < K << M. In
the integral involving only the first term in the brackets of (12.567), the energies En′
are absent and we can replace the sum over all wave functions |n′ i by an integral
over plane waves. Then we obtain the expectation value of the energy calculated for
a free electron in Eq. (12.560):

1
− hn|p2 |niJM (0). (12.568)
2M
This produces the same mass renormalization as before. Subtracting this from ∆En ,
we are left with
e2 X
∆En′ = (p̂i eikx )nn′ (p̂j e−ikx )n′ n
2M 2 n′
d3 k 1
!
ki kj En′ − En
Z
× 3 2
δij − 2 . (12.569)
(2π) k k En′ − En + k

In order to understand the behavior of the integral, it is useful to further split the
k-integral in the soft-photon regime. Thus we introduce the eneries Mα2 ≪ K ′ ≪
K = Mα, and distinguish an upper regime with wave number K ′ < k < K and a
lower regime with Mα2 ≪ k < K ′ . The corresponding energy shifts are denoted by
∆low En′ and ∆up En′ , respectively. In the upper regime we may approximate (12.569)
by

e2 X
" #
∆up En′ = (p̂i eikx )nn′ (p̂j e−ikx )n′ n (En′ − En ) + (i ↔ j) (12.570)
4M 2 n′
d3 k 1
!
Z
ki kj
× 3 2
δij − 2 .
(2π) k k

The sum over n′ can now be expressed with the help of the Hamilton operator ĤC
of the Coulomb system as
X   
p̂i eikx p̂j e−ikx (En′ − En ) + (i ↔ j) = hn|[p̂i eikx [ĤC , p̂j e−ikx ]]|ni.
nn′ n′ n
n′
(12.571)
Working out the commutators, this reduces to

[p̂i eikx , [ĤC , p̂j e−ikx ]] = ∂i ∂j VC (x) + . . . , (12.572)

888 12 Quantum Electrodynamics

where the omitted terms contain a factor ki , kj , or both. Due to their longitudinal
nature, they do not contribute to (12.571), where they are contracted with the
transverse projection tensor δij − ki kj /k 2 . Thus we find

e2 d3 k 1
!
ki kj
Z
∆up En′ = hn|∂i ∂j VC (x)|ni δij − . (12.573)
4M 2 K>|k|>K ′ (2π)3 k 3 k2

Doing the integral over all k-directions yields

e2 1 K dk
Z
∆up En′ = 2
hn|∇2 VC (x)|ni 2 . (12.574)
4M 3π K′ k
Inserting
∇2 VC (x) = 4πZαδ (3) (x), (12.575)
this can be rewritten as
4αZ K
∆up En′ = |ψn (0)| 2
log . (12.576)
3πM 2 K′
Note that this expression matches smoothly with the energy shift (12.544) caused by
hard photons. By adding the two results, the intermediate energy scale K cancels,
producing an energy shift depending only on the separation parameter K ′ :

4α M 3 1
 
(∆hard + ∆up )En′ = |ψn (0)| 2
log − − . (12.577)
3πM 2 K′ 8 5
Consider now the lower soft-photon part of the integral (12.569). Here we take
advantage of the fact that, due to the presence of the atomic probability distribution
|ψn (x)|2 , the integration over x is limited to a range |x| ∼< a
B = 1/αM. Since
′
|k| < K ≪ αM we see that
|kx| ≪ 1, (12.578)
so that we can neglect the exponential eikx in the matrix elements (p̂eikx )nn′ . Thus
we may evaluate the simpler expression

e2 X d3 k 1
!
Z K
ki kj En′ − En
∆low En′ ≈ (p i ) nn ′ (p j ) n ′ n δij − . (12.579)
2M 2 n′ 0 (2π)3 k 2 k2 En′ − En + k

This approximation amounts to neglecting the recoil of the atom. Performing the
integral over all k-directions yields

|pnn′ |2
∆low En′ ≈ JM (n, n′ )
X
(12.580)
n′
2M

with
2α 2
Z K′ En′ − En
′
JM (n, n ) = dk . (12.581)
πM 3 0 En′ − En + k
12.21 Lamb Shift 889

After doing the k-integral, this becomes

K′
" #
4α
JM (n, n′ ) ≈ (En′ − En ) log + iπΘ(En − En′ ) . (12.582)
3πM |En′ − En |

The imaginary part contributes to the decay rate of an atom from the state n
into a lower state n′ . It can be dropped in the final formula for the energy shift.
Let us decompose

K′ K′ En′ − En
log = log − log , (12.583)
En′ − En 2E1S 2E1S
(1)
thus separating (12.580) into two sums ∆low En and ∆low En . The first of these sums
can be treated the same way as in (12.571) and yields a contribution

(1) 4αZ K′
∆low En′ = |ψn (0)| 2
log . (12.584)
3πM 2 2E1S

Together with the energy shift (12.577), this becomes

(1) 4αZ 4α3Z 4 2

(∆hard + ∆up + ∆low )En′ = |ψn (0)| 2
C̄ = α M C̄ (12.585)
3πM 2 3πn3
with the constant
M 3 1 1 3 1
C̄ = log − − = log 2 − − ≈ 9.265. (12.586)
2E1S 8 5 α 8 5
The separation parameter K ′ has disappeared, and the result is unambiguous.
For the n = 2 -state of the hydrogen atom, the numerical value is
(1)
(∆hard + ∆up + ∆low )En′ ≈ 135.6 MHz × C̄
≈ (1 334 − 51) MHz ≈ 1 283 MHz. (12.587)

This is still larger than the experimental value 1 057 MHz. The relativistic treatment
of the hard regime together with the upper soft-photon regime have produced a
number which lies about 4% below the upper rough estimate (12.538).
It remains to calculate the second sum ∆low En′ in the lower soft-photon regime
which contains the effect of the wave functions in an essential way. This sum is
slightly more involved and requires evaluating a detailed spectral sum. We shall
write it as
2α
∆low En′ = − M 2 Sn (12.588)
3π
where Sn denotes the sum

|En′ − En |
|pnn′ |2 (En′ − En ) log
X
Sn = . (12.589)
n′ 2E1S
890 12 Quantum Electrodynamics

It is convenient to define an average energy E aν by the relation

E av
" #
2
X
Sn = |pnn′ | (En′ − En ) log . (12.590)
n′ 2E1S

The bracket is, by virtue of (12.571), equal to

1 1
|pnn′ |2 (En′ − En ) = hn|[p̂eikx [ĤC , p̂e−ikx ]]|ni = hn|∇2 VC |ni
X

n′
2 2
1 2 2Z 4 α
= 4πZα|ψn (0)| = . (12.591)
2 n3
This provides a useful check for the convergence of the calculation. An explicit
evaluation of the sum gives, for the 2S level [32]:

E av = 8.320 α2 M. (12.592)

By writing Sn in the form (12.590), the effect of the correction is to subtract a term
log(2E1S /E av ) from C̄ in (12.585), thus producing the result

4α 2 4α3 Z 4 2
∆En = |ψn (0)| C 2 = α MC2 , (12.593)
3πM 2 3πn3
with
tot 2E1S 1 3 1 2E1S
C2S 1/2 = C̄ − log
av
= log 2 − − − log av ≈ 7.146. (12.594)
E α 8 5 E
In combining the hard-photon with the two soft-photon results we have been
a bit careless since the first result (12.544) was derived with a finite photon mass
α2 M ≫ µ ≪ αM as an infrared cutoff parameter. The calculation of the finite
correction (12.595) should be done in the same way, i.e., we should integrate

e2 X
∆low En′ ≈ (pi )nn′ (pj )n′ n
2M 2 n′
d3 k
!
K 1 ki kj En′ − En
Z
× 3 2 2
δij − 2 √ . (12.595)
0 (2π) k + µ k + µ2 En′ − En + k 2 + µ2

The photon mass changes (12.581) into

1 k2
!
2α
Z K′ En′ − En
′
JM (n, n′ ) = dk 1 − √ . (12.596)
πM 0 3 k 2 + µ2 En′ − En + k 2 + µ2
The difference between the two integrals is
" #
′ 4α µ 5
JM (n, n′ ) − JM (n, n ) = ′
(En′ − En ) log + + iπΘ(En − En′ ) .
3πM 2|En′ − En | 6
(12.597)
12.21 Lamb Shift 891

tot
A comparison with (12.582) shows that the constant C2S 1/2 in (12.594) receives an

additional correction of 5/6 − log 2 ≈ 0.140, bringing it up to

tot M 3 1 2E1S 5
C2S 1/2 = log − − − log av + − log 2 ≈ 7.286, (12.598)
2E1S 8 5 E 6
corresponding to an energy shift

∆E2S 1/2 ≈ (1 334.3 − 51 − 27.3 − 287 + 20) MHz ≈ 989 MHz, (12.599)

which is smaller than the experimental value 1 057 MHz.

An important correction is missing in this calculation: the one caused by the
anomalous magnetic moment of the electron. This produces an energy shift of
about 68.5 MHz, which brings the Lamb shift from the value (12.599) up to 1 057.5
MHz, in excellent agreement with the experimental number 1 057 MHz.
This contribution will be calculated in the next subsection. Before we come to
that, however, we want to observe that, while the relativistic Lamb shift was applied
only to s-waves, the effect of the wave functions changes also the energy of states
with orbital angular momenta l ≥ 0. For such wave functions, we may define an
average energy analog to (12.590) as
" #
X
2 Eav
Sn = |pnn′ | (En′ − En ) log . (12.600)
n′ l=0
2E1S

The bracket must be taken for l = 0 since it vanishes for l > 0 by virtue of the same
commutator calculation as in (12.591), to be evaluated between l 6= 0-states. By
doing the spectral sum one finds the average energy [33]
av
E2P = 0.9704 α2 M. (12.601)

This raises the p-wave slightly by

av α3 2 av
∆ E2P = α MC2P , (12.602)
6π
with
av 2E1S
C2P = log av
≈ 0.03, (12.603)
2E2P
i.e., by
∆av E2P ≈ 135.6 MHz × C2P
av
≈ 4 MHz. (12.604)

12.21.4 Effect of the Anomalous Magnetic Moment

The relativistic current of the electron was found in Eq. (12.401) to have the form
i µν
 
′ µ ′ µ 2
hp |j |pi = eū(p ) γ F (q ) + σ qν G(q 2 ) u(p′ ), (12.605)
2M
892 12 Quantum Electrodynamics

with the form factors F (q 2 ) and G(q 2 ) given by Eqs. (12.403) and (12.391). For small
momentum transfers, these can be approximated by [recall (12.407) and (12.418)]
!
2 2 α M 3
F (q ) ≈ 1 + q Revc 2 , Revc 2 = log − , (12.606)
3πM 2 µ 8
α
G(q 2 ) ≈ .
2π
In configuration space, this amounts to an effective extra electromagnetic interaction
of the Dirac field of the electron which can be written as a Lagrangian density
i µν α
 h i 
Leff = −eψ̄(x) γ µ 1 − Revc 2 Aµ (x) + σ i ∂ν Aµ (x) ψ(x). (12.607)
2M 2π
In a static electric field, the radiative corrections provide a solution of the Dirac
equation ψnjm (x) with an additional energy
iα
 
∆E = eψ̄(x) γ 0 Revc 2 ∇2 A0 (x) + · E(x) ψ(x). (12.608)
4πM
To lowest order in α, we approximate the solutions to the Dirac equation by com-
binations of the nonrelativistic Schrödinger wave functions ψnlm with rest spinors
u(0, s3 ), combining them to state of total angular momentum j with the help of
Clebsch-Gordan coefficients, as shown in Eq. (6.184). The first term in (12.608)
leads precisely to the relativistic energy shift ∆hard E calculated in (12.544). How-
ever, that calculation lacked the energy shift due to the second term, which arises
from the anomalous magnetic moment:
eα
Z
∆E a = i d3 x ψ̄(x) ψ(x)E(x). (12.609)
4πM
In order to calculate this we have to approximate the solutions of the Dirac
equations a little better than so far. We go to the Dirac representation of the gamma
matrices and decompose the bispinor into two simple spinors, as in Eq. (6.184),
!
ξ(x)
ψ(x) = . (12.610)
η(x)

These satisfy the Dirac equation (4.500). We observed in (4.580) that the lower,
small components are related to the upper large ones by [see also (6.113)]
∇·
η(x) = −i ξ(x) + O(α). (12.611)
M
Neglecting the corrections of the order of α and inserting (12.611) into (12.610) and
(12.609), thereby using the explicit form of the electric field
x
E(x) = −Zα , (12.612)
|x|3
12.21 Lamb Shift 893

the integrand is found to contain an expression

x ·x ·x
ψ̄(x) · 3
ψ(x) = ξ † (x) 3 η(x) − η † (x) 3 ξ(x)
|x| |x| |x|
·x
" #
1
≈ ξ (x) 3  · ∇ξ(x) + h.c. .
†
(12.613)
2Mi |x|

After an integration by parts this leads to

·x
" #
x
Z
d xψ̄(x) · 3 ψ(x) ≈ ξ † (x)  · ∇,
3
ξ(x). (12.614)
|x| |x|3

We now observe that

x 1
∇· 3
= −∇2 = 4πδ (3) (x), (12.615)
|x| |x|

which allows us to calculate the commutator, using the chain rule, as being

·x
" #
1 1 1
 · ∇, = −∇2 − 2i · (x × ∇) 3 = 4πδ (3) (x) + 4 3 L · S. (12.616)
|x|3 |x| |x| |x|

The first term gives rise to another energy shift for s-waves, which may be written
in the same general form as the previous ones:
4α
∆En = 2
Zα|ψn (0)|2 C a , (12.617)
3M
with the constant being now
3
Ca ≡ . (12.618)
8
The second term depends on the angular momentum of the Dirac wave function.
Writing 2L · S = J2 − L2 − S2 , the eigenvalues of L · S are j(j + 1) − l(l + 1) − 3/4.
The expectation value of 1/|x|3 in l 6= 0 -states is

1 2
hnlm| 3
|nlmi = 3
(ZαM)3 . (12.619)
|x| l(l + 1)(2l + 1)n

Thus we obtain an energy shift for l 6= 0-states:

4α 2 a 4α3 Z 4 2 a
∆En = Zα|ψn (0)| C j,lj = α MCj,l (12.620)
3M 2 3πn3
with
1

1
3 1

 j =l+ ,
l+1 2

a
Cj,l ≡ for (12.621)
 −1
8 2l + 1 
 1
j = l − ; (l ≥ 1).
l 2
894 12 Quantum Electrodynamics

This leads for a level shift of the s-wave

α3 2
∆a E2S 1/2 = a
α MC1/2,0 , (12.622)
6π
with
a 3
C1/2,0 = . (12.623)
8
Remarkably, the anomalous magnetic moment causes a small energy shift also in
states of higher angular momenta. The 2P 1/2 -state of the relativistic hydrogen atom
is shifted opposite to the 2S 1/2 -state by
α3 2
∆a E2P 1/2 = a
α MC1/2,1 , (12.624)
6π
with
a 1
C1/2,1 ≡− . (12.625)
8
Together with the s-wave splitting (12.623) this leads to the total Lamb shift:

α3 2 α3 2
∆a E = a
α M(C1/2,0 a
− C1/2,1 )= a
α MCLamb , (12.626)
6π 6π
with
a 3 1 1
CLamb = + = , (12.627)
8 8 2
i.e., by
∆a E ≈ 135.6 MHz × CLamb
a
≈ 68.9 MHz. (12.628)
Recently, Lamb shifts have been measured also in heavy atoms where the atomic
levels are accessible to X-ray spectroscopy. There the theoretical analysis is much
more difficult due to the narrowness of the atomic wave functions and the large size
of the nucleus [35].

Appendix 12A Calculation of the Dirac Trace in

the Klein-Nishina Formula
The trace (12.289) can be expanded as

1 t8 + M 2 t61 t82 + M 2 t62 t83 + M 2 t63 t84 + M 2 t64

 
F = 2 1 + + + , (12A.1)
M (2pk)2 (2pk)(2p′ k ′ ) (2pk)(2pk ′ ) (2pk ′ )2

where tij denotes the following Dirac traces involving i gamma matrices:
1 1
t81 = ǫ ′ k/ ǫ/ p/ ǫ/ k/ ǫ/′ p/ ′ ),
4 tr(/ t82 = ǫ ′ k/ ǫ/ p/ ǫ/ ′ k/ ′ ǫ/ p/ ′ ),
4 tr(/
1 1
t83 = ǫ k/ ′ ǫ/′ p/ ǫ/ k/ ǫ/′ p/ ′ ),
4 tr(/ t84 = ǫ k/ ′ ǫ/′ p/ ǫ/′ k/ ′ ǫ/ p/ ′ ),
4 tr(/
1 1 (12A.2)
t61 = ǫ ′ k/ ǫ/ ǫ/ k/ ǫ/′ ),
4 tr(/ t62 = ǫ ′ k/ ǫ/ ǫ/′ k/ ′ ǫ/),
4 tr(/
1 1
t63 = ǫ k/ ′ ǫ/′ ǫ/ k/ ǫ/ ′ ),
4 tr(/ t64 = ǫ k/ ′ ǫ/′ ǫ/′ k/ ′ ǫ/).
4 tr(/

For brevity, we have omitted the symbols of complex conjugation on the outgoing polarization
vectors, which may be taken to be real and purely spatial, corresponding to linear polarizations.
Appendix 12A Calculation of the Dirac Trace in Klein-Nishina Formula 895

Traces involving an odd number of gamma matrices have been omitted, since they vanish. We now
use the crossing symmetry (12.286) to rewrite F as

1 t8 + M 2 t61 t82 + M 2 t62

 
F = 2 1 + + (ǫ ↔ ǫ ′
, k ↔ −k ′
) . (12A.3)
M (2pk)2 (2pk)(2p′ k ′ )
The traces are evaluated using the Wick expansion (12.253), together with the properties

p2 = p′2 = M 2 , k 2 = k ′2 = 0, ǫ2 = ǫ′2 = −1. (12A.4)

Using further the properties in the laboratory frame with p = (M, 0, 0, 0):

ǫp = ǫ′ p = 0,
ǫ ′ p′ = ǫ′ (p + k − k ′ ) = ǫ′ k,

we calculate t81 as follows: First we use ǫ2 = −1, and the Dirac relation (12.249) to reduce

ǫ/ p/ ǫ/ = − p/ ǫ/ ǫ/ = p/ . (12A.5)

Similarly, k 2 = 0 allows us to rewrite

k/ p/ k/ = − k/ k/ p/ + 2(kp)/
k = 2(kp)/
k. (12A.6)

Then t81 becomes

1
t81 = 2(pk) tr(/
ǫ ′ k/ ǫ/′ p/ ′ ). (12A.7)
4
Now there are only three Wick contractions in the expansion à la (12.253):

t81 = 2(pk) [(ǫ′ k)(ǫ′ p′ ) − (ǫ′ ǫ′ )(kp′ ) + (ǫ′ p′ )(kǫ′ )] , (12A.8)

which yield
t81 = 2(pk) [2(ǫ′ k)(ǫ′ p′ ) + (kp′ )] . (12A.9)
With the help of the substitution

kp′ = − 21 [(p′ − k)2 − M 2 ] = − 12 [(p − k ′ )2 − M 2 ] = pk ′ , (12A.10)

this becomes
t81 = 4(pk)(ǫ′ k)2 + 2(pk ′ )(pk). (12A.11)
Note that the similar relation

kp = 21 [(p + k)2 − M 2 ] = 12 [(p′ + k ′ )2 − M 2 ] = p′ k ′ , (12A.12)

leads to
t84 = −4(pk ′ )(ǫk ′ )2 + 2(pk ′ )(pk) (12A.13)
arising from t81 via the crossing operation (12.286). The other traces are

t61 = t64 = 0, (12A.14)

since they contain, in the middle, the products k/ ǫ/ ǫ/ k/ = − k/ k/ = −k 2 = 0 and k/ ′ ǫ/′ ǫ/′ k/ ′ =
− k/ ′ k/ ′ = −k ′2 = 0. We further find

ǫ ′ ǫ/ ǫ/′ ǫ/ k/ ′ k/ ) = M 2 2(kk ′ )(ǫǫ′ )2 −2(ǫǫ′ )(kǫ′ )(k ′ ǫ)−(kk ′ ) ,

t62 = t63 = 14 tr(/
 
(12A.15)

and finally,

t82 = t83 = −4(ǫǫ′ )2 (kp)(k ′ p) + 2(ǫǫ′ )2 (kk ′ )M 2 − 2(ǫǫ′ )(ǫk ′ )(ǫ′ k)M 2 − 2(ǫk ′ )2 (kp)
+ 2(ǫk)2 (k ′ p) − (kk ′ )M 2 + 2(kp)(k ′ p). (12A.16)
896 12 Quantum Electrodynamics

Hence
 8
t1 + M 2 t61 t82 + M 2 t62
 
1 1
+ ′ ′
(ǫ ↔ ǫ , k ↔ k ) = 2(ǫǫ′ )2
2M 2 (2pk)2 (2pk)(2p′ k ′ ) 2M 2
 
1 ′ 2 ′ 2 ′ ′ ′

′ ′
+ (ǫk ) (kp) + (ǫ k) (k p) − (kk )(k p) + (ǫ ↔ ǫ , k ↔ k ) .
2(pk)(pk ′ )

The bracket is, explicitly,

1
{(kk ′ )[(kp) − (k ′ p)]} , (12A.17)
2(pk)(pk ′ )

which can be simplified to

1
(kk ′ )2 , (12A.18)
2(pk)(pk ′ )

using the equation

kk ′ = k(p + k − p′ ) = kp − kp′ = kp − k ′ p. (12A.19)

Thus we obtain [34]

(k ′ k)2
 
1 ′ 2
F = + 4(ǫ ǫ) . (12A.20)
2M 2 (pk)(pk ′ )

In the laboratory frame where

pk ′ = M ω ′ , pk = M ω, (12A.21)

and Compton’s relation ensures that (12.264),

 
1 1
k ′ k = ω ′ ω(1 − cos θ) = M ω ′ ω − = M (ω − ω ′ ) ,
ω ′ ω

pk ′ = M ω ′ , pk = M ω, (12A.22)

and expression (12A.20) reduces to (12.254).

In the electron-positron annihilation process, the trace in Eq. (12.289) becomes

1 t81 − M 2 t61 t82 − M 2 t62

 
′ ′
F = + + (ǫ ↔ ǫ , k ↔ k ) , (12A.23)
M2 (2pk)2 (2pk)(2p′ k ′ )

and we find, after the replacement p′ = k + k ′ − p,

 
1 (k ′ k) [(kp) + (k ′ p)] ′ 2
F = − 4(ǫ ǫ) . (12A.24)
2M 2 (kp)(k ′ p)

Inserting (12.305) and using k p = ωM and k ′ p = ω ′ M , this becomes

(k ′ k)2
 
1 ′ 2
F = − 4(ǫ ǫ) , (12A.25)
2M 2 4(kp)(k ′ p)

instead of (12A.20), thus yielding Eq. (12.290) which we wanted to derive.

Notes and References 897

Notes and References

More on this subject can be found in the textbooks
S. Schweber, Relativistic Quantum Fields, Harper and Row, N.Y., N.Y., 1961;
J.D. Bjorken and S.D. Drell, Vol. I: Relativistic Quantum Mechanics, Vol. II: Relativistic Quantum
Fields, McGraw-Hill, New York, 1965;
J.M. Jauch and F. Rohrlich, The Theory of Photons and Electrons, Springer, Second Edition,
Berlin 1975;
L.H. Ryder, Quantum Field Theory, Second edition, Cambridge University Press, 1996, Cambridge;
M.E. Peskin and D.V. Schroeder, An Introduction to Quantum Field Theory, Addison-Wesley, New
York, 1995;
and, most of all,
C. Itzykson and J.-B. Zuber, Quantum Field Theory, McGraw-Hill, New York (1985).
The individual citations refer to:
[1] L.D. Faddeev and V.N. Popov, Phys. Lett. B 25 , 29 (1967).
[2] J.C. Ward, Phys. Rev. 78, 182 (1959).
[3] Y. Takahashi, Nuovo Cimento 6, 371 (1957).
[4] F. Rohrlich, Phys. Rev. 80, 666 (1950).
[5] See the third of the above textbooks;
[6] J. Schwinger, Phys. Rev. 125, 397 (1962); ibid. 128, 2425 (1962).
[7] A. Ashkin, L.A. Page, and W.M. Woodward, Phys. Rev. 94, 1974 (p. 357).
[8] K. Strauch in Proceedings of the Sixth Symposium on Electron and Photon Interactions at
High Energies, edited by H. Rollnik and W. Pfeil, North-Holland, Amsterdam, 1974.
[9] S. Coleman, R. Jackiw, and L. Susskind, Ann. Phys. 93, 267 (1975);
S. Coleman, ibid. 101, 239 (1976);
C. Adam, Nucl. Phys. B 54, 198 (1997).
[10] G. ’t Hooft and M.T. Veltman, Nucl. Phys. B 44, 189 (1972).
For earlier related publications see:
C. Bollini, J. Giambagi, and A.G. Dominguez, Nuovo Cimento 31, 550 (1964);
P. Breitenlohner and H. Mitter, Nucl. Phys. B 7, 443 (1968);
C. Bollini and J. Giambagi, Nuovo Cimento B 12 (1972);
J.F. Ashmore, Lett. Nuovo Cim. 4, 289 (1972).
[11] See the textbook
H. Kleinert, Multivalued Fields in Condensed Matter, Electromagnetism, and Gravitation,
World Scientific, Singapore 2008, pp. 1-497, (klnrt.de/b11)
[12] See for example
J.D. Jackson, Classical Electrodynamics, Wiley and Sons, New York, 1967, or
H. Ohanian, Classical Electrodynamics, Allyn and Bacon, Boston, Mass., 1988.
[13] W. Gordon, Ann. d. Phys. 2, 1031 (1929).
The matrix elements (12.177) can also be found very simply by group theory, as shown in
H. Kleinert, Fortschr. Phys. 6, 1 (1968), and Group Dynamics of the Hydrogen Atom,
Lectures presented at the 1967 Boulder Summer School, published in Lectures in Theoretical
Physics, Vol. X B, pp. 427–482, ed. by A.O. Barut and W.E. Brittin, Gordon and Breach,
New York, 1968.
[14] G.S. Atkins and E.D. Pfahl, Phys. Rev. A 59, R915 (1999);
G.S. Atkins and F.R. Brown, Phys. Rev. A 28, 1164 (1983).
898 12 Quantum Electrodynamics

[15] For a review of the experimental situation see the internet page http://int.phys.
washington.edu/~int talk/WorkShops/int 02 3/People/Vetter P/pstalk.pdf.
[16] See
R.S. Van Dyck, in Quantum Electrodynamics, ed. by T. Kinoshita, World Scientific, Singa-
pore, 1990, p. 322.
[17] The numerical value uses the fine-structure constant α−1 = 137.035 997 9(3 2) found with
the help of the quantum Hall effect. See M.E. Cage et al., IEEE Trans. Instrum. Meas. 38,
284 (1989).
[18] J. Schwinger, Phys. Rev. 73, 416 (1948).
[19] For a discussion see the review by
J. Gailey and E. Picasso, Progr. Nucl. Phys. 12, Part 1, eds. D.M. Brink and J.H. Mulvey,
Pergamon, Oxford, 1970, pp. 43-.
[20] The calculation of c2 was performed in the Feynman gauge by
A. Petermann, Helv. Phys. Acta 30, 407 (1957);
C.M. Sommerfield, Phys. Rev. 107, 328 (1957);
and in another gauge by
G.S. Adkins, Phys. Rev. D 39, 3798 (1989).
The number for c3 was published in T. Kinoshita, IEEE Trans. Instrum. Meas. 38, 172
(1989).
See also the review articles by T. Kinoshita as well as by
R.Z. Roskies, E. Remiddi, and M.J. Levine, in Quantum Electrodynamics, ed. by T. Ki-
noshita, World Scientific, Singapore, 1990, p. 218 and p. 162, respectively.
[21] T. Kinoshita in Quantum Electrodynamics, ed. by T. Kinoshita, World Scientific, Singapore,
1990, p. 419; A. Czarnecki and W.J. Marciano (hep-ph/0102122).
[22] The small errors are estimated by
M. Davier, Nucl. Phys. B (Proc. Suppl.) 76, 327 (1999) (hep-ex/9912044).
The second, more conservative estimate is due to
F. Jegerlehner, in Radiative Corrections, edited by J. Solà (World Scientific, Singapore,
1999), pp. 75–89. See also his lectures in a seminar at New York University in honor of A.
Sirlin’s 70th Birthday.
[23] R. Jackiw and S. Weinberg, Phys. Rev. D 5, 2473 (1972);
W.A. Bardeen, R. Gastmans, and B.E. Lautrup, Nucl. Phys. B 46, 315 (1972).
[24] J. Bailey et al., Phys. Lett. B 68, 191 (1977);
F.J.M. Farley and E. Picasso, in Quantum Electrodynamics, ed. by T. Kinoshita, World
Scientific, Singapore, 1990, p. 479.
H.N. Brown et al., Phys. Rev. Lett. 86, 2227 (2001) (hep-ex/0102017);
G.W. Bennett et al., Phys. Rev. Lett. 89, 101804 (2002); Erratum-ibid. 89, 129903 (2002)
(hep-ex/0208001);
See also http://phyppro1.phy.bnl.gov/g2muon.
[25] W. Gordon, Ann. d. Phys. 2, 1031 (1929).
The matrix elements (12.177) can also be found very simply by group theory, as shown by
H. Kleinert, Fortschr. Phys. 6, 1, (1968), and Group Dynamics of the Hydrogen Atom,
Lectures presented at the 1967 Boulder Summer School, published in Lectures in Theoretical
Physics, Vol. X B, pp. 427–482, ed. by A.O. Barut and W.E. Brittin, Gordon and Breach,
New York, 1968.
[26] For details the reader is referred to Bernard de Wit’s lecture on this subject in
http://www.phys.uu.nl/~bdewit/ftip/AppendixE.pdf.
Notes and References 899

[27] The Schwinger model with a nonzero initial electron mass is discussed in Ref. [9]
[28] Y. Takahashi, Nuovo Cimento 6, 370 (1957).
[29] H.A. Bethe, Phys. Rev. 72, 339 (1947);
J. Schwinger and V. Weisskopf, Phys. Rev. 73, 1272A (1948);
R.P. Feynman, Phys. Rev. 76, 939 (1948);
N.M. Kroll and W.E. Lamb, Phys. Rev. 75, 388 (1949);
J.B. French and V. Weisskopf, Phys. Rev. 75, 1240 (1949);
H.A. Bethe, L.M. Brown, and J.R. Stehn, Phys. Rev. 77, 370 (1950);
A.J. Layzer, Phys. Rev. Lett. 4, 580 (1960);
G.W. Erickson and D.R. Yennie, Annals of Phys. 35, 271, 447 (1965);
G.W. Erickson, Phys. Rev. Lett. 27, 780 (1971);
P.J. Mohr, Phys. Rev. Lett. 34, 1050 (1975).
[30] The first experiment was done by
W.E. Lamb and R.C. Retherford, Phys. Rev. 72, 241 (1947); 79, 549 (1950); 81, 222 (1951);
86, 1014 (1951).
See also the review
W.E. Lamb, Rep. Progr. Phys. 14, 19 (1951).
A more accurate value ∆E = 1 057.8 ± 0.1 MHz was found by
S. Triebwasser, E.E. Dayhoff, and W.E. Lamb, Phys. Rev. 89, 98 (1953).
[31] E.A. Uehling, Phys. Rev. 49, 55 (1935).
[32] H.A. Bethe and E.E. Salpeter in Encyclopedia of Physics (Handbuch der Physik), Springer,
Berlin, 1957, p. 405.
[33] For more average energy values see J.M. Harriman, Phys. Rev. 101, 594 (1956).
[34] The calculation can actually be performed with a simple reduce program which can be
downloaded from http://www.physik.fuberlin.de/~kleinert/b6/programs/reduce.
[35] For a review see
P.J. Mohr, G. Plunien, and G. Seff, QED-Corrections in Heavy Atoms, Phys. Rep 293, 227
(1998).