5 Gauge Fields

In this chapter, we will quantize the Maxwell Lagrangian. This will lead to massless spin-1
particles. As we will see in the next chapter, these particles mediate long range forces between
charged particles.

5.1 Wigner’s Classification

In the previous chapter, we showed how particles can be identified with unitary irreducible
representations of the Poincaré group. These representations are labeled by the mass m and the
spin j of the particles. Since the mass is a Lorentz invariant, it is obviously a good quantum
number to describe particles. To understand the role of spin, we go to the rest frame of the
particle, where pµ = (m, 0, 0, 0), for m > 0. Transformations associated with the rotation group
SO(3) (also called the little group) leave the four-momentum invariant. The quantum number
associated with the group SO(3) is the spin j. States within each representation are labeled
by jz = j, j + 1, . . . , +j. This means that massive particles of spin j have 2j + 1 degrees of
freedom.
For massless particles, on the other hand, the rest frame does not exist. Choosing the direction
of propagation to be the z-axis, we can write pµ = (E, 0, 0, E). The little group that leaves the
four-momentum invariant is SO(2), the group of rotations in the (x, y) plane. The generator of
the little group is Jz . The eigenvalues of Jz represent the angular momentum in the direction
of propagation of the particles (here, chosen to be the z direction), also called the helicity. It
can be shown that these eigenvalues are quantized, h = 0, ± 12 , ±1, . . .. The two massless states
with helicity ±1 describe the photon and those with helicity ±2 correspond to the graviton.
Neutrinos and antineutrinos have helicity h = 12 and h = + 12 , respectively.1
Particles with integer spin j are naturally identified with tensor fields Tµ1 ...µj . However, in
general, the tensor Tµ1 ...µj has 4j components, while a massive spin-j particle has 2j + 1 degrees
of freedom (and a massless particle with spin has only two degrees of freedom). The tensor
Tµ1 ...µj is not yet an irreducible representation of spin-j particles. We need to impose extra
constraints to isolate the spin-j component of the field. We will illustrate this for the case of
spin-1 particles.
1
Strictly speaking, a massless particles with helicity +h and a massless particle with helicity h are di↵erent
particles. However, we may write the helicity operator as h = p̂ · J, where p̂ is the unit vector in the direction of
propagation. Helicity is therefore a pseudo-scalar, i.e. it changes sign under parity. If the interaction conserved
parity, then particle of helicity +h and h must appear symmetrically in the theory. It is then natural to treat
the two helicities as two degrees of freedom of the same particle, e.g. the two polarizations of the photon and the
graviton.

89
 5.2 Rediscovering Maxwell 90

5.2 Rediscovering Maxwell

Our goal is to construct a Lagrangian for Aµ that propagates the right number of degrees of
freedom, i.e. three for a massive field and two for a massless field.

Massless spin-1 particles

Lorentz invariance and locality allow two possible kinetic terms for a vector field

Lkin = a1 L1 + a2 L2 , (5.2.1)

where

L 1 = @ µ A⌫ @ µ A⌫ , (5.2.2)
µ ⌫ µ ⌫
L2 = @µ A @⌫ A = @⌫ A @µ A + boundary terms . (5.2.3)

Let us write Aµ ⌘ µ + @µ , where @ µ µ = 0. The fields µ and are referred to as the
transverse mode and the longitudinal mode, respectively. The Lagrangian (5.2.1) implies that
the longitudinal mode satisfies
Lkin = (a1 + a2 )(2 )2 , (5.2.4)
which is equivalent to
✓ ◆2
˜2 1 ˜2
Lkin = (a1 + a2 ) , (5.2.5)
4
after integrating out the Lagrange multiplier field ˜ = 22 . Defining ⌘ 1 + 2 and ˜ ⌘ 1 2,
this becomes ✓ ◆2
1
Lkin = (a1 + a2 ) + 1 2 1 22 2 ( 1 2) . (5.2.6)
4
We see that one of the two fields is always a ghost, unless a1 + a2 = 0. In that case, the
longitudinal mode is non-dynamical and the transverse mode satisfies

LÂ 2
kin = a1 (@µ Â⌫ ) , (5.2.7)

which has the right normalization for a1 = 12 . The only allowed kinetic term therefore is
1 1 2
Lkin = (@µ A⌫ @ µ A⌫ @µ Aµ @⌫ A⌫ ) = F , (5.2.8)
2 4 µ⌫
where Fµ⌫ ⌘ @µ A⌫ @ ⌫ Aµ .
What we have just discovered is rather remarkable: A massless spin-1 field must satisfy the
Maxwell Lagrangian
1 2
LMaxwell = F , (5.2.9)
4 µ⌫
if we want the theory to be ghost-free.
Notice that the field strength Fµ⌫ , and hence the Lagrangian (5.2.9), is invariant under the
transformation
Aµ (x) ! Aµ (x) + @µ ↵ , (5.2.10)
for any function ↵(x). This is called a gauge symmetry of the theory. Field configurations Aµ (x)
that di↵er by the derivative of a scalar function are physically equivalent. As we will now show,
 5.2 Rediscovering Maxwell 91

the gauge symmetry removes two degrees of freedom from the vector field Aµ , leaving precisely
the two degrees of freedom of a massless spin-1 field.
The equation of motion associated with (5.2.9) is

2Aµ @µ (@⌫ A⌫ ) = 0 . (5.2.11)

It is useful to separate this into the 0 and i components:

@j2 A0 + @0 @j Aj = 0 , (5.2.12)
2Ai @i (@0 A0 @j Aj ) = 0 . (5.2.13)

Under the gauge symmetry (5.2.10), we have @j Aj ! @j Aj +@j2 ↵. This allows us to set @j Aj = 0.
This choice is called Coulomb gauge and removes one degree of freedom. The equation of motion
(5.2.12) then becomes
@j2 A0 = 0 . (5.2.14)
The field component A0 transforms as A0 ! A0 + @0 ↵. Note that this doesn’t change the
equation of motion (5.2.14), as long as @j2 ↵ = 0, so that we remain in Coulomb gauge. We can
use this residual gauge freedom to set A0 ⌘ 0. The equation of motion for Ai then reduces to

2Ai = 0 , @i Ai = 0 . (5.2.15)

This describes the two degrees of freedom associated with a massless spin-1 particle.

Massive spin-1 particles

A masssive spin-1 field must satisfy the Proca Lagrangian

1 2 1
LProca = Fµ⌫ + m2 A2µ , (5.2.16)
4 2

where the presence of a mass term does not change the fact that the kinetic has been uniquely
fixed by the requirement of the absence of a ghost. The corresponding equation of motion is

@ µ Fµ⌫ = m2 A⌫ . (5.2.17)

Acting on this with @ ⌫ , the left-hand side vanishes, @ ⌫ @ µ Fµ⌫ because the field strength Fµ⌫ is
anti-symmetric. The right-hand side then implies the following constraint

@ ⌫ A⌫ = 0 . (5.2.18)

This removes one degree of freedom from Aµ , so that the theory has the expected three prop-
agating degrees of freedom. This can also be seen by substituting Aµ ⌘ µ + m 1 @µ into
(5.2.16), which gives
1 1 1
LProca = (@µ Â⌫ )2 + m2 Â2µ + (@µ )2 . (5.2.19)
2 2 2
We see that the longitudinal mode has been revived by the mass term. A massive vector field
thus propagates three degrees of freedom: two transverse modes µ and a longitudinal mode .
 5.3 Gauge Symmetry 92

5.3 Gauge Symmetry

It is worth saying a few more words about the gauge symmetry of the Maxwell Lagrangian.
First of all, “gauge symmetry” is a bad term, since it isn’t really a symmetry at all, but rather
a redundancy of description. Ordinary symmetries map a physical state into another physical
state with the same properties. States that are related by gauge symmetries, on the other hand,
are to be identified: they are the same physical state.
It is often helpful to describe the physics by selecting a representative field configuration from
those related by the gauge symmetry. However, this procedure, called gauge fixing, is not unique.
Di↵erent gauge choices can be useful for di↵erent purposes. We will present the quantization of
the theory in two di↵erent gauges:
• Lorentz gauge: We can always pick a representative field configuration satisfying

@ µ Aµ = 0 . (5.3.20)

To see this, suppose that the gauge field A0µ (x) does not satisfy (5.3.20), but instead we
have @ µ A0µ = f (x), for some function f (x). Applying the gauge transformation (5.2.10)
leads to (5.3.20) if we can chose ↵(x) such that

@µ @ µ ↵ = f. (5.3.21)

This equation is guaranteed to have a solution, so we can always impose the Lorentz
gauge condition (5.3.20). In fact, the constraint (5.3.21) still does not pick a unique
gauge, because we are always free to make further gauge transformations with

@µ @ µ ↵ = 0 , (5.3.22)

which also has non-trivial solutions. This residual gauge symmetry can sometimes lead
to confusions. As we will see, it leads to subtleties in the quantization of the theory. On
the other hand, an advantage of working in Lorentz gauge is that it is manifestly Lorentz
invariant.

• Coulomb gauge: The residual gauge transformations (5.3.22) allow us to set r · A = 0.

By the equations of motion, this implies

A0 = 0 . (5.3.23)

A disadvantage of working on Coulomb gauge is that it breaks Lorentz invariance. Al-

though the final answer for any physical observables will be Lorentz invariant, this will
not be obvious at intermediate steps in the calculation. On the other hand, the physical
degrees of freedom are most transparent in Coulomb gauge: the three components of the
vector A satisfy r · A = 0, which leaves two physical degrees of freedom which after
quantization become the two polarization states of the photon.

5.4 Quantization⇤
We will present the quantization of the Maxwell Lagrangian twice: first in Coulomb gauge and
then in Lorentz gauge. We will get the same answers, but in each case we will face di↵erent
subtleties.
 5.4 Quantization⇤ 93

To prepare for the quantization procedure, we derive the momentum ⇧µ conjugate to Aµ :

@L
⇧0 = = 0, (5.4.24)
@ Ȧ0
@L
⇧i = = F 0i = E i . (5.4.25)
@ Ȧi
The fact that ⇧0 vanishes is consistent with the fact that A0 is a non-dynamical field, but it
highlights the problem of maintaining covariance in the quantization procedure. The remaining
components of the momentum are simply the electric field E i . We derive the Hamiltonian in
the standard way
Z ⇣ ⌘
H = d3 x ⇧i Ȧi L
Z 
1
= d x (Ei2 + Bi2 ) A0 (r · E) ,
3
(5.4.26)
2
where Ei ⌘ @0 Ai @i A0 and Bi ⌘ ✏ijk @j Ak . This agrees with the conserved energy derived via
Noether’s theorem in (1.2.48). We see that A0 plays the role of a Lagrange multiplier imposing
Gauss’ law, r · E = 0.

Coulomb gauge
In §5.2, we showed that the Maxwell equation in Coulomb gauge becomes

2A = 0 , (5.4.27)

subject to the constraint r · A = 0. Solutions to this equation can be written as

ipx
A(x) = ✏(p) e , (5.4.28)

which is analogous to the positive frequency spinor solutions discussed in the previous chapter,
cf. eq. (4.3.85). In order for the gauge condition r · A = 0 to hold, the polarization vectors
✏(p) must satisfy p · ✏ = 0, i.e. they are transverse polarizations. For p = (0, 0, p), the two
independent polarization vectors can be chosen to be ✏1 = p12 (1, +i, 0) and ✏2 = p12 (1, i, 0).
The mode expansion of the field operator and its conjugate momentum therefore are
Z
d3 p 1 Xh +ip·x ⇤ † ip·x
i
A(x) = p ✏ (p) a p e + ✏ (p) a p e , (5.4.29)
(2⇡)3 2Ep
Z r
d3 p Ep X h +ip·x ⇤ † ip·x
i
E(x) = ( i) ✏ (p) a p e ✏ (p) a p e . (5.4.30)
(2⇡)3 2

where the sum runs over the two polarizations = 1, 2. We impose the usual commutation
relations on the operators ap and ap† :
⇥ 0 ⇤ 0
ap , aq † = (2⇡)3 (3) (p q) , (5.4.31)
⇥ 0⇤ ⇥ † 0†⇤
a p , aq = a p , aq = 0 . (5.4.32)

Note that this implies a somewhat unusual commutation relation for Ai and ⇧i = E i , namely
Z ✓ ◆
i j d3 p ip·(x y) ij pi pj ij
[A (t, x), E (t, y)] = i e ⌘ i ? (x y) , (5.4.33)
(2⇡)3 |p|2
 5.4 Quantization⇤ 94

ij
where ? (x y) is called the “transverse” Dirac delta function. This would be an ordinary
delta function if we didn’t have the pi pj /p2 term in the integrand. To see that this extra term is
necessary, let us take the divergence of both sides of (5.4.33) with respect to x. On the left-hand
side, we find [r · A(t, x), E(t, y)] which vanishes because r · A = 0 in Coulomb gauge. On the
right-hand side, the integrand becomes pi ( ij pi pj /p2 ) = pj pj = 0. We see that without
the pi pj /p2 term in the integrand the commutator would not be consistent with the constraint
r · A = 0. Taking the divergence with respect to y would give [A(t, x), r · E(t, y)] on the
left-hand side, which also has to vanish because of Gauss’ law r · E = 0.

Exercise.—Prove the following identity

X p i pj
✏i (p)✏j (p) = ij
. (5.4.34)
|p|2

Use it to show that (5.4.31) and (5.4.32) imply (5.4.33).

To find the Hamiltonian operator, we substitute the mode expansions (5.4.29) and (5.4.30)
into (5.4.26). After normal ordering, this gives
Z
d3 p X
H= Ep ap† ap . (5.4.35)
(2⇡)3

Similarly, the momentum operator becomes

Z
d3 p X
P= p ap† ap , (5.4.36)
(2⇡)3

where |p| = Ep . This shows that a†p, |0i creates particles with momentum p and energy Ep =
|p|. With a bit more work, we could also show that these particles carry spin 1 and helicity
= ±1. We have discovered photons!
Finally, let us consider the Feynman propagator for the fields Ai (x):
ij
? (x y) ⌘ h0|T {Ai (x)Aj (y)}|0i . (5.4.37)

which sometimes is referred to as the “transverse” propagator. Substituting the mode expansion
(5.4.29) into (5.4.37), we find
Z ✓ ◆
ij d4 p i ij pi pj
? (x y) = e ip·(x y) . (5.4.38)
(2⇡)4 p2 + i✏ |p|2

The lack of manifest Lorentz symmetry2 of this form of the propagator is the price that we have
to pay for working in Coulomb gauge. However, in §5.5, we will see that this propagator can be
massaged into a much nicer, and more manifestly Lorentz-invariant form, once we couple the
theory to matter.
2
To prove that the theory secretly is still Lorentz invariant, we could express all generators of the Poincaré
group in terms of the creation and annihilation operators (we already did this explicitly for the four-momentum).
Using the commutation relations for the creation and annihilation operators, we could then show that these
generators indeed satisfy the Lorentz algebra. This proves covariance of the quantization in Coulomb gauge.
 5.4 Quantization⇤ 95

Lorentz gauge
To maintain Lorentz invariance throughout the quantization procedure, we must be able to treat
all components of Aµ (x) on an equal footing and impose the following commutation relations

[Aµ (t, x), ⇧⌫ (t, y)] = i⌘ µ⌫ (3)

(x y) , (5.4.39)
[Aµ (t, x), A⌫ (t, y)] = [⇧µ (t, x), ⇧⌫ (t, y)] = 0 . (5.4.40)

However, this is in conflict with the fact that we have ⇧0 = 0 for the Maxwell Lagrangian,
cf. eq. (5.4.24). The solution will be to modify the Lagrangian in such a way that ⇧0 6= 0. This
procedure will introduce spurious degrees of freedom that in the end have to be removed by
hand from the spectrum of states.
Consider the following Lagrangian
1 2 ⇠
L0 = F (@µ Aµ )2 , (5.4.41)
4 µ⌫ 2
where ⇠ is a parameter that, in principle, can be chosen freely. Classically, the Lagrangian L0
reduces to the Maxwell Lagrangian if we impose the Lorentz gauge condition @µ Aµ = 0. The
extra contribution in (5.4.41) is called a gauge-fixing term. It plays the role of a Lagrange
multiplier that imposed the Lorentz constraint on the field. We will see, however, the quantum-
mechanically, the Lorentz condition cannot hold as a operator statement, i.e. we are forced
to @µ µ (x) 6= 0. Instead, we will impose a weaker condition on the physical states, namely,
h |@µ µ (x)| i = 0.
The conjugate momentum now is
@L0
⇧µ = = F 0µ ⇠⌘ 0µ (@⌫ A⌫ ) . (5.4.42)
@ Ȧµ
In particular, the field A0 has the conjugate partner ⇧0 = ⇠(@⌫ A⌫ ). The equation of motion
corresponding to L0 is
2Aµ (1 ⇠)@ µ (@⌫ A⌫ ) = 0 . (5.4.43)
Classically, this is equivalent to the Maxwell equation in Lorentz gauge, i.e. 2Aµ = 0, with
@µ Aµ = 0.
In the following, we will specialize to the case ⇠ = 1 (sometimes confusingly called Feynman
gauge). The Lagrangian can then be brought into a particularly simple form after integration
by parts in the action integral:
1 1 1
L0 = @µ A⌫ @ µ A⌫ + @µ A⌫ @ ⌫ Aµ (@µ Aµ )(@⌫ A⌫ )
2 2 2
1 1
= @µ A⌫ @ µ A⌫ + @µ [A⌫ (@ ⌫ Aµ ) (@⌫ A⌫ )Aµ ] . (5.4.44)
2 2
The last term is a total divergence and therefore does not contribute to the equation of motion.
The dynamics can therefore be described by
1
L00 = @ µ A⌫ @ µ A⌫ , (5.4.45)
2
and the conjugate momentum is
@L00
⇧µ = = Ȧµ . (5.4.46)
@ Ȧµ
 5.4 Quantization⇤ 96

It is now manifest that the time and space components of the field Aµ appear on an equal
footing. The Hamiltonian density is
1 µ 1
H00 = ⇧µ Ȧµ L00 = ⇧ ⇧µ + @i Aµ @ i Aµ . (5.4.47)
2 2
Written out in components, this becomes
1h 0 2 i 1h i
H00 = (Ȧ ) + (rA0 )2 + (Ȧi )2 + (rAi )2 . (5.4.48)
2 2
Notice the wrong sign for the scalar part A0 ! The Hamiltonian isn’t positive definite. This
shouldn’t come as a surprise. In §5.2, we worked hard to show that the Maxwell Lagrangian
is the unique Lagrangian for a massless vector field with positive definite Hamiltonian. By
adding the extra term in (5.4.41) we undid all that good work. As we will see, imposing the
gauge condition will fix the problem and ensure that the Hamiltonian of the physical degrees of
freedom is positive definite. The ghost degree of freedom in (5.4.48) will be removed from the
spectrum of allowed states.
Turning the classical fields into operators, we impose the commutation relation (5.4.39), or

[Aµ (t, x), Ȧ⌫ (t, y)] = i⌘ µ⌫ (3)

(x y) . (5.4.49)

For the spatial components Ai these just look like three copies of the commutation relations for
a scalar field, [ µ (t, x), ˙ (t, y)] = i (3) (x y). The commutation relation for A0 , on the other
hand, has the wrong sign! This is unavoidable if the commutation relation is written in covariant
form: the right-hand side must contain the metric tensor ⌘ µ⌫ which contains both signs. States
associates with the operator A0 will have negative norm. It is a good thing that we already
decided that we will get ride of these states eventually, i.e. once the gauge condition is imposed.
Fearlessly, we march on. Expanding the operator Aµ (x) in plane wave solutions, we get
Z
1 Xh µ i
3
d3 p
µ
A (x) = p ✏ (p) ap e ip·x
+ ✏µ⇤ (p) ap† eip·x . (5.4.50)
(2⇡)3 2Ep
=0

This time we have four polarization vectors ✏µ (p) instead of the two polarization vectors we
encountered in Coulomb gauge. This is because we are working with all four components of the
redundant field Aµ and haven’t yet imposed a gauge condition. We choose ✏µ0 to be timelike and
✏µ1,2,3 to be spacelike. We select ✏µ1 and ✏µ2 as the two transverse polarizations

pµ ✏µ1 = pµ ✏µ2 = 0 . (5.4.51)

The polarization vector ✏µ3 is then longitudinal. Finally, we normalize the polarization vectors,
so that
⌘µ⌫ ✏µ ✏⌫ 0 = ⌘ 0 . (5.4.52)
Substituting (5.4.50) into (5.4.49), we get
⇥ 0 ⇤ 0
a p , aq † = ⌘ (2⇡)3 (3)
(p q) . (5.4.53)

As we already anticipated, the signs are weird. For = 1, 2, 3, everything looks fine, but for
= 0, we get the wrong sign
⇥ 0 ⇤ 0
a p , aq † = (2⇡)3 (3) (p q) , , 0 = 1, 2, 3,
⇥ 0 0† ⇤ (5.4.54)
ap , aq = (2⇡)3 (3) (p q) .
 5.4 Quantization⇤ 97

Naively, this spells disaster for the stability of the theory. Consider the vacuum and one-particle
excitations defined in the usual way

ap |0i = 0 ,
(5.4.55)
|p, i ⌘ ap† |0i .

For states with timelike polarization, we then get

hp, 0|p, 0i = h0|a0p a0†

p |0i = (2⇡)3 (3)
(0) . (5.4.56)

What is the negative sign doing there? A Hilbert space with negative norm means negative
probability which doesn’t make any sense.

Gupta-Bleuler resolution
Imposing the Lorentz condition @µ Aµ = 0 will come to our rescue. How this is done, however,
is a bit subtle. Asking for the operator Aµ to satisfy @µ Aµ = 0 would not be consistent with
the commutation relations. Its the old issue that ⇧0 = (@µ Aµ ) must be non-zero. A weaker
condition would be to impose
@ µ Aµ | i = 0 , (5.4.57)
and hope that the states that satisfy this would exclude the negative norm states. Unfor-
tunately, this also doesn’t work. To see this, let us split the operator Aµ (x) into positive and
negative frequency more, i.e. Aµ (x) = A+ +
µ (x) + Aµ (x), where Aµ (x)|0i = 0. The problem is
then apparent: since @ µ Aµ |0i =
6 0, not even the vacuum would be a physically allowed state if
we impose the condition (5.4.57). To keep the vacuum alive, we try the so-called Gupta-Bleuler
condition
@ µ A+
µ| i = 0, (5.4.58)
Note that this implies h |@ µ A+ µ
µ = 0, so that the operator @µ A has vanishing matrix elements
between physical states
h 0 |@µ Aµ | i = 0 . (5.4.59)
Unfortunately, even the Gupta-Bleuler condition doesn’t remove the negative norm states from
the physical Hilbert space. But it is close, so let’s not give up prematurely.
Let us split the states of the Hilbert space into states | ? i containing only transverse photons
(created by a1† 2†
p and ap ) and states | i including both timelike photons (created by ap ) and
0†
3†
the longitudinal photons (created by ap ). The Gupta-Bleuler condition (5.4.58) requires that

(a3p a0p )| i = 0 . (5.4.60)

In words, physical states must contain combinations of timelike and longitudinal photons: a
state with a timelike photon of momentum p, must also contain a longitudinal photon with the
same momentum. A general state | i will be a linear combination of states n with n pairs of
timelike and longitudinal photons:
1
X
| i= cn | n i , (5.4.61)
n=0

where | 0i = |0i. For n > 0, the states | ni have zero norm

h n| i= 0n 0m . (5.4.62)
 5.5 Photon Propagator 98

This is progress, since the inner product of all states is now positive semi-definite. However,
nobody has ever seen timelike or longitudinal photons, so what are we to make of the state
zero-norm states | n i? The answer is that they are gauge equivalent to the vacuum state. In
other words, two states that di↵er only by | i are physically equivalent, i.e. | ? i + | i ⇠ | ? i.
This means that we will get the same expectation values for all physical observables if we simply
ignore the states | i. For example, consider the Hamiltonian
Z 3
!
d3 p X
† 0† 0
H= |p| ap ap ap ap . (5.4.63)
(2⇡)3
=1

The Gupta-Bleuler condition (5.4.60) implies h |a3† 3 0† 0

p ap | i = h |ap ap | i, so that the con-
tribution from the timelike and longitudinal photons cancel in the expectation value of the
Hamiltonian. This also makes the Hamiltonian positive definite. Life is good again.

5.5 Photon Propagator

The Feynman propagator of the vector field Aµ is defined as
µ⌫
F (x y) ⌘ h0|T {Aµ (x)A⌫ (y)}|0i . (5.5.64)

In this section, we compute this propagator in both Lorentz and Coloumb gauge.

Lorentz gauge
Direct substitution of the mode expansions (5.4.50) into (5.5.64) gives
Z
µ⌫ d4 p i⌘ µ⌫ ip·(x y)
F (x y) = e , (5.5.65)
(2⇡)4 p2 + i✏

or, in momentum space,

µ⌫ i⌘ µ⌫
F (p) = . (5.5.66)
p2 + i✏
Unlike the propagator in Coulomb gauge, this form of the propagator is manifestly Lorentz
invariant.

Coulomb gauge⇤
In §6.1, we will see that gauge invariance demands that the coupling between light and matter is
through couplings of the vector potential Aµ to conserved currents J µ . Including this interaction
into the Maxwell Lagrangian, we obtain
1 2
L= F Aµ J µ + · · · , (5.5.67)
4 µ⌫
where J µ is some function of the matter fields (to be discussed in the next chapter) and the
ellipses denote additional terms that characterize the dynamics of the matter fields (e.g. the
Dirac Lagrangian for fermions). Let us use this Lagrangian to show that the ugly looking
transverse propagator in Coulomb gauge, cf. eq. (5.4.38), can be written in a nicer form.
 5.5 Photon Propagator 99

We first note that, in the Coulomb gauge, the Lagrangian (5.5.67) becomes
1 1 1
L = Ȧ2i (@j Ai )2 A0 r2 A0 A0 J 0 + Ai J i . (5.5.68)
2 2 2
Unlike before, we are not allowed to set A0 = 0 in the interacting theory. Instead, the equation
of motion implies Z
2 0 J 0 (t, y)
r A0 = J ) A0 (t, x) = d3 y . (5.5.69)
4⇡|x y|
Since A0 is non-dynamical, we can plug this back into the action to get
1 1
L = (@µ Ai )2 + Ai J i A0 J 0
2 2
Z
1 1 J 0 (t, x)J 0 (t, y)
= (@µ Ai )2 + Ai J i + d3 y . (5.5.70)
2 2 4⇡|x y|

The last there is called the Coulomb term. It looks nonlocal, but this is an artefact of working
in Coulomb gauge and will not show up in physical observables.
We could capture the Coulomb term by defining the following propagator for the field A0 (x):3
Z
00 (x0 y 0 ) d4 p e ip·(x y)
F (x y) = = . (5.5.71)
4⇡|x y| (2⇡)4 |p|2

This can be combined with the propagator for the field Ai (x), cf. eq. (5.4.38), by writing
8 i
>
> µ=⌫=0
>
> |p|2
>
>
>
< ✓ ◆
µ⌫
(p) = i ij pi pj (5.5.72)
F > µ = i, ⌫ = j
>
> p2 + i✏ |p|2
>
>
>
>
:
0 otherwise

or, more compactly, ✓ ◆

µ⌫ i p2 µ ⌫
F (p) = 2 P?µ⌫ (p) + n n , (5.5.73)
p + i✏ |p|2
where P?0⌫ ⌘ 0, P?ij ⌘ ij pi pj /|p|2 , and nµ ⌘ (1, 0, 0, 0). To massage this further, let us define
p̃µ as the vector consisting of only the spatial components of pµ , i.e. p̃µ ⌘ (0, pi ). We can then
write (5.5.73) as
✓ ◆
µ⌫ i µ⌫ µ ⌫ p̃µ p̃⌫ (p0 )2 |p|2 µ ⌫
F (p) = 2 ⌘ +n n + n n
p + i✏ |p|2 |p|2
✓ ◆
i µ⌫ p̃µ p̃⌫ (p0 nµ )(p0 n⌫ )
= 2 ⌘ +
p + i✏ |p|2 |p|2
✓ µ ⌫ µ
◆
i µ⌫ p̃ p̃ (p p̃µ )(p⌫ p̃⌫ )
= 2 ⌘ +
p + i✏ |p|2 |p|2
✓ ◆
i µ⌫ pµ p⌫ pµ p̃⌫ p̃µ p⌫
= 2 ⌘ + . (5.5.74)
p + i✏ |p|2
3
The field A0 mediates the Coulomb interaction and sometime is referred to as the Coulomb photon.
 5.5 Photon Propagator 100

The crucial feature of the second term in (5.5.74) is that it is proportional to pµ , so it doesn’t
contribute when contracted with a conserved current:
µ⌫ i ⇤
Jµ⇤ (p) F (p)J⌫ (p) = J (p)J µ (p) . (5.5.75)
p2 µ
As we will see, physical observables will always involve these types of contractions, so without
loss of generality we can work with the simpler propagator

µ⌫ i⌘ µ⌫
F (p) = . (5.5.76)
p2 + i✏

Of course, this is the same propagator we found more directly in Lorentz gauge.
 5.6 Problems 101

5.6 Problems
1. A natural guess for the Lagrangian of a massive spin-1 field would be

? 1 1
L= @⌫ Aµ @ ⌫ Aµ + m2 A2µ ,
2 2
where A2µ ⌘ Aµ Aµ . Show that this would lead to a Hamiltonian that is unbounded from
below. In the quantum theory this would lead to a catastrophic instability.
Then, try the Lagrangian

? 1 b 1
L= @⌫ Aµ @ ⌫ Aµ + Aµ @µ @⌫ A⌫ + m2 A2µ ,
2 2 2
where b is an arbitrary constant. Determine the value of b for which this theory propagates
exactly three degrees of freedom. Show that the corresponding Hamiltonian is bounded
from below.

2. Consider the Proca Lagrangian for a massive vector field

1 2 1
LProca = Fµ⌫ + m2 A2µ .
4 2
i) Verify that this theory is not gauge invariant and show that the equations of motion
derived from this Lagrangian are

(2 + m2 )Aµ = 0 , @ µ Aµ = 0 .

ii) Perform the canonical quantization of the Proca theory and verify that it leads to
massive particles of spin 1.