Lorentz Covariance in the Coulomb

Gauge

In QED, the Coulomb gauge is specified by ∂i Ai = 0, A0 = 0 for Aµ the

Maxwell potential. Under a Lorentz transformation Λ with infinitesimal
parameter ,
∂
U ()Aµ (x)U −1 () = Aµ (x) − µν Aν (x) + λ(x, )
∂xµ

for some operator gauge function λ(x, ).∗

To see this, we first note that for A(x) a solution of ∂µ F µν = 0, there
exists A0 (x0 ) a solution to the Lorentz transformed system. However, Aµ does
not quite transform as a 4-vector under a Lorentz transformation and we
must also consider a supplementary gauge term which admits the ∂ µ λ(x, ).
Aµ (x) is an operator in some Hilbert space H with a unitary representation
U of the Lorentz group. Thus,
Λ
Aµ (x) −→ A0µ (x0 ) = U ()Aµ (x)U −1 ().

So,

A →A0 = (1 − )A + [gauge terms],

Aµ →A0µ (x0 ) = Aµ (x) − µ ν Aν (x) + ∂ µ λ(x, )

and similarly,
x → x0 = (1 − )x.
Note that by Lorentz invariance of the metric g µν, µν = −νµ is an antisym-
metric tensor. This is because since we demand invariance, g = g 0 ⇒ g = 0
necessarily. This determines the antisymmetric form of  from the symmetry
of g.
∗
Bjorken & Drell associate primed fields A0 with x and unprimed fields A with x0 . This
is counter-intuitive, so I’ve reversed the notation. This explains the sign differences in, for
example, B&D 11.51.

1
 Any field φ transforms under a Lorentz transformation Λ as
φ(x) → φ0 (x0 ) = ρ(Λ)φ(Λ−1 x0 ). (‡)
We are therefore interested in the inverse transformation for x.
∞
(1 − )−1 = k = 1 +  + 2 + · · ·
X

k=0
= 1 +  + O(2 ) ≈ 1 + .
∴ x = (1 + )x0 .
In component form, xκ = x0κ + κ σ x0σ.
As A0 = 0, its Lorentz transformation must also be zero. (Impossible to
boost into a non-zero reference frame, etc.) Taking this transformation,
!
A00 (x0 ) = A0 (x) − 0 ν Aν (x) + ∂ 0 λ(x, ) = 0.
This implies
∂ 0 λ = 0 ν Aν
= 0 i Ai .
Looking at the Lorentz transformation of ∂i Ai ,
h i
∂i0 A0i (x0 ) = ∂i0 Ai (x) − i ν Aν (x) + ∂ i λ . (♥)
d dxκ
Recall that d
dx0i
= dxκ dx0i
, so we must find the Jacobian terms.†
∂ κ ∂ h 0κ i
x = x + κ
σ x 0σ
∂x0i ∂x0i
∂x0κ κ ∂x
0σ
= +  σ
∂x0i ∂x0i
= δi +  σ δi .
κ κ σ

We can now write ♥ as

h i 
∂i0 A0i (x0 ) = ∂κ Ai (x) − i ν Aν (x) + ∂ i λ δiκ + κ σ δiσ
= δiκ ∂κ Ai (x) − δiκ i ν ∂κ Aν (x) + δiκ ∂κ ∂ i λ
+ κ σ δiσ ∂κ Ai (x) − O(2 ) + κ σ δiσ ∂κ ∂ i λ
= ∂i Ai (x) − i ν ∂i Aν (x) + ∂i ∂ i λ + κ i ∂κ Ai (x) + κ i ∂κ ∂ i λ.
Introducing some awful notation, we’ll define the difference between the
Lorentz transformed ∂i Ai and the original by
†
This is kind of weird. We can make sense of it by not thinking about (‡) and noticing
that Λ : A → A0 = (1 − )A + · · · , so the argument of A must invert this, x0 → x = (1 + )x0 .

2
 ∆∂A ≡ ∂i0 A0i (x0 ) − ∂i Ai (x)
= −i ν ∂i Aν (x) + ∂i ∂ i λ +κ i ∂κ Ai (x) + κ i ∂κ ∂ i λ
| {z } | {z }
−4λ O(2 )

= −i ν ∂i Aν (x) + κ i ∂κ Ai (x) − 4λ

= κ i ∂κ Ai (x) − i ν ∂i Aν (x) − 4λ

by the definition of the Laplacian 4 ≡ −∂i ∂ i . Decomposing κ, ν into space

and time,

= 0 i ∂0 Ai (x) + j i ∂j Ai (x) − i 0 ∂i A0 (x) −i j ∂i Aj (x) − 4λ

| {z }
0
= 0 i ∂0 Ai (x) + j i ∂j Ai (x) − i j ∂i Aj (x) −4λ.
| {z }
a

Examining the a-term,

a = ji ∂j Ai − ij ∂i Aj
 
= ji ∂j Ai + ∂i Aj = 0

by symmetry. As ∆∂A = 0, this implies

0 i ∂0 Ai = 4λ. (l)

Choose 4λ(x, ) = φ(x, ). Then φ = 0 i ∂0 Ai (x) and

Z
λ= d3 x̄ G(x − x̄)φ(x̄)

where G is the Green function such that 4G(x) ≡ δ (3) (x). Several Fourier
transformations later (cf. Nigel Buttimore’s CED course), we arrive at the
rather Coulombic result
1 √
G(x) = − , r = |x| = x2
4πr
and G is independent of t. By l,
Z
0 i ∂0 Ai (t, x̄)
λ(t, x) = − d3 x̄ .
4π|x − x̄|

Corrections to fionnf@maths.tcd.ie.