SciPost Phys.

16, 142 (2024)

Asymptotic structure of higher dimensional Yang-Mills theory

Temple He1 and Prahar Mitra2

1 Walter Burke Institute for Theoretical Physics, California Institute of Technology,

Pasadena, CA 91125 USA
2 Department of Applied Mathematics and Theoretical Physics,
University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, UK

Abstract
Using the covariant phase space formalism, we construct the phase space for non-Abelian
gauge theories in (d + 2)-dimensional Minkowski spacetime for any d ≥ 2, including the
edge modes that symplectically pair to the low energy degrees of freedom of the gauge
field. Despite the fact that the symplectic form in odd and even-dimensional spacetimes
appear ostensibly different, we demonstrate that both cases can be treated in a unified
manner by utilizing the shadow transform. Upon quantization, we recover the algebra
of the vacuum sector of the Hilbert space and derive a Ward identity that implies the
leading soft gluon theorem in (d + 2)-dimensional spacetime.

Copyright T. He and P. Mitra. Received 29-06-2023

This work is licensed under the Creative Commons Accepted 07-05-2024
Check for
Attribution 4.0 International License. Published 30-05-2024 updates

Published by the SciPost Foundation. doi:10.21468/SciPostPhys.16.5.142

Contents
1 Introduction 2

2 Asymptotic phase space 3

2.1 Covariant phase space 3
2.2 Symplectic form on I ± 5

3 Asymptotic Hilbert space 8

3.1 Quantum commutators 8
3.2 Canonical quantization 9

4 Ward identity and the leading soft gluon theorem 11

4.1 Ward identity 11
4.2 Leading soft gluon theorem 12

A Definitions and conventions 13

A.1 Differential geometry 14
A.1.1 Spacetime manifold 14
A.1.2 Phase space manifold 16

B Radial electric field as shadow transform 16

References 19

1
 SciPost Phys. 16, 142 (2024)

1 Introduction
In recent years, the discovery of the equivalence between asymptotic symmetries, soft theo-
rems, and the memory effect has led to a resurgence in the study of the infrared sector of quan-
tum field theories (QFTs) (for a review, see [1], and references therein). This triangle of equiv-
alence, dubbed the infrared triangle, was first shown in the context of four-dimensional gravity
in asymptotically flat spacetimes [2–4], where it was established that the leading soft graviton
theorem [5] is the Ward identity for BMS supertranslations [6, 7], which in turn is associated
to a gravitational memory effect [8–11]. This was later extended to four-dimensional gauge
theories [12–18], as well as higher even-dimensional gravity and gauge theories [19–21]. Fur-
thermore, beyond the leading soft theorems, there has also been extensive research connecting
the subleading (and sub-subleading) soft theorems to asymptotic symmetries and memory ef-
fects, e.g. see [22–28].
Although soft theorems exist in all spacetime dimensions, it was only recently that the con-
nection between soft theorems and asymptotic symmetries was established in odd spacetime
dimensions. The main technical challenge was due to the qualitatively different properties of
massless wave propagation in odd and even-dimensional spacetimes. This difference is often
referred to as the failure of Huygens’ principle in odd-dimensional spacetimes. For the case of
massless gauge theories, this difficulty was surmounted in [29–31], where it was demonstrated
that the charge generating large gauge transforms (LGTs) in odd-dimensional spacetime has
a somewhat different form from those in even-dimensional spacetime.
Meanwhile, in order to study the vacuum sector of the Hilbert space in gauge theories, an
analysis was carried out in [32] using the covariant phase space formalism to construct the
symplectic form of four-dimensional non-Abelian gauge theories. This allowed for a deriva-
tion of the Dirac brackets pertaining to the soft modes, i.e. the soft gluon mode N and its
symplectic partner C, which is the gauge field edge mode. Upon canonical quantization, this
led unambiguously to the algebra of soft operators in the vacuum Hilbert space, and the corre-
sponding Ward identity was shown to give rise to the leading soft gluon theorem. In addition,
it was identified that the charge generating LGTs are canonical transformations preserving the
symplectic form.
The fact that the large gauge charge takes on different forms in odd and even spacetime
dimensions suggests that it might be difficult to extend the analysis of [32] in a uniform man-
ner to both odd and even-dimensional non-Abelian gauge theories. However, inspired by the
analysis of [33], where it was shown that the soft effective action of gravity and gauge theories
in any dimension does not involve the soft operator, but rather its shadow transform, we are
led to wonder if it is perhaps more natural to write the symplectic form not in terms of the
soft gluon operator N , but rather its shadow transform N e . We demonstrate in this paper that
this is indeed the case. By writing the symplectic form in terms of the shadow transform of
the soft operator, all differences between odd and even dimensions disappear, leading us to
a uniform treatment of both odd and even-dimensional non-Abelian gauge theories via the
covariant phase space formalism. This extends the analysis using covariant phase space for-
malism initiated in [32] to theories with dimensions greater than four, and also confirms the
results of [34] pertaining to non-Abelian gauge theories.
Because this paper is a direct extension of the analysis performed in [32], we will often-
times neglect details and refer the reader to [32] for a more in-depth analysis and treatment.
In Section 2, we use the covariant phase space formalism to construct the symplectic form for
all (d + 2)-dimensional theories for d ≥ 2. In Section 3, we canonically quantize the phase
space and obtain the quantum commutators by inverting the symplectic form, and then con-
struct the Hilbert space, including both the vacuum sector as well as the radiative Fock space.
Finally, in Section 4, we write down the Ward identity associated with LGTs and use it to derive
the leading soft gluon theorem.

2
 SciPost Phys. 16, 142 (2024)

2 Asymptotic phase space

In this paper, we use the covariant phase space formalism to construct the symplectic form
for non-Abelian gauge theories in d + 2 dimensions. The notations and conventions employed
in this paper are those used in [32]. For completeness, we review some of them in Appendix
A and introduce others as needed in the main text, but for more details, we refer the reader
to [32].

2.1 Covariant phase space

Consider a non-Abelian gauge theory in a (d + 2)-dimensional spacetime (M, g), where M is
the Lorentzian spacetime manifold and g the metric, with compact semi-simple gauge group
G and corresponding Lie algebra g. On (M, g), we set up a generalized coordinate sys-
tem {x µ }, and on g, we install a basis of generators T I , where I ∈ {1, . . . , |g|}, that satisfy
[T I , T J ] = f I J K T K . The dynamics of a non-Abelian gauge field A = AIµ (x) d x µ ⊗ T I in d + 2
dimensions is described minimally by a Lagrangian form

1
L= Tr[F ∧ ⋆F] , F = dA + A ∧ A , (1)
2g 2

where Tr is the trace in the adjoint representation normalized so that Tr[T I T J ] = −δ I J . More
generally, we could also add higher derivative terms, as well as terms involving other matter
fields, to the Lagrangian (1). This generalization was studied in great detail in [32], and as
those extra terms do not play a role in the structure of the phase space on I ± , we will safely
ignore them for the rest of our analysis.
The Lagrangian (1) is a (d + 2)-form in spacetime and a 0-form (function) in field config-
uration space F.1 A generic vector field X on F is given by

δ
Z
dd+2 x −g δAIµ (x) I
p
X≡ , (2)
M δAµ (x)

and the action of such a vector on L is

1  1

  ‹
d
X(L) = 2 Tr X(A) ∧ d ⋆ F + A ∧ ⋆F − (−1) ⋆ F ∧ A + d 2 Tr[X(A) ∧ ⋆F] . (3)
g g

The first term is used to define the solution space S, which is the subspace of F

S ≡ A ∈ F | d ⋆ F + A ∧ ⋆F − (−1)d ⋆ F ∧ A = 0 . (4)

In the rest of this paper, we will work exclusively on S, which is known as going on-shell.
In the covariant phase space formalism [32, 35–39], the pre-symplectic potential is con-
structed by integrating the second term in (3) over a Cauchy slice Σ, i.e.
Z
1
ΘΣ (X) = 2
e Tr[X(A) ∧ ⋆F] . (5)
g Σ

The pre-symplectic form is then constructed by taking an exterior derivative of this on S, and
is given by Z
1
ΩΣ (X, Y) = − 2
e Tr[X(A) ∧ ⋆Y(F) − Y(A) ∧ ⋆X(F)] . (6)
g Σ
1
This is the space of allowed field configurations of the gauge field A, which is defined by choosing appropriate
boundary conditions.

3
 SciPost Phys. 16, 142 (2024)

Without any specific choice of boundary conditions for the gauge field on ∂ Σ, the pre-
symplectic form generically depends on the choice of Σ. This can be fixed (as shown in [32])
by restricting to a solution space satisfying
F ∂ Σ = 0 =⇒ A ∂ Σ = C dC −1 , C ∈G. (7)
On this solution space, it is convenient to write the gauge field as2

A = C ĀC −1 + C dC −1 , Ā|∂ Σ = 0 . (8)

Substituting (8) into (5), we find that the symplectic form takes the form
Z
1
Ω(X, Y) = − 2
e Tr[X(Ā) ∧ ⋆Y(F̄) − Y(Ā) ∧ ⋆X(F̄)]
g Σ
I (9)
1
+ 2 Tr[X(C) ⋆ Y(F̄C −1 ) − Y(C) ⋆ X(F̄C −1 )] ,
g ∂Σ

where F̄ = dĀ + Ā ∧ Ā. Note that we have now dropped the Σ subscript on Ω e.
Given the pre-symplectic form (9), the phase space of the theory is given by Γ = S/ ker Ω e,
and the corresponding symplectic form is defined by Ω = Ω e |Γ . One class of elements of the
kernel is given by the vector field that generates infinitesimal gauge transformations3
δ
Z
Xϵ = − dd+2 x −g (Dµ ϵ) I I , Dµ ϵ ≡ ∂µ ϵ + [Aµ , ϵ] .
p
(10)
M δA µ

It can be verified via direct computation that the action of this vector on Ā and C is given by
Xϵ (Ā) = 0 , Xϵ (C) = ϵC . (11)

In particular, we see that Ā is gauge invariant. Using this, we find

I
1
Ω
e (Y, Xϵ ) = − Tr[ϵ ⋆ Y(F)] . (12)
g2 ∂ Σ

We see from this that if ϵ|∂ Σ = 0, then Xϵ lives in the kernel of Ω e . Therefore, these are
the small gauge transformations, and the phase space Γ is constructed by identifying all the
transformations generated by this vector. In practice, this is achieved by imposing a gauge
fixing condition f [Aµ ] = 0 that removes these redundancies.
On the other hand, we see from (12) that if ϵ|∂ Σ ̸= 0 and ϵ is field-independent, i.e.
Y(ϵ) = 0, then we can rewrite (12) as
I
1
 ‹
Ω
e (Y, Xϵ ) = Y − Tr[ϵ ⋆ F ] , (13)
g2 ∂ Σ
which means Xϵ is a Hamiltonian vector field that generates a canonical transformation. These
are known as large gauge transformations (LGTs), and the corresponding Hamiltonian charge
is I
1
Qϵ = − 2 Tr[ϵ ⋆ F] . (14)
g ∂Σ
Upon canonical quantization, the large gauge charge obeys the quantum commutation rela-
tion4
[Q ϵ , Q ϵ′ ] = −iΩ(Xϵ , Xϵ′ ) = iQ [ϵ,ϵ′ ] . (15)
2
Note that (7) defines C only on ∂ Σ. To decompose the gauge field as in (8), we need to smoothly extend C
into Σ. This extension is not unique, but as we will see from (19), the bulk extension of C does not enter the
symplectic form, so all extensions are physically equivalent.
3
Finite gauge transformations act on the gauge field as A → g A g −1 + g d g −1 , or equivalently Ā → Ā, C → gC.
4
The quantum commutator is related to the Poisson/Dirac bracket by a factor of i, such that [·, ·] = i{·, ·}.

4
 SciPost Phys. 16, 142 (2024)

2.2 Symplectic form on I ±

Thus far, our results have been rather general. We now focus on the case of interest. We
take M to be (d +2)-dimensional Minkowski spacetime and consider the symplectic form on
two distinct Cauchy slices Σ± = I ± ∪ i ± with boundaries ∂ Σ± = I∓± , where I + (I − ) is
the future (past) null infinity, i + (i − ) the future (past) timelike infinity, and I−+ (I+− ) the past
(future) boundary of I + (I − ). To describe these surfaces, we will work in flat null coordinates
(u, x a , r), which are related to the usual Cartesian coordinates via
 ‹
µ r 2 u 2 u
X = a
1 + x + , 2x , 1 − x − , x 2 = δab x a x b , (16)
2 r r
where µ ∈ {0, 1, · · · , d + 1} are the spacetime coordinates and a, b ∈ {1, · · · , d} are the trans-
verse spatial coordinates.5 Transverse spatial indices are raised and lowered by the Kronecker
delta δ a b and δa b , respectively. In flat null coordinates, Minkowski spacetime is then given by
the metric

ds2 = −du dr + r 2 δab dx a dx b . (17)

The asymptotic null boundaries I ± of Minkowski spacetime are located at r → ±∞ with

(u, x a ) fixed. Their respective past and future boundaries I∓± are located by further taking the
limit u → ∓∞. For more details, we refer the reader to Appendix A of [29].
With the identification Σ± = I ± ∪ i ± , the boundary condition (7) is then the condition

A|I−+ = A|I+− = CdC −1 , (18)

so the gauge field is flat and identified across spatial infinity i 0 .6 This is the matching con-
dition that has been imposed in previous works, e.g. [3, 4, 29, 30], and as argued in [32] is
sufficient for the symplectic form on I + to be the same as that on I − .7 This is equivalent to
the requirement that there is no flux leaving the system through spatial infinity i 0 .
To define the phase space with an invertible symplectic form, we need to fix the small
gauge symmetry. Following [32], we do this by setting Au = 0. The symplectic form (9) on Σ±
in flat null coordinates then works out to be
Z
1
Ω(X, Y) = 2 du dd x Tr X(±a )Y(∂u ±a ) − (X ↔ Y)
 
g
Z (19)
2  ± −1

d
∓ 2 d x Tr X(C)Y(E C ) − (X ↔ Y) ,
g
where  
d
±a ≡C lim |r| 2 −1 Āa C −1 ,
r→±∞
  1
Z (20)
±
 
E ≡ d
lim lim |r| ∂u Ā r ± du lim |r|d−2 Āa , ∂u Āa .
u∓∞ r→±∞ 2 r→±∞

Notice that even though the right-hand side of (19) appears to depend on I ± , the boundary
condition (18) ensures that the symplectic form Ω does not.
To simplify this further, we invoke the results of [29–31]. In those papers, the authors
decompose the gauge field into a radiative part and a Coulombic part so that
Āµ = ĀR± C±
µ + Āµ , (21)
5
We will throughout this paper use lowercase Greek indices to denote spacetime coordinates, and lowercase
Latin indices to denote transverse spatial coordinates.
6
Note that the coordinate x a describes anti-podal points on I + and I − , so (18) is actually an anti-podal match-
ing condition.
7
Thus, we do not need to include ± superscripts on the field C.

5
 SciPost Phys. 16, 142 (2024)

where the radiative part (R) satisfies the free Maxwell’s equations, whereas the Coulombic part
(C) is the inhomogeneous solution to Maxwell’s equations and describes the self-interaction
of the gauge field and, more generally, the interaction of the gauge field with the charged
matter fields. This is defined by integrating the charge current against a Green’s function. In
the above equation, the ± in the superscript does not imply we are taking the r → ±∞ limit,
but rather distinguishes the Green’s function that is used to extract the Coulombic part: + for
the advanced Green’s function and − for the retarded Green’s function. This choice ensures
that in the far future ĀC+ C− R+
µ → 0 and in the far past Āµ → 0. Consequently, Āµ describes the
outgoing radiative mode and ĀR− µ the incoming radiative mode. In the quantum theory, they
create the out and in one-particle gluon states respectively, and [30] showed that these gauge
fields fall off at large |r| as

− d2 −1
€ Š
+ O |r|−d , −d



ĀR±
r = O |r| ĀC±
r = O |r| , (22)
− d2 +1
€ Š
+ O |r|−d+1 , −d+1



ĀR±
a = O |r| ĀC±
a = O |r| . (23)

It is important to note that in [30], the authors considered the large r expansion of the gauge
fields in a frame where C = 1, so the expansions, in fact, apply to the gauge field Āµ , not Aµ .
We can now use (21) to simplify (20). First, it is clear from (23) that ±a is well-defined
and only receives a contribution from the radiative part of the gauge field, as the Coulombic
part falls off too quickly at large |r|. This means we can write
d
€ Š
±a = C lim |r| 2 −1 ĀR±
a C
−1
. (24)
r→±∞

Next, we decompose E ± into radiative modes, resulting in

Z
±
€ Š € Š 1  
E = lim lim |r| d
∂u ĀR±
r + lim lim |r| d
∂u ĀC±
r ± du lim |r|d−2 ĀR±a , ∂u ĀR±
a .
u∓∞ r→±∞ u∓∞ r→±∞ 2 r→±∞
(25)
Note that in the last term, we have only kept the radiative contribution since the Coulombic
part again falls off too quickly at large |r|. To further simplify this, we use (4.7) and (4.8)
of [30], which in our notation is given by8

2 lim |r|d ∂u2 ĀC±

 
r = lim |r| d−2 R±a
Ā , ∂u ĀR±
a . (26)
r→±∞ r→±∞

±
Integrating over u and using the fact that AC±
r vanishes on I± , we find
Z
d 1  
lim lim |r| ∂u ĀC±
r =∓ du lim |r|d−2 ĀR±a , ∂u ĀR±
a . (27)
u→∓∞ r→±∞ 2 r→±∞

Using this, we find that the second line in (25) cancels exactly and we have

E ± = lim lim |r|d ∂u ĀR±

r . (28)
u∓∞ r→±∞
8
The exact form of (4.7), given (4.8), in [30] is
(R±, d −1)
h d
i
(C±,d)
2∂u Fur = Aa(R±, 2 −1) , Fua 2 ,

where we dropped the matter current contribution since we are considering a pure Yang-Mills theory. The notation
used in that paper translates to one used here as
(R±, d2 −1) d
€ Š € Š
(C±,d)
Fur =C lim |r|d ∂u ĀC±
r
C −1 , Aa =C lim |r| 2 −1 ĀR±
a
C −1 .
r→±∞ r→±∞

6
 SciPost Phys. 16, 142 (2024)

Note that given (22), it seems like the limit here is divergent. However, it turns out that all
these divergent terms die off at large |u| [29], so the final result here is actually finite.
We can now evaluate the explicit form of the limit of ĀR± a in (24) and (28). Since the
radiative part satisfies the homogeneous Maxwell’s equation, it admits a mode expansion.
Thus, we can write

dd+1 p 1 a
Z
” —
R±
Āµ (X ) = g ϵ (⃗
p ) Ō ±
(⃗
p )e ip·X
+ c.c. , pµ = (|⃗p |, ⃗p ) , (29)
(2π)d+1 2p0 µ a

where the polarization tensor satisfies ϵua (p) = 0 (by our gauge condition) and pµ ϵµa (⃗p ) = 0
(by transversality condition). We can now perform a large r expansion of (29) to extract the
leading large r term and determine both ±a in (24) and E ± in (28). This involves changing the
integration variable from ⃗p to (ω, y a ), so that the momentum vector and polarization tensor
are now written as
2
 
a 1− y
⃗p(ω, y) = ω y , , ϵµa (⃗p ) = (− y a , δ ab , − y a ) . (30)
2

We can then use the stationary phase approximation to localize the integral over y to x in the
large |r| limit. This process is explained in detail in [29–31], and the limit in (24) simplifies
to Z∞
g d
” i iπd
—
±
Âa (u, x) = ± d
dω ω 2 −1 Oa± (ω, x)e− 2 ωu∓ 4 + c.c. , (31)
2(2π) 2 +1 0
where
Oa± (ω, x) ≡ C Ōa± (ω, x)C −1 . (32)
Once we quantize the theory in Section 3, we will see that for ω > 0, Oa± (ω, x) is the annihi-
lation operator of a gluon with polarization a and momentum ⃗p(ω, x), while Oa± (ω, x)† is the
corresponding creation operator.
The limit in (28) is more complicated to evaluate and requires one to consider the large
r expansion separately for d odd and even.9 This is done explicitly in [29], where it was
determined that
g2 2 d2 −1 a ±

± 2(4π) d2 Γ d (−∂ )
 ∂ Na (x) , d even,
± (2)
E (x) = d−1
Z
∂ a Na± ( y) (33)
g 2 (−1) 2 Γ (d−1) d
±

8π d+1 d y , d odd,
[(x − y)2 ]d−1

where10
1
Na± (x) ≡ lim ωŌa± (ω, x) .
 
(34)
g ω→0
Thus, Na± creates a soft gluon with polarization a. Furthermore, it was shown in [31] that this
operator is Hermitian
Na± (x)† = Na± (x) , (35)
and satisfies a flatness constraint
±
∂[a Nb] (x) = 0 =⇒ Na± (x) = ∂a N ± (x) . (36)
p p
9
It can be shown that the mode expansion can be recast in terms of Bessel functions Kν ( iu/ ir) with
ν = d2 − 1 + Z. The large r expansion then corresponds to the asymptotic expansion of Kν (z) at small z. This
is qualitatively different depending on whether ν is an integer or half-integer.
10
A similar soft operator, which we denote as N̂a± , was defined in (3.75) of [32], and it is related to the one
g2
defined in (34) via N̂a± = − 4p2 C Na± C −1 .

7
 SciPost Phys. 16, 142 (2024)

The Hermiticity constraint arises from the requirement that the asymptotic expansion of the
gauge field is analytic, and the flatness constraint arises from the requirement that the sym-
plectic form is invertible, as was discussed in [32].11
We see from (33) that there is a qualitatively different structure in odd and even dimen-
sions. The relationship between E ± and Na± is local in even dimensions but involves a non-local
integral over y in odd dimensions. This is due to the fact that Huygens’s principle is satisfied
in even dimensions but not in odd dimensions. More precisely, the Green’s function in even
dimensions is localized on the light cone whereas in odd dimensions it has support everywhere
in the interior of the light cone. Nevertheless, despite these different structures in odd and
even dimensions, it is possible to unify them via a shadow transform. This is an integral trans-
form that maps a conformal primary operator (in a CFT) of scaling dimension ∆ to another
primary operator with dimension d − ∆. For vector primaries, this is defined as

Ia b (x − y)
Z
x x
ea (x) ≡ dd y
V V b ( y) , Iab (x) = δab − 2 a 2 b . (37)
[(x − y) ]2 d−∆ x

Up to a normalization factor, the shadow transform is its own inverse, i.e.

πd (∆ − 1)(d − ∆ − 1)Γ ( d2 − ∆)Γ (∆ − d2 )

ea (x) = c∆,1 Va (x) ,
Ve c∆,1 = . (38)
Γ (∆ + 1)Γ (d − ∆ + 1)

Using the shadow transform, we prove in Appendix B that both cases of (33) can be unified
via

g2 2 ±
E± = ± ∂ N , ∂a N ± = ∂á ±
aN . (39)
4c1,1

Substituting (31) and (39) into (19), the full symplectic form becomes
Z Z ∞
i
dω ωd−1 Tr δ ab X(Oa± )Y(O±†
 
Ω(X, Y) = d
d x b
) − (X ↔ Y)
2(2π)d+1 0
Z (40)
1 d
 2 ± −1

− d x Tr X(C)Y(∂ N C ) − (X ↔ Y) .
2c1,1

3 Asymptotic Hilbert space

In the previous section, we have restricted ourselves to a purely classical analysis. We now
promote the classical fields to quantum fields via canonical quantization. In Section 3.1, we
extract the quantum commutators from the symplectic form. Then, in Section 3.2, we construct
the full Hilbert space involving the gauge field, including the vacuum sector.

3.1 Quantum commutators

By inverting the symplectic form (40), we can determine the quantum commutators of the op-
erators. To do this, we expand out the Lie algebra indices in (40) in the adjoint representation.
Since N ± , Oa± ∈ g, we can write

Oa± = Oa±I T I , N ± = N ±I T I , (T I )J K = − f I J K . (41)

In [31], the operator Na± (x) was called Oa(±,0) (x), and the Hermiticity and flatness conditions are discussed in
11

equations (2.21) and (5.14) of that paper, respectively.

8
 SciPost Phys. 16, 142 (2024)

Furthermore, since C ∈ G , it has indices C I J , and its inverse is given by (C −1 ) I J =(C T ) I J = C J I .12
It also satisfies the identity13
′ ′
J K′ ′ ′ ′ ′ ′
J K′ ′ ′ ′
fI C I I C JJ C KK = f I CI ICJ J CK K
= f IJK . (42)

Using these relations, we can explicitly take the trace of (40), so that it becomes
Z Z∞
i ±I ±I†


Ω(X, Y) = d d
x dω ω d−1 ab
δ X( O a )Y(O b
) − (X ↔ Y)
2(2π)d+1 0
Z (43)
1
 ‹
JKL 2 ±K I L
− f d IJ
d x X(C )Y(∂ N C ) − (X ↔ Y) .
2c1,1
With everything written out explicitly, we can now invert this following the procedure
given in Section 3.3.6 of [32] to obtain the Dirac brackets, which then give rise to the quantum
commutators (see Footnote 4)
 2(2π)d+1
Oa±I (ω, x), O±J ′ †
δab δ I J δ(ω − ω′ )δ d (x − y) ,

b (ω , y) = (44)
 IJ  ω d−1

C (x), C K L ( y) = 0 , (45)
 ±I 
N (x), C J K ( y) = 2ic1,1 f I K L C J L ( y)G(x − y) , (46)
Z
 ±I ±J
dd z G(x − z)G( y − z)∂ 2 N ±K (z) ,

N (x), N ( y) = 2ic1,1 f IJK
(47)

where G(x) is the scalar Green’s function satisfying ∂ 2 G(x) = δ d (x) and is explicitly given by

1
 4π ln(x 2 ) , d = 2,
G(x) = Γ ( d2 −1) 1 (48)
− d d −1 , d > 2.
4π 2 (x 2 ) 2

Now, since the momentum ⃗p(ω, x) in flat null coordinates is parametrized via (30), it can be
deduced that
2
δ(ω − ω′ )δ d (x − x ′ ) = 2|⃗p | δ d+1 (⃗p − ⃗p ′ ) . (49)
ω d−1

Consequently, (44) is precisely the commutation relation for a pair of creation and annihilation
operators, and so we interpret Oa±I (ω, x) as an operator that annihilates (by acting on the vac-
uum state) an outgoing (+) or incoming (−) one-particle gluon state with color I, polarization
a, and momentum ⃗p(ω, x).

3.2 Canonical quantization

Having derived the quantum commutators in the previous section, we now proceed to canon-
ically quantize the theory. We start with the vacuum sector of the theory, which is spanned by
the operators C and N ± . Since C commutes with itself, we can work in a basis of states |U, ±〉
that diagonalizes this operator:
C(x)| U, ± 〉 = U(x)| U, ± 〉 , 〈 U, ± | U ′ , ± 〉 = δ(U − U ′ ) . (50)

As discussed in [32], the state with U(x) = 1 is Lorentz invariant, and more general vacuum
states are given by Z
| f,±〉 = [DU] f (U)| U, ± 〉 . (51)

I I I I I I T I I
12
In the adjoint representation, (T I ) T = −T I , and so (eα T )−1 = e−α T = eα (T ) = (eα T ) T .
13
The general identity that applies to any representation R i of G is R i (C)−1 R i (T I )R i (C) = C I J R i (T J ). Note then
that (42) is the special case of this applied to the adjoint representation.

9
 SciPost Phys. 16, 142 (2024)

To complete the description of the vacuum Hilbert space, we determine after some algebra the
action of N ± on the basis states via (46) to be
Z
N ±I (x)| U, ± 〉 = 2ic1,1 dd y G(x − y)U J I ( y)DJU( y) | U, ± 〉 , (52)

where DUI is a derivative operator defined by

I
DU(x) U( y) = −T I U(x)δ d (x − y) . (53)

This operator was introduced and extensively studied in Appendix B of [32], and we refer the
reader there for more details regarding its explicit form and properties. It can then be easily
verified that (52) is consistent with (47).
With the vacuum sector fully characterized, the rest of the Hilbert space is constructed as
follows. The annihilation operators by definition annihilate all the vacuum states, so that

Oa±I (ω, x)| U, ± 〉 = 0 . (54)

The remaining states are then constructed as a Fock space by repeatedly acting on the vacuum
states with the creation operators Oa±I (ω, x)† . This completes our description of the full Hilbert
space.
Analogous to the four-dimensional case considered in [32], the charge that generates LGTs
is from (14)
I
1
Qϵ = ± 2 Tr[ϵ ⋆ F]
g I±
Z∓ Z
1 1
d x ϵ (x)C (x)∂ N (x) − 2 du dd x Tr ϵ(x) ±a (u, x), ∂u ±a (u, x) ,
2 ±J
  
=− d I IJ
2c1,1 g
(55)

where the commutator in the second term is the Lie algebra commutator and not the quantum
commutator. Notice that Q ϵ does not require a ± superscript since it is constructed from the
symplectic form, which itself is equal on I + and I − due to the boundary condition we imposed
in (7). Using (31), the second term can be rewritten as
Z
1 d
  ±a ±

du d x Tr ϵ(x)  (u, x), ∂u  a (u, x)
g2
Z∞ (56)
i f I J K δa b
Z
±J † ±K
= dω ω d−1 d I
d x ϵ (x)Oa (ω, x) O b (ω, x) .
2(2π)d+1 0

Using this and the commutators (44)–(47), it is easy to check that [Q ϵ , ·] = −iδϵ (·), so that

Q ϵ , Oa±I (ω, x) = −i f I J K ϵ J (x)Oa±K (ω, x) ,

 
(57)
±I † ±K †
 
Q ϵ , Oa (ω, x) = −i f IJK J
ϵ (x)Oa (ω, x) , (58)
 
Q ϵ , C(x) = −iϵ(x)C(x) , (59)
±
 
Q ϵ , N (x) = 0 . (60)

Lastly, by (52) and (54), we find

Z
Q ϵ | U, ± 〉 = −i dd x ϵ I (x)DU(x)
I
| U, ± 〉 . (61)

10
 SciPost Phys. 16, 142 (2024)

4 Ward identity and the leading soft gluon theorem

Thus far, our discussion has been limited to pure Yang-Mills theory. However, for the rest of the
paper, we will generalize our analysis to include massless states that transform under arbitrary
representations R i of G . In Section 4.1, we will derive the Ward identity associated with LGTs.
We then show that the Ward identity implies the leading soft gluon theorem in Section 4.2.

4.1 Ward identity

Consider an n-point scattering amplitude,14

〈 U, + |T {O1 · · · On } | U ′ , − 〉 , (62)

where T is the time-ordering operator and Oi ≡ Oi (ωi , x i ) is defined by15

Oi ≡ R i (C(x i )) θ (ωi ) Ōi+ (ωi , x i ) − Ōi− (ωi , x i ) + θ (−ωi ) Ōi− (−ωi , x i )† − Ōi+ (−ωi , x i )† ,
   

(63)
where θ is the Heaviside step function. These operators in the scattering amplitude appear
rather complicated, but for hard operators (63) is rather simple. For instance, consider ωi > 0.
In this case, only the first term in (63) is non-zero. When inserted into (62), the time-ordering
operator moves Ō− to the right where it annihilates the in-vacuum. Consequently, if ωi > 0,
we have Oi = R i (C(x i ))Ōi+ (ωi , x i ). Likewise, a similar argument shows that for ωi < 0 we
have Oi = R i (C(x i ))Ōi− (−ωi , x i )† . To summarize, we have

R i (C(x))Ōi+ (ω, x) ,
¨
ω > 0,
Oi (ω, x) = (64)
R i (C(x))Ōi− (−ω, x)† , ω < 0.

The complicated expression (63) is necessary only for the case ωi = 0. For instance, the soft
gluon operator [33, 40] is defined by

SaI (x) ≡ lim ωOaI (ω, x) = C I J (x) Na+J (x) − Na−J (x) ,
 
ω→0
(65)

where we have used (34) and (35), as well as the regulated Heaviside step function θ (0) = 12 ,
for the second equality.
To derive the Ward identity associated with LGTs, first note that C(x) commutes with all
operators with energies not strictly zero. Hence, we have

〈U, +|C(x)T {O1 · · · On }|U ′ , −〉 = 〈U, +|T {O1 · · · On }C(x)|U ′ , −〉 . (66)

Because the vacua are C eigenstates, it follows that

(U(x) − U ′ (x))〈U, +|T {O1 · · · On }|U ′ , −〉 = 0 , (67)

which immediately implies

〈U, +|T {O1 · · · On }|U ′ , −〉 = δ(U − U ′ )〈O1 · · · On 〉U , (68)

where the correlator is implicitly time-ordered.

14
More generally, we can consider scattering amplitudes with arbitrary in and out-vacuum states given in (51),
but they can all be described in terms of (62).
15
Comparing this definition of Oi to that given in Section 4.3 of [32], we see that (63) has an extra factor of
R i (C(x i )). This is because Ōi± are gauge-invariant operators, whereas Oi± defined in [32] are gauge-covariant
operators, and the relationship between them is captured via multiplication by R i (C(x i )).

11
 SciPost Phys. 16, 142 (2024)

Next, recall that the large gauge charge Q ϵ given in (14) generates LGTs (see (57)). This
means the action of Q ϵ on an arbitrary field Oi in representation R i is given by
 
Q ϵ , Oi = −iR i (ϵ(x i ))Oi . (69)

It follows that when inserting Q ϵ on the left of all the hard operators Oi and then commuting
it past all the Oi ’s, we obtain
n
X
 ′
R i (ϵ(x i ))〈 U, + |T {O1 · · · On }| U ′ , − 〉 .

〈 U, + | Q ϵ , T {O1 · · · On } | U , − 〉 = −i (70)
i=1

Using (61), the above equality becomes

Z
+ DUI ′ ( y) 〈 U, + |T {O1 · · · On }| U ′ , − 〉


dd y ϵ I ( y) DU(
I
y)

n
X (71)
=− R i (ϵ(x i ))〈 U, + |T {O1 · · · On }| U ′ , − 〉 .
i=1

+ DUI ′ ( y) δ(U − U ′ ) = 0 derived in [32], we

I


Substituting in (68) and using the property DU( y)
find
Z Xn
dd y ϵ I ( y)DU(
I
y)
〈O1 · · · On 〉U = − R i (ϵ(x i ))〈O1 · · · On 〉U . (72)
i=1

Finally, setting ϵ( y) = T I δ d (x − y), we obtain the differential equation

n
X
I
DU(x) 〈 O1 · · · O n 〉 U = − δ d (x − x i )R i (T I )〈 O1 · · · On 〉U . (73)
i=1

This is a first-order differential equation, which we can solve to get the Ward identity

〈 O1 · · · On 〉U = R1 (U(x 1 )) · · · R n (U(x n ))〈 O1 · · · On 〉U=1 , (74)

where 〈 O1 · · · On 〉U=1 is an integration constant, and is the scattering amplitude evaluated in

standard QFT. This is the main result of this section, and we will use it to derive the leading
soft gluon theorem in the next subsection.

4.2 Leading soft gluon theorem

The leading soft gluon theorem states that the insertion of the soft gluon operator (65) is given
by
n
X pi · ϵa (x)
I
〈 Sa (x)O1 · · · On 〉U=1 = i g ω R i (T I )〈 O1 · · · On 〉U=1
i=1
p i · p
n (75)
X
∂a ln (x − x i )2 R i (T I )〈 O1 · · · On 〉U=1 ,
 
= ig
i=1

where in the second equality, we have used the explicit parameterization (30) for both the
momenta and polarization. In this subsection, we derive this from (74) (or equivalently (73)).
Let us start with the left-hand side of the above equation along with the definition (65).
We need to determine the action of Na± (x) on the vacuum state. Recalling (38), we can easily
invert (39) so that
1 ã
Na±I (x) = ∂a N ±I (x) . (76)
c1,1

12
 SciPost Phys. 16, 142 (2024)

Using the fact that ∂a G(x) has conformal dimension ∆ = d − 1, we can evaluate its shadow
transform (37) to be
1 2
∂Þ
a G(x) = ∂a ln(x ) . (77)
2
It then follows from this and (52) that the action of Na± on the vacuum state is given by
Z
±I
Na (x)| U, ± 〉 = i dd y∂a ln (x − y)2 U J I ( y)DJU( y) | U, ± 〉 .
 
(78)

Hence, upon inserting SaI (x) between any two vacua U and U ′ , we get
〈U, +|T {SaI (x)O1 · · · On }|U ′ , −〉
= 〈U, +|C I J (x)T {(NaJ+ (x) − NaJ− (x))O1 · · · On }|U ′ , −〉
Z
(79)
= −i g U (x) dd y ∂a ln (x − y)2 U K J ( y)DKU( y) + U ′K J ( y)DKU ′ ( y)
IJ
 

× 〈U, +|T {O1 · · · On }|U ′ , −〉 ,

where the time-ordering operator T moves Na+ to the left to act on the out-vacuum and Na−
to the right to act on the in-vacuum. Substituting in (68) on both sides of the above equation
and simplifying, we arrive at the expression
Z
〈 Sa (x)O1 · · · On 〉U = −i g U (x) dd y ∂a ln (x − y)2 U K J ( y)DKU( y) 〈 O1 · · · On 〉U .
I IJ
 
(80)

Finally, substituting in (73), we get

n
X
∂a ln (x − x k )2 U K J (x k )R i (T K )〈 O1 · · · On 〉U ,
 
〈 SaI (x)O1 · · · On 〉U IJ
= i g U (x) (81)
i=1

and then setting U = 1, we reproduce the leading soft gluon theorem (75). Of course, the
result we have just derived is more general than (75), since it can be evaluated for any U
whereas the soft theorem applies only for U = 1. However, we can also consider scattering
amplitudes with multiple soft gluons where the soft limits are taken consecutively.16 It can be
shown that this general multiple soft gluon theorem is in fact completely equivalent to (81).

Acknowledgments
We would like to thank Daniel Kapec for useful conversations that initiated this work.

Funding information T.H. has been supported by the Heising-Simons Foundation “Obser-
vational Signatures of Quantum Gravity” collaboration grant 2021-2817, the U.S. Depart-
ment of Energy grant DE-SC0011632, and the Walter Burke Institute for Theoretical Physics.
P.M. gratefully acknowledges support from the STFC consolidated grants ST/P000681/1 and
ST/T000694/1.

A Definitions and conventions

In this appendix, we provide a lightning overview of the conventions we use from differential
geometry and Lie algebra. We will be following the conventions outlined in [32], in which
more details can be found (see also Chapter 20 of [41]).
16
The symmetry interpretation of simultaneous soft limits is still an open problem.

13
 SciPost Phys. 16, 142 (2024)

A.1 Differential geometry

We will mainly be interested in two manifolds – the phase space symplectic manifold and the
spacetime Lorentzian manifold. Boldfaced letters are used to denote forms and vectors on the
phase space manifold, and regular letters denote those on the spacetime manifold.

A.1.1 Spacetime manifold

Let our spacetime M to be a (d +2)-dimensional globally hyperbolic Lorentzian manifold with
coordinates x µ and metric gµν , where µ, ν ∈ {0, 1, . . . , d + 1}. At every point p ∈ M, we adopt
the canonical basis ∂µ on the tangent space Tp M and dx µ on the cotangent space Tp∗ M. The
wedge product is defined as

dx µ1 ∧ · · · ∧ dx µq ≡ q! dx [µ1 ⊗ · · · ⊗ dx µq ] , (A.1)

where [ · · · ] denote the weighted antisymmetrization of indices, e.g. ω[µν] = 2!

1
(ωµν − ωνµ ).
This is a basis on the space of q-forms Ωq (M), which means any q-form Cq can be written as

1
Cq = (Cq )µ1 ···µq (x) dx µ1 ∧ · · · ∧ dx µq , Cq ∈ Ωq (M) . (A.2)
q!

This implies that for any p-form C p and q-form Cq , we have

(p + q)!
(C p ∧ Cq )µ1 ···µp+q = (C p )[µ1 ···µp (Cq )µp+1 ···µp+q ] . (A.3)
p!q!

The volume form on M is a (d + 2)-form defined as

Æ
ε ≡ − det g dx 0 ∧ · · · ∧ dx d+1 , (A.4)

and the choice of the sign above fixes an orientation for the spacetime. It obeys the useful
identity
ν νp
εµ1 ···µp αp+1 ···αd+2 εαp+1 ···αd+2 ν1 ···νp = (−1)s p!(d + 2 − p)!δ[µ1 · · · δµ ] , (A.5)
1 p

where s is the number of negative eigenvalues in the metric, as well as

µ ···µd+2 µ ···µd+2
εµ1 ···µd+2 ∂ µ1 Cd+1
2
= ∇µ1 (εµ1 ···µd+2 Cd+1
2
), (A.6)

for any (d + 1)-form Cd+1 .

The following conventions are used for standard operations on forms, with Cq being a
q-form:

(iξ Cq )µ1 ···µq−1 ≡ ξµ (Cq )µµ1 ···µq−1 ,

(dCq )µ1 ···µq+1 ≡ (q + 1)∂[µ1 (Cq )µ2 ···µq+1 ] ,
(A.7)
1
(⋆Cq )µ1 ···µd+2−q ≡ εµ1 ···µd+2−q ν1 ···νq (Cq )ν1 ···νq .
q!

The exterior derivatives are nilpotent, i.e. d2 = 0, which implies that all exact forms
(Cq = dCq−1 ) are closed (dCq = 0). We can also verify

⋆2 Cq = (−1)s+q(d+2−q) Cq , (A.8)

where again s is the number of negative eigenvalues in the metric.

14
 SciPost Phys. 16, 142 (2024)

Given a q-form, we can integrate it over a q-dimensional submanifold of M. In particular,

we are mainly concerned with the cases q = d, d + 1 and d + 2, for which
Z Z
Æ
Cd+2 = (−1) s
dd+2 x |g|(⋆Cd+2 ) ,
Σd+2 Σd+2
Z Z
Cd+1 = − dΣµ (⋆Cd+1 )µ , (A.9)
Σd+1 Σd+1
Z Z
1
Cd = − dSµν (⋆Cd )µν ,
Σd 2 Σd

where dΣµ and dSµν are respectively the directed area elements corresponding to the subman-
ifolds Σd+1 and Σd . Stokes’ theorem for a q-form is given by
Z I
dCq−1 = Cq−1 , (A.10)
Σq ∂ Σq

where we orient ∂ Σq to be outwardly-directed with respect to Σq . It is useful for us to focus

on the special cases where q = d + 2 and q = d + 1, in which case Stokes’ theorem is given to
be
Z I Z I
µ µ [µν] 1
ε∇µ C = dΣµ C , dΣµ ∇ν C = dSµν C µν , (A.11)
Σ ∂Σ Σ 2 ∂Σ
d+2 d+2 d+1 d+1

where ∇µ is the covariant derivative with respect to the metric g.

One of the (slight) challenges pertaining to working with forms is settling the explicit signs
that determine the orientation of various area elements. We therefore discuss the signs that
will be relevant for our work here. Flat null coordinates are related to Cartesian coordinates
by
r u u
Xµ = 1 + x 2 + , 2x a , 1 − x 2 − . (A.12)
2 r r
The oriented volume form is
1
ε = dX 0 ∧ · · · ∧ dX d+1 = r d du ∧ dd x ∧ dr . (A.13)
2
Given this and recalling the first equation in (A.9), the integral over any (d + 2)-form is given
by Z Z
1
Cd+2 = − du dr dd x |r|d (⋆Cd+2 ) . (A.14)
M 2
The area elements on I ± and I∓± can be determined by requiring that Stokes’ theorem
(A.10) is satisfied. This means we require
Z Z Z
dCd+1 = Cd+1 − Cd+1 + (other boundary terms) ,
M I+ I−
Z I (A.15)
dCd = ∓ Cd .
I ± ∪i ± I∓±

Note that the ± signs above are required since I ± are future/past boundaries of M respec-
tively, and I∓± are past/future boundaries of I ± respectively. It can now be checked that
(A.15) implies Z Z
1  
Cd+1 = − du dd x lim |r|d (⋆Cd+1 ) r ,
I± 2 I± r→±∞
Z Z (A.16)
1  
Cd = d d
d x lim lim |r| (⋆Cd ) ur
.
I± 2 I± u→∓∞ r→±∞
∓ ∓

15
 SciPost Phys. 16, 142 (2024)

A.1.2 Phase space manifold

The phase space is a symplectic manifold (Γ , Ω), where Γ is a smooth manifold and Ω is a
closed non-degenerate two-form known as the symplectic form, i.e. it obeys
dΩ = 0 , iX Ω = 0 =⇒ X = 0, for all X ∈ TΓ . (A.17)
As in [32], we assume both the first and second cohomology groups vanish, i.e.
H1 (Γ ) = H2 (Γ ) = 0, so that there exists the one-form Θ called the symplectic potential such
that Ω = dΘ. Notice that the symplectic potential is defined only up to an exact one-form.
The phase space for a particular field theory living on a spacetime manifold M is specified
in terms of the dynamical fields ϕ i in the field theory. Typically, we impose boundary conditions
on the fields ϕ i to ensure finite energy flux through the boundaries of the spacetime manifold.
The configuration space F is then defined to be the space of all field configurations that satisfy
the boundary conditions. A vector X ∈ T F is defined in component notation to be
XZ δ
X= ε Xi (ϕ i , ∇µ ϕ i , . . .) , (A.18)
i M δϕ i

where Xi is in general a function of the dynamical fields (as well as any possible background
fields), the metric, and spacetime coordinates. Given any function f in terms of the fields, we
refer to X( f ) as the variation of f with respect to X.
To obtain the symplectic form from a Lagrangian, we utilize the covariant phase space
formalism. Given a Lagrangian spacetime (d + 2)-form L involving fields ϕ i , we have
X
X(L) = Ei X(ϕ i ) + dθ (X) , (A.19)
i

where θ is known as the symplectic potential current density, and Ei are the equations of
motion. The solution space S is defined to be the subspace of F where the fields satisfy Ei = 0,
and the tangent space T S consists of vector satisfying the linearized equations of motion
X(Ei ) = 0. In this paper, we restrict ourselves to the solution space S, and field configurations
that live in S are known as on-shell configurations.
The symplectic current density ω is defined to be ω = dθ . The pre-symplectic potential
and pre-symplectic form can then be obtained by respectively integrating θ and ω over a
future-directed (d + 1)-dimensional Cauchy slice Σ, i.e.
Z Z
Θe Σ (X) = θ (X) , Ωe Σ (X, Y) = ω(X, Y) . (A.20)
Σ Σ
The pre-symplectic form is a closed two-form on S, but it is not necessarily invertible. To
remedy this, we determine the kernel of Ωe , and we identify X ∼ X′ if X − X′ ∈ ker ΩΣ . The
phase space is then Γ ≡ S/ ∼, and we have ΘΣ = Θ̃|Γ and Ω = Ω e |Γ . Note that by construction,
Ω is both closed and non-degenerate on Γ . At a practical level, the equivalence relation on
Γ is imposed by a gauge fixing condition f [ϕ] = 0, which maps each equivalence class to a
particular representative.

B Radial electric field as shadow transform

In this appendix, we prove that (33), which we reproduce here for convenience,
g2 2 d2 −1 a (±)

 2(4π) d2 Γ ( d ) (−∂ )
 ± ∂ Na (x) , d even,
± 2
E (x) = 2
d−1
Z
∂ a Na(±) ( y) (B.1)
± g (−1)8πd+1Γ (d−1) dd y
2
 , d odd,
[(x − y)2 ]d−1

16
 SciPost Phys. 16, 142 (2024)

can be unified by utilizing the shadow transform so that

g2 2 ±
E± = ± ∂ N , ∂a N ± = ∂á ±
aN . (B.2)
4c1,1
We recall from (37) that the definition of the shadow transform for vector primaries is
Ia b (x − y)
Z
x x
ea (x) ≡ dd y
V V b ( y) , Iab (x) = δab − 2 a 2 b . (B.3)
[(x − y) ]2 d−∆ x

ea (x) = c∆,1 Va (x), where

This satisfies Ve

πd (∆ − 1)(d − ∆ − 1)Γ ( d2 − ∆)Γ (∆ − d2 )

c∆,1 = . (B.4)
Γ (∆ + 1)Γ (d − ∆ + 1)
Note that if ∆ = 1, then one of the terms in the numerator vanishes. If d is odd, this means
c1,1 = 0, although we will see that the final result E ± is still finite. If d is even, then Γ (1 − d2 )
is divergent for d ≥ 2, and so the numerator of (B.4) is rendered finite.
We will now show that (B.2) reproduces (B.1). Let us check the odd-dimensional case
first. Because we have c1,1 = 0, we will need to carefully regulate the integral. We do so by
evaluating the shadow transform for a field with generic ∆ and then taking the limit ∆ → 1.
The regulated definition of E ± (x) then becomes
g2 aá
E ± (x) = ± lim ∂ ∂a N ± (x)
∆→1 4c∆,1

g2 a Iab (x − y)
Z
= ± lim ∂ dd y ∂ b N ± ( y)
∆→1 4c∆,1 [(x − y)2 ]d−∆
(B.5)
g 2 Γ (d − 1)
Z
1
=± dd y ∂ 2 N ± ( y)
d
4π (d − 2)Γ ( 2 − 1)Γ (1 − 2 )
d d [(x − y)2 ]d−1

d−1
g 2 (−1) 2 Γ (d − 1) ∂ a Na± ( y)
Z
=± d d
y ,
8πd+1 [(x − y)2 ]d−1
where in the second equality we used the definition of the shadow transform for an operator
of dimension ∆; in the third equality we used “integration-by-parts” style manipulation (IBP)
to move the partial derivative onto the other term in the integrand, applied the identity [33]
Ia b (x) ∆−1 1
§ ª § ª
∂ b
= ∂a , (B.6)
(x 2 )d−∆ d −∆ (x 2 )d−∆
for any ∆ not an integer,17 and moved the derivative ∂ a on x into the integral and changed
it into a y derivative; and in the final equality we used IBP again to move all the derivatives
back onto N ± ( y) and Na± = ∂a N ± . This is in agreement with (B.1).
Next, let us check the even-dimensional case in (B.1). It will be useful to first prove the
following useful identity:

b Ia b (x)
2c1,1
§ ª
d
∂ = (−∂ 2 ) 2 −1 ∂a δ d (x) . (B.7)
(x )
2 d−1 d d
(4π) 2 Γ ( 2 )
To prove this, we want to write the right-hand side in a way that can be easily regularized. We
claim that we can rewrite this identity as
Ia b (x) 2c1,1
§ ª
∂b = − (−∂ 2 )d−1 ∂a log(x 2 ) . (B.8)
(x 2 )d−1 d d
(4π) 2 Γ ( 2 )2

17
As we’re taking the limit ∆ → 1, we can assume ∆ is indeed not an integer and hence apply this equation.

17
 SciPost Phys. 16, 142 (2024)

To see why, we first note for any positive integer m we have

1 2mx a 1
§ ª
−∂ 2 = ∂ a = 2m[d − 2(m + 1)] 2 m+1 . (B.9)
(x )
2 m (x )
2 m+1 (x )
Using this recursively implies that for any positive integer n we have

1 4n Γ (m + n)Γ ( d2 − m) 1
§ ª
(−∂ 2 )n = . (B.10)
(x 2 )m Γ (m)Γ ( d2 − m − n) (x 2 )m+n

It follows

1 2d−3 (d − 2)Γ ( d2 − 1)2

 ‹
d d
(−∂ ) 2 2 −1 2
log(x ) = −2(d − 2)(−∂ ) 2 2 −2 =− , (B.11)
x2 d
(x 2 ) 2 −1

where in the second equality we applied (B.10) with m = 1 and n = d2 − 2. Now, if we act on
this with −∂ 2 one more time, it is clear from (B.9) with m = d2 − 1 that the result vanishes.
However, this is only true if x ̸= 0, and so acting on both sides of (B.11) with −∂ 2 can give
something proportional to the delta function. To determine the coefficient, we first note that
we can write the delta function as
Γ ( d2 + 1) ε2
δ d (x) = d
lim d
. (B.12)
π2 ε→0 (x 2 + ε2 ) 2 +1
To check this is indeed the case, note that for x ̸= 0, the right-hand side obviously vanishes.
Furthermore, integrating it over all of space yields
Z∞
Γ ( d2 + 1) ε2 ε2
Z
d
d
lim d x d
= lim d dr r d−1 d
= 1, (B.13)
π2 ε→0 (x 2 + ε2 ) 2 ε→0
0 (x 2 + ε2 ) 2 +1
which proves (B.12).
Similarly regulating the right-hand side of (B.11), we obtain
‹2 ¨ «
1

d d
(−∂ 2 ) 2 log(x 2 ) = −2d−3 (d − 2)Γ − 1 lim (−∂ 2 ) d
2 ε→0 (x 2 + ε2 ) 2 −1
– ™
2
1
 ‹  ‹
d d x
= −2d Γ Γ + 1 lim −
2 2 ε→0 (x 2 + ε2 ) d2 d
(x 2 + ε2 ) 2 +1 (B.14)
ε2
 ‹  ‹
d d
= −2d Γ Γ + 1 lim
2 2 ε→0 (x 2 + ε2 ) d2 +1
 ‹
d d
= −(4π) 2 Γ δ d (x) ,
2
where in the last equality we used (B.12). Substituting the delta function in (B.7) with this
expression, we obtain

b Ia b (x)
2c1,1
§ ª
∂ =− (−∂ 2 )d−1 ∂a log(x 2 ) , (B.15)
(x )
2 d−1
(4π)d Γ ( d )2 2

which is exactly (B.8), as claimed. This means to prove (B.7), it suffices to prove (B.15).
Now, note that c1,1 is ill-defined since there are both zeros and divergences in the numer-
ator, so we regulate (B.15) and rewrite it as
Ia b (x) 2c∆,1 1
§ ª § ª
2 d−1
lim ∂ b
= − lim
2 (−∂ ) ∂a , (B.16)
∆→1 (x 2 )d−∆ ∆→1
(4π)d Γ d2 (∆ − 1)(x 2 )1−∆

18
 SciPost Phys. 16, 142 (2024)

where we noted
1 1
lim = + log(x 2 ) , (B.17)
∆→1 (∆ − 1)(x )
2 1−∆ ∆−1
1
and the leading ∆−1 is eliminated by the derivative ∂a . It follows using (B.6) that the left-hand
side of (B.16) is
Ia b (x) ∆−1 1
§ ª § ª
lim ∂ b
= lim ∂a , (B.18)
∆→1 (x 2 )d−∆ ∆→1 d − ∆ (x 2 )d−∆
whereas the right-hand side of (B.16) is
2c∆,1 1
§ ª
2 d−1
− lim (−∂ ) ∂ a
(∆ − 1)(x 2 )1−∆
∆→1

2
(4π)d Γ d2
4d−1 Γ (d − ∆)Γ d2 + ∆ − 1

§
2c∆,1 1
ª
= − lim ∂ a
(B.19)
(x 2 )d−∆

2
Γ (1 − ∆)Γ ∆ − d2


∆→1
(4π)d Γ d2 (∆ − 1)
∆−1 1
§ ª
= lim ∂a ,
∆→1 d − 1 (x 2 )d−∆
where in the first equality we used (B.10) with n = d − 1 and m = 1 − ∆, and in the last
equality we substituted in c∆,1 from (B.4) and then expanded about ∆ = 1. This is exactly
(B.18), thus proving (B.16). As (B.16) is a rewriting of (B.15), which as we argued above is
equivalent to (B.7), this completes the proof of (B.7).
Having proved the identity, we return to evaluating E ± for the even-dimensional case,
which is given in (B.2) to be

g2 aá 2
Iab (x − y)
Z § ª
± g
E =± ∂ ∂a N = ±
± ∂ a d
d y ∂ b N ± ( y) . (B.20)
4c1,1 4c1,1 [(x − y)2 ]d−1

It follows
g2 a Iab (x − y)
Z § ª
E± = ∓ ∂ d
d y∂ b
N ± ( y)
4c1,1 [(x − y)2 ]d−1
g2
Z
=± d
dd y(−∂ 2 )d−1 ∂ a ∂a δ d (x − y)N ± ( y) (B.21)
2(4π) 2 Γ ( d2 )2
g2 d
=± d
(−∂ 2 ) 2 −1 ∂ a Na± (x) ,
2(4π) 2 Γ ( d2 )
where in the first equality we used IBP, in the second equality we used (B.7) and moved the
derivative ∂ a on x into the integral and changed it into a y derivative, and in the final equality
we used IBP again to move derivatives back onto N ± ( y). This is in agreement with (B.1), and
completes the proof that (B.2) is indeed correct for both odd and even-dimensional cases.

References
[1] A. Strominger, Lectures on the infrared structure of gravity and gauge the-
ory, Princeton University Press, Princeton, USA, ISBN 9781400889853 (2018),
doi:10.23943/9781400889853.

[2] A. Strominger, On BMS invariance of gravitational scattering, J. High Energy Phys. 07,
152 (2014), doi:10.1007/JHEP07(2014)152.

19
 SciPost Phys. 16, 142 (2024)

[3] T. He, V. Lysov, P. Mitra and A. Strominger, BMS supertranslations and Weinberg’s soft
graviton theorem, J. High Energy Phys. 05, 151 (2015), doi:10.1007/JHEP05(2015)151.

[4] A. Strominger and A. Zhiboedov, Gravitational memory, BMS supertranslations and soft
theorems, J. High Energy Phys. 01, 086 (2016), doi:10.1007/JHEP01(2016)086.

[5] S. Weinberg, Infrared photons and gravitons, Phys. Rev. 140, B516 (1965),
doi:10.1103/PhysRev.140.B516.

[6] H. Bondi, M. G. J. van der Burg, and A. W. K. Metzner, Gravitational waves in general
relativity, VII. Waves from axi-symmetric isolated system, Proc. R. Soc. Lond., A. Math.
Phys. Sci. 269, 21 (1962), doi:10.1098/rspa.1962.0161.

[7] R. K. Sachs, Gravitational waves in general relativity VIII. Waves in asymptoti-

cally flat space-time, Proc. R. Soc. Lond., A. Math. Phys. Sci. 270, 103 (1962),
doi:10.1098/rspa.1962.0206.

[8] V. B. Braginsky and L. P. Grishchuk, Kinematic resonance and memory effect in free mass
gravitational antennas, Sov. J. Exp. Theor. Phys. 62, 427 (1985).

[9] A. G. Wiseman and C. M. Will, Christodoulou’s nonlinear gravitational-wave mem-

ory: Evaluation in the quadrupole approximation, Phys. Rev. D 44, R2945 (1991),
doi:10.1103/PhysRevD.44.R2945.

[10] D. Christodoulou, Nonlinear nature of gravitation and gravitational-wave experiments,

Phys. Rev. Lett. 67, 1486 (1991), doi:10.1103/PhysRevLett.67.1486.

[11] K. S. Thorne, Gravitational-wave bursts with memory: The Christodoulou effect, Phys. Rev.
D 45, 520 (1992), doi:10.1103/PhysRevD.45.520.

[12] T. He, P. Mitra, A. P. Porfyriadis and A. Strominger, New symmetries of massless QED, J.
High Energy Phys. 10, 112 (2014), doi:10.1007/JHEP10(2014)112.

[13] T. He, P. Mitra and A. Strominger, 2D Kac-Moody symmetry of 4D Yang-Mills theory, J.

High Energy Phys. 10, 137 (2016), doi:10.1007/JHEP10(2016)137.

[14] S. Pasterski, Asymptotic symmetries and electromagnetic memory, J. High Energy Phys. 09,
154 (2017), doi:10.1007/JHEP09(2017)154.

[15] M. Campiglia and A. Laddha, Asymptotic symmetries of QED and Weinberg’s soft photon
theorem, J. High Energy Phys. 07, 115 (2015), doi:10.1007/JHEP07(2015)115.

[16] T. T. Dumitrescu, T. He, P. Mitra and A. Strominger, Infinite-dimensional fermionic

symmetry in supersymmetric gauge theories, J. High Energy Phys. 08, 051 (2021),
doi:10.1007/JHEP08(2021)051.

[17] M. Pate, A.-M. Raclariu and A. Strominger, Color memory: A Yang-Mills ana-
log of gravitational wave memory, Phys. Rev. Lett. 119, 261602 (2017),
doi:10.1103/PhysRevLett.119.261602.

[18] A. Ball, M. Pate, A.-M. Raclariu, A. Strominger and R. Venugopalan, Measuring color
memory in a color glass condensate at electron-ion colliders, Ann. Phys. 407, 15 (2019),
doi:10.1016/j.aop.2019.04.010.

[19] D. Kapec, V. Lysov and A. Strominger, Asymptotic symmetries of massless QED in even di-
mensions, Adv. Theor. Math. Phys. 21, 1747 (2017), doi:10.4310/ATMP.2017.v21.n7.a6.

20
 SciPost Phys. 16, 142 (2024)

[20] D. Kapec, V. Lysov, S. Pasterski and A. Strominger, Higher-dimensional supertrans-

lations and Weinberg’s soft graviton theorem, Ann. Math. Sci. Appl. 2, 69 (2017),
doi:10.4310/AMSA.2017.v2.n1.a2.

[21] M. Pate, A.-M. Raclariu and A. Strominger, Gravitational memory in higher dimensions, J.
High Energy Phys. 06, 138 (2018), doi:10.1007/JHEP06(2018)138.

[22] M. Campiglia and M. Varadarajan, A quantum kinematics for asymptotically flat gravity,
Class. Quantum Gravity 32, 135011 (2015), doi:10.1088/0264-9381/32/13/135011.

[23] D. Kapec, V. Lysov, S. Pasterski and A. Strominger, Semiclassical Virasoro sym-

metry of the quantum gravity S -matrix, J. High Energy Phys. 08, 058 (2014),
doi:10.1007/JHEP08(2014)058.

[24] V. Lysov, S. Pasterski and A. Strominger, Low’s subleading soft theorem as a symmetry of
QED, Phys. Rev. Lett. 113, 111601 (2014), doi:10.1103/PhysRevLett.113.111601.

[25] S. Pasterski, A. Strominger and A. Zhiboedov, New gravitational memories, J. High Energy
Phys. 12, 053 (2016), doi:10.1007/JHEP12(2016)053.

[26] M. Campiglia and A. Laddha, Asymptotic symmetries of gravity and soft theorems for mas-
sive particles, J. High Energy Phys. 12, 001 (2015), doi:10.1007/JHEP12(2015)094.

[27] M. Campiglia and A. Laddha, Subleading soft photons and large gauge transformations, J.
High Energy Phys. 11, 012 (2016), doi:10.1007/JHEP11(2016)012.

[28] M. Campiglia and A. Laddha, Sub-subleading soft gravitons: New symmetries of quantum
gravity?, Phys. Lett. B 764, 218 (2017), doi:10.1016/j.physletb.2016.11.046.

[29] T. He and P. Mitra, Asymptotic symmetries and Weinberg’s soft photon theorem in Minkd+2 ,
J. High Energy Phys. 10, 213 (2019), doi:10.1007/JHEP10(2019)213.

[30] T. He and P. Mitra, Asymptotic symmetries in (d + 2)-dimensional gauge theories, J. High

Energy Phys. 10, 277 (2019), doi:10.1007/JHEP10(2019)277.

[31] T. He and P. Mitra, New magnetic symmetries in (d + 2)-dimensional QED, J. High Energy
Phys. 01, 122 (2021), doi:10.1007/JHEP01(2021)122.

[32] T. He and P. Mitra, Covariant phase space and soft factorization in non-Abelian gauge the-
ories, J. High Energy Phys. 03, 015 (2021), doi:10.1007/JHEP03(2021)015.

[33] D. Kapec and P. Mitra, Shadows and soft exchange in celestial CFT, Phys. Rev. D 105,
026009 (2022), doi:10.1103/PhysRevD.105.026009.

[34] D. Kapec, Soft particles and infinite-dimensional geometry, Class. Quantum Gravity 41,
015001 (2023), doi:10.1088/1361-6382/ad0514.

[35] C. Crnkovic and E. Witten, Covariant description of canonical formalism in geometrical

theories, in Three hundred years of gravitation, Cambridge Univeristy Press, Cambridge,
UK, ISBN 9780521379762 (1987).

[36] J. Lee and R. M. Wald, Local symmetries and constraints, J. Math. Phys. 31, 725 (1990),
doi:10.1063/1.528801.

[37] R. M. Wald and A. Zoupas, General definition of “conserved quantities” in gen-

eral relativity and other theories of gravity, Phys. Rev. D 61, 084027 (2000),
doi:10.1103/PhysRevD.61.084027.

21
 SciPost Phys. 16, 142 (2024)

[38] V. Iyer and R. M. Wald, Some properties of the Noether charge and a proposal for dynamical
black hole entropy, Phys. Rev. D 50, 846 (1994), doi:10.1103/PhysRevD.50.846.

[39] D. Harlow and J.-q. Wu, Covariant phase space with boundaries, J. High Energy Phys. 10,
146 (2020), doi:10.1007/JHEP10(2020)146.

[40] D. Kapec and P. Mitra, A d-dimensional stress tensor for Minkd+2 gravity, J. High Energy
Phys. 05, 186 (2018), doi:10.1007/JHEP05(2018)186.

[41] V. P. Nair, Quantum field theory: A modern perspective, Springer, New York, USA, ISBN
9780387213866 (2005), doi:10.1007/b106781.

22