Introduction to Celestial Holography

Ana Raclariu
Perimeter Institute for Theoretical Physics, ON, Canada
araclariu@[Link]

June 18, 2021

Abstract
These notes consist of 3 lectures on celestial holography given at the Pre-Strings
school 2021. We start by reviewing how the subleading soft graviton theorem implies
that Lorentz symmetry in asympotically flat spacetimes is enhanced to Virasoro.
Celestial amplitudes defined as S-matrices computed in a basis of boost eigenstates
are shown to make this symmetry manifest. Both massless and massive asymptotic
states are recast as insertions on the celestial sphere transforming as global confor-
mal primaries under the Lorentz SL(2, C). We conclude with a review of celestial
symmetries and the constraints they impose on celestial scattering.

1 Introduction
AdS/CFT [1, 2, 3] provided a concrete realization of the holographic principle [4, 5]: a
theory of gravity in D dimensional spacetimes should be dual to a quantum theory in one
dimension less. A realization of this duality in any but asymptotically negatively curved
backgrounds remains an important open problem. The goal of these lectures is to review
some of the recent progress in the case of asymptotically flat spacetimes.
In the past decade we learned that gravity and gauge theory in asymptotically flat
spacetimes (AFS) in the infrared are governed by a triangular equivalence: soft theorems
can be recast as conservation laws associated with large gauge symmetries [6, 7], while
memory effects are an observable signature thereof [8, 9] (see [10] for a detailed review).
These developments lead to the proposal that gravity in four-dimensional asymptotically
flat spacetimes may be dual to a theory living on the “celestial sphere” at infinity [11].
This proposal is backed up by evidence that Lorentz symmetry is enhanced to Virasoro
[12], as well as the existence of a stress tensor constructed from a particular subleading
soft graviton mode in the bulk [13]. In the first lecture we will see how all this is a direct
consequence of the subleading soft graviton theorem. In the second lecture we formulate
scattering in AFS in terms of a new observable: the celestial amplitude. We show how
celestial amplitudes re-express the S-matrix in a basis of boost eigenstates (in contrast
to conventional scattering amplitudes which are computed in a plane wave basis). Such
a construction can be done for scattering of both massive and massless particles which

1
we illustrate in the case of two massless and one massive particles. In the final lecture
we describe some recent developments centered around the theme of celestial symmetries.
We show that both bulk translation symmetry as well as the soft theorems imply the
existence of currents on the celestial sphere which constrain the celestial amplitudes.
We will see for example that Poincaré symmetry can be used to completely fix celestial
three point functions and strongly constrain 4-point functions [14], while subleading and
subsubleading soft theorems can be used to completely fix the leading OPE coefficients in
a collinear expansion of gluons and gravitons [15]. We finally show that celestial theories
contain an infinity of soft currents and compute their algebra in some examples [16, 17].
We have tried to give a self-contained overview of this rapidly growing field by choosing
a particular path through the subject. Many fascinating recent developments have been
left out. Explicit constructions of tree-level celestial amplitudes have appeared in [18,
11, 19, 20, 21, 22, 23]. Loop corrections were addressed in [24] while properties of bulk
scattering such as the double copy have been worked out in [25, 26]. Celestial symmetries
in both gravity and gauge theories have been discussed in [27, 28, 29, 30, 14, 31, 32,
33, 16, 17], while their constraints have been studied in [34, 35]. Analytic properties of
celestial 4-point functions in the complex boost-weight plane have been worked out in [36]
and conformal block expansions were computed in [37, 38, 39, 40]. Infrared divergences
and related aspects were discussed in [41, 42, 43, 44, 45, 46, 47, 48, 49]. We hope these
lectures provide a bridge between the earlier developments reviewed in [10] and more
recent results.

2 Soft theorems and asymptotic symmetries

2.1 Soft theorems
In this section we introduce a universal1 relation obeyed by scattering amplitudes in any
theory with massless particles. For simplicity, we focus on tree-level (classical) scattering.
In gravity and gauge theory, the scattering of high-energy charged particles is accompanied
by radiation. The radiation can be described as a collection of quanta (eg. photons,
gravitons,...) of different energies. When the energy carried away by one such quantum
is small, the scattering amplitude factorizes

lim A±
 (0)± (1)±

n+1 (q) = Sn + S n + O(ω) An , (1)
ω→0

where An+1 is a scattering amplitude of n generic particles of four-momenta p1 , ..., pn and

one massless particle of four-momentum q = (ω, ~q) and positive or negative helicity. An
is the same scattering amplitude in the absence of the massless particle. This limit is
illustrated in figure 1 and will be referred to as the soft limit.
(0)± (1)±
Sn and Sn are the leading and subleading soft factors respectively, which take
the form [50, 51, 52]
1
Universal here means independent of the nature of other particles involved in the scattering process.

2
Figure 1: The soft limit relates an amplitude with a low-energy massless particle to the
same amplitude without the massless particle.
n n
κ X (pk · ε± (q))
2
iκ X ε± (q) · pk √
Sn(0)± = , Sn(1)± = − q · Jk · ε± (q) , κ= 32πG
2 k=1 pk · q 2 k=1 pk · q
(2)
in gravity2
n n
X pk · ε± (q) X q · Jk · ε± (q)
Sn(0)± = Qk , Sn(1)± = −i Qk (3)
k=1
pk · q k=1
pk · q

in quantum electrodynamics (QED). G and Qk are Newton’s constant and the charges of
the n particles respectively. We expressed the polarization tensor ε±
µν of the graviton as
3

ε± ± ±
µν (q) = εµ (q)εν (q), (4)
where ε±
µ (q) is the polarization of a helicity-1 particle obeying

ε± (q) · q = 0, ε± (q) · ε± (q) = 0, ε± (q) · ε̄ ± (q) = 1. (5)

Jk is the total angular momentum of particle k. One sometimes works in units where
√
8πG = 1, κ = 32πG = 2. (6)
Notice that the soft theorem (1) captures the behavior of the scattering amplitude in
an expansion around ω = 0. The leading term in (1) has a simple pole at ω = 0 which can
be understood by considering the Feynman diagrams contributing to the scattering of n+1
particles, as shown in figure 2. In particular, as ω → 0, the leading order contribution
comes from diagrams where the massless particle attaches to an external line. In this
limit, an internal propagator goes on-shell and the amplitude develops a pole in q
" n #
X Vk (ε, pk )
lim An+1 (q) = −i + O(ω 0 ) An , (7)
ω→0 2p k ·q
k=1
± 2
(2)± k ·ε )
Pn
2
In gravity there is also a sub-subleading soft theorem [52] with Sn = − κ4 k=1 (q·Jq·p k
.
3
We pick a gauge such that the graviton is transverse and traceless, q µ εµν = q ν εµν = εµµ = 0.

3
Figure 2: In the soft limit, the amplitude will include contributions from Feynman dia-
grams where the soft particle attaches to external and internal lines. Diagrams where the
soft particle attaches to an internal line are subleading in the soft limit.

where Vk (ε, pk ) is the leading term as ω → 0 in the (momentum-space) coupling at vertex

k. The remaining diagrams, where the massless particle attaches to an internal line remain
finite as ω → 0.
The analysis of subleading terms in the (tree-level) soft expansion was carried out
explicitly in gauge theory [51] and, more recently in gravity using on-shell amplitudes
techniques [52] . The “brute-force” computation is lengthy and subtle,4 yet a number of
apparently miraculous cancellations yield the final result (1) universal in gravity [53, 54],5
with the subleading soft graviton factor taking the simple form in (2).
It is often the case in physics that simple answers found as a result of complicated
calculations point towards an underlying symmetry of the theory. Indeed, as we will review
in section 2.5, the subleading soft graviton theorem is nothing but a consequence of an
infinite-dimensional symmetry of the S-matrix [12] which enhances the Lorentz symmetry
discussed before. Moreover, a certain mode of the soft graviton will be identitfied with the
generator of this symmetry in section 2.6 by recasting the subleading soft graviton theorem
as the Virasoro-Ward identities of an insertion of the stress-tensor in a two-dimensional
(2D) conformal correlation function [13].

2.2 Penrose diagram of Minkowski space

These lectures are about physics at “infinity”. Penrose diagrams are a convenient tool in
studying this as they preserve the causal structure of spacetime while distorting distances.
In particular, infinite distances can become finite with “infinity” coinciding with the
boundary of a finite region. In this section we review how this works in Minkowski space.
4
This is not only because of many sources of subleading corrections coming from both classes of
diagrams in figure 2, but also because at subleading order in the soft expansion, momenta of other
external particles have to be deformed to obey overall momentum conservation.
5
The subleading soft photon theorem may receive non-universal corrections from a short list of oper-
ators [54, 55].

4
 Figure 3: Penrose diagram of Minkowski space.

The Minkowski metric takes the form

ds2 = −dt2 + d~x2 = −dt2 + dr2 + r2 dΩ22 , (8)

where
dΩ22 = dθ2 + (sin θ)2 dϕ2 (9)
is the metric on the unit two-sphere. It will be convenient to introduce retarded and
advanced coordinates u, v
u = t − r, v = t + r, (10)
and coordinates (z, z̄) related to the angular coordinates (θ, φ) by a stereographic projec-
tion
θ θ
z = cot eiϕ , z̄ = cot e−iϕ . (11)
2 2
In retarded coordinates (u, r, z, z̄) the metric (8) becomes
2
ds2 = −du2 − 2dudr + 2r2 γzz̄ dzdz̄, γzz̄ = (12)
(1 + z z̄)2

and similarly, in advanced coordinates (v, r, z, z̄)

ds2 = −dv 2 + 2dvdr + 2r2 γzz̄ dzdz̄. (13)

The asymptotic structure of (8) can be understood by introducing coordinates (T, R)

related to (t, r) by

u = tan U, v = tan V, T = U + V, R = V − U, (14)

5
Figure 4: Penrose diagram of Minkowski space where each pair of points represents a
two-sphere. Massive particles come in from i− and go out at i+ , while massless particles
enter and exit spacetime at I ± .

in which case (8) reduces to

ds2 = Ω2 (T, R) −dT 2 + dR2 + 2r2 (R, T ) sin2 Rγzz̄ dzdz̄ ,

1 1 (15)
Ω−2 (T, R) = 4 cos2 (T − R) cos2 (T + R).
2 2
In the original coordinates, Minkowski space is covered by r > 0, −∞ < u < v < ∞,
therefore in the new coordinates (14), − π2 < U < V < π2 and 0 < R < π. This is shown
in figure 3.
Exercise 2.1. a) Plot the lines of constant r and t in the (R, T ) plane.
r−r0
b) Plot the lines of t
for different values of r0 in the (R, T ) plane.
It will be convenient to unfold this diagram to represent antipodal points on the
spheres. Future null infinity (I + ) is defined by taking r → ∞ for fixed u, while past
null infinity (I − ) is reached by taking r → ∞ for fixed v. In a free theory, outgoing
massless particles follow lines of unit slope and cross a point on the sphere at I + at some
retarded time u. This is illustrated in figure 4. Similarly, incoming massless particles
start at I − . Massive particles never reach I ± , but only past and future timelike infinities
i∓ (t → ∓∞).

2.3 Asymptotically flat spacetimes

Asymptotically flat spacetimes have the same causal structure as Minkowski space at
infinity. An asymptotically flat spacetime admits an expansion in powers of r−1 around

6
the Minkowski metric (12) at I +

ds2 = − du2 − 2dudr + 2r2 γzz̄ dzdz̄,

2mB 2 (16)
+ du + rCzz dz 2 + rCz̄z̄ dz̄ 2 + 2guz dudz + 2guz̄ dudz̄ + ....
r
Solving the Einstein equations6
1 M
Rµν − gµν R = Tµν (17)
2
order by order in a large-r expansion7 one finds
1 1 2
guz = Dz Czz + Czz Dz C zz + Nz + O(r−2 ). (18)
2 6r 3r
Here mB and Nz are the Bondi mass aspect and angular momentum aspect respectively,
while
Nzz = ∂u Czz (19)
is the outgoing news tensor. They are all functions of (u, z, z̄).
mB , Czz , Nz are not independent. They are related by the constraint equations8 of
(17). The uu constraint gives
1 1 1 M (2) 1
∂u mB = Dz2 N zz + Dz̄2 N z̄z̄ − Tuu − Nzz N zz , (20)
4 4 2 4
while the uz constraint reduces to
1 1
∂u Nz = Dz Dz2 C zz − Dz̄2 C z̄z̄ − Tuz
M
+ ∂z mB + Dz ∂u (Czz C zz )


4 16 (21)
1 zz 1 1
− N Dz Czz − Nzz Dz C zz − Dz (C zz Nzz − N zz Czz ) .
4 4 4
We defined
M (2)
Tuu = lim r2 Tuu
M
. (22)
r→∞

Exercise 2.2. **Optional** Verify (20) and (21).

The analogous analysis near I − can be found in [12].

2.4 Asymptotic symmetries

The asymptotic symmetry group of (16) has beeen proposed to be the extended BMS+
group in [56]. This is generated by vector fields ξ + that preserve the asymptotic fall-off
of (16) with r

Lξ+ gur = O(r−2 ), Lξ+ guz = O(1), Lξ+ gzz = O(r), Lξ+ guu = O(r−1 ). (23)
6
We set κ = 2.
7
The leading term in the matter stress tensor is O(r−2 )
8
These are the components of (17) in a direction tangent to I + .

7
Such vector fields are of the form
 u  +z u 1 u
ξ+ = 1 + Y ∂z − Dz̄ Dz Y +z ∂z̄ − (u + r)Dz Y +z ∂r + Dz Y +z ∂u + c.c.
2r 2r 2 2 (24)
+ 1 z + z̄ +

z +
+ f ∂u − D f ∂z + D f ∂z̄ + D Dz f ∂r ,
r
where f + (z, z̄) is an arbitrary function on S 2 and Y + (z, z̄) is a conformal Killing vector
on S 2
∂z̄ Y +z = 0. (25)
Exercise 2.3. Use (23) and (16) to derive (24).
One easy way to see that (25) ought to hold is to notice that under Lorentz transfor-
mations,
LY + gzz = 2r2 γzz̄ ∂z Y z̄ + O(r). (26)
Imposing that (23) are obeyed leads to (25).
Globally, (25) admits six solutions
1 i
Y13z = − (1 + z 2 ), Y23z = − (1 − z 2 ),
Y12z = iz,
2 2 (27)
z z 1 2 z i 2
Y03 = z, Y01 = − (1 − z ), Y02 = − (1 + z )
2 2
corresponding to the three Lorentz rotations and three boosts (see appendix A). Locally,
there are infinitely many solutions Y z ∝ z n .
For the remainder of this section, we restrict to the subgroup of symmetries of (16)
generated by (24) with f = 0. In this case, the vector fields (24) that map I + to itself are
u
ξ+ = Y +z ∂z + Dz Y +z ∂u + c.c.. (28)
I + 2
The infinitesimal BMS+ transformations (28) act on the metric components as follows
1
Dz Y +z + Dz̄ Y +z̄ (u∂u − 1) Czz + LY + Czz − uDz3 Y +z ,


δY + Czz =
2 (29)
u
Dz Y +z + Dz̄ Y +z̄ ∂u Nzz + LY + Nzz − Dz3 Y +z .


δY + Nzz = ∂u δY + Czz =
2
Upon quantization, (29) imply the existence of “charges”9 under which an outgoing Fock
state transforms as
Q+ (Y + )|outi = −iδY + |outi, (30)
where
Q+ (Y + ) = Q+ +
H + QS (31)
and [57]
Z
1
Q+
H = dud2 zγzz̄ (uDz Y z Nzz N zz + Y z Dz (Czz N zz ) + 2Y z Czz Dz N zz ) + c.c. + matter,
4 I+ Z
1
Q+
S =− dud2 zDz3 Y +z uN zz̄ + c.c..
2 I+
(32)
9
We haven’t shown these are conserved yet.

8
Using the canonical commutation relations [58]

[Nz̄z̄ (u, z, z̄), Cww (u0 , w, w̄)] = −2iγzz̄ δ (2) (z − w)δ(u − u0 ), (33)

one can show that

X ωk 
Q+
H |outi = −i +z
LY + (zk ) − Dz Y (zk )∂ωk |outi. (34)
k∈out
2

Similar formulas hold near I − .

2.5 Recovering the Virasoro symmetry from the soft theorem

An independent action of BMS+ and BMS− on I + and I − leads to an ambiguity in defining
scattering in AFS. In particular, upon specifying data at I + , the S-matrix provides a map
between in and out states up to a BMS transformation.
A solution to this problem was proposed in [59] where it was shown that the grav-
itational scattering problem in AFS becomes well defined upon imposing the antipodal
matching conditions

f (z, z̄) +
= f (z, z̄) −
, Y + (z, z̄) +
= Y − (z, z̄) −
. (35)
I− I+ I− I+

Here, points on the sphere at I−+ are antipodally related to points at I+− , (z, z̄) +
=
I−

(− z̄1 , − z1 ) −
. Moreover, upon imposing the boundary condition10
I+

Nz (z, z̄) +
= Nz (z, z̄) −
, (36)
I− I+

the charges at I+− and I−+ obey

Q+ = Q− . (37)
It then makes sense to study the constraint imposed by conservation of Q on the S-matrix

hout|Q+ S − SQ− |ini = 0. (38)

Using the split (31) into soft and hard parts, (38) becomes
− −
hout|Q+ +
S S − SQS |ini = −hout|QH S − SQH |ini (39)

Since QS is linear in the news, it can be shown to correspond to a graviton mode. In

particular, one finds [12]
Z ∞
(1) i h
out†
i
Nz̄z̄ = duuNz̄z̄ = ˆz̄z̄ lim (1 + ω∂ω ) aout
− (ωx̂) − a+ (ωx̂) (40)
−∞ 4π ω→0
10
Using the constraint (21) the superrotation charges can be put into the simpler form Q+ (Y + ) =
2
R
d z (Yz̄ Nz + Yz Nz̄ ) .

9
and hence QS picks out a particular subleading11 soft graviton mode. Using the subleading
soft relation (2) under the assumption that all other external states are massless scalars,
we find the LHS of (39) reduces to
Z
+ − i
hout|QS S − SQS |ini = − d2 zDz3 Y z ˆz̄z̄ S (1)− hout|S|ini
4π
X  ωk  (41)
= −i Y zk ∂zk − Dzk Y zk ∂ωk hout|S|ini.
k∈in,out
2

In the second line, we integrated by parts and used the parameterizations of momenta
ω
q(ω, z, z̄) = ω q̂(z, z̄) = (1 + z z̄, z + z̄, −i(z − z̄), 1 − z z̄) (42)
1 + z z̄
for which the subleading soft factor of a negative helicity graviton becomes
X  (z − zk )(1 + z z̄k ) (z − zk )2

(1)−
S = ω k ∂ ωk + ∂zk (43)
(z̄k − z̄)(1 + zk z̄k ) z̄k − z̄
k

The RHS of (41) is nothing but the action of the hard charge QH on scalar asymptotic
states. We conclude the subleading soft theorem implies conservation of the charges (31),
hence the enhancement of Lorentz symmetry to Virasoro.

2.6 A 2D stress tensor for 4D gravity

(41) implies that gravity in AFS has a remarkable feature: the existence of a subleading
soft graviton mode whose insertions in the quantum gravity S-matrix behaves like the
stress tensor of a 2D CFT! To see this, we simply set
1
Y zk = (44)
z − zk
in (41). Upon defining12
Z
1 (1)
Tzz ≡ i d2 w Dw2 Dw̄ Nw̄w̄ , (45)
z−w
(41) reduces to
n
" #
X ĥk Γzzkk zk 1
hTzz O1 ...On i = + ĥk + ∂z hO1 ...On i. (46)
k=1
(z − zk )2 z − zk z − zk k
This is nothing but the Ward identity of a stress tensor in a conformal field theory on a
curved background. Note however that the weight13 ĥk = − 12 ωk ∂ωk is a differential opera-
tor which acts non-diagonally on S-matrix elements in a basis of momentum eigenstates.
In the next section we will introduce a new basis of asymptotic states which diagonalize
the action of ĥk . The scattering problem in AFS will be then reformulated in terms of an
observable living on the celestial sphere: the celestial amplitude.
11
1 + ω∂ω projects out the leading soft pole.
12
Note that this operator is directly related to the soft charges Q± z 1
S evaluated at Y = z−w .
13 s −ω ∂
For external states of spin sk , the weights generalize to ĥk = k 2k ωk and (46) gets corrected by a
spin connection term, see [13] for the general formula.

10
3 Celestial amplitudes
In this section we introduce a basis for scattering in AFS which diagonalizes asymptotic
boosts as opposed to momentum generators. We show how to formulate scattering in this
basis, with celestial amplitudes defining observables living on the sphere at infinity. We
illustrate this construction by computing the celestial amplitude of two massless and one
massive scalars.

3.1 Conformal primary wavefunctions

Scalar conformal primary wavefunctions are solutions to the wave equation

∇2 − m2 Ψ = 0,


(47)

which are “highest weight” with respect to the Lorentz SO(1, 3). We start by identi-
fying the associated highest weight states, then impose they are solutions to (47). A
representation of the Lorentz generators is

J µν = Lµν + S µν , (48)

where
Lµν = i (xµ ∂ ν − xν ∂ µ ) (49)
is the orbital angular momentum generator and S µν is the spin generator. For simplicity,
in this section we focus on scalars in which case S µν = 0.
(49) consist of rotations

J1 = −i(x2 ∂x3 − x3 ∂x2 ), J2 = i(x1 ∂x3 − x3 ∂x1 ), J3 = −i(x1 ∂x2 − x2 ∂x1 ) (50)

and boosts

K1 = i(x0 ∂x1 + x1 ∂x0 ), K2 = i(x0 ∂x2 + x2 ∂x0 ), K3 = i(x0 ∂x3 + x3 ∂x0 ). (51)

These generators obey the standard Lorentz algebra

[Ji , Jj ] = iijk Jk ,
[Ki , Kj ] = −iijk Jk , (52)
[Ji , Kj ] = iijk Kk ,
while
J3 + iK3 J1 + iK1 + i(J2 + iK2 ) J1 + iK1 − i(J2 + iK2 )
L0 = , L−1 = , L1 = ,
2 2 2
J3 − iK3 J1 − iK1 − i(J2 − iK2 ) J1 − iK1 + i(J2 − iK2 )
L̄0 = − , L̄−1 = − , L̄1 = −
2 2 2
(53)

obey the SL(2, C) commutation relations

[Lm , Ln ] = (m − n)Lm+n , [L̄m , L̄n ] = (m − n)L̄m+n . (54)

11
Exercise 3.1. Check (52) and (54).
We now notice that
1
Ψ∆ ∝ (55)
(x0 + x3 ) ∆
obeys
(L0 + L̄0 )Ψ∆ = ∆Ψ∆ , (L0 − L̄0 )Ψ∆ = 0 (56)
as well as
L1 Ψ∆ = L̄1 Ψ∆ = 0. (57)
In other words, (55) diagonalizes boosts in the x3 direction and obeys a highest weight
condition (57).
One could have done the same analysis starting with a set of rotated bulk Lorentz
generators,
Ji0 = Rij Jj , Ki0 = Rij Kj , (58)
where    
cos ϕ cos θ sin ϕ cos θ − sin θ n̂1
R=  − sin ϕ cos ϕ 0  ≡ n̂2  .

cos ϕ sin θ sin ϕ sin θ cos θ n̂3

Multiplication by an arbitrary function f of the Lorentz invariant x2 will preserve both

the eigenvalue and highest weight conditions (56), (57), hence in general, a highest weight
solution diagonal with respect to K30 will be
f (x2 )
Ψ∆ (q̂; x) = , q̂ = (1, n̂3 ) = q̂(z, z̄), (59)
(q̂ · x)∆
with q̂(z, z̄) in (42).
Finally, we require that (59) obeys the wave equation. Plugging (59) into (47), we
find the following differential equation for f

4x2 f 00 (x2 ) − 4(∆ − 2)f 0 (x2 ) − m2 f (x2 ) = 0. (60)

The solutions to (60) are linear combination of Bessel functions (of first kind)
√ ∆−1 h √ √ i
f (x2 ) = −x2 c1 I∆−1 (m x2 ) + c2 I−∆+1 (m x2 ) . (61)

Imposing that (61) decays to 0 as x2 → ∞ picks out the linear combination proportional
to the Bessel function of second kind
√ ∆−1 √
2 2
f (x ) ∝ −x K∆−1 (m x2 ). (62)

We conclude that up to normalization, the massive conformal primary wavefunctions take

the form14 √
∆−1
−x2 √
Ψ∆ (q̂; x) ∝ K ∆−1 (m x2 ). (63)
(q̂ · x)∆
14
An i prescription distinguishes between in and out states [60].

12
 Under Lorentz transformations, both q̂ and x transform linearly
1/2
µ 0 0 ∂~z0
q̂ (z , z̄ ) = Λµν q̂ ν (z, z̄),
∂~z (64)
µ0
x = Λµν xν

hence (63) obeys

∂~z0 −∆/2
Ψ∆ (Λµν x; ~z0 (~z)) = Ψ∆ (x; ~z). (65)
∂~z
Exercise 3.2. Show that the conformal primary wavefunctions obey (65).

In the next section we give an alternative derivation of (63).

3.2 Milne slicing

The conformal compactification of Minkowski space in section 2.2 obscures one aspect of
scattering in AFS: all massive particles enter (exit) spacetime at a point, i− (i+ ), so how
are we supposed to distingush between different asymptotics? The key is to resolve past
and future timelike infinities by introducing the new coordinates [61, 62]

t2 − r2 = τ 2 ,
(66)
ρτ = r.

In (τ, ρ, z, z̄) coordinates, (8) becomes

dρ2
 
2 2 2
ds = −dτ + τ 2
+ 2ρ γzz̄ dzdz̄ = −dτ 2 + τ 2 dsH3 .
2
(67)
1+ρ

We learn that slices of constant τ correspond to hyperboloids of radius τ , while constant

−1/2
r2

r p
ρ = ± 1− 2 corresponds to constant r/t. Since t = τ 1 + ρ2 , as τ → ∞,
t t
t → ∞ for fixed (ρ, z, z̄). We illustrate the foliations of the past and future light-cones
(also known as Milne wedges) with H3 slices in figure 5.
Parameterizing the trajectory of massive particle of constant momentum p~ and energy
E by
t
~r = ~r0 + p~, (68)
E
we find that as t → ∞,
|~p| ~r p~
ρ→ , → . (69)
m r p
Hence at late times, massive particles asymptote to fixed (ρ, z, z̄), or equivalently points
on the unit hyperboloid at i+ . Similarily, the Rindler wedges can be foliated with dS3
slices. This slicing is easily obtained by letting

τ = ie
τ, ρ = −ie
ρ (70)

13
Figure 5: Minkowski space split into four regions: the past and future lightcones are
covered by H3 slices while the causally disconnected Rindler regions are covered by dS3
slices.

in (67) in which case

ρ2
 
2 2 2 de 2
ds = de
τ + τe + 2e
ρ γzz̄ dzdz̄ . (71)
1 − ρe2
The proper, orthocronous Lorentz group then acts as the group of isometries on the H3
slices for t > 0. To find solutions to (47) that preserve slices of constant τ it is convenient
to express ∇2 with respect to the coordinates in (67) .
Exercise 3.3. Using
1 √
∇2 Ψ = √ ∂µ (g µν g∂ν Ψ) (72)
g
show that (47) reduces to
 
1

S 2
3ρ + 2 ∂ρ + ρ ρ + 1 ∂ρ − ρτ 3∂τ + τ ∂τ + 2 2 Ψ = m2 Ψ.
2 2


2 2
2
(73)
ρτ ρτ
Setting
ρ = sinh η, (74)
(73) becomes
 
1 2 ∂τ 2 S 2
Ψ = m2 Ψ.


∂ + 2 coth η∂η − 3 − ∂τ + (75)
τ2 η τ 2
sinh ητ 2

This equation can be solved by separation of variables [63]

Ψ = φp,l (η)ϕp (τ )Ylm (z, z̄) (76)

14
where
 
2 −l(l + 1) 2
∂η + 2 coth η∂η + − p φp,l (η) = 0,
sinh2 η
(77)
p2
 
∂τ 2 2
−3 − ∂τ + 2 − m ϕp (τ ) = 0
τ τ
and
S 2 Ylm = −l(l + 1)Ylm . (78)
Note that p is a free parameter which cancels in (75). We recognize the first equation
in (77) as the massive wave equation in AdS3 while the second equation has two linearly
independent solutions
I√1+p2 (mτ ) K√1+p2 (mτ )
ϕp (τ ) = , . (79)
τ τ
Choosing the second solution as it decays at τ → ∞, we recover the τ dependence in (63)
upon identifying p
∆ − 1 = 1 + p2 (80)
Using (80) it is a standard exercise in AdS3 to show that the first equation in (77) can
be written in terms of the SL(2, C) generators (53) (upon the appropriate coordinate
transformation, see appendix A)

4L20 − 2L−1 L1 − 2L1 L−1 φp,l = 4L̄20 − 2L̄−1 L̄1 − 2L̄1 L̄−1 φp,l = ∆(∆ − 2)φp,l . (81)

Using the SL(2, C) commutation relations (54) and imposing that L1 φp,l = 0, (81) reduces
to (56), (57) and we recover the solutions (63).
Note that (77) admit more general solutions which don’t obey the highest weight
condition (57). These can be used to construct the unitary principal series representations
of SL(2, C). This complementary calculation is detailed for the dS slicing of Minkowski
space (71) in [63].

3.3 Integral representation

The conformal primary wavefunctions (63) admit the Fourier representation
Z
Ψ∆ (x; ~z) = d3 p̂G∆ (p̂; ~z)eimp̂·X , (82)
H3

where the momenta

1
1 + y 2 + ww̄, w + w̄, −i(w − w̄), 1 − y 2 − ww̄


p̂(y, w, w̄) = (83)
2y
are in one-to-one correspondence with points on the unit hyperboloid at i+ and
 ∆
y
G∆ (y, w, w̄; z, z̄) = (84)
y 2 + |z − w|2

15
is the bulk-to-boundary propagator in AdS3 [64]. As they are weighted integrals of plane
waves, they automatically solve the wave equation. That they transform as (65) under
SL(2, C) follows from the transformation property
−∆/2
∂~z0
G∆ (p̂0 ; q̂ 0 ) = G∆ (p̂; q̂) (85)
∂~z

of (84). The Fourier transform (82) can be evaluated to recover the conformal primary
wavefunctions (63).

3.4 Massless particles

The massless conformal primary wavefunctions can be obtained from (63) by taking the
m → 0 limit (assuming Re(∆) > 1). Using the expansion

K∆−1 (x) = x−∆ 2∆−2 Γ(∆ − 1)x + O(x2 )

(86)
 
∆ −∆ −1 −2−∆ Γ(1 − ∆) 2
+x 2 Γ(1 − ∆)x + 2 x + O(x ) ,
∆

we find
1
ϕ∆ (q̂; x) = lim Ψ∆ (q̂; x) ∝ . (87)
m→0 (q̂ · x)∆
The integral representation of the massless conformal primary wavefunctions can also be
derived from (82) by taking the limit m → 0 for fixed ω ≡ m/(2y). In this limit, (84)
becomes proportional to [65]

π y∆
lim G∆ (y, z, z̄; w, w̄) = y 2−∆ δ (2) (z − w) + + O(y 4−∆ ) (88)
y→0 ∆−1 (w − z)2∆

and upon evaluating the integral in (82) we recover (87).15

Exercise 3.4. Evaluate (82) in the limit (88) and show that the massless conformal
wavefunction indeed reduces to (87).

3.5 A conformal primary basis for scattering in AFS

Using the map (82) from plane wave solutions to conformal primary solutions of the scalar
wave equation implies that the S-matrix elements in the conformal primary basis take
the form n Z
Y d3 p̂i
A(∆i , zi , z̄i ) =
e G∆i (p̂i ; zi , z̄i )A(i mi p̂i ), (89)
i=1 H3
p0i
where i = ±1 depending on whether the particle is in- or outgoing. (65) implies that (89)
transform like conformal correlation functions of primary operators under the 2D global
15
In fact there are two contributions in the massless limit: the conformal primary (87) and its shadow.
Since these solutions are not linearly independent, [60] argued it is sufficient to restrict to (87).

16
conformal symmetry action
n −∆i /2
e i , ~zi0 (~zi )) =
Y ∂~zi0
A(∆ A(∆
e i , ~zi ). (90)
i=1
∂~zi

This is the definition of a celestial amplitude.

It can be shown that [60]
• The massive conformal primary wavefunctions (82) form a basis of solutions to the
wave equation provided ∆ = 1+iλ, λ ≥ 0. For such ∆, these solutions are complete,
linearly independent and orthogonal.16

• The massless conformal primary wavefunctions form a basis of solutions to the wave
equation provided ∆ = 1 + iλ, λ ∈ R.

3.6 Example: celestial 3-point scattering of 2 massless and 1

massive scalars
In this section we compute the celestial amplitude for two massless and one massive scalar
[66]. We start with the momentum space 3-point interaction

A(p̂i ) = gδ (4) (ω1 q̂1 + ω2 q̂2 − mp̂). (91)

The associated celestial amplitude is then

2 Z ∞ Z ∞ Z  ∆3
Y dy y
A(∆
e i , zi , z̄i ) = g dωi ωi∆i −1 2
dw
i=1 0 0 y3 y + |z3 − w|2
2
(92)
(4)
× δ (ω1 q̂1 + ω2 q̂2 − mp̂).

Using the parameterizations of momenta

q̂ = (1 + z z̄, z + z̄, −i(z − z̄), 1 − z z̄) ,

1 (93)
1 + y 2 + ww̄, w + w̄, −i(w − w̄), 1 − y 2 − ww̄


p̂ =
2y
and evaluating the integrals over y, w ~ and ω2 , (92) reduces to
2∆2 +∆3 −4 Z ∞
gm ω ∆1 −∆2 +∆3 −1
A(∆
e i , zi , z̄i ) = dω . (94)
22∆2 −∆3 −1 |z12 |2∆2 −2∆3 0 (m2 |z23 |2 + 4|z12 |2 |z13 |2 ω 2 )∆3
Exercise 3.5. Check this.
Upon a change of variables, the remaining integral becomes proportional to the stan-
dard integral Z 1
Γ(α)Γ(β)
dttα−1 (1 − t)β−1 = ≡ B(α, β) (95)
0 Γ(α + β)
16
Notice that in this case, p2 (the effective “mass” on the AdS3 slices in (77)) will be complex

17
and we conclude
C(∆1 , ∆2 , ∆3 )
A(∆
e i , zi , z̄i ) = ,
|z12 |∆1 +∆2 −∆3 |z13 |∆1 +∆3 −∆2 |z23 |∆2 +∆3 −∆1
(96)
m∆1 +∆2 −4
 
∆12 + ∆3 −∆12 + ∆3
C(∆1 , ∆2 , ∆3 ) = g ∆1 +∆2 B , .
2 2 2

4 Symmetries
Celestial amplitudes obey a wealth of constraints which we discuss next.

4.1 Poincaré symmetry

By construction, they transform covariantly under the Lorentz group. As shown in [29]
the Lorentz generators act on operators Oh,h̄ (z, z̄) on the celestial sphere as
L0 = 2(z∂z + h), L− = ∂z , L+ = z 2 ∂z + 2zh,
(97)
L̄0 = 2(z̄∂z̄ + h̄), L̄− = ∂z̄ , L̄+ = z̄ 2 ∂z̄ + 2z̄ h̄.
Global conformal symmetry of celestial amplitudes is the constraint that
LI AeN = L̄I AeN = 0, (98)
where AN is an N-point celestial amplitude and
N
X N
X
LI = LI,n , L̄I ≡ L̄I,n . (99)
n=1 n=1

(98) is a familiar property of correlation functions in 2D CFT.

Additionally, bulk translation invariance implies that
N
X
Pµ AeN = 0, Pµ ≡ Pµ,n . (100)
n=1

The celestial translation generators of massless particles are the weight shifting operators
Pµ,i = i q̂µ (zi , z̄i )e∂∆i . (101)
To see this, consider the action of momenta on momentum space amplitudes
P̂i A(q1 , ..., qn ) = i ωi q̂i A(q1 , ..., qn ) (102)
and transform it to the conformal primary basis by taking a Mellin transform
n Z ∞ 
∆ −1
Y
j
Pi A(∆
e 1 , ...∆n ) = dωj ωj i ωi q̂i A(q1 , ..., qn )
j=1 0
n Z ∞ Z ∞
∆ −1 (103)
Y
= dωj ωj j dωi ωi∆i +1−1 i q̂i A(q1 , ..., qn )
j=1 0 0
j6=i

= i q̂i A(∆
e 1 , ..., ∆i + 1, ..., ∆n ).

18
We conclude that for massless scattering, (100) relates celestial amplitudes involving op-
erators of shifted weights

AeN (∆1 + 1, ∆2 , ..., ∆N ) + AeN (∆1 , ∆2 + 1, ..., ∆N ) + ...AeN (∆1 , ..., ∆N + 1) = 0. (104)

For massive scalars, (101) are replaced by

∂z̄ q µ ∂z + ∂z q µ ∂z̄ q µ ∂z ∂z̄ ∆q µ ∂∆

  
µ m µ −∂∆
P = ∂z ∂z̄ q + + e + e . (105)
2 ∆−1 (∆ − 1)2 ∆−1

(105) is determined by imposing the on-shell condition

Pµ P µ = −m2 , (106)

as well as the Poincaré algebra

[Pµ , Pν ] = 0, [Mµν , Pρ ] = ηµρ Pν − ηνρ Pµ . (107)

Exercise 4.1. Verify (105) satisfy (106) and (107).

The momentum generators for spinning particles can be found in [31]. It is interesting
to notice that in addition to off-diagonal terms in dimension ∆ = h + h̄, they contain
off-diagonal terms in spin J = h − h̄.

4.2 Constraints from Poincaré symmetry

(97) and (101) imply that any celestial 4-point function can be put into the form

Ae4 = Khi ,h̄i (zi , z̄i )δ(z − z̄)f hi ,h̄i (z, z̄), (108)

where
4
h/3−hi −hj h̄/3−h̄i −h̄j
Y
Khi ,h̄i (zi , z̄i ) = zij z̄ij (109)
i<j=1

and
z13 z24 z̄13 z̄24
z= , z̄ = (110)
z12 z34 z̄12 z̄34
are the conformally invariant cross-ratios in 2D. For massless scattering, (110) are related
to the bulk variables (Mandelstam invariants) by
t
z=− , s = −(p1 + p2 )2 , t = −(p1 + p3 )2 . (111)
s
Additionally, in the case of massless scattering (100) implies that
4
1 1
X
Khj + 1 ,h̄j + 1 (zi , z̄i )f hj + 2 ,h̄j + 2 (z, z̄) = 0. (112)
2 2
j=1

19
Exercise 4.2. Show that
4
X
Khj + 1 ,h̄j + 1 (zi , z̄i ) = 0. (113)
2 2
j=1

Since the conformally covariant factor is translationally invariant by itself, (112) can
be non-trivially obeyed if
1 1 1 1
f hi + 2 ,h̄i + 2 (z, z̄) = f hj + 2 ,h̄j + 2 (z, z̄), ∀i, j. (114)
By induction it can be shown that (114) implies that
4
X 4
X
hi ,h̄i β,Ji
f (z, z̄) = f (z, z̄), β= (hi + h̄i ) = ∆i . (115)
i=1 i=1

Exercise 4.3. Convince yourself of (115).

More generally, (100) will be obeyed by celestial amplitudes with both massive and
massless particles, provided Pµ,n are chosen appropriately. For example, consider the
three point amplitude of two massless and one massive scalar computed in section 3.6.
The same result can be easily recovered by considering the constraint
 
(m)
P1 + P 2 + P3 Ae3 (1, 2, 3(m) ) = 0, (116)
(m)
where P1 , P2 are the massless momentum generators in (101) while P3 is the massive
momentum (105). Global conformal invariance fixes
e 2, 3(m) ) = C(∆1 , ∆2 , ∆3 )
A(1, , (117)
|z12 |∆ 1 +∆ 2 +∆3 |z23 ||∆2 +∆3 −∆1 |z13 |∆1 +∆3 −∆1
hence (116) leads to the following recursion relations on the 3-point coefficients [14]
 2
∆12 (∆3 − 1)2

− C∆1 ,∆2 ,∆3 −1 + ∆3 (∆3 − 1)C∆1 ,∆2 ,∆3 +1 = 0,
4 4
(118)
42 (∆3 − 1)C∆1 ,∆2 +1,∆3 + m3 (∆3 − 1 − ∆12 )C∆1 ,∆2 ,∆3 −1 = 0,
41 (∆3 − 1)C∆1 +1,∆2 ,∆3 + m3 (∆3 − 1 + ∆12 )C∆1 ,∆2 ,∆3 −1 = 0,
where ∆12 ≡ ∆1 − ∆2 .
Exercise 4.4. Show that (118) are solved by
 
∆1 − ∆2 + ∆3 ∆2 − ∆1 + ∆3
C(∆1 , ∆2 , ∆3 ) = B , c∆1 ,∆2 ,∆3 , (119)
2 2
where
c∆1 ,∆2 ,∆3 −1 = c∆1 ,∆2 ,∆3 +1 ,
c∆1 +1,∆2 ,∆3 = c∆1 ,∆2 +1,∆3 , (120)
c∆1 +1,∆2 ,∆3 = c∆1 ,∆2 ,∆3 −1 .
The first constraint in (120) implies c is perodic in ∆3 of period 2, the second implies
periodicity of period 1 in ∆1 and ∆2 up to dependence of ∆1 + ∆2 while additionally,
P3 the
last constraint implies c is periodic of period 1 in ∆1 , ∆3 up to dependence of i=1 ∆i
in which case the period becomes 2. The periodic function is again set to a constant by
requiring the inverse Mellin transform is well defined.

20
4.3 Conformally soft symmetries
We saw in section 2.5 that the subleading soft theorem implies a conservation law asso-
ciated with the enhancement of Lorentz symmetries to an infinite-dimensional Virasoro
symmetry of the S-matrix. This is one example of a more general equivalence between
soft theorem and conservation laws associated with large gauge symmetries. For example,
it was shown in [7] that the leading soft photon theorem implies that soft photons behave
as U (1) currents. As such, their insertions into the S-matrix obey Ward identities of the
form
hJz O1 (ω1 , z1 , z̄1 )...On (ωn , zn , z̄n )i ≡ lim ωhO+ (ω, z, z̄)O1 (ω1 , z1 , z̄1 )...On (ωn , zn , z̄n )i
ω→0
n
X Qk
= hO1 (ω1 , z1 , z̄1 )...On (ωn , zn , z̄n )i
k=1
z − zk

(121)
Similarly, the leading soft gluon theorem is associated with a Kac-Moody symmetry gen-
erated by non-abelian currents obeying the Ward identities [67]
hJza O1 (ω1 , z1 , z̄1 )...On (ωn , zn , z̄n )i ≡ lim ωhO+,a (ω, z, z̄)O1 (ω1 , z1 , z̄1 )...On (ωn , zn , z̄n )i
ω→0
n
X 1
= hO1 (ω1 , z1 , z̄1 )...Tka Ok ...On (ωn , zn , z̄n )i.
k=1
z − zk
(122)
We would like to recast (121) and (122) as identities obeyed by celestial amplitudes.
Since the currents were constructed from low-energy limits of bulk photons and gluons,
while celestial operators involve integrals over photons and gluons of all energies, it is
a-priori not immediately obvious how to construct celestial currents. One hint is that
in conventional CFTd , currents saturate unitarity bounds and hence the dimension of a
spin-j current is constrained to be [68]
∆ = d + j − 2. (123)
In particular, positive helicity conformally soft photons and gluons should correspond to
operators of weights17 (h, h̄) = (1, 0), while negative helicity ones should have (h, h̄) =
(0, 1). They should be associated with abelian and non-abelian symmetries on the celestial
sphere.
The simplest way to show that this guess is indeed correct is to start with the Mellin
representation Z ∞
+
O∆ (z, z̄) = dωω ∆−1 O+ (ω, z, z̄) (124)
0
and notice that [69]
Z ∞
+
lim (∆ − 1)O∆ (z, z̄) = lim dω(∆ − 1)ω ∆−1 O+ (ω, z, z̄)
∆→1 ∆→1 0
Z ∞ (125)
=2 dωδ(ω)ωO+ (ω, z, z̄) = lim ωO+ (ω, z, z̄).
0 ω→0

17 ∆+J
The conformal weights are related to the conformal dimensions ∆ and the spin J by h = 2 , h̄ =
∆−J
2 .

21
In the last line we have used the identity18

lim |x|−1 = δ(x). (126)
→0 2

More generally
Z ω∗
+
lim (∆ + n)O∆ (z, z̄) = lim (∆ + n) dωω ∆−1 O+ (ω, z, z̄)
∆→−n ∆→−n 0
X Z ω∗
= lim (∆ + n) dωω ∆+k−1 Ok+ (z, z̄) (127)
∆→−n 0
k
= On+ (z, z̄),
where we expanded X
O+ (ω, z, z̄) = ω k Ok+ (z, z̄) (128)
k

for ω  ω∗ and assumed that insertions of O(ω, z, z̄) decay fast enough19 with energy in
which case the high-energy part of the Mellin integral will be free of poles in ∆ + n. We
conclude that the ∆ → −n limit of a celestial operators for n = −1, 0, 1, ... picks out the
O(ω n ) term in an expansion around ω = 0. For example, a subleading soft photon will
correspond to the celestial operator
+
lim ∆O∆ (z, z̄). (129)
∆→0

This infinity of soft currents has been studied in [70, 71, 16]. There exists a comple-
mentary tower of positive integer-dimension operators (also known as Goldstone modes),
canonically conjugate to the conformally soft modes above [72]. Their combined Ward
identities are expected to highly constrain celestial amplitudes, but a complete under-
standing of these symmetries and their implications remains an open problem. In the
next section we discuss a few instances in which soft celestial symmetries were used to
derive non-trivial properties of celestial amplitudes.

4.4 Applications
4.4.1 Celestial operator products of gluons
In this section we assume positive helicity gluons admit the holomorphic collinear expan-
sion
+,a +,b if abc +,c
O∆ (z ,
1 1z̄ )O∆2 (z ,
2 2z̄ ) ∼ C(∆1 , ∆2 )O∆ 1 +∆2 −1
(z2 , z̄2 ) + ..., (130)
1
z12
where ... include contributions from SL(2, C) descendants. The form of the OPE is fixed
by the leading soft theorem and SL(2, C) up to a coefficient C(∆1 , ∆2 ). We now show that
the subleading conformally soft gluon theorem determines this leading OPE coefficient up
to a normalization fixed by the leading soft gluon theorem [15].
18
This holds provided that x has compact support.
19
An exponential fall-off limω→∞ O(ω, z, z̄) ∼ e−ω will ensure this limit is well defined for any negative
integer ∆.

22
 The subleading soft gluon theorem can be recast as a “symmetry”20 under which
gluons transform as follows
±,a ±,c
δb O ∆ (z, z̄) = −(∆ − 1 ± 1 + z∂z )if abc O∆−1 ,
±,a ±,c (131)
δ̄b O∆ (z, z̄) = −(∆ − 1 ∓ 1 + z̄∂z̄ )if abc O∆−1 .

Acting with δ̄ on both sides of (130) leads and comparing the two sides, we deduce that
C(∆1 , ∆2 ) obey the recursion relation

(∆1 − 2)C(∆1 − 1, ∆2 ) = (∆1 + ∆2 − 3)C(∆1 , ∆2 ), (132)

which has the unique21 solution

C(∆1 , ∆2 ) = B(∆1 − 1, ∆2 − 1). (133)

Exercise 4.5. Write down the action of δ̄ on (130) and deduce (132).

4.4.2 Holographic symmetry algebras

(130) can be generalized to include contributions from all SL(2, R)R descendants. One
finds
∞
−if abc X z̄ n
+,a
O∆ +,b
(z1 , z̄1 )O∆ (z2 , z̄2 ) ∼ B(∆1 − 1 + n, ∆2 − 1) 12 ∂¯n O∆
+,c
1 +∆2 −1
(z2 , z̄2 ). (134)
1 2
z12 n=0 n!

(134) follows by resuming contributions from the right-moving descendants through the
OPE block [16]
+,c
−if abc 1 O∆P (z2 , z̄2 + tz̄12 )
Z
+,a +,b
O∆1 (z1 , z̄1 )O∆2 (z2 , z̄2 ) ∼ dt 2−∆1 . (135)
z12 0 t (1 − t)2−∆2
It is interesting to consider the algebra of the soft operators defined in section 4.3. First
notice that if ∆1 , ∆2 ∈ {1, 0, −1, ...}, ∆1 + ∆2 − 1 ∈ {1, 0, −1, ...} and the algebra of
soft operators closes. Then mode expanding such an operator on the right, we find for
k = 1, 0, −1, ...
+,a
X Ok+,n
+,a (z)
lim Ok+ (z, z̄) = lim k−1 . (136)
→0 →0 n+ 2
n z̄
Defining
+,a +,a
Rk,a (z, z̄) = lim Ok+ (z, z̄), Rnk,a (z) = lim Ok+ (z), (137)
→0 →0

we see that for

k−1 1−k
≤n≤ , (138)
2 2
Rnk,a (z) organize into (2 − k)-dimensional SL(2, R)R representations as

∂¯2−k Rk,a (z, z̄) = 0. (139)

20
These have not been shown to be associated with asymptotic charges.
21
By Wieland’s theorem, see appendix E of [15]. The normalization is fixed by the leading soft theorem.

23
From (134), the OPE of soft currents is found to be
1−k 
−if abc X 2 − k − l − n z̄12
 n
k,a l,b
R (z1 , z̄1 )R (z2 , z̄2 ) ∼ ∂¯n Rk+l−1,c (z2 , z̄2 ). (140)
z12 n=0 1 − l n!

This follows from setting ∆1 = k, ∆2 = l in (134) and using

Γ(k − 1 + n)Γ(l − 1) 1 Γ(3 − k − l − n)
lim  = . (141)
→0 Γ(k + l + n − 2) (1 − l)! Γ(2 − k − n)
Finally, (140) can be used to derive the algebra of the soft operators Rnk,a
− n0 + n0
 1−k
− n + 1−l
  1−k
+ n + 1−l

k,a l,b ab k+l−1,c
[Rn , Rn0 ] = −if c 2
1−k
2 2
1−k
2 Rn+n 0 , (142)
2
−n 2
+n
which follows from I
dz̄ n+ k−1 −1 k,a
Rnk,a (z) z̄ = 2 R (z, z̄) (143)
2πi
and the commutator of holomorphic operators
I
dw
[A, B](z) = A(w)B(z) (144)
2πi
applied to (140).
Upon redefining [17]
   
1−k 1−k
R̂nk,a ≡ −n ! + n !Rnk,a , (145)
2 2
(142) simplifies to
[R̂nk,a , R̂nl,b0 ] = −if abc R̂n+n
k+l−1,c
0 . (146)
A similar analysis can be done for gravitons, where the analog of (146) was identified with
a w1+∞ algebra in [17]
p p+q−2
[wm , wnq ] = [m(q − 1) − n(p − 1)]wm+n . (147)
The implications of (147) for gravity in AFS remains a fascinating open problem.

A Milne wedges and conformal primaries

The form (67) of the Minkowski metric with ρ = sinh η can be obtained directly from
ds2 = −(dx0 )2 + (dx1 )2 + (dx2 )2 + (dx3 )2 (148)
via the coordinate transform
x0 = τ cosh η,
x3 = τ cos θ sinh η,
(149)
x2 = τ sin θ cos ϕ sinh η,
x1 = τ sin θ sin ϕ sinh η

24
 The isometries of (67) are inherited from isometries of (148) which preserve the slices
of constant τ which coincide with the Lorentz transformations (50), (51). In (τ, η, θ, ϕ)
coordinates, the Lorentz generators generators take the form

J3 = i∂ϕ , J2 = −i(sin ϕ∂θ + cos ϕ cot θ∂ϕ ), J1 = −i(− cos ϕ∂θ + sin ϕ cot θ∂ϕ ),
K3 = −i(cos θ∂η − sin θ coth η∂θ ),
(150)
K2 = −i(cos ϕ sin θ∂η + cos θ cos ϕ coth η∂θ − coth η csc θ sin ϕ∂ϕ ),
K1 = −i(sin θ sin ϕ∂η + cos θ coth η sin ϕ∂θ + coth η cos ϕ csc θ∂ϕ ).

In the limit η → ∞, (150) reduce to

J3 = i∂ϕ , J2 = −i(sin ϕ∂θ + cos ϕ cot θ∂ϕ ), J1 = −i(− cos ϕ∂θ + sin ϕ cot θ∂ϕ ),
K3 = i sin θ∂θ , K2 = −i(cos θ cos ϕ∂θ − csc θ sin ϕ∂ϕ ), K1 = −i(cos θ sin ϕ∂θ + cos ϕ csc θ∂ϕ )
(151)

or equivalently in (z, z̄) coordinates

θ θ
z = cot eiϕ , z̄ = cot e−iϕ (152)
2 2
and using identities such as
1
= 1 + z z̄, (153)
(sin θ/2)2
1 i
J3 = (z∂z − z̄∂z̄ ), J2 = − (z 2 − 1)∂z − (z̄ 2 − 1)∂z̄ , J1 = − (z 2 + 1)∂z + (z̄ 2 + 1)∂z̄ ,
 
2 2
i 2 1 2
K3 = −i(z∂z + z̄∂z̄ ), K2 = − (z − 1)∂z + (z̄ 2 − 1)∂z̄ , K1 = (z + 1)∂z − (z̄ 2 + 1)∂z̄ .
 
2 2
(154)
These precisely agree with (24) with f = 0 and Y z in (27).
As before,
f (τ 2 ) f (τ 2 )
Ψ∆ = 0 = (155)
(x + x3 )∆ (τ (cosh η + sin θ sin ϕ sinh η))∆
obeys
(L0 + L̄0 )Ψ∆ = ∆Ψ∆ , (L0 − L̄0 )Ψ∆ = 0 (156)
and
L1 Ψ∆ = L̄1 Ψ∆ = 0. (157)
(155) diagonalizes boosts along the x3 axis and obeys the highest weight condition (157).

References
[1] Juan Martin Maldacena. The Large N limit of superconformal field theories and
supergravity. Int. J. Theor. Phys., 38:1113–1133, 1999. [Adv. Theor. Math.
Phys.2,231(1998)].

25
 [2] Ofer Aharony, Steven S. Gubser, Juan Martin Maldacena, Hirosi Ooguri, and Yaron
Oz. Large N field theories, string theory and gravity. Phys. Rept., 323:183–386, 2000.

[3] Shinsei Ryu and Tadashi Takayanagi. Holographic derivation of entanglement en-
tropy from AdS/CFT. Phys. Rev. Lett., 96:181602, 2006.

[4] Leonard Susskind. The World as a hologram. J. Math. Phys., 36:6377–6396, 1995.

[5] Raphael Bousso. The Holographic principle. Rev. Mod. Phys., 74:825–874, 2002.

[6] Temple He, Vyacheslav Lysov, Prahar Mitra, and Andrew Strominger. BMS super-
translations and Weinberg’s soft graviton theorem. JHEP, 05:151, 2015.

[7] Temple He, Prahar Mitra, Achilleas P. Porfyriadis, and Andrew Strominger. New
Symmetries of Massless QED. JHEP, 10:112, 2014.

[8] Andrew Strominger and Alexander Zhiboedov. Gravitational Memory, BMS Super-
translations and Soft Theorems. JHEP, 01:086, 2016.

[9] Sabrina Pasterski. Asymptotic Symmetries and Electromagnetic Memory. JHEP,

09:154, 2017.

[10] Andrew Strominger. Lectures on the Infrared Structure of Gravity and Gauge The-
ory. 2017.

[11] Sabrina Pasterski, Shu-Heng Shao, and Andrew Strominger. Gluon Amplitudes as
2d Conformal Correlators. Phys. Rev., D96(8):085006, 2017.

[12] Daniel Kapec, Vyacheslav Lysov, Sabrina Pasterski, and Andrew Strominger. Semi-
classical Virasoro symmetry of the quantum gravity S-matrix. JHEP, 08:058, 2014.

[13] Daniel Kapec, Prahar Mitra, Ana-Maria Raclariu, and Andrew Strominger. 2D Stress
Tensor for 4D Gravity. Phys. Rev. Lett., 119(12):121601, 2017.

[14] Y. T. Albert Law and Michael Zlotnikov. Poincarà c constraints on celestial ampli-
tudes. JHEP, 03:085, 2020.

[15] Monica Pate, Ana-Maria Raclariu, Andrew Strominger, and Ellis Ye Yuan. Celestial
Operator Products of Gluons and Gravitons. 2019.

[16] Alfredo Guevara, Elizabeth Himwich, Monica Pate, and Andrew Strominger. Holo-
graphic Symmetry Algebras for Gauge Theory and Gravity. 3 2021.

[17] Andrew Strominger. w(1+infinity) and the Celestial Sphere. 5 2021.

[18] Sabrina Pasterski, Shu-Heng Shao, and Andrew Strominger. Flat Space Amplitudes
and Conformal Symmetry of the Celestial Sphere. Phys. Rev., D96(6):065026, 2017.

[19] Nabamita Banerjee, Shamik Banerjee, Sayali Atul Bhatkar, and Sachin Jain. Con-
formal Structure of Massless Scalar Amplitudes Beyond Tree level. JHEP, 04:039,
2018.

26
[20] Anders Schreiber, Anastasia Volovich, and Michael Zlotnikov. Tree-level gluon am-
plitudes on the celestial sphere. Phys. Lett., B781:349–357, 2018.

[21] Stephan Stieberger and Tomasz R. Taylor. Strings on Celestial Sphere. Nucl. Phys.,
B935:388–411, 2018.

[22] Andreas Brandhuber, Graham R. Brown, Joshua Gowdy, Bill Spence, and Gabriele
Travaglini. Celestial Superamplitudes. 5 2021.

[23] Hongliang Jiang. Celestial superamplitude in N = 4 SYM theory. 5 2021.

[24] Hernán A. González, Andrea Puhm, and Francisco Rojas. Loop corrections to celes-
tial amplitudes. Phys. Rev. D, 102(12):126027, 2020.

[25] Eduardo Casali and Andrea Puhm. Double Copy for Celestial Amplitudes. Phys.
Rev. Lett., 126(10):101602, 2021.

[26] Eduardo Casali and Atul Sharma. Celestial double copy from the worldsheet. JHEP,
05:157, 2021.

[27] Jacques Distler, Raphael Flauger, and Bart Horn. Double-soft graviton amplitudes
and the extended BMS charge algebra. JHEP, 08:021, 2019.

[28] Shamik Banerjee. Null Infinity and Unitary Representation of The Poincare Group.
JHEP, 01:205, 2019.

[29] Stephan Stieberger and Tomasz R. Taylor. Symmetries of Celestial Amplitudes.

Phys. Lett., B793:141–143, 2019.

[30] Shamik Banerjee, Pranjal Pandey, and Partha Paul. Conformal properties of soft-
operators - 1 : Use of null-states. 2019.

[31] Y. T. Albert Law and Michael Zlotnikov. Massive Spinning Bosons on the Celestial
Sphere. JHEP, 06:079, 2020.

[32] Andrea Puhm. Conformally Soft Theorem in Gravity. 2019.

[33] Shamik Banerjee and Sudip Ghosh. MHV Gluon Scattering Amplitudes from Celes-
tial Current Algebras. 10 2020.

[34] Angelos Fotopoulos, Stephan Stieberger, Tomasz R. Taylor, and Bin Zhu. Extended
BMS Algebra of Celestial CFT. 2019.

[35] Wei Fan, Angelos Fotopoulos, Stephan Stieberger, and Tomasz R. Taylor. On Sug-
awara construction on Celestial Sphere. JHEP, 09:139, 2020.

[36] Nima Arkani-Hamed, Monica Pate, Ana-Maria Raclariu, and Andrew Strominger.
Celestial Amplitudes from UV to IR. 12 2020.

27
[37] Dhritiman Nandan, Anders Schreiber, Anastasia Volovich, and Michael Zlotnikov.
Celestial Amplitudes: Conformal Partial Waves and Soft Limits. JHEP, 10:018,
2019.

[38] Y. T. Albert Law and Michael Zlotnikov. Relativistic partial waves for celestial
amplitudes. JHEP, 11:149, 2020.

[39] Wei Fan, Angelos Fotopoulos, Stephan Stieberger, Tomasz R. Taylor, and Bin Zhu.
Conformal blocks from celestial gluon amplitudes. JHEP, 05:170, 2021.

[40] Alexander Atanasov, Walker Melton, Ana-Maria Raclariu, and Andrew Strominger.
Conformal Block Expansion in Celestial CFT. 4 2021.

[41] Sangmin Choi and Ratindranath Akhoury. Magnetic soft charges, dual supertrans-
lations, and ’t Hooft line dressings. Phys. Rev. D, 102(2):025001, 2020.

[42] Alok Laddha, Siddharth G. Prabhu, Suvrat Raju, and Pushkal Shrivastava. The
Holographic Nature of Null Infinity. SciPost Phys., 10:041, 2021.

[43] Eliot Hijano and Dominik Neuenfeld. Soft photon theorems from CFT Ward identites
in the flat limit of AdS/CFT. JHEP, 11:009, 2020.

[44] Elizabeth Himwich, Sruthi A. Narayanan, Monica Pate, Nisarga Paul, and Andrew
Strominger. The Soft S-Matrix in Gravity. JHEP, 09:129, 2020.

[45] Hayato Hirai and Sotaro Sugishita. IR finite S-matrix by gauge invariant dressed
states. JHEP, 02:025, 2021.

[46] Laurent Freidel, Jerzy Kowalski-Glikman, Robert G. Leigh, and Djordje Minic.
Quantum Gravity Phenomenology in the Infrared. 4 2021.

[47] Hernán A. González and Francisco Rojas. The structure of IR divergences in celestial
gluon amplitudes. 4 2021.

[48] Kevin Nguyen and Jakob Salzer. Celestial IR divergences and the effective action of
supertranslation modes. 5 2021.

[49] Luke Lippstreu. Zwanziger’s pairwise little group on the celestial sphere. 5 2021.

[50] Steven Weinberg. Infrared photons and gravitons. Phys. Rev., 140:B516–B524, Oct
1965.

[51] F. E. Low. Bremsstrahlung of very low-energy quanta in elementary particle colli-

sions. Phys. Rev., 110:974–977, 1958.

[52] Freddy Cachazo and Andrew Strominger. Evidence for a New Soft Graviton Theorem.
2014.

[53] Song He, Yu-tin Huang, and Congkao Wen. Loop Corrections to Soft Theorems in
Gauge Theories and Gravity. JHEP, 12:115, 2014.

28
[54] Henriette Elvang, Callum R. T. Jones, and Stephen G. Naculich. Soft Photon and
Graviton Theorems in Effective Field Theory. Phys. Rev. Lett., 118(23):231601, 2017.

[55] Alok Laddha and Prahar Mitra. Asymptotic Symmetries and Subleading Soft Photon
Theorem in Effective Field Theories. JHEP, 05:132, 2018.

[56] Glenn Barnich and Cedric Troessaert. BMS charge algebra. JHEP, 12:105, 2011.

[57] Glenn Barnich and Cedric Troessaert. Supertranslations call for superrotations. PoS,
CNCFG2010:010, 2010. [Ann. U. Craiova Phys.21,S11(2011)].

[58] A. Ashtekar. Asymptotic quantization: based on 1984 Naples lectures. 1987.

[59] Andrew Strominger. On BMS Invariance of Gravitational Scattering. JHEP, 07:152,

2014.

[60] Sabrina Pasterski and Shu-Heng Shao. Conformal basis for flat space amplitudes.
Phys. Rev., D96(6):065022, 2017.

[61] Jan de Boer and Sergey N. Solodukhin. A holographic reduction of minkowski space-
time. Nuclear Physics B, 665:545 – 593, 2003.

[62] Miguel Campiglia and Alok Laddha. Asymptotic symmetries of QED and Weinberg’s
soft photon theorem. JHEP, 07:115, 2015.

[63] Chang Liu and David A. Lowe. Conformal Wave Expansions for Flat Space Ampli-
tudes. 5 2021.

[64] Edward Witten. Anti-de Sitter space and holography. Adv. Theor. Math. Phys.,
2:253–291, 1998.

[65] David Kutasov and Nathan Seiberg. More comments on string theory on AdS(3).
JHEP, 04:008, 1999.

[66] Ho Tat Lam and Shu-Heng Shao. Conformal Basis, Optical Theorem, and the Bulk
Point Singularity. Phys. Rev., D98(2):025020, 2018.

[67] Temple He, Prahar Mitra, and Andrew Strominger. 2D Kac-Moody Symmetry of 4D
Yang-Mills Theory. JHEP, 10:137, 2016.

[68] Joshua D. Qualls. Lectures on Conformal Field Theory. 11.

[69] Monica Pate, Ana-Maria Raclariu, and Andrew Strominger. Conformally Soft The-
orem in Gauge Theory. Phys. Rev., D100(8):085017, 2019.

[70] Alfredo Guevara. Notes on Conformal Soft Theorems and Recursion Relations in
Gravity. 2019.

[71] Tim Adamo, Lionel Mason, and Atul Sharma. Celestial amplitudes and conformal
soft theorems. Class. Quant. Grav., 36(20):205018, 2019.

29
[72] Laura Donnay, Sabrina Pasterski, and Andrea Puhm. Asymptotic Symmetries and
Celestial CFT. JHEP, 09:176, 2020.

30