Prepared for submission to JHEP
Motivation and summary
What is the simplest non-trivial observable of four-dimensional N = 4 \mathcal{N}=4 N = 4 super Yang–Mills (SYM) theory at finite coupling and finite N N N ? By non-trivial we mean sensitive to the unprotected dynamics: quantities that are protected, or computable by supersymmetric localization, are not what we are after. Correlation functions of local operators, such as the four-point function of the stress-tensor multiplet, are the obvious candidates [1 , 2 , 3 , 4 ] , but they are intricate objects: already at zero coupling, the intermediate spectrum is highly degenerate. The 2 → 2 2 \to2 2 → 2 scattering amplitude of the elementary massless states is intrinsically simpler, but it is not an observable.1 Soft and collinear divergences, organized by gauge invariance and factorization into universal structures [10 , 11 , 12 , 13 ] , must be regulated before the amplitude can be defined at all.
The standard regulators do not turn the amplitude into a non-perturbative object. Dimensional regularization [14 ] is defined order by order in perturbation theory. Moving onto the Coulomb branch gives masses to some of the states and regulates planar, color-ordered amplitudes [15 , 16 , 17 ] , but at finite N N N the massless fields of the unbroken gauge group still run in loops and the divergences return.
In this paper we study a different possibility: placing the theory on a rigid A d S 4 \mathrm{AdS}_4 AdS 4 background [18 ] . The idea that negative curvature tames the infrared goes back to Callan and Wilczek [19 ] , and it is common lore in the study of the flat-space limit [20 , 21 , 22 ] that the AdS radius acts as an infrared regulator. Once boundary conditions are specified, the natural observables are correlation functions of operators at the conformal boundary [23 ] . They are finite, they are defined for any coupling and any gauge group without reference to a planar limit, and their large-radius limit reproduces flat-space scattering [24 , 25 , 26 ] . Since N = 4 \mathcal{N}=4 N = 4 SYM is conformal, there is a sharper way to say this. The theory on A d S 4 \mathrm{AdS}_4 AdS 4 is Weyl-equivalent to the theory on the flat half-space, and its boundary correlators do not depend on the radius L L L . The radius enters only when the dimensionless boundary data are converted into flat-space momenta, so that the scattering amplitude is recovered from the regime of large Mellin variables. Nothing is cut off: the boundary correlator is itself the infrared-safe observable, and the divergences of the amplitude reappear as logarithms in that regime.
The price is a choice of boundary conditions, but there is a canonical one. Among the half-BPS boundary conditions classified in [27 , 28 ] , the Dirichlet condition introduces no boundary degrees of freedom and preserves the whole gauge group G G G as a global symmetry of the boundary theory, a general feature of gauge theories on A d S 4 \mathrm{AdS}_4 AdS 4 with Dirichlet conditions [29 ] . The associated conserved current sits in a short multiplet of the preserved 3 d 3d 3 d N = 4 \mathcal{N}=4 N = 4 superconformal algebra, whose superconformal primary is the moment map μ \mu μ , a scalar of dimension one in the adjoint of G G G . Its four-point function is the boundary counterpart of 2 → 2 2 \to2 2 → 2 scattering of the elementary colored states. We propose it as the simplest non-trivial non-perturbative observable of N = 4 \mathcal{N}=4 N = 4 SYM. At zero coupling, it is a generalized free field, with a single single-particle multiplet and non-degenerate two-particle states, in contrast with the free limit of local correlators. At finite coupling, it defines consistent boundary conformal data [30 , 31 ] , although not a local three-dimensional Conformal Field Theory (CFT), since there is no boundary stress tensor (see [32 , 33 ] for related examples).
This makes the correlator a natural target for the conformal bootstrap. A numerical study at finite coupling, leveraging localization results for the protected conformal data, will be reported elsewhere [34 ] . Boundary correlators of Yang–Mills theory and related models in AdS have been studied in connection with confinement [29 , 35 , 36 ] , including with the numerical bootstrap [37 ] . Here we focus on the weak-coupling expansion of the correlator, which we carry to one loop. There are two kinds of lessons. The first is conceptual: we show explicitly how the infrared structure of the flat-space amplitude emerges from the boundary correlator. The divergent terms have the universal color structure of one-loop infrared divergences, and a natural subtraction prescription can be compared with dimensional regularization. The second is technical: we adapt the AdS unitarity method to a three-dimensional setting in which the reduced correlator is non-local in position space, and find that the one-loop answer is nonetheless remarkably simple in Mellin space.
Main results
We begin by deriving the tree-level moment-map four-point function directly from the microscopic N = 4 \mathcal{N}=4 N = 4 SYM theory with Dirichlet boundary conditions. Three ingredients contribute: bulk gauge-field exchange, the quartic scalar interaction, and a scalar exchange induced by the Nahm supersymmetric boundary condition. We show that these contributions combine exactly into the d = 3 d=3 d = 3 , p = 2 p=2 p = 2 tree-level supergluon Mellin amplitude of [38 ] , which was fixed by superconformal symmetry, crossing, factorization, and the exchanged spectrum. Schematically,
M N = 4 S Y M , A d S 4 t r e e = M g a u g e ⏟ vector exchange + M q u a r t i c + M N a h m ⏟ scalar sector = M ( p = 2 ) s u p e r g l u o n t r e e (1) \mathcal{M}^{\mathrm{tree}}_{\mathcal{N}=4\ \mathrm{SYM},\,\mathrm{AdS}_4}
=
\underbrace{\mathcal{M}_{\mathrm{gauge}}}_{\text{vector exchange}}
+
\underbrace{\mathcal{M}_{\mathrm{quartic}}+\mathcal{M}_{\mathrm{Nahm}}}_{\text{scalar sector}}
=
\mathcal{M}^{\mathrm{tree}}_{(p=2)\,\mathrm{supergluon}} \tag*{(1)} M N = 4 SYM , AdS 4 tree = vector exchange M gauge + scalar sector M quartic + M Nahm = M ( p = 2 ) supergluon tree (1)
The Nahm term is the one that is easy to miss: it comes from the non-linear scalar boundary condition rather than from a bulk vertex, and it is subleading in the flat-space limit. It is nevertheless required at finite AdS radius by supersymmetry.
The passage to one loop is where the spectrum of the microscopic theory becomes essential. A direct evaluation of loop Witten diagrams is generally cumbersome. Instead, we use AdS unitarity [39 , 40 , 41 , 42 , 43 ] , which reconstructs the non-local loop amplitude from lower-order conformal data. In our theory, the two-particle cut is obtained by gluing two copies of the tree-level moment-map amplitude. The closest precedent is the AdS5 _5 5 toy model of [44 ] , obtained by truncating the AdS5 × S 3 _5 \times S^3 5 × S 3 supergluon theory to its lowest p = 2 p=2 p = 2 Kaluza–Klein (KK) sector. In that model, the one-loop cut is likewise obtained from the p = 2 p=2 p = 2 sector alone, whereas the full AdS5 _5 5 × \times × S3 ^3 3 theory requires the tree-level ⟨ 22 p p ⟩ \langle22pp\rangle ⟨ 22 pp ⟩ amplitudes for all allowed p p p . By contrast, our setup naturally isolates the loop dynamics of a supersymmetric gauge theory in AdS without additional compactification or gravitational sectors.
A remarkable finding is that the one-loop double-discontinuity, obtained directly from the full tree-level correlator, admits a simple reduced Mellin representation involving only simultaneous poles. This mirrors structures found in higher-dimensional holographic one-loop correlators [44 , 45 , 46 ] . The answer can be organized in terms of a single simultaneous-pole kernel,
B ( s , t ) = ∑ m , n ≥ 1 c m n ( s − 2 m ) ( t − 2 n ) , (2) B(s,t)=\sum_{m,n\geq1}\frac{c_{mn}}{(s-2m)(t-2n)}, \tag*{(2)} B ( s , t ) = m , n ≥ 1 ∑ ( s − 2 m ) ( t − 2 n ) c mn , (2)
whose coefficients admit the closed form
c m n = 1 9 π 4 [ π Γ ( m ) Γ ( n ) Γ ( m + n + 1 2 ) Γ ( m + 1 2 ) Γ ( n + 1 2 ) Γ ( m + n + 1 ) − 1 m n ] , (3) c_{mn}=\frac{1}{9\pi^4}\left[\frac{\sqrt{\pi}\,\Gamma(m)\Gamma(n)\Gamma\left(m+n+\frac{1}{2}\right)}{\Gamma\left(m+\frac{1}{2}\right)\Gamma\left(n+\frac{1}{2}\right)\Gamma(m+n+1)}-\frac{1}{mn}\right], \tag*{(3)} c mn = 9 π 4 1 [ Γ ( m + 2 1 ) Γ ( n + 2 1 ) Γ ( m + n + 1 ) π Γ ( m ) Γ ( n ) Γ ( m + n + 2 1 ) − mn 1 ] , (3)
which is manifestly symmetric in m m m and n n n . The complete reduced amplitude takes the familiar color-box form: it is the sum over the three channels of the kernel B B B , each dressed by a one-loop color-box tensor, see (124) . The full R-symmetry dependence is restored by the finite-difference operators that implement the 3d N = 4 \mathcal{N}=4 N = 4 superconformal Ward identities in Mellin space [47 ] , see (126) . We also resum the elementary Ward-lifted block exactly: (142) gives its highest R-symmetry component, while the identities (6.19) determine the remaining two.
We conclude by studying the flat-space limit of the reconstructed one-loop correlator. Since the boundary correlator does not depend on the AdS radius L L L , this is the regime of large Mellin variables: L L L enters only through the conversion of Mellin variables into flat-space momenta. The discrete tower of AdS Mellin poles becomes dense and, upon resummation, gives rise to the logarithmic branch cuts of the four-dimensional massless box. The infrared divergences of the flat-space amplitude reappear as logarithms of L L L , which diverge at fixed flat-space kinematics as L → ∞ L\to\infty L → ∞ . In this sense, the AdS radius acts as an infrared regulator for a four-dimensional massless gauge theory, for any gauge group and without reference to the planar limit. The divergent terms have the universal color structure of one-loop infrared divergences: the double logarithm is proportional to the tree-level amplitude, and the single logarithms take the color-dipole form. A natural AdS minimal-subtraction scheme is then defined, and the resulting finite remainder is compared with the dimensionally regulated flat-space box [48 , 49 ] and with the Higgs regulator [15 ] .
Physical setup and conventions
We start from four-dimensional N = 4 \mathcal{N}=4 N = 4 super Yang–Mills theory with gauge group G G G in flat space. Since the theory is conformal, placing it on AdS4 _4 4 is equivalent, up to a Weyl transformation, to placing it on the flat half-space. In Poincaré coordinates,
d s A d S 4 2 = L 2 z 2 ( d z 2 + d x i d x i ) , (4) ds^{2}_{\mathrm{AdS}_4} = \frac{L^{2}}{z^{2}}\left(dz^{2} + dx_i dx^i\right), \tag*{(4)} d s AdS 4 2 = z 2 L 2 ( d z 2 + d x i d x i ) , (4)
so that A d S 4 \mathrm{AdS}_4 AdS 4 is conformal to the region z > 0 z > 0 z > 0 of flat space. Throughout, the A d S 4 \mathrm{AdS}_4 AdS 4 metric is a fixed background and gravity is non-dynamical.
A quantum field theory on AdS is not completely specified until boundary conditions are chosen at the conformal boundary. In Lorentzian signature, this boundary is timelike, so AdS is not globally hyperbolic and initial data alone do not determine the evolution. A well-posed dynamical problem therefore requires an admissible choice of boundary conditions.2 Since Poincaré A d S 4 \mathrm{AdS}_4 AdS 4 is conformally equivalent to the flat half-space, conformally invariant boundary conditions for N = 4 \mathcal{N}=4 N = 4 SYM on A d S 4 \mathrm{AdS}_4 AdS 4 can equivalently be described as conformal boundary conditions for the same theory on R + 4 \mathbb{R}^4_+ R + 4 .
The four-dimensional N = 4 \mathcal{N}=4 N = 4 vector multiplet contains the gauge field, fermions, and six real adjoint scalars Φ M \Phi^M Φ M , M = 1 , … , 6 M=1,\ldots,6 M = 1 , … , 6 , transforming in the 6 of the S O ( 6 ) R SO(6)_R S O ( 6 ) R R-symmetry. Boundary conditions preserving the maximal supersymmetry compatible with the presence of a boundary were classified in [27 , 28 ] . A half-BPS boundary admits the three-dimensional superconformal algebra O S p ( 4 ∣ 4 ) OSp(4|4) O S p ( 4∣4 ) , with sixteen fermionic generators, corresponding to eight Poincaré and eight conformal supercharges. Its bosonic subalgebra is
S O ( 3 , 2 ) ⏟ 3d conformal group = A d S 4 isometries × S O ( 4 ) R ⏟ R-symmetry . (5) \underbrace{SO(3,2)}_{\substack{\text{3d conformal group}\\=\mathrm{AdS}_4\text{ isometries}}}
\times
\underbrace{SO(4)_R}_{\text{R-symmetry}}. \tag*{(5)} 3d conformal group = AdS 4 isometries S O ( 3 , 2 ) × R-symmetry S O ( 4 ) R . (5)
Thus the boundary condition selects the R-symmetry subgroup
S O ( 6 ) R ⟶ S O ( 4 ) R ≃ S U ( 2 ) H × S U ( 2 ) C . (6) SO(6)_R \longrightarrow SO(4)_R \simeq SU(2)_H \times SU(2)_C. \tag*{(6)} S O ( 6 ) R ⟶ S O ( 4 ) R ≃ S U ( 2 ) H × S U ( 2 ) C . (6)
Accordingly, the 6 of S O ( 6 ) R SO(6)_R S O ( 6 ) R branches as
6 ⟶ ( 3 , 1 ) ⊕ ( 1 , 3 ) , (7) \mathbf{6} \longrightarrow(\mathbf{3},\mathbf{1}) \oplus(\mathbf{1},\mathbf{3}), \tag*{(7)} 6 ⟶ ( 3 , 1 ) ⊕ ( 1 , 3 ) , (7)
and the six scalars naturally split into two triplets,
X a ∈ ( 3 , 1 ) , Y a ˙ ∈ ( 1 , 3 ) , a , a ˙ = 1 , 2 , 3. (8) X^a \in(\mathbf{3},\mathbf{1}), \qquad Y^{\dot a} \in(\mathbf{1},\mathbf{3}), \qquad a,\dot a = 1,2,3. \tag*{(8)} X a ∈ ( 3 , 1 ) , Y a ˙ ∈ ( 1 , 3 ) , a , a ˙ = 1 , 2 , 3. (8)
Our convention is that the first and second factors refer to S U ( 2 ) H SU(2)_H S U ( 2 ) H and S U ( 2 ) C SU(2)_C S U ( 2 ) C , respectively. The labels H H H and C C C follow the standard three-dimensional N = 4 \mathcal{N}=4 N = 4 terminology, where the two R-symmetry factors are associated with the Higgs and Coulomb branches.
Equivalently, the triplets may be represented as symmetric rank-two tensors,
X A B = X B A , Y A ˙ B ˙ = Y B ˙ A ˙ , A , A ˙ = 1 , 2 , (9) X^{AB}=X^{BA},\qquad Y^{\dot A\dot B}=Y^{\dot B\dot A},\qquad A,\dot A=1,2, \tag*{(9)} X A B = X B A , Y A ˙ B ˙ = Y B ˙ A ˙ , A , A ˙ = 1 , 2 , (9)
transforming under S U ( 2 ) H SU(2)_H S U ( 2 ) H and S U ( 2 ) C SU(2)_C S U ( 2 ) C , respectively.
Dirichlet boundary conditions and the flavor-current multiplet. Among the half-BPS boundary conditions described above, we choose the simplest Dirichlet/D-type boundary condition, with no additional boundary degrees of freedom. This means that the local operators in the 3d boundary operator algebra arise from the allowed boundary data of the bulk fields, as explained below.
For a bulk gauge theory, the D-type choice also specifies how the bulk gauge symmetry is realized at the boundary. Gauge transformations that vanish at the boundary remain gauge redundancies, whereas transformations approaching a constant non-trivial value at z = 0 z=0 z = 0 act as the global G G G flavor symmetry on boundary operators. In the flat-half-space frame, the D-type boundary conditions for the bulk fields take the form3
A i I ∣ z = 0 = a i I ( x ) , Y a ˙ ∣ z = 0 = 0 , ( D z X a − i 2 ϵ a b c [ X b , X c ] ) ∣ z = 0 = 0. (10) \left.A_i^I\right|_{z=0}=a_i^I(x),\qquad
\left.Y^{\dot a}\right|_{z=0}=0,\qquad
\left.\left(D_zX^a-\frac{i}{2}\epsilon^{abc}[X^b,X^c]\right)\right|_{z=0}=0. \tag*{(10)} A i I z = 0 = a i I ( x ) , Y a ˙ z = 0 = 0 , ( D z X a − 2 i ϵ ab c [ X b , X c ] ) z = 0 = 0. (10)
Here, a i I ( x ) a_i^I(x) a i I ( x ) is a non-dynamical background source for the conserved current J i I J_i^I J i I associated with the boundary flavor symmetry. More specifically, in radial gauge, A z = 0 A_z=0 A z = 0 , the bulk equations of motion give the near-boundary expansion4
A i I ( z , x ) = a i I ( x ) + z b i I ( x ) + ⋯ . (11) A_i^I(z,x)=a_i^I(x)+z\,b_i^I(x)+\cdots. \tag*{(11)} A i I ( z , x ) = a i I ( x ) + z b i I ( x ) + ⋯ . (11)
Since F z i I = ∂ z A i I F_{zi}^I=\partial_zA_i^I F z i I = ∂ z A i I , the coefficient b i I ( x ) b_i^I(x) b i I ( x ) is precisely the boundary limit of the radial field strength. Up to normalization, this mode represents microscopically the corresponding dimension-two boundary current,
J i I ∼ F z i I ∣ z = 0 . (12) J_i^I\sim\left.F_{zi}^I\right|_{z=0}. \tag*{(12)} J i I ∼ F z i I z = 0 . (12)
The two bulk scalar triplets instead obey Dirichlet boundary conditions for Y a ˙ Y^{\dot a} Y a ˙ , while X a X^a X a satisfies a supersymmetric generalized Neumann/Nahm condition [27 , 50 ] . We choose the trivial Nahm background, as a pole X a ∼ 1 / z X^a\sim1/z X a ∼ 1/ z generally breaks the boundary flavor symmetry G G G to a subgroup.
Microscopically, the boundary value of X X X defines a scalar operator
μ a I ∼ X a I ∣ z = 0 , (13) \mu^{aI}\sim X^{aI}\big|_{z=0}, \tag*{(13)} μ a I ∼ X a I z = 0 , (13)
while the Dirichlet triplet Y Y Y is fixed to vanish at z = 0 z=0 z = 0 and therefore does not define an independent boundary operator. The first non-vanishing boundary mode is instead its normal derivative,
O Y a I ∼ D z Y a I ∣ z = 0 . (14) \mathcal{O}_{Y}^{aI}\sim D_zY^{aI}\big|_{z=0}. \tag*{(14)} O Y a I ∼ D z Y a I z = 0 . (14)
Supersymmetry organizes these boundary modes into the 3 d 3d 3 d N = 4 \mathcal{N}=4 N = 4 flavor-current multiplet B [ 0 ] ( 2 ; 0 ) \mathcal{B}_{[0]}^{(2;0)} B [ 0 ] ( 2 ; 0 ) , in the notation of [51 ] .5 Here, the superscripts denote the Dynkin labels of S U ( 2 ) H × S U ( 2 ) C SU(2)_H\times SU(2)_C S U ( 2 ) H × S U ( 2 ) C , and the superconformal primary of the BPS multiplet is a scalar of dimension one. This is the moment-map operator,
μ A B I = μ B A I , Δ μ = 1 , μ I ∈ ( 3 , 1 ) , (15) \mu^I_{AB}=\mu^I_{BA},\qquad\Delta_\mu=1,\qquad\mu^I\in(3,1), \tag*{(15)} μ A B I = μ B A I , Δ μ = 1 , μ I ∈ ( 3 , 1 ) , (15)
generated microscopically by (13) . We package its S U ( 2 ) H SU(2)_H S U ( 2 ) H indices with a polarization spinor v A v^A v A ,
O 2 I ( x ; v ) ≡ μ A B I ( x ) v A v B , Δ O 2 = 1 , (16) \mathcal{O}_2^I(x;v)\equiv\mu^I_{AB}(x)v^Av^B,\qquad\Delta_{\mathcal{O}_2}=1, \tag*{(16)} O 2 I ( x ; v ) ≡ μ A B I ( x ) v A v B , Δ O 2 = 1 , (16)
where the subscript refers to its S U ( 2 ) H SU(2)_H S U ( 2 ) H Dynkin label (equivalently, its degree in the polarization), rather than its conformal dimension. Explicitly, μ A B I v A v B = y a ( v ) μ a I \mu^I_{AB}v^Av^B=y^a(v)\,\mu^{aI} μ A B I v A v B = y a ( v ) μ a I , where y a ( v ) ≡ i 2 v A ( σ a ϵ ) A B v B y^a(v)\equiv\frac{i}{\sqrt{2}}v^A(\sigma^a\epsilon)_{AB}v^B y a ( v ) ≡ 2 i v A ( σ a ϵ ) A B v B is a null vector built from the Pauli matrices σ a \sigma^a σ a and the antisymmetric tensor ϵ A B \epsilon_{AB} ϵ A B , normalized so that y ( v 1 ) ⋅ y ( v 2 ) = ( ϵ A B v 1 A v 2 B ) 2 y(v_1)\cdot y(v_2)=(\epsilon_{AB}v_1^Av_2^B)^2 y ( v 1 ) ⋅ y ( v 2 ) = ( ϵ A B v 1 A v 2 B ) 2 . Then, representation theory identifies the dimension-two ( 1 , 3 ) (1,3) ( 1 , 3 ) scalar in (14) as a superconformal descendant of the moment map, appearing at the same top level of the multiplet as the dimension-two ( 1 , 1 ) (1,1) ( 1 , 1 ) conserved current (12) . The complete multiplet, including its fermionic component Ψ I \Psi^I Ψ I , is
μ I : Δ = 1 , j = 0 , ( 3 , 1 ) ↓ Q Ψ I : Δ = 3 2 , j = 1 2 , ( 2 , 2 ) ↙ Q ↘ Q J i I : Δ = 2 , j = 1 , ( 1 , 1 ) O Y I : Δ = 2 , j = 0 , ( 1 , 3 ) (17) \begin{aligned}
\mu^I &: \ \Delta=1,\ j=0,\ (3,1)\\
&\downarrow Q\\
\Psi^I &: \ \Delta=\frac{3}{2},\ j=\frac{1}{2},\ (2,2)\\
&\swarrow Q \qquad\searrow Q\\
J_i^I &: \ \Delta=2,\ j=1,\ (1,1) \qquad\mathcal{O}_Y^I:\ \Delta=2,\ j=0,\ (1,3) \tag*{(17)}
\end{aligned} μ I Ψ I J i I : Δ = 1 , j = 0 , ( 3 , 1 ) ↓ Q : Δ = 2 3 , j = 2 1 , ( 2 , 2 ) ↙ Q ↘ Q : Δ = 2 , j = 1 , ( 1 , 1 ) O Y I : Δ = 2 , j = 0 , ( 1 , 3 ) (17)
where each arrow denotes the action of a Poincaré supercharge Q Q Q , and for each component we list its conformal dimension Δ \Delta Δ , its spin j j j and its representation under S U ( 2 ) H × S U ( 2 ) C SU(2)_H\times SU(2)_C S U ( 2 ) H × S U ( 2 ) C .
In the equivalent A d S 4 \mathrm{AdS}_4 AdS 4 description, the bulk scalars X , Y X,Y X , Y are conformally coupled to the curvature, with
m 2 L 2 = Δ ( Δ − 3 ) = − 2 , Δ − = 1 , Δ + = 2. (18) m^2 L^2 = \Delta(\Delta- 3) = -2,\qquad\Delta_- = 1,\qquad\Delta_+ = 2. \tag*{(18)} m 2 L 2 = Δ ( Δ − 3 ) = − 2 , Δ − = 1 , Δ + = 2. (18)
Therefore, under the Weyl map, the Nahm field X a X^a X a gives the Δ = 1 \Delta= 1 Δ = 1 quantization, while the Dirichlet field Y a ˙ Y^{\dot a} Y a ˙ gives the Δ = 2 \Delta= 2 Δ = 2 quantization. Equivalently,
X f l a t a = X ( 0 ) a + ⋯ ⟹ X A d S a ∼ z X ( 0 ) a + ⋯ , X^a_{\mathrm{flat}} = X^a_{(0)} + \cdots\quad\Longrightarrow\quad X^a_{\mathrm{AdS}} \sim z X^a_{(0)} + \cdots, X flat a = X ( 0 ) a + ⋯ ⟹ X AdS a ∼ z X ( 0 ) a + ⋯ ,
Y f l a t a ˙ = z Y ( 1 ) a ˙ + ⋯ ⟹ Y A d S a ˙ ∼ z 2 Y ( 1 ) a ˙ + ⋯ . (19) Y^{\dot a}_{\mathrm{flat}} = z Y^{\dot a}_{(1)} + \cdots\quad\Longrightarrow\quad Y^{\dot a}_{\mathrm{AdS}} \sim z^2 Y^{\dot a}_{(1)} + \cdots. \tag*{(19)} Y flat a ˙ = z Y ( 1 ) a ˙ + ⋯ ⟹ Y AdS a ˙ ∼ z 2 Y ( 1 ) a ˙ + ⋯ . (19)
The correlators of the boundary theory are functions of the complexified coupling
τ = θ 2 π + 4 π i g 2 , (20) \tau= \frac{\theta}{2\pi} + \frac{4\pi i}{g^2}, \tag*{(20)} τ = 2 π θ + g 2 4 π i , (20)
where g g g is the conventionally normalized gauge coupling. They are not invariant under S-duality, which maps the Dirichlet condition to a different boundary condition, in which the dual gauge fields are coupled to the three-dimensional theory T [ G ] T[G] T [ G ] [28 ] . The strong-coupling limit of our observable is therefore governed by the weakly coupled dual bulk theory interacting with T [ G ] T[G] T [ G ] .
The displacement multiplet. Every genuine boundary CFT has a displacement operator because translations normal to the boundary are broken [52 , 53 ] . In the flat half-space description, this breaking is encoded in the Ward identity
∂ μ T μ z ( x , z ) = δ ( z ) D ( x ) , (21) \partial_\mu T^{\mu z}(x,z) = \delta(z) D(x), \tag*{(21)} ∂ μ T μ z ( x , z ) = δ ( z ) D ( x ) , (21)
where D is the displacement operator, a scalar of dimension four. For a half-BPS boundary preserving O S p ( 4 ∣ 4 ) OSp(4|4) O S p ( 4∣4 ) , the displacement is a descendant of the short multiplet B [ 0 ] 2 ( 4 ; 0 ) B[0]^{(4;0)}_2 B [ 0 ] 2 ( 4 ; 0 ) [31 , 51 ] . In our conventions, in which the moment map is charged under S U ( 2 ) H SU(2)_H S U ( 2 ) H , its superconformal primary is a scalar of dimension two, transforming with spin two under S U ( 2 ) H SU(2)_H S U ( 2 ) H and as a singlet under S U ( 2 ) C SU(2)_C S U ( 2 ) C and the boundary flavor group G G G .6
The moment-map OPE. Our main observable is the four-point function of moment-map operators,
⟨ O 2 I 1 ( x 1 ; v 1 ) O 2 I 2 ( x 2 ; v 2 ) O 2 I 3 ( x 3 ; v 3 ) O 2 I 4 ( x 4 ; v 4 ) ⟩ . (22) \left\langle O^{I_1}_2(x_1;v_1) O^{I_2}_2(x_2;v_2) O^{I_3}_2(x_3;v_3) O^{I_4}_2(x_4;v_4) \right\rangle. \tag*{(22)} ⟨ O 2 I 1 ( x 1 ; v 1 ) O 2 I 2 ( x 2 ; v 2 ) O 2 I 3 ( x 3 ; v 3 ) O 2 I 4 ( x 4 ; v 4 ) ⟩ . (22)
Supersymmetry relates it to correlation functions of the other components of the flavor-current multiplet, including those of the conserved currents. The boundary operator algebra contains many O S p ( 4 ∣ 4 ) OSp(4|4) O S p ( 4∣4 ) multiplets. For our purposes, we focus on those appearing in the OPE of two moment maps, or equivalently of two flavor-current multiplets. We follow the conventions of [54 ] , with the two S U ( 2 ) SU(2) S U ( 2 ) R-symmetry factors exchanged in order.
In addition to the superconformal selection rules, Bose symmetry under exchange of the two external moment-map operators, 1 ↔ 2 1 \leftrightarrow2 1 ↔ 2 , constrains the quantum numbers of the operator appearing in their OPE. Since the external operators transform in the adjoint of the flavor group,
Adj ⊗ Adj = Sym 2 ( Adj ) ⊕ ∧ 2 ( Adj ) . (23) \operatorname{Adj} \otimes\operatorname{Adj} = \operatorname{Sym}^{2}(\operatorname{Adj}) \oplus\wedge^{2}(\operatorname{Adj}). \tag*{(23)} Adj ⊗ Adj = Sym 2 ( Adj ) ⊕ ∧ 2 ( Adj ) . (23)
We denote by σ R \sigma_{\mathcal R} σ R the parity of the corresponding flavor tensor under I 1 ↔ I 2 I_{1} \leftrightarrow I_{2} I 1 ↔ I 2 ,
σ R ≡ { + 1 , R ⊂ Sym 2 ( Adj ) , − 1 , R ⊂ ∧ 2 ( Adj ) . (24) \sigma_{\mathcal R} \equiv
\begin{cases}
+1, & \mathcal R \subset\operatorname{Sym}^{2}(\operatorname{Adj}),\\
-1, & \mathcal R \subset\wedge^{2}(\operatorname{Adj}).
\end{cases} \tag*{(24)} σ R ≡ { + 1 , − 1 , R ⊂ Sym 2 ( Adj ) , R ⊂ ∧ 2 ( Adj ) . (24)
For an exchanged superconformal primary of spacetime spin ℓ \ell ℓ and S U ( 2 ) H SU(2)_{H} S U ( 2 ) H isospin j H j_{H} j H , the spacetime and R-symmetry structures contribute parities ( − 1 ) ℓ (-1)^{\ell} ( − 1 ) ℓ and ( − 1 ) j H (-1)^{j_{H}} ( − 1 ) j H , respectively. Bose symmetry therefore requires
( − 1 ) ℓ + j H σ R = 1. (25) (-1)^{\ell+j_{H}}\sigma_{\mathcal R}=1. \tag*{(25)} ( − 1 ) ℓ + j H σ R = 1. (25)
To implement this constraint compactly in the OPE below, we define the allowed flavor sectors at fixed spacetime spin,
F ℓ ( + ) ≡ { Sym 2 ( Adj ) , ℓ even , ∧ 2 ( Adj ) , ℓ odd , F ℓ ( − ) ≡ { ∧ 2 ( Adj ) , ℓ even , Sym 2 ( Adj ) , ℓ odd . (26) \mathcal F_{\ell}^{(+)} \equiv
\begin{cases}
\operatorname{Sym}^{2}(\operatorname{Adj}), & \ell\ \text{even},\\
\wedge^{2}(\operatorname{Adj}), & \ell\ \text{odd},
\end{cases}
\qquad
\mathcal F_{\ell}^{(-)} \equiv
\begin{cases}
\wedge^{2}(\operatorname{Adj}), & \ell\ \text{even},\\
\operatorname{Sym}^{2}(\operatorname{Adj}), & \ell\ \text{odd}.
\end{cases} \tag*{(26)} F ℓ ( + ) ≡ { Sym 2 ( Adj ) , ∧ 2 ( Adj ) , ℓ even , ℓ odd , F ℓ ( − ) ≡ { ∧ 2 ( Adj ) , Sym 2 ( Adj ) , ℓ even , ℓ odd . (26)
Then, the moment-map OPE takes the schematic form
( B [ 0 ] 1 ( 2 ; 0 ) , Adj ) × ( B [ 0 ] 1 ( 2 ; 0 ) , Adj ) ∼ ( B [ 0 ] 1 ( 0 ; 0 ) , 1 ) + ( B [ 0 ] 1 ( 2 ; 0 ) , Adj A ) + ∑ R ⊂ Sym 2 ( Adj ) ( B [ 0 ] 2 ( 4 ; 0 ) , R ) + ∑ ℓ = 0 ∞ ∑ R ⊂ F ℓ ( + ) ( A [ 2 ℓ ] ℓ + 1 ( 0 ; 0 ) , R ) + ∑ ℓ = 0 ∞ ∑ R ⊂ F ℓ ( − ) ( A [ 2 ℓ ] ℓ + 2 ( 2 ; 0 ) , R ) + ∑ ℓ = 0 ∞ ∑ Δ > ℓ + 1 ∑ R ⊂ F ℓ ( + ) ( L [ 2 ℓ ] Δ ( 0 ; 0 ) , R ) . (27) \begin{aligned}
\bigl(\mathcal B[0]_{1}^{(2;0)},\operatorname{Adj}\bigr)\times
\bigl(\mathcal B[0]_{1}^{(2;0)},\operatorname{Adj}\bigr)
&\sim
\bigl(\mathcal B[0]_{1}^{(0;0)},\mathbf1\bigr)
+\bigl(\mathcal B[0]_{1}^{(2;0)},\operatorname{Adj}_{A}\bigr)\\
&\quad+\sum_{\mathcal R\subset\operatorname{Sym}^{2}(\operatorname{Adj})}
\bigl(\mathcal B[0]_{2}^{(4;0)},\mathcal R\bigr)
+\sum_{\ell=0}^{\infty}\sum_{\mathcal R\subset\mathcal F_{\ell}^{(+)}}
\bigl(\mathcal A[2\ell]_{\ell+1}^{(0;0)},\mathcal R\bigr)\\
&\quad+\sum_{\ell=0}^{\infty}\sum_{\mathcal R\subset\mathcal F_{\ell}^{(-)}}
\bigl(\mathcal A[2\ell]_{\ell+2}^{(2;0)},\mathcal R\bigr)\\
&\quad+\sum_{\ell=0}^{\infty}\sum_{\Delta>\ell+1}
\sum_{\mathcal R\subset\mathcal F_{\ell}^{(+)}}
\bigl(\mathcal L[2\ell]_{\Delta}^{(0;0)},\mathcal R\bigr). \tag*{(27)}
\end{aligned} ( B [ 0 ] 1 ( 2 ; 0 ) , Adj ) × ( B [ 0 ] 1 ( 2 ; 0 ) , Adj ) ∼ ( B [ 0 ] 1 ( 0 ; 0 ) , 1 ) + ( B [ 0 ] 1 ( 2 ; 0 ) , Adj A ) + R ⊂ Sym 2 ( Adj ) ∑ ( B [ 0 ] 2 ( 4 ; 0 ) , R ) + ℓ = 0 ∑ ∞ R ⊂ F ℓ ( + ) ∑ ( A [ 2 ℓ ] ℓ + 1 ( 0 ; 0 ) , R ) + ℓ = 0 ∑ ∞ R ⊂ F ℓ ( − ) ∑ ( A [ 2 ℓ ] ℓ + 2 ( 2 ; 0 ) , R ) + ℓ = 0 ∑ ∞ Δ > ℓ + 1 ∑ R ⊂ F ℓ ( + ) ∑ ( L [ 2 ℓ ] Δ ( 0 ; 0 ) , R ) . (27)
Some comments are in order. The flavor-current multiplet is exchanged in the antisymmetric adjoint channel. The protected B [ 0 ] 2 ( 4 ; 0 ) \mathcal B[0]_{2}^{(4;0)} B [ 0 ] 2 ( 4 ; 0 ) sector also contains a distinguished flavor-singlet multiplet whose descendants include the displacement operator. The family A [ 2 ℓ ] ℓ + 1 ( 0 ; 0 ) \mathcal A[2\ell]_{\ell+1}^{(0;0)} A [ 2 ℓ ] ℓ + 1 ( 0 ; 0 ) contains conserved-current multiplets. Its ℓ = 0 \ell=0 ℓ = 0 member, A [ 0 ] 1 ( 0 ; 0 ) \mathcal A[0]_{1}^{(0;0)} A [ 0 ] 1 ( 0 ; 0 ) , is the three-dimensional stress-tensor multiplet. These multiplets are allowed by the selection rules, but they do not appear in our correlator. The three-dimensional superconformal symmetry is inherited from the four-dimensional bulk, and there is no conserved boundary stress tensor, nor any conserved higher-spin current, at any finite value of the bulk coupling. This is already the case at zero coupling, where the moment map is a generalized free field of dimension one: the R-symmetry singlets of spin ℓ \ell ℓ in its OPE start at dimension ℓ + 2 \ell+2 ℓ + 2 , above the dimension ℓ + 1 \ell+1 ℓ + 1 of the superconformal primary of A [ 2 ℓ ] ℓ + 1 ( 0 ; 0 ) \mathcal{A}[2\ell]_{\ell+1}^{(0;0)} A [ 2 ℓ ] ℓ + 1 ( 0 ; 0 ) . The unprotected sector consists of an infinite family of long multiplets, whose superconformal primaries are R-symmetry singlets. These will be the main focus of our one-loop analysis.
Finally, the moment-map operators generate a protected one-dimensional topological subsector of the boundary theory through Higgs-branch cohomology [50 , 55 , 56 , 57 ] . This sector provides exact constraints on protected OPE coefficients, complementary to the perturbative data extracted from the four-point function.
Tree-level AdS4 amplitude
In this section we derive the tree-level four-point function of the boundary moment-map operators from Dirichlet 4 d 4d 4 d N = 4 \mathcal{N}=4 N = 4 SYM on AdS4 _4 4 . We normalize the gauge generators as
Tr ( T I T J ) = κ δ I J , [ T I , T J ] = i f I J K T K . (28) \operatorname{Tr}(T^I T^J)=\kappa\,\delta^{IJ},\qquad[T^I,T^J]=i f^{IJK}T^K. \tag*{(28)} Tr ( T I T J ) = κ δ I J , [ T I , T J ] = i f I J K T K . (28)
In the flat-half-space frame, the bosonic part of the Euclidean action is
S = 1 g Y M 2 ∫ z > 0 d 4 x Tr [ 1 4 F μ ν F μ ν + 1 2 D μ Φ M D μ Φ M − 1 4 [ Φ M , Φ N ] 2 ] , Φ M = ( X a , Y a ˙ ) , (29) S=\frac{1}{g_{\mathrm{YM}}^2}\int_{z>0}d^4x\,\operatorname{Tr}\left[\frac{1}{4}F_{\mu\nu}F^{\mu\nu}+\frac{1}{2}D_\mu\Phi^M D^\mu\Phi^M-\frac{1}{4}[\Phi^M,\Phi^N]^2\right],\qquad\Phi^M=(X^a,Y^{\dot a}), \tag*{(29)} S = g YM 2 1 ∫ z > 0 d 4 x Tr [ 4 1 F μν F μν + 2 1 D μ Φ M D μ Φ M − 4 1 [ Φ M , Φ N ] 2 ] , Φ M = ( X a , Y a ˙ ) , (29)
with D μ = ∂ μ − i [ A μ , ⋅ ] D_\mu=\partial_\mu-i[A_\mu,\cdot] D μ = ∂ μ − i [ A μ , ⋅ ] and all fields expanded on the generators, Φ M = Φ M I T I \Phi^M=\Phi^{MI}T^I Φ M = Φ M I T I . It is supplemented by the boundary term (46) discussed below. In components, the action carries an overall factor κ / g Y M 2 \kappa/g_{\mathrm{YM}}^2 κ / g YM 2 , so that the canonically normalized coupling is g Y M 2 / κ g_{\mathrm{YM}}^2/\kappa g YM 2 / κ .7 We strip this common normalization from the intermediate Witten-diagram kernels and restore it in the complete correlator below.
Momentum-space representation
Boundary correlators in the AdS4 _4 4 theory admit a perturbative expansion in momentum-space Witten diagrams. We work in Euclidean Poincaré AdS4 _4 4 and set the AdS radius to one. Translation invariance along the boundary allows us to Fourier transform the boundary coordinates and define the 3 d 3d 3 d momenta k i k_i k i , with ∑ i k i = 0 \sum_i k_i=0 ∑ i k i = 0 .8 We denote
k i ≡ ∣ k i ∣ , k i j ≡ k i + k j , k ‾ i j ≡ ∣ k i + k j ∣ , E ≡ ∑ i = 1 4 k i . (30) k_i\equiv|k_i|,\qquad k_{ij}\equiv k_i+k_j,\qquad\underline{k}_{ij}\equiv|k_i+k_j|,\qquad E\equiv\sum_{i=1}^{4}k_i. \tag*{(30)} k i ≡ ∣ k i ∣ , k ij ≡ k i + k j , k ij ≡ ∣ k i + k j ∣ , E ≡ i = 1 ∑ 4 k i . (30)
The moment-map four-point function also carries adjoint gauge-group indices and S U ( 2 ) H SU(2)_H S U ( 2 ) H quantum numbers. We organize the former using the three cubic-diagram color factors,
c s I 1 I 2 I 3 I 4 ≡ f I 1 I 2 J f J I 3 I 4 , c t I 1 I 2 I 3 I 4 ≡ f I 1 I 4 J f J I 2 I 3 , c u I 1 I 2 I 3 I 4 ≡ f I 1 I 3 J f J I 4 I 2 , (31) c_s^{I_1 I_2 I_3 I_4} \equiv f^{I_1 I_2 J} f^{J I_3 I_4}, \qquad c_t^{I_1 I_2 I_3 I_4} \equiv f^{I_1 I_4 J} f^{J I_2 I_3}, \qquad c_u^{I_1 I_2 I_3 I_4} \equiv f^{I_1 I_3 J} f^{J I_4 I_2}, \tag*{(31)} c s I 1 I 2 I 3 I 4 ≡ f I 1 I 2 J f J I 3 I 4 , c t I 1 I 2 I 3 I 4 ≡ f I 1 I 4 J f J I 2 I 3 , c u I 1 I 2 I 3 I 4 ≡ f I 1 I 3 J f J I 4 I 2 , (31)
satisfying the Jacobi identity
c s + c t + c u = 0. (32) c_s+c_t+c_u=0. \tag*{(32)} c s + c t + c u = 0. (32)
The S U ( 2 ) H SU(2)_H S U ( 2 ) H dependence is instead encoded using the polarization spinors introduced in (16) . We define the three S U ( 2 ) H SU(2)_H S U ( 2 ) H -invariant tensor structures
R s ≡ ( v 1 ⋅ v 2 ) 2 ( v 3 ⋅ v 4 ) 2 , R t ≡ ( v 1 ⋅ v 4 ) 2 ( v 2 ⋅ v 3 ) 2 , R u ≡ ( v 1 ⋅ v 3 ) 2 ( v 2 ⋅ v 4 ) 2 , (33) R_s \equiv(v_1\cdot v_2)^2(v_3\cdot v_4)^2,\qquad R_t \equiv(v_1\cdot v_4)^2(v_2\cdot v_3)^2,\qquad R_u \equiv(v_1\cdot v_3)^2(v_2\cdot v_4)^2, \tag*{(33)} R s ≡ ( v 1 ⋅ v 2 ) 2 ( v 3 ⋅ v 4 ) 2 , R t ≡ ( v 1 ⋅ v 4 ) 2 ( v 2 ⋅ v 3 ) 2 , R u ≡ ( v 1 ⋅ v 3 ) 2 ( v 2 ⋅ v 4 ) 2 , (33)
where ( v i ⋅ v j ) ≡ ϵ A B v i A v j B (v_i\cdot v_j)\equiv\epsilon_{AB}v_i^A v_j^B ( v i ⋅ v j ) ≡ ϵ A B v i A v j B . The R-symmetry cross-ratio is instead
α ≡ ( v 1 ⋅ v 3 ) ( v 2 ⋅ v 4 ) ( v 1 ⋅ v 2 ) ( v 3 ⋅ v 4 ) . (34) \alpha\equiv\frac{(v_1\cdot v_3)(v_2\cdot v_4)}{(v_1\cdot v_2)(v_3\cdot v_4)}. \tag*{(34)} α ≡ ( v 1 ⋅ v 2 ) ( v 3 ⋅ v 4 ) ( v 1 ⋅ v 3 ) ( v 2 ⋅ v 4 ) . (34)
In particular, notice that
R u / R s = α 2 , R t / R s = ( 1 − α ) 2 . (35) R_u/R_s=\alpha^2,\qquad R_t/R_s=(1-\alpha)^2. \tag*{(35)} R u / R s = α 2 , R t / R s = ( 1 − α ) 2 . (35)
We now collect the bulk ingredients needed to construct the tree-level Witten diagrams.
External scalar wavefunctions. As discussed in § 2 , the X X X scalars have m 2 L 2 = − 2 m^2L^2=-2 m 2 L 2 = − 2 and the boundary moment-map operator corresponds to their alternate Δ = 1 \Delta=1 Δ = 1 quantization. Near the AdS boundary, a Fourier mode has the general asymptotic form
ϕ k ( z ) = a ( k ) z + b ( k ) z 2 + ⋯ . (36) \phi_k(z)=a(k)z+b(k)z^2+\cdots. \tag*{(36)} ϕ k ( z ) = a ( k ) z + b ( k ) z 2 + ⋯ . (36)
In the standard quantization, Δ = 2 \Delta=2 Δ = 2 , a ( k ) a(k) a ( k ) is the source and b ( k ) b(k) b ( k ) the response, while in the alternate quantization, Δ = 1 \Delta=1 Δ = 1 , their roles are exchanged [58 , 59 ] . Modified Bessel functions solve the full radial equation. Regularity in the AdS interior (z → ∞ z\to\infty z → ∞ ) then selects
ϕ k ( z ) ∝ z 3 / 2 K 1 / 2 ( k z ) ∝ z e − k z = z − k z 2 + ⋯ , (37) \phi_k(z)\propto z^{3/2}K_{1/2}(kz)\propto ze^{-kz}=z-kz^2+\cdots, \tag*{(37)} ϕ k ( z ) ∝ z 3/2 K 1/2 ( k z ) ∝ z e − k z = z − k z 2 + ⋯ , (37)
and therefore relates the two near-boundary coefficients by b = − k a b=-ka b = − k a . With unit source normalization, the bulk-to-boundary wavefunctions are
K Δ = 1 ( k , z ) = − z k e − k z = − 1 k K Δ = 2 ( k , z ) . (38) K_{\Delta=1}(k,z)=-\frac{z}{k}e^{-kz}=-\frac{1}{k}K_{\Delta=2}(k,z). \tag*{(38)} K Δ = 1 ( k , z ) = − k z e − k z = − k 1 K Δ = 2 ( k , z ) . (38)
At tree level, the boundary moment-map four-point function receives bulk contributions from the gauge-field exchange and the quartic scalar interaction of the four-dimensional N = 4 \mathcal{N}=4 N = 4 SYM action.
Gauge-field exchange. The momentum-space vertex takes the form
V X X A i = f I 1 I 2 J δ a 1 a 2 ( k 1 − k 2 ) i , (39) V_{X X A}^{i} = f^{I_{1}I_{2}J}\delta^{a_{1}a_{2}}(k_{1}-k_{2})^{i}, \tag*{(39)} V X X A i = f I 1 I 2 J δ a 1 a 2 ( k 1 − k 2 ) i , (39)
where i = 1 , 2 , 3 i=1,2,3 i = 1 , 2 , 3 labels directions tangent to the AdS boundary, and the gauge-field bulk-to-bulk propagator admits the spectral representation [60 ]
G i j I J ( q ; z , z ′ ) = δ I J ∫ 0 ∞ p d p ( z z ′ ) 1 / 2 J 1 / 2 ( p z ) J 1 / 2 ( p z ′ ) q 2 + p 2 ( δ i j + q i q j p 2 ) . (40) G_{ij}^{IJ}(q;z,z')=\delta^{IJ}\int_{0}^{\infty}p\,dp\frac{(zz')^{1/2}J_{1/2}(pz)J_{1/2}(pz')}{q^{2}+p^{2}}\left(\delta_{ij}+\frac{q_iq_j}{p^{2}}\right). \tag*{(40)} G ij I J ( q ; z , z ′ ) = δ I J ∫ 0 ∞ p d p q 2 + p 2 ( z z ′ ) 1/2 J 1/2 ( p z ) J 1/2 ( p z ′ ) ( δ ij + p 2 q i q j ) . (40)
For the s s s -channel, the momentum flowing through the internal line is k 1 + k 2 k_1+k_2 k 1 + k 2 , with magnitude k 12 k_{12} k 12 . Using the external Δ = 1 \Delta=1 Δ = 1 wavefunctions (38) and performing the radial and spectral integrals gives the exchange contribution
W s = − 1 k 1 k 2 k 3 k 4 1 E s [ ( k 1 − k 2 ) ⋅ ( k 3 − k 4 ) − ( k 1 2 − k 2 2 ) ( k 3 2 − k 4 2 ) E + k 12 k 12 k 34 k 12 ] , (41) W_s=-\frac{1}{k_1k_2k_3k_4}\frac{1}{E s}\left[(k_1-k_2)\cdot(k_3-k_4)-(k_1^2-k_2^2)(k_3^2-k_4^2)\frac{E+k_{12}}{k_{12}k_{34}k_{12}}\right], \tag*{(41)} W s = − k 1 k 2 k 3 k 4 1 E s 1 [ ( k 1 − k 2 ) ⋅ ( k 3 − k 4 ) − ( k 1 2 − k 2 2 ) ( k 3 2 − k 4 2 ) k 12 k 34 k 12 E + k 12 ] , (41)
where the Mandelstam-like variable is defined by s ≡ ( k 12 + k 12 ) ( k 34 + k 34 ) s\equiv(k_{12}+k_{12})(k_{34}+k_{34}) s ≡ ( k 12 + k 12 ) ( k 34 + k 34 ) .9 The t t t - and u u u -channel diagrams follow by crossing, and the corresponding R R R -symmetry structures are the R s R_s R s , R t R_t R t , and R u R_u R u introduced in (33) , respectively.
Scalar contact interaction. The scalar potential of N = 4 \mathcal{N}=4 N = 4 SYM contains the quartic interaction f I J M f K L M X a I X b J X a K X b L f^{IJM}f^{KLM}X^{aI}X^{bJ}X^{aK}X^{bL} f I J M f K L M X a I X b J X a K X b L . Contracting the external S U ( 2 ) H SU(2)_H S U ( 2 ) H indices with the polarization spinors, the corresponding gauge and R R R -symmetry tensor is
T 4 = c s ( R u − R t ) + c t ( R s − R u ) + c u ( R t − R s ) . (42) T_4=c_s(R_u-R_t)+c_t(R_s-R_u)+c_u(R_t-R_s). \tag*{(42)} T 4 = c s ( R u − R t ) + c t ( R s − R u ) + c u ( R t − R s ) . (42)
Since the quartic vertex carries no derivatives, the remaining momentum dependence comes entirely from the radial contact integral, giving
W c = − 1 E k 1 k 2 k 3 k 4 . (43) W_c=-\frac{1}{E k_1k_2k_3k_4}. \tag*{(43)} W c = − E k 1 k 2 k 3 k 4 1 . (43)
Nahm-induced scalar exchange. For the specific D-type theory considered here, the two bulk contributions above actually do not exhaust the tree-level correlator. The supersymmetric boundary condition obeyed by the X X X scalars is non-linear,
( D z X a − i 2 ϵ a b c [ X b , X c ] ) ∣ z = 0 = 0. (44) \left(D_zX^a-\frac{i}{2}\epsilon^{abc}[X^b,X^c]\right)\bigg|_{z=0}=0. \tag*{(44)} ( D z X a − 2 i ϵ ab c [ X b , X c ] ) z = 0 = 0. (44)
Around the trivial background X a = 0 X^a=0 X a = 0 , the boundary condition linearizes to the ordinary Neumann condition. Together with the free bulk equation of motion, this selects the Neumann scalar propagator. In the flat-half-space frame, and stripping the overall coupling normalization as above, it takes the form
G N α I , b J ( q ; z , z ′ ) = δ a b δ I J 1 2 q ( e − q ∣ z − z ′ ∣ + e − q ( z + z ′ ) ) . (45) G_N^{\alpha I,bJ}(\mathbf{q};z,z')=\delta^{ab}\delta^{IJ}\frac{1}{2q}\left(e^{-q\lvert z-z'\rvert}+e^{-q(z+z')}\right). \tag*{(45)} G N α I , b J ( q ; z , z ′ ) = δ ab δ I J 2 q 1 ( e − q ∣ z − z ′ ∣ + e − q ( z + z ′ ) ) . (45)
The non-linear term in (44) can be implemented by the variation of the finite boundary functional
S N a h m = i 6 g Y M 2 ∫ z = 0 d 3 x ϵ a b c Tr ( X a [ X b , X c ] ) = − κ 6 g Y M 2 ∫ z = 0 d 3 x ϵ a b c f I J K X a I X b J X c K . (46) S_{\mathrm{Nahm}}=\frac{i}{6g_{\mathrm{YM}}^2}\int_{z=0}d^3x\,\epsilon^{abc}\operatorname{Tr}\left(X^a[X^b,X^c]\right)=-\frac{\kappa}{6g_{\mathrm{YM}}^2}\int_{z=0}d^3x\,\epsilon^{abc}f_{IJK}X^{aI}X^{bJ}X^{cK}. \tag*{(46)} S Nahm = 6 g YM 2 i ∫ z = 0 d 3 x ϵ ab c Tr ( X a [ X b , X c ] ) = − 6 g YM 2 κ ∫ z = 0 d 3 x ϵ ab c f I J K X a I X b J X cK . (46)
This is a perturbative implementation of the supersymmetric boundary condition, rather than an independent boundary interaction.10
We now evaluate the scalar-exchange contribution generated by two insertions of the Nahm boundary vertex. Since both vertices are localized at z = 0 z=0 z = 0 , the internal X X X line is obtained by restricting both endpoints of the Neumann propagator to the boundary,
G N α I , b J ( q ; 0 , 0 ) = δ a b δ I J q . (47) G_N^{\alpha I,bJ}(\mathbf{q};0,0)=\frac{\delta^{ab}\delta^{IJ}}{q}. \tag*{(47)} G N α I , b J ( q ; 0 , 0 ) = q δ ab δ I J . (47)
The external legs are the corresponding Neumann source wavefunctions, K N ( k , 0 ) = − 1 / k K_N(k,0)=-1/k K N ( k , 0 ) = − 1/ k , related by the Weyl transformation to the alternate-quantization wavefunction (38) . Joining the two boundary vertices with the internal propagator and attaching the four external legs gives the s s s -channel contribution
B s = 1 k 1 k 2 k 3 k 4 k 12 . (48) B_s=\frac{1}{k_1k_2k_3k_4k_{12}}. \tag*{(48)} B s = k 1 k 2 k 3 k 4 k 12 1 . (48)
The color and R R R -symmetry tensors follow directly from the two cubic vertices: B s B_s B s multiplies c s ( R u − R t ) c_s(R_u-R_t) c s ( R u − R t ) , with the t t t - and u u u -channel structures obtained by crossing. Notice that this is the same tensorial structure as the quartic interaction in (42) . Their relative coefficient follows from the common microscopic N = 4 \mathcal{N}=4 N = 4 SYM action and its supersymmetric boundary completion. We can then combine the scalar sector of the s s s -channel tree-level correlator into
B ~ s ≡ B s + W c = E − k 12 E k 1 k 2 k 3 k 4 k 12 . (49) \widetilde{B}_s\equiv B_s+W_c=\frac{E-k_{12}}{E k_1k_2k_3k_4k_{12}}. \tag*{(49)} B s ≡ B s + W c = E k 1 k 2 k 3 k 4 k 12 E − k 12 . (49)
Complete result. To conclude, let us restore the coupling normalization. We normalize the moment map as μ a I ≡ ( κ / g Y M 2 ) X a I ∣ z = 0 \mu^{aI}\equiv(\kappa/g_{\mathrm{YM}}^2)X^{aI}\big|_{z=0} μ a I ≡ ( κ / g YM 2 ) X a I z = 0 . At leading order in weak coupling, its twopoint function is
⟨ μ a I ( k ) μ b J ( − k ) ⟩ = κ g Y M 2 δ a b δ I J k (50) \left\langle\mu^{aI}(\mathbf{k})\mu^{bJ}(-\mathbf{k}) \right\rangle= \frac{\kappa}{g_{\mathrm{YM}}^2}\frac{\delta^{ab}\delta^{IJ}}{k} \tag*{(50)} ⟨ μ a I ( k ) μ b J ( − k ) ⟩ = g YM 2 κ k δ ab δ I J (50)
where we have suppressed the overall 3 d 3d 3 d momentum-conserving delta function. It is convenient to define momentum-normalized operators
μ ^ a I ≡ g Y M κ μ a I , ⟨ μ ^ a I ( k ) μ ^ b J ( − k ) ⟩ = δ a b δ I J k . (51) \widehat{\mu}^{aI} \equiv\frac{g_{\mathrm{YM}}}{\sqrt{\kappa}}\,\mu^{aI}, \qquad\left\langle\widehat{\mu}^{aI}(\mathbf{k})\widehat{\mu}^{bJ}(-\mathbf{k}) \right\rangle= \frac{\delta^{ab}\delta^{IJ}}{k}. \tag*{(51)} μ a I ≡ κ g YM μ a I , ⟨ μ a I ( k ) μ b J ( − k ) ⟩ = k δ ab δ I J . (51)
Following our definition (16) , the tree-level four-point function can be written as
⟨ O ^ 2 I 1 O ^ 2 I 2 O ^ 2 I 3 O ^ 2 I 4 ⟩ t r e e = g Y M 2 κ [ c s ( R s W s + ( R u − R t ) B ~ s ) + c r o s s e d ] . (52) \left\langle\widehat{\mathcal{O}}^{I_1}_2\widehat{\mathcal{O}}^{I_2}_2\widehat{\mathcal{O}}^{I_3}_2\widehat{\mathcal{O}}^{I_4}_2 \right\rangle_{\mathrm{tree}} = \frac{g_{\mathrm{YM}}^2}{\kappa}\left[c_s\left(R_sW_s+(R_u-R_t)\widetilde{B}_s\right)+\mathrm{crossed}\right]. \tag*{(52)} ⟨ O 2 I 1 O 2 I 2 O 2 I 3 O 2 I 4 ⟩ tree = κ g YM 2 [ c s ( R s W s + ( R u − R t ) B s ) + crossed ] . (52)
Matching to the supergluon Mellin amplitude
The moment map belongs to the same short multiplet as the conserved current, whose AdS realization is the bulk gauge field. It is therefore useful to place our observable in the broader context of gauge-theory correlators in AdS. Tree-level current correlators in pure Yang–Mills theory were studied in [60 , 63 ] , and subsequent developments introduced compact momentum-space representations, recursive constructions, and generalized color–kinematics relations [61 , 64 , 65 , 66 ] ; see also [67 ] for the self-dual sector. On the supersymmetric side, four-point amplitudes of AdS supergluons were obtained in a broad class of holographic theories and shown to exhibit related universal and color–kinematic structures [38 ] . Explicit double-copy relations have also been obtained in special four-point AdS settings [68 ] .11 In these holographic constructions, however, the gauge fields typically arise from flavor-brane or compactification sectors of a higher-dimensional gravitational theory. The problem considered here is different: the gauge field is that of four-dimensional N = 4 \mathcal{N}=4 N = 4 SYM itself, propagating on rigid A d S 4 \mathrm{AdS}_4 AdS 4 .
We now introduce the tree-level d = 3 d=3 d = 3 , p = 2 p=2 p = 2 supergluon Mellin amplitude of [38 ] , while establishing the relation to the microscopic momentum-space result derived above. These supergluons are scalar superprimaries of short superconformal multiplets whose descendants have spin at most one. We denote their Kaluza–Klein label by p p p , to avoid confusion with the momentum magnitudes k i k_i k i introduced above.
The lowest multiplet, p = 2 p=2 p = 2 , is the flavor-current multiplet. In the 3 d 3d 3 d N = 4 \mathcal{N}=4 N = 4 description, its bosonic components include the dimension-one scalar superprimary s 2 I s^I_2 s 2 I , the dimension-two vector A 2 , μ I A^I_{2,\mu} A 2 , μ I , and the dimension-two scalar t 2 I t^I_2 t 2 I , transforming respectively in ( 3 , 1 ) (3,1) ( 3 , 1 ) , ( 1 , 1 ) (1,1) ( 1 , 1 ) , and ( 1 , 3 ) (1,3) ( 1 , 3 ) under the R-symmetry S U ( 2 ) a × S U ( 2 ) b SU(2)_a\times SU(2)_b S U ( 2 ) a × S U ( 2 ) b .12 Here I I I is an adjoint flavor index. We identify the two R-symmetry factors with those preserved by our boundary conditions, S U ( 2 ) H × S U ( 2 ) C SU(2)_H \times SU(2)_C S U ( 2 ) H × S U ( 2 ) C . At the level of the AdS fields,
s 2 I ↔ X a , I ( Δ = 1 ) , A 2 , μ I ↔ A μ I , t 2 I ↔ Y a ˙ , I ( Δ = 2 ) . (53) s_2^I \leftrightarrow X^{a,I}\quad(\Delta= 1),\qquad A_{2,\mu}^I \leftrightarrow A_\mu^I,\qquad t_2^I \leftrightarrow Y^{\dot a,I}\quad(\Delta= 2). \tag*{(53)} s 2 I ↔ X a , I ( Δ = 1 ) , A 2 , μ I ↔ A μ I , t 2 I ↔ Y a ˙ , I ( Δ = 2 ) . (53)
For the four-point function of the p = 2 p=2 p = 2 superprimaries, the exchanged current multiplet contributes through the scalar superprimary s 2 s_2 s 2 and the vector descendant A 2 A_2 A 2 , while t 2 t_2 t 2 transforms non-trivially under S U ( 2 ) C SU(2)_C S U ( 2 ) C and is therefore not exchanged in the s 2 × s 2 s_2 \times s_2 s 2 × s 2 OPE.
The tree-level answer of [38 ] is fixed by superconformal symmetry, crossing, and factorization. In this sense, it is universal. Our microscopic calculation provides a direct realization of this structure in N = 4 \mathcal{N}=4 N = 4 SYM. Moreover, in the holographic supergluon constructions, the leading large-N N N dynamics is governed by the self-interactions of the vector multiplet, while graviton exchange and other contributions from the gravitational sector are subleading. In our setup, gravity is instead non-dynamical from the outset.
Let us now make the comparison explicit by introducing the Mellin representation, which will also provide the framework for the one-loop analysis. For four external operators of dimension one, we write
⟨ ∏ i = 1 4 O 2 I i ( x i ; v i ) ⟩ = ( ( v 1 ⋅ v 2 ) ( v 3 ⋅ v 4 ) x 12 x 34 ) 2 G I 1 I 2 I 3 I 4 ( U , V ; α ) , (54) \left\langle\prod_{i=1}^{4}\mathcal{O}_2^{I_i}(x_i;v_i)\right\rangle=
\left(\frac{(v_1\cdot v_2)(v_3\cdot v_4)}{x_{12}x_{34}}\right)^2
\mathcal{G}^{I_1I_2I_3I_4}(U,V;\alpha), \tag*{(54)} ⟨ i = 1 ∏ 4 O 2 I i ( x i ; v i ) ⟩ = ( x 12 x 34 ( v 1 ⋅ v 2 ) ( v 3 ⋅ v 4 ) ) 2 G I 1 I 2 I 3 I 4 ( U , V ; α ) , (54)
with the conformal cross-ratios
U = x 12 2 x 34 2 x 13 2 x 24 2 , V = x 14 2 x 23 2 x 13 2 x 24 2 . (55) U=\frac{x_{12}^2x_{34}^2}{x_{13}^2x_{24}^2},\qquad
V=\frac{x_{14}^2x_{23}^2}{x_{13}^2x_{24}^2}. \tag*{(55)} U = x 13 2 x 24 2 x 12 2 x 34 2 , V = x 13 2 x 24 2 x 14 2 x 23 2 . (55)
The connected part of the dynamical function admits the Mellin representation [71 , 72 ]
G I 1 I 2 I 3 I 4 ( U , V ; α ) = ∫ − i ∞ i ∞ d s d t ( 4 π i ) 2 U s / 2 V t / 2 − 1 K ( s , t ) M I 1 I 2 I 3 I 4 ( s , t ; α ) , (56) \mathcal{G}^{I_1I_2I_3I_4}(U,V;\alpha)
=\int_{-i\infty}^{i\infty}\frac{ds\,dt}{(4\pi i)^2}
U^{s/2}V^{t/2-1}\mathcal{K}(s,t)\mathcal{M}^{I_1I_2I_3I_4}(s,t;\alpha), \tag*{(56)} G I 1 I 2 I 3 I 4 ( U , V ; α ) = ∫ − i ∞ i ∞ ( 4 π i ) 2 d s d t U s /2 V t /2 − 1 K ( s , t ) M I 1 I 2 I 3 I 4 ( s , t ; α ) , (56)
where s + t + u = 4 s+t+u=4 s + t + u = 4 , and the corresponding Mellin measure reads
K ( s , t ) ≡ Γ ( 2 − s 2 ) 2 Γ ( 2 − t 2 ) 2 Γ ( 2 − u 2 ) 2 . (57) \mathcal{K}(s,t)\equiv
\Gamma\left(\frac{2-s}{2}\right)^2
\Gamma\left(\frac{2-t}{2}\right)^2
\Gamma\left(\frac{2-u}{2}\right)^2. \tag*{(57)} K ( s , t ) ≡ Γ ( 2 2 − s ) 2 Γ ( 2 2 − t ) 2 Γ ( 2 2 − u ) 2 . (57)
The tree-level supergluon amplitude takes the color decomposition
M I 1 I 2 I 3 I 4 ( s , t ; α ) = ( C 222 ) 2 ( c s M s + c t M t + c u M u ) , (58) \mathcal{M}^{I_1I_2I_3I_4}(s,t;\alpha)
=(C_{222})^2\left(c_sM_s+c_tM_t+c_uM_u\right), \tag*{(58)} M I 1 I 2 I 3 I 4 ( s , t ; α ) = ( C 222 ) 2 ( c s M s + c t M t + c u M u ) , (58)
where C 222 C_{222} C 222 is the OPE coefficient of the unit-normalized flavor-current multiplet, and
M s ( s , t ; α ) = [ ( 1 − α ) s + t − 2 ] [ 1 π 2 s − 1 π 5 / 2 s Γ ( 1 2 − s 2 ) Γ ( 1 − s 2 ) ] , (59) M_s(s,t;\alpha) = \left[(1-\alpha)s+t-2\right]\left[\frac{1}{\pi^2s}-\frac{1}{\pi^{5/2}s}\frac{\Gamma\left(\frac{1}{2}-\frac{s}{2}\right)}{\Gamma\left(1-\frac{s}{2}\right)}\right], \tag*{(59)} M s ( s , t ; α ) = [ ( 1 − α ) s + t − 2 ] [ π 2 s 1 − π 5/2 s 1 Γ ( 1 − 2 s ) Γ ( 2 1 − 2 s ) ] , (59)
while M t M_t M t and M u M_u M u are obtained by crossing. In the conventions of appendix A of [38 ] , we can read the contributions of the vector and scalar exchanges, M Δ , ℓ \mathcal{M}_{\Delta,\ell} M Δ , ℓ , whose relative coefficient is fixed by superconformal symmetry:
M s ( s , t ; α ) = − 1 16 M 2 , 1 ( s ) ( s , t ) + ( 2 α − 1 ) [ 1 2 M 1 , 0 ( s ) ( s ) − 1 2 π 2 ] . (60) M_s(s,t;\alpha)=-\frac{1}{16}\mathcal{M}^{(s)}_{2,1}(s,t)+(2\alpha-1)\left[\frac{1}{2}\mathcal{M}^{(s)}_{1,0}(s)-\frac{1}{2\pi^2}\right]. \tag*{(60)} M s ( s , t ; α ) = − 16 1 M 2 , 1 ( s ) ( s , t ) + ( 2 α − 1 ) [ 2 1 M 1 , 0 ( s ) ( s ) − 2 π 2 1 ] . (60)
The comparison with our microscopic result (52) is now immediate. The three contributions are identified as13
W s ⟷ − 1 16 M 2 , 1 ( s ) ( s , t ) , B s ⟷ 1 2 M 1 , 0 ( s ) ( s ) , W c ⟷ − 1 2 π 2 . (61) W_s\longleftrightarrow-\frac{1}{16}\mathcal{M}^{(s)}_{2,1}(s,t),\qquad B_s\longleftrightarrow\frac{1}{2}\mathcal{M}^{(s)}_{1,0}(s),\qquad W_c\longleftrightarrow-\frac{1}{2\pi^2}. \tag*{(61)} W s ⟷ − 16 1 M 2 , 1 ( s ) ( s , t ) , B s ⟷ 2 1 M 1 , 0 ( s ) ( s ) , W c ⟷ − 2 π 2 1 . (61)
Thus the bulk gauge-field exchange supplies the vector component A 2 A_2 A 2 of the exchanged current multiplet, while the boundary scalar exchange induced by the non-linear Nahm condition supplies precisely its scalar component s 2 s_2 s 2 . The quartic scalar interaction provides the corresponding contact contribution. Together, these three terms reproduce (60) , with the relative R-symmetry factor ( R u − R t ) / R s = 2 α − 1 (R_u-R_t)/R_s=2\alpha-1 ( R u − R t ) / R s = 2 α − 1 following from our momentum-space result (52) .
Finally, converting from our momentum normalization (51) to the unit position-space normalization, O 2 = 2 π O ^ 2 \mathcal{O}_2=\sqrt{2\pi}\,\widehat{\mathcal{O}}_2 O 2 = 2 π O 2 , the tree-level matching fixes the leading weak-coupling normalization,
( C 222 ) 2 = g Y M 2 4 κ + O ( g Y M 4 ) . (62) (C_{222})^2=\frac{g_{\mathrm{YM}}^2}{4\kappa}+O(g_{\mathrm{YM}}^4). \tag*{(62)} ( C 222 ) 2 = 4 κ g YM 2 + O ( g YM 4 ) . (62)
Tree-level CFT data and one-loop strategy
At tree level, the four-point function is determined entirely by the external current multiplet and its allowed cubic and quartic interactions. At one loop, by contrast, the answer is sensitive to the complete spectrum of states that can propagate as intermediate states. We expand the four-point correlator at weak coupling as
G = G G F F + ε G t r e e + ε 2 G 1 − l o o p + ⋯ , (63) \mathcal{G}=\mathcal{G}_{\mathrm{GFF}}+\varepsilon\mathcal{G}_{\mathrm{tree}}+\varepsilon^2\mathcal{G}_{\mathrm{1-loop}}+\cdots, \tag*{(63)} G = G GFF + ε G tree + ε 2 G 1 − loop + ⋯ , (63)
where ε = ( C 222 ) 2 = 8 / C J \varepsilon=(C_{222})^2=8/C_J ε = ( C 222 ) 2 = 8/ C J , and C J C_J C J is the flavor central charge in the conventions of [38 ] .14 The leading term is the disconnected generalized-free-field (GFF) correlator, which contains the double-trace families of the dimension-one moment-map operator,
[ O 2 O 2 ] n , ℓ , Δ n , ℓ ( 0 ) = 2 Δ O 2 + 2 n + ℓ = 2 + 2 n + ℓ . (64) [\mathcal{O}_2\mathcal{O}_2]_{n,\ell},\qquad\Delta_{n,\ell}^{(0)}=2\Delta_{\mathcal{O}_2}+2n+\ell=2+2n+\ell. \tag*{(64)} [ O 2 O 2 ] n , ℓ , Δ n , ℓ ( 0 ) = 2 Δ O 2 + 2 n + ℓ = 2 + 2 n + ℓ . (64)
Interactions perturb these data starting at order ε \varepsilon ε . For a double-trace family of spin ℓ \ell ℓ , the dimension and squared OPE coefficient admit the expansions
Δ n , ℓ = Δ n , ℓ ( 0 ) + ε γ n , ℓ ( 1 ) + ε 2 γ n , ℓ ( 2 ) + ⋯ , (65) \Delta_{n,\ell}=\Delta_{n,\ell}^{(0)}+\varepsilon\gamma_{n,\ell}^{(1)}+\varepsilon^2\gamma_{n,\ell}^{(2)}+\cdots, \tag*{(65)} Δ n , ℓ = Δ n , ℓ ( 0 ) + ε γ n , ℓ ( 1 ) + ε 2 γ n , ℓ ( 2 ) + ⋯ , (65)
a n , ℓ = A n , ℓ + ε a n , ℓ ( 1 ) + ε 2 a n , ℓ ( 2 ) + ⋯ . (66) a_{n,\ell}=A_{n,\ell}+\varepsilon a_{n,\ell}^{(1)}+\varepsilon^2a_{n,\ell}^{(2)}+\cdots. \tag*{(66)} a n , ℓ = A n , ℓ + ε a n , ℓ ( 1 ) + ε 2 a n , ℓ ( 2 ) + ⋯ . (66)
Here A n , ℓ A_{n,\ell} A n , ℓ denotes the GFF coefficient fixed by the disconnected correlator, while γ n , ℓ ( k ) \gamma_{n,\ell}^{(k)} γ n , ℓ ( k ) are the interaction-induced anomalous dimensions of the exchanged long multiplets. The dimensions of protected short multiplets, by contrast, are fixed by supersymmetry.
These anomalous dimensions generate logarithmic terms in the position-space correlator. To see this, recall that a bosonic conformal block behaves in the OPE limit U → 0 U\to0 U → 0 at fixed V V V as
G Δ , ℓ ( U , V ) = U Δ − ℓ 2 [ g Δ , ℓ ( V ) + O ( U ) ] . (67) G_{\Delta,\ell}(U,V)=U^{\frac{\Delta-\ell}{2}}\left[g_{\Delta,\ell}(V)+O(U)\right]. \tag*{(67)} G Δ , ℓ ( U , V ) = U 2 Δ − ℓ [ g Δ , ℓ ( V ) + O ( U ) ] . (67)
Substituting the perturbative expansion of the dimension into the leading power of U U U gives
U Δ n , ℓ − ℓ 2 = U Δ n , ℓ ( 0 ) − ℓ 2 [ 1 + ε γ n , ℓ ( 1 ) 2 log U + ε 2 ( γ n , ℓ ( 2 ) 2 log U + ( γ n , ℓ ( 1 ) ) 2 8 log 2 U ) + ⋯ ] . (68) U^{\frac{\Delta_{n,\ell}-\ell}{2}}=U^{\frac{\Delta_{n,\ell}^{(0)}-\ell}{2}}\left[1+\varepsilon\frac{\gamma_{n,\ell}^{(1)}}{2}\log U+\varepsilon^2\left(\frac{\gamma_{n,\ell}^{(2)}}{2}\log U+\frac{\left(\gamma_{n,\ell}^{(1)}\right)^2}{8}\log^2 U\right)+\cdots\right]. \tag*{(68)} U 2 Δ n , ℓ − ℓ = U 2 Δ n , ℓ ( 0 ) − ℓ 1 + ε 2 γ n , ℓ ( 1 ) log U + ε 2 2 γ n , ℓ ( 2 ) log U + 8 ( γ n , ℓ ( 1 ) ) 2 log 2 U + ⋯ . (68)
The tree-level anomalous dimensions γ n , ℓ ( 1 ) \gamma_{n,\ell}^{(1)} γ n , ℓ ( 1 ) generate both the single logarithm at order ε \varepsilon ε and the double logarithm at order ε 2 \varepsilon^2 ε 2 . In particular, the corresponding coefficients for an individual exchanged family are
G t r e e ⊃ 1 2 A n , ℓ γ n , ℓ ( 1 ) log U , G 1 -loop ⊃ 1 8 A n , ℓ ( γ n , ℓ ( 1 ) ) 2 log 2 U , (69) \mathcal{G}_{\mathrm{tree}}\supset\frac{1}{2}A_{n,\ell}\gamma_{n,\ell}^{(1)}\log U,\qquad\mathcal{G}_{1\text{-loop}}\supset\frac{1}{8}A_{n,\ell}\left(\gamma_{n,\ell}^{(1)}\right)^2\log^2 U, \tag*{(69)} G tree ⊃ 2 1 A n , ℓ γ n , ℓ ( 1 ) log U , G 1 -loop ⊃ 8 1 A n , ℓ ( γ n , ℓ ( 1 ) ) 2 log 2 U , (69)
where the appropriate unperturbed conformal-block dependence is understood, and flavor labels are suppressed for the moment. At one loop, the correlator therefore has the schematic expansion
G 1 -loop ( U , V ) = F 2 ( U , V ) log 2 U + F 1 ( U , V ) log U + F 0 ( U , V ) . (70) \mathcal{G}_{1\text{-loop}}(U,V)=F_2(U,V)\log^2 U+F_1(U,V)\log U+F_0(U,V). \tag*{(70)} G 1 -loop ( U , V ) = F 2 ( U , V ) log 2 U + F 1 ( U , V ) log U + F 0 ( U , V ) . (70)
The crucial observation is that the coefficient F 2 ( U , V ) F_2(U,V) F 2 ( U , V ) is determined entirely by the tree-level anomalous dimensions and the leading GFF OPE coefficients. By contrast, F 1 ( U , V ) F_1(U,V) F 1 ( U , V ) and F 0 ( U , V ) F_0(U,V) F 0 ( U , V ) also contain genuinely new one-loop CFT data, including the second-order anomalous dimensions and corrections to the OPE coefficients. To isolate the information contained in F 2 ( U , V ) \mathcal{F}_2(U,V) F 2 ( U , V ) , we can take the double-discontinuity of the one-loop expression.15
dDisc U G 1 -loop ( U , V ) = 4 π 2 F 2 ( U , V ) . (71) \operatorname{dDisc}_{U}\mathcal{G}_{1\text{-loop}}(U,V)=4\pi^2\mathcal{F}_2(U,V). \tag*{(71)} dDisc U G 1 -loop ( U , V ) = 4 π 2 F 2 ( U , V ) . (71)
Indeed, the unperturbed double-trace operators contribute with integer powers of U U U , and the double-discontinuity annihilates analytic and single-logarithmic terms.
This is the basis of the AdS unitarity method [39 , 43 ] . In flat space, unitarity relates the discontinuity of a loop amplitude to intermediate physical states going on shell. The corresponding unitarity cut is computed by replacing the cut propagators with on-shell states and gluing lower-order amplitudes [73 , 74 ] . Although AdS has no conventional S S S -matrix, the analogous information is encoded in the double-discontinuity of boundary correlators. At one loop, the two-particle cut is obtained by gluing tree-level boundary correlators and summing over the intermediate double-trace states (64) . This is where the microscopic spectrum of the theory becomes essential. In the A d S d + 1 × S 3 \mathrm{AdS}_{d+1}\times S^3 AdS d + 1 × S 3 supergluon constructions, four external p = 2 p=2 p = 2 operators can couple across the cut to higher Kaluza–Klein multiplets. The one-loop ⟨ 2222 ⟩ \langle2222\rangle ⟨ 2222 ⟩ correlator therefore requires tree-level amplitudes of the form ⟨ 22 p p ⟩ \langle22pp\rangle ⟨ 22 pp ⟩ for all allowed p p p , and operators belonging to different KK levels can mix. In contrast, our theory has no internal compact space and hence no KK tower. Under the identification established at tree level, the two-particle cut is obtained by gluing the p = 2 p=2 p = 2 sector of [38 ] to itself.
Moreover, there is no mixing problem to solve. At order ε 0 \varepsilon^0 ε 0 , the states that contribute to the cut are two-particle states of the flavor-current multiplet, which transform in the tensor product of two copies of B [ 0 ] ( 2 ; 0 ) \mathcal{B}_{[0]}^{(2;0)} B [ 0 ] ( 2 ; 0 ) . In this tensor product, each long multiplet L Δ [ 2 ℓ ] ( 0 ; 0 ) \mathcal{L}_{\Delta}^{[2\ell](0;0)} L Δ [ 2 ℓ ] ( 0 ; 0 ) in a given flavor representation appears at most once. Indeed, such a multiplet contains exactly one conformal primary with S U ( 2 ) H SU(2)_H S U ( 2 ) H spin two and S U ( 2 ) C SU(2)_C S U ( 2 ) C spin zero, of dimension Δ + 2 \Delta+2 Δ + 2 and spin ℓ \ell ℓ : this is the top component discussed below, see (72) . In the two-particle space, S U ( 2 ) H SU(2)_H S U ( 2 ) H spin two can only be reached with two moment maps, since all other components of the multiplet carry S U ( 2 ) H SU(2)_H S U ( 2 ) H spin at most one half, and there is exactly one such generalized-free double-trace primary for each dimension, spin and flavor symmetry. The double-trace long multiplets are therefore non-degenerate, and the one-loop double logarithm is controlled by A n , ℓ ( γ n , ℓ ( 1 ) ) 2 A_{n,\ell}(\gamma_{n,\ell}^{(1)})^2 A n , ℓ ( γ n , ℓ ( 1 ) ) 2 for each individual multiplet, with no average over degenerate operators. As usual in the AdS unitarity method [39 ] , states with more than two particles do not contribute to the double-discontinuity at this order.
The double-discontinuity alone does not determine the remaining single-logarithmic and analytic terms. Together with crossing symmetry and analyticity, however, the cut data can be used to reconstruct the non-local part of the correlator, up to the familiar low-spin or contact-term ambiguities. In position space, this can be implemented through the Lorentzian inversion formula [75 ] or conformal dispersion relations [76 ] . In the next section, we will instead avoid performing this reconstruction directly in position space and translate the double-discontinuity data into Mellin space, where the problem becomes a much simpler algebraic reconstruction of the one-loop amplitude.
The highest R-symmetry channel
The four-point function of moment-map operators is naturally decomposed into O S p ( 4 ∣ 4 ) OSp(4|4) O S p ( 4∣4 ) superconformal blocks rather than ordinary conformal blocks. Each superconformal block is a fixed linear combination of three-dimensional bosonic conformal blocks G Δ , ℓ ( U , V ) G_{\Delta,\ell}(U,V) G Δ , ℓ ( U , V ) , multiplied by S U ( 2 ) H SU(2)_{H} S U ( 2 ) H R-symmetry harmonics Y 2 j ( α ) \mathcal{Y}_{2j}(\alpha) Y 2 j ( α ) , with relative coefficients determined by superconformal symmetry [54 ] . Since the dependence on the R-symmetry cross-ratio is a polynomial of degree two, it is particularly convenient to focus on the highest R-symmetry channel, namely the coefficient of α 2 \alpha^{2} α 2 . Equivalently, in the harmonic basis this isolates the highest R-symmetry harmonic Y 4 ( α ) = α 2 / 2 − α / 2 + 1 / 12 \mathcal{Y}_{4}(\alpha)=\alpha^{2}/2-\alpha/2+1/12 Y 4 ( α ) = α 2 /2 − α /2 + 1/12 , in the conventions of [38 ] .
The unprotected long multiplets allowed by the moment-map OPE have R-symmetry-singlet superconformal primaries (27) . Although their full superconformal blocks contain several bosonic conformal blocks, the highest R-symmetry component contains only one: a bosonic conformal primary at the second bosonic level of the supermultiplet, with dimension Δ + 2 \Delta+2 Δ + 2 and the same spacetime spin ℓ \ell ℓ . Therefore, the projection of the superconformal block is
[ α 2 ] G [ 2 ℓ ] Δ ( 0 ; 0 ) ( U , V ; α ) = q Δ , ℓ G Δ + 2 , ℓ ( U , V ) , (72) [\alpha^{2}]\,\mathcal{G}_{[2\ell]_{\Delta}}^{(0;0)}(U,V;\alpha)=q_{\Delta,\ell}G_{\Delta+2,\ell}(U,V), \tag*{(72)} [ α 2 ] G [ 2 ℓ ] Δ ( 0 ; 0 ) ( U , V ; α ) = q Δ , ℓ G Δ + 2 , ℓ ( U , V ) , (72)
where q Δ , ℓ q_{\Delta,\ell} q Δ , ℓ includes the fixed superconformal-block prefactor and the normalization of the R-symmetry harmonic. For the double-trace long operators appearing perturbatively in our correlator (64) , the selected bosonic block has dimension 4 + 2 n + ℓ 4+2n+\ell 4 + 2 n + ℓ . We therefore introduce the shorthand
G n , ℓ ( U , V ) ≡ G 4 + 2 n + ℓ , ℓ ( U , V ) . (73) G_{n,\ell}(U,V)\equiv G_{4+2n+\ell,\ell}(U,V). \tag*{(73)} G n , ℓ ( U , V ) ≡ G 4 + 2 n + ℓ , ℓ ( U , V ) . (73)
This simplification applies independently in each allowed flavor representation. After projecting onto a fixed flavor channel and stripping the corresponding group-theory factor, the tree-level log U \log U log U term can be expanded directly in the bosonic top-block basis,
[ α 2 ] G t r e e ( U , V ) ∣ log U = ∑ n , ℓ ω n , ℓ G n , ℓ ( U , V ) . (74) [\alpha^{2}]\,\mathcal{G}_{\mathrm{tree}}(U,V)\big|_{\log U}=\sum_{n,\ell}\omega_{n,\ell}\,G_{n,\ell}(U,V). \tag*{(74)} [ α 2 ] G tree ( U , V ) l o g U = n , ℓ ∑ ω n , ℓ G n , ℓ ( U , V ) . (74)
The coefficients ω n , ℓ \omega_{n,\ell} ω n , ℓ are the universal color-stripped coefficients of these bosonic top blocks. After restoring the flavor projection, they are related to the physical CFT data by ω n , ℓ ∝ A n , ℓ ( 1 ) γ n , ℓ \omega_{n,\ell}\propto A_{n,\ell}^{(1)}\gamma_{n,\ell} ω n , ℓ ∝ A n , ℓ ( 1 ) γ n , ℓ , as discussed below.16 To determine these coefficients, we first rewrite the tree-level amplitude in position space and extract its log U \log U log U term. Then, expanding at small U U U and around V = 1 V=1 V = 1 , we match to the corresponding lightcone expansion of the 3 d 3d 3 d bosonic blocks G n , ℓ G_{n,\ell} G n , ℓ , determining ω n , ℓ \omega_{n,\ell} ω n , ℓ recursively.
Position-space representation
In many familiar holographic examples, exchange Witten diagrams truncate to finite sums of contact diagrams, or D D D -functions [77 , 78 , 79 ] . Although the usual truncation condition is not satisfied in the present AdS4 _4 4 problem, the tree-level Mellin amplitude still admits a finite representation in terms of normalized D ˉ \bar{D} D ˉ -functions, similarly to other 3 d 3d 3 d holographic correlators [80 ] . The required weights include half-integer and vanishing values, which are defined by analytic continuation of their Mellin–Barnes representation. Our conventions and the corresponding Mellin-to-position-space dictionary are collected in appendix § 10 . The α 2 \alpha^2 α 2 projection of the Mellin amplitude (59) is
[ α 2 ] M s = 0 , [ α 2 ] M t = − 2 ( s − 2 ) π 3 [ π 2 t + π Γ ( 1 − t 2 ) t 2 Γ ( − t 2 ) ] , [ α 2 ] M u = 2 ( s − 2 ) π 3 [ π 2 u + π Γ ( 1 − u 2 ) u 2 Γ ( − u 2 ) ] . (75) \begin{aligned}
\left[\alpha^2\right]M_s &= 0,\\
\left[\alpha^2\right]M_t &= -\frac{2(s-2)}{\pi^3}\left[\frac{\pi}{2t}+\frac{\sqrt{\pi}\,\Gamma\left(\frac{1-t}{2}\right)}{t^2\Gamma\left(-\frac{t}{2}\right)}\right],\\
\left[\alpha^2\right]M_u &= \frac{2(s-2)}{\pi^3}\left[\frac{\pi}{2u}+\frac{\sqrt{\pi}\,\Gamma\left(\frac{1-u}{2}\right)}{u^2\Gamma\left(-\frac{u}{2}\right)}\right]. \tag*{(75)}
\end{aligned} [ α 2 ] M s [ α 2 ] M t [ α 2 ] M u = 0 , = − π 3 2 ( s − 2 ) [ 2 t π + t 2 Γ ( − 2 t ) π Γ ( 2 1 − t ) ] , = π 3 2 ( s − 2 ) [ 2 u π + u 2 Γ ( − 2 u ) π Γ ( 2 1 − u ) ] . (75)
and the corresponding position-space correlator takes the form
[ α 2 ] G t r e e ( U , V ) = U 2 [ c t T t ( U , V ) + c u T u ( U , V ) ] , (76) \left[\alpha^2\right]\mathcal{G}_{\mathrm{tree}}(U,V)=U^2\left[c_t T_t(U,V)+c_u T_u(U,V)\right], \tag*{(76)} [ α 2 ] G tree ( U , V ) = U 2 [ c t T t ( U , V ) + c u T u ( U , V ) ] , (76)
with the two channel functions
T u ( U , V ) ≡ 1 π 3 ( π D ˉ 2 , 1 , 1 , 0 ( U , V ) − π D ˉ 3 , 1 2 , 1 , 1 2 ( U , V ) ) , T t ( U , V ) ≡ − 1 π 3 ( π V − 1 D ˉ 2 , 1 , 0 , 1 ( U , V ) − π V − 1 / 2 D ˉ 1 , 3 2 , 1 2 , 0 ( U , V ) ) . (77) \begin{aligned}
T_u(U,V)&\equiv\frac{1}{\pi^3}\left(\pi\bar{D}_{2,1,1,0}(U,V)-\sqrt{\pi}\bar{D}_{3,\frac{1}{2},1,\frac{1}{2}}(U,V)\right),\\
T_t(U,V)&\equiv-\frac{1}{\pi^3}\left(\pi V^{-1}\bar{D}_{2,1,0,1}(U,V)-\sqrt{\pi}V^{-1/2}\bar{D}_{1,\frac{3}{2},\frac{1}{2},0}(U,V)\right). \tag*{(77)}
\end{aligned} T u ( U , V ) T t ( U , V ) ≡ π 3 1 ( π D ˉ 2 , 1 , 1 , 0 ( U , V ) − π D ˉ 3 , 2 1 , 1 , 2 1 ( U , V ) ) , ≡ − π 3 1 ( π V − 1 D ˉ 2 , 1 , 0 , 1 ( U , V ) − π V − 1/2 D ˉ 1 , 2 3 , 2 1 , 0 ( U , V ) ) . (77)
Here, we have used the color Jacobi identity (32) and chosen ( c t , c u ) (c_t,c_u) ( c t , c u ) as independent color factors.
Logarithmic term. To extract the tree-level CFT data entering the one-loop unitarity construction, we only need the coefficient of log U \log U log U in (76) . It can be obtained directly from the Mellin–Barnes representation of the analytically continued D ˉ \bar{D} D ˉ -functions, giving
D ˉ r 1 r 2 r 3 r 4 ( U , V ) ∣ log U = ∑ m , n ≥ 0 d m , n ( r i ) U m ( 1 − V ) n , (78) \left.\bar{D}_{r_1r_2r_3r_4}(U,V)\right|_{\log U}=\sum_{m,n\geq0}d_{m,n}(r_i)U^m(1-V)^n, \tag*{(78)} D ˉ r 1 r 2 r 3 r 4 ( U , V ) l o g U = m , n ≥ 0 ∑ d m , n ( r i ) U m ( 1 − V ) n , (78)
where the explicit coefficients d m , n ( r i ) d_{m,n}(r_i) d m , n ( r i ) are given in appendix § 10 .
Tree-level data
The two functions T t T_t T t and T u T_u T u introduced above are convenient for the explicit D ˉ \bar{D} D ˉ -function representation, since they multiply the two independent color factors c t c_t c t and c u c_u c u . However, before performing the conformal-block decomposition, it is useful to restore a Jacobi-symmetric three-channel representation,
[ α 2 ] G t r e e U 2 = c s H s + c t H t + c u H u . (79) \frac{[\alpha^2]\mathcal{G}_{\mathrm{tree}}}{U^2}=c_sH_s+c_tH_t+c_uH_u. \tag*{(79)} U 2 [ α 2 ] G tree = c s H s + c t H t + c u H u . (79)
The functions H x H_x H x are defined only up to a common shift H x → H x + χ H_x\to H_x+\chi H x → H x + χ , which drops out by the color Jacobi identity (32) . We fix this redundancy by imposing
H s + H t + H u = 0. (80) H_s+H_t+H_u=0. \tag*{(80)} H s + H t + H u = 0. (80)
With this choice, T t = H t − H s T_t=H_t-H_s T t = H t − H s and T u = H u − H s T_u=H_u-H_s T u = H u − H s .
The coefficient of log U \log U log U is decomposed in the basis of three-dimensional bosonic top blocks (73) . We define the universal dynamical coefficient ω n , ℓ \omega_{n,\ell} ω n , ℓ through the Jacobi-symmetric channel decomposition,
U 2 H x ( U , V ) ∣ log U = ∑ n , ℓ ≥ 0 q x ( ℓ ) ω n , ℓ G n , ℓ ( U , V ) , x = s , t , u , (81) \left.U^2H_x(U,V)\right|_{\log U}=\sum_{n,\ell\geq0}q_x(\ell)\,\omega_{n,\ell}\,G_{n,\ell}(U,V),\qquad x=s,t,u, \tag*{(81)} U 2 H x ( U , V ) l o g U = n , ℓ ≥ 0 ∑ q x ( ℓ ) ω n , ℓ G n , ℓ ( U , V ) , x = s , t , u , (81)
where
q s ( ℓ ) = 1 − ( − 1 ) ℓ 2 , q t ( ℓ ) = − ( 1 + ( − 1 ) ℓ 2 ) , q u ( ℓ ) = 1 2 + ( − 1 ) ℓ . (82) q_s(\ell)=\frac{1-(-1)^\ell}{2},\qquad q_t(\ell)=-\left(\frac{1+(-1)^\ell}{2}\right),\qquad q_u(\ell)=\frac{1}{2}+(-1)^\ell. \tag*{(82)} q s ( ℓ ) = 2 1 − ( − 1 ) ℓ , q t ( ℓ ) = − ( 2 1 + ( − 1 ) ℓ ) , q u ( ℓ ) = 2 1 + ( − 1 ) ℓ . (82)
Notice that q s ( ℓ ) + q t ( ℓ ) + q u ( ℓ ) = 0 q_s(\ell)+q_t(\ell)+q_u(\ell)=0 q s ( ℓ ) + q t ( ℓ ) + q u ( ℓ ) = 0 , consistently with the Jacobi-symmetric choice above. All explicit channel-dependent spin-parity normalization is carried by q x ( ℓ ) q_x(\ell) q x ( ℓ ) ; the closed formula for ω n , ℓ \omega_{n,\ell} ω n , ℓ contains no additional piecewise even/odd prefactor. Translating to the two functions used in the explicit D ˉ \bar{D} D ˉ representation, we obtain
U 2 T t ( U , V ) ∣ log U = − 3 2 ∑ n , ℓ ≥ 0 ω n , ℓ G n , ℓ ( U , V ) , (83) \left.U^2T_t(U,V)\right|_{\log U}=-\frac{3}{2}\sum_{n,\ell\geq0}\omega_{n,\ell}\,G_{n,\ell}(U,V), \tag*{(83)} U 2 T t ( U , V ) l o g U = − 2 3 n , ℓ ≥ 0 ∑ ω n , ℓ G n , ℓ ( U , V ) , (83)
U 2 T u ( U , V ) ∣ log U = 3 2 ∑ n , ℓ ≥ 0 ( − 1 ) ℓ ω n , ℓ G n , ℓ ( U , V ) . (84) \left.U^2T_u(U,V)\right|_{\log U}=\frac{3}{2}\sum_{n,\ell\geq0}(-1)^\ell\omega_{n,\ell}\,G_{n,\ell}(U,V). \tag*{(84)} U 2 T u ( U , V ) l o g U = 2 3 n , ℓ ≥ 0 ∑ ( − 1 ) ℓ ω n , ℓ G n , ℓ ( U , V ) . (84)
The recursive block extraction can be organized using the small-U U U lightcone expansion of the three-dimensional scalar blocks, which gives
ω n , ℓ = 2 4 − 2 n 3 π ( 2 n + 1 ) ! ( n ! ) 2 ( 2 ℓ + 1 ) ( 2 ℓ + 2 n + 3 ) ℓ + 1 Γ ( ℓ + n + 2 ) Γ ( ℓ + 2 n + 2 ) Γ ( ℓ + n + 5 2 ) Γ ( ℓ + 2 n + 5 2 ) Q n , ℓ . (85) \omega_{n,\ell}=\frac{2^{4-2n}}{3\pi}\frac{(2n+1)!}{(n!)^2}\frac{(2\ell+1)(2\ell+2n+3)}{\ell+1}\frac{\Gamma(\ell+n+2)\Gamma(\ell+2n+2)}{\Gamma\left(\ell+n+\frac{5}{2}\right)\Gamma\left(\ell+2n+\frac{5}{2}\right)}Q_{n,\ell}. \tag*{(85)} ω n , ℓ = 3 π 2 4 − 2 n ( n ! ) 2 ( 2 n + 1 )! ℓ + 1 ( 2 ℓ + 1 ) ( 2 ℓ + 2 n + 3 ) Γ ( ℓ + n + 2 5 ) Γ ( ℓ + 2 n + 2 5 ) Γ ( ℓ + n + 2 ) Γ ( ℓ + 2 n + 2 ) Q n , ℓ . (85)
where Q n , ℓ Q_{n,\ell} Q n , ℓ is a terminating hypergeometric function,
Q n , ℓ ≡ 4 F 3 ( 1 , 1 , − n , − n − 1 2 3 2 , ℓ + 2 , − ℓ − 2 n − 1 ; 1 ) . (86) Q_{n,\ell} \equiv{}_4F_3\left(\begin{matrix} 1,\,1,\,-n,\,-n-\frac{1}{2} \\ \frac{3}{2},\,\ell+2,\,-\ell-2n-1 \end{matrix};1\right). \tag*{(86)} Q n , ℓ ≡ 4 F 3 ( 1 , 1 , − n , − n − 2 1 2 3 , ℓ + 2 , − ℓ − 2 n − 1 ; 1 ) . (86)
Generalized-free data
For four identical external scalars of dimension one, the GFF double-trace family has
Δ m , ℓ ( 0 ) = 2 + 2 m + ℓ , m ∈ Z ≥ 0 . (87) \Delta_{m,\ell}^{(0)} = 2+2m+\ell,\qquad m\in\mathbb{Z}_{\geq0}. \tag*{(87)} Δ m , ℓ ( 0 ) = 2 + 2 m + ℓ , m ∈ Z ≥ 0 . (87)
In the 12 → 34 12\to34 12 → 34 channel, the complete disconnected α 2 \alpha^2 α 2 component is
[ α 2 ] G G F F I 1 I 2 I 3 I 4 ( U , V ) = U δ I 1 I 3 δ I 2 I 4 + U V δ I 1 I 4 δ I 2 I 3 . (88) [\alpha^2]G_{\mathrm{GFF}}^{I_1I_2I_3I_4}(U,V)=U\delta^{I_1I_3}\delta^{I_2I_4}+\frac{U}{V}\delta^{I_1I_4}\delta^{I_2I_3}. \tag*{(88)} [ α 2 ] G GFF I 1 I 2 I 3 I 4 ( U , V ) = U δ I 1 I 3 δ I 2 I 4 + V U δ I 1 I 4 δ I 2 I 3 . (88)
We define the coefficients A m , ℓ W i c k A_{m,\ell}^{\mathrm{Wick}} A m , ℓ Wick of the single U / V U/V U / V Wick contraction by
U V = ∑ m , ℓ ≥ 0 A m , ℓ W i c k G 2 + 2 m + ℓ , ℓ ( U , V ) , (89) \frac{U}{V}=\sum_{m,\ell\geq0}A_{m,\ell}^{\mathrm{Wick}}G_{2+2m+\ell,\ell}(U,V), \tag*{(89)} V U = m , ℓ ≥ 0 ∑ A m , ℓ Wick G 2 + 2 m + ℓ , ℓ ( U , V ) , (89)
while exchanging points 3 and 4 gives the other contraction, with a parity factor ( − 1 ) ℓ (-1)^\ell ( − 1 ) ℓ . In the normalization used for the 3 d 3d 3 d blocks, for m ≥ 1 m\geq1 m ≥ 1 the single-channel coefficients are17
A m , ℓ W i c k = 16 ( 2 ℓ + 1 ) π Γ ( m + 1 2 ) Γ ( m + ℓ + 1 ) Γ ( 2 m + ℓ + 1 ) Γ ( m + 1 ) Γ ( m + ℓ + 3 2 ) Γ ( 2 m + ℓ + 1 2 ) . (90) A_{m,\ell}^{\mathrm{Wick}}=\frac{16(2\ell+1)\sqrt{\pi}\,\Gamma\left(m+\frac{1}{2}\right)\Gamma(m+\ell+1)\Gamma(2m+\ell+1)}{\Gamma(m+1)\Gamma\left(m+\ell+\frac{3}{2}\right)\Gamma\left(2m+\ell+\frac{1}{2}\right)}. \tag*{(90)} A m , ℓ Wick = Γ ( m + 1 ) Γ ( m + ℓ + 2 3 ) Γ ( 2 m + ℓ + 2 1 ) 16 ( 2 ℓ + 1 ) π Γ ( m + 2 1 ) Γ ( m + ℓ + 1 ) Γ ( 2 m + ℓ + 1 ) . (90)
The top blocks relevant to the long sector of the α 2 \alpha^2 α 2 component have dimension 4 + 2 n + ℓ 4+2n+\ell 4 + 2 n + ℓ , implying that m = n + 1 m=n+1 m = n + 1 . We therefore define the single-channel coefficient
A n , ℓ c h = A n + 1 , ℓ W i c k . (91) A_{n,\ell}^{\mathrm{ch}}=A_{n+1,\ell}^{\mathrm{Wick}}. \tag*{(91)} A n , ℓ ch = A n + 1 , ℓ Wick . (91)
Flavor projection and physical anomalous dimensions
The coefficient ω n , ℓ \omega_{n,\ell} ω n , ℓ extracted above is a universal color-stripped dynamical quantity. To obtain physical CFT data, we must resolve the flavor structure in the 12 → 34 12\to34 12 → 34 channel. The Jacobi-symmetric decomposition (81) already determines the color operator multiplying a spin-ℓ \ell ℓ block,
C ℓ ≡ ∑ x = s , t , u q x ( ℓ ) c x = { 3 2 ( c u − c t ) , ℓ even , 3 2 c s , ℓ odd . (92) C_\ell\equiv\sum_{x=s,t,u}q_x(\ell)c_x=
\begin{cases}
\frac{3}{2}(c_u-c_t), & \ell\ \text{even},\\
\frac{3}{2}c_s, & \ell\ \text{odd}.
\end{cases} \tag*{(92)} C ℓ ≡ x = s , t , u ∑ q x ( ℓ ) c x = { 2 3 ( c u − c t ) , 2 3 c s , ℓ even , ℓ odd . (92)
We therefore only need to diagonalize C ℓ C_\ell C ℓ on the irreducible components R ⊂ A d j ⊗ A d j \mathcal{R} \subset\mathrm{Adj} \otimes\mathrm{Adj} R ⊂ Adj ⊗ Adj ,
C ℓ P R = λ R , ℓ P R , (93) C_\ell P_{\mathcal{R}} = \lambda_{\mathcal{R},\ell}P_{\mathcal{R}}, \tag*{(93)} C ℓ P R = λ R , ℓ P R , (93)
where P R P_{\mathcal{R}} P R is the projector onto R \mathcal{R} R . The color tensors naturally act as operators on A d j ⊗ A d j \mathrm{Adj}\otimes\mathrm{Adj} Adj ⊗ Adj . With the conventions of (28) , one has, for instance c u = T 1 ⋅ T 2 c_u = T_1\cdot T_2 c u = T 1 ⋅ T 2 , so their action on a fixed irreducible representation is determined by the corresponding quadratic Casimirs. Using also the Bose-symmetry selection rule σ R = ( − 1 ) ℓ \sigma_{\mathcal{R}} = (-1)^\ell σ R = ( − 1 ) ℓ , we obtain
λ R , ℓ = { 3 2 ( C R − 2 C A ) , ℓ even , R ⊂ Sym 2 ( A d j ) , 3 2 C A , ℓ odd , R = A d j A , (94) \lambda_{\mathcal{R},\ell} =
\begin{cases}
\frac{3}{2}\left(C_{\mathcal{R}} - 2C_A\right), & \ell\ \text{even},\quad\mathcal{R} \subset\operatorname{Sym}^2(\mathrm{Adj}),\\
\frac{3}{2}C_A, & \ell\ \text{odd},\quad\mathcal{R} = \mathrm{Adj}_A,
\end{cases} \tag*{(94)} λ R , ℓ = { 2 3 ( C R − 2 C A ) , 2 3 C A , ℓ even , R ⊂ Sym 2 ( Adj ) , ℓ odd , R = Adj A , (94)
where C A C_A C A and C R C_{\mathcal{R}} C R denote the quadratic Casimirs in the adjoint representation and in R \mathcal{R} R , respectively, with f I K L f J K L = C A δ I J f^{IKL}f^{JKL} = C_A\delta^{IJ} f I K L f J K L = C A δ I J . For odd spin, all other flavor representations have vanishing eigenvalue. Therefore,
ω n , ℓ R = λ R , ℓ ω n , ℓ . (95) \omega^{\mathcal{R}}_{n,\ell} = \lambda_{\mathcal{R},\ell}\,\omega_{n,\ell}. \tag*{(95)} ω n , ℓ R = λ R , ℓ ω n , ℓ . (95)
The same flavor projection must be applied to the generalized-free correlator. The two non-trivial Wick contractions give the flavor operator ( − 1 ) ℓ 1 + Π (-1)^\ell\mathbf{1}+\Pi ( − 1 ) ℓ 1 + Π , where Π \Pi Π exchanges the two adjoint indices. Using the Bose-symmetry selection rule, we obtain
A n , ℓ R = 2 ( − 1 ) ℓ A n , ℓ c h , (96) A^{\mathcal{R}}_{n,\ell} = 2(-1)^\ell A^{\mathrm{ch}}_{n,\ell}, \tag*{(96)} A n , ℓ R = 2 ( − 1 ) ℓ A n , ℓ ch , (96)
which is the signed coefficient of the projected top block. To extract the anomalous dimensions, recall that the bosonic top block selected by the α 2 \alpha^2 α 2 projection is a descendant of a long double-trace superprimary with
Δ n , ℓ , R s p = 2 + 2 n + ℓ + ϵ γ n , ℓ , R ( 1 ) + O ( ϵ 2 ) . (97) \Delta^{\mathrm{sp}}_{n,\ell,\mathcal{R}} = 2 + 2n + \ell+ \epsilon\,\gamma^{(1)}_{n,\ell,\mathcal{R}} + O(\epsilon^2). \tag*{(97)} Δ n , ℓ , R sp = 2 + 2 n + ℓ + ϵ γ n , ℓ , R ( 1 ) + O ( ϵ 2 ) . (97)
Its dimension is shifted by two units relative to the superprimary, so both acquire the same anomalous dimension. The coefficient of the tree-level logarithm is therefore
ω n , ℓ R = 1 2 A n , ℓ R γ n , ℓ , R ( 1 ) . (98) \omega^{\mathcal{R}}_{n,\ell} = \frac{1}{2}A^{\mathcal{R}}_{n,\ell}\gamma^{(1)}_{n,\ell,\mathcal{R}}. \tag*{(98)} ω n , ℓ R = 2 1 A n , ℓ R γ n , ℓ , R ( 1 ) . (98)
Combining the flavor projections of the logarithmic and generalized-free coefficients, we obtain
γ n , ℓ , R ( 1 ) = ( − 1 ) ℓ λ R , ℓ ω n , ℓ A n , ℓ c h . (99) \gamma^{(1)}_{n,\ell,\mathcal{R}} = (-1)^\ell\lambda_{\mathcal{R},\ell}\frac{\omega_{n,\ell}}{A^{\mathrm{ch}}_{n,\ell}}. \tag*{(99)} γ n , ℓ , R ( 1 ) = ( − 1 ) ℓ λ R , ℓ A n , ℓ ch ω n , ℓ . (99)
As a normalization check, at the lowest level we find γ 0 , 0 , R ( 1 ) = λ R , 0 / π 2 \gamma^{(1)}_{0,0,\mathcal{R}} = \lambda_{\mathcal{R},0}/\pi^2 γ 0 , 0 , R ( 1 ) = λ R , 0 / π 2 , corresponding to a long double-trace superprimary of unperturbed dimension two, whose bosonic top component has dimension four.
One-loop AdS4 _4 4 amplitude
Algorithm for the one-loop coefficients
We are now ready to use the tree-level CFT data to fix the non-local part of the one-loop amplitude, via the AdS unitarity method.
In the higher-dimensional examples where this construction was first implemented, the superconformal Ward identity can be solved in terms of a single reduced correlator, with the full R-symmetry dependence generated by differential operators in position space, or equivalently by finite-difference operators in Mellin space [78 , 79 , 81 ] . In three dimensions, the standard reduced position-space representation involves non-local operators [82 ] . We therefore construct the one-loop double-discontinuity directly from the full α 2 \alpha^2 α 2 moment-map correlator, and only afterwards reorganize the result in terms of a reduced Mellin amplitude. This step uses the recent reduced Mellin-space construction of [83 ] , adapted to three-dimensional theories with eight supercharges in [47 ] .18
The one-loop input is the log 2 U \log^2 U log 2 U coefficient in the highest R-symmetry channel of the full correlator. For a fixed flavor irreducible representation R \mathcal{R} R , the unitarity relation gives
[ α 2 ] G R 1 -loop ( U , V ) ∣ log 2 U = 1 2 ∑ n , ℓ ( ω n , ℓ R ) 2 A n , ℓ R G n , ℓ ( U , V ) . (100) [\alpha^2]\mathcal{G}^{1\text{-loop}}_{\mathcal{R}}(U,V)\big|_{\log^2 U}
= \frac{1}{2}\sum_{n,\ell}\frac{\left(\omega^{\mathcal{R}}_{n,\ell}\right)^2}{A^{\mathcal{R}}_{n,\ell}}G_{n,\ell}(U,V). \tag*{(100)} [ α 2 ] G R 1 -loop ( U , V ) l o g 2 U = 2 1 n , ℓ ∑ A n , ℓ R ( ω n , ℓ R ) 2 G n , ℓ ( U , V ) . (100)
As in the tree-level analysis, we now separate the universal kinematic spin sums from their flavor dressing. We define
F e v e n , o d d ( U , V ) ≡ 1 4 ∑ n , ℓ ≥ 0 1 ± ( − 1 ) ℓ 2 ω n , ℓ 2 A n , ℓ c h G n , ℓ ( U , V ) . (101) F_{\mathrm{even,odd}}(U,V) \equiv\frac{1}{4}\sum_{n,\ell\geq0}\frac{1\pm(-1)^\ell}{2}\frac{\omega^2_{n,\ell}}{A^{\mathrm{ch}}_{n,\ell}}G_{n,\ell}(U,V). \tag*{(101)} F even , odd ( U , V ) ≡ 4 1 n , ℓ ≥ 0 ∑ 2 1 ± ( − 1 ) ℓ A n , ℓ ch ω n , ℓ 2 G n , ℓ ( U , V ) . (101)
These are the two universal functions that we need to reconstruct. The flavor dependence is restored by multiplying each spin-ℓ \ell ℓ block by ( − 1 ) ℓ λ R , ℓ (-1)^\ell\lambda_{\mathcal{R},\ell} ( − 1 ) ℓ λ R , ℓ . Thus, an allowed even-spin channel is proportional to F e v e n F_{\mathrm{even}} F even , whereas an allowed odd-spin channel is proportional to − F o d d -F_{\mathrm{odd}} − F odd . This provides the position-space input for the AdS unitarity reconstruction.
Summary of the strategy. To expose the analytic structure of the one-loop double-discontinuity, we expand the universal functions at small U U U ,
F e v e n , o d d ( U , V ) = ∑ p ≥ 2 U p h p ± ( V ) . (102) F_{\mathrm{even,odd}}(U,V)=\sum_{p\geq2}U^p h_p^{\pm}(V). \tag*{(102)} F even , odd ( U , V ) = p ≥ 2 ∑ U p h p ± ( V ) . (102)
At fixed p p p , only finitely many double-trace levels contribute, while the spin sum remains infinite. The highest logarithmic singularity in the small-V V V expansion of the functions h p ± ( V ) h_p^{\pm}(V) h p ± ( V ) is proportional to log 2 V \log^2 V log 2 V . Since F e v e n , o d d F_{\mathrm{even,odd}} F even , odd determine the coefficient of log 2 U \log^2 U log 2 U in the one-loop correlator, these terms correspond to log 2 U log 2 V \log^2 U\,\log^2 V log 2 U log 2 V contributions.
Their Mellin-space interpretation is particularly simple. The Mellin measure has double poles at the double-trace locations (57) , while an additional simple pole of the Mellin amplitude at the same location produces a triple pole and hence a squared logarithm. Simultaneous log 2 U \log^2 U log 2 U and log 2 V \log^2 V log 2 V terms therefore signal simultaneous simple poles in the full Mellin amplitude. Motivated by the remarkably simple analytic structure found for one-loop reduced Mellin amplitudes of AdS supergluons (see, for example, [44 , 46 ] ), and the recent Mellin reduced formalism proposed in [47 ] , we ask whether the present full-correlator data admit an analogous reduced representation. We therefore make the minimal reduced ansatz19
B ( s , t ) = ∑ m , n ≥ 1 c m n ( s − 2 m ) ( t − 2 n ) . (103) B(s,t)=\sum_{m,n\geq1}\frac{c_{mn}}{(s-2m)(t-2n)}. \tag*{(103)} B ( s , t ) = m , n ≥ 1 ∑ ( s − 2 m ) ( t − 2 n ) c mn . (103)
The even- and odd-spin reduced amplitudes are obtained from the appropriate crossed sum and difference of this kernel. They are then mapped to the full α 2 \alpha^2 α 2 component by the Mellin-space Ward/difference operator of [47 ] .
The coefficients c m n c_{mn} c mn are first fixed by requiring the Ward-lifted amplitude to reproduce the highest crossed-channel logarithms, log 2 U log 2 V \log^2 U\,\log^2 V log 2 U log 2 V , of the full correlator. Since the Ward operator shifts the Mellin variables only by finite amounts, a given c m n c_{mn} c mn contributes to a finite set of neighboring U U U -slices. The completeness of the simultaneous-pole ansatz is then tested by comparing the reconstructed amplitude with the full log 2 U \log^2 U log 2 U data, including the terms subleading in log V \log V log V .
The conceptual reconstruction can be summarized as
ω n , ℓ ⟶ 1 4 ω n , ℓ 2 A n , ℓ c h ⟶ F e v e n , F o d d ⟶ h p ± ( V ) ⟶ [ log 2 V ] h p ± ( V ) ⟶ c m n . (104) \omega_{n,\ell}\longrightarrow\frac{1}{4}\frac{\omega_{n,\ell}^{2}}{A_{n,\ell}^{\mathrm{ch}}}\longrightarrow F_{\mathrm{even}},F_{\mathrm{odd}}\longrightarrow h_p^{\pm}(V)\longrightarrow[\log^2 V]h_p^{\pm}(V)\longrightarrow c_{mn}. \tag*{(104)} ω n , ℓ ⟶ 4 1 A n , ℓ ch ω n , ℓ 2 ⟶ F even , F odd ⟶ h p ± ( V ) ⟶ [ log 2 V ] h p ± ( V ) ⟶ c mn . (104)
The first steps are statements about CFT data and the one-loop double-discontinuity, followed by the reconstruction of the non-local amplitude. We now report the explicit results.
Step 1: U U U -slices. The starting point is the small-U U U expansion (102) . We define h ^ p ± ( V ) ≡ π 4 h p ± ( V ) \widehat{h}_p^{\pm}(V)\equiv\pi^4 h_p^{\pm}(V) h p ± ( V ) ≡ π 4 h p ± ( V ) . The spin-summed slices take the finite rational–polylogarithmic form
h ^ p ± ( V ) = 1 ( 1 − V ) 2 p − 2 [ R p , 0 ± ( V ) + R p , 1 ± ( V ) log V + R p , 2 ± ( V ) log 2 V + R p , 3 ± ( V ) Li 2 ( 1 − V ) ] . (105) \widehat{h}_p^{\pm}(V)=\frac{1}{(1-V)^{2p-2}}\left[R_{p,0}^{\pm}(V)+R_{p,1}^{\pm}(V)\log V+R_{p,2}^{\pm}(V)\log^2 V+R_{p,3}^{\pm}(V)\operatorname{Li}_2(1-V)\right]. \tag*{(105)} h p ± ( V ) = ( 1 − V ) 2 p − 2 1 [ R p , 0 ± ( V ) + R p , 1 ± ( V ) log V + R p , 2 ± ( V ) log 2 V + R p , 3 ± ( V ) Li 2 ( 1 − V ) ] . (105)
where the R p , i ± ( V ) R_{p,i}^{\pm}(V) R p , i ± ( V ) are polynomials. For example, for the first two slices we find
h ^ 2 + ( V ) = log 2 V 36 ( 1 − V ) 2 , (106) \hat{h}_{2}^{+}(V)=\frac{\log^{2}V}{36(1-V)^{2}}, \tag*{(106)} h ^ 2 + ( V ) = 36 ( 1 − V ) 2 log 2 V , (106)
h ^ 2 − ( V ) = log V ( 4 + log V ) + 4 Li 2 ( 1 − V ) 36 ( 1 − V ) 2 , (107) \hat{h}_{2}^{-}(V)=\frac{\log V(4+\log V)+4\operatorname{Li}_{2}(1-V)}{36(1-V)^{2}}, \tag*{(107)} h ^ 2 − ( V ) = 36 ( 1 − V ) 2 log V ( 4 + log V ) + 4 Li 2 ( 1 − V ) , (107)
and
h ^ 3 + ( V ) = 5 ( 1 − V ) log V + ( 2 + V ) log 2 V + 2 ( 1 − V ) Li 2 ( 1 − V ) 36 ( 1 − V ) 4 , (108) \hat{h}_{3}^{+}(V)=\frac{5(1-V)\log V+(2+V)\log^{2}V+2(1-V)\operatorname{Li}_{2}(1-V)}{36(1-V)^{4}}, \tag*{(108)} h ^ 3 + ( V ) = 36 ( 1 − V ) 4 5 ( 1 − V ) log V + ( 2 + V ) log 2 V + 2 ( 1 − V ) Li 2 ( 1 − V ) , (108)
h ^ 3 − ( V ) = 10 ( 1 − V ) + 11 ( 1 + V ) log V + ( 2 + V ) log 2 V + 6 ( 1 + V ) Li 2 ( 1 − V ) 36 ( 1 − V ) 4 . (109) \hat{h}_{3}^{-}(V)=\frac{10(1-V)+11(1+V)\log V+(2+V)\log^{2}V+6(1+V)\operatorname{Li}_{2}(1-V)}{36(1-V)^{4}}. \tag*{(109)} h ^ 3 − ( V ) = 36 ( 1 − V ) 4 10 ( 1 − V ) + 11 ( 1 + V ) log V + ( 2 + V ) log 2 V + 6 ( 1 + V ) Li 2 ( 1 − V ) . (109)
Step 2: The log 2 V \log^{2}V log 2 V coefficient. Despite their different finite parts, the two parity sectors have the same highest crossed-channel logarithm. We find the general form
[ log 2 V ] h ^ p + ( V ) = [ log 2 V ] h ^ p − ( V ) ≡ L ^ p ( V ) = P p ( V ) 36 ( 1 − V ) 2 p − 2 , (110) [\log^{2}V]\hat{h}_{p}^{+}(V)=[\log^{2}V]\hat{h}_{p}^{-}(V)\equiv\hat{L}_{p}(V)=\frac{P_{p}(V)}{36(1-V)^{2p-2}}, \tag*{(110)} [ log 2 V ] h ^ p + ( V ) = [ log 2 V ] h ^ p − ( V ) ≡ L ^ p ( V ) = 36 ( 1 − V ) 2 p − 2 P p ( V ) , (110)
where
P p ( V ) = ∑ k = 0 p − 2 ( p − 1 k + 1 ) ( p − 2 k ) V k = ( p − 1 ) 2 F 1 ( 2 − p , 2 − p ; 2 ; V ) . (111) P_{p}(V)=\sum_{k=0}^{p-2}\binom{p-1}{k+1}\binom{p-2}{k}V^{k}=(p-1)\,{}_2F_{1}(2-p,2-p;2;V). \tag*{(111)} P p ( V ) = k = 0 ∑ p − 2 ( k + 1 p − 1 ) ( k p − 2 ) V k = ( p − 1 ) 2 F 1 ( 2 − p , 2 − p ; 2 ; V ) . (111)
Step 3: Triangular recursion for the Mellin coefficients. The last step is to determine the coefficients of the reduced simultaneous-pole ansatz from the highest crossed-channel logarithms. Since the log 2 U log 2 V \log^{2}U\log^{2}V log 2 U log 2 V terms probe simultaneous poles in the s s s and t t t channels, it is sufficient at this stage to consider the common s t st s t kernel B ( s , t ) B(s,t) B ( s , t ) . Its contribution to the full α 2 \alpha^{2} α 2 Mellin amplitude is obtained by the Ward lift [47 ] ,
D 2 B ( s , t ) = ( 2 − s ) 2 ( 4 − s ) B ( s − 4 , t ) + ( 2 − s ) ( 3 − s ) ( 2 − t ) B ( s − 2 , t − 2 ) + ( 3 − s ) ( s − 2 ) ( 2 − s − t ) B ( s − 2 , t ) . (112) \mathcal{D}_{2}B(s,t)=(2-s)^{2}(4-s)B(s-4,t)+(2-s)(3-s)(2-t)B(s-2,t-2)
+(3-s)(s-2)(2-s-t)B(s-2,t). \tag*{(112)} D 2 B ( s , t ) = ( 2 − s ) 2 ( 4 − s ) B ( s − 4 , t ) + ( 2 − s ) ( 3 − s ) ( 2 − t ) B ( s − 2 , t − 2 ) + ( 3 − s ) ( s − 2 ) ( 2 − s − t ) B ( s − 2 , t ) . (112)
To make the matching explicit, suppose that near a simultaneous double-trace location the full α 2 \alpha^{2} α 2 Mellin amplitude behaves as
[ α 2 ] M f u l l ( s , t ) ∼ R K , L ( s − 2 K ) ( t − 2 L ) . (113) [\alpha^{2}]\mathcal{M}^{\mathrm{full}}(s,t)\sim\frac{R_{K,L}}{(s-2K)(t-2L)}. \tag*{(113)} [ α 2 ] M full ( s , t ) ∼ ( s − 2 K ) ( t − 2 L ) R K , L . (113)
The Mellin measure (57) has double poles at the same locations, so the full integrand has triple poles in both s s s and t t t . Their inverse Mellin transform therefore produces
ρ K , L R K , L U K V L − 1 log 2 U log 2 V , ρ K , L = Γ ( K + L − 1 ) 2 16 [ ( K − 1 ) ! ] 2 [ ( L − 1 ) ! ] 2 , (114) \rho_{K,L}R_{K,L}U^{K}V^{L-1}\log^{2}U\log^{2}V,\qquad
\rho_{K,L}=\frac{\Gamma(K+L-1)^{2}}{16[(K-1)!]^{2}[(L-1)!]^{2}}, \tag*{(114)} ρ K , L R K , L U K V L − 1 log 2 U log 2 V , ρ K , L = 16 [( K − 1 )! ] 2 [( L − 1 )! ] 2 Γ ( K + L − 1 ) 2 , (114)
where ρ K , L \rho_{K,L} ρ K , L follows from the Mellin measure.
On the CFT side, the log 2 U log 2 V \log^2 U\log^2 V log 2 U log 2 V part of the full α 2 \alpha^2 α 2 correlator is encoded in the functions L ^ p ( V ) \widehat{L}_p(V) L p ( V ) (110) . Expanding
L ^ p ( V ) = ∑ r ≥ 0 L ^ p , r V r , (115) \widehat{L}_p(V)=\sum_{r\geq0}\widehat{L}_{p,r}V^r, \tag*{(115)} L p ( V ) = r ≥ 0 ∑ L p , r V r , (115)
the coefficient L ^ p , r \widehat{L}_{p,r} L p , r multiplies U p V r log 2 U log 2 V U^pV^r\log^2 U\log^2 V U p V r log 2 U log 2 V in the full correlator. The Mellin and CFT descriptions are therefore related by
L ^ K , L − 1 = ρ K , L R ^ K , L , (116) \widehat{L}_{K,L-1}=\rho_{K,L}\widehat{R}_{K,L}, \tag*{(116)} L K , L − 1 = ρ K , L R K , L , (116)
where we have rescaled the coefficients by a common factor of π 4 \pi^4 π 4 , denoting the resulting quantities by c ^ m n \widehat{c}_{mn} c mn and R ^ K , L \widehat{R}_{K,L} R K , L , respectively. Combining the full simultaneous-pole coefficient generated by (112) with (116) gives
L ^ m + 1 , n − 1 = ρ m + 1 , n [ − 8 m 2 ( m − 1 ) c ^ m − 1 , n − 4 m ( 2 m − 1 ) ( n − 1 ) c ^ m , n − 1 + 4 m ( 2 m − 1 ) ( m + n ) c ^ m , n ] , (117) \widehat{L}_{m+1,n-1}=\rho_{m+1,n}\left[-8m^2(m-1)\widehat{c}_{m-1,n}-4m(2m-1)(n-1)\widehat{c}_{m,n-1}+4m(2m-1)(m+n)\widehat{c}_{m,n}\right], \tag*{(117)} L m + 1 , n − 1 = ρ m + 1 , n [ − 8 m 2 ( m − 1 ) c m − 1 , n − 4 m ( 2 m − 1 ) ( n − 1 ) c m , n − 1 + 4 m ( 2 m − 1 ) ( m + n ) c m , n ] , (117)
with
ρ m + 1 , n = Γ ( m + n ) 2 16 ( m ! ) 2 ( ( n − 1 ) ! ) 2 , c ^ 0 , n = c ^ m , 0 = 0. (118) \rho_{m+1,n}=\frac{\Gamma(m+n)^2}{16(m!)^2((n-1)!)^2},\qquad\widehat{c}_{0,n}=\widehat{c}_{m,0}=0. \tag*{(118)} ρ m + 1 , n = 16 ( m ! ) 2 (( n − 1 )! ) 2 Γ ( m + n ) 2 , c 0 , n = c m , 0 = 0. (118)
Since the first two terms in (117) involve coefficients with lower indices, while the last contains the new coefficient c ^ m , n \widehat{c}_{m,n} c m , n , the system is triangular and determines the residue matrix recursively. The resulting coefficients admit the compact symmetric expression
c ^ m n = 1 9 [ π Γ ( m ) Γ ( n ) Γ ( m + n + 1 2 ) Γ ( m + 1 2 ) Γ ( n + 1 2 ) Γ ( m + n + 1 ) − 1 m n ] . (119) \widehat{c}_{mn}=\frac{1}{9}\left[\frac{\sqrt{\pi}\,\Gamma(m)\Gamma(n)\Gamma\left(m+n+\frac{1}{2}\right)}{\Gamma\left(m+\frac{1}{2}\right)\Gamma\left(n+\frac{1}{2}\right)\Gamma(m+n+1)}-\frac{1}{mn}\right]. \tag*{(119)} c mn = 9 1 [ Γ ( m + 2 1 ) Γ ( n + 2 1 ) Γ ( m + n + 1 ) π Γ ( m ) Γ ( n ) Γ ( m + n + 2 1 ) − mn 1 ] . (119)
The manifest symmetry c ^ m n = c ^ n m \widehat{c}_{mn}=\widehat{c}_{nm} c mn = c nm is a highly non-trivial consistency check of the ansatz. It is not imposed by the triangular recursion itself: the recursion determines entries row by row from the U U U -slice data. The fact that the resulting matrix is symmetric is therefore strong evidence that the full α 2 \alpha^2 α 2 data really come from a crossing-symmetric reduced Mellin object. Since the two parity sectors have the same highest-logarithmic data (110) , the same residue matrix c ^ m n \widehat{c}_{mn} c mn governs both.
Although the coefficients c ^ m n \widehat{c}_{mn} c mn were fixed using only the log 2 U log 2 V \log^2 U\log^2 V log 2 U log 2 V terms, after the Ward lift and inverse Mellin transform the reconstructed amplitudes reproduce the complete functions F e v e n ( U , V ) F_{\mathrm{even}}(U,V) F even ( U , V ) and F o d d ( U , V ) F_{\mathrm{odd}}(U,V) F odd ( U , V ) , including the single-logarithmic and regular terms in V V V . We have checked the reconstruction in exact arithmetic for both parity sectors: for p = 2 , … , 10 p=2,\ldots,10 p = 2 , … , 10 , the inverse Mellin transform of the Ward-lifted kernels D 2 [ B ( s , t ) ± B ( s , u ~ ) ] \mathcal{D}_2[B(s,t) \pm B(s,\widetilde{u})] D 2 [ B ( s , t ) ± B ( s , u )] , with u ~ = − s − t \widetilde{u}=-s-t u = − s − t , agrees with the independently computed CFT block sum through order ( 1 − V ) 12 (1-V)^{12} ( 1 − V ) 12 , including the full slice rather than only its log 2 V \log^2 V log 2 V coefficient. The closed position-space expressions for p = 2 , … , 7 p=2,\ldots,7 p = 2 , … , 7 also agree identically with the inverse Mellin result. These checks support the conclusion that no additional reduced single-pole contribution is required.
Flavor projection and complete one-loop amplitude
We now restore the flavor structure and crossing symmetry and assemble the complete reduced one-loop amplitude. The Mellin variables of the full and reduced correlators obey different constraints. For the full correlator, s + t + u = 4 s+t+u=4 s + t + u = 4 , while the modified Mellin transform defining the reduced amplitude is naturally parametrized by
u ~ ≡ u − 4 , s + t + u ~ = 0 , (120) \widetilde{u} \equiv u-4,\qquad s+t+\widetilde{u}=0, \tag*{(120)} u ≡ u − 4 , s + t + u = 0 , (120)
in terms of which the crossing properties of the reduced amplitude take their simplest form [47 ] . Using the common kernel B ( s , t ) B(s,t) B ( s , t ) reconstructed above, we introduce the crossed parity combinations entering the s s s -channel cut,
f e v e n ≡ B ( s , t ) + B ( s , u ~ ) , f o d d ≡ B ( s , t ) − B ( s , u ~ ) . (121) f_{\mathrm{even}} \equiv B(s,t)+B(s,\widetilde{u}),\qquad f_{\mathrm{odd}} \equiv B(s,t)-B(s,\widetilde{u}). \tag*{(121)} f even ≡ B ( s , t ) + B ( s , u ) , f odd ≡ B ( s , t ) − B ( s , u ) . (121)
Recall from (92) that the tree-level flavor operator C ℓ \mathcal{C}_{\ell} C ℓ takes different forms for even and odd spins. The one-loop unitarity relation involves the squares of the corresponding flavor eigenvalues. Combining these with the even- and odd-spin reduced kernels, and using the Jacobi identity, the s s s -channel cut can be reorganized in terms of color-box tensors:
C e v e n 2 f e v e n − C o d d 2 f o d d = 9 [ d s t B ( s , t ) + d s u B ( s , u ~ ) ] , (122) \mathcal{C}_{\mathrm{even}}^{2}f_{\mathrm{even}}-\mathcal{C}_{\mathrm{odd}}^{2}f_{\mathrm{odd}}=9\left[d_{st}B(s,t)+d_{su}B(s,\widetilde{u})\right], \tag*{(122)} C even 2 f even − C odd 2 f odd = 9 [ d s t B ( s , t ) + d s u B ( s , u ) ] , (122)
where
d s t I 1 I 2 I 3 I 4 = f J I 1 K f K I 2 L f L I 3 M f M I 4 J , (123) d_{st}^{I_{1}I_{2}I_{3}I_{4}}=f^{JI_{1}K}f^{KI_{2}L}f^{LI_{3}M}f^{MI_{4}J}, \tag*{(123)} d s t I 1 I 2 I 3 I 4 = f J I 1 K f K I 2 L f L I 3 M f M I 4 J , (123)
along with the crossed expressions. The complete non-local reduced one-loop amplitude is
M ~ c u t 1 -loop ( s , t ) = 9 [ d s t B ( s , t ) + d s u B ( s , u ~ ) + d t u B ( t , u ~ ) ] , (124) \widetilde{\mathcal{M}}_{\mathrm{cut}}^{1\text{-loop}}(s,t)=9\left[d_{st}B(s,t)+d_{su}B(s,\widetilde{u})+d_{tu}B(t,\widetilde{u})\right], \tag*{(124)} M cut 1 -loop ( s , t ) = 9 [ d s t B ( s , t ) + d s u B ( s , u ) + d t u B ( t , u ) ] , (124)
where the kernel B B B is given by (103) , with coefficients (119) . This is the same color-box organization found in holographic studies.
As a final remark, notice that the double-discontinuity determines the non-local part of the reduced amplitude but is insensitive to regular Mellin polynomials. In appendix C we argue that such polynomial local completions are excluded: requiring that the large-radius limit of the correlator reproduces the flat-space one-loop amplitude, any polynomial term would produce a power-law growth in L L L that the amplitude does not have. Protected OPE data provide an independent check at finite AdS radius. We therefore take
M ~ t o t 1 -loop = M ~ c u t 1 -loop . (125) \widetilde{\mathcal{M}}_{\mathrm{tot}}^{1\text{-loop}}=\widetilde{\mathcal{M}}_{\mathrm{cut}}^{1\text{-loop}}. \tag*{(125)} M tot 1 -loop = M cut 1 -loop . (125)
The full R-symmetry-dependent amplitude is obtained by the Ward lift,
M f u l l 1 -loop ( s , t ; α ) = ∑ J = 0 2 α J D J M ~ t o t 1 -loop ( s , t ) , (126) \mathcal{M}_{\mathrm{full}}^{1\text{-loop}}(s,t;\alpha)=\sum_{J=0}^{2}\alpha^J\mathcal{D}_J\widetilde{\mathcal{M}}_{\mathrm{tot}}^{1\text{-loop}}(s,t), \tag*{(126)} M full 1 -loop ( s , t ; α ) = J = 0 ∑ 2 α J D J M tot 1 -loop ( s , t ) , (126)
where the finite-difference operators D J \mathcal{D}_J D J implement the superconformal Ward identities. The highest R-symmetry operator D 2 \mathcal{D}_2 D 2 was given in (112) , while the remaining two read [47 eq. (B.11)]
D 0 = ( s + t ) ( s + t − 2 ) 2 P 0 , 0 + ( 2 − s ) ( s + t − 1 ) ( s + t − 2 ) P − 2 , 0 + ( 2 − t ) ( s + t − 1 ) ( s + t − 2 ) P 0 , − 2 , (127) \mathcal{D}_0=(s+t)(s+t-2)^2P_{0,0}+(2-s)(s+t-1)(s+t-2)P_{-2,0}
+(2-t)(s+t-1)(s+t-2)P_{0,-2}, \tag*{(127)} D 0 = ( s + t ) ( s + t − 2 ) 2 P 0 , 0 + ( 2 − s ) ( s + t − 1 ) ( s + t − 2 ) P − 2 , 0 + ( 2 − t ) ( s + t − 1 ) ( s + t − 2 ) P 0 , − 2 , (127)
D 1 = − ( s + t ) ( s + t − 2 ) 2 P 0 , 0 + ( t − 2 ) ( s + 2 t − 4 ) ( s + t − 2 ) P 0 , − 2 + ( s − 2 ) ( t + 2 ) ( s + t − 2 ) P − 2 , 0 + ( s − 2 ) ( t − 2 ) ( s − t ) P − 2 , − 2 + ( s − 2 ) 2 ( s − 4 ) P − 4 , 0 + ( 4 − t ) ( t − 2 ) 2 P 0 , − 4 , (128) \begin{aligned}
\mathcal{D}_1={}&-(s+t)(s+t-2)^2P_{0,0}+(t-2)(s+2t-4)(s+t-2)P_{0,-2}\\
&+(s-2)(t+2)(s+t-2)P_{-2,0}+(s-2)(t-2)(s-t)P_{-2,-2}\\
&+(s-2)^2(s-4)P_{-4,0}+(4-t)(t-2)^2P_{0,-4}, \tag*{(128)}
\end{aligned} D 1 = − ( s + t ) ( s + t − 2 ) 2 P 0 , 0 + ( t − 2 ) ( s + 2 t − 4 ) ( s + t − 2 ) P 0 , − 2 + ( s − 2 ) ( t + 2 ) ( s + t − 2 ) P − 2 , 0 + ( s − 2 ) ( t − 2 ) ( s − t ) P − 2 , − 2 + ( s − 2 ) 2 ( s − 4 ) P − 4 , 0 + ( 4 − t ) ( t − 2 ) 2 P 0 , − 4 , (128)
where P a , b P_{a,b} P a , b denotes the shift operator, P a , b f ( s , t ) ≡ f ( s + a , t + b ) P_{a,b}f(s,t)\equiv f(s+a,t+b) P a , b f ( s , t ) ≡ f ( s + a , t + b ) .
Notice that the simplicity of the reduced amplitude does not imply the absence of single poles in the full Mellin amplitude. The finite-difference Ward lift can generate single-pole terms and Mellin-variable-dependent residues from the simultaneous-pole kernel.
The flat-space limit
We now study the large-radius limit of the reconstructed one-loop amplitude. The external moment-map operators select bulk scalar states belonging to the same four-dimensional N = 4 \mathcal{N}=4 N = 4 SYM multiplet as the gluons. Their effective AdS mass satisfies m 2 = − 2 / L 2 m^2=-2/L^2 m 2 = − 2/ L 2 , so these states become massless in the flat-space limit. We denote the Mellin variables by s , t , u s,t,u s , t , u , while S , T , U S,T,U S , T , U are the four-dimensional Mandelstam invariants,
S = − ( p 1 + p 2 ) 2 , T = − ( p 1 + p 4 ) 2 , U = − ( p 1 + p 3 ) 2 , S + T + U = 0. (129) S=-(p_1+p_2)^2,\qquad T=-(p_1+p_4)^2,\qquad U=-(p_1+p_3)^2,\qquad S+T+U=0. \tag*{(129)} S = − ( p 1 + p 2 ) 2 , T = − ( p 1 + p 4 ) 2 , U = − ( p 1 + p 3 ) 2 , S + T + U = 0. (129)
In this section, U U U always denotes the Mandelstam invariant, not the conformal cross-ratio of the previous sections, which does not appear here. All external momenta are incoming and obey p i 2 = 0 p_i^2=0 p i 2 = 0 . We use a mostly-plus Lorentzian signature. The prescription originally proposed in [24 ] extracts a local flat-space scattering amplitude from the large-radius limit of the Mellin representation of an AdS boundary correlator,
A 4 ( S , T ; α ) ≡ 8 π 2 lim L → ∞ I β [ M ( L 2 S 2 β , L 2 T 2 β ; α ) ] (130) \mathcal{A}_4(S,T;\alpha) \equiv8\pi^2 \lim_{L\to\infty} \mathcal{I}_{\beta}\left[\mathcal{M}\left(\frac{L^2S}{2\beta},\frac{L^2T}{2\beta};\alpha\right)\right] \tag*{(130)} A 4 ( S , T ; α ) ≡ 8 π 2 L → ∞ lim I β [ M ( 2 β L 2 S , 2 β L 2 T ; α ) ] (130)
where we have stripped the momentum-conserving delta function and the external polarization factor R s R_s R s , and
I β [ F ] ≡ π ∫ C d β 2 π i e β β − 1 / 2 F ( β ) . (131) \mathcal{I}_{\beta}[F] \equiv\sqrt{\pi}\int_C\frac{d\beta}{2\pi i}e^\beta\beta^{-1/2}F(\beta). \tag*{(131)} I β [ F ] ≡ π ∫ C 2 π i d β e β β − 1/2 F ( β ) . (131)
Following [25 ] , the contour C C C is oriented upward and chosen to lie to the right of the poles inherited from the Mellin amplitude and of the branch cut generated by β − 1 / 2 \beta^{-1/2} β − 1/2 . We use the principal branch, − π < arg β < π -\pi< \arg\beta< \pi − π < arg β < π , with the cut along the negative real axis, and the same branch for log β \log\beta log β below. Introducing
λ ≡ L 2 2 β , (132) \lambda\equiv\frac{L^2}{2\beta}, \tag*{(132)} λ ≡ 2 β L 2 , (132)
for fixed flat-space kinematics, the Mellin amplitude is evaluated at
s = λ S , t = λ T , u = 4 + λ U . (133) s = \lambda S,\qquad t = \lambda T,\qquad u = 4+\lambda U. \tag*{(133)} s = λ S , t = λ T , u = 4 + λ U . (133)
As the radius of AdS goes to infinity, λ → ∞ \lambda\to\infty λ → ∞ and the transform probes the simultaneous large-Mellin regime at fixed ratios s / t = S / T s/t = S/T s / t = S / T . However, in the present massless one-loop problem, the strict limit in (130) is infrared divergent. We therefore define the auxiliary large-radius asymptotic transform
A 4 , L ( S , T ; α ) ≡ 8 π 2 I β [ M ( L 2 S 2 β , L 2 T 2 β ; α ) ] a s y m p . (134) \mathcal{A}_{4,L}(S,T;\alpha) \equiv8\pi^2\mathcal{I}_{\beta}\left[\mathcal{M}\left(\frac{L^2S}{2\beta},\frac{L^2T}{2\beta};\alpha\right)\right]_{\mathrm{asymp}}. \tag*{(134)} A 4 , L ( S , T ; α ) ≡ 8 π 2 I β [ M ( 2 β L 2 S , 2 β L 2 T ; α ) ] asymp . (134)
We first take the large-L L L asymptotic expansion of the integrand at fixed β \beta β and fixed S , T , U S,T,U S , T , U , and then compute the inverse Laplace moments. At large but finite L L L , the result is finite, and in this sense the AdS radius regulates the infrared. We stress, however, that L L L is not a parameter of the boundary correlator: the theory is conformal, and the Mellin amplitude M ( s , t ; α ) \mathcal{M}(s,t;\alpha) M ( s , t ; α ) does not depend on L L L . The radius enters only through the scaling (133) , which converts Mellin variables into flat-space momenta, so that the logarithms of L L L found below are logarithms of large Mellin variables. The order in which the Mellin sums, the Ward lift, and this asymptotic expansion must be performed is important and will be specified below.
We first derive the large-Mellin asymptotics in a region that does not cross the Mellin poles, and then analytically continue the result to physical Lorentzian kinematics.
Reduced versus full Mellin amplitude. Recall the one-loop reduced kernel
B ( s , t ) = ∑ m , n ≥ 1 c m n ( s − 2 m ) ( t − 2 n ) , c m n = 1 9 π 4 ( a m n − 1 m n ) , (135) B(s,t)=\sum_{m,n\ge1}\frac{c_{mn}}{(s-2m)(t-2n)},\qquad c_{mn}=\frac{1}{9\pi^{4}}\left(a_{mn}-\frac{1}{mn}\right), \tag*{(135)} B ( s , t ) = m , n ≥ 1 ∑ ( s − 2 m ) ( t − 2 n ) c mn , c mn = 9 π 4 1 ( a mn − mn 1 ) , (135)
where we have defined
a m n ≡ π Γ ( m ) Γ ( n ) Γ ( m + n + 1 2 ) Γ ( m + 1 2 ) Γ ( n + 1 2 ) Γ ( m + n + 1 ) . (136) a_{mn}\equiv\frac{\sqrt{\pi}\,\Gamma(m)\Gamma(n)\Gamma\left(m+n+\frac{1}{2}\right)}{\Gamma\left(m+\frac{1}{2}\right)\Gamma\left(n+\frac{1}{2}\right)\Gamma(m+n+1)}. \tag*{(136)} a mn ≡ Γ ( m + 2 1 ) Γ ( n + 2 1 ) Γ ( m + n + 1 ) π Γ ( m ) Γ ( n ) Γ ( m + n + 2 1 ) . (136)
The discrete AdS poles have a direct flat-space interpretation. At fixed β \beta β , the scaling s = λ S s=\lambda S s = λ S maps the pole s = 2 m s=2m s = 2 m to S m = 2 m / λ S_m=2m/\lambda S m = 2 m / λ . The spacing between neighboring poles is therefore
S m + 1 − S m = 2 λ → λ → ∞ 0 , (137) S_{m+1}-S_m=\frac{2}{\lambda}\xrightarrow{\lambda\to\infty}0, \tag*{(137)} S m + 1 − S m = λ 2 λ → ∞ 0 , (137)
and similarly in the other channels. As the curvature is removed, the discrete AdS poles become dense in S S S . Conversely, probing a fixed nonzero value of S S S as λ → ∞ \lambda\to\infty λ → ∞ requires m ∼ λ S / 2 m\sim\lambda S/2 m ∼ λ S /2 , so that the relevant pole levels become parametrically large. For simultaneous poles, the corresponding scaling region has m , n = O ( λ ) m,n=O(\lambda) m , n = O ( λ ) . This motivates a continuum approximation to the large-level pole sums, whose resummation is expected to reproduce the logarithmic non-analyticities and branch cuts characteristic of flat-space loop amplitudes.
There is, however, an important subtlety: the leading continuum approximation to the reduced amplitude does not necessarily capture the full flat-space limit. The outcome also depends on the pole residues and, crucially, on the action of the superconformal Ward identities. The flat-space scaling (133) is formulated in terms of the full Mellin variables, whereas the reduced amplitude introduced in the preceding section uses a shifted third variable, u ~ = u − 4 \widetilde{u}=u-4 u = u − 4 . Although this constant shift is subleading at large Mellin variables, the relation between the reduced and full amplitudes involves a non-trivial finite-difference Ward operator, whose action must be taken into account.
To see why this matters, consider the asymptotic behavior of the reduced coefficients. At simultaneously large m , n m,n m , n with fixed ratio, we find
a m n ∼ π m n ( m + n ) . (138) a_{mn}\sim\frac{\sqrt{\pi}}{\sqrt{mn(m+n)}}. \tag*{(138)} a mn ∼ mn ( m + n ) π . (138)
Replacing the double sum by its continuum approximation then gives a leading contribution to the reduced Mellin amplitude that is homogeneous of degree − 3 / 2 -3/2 − 3/2 , as in the five-dimensional massless box of [44 ] .
In many familiar holographic examples, the superconformal Ward difference operator reduces, in the flat-space limit, to multiplication by a universal kinematic polynomial [45 ] . Here, instead, its leading large-Mellin action retains a differential operator. Defining $\mathcal{E}\equiv s ∂ s + t ∂ t s\partial_s+t\partial_t s ∂ s + t ∂ t , we find
D 0 + α D 1 + α 2 D 2 ∼ ( u ~ + α s ) 2 ( 3 + 2 E ) . (139) \mathcal{D}_0+\alpha\mathcal{D}_1+\alpha^2\mathcal{D}_2\sim(\widetilde{u}+\alpha s)^2(3+2\mathcal{E}). \tag*{(139)} D 0 + α D 1 + α 2 D 2 ∼ ( u + α s ) 2 ( 3 + 2 E ) . (139)
In particular, for a homogeneous function f − 3 / 2 f_{-3/2} f − 3/2 ,
( 3 + 2 E ) f − 3 / 2 = 0. (140) (3+2\mathcal{E})f_{-3/2}=0. \tag*{(140)} ( 3 + 2 E ) f − 3/2 = 0. (140)
Thus, the leading continuum contribution to the reduced amplitude is annihilated by the leading Ward operator in every R-symmetry component. The naive leading large-Mellin behavior of the reduced amplitude is therefore insufficient to determine the physical four-dimensional result. Subleading contributions, including those associated with the 1 / ( m n ) 1/(mn) 1/ ( mn ) term and the endpoints of the pole sums, cannot be discarded a priori. To retain their combined effect, we must first apply the exact Ward lift and resum the resulting full Mellin amplitude, and only then take the large-Mellin limit.
Exact resummation of the full amplitude. Let us define the α 2 \alpha^2 α 2 -component of the elementary Ward-lifted s t st s t block
V ^ s t ( s , t ) ≡ π 4 D 2 B ( s , t ) . (141) \widehat{\mathcal{V}}_{st}(s,t)\equiv\pi^4\mathcal{D}_2B(s,t). \tag*{(141)} V s t ( s , t ) ≡ π 4 D 2 B ( s , t ) . (141)
We find the exact resummation
V ^ s t ( s , t ) = ( s − 2 ) [ 1 9 H ( 1 − s 2 ) H ( − t / 2 ) t + τ ( t ) ] . (142) \widehat{\mathcal{V}}_{st}(s,t)=(s-2)\left[\frac{1}{9}H\left(1-\frac{s}{2}\right)\frac{H(-t/2)}{t}+\tau(t)\right]. \tag*{(142)} V s t ( s , t ) = ( s − 2 ) [ 9 1 H ( 1 − 2 s ) t H ( − t /2 ) + τ ( t ) ] . (142)
where H ( z ) = ψ ( z + 1 ) + γ E H(z)=\psi(z+1)+\gamma_E H ( z ) = ψ ( z + 1 ) + γ E is the analytic continuation of the harmonic numbers, ψ ( z ) = Γ ′ ( z ) / Γ ( z ) \psi(z)=\Gamma'(z)/\Gamma(z) ψ ( z ) = Γ ′ ( z ) /Γ ( z ) is the digamma function, and γ E \gamma_E γ E is the Euler–Mascheroni constant. The one-variable completion τ ( t ) \tau(t) τ ( t ) admits the absolutely convergent pole representation
τ ( t ) = 1 9 ∑ n = 1 ∞ b n t − 2 n , b n = 1 n 2 ( 4 n ( 2 n n ) − 1 ) . (143) \tau(t)=\frac{1}{9}\sum_{n=1}^{\infty}\frac{b_n}{t-2n},\qquad b_n=\frac{1}{n^2}\left(\frac{4^n}{\binom{2n}{n}}-1\right). \tag*{(143)} τ ( t ) = 9 1 n = 1 ∑ ∞ t − 2 n b n , b n = n 2 1 ( ( n 2 n ) 4 n − 1 ) . (143)
Equivalently, it can be written in closed hypergeometric form as
τ ( t ) = − 2 9 t 2 H ( − t 2 ) + π 2 27 t + 2 π 3 / 2 9 t 2 Γ ( 1 2 − t 2 ) Γ ( − t 2 ) − 4 9 ( t − 1 ) t 3 F 2 ( 1 2 , 1 2 , 1 3 2 , 3 2 − t 2 ; 1 ) . (144) \tau(t)=-\frac{2}{9t^2}H\left(-\frac{t}{2}\right)+\frac{\pi^2}{27t}+\frac{2\pi^{3/2}}{9t^2}\frac{\Gamma\left(\frac{1}{2}-\frac{t}{2}\right)}{\Gamma\left(-\frac{t}{2}\right)}-\frac{4}{9(t-1)t}\,{}_3F_2\left(\begin{matrix}\frac{1}{2},\frac{1}{2},1\\\frac{3}{2},\frac{3}{2}-\frac{t}{2}\end{matrix};1\right). \tag*{(144)} τ ( t ) = − 9 t 2 2 H ( − 2 t ) + 27 t π 2 + 9 t 2 2 π 3/2 Γ ( − 2 t ) Γ ( 2 1 − 2 t ) − 9 ( t − 1 ) t 4 3 F 2 ( 2 1 , 2 1 , 1 2 3 , 2 3 − 2 t ; 1 ) . (144)
The product of harmonic functions in (142) carries the simultaneous poles, while τ \tau τ contains the simple-pole terms generated by the same Ward lift. It is therefore fixed by the reduced kernel and is not an independent addition. The pole representation implies t τ ( t ) → π 2 / 27 t\,\tau(t)\to\pi^2/27 t τ ( t ) → π 2 /27 for large negative t t t . Together with the large-argument expansion of the harmonic functions, this gives the flat-space scaling
W ^ s t ( λ S , λ T ) = 1 9 S T [ L S L T + π 2 3 ] + o ( 1 ) , (145) \widehat{\mathcal{W}}_{st}(\lambda S,\lambda T)=\frac{1}{9}\frac{S}{T}\left[\mathcal{L}_{S}\mathcal{L}_{T}+\frac{\pi^{2}}{3}\right]+o(1), \tag*{(145)} W s t ( λ S , λ T ) = 9 1 T S [ L S L T + 3 π 2 ] + o ( 1 ) , (145)
where we have defined
L X ( β , L ) ≡ log ( L 2 ( − X − i 0 ) 4 ) − log β + γ E , X ∈ { S , T , U } . (146) \mathcal{L}_{X}(\beta,L)\equiv\log\left(\frac{L^{2}(-X-i0)}{4}\right)-\log\beta+\gamma_{E},\qquad X\in\{S,T,U\}. \tag*{(146)} L X ( β , L ) ≡ log ( 4 L 2 ( − X − i 0 ) ) − log β + γ E , X ∈ { S , T , U } . (146)
The other two R-symmetry components do not require a separate resummation. Writing W ^ J ≡ π 4 D J B \widehat{\mathcal{W}}_{J}\equiv\pi^{4}D_{J}B W J ≡ π 4 D J B , so that W ^ 2 = W ^ s t \widehat{\mathcal{W}}_{2}=\widehat{\mathcal{W}}_{st} W 2 = W s t , the symmetry B ( s , t ) = B ( t , s ) B(s,t)=B(t,s) B ( s , t ) = B ( t , s ) and the explicit form of the Ward operators imply the exact identities
W ^ 0 ( s , t ) = ( s + t − 2 ) [ W ^ 2 ( s + 2 , t ) s + W ^ 2 ( t + 2 , s ) t ] , (147) \widehat{\mathcal{W}}_{0}(s,t)=(s+t-2)\left[\frac{\widehat{\mathcal{W}}_{2}(s+2,t)}{s}+\frac{\widehat{\mathcal{W}}_{2}(t+2,s)}{t}\right], \tag*{(147)} W 0 ( s , t ) = ( s + t − 2 ) [ s W 2 ( s + 2 , t ) + t W 2 ( t + 2 , s ) ] , (147)
W ^ 1 ( s , t ) = W ^ 2 ( t , s ) − W ^ 2 ( s , t ) − W ^ 0 ( s , t ) . (148) \widehat{\mathcal{W}}_{1}(s,t)=\widehat{\mathcal{W}}_{2}(t,s)-\widehat{\mathcal{W}}_{2}(s,t)-\widehat{\mathcal{W}}_{0}(s,t). \tag*{(148)} W 1 ( s , t ) = W 2 ( t , s ) − W 2 ( s , t ) − W 0 ( s , t ) . (148)
Together with (142) , they determine the other two components exactly, including their single-pole completions. In the flat-space scaling, (145) then gives
π 4 [ ( D 0 + α D 1 + α 2 D 2 ) B ( s , t ) ] ∣ s = λ S , t = λ T = K ( S , T ; α ) 9 S T [ L S L T + π 2 3 ] + o ( 1 ) , (149) \left.\pi^{4}\left[(D_{0}+\alpha D_{1}+\alpha^{2}D_{2})B(s,t)\right]\right|_{s=\lambda S,\,t=\lambda T}=\frac{K(S,T;\alpha)}{9ST}\left[\mathcal{L}_{S}\mathcal{L}_{T}+\frac{\pi^{2}}{3}\right]+o(1), \tag*{(149)} π 4 [ ( D 0 + α D 1 + α 2 D 2 ) B ( s , t ) ] s = λ S , t = λ T = 9 S T K ( S , T ; α ) [ L S L T + 3 π 2 ] + o ( 1 ) , (149)
where K ( S , T ; α ) = ( U + α S ) 2 K(S,T;\alpha)=(U+\alpha S)^{2} K ( S , T ; α ) = ( U + α S ) 2 . The crossed blocks have this same overall scalar kinematic factor, with the pair S T ST S T in the denominator and logarithms replaced by S U SU S U and T U TU T U , respectively. The equation above makes the four-dimensional box structure manifest. Although the α 2 \alpha^{2} α 2 -component in (145) scales as S / T S/T S / T , the Ward lift supplies the scalar kinematic factor, with [ α 2 ] K = S 2 [\alpha^{2}]K=S^{2} [ α 2 ] K = S 2 . After stripping this factor, the result is proportional to 1 / ( S T ) 1/(ST) 1/ ( S T ) , as required for the four-dimensional massless scalar box.
The AdS radius as an infrared regulator
Applying the flat-space transform to the resummed Mellin block (145) fixes the dimensionless AdS-regulated box to
I X Y , L A d S = Λ X ( L ) Λ Y ( L ) − π 2 6 ( X Y ) = ( S T ) , ( S U ) , ( T U ) , (150) \mathcal{I}^{\mathrm{AdS}}_{XY,L}=\Lambda_{X}(L)\Lambda_{Y}(L)-\frac{\pi^{2}}{6}\qquad(XY)=(ST),(SU),(TU), \tag*{(150)} I X Y , L AdS = Λ X ( L ) Λ Y ( L ) − 6 π 2 ( X Y ) = ( S T ) , ( S U ) , ( T U ) , (150)
where we have defined
Λ X ( L ) ≡ log ( L 2 ( − X − i 0 ) ) + 2 γ E . (151) \Lambda_{X}(L)\equiv\log\left(L^{2}(-X-i0)\right)+2\gamma_{E}. \tag*{(151)} Λ X ( L ) ≡ log ( L 2 ( − X − i 0 ) ) + 2 γ E . (151)
At fixed kinematics, the box diverges logarithmically as L → ∞ L\to\infty L → ∞ : the radius L L L plays the role of an infrared regulator. The constants that accompany the logarithms, namely the 2 γ E 2\gamma_{E} 2 γ E in (151) and the − π 2 / 6 -\pi^{2}/6 − π 2 /6 in (150) , are tied to the prescription (130) –(131) that converts Mellin variables into flat-space momenta. They should be regarded as part of the definition of the regulator, in the same way as the scheme-dependent constants of dimensional regularization. The regulator-independent information is the structure of the divergent terms and the kinematic dependence of the finite remainder. Restoring the flavor dressing and coupling normalization gives the one-loop regularized amplitude
A 4 , L 1 -loop = 8 ( C 222 ) 4 π 2 K ( S , T ; α ) ∑ ( X Y ) d X Y X Y I X Y , L A d S + o ( 1 ) . (152) \mathcal{A}^{1\text{-loop}}_{4,L}=\frac{8(C_{222})^4}{\pi^2}K(S,T;\alpha)\sum_{(XY)}\frac{d_{XY}}{XY}\mathcal{I}^{\mathrm{AdS}}_{XY,L}+o(1). \tag*{(152)} A 4 , L 1 -loop = π 2 8 ( C 222 ) 4 K ( S , T ; α ) ( X Y ) ∑ X Y d X Y I X Y , L AdS + o ( 1 ) . (152)
Here d S T ≡ d s t d_{ST}\equiv d_{st} d S T ≡ d s t , defined in (123) , and similarly in the crossed channels. This has the structure of the one-loop four-point amplitude of N = 4 \mathcal{N}=4 N = 4 SYM in flat space [85 ] : a universal kinematic factor multiplying a sum of scalar box integrals, one for each pair of channels, dressed by the color factors d X Y d_{XY} d X Y .
AdS minimal-subtraction scheme. Now, to isolate the dependence on the geometric infrared regulator from the kinematic logarithms, let us introduce an arbitrary reference scale μ \mu μ .20 We define
η ≡ log ( L 2 μ 2 ) , ℓ X ≡ log ( − X − i 0 μ 2 ) . (153) \eta\equiv\log(L^2\mu^2),\qquad\ell_X\equiv\log\left(\frac{-X-i0}{\mu^2}\right). \tag*{(153)} η ≡ log ( L 2 μ 2 ) , ℓ X ≡ log ( μ 2 − X − i 0 ) . (153)
The quantity η \eta η contains all the dependence on the AdS radius. In particular, η → + ∞ \eta\to+\infty η → + ∞ when L → ∞ L\to\infty L → ∞ , namely when the AdS infrared regulator is removed. On the other hand, ℓ X \ell_X ℓ X contains only the dependence on the flat-space kinematics. Since Λ X ( L ) = η + ℓ X + 2 γ E \Lambda_X(L)=\eta+\ell_X+2\gamma_E Λ X ( L ) = η + ℓ X + 2 γ E , equation (150) becomes
I X Y , L A d S = η 2 + η ( ℓ X + ℓ Y + 4 γ E ) + ( ℓ X + 2 γ E ) ( ℓ Y + 2 γ E ) − π 2 6 . (154) \mathcal{I}^{\mathrm{AdS}}_{XY,L}=\eta^2+\eta(\ell_X+\ell_Y+4\gamma_E)+(\ell_X+2\gamma_E)(\ell_Y+2\gamma_E)-\frac{\pi^2}{6}. \tag*{(154)} I X Y , L AdS = η 2 + η ( ℓ X + ℓ Y + 4 γ E ) + ( ℓ X + 2 γ E ) ( ℓ Y + 2 γ E ) − 6 π 2 . (154)
We define a natural geometric minimal-subtraction scheme by removing the terms that diverge with η \eta η ,
S X Y A d S = η 2 + η ( ℓ X + ℓ Y + 4 γ E ) , (155) \mathcal{S}^{\mathrm{AdS}}_{XY}=\eta^2+\eta(\ell_X+\ell_Y+4\gamma_E), \tag*{(155)} S X Y AdS = η 2 + η ( ℓ X + ℓ Y + 4 γ E ) , (155)
which leaves the finite remainder
H X Y A d S ( μ ) = ( ℓ X + 2 γ E ) ( ℓ Y + 2 γ E ) − π 2 6 . (156) \mathcal{H}^{\mathrm{AdS}}_{XY}(\mu)=(\ell_X+2\gamma_E)(\ell_Y+2\gamma_E)-\frac{\pi^2}{6}. \tag*{(156)} H X Y AdS ( μ ) = ( ℓ X + 2 γ E ) ( ℓ Y + 2 γ E ) − 6 π 2 . (156)
Universal infrared structure. The divergent terms (155) have the color structure required by infrared factorization. The color-box tensors obey
d s t − d t u = C A 2 c t , d s u − d t u = − C A 2 c u , (157) d_{st}-d_{tu}=\frac{C_A}{2}c_t,\qquad d_{su}-d_{tu}=-\frac{C_A}{2}c_u, \tag*{(157)} d s t − d t u = 2 C A c t , d s u − d t u = − 2 C A c u , (157)
where we recall that C A C_A C A is the quadratic Casimir of the adjoint representation. Moreover, the flat-space limit of the tree-level amplitude (58) is
A 4 t r e e = − 8 ( C 222 ) 2 K ( S , T ; α ) ( c t S T − c u S U ) . (158) \mathcal{A}^{\mathrm{tree}}_4=-8(C_{222})^2K(S,T;\alpha)\left(\frac{c_t}{ST}-\frac{c_u}{SU}\right). \tag*{(158)} A 4 tree = − 8 ( C 222 ) 2 K ( S , T ; α ) ( S T c t − S U c u ) . (158)
Using S + T + U = 0 S+T+U=0 S + T + U = 0 , the divergent part of (152) can then be written as
A 4 , L 1 − l o o p ∣ d i v = ( C 222 ) 2 π 2 ∑ j = 2 4 T 1 ⋅ T j [ 1 2 η 2 + η ( ℓ 1 j + 2 γ E ) ] A 4 t r e e , (159) \left.\mathcal{A}^{1-\mathrm{loop}}_{4,L}\right|_{\mathrm{div}}=\frac{(C_{222})^2}{\pi^2}\sum_{j=2}^{4}\mathbf{T}_1\cdot\mathbf{T}_j\left[\frac{1}{2}\eta^2+\eta\left(\ell_{1j}+2\gamma_E\right)\right]\mathcal{A}^{\mathrm{tree}}_4, \tag*{(159)} A 4 , L 1 − loop div = π 2 ( C 222 ) 2 j = 2 ∑ 4 T 1 ⋅ T j [ 2 1 η 2 + η ( ℓ 1 j + 2 γ E ) ] A 4 tree , (159)
where ℓ 12 ≡ ℓ S \ell_{12}\equiv\ell_S ℓ 12 ≡ ℓ S , ℓ 14 ≡ ℓ T \ell_{14}\equiv\ell_T ℓ 14 ≡ ℓ T , ℓ 13 ≡ ℓ U \ell_{13}\equiv\ell_U ℓ 13 ≡ ℓ U , and T i ⋅ T j ≡ ∑ J T i J T j J \mathbf{T}_i\cdot\mathbf{T}_j\equiv\sum_J T_i^J T_j^J T i ⋅ T j ≡ ∑ J T i J T j J , with T i J T_i^J T i J the adjoint generator ( T J ) I K = − i f J I K (T^J)_{IK}=-if^{JIK} ( T J ) I K = − i f J I K acting on the color index of the i i i -th particle. Color conservation gives
∑ j = 2 4 T 1 ⋅ T j = − C A , (160) \sum_{j=2}^{4}\mathbf{T}_1\cdot\mathbf{T}_j=-C_A, \tag*{(160)} j = 2 ∑ 4 T 1 ⋅ T j = − C A , (160)
so that the double logarithm is proportional to the tree-level amplitude,
A 4 , L 1 − l o o p ∣ η 2 = − ( C 222 ) 2 C A 2 π 2 η 2 A 4 t r e e , (161) \left.\mathcal{A}^{1-\mathrm{loop}}_{4,L}\right|_{\eta^2}=-\frac{(C_{222})^2C_A}{2\pi^2}\eta^2\mathcal{A}^{\mathrm{tree}}_4, \tag*{(161)} A 4 , L 1 − loop η 2 = − 2 π 2 ( C 222 ) 2 C A η 2 A 4 tree , (161)
while the single logarithms have the color-dipole form familiar from the infrared structure of gauge-theory amplitudes [10 , 11 , 12 ] . This holds for any gauge group and does not rely on the planar limit.
Comparison with dimensional regularization. Translating the AdS infrared regularization to a standard scheme such as dimensional regularization is not automatic. In general, there is no universal identification of the geometric logarithm with a Laurent pole in ϵ I R \epsilon_{\mathrm{IR}} ϵ IR . We now present an explicit comparison between two infrared subtraction schemes. Let D = 4 − 2 ϵ I R D=4-2\epsilon_{\mathrm{IR}} D = 4 − 2 ϵ IR . Factoring out the standard kinematic and one-loop prefactor, we define the dimensionless box function I X Y D R I^{\mathrm{DR}}_{XY} I X Y DR by
I b o x D R ( X , Y ) ≡ I X Y D R 8 π 2 X Y . (162) I^{\mathrm{DR}}_{\mathrm{box}}(X,Y)\equiv\frac{\mathcal{I}^{\mathrm{DR}}_{XY}}{8\pi^2XY}. \tag*{(162)} I box DR ( X , Y ) ≡ 8 π 2 X Y I X Y DR . (162)
In the dimensionally regulated normalization of [48 , 49 ] , the massless scalar box reads21
I X Y D R = 2 ϵ I R 2 − ℓ X + ℓ Y ϵ I R + ℓ X ℓ Y − 2 π 2 3 + O ( ϵ I R ) . (163) \mathcal{I}^{\mathrm{DR}}_{XY} = \frac{2}{\epsilon_{\mathrm{IR}}^2} - \frac{\ell_X+\ell_Y}{\epsilon_{\mathrm{IR}}} + \ell_X\ell_Y - \frac{2\pi^2}{3} + O(\epsilon_{\mathrm{IR}}). \tag*{(163)} I X Y DR = ϵ IR 2 2 − ϵ IR ℓ X + ℓ Y + ℓ X ℓ Y − 3 2 π 2 + O ( ϵ IR ) . (163)
Dimensional minimal subtraction is defined by removing only the Laurent poles. The resulting finite remainder is
H X Y D R ( μ ) = ℓ X ℓ Y − 2 π 2 3 . (164) \mathcal{H}^{\mathrm{DR}}_{XY}(\mu) = \ell_X\ell_Y - \frac{2\pi^2}{3}. \tag*{(164)} H X Y DR ( μ ) = ℓ X ℓ Y − 3 2 π 2 . (164)
To compare with the finite remainder in the AdS scheme, we allow for a finite conversion between the reference scales of the two regulators. We write
ℓ X A d S ≡ log ( − X − i 0 μ A d S 2 ) , ℓ X D R ≡ log ( − X − i 0 μ D R 2 ) . (165) \ell_X^{\mathrm{AdS}} \equiv\log\left(\frac{-X-i0}{\mu_{\mathrm{AdS}}^2}\right), \qquad\ell_X^{\mathrm{DR}} \equiv\log\left(\frac{-X-i0}{\mu_{\mathrm{DR}}^2}\right). \tag*{(165)} ℓ X AdS ≡ log ( μ AdS 2 − X − i 0 ) , ℓ X DR ≡ log ( μ DR 2 − X − i 0 ) . (165)
We choose μ D R = e − γ E μ A d S \mu_{\mathrm{DR}}=e^{-\gamma_E}\mu_{\mathrm{AdS}} μ DR = e − γ E μ AdS . This aligns the kinematic logarithms, while clearly the remaining constant depends on the infrared subtraction scheme,
H X Y A d S ( μ A d S ) = H X Y D R ( e − γ E μ A d S ) + π 2 2 . (166) \mathcal{H}^{\mathrm{AdS}}_{XY}(\mu_{\mathrm{AdS}})=\mathcal{H}^{\mathrm{DR}}_{XY}(e^{-\gamma_E}\mu_{\mathrm{AdS}})+\frac{\pi^2}{2}. \tag*{(166)} H X Y AdS ( μ AdS ) = H X Y DR ( e − γ E μ AdS ) + 2 π 2 . (166)
Comparison with the Higgs regulator. It is also instructive to compare with the Higgs regulator of [15 ] , in which planar color-ordered amplitudes are regulated by masses m m m on the Coulomb branch. In the conventions of that reference, the one-loop four-point amplitude is governed by the box function
I ( 1 ) ( s , t ; m 2 ) = 2 log m 2 s log m 2 t − π 2 + O ( m 2 ) , (167) I^{(1)}(s,t;m^2)=2\log\frac{m^2}{s}\log\frac{m^2}{t}-\pi^2+O(m^2), \tag*{(167)} I ( 1 ) ( s , t ; m 2 ) = 2 log s m 2 log t m 2 − π 2 + O ( m 2 ) , (167)
written in the Euclidean region, where the logarithms are real. This has the same bilinear form as (150) . Identifying their s , t s,t s , t , which are positive in the Euclidean region, with − X , − Y -X,-Y − X , − Y , we find
I X Y , L A d S = 1 2 I ( 1 ) ∣ m 2 = e − 2 γ E / L 2 + π 2 3 + o ( 1 ) , (168) \mathcal{I}^{\mathrm{AdS}}_{XY,L}=\left.\frac{1}{2}I^{(1)}\right|_{m^2=e^{-2\gamma_E}/L^2}+\frac{\pi^2}{3}+o(1), \tag*{(168)} I X Y , L AdS = 2 1 I ( 1 ) m 2 = e − 2 γ E / L 2 + 3 π 2 + o ( 1 ) , (168)
so that the AdS radius plays the role of an inverse mass, m ↔ e − γ E / L m\leftrightarrow e^{-\gamma_E}/L m ↔ e − γ E / L , and the two regulators differ by a finite scheme constant. The overall normalization agrees as well.
In the planar limit, each color-ordered amplitude receives a single box, whose coefficient relative to the tree-level amplitude is given by (161) . For S U ( N ) SU(N) S U ( N ) , using C A = 2 κ N C_A=2\kappa N C A = 2 κ N and (62) , this becomes, to leading order in the coupling,
− ( C 222 ) 2 C A 2 π 2 = − g Y M 2 N 4 π 2 = − a , a = g 2 N 8 π 2 , g 2 = 2 g Y M 2 , (169) -\frac{(C_{222})^2 C_A}{2\pi^2}=-\frac{g_{\mathrm{YM}}^2N}{4\pi^2}=-a,\qquad a=\frac{g^2N}{8\pi^2},\qquad g^2=2g_{\mathrm{YM}}^2, \tag*{(169)} − 2 π 2 ( C 222 ) 2 C A = − 4 π 2 g YM 2 N = − a , a = 8 π 2 g 2 N , g 2 = 2 g YM 2 , (169)
where g g g is the conventionally normalized gauge coupling. Given (168) , this reproduces the coefficient of I ( 1 ) I^{(1)} I ( 1 ) in the planar amplitude M 4 = 1 − a 2 I ( 1 ) + ⋯ M_4 = 1 - \frac{a}{2} I^{(1)} + \cdots M 4 = 1 − 2 a I ( 1 ) + ⋯ of [15 ] . There is, however, an important difference of scope. The Higgs regulator renders finite only planar color-ordered amplitudes: beyond the planar limit, the massless fields of the unbroken gauge group run in loops and their infrared divergences remain. The AdS-regulated result (152) , by contrast, holds for the complete color structure.
Discussion
We have studied the four-point function of boundary moment-map operators in four-dimensional N = 4 \mathcal{N}=4 N = 4 SYM on fixed A d S 4 \mathrm{AdS}_4 AdS 4 with Dirichlet boundary conditions. Starting from the microscopic theory, we have reproduced the d = 3 d=3 d = 3 , p = 2 p=2 p = 2 tree-level supergluon amplitude of [38 ] and then used it to reconstruct the non-local one-loop correlator via AdS unitarity. Remarkably, the double-discontinuity computed from the full tree-level correlator admits a simple reduced Mellin representation involving only simultaneous poles, with manifestly symmetric coefficients. The agreement with the complete one-loop double-discontinuity and the flat-space limit provides non-trivial checks of this structure. A further independent check at finite AdS radius comes from the matching of protected OPE coefficients, as discussed in appendix § 11 .
The main conceptual lesson of our calculation is an explicit realization of the AdS radius as an infrared regulator. The boundary correlator itself is infrared safe and independent of L L L ; the radius enters only when the Mellin variables are converted into flat-space momenta. In the resulting large-radius limit, the discrete Mellin poles condense into the branch cuts of the four-dimensional massless box, and the infrared divergences of the flat-space amplitude reappear as logarithms of L L L , with the regulated one-loop answer containing both log 2 L \log^2 L log 2 L and log L \log L log L terms. Their color structure is the universal one dictated by infrared factorization, for any gauge group. The comparison with dimensional regularization and with the Higgs regulator makes this correspondence quantitative, including the finite differences between the infrared subtraction schemes. Our result thus provides a concrete example of how the infrared singularities of a massless gauge theory are regulated by placing the theory on AdS and recovered as L → ∞ L \to\infty L → ∞ .
Beyond perturbation theory, the main virtue of this observable is that nothing in its definition refers to weak coupling or to the planar limit. For a given gauge group, the moment-map correlator is a single function of the cross-ratios and of the complexified coupling τ \tau τ , and the results of this paper describe its expansion around one corner of the conformal manifold. Other corners can be reached by different means. S-duality maps the Dirichlet condition to the coupling to T [ G ] T[G] T [ G ] [28 ] , so that the strong-coupling regime is described by a weakly coupled bulk theory interacting with a strongly coupled boundary theory. At large N N N and large ’t Hooft coupling, N = 4 \mathcal{N}=4 N = 4 SYM on a half-space, or equivalently on A d S 4 \mathrm{AdS}_{4} AdS 4 [18 ] , with half-BPS boundary conditions is dual to type IIB supergravity on the A d S 4 × S 2 × S 2 × Σ \mathrm{AdS}_{4} \times S^{2} \times S^{2} \times\Sigma AdS 4 × S 2 × S 2 × Σ solutions of [86 , 87 ] , as identified in [88 ] . Computing the moment-map four-point function in the solution dual to the Dirichlet condition would give the strong-coupling counterpart of the correlator studied here. In between, the correlator is constrained by crossing symmetry and unitarity, together with protected data that can be computed exactly. The one-loop anomalous dimensions and OPE-coefficient corrections, which can be extracted from our result, will provide, with the tree-level data and the protected OPE coefficients, weak-coupling benchmarks for the numerical conformal bootstrap of this correlator at finite coupling, to which we will return in [34 ] .
Within perturbation theory, a natural next step is to extend this analysis to higher loops. In flat space, infrared singularities of gauge-theory amplitudes obey universal factorization and exponentiation properties. It would be interesting to understand how these structures are realized in AdS, and whether higher-loop infrared divergences can be systematically organized in terms of powers of log L \log L log L . In particular, one may ask how the usual infrared anomalous dimensions and the associated scheme dependence emerge in this geometric regularization. This perspective may also provide a useful starting point for studying infrared physics beyond weak coupling. In particular, it is natural to conjecture that the double logarithms of (159) exponentiate, as they do in flat space [11 , 13 ] . At finite coupling, the boundary correlator would then be Sudakov suppressed at large Mellin variables and fixed angle, with a color structure that reduces at one loop to the dipole form found here. The exponent would be governed by a function of the coupling whose leading term is fixed by our one-loop result. In dimensional regularization, this function is the cusp anomalous dimension [89 , 90 , 91 ] , but it is not regulator independent: on the Coulomb branch it interpolates between the cusp and the octagon anomalous dimensions, which differ beyond one loop [17 , 92 , 93 ] . Which function arises on A d S 4 \mathrm{AdS}_{4} AdS 4 is an open question. We have not tested this beyond one loop.
It would also be interesting to obtain a closed position-space representation of the complete one-loop correlator. Although the individual U U U -slices of the double-discontinuity involve only ordinary logarithms and dilogarithms, an integral representation of the resummed answer reveals a richer analytic structure. As shown in appendix D, the inverse Mellin transform of the one-variable completion τ ( t ) \tau(t) τ ( t ) in (143) involves an algebraic kernel defining a genus-one curve at generic cross-ratios. After a change of variables, this curve coincides with the one appearing in the one-loop Witten diagram of a conformally coupled scalar with Δ = 1 \Delta= 1 Δ = 1 in A d S 4 \mathrm{AdS}_{4} AdS 4 [94 , 95 ] . Whether an irreducible elliptic contribution survives in the complete crossing-symmetric correlator, and whether the answer can be expressed in terms of elliptic multiple polylogarithms, remains to be understood.
Another direction is to investigate the one-loop correlator directly in momentum space, providing an independent approach to its analytic structure and flat-space limit. Supersymmetry could also be used to reconstruct correlators of the conserved-current descendants of the moment map, allowing a more direct comparison with gluon amplitudes [96 ] . It would furthermore be interesting to extend the analysis to other supersymmetric boundary conditions, possibly involving additional three-dimensional degrees of freedom, and ultimately beyond supersymmetry. In particular, generalizing the construction to pure Yang–Mills theory on AdS would help identify which features of the one-loop answer follow from the geometry and which rely on supersymmetric constraints.
The relation to flat-space scattering also raises the question of whether additional symmetries of N = 4 \mathcal{N}=4 N = 4 SYM amplitudes survive, in a generalized form, at finite AdS radius. Planar flat-space amplitudes possess dual conformal and Yangian symmetries, closely related to their integrable structure [97 , 98 ] . Related Yangian structures have also been identified for certain Witten diagrams [99 ] . It would be interesting to investigate whether a deformed or twisted realization of these symmetries acts on the full boundary correlator, with the effects of the AdS curvature and boundary conditions encoded in the corresponding Ward identities. At strong coupling, planar flat-space amplitudes admit a description in terms of minimal surfaces in AdS, where integrability plays a central role [100 , 101 ] . Although this holographic construction is distinct from gauge theory on rigid A d S 4 \mathrm{AdS}_4 AdS 4 , understanding whether any of these structures have a counterpart in the present setting is an intriguing open problem.
We close by returning to the question posed at the beginning. The scattering amplitudes of N = 4 \mathcal{N}=4 N = 4 SYM have been computed to remarkably high orders and have revealed a wealth of hidden structure, yet they are infrared divergent, they are defined only order by order in perturbation theory, and their most striking properties are confined to the planar limit. The boundary correlator studied here encodes the amplitude in its behavior at large Mellin variables, but it is a well-defined function of the coupling for any gauge group. It obeys crossing symmetry and unitarity, it is constrained by superconformal symmetry and by exact results from localization, and it can be approached from weak coupling, from its S-dual description and, at large N N N , from supergravity. It is natural to hope that this makes it one of the few observables of an interacting four-dimensional gauge theory that can be determined at finite coupling. The one-loop analysis presented here is a first step in that direction.
Acknowledgments
We are grateful to Clément Virally for helpful discussions on his reduced Mellin-space construction of the superconformal Ward identities. We also thank Lorenzo Di Pietro for stimulating conversations, and Matteo Sacchi for insightful discussions on related topics. We acknowledge the use of AI agents for assistance with some technical calculations and some editing of the draft. The work of LR is supported in part by NSF grant PHY-2513893 and by the Simons Foundation grant 681267 (Simons Investigator Award).
Three-dimensional conformal blocks
Generic three-dimensional scalar blocks do not admit the finite holomorphic factorization familiar in even dimensions. It is therefore useful to introduce a systematic lightcone expansion at small U U U with fixed W = 1 − V W = 1 - V W = 1 − V :
G Δ , ℓ ( U , V ) = ∑ q = 0 ∞ U Δ − ℓ 2 + q g Δ , ℓ ( q ) ( W ) . (170) G_{\Delta,\ell}(U,V)=\sum_{q=0}^{\infty}U^{\frac{\Delta-\ell}{2}+q}g_{\Delta,\ell}^{(q)}(W). \tag*{(170)} G Δ , ℓ ( U , V ) = q = 0 ∑ ∞ U 2 Δ − ℓ + q g Δ , ℓ ( q ) ( W ) . (170)
We normalize the blocks such that, in the radial OPE limit,
G Δ , ℓ ( U , V ) ∼ U Δ / 2 4 Δ P ℓ ( 1 + U − V 2 U ) , (171) G_{\Delta,\ell}(U,V)\sim\frac{U^{\Delta/2}}{4^\Delta}P_\ell\left(\frac{1+U-V}{2\sqrt{U}}\right), \tag*{(171)} G Δ , ℓ ( U , V ) ∼ 4 Δ U Δ/2 P ℓ ( 2 U 1 + U − V ) , (171)
with the argument of the Legendre polynomial held fixed. In this convention, the leading lightcone block is
g Δ , ℓ ( 0 ) ( W ) = ( 1 2 ) ℓ 4 Δ ℓ ! W ℓ 2 F 1 ( Δ + ℓ 2 , Δ + ℓ 2 ; Δ + ℓ ; W ) . (172) g_{\Delta,\ell}^{(0)}(W)=\frac{\left(\frac{1}{2}\right)_\ell}{4^\Delta\ell!}W^\ell\,{}_2F_1\left(\frac{\Delta+\ell}{2},\frac{\Delta+\ell}{2};\Delta+\ell;W\right). \tag*{(172)} g Δ , ℓ ( 0 ) ( W ) = 4 Δ ℓ ! ( 2 1 ) ℓ W ℓ 2 F 1 ( 2 Δ + ℓ , 2 Δ + ℓ ; Δ + ℓ ; W ) . (172)
The first subleading lightcone correction can be written as
g Δ , ℓ ( 1 ) ( W ) = ( 1 2 ) ℓ 4 Δ ℓ ! [ a 10 ( Δ , ℓ ) W ℓ 2 F 1 ( Δ + ℓ 2 + 1 , Δ + ℓ 2 + 1 ; Δ + ℓ + 2 ; W ) + a 12 ( ℓ ) W ℓ − 2 2 F 1 ( Δ + ℓ 2 − 1 , Δ + ℓ 2 − 1 ; Δ + ℓ − 2 ; W ) ] . (173) g_{\Delta,\ell}^{(1)}(W)=\frac{\left(\frac{1}{2}\right)_\ell}{4^\Delta\ell!}\left[a_{10}(\Delta,\ell)W^\ell\,{}_2F_1\left(\frac{\Delta+\ell}{2}+1,\frac{\Delta+\ell}{2}+1;\Delta+\ell+2;W\right)+a_{12}(\ell)W^{\ell-2}\,{}_2F_1\left(\frac{\Delta+\ell}{2}-1,\frac{\Delta+\ell}{2}-1;\Delta+\ell-2;W\right)\right]. \tag*{(173)} g Δ , ℓ ( 1 ) ( W ) = 4 Δ ℓ ! ( 2 1 ) ℓ [ a 10 ( Δ , ℓ ) W ℓ 2 F 1 ( 2 Δ + ℓ + 1 , 2 Δ + ℓ + 1 ; Δ + ℓ + 2 ; W ) + a 12 ( ℓ ) W ℓ − 2 2 F 1 ( 2 Δ + ℓ − 1 , 2 Δ + ℓ − 1 ; Δ + ℓ − 2 ; W ) ] . (173)
where
a 10 ( Δ , ℓ ) = ( Δ − 1 ) 2 ( Δ + ℓ 2 ) 4 ( Δ + ℓ − 1 ) 2 ( Δ + ℓ ) 2 ( Δ − 1 2 ) , (174) a_{10}(\Delta,\ell)=\frac{(\Delta-1)_2\left(\frac{\Delta+\ell}{2}\right)_4}{(\Delta+\ell-1)_2(\Delta+\ell)_2\left(\Delta-\frac{1}{2}\right)}, \tag*{(174)} a 10 ( Δ , ℓ ) = ( Δ + ℓ − 1 ) 2 ( Δ + ℓ ) 2 ( Δ − 2 1 ) ( Δ − 1 ) 2 ( 2 Δ + ℓ ) 4 , (174)
while
a 12 ( ℓ ) = { − 2 ℓ ( ℓ − 1 ) 2 ℓ − 1 , ℓ ≥ 2 , 0 , ℓ = 0 , 1. (175) a_{12}(\ell)=
\begin{cases}
-\dfrac{2\ell(\ell-1)}{2\ell-1}, & \ell\ge2,\\
0, & \ell=0,1.
\end{cases} \tag*{(175)} a 12 ( ℓ ) = ⎩ ⎨ ⎧ − 2 ℓ − 1 2 ℓ ( ℓ − 1 ) , 0 , ℓ ≥ 2 , ℓ = 0 , 1. (175)
Higher lightcone orders can be generated recursively from the three-dimensional quadratic Casimir equation or from the closed-form coefficients of [102 ] . For the blocks relevant to our calculation, the contribution at order U 2 + p U^{2+p} U 2 + p receives terms with n + q = p n+q=p n + q = p . Thus, after subtracting the contributions of lower-twist families, the leading lightcone blocks determine the coefficients ω p , ℓ \omega_{p,\ell} ω p , ℓ recursively. This is the procedure used to match the position-space D ˉ \bar{D} D ˉ representation to the conformal block decomposition.
Mellin to position space
In this appendix we collect the Mellin–Barnes dictionary used in § 4.2 to rewrite the tree-level Mellin amplitude in terms of normalized D ˉ \bar{D} D ˉ -functions and to extract its log U U U contribution.
For generic positive weights r i r_i r i , a scalar contact Witten diagram can be written as
D r 1 r 2 r 3 r 4 ( x i ) ≡ ∫ A d S d + 1 d z 0 d d z z 0 d + 1 ∏ i = 1 4 ( z 0 z 0 2 + ( z ⃗ − x ⃗ i ) 2 ) r i . (176) D_{r_1r_2r_3r_4}(x_i) \equiv\int_{\mathrm{AdS}_{d+1}} \frac{\mathrm{d}z_0 \mathrm{d}^d z}{z_0^{d+1}} \prod_{i=1}^{4} \left(\frac{z_0}{z_0^2+(\vec{z}-\vec{x}_i)^2}\right)^{r_i}. \tag*{(176)} D r 1 r 2 r 3 r 4 ( x i ) ≡ ∫ AdS d + 1 z 0 d + 1 d z 0 d d z i = 1 ∏ 4 ( z 0 2 + ( z − x i ) 2 z 0 ) r i . (176)
After extracting the standard conformal prefactor, the normalized function D ˉ r 1 r 2 r 3 r 4 \bar{D}_{r_1r_2r_3r_4} D ˉ r 1 r 2 r 3 r 4 is defined by
2 π − d / 2 ∏ i = 1 4 Γ ( r i ) Γ ( 1 2 ∑ i r i − d 2 ) D r 1 r 2 r 3 r 4 ( x i ) ≡ ( x 14 2 ) 1 2 ∑ i r i − r 1 − r 4 ( x 34 2 ) 1 2 ∑ i r i − r 3 − r 4 ( x 13 2 ) 1 2 ∑ i r i − r 4 ( x 24 2 ) r 2 D ˉ r 1 r 2 r 3 r 4 ( U , V ) . (177) \frac{2\pi^{-d/2}\prod_{i=1}^{4}\Gamma(r_i)}{\Gamma\left(\frac{1}{2}\sum_i r_i-\frac{d}{2}\right)}D_{r_1r_2r_3r_4}(x_i) \equiv\frac{(x_{14}^2)^{\frac{1}{2}\sum_i r_i-r_1-r_4}(x_{34}^2)^{\frac{1}{2}\sum_i r_i-r_3-r_4}}{(x_{13}^2)^{\frac{1}{2}\sum_i r_i-r_4}(x_{24}^2)^{r_2}}\bar{D}_{r_1r_2r_3r_4}(U,V). \tag*{(177)} Γ ( 2 1 ∑ i r i − 2 d ) 2 π − d /2 ∏ i = 1 4 Γ ( r i ) D r 1 r 2 r 3 r 4 ( x i ) ≡ ( x 13 2 ) 2 1 ∑ i r i − r 4 ( x 24 2 ) r 2 ( x 14 2 ) 2 1 ∑ i r i − r 1 − r 4 ( x 34 2 ) 2 1 ∑ i r i − r 3 − r 4 D ˉ r 1 r 2 r 3 r 4 ( U , V ) . (177)
With this normalization, the reduced function D ˉ \bar{D} D ˉ is independent of the spacetime dimension d d d . It admits the Mellin–Barnes representation
D ˉ r 1 r 2 r 3 r 4 ( U , V ) = ∫ d σ d τ ( 4 π i ) 2 U σ / 2 V τ / 2 D r i ( σ , τ ) , D r i ( σ , τ ) = Γ ( − σ 2 ) Γ ( − τ 2 ) Γ ( r 2 + σ + τ 2 ) Γ ( − r 1 − r 2 + r 3 + r 4 − σ 2 ) × Γ ( r 1 − r 2 − r 3 + r 4 − τ 2 ) Γ ( r 1 + r 2 + r 3 − r 4 + σ + τ 2 ) . (178) \begin{aligned}
\bar{D}_{r_1r_2r_3r_4}(U,V) &= \int\frac{\mathrm{d}\sigma\,\mathrm{d}\tau}{(4\pi i)^2}U^{\sigma/2}V^{\tau/2}\mathcal{D}_{r_i}(\sigma,\tau),\\
\mathcal{D}_{r_i}(\sigma,\tau) &= \Gamma\left(-\frac{\sigma}{2}\right)\Gamma\left(-\frac{\tau}{2}\right)\Gamma\left(r_2+\frac{\sigma+\tau}{2}\right)\Gamma\left(\frac{-r_1-r_2+r_3+r_4-\sigma}{2}\right)\\
&\quad\times\Gamma\left(\frac{r_1-r_2-r_3+r_4-\tau}{2}\right)\Gamma\left(\frac{r_1+r_2+r_3-r_4+\sigma+\tau}{2}\right). \tag*{(178)}
\end{aligned} D ˉ r 1 r 2 r 3 r 4 ( U , V ) D r i ( σ , τ ) = ∫ ( 4 π i ) 2 d σ d τ U σ /2 V τ /2 D r i ( σ , τ ) , = Γ ( − 2 σ ) Γ ( − 2 τ ) Γ ( r 2 + 2 σ + τ ) Γ ( 2 − r 1 − r 2 + r 3 + r 4 − σ ) × Γ ( 2 r 1 − r 2 − r 3 + r 4 − τ ) Γ ( 2 r 1 + r 2 + r 3 − r 4 + σ + τ ) . (178)
For the non-positive and half-integer weights encountered below, (B.2) should not be interpreted as a literal convergent contact integral. Instead, we define the corresponding D ˉ \bar{D} D ˉ -functions by analytic continuation in the parameters r i r_i r i of the Mellin–Barnes representation (B.3).
Our four-point function is written using the Mellin convention
G ( U , V ) = ∫ d s d t ( 4 π i ) 2 U s / 2 V t / 2 − 1 K ( s , t ) M ( s , t ) , (179) \mathcal{G}(U,V)=\int\frac{\mathrm{d}s\,\mathrm{d}t}{(4\pi i)^2}U^{s/2}V^{t/2-1}\mathcal{K}(s,t)\mathcal{M}(s,t), \tag*{(179)} G ( U , V ) = ∫ ( 4 π i ) 2 d s d t U s /2 V t /2 − 1 K ( s , t ) M ( s , t ) , (179)
where
K ( s , t ) ≡ Γ ( 2 − s 2 ) 2 Γ ( 2 − t 2 ) 2 Γ ( 2 − u 2 ) 2 , s + t + u = 4. (180) \mathcal{K}(s,t)\equiv\Gamma\left(\frac{2-s}{2}\right)^2\Gamma\left(\frac{2-t}{2}\right)^2\Gamma\left(\frac{2-u}{2}\right)^2,\qquad s+t+u=4. \tag*{(180)} K ( s , t ) ≡ Γ ( 2 2 − s ) 2 Γ ( 2 2 − t ) 2 Γ ( 2 2 − u ) 2 , s + t + u = 4. (180)
Consider now a position-space term
U a V b D ˉ r 1 r 2 r 3 r 4 ( U , V ) . (181) U^aV^b\bar{D}_{r_1r_2r_3r_4}(U,V). \tag*{(181)} U a V b D ˉ r 1 r 2 r 3 r 4 ( U , V ) . (181)
Using (B.3), it takes the form
U a V b D ˉ r 1 r 2 r 3 r 4 ( U , V ) = ∫ d σ d τ ( 4 π i ) 2 U σ / 2 + a V τ / 2 + b D r i ( σ , τ ) . (182) U^{a}V^{b}\bar{D}_{r_{1}r_{2}r_{3}r_{4}}(U,V)=\int\frac{\mathrm{d}\sigma\,\mathrm{d}\tau}{(4\pi i)^{2}}U^{\sigma/2+a}V^{\tau/2+b}\mathcal{D}_{r_i}(\sigma,\tau). \tag*{(182)} U a V b D ˉ r 1 r 2 r 3 r 4 ( U , V ) = ∫ ( 4 π i ) 2 d σ d τ U σ /2 + a V τ /2 + b D r i ( σ , τ ) . (182)
Matching the powers of the cross-ratios with U s / 2 V t / 2 − 1 U^{s/2}V^{t/2-1} U s /2 V t /2 − 1 gives
σ = s − 2 a , τ = t − 2 − 2 b . (183) \sigma=s-2a,\qquad\tau=t-2-2b. \tag*{(183)} σ = s − 2 a , τ = t − 2 − 2 b . (183)
The Mellin amplitude represented by this term is therefore
U a V b D ˉ r 1 r 2 r 3 r 4 ( U , V ) ⟷ D r i ( s − 2 a , t − 2 − 2 b ) K ( s , t ) , (184) U^{a}V^{b}\bar{D}_{r_{1}r_{2}r_{3}r_{4}}(U,V)\quad\longleftrightarrow\quad\frac{\mathcal{D}_{r_i}(s-2a,t-2-2b)}{\mathcal{K}(s,t)}, \tag*{(184)} U a V b D ˉ r 1 r 2 r 3 r 4 ( U , V ) ⟷ K ( s , t ) D r i ( s − 2 a , t − 2 − 2 b ) , (184)
with u = 4 − s − t u=4-s-t u = 4 − s − t understood throughout. This formula is particularly useful because ratios of the Gamma functions in D r i \mathcal{D}_{r_i} D r i against the universal external kernel K ( s , t ) \mathcal{K}(s,t) K ( s , t ) frequently collapse to the rational and Gamma-function structures appearing directly in the tree-level Mellin amplitude.
Applying this dictionary to the tree-level Mellin amplitude gives a finite D ˉ \bar{D} D ˉ -function representation for each of its three R-symmetry components. Suppressing the common arguments ( U , V ) (U,V) ( U , V ) of the D ˉ \bar{D} D ˉ -functions, we find
[ α 0 ] G t r e e = 1 π 3 [ − c s ( π U D ˉ 2 , 1 , 1 , 0 − π U 1 / 2 D ˉ 3 2 , 1 2 , 1 , 0 ) + c t ( π U V D ˉ 1 , 1 , 0 , 2 − π U V D ˉ 3 2 , 1 , 0 , 3 2 ) ] , [\alpha^{0}]\mathcal{G}_{\mathrm{tree}}=\frac{1}{\pi^{3}}\left[-c_{s}\left(\pi U\bar{D}_{2,1,1,0}-\sqrt{\pi}U^{1/2}\bar{D}_{\frac{3}{2},\frac{1}{2},1,0}\right)+c_{t}\left(\pi\frac{U}{V}\bar{D}_{1,1,0,2}-\sqrt{\pi}\frac{U}{V}\bar{D}_{\frac{3}{2},1,0,\frac{3}{2}}\right)\right], [ α 0 ] G tree = π 3 1 [ − c s ( π U D ˉ 2 , 1 , 1 , 0 − π U 1/2 D ˉ 2 3 , 2 1 , 1 , 0 ) + c t ( π V U D ˉ 1 , 1 , 0 , 2 − π V U D ˉ 2 3 , 1 , 0 , 2 3 ) ] ,
[ α ] G t r e e = 1 π 3 { c s [ π U ( D ˉ 2 , 1 , 0 , 1 + D ˉ 2 , 1 , 1 , 0 ) − π U 1 / 2 ( D ˉ 3 2 , 1 2 , 0 , 1 + D ˉ 3 2 , 1 2 , 1 , 0 ) ] + c t [ π U 2 V D ˉ 2 , 1 , 0 , 1 − π U 2 V − 1 / 2 D ˉ 3 2 , 1 2 , 0 − π U V − 1 D ˉ 1 , 1 , 0 , 2 + π U V D ˉ 3 2 , 1 , 0 , 3 2 ] − c u [ π U 2 D ˉ 2 , 1 , 1 , 0 − π U 2 D ˉ 3 2 , 1 , 1 , 1 2 , 0 + π U D ˉ 2 , 0 , 1 , 1 − π U D ˉ 3 2 , 0 , 1 2 , 1 ] } , \begin{aligned}
[\alpha]\mathcal{G}_{\mathrm{tree}}={}&\frac{1}{\pi^{3}}\Bigg\{c_{s}\left[\pi U\left(\bar{D}_{2,1,0,1}+\bar{D}_{2,1,1,0}\right)-\sqrt{\pi}U^{1/2}\left(\bar{D}_{\frac{3}{2},\frac{1}{2},0,1}+\bar{D}_{\frac{3}{2},\frac{1}{2},1,0}\right)\right]\\
&+c_{t}\left[\pi\frac{U^{2}}{V}\bar{D}_{2,1,0,1}-\sqrt{\pi}U^{2}V^{-1/2}\bar{D}_{\frac{3}{2},\frac{1}{2},0}-\pi U V^{-1}\bar{D}_{1,1,0,2}+\sqrt{\pi}\frac{U}{V}\bar{D}_{\frac{3}{2},1,0,\frac{3}{2}}\right]\\
&-c_{u}\left[\pi U^{2}\bar{D}_{2,1,1,0}-\sqrt{\pi}U^{2}\bar{D}_{\frac{3}{2},1,1,\frac{1}{2},0}+\pi U\bar{D}_{2,0,1,1}-\sqrt{\pi}U\bar{D}_{\frac{3}{2},0,\frac{1}{2},1}\right]\Bigg\},
\end{aligned} [ α ] G tree = π 3 1 { c s [ π U ( D ˉ 2 , 1 , 0 , 1 + D ˉ 2 , 1 , 1 , 0 ) − π U 1/2 ( D ˉ 2 3 , 2 1 , 0 , 1 + D ˉ 2 3 , 2 1 , 1 , 0 ) ] + c t [ π V U 2 D ˉ 2 , 1 , 0 , 1 − π U 2 V − 1/2 D ˉ 2 3 , 2 1 , 0 − π U V − 1 D ˉ 1 , 1 , 0 , 2 + π V U D ˉ 2 3 , 1 , 0 , 2 3 ] − c u [ π U 2 D ˉ 2 , 1 , 1 , 0 − π U 2 D ˉ 2 3 , 1 , 1 , 2 1 , 0 + π U D ˉ 2 , 0 , 1 , 1 − π U D ˉ 2 3 , 0 , 2 1 , 1 ] } ,
[ α 2 ] G t r e e = U 2 π 3 [ c u ( π D ˉ 2 , 1 , 1 , 0 − π D ˉ 3 2 , 1 , 1 , 1 2 , 0 ) − c t ( π V − 1 D ˉ 2 , 1 , 0 , 1 − π V − 1 / 2 D ˉ 3 2 , 1 2 , 0 ) ] . (185) [\alpha^{2}]\mathcal{G}_{\mathrm{tree}}=\frac{U^{2}}{\pi^{3}}\left[c_{u}\left(\pi\bar{D}_{2,1,1,0}-\sqrt{\pi}\bar{D}_{\frac{3}{2},1,1,\frac{1}{2},0}\right)-c_{t}\left(\pi V^{-1}\bar{D}_{2,1,0,1}-\sqrt{\pi}V^{-1/2}\bar{D}_{\frac{3}{2},\frac{1}{2},0}\right)\right]. \tag*{(185)} [ α 2 ] G tree = π 3 U 2 [ c u ( π D ˉ 2 , 1 , 1 , 0 − π D ˉ 2 3 , 1 , 1 , 2 1 , 0 ) − c t ( π V − 1 D ˉ 2 , 1 , 0 , 1 − π V − 1/2 D ˉ 2 3 , 2 1 , 0 ) ] . (185)
The highest R-symmetry component in the last line is the representation used in the main text to extract the logarithmic term required for the tree-level CFT data.
Extraction of the log U U U term. The coefficient of log U \log U log U can be obtained directly from the Mellin–Barnes representation, without requiring a closed form for the complete D ˉ \bar{D} D ˉ functions. Following [80 , 103 ] ,
D ˉ r 1 r 2 r 3 r 4 ( U , V ) ∣ log U = ∑ m , n ≥ 0 d m , n ( r i ) U m ( 1 − V ) n , (186) \left.\bar{D}_{r_1r_2r_3r_4}(U,V)\right|_{\log U}=\sum_{m,n\geq0}d_{m,n}(r_i)U^m(1-V)^n, \tag*{(186)} D ˉ r 1 r 2 r 3 r 4 ( U , V ) l o g U = m , n ≥ 0 ∑ d m , n ( r i ) U m ( 1 − V ) n , (186)
where the coefficients obtained from the Mellin–Barnes residues are
d m , n ( r i ) = − exp [ i π 2 ( r 1 + r 2 − r 3 − r 4 ) ] Γ ( r 1 + m ) Γ ( r 2 + m + n ) Γ ( m + 1 ) Γ ( n + 1 ) Γ ( r 1 + r 2 + 2 m + n ) Γ ( r 1 + r 2 − r 3 + r 4 2 + m ) Γ ( r 1 + r 2 + r 3 − r 4 2 + m + n ) Γ ( r 1 + r 2 − r 3 − r 4 + 2 2 + m ) . (187) d_{m,n}(r_i)=-\exp\left[\frac{i\pi}{2}(r_1+r_2-r_3-r_4)\right]\frac{\Gamma(r_1+m)\Gamma(r_2+m+n)}{\Gamma(m+1)\Gamma(n+1)\Gamma(r_1+r_2+2m+n)}
\frac{\Gamma\left(\frac{r_1+r_2-r_3+r_4}{2}+m\right)\Gamma\left(\frac{r_1+r_2+r_3-r_4}{2}+m+n\right)}{\Gamma\left(\frac{r_1+r_2-r_3-r_4+2}{2}+m\right)}. \tag*{(187)} d m , n ( r i ) = − exp [ 2 iπ ( r 1 + r 2 − r 3 − r 4 ) ] Γ ( m + 1 ) Γ ( n + 1 ) Γ ( r 1 + r 2 + 2 m + n ) Γ ( r 1 + m ) Γ ( r 2 + m + n ) Γ ( 2 r 1 + r 2 − r 3 − r 4 + 2 + m ) Γ ( 2 r 1 + r 2 − r 3 + r 4 + m ) Γ ( 2 r 1 + r 2 + r 3 − r 4 + m + n ) . (187)
As above, this expression is understood by analytic continuation in the weights whenever one or more of the r i r_i r i are non-positive or half-integer. For the four combinations in (77) , the continuation gives real coefficients in the branch convention used throughout the paper.
Local ambiguities
The simultaneous-pole reconstruction provides strong evidence that the complete one-loop log 2 U \log^2 U log 2 U coefficient is captured without additional reduced single-pole terms. We now examine possible local contact-term ambiguities at separated boundary points. Finite-derivative contact interactions are represented by polynomial full Mellin amplitudes [38 ] . In the reduced representation used for the cut, we first consider a polynomial completion (with flavor indices understood),
M ~ t o t 1 -loop = M ~ c u t 1 -loop + P l o c ( s , t ) . (188) \widetilde{\mathcal{M}}_{\mathrm{tot}}^{1\text{-loop}}=\widetilde{\mathcal{M}}_{\mathrm{cut}}^{1\text{-loop}}+P_{\mathrm{loc}}(s,t). \tag*{(188)} M tot 1 -loop = M cut 1 -loop + P loc ( s , t ) . (188)
We do not assume that every polynomial full Ward solution admits a polynomial reduced primitive.22
Visibility under the Ward lift. The full amplitude is reconstructed from the reduced one by the superconformal Ward operators [47 ] ,
M f u l l ( s , t ; α ) = ∑ J = 0 2 α J D J M ~ ( s , t ) . (189) M_{\mathrm{full}}(s,t;\alpha)=\sum_{J=0}^{2}\alpha^J D_J\widetilde{\mathcal{M}}(s,t). \tag*{(189)} M full ( s , t ; α ) = J = 0 ∑ 2 α J D J M ( s , t ) . (189)
Let P q P_q P q be the nonzero homogeneous part of highest degree q ≥ 0 q\geq0 q ≥ 0 of P l o c P_{\mathrm{loc}} P loc , so that
P l o c ( λ s , λ t ) = λ q P q ( s , t ) + O ( λ q − 1 ) . P_{\mathrm{loc}}(\lambda s,\lambda t)=\lambda^qP_q(s,t)+O(\lambda^{q-1}). P loc ( λ s , λ t ) = λ q P q ( s , t ) + O ( λ q − 1 ) .
Expanding the finite shifts in the Ward operators at large Mellin variables, the nominally leading cubic terms cancel. Defining E = s ∂ s + t ∂ t \mathcal{E}=s\partial_s+t\partial_t E = s ∂ s + t ∂ t , the first non-vanishing contribution is
∑ J = 0 2 α J D J ∼ ( u ~ + α s ) 2 ( 3 + 2 E ) , u ~ = − s − t . (190) \sum_{J=0}^{2}\alpha^J\mathcal{D}_J\sim(\widetilde{u}+\alpha s)^2(3+2\mathcal{E}),\qquad\widetilde{u}=-s-t. \tag*{(190)} J = 0 ∑ 2 α J D J ∼ ( u + α s ) 2 ( 3 + 2 E ) , u = − s − t . (190)
In particular,
[ D 2 P l o c ] q + 2 = ( 3 + 2 q ) s 2 P q ( s , t ) , (191) \left[\mathcal{D}_2P^{\mathrm{loc}}\right]_{q+2}=(3+2q)s^2P_q(s,t), \tag*{(191)} [ D 2 P loc ] q + 2 = ( 3 + 2 q ) s 2 P q ( s , t ) , (191)
which cannot vanish for a nonzero P q P_q P q . Thus a reduced polynomial of degree q q q produces a full Mellin polynomial of degree q + 2 q+2 q + 2 , and the Ward lift is injective on reduced polynomials, already in the α 2 \alpha^2 α 2 component. The first cases, q = 0 , 1 q=0,1 q = 0 , 1 , have the derivative counting of supersymmetric F 4 F^4 F 4 and D 2 F 4 D^2F^4 D 2 F 4 contact interactions, respectively.
Exclusion from the flat-space limit. To exclude contact terms without assuming that all polynomial Ward solutions have reduced polynomial primitives, consider directly an arbitrary nonzero polynomial addition to the full Mellin amplitude, with highest homogeneous part Q d ( s , t ; α ) Q_d(s,t;\alpha) Q d ( s , t ; α ) of degree d ≥ 0 d\geq0 d ≥ 0 . Any polynomial decomposes into homogeneous parts in s , t s,t s , t , and its highest-degree part controls the leading large-L L L behavior. In the large-radius scaling of § 6 ,
s = L 2 S 2 β , t = L 2 T 2 β , s=\frac{L^2S}{2\beta},\qquad t=\frac{L^2T}{2\beta}, s = 2 β L 2 S , t = 2 β L 2 T ,
its leading contribution under the normalized Penedones transform (130) is
δ A 4 , L p o l y = 8 π 2 ( 2 d − 1 ) ! ! L 2 d Q d ( S , T ; α ) + O ( L 2 d − 2 ) , (192) \delta\mathcal{A}^{\mathrm{poly}}_{4,L}=\frac{8\pi^2}{(2d-1)!!}L^{2d}Q_d(S,T;\alpha)+O(L^{2d-2}), \tag*{(192)} δ A 4 , L poly = ( 2 d − 1 )!! 8 π 2 L 2 d Q d ( S , T ; α ) + O ( L 2 d − 2 ) , (192)
with ( − 1 ) ! ! = 1 (-1)!!=1 ( − 1 )!! = 1 . Here Q d Q_d Q d is evaluated at generic fixed four-dimensional kinematics. For a reduced polynomial, (C.4) gives d = q + 2 d=q+2 d = q + 2 and
Q q + 2 ( S , T ; α ) = ( 3 + 2 q ) K ( S , T ; α ) P q ( S , T ) , K ( S , T ; α ) = ( ( α − 1 ) S − T ) 2 , Q_{q+2}(S,T;\alpha)=(3+2q)K(S,T;\alpha)P_q(S,T),\qquad K(S,T;\alpha)=\left((\alpha-1)S-T\right)^2, Q q + 2 ( S , T ; α ) = ( 3 + 2 q ) K ( S , T ; α ) P q ( S , T ) , K ( S , T ; α ) = ( ( α − 1 ) S − T ) 2 ,
which is nonzero for P q ≠ 0 P_q\ne0 P q = 0 at generic kinematics.
For d ≥ 1 d\geq1 d ≥ 1 , (C.5) produces a power divergence in L L L . By contrast, the massless one-loop four-point amplitude of four-dimensional N = 4 \mathcal{N}=4 N = 4 SYM is cut-constructible, built from scalar boxes, with no independent rational remainder [74 , 85 , 104 ] . Its infrared singularities are logarithmic, in agreement with our reconstructed cut in § 6 :
A 4 , L c u t = O ( log 2 L ) . \mathcal{A}^{\mathrm{cut}}_{4,L}=O(\log^2 L). A 4 , L cut = O ( log 2 L ) .
Thus a nonzero Q d Q_d Q d with d ≥ 1 d\geq1 d ≥ 1 would introduce an uncancelled power divergence absent from the microscopic amplitude. A constant full polynomial (d = 0 d = 0 d = 0 ) would instead give an independent rational flat-space term, which is likewise absent; such a polynomial is also forbidden by the full Ward identities together with crossing symmetry, as found in appendix D of [38 ] . UV finiteness [105 , 106 ] by itself would not exclude finite local terms.
Nor can the coefficient of a polynomial addition be made to vanish by an inverse power of L L L . Since the theory is Weyl-equivalent to N = 4 \mathcal{N} = 4 N = 4 SYM on the flat half-space, with fixed conformal boundary conditions, the normalized boundary correlator is independent of L L L . Its dimensionless Mellin coefficients therefore cannot acquire inverse powers of L L L . In particular, independent higher-derivative curvature-suppressed contact terms would require an additional dimensionful coupling, absent from the microscopic theory. This does not exclude curvature couplings fixed by the conformal completion of the action, which are already part of the theory. Flat-space matching therefore excludes every nonzero full polynomial contact term. Combined with the injectivity of the Ward lift, this also requires P l o c = 0 P_{\mathrm{loc}} = 0 P loc = 0 .
We stress the status of this argument. Because L L L enters only through the scaling of the Mellin variables, the absence of power divergences in L L L is the same statement as a bound on the growth of the full Mellin amplitude at large s , t s,t s , t with fixed ratio. This bound is not an independent assumption: it follows from requiring that the large-radius limit of the correlator reproduce the flat-space one-loop amplitude, as in § 6 . That requirement is a physical input. Crossing symmetry and the double-discontinuity alone do not fix local terms.
Independent protected OPE check. An independent check at finite AdS radius comes from the protected B [ 0 ] 2 ( 4 ; 0 ) \mathcal{B}[0]_{2}^{(4;0)} B [ 0 ] 2 ( 4 ; 0 ) multiplets in the moment-map OPE. Their squared OPE coefficients can be computed independently using localization of the mass-deformed hemisphere partition function and the associated protected twisted-trace algebra [50 ] , as we show explicitly in [34 ] . For S U ( N ≥ 4 ) SU(N \ge4) S U ( N ≥ 4 ) , these multiplets appear in the symmetric flavor representations
Sym 2 ( Adj ) = 1 ⊕ Adj S ⊕ R − ⊕ R + , (193) \operatorname{Sym}^{2}(\operatorname{Adj}) = 1 \oplus\operatorname{Adj}_{S} \oplus\mathcal{R}_{-} \oplus\mathcal{R}_{+}, \tag*{(193)} Sym 2 ( Adj ) = 1 ⊕ Adj S ⊕ R − ⊕ R + , (193)
where R − \mathcal{R}_{-} R − and R + \mathcal{R}_{+} R + denote the two additional irreducible representations, characterized by the Casimirs
C 2 ( R ± ) = 4 ( N ± 1 ) , C A = 2 N , (194) C_{2}(\mathcal{R}_{\pm}) = 4(N \pm1), \qquad C_{A} = 2N, \tag*{(194)} C 2 ( R ± ) = 4 ( N ± 1 ) , C A = 2 N , (194)
in our normalization κ = 1 \kappa= 1 κ = 1 . In the following, we denote by G R ( 2 ) \mathcal{G}_{\mathcal{R}}^{(2)} G R ( 2 ) and M ~ R ( 2 ) \widetilde{\mathcal{M}}_{\mathcal{R}}^{(2)} M R ( 2 ) the order-ε 2 \varepsilon^{2} ε 2 coefficients of the correlator and reduced Mellin amplitude, respectively, in the flavor representation R \mathcal{R} R , with the overall factor ε 2 \varepsilon^{2} ε 2 extracted.
The protected multiplet B [ 0 ] 2 ( 4 ; 0 ) \mathcal{B}[0]_{2}^{(4;0)} B [ 0 ] 2 ( 4 ; 0 ) has a scalar superprimary of dimension Δ = 2 \Delta= 2 Δ = 2 and S U ( 2 ) H SU(2)_H S U ( 2 ) H spin two. In our normalization, its superconformal block starts at V = 1 V=1 V = 1 as
U ( 1 − 6 α + 6 α 2 ) = 12 U Y 4 ( α ) , (195) U\left(1-6\alpha+6\alpha^2\right)=12UY_4(\alpha), \tag*{(195)} U ( 1 − 6 α + 6 α 2 ) = 12 U Y 4 ( α ) , (195)
in terms of the R-symmetry harmonic introduced in § 4 . The leading U U U term is therefore extracted from the s = 2 s=2 s = 2 pole of the Mellin measure. Near this pole, the Ward lift gives
D 2 M ~ R ( 2 ) = ( s − 2 ) [ ( t − 2 ) M ~ R ( 2 ) ( 0 , t − 2 ) − t M ~ R ( 2 ) ( 0 , t ) ] + O ( ( s − 2 ) 2 ) . D_2\widetilde{\mathcal{M}}_{\mathcal{R}}^{(2)}=(s-2)\left[(t-2)\widetilde{\mathcal{M}}_{\mathcal{R}}^{(2)}(0,t-2)-t\widetilde{\mathcal{M}}_{\mathcal{R}}^{(2)}(0,t)\right]+O\left((s-2)^2\right). D 2 M R ( 2 ) = ( s − 2 ) [ ( t − 2 ) M R ( 2 ) ( 0 , t − 2 ) − t M R ( 2 ) ( 0 , t ) ] + O ( ( s − 2 ) 2 ) .
The factor ( s − 2 ) (s-2) ( s − 2 ) cancels one of the double poles of the Mellin measure, leaving a term proportional to U U U , without log U \log U log U . At V = 1 V=1 V = 1 , the remaining Gamma-function weight is periodic under t → t + 2 t\to t+2 t → t + 2 , allowing the contour integral to be reduced to the residue at t = 0 t=0 t = 0 . For the contact polynomials and the reconstructed cut kernel, this gives
[ U α 2 ] G R ( 2 ) ( U , 1 ) = 4 M ~ R ( 2 ) ( 0 , 0 ) . [U\alpha^2]\mathcal{G}_{\mathcal{R}}^{(2)}(U,1)=4\widetilde{\mathcal{M}}_{\mathcal{R}}^{(2)}(0,0). [ U α 2 ] G R ( 2 ) ( U , 1 ) = 4 M R ( 2 ) ( 0 , 0 ) .
Comparing with the coefficient 6 6 6 of U α 2 U\alpha^2 U α 2 in the protected superblock, we obtain
a B , R ( 2 ) = 2 3 M ~ R ( 2 ) ( 0 , 0 ) , (196) a_{\mathcal{B},\mathcal{R}}^{(2)}=\frac{2}{3}\widetilde{\mathcal{M}}_{\mathcal{R}}^{(2)}(0,0), \tag*{(196)} a B , R ( 2 ) = 3 2 M R ( 2 ) ( 0 , 0 ) , (196)
where a B , R ( 2 ) a_{\mathcal{B},\mathcal{R}}^{(2)} a B , R ( 2 ) denotes the order-ϵ 2 \epsilon^2 ϵ 2 correction to the squared OPE coefficient of the protected multiplet B = B [ 0 ] ( 4 ; 0 ) \mathcal{B}=\mathcal{B}_{[0]}^{(4;0)} B = B [ 0 ] ( 4 ; 0 ) in the flavor representation R \mathcal{R} R . For the reconstructed cut, B ( 0 , 0 ) = 1 / 1620 B(0,0)=1/1620 B ( 0 , 0 ) = 1/1620 , and the flavor projections give
( a B , 1 ( 2 ) , a B , A d j S ( 2 ) , a B , R − ( 2 ) , a B , R + ( 2 ) ) c u t = ( N 2 27 , N 2 135 , 6 − N 135 , 6 + N 135 ) , (197) \left(a_{\mathcal{B},1}^{(2)},a_{\mathcal{B},\mathrm{Adj}_S}^{(2)},a_{\mathcal{B},\mathcal{R}_{-}}^{(2)},a_{\mathcal{B},\mathcal{R}_{+}}^{(2)}\right)_{\mathrm{cut}}=\left(\frac{N^2}{27},\frac{N^2}{135},\frac{6-N}{135},\frac{6+N}{135}\right), \tag*{(197)} ( a B , 1 ( 2 ) , a B , Adj S ( 2 ) , a B , R − ( 2 ) , a B , R + ( 2 ) ) cut = ( 27 N 2 , 135 N 2 , 135 6 − N , 135 6 + N ) , (197)
in agreement with the independent protected-sector calculation presented in [34 ] . An additional reduced contact polynomial P R ( s , t ) P_{\mathcal{R}}(s,t) P R ( s , t ) would shift these OPE coefficients by 2 3 P R ( 0 , 0 ) \frac{2}{3}P_{\mathcal{R}}(0,0) 3 2 P R ( 0 , 0 ) . The exact agreement therefore requires P R ( 0 , 0 ) = 0 P_{\mathcal{R}}(0,0)=0 P R ( 0 , 0 ) = 0 in every flavor channel.
To see what this constrains, a crossing-symmetric constant contact for S U ( N ≥ 4 ) SU(N\geq4) S U ( N ≥ 4 ) has the form
P 0 = a T δ + b T t r , P_0=aT_\delta+bT_{\mathrm{tr}}, P 0 = a T δ + b T tr ,
where
T δ I J K L = δ I J δ K L + δ I K δ J L + δ I L δ J K , T_\delta^{IJKL}=\delta^{IJ}\delta^{KL}+\delta^{IK}\delta^{JL}+\delta^{IL}\delta^{JK}, T δ I J K L = δ I J δ K L + δ I K δ J L + δ I L δ J K ,
T t r I J K L = Tr f ( t I t J t K t L ) , T_{\mathrm{tr}}^{IJKL}=\operatorname{Tr}_{f}\left(t^I t^J t^K t^L\right), T tr I J K L = Tr f ( t I t J t K t L ) ,
with averaged symmetrization and Tr f ( t I t J ) = δ I J \operatorname{Tr}_{f}(t^I t^J)=\delta^{IJ} Tr f ( t I t J ) = δ I J . Its projections onto R − \mathcal{R}_{-} R − and R + \mathcal{R}_{+} R + are
P 0 , R − = 2 a − b 3 , P 0 , R + = 2 a + b 3 . P_{0,\mathcal{R}_{-}}=2a-\frac{b}{3},\qquad P_{0,\mathcal{R}_{+}}=2a+\frac{b}{3}. P 0 , R − = 2 a − 3 b , P 0 , R + = 2 a + 3 b .
Requiring both to vanish fixes a = b = 0 a=b=0 a = b = 0 . For S U ( 2 ) SU(2) S U ( 2 ) and S U ( 3 ) SU(3) S U ( 3 ) , the single independent constant structure is likewise fixed by the protected data.
To summarize, within polynomial Mellin contact ambiguities, the flat-space limit of the microscopic four-dimensional theory excludes any nonzero local completion: the Ward lift turns a degree-q q q reduced polynomial into a term whose Penedones transform grows as L 2 q + 4 L^{2q+4} L 2 q + 4 , whereas the massless one-loop amplitude has only logarithmic infrared divergences. The exact protected OPE matching independently eliminates the constant contact sector directly at finite AdS radius. We therefore conclude, for the polynomial local completions considered here, that
M ~ t o t 1 -loop = M ~ c u t 1 -loop . (198) \widetilde{\mathcal{M}}_{\mathrm{tot}}^{1\text{-loop}}=\widetilde{\mathcal{M}}_{\mathrm{cut}}^{1\text{-loop}}. \tag*{(198)} M tot 1 -loop = M cut 1 -loop . (198)
An elliptic curve in the position-space representation
We derive an integral representation for the contribution of τ ( t ) \tau(t) τ ( t ) to the inverse Mellin transform of the Ward-lifted s t st s t block (142) . Throughout this appendix, U , V U,V U , V denote the conformal cross-ratios. We work with W ^ s t = π 4 D 2 B \widehat{\mathcal{W}}_{st}=\pi^4D_2B W s t = π 4 D 2 B , suppressing the overall color and coupling factors.
The pole representation (143) of τ \tau τ can be converted into an integral using the generating function
g ( x ) ≡ ∑ n = 1 ∞ x n n 2 ( 4 n ( 2 n n ) − 1 ) = 2 arcsin 2 x − Li 2 ( x ) . (199) g(x)\equiv\sum_{n=1}^{\infty}\frac{x^n}{n^2}\left(\frac{4^n}{\binom{2n}{n}}-1\right)=2\arcsin^2\sqrt{x}-\operatorname{Li}_2(x). \tag*{(199)} g ( x ) ≡ n = 1 ∑ ∞ n 2 x n ( ( n 2 n ) 4 n − 1 ) = 2 arcsin 2 x − Li 2 ( x ) . (199)
Indeed, integrating each term gives
τ ( t ) = − 1 18 ∫ 0 1 d x x x − t / 2 g ( x ) , Re t < 2 , (200) \tau(t)=-\frac{1}{18}\int_0^1\frac{\mathrm{d}x}{x}x^{-t/2}g(x),\qquad\operatorname{Re}t<2, \tag*{(200)} τ ( t ) = − 18 1 ∫ 0 1 x d x x − t /2 g ( x ) , Re t < 2 , (200)
with other values obtained by analytic continuation.
Let Φ ( U , V ) ≡ D ˉ 1111 ( U , V ) \Phi(U,V)\equiv\bar{D}_{1111}(U,V) Φ ( U , V ) ≡ D ˉ 1111 ( U , V ) denote the conformal scalar box. In our Mellin convention (56) , the inverse transform of the Gamma-function measure alone is U Φ ( U , V ) U\Phi(U,V) U Φ ( U , V ) . Inserting (D.2) therefore replaces V V V by V / x V/x V / x in this kernel and introduces an additional factor of 1 / x 1/x 1/ x , due to the power V t / 2 − 1 V^{t/2-1} V t /2 − 1 , while multiplication by s − 2 s-2 s − 2 acts as 2 U ∂ U − 2 2U\partial_U-2 2 U ∂ U − 2 . The contribution of ( s − 2 ) τ ( t ) (s-2)\tau(t) ( s − 2 ) τ ( t ) to the position-space transform of W ^ s t \widehat{\mathcal{W}}_{st} W s t is consequently
G τ s t ( U , V ) = − 1 18 ∫ 0 1 d x x g ( x ) ( 2 U ∂ U − 2 ) [ U x Φ ( U , V x ) ] = − U 2 9 ∂ U ∫ 0 1 d x x 2 g ( x ) Φ ( U , V x ) . (201) \begin{aligned}
\mathcal{G}_{\tau}^{st}(U,V)=-\frac{1}{18}\int_0^1\frac{\mathrm{d}x}{x}g(x)(2U\partial_U-2)\left[\frac{U}{x}\Phi\left(U,\frac{V}{x}\right)\right] \\
&=-\frac{U^2}{9}\partial_U\int_0^1\frac{\mathrm{d}x}{x^2}g(x)\Phi\left(U,\frac{V}{x}\right). \tag*{(201)}
\end{aligned} G τ s t ( U , V ) = − 18 1 ∫ 0 1 x d x g ( x ) ( 2 U ∂ U − 2 ) [ x U Φ ( U , x V ) ] = − 9 U 2 ∂ U ∫ 0 1 x 2 d x g ( x ) Φ ( U , x V ) . (201)
The algebraic structure follows directly from the scalar-box kernel. A useful representation is
Φ ( U , V ) = ∫ 0 ∞ d ξ log [ ( ξ + V ) ( 1 + ξ ) U ξ ] ξ 2 + ( 1 + V − U ) ξ + V . (202) \Phi(U,V)=\int_{0}^{\infty}d\xi\,\frac{\log\left[\frac{(\xi+V)(1+\xi)}{U\xi}\right]}{\xi^{2}+(1+V-U)\xi+V}. \tag*{(202)} Φ ( U , V ) = ∫ 0 ∞ d ξ ξ 2 + ( 1 + V − U ) ξ + V log [ U ξ ( ξ + V ) ( 1 + ξ ) ] . (202)
The quadratic denominator has discriminant
λ ( 1 , U , V ) = ( 1 − U − V ) 2 − 4 U V . (203) \lambda(1,U,V)=(1-U-V)^{2}-4UV. \tag*{(203)} λ ( 1 , U , V ) = ( 1 − U − V ) 2 − 4 U V . (203)
In (D.3), the substitution
x = r 2 1 + r 2 , 0 < r < ∞ , (204) x=\frac{r^{2}}{1+r^{2}},\qquad0<r<\infty, \tag*{(204)} x = 1 + r 2 r 2 , 0 < r < ∞ , (204)
gives arcsin x = arctan r \arcsin\sqrt{x}=\arctan r arcsin x = arctan r and
λ ( 1 , U , V x ) = P 4 ( r ) r 4 , P 4 ( r ) = λ ( 1 , U , V ) r 4 + 2 V ( V − 1 − U ) r 2 + V 2 . (205) \lambda\left(1,U,\frac{V}{x}\right)=\frac{P_{4}(r)}{r^{4}},\qquad P_{4}(r)=\lambda(1,U,V)r^{4}+2V(V-1-U)r^{2}+V^{2}. \tag*{(205)} λ ( 1 , U , x V ) = r 4 P 4 ( r ) , P 4 ( r ) = λ ( 1 , U , V ) r 4 + 2 V ( V − 1 − U ) r 2 + V 2 . (205)
The resulting algebraic kernel thus involves the curve y 2 = P 4 ( r ) y^{2}=P_{4}(r) y 2 = P 4 ( r ) . Its discriminant is
Discr P 4 = 4096 U 2 V 6 λ ( 1 , U , V ) , (206) \operatorname{Discr}P_{4}=4096U^{2}V^{6}\lambda(1,U,V), \tag*{(206)} Discr P 4 = 4096 U 2 V 6 λ ( 1 , U , V ) , (206)
which is nonzero at generic cross-ratios, so the smooth compactification of this curve has genus one.
There is a direct relation to the curve found in [95 ] . Denoting the monic quartic in eq. (4.2) of that reference by P 4 S t ( σ ; x , y , z ) P_{4}^{\mathrm{St}}(\sigma;x,y,z) P 4 St ( σ ; x , y , z ) , one finds
P 4 ( i σ ) λ ( 1 , U , V ) = P 4 S t ( σ ; V , U , 1 ) . (207) \frac{P_{4}(i\sigma)}{\lambda(1,U,V)}=P_{4}^{\mathrm{St}}(\sigma;V,U,1). \tag*{(207)} λ ( 1 , U , V ) P 4 ( iσ ) = P 4 St ( σ ; V , U , 1 ) . (207)
Thus the two representations involve the same complex elliptic curve, after a rescaling of the vertical coordinate. This identifies the elliptic geometry of the kernel in (D.3); it does not establish that an irreducible elliptic contribution survives in the complete crossing-symmetric correlator.