Introduction
Hard-exclusive Compton scattering provides a framework for studying the partonic structure of hadrons in quantum chromodynamics (QCD). In the generalized Bjorken regime, the amplitude separates into a perturbatively calculable hard coefficient and nonperturbative off-forward quark and gluon matrix elements, parameterized by generalized parton distributions (GPDs). GPDs correlate parton momentum with momentum transfer to the hadron; their moments give form factors, while at zero skewness their momentum-transfer dependence provides information about the transverse spatial distribution of partons[1 , 2 , 3 , 4 , 5 , 6 , 7 ] .
A growing direction in the GPD program is to use hard-exclusive Compton amplitudes not only to map hadron structure, but also to investigate how gluon fields shape some of QCD’s most fundamental phenomena. The chiral anomaly connects gluon topology to the unexpectedly large mass of the η ′ \eta^\prime η ′ through the pseudoscalar operator F ~ μ ν F μ ν \tilde F^{\mu\nu}F_{\mu\nu} F ~ μν F μν [8 , 9 , 10 , 11 , 12 , 13 ] . The trace anomaly connects the scale dependence of QCD to hadron masses through the scalar gluonic operator F μ ν F μ ν F^{\mu\nu}F_{\mu\nu} F μν F μν , and is central to understanding the origin of the proton mass[14 , 15 ] . Hard-exclusive Compton amplitudes therefore offer a way to access off-forward matrix elements of these operators, bringing anomaly physics into the study of hadron structure.
These gluonic operators also appear in amplitudes as residues of infrared poles. Such poles have long been studied in three-point functions. In the chiral channel, the massless axial-vector/vector/vector triangle produces a pole contribution to the off-forward matrix element of the flavor-singlet axial current J 5 μ = ∑ f q ˉ f γ μ γ 5 q f J^\mu_5=\sum_f \bar q_f\gamma^\mu\gamma_5q_f J 5 μ = ∑ f q ˉ f γ μ γ 5 q f , where the sum runs over quark flavors. In schematic form,
⟨ p 2 ∣ J 5 μ ∣ p 1 ⟩ p o l e ∝ ℓ μ t ⟨ p 2 ∣ i F α β F ~ α β ∣ p 1 ⟩ ; ℓ = p 2 − p 1 , t = ℓ 2 . (1) \langle p_2|J^\mu_5|p_1\rangle_{\mathrm{pole}} \propto\frac{\ell^\mu}{t}\langle p_2|iF^{\alpha\beta}\widetilde F_{\alpha\beta}|p_1\rangle;\qquad\ell=p_2-p_1,\qquad t=\ell^2. \tag*{(1)} ⟨ p 2 ∣ J 5 μ ∣ p 1 ⟩ pole ∝ t ℓ μ ⟨ p 2 ∣ i F α β F α β ∣ p 1 ⟩ ; ℓ = p 2 − p 1 , t = ℓ 2 . (1)
Here, 1 / t 1/t 1/ t is the infrared pole, whose residue carries the gluonic operator structure of the chiral anomaly, a connection discussed since the classic analyses of the axial triangle[8 , 9 , 16 ] .
An analogous pole occurs in the trace-anomaly channel. In the massless limit, the triangle contribution to the off-forward energy-momentum-tensor Θ μ ν \Theta^{\mu\nu} Θ μν matrix element has the schematic form
⟨ p 2 ∣ Θ μ ν ∣ p 1 ⟩ p o l e ∝ 1 t ⟨ p 2 ∣ F α β F α β ∣ p 1 ⟩ ; ℓ = p 2 − p 1 , t = ℓ 2 . (2) \langle p_2|\Theta^{\mu\nu}|p_1\rangle_{\mathrm{pole}} \propto\frac{1}{t}\langle p_2|F^{\alpha\beta}F_{\alpha\beta}|p_1\rangle;\qquad\ell=p_2-p_1,\qquad t=\ell^2. \tag*{(2)} ⟨ p 2 ∣ Θ μν ∣ p 1 ⟩ pole ∝ t 1 ⟨ p 2 ∣ F α β F α β ∣ p 1 ⟩ ; ℓ = p 2 − p 1 , t = ℓ 2 . (2)
Giannotti and Mottola first identified the corresponding trace-anomaly pole in QED [17 ] . Subsequent work studied its structure in other gauge theories [18 , 19 ] . More recent dispersive analyses have examined how the pole strength is shared between the pole and continuum as the kinematics and quark mass vary [20 ] .
At the level of scattering amplitudes, significant progress has been made. Tarasov and Venugopalan traced the chiral-anomaly contribution in polarized deep-inelastic scattering (DIS) to a triangle mechanism within the one-loop box amplitude, in both the Bjorken and Regge limits [12 , 13 ] . In deeply virtual Compton scattering (DVCS), Bhattacharya, Hatta, and Vogelsang identified chiral-anomaly-related and trace-anomaly-related poles in the polarized and unpolarized amplitudes, respectively, and provided a prescription for relating their residues to the corresponding gluonic operators [21 , 22 ] . Going beyond perturbation theory, they discussed the implications of these anomaly poles for the nonperturbative structure of GPDs [23 ] . Lending support to this interpretation, a recent chiral quark soliton model calculation finds that singlet pseudoscalar exchange gives a sizable contribution to the flavor-singlet polarized GPD E ~ \widetilde{E} E [24 , 25 ] . Complementary operator-level studies have examined gluon-state matrix elements, their relation to GPDs, and the role of infrared regulators [26 , 27 ] .
In this work, we address the complementary question of why and how the pole-producing region arises within the Compton loop integrals and how its contribution depends on the infrared prescription. We study this question in the gluon channel of double deeply virtual Compton scattering (DDVCS) using a systematic Landau analysis of the Passarino–Veltman coefficient functions in the one-loop box amplitude. The analysis identifies the singular region as the boundary of the box’s Feynman-parameter integration domain on which the parameter associated with the internal quark propagator joining the two photon vertices vanishes. In the Landau classification, this is a subleading singularity: the propagator associated with the vanishing parameter is inactive, while the remaining three propagators define a reduced diagram with triangle topology. The original integral remains a four-point function; the triangle describes the boundary configuration that controls its singular behavior. Our Landau analysis thus identifies the triangle boundary responsible for the pole in DDVCS. A related triangle mechanism was discussed by Tarasov and Venugopalan in the worldline approach to polarized DIS. We further characterize this singular configuration and its dependence on the infrared prescription in DDVCS.
We examine this boundary configuration using three infrared prescriptions: dimensional regularization (DR), nonzero momentum transfer (t ≠ 0 t\ne0 t = 0 ), and a finite quark mass (m ≠ 0 m\ne0 m = 0 ). At t = m = 0 t=m=0 t = m = 0 , the relevant Passarino–Veltman coefficient functions become singular as this boundary is approached. Introducing a cutoff on the vanishing Feynman parameter reveals a linear divergence in the inverse cutoff. DR assigns the associated contribution zero by analytic continuation, so this power divergence does not appear as a 1 / ϵ 1/\epsilon 1/ ϵ pole at d = 4 d=4 d = 4 . The analytically continued parameter integral instead has a pole at d = 6 d=6 d = 6 , where the divergence becomes logarithmic. This pole identifies the critical dimension of the same boundary contribution. Keeping t t t or m m m nonzero regulates this boundary region and exposes contributions proportional to 1 / t 1/t 1/ t or 1 / m 2 1/m^2 1/ m 2 , respectively. To our knowledge, the 1 / m 2 1/m^2 1/ m 2 behavior has not previously been reported for these amplitudes. The triangle boundary thus provides a common geometric origin for these behaviors, and its role remains robust across all three infrared prescriptions.
We also show that the pole residue depends on how the external gluon polarizations are treated. Contracting the box amplitude with a transverse projector can remove an anomaly-related contribution that survives when the external gluon polarization vectors are kept explicit. This sensitivity may help explain the absence of such contributions in earlier calculations. We also present complete one-loop results for polarized and unpolarized DDVCS with all three infrared prescriptions. After summing the diagrams, the corresponding terms 1 / ϵ I R 1/\epsilon_{\mathrm{IR}} 1/ ϵ IR , ln ( Q 2 / ( − t ) ) \ln(Q^2/(-t)) ln ( Q 2 / ( − t )) , and ln ( Q 2 / m 2 ) \ln(Q^2/m^2) ln ( Q 2 / m 2 ) multiply the same evolution kernel, providing a consistency check across the three calculations.
The paper is organized as follows. Section II introduces the Passarino–Veltman notation and the Landau equations. Section III identifies the triangle boundary and examines how infrared regulators and gluon-polarization projections affect its contribution. Section IV presents the complete one-loop results for polarized and unpolarized DDVCS. Section V summarizes our findings and outlines prospects for future work.
Preliminaries
Passarino–Veltman functions
We express the one-loop tensor integrals in terms of the conventional Passarino–Veltman coefficient functions A A A , B B B , C C C , and D D D ; see, for example, Refs. [28 , 29 ] . We work in
d = 4 − 2 ϵ . (3) d = 4 - 2\epsilon. \tag*{(3)} d = 4 − 2 ϵ . (3)
We write ϵ = ϵ I R < 0 \epsilon= \epsilon_{\mathrm{IR}} < 0 ϵ = ϵ IR < 0 when regulating infrared divergences and ϵ = ϵ U V > 0 \epsilon= \epsilon_{\mathrm{UV}} > 0 ϵ = ϵ UV > 0 when regulating ultraviolet divergences. In the Passarino–Veltman tensor decomposition, every pair of zero indices represents one metric tensor, whereas the indices 1 , 2 , 3 1,2,3 1 , 2 , 3 label the corresponding independent external momenta. We introduce the nonnegative integers r , n i = 0 , 1 , 2 , … r,n_i = 0,1,2,\ldots r , n i = 0 , 1 , 2 , … , where r r r counts the number of metric tensors and n i n_i n i counts the number of factors of the i i i th external momentum. (Appendix A illustrates the index conventions in detail using two-point functions as examples.) We then use the compact notation
A 0 2 r ≡ A 00 ⋯ 0 ⏟ 2 r , A_{0\,2r} \equiv A_{\underbrace{00\cdots0}_{2r}}, A 0 2 r ≡ A 2 r 00 ⋯ 0 ,
B 0 2 r 1 n 1 ≡ B 00 ⋯ 0 ⏟ 2 r 11 ⋯ 1 ⏟ n 1 , B_{0\,2r\,1\,n_1} \equiv B_{\underbrace{00\cdots0}_{2r}\underbrace{11\cdots1}_{n_1}}, B 0 2 r 1 n 1 ≡ B 2 r 00 ⋯ 0 n 1 11 ⋯ 1 ,
C 0 2 r 1 n 1 2 n 2 ≡ C 00 ⋯ 0 ⏟ 2 r 11 ⋯ 1 ⏟ n 1 22 ⋯ 2 ⏟ n 2 , C_{0\,2r\,1\,n_1\,2\,n_2} \equiv C_{\underbrace{00\cdots0}_{2r}\underbrace{11\cdots1}_{n_1}\underbrace{22\cdots2}_{n_2}}, C 0 2 r 1 n 1 2 n 2 ≡ C 2 r 00 ⋯ 0 n 1 11 ⋯ 1 n 2 22 ⋯ 2 ,
D 0 2 r 1 n 1 2 n 2 3 n 3 ≡ D 00 ⋯ 0 ⏟ 2 r 11 ⋯ 1 ⏟ n 1 22 ⋯ 2 ⏟ n 2 33 ⋯ 3 ⏟ n 3 . (4) D_{0\,2r\,1\,n_1\,2\,n_2\,3\,n_3} \equiv D_{\underbrace{00\cdots0}_{2r}\underbrace{11\cdots1}_{n_1}\underbrace{22\cdots2}_{n_2}\underbrace{33\cdots3}_{n_3}}. \tag*{(4)} D 0 2 r 1 n 1 2 n 2 3 n 3 ≡ D 2 r 00 ⋯ 0 n 1 11 ⋯ 1 n 2 22 ⋯ 2 n 3 33 ⋯ 3 . (4)
The corresponding tensor rank is
R = 2 r + ∑ i n i . (5) R = 2r + \sum_i n_i. \tag*{(5)} R = 2 r + i ∑ n i . (5)
In the Feynman-parameter representations below, the integers n i n_i n i also determine the powers of the corresponding Feynman parameters. The scalar functions are obtained by setting r = n i = 0 r=n_i=0 r = n i = 0 and are denoted by A 0 A_0 A 0 , B 0 B_0 B 0 , C 0 C_0 C 0 , and D 0 D_0 D 0 .
The general one-point coefficient is
A 0 2 r ( m 0 ) = μ 2 ϵ e γ E ϵ ( − 1 ) r + 1 2 r Γ ( − 1 − r + ϵ ) ( m 0 2 ) r ( 1 m 0 2 ) − 1 + ϵ . (6) A_{0\,2r}(m_0) = \mu^{2\epsilon}e^{\gamma_E\epsilon}\frac{(-1)^{r+1}}{2^r}\Gamma(-1-r+\epsilon)(m_0^2)^r\left(\frac{1}{m_0^2}\right)^{-1+\epsilon}. \tag*{(6)} A 0 2 r ( m 0 ) = μ 2 ϵ e γ E ϵ 2 r ( − 1 ) r + 1 Γ ( − 1 − r + ϵ ) ( m 0 2 ) r ( m 0 2 1 ) − 1 + ϵ . (6)
The symbol μ \mu μ denotes the dimensional regularization scale, and γ E \gamma_E γ E is the Euler–Mascheroni constant. The general two-point coefficient is
B 0 2 r 1 n 1 ( s ; m 0 , m 1 ) = μ 2 ϵ e γ E ϵ ( − 1 ) 2 + n 1 + r 2 r Γ ( ϵ − r ) ∫ 0 1 d x x n 1 [ F B ( x ) − i ϵ ] − ϵ + r , (7) B_{0\,2r\,1\,n_1}(s;m_0,m_1)
= \mu^{2\epsilon}e^{\gamma_E\epsilon}\frac{(-1)^{2+n_1+r}}{2^r}\Gamma(\epsilon-r)\int_0^1 dx\,x^{n_1}\left[F_B(x)-i\epsilon\right]^{-\epsilon+r}, \tag*{(7)} B 0 2 r 1 n 1 ( s ; m 0 , m 1 ) = μ 2 ϵ e γ E ϵ 2 r ( − 1 ) 2 + n 1 + r Γ ( ϵ − r ) ∫ 0 1 d x x n 1 [ F B ( x ) − i ϵ ] − ϵ + r , (7)
where
F B ( x ) = s x 2 + ( − s + m 1 2 − m 0 2 ) x + m 0 2 . (8) F_B(x) = sx^2 + \left(-s+m_1^2-m_0^2\right)x+m_0^2. \tag*{(8)} F B ( x ) = s x 2 + ( − s + m 1 2 − m 0 2 ) x + m 0 2 . (8)
Here and below, s s s , s i s_i s i , and s i j s_{ij} s ij denote external kinematic invariants. Their explicit expressions for the process considered here will be specified in the following sections. The general three-point coefficient is
C 0 2 r 1 n 1 2 n 2 ( s 1 , s 12 , s 2 ; m 0 , m 1 , m 2 ) = μ 2 ϵ e γ E ϵ ( − 1 ) 3 + n 1 + n 2 + r 2 r Γ ( 1 + ϵ − r ) ∫ 0 1 d x ∫ 0 1 − x d y x n 1 y n 2 [ F C ( x , y ) − i ϵ ] − 1 − ϵ + r , (9) C_{0\,2r\,1\,n_1\,2\,n_2}(s_1,s_{12},s_2;m_0,m_1,m_2)
= \mu^{2\epsilon}e^{\gamma_E\epsilon}\frac{(-1)^{3+n_1+n_2+r}}{2^r}\Gamma(1+\epsilon-r)\int_0^1 dx\int_0^{1-x}dy\,x^{n_1}y^{n_2}\left[F_C(x,y)-i\epsilon\right]^{-1-\epsilon+r}, \tag*{(9)} C 0 2 r 1 n 1 2 n 2 ( s 1 , s 12 , s 2 ; m 0 , m 1 , m 2 ) = μ 2 ϵ e γ E ϵ 2 r ( − 1 ) 3 + n 1 + n 2 + r Γ ( 1 + ϵ − r ) ∫ 0 1 d x ∫ 0 1 − x d y x n 1 y n 2 [ F C ( x , y ) − i ϵ ] − 1 − ϵ + r , (9)
where
F C ( x , y ) = s 1 x 2 + s 2 y 2 + ( s 1 + s 2 − s 12 ) x y + ( − s 1 + m 1 2 − m 0 2 ) x + ( − s 2 + m 2 2 − m 0 2 ) y + m 0 2 . (10) F_C(x,y) = s_1x^2+s_2y^2+(s_1+s_2-s_{12})xy+\left(-s_1+m_1^2-m_0^2\right)x+\left(-s_2+m_2^2-m_0^2\right)y+m_0^2. \tag*{(10)} F C ( x , y ) = s 1 x 2 + s 2 y 2 + ( s 1 + s 2 − s 12 ) x y + ( − s 1 + m 1 2 − m 0 2 ) x + ( − s 2 + m 2 2 − m 0 2 ) y + m 0 2 . (10)
The Feynman parameters x x x and y y y span the triangular domain
x ≥ 0 , y ≥ 0 , x + y ≤ 1. (11) x \ge0,\qquad y \ge0,\qquad x+y \le1. \tag*{(11)} x ≥ 0 , y ≥ 0 , x + y ≤ 1. (11)
The general four-point coefficient is
D 02 r 1 n 1 n 2 n 3 ( s 1 , s 2 , s 3 , s 4 ; s 12 , s 23 ; m 0 , m 1 , m 2 , m 3 ) = μ 2 ϵ e γ E ϵ ( − 1 ) 4 + n 1 + n 2 + n 3 + r 2 r Γ ( 2 + ϵ − r ) ∫ 0 1 d x ∫ 0 1 − x d y ∫ 0 1 − x − y d z x n 1 y n 2 z n 3 [ F D ( x , y , z ) − i ϵ ] − 2 − ϵ + r , (12) \mathcal{D}_{0 2 r_{1} n_{1} n_{2} n_{3}}(s_{1},s_{2},s_{3},s_{4};s_{12},s_{23};m_{0},m_{1},m_{2},m_{3})=\mu^{2\epsilon}e^{\gamma_{E}\epsilon}\frac{(-1)^{4+n_{1}+n_{2}+n_{3}+r}}{2^{r}}\Gamma(2+\epsilon-r)\int_{0}^{1}dx\int_{0}^{1-x}dy\int_{0}^{1-x-y}dz\,x^{n_{1}}y^{n_{2}}z^{n_{3}}\left[F_{D}(x,y,z)-i\epsilon\right]^{-2-\epsilon+r}, \tag*{(12)} D 02 r 1 n 1 n 2 n 3 ( s 1 , s 2 , s 3 , s 4 ; s 12 , s 23 ; m 0 , m 1 , m 2 , m 3 ) = μ 2 ϵ e γ E ϵ 2 r ( − 1 ) 4 + n 1 + n 2 + n 3 + r Γ ( 2 + ϵ − r ) ∫ 0 1 d x ∫ 0 1 − x d y ∫ 0 1 − x − y d z x n 1 y n 2 z n 3 [ F D ( x , y , z ) − i ϵ ] − 2 − ϵ + r , (12)
where
F D ( x , y , z ) = s 1 x 2 + s 12 y 2 + s 4 z 2 + ( s 1 + s 12 − s 2 ) x y + ( s 1 − s 23 + s 4 ) x z + ( s 12 − s 3 + s 4 ) y z + ( − s 1 + m 1 2 − m 0 2 ) x + ( − s 12 + m 2 2 − m 0 2 ) y + ( − s 4 + m 3 2 − m 0 2 ) z + m 0 2 . (13) \begin{aligned}
F_{D}(x,y,z)={}&s_{1}x^{2}+s_{12}y^{2}+s_{4}z^{2}+(s_{1}+s_{12}-s_{2})xy+(s_{1}-s_{23}+s_{4})xz+(s_{12}-s_{3}+s_{4})yz\\
&+(-s_{1}+m_{1}^{2}-m_{0}^{2})x+(-s_{12}+m_{2}^{2}-m_{0}^{2})y+(-s_{4}+m_{3}^{2}-m_{0}^{2})z+m_{0}^{2}. \tag*{(13)}
\end{aligned} F D ( x , y , z ) = s 1 x 2 + s 12 y 2 + s 4 z 2 + ( s 1 + s 12 − s 2 ) x y + ( s 1 − s 23 + s 4 ) x z + ( s 12 − s 3 + s 4 ) y z + ( − s 1 + m 1 2 − m 0 2 ) x + ( − s 12 + m 2 2 − m 0 2 ) y + ( − s 4 + m 3 2 − m 0 2 ) z + m 0 2 . (13)
The Feynman parameters x , y , z x,y,z x , y , z span the tetrahedral domain
x ≥ 0 , y ≥ 0 , z ≥ 0 , x + y + z ≤ 1. (14) x \ge0,\qquad y \ge0,\qquad z \ge0,\qquad x+y+z \le1. \tag*{(14)} x ≥ 0 , y ≥ 0 , z ≥ 0 , x + y + z ≤ 1. (14)
For the present analysis, the relation between the three-point and four-point functions is particularly important. On the boundary z = 0 z=0 z = 0 of the box parameter domain,
F D ( x , y , 0 ) = s 1 x 2 + s 12 y 2 + ( s 1 + s 12 − s 2 ) x y + ( − s 1 + m 1 2 − m 0 2 ) x + ( − s 12 + m 2 2 − m 0 2 ) y + m 0 2 . (15) F_{D}(x,y,0)=s_{1}x^{2}+s_{12}y^{2}+(s_{1}+s_{12}-s_{2})xy+(-s_{1}+m_{1}^{2}-m_{0}^{2})x+(-s_{12}+m_{2}^{2}-m_{0}^{2})y+m_{0}^{2}. \tag*{(15)} F D ( x , y , 0 ) = s 1 x 2 + s 12 y 2 + ( s 1 + s 12 − s 2 ) x y + ( − s 1 + m 1 2 − m 0 2 ) x + ( − s 12 + m 2 2 − m 0 2 ) y + m 0 2 . (15)
This is precisely the three-point (Symanzik) polynomial in Eq. (10) , with the triangle invariants identified as s 1 ( C ) = s 1 s_{1}^{(C)}=s_{1} s 1 ( C ) = s 1 , s 12 ( C ) = s 2 s_{12}^{(C)}=s_{2} s 12 ( C ) = s 2 , and s 2 ( C ) = s 12 s_{2}^{(C)}=s_{12} s 2 ( C ) = s 12 , while the masses m 0 , m 1 , m 2 m_{0},m_{1},m_{2} m 0 , m 1 , m 2 remain unchanged. Thus,
F D ( x , y , 0 ) = F C ( x , y ) ∣ ( s 1 ( C ) , s 12 ( C ) , s 2 ( C ) ) = ( s 1 , s 2 , s 12 ) . (16) F_{D}(x,y,0)=F_{C}(x,y)\big|_{(s_{1}^{(C)},s_{12}^{(C)},s_{2}^{(C)})=(s_{1},s_{2},s_{12})}. \tag*{(16)} F D ( x , y , 0 ) = F C ( x , y ) ( s 1 ( C ) , s 12 ( C ) , s 2 ( C ) ) = ( s 1 , s 2 , s 12 ) . (16)
In the argument ordering of Eq. (9) , this boundary is therefore associated with the triangle coefficient
C 02 r 1 n 1 n 2 ( s 1 , s 2 , s 12 ; m 0 , m 1 , m 2 ) . (17) C_{0 2 r_{1} n_{1} n_{2}}(s_{1},s_{2},s_{12};m_{0},m_{1},m_{2}). \tag*{(17)} C 02 r 1 n 1 n 2 ( s 1 , s 2 , s 12 ; m 0 , m 1 , m 2 ) . (17)
Thus, the z = 0 z=0 z = 0 boundary of the box is described by an embedded triangle subdiagram. This relation will be important for the anomaly-related pole studied below.
Landau equations
Feynman integrals are generally complicated functions of the external kinematic invariants, and their singularities can be difficult to identify by direct integration. The Landau equations provide necessary conditions for locating possible singularities directly from the propagator denominators [30 ] . They may be expressed either in momentum space or, equivalently, in Feynman-parameter space.
Consider a one-loop integral with N N N propagators, labeled by i = 0 , … , N − 1 i=0,\ldots,N-1 i = 0 , … , N − 1 , with denominators
D i ( k ) + i ϵ , D i ( k ) = ( k + κ i ) 2 − m i 2 , ϵ → 0 + , (18) \mathcal{D}_{i}(k)+i\epsilon,\qquad\mathcal{D}_{i}(k)=(k+\kappa_{i})^{2}-m_{i}^{2},\qquad\epsilon\to0^{+}, \tag*{(18)} D i ( k ) + i ϵ , D i ( k ) = ( k + κ i ) 2 − m i 2 , ϵ → 0 + , (18)
where k k k is the loop momentum and the κ i \kappa_{i} κ i are fixed combinations of external momenta. Introducing nonnegative Feynman parameters α i \alpha_{i} α i , the momentum-space Landau equations are
α i D i ( k ) = 0 , i = 0 , … , N − 1 , (19) \alpha_{i}\mathcal{D}_{i}(k)=0,\qquad i=0,\ldots,N-1, \tag*{(19)} α i D i ( k ) = 0 , i = 0 , … , N − 1 , (19)
∑ i = 0 N − 1 α i ( k + κ i ) μ = 0. (20) \sum_{i=0}^{N-1}\alpha_{i}(k+\kappa_{i})^{\mu}=0. \tag*{(20)} i = 0 ∑ N − 1 α i ( k + κ i ) μ = 0. (20)
If α i > 0 \alpha_{i}>0 α i > 0 , the corresponding propagator is active and must satisfy D i ( k ) = 0 \mathcal{D}_{i}(k)=0 D i ( k ) = 0 . If α i = 0 \alpha_{i}=0 α i = 0 , that propagator is inactive. The second equation is the stationarity condition with respect to the loop momentum and follows from ∂ k μ ∑ i α i D i ( k ) = 0 \partial_{k_{\mu}}\sum_{i}\alpha_{i}\mathcal{D}_{i}(k)=0 ∂ k μ ∑ i α i D i ( k ) = 0 . With the normalization
α i ≥ 0 , ∑ i = 0 N − 1 α i = 1 , (21) \alpha_i \ge0,\qquad\sum_{i=0}^{N-1}\alpha_i=1, \tag*{(21)} α i ≥ 0 , i = 0 ∑ N − 1 α i = 1 , (21)
the Feynman parameters span an ( N − 1 ) (N-1) ( N − 1 ) -dimensional simplex. After eliminating α 0 \alpha_0 α 0 , the Landau equations at an interior point take the equivalent parameter-space form
F = 0 , ∂ F ∂ α j = 0 , j = 1 , … , N − 1 , (22) \mathcal{F}=0,\qquad\frac{\partial\mathcal{F}}{\partial\alpha_j}=0,\qquad j=1,\ldots,N-1, \tag*{(22)} F = 0 , ∂ α j ∂ F = 0 , j = 1 , … , N − 1 , (22)
where the derivatives are taken with respect to the remaining independent parameters. Note that F \mathcal{F} F is the usual one-loop Symanzik polynomial,
F ( α ) = ∑ i = 0 N − 1 α i m i 2 − ∑ 0 ≤ i < j ≤ N − 1 α i α j ( κ i − κ j ) 2 , (23) \mathcal{F}(\alpha)=\sum_{i=0}^{N-1}\alpha_i m_i^2-\sum_{0\le i<j\le N-1}\alpha_i\alpha_j(\kappa_i-\kappa_j)^2, \tag*{(23)} F ( α ) = i = 0 ∑ N − 1 α i m i 2 − 0 ≤ i < j ≤ N − 1 ∑ α i α j ( κ i − κ j ) 2 , (23)
whose explicit two-, three-, and four-point forms are the polynomials F B F_B F B , F C F_C F C , and F D F_D F D introduced above. If one or more parameters vanish, the Landau conditions apply to F \mathcal{F} F restricted to the corresponding boundary face, with stationarity imposed only along that face. Such solutions are governed by lower-point subdiagrams. In particular, the z = 0 z=0 z = 0 boundary of the box is described by an embedded triangle, as shown in Eq. (16) .
The primary goal of this section is to explain how the 1 / t 1/t 1/ t poles reported in the gluon channel of polarized and unpolarized DDVCS arise from the one-loop integrals [21 , 22 ] . Throughout this section (and the remaining sections), we use “polarized” and “unpolarized” interchangeably with “antisymmetric” and “symmetric,” respectively, referring to the symmetry of the Compton amplitude under interchange of the photon Lorentz indices. The residues of these poles in the polarized and unpolarized amplitudes carry the gluonic operator structures associated with the chiral and trace anomalies, respectively. Despite these distinct residue structures, both poles arise from the same subleading Landau singularity of the one-loop box. At this singularity, the Feynman parameter associated with the quark propagator joining the two photon vertices vanishes. This places the singularity on a boundary face of the box’s Feynman-parameter domain, where the other three propagators form a triangle subdiagram. We examine this boundary region using three infrared prescriptions: dimensional regularization (DR), nonzero momentum transfer (t ≠ 0 t\ne0 t = 0 ), and a finite quark mass (m ≠ 0 m\ne0 m = 0 ). The comparison shows how each prescription treats the same singular region.
Below, we carry out the calculation for the s s s -channel box, shown as diagram (1) in Figure 1 . The same calculation and conclusions apply to the u u u -channel hourglass, shown as diagram (3) in the same figure. By contrast, the cat’s-eye topology shown as diagram (2) does not generate an anomaly-related pole. To expose the dependence on the momentum transfer, we first state the partonic kinematics at finite t t t . We define the average photon and parton momenta (see Refs. [21 , 22 ] )
Figure 1. The three gluon-channel diagrams relevant for DDVCS: the box (1), cat’s-eye (2), and hourglass (3) topologies.
q = q 1 + q 2 2 , p = p 1 + p 2 2 , (24) q=\frac{q_1+q_2}{2},\qquad p=\frac{p_1+p_2}{2}, \tag*{(24)} q = 2 q 1 + q 2 , p = 2 p 1 + p 2 , (24)
and parametrize the incoming and outgoing momenta as
q 1 = q + ℓ 2 , q 2 = q − ℓ 2 , p 1 = p − ℓ 2 , p 2 = p + ℓ 2 . (25) q_1=q+\frac{\ell}{2},\qquad q_2=q-\frac{\ell}{2},\qquad p_1=p-\frac{\ell}{2},\qquad p_2=p+\frac{\ell}{2}. \tag*{(25)} q 1 = q + 2 ℓ , q 2 = q − 2 ℓ , p 1 = p − 2 ℓ , p 2 = p + 2 ℓ . (25)
The momentum transfer is therefore
ℓ = q 1 − q 2 = p 2 − p 1 , t = ℓ 2 . (26) \ell=q_1-q_2=p_2-p_1,\qquad t=\ell^2. \tag*{(26)} ℓ = q 1 − q 2 = p 2 − p 1 , t = ℓ 2 . (26)
The incoming and outgoing partons are taken to be massless, p 1 2 = p 2 2 = 0 p_1^2=p_2^2=0 p 1 2 = p 2 2 = 0 , which implies
p 2 = − t 4 , p ⋅ ℓ = 0. (27) p^2=-\frac{t}{4},\qquad p\cdot\ell=0. \tag*{(27)} p 2 = − 4 t , p ⋅ ℓ = 0. (27)
Following the definitions used at the hadronic level in Ref. [22 ] , we introduce the corresponding partonic variables
Q 2 = − q 2 , x ^ = Q 2 2 p ⋅ q , ξ ^ = q 2 2 − q 1 2 4 p ⋅ q . (28) Q^2=-q^2,\qquad\hat{x}=\frac{Q^2}{2p\cdot q},\qquad\hat{\xi}=\frac{q_2^2-q_1^2}{4p\cdot q}. \tag*{(28)} Q 2 = − q 2 , x ^ = 2 p ⋅ q Q 2 , ξ ^ = 4 p ⋅ q q 2 2 − q 1 2 . (28)
The photon virtualities can then be written as
q 1 2 = − Q a 2 = − Q 2 x ^ + ξ ^ x ^ + t 4 , q 2 2 = − Q b 2 = − Q 2 x ^ − ξ ^ x ^ + t 4 . (29) q_1^2=-Q_a^2=-Q^2\frac{\hat{x}+\hat{\xi}}{\hat{x}}+\frac{t}{4},\qquad q_2^2=-Q_b^2=-Q^2\frac{\hat{x}-\hat{\xi}}{\hat{x}}+\frac{t}{4}. \tag*{(29)} q 1 2 = − Q a 2 = − Q 2 x ^ x ^ + ξ ^ + 4 t , q 2 2 = − Q b 2 = − Q 2 x ^ x ^ − ξ ^ + 4 t . (29)
For DDVCS, x ^ \hat{x} x ^ and ξ ^ \hat{\xi} ξ ^ are treated as independent partonic variables. The t = 0 t=0 t = 0 kinematics used in DR and finite-quark-mass (m ≠ 0 m\ne0 m = 0 ) calculations follow directly from the above expressions.
With the kinematics specified, we now turn to the partonic Compton amplitude and consider two treatments of the external gluon polarizations. First, we retain the polarization vectors explicitly and write
T μ ν = T μ ν α β ϵ α ( p 1 ) ϵ β ∗ ( p 2 ) . (30) \mathcal{T}^{\mu\nu}=\mathcal{T}^{\mu\nu\alpha\beta}\epsilon_\alpha(p_1)\epsilon_\beta^*(p_2). \tag*{(30)} T μν = T μν α β ϵ α ( p 1 ) ϵ β ∗ ( p 2 ) . (30)
Here, μ \mu μ and ν \nu ν are the Lorentz indices of the incoming and outgoing photons in Figure 1 , respectively. The vectors ϵ α ( p 1 ) \epsilon_\alpha(p_1) ϵ α ( p 1 ) and ϵ β ∗ ( p 2 ) \epsilon_\beta^*(p_2) ϵ β ∗ ( p 2 ) describe the polarizations of the incoming and outgoing gluons, respectively. Second, we contract the gluon Lorentz indices α \alpha α and β \beta β of the amplitude with the transverse projector g ⊥ α β g_{\perp\alpha\beta} g ⊥ α β (see Eq. (45) ), giving
T ⊥ μ ν = T μ ν α β g ⊥ α β . (31) \mathcal{T}_{\perp}^{\mu\nu}=\mathcal{T}^{\mu\nu\alpha\beta}g_{\perp\alpha\beta}. \tag*{(31)} T ⊥ μν = T μν α β g ⊥ α β . (31)
Below, we identify the anomaly-related terms with the gluon polarization vectors kept explicit and examine how the transverse projection affects their residues. Finally, we note that the diagrams are evaluated using the Mathematica package Package-X [31 ] .
Dimensional Regularization (DR)
Identifying the triangle singularity in the box
We begin with the t = m = 0 t=m=0 t = m = 0 calculation in DR. The relevant loop integrals have been evaluated previously [32 , 33 , 34 ] . Here we examine the analytic structure of their Passarino–Veltman coefficient functions to locate the singular region within the box. In the Feynman-parameter representation, this region lies near the boundary z = 0 z=0 z = 0 . As we will show, in four dimensions a lower cutoff δ \delta δ on z z z exposes a power divergence proportional to 1 / δ 1/\delta 1/ δ . Analytic continuation in DR conceals this endpoint divergence, which produces no 1 / ϵ 1/\epsilon 1/ ϵ pole in the expansion about d = 4 d=4 d = 4 . In the following subsections, we show that nonzero momentum transfer and a finite quark mass regulate the same boundary region, yielding the anomaly-related 1 / t 1/t 1/ t pole and the 1 / m 2 1/m^2 1/ m 2 behavior, respectively.
After tensor reduction, both the symmetric and antisymmetric parts of the Compton amplitude, T S μ ν \mathcal{T}^{\mu\nu}_S T S μν and T A μ ν \mathcal{T}^{\mu\nu}_A T A μν , contain four-point coefficient functions of the form
D I ( 0 , 0 , − Q b 2 , − Q a 2 ; 0 , s ; 0 , 0 , 0 , 0 ) , I ∈ I D . (32) D_I(0,0,-Q_b^2,-Q_a^2;0,s;0,0,0,0),\qquad I\in\mathcal{I}_D. \tag*{(32)} D I ( 0 , 0 , − Q b 2 , − Q a 2 ; 0 , s ; 0 , 0 , 0 , 0 ) , I ∈ I D . (32)
where I I I is the Passarino–Veltman coefficient multi-index and
I D = { 0 , 1 , 2 , 3 , 11 , 12 , 13 , 22 , 23 , 33 , \mathcal{I}_D=\left\{0,1,2,3,11,12,13,22,23,33,\right. I D = { 0 , 1 , 2 , 3 , 11 , 12 , 13 , 22 , 23 , 33 ,
112 , 113 , 122 , 123 , 133 , 223 , 233 , 333 , \left.112,113,122,123,133,223,233,333,\right. 112 , 113 , 122 , 123 , 133 , 223 , 233 , 333 ,
00 , 001 , 002 , 003 } . (33) \left.00,001,002,003\right\}. \tag*{(33)} 00 , 001 , 002 , 003 } . (33)
Although these coefficients differ in tensor rank and numerator structure, they share the same four-point denominator polynomial (see below) and therefore the same candidate Landau singularities. Their tensor indices determine the strength with which a singular region contributes, but not its location. At t = 0 t = 0 t = 0 , the four-point coefficient functions D I D_I D I share the denominator polynomial
F D ( x , y , z ) = z [ a ( x , y ) − Q a 2 z ] . (34) F_D(x,y,z) = z\left[a(x,y)-Q_a^2z\right]. \tag*{(34)} F D ( x , y , z ) = z [ a ( x , y ) − Q a 2 z ] . (34)
where
a ( x , y ) = Q a 2 − ( Q a 2 + s ) x + ( Q b 2 − Q a 2 ) y . (35) a(x,y) = Q_a^2-(Q_a^2+s)x+(Q_b^2-Q_a^2)y. \tag*{(35)} a ( x , y ) = Q a 2 − ( Q a 2 + s ) x + ( Q b 2 − Q a 2 ) y . (35)
For the momentum routing used here, z z z is the Feynman parameter associated with the internal quark propagator joining the two photon vertices. The overall factor of z z z in F D F_D F D implies that
F D ( x , y , 0 ) = 0 , ∂ F D ∂ x ∣ z = 0 = ∂ F D ∂ y ∣ z = 0 = 0 (36) F_D(x,y,0)=0,\qquad\left.\frac{\partial F_D}{\partial x}\right|_{z=0}=\left.\frac{\partial F_D}{\partial y}\right|_{z=0}=0 \tag*{(36)} F D ( x , y , 0 ) = 0 , ∂ x ∂ F D z = 0 = ∂ y ∂ F D z = 0 = 0 (36)
throughout the face z = 0 z = 0 z = 0 . These are the Landau conditions on that face, with stationarity imposed in the x x x and y y y directions. The derivative in the z z z direction is
∂ F D ∂ z ∣ z = 0 = a ( x , y ) . (37) \left.\frac{\partial F_D}{\partial z}\right|_{z=0}=a(x,y). \tag*{(37)} ∂ z ∂ F D z = 0 = a ( x , y ) . (37)
Since z z z is fixed to zero on the face, this derivative is unconstrained by the boundary Landau conditions. In the Landau classification, the solution at z = 0 z = 0 z = 0 is a subleading singularity of the box. The propagator associated with the vanishing parameter z z z is inactive, while the remaining three propagators form the triangle subdiagram shown in Figure 2 . The loop integral is still a four-point function. The triangle identifies the boundary configuration that controls its singular behavior and, as we show below, gives rise to the anomaly-related 1 / t 1/t 1/ t and 1 / m 2 1/m^2 1/ m 2 pole in DDVCS. A related triangle mechanism was discussed by Tarasov and Venugopalan in their worldline analysis of polarized DIS [12 , 13 ] .
Figure 2. Box-to-triangle reduction on the Landau boundary z = 0 z = 0 z = 0 . The red dashed line marks the quark propagator joining the two photon vertices, whose Feynman parameter z z z vanishes on this boundary. The remaining three propagators then form the triangle subdiagram responsible for the anomaly-related Landau singularity. The same boundary reduction applies to the u u u -channel hourglass diagram.
Mathematically, the box-to-triangle reduction follows by restricting the box denominator polynomial to the Landau boundary z = 0 z = 0 z = 0 . As shown in Eq. (16) , setting z = 0 z = 0 z = 0 removes the terms containing the Feynman parameter of the inactive propagator and leaves the triangle polynomial of the three active propagators,
F D ( x , y , 0 ) = F C ( x , y ) . (38) F_D(x,y,0)=F_C(x,y). \tag*{(38)} F D ( x , y , 0 ) = F C ( x , y ) . (38)
For the massless, zero-transfer kinematics of the box in Eq. (32) , all three external invariants and all three internal masses of the triangle vanish. The associated scalar three-point function is C 0 ( 0 , 0 , 0 ; 0 , 0 , 0 ) C_0(0,0,0;0,0,0) C 0 ( 0 , 0 , 0 ; 0 , 0 , 0 ) . The corresponding tensor three-point coefficients have the same kinematic arguments. By infrared power counting, the scalar triangle is power divergent in four dimensions. The calculation below shows how this infrared behavior is explicitly realized in the original box integral as z → 0 z \to0 z → 0 . We examine the four-point integrand as z → 0 z \to0 z → 0 and find that a lower cutoff δ \delta δ on z z z exposes a 1 / δ 1/\delta 1/ δ divergence in the scalar box. Analytic continuation in DR conceals this power divergence, which produces no 1 / ϵ 1/\epsilon 1/ ϵ pole in the expansion about d = 4 d = 4 d = 4 . At the critical dimension d = 6 d = 6 d = 6 , the same endpoint diverges logarithmically, giving a pole in the analytically continued endpoint integral.
We now examine how the Feynman-parameter integrands of the four-point coefficient functions D I D_I D I behave as z → 0 z \to0 z → 0 . In d = 4 − 2 ϵ d = 4 - 2\epsilon d = 4 − 2 ϵ , a general coefficient has the form
D 0 2 r 1 r 1 2 n 2 3 n 3 ∝ ∫ d x d y d z x n 1 y n 2 z n 3 [ F D ( x , y , z ) − i ϵ ] − 2 − ϵ + r . (39) D_{0^{2r_1r_1 2n_2 3n_3}} \propto\int dx\,dy\,dz\,x^{n_1}y^{n_2}z^{n_3}\left[F_D(x,y,z)-i\epsilon\right]^{-2-\epsilon+r}. \tag*{(39)} D 0 2 r 1 r 1 2 n 2 3 n 3 ∝ ∫ d x d y d z x n 1 y n 2 z n 3 [ F D ( x , y , z ) − i ϵ ] − 2 − ϵ + r . (39)
At a generic point of the z = 0 z = 0 z = 0 boundary, where a ( x , y ) ≠ 0 a(x,y) \ne0 a ( x , y ) = 0 , Eq. (34) gives
F D ( x , y , z ) = a ( x , y ) z + O ( z 2 ) . (40) F_D(x,y,z) = a(x,y)z + O(z^2). \tag*{(40)} F D ( x , y , z ) = a ( x , y ) z + O ( z 2 ) . (40)
Thus, for power counting normal to the boundary, the integrand behaves as
z n 3 [ F D − i ϵ ] − 2 − ϵ + r ∼ [ a ( x , y ) ] − 2 − ϵ + r z n 3 − 2 − ϵ + r . (41) z^{n_3}\left[F_D - i\epsilon\right]^{-2-\epsilon+r} \sim\left[a(x,y)\right]^{-2-\epsilon+r}z^{n_3-2-\epsilon+r}. \tag*{(41)} z n 3 [ F D − i ϵ ] − 2 − ϵ + r ∼ [ a ( x , y ) ] − 2 − ϵ + r z n 3 − 2 − ϵ + r . (41)
Here the causal prescription is understood; the displayed scaling describes the endpoint power away from the zero of a ( x , y ) a(x,y) a ( x , y ) .
For the scalar coefficient, r = n 3 = 0 r = n_3 = 0 r = n 3 = 0 , the generic z = 0 z = 0 z = 0 boundary gives the endpoint behavior
∫ 0 1 d z z − 2 − ϵ . (42) \int_0^1 \mathrm{d}z\,z^{-2-\epsilon}. \tag*{(42)} ∫ 0 1 d z z − 2 − ϵ . (42)
At d = 4 d = 4 d = 4 , or ϵ = 0 \epsilon= 0 ϵ = 0 , this endpoint is not integrable in the ordinary sense. Introducing a dimensionless lower cutoff δ \delta δ on the Feynman parameter makes the degree of divergence explicit:
∫ δ 1 d z z 2 = 1 δ − 1. (43) \int_\delta^1 \frac{\mathrm{d}z}{z^2} = \frac{1}{\delta} - 1. \tag*{(43)} ∫ δ 1 z 2 d z = δ 1 − 1. (43)
Thus, in a cutoff description of the parameter endpoint, this contribution diverges linearly in 1 / δ 1/\delta 1/ δ .
In DR, on the other hand, we evaluate the model endpoint integral in its region of convergence, Re ϵ < − 1 \operatorname{Re}\epsilon< -1 Re ϵ < − 1 , where
∫ 0 1 d z z − 2 − ϵ = − 1 1 + ϵ . (44) \int_0^1 \mathrm{d}z\,z^{-2-\epsilon} = -\frac{1}{1+\epsilon}. \tag*{(44)} ∫ 0 1 d z z − 2 − ϵ = − 1 + ϵ 1 . (44)
Analytic continuation to ϵ = 0 \epsilon= 0 ϵ = 0 gives a finite value, even though the original four-dimensional endpoint integral diverges as 1 / δ 1/\delta 1/ δ when the lower cutoff δ \delta δ tends to zero. DR thus conceals the four-dimensional power divergence, which gives no 1 / ϵ 1/\epsilon 1/ ϵ pole in the expansion about d = 4 d = 4 d = 4 . The continued endpoint integral has a pole at ϵ = − 1 \epsilon= -1 ϵ = − 1 , which singles out d = 6 d = 6 d = 6 as the critical dimension for this endpoint. At d = 6 d = 6 d = 6 , the integrand scales as 1 / z 1/z 1/ z , and the endpoint integral diverges logarithmically. The pole at d = 6 d = 6 d = 6 therefore marks the transition from power-divergent behavior below six dimensions to an integrable endpoint above six dimensions. As we show below, nonzero momentum transfer and a finite quark mass regulate the same boundary region, yielding 1 / t 1/t 1/ t and 1 / m 2 1/m^2 1/ m 2 terms, respectively. After summing the relevant diagrams and projecting onto the appropriate tensor structures, the residues of these terms carry the gluonic operator structures associated with F μ ν F μ ν F^{\mu\nu}F_{\mu\nu} F μν F μν and F μ ν F ~ μ ν F^{\mu\nu}\widetilde{F}_{\mu\nu} F μν F μν . Note that at a generic point on the boundary, additional powers of z z z in the numerator or metric contributions with r > 0 r > 0 r > 0 soften the endpoint behavior, as shown in Eq. (41) .
Effect of the transverse gluon projector
The calculation of the unpolarized DVCS amplitude in Ref. [34 ] uses the following transverse polarization sum for the external gluons:
g ⊥ α β = g α β − 1 Q 2 ( q α + Q 2 p ⋅ q p α ) ( q β + Q 2 p ⋅ q p β ) + q α q β Q 2 . (45) g_{\perp}^{\alpha\beta} = g^{\alpha\beta} - \frac{1}{Q^2}\left(q^\alpha+ \frac{Q^2}{p \cdot q}p^\alpha\right)\left(q^\beta+ \frac{Q^2}{p \cdot q}p^\beta\right) + \frac{q^\alpha q^\beta}{Q^2}. \tag*{(45)} g ⊥ α β = g α β − Q 2 1 ( q α + p ⋅ q Q 2 p α ) ( q β + p ⋅ q Q 2 p β ) + Q 2 q α q β . (45)
We loosely refer to the amplitude obtained by contracting the two gluon Lorentz indices with g ⊥ α β g_{\perp}^{\alpha\beta} g ⊥ α β in Eq. (31) as the “projected amplitude”. We compare it with the amplitude in which the polarization product ϵ α ( p 1 ) ϵ β ∗ ( p 2 ) \epsilon_\alpha(p_1)\epsilon_\beta^*(p_2) ϵ α ( p 1 ) ϵ β ∗ ( p 2 ) is kept explicit, as in Eq. (30) . As we show below, contracting with g ⊥ α β g_{\perp}^{\alpha\beta} g ⊥ α β changes the numerator and can modify the residue associated with the Landau singularity identified by the propagator denominators.
To illustrate this effect, consider the box diagram in Figure 1 . We use the loop momentum routing of Ref. [34 ] : the left and right vertical lines carry k + ξ p k + \xi p k + ξ p and k − ξ p k - \xi p k − ξ p , respectively, and the upper and lower horizontal lines carry k + q k + q k + q and k − p k - p k − p , respectively. This routing follows from the lightlike momentum convention of Ref. [34 ] ,
ℓ μ ⟶ − 2 ξ p μ , (46) \ell^\mu\longrightarrow-2\xi p^\mu, \tag*{(46)} ℓ μ ⟶ − 2 ξ p μ , (46)
so ℓ 2 = 0 \ell^2 = 0 ℓ 2 = 0 , and
k + ξ p = k − ℓ 2 , k − ξ p = k + ℓ 2 . (47) k + \xi p = k - \frac{\ell}{2}, \qquad k - \xi p = k + \frac{\ell}{2}. \tag*{(47)} k + ξ p = k − 2 ℓ , k − ξ p = k + 2 ℓ . (47)
Following the notation of Ref. [34 ] , we denote the left and right vertical propagator denominators by D − D_{-} D − and D + D_{+} D + , respectively, and the upper horizontal denominator by D q D_q D q . The lower horizontal propagator denominator is D p D_p D p . Thus, the four denominators read
D ± = ( k ± ℓ 2 ) 2 , D q = ( k + q ) 2 , D p = ( k − p ) 2 . (48) D_{\pm}=\left(k\pm\frac{\ell}{2}\right)^2,\qquad D_q=(k+q)^2,\qquad D_p=(k-p)^2. \tag*{(48)} D ± = ( k ± 2 ℓ ) 2 , D q = ( k + q ) 2 , D p = ( k − p ) 2 . (48)
The transverse projection gives a loop numerator of the form
N = ( k ⋅ q ) [ A k 2 + B ( k ⋅ p ) ] , (49) N=(k\cdot q)\left[Ak^2+B(k\cdot p)\right], \tag*{(49)} N = ( k ⋅ q ) [ A k 2 + B ( k ⋅ p ) ] , (49)
where A A A and B B B are kinematic coefficients. Thus the projected box contribution is the loop integral
I T ≡ ∫ k ( k ⋅ q ) [ A k 2 + B ( k ⋅ p ) ] D + D − D q D p = I T 1 + I T 2 , (50) I_T\equiv\int_k\frac{(k\cdot q)\left[Ak^2+B(k\cdot p)\right]}{D_+D_-D_qD_p}=I_{T1}+I_{T2}, \tag*{(50)} I T ≡ ∫ k D + D − D q D p ( k ⋅ q ) [ A k 2 + B ( k ⋅ p ) ] = I T 1 + I T 2 , (50)
where
I T 1 = A ∫ k ( k ⋅ q ) k 2 D + D − D q D p , I T 2 = B ∫ k ( k ⋅ q ) ( k ⋅ p ) D + D − D q D p . (51) I_{T1}=A\int_k\frac{(k\cdot q)k^2}{D_+D_-D_qD_p},\qquad I_{T2}=B\int_k\frac{(k\cdot q)(k\cdot p)}{D_+D_-D_qD_p}. \tag*{(51)} I T 1 = A ∫ k D + D − D q D p ( k ⋅ q ) k 2 , I T 2 = B ∫ k D + D − D q D p ( k ⋅ q ) ( k ⋅ p ) . (51)
Here ∫ k \int_k ∫ k denotes the dimensionally regulated loop integration; the causal + i ε +i\varepsilon + i ε prescription in each propagator is understood.
The scalar products in the numerator can be rewritten in terms of the denominators:
k ⋅ q = 1 2 D q − 1 4 D + − 1 4 D − − q 2 2 , k 2 = 1 2 ( D + + D − ) , k ⋅ p = 1 4 ( D + + D − ) − 1 2 D p . (52) \begin{aligned}
k\cdot q&=\frac{1}{2}D_q-\frac{1}{4}D_+-\frac{1}{4}D_--\frac{q^2}{2},\\
k^2&=\frac{1}{2}(D_++D_-),\\
k\cdot p&=\frac{1}{4}(D_++D_-)-\frac{1}{2}D_p. \tag*{(52)}
\end{aligned} k ⋅ q k 2 k ⋅ p = 2 1 D q − 4 1 D + − 4 1 D − − 2 q 2 , = 2 1 ( D + + D − ) , = 4 1 ( D + + D − ) − 2 1 D p . (52)
Inserting these identities into the two terms in Eq. (51) yields
I T 1 = A ∫ k [ 1 4 D − D p + 1 4 D + D p − 1 4 D q D p − D + 8 D − D q D p − D − 8 D + D q D p − q 2 4 D − D q D p − q 2 4 D + D q D p ] , (53) I_{T1}=A\int_k\left[\frac{1}{4D_-D_p}+\frac{1}{4D_+D_p}-\frac{1}{4D_qD_p}-\frac{D_+}{8D_-D_qD_p}-\frac{D_-}{8D_+D_qD_p}-\frac{q^2}{4D_-D_qD_p}-\frac{q^2}{4D_+D_qD_p}\right], \tag*{(53)} I T 1 = A ∫ k [ 4 D − D p 1 + 4 D + D p 1 − 4 D q D p 1 − 8 D − D q D p D + − 8 D + D q D p D − − 4 D − D q D p q 2 − 4 D + D q D p q 2 ] , (53)
I T 2 = B ∫ k [ 1 8 D − D p + 1 8 D + D p − 1 4 D + D − − 1 8 D q D p + 1 8 D − D q + 1 8 D + D q − D + 16 D − D q D p − D − 16 D + D q D p − q 2 8 D − D q D p − q 2 8 D + D q D p + q 2 4 D + D − D q ] . (54) I_{T2}=B\int_k\left[\frac{1}{8D_-D_p}+\frac{1}{8D_+D_p}-\frac{1}{4D_+D_-}-\frac{1}{8D_qD_p}+\frac{1}{8D_-D_q}+\frac{1}{8D_+D_q}-\frac{D_+}{16D_-D_qD_p}-\frac{D_-}{16D_+D_qD_p}-\frac{q^2}{8D_-D_qD_p}-\frac{q^2}{8D_+D_qD_p}+\frac{q^2}{4D_+D_-D_q}\right]. \tag*{(54)} I T 2 = B ∫ k [ 8 D − D p 1 + 8 D + D p 1 − 4 D + D − 1 − 8 D q D p 1 + 8 D − D q 1 + 8 D + D q 1 − 16 D − D q D p D + − 16 D + D q D p D − − 8 D − D q D p q 2 − 8 D + D q D p q 2 + 4 D + D − D q q 2 ] . (54)
The numerator factors cancel one or more propagators, reducing the projected loop integral to a linear combination of bubble (first line of the above equations) and triangle (second line of the above equations) integrals. The triangle integrals in this reduction do not generate the power infrared divergence associated with the z = 0 z=0 z = 0 boundary of the original box, where the quark propagator joining the two photon vertices becomes inactive. See Figure 3 for a schematic illustration of this. So, in essence, the g ⊥ α β g_{\perp}^{\alpha\beta} g ⊥ α β projection changes the numerator and hence the residue with which this boundary contributes; a suppressed or vanishing residue in the projected amplitude does not mean that the triangle boundary is absent. The bubble and triangle integrals in Eqs. (53) and (54) must accordingly be understood as the projected combination, rather than as evidence that the underlying Landau surface has disappeared.
Figure 3. Schematic illustration of the effect of a transverse gluon-polarization projector (g ⊥ α β g_{\perp}^{\alpha\beta} g ⊥ α β or ϵ ⊥ α β \epsilon_{\perp}^{\alpha\beta} ϵ ⊥ α β ) on the box numerator. In the projected integrand, numerator factors cancel propagator denominators, allowing the box contribution to be expressed as a sum of bubble and triangle integrals. The resulting combination is free of the anomaly-related pole.
Finally, we find the same pattern when we apply the analogous transverse projection to the polarized DVCS amplitude. In this case, the transverse projector is
ϵ ⊥ α β ∝ ϵ α β ρ σ p ρ q σ . (55) \epsilon_{\perp}^{\alpha\beta} \propto\epsilon^{\alpha\beta\rho\sigma}p_{\rho}q_{\sigma}. \tag*{(55)} ϵ ⊥ α β ∝ ϵ α β ρ σ p ρ q σ . (55)
We refrain from showing here the explicit reduction of the projected loop integrals to bubble and triangle integrals, as it follows the same pattern as in the unpolarized case.
The calculations in this section establish two results that, to our knowledge, have not previously been made explicit. First, we identify the z = 0 z=0 z = 0 boundary of the box’s Feynman-parameter domain as the source of a power infrared divergence in the relevant Passarino–Veltman coefficients. A lower cutoff on z z z makes this divergence explicit in four dimensions. In DR, analytic continuation conceals the power divergence, which produces no 1 / ϵ 1/\epsilon 1/ ϵ pole in the expansion about d = 4 d=4 d = 4 . Second, we show that contracting the box amplitude with a transverse gluon projector can suppress the associated pole residue through its effect on the numerator. The propagator denominators still satisfy the Landau conditions on the same boundary. This sensitivity to the polarization projection may help explain why the contribution was absent in earlier calculations. In the following subsections, we show that nonzero momentum transfer and a finite quark mass regulate this same boundary region, yielding the anomaly-related 1 / t 1/t 1/ t pole and the 1 / m 2 1/m^2 1/ m 2 behavior, respectively. The triangle mechanism is thus robust across all three infrared prescriptions: the same boundary controls the singular behavior in each case.
Nonzero momentum transfer (t=0)
We first regulate the z = 0 z=0 z = 0 boundary by keeping the momentum transfer nonzero. The tensor-reduced amplitude contains the four-point coefficient functions
D I ( 0 , 0 , − Q b 2 , − Q a 2 ; t , s ; 0 , 0 , 0 , 0 ) , I ∈ I D , (56) D_I(0,0,-Q_b^2,-Q_a^2;t,s;0,0,0,0), \qquad I \in\mathcal{I}_D, \tag*{(56)} D I ( 0 , 0 , − Q b 2 , − Q a 2 ; t , s ; 0 , 0 , 0 , 0 ) , I ∈ I D , (56)
with the coefficient-index set I D \mathcal{I}_D I D defined in Eq. (33) . For this family, the denominator polynomial is
F D ( x , y , z ) = a ( x , y ) z t + b ( x , y ) t + c z 2 + d ( x , y ) t z , (57) F_D(x,y,z)=a(x,y)zt+b(x,y)t+cz^2+d(x,y)tz, \tag*{(57)} F D ( x , y , z ) = a ( x , y ) z t + b ( x , y ) t + c z 2 + d ( x , y ) t z , (57)
where a ( x , y ) a(x,y) a ( x , y ) is defined in Eq. (35) , and
b ( x , y ) = − y ( 1 − x − y ) , c = − Q a 2 , d ( x , y ) = y . (58) b(x,y)=-y(1-x-y), \qquad c=-Q_a^2, \qquad d(x,y)=y. \tag*{(58)} b ( x , y ) = − y ( 1 − x − y ) , c = − Q a 2 , d ( x , y ) = y . (58)
At t = 0 t=0 t = 0 , this reduces to
F D ( x , y , z ) = z [ a ( x , y ) − Q a 2 z ] , (59) F_D(x,y,z)=z\left[a(x,y)-Q_a^2z\right], \tag*{(59)} F D ( x , y , z ) = z [ a ( x , y ) − Q a 2 z ] , (59)
which vanishes along the full z = 0 z=0 z = 0 face. This is the boundary associated with the reduced triangle shown in Figure 2 .
To analyze the contribution from this boundary, choose a fixed, small z 0 z_0 z 0 , independent of t t t , and restrict z z z to 0 ≤ z ≤ z 0 0\leq z\leq z_0 0 ≤ z ≤ z 0 . The value z 0 z_0 z 0 only marks the upper end of the local boundary region; it is not a physical scale or a lower cutoff. At generic fixed x , y x,y x , y , with a ( x , y ) , b ( x , y ) ≠ 0 a(x,y),b(x,y)\ne0 a ( x , y ) , b ( x , y ) = 0 2 , the terms a ( x , y ) z a(x,y)z a ( x , y ) z and b ( x , y ) t b(x,y)t b ( x , y ) t become comparable at
z ∼ z t , z t ≡ ∣ b ( x , y ) t a ( x , y ) ∣ . (60) z\sim z_t, \qquad z_t \equiv\left|\frac{b(x,y)t}{a(x,y)}\right|. \tag*{(60)} z ∼ z t , z t ≡ a ( x , y ) b ( x , y ) t . (60)
The scale z t z_t z t divides the boundary region into two ranges in which different terms dominate. The original integral continues to run from z = 0 z=0 z = 0 ; we split it at z t z_t z t only to estimate the contribution on each side of the crossover. For r = 0 r=0 r = 0 , the scalar coefficient has a local integrand proportional to F D − 2 F_D^{-2} F D − 2 . Keeping the two competing terms gives
I t ≡ ∫ 0 z 0 d z [ a ( x , y ) z + b ( x , y ) t ] 2 = 1 a ( x , y ) b ( x , y ) t − 1 a ( x , y ) [ a ( x , y ) z 0 + b ( x , y ) t ] . (61) I_t \equiv\int_0^{z_0} \frac{\mathrm{d}z}{[a(x,y)z+b(x,y)t]^2}
= \frac{1}{a(x,y)b(x,y)t}-\frac{1}{a(x,y)[a(x,y)z_0+b(x,y)t]}. \tag*{(61)} I t ≡ ∫ 0 z 0 [ a ( x , y ) z + b ( x , y ) t ] 2 d z = a ( x , y ) b ( x , y ) t 1 − a ( x , y ) [ a ( x , y ) z 0 + b ( x , y ) t ] 1 . (61)
For generic a ( x , y ) b ( x , y ) ≠ 0 a(x,y)b(x,y)\ne0 a ( x , y ) b ( x , y ) = 0 , the second term remains finite as t → 0 t\to0 t → 0 , whereas the first is proportional to 1 / t 1/t 1/ t . The two sides of the crossover both contribute at this inverse-power order. For 0 ≤ z ≲ z t 0\le z\lesssim z_t 0 ≤ z ≲ z t , the term b ( x , y ) t b(x,y)t b ( x , y ) t dominates, so
∫ 0 z t d z [ a ( x , y ) z + b ( x , y ) t ] 2 ∼ z t [ b ( x , y ) t ] 2 ∼ 1 ∣ a ( x , y ) b ( x , y ) ∣ ∣ t ∣ . (62) \int_0^{z_t}\frac{\mathrm{d}z}{[a(x,y)z+b(x,y)t]^2}\sim\frac{z_t}{[b(x,y)t]^2}\sim\frac{1}{|a(x,y)b(x,y)||t|}. \tag*{(62)} ∫ 0 z t [ a ( x , y ) z + b ( x , y ) t ] 2 d z ∼ [ b ( x , y ) t ] 2 z t ∼ ∣ a ( x , y ) b ( x , y ) ∣∣ t ∣ 1 . (62)
For z t ≲ z ≤ z 0 z_t\lesssim z\le z_0 z t ≲ z ≤ z 0 , the term a ( x , y ) z a(x,y)z a ( x , y ) z dominates, giving
∫ z t z 0 d z [ a ( x , y ) z + b ( x , y ) t ] 2 ∼ 1 a ( x , y ) 2 ( 1 z t − 1 z 0 ) ∼ 1 ∣ a ( x , y ) b ( x , y ) ∣ ∣ t ∣ . (63) \int_{z_t}^{z_0}\frac{\mathrm{d}z}{[a(x,y)z+b(x,y)t]^2}\sim\frac{1}{a(x,y)^2}\left(\frac{1}{z_t}-\frac{1}{z_0}\right)\sim\frac{1}{|a(x,y)b(x,y)||t|}. \tag*{(63)} ∫ z t z 0 [ a ( x , y ) z + b ( x , y ) t ] 2 d z ∼ a ( x , y ) 2 1 ( z t 1 − z 0 1 ) ∼ ∣ a ( x , y ) b ( x , y ) ∣∣ t ∣ 1 . (63)
The second estimate is controlled by its lower endpoint, z ∼ z t z\sim z_t z ∼ z t . Thus, the 1 / t 1/t 1/ t behavior is generated by the neighborhood of the crossover and receives contributions from both z ≲ z t z\lesssim z_t z ≲ z t and z ≳ z t z\gtrsim z_t z ≳ z t .
The same r = 0 r=0 r = 0 coefficient also contains a logarithm, which comes from a subleading term in the local expansion. Define
G t = a ( x , y ) z + b ( x , y ) t , H t = c z 2 + d ( x , y ) t z , F D = G t + H t . (64) G_t=a(x,y)z+b(x,y)t,\qquad H_t=cz^2+d(x,y)tz,\qquad F_D=G_t+H_t. \tag*{(64)} G t = a ( x , y ) z + b ( x , y ) t , H t = c z 2 + d ( x , y ) t z , F D = G t + H t . (64)
Where ∣ H t ∣ ≪ ∣ G t ∣ |H_t|\ll|G_t| ∣ H t ∣ ≪ ∣ G t ∣ , the scalar integrand becomes
1 F D 2 = 1 G t 2 − 2 [ c z 2 + d ( x , y ) t z ] G t 3 + ⋯ . (65) \frac{1}{F_D^2}=\frac{1}{G_t^2}-\frac{2[cz^2+d(x,y)tz]}{G_t^3}+\cdots. \tag*{(65)} F D 2 1 = G t 2 1 − G t 3 2 [ c z 2 + d ( x , y ) t z ] + ⋯ . (65)
For z ≲ z t z\lesssim z_t z ≲ z t , G t ≃ b ( x , y ) t G_t\simeq b(x,y)t G t ≃ b ( x , y ) t . The integrals of both correction terms over this range remain finite as t → 0 t\to0 t → 0 ; this range does not generate the logarithm. For z ≳ z t z\gtrsim z_t z ≳ z t , G t ≃ a ( x , y ) z G_t\simeq a(x,y)z G t ≃ a ( x , y ) z , and the c z 2 cz^2 c z 2 correction behaves as 1 / z 1/z 1/ z . It therefore gives
− 2 c ∫ z t z 0 d z z 2 [ a ( x , y ) z + b ( x , y ) t ] 3 ∼ − 2 c a ( x , y ) 3 ln ( z 0 z t ) . (66) -2c\int_{z_t}^{z_0}\mathrm{d}z\,\frac{z^2}{[a(x,y)z+b(x,y)t]^3}\sim-\frac{2c}{a(x,y)^3}\ln\left(\frac{z_0}{z_t}\right). \tag*{(66)} − 2 c ∫ z t z 0 d z [ a ( x , y ) z + b ( x , y ) t ] 3 z 2 ∼ − a ( x , y ) 3 2 c ln ( z t z 0 ) . (66)
Since z t ∝ ∣ t ∣ z_t\propto|t| z t ∝ ∣ t ∣ at fixed x , y x,y x , y , this contains a logarithm of ∣ t ∣ |t| ∣ t ∣ . The causal branch is understood. The term proportional to d ( x , y ) t z d(x,y)tz d ( x , y ) t z behaves as t / z 2 t/z^2 t / z 2 above the crossover, and its integral is of order t / z t ∼ t 0 t/z_t\sim t^0 t / z t ∼ t 0 ; it does not produce an unsuppressed ln ∣ t ∣ \ln|t| ln ∣ t ∣ term. The pole and logarithm therefore probe the same triangle boundary but arise from different terms in the local expansion.
For r = 1 r=1 r = 1 , the local integrand instead contains F D − 1 F_D^{-1} F D − 1 . The leading boundary form already yields a logarithm:
∫ 0 z 0 d z a ( x , y ) z + b ( x , y ) t = 1 a ( x , y ) ln [ a ( x , y ) z 0 + b ( x , y ) t b ( x , y ) t ] . (67) \int_0^{z_0}\frac{\mathrm{d}z}{a(x,y)z+b(x,y)t}=\frac{1}{a(x,y)}\ln\left[\frac{a(x,y)z_0+b(x,y)t}{b(x,y)t}\right]. \tag*{(67)} ∫ 0 z 0 a ( x , y ) z + b ( x , y ) t d z = a ( x , y ) 1 ln [ b ( x , y ) t a ( x , y ) z 0 + b ( x , y ) t ] . (67)
Its nonanalytic dependence as t → 0 t\to0 t → 0 is logarithmic, with the causal prescription understood.
Recall that at t = 0 t=0 t = 0 , the box endpoint is power divergent in four dimensions. Analytic continuation in DR conceals this divergence, which gives no 1 / ϵ 1/\epsilon 1/ ϵ pole in the expansion about d = 4 d=4 d = 4 . At the critical dimension d = 6 d=6 d = 6 , the same endpoint is logarithmically divergent, and the analytically continued endpoint integral has a pole. Keeping t ≠ 0 t\ne0 t = 0 regulates the same boundary region and makes the 1 / t 1/t 1/ t behavior of the four-point coefficient functions explicit as t → 0 t\to0 t → 0 . After the relevant diagrams are combined and the appropriate tensor structures are projected, the residues are proportional to matrix elements of the gluonic operators F μ ν F μ ν F^{\mu\nu}F_{\mu\nu} F μν F μν and F μ ν F ~ μ ν F^{\mu\nu}\widetilde{F}_{\mu\nu} F μν F μν . Thus, the boundary analysis identifies the geometry that generates the anomaly-related pole residues found in earlier work.
The Compton amplitude contains 22 distinct four-point coefficient functions from the family in Eq. (33) . We focus on three representative coefficients to illustrate how their nonanalytic dependence on t t t is governed by the powers of z z z in the numerator and of F D F_D F D in the denominator of the parameter representation. Define
Δ t = ( Q a 2 + s ) ( Q b 2 + s ) + s t , λ t = λ ( − Q a 2 , − Q b 2 , t ) , (68) \Delta_t=(Q_a^2+s)(Q_b^2+s)+st,\qquad\lambda_t=\lambda(-Q_a^2,-Q_b^2,t), \tag*{(68)} Δ t = ( Q a 2 + s ) ( Q b 2 + s ) + s t , λ t = λ ( − Q a 2 , − Q b 2 , t ) , (68)
where the Källén function
λ ( x , y , z ) = x 2 + y 2 + z 2 − 2 x y − 2 x z − 2 y z . (69) \lambda(x,y,z)=x^2+y^2+z^2-2xy-2xz-2yz. \tag*{(69)} λ ( x , y , z ) = x 2 + y 2 + z 2 − 2 x y − 2 x z − 2 y z . (69)
We begin with the scalar coefficient, r = 0 r=0 r = 0 and n 3 = 0 n_3=0 n 3 = 0 :
D 0 ( 0 , 0 , − Q b 2 , − Q a 2 ; t , s ; 0 , 0 , 0 , 0 ) , (70) D_0(0,0,-Q_b^2,-Q_a^2;t,s;0,0,0,0), \tag*{(70)} D 0 ( 0 , 0 , − Q b 2 , − Q a 2 ; t , s ; 0 , 0 , 0 , 0 ) , (70)
which has the expansion
D 0 = 1 ϵ 2 s t − π 2 12 s t − 2 s t Li 2 ( Q a 2 + s s ) − 2 s t Li 2 ( Q b 2 + s s ) − 1 s t ln 2 ( t s ) − 1 ϵ s t ln ( μ 2 Q a 2 ) − 1 2 s t ln 2 ( μ 2 Q a 2 ) − 1 2 s t ln 2 ( μ 2 Q b 2 ) + 1 s t ln 2 ( − μ 2 s ) − 1 s t ln ( μ 2 Q b 2 ) ln ( − μ 2 t ) − 1 s t ln 2 ( − μ 2 t ) + 3 2 s t ln 2 ( − μ 2 t ) + 1 s t ln ( μ 2 Q a 2 ) [ ln ( μ 2 Q b 2 ) − ln ( − μ 2 t ) ] + 1 ϵ s t [ − ln ( μ 2 Q b 2 ) + 2 ln ( − μ 2 s ) + ln ( − μ 2 t ) ] + O ( ϵ ) . (71) \begin{aligned}
D_0={}&\frac{1}{\epsilon^2st}-\frac{\pi^2}{12st}-\frac{2}{st}\operatorname{Li}_2\left(\frac{Q_a^2+s}{s}\right)-\frac{2}{st}\operatorname{Li}_2\left(\frac{Q_b^2+s}{s}\right)\\
&-\frac{1}{st}\ln^2\left(\frac{t}{s}\right)-\frac{1}{\epsilon st}\ln\left(\frac{\mu^2}{Q_a^2}\right)-\frac{1}{2st}\ln^2\left(\frac{\mu^2}{Q_a^2}\right)-\frac{1}{2st}\ln^2\left(\frac{\mu^2}{Q_b^2}\right)\\
&+\frac{1}{st}\ln^2\left(-\frac{\mu^2}{s}\right)-\frac{1}{st}\ln\left(\frac{\mu^2}{Q_b^2}\right)\ln\left(-\frac{\mu^2}{t}\right)-\frac{1}{st}\ln^2\left(-\frac{\mu^2}{t}\right)+\frac{3}{2st}\ln^2\left(-\frac{\mu^2}{t}\right)\\
&+\frac{1}{st}\ln\left(\frac{\mu^2}{Q_a^2}\right)\left[\ln\left(\frac{\mu^2}{Q_b^2}\right)-\ln\left(-\frac{\mu^2}{t}\right)\right]+\frac{1}{\epsilon st}\left[-\ln\left(\frac{\mu^2}{Q_b^2}\right)+2\ln\left(-\frac{\mu^2}{s}\right)+\ln\left(-\frac{\mu^2}{t}\right)\right]+\mathcal{O}(\epsilon). \tag*{(71)}
\end{aligned} D 0 = ϵ 2 s t 1 − 12 s t π 2 − s t 2 Li 2 ( s Q a 2 + s ) − s t 2 Li 2 ( s Q b 2 + s ) − s t 1 ln 2 ( s t ) − ϵs t 1 ln ( Q a 2 μ 2 ) − 2 s t 1 ln 2 ( Q a 2 μ 2 ) − 2 s t 1 ln 2 ( Q b 2 μ 2 ) + s t 1 ln 2 ( − s μ 2 ) − s t 1 ln ( Q b 2 μ 2 ) ln ( − t μ 2 ) − s t 1 ln 2 ( − t μ 2 ) + 2 s t 3 ln 2 ( − t μ 2 ) + s t 1 ln ( Q a 2 μ 2 ) [ ln ( Q b 2 μ 2 ) − ln ( − t μ 2 ) ] + ϵs t 1 [ − ln ( Q b 2 μ 2 ) + 2 ln ( − s μ 2 ) + ln ( − t μ 2 ) ] + O ( ϵ ) . (71)
The origin of 1 / t 1/t 1/ t is the z = 0 z=0 z = 0 boundary identified above, where the scalar parameter integrand has no numerator factor that vanishes with z z z . The integrand therefore retains its strongest singular behavior at the z = 0 z=0 z = 0 boundary. For comparison, consider the r = 0 r=0 r = 0 coefficient with n 3 = 1 n_3=1 n 3 = 1 , denoted D 3 D_3 D 3 . Its expansion is
D 3 = Li 2 ( Q a 2 + s s ) + Li 2 ( Q b 2 + s s ) Δ t + 1 2 Δ t ln 2 ( t s ) + 1 2 Δ t ln ( μ 2 Q b 2 ) ln ( − μ 2 t ) − 1 2 Δ t ln 2 ( − μ 2 t ) + ln ( μ 2 Q a 2 ) [ − 1 2 Δ t ln ( μ 2 Q b 2 ) + 1 2 Δ t ln ( − μ 2 t ) ] − 1 2 Δ t + ( Q a 2 + Q b 2 + 2 s + t ) B ( Q a 2 , Q b 2 , t ) , (72) \begin{aligned}
D_3={}&\frac{\operatorname{Li}_2\left(\frac{Q_a^2+s}{s}\right)+\operatorname{Li}_2\left(\frac{Q_b^2+s}{s}\right)}{\Delta_t}+\frac{1}{2\Delta_t}\ln^2\left(\frac{t}{s}\right)+\frac{1}{2\Delta_t}\ln\left(\frac{\mu^2}{Q_b^2}\right)\ln\left(-\frac{\mu^2}{t}\right)-\frac{1}{2\Delta_t}\ln^2\left(-\frac{\mu^2}{t}\right)\\
&+\ln\left(\frac{\mu^2}{Q_a^2}\right)\left[-\frac{1}{2\Delta_t}\ln\left(\frac{\mu^2}{Q_b^2}\right)+\frac{1}{2\Delta_t}\ln\left(-\frac{\mu^2}{t}\right)\right]-\frac{1}{2\Delta_t}+(Q_a^2+Q_b^2+2s+t)\mathcal{B}(Q_a^2,Q_b^2,t), \tag*{(72)}
\end{aligned} D 3 = Δ t Li 2 ( s Q a 2 + s ) + Li 2 ( s Q b 2 + s ) + 2 Δ t 1 ln 2 ( s t ) + 2 Δ t 1 ln ( Q b 2 μ 2 ) ln ( − t μ 2 ) − 2 Δ t 1 ln 2 ( − t μ 2 ) + ln ( Q a 2 μ 2 ) [ − 2 Δ t 1 ln ( Q b 2 μ 2 ) + 2 Δ t 1 ln ( − t μ 2 ) ] − 2 Δ t 1 + ( Q a 2 + Q b 2 + 2 s + t ) B ( Q a 2 , Q b 2 , t ) , (72)
where the dilogarithm combination is
B ( Q a 2 , Q b 2 , t ) ≡ 1 λ t { − Li 2 ( − Q a 2 − Q b 2 − t − λ t − Q a 2 − Q b 2 − t + λ t ) + Li 2 ( − Q a 2 − Q b 2 − t + λ t − Q a 2 − Q b 2 − t − λ t ) − Li 2 ( Q a 2 − Q b 2 + t − λ t Q a 2 − Q b 2 + t + λ t ) + Li 2 ( Q a 2 − Q b 2 + t + λ t − Q a 2 + Q b 2 − t + λ t ) − Li 2 ( − Q a 2 + Q b 2 + t − λ t − Q a 2 + Q b 2 + t + λ t ) + Li 2 ( − Q a 2 + Q b 2 + t + λ t − Q a 2 + Q b 2 + t − λ t ) } . (73) \begin{aligned}
\mathcal{B}(Q_a^2,Q_b^2,t)\equiv\frac{1}{\sqrt{\lambda_t}}\Bigg\{&
-\operatorname{Li}_2\left(\frac{-Q_a^2-Q_b^2-t-\sqrt{\lambda_t}}{-Q_a^2-Q_b^2-t+\sqrt{\lambda_t}}\right)
+\operatorname{Li}_2\left(\frac{-Q_a^2-Q_b^2-t+\sqrt{\lambda_t}}{-Q_a^2-Q_b^2-t-\sqrt{\lambda_t}}\right)\\
&-\operatorname{Li}_2\left(\frac{Q_a^2-Q_b^2+t-\sqrt{\lambda_t}}{Q_a^2-Q_b^2+t+\sqrt{\lambda_t}}\right)
+\operatorname{Li}_2\left(\frac{Q_a^2-Q_b^2+t+\sqrt{\lambda_t}}{-Q_a^2+Q_b^2-t+\sqrt{\lambda_t}}\right)\\
&-\operatorname{Li}_2\left(\frac{-Q_a^2+Q_b^2+t-\sqrt{\lambda_t}}{-Q_a^2+Q_b^2+t+\sqrt{\lambda_t}}\right)
+\operatorname{Li}_2\left(\frac{-Q_a^2+Q_b^2+t+\sqrt{\lambda_t}}{-Q_a^2+Q_b^2+t-\sqrt{\lambda_t}}\right)\Bigg\}. \tag*{(73)}
\end{aligned} B ( Q a 2 , Q b 2 , t ) ≡ λ t 1 { − Li 2 ( − Q a 2 − Q b 2 − t + λ t − Q a 2 − Q b 2 − t − λ t ) + Li 2 ( − Q a 2 − Q b 2 − t − λ t − Q a 2 − Q b 2 − t + λ t ) − Li 2 ( Q a 2 − Q b 2 + t + λ t Q a 2 − Q b 2 + t − λ t ) + Li 2 ( − Q a 2 + Q b 2 − t + λ t Q a 2 − Q b 2 + t + λ t ) − Li 2 ( − Q a 2 + Q b 2 + t + λ t − Q a 2 + Q b 2 + t − λ t ) + Li 2 ( − Q a 2 + Q b 2 + t − λ t − Q a 2 + Q b 2 + t + λ t ) } . (73)
This coefficient contains logarithmic dependence on t t t , but not the same 1 / t 1/t 1/ t enhancement as D 0 D_0 D 0 . This is consistent with Eq. (41) : for n 3 = 1 n_3=1 n 3 = 1 , the numerator supplies a factor of z z z , which vanishes on the z = 0 z=0 z = 0 boundary and softens the endpoint behavior.
Finally, the r = 1 r=1 r = 1 , n 3 = 0 n_3=0 n 3 = 0 example, denoted D 00 D_{00} D 00 , has the expansion
D 00 = s 2 Δ t Li 2 ( Q a 2 + s s ) + s 2 Δ t Li 2 ( Q b 2 + s s ) + s 4 Δ t ln 2 ( t s ) + s 4 Δ t ln ( μ 2 Q a 2 ) ln ( − μ 2 t ) − s 4 Δ t ln 2 ( − μ 2 t ) + ln ( μ 2 Q b 2 ) [ − s 4 Δ t ln ( μ 2 Q a 2 ) + s 4 Δ t ln ( − μ 2 t ) ] + 2 Q a 2 Q b 2 + Q a 2 s + Q b 2 s + s t 4 Δ t + B ( Q a 2 , Q b 2 , t ) . (74) \begin{aligned}
D_{00}={}&\frac{s}{2\Delta_t}\operatorname{Li}_2\left(\frac{Q_a^2+s}{s}\right)+\frac{s}{2\Delta_t}\operatorname{Li}_2\left(\frac{Q_b^2+s}{s}\right)+\frac{s}{4\Delta_t}\ln^2\left(\frac{t}{s}\right)+\frac{s}{4\Delta_t}\ln\left(\frac{\mu^2}{Q_a^2}\right)\ln\left(-\frac{\mu^2}{t}\right)-\frac{s}{4\Delta_t}\ln^2\left(-\frac{\mu^2}{t}\right)\\
&+\ln\left(\frac{\mu^2}{Q_b^2}\right)\left[-\frac{s}{4\Delta_t}\ln\left(\frac{\mu^2}{Q_a^2}\right)+\frac{s}{4\Delta_t}\ln\left(-\frac{\mu^2}{t}\right)\right]+\frac{2Q_a^2Q_b^2+Q_a^2s+Q_b^2s+st}{4\Delta_t}+\mathcal{B}(Q_a^2,Q_b^2,t). \tag*{(74)}
\end{aligned} D 00 = 2 Δ t s Li 2 ( s Q a 2 + s ) + 2 Δ t s Li 2 ( s Q b 2 + s ) + 4 Δ t s ln 2 ( s t ) + 4 Δ t s ln ( Q a 2 μ 2 ) ln ( − t μ 2 ) − 4 Δ t s ln 2 ( − t μ 2 ) + ln ( Q b 2 μ 2 ) [ − 4 Δ t s ln ( Q a 2 μ 2 ) + 4 Δ t s ln ( − t μ 2 ) ] + 4 Δ t 2 Q a 2 Q b 2 + Q a 2 s + Q b 2 s + s t + B ( Q a 2 , Q b 2 , t ) . (74)
Here the reduced power of the denominator in the parameter representation gives logarithmic dependence on t t t already at leading order near the boundary.
To summarize, for r = 1 r=1 r = 1 , the parameter representation contains one fewer power of F D F_D F D in the denominator. Near the z = 0 z=0 z = 0 boundary, its leading local behavior is therefore proportional to ( a ( x , y ) z + b ( x , y ) t ) − 1 (a(x,y)z+b(x,y)t)^{-1} ( a ( x , y ) z + b ( x , y ) t ) − 1 . Integration over z z z produces logarithmic dependence on t t t . By contrast, when r = 0 r=0 r = 0 and n 3 = 1 n_3=1 n 3 = 1 , the numerator factor z z z suppresses the 1 / t 1/t 1/ t enhancement found for n 3 = 0 n_3=0 n 3 = 0 , although logarithmic dependence on t t t may remain. These three examples illustrate the distinct boundary behaviors within the four-point coefficient family. The remaining coefficients can be classified by the same parameter-space power counting.3
As discussed above, the propagator denominators determine the location of the Landau surface, whereas the residue depends on the numerator and on how the gluon polarizations are treated. The DVCS analyses of Refs. [21 , 22 ] kept the gluon polarization vectors explicit and identified the tensor structures
( ϵ 1 ⋅ ℓ ) ( ϵ 2 ∗ ⋅ ℓ ) − 1 2 ( ϵ 1 ⋅ ϵ 2 ∗ ) ℓ 2 ∝ ⟨ p 2 ∣ F μ ν F μ ν ∣ p 1 ⟩ , 2 i ϵ ϵ 1 ϵ 2 ℓ p ∝ ⟨ p 2 ∣ F μ ν F ~ μ ν ∣ p 1 ⟩ . (75) (\epsilon_1 \cdot\ell)(\epsilon_2^* \cdot\ell) - \frac{1}{2}(\epsilon_1 \cdot\epsilon_2^*)\ell^2 \propto\langle p_2|F^{\mu\nu}F_{\mu\nu}|p_1\rangle,
\\
2i\epsilon^{\epsilon_1\epsilon_2\ell p} \propto\langle p_2|F^{\mu\nu}\widetilde{F}_{\mu\nu}|p_1\rangle. \tag*{(75)} ( ϵ 1 ⋅ ℓ ) ( ϵ 2 ∗ ⋅ ℓ ) − 2 1 ( ϵ 1 ⋅ ϵ 2 ∗ ) ℓ 2 ∝ ⟨ p 2 ∣ F μν F μν ∣ p 1 ⟩ , 2 i ϵ ϵ 1 ϵ 2 ℓ p ∝ ⟨ p 2 ∣ F μν F μν ∣ p 1 ⟩ . (75)
Here ϵ ϵ 1 ϵ 2 ℓ p \epsilon^{\epsilon_1\epsilon_2\ell p} ϵ ϵ 1 ϵ 2 ℓ p denotes the Levi-Civita contraction of the four indicated vectors, with ϵ 1 = ϵ ( p 1 ) \epsilon_1 = \epsilon(p_1) ϵ 1 = ϵ ( p 1 ) and ϵ 2 ∗ = ϵ ∗ ( p 2 ) \epsilon_2^* = \epsilon^*(p_2) ϵ 2 ∗ = ϵ ∗ ( p 2 ) . The first structure is associated with the trace anomaly and the second with the chiral anomaly. In the loop amplitude, the singular Passarino–Veltman coefficients identified above generate 1 / t 1/t 1/ t terms with these scalar and pseudoscalar structures.
To see why keeping the gluon polarizations explicit matters, consider the numerator tensor k α k β k^\alpha k^\beta k α k β (where k k k is the loop momentum). Its contraction with the polarization vectors gives
ϵ α ( p 1 ) ϵ β ∗ ( p 2 ) k α k β = ( ϵ ( p 1 ) ⋅ k ) ( ϵ ∗ ( p 2 ) ⋅ k ) . (76) \epsilon_\alpha(p_1)\epsilon_\beta^*(p_2)k^\alpha k^\beta= \bigl(\epsilon(p_1)\cdot k\bigr)\bigl(\epsilon^*(p_2)\cdot k\bigr). \tag*{(76)} ϵ α ( p 1 ) ϵ β ∗ ( p 2 ) k α k β = ( ϵ ( p 1 ) ⋅ k ) ( ϵ ∗ ( p 2 ) ⋅ k ) . (76)
Tensor reduction can generate a term proportional to ℓ α ℓ β \ell^\alpha\ell^\beta ℓ α ℓ β . Thus, its contraction gives ( ϵ 1 ⋅ ℓ ) ( ϵ 2 ∗ ⋅ ℓ ) (\epsilon_1\cdot\ell)(\epsilon_2^*\cdot\ell) ( ϵ 1 ⋅ ℓ ) ( ϵ 2 ∗ ⋅ ℓ ) , which enters the scalar structure in Eq. (75) .
We now examine how the transverse projectors act on this numerator tensor. The symmetric projector g ⊥ α β g_\perp^{\alpha\beta} g ⊥ α β gives a scalar that is invariant under rotations in the transverse plane:
g ⊥ α β k α k β = k ⊥ 2 . (77) g_{\perp\alpha\beta}k^\alpha k^\beta= k_\perp^2. \tag*{(77)} g ⊥ α β k α k β = k ⊥ 2 . (77)
Using Eq. (52) , we can express the resulting numerator in terms of propagator denominators. The cancellations reorganize the box contribution into bubble and triangle integrals and remaining box terms, none of which generates the anomaly-related 1 / t 1/t 1/ t pole. Likewise, the antisymmetric projector annihilates this particular numerator tensor by symmetry:
ϵ ⊥ α β k α k β = 0. (78) \epsilon_{\perp\alpha\beta}k^\alpha k^\beta= 0. \tag*{(78)} ϵ ⊥ α β k α k β = 0. (78)
The other numerator terms can contribute to the projected antisymmetric amplitude and give no 1 / t 1/t 1/ t pole in the calculation considered here.
Thus, keeping the gluon polarization vectors explicit allows us to track the tensor structures that carry the anomaly-related residues (see Eq. (75) ). A transverse projector can suppress these residues by changing the numerator and reducing the box contribution to a linear combination of bubble, triangle, and box integrals that yield no anomaly-related pole. We emphasize that the suppression concerns the residue in the projected amplitude. The original box denominators still satisfy the Landau conditions on the same boundary.
Nonzero quark mass (m ≠ 0 m \ne0 m = 0 )
We next regulate the same boundary by assigning a common mass m m m to the four internal propagators and setting t = 0 t = 0 t = 0 . The resulting mass-regulated expressions also describe the leading behavior in the hierarchy ∣ t ∣ ≪ m 2 ≪ Q 2 |t| \ll m^2 \ll Q^2 ∣ t ∣ ≪ m 2 ≪ Q 2 . The case with both t t t and m m m nonzero will be discussed briefly later in this subsection; a systematic analysis of the joint limit is deferred to a separate work.
The tensor-reduced amplitude then contains
D I ( 0 , 0 , − Q b 2 , − Q a 2 ; 0 , s ; m 2 , m 2 , m 2 , m 2 ) , I ∈ I D . (79) D_I(0,0,-Q_b^2,-Q_a^2;0,s;m^2,m^2,m^2,m^2), \qquad I \in\mathcal{I}_D. \tag*{(79)} D I ( 0 , 0 , − Q b 2 , − Q a 2 ; 0 , s ; m 2 , m 2 , m 2 , m 2 ) , I ∈ I D . (79)
Because the Feynman parameters satisfy ∑ i α i = 1 \sum_i \alpha_i = 1 ∑ i α i = 1 , the mass contribution to the denominator polynomial is m 2 ∑ i α i = m 2 m^2\sum_i\alpha_i = m^2 m 2 ∑ i α i = m 2 . Hence
F D ( m ) ( x , y , z ) = m 2 + a ( x , y ) z + c z 2 , (80) F_D^{(m)}(x,y,z) = m^2 + a(x,y)z + cz^2, \tag*{(80)} F D ( m ) ( x , y , z ) = m 2 + a ( x , y ) z + c z 2 , (80)
where a ( x , y ) a(x,y) a ( x , y ) is defined in Eq. (35) , and c c c is given in Eq. (58) . In the massless limit this polynomial vanishes on the z = 0 z=0 z = 0 face associated with the reduced triangle in Figure 2 . A nonzero quark mass lifts the boundary value of the polynomial from zero to m 2 m^2 m 2 , giving F D ( m ) ( x , y , 0 ) = m 2 F_D^{(m)}(x,y,0)=m^2 F D ( m ) ( x , y , 0 ) = m 2 . At generic fixed x , y x,y x , y , the mass term competes with the term linear in z z z at
z ∼ z m , z m ≡ m 2 ∣ a ( x , y ) ∣ . (81) z \sim z_m,\qquad z_m \equiv\frac{m^2}{\lvert a(x,y)\rvert}. \tag*{(81)} z ∼ z m , z m ≡ ∣ a ( x , y )∣ m 2 . (81)
As in the finite-t t t case, choose a fixed, small upper limit z 0 z_0 z 0 for the boundary region. This does not change the lower endpoint of the original integral, which remains z = 0 z=0 z = 0 . The split at z m z_m z m only separates the ranges in which m 2 m^2 m 2 and a ( x , y ) z a(x,y)z a ( x , y ) z dominate.
For r = 0 r=0 r = 0 , the scalar coefficient has a local integrand proportional to ( F D ( m ) ) − 2 \left(F_D^{(m)}\right)^{-2} ( F D ( m ) ) − 2 . Keeping the leading terms gives
I m ≡ ∫ 0 z 0 d z [ a ( x , y ) z + m 2 ] 2 = 1 a ( x , y ) m 2 − 1 a ( x , y ) [ a ( x , y ) z 0 + m 2 ] . (82) I_m \equiv\int_0^{z_0}\frac{\mathrm{d}z}{\left[a(x,y)z+m^2\right]^2}
= \frac{1}{a(x,y)m^2}-\frac{1}{a(x,y)\left[a(x,y)z_0+m^2\right]}. \tag*{(82)} I m ≡ ∫ 0 z 0 [ a ( x , y ) z + m 2 ] 2 d z = a ( x , y ) m 2 1 − a ( x , y ) [ a ( x , y ) z 0 + m 2 ] 1 . (82)
For generic a ( x , y ) ≠ 0 a(x,y)\ne0 a ( x , y ) = 0 , the second term remains finite as m 2 → 0 m^2\to0 m 2 → 0 , while the first is proportional to 1 / m 2 1/m^2 1/ m 2 . Both sides of z m z_m z m contribute at this inverse-mass order. Below the crossover,
∫ 0 z m d z [ a ( x , y ) z + m 2 ] 2 ∼ ∫ 0 z m d z m 4 = z m m 4 ∼ 1 ∣ a ( x , y ) ∣ m 2 . (83) \int_0^{z_m}\frac{\mathrm{d}z}{\left[a(x,y)z+m^2\right]^2}
\sim\int_0^{z_m}\frac{\mathrm{d}z}{m^4}
= \frac{z_m}{m^4}
\sim\frac{1}{\lvert a(x,y)\rvert m^2}. \tag*{(83)} ∫ 0 z m [ a ( x , y ) z + m 2 ] 2 d z ∼ ∫ 0 z m m 4 d z = m 4 z m ∼ ∣ a ( x , y )∣ m 2 1 . (83)
Above the crossover, the denominator is approximately a ( x , y ) z a(x,y)z a ( x , y ) z :
∫ z m z 0 d z [ a ( x , y ) z + m 2 ] 2 ∼ 1 a ( x , y ) 2 ( 1 z m − 1 z 0 ) ∼ 1 ∣ a ( x , y ) ∣ m 2 . (84) \int_{z_m}^{z_0}\frac{\mathrm{d}z}{\left[a(x,y)z+m^2\right]^2}
\sim\frac{1}{a(x,y)^2}\left(\frac{1}{z_m}-\frac{1}{z_0}\right)
\sim\frac{1}{\lvert a(x,y)\rvert m^2}. \tag*{(84)} ∫ z m z 0 [ a ( x , y ) z + m 2 ] 2 d z ∼ a ( x , y ) 2 1 ( z m 1 − z 0 1 ) ∼ ∣ a ( x , y )∣ m 2 1 . (84)
This second integral is dominated by its lower endpoint, z ∼ z m z\sim z_m z ∼ z m . Thus, the inverse-mass behavior receives contributions from both z ≲ z m z\lesssim z_m z ≲ z m and z ≳ z m z\gtrsim z_m z ≳ z m . As we will show in the next section (Sec. IV), after summing the contributing diagrams and projecting onto the relevant tensor structures, the 1 / m 2 1/m^2 1/ m 2 terms do not cancel. We therefore find an inverse-mass-squared contribution in the complete amplitude. To our knowledge, this behavior has not previously been reported for these amplitudes. The parameter analysis traces this term to the neighborhood of the z = 0 z=0 z = 0 boundary associated with the reduced triangle. The origin of this contribution at the level of the full QCD operator expansion is not addressed here. In particular, we do not identify it with a specific operator matrix element or assign it an independent operator interpretation. We regard the result as a feature of the calculated amplitude whose interpretation requires further study.
The r = 0 r=0 r = 0 coefficient also contains a logarithm from the first correction to the local denominator. Write
F D ( m ) = G m + H m , G m = a ( x , y ) z + m 2 , H m = c z 2 . (85) F_D^{(m)}=G_m+H_m,\qquad G_m=a(x,y)z+m^2,\qquad H_m=cz^2. \tag*{(85)} F D ( m ) = G m + H m , G m = a ( x , y ) z + m 2 , H m = c z 2 . (85)
The expansion gives
1 ( F D ( m ) ) 2 = 1 G m 2 − 2 c z 2 G m 3 + ⋯ . (86) \frac{1}{\left(F_D^{(m)}\right)^2}
=\frac{1}{G_m^2}-\frac{2cz^2}{G_m^3}+\cdots. \tag*{(86)} ( F D ( m ) ) 2 1 = G m 2 1 − G m 3 2 c z 2 + ⋯ . (86)
For z ≲ z m z\lesssim z_m z ≲ z m , the correction is proportional to z 2 / m 6 z^2/m^6 z 2 / m 6 , and its integral remains finite as m 2 → 0 m^2\to0 m 2 → 0 . For z ≳ z m z\gtrsim z_m z ≳ z m , it behaves as 1 / z 1/z 1/ z , giving
− 2 c ∫ z m z 0 z 2 d z [ a ( x , y ) z + m 2 ] 3 ∼ − 2 c a ( x , y ) 3 ln ( z 0 z m ) . (87) -2c\int_{z_m}^{z_0}\frac{z^2\,\mathrm{d}z}{\left[a(x,y)z+m^2\right]^3}
\sim-\frac{2c}{a(x,y)^3}\ln\left(\frac{z_0}{z_m}\right). \tag*{(87)} − 2 c ∫ z m z 0 [ a ( x , y ) z + m 2 ] 3 z 2 d z ∼ − a ( x , y ) 3 2 c ln ( z m z 0 ) . (87)
Since z m ∝ m 2 z_m\propto m^2 z m ∝ m 2 , this contains a logarithm of m 2 m^2 m 2 . The logarithm accumulates over the interval above the crossover; it is not generated by the region z ≪ z m z\ll z_m z ≪ z m .
For r = 1 r=1 r = 1 , the local integrand instead contains ( F D ( m ) ) − 1 \left(F_D^{(m)}\right)^{-1} ( F D ( m ) ) − 1 . The leading boundary form already gives
∫ 0 z 0 d z a ( x , y ) z + m 2 = 1 a ( x , y ) ln [ a ( x , y ) z 0 + m 2 m 2 ] , (88) \int_0^{z_0}\frac{\mathrm{d}z}{a(x,y)z+m^2}
=\frac{1}{a(x,y)}\ln\left[\frac{a(x,y)z_0+m^2}{m^2}\right], \tag*{(88)} ∫ 0 z 0 a ( x , y ) z + m 2 d z = a ( x , y ) 1 ln [ m 2 a ( x , y ) z 0 + m 2 ] , (88)
and hence a logarithm as m 2 → 0 m^2 \to0 m 2 → 0 .
The comparison with nonzero momentum transfer is direct. At generic fixed x , y x,y x , y , the two regulators enter the leading local denominator as
a ( x , y ) z + b ( x , y ) t ⟷ a ( x , y ) z + m 2 . (89) a(x,y)z+b(x,y)t \longleftrightarrow a(x,y)z+m^2. \tag*{(89)} a ( x , y ) z + b ( x , y ) t ⟷ a ( x , y ) z + m 2 . (89)
They set different crossover scales, z t = ∣ b ( x , y ) t / a ( x , y ) ∣ z_t=\lvert b(x,y)t/a(x,y)\rvert z t = ∣ b ( x , y ) t / a ( x , y )∣ and z m = m 2 / ∣ a ( x , y ) ∣ z_m=m^2/\lvert a(x,y)\rvert z m = m 2 / ∣ a ( x , y )∣ , but regulate the same endpoint region near z = 0 z=0 z = 0 . For r = 0 r=0 r = 0 , both sides of the crossover contribute to the inverse-power behavior, 1 / t 1/t 1/ t or 1 / m 2 1/m^2 1/ m 2 , while the logarithm comes from the correction that produces a 1 / z 1/z 1/ z integrand above the crossover. For r = 1 r=1 r = 1 , the leading local integrand already behaves as 1 / z 1/z 1/ z above the crossover and therefore produces a logarithm.
In the same spirit as in the previous subsection, we now examine the same four-point coefficient functions with a common nonzero internal mass m m m , setting t = 0 t=0 t = 0 . For compactness, define
R a ≡ Q a 2 ( 4 m 2 + Q a 2 ) , R b ≡ Q b 2 ( 4 m 2 + Q b 2 ) , R s ≡ s ( s − 4 m 2 ) , L a ≡ ln ( 2 m 2 + Q a 2 + R a 2 m 2 ) , L b ≡ ln ( 2 m 2 + Q b 2 + R b 2 m 2 ) , L s ≡ ln ( 2 m 2 − s + R s 2 m 2 ) . (90) \begin{aligned}
R_a &\equiv\sqrt{Q_a^2(4m^2+Q_a^2)}, &
R_b &\equiv\sqrt{Q_b^2(4m^2+Q_b^2)}, &
R_s &\equiv\sqrt{s(s-4m^2)},\\
L_a &\equiv\ln\left(\frac{2m^2+Q_a^2+R_a}{2m^2}\right), &
L_b &\equiv\ln\left(\frac{2m^2+Q_b^2+R_b}{2m^2}\right), &
L_s &\equiv\ln\left(\frac{2m^2-s+R_s}{2m^2}\right). \tag*{(90)}
\end{aligned} R a L a ≡ Q a 2 ( 4 m 2 + Q a 2 ) , ≡ ln ( 2 m 2 2 m 2 + Q a 2 + R a ) , R b L b ≡ Q b 2 ( 4 m 2 + Q b 2 ) , ≡ ln ( 2 m 2 2 m 2 + Q b 2 + R b ) , R s L s ≡ s ( s − 4 m 2 ) , ≡ ln ( 2 m 2 2 m 2 − s + R s ) . (90)
We first consider the scalar coefficient, r = 0 r=0 r = 0 and n 3 = 0 n_3=0 n 3 = 0 :
D 0 ( m ) ≡ D 0 ( 0 , 0 , − Q b 2 , − Q a 2 ; 0 , s ; m 2 , m 2 , m 2 , m 2 ) . (91) D_0^{(m)}\equiv D_0(0,0,-Q_b^2,-Q_a^2;0,s;m^2,m^2,m^2,m^2). \tag*{(91)} D 0 ( m ) ≡ D 0 ( 0 , 0 , − Q b 2 , − Q a 2 ; 0 , s ; m 2 , m 2 , m 2 , m 2 ) . (91)
Its mass dependence is
D 0 ( m ) = R a L a m 2 ( Q a 2 − Q b 2 ) ( Q a 2 + s ) + R b L b m 2 ( − Q a 2 + Q b 2 ) ( Q b 2 + s ) + R s L s m 2 ( Q a 2 + s ) ( Q b 2 + s ) . (92) D_0^{(m)}=\frac{R_aL_a}{m^2(Q_a^2-Q_b^2)(Q_a^2+s)}+\frac{R_bL_b}{m^2(-Q_a^2+Q_b^2)(Q_b^2+s)}+\frac{R_sL_s}{m^2(Q_a^2+s)(Q_b^2+s)}. \tag*{(92)} D 0 ( m ) = m 2 ( Q a 2 − Q b 2 ) ( Q a 2 + s ) R a L a + m 2 ( − Q a 2 + Q b 2 ) ( Q b 2 + s ) R b L b + m 2 ( Q a 2 + s ) ( Q b 2 + s ) R s L s . (92)
The explicit factor 1 / m 2 1/m^2 1/ m 2 displays the power enhancement in this coefficient4 . For comparison, the r = 0 r=0 r = 0 , n 3 = 1 n_3=1 n 3 = 1 coefficient is
D 3 ( m ) = − L a 2 + L b 2 2 ( Q a 2 − Q b 2 ) ( Q a 2 + s ) + − L b 2 + L s 2 2 ( Q b 2 + s ) ( Q a 2 + s ) . (93) D_3^{(m)}=-\frac{L_a^2+L_b^2}{2(Q_a^2-Q_b^2)(Q_a^2+s)}+\frac{-L_b^2+L_s^2}{2(Q_b^2+s)(Q_a^2+s)}. \tag*{(93)} D 3 ( m ) = − 2 ( Q a 2 − Q b 2 ) ( Q a 2 + s ) L a 2 + L b 2 + 2 ( Q b 2 + s ) ( Q a 2 + s ) − L b 2 + L s 2 . (93)
This expression contains logarithms of the mass-regulated invariants, but no explicit 1 / m 2 1/m^2 1/ m 2 prefactor. The factor of z z z in its parameter numerator vanishes at the z = 0 z=0 z = 0 boundary and softens the power enhancement found for D 0 ( m ) D_0^{(m)} D 0 ( m ) .
The r = 1 r=1 r = 1 , n 3 = 0 n_3=0 n 3 = 0 coefficient provides a third example:
D 00 ( m ) = R a L a ( Q a 2 − Q b 2 ) ( Q a 2 + s ) + R b L b ( − Q a 2 + Q b 2 ) ( Q b 2 + s ) + R s L s ( Q a 2 + s ) ( Q b 2 + s ) + Q a 2 2 ( Q a 2 + s ) [ − L a 2 2 ( Q a 2 − Q b 2 ) + L b 2 2 ( Q a 2 − Q b 2 ) ] + s 2 ( Q a 2 + s ) [ − L b 2 2 ( Q b 2 + s ) + L s 2 2 ( Q b 2 + s ) ] . (94) \begin{aligned}
D_{00}^{(m)}={}&\frac{R_aL_a}{(Q_a^2-Q_b^2)(Q_a^2+s)}+\frac{R_bL_b}{(-Q_a^2+Q_b^2)(Q_b^2+s)}+\frac{R_sL_s}{(Q_a^2+s)(Q_b^2+s)}\\
&+\frac{Q_a^2}{2(Q_a^2+s)}\left[-\frac{L_a^2}{2(Q_a^2-Q_b^2)}+\frac{L_b^2}{2(Q_a^2-Q_b^2)}\right]\\
&+\frac{s}{2(Q_a^2+s)}\left[-\frac{L_b^2}{2(Q_b^2+s)}+\frac{L_s^2}{2(Q_b^2+s)}\right]. \tag*{(94)}
\end{aligned} D 00 ( m ) = ( Q a 2 − Q b 2 ) ( Q a 2 + s ) R a L a + ( − Q a 2 + Q b 2 ) ( Q b 2 + s ) R b L b + ( Q a 2 + s ) ( Q b 2 + s ) R s L s + 2 ( Q a 2 + s ) Q a 2 [ − 2 ( Q a 2 − Q b 2 ) L a 2 + 2 ( Q a 2 − Q b 2 ) L b 2 ] + 2 ( Q a 2 + s ) s [ − 2 ( Q b 2 + s ) L b 2 + 2 ( Q b 2 + s ) L s 2 ] . (94)
Like D 3 ( m ) D_3^{(m)} D 3 ( m ) , this coefficient has logarithmic mass dependence but no explicit 1 / m 2 1/m^2 1/ m 2 prefactor.
To summarize, introducing a nonzero mass exposes the same z = 0 z=0 z = 0 triangle boundary in the four-point coefficient functions that is regulated by nonzero t t t . After the relevant contributions are combined, this boundary produces a 1 / m 2 1/m^2 1/ m 2 term. To repeat again, in the massless, zero-transfer limit, the corresponding triangle vanishes in DR at d = 4 d=4 d = 4 by analytic continuation. This vanishing does not remove the underlying power-divergent region. A cutoff regulator can make its power behavior explicit, while in DR the same region becomes logarithmically divergent at the critical dimension d = 6 d=6 d = 6 .
Note that in the preceding subsections, we considered the cases in which t t t or m m m separately regulates the boundary. We can also obtain the exact scalar coefficient with both regulators nonzero. The exact expression involves square roots and continued dilogarithms. We present it below in its unexpanded form for illustration:
D 0 ( 0 , 0 , − Q b 2 , − Q a 2 ; t , s ; m 2 , m 2 , m 2 , m 2 ) . (95) D_0(0,0,-Q_b^2,-Q_a^2;t,s;m^2,m^2,m^2,m^2). \tag*{(95)} D 0 ( 0 , 0 , − Q b 2 , − Q a 2 ; t , s ; m 2 , m 2 , m 2 , m 2 ) . (95)
We use the abbreviations
M ≡ m 2 , A ≡ Q a 2 , B ≡ Q b 2 , K ≡ 2 M + A + A ( A + 4 M ) , (96) M \equiv m^2,\qquad A \equiv Q_a^2,\qquad B \equiv Q_b^2,\qquad K \equiv2M+A+\sqrt{A(A+4M)}, \tag*{(96)} M ≡ m 2 , A ≡ Q a 2 , B ≡ Q b 2 , K ≡ 2 M + A + A ( A + 4 M ) , (96)
and
U ≡ 2 M ( 2 M + B ) − ( 2 M − s ) ( 2 M − t ) , (97) U \equiv2M(2M+B)-(2M-s)(2M-t), \tag*{(97)} U ≡ 2 M ( 2 M + B ) − ( 2 M − s ) ( 2 M − t ) , (97)
V ≡ K 2 ( B + s ) + M t , (98) V \equiv\frac{K}{2}(B+s)+Mt, \tag*{(98)} V ≡ 2 K ( B + s ) + M t , (98)
W ≡ − 2 M ( B + s ) K − t , (99) W \equiv-\frac{2M(B+s)}{K}-t, \tag*{(99)} W ≡ − K 2 M ( B + s ) − t , (99)
Θ X ≡ U + 2 V X , χ ≡ K 2 − 2 M 2 K . (100) \Theta_X \equiv U+2VX,\qquad\chi\equiv\frac{K}{2}-\frac{2M^2}{K}. \tag*{(100)} Θ X ≡ U + 2 V X , χ ≡ 2 K − K 2 M 2 . (100)
The roots in the outer sum are specified by
P ( X ) ≡ V X 2 + U X − M W = 0. (101) P(X) \equiv VX^2+UX-MW=0. \tag*{(101)} P ( X ) ≡ V X 2 + U X − M W = 0. (101)
Here D ( u , v ) \mathcal{D}(u,v) D ( u , v ) and L ( u , v ) \mathcal{L}(u,v) L ( u , v ) denote the branch-prescribed two-argument dilogarithm and logarithm, respectively, in the convention of the loop-integral result. We write D c ( { ⋯ } , { ⋯ } ) \mathcal{D}_{c}(\{\cdots\},\{\cdots\}) D c ({ ⋯ } , { ⋯ }) for its continued dilogarithm. Define
R X ≡ M ( 1 + X 2 ) + X ( 2 M + B ) M ( 1 + X 2 ) + X ( 2 M − s ) , (102) R_X \equiv\frac{M(1+X^2)+X(2M+B)}{M(1+X^2)+X(2M-s)}, \tag*{(102)} R X ≡ M ( 1 + X 2 ) + X ( 2 M − s ) M ( 1 + X 2 ) + X ( 2 M + B ) , (102)
S X ≡ 2 M + X χ − 2 M ( 2 M + B ) K − t 2 M + X χ − 2 M ( 2 M − s ) K , (103) S_X \equiv\frac{2M+X\chi-\frac{2M(2M+B)}{K}-t}{2M+X\chi-\frac{2M(2M-s)}{K}}, \tag*{(103)} S X ≡ 2 M + X χ − K 2 M ( 2 M − s ) 2 M + X χ − K 2 M ( 2 M + B ) − t , (103)
J X ≡ − L ( − X , W Θ X ) + L ( − X K 2 M , K W 2 M Θ X ) − ln ( K 2 M ) . (104) J_X \equiv-\mathcal{L}\left(-X,\frac{W}{\Theta_X}\right)+\mathcal{L}\left(-\frac{XK}{2M},\frac{KW}{2M\Theta_X}\right)-\ln\left(\frac{K}{2M}\right). \tag*{(104)} J X ≡ − L ( − X , Θ X W ) + L ( − 2 M X K , 2 M Θ X K W ) − ln ( 2 M K ) . (104)
The inner sums are
R B ( X ) ≡ ∑ r : M r 2 + ( 2 M + B ) r + M = 0 D c ( { − 1 r , − M r + M r } , { − X , W 2 M Θ X } ) , (105) \mathcal{R}_B(X) \equiv\sum_{r: \, Mr^2+(2M+B)r+M=0}\mathcal{D}_{c}\left(\left\{-\frac{1}{r},-\frac{M}{r}+Mr\right\},\left\{-X,\frac{W}{2M\Theta_X}\right\}\right), \tag*{(105)} R B ( X ) ≡ r : M r 2 + ( 2 M + B ) r + M = 0 ∑ D c ( { − r 1 , − r M + M r } , { − X , 2 M Θ X W } ) , (105)
R s ( X ) ≡ ∑ r : M r 2 + ( 2 M − s ) r + M = 0 D c ( { − 1 r , − M r + M r } , { − X , W 2 M Θ X } ) , (106) \mathcal{R}_s(X) \equiv\sum_{r: \, Mr^2+(2M-s)r+M=0}\mathcal{D}_{c}\left(\left\{-\frac{1}{r},-\frac{M}{r}+Mr\right\},\left\{-X,\frac{W}{2M\Theta_X}\right\}\right), \tag*{(106)} R s ( X ) ≡ r : M r 2 + ( 2 M − s ) r + M = 0 ∑ D c ( { − r 1 , − r M + M r } , { − X , 2 M Θ X W } ) , (106)
R t ( X ) ≡ ∑ r : M r 2 + ( 2 M − t ) r + M = 0 D c ( { − 1 r , − M r + M r } , { − X K 2 M , K W 2 M Θ X } ) . (107) \mathcal{R}_t(X) \equiv\sum_{r: \, Mr^2+(2M-t)r+M=0}\mathcal{D}_{c}\left(\left\{-\frac{1}{r},-\frac{M}{r}+Mr\right\},\left\{-\frac{XK}{2M},\frac{KW}{2M\Theta_X}\right\}\right). \tag*{(107)} R t ( X ) ≡ r : M r 2 + ( 2 M − t ) r + M = 0 ∑ D c ( { − r 1 , − r M + M r } , { − 2 M X K , 2 M Θ X K W } ) . (107)
Then
D 0 = ∑ X : P ( X ) = 0 Φ ( X ) Θ X , (108) D_0=\sum_{X: \, P(X)=0}\frac{\Phi(X)}{\Theta_X}, \tag*{(108)} D 0 = X : P ( X ) = 0 ∑ Θ X Φ ( X ) , (108)
Φ ( X ) ≡ D ( 1 + X K 2 M , − X K 2 M + K W 2 M Θ X ) + D ( 1 + X K 2 M , X K 2 M + K W 2 M Θ X ) − L ( R X , S X ) J X + R B ( X ) − R s ( X ) − R t ( X ) . (109) \Phi(X)\equiv\mathcal{D}\left(1+\frac{XK}{2M},-\frac{XK}{2M}+\frac{KW}{2M\Theta_X}\right)+\mathcal{D}\left(1+\frac{XK}{2M},\frac{XK}{2M}+\frac{KW}{2M\Theta_X}\right)-\mathcal{L}(R_X,S_X)J_X+\mathcal{R}_B(X)-\mathcal{R}_s(X)-\mathcal{R}_t(X). \tag*{(109)} Φ ( X ) ≡ D ( 1 + 2 M X K , − 2 M X K + 2 M Θ X K W ) + D ( 1 + 2 M X K , 2 M X K + 2 M Θ X K W ) − L ( R X , S X ) J X + R B ( X ) − R s ( X ) − R t ( X ) . (109)
The branch prescriptions in D \mathcal{D} D , L \mathcal{L} L , and D c \mathcal{D}_c D c are inherited from the loop integral. Although the expression is exact for nonzero t t t and m m m , extracting its asymptotic behavior requires careful control of the branches and the order of limits. The two relevant hierarchies suggest the following leading behavior for the complete amplitude:
A ( t , m 2 ) ∼ { R t + O ( m 2 ∣ t ∣ ) , m 2 ≪ ∣ t ∣ ≪ Q 2 , R m 2 + O ( ∣ t ∣ m 2 ) , ∣ t ∣ ≪ m 2 ≪ Q 2 . (110) \mathcal{A}(t,m^2)\sim
\begin{cases}
\frac{\mathcal{R}}{t}+\mathcal{O}\left(\frac{m^2}{\lvert t\rvert}\right), & m^2\ll\lvert t\rvert\ll Q^2,\\[6pt]
\frac{\mathcal{R}}{m^2}+\mathcal{O}\left(\frac{\lvert t\rvert}{m^2}\right), & \lvert t\rvert\ll m^2\ll Q^2.
\end{cases} \tag*{(110)} A ( t , m 2 ) ∼ ⎩ ⎨ ⎧ t R + O ( ∣ t ∣ m 2 ) , m 2 R + O ( m 2 ∣ t ∣ ) , m 2 ≪ ∣ t ∣ ≪ Q 2 , ∣ t ∣ ≪ m 2 ≪ Q 2 . (110)
Here R \mathcal{R} R denotes the common residue (chiral or trace anomaly) in the complete amplitude. Thus, the leading singular contribution is associated with the same anomaly-related tensor structure in either hierarchy. A controlled derivation of these limits, including the branch choices and their relation to dimensional regularization, will be presented in a separate work. For clarity, we focus here on the cases in which t t t or m m m separately regulates the boundary.
Complete one-loop results in DDVCS
We now combine the box, cat’s-eye, and hourglass contributions shown in Figure 1 and present the complete one-loop results using three complementary infrared prescriptions: DR, nonzero momentum transfer (t ≠ 0 t \ne0 t = 0 ), and a finite quark mass (m ≠ 0 m \ne0 m = 0 ). The comparison shows how the anomaly-related terms and the infrared singularities associated with evolution are manifested in each prescription. The regulator dependence of the complete one-loop result can be summarized schematically as follows:
R b d r y D R + P l i n D R ⏟ = 0 at d = 4 + K e v o l ϵ I R + T r e m D R . (111) \mathcal{R}_{\mathrm{bdry}}^{\mathrm{DR}}+\underbrace{\mathcal{P}_{\mathrm{lin}}^{\mathrm{DR}}}_{=0\ \text{at}\ d=4}+\frac{\mathcal{K}_{\mathrm{evol}}}{\epsilon_{\mathrm{IR}}}+\mathcal{T}_{\mathrm{rem}}^{\mathrm{DR}}. \tag*{(111)} R bdry DR + = 0 at d = 4 P lin DR + ϵ IR K evol + T rem DR . (111)
Here P l i n D R \mathcal{P}_{\mathrm{lin}}^{\mathrm{DR}} P lin DR is a bookkeeping symbol for the power-divergent boundary sector, which diverges as 1 / δ 1/\delta 1/ δ at d = 4 d=4 d = 4 with a lower endpoint cutoff δ \delta δ . We set this symbol to zero in the DR result because analytic continuation yields no 1 / ϵ 1/\epsilon 1/ ϵ pole from this sector in the expansion about d = 4 d=4 d = 4 . This does not mean that the underlying singular region is absent: the 1 / δ 1/\delta 1/ δ divergence is explicit in the cutoff calculation. Its absence from the pole part at d = 4 d=4 d = 4 is a consequence of analytic continuation. As discussed in Sec. III A, the same endpoint diverges logarithmically at the critical dimension d = 6 d=6 d = 6 , where the analytically continued integral has a pole at ϵ = − 1 \epsilon=-1 ϵ = − 1 .
R t t + K e v o l ln ( Q 2 − t ) + T r e m t . (112) \frac{\mathcal{R}_{t}}{t}+\mathcal{K}_{\mathrm{evol}}\ln\left(\frac{Q^{2}}{-t}\right)+\mathcal{T}_{\mathrm{rem}}^{t}. \tag*{(112)} t R t + K evol ln ( − t Q 2 ) + T rem t . (112)
R m m 2 + K e v o l ln ( Q 2 m 2 ) + T r e m m . (113) \frac{\mathcal{R}_{m}}{m^{2}}+\mathcal{K}_{\mathrm{evol}}\ln\left(\frac{Q^{2}}{m^{2}}\right)+\mathcal{T}_{\mathrm{rem}}^{m}. \tag*{(113)} m 2 R m + K evol ln ( m 2 Q 2 ) + T rem m . (113)
A few comments are in order. First, in the nonzero-t t t calculation, we do not impose the lightlike relation ℓ μ = − 2 ξ p μ \ell^{\mu}=-2\xi p^{\mu} ℓ μ = − 2 ξ p μ [21 , 22 ] , which would set t = 0 t=0 t = 0 . In the mass-regulated calculation as well we do not use ℓ μ = − 2 ξ p μ \ell^{\mu}=-2\xi p^{\mu} ℓ μ = − 2 ξ p μ . This choice allows us to retain the polarization-dependent tensor structures that carry the anomaly-related contribution (see Eq. (75) ). Second, we emphasize again that the symbol P l i n D R = 0 \mathcal{P}_{\mathrm{lin}}^{\mathrm{DR}}=0 P lin DR = 0 records how DR treats the boundary contribution. It does not imply that the underlying power-divergent region is absent. Third, the 1 / t 1/t 1/ t and 1 / m 2 1/m^{2} 1/ m 2 terms expose the anomaly-related residues R t \mathcal{R}_{t} R t and R m \mathcal{R}_{m} R m , respectively. The gluonic tensor structures identified in Eq. (75) allow us to write the residues of the 1 / t 1/t 1/ t and 1 / m 2 1/m^{2} 1/ m 2 terms schematically as
R ρ X = C ρ X ⟨ p 2 ∣ O X ∣ p 1 ⟩ , ρ = t , m , X = A , S , (114) \mathcal{R}_{\rho}^{X}=C_{\rho}^{X}\langle p_{2}|\mathcal{O}_{X}|p_{1}\rangle,\qquad\rho=t,m,\quad X=A,S, \tag*{(114)} R ρ X = C ρ X ⟨ p 2 ∣ O X ∣ p 1 ⟩ , ρ = t , m , X = A , S , (114)
where A A A and S S S denote the antisymmetric (polarized) and symmetric (unpolarized) amplitudes, respectively, and O A = F μ ν F ~ μ ν \mathcal{O}_{A}=F^{\mu\nu}\widetilde{F}_{\mu\nu} O A = F μν F μν and O S = F μ ν F μ ν \mathcal{O}_{S}=F^{\mu\nu}F_{\mu\nu} O S = F μν F μν . Within each polarization channel (polarized or unpolarized), the 1 / t 1/t 1/ t and 1 / m 2 1/m^{2} 1/ m 2 residues share the same gluonic operator structure, while the coefficient functions C t X C_{t}^{X} C t X and C m X C_{m}^{X} C m X may differ. This shared operator structure is consistent with the common origin of the two inverse-power terms. We leave a comparison of the x x x -integrals of the coefficient functions, along the lines of Refs. [21 , 22 ] , to future work. Finally, the infrared pole 1 / ϵ I R 1/\epsilon_{\mathrm{IR}} 1/ ϵ IR and the regulator-dependent logarithms share the same evolution-kernel coefficient, K e v o l \mathcal{K}_{\mathrm{evol}} K evol . Moreover, each T r e m \mathcal{T}_{\mathrm{rem}} T rem denotes the remainder finite in the corresponding regulator limit; its dependence on the polarization channel and external kinematics is implicit.
Note that the ultraviolet-pole cancellations follow a different, channel-dependent pattern. In the symmetric amplitude, the ultraviolet poles from the box and hourglass diagrams cancel against the pole from the cat’s-eye diagram. In the antisymmetric amplitude, the ultraviolet poles from the box and hourglass diagrams cancel each other. We are now ready to present the complete results explicitly.
Polarized case
For the polarized amplitude, the evolution kernel is
K ~ e v o l ( x ^ , ξ ^ ) = 2 x ^ − 1 − ξ ^ 2 ( 1 − ξ ^ 2 ) 2 ln ( x ^ − 1 x ^ ) − 2 x ^ − ξ ^ ( 1 − ξ ^ 2 ) 2 ln ( x ^ − ξ ^ x ^ ) − ( x ^ → − x ^ ) . (115) \widetilde{\mathcal{K}}_{\mathrm{evol}}(\hat{x},\hat{\xi})=\frac{2\hat{x}-1-\hat{\xi}^{2}}{(1-\hat{\xi}^{2})^{2}}\ln\left(\frac{\hat{x}-1}{\hat{x}}\right)-2\frac{\hat{x}-\hat{\xi}}{(1-\hat{\xi}^{2})^{2}}\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)-(\hat{x}\to-\hat{x}). \tag*{(115)} K evol ( x ^ , ξ ^ ) = ( 1 − ξ ^ 2 ) 2 2 x ^ − 1 − ξ ^ 2 ln ( x ^ x ^ − 1 ) − 2 ( 1 − ξ ^ 2 ) 2 x ^ − ξ ^ ln ( x ^ x ^ − ξ ^ ) − ( x ^ → − x ^ ) . (115)
This expression agrees with the previously reported polarized result [21 , 22 , 32 , 33 , 34 ] . The same kernel multiplies the infrared pole 1 / ϵ I R 1/\epsilon_{\mathrm{IR}} 1/ ϵ IR in dimensional regularization and the regulator logarithms in the nonzero-t t t and nonzero-m m m calculations.
In the notation of Ref. [22 ] , the coefficient of the anomaly-related 1 / t 1/t 1/ t term is given by:
R t A ( x ^ , ξ ^ ) ∼ 8 ⟨ p 2 ∣ F μ ν F ~ μ ν ∣ p 1 ⟩ ( 1 − ξ ^ 2 ) { ( 1 − x ^ ) ln ( x ^ − 1 x ) + ( x ^ − ξ ^ ) ln ( x ^ − ξ ^ x ^ ) − ( x ^ → − x ^ ) } . (116) \mathcal{R}^{A}_{t}(\hat{x},\hat{\xi}) \sim\frac{8\langle p_{2}|F^{\mu\nu}\widetilde{F}_{\mu\nu}|p_{1}\rangle}{(1-\hat{\xi}^{2})}\left\{(1-\hat{x})\ln\left(\frac{\hat{x}-1}{x}\right)+(\hat{x}-\hat{\xi})\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)-(\hat{x}\to-\hat{x})\right\}. \tag*{(116)} R t A ( x ^ , ξ ^ ) ∼ ( 1 − ξ ^ 2 ) 8 ⟨ p 2 ∣ F μν F μν ∣ p 1 ⟩ { ( 1 − x ^ ) ln ( x x ^ − 1 ) + ( x ^ − ξ ^ ) ln ( x ^ x ^ − ξ ^ ) − ( x ^ → − x ^ ) } . (116)
With the mass regulator, the coefficient of the 1 / m 2 1/m^{2} 1/ m 2 term is given by:
R m A ( x ^ , ξ ^ ) ∼ 4 ⟨ p 2 ∣ F μ ν F ~ μ ν ∣ p 1 ⟩ 3 ξ ^ ( ξ ^ 2 − 1 ) 2 { − 4 x ^ ξ ^ + 4 x ^ ξ ^ 3 + 2 ξ ^ ( x ^ − 1 ) 3 ln ( x ^ − 1 ) + 2 ξ ^ ( x ^ + 1 ) 3 ln ( x ^ + 1 ) + [ 3 x ^ 2 − 6 x ^ ξ ^ − 2 x ^ 3 ξ ^ + 3 ξ ^ 2 + 3 x ^ 2 ξ ^ 2 − ξ ^ 4 ] ln ( x ^ − ξ ^ ) + [ − 3 x ^ 2 − 6 x ^ ξ ^ − 2 x ^ 3 ξ ^ − 3 ξ ^ 2 − 3 x ^ 2 ξ ^ 2 + ξ ^ 4 ] ln ( x ^ + ξ ^ ) } (117) \begin{aligned}
\mathcal{R}^{A}_{m}(\hat{x},\hat{\xi}) \sim{}& \frac{4\langle p_{2}|F^{\mu\nu}\widetilde{F}_{\mu\nu}|p_{1}\rangle}{3\hat{\xi}(\hat{\xi}^{2}-1)^{2}}\left\{-4\hat{x}\hat{\xi}+4\hat{x}\hat{\xi}^{3}+2\hat{\xi}(\hat{x}-1)^{3}\ln(\hat{x}-1)+2\hat{\xi}(\hat{x}+1)^{3}\ln(\hat{x}+1)\right.\\
&\left.+\left[3\hat{x}^{2}-6\hat{x}\hat{\xi}-2\hat{x}^{3}\hat{\xi}+3\hat{\xi}^{2}+3\hat{x}^{2}\hat{\xi}^{2}-\hat{\xi}^{4}\right]\ln(\hat{x}-\hat{\xi})\right.\\
&\left.\left.+\left[-3\hat{x}^{2}-6\hat{x}\hat{\xi}-2\hat{x}^{3}\hat{\xi}-3\hat{\xi}^{2}-3\hat{x}^{2}\hat{\xi}^{2}+\hat{\xi}^{4}\right]\ln(\hat{x}+\hat{\xi})\right\}\right. \tag*{(117)}
\end{aligned} R m A ( x ^ , ξ ^ ) ∼ 3 ξ ^ ( ξ ^ 2 − 1 ) 2 4 ⟨ p 2 ∣ F μν F μν ∣ p 1 ⟩ { − 4 x ^ ξ ^ + 4 x ^ ξ ^ 3 + 2 ξ ^ ( x ^ − 1 ) 3 ln ( x ^ − 1 ) + 2 ξ ^ ( x ^ + 1 ) 3 ln ( x ^ + 1 ) + [ 3 x ^ 2 − 6 x ^ ξ ^ − 2 x ^ 3 ξ ^ + 3 ξ ^ 2 + 3 x ^ 2 ξ ^ 2 − ξ ^ 4 ] ln ( x ^ − ξ ^ ) + [ − 3 x ^ 2 − 6 x ^ ξ ^ − 2 x ^ 3 ξ ^ − 3 ξ ^ 2 − 3 x ^ 2 ξ ^ 2 + ξ ^ 4 ] ln ( x ^ + ξ ^ ) } (117)
Thus, the coefficients of the 1 / t 1/t 1/ t and 1 / m 2 1/m^{2} 1/ m 2 terms multiply the same pseudoscalar gluonic tensor structure, which appears in the matrix element of F μ ν F ~ μ ν F^{\mu\nu}\widetilde{F}_{\mu\nu} F μν F μν and is associated with the chiral anomaly. As discussed above, keeping the gluon polarization vectors explicit makes this structure manifest in the residue. Contracting with a transverse projector can reorganize the numerator and suppress the residue, while the original propagator denominators still satisfy the Landau conditions on the same boundary. Both regulators, t ≠ 0 t\ne0 t = 0 and m ≠ 0 m\ne0 m = 0 , therefore, expose the same Landau boundary, although the associated ( x ^ , ξ ^ ) (\hat{x},\hat{\xi}) ( x ^ , ξ ^ ) -dependent coefficient functions may differ.
We now turn to the finite terms. The DR result contains the following coefficient function:
T r e m A , D R ( x ^ , ξ ^ ) ∼ − 4 Q 2 x ^ ( ξ ^ 2 − 1 ) { − 4 ( x ^ − ξ ^ 2 ) ln ( x ^ x ^ − 1 ) + ( − 1 + 2 x ^ − ξ ^ 2 ) ln 2 ( x ^ x ^ − 1 ) − 4 ( x ^ + ξ ^ 2 ) ln ( x ^ 1 + x ^ ) + ( 1 + 2 x ^ + ξ ^ 2 ) ln 2 ( x ^ 1 + x ^ ) + 4 ( x ^ − ξ ^ ) ln ( x ^ x ^ − ξ ^ ) − 2 ( x ^ − ξ ^ ) ln 2 ( x ^ x ^ − ξ ^ ) − 2 ( x ^ + ξ ^ ) [ − 2 + ln ( x ^ x ^ + ξ ^ ) ] ln ( x ^ x ^ + ξ ^ ) } . (118) \begin{aligned}
\mathcal{T}^{A,\mathrm{DR}}_{\mathrm{rem}}(\hat{x},\hat{\xi}) \sim{}&-\frac{4Q^{2}}{\hat{x}(\hat{\xi}^{2}-1)}\left\{-4(\hat{x}-\hat{\xi}^{2})\ln\left(\frac{\hat{x}}{\hat{x}-1}\right)+(-1+2\hat{x}-\hat{\xi}^{2})\ln^{2}\left(\frac{\hat{x}}{\hat{x}-1}\right)\right.\\
&-4(\hat{x}+\hat{\xi}^{2})\ln\left(\frac{\hat{x}}{1+\hat{x}}\right)+(1+2\hat{x}+\hat{\xi}^{2})\ln^{2}\left(\frac{\hat{x}}{1+\hat{x}}\right)\\
&+4(\hat{x}-\hat{\xi})\ln\left(\frac{\hat{x}}{\hat{x}-\hat{\xi}}\right)-2(\hat{x}-\hat{\xi})\ln^{2}\left(\frac{\hat{x}}{\hat{x}-\hat{\xi}}\right)\\
&\left.-2(\hat{x}+\hat{\xi})\left[-2+\ln\left(\frac{\hat{x}}{\hat{x}+\hat{\xi}}\right)\right]\ln\left(\frac{\hat{x}}{\hat{x}+\hat{\xi}}\right)\right\}. \tag*{(118)}
\end{aligned} T rem A , DR ( x ^ , ξ ^ ) ∼ − x ^ ( ξ ^ 2 − 1 ) 4 Q 2 { − 4 ( x ^ − ξ ^ 2 ) ln ( x ^ − 1 x ^ ) + ( − 1 + 2 x ^ − ξ ^ 2 ) ln 2 ( x ^ − 1 x ^ ) − 4 ( x ^ + ξ ^ 2 ) ln ( 1 + x ^ x ^ ) + ( 1 + 2 x ^ + ξ ^ 2 ) ln 2 ( 1 + x ^ x ^ ) + 4 ( x ^ − ξ ^ ) ln ( x ^ − ξ ^ x ^ ) − 2 ( x ^ − ξ ^ ) ln 2 ( x ^ − ξ ^ x ^ ) − 2 ( x ^ + ξ ^ ) [ − 2 + ln ( x ^ + ξ ^ x ^ ) ] ln ( x ^ + ξ ^ x ^ ) } . (118)
For t ≠ 0 t\ne0 t = 0 , the finite term in the notation of Ref. [22 ] reads:
T r e m A , t ( x ^ , ξ ^ ) ∼ 1 ( 1 − ξ ^ 2 ) 2 { ( 2 x ^ − 1 − ξ ^ 2 ) [ − ln ( x ^ − 1 x ^ ) + 1 2 ln 2 ( x ^ − 1 x ^ ) ] + 2 ( x ^ − ξ ^ ) ln ( x ^ − ξ ^ x ^ ) + ξ ^ ln 2 ( x ^ − ξ ^ x ^ ) − x ^ ln ( x ^ − ξ ^ x ^ ) ln ( x ^ + ξ ^ x ^ ) + 2 ξ ^ Li 2 ( − 2 ξ ^ x ^ − ξ ^ ) + ( 2 x ^ − 1 − ξ ^ 2 ) [ Li 2 ( 1 − ξ ^ 1 − x ^ ) + Li 2 ( 1 + ξ ^ 1 − x ^ ) ] − ( x ^ → − x ^ ) } . (119) \begin{aligned}
\mathcal{T}^{A,t}_{\mathrm{rem}}(\hat{x},\hat{\xi}) \sim{}&\frac{1}{(1-\hat{\xi}^{2})^{2}}\left\{(2\hat{x}-1-\hat{\xi}^{2})\left[-\ln\left(\frac{\hat{x}-1}{\hat{x}}\right)+\frac{1}{2}\ln^{2}\left(\frac{\hat{x}-1}{\hat{x}}\right)\right]\right.\\
&+2(\hat{x}-\hat{\xi})\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)+\hat{\xi}\ln^{2}\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)-\hat{x}\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)\ln\left(\frac{\hat{x}+\hat{\xi}}{\hat{x}}\right)\\
&\left.+2\hat{\xi}\operatorname{Li}_{2}\left(\frac{-2\hat{\xi}}{\hat{x}-\hat{\xi}}\right)+(2\hat{x}-1-\hat{\xi}^{2})\left[\operatorname{Li}_{2}\left(\frac{1-\hat{\xi}}{1-\hat{x}}\right)+\operatorname{Li}_{2}\left(\frac{1+\hat{\xi}}{1-\hat{x}}\right)\right]-(\hat{x}\to-\hat{x})\right\}. \tag*{(119)}
\end{aligned} T rem A , t ( x ^ , ξ ^ ) ∼ ( 1 − ξ ^ 2 ) 2 1 { ( 2 x ^ − 1 − ξ ^ 2 ) [ − ln ( x ^ x ^ − 1 ) + 2 1 ln 2 ( x ^ x ^ − 1 ) ] + 2 ( x ^ − ξ ^ ) ln ( x ^ x ^ − ξ ^ ) + ξ ^ ln 2 ( x ^ x ^ − ξ ^ ) − x ^ ln ( x ^ x ^ − ξ ^ ) ln ( x ^ x ^ + ξ ^ ) + 2 ξ ^ Li 2 ( x ^ − ξ ^ − 2 ξ ^ ) + ( 2 x ^ − 1 − ξ ^ 2 ) [ Li 2 ( 1 − x ^ 1 − ξ ^ ) + Li 2 ( 1 − x ^ 1 + ξ ^ ) ] − ( x ^ → − x ^ ) } . (119)
Moving onto the m ≠ 0 m\ne0 m = 0 case:
T r e m A , m ( x ^ , ξ ^ ) ∼ 4 ξ ^ 2 − 1 { ( 1 − 2 x ^ + ξ ^ 2 ) ln 2 ( x ^ − 1 ) + ln ( x ^ − 1 ) [ − 6 + 8 x ^ − 2 ξ ^ 2 + ( − 2 + 4 x ^ − 2 ξ ^ 2 ) ln x ^ ] + 2 [ 3 + 4 x ^ + ξ ^ 2 + ( 1 + 2 x ^ + ξ ^ 2 ) ln x ^ ] ln ( 1 + x ^ ) − ( 1 + 2 x ^ + ξ ^ 2 ) ln 2 ( 1 + x ^ ) − 4 ( x ^ − ξ ^ ) ( 2 + ln x ^ ) ln ( x ^ − ξ ^ ) + 2 ( x ^ − ξ ^ ) ln 2 ( x ^ − ξ ^ ) − 2 ( x ^ + ξ ^ ) [ 4 + 2 ln x ^ − ln ( x ^ + ξ ^ ) ] ln ( x ^ + ξ ^ ) } . (120) \begin{aligned}
\mathcal{T}^{A,m}_{\mathrm{rem}}(\hat{x},\hat{\xi}) \sim{}&\frac{4}{\hat{\xi}^{2}-1}\left\{(1-2\hat{x}+\hat{\xi}^{2})\ln^{2}(\hat{x}-1)+\ln(\hat{x}-1)\left[-6+8\hat{x}-2\hat{\xi}^{2}+(-2+4\hat{x}-2\hat{\xi}^{2})\ln\hat{x}\right]\right.\\
&+2\left[3+4\hat{x}+\hat{\xi}^{2}+(1+2\hat{x}+\hat{\xi}^{2})\ln\hat{x}\right]\ln(1+\hat{x})-(1+2\hat{x}+\hat{\xi}^{2})\ln^{2}(1+\hat{x})\\
&-4(\hat{x}-\hat{\xi})(2+\ln\hat{x})\ln(\hat{x}-\hat{\xi})+2(\hat{x}-\hat{\xi})\ln^{2}(\hat{x}-\hat{\xi})\\
&\left.-2(\hat{x}+\hat{\xi})\left[4+2\ln\hat{x}-\ln(\hat{x}+\hat{\xi})\right]\ln(\hat{x}+\hat{\xi})\right\}. \tag*{(120)}
\end{aligned} T rem A , m ( x ^ , ξ ^ ) ∼ ξ ^ 2 − 1 4 { ( 1 − 2 x ^ + ξ ^ 2 ) ln 2 ( x ^ − 1 ) + ln ( x ^ − 1 ) [ − 6 + 8 x ^ − 2 ξ ^ 2 + ( − 2 + 4 x ^ − 2 ξ ^ 2 ) ln x ^ ] + 2 [ 3 + 4 x ^ + ξ ^ 2 + ( 1 + 2 x ^ + ξ ^ 2 ) ln x ^ ] ln ( 1 + x ^ ) − ( 1 + 2 x ^ + ξ ^ 2 ) ln 2 ( 1 + x ^ ) − 4 ( x ^ − ξ ^ ) ( 2 + ln x ^ ) ln ( x ^ − ξ ^ ) + 2 ( x ^ − ξ ^ ) ln 2 ( x ^ − ξ ^ ) − 2 ( x ^ + ξ ^ ) [ 4 + 2 ln x ^ − ln ( x ^ + ξ ^ ) ] ln ( x ^ + ξ ^ ) } . (120)
Unpolarized case
For the unpolarized amplitude, the evolution kernel is
K e v o l ( x ^ , ξ ^ ) = 1 − 2 x ^ + 2 x ^ 2 − ξ ^ 2 ( 1 − ξ ^ 2 ) 2 ln ( x ^ − 1 x ^ ) + ( x ^ − ξ ^ ) ( 1 − 2 x ^ ξ ^ − ξ ^ 2 ) ξ ^ ( 1 − ξ ^ 2 ) 2 ln ( x ^ − ξ ^ x ^ ) + ( x ^ → − x ^ ) . (121) \mathcal{K}_{\mathrm{evol}}(\hat{x},\hat{\xi}) = \frac{1-2\hat{x}+2\hat{x}^{2}-\hat{\xi}^{2}}{(1-\hat{\xi}^{2})^{2}}\ln\left(\frac{\hat{x}-1}{\hat{x}}\right)+\frac{(\hat{x}-\hat{\xi})(1-2\hat{x}\hat{\xi}-\hat{\xi}^{2})}{\hat{\xi}(1-\hat{\xi}^{2})^{2}}\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)+(\hat{x}\to-\hat{x}). \tag*{(121)} K evol ( x ^ , ξ ^ ) = ( 1 − ξ ^ 2 ) 2 1 − 2 x ^ + 2 x ^ 2 − ξ ^ 2 ln ( x ^ x ^ − 1 ) + ξ ^ ( 1 − ξ ^ 2 ) 2 ( x ^ − ξ ^ ) ( 1 − 2 x ^ ξ ^ − ξ ^ 2 ) ln ( x ^ x ^ − ξ ^ ) + ( x ^ → − x ^ ) . (121)
This expression agrees with the previously reported unpolarized result [21 , 22 , 32 , 33 , 34 ] . As in the polarized case, the same kernel multiplies the infrared pole 1 / ϵ I R 1/\epsilon_{\mathrm{IR}} 1/ ϵ IR and the regulator logarithms in all three calculations. In each channel, this agreement across the three regularizations provides a cross-check of the complete one-loop calculation.
In the notation of Ref. [22 ] , the coefficient of the anomaly-related 1 / t 1/t 1/ t is given by:
R t S ( x ^ , ξ ^ ) ∼ 2 ⟨ p 2 ∣ F μ ν F μ ν ∣ p 1 ⟩ { 1 + 1 1 − ξ ^ 2 [ x ^ ( 1 − ξ ^ ) ln ( x ^ − 1 x ^ ) + x ^ ( x ^ − ξ ^ ) ln ( x ^ − ξ ^ x ^ ) + ( x ^ → − x ^ ) ] } . (122) \mathcal{R}^{S}_{t}(\hat{x},\hat{\xi})\sim2\langle p_{2}|F^{\mu\nu}F_{\mu\nu}|p_{1}\rangle\left\{1+\frac{1}{1-\hat{\xi}^{2}}\left[\hat{x}(1-\hat{\xi})\ln\left(\frac{\hat{x}-1}{\hat{x}}\right)+\hat{x}(\hat{x}-\hat{\xi})\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)+(\hat{x}\to-\hat{x})\right]\right\}. \tag*{(122)} R t S ( x ^ , ξ ^ ) ∼ 2 ⟨ p 2 ∣ F μν F μν ∣ p 1 ⟩ { 1 + 1 − ξ ^ 2 1 [ x ^ ( 1 − ξ ^ ) ln ( x ^ x ^ − 1 ) + x ^ ( x ^ − ξ ^ ) ln ( x ^ x ^ − ξ ^ ) + ( x ^ → − x ^ ) ] } . (122)
On the other hand, with the mass regulator, the coefficient of the 1 / m 2 1/m^{2} 1/ m 2 term reads:
R m S ( x ^ , ξ ^ ) ∼ − 8 ⟨ p 2 ∣ F μ ν F μ ν ∣ p 1 ⟩ 3 ξ ^ ( ξ ^ 2 − 1 ) 2 { ξ ^ + 4 x ^ 2 ξ ^ − 2 ξ ^ 3 − 4 x ^ 2 ξ ^ 3 + ξ ^ 5 − 2 x ^ ξ ^ ( x ^ − 1 ) 3 ln ( x ^ − 1 ) − 2 x ^ ξ ^ ( x ^ + 1 ) 3 ln ( x ^ + 1 ) \mathcal{R}^{S}_{m}(\hat{x},\hat{\xi})\sim-\frac{8\langle p_{2}|F^{\mu\nu}F_{\mu\nu}|p_{1}\rangle}{3\hat{\xi}(\hat{\xi}^{2}-1)^{2}}\left\{\hat{\xi}+4\hat{x}^{2}\hat{\xi}-2\hat{\xi}^{3}-4\hat{x}^{2}\hat{\xi}^{3}+\hat{\xi}^{5}-2\hat{x}\hat{\xi}(\hat{x}-1)^{3}\ln(\hat{x}-1)-2\hat{x}\hat{\xi}(\hat{x}+1)^{3}\ln(\hat{x}+1)
\right. R m S ( x ^ , ξ ^ ) ∼ − 3 ξ ^ ( ξ ^ 2 − 1 ) 2 8 ⟨ p 2 ∣ F μν F μν ∣ p 1 ⟩ { ξ ^ + 4 x ^ 2 ξ ^ − 2 ξ ^ 3 − 4 x ^ 2 ξ ^ 3 + ξ ^ 5 − 2 x ^ ξ ^ ( x ^ − 1 ) 3 ln ( x ^ − 1 ) − 2 x ^ ξ ^ ( x ^ + 1 ) 3 ln ( x ^ + 1 )
+ [ − 3 x ^ 3 + 6 x ^ 2 ξ ^ + 2 x ^ 4 ξ ^ − 3 x ^ ξ ^ 2 − 3 x ^ 3 ξ ^ 2 + x ^ ξ ^ 4 ] ln ( x ^ − ξ ^ ) + [ 3 x ^ 3 + 6 x ^ 2 ξ ^ + 2 x ^ 4 ξ ^ + 3 x ^ ξ ^ 2 + 3 x ^ 3 ξ ^ 2 − x ^ ξ ^ 4 ] ln ( x ^ + ξ ^ ) } . (123) \left.
+\left[-3\hat{x}^{3}+6\hat{x}^{2}\hat{\xi}+2\hat{x}^{4}\hat{\xi}-3\hat{x}\hat{\xi}^{2}-3\hat{x}^{3}\hat{\xi}^{2}+\hat{x}\hat{\xi}^{4}\right]\ln(\hat{x}-\hat{\xi})
+\left[3\hat{x}^{3}+6\hat{x}^{2}\hat{\xi}+2\hat{x}^{4}\hat{\xi}+3\hat{x}\hat{\xi}^{2}+3\hat{x}^{3}\hat{\xi}^{2}-\hat{x}\hat{\xi}^{4}\right]\ln(\hat{x}+\hat{\xi})\right\}. \tag*{(123)} + [ − 3 x ^ 3 + 6 x ^ 2 ξ ^ + 2 x ^ 4 ξ ^ − 3 x ^ ξ ^ 2 − 3 x ^ 3 ξ ^ 2 + x ^ ξ ^ 4 ] ln ( x ^ − ξ ^ ) + [ 3 x ^ 3 + 6 x ^ 2 ξ ^ + 2 x ^ 4 ξ ^ + 3 x ^ ξ ^ 2 + 3 x ^ 3 ξ ^ 2 − x ^ ξ ^ 4 ] ln ( x ^ + ξ ^ ) } . (123)
The coefficients of the 1 / t 1/t 1/ t and 1 / m 2 1/m^{2} 1/ m 2 terms multiply the same scalar gluonic tensor structure, which appears in the matrix element of F μ ν F μ ν F^{\mu\nu}F_{\mu\nu} F μν F μν and is associated with the trace anomaly. Keeping the gluon polarization vectors explicit makes this structure manifest in the residue. Setting ℓ μ = − 2 ξ p μ \ell^{\mu}=-2\xi p^{\mu} ℓ μ = − 2 ξ p μ with p 2 = 0 p^{2}=0 p 2 = 0 imposes t = ℓ 2 = 0 t=\ell^{2}=0 t = ℓ 2 = 0 from the outset, obscuring the 1 / t 1/t 1/ t or 1 / m 2 1/m^{2} 1/ m 2 behavior. As in the polarized case, contraction with a transverse projector can suppress the residue through its effect on the numerator. The Landau conditions continue to select the same boundary of the box integral. To summarize, both regulators expose this boundary, although the coefficient functions may differ.
Finally, we turn to the finite terms. The DR result contains the following coefficient function:
T r e m S , D R ( x ^ , ξ ^ ) ∼ 8 ξ ^ ( ξ ^ 2 − 1 ) { − 2 ( ξ ^ + ξ ^ 3 ) ln ( x ^ x ^ − 1 ) + ξ ^ ( − 1 + 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln 2 ( x ^ x ^ − 1 ) \mathcal{T}^{S,\mathrm{DR}}_{\mathrm{rem}}(\hat{x},\hat{\xi})\sim\frac{8}{\hat{\xi}(\hat{\xi}^{2}-1)}\left\{-2(\hat{\xi}+\hat{\xi}^{3})\ln\left(\frac{\hat{x}}{\hat{x}-1}\right)+\hat{\xi}(-1+2\hat{x}-2\hat{x}^{2}+\hat{\xi}^{2})\ln^{2}\left(\frac{\hat{x}}{\hat{x}-1}\right)
\right. T rem S , DR ( x ^ , ξ ^ ) ∼ ξ ^ ( ξ ^ 2 − 1 ) 8 { − 2 ( ξ ^ + ξ ^ 3 ) ln ( x ^ − 1 x ^ ) + ξ ^ ( − 1 + 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln 2 ( x ^ − 1 x ^ )
− 2 ξ ^ ( ξ ^ 2 − 1 ) ln ( x ^ 1 + x ^ ) + ξ ^ ( − 1 − 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln 2 ( x ^ 1 + x ^ ) \left.
-2\hat{\xi}(\hat{\xi}^{2}-1)\ln\left(\frac{\hat{x}}{1+\hat{x}}\right)+\hat{\xi}(-1-2\hat{x}-2\hat{x}^{2}+\hat{\xi}^{2})\ln^{2}\left(\frac{\hat{x}}{1+\hat{x}}\right)
\right. − 2 ξ ^ ( ξ ^ 2 − 1 ) ln ( 1 + x ^ x ^ ) + ξ ^ ( − 1 − 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln 2 ( 1 + x ^ x ^ )
+ 2 ( x ^ + ξ ^ − x ^ ξ ^ 2 + ξ ^ 3 ) ln ( x ^ x ^ − ξ ^ ) + ( − x ^ + ξ ^ + 2 x ^ 2 ξ ^ − x ^ ξ ^ 2 − ξ ^ 3 ) ln 2 ( x ^ x ^ − ξ ^ ) \left.
+2(\hat{x}+\hat{\xi}-\hat{x}\hat{\xi}^{2}+\hat{\xi}^{3})\ln\left(\frac{\hat{x}}{\hat{x}-\hat{\xi}}\right)+(-\hat{x}+\hat{\xi}+2\hat{x}^{2}\hat{\xi}-\hat{x}\hat{\xi}^{2}-\hat{\xi}^{3})\ln^{2}\left(\frac{\hat{x}}{\hat{x}-\hat{\xi}}\right)
\right. + 2 ( x ^ + ξ ^ − x ^ ξ ^ 2 + ξ ^ 3 ) ln ( x ^ − ξ ^ x ^ ) + ( − x ^ + ξ ^ + 2 x ^ 2 ξ ^ − x ^ ξ ^ 2 − ξ ^ 3 ) ln 2 ( x ^ − ξ ^ x ^ )
+ 4 ξ ^ ln ( x ^ − ξ ^ x ^ − 1 ) + 2 ( − x ^ − ξ ^ + x ^ ξ ^ 2 + ξ ^ 3 ) ln ( x ^ x ^ + ξ ^ ) + ( x ^ + ξ ^ + 2 x ^ 2 ξ ^ + x ^ ξ ^ 2 − ξ ^ 3 ) ln 2 ( x ^ x ^ + ξ ^ ) } . (124) \left.
+4\hat{\xi}\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}-1}\right)+2(-\hat{x}-\hat{\xi}+\hat{x}\hat{\xi}^{2}+\hat{\xi}^{3})\ln\left(\frac{\hat{x}}{\hat{x}+\hat{\xi}}\right)+(\hat{x}+\hat{\xi}+2\hat{x}^{2}\hat{\xi}+\hat{x}\hat{\xi}^{2}-\hat{\xi}^{3})\ln^{2}\left(\frac{\hat{x}}{\hat{x}+\hat{\xi}}\right)\right\}. \tag*{(124)} + 4 ξ ^ ln ( x ^ − 1 x ^ − ξ ^ ) + 2 ( − x ^ − ξ ^ + x ^ ξ ^ 2 + ξ ^ 3 ) ln ( x ^ + ξ ^ x ^ ) + ( x ^ + ξ ^ + 2 x ^ 2 ξ ^ + x ^ ξ ^ 2 − ξ ^ 3 ) ln 2 ( x ^ + ξ ^ x ^ ) } . (124)
For t ≠ 0 t\ne0 t = 0 , the finite term in the notation of Ref. [22 ] is:
T r e m S , t ( x ^ , ξ ^ ) ∼ 1 ( 1 − ξ ^ 2 ) 2 { ( 1 − 2 x ^ + 2 x ^ 2 − ξ ^ 2 ) [ − ln ( x ^ − 1 x ^ ) + 1 2 ln 2 ( x ^ − 1 x ^ ) ] \mathcal{T}^{S,t}_{\mathrm{rem}}(\hat{x},\hat{\xi})\sim\frac{1}{(1-\hat{\xi}^{2})^{2}}\left\{(1-2\hat{x}+2\hat{x}^{2}-\hat{\xi}^{2})\left[-\ln\left(\frac{\hat{x}-1}{\hat{x}}\right)+\frac{1}{2}\ln^{2}\left(\frac{\hat{x}-1}{\hat{x}}\right)\right]
\right. T rem S , t ( x ^ , ξ ^ ) ∼ ( 1 − ξ ^ 2 ) 2 1 { ( 1 − 2 x ^ + 2 x ^ 2 − ξ ^ 2 ) [ − ln ( x ^ x ^ − 1 ) + 2 1 ln 2 ( x ^ x ^ − 1 ) ]
− ( x ^ − ξ ^ ) ( 1 − 2 x ^ ξ ^ − ξ ^ 2 ) ξ ^ ln ( x ^ − ξ ^ x ^ ) + x ^ ( 1 + ξ ^ 2 ) 2 ξ ^ ln 2 ( x ^ − ξ ^ x ^ ) − 1 + 2 x ^ 2 − ξ ^ 2 2 ln ( x ^ − ξ ^ x ^ ) ln ( x ^ + ξ ^ x ^ ) \left.
-\frac{(\hat{x}-\hat{\xi})(1-2\hat{x}\hat{\xi}-\hat{\xi}^{2})}{\hat{\xi}}\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)+\frac{\hat{x}(1+\hat{\xi}^{2})}{2\hat{\xi}}\ln^{2}\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)-\frac{1+2\hat{x}^{2}-\hat{\xi}^{2}}{2}\ln\left(\frac{\hat{x}-\hat{\xi}}{\hat{x}}\right)\ln\left(\frac{\hat{x}+\hat{\xi}}{\hat{x}}\right)
\right. − ξ ^ ( x ^ − ξ ^ ) ( 1 − 2 x ^ ξ ^ − ξ ^ 2 ) ln ( x ^ x ^ − ξ ^ ) + 2 ξ ^ x ^ ( 1 + ξ ^ 2 ) ln 2 ( x ^ x ^ − ξ ^ ) − 2 1 + 2 x ^ 2 − ξ ^ 2 ln ( x ^ x ^ − ξ ^ ) ln ( x ^ x ^ + ξ ^ )
+ x ^ ( 1 + ξ ^ 2 ) ξ ^ Li 2 ( − 2 ξ ^ x ^ − ξ ^ ) + ( 1 − 2 x ^ + 2 x ^ 2 − ξ ^ 2 ) [ Li 2 ( 1 − ξ ^ 1 − x ^ ) + Li 2 ( 1 + ξ ^ 1 − x ^ ) ] + ( x ^ → − x ^ ) } . (125) \left.
+\frac{\hat{x}(1+\hat{\xi}^{2})}{\hat{\xi}}\operatorname{Li}_{2}\left(\frac{-2\hat{\xi}}{\hat{x}-\hat{\xi}}\right)+(1-2\hat{x}+2\hat{x}^{2}-\hat{\xi}^{2})\left[\operatorname{Li}_{2}\left(\frac{1-\hat{\xi}}{1-\hat{x}}\right)+\operatorname{Li}_{2}\left(\frac{1+\hat{\xi}}{1-\hat{x}}\right)\right]+(\hat{x}\to-\hat{x})\right\}. \tag*{(125)} + ξ ^ x ^ ( 1 + ξ ^ 2 ) Li 2 ( x ^ − ξ ^ − 2 ξ ^ ) + ( 1 − 2 x ^ + 2 x ^ 2 − ξ ^ 2 ) [ Li 2 ( 1 − x ^ 1 − ξ ^ ) + Li 2 ( 1 − x ^ 1 + ξ ^ ) ] + ( x ^ → − x ^ ) } . (125)
Finally, moving onto the m ≠ 0 m \ne0 m = 0 case, the finite terms read:
T r e m S , m ( x ^ , ξ ^ ) ∼ − 4 9 ξ ^ ( ξ ^ 2 − 1 ) { − 52 ξ ^ + 52 ξ ^ 3 − 9 ξ ^ ( − 1 + 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln 2 ( x ^ − 1 ) + 6 ξ ^ ln ( x ^ − 1 ) [ 2 ( x ^ − 1 ) 2 + ( − 3 + 6 x ^ − 6 x ^ 2 + 3 ξ ^ 2 ) ln x ^ ] + 12 ξ ^ ( 1 + 2 x ^ + x ^ 2 ) ln ( 1 + x ^ ) + 9 ξ ^ ( 1 + 2 x ^ + 2 x ^ 2 − ξ ^ 2 ) ln 2 ( 1 + x ^ ) + ( − 12 x ^ 2 ξ ^ + 24 x ^ ξ ^ 2 − 12 ξ ^ 3 ) ln ( x ^ − ξ ^ ) + ( 9 x ^ − 9 ξ ^ − 18 x ^ 2 ξ ^ + 9 x ^ ξ ^ 2 + 9 ξ ^ 3 ) ln 2 ( x ^ − ξ ^ ) + ( − 12 x ^ 2 ξ ^ − 24 x ^ ξ ^ 2 − 12 ξ ^ 3 ) ln ( x ^ + ξ ^ ) + ( − 9 x ^ − 9 ξ ^ − 18 x ^ 2 ξ ^ − 9 x ^ ξ ^ 2 + 9 ξ ^ 3 ) ln 2 ( x ^ + ξ ^ ) + 6 ln x ^ [ − 4 ξ ^ + 4 ξ ^ 3 + 3 ξ ^ ( − 1 − 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln ( 1 + x ^ ) − 3 ( x ^ − ξ ^ − 2 x ^ 2 ξ ^ + x ^ ξ ^ 2 + ξ ^ 3 ) ln ( x ^ − ξ ^ ) \mathcal{T}_{\mathrm{rem}}^{S,m}(\hat{x},\hat{\xi}) \sim-\frac{4}{9\hat{\xi}(\hat{\xi}^{2}-1)}\left\{
-52\hat{\xi}+52\hat{\xi}^{3}
-9\hat{\xi}(-1+2\hat{x}-2\hat{x}^{2}+\hat{\xi}^{2})\ln^{2}(\hat{x}-1)
+6\hat{\xi}\ln(\hat{x}-1)\left[2(\hat{x}-1)^{2}+(-3+6\hat{x}-6\hat{x}^{2}+3\hat{\xi}^{2})\ln\hat{x}\right]
+12\hat{\xi}(1+2\hat{x}+\hat{x}^{2})\ln(1+\hat{x})
+9\hat{\xi}(1+2\hat{x}+2\hat{x}^{2}-\hat{\xi}^{2})\ln^{2}(1+\hat{x})
+(-12\hat{x}^{2}\hat{\xi}+24\hat{x}\hat{\xi}^{2}-12\hat{\xi}^{3})\ln(\hat{x}-\hat{\xi})
+(9\hat{x}-9\hat{\xi}-18\hat{x}^{2}\hat{\xi}+9\hat{x}\hat{\xi}^{2}+9\hat{\xi}^{3})\ln^{2}(\hat{x}-\hat{\xi})
+(-12\hat{x}^{2}\hat{\xi}-24\hat{x}\hat{\xi}^{2}-12\hat{\xi}^{3})\ln(\hat{x}+\hat{\xi})
+(-9\hat{x}-9\hat{\xi}-18\hat{x}^{2}\hat{\xi}-9\hat{x}\hat{\xi}^{2}+9\hat{\xi}^{3})\ln^{2}(\hat{x}+\hat{\xi})
+6\ln\hat{x}\left[-4\hat{\xi}+4\hat{\xi}^{3}+3\hat{\xi}(-1-2\hat{x}-2\hat{x}^{2}+\hat{\xi}^{2})\ln(1+\hat{x})-3(\hat{x}-\hat{\xi}-2\hat{x}^{2}\hat{\xi}+\hat{x}\hat{\xi}^{2}+\hat{\xi}^{3})\ln(\hat{x}-\hat{\xi})\right.
\right. T rem S , m ( x ^ , ξ ^ ) ∼ − 9 ξ ^ ( ξ ^ 2 − 1 ) 4 { − 52 ξ ^ + 52 ξ ^ 3 − 9 ξ ^ ( − 1 + 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln 2 ( x ^ − 1 ) + 6 ξ ^ ln ( x ^ − 1 ) [ 2 ( x ^ − 1 ) 2 + ( − 3 + 6 x ^ − 6 x ^ 2 + 3 ξ ^ 2 ) ln x ^ ] + 12 ξ ^ ( 1 + 2 x ^ + x ^ 2 ) ln ( 1 + x ^ ) + 9 ξ ^ ( 1 + 2 x ^ + 2 x ^ 2 − ξ ^ 2 ) ln 2 ( 1 + x ^ ) + ( − 12 x ^ 2 ξ ^ + 24 x ^ ξ ^ 2 − 12 ξ ^ 3 ) ln ( x ^ − ξ ^ ) + ( 9 x ^ − 9 ξ ^ − 18 x ^ 2 ξ ^ + 9 x ^ ξ ^ 2 + 9 ξ ^ 3 ) ln 2 ( x ^ − ξ ^ ) + ( − 12 x ^ 2 ξ ^ − 24 x ^ ξ ^ 2 − 12 ξ ^ 3 ) ln ( x ^ + ξ ^ ) + ( − 9 x ^ − 9 ξ ^ − 18 x ^ 2 ξ ^ − 9 x ^ ξ ^ 2 + 9 ξ ^ 3 ) ln 2 ( x ^ + ξ ^ ) + 6 ln x ^ [ − 4 ξ ^ + 4 ξ ^ 3 + 3 ξ ^ ( − 1 − 2 x ^ − 2 x ^ 2 + ξ ^ 2 ) ln ( 1 + x ^ ) − 3 ( x ^ − ξ ^ − 2 x ^ 2 ξ ^ + x ^ ξ ^ 2 + ξ ^ 3 ) ln ( x ^ − ξ ^ )
+ 3 x ^ ln ( x ^ + ξ ^ ) + 3 ξ ^ ln ( x ^ + ξ ^ ) + 6 x ^ 2 ξ ^ ln ( x ^ + ξ ^ ) + 3 x ^ ξ ^ 2 ln ( x ^ + ξ ^ ) − 3 ξ ^ 3 ln ( x ^ + ξ ^ ) ] } . (126) \left.
\left.
+3\hat{x}\ln(\hat{x}+\hat{\xi})+3\hat{\xi}\ln(\hat{x}+\hat{\xi})+6\hat{x}^{2}\hat{\xi}\ln(\hat{x}+\hat{\xi})+3\hat{x}\hat{\xi}^{2}\ln(\hat{x}+\hat{\xi})-3\hat{\xi}^{3}\ln(\hat{x}+\hat{\xi})\right]
\right\}. \tag*{(126)} + 3 x ^ ln ( x ^ + ξ ^ ) + 3 ξ ^ ln ( x ^ + ξ ^ ) + 6 x ^ 2 ξ ^ ln ( x ^ + ξ ^ ) + 3 x ^ ξ ^ 2 ln ( x ^ + ξ ^ ) − 3 ξ ^ 3 ln ( x ^ + ξ ^ ) ] } . (126)
Conclusions
We have traced the anomaly-related poles in the gluon channel of double deeply virtual Compton scattering (DDVCS) to a subleading Landau singularity of the one-loop box amplitude. The singularity lies on the boundary of the Feynman-parameter domain where the parameter associated with the internal quark propagator joining the two photon vertices vanishes. This propagator is inactive in the Landau equations, and the other three form a triangle subdiagram. The triangle (see Figure 2 ) thus identifies the specific boundary configuration of the four-point integral responsible for the poles.
We identify the same pole-producing boundary of the one-loop box across three infrared prescriptions: dimensional regularization (DR), nonzero momentum transfer (t ≠ 0 t \ne0 t = 0 ), and finite quark mass (m ≠ 0 m \ne0 m = 0 ). At t = m = 0 t=m=0 t = m = 0 , the box endpoint has a power (linear) infrared divergence in four dimensions. By analytic continuation, DR conceals this power divergence, which produces no 1 / ϵ 1/\epsilon 1/ ϵ pole in the expansion about d = 4 d=4 d = 4 . If evaluated in six dimensions, the same endpoint would diverge logarithmically, giving a pole in the expansion about d = 6 d=6 d = 6 . Nonzero t t t and finite m m m expose this boundary contribution through the known 1 / t 1/t 1/ t anomaly-related pole and a previously unreported 1 / m 2 1/m^{2} 1/ m 2 behavior, respectively. After summing the contributing diagrams, we find that the two inverse-power terms carry the same anomaly-related residue. Table I summarizes the boundary scaling for all three infrared prescriptions. These results establish the triangle boundary of the one-loop box as the common geometric origin of the anomaly-related poles.
Table I. Schematic behavior of the four-point coefficient functions near the triangle boundary z = 0 z=0 z = 0 , where z z z is the Feynman parameter of the quark propagator joining the two photon vertices and z 0 z_{0} z 0 is the upper cutoff defining the near-boundary region. After summing the contributing diagrams, the t t t - and m m m -regulated residues contain the same anomaly operator structure, with potentially different coefficient functions.
We also show that the pole residue depends on how the external gluon polarizations are treated. One can retain the polarization vectors explicitly or contract the two gluon Lorentz indices of the box amplitude with a transverse projector. If one uses the transverse projector, the box numerator can be rewritten in terms of inverse propagators. These factors cancel the corresponding propagators in the integrand, suppressing the pole residue associated with the box boundary and leaving bubble and triangle integrals (the latter do not generate the anomaly-related pole). Retaining explicit polarization vectors preserves the tensor structures associated with the scalar and pseudoscalar gluonic operators, allowing us to identify the corresponding anomaly-related terms. This projection dependence may help explain why some of these terms were absent in earlier calculations.
We also obtained the complete one-loop results for polarized and unpolarized DDVCS with each infrared prescription. After summing the box, cat’s-eye, and hourglass diagrams, we find that, within each polarization channel, the same evolution kernel governs the infrared pole in DR and the logarithms ln ( Q 2 / ( − t ) ) \ln(Q^2/(-t)) ln ( Q 2 / ( − t )) and ln ( Q 2 / m 2 ) \ln(Q^2/m^2) ln ( Q 2 / m 2 ) . The recovery of the same kernel with all three prescriptions provides a nontrivial consistency check of our one-loop results.
In conclusion, we have identified the triangle boundary of the one-loop box as the common geometric origin of the anomaly-related poles. We leave a systematic study of the joint t , m → 0 t,m \to0 t , m → 0 limit, its relation to DR, and the analytic continuation of the associated loop functions to future work. Our results provide a starting point for identifying analogous poles in other hard-exclusive processes.
Acknowledgment
S. B. thanks Yoshitaka Hatta for numerous helpful suggestions that improved the presentation of this manuscript. S. B. also gratefully acknowledges stimulating discussions with Yoshitaka Hatta, Zhong-Bo Kang, Dmitri Kharzeev, Raju Venugopalan, and Werner Vogelsang. This work was supported by the U.S. National Science Foundation under Grant No. PHY-2609761. A. Y. gratefully acknowledges support from the Mark Miller Award of the Department of Physics at the University of Connecticut, which made this undergraduate research contribution possible.
Passarino–Veltman functions
Passarino–Veltman reduction expresses a one-loop tensor integral as a sum of Lorentz tensors multiplied by scalar coefficient functions. The letters A A A , B B B , C C C , and D D D label integrals with one, two, three, and four propagators, respectively. We use a two-point (B B B -function) integral to illustrate how the tensor coefficients and their indices are defined.
For the routing in Figure 4 , define
Figure 4. Two-point, or bubble, topology defining the B B B -function family. The external momentum satisfies p 2 = s p^2=s p 2 = s ; the two internal propagators have masses m 0 m_0 m 0 and m 1 m_1 m 1 .
D 0 = k 2 − m 0 2 + i 0 , D 1 = ( k + p ) 2 − m 1 2 + i 0. (127) D_0 = k^2 - m_0^2 + i0,\qquad D_1 = (k+p)^2 - m_1^2 + i0. \tag*{(127)} D 0 = k 2 − m 0 2 + i 0 , D 1 = ( k + p ) 2 − m 1 2 + i 0. (127)
Here i 0 i0 i 0 denotes i ε i\varepsilon i ε with ε → 0 + \varepsilon\to0^+ ε → 0 + , using the same causal prescription as in the Feynman-parameter representations. The scalar B B B -function is the loop integral with numerator 1 1 1 and denominator D 0 D 1 D_0D_1 D 0 D 1 . Tensor B B B -functions have factors of the loop momentum k μ k^\mu k μ in the numerator. We write the scalar and tensor integrals as
B 0 ( s ; m 0 , m 1 ) ≡ ∫ k 1 D 0 D 1 , (128) B_0(s;m_0,m_1) \equiv\int_k \frac{1}{D_0D_1}, \tag*{(128)} B 0 ( s ; m 0 , m 1 ) ≡ ∫ k D 0 D 1 1 , (128)
B μ ( p ) ≡ ∫ k k μ D 0 D 1 , (129) B^\mu(p) \equiv\int_k \frac{k^\mu}{D_0D_1}, \tag*{(129)} B μ ( p ) ≡ ∫ k D 0 D 1 k μ , (129)
B μ ν ( p ) ≡ ∫ k k μ k ν D 0 D 1 . (130) B^{\mu\nu}(p) \equiv\int_k \frac{k^\mu k^\nu}{D_0D_1}. \tag*{(130)} B μν ( p ) ≡ ∫ k D 0 D 1 k μ k ν . (130)
Here ∫ k \int_k ∫ k denotes the dimensionally regulated loop integration with the normalization used in Eq. (5) . Since p μ p^\mu p μ is the only external vector in a two-point integral, Lorentz covariance gives
B μ ( p ) = p μ B 1 ( s ; m 0 , m 1 ) , (131) B^\mu(p) = p^\mu B_1(s;m_0,m_1), \tag*{(131)} B μ ( p ) = p μ B 1 ( s ; m 0 , m 1 ) , (131)
B μ ν ( p ) = g μ ν B 00 ( s ; m 0 , m 1 ) + p μ p ν B 11 ( s ; m 0 , m 1 ) . (132) B^{\mu\nu}(p) = g^{\mu\nu}B_{00}(s;m_0,m_1) + p^\mu p^\nu B_{11}(s;m_0,m_1). \tag*{(132)} B μν ( p ) = g μν B 00 ( s ; m 0 , m 1 ) + p μ p ν B 11 ( s ; m 0 , m 1 ) . (132)
Thus, B 1 B_1 B 1 is the scalar coefficient multiplying one external momentum. At rank two, B 00 B_{00} B 00 multiplies the metric tensor, while B 11 B_{11} B 11 multiplies two external momenta. The scalar function B 0 B_0 B 0 has no Lorentz indices. Thus, the subscripts record these Lorentz structures. In the notation of Eq. (2) , each pair of zero indices represents one metric tensor, and each index 1 represents one factor of p μ p^\mu p μ . The following examples make the counting explicit:
Table A.
Here r r r counts metric tensors and n 1 n_1 n 1 counts factors of the external momentum. The tensor rank is 2 r + n 1 2r+n_1 2 r + n 1 .
The parameter representation follows by combining the two propagators:
1 D 0 D 1 = ∫ 0 1 d x 1 [ ( k + x p ) 2 − F B ( x ) + i 0 ] 2 , (133) \frac{1}{D_0D_1}=\int_0^1\mathrm{d}x\,\frac{1}{\left[(k+xp)^2-F_B(x)+\mathrm{i}0\right]^2}, \tag*{(133)} D 0 D 1 1 = ∫ 0 1 d x [ ( k + x p ) 2 − F B ( x ) + i 0 ] 2 1 , (133)
where
F B ( x ) = s x 2 + ( − s + m 1 2 − m 0 2 ) x + m 0 2 . (134) F_B(x)=sx^2+(-s+m_1^2-m_0^2)x+m_0^2. \tag*{(134)} F B ( x ) = s x 2 + ( − s + m 1 2 − m 0 2 ) x + m 0 2 . (134)
After shifting the loop momentum to q = k + x p q=k+xp q = k + x p , the rank-one numerator becomes
k μ = q μ − x p μ . (135) k^\mu=q^\mu-xp^\mu. \tag*{(135)} k μ = q μ − x p μ . (135)
The term proportional to q μ q^\mu q μ vanishes upon integration because it is odd in the shifted loop momentum. The remaining term is proportional to − x p μ -xp^\mu − x p μ . This explains why the parameter representation for B 1 B_1 B 1 contains one power of x x x , with a minus sign relative to B 0 B_0 B 0 . At rank two, the numerator contains
k μ k ν = q μ q ν − x ( q μ p ν + p μ q ν ) + x 2 p μ p ν . k^\mu k^\nu=q^\mu q^\nu-x(q^\mu p^\nu+p^\mu q^\nu)+x^2p^\mu p^\nu. k μ k ν = q μ q ν − x ( q μ p ν + p μ q ν ) + x 2 p μ p ν .
The terms linear in q μ q^\mu q μ again vanish. The q μ q ν q^\mu q^\nu q μ q ν term generates the metric contribution B 00 B_{00} B 00 , while the x 2 p μ p ν x^2p^\mu p^\nu x 2 p μ p ν term generates B 11 B_{11} B 11 . For completeness, the resulting coefficients in the normalization of Eq. (5) are
B 0 ( s ; m 0 , m 1 ) = μ 2 ϵ e γ E ϵ Γ ( ϵ ) ∫ 0 1 d x [ F B ( x ) − i ϵ ] − ϵ , B_0(s;m_0,m_1)=\mu^{2\epsilon}e^{\gamma_E\epsilon}\Gamma(\epsilon)\int_0^1\mathrm{d}x\,[F_B(x)-\mathrm{i}\epsilon]^{-\epsilon}, B 0 ( s ; m 0 , m 1 ) = μ 2 ϵ e γ E ϵ Γ ( ϵ ) ∫ 0 1 d x [ F B ( x ) − i ϵ ] − ϵ ,
B 1 ( s ; m 0 , m 1 ) = − μ 2 ϵ e γ E ϵ Γ ( ϵ ) ∫ 0 1 d x x [ F B ( x ) − i ϵ ] − ϵ , B_1(s;m_0,m_1)=-\mu^{2\epsilon}e^{\gamma_E\epsilon}\Gamma(\epsilon)\int_0^1\mathrm{d}x\,x[F_B(x)-\mathrm{i}\epsilon]^{-\epsilon}, B 1 ( s ; m 0 , m 1 ) = − μ 2 ϵ e γ E ϵ Γ ( ϵ ) ∫ 0 1 d x x [ F B ( x ) − i ϵ ] − ϵ ,
B 00 ( s ; m 0 , m 1 ) = − μ 2 ϵ e γ E ϵ 2 Γ ( ϵ − 1 ) ∫ 0 1 d x [ F B ( x ) − i ϵ ] 1 − ϵ , B_{00}(s;m_0,m_1)=-\frac{\mu^{2\epsilon}e^{\gamma_E\epsilon}}{2}\Gamma(\epsilon-1)\int_0^1\mathrm{d}x\,[F_B(x)-\mathrm{i}\epsilon]^{1-\epsilon}, B 00 ( s ; m 0 , m 1 ) = − 2 μ 2 ϵ e γ E ϵ Γ ( ϵ − 1 ) ∫ 0 1 d x [ F B ( x ) − i ϵ ] 1 − ϵ ,
B 11 ( s ; m 0 , m 1 ) = μ 2 ϵ e γ E ϵ Γ ( ϵ ) ∫ 0 1 d x x 2 [ F B ( x ) − i ϵ ] − ϵ . (136) B_{11}(s;m_0,m_1)=\mu^{2\epsilon}e^{\gamma_E\epsilon}\Gamma(\epsilon)\int_0^1\mathrm{d}x\,x^2[F_B(x)-\mathrm{i}\epsilon]^{-\epsilon}. \tag*{(136)} B 11 ( s ; m 0 , m 1 ) = μ 2 ϵ e γ E ϵ Γ ( ϵ ) ∫ 0 1 d x x 2 [ F B ( x ) − i ϵ ] − ϵ . (136)
These are the cases r = 0 , n 1 = 0 r=0,n_1=0 r = 0 , n 1 = 0 , r = 0 , n 1 = 1 r=0,n_1=1 r = 0 , n 1 = 1 , r = 1 , n 1 = 0 r=1,n_1=0 r = 1 , n 1 = 0 , and r = 0 , n 1 = 2 r=0,n_1=2 r = 0 , n 1 = 2 , respectively, of the general two-point representation in Eq. (5) .
The same index-counting rule applies to the three- and four-point functions. Their indices 1, 2, 3 label the independent external momenta used in the tensor decomposition. For example, D 00123 D_{00123} D 00123 contains one pair of zero indices, corresponding to one metric tensor, and one factor of each of the momenta labelled 1, 2, 3. Its tensor rank is therefore five. In the parameter representation of Eq. (10) , the momentum labels 1, 2, 3 correspond to the factors x x x , y y y , z z z , respectively, so this coefficient carries the numerator factor x y z xyz x y z .
References