Introduction
Two-loop amplitudes with final-state jets are among the missing ingredients needed for precision Higgs studies at the Large Hadron Collider (LHC) [1 ] . In this paper we study the two-loop amplitudes for the process
q ( p 1 ) + q ˉ ( p 2 ) → g ( p 3 ) + h ( q 1 ) … h ( q m ) , (1) q(p_1) + \bar{q}(p_2) \to g(p_3) + h(q_1) \ldots h(q_m), \tag*{(1)} q ( p 1 ) + q ˉ ( p 2 ) → g ( p 3 ) + h ( q 1 ) … h ( q m ) , (1)
and for its crossed channel, q + g → q + m H q+g \to q+mH q + g → q + m H . To carry out the two-loop integral numerically in four dimensions, one first has to rewrite the integrand in a form that is infrared (IR) finite.
Loop amplitudes in massless gauge theory are studied using an infrared regulator. A common choice is dimensional regularization in 4 + 2 ϵ 4 + 2\epsilon 4 + 2 ϵ dimensions. Infrared divergences of the amplitude appear as poles in ϵ \epsilon ϵ when using such a regulator. While final-state singularities cancel after summing over all final states (including unresolved parton phase space), initial-state singularities require collinear subtractions in the cross section [2 ] . Simple two-loop integrals with a small number of mass scales can be carried out analytically in terms of known functions using dimensional regularization, more complicated integrals with multiple scales remain a bottle neck at two-loops and beyond [3 ] .
An alternative is to remove the IR divergences before integration, so that the remaining loop integrand can be integrated numerically in exactly d = 4 d = 4 d = 4 . Related numerical and local approaches to loop integrals have been developed in Refs. [4 , 5 , 6 , 7 , 8 , 9 ] .
One way to deal with final-state singularities is the use of time-ordered perturbation theory [10 ] and the related family of loop-tree duality methods, wherein final-state cancellations can be made manifestly local in momentum space using the unitarity algebra [11 , 12 , 13 , 14 , 15 , 16 , 17 ] .
For real radiation, a class of covariant local subtraction algorithms handles both initial- and final-state singularities in cross sections [18 , 19 , 20 , 21 , 22 , 23 ] .
Focusing on amplitude-level methods, particularly relevant for fully differential predictions with electroweak bosons and jets in the final-state, one would like a locally finite amplitude integrand when the number of scales becomes large [24 ] . One systematic approach to constructing such integrands in gauge theories is the local finiteness program of Anastasiou and Sterman. The authors identify pinch surfaces using the Landau equations [25 ] , which have a simple interpretation in terms of soft lines and jets [26 , 27 , 28 , 29 ] , and construct local counterterms near each pinch surface. The remainder is finite in four dimensions, while the counterterms are simple enough to integrate analytically. The idea was first demonstrated for massless 2 → 2 2 \to2 2 → 2 master integrals [30 ] , then for two-loop QED amplitudes with several off-shell photons [31 ] , and subsequently for two-loop QCD amplitudes for q q ˉ → V 1 … V n q\bar{q} \to V_1 \ldots V_n q q ˉ → V 1 … V n [32 ] . It has since been extended to gluon-initiated electroweak and multi-Higgs production through massive quark loops [33 ] and to real-virtual NNLO corrections [34 ] . In the QED and QCD constructions, local Ward identities are needed to line up the integrands of different diagrams. The loop-momentum shifts required for this are compensated by explicit “shift counterterms”, and the subtractions are constructed to be local in loop momentum near each pinch surface. A main drawback of this method is the need for increasingly complicated shift counterterms and, at higher loops, the treatment of jet-soft interactions [35 ] .
In this paper, we study a different construction for the next step in this program: two-loop amplitudes for multi-Higgs production through a massive top-quark loop in association with one jet. We consider the quark-gluon channel with a final-state quark, q g → q + m H qg \to q + m_H q g → q + m H , and the crossed quark-antiquark channel with a final-state gluon, q q ˉ → g + m H q\bar{q} \to g + m_H q q ˉ → g + m H . Our starting point is the G-K decomposition, used in the spirit of the Grammer–Yennie method in QED [36 ] , and applied through exact identities between Feynman graphs as suggested in Ref. [37 ] . Using Ward identities between Feynman graphs, without making any reference to an infrared region, we separate the longitudinal and gauge degrees of freedom of the internal gluons from the rest. We then isolate the contribution involving only the physical degrees of freedom of internal gluons. We show that this contribution is free of collinear singularities. The only remaining IR divergence is an overall soft singularity, which we identify with a change of the reference gauge vector, and we construct explicit soft counterterms for it. This gives a locally IR-finite representation of the physical contribution. As a check, we verify that the factorized longitudinal degrees of freedom reproduce the remaining IR poles of the original Feynman graphs.
The main difference from the Anastasiou–Sterman construction is that our method decomposes the amplitude using exact identities that are non-local in momentum space. As a result, we require no shift counterterms. We use the analysis of individual IR regions only to establish that the modified integrand, obtained from our physical-gauge Feynman rules, is finite in collinear limits, without having to construct counterterms from the limits themselves. We expect this method to carry over to more complicated processes and to eliminate all collinear singularities there.
The paper is organized as follows. In § 2 , we explain the kinematics, introduce a convenient graph classification, and describe some useful graphical tools. In § 3 , we use our modified Feynman rules to eliminate “transient singularities”—singularities that appear in individual graphs but vanish after summing over a gauge-invariant subset. In § 4 , we consider the remaining genuinely IR-divergent graphs, which are the main focus of this paper. We show that the two-sided physical-propagator GG terms are collinear finite and construct their soft counterterms. We then show that the GK and KK terms factorize into products of lower-loop amplitudes and can be evaluated consistently. In App. § 7 , we verify that the longitudinal contributions reproduce the IR poles of the original graphs. Finally, § 5 contains our conclusions.
Figure 1. Set of Feynman graphs that describe the process of interest at one loop. Throughout, we will utilize H \mathcal{H} H to represent a fermion loop, which consists of all topologically distinct permutations of attachments on the top loop.
Kinematics and conventions
In this section, we will describe the process of interest, list families of graphs that contribute to it, and finally end with a definition of the scalar-scalar-gluon decomposition as well as the G-K decomposition which will be our main tools in the sections that follow.
In this paper, we will study the process
q ( p 1 , s 1 , i 1 ) + q ˉ ( p 2 , s 2 , i 2 ) → g ( p 3 , ϵ 3 , x ) + h ( q 1 ) ⋯ h ( q m ) , (2) q(p_1,s_1,i_1) + \bar{q}(p_2,s_2,i_2) \to g(p_3,\epsilon_3,x) + h(q_1)\cdots h(q_m), \tag*{(2)} q ( p 1 , s 1 , i 1 ) + q ˉ ( p 2 , s 2 , i 2 ) → g ( p 3 , ϵ 3 , x ) + h ( q 1 ) ⋯ h ( q m ) , (2)
where s 1 , s 2 s_1,s_2 s 1 , s 2 are the spins of the quark and anti-quark and i 1 , i 2 i_1,i_2 i 1 , i 2 are the color indices of the quark and anti-quark respectively. We will denote by ϵ 3 μ ( p 3 ; p 1 ) \epsilon_3^\mu(p_3;p_1) ϵ 3 μ ( p 3 ; p 1 ) , the polarization vector of the gluon which satisfies ϵ 3 μ ( p 3 ; p 1 ) ⋅ p 3 = ϵ 3 μ ( p 3 ; p 1 ) ⋅ p 1 = 0 \epsilon_3^\mu(p_3;p_1)\cdot p_3=\epsilon_3^\mu(p_3;p_1)\cdot p_1=0 ϵ 3 μ ( p 3 ; p 1 ) ⋅ p 3 = ϵ 3 μ ( p 3 ; p 1 ) ⋅ p 1 = 0 . We will also consistently use x x x for the (adjoint) color index of the external gluon. All partons will be assumed to be massless, p i 2 = 0 p_i^2=0 p i 2 = 0 . The one-loop Feynman graphs that contribute to the process are displayed in Fig. 1. It will be useful to represent the fermion loop by the number of distinct electroweak and gluon attachments. If we denote by H μ 1 … μ n a 1 … a n [ n , m ] ( p 1 … p n , q 1 … q m ) \mathcal{H}^{a_1\ldots a_n}_{\mu_1\ldots\mu_n}[n,m](p_1\ldots p_n,q_1\ldots q_m) H μ 1 … μ n a 1 … a n [ n , m ] ( p 1 … p n , q 1 … q m ) the single fermion loop and by M ( L ) ( p 1 , p 2 , p 3 , { q i } ; s 1 , s 2 , ϵ 3 , i 1 , i 2 , x ) \mathcal{M}^{(L)}(p_1,p_2,p_3,\{q_i\};s_1,s_2,\epsilon_3,i_1,i_2,x) M ( L ) ( p 1 , p 2 , p 3 , { q i } ; s 1 , s 2 , ϵ 3 , i 1 , i 2 , x ) the L L L loop amplitude, which we will sometimes shorten to M ( L ) \mathcal{M}^{(L)} M ( L ) , the one loop amplitude is given by
M ( 1 ) = ϵ 3 μ 2 ( p 3 ; p 1 ) H μ 1 μ 2 a , x [ 2 , m ] ( P , − p 3 , q 1 … q m ) − i δ ν μ 1 P 2 + i 0 v ˉ ( p 2 , s 2 ) − i g s T a γ ν u ( p 1 , s 1 ) . (3) \mathcal{M}^{(1)}=\epsilon_3^{\mu_2}(p_3;p_1)\mathcal{H}^{a,x}_{\mu_1\mu_2}[2,m](P,-p_3,q_1\ldots q_m)\frac{-i\delta^{\mu_1}_{\nu}}{P^2+i0}\bar{v}(p_2,s_2)-ig_sT^{a}\gamma^\nu u(p_1,s_1). \tag*{(3)} M ( 1 ) = ϵ 3 μ 2 ( p 3 ; p 1 ) H μ 1 μ 2 a , x [ 2 , m ] ( P , − p 3 , q 1 … q m ) P 2 + i 0 − i δ ν μ 1 v ˉ ( p 2 , s 2 ) − i g s T a γ ν u ( p 1 , s 1 ) . (3)
Figure 2. Two examples of topologically factorized graphs. (a) is a vertex correction of the current which factorizes from the fermion loop on account of the intermediate gluon propagator being a one particle reducible cut. (b) represents the self-energy of the intermediate gluon which also factorizes from the fermion loop.
Throughout this paper, the fermion-loop function H \mathcal{H} H includes all coupling factors, fermion propagators, and sums over distinct permutations of gluon and Higgs attachments within the massive top-quark loop. The fermion loop is defined with incoming gluons and outgoing electroweak bosons. We will also find it useful to use the shorthand P = p 1 + p 2 P = p_1 + p_2 P = p 1 + p 2 . As an example, H [ 2 , 1 ] \mathcal{H}[2,1] H [ 2 , 1 ] is simply the familiar one-loop top-quark contribution to Higgs production through gluon fusion, with the two gluons allowed to be off shell. Explicitly,
H μ ν a b [ 2 , 1 ] ( k 1 , k 2 ; q ) = − ∫ d d ℓ ( 2 π ) d Tr c , D [ ( − i g s T a γ μ ) S t ( ℓ + k 1 ) ( − i g s T b γ ν ) S t ( ℓ + q ) ( − i y t ) S t ( ℓ ) ] + ( ( a , μ , k 1 ) ↔ ( b , ν , k 2 ) ) . (4) \mathcal{H}^{ab}_{\mu\nu}[2,1](k_1,k_2;q)=-\int\frac{\mathrm{d}^{d}\ell}{(2\pi)^{d}}\operatorname{Tr}_{c,D}\left[(-\mathrm{i}g_sT^a\gamma_\mu)S_t(\ell+k_1)(-\mathrm{i}g_sT^b\gamma_\nu)S_t(\ell+q)(-\mathrm{i}y_t)S_t(\ell)\right]+((a,\mu,k_1)\leftrightarrow(b,\nu,k_2)). \tag*{(4)} H μν ab [ 2 , 1 ] ( k 1 , k 2 ; q ) = − ∫ ( 2 π ) d d d ℓ Tr c , D [ ( − i g s T a γ μ ) S t ( ℓ + k 1 ) ( − i g s T b γ ν ) S t ( ℓ + q ) ( − i y t ) S t ( ℓ ) ] + (( a , μ , k 1 ) ↔ ( b , ν , k 2 )) . (4)
where q = k 1 + k 2 q=k_1+k_2 q = k 1 + k 2 , y t y_t y t , and S t ( k ) = i ( k̸ + m t ) / ( k 2 − m t 2 + i 0 ) S_t(k)=\mathrm{i}(\not{k}+m_t)/(k^2-m_t^2+\mathrm{i}0) S t ( k ) = i ( k + m t ) / ( k 2 − m t 2 + i 0 ) is the massive top-quark propagator. The trace runs over Dirac and color indices and the minus sign comes from the fermion loop.
Let us now classify all graphs that appear in the two-loop amplitude. First, we decompose the two-loop amplitude into three pieces:
M ( 2 ) = M I R ( 2 ) + M I R ‾ ( 2 ) + M f a c t ( 2 ) . (5) \mathcal{M}^{(2)}=\mathcal{M}_{\mathrm{IR}}^{(2)}+\mathcal{M}_{\overline{\mathrm{IR}}}^{(2)}+\mathcal{M}_{\mathrm{fact}}^{(2)}. \tag*{(5)} M ( 2 ) = M IR ( 2 ) + M IR ( 2 ) + M fact ( 2 ) . (5)
The first two terms contain connected two-loop graphs. We distinguish between graphs in M I R ( 2 ) \mathcal{M}_{\mathrm{IR}}^{(2)} M IR ( 2 ) , which require an explicit treatment of their infrared singularities, and graphs in M I R ‾ ( 2 ) \mathcal{M}_{\overline{\mathrm{IR}}}^{(2)} M IR ( 2 ) , whose infrared singularities cancel after summing over the appropriate families of graphs using Ward identities. The graphs in M f a c t ( 2 ) \mathcal{M}_{\mathrm{fact}}^{(2)} M fact ( 2 ) are topologically factorized. They satisfy
M f a c t ( 2 ) = ∑ G f a c t = G 1 × G 2 [ ∫ d d k ( 2 π ) d G 1 ( k ) ] [ ∫ d d ℓ ( 2 π ) d G 2 ( ℓ ) ] . (6) \mathcal{M}_{\mathrm{fact}}^{(2)}=\sum_{G_{\mathrm{fact}}=G_1\times G_2}\left[\int\frac{\mathrm{d}^{d}k}{(2\pi)^{d}}G_1(k)\right]\left[\int\frac{\mathrm{d}^{d}\ell}{(2\pi)^{d}}G_2(\ell)\right]. \tag*{(6)} M fact ( 2 ) = G fact = G 1 × G 2 ∑ [ ∫ ( 2 π ) d d d k G 1 ( k ) ] [ ∫ ( 2 π ) d d d ℓ G 2 ( ℓ ) ] . (6)
Two representative examples of topologically factorized graphs are drawn in Fig. 2.
After introducing the scalar-scalar-gluon decomposition, we will be able to classify the remaining graphs into M I R ( 2 ) \mathcal{M}_{\mathrm{IR}}^{(2)} M IR ( 2 ) , M I R ‾ ( 2 ) \mathcal{M}_{\overline{\mathrm{IR}}}^{(2)} M IR ( 2 ) which we turn to now.
Scalar-Scalar-Gluon decomposition
Let us now discuss the first of two tools that we will make use of extensively. It will be convenient to decompose the triple gluon vertex into its three constituent terms. Such a decomposition has already appeared in the literature in [33 ] and was called the scalar-scalar gluon decomposition there. In conventions where all three momenta are incoming into a three gluon vertex, the momenta satisfy
k 1 + k 2 + k 3 = 0. (7) k_1+k_2+k_3=0. \tag*{(7)} k 1 + k 2 + k 3 = 0. (7)
The full three gluon vertex is given by
V α β γ a b c ( k 1 , k 2 , k 3 ) = − g s f a b c [ g α β ( k 1 − k 2 ) γ + g β γ ( k 2 − k 3 ) α + g γ α ( k 3 − k 1 ) β ] . (8) V^{abc}_{\alpha\beta\gamma}(k_1,k_2,k_3)=-g_s f^{abc}\left[g_{\alpha\beta}(k_1-k_2)_\gamma+g_{\beta\gamma}(k_2-k_3)_\alpha+g_{\gamma\alpha}(k_3-k_1)_\beta\right]. \tag*{(8)} V α β γ ab c ( k 1 , k 2 , k 3 ) = − g s f ab c [ g α β ( k 1 − k 2 ) γ + g β γ ( k 2 − k 3 ) α + g γ α ( k 3 − k 1 ) β ] . (8)
Each term in this vertex will now be represented by a distinct graph using the decomposition
Figure 1 (PDF p. 6).
Here, each graph is given by only one term in Eq. (8) . Explicitly, we have
Figure 2 (PDF p. 6).
[ V α β γ a b c ] ( 12 ) = − g s f a b c g α β ( k 1 − k 2 ) γ . (9) \left[V^{abc}_{\alpha\beta\gamma}\right]^{(12)}=-g_s f^{abc}g_{\alpha\beta}(k_1-k_2)_\gamma. \tag*{(9)} [ V α β γ ab c ] ( 12 ) = − g s f ab c g α β ( k 1 − k 2 ) γ . (9)
Throughout this paper, we will use the notation [ V ] [V] [ V ] to remind the reader that the vertex is not the full three gluon vertex, but rather a specific piece picked out through this decomposition. The specific term picked out in Eq. (8) will be made clear using a graph. With this graph decomposition on hand, we are now ready to split all topologies further into M I R ( 2 ) , M I R ( 2 ) \mathcal{M}^{(2)}_{\mathrm{IR}},\mathcal{M}^{(2)}_{\mathrm{IR}} M IR ( 2 ) , M IR ( 2 ) . A further subclassification within M I R ( 2 ) \mathcal{M}^{(2)}_{\mathrm{IR}} M IR ( 2 ) distinguishes between the pinched graphs which need a graph sum to manifest finiteness from the unpinched graphs which are automatically infrared finite.
M I R ( 2 ) = M I R , p i n c h e d ( 2 ) + M I R , u n p i n c h e d ( 2 ) (10) \mathcal{M}^{(2)}_{\mathrm{IR}}=\mathcal{M}^{(2)}_{\mathrm{IR,pinched}}+\mathcal{M}^{(2)}_{\mathrm{IR,unpinched}} \tag*{(10)} M IR ( 2 ) = M IR , pinched ( 2 ) + M IR , unpinched ( 2 ) (10)
List of non-factorized two-loop graphs
Using the scalar-scalar-gluon decomposition, we can now classify all the remaining non-factorized two-loop graphs into the two families M I R ( 2 ) \mathcal{M}_{\mathrm{IR}}^{(2)} M IR ( 2 ) and M I R ( 2 ) \mathcal{M}_{\mathrm{IR}}^{(2)} M IR ( 2 ) introduced in Eq. (5) . Throughout this catalogue, H \mathcal{H} H represents the complete massive fermion loop, including all distinct permutations of gluon and Higgs attachments.
The first family is shown in Figure 3 . These five graphs have genuine soft or collinear singularities, whose leading contributions must be isolated by factorization. We return to them in § 4 .
Figure 3. The five graph topologies contributing to M I R ( 2 ) \mathcal{M}_{\mathrm{IR}}^{(2)} M IR ( 2 ) . Dashed lines denote scalar legs of the scalar-scalar-gluon decomposition.
The remaining graphs belong to M I R ( 2 ) \mathcal{M}_{\mathrm{IR}}^{(2)} M IR ( 2 ) . It is useful to separate them further according to whether their massless propagators admit infrared pinch surfaces as in Eq. (10) .
The ten pinched graphs are displayed in Figure 4 . Although their denominators admit soft or collinear pinches, their infrared singularities disappear after the appropriate Ward-identity rearrangements and graph sums. The corresponding analysis is given in § 3 .
Figure 4. The ten pinched graphs contributing to M I R ( 2 ) \mathcal{M}^{(2)}_{\mathrm{IR}} M IR ( 2 ) , labeled (i)–(x).
Finally, Figure 5 contains the unpinched graphs, which require no infrared rearrangement. The last graph, labeled ( α ) (\alpha) ( α ) , denotes the compact two-loop hard-loop contribution.
Figure 5. The unpinched contribution to M I R ( 2 ) \mathcal{M}^{(2)}_{\mathrm{IR}} M IR ( 2 ) .
G-K decomposition
We will also make extensive use of the G-K decomposition of the gluon propagator. For a gluon carrying momentum k k k , we define
δ ν μ = G ν μ ( k ; q ) + K ν μ ( k ; q ) , (11) \delta^\mu_\nu= G^\mu_\nu(k;q) + K^\mu_\nu(k;q), \tag*{(11)} δ ν μ = G ν μ ( k ; q ) + K ν μ ( k ; q ) , (11)
K ν μ ( k ; q ) = ( 2 q + k ) μ k ν k 2 + 2 q ⋅ k + i 0 , (12) K^\mu_\nu(k;q) = \frac{(2q+k)^\mu k_\nu}{k^2+2q\cdot k+i0}, \tag*{(12)} K ν μ ( k ; q ) = k 2 + 2 q ⋅ k + i 0 ( 2 q + k ) μ k ν , (12)
G ν μ ( k ; q ) = δ ν μ − K ν μ ( k ; q ) . (13) G^\mu_\nu(k;q) = \delta^\mu_\nu-K^\mu_\nu(k;q). \tag*{(13)} G ν μ ( k ; q ) = δ ν μ − K ν μ ( k ; q ) . (13)
Here q q q is a reference momentum and will be taken to be time-like throughout, satisfying q 2 > 0 q^2>0 q 2 > 0 . A specific choice that works well is q = P = p 1 + p 2 q=P=p_1+p_2 q = P = p 1 + p 2 . With this choice the auxilliary denominator is, ( k + P ) 2 − P 2 + i 0 (k+P)^2-P^2+i0 ( k + P ) 2 − P 2 + i 0 . The K K K term is longitudinal with respect to the gluon momentum k k k and will be treated using Ward identities. Away from the pole of
the projector, we have
k μ G ν μ ( k ; q ) = 0 , (14) k_\mu G^\mu_{\ \nu}(k;q)=0, \tag*{(14)} k μ G ν μ ( k ; q ) = 0 , (14)
G ν μ ( k ; q ) ( 2 q + k ) ν = 0. (15) G^\mu_{\ \nu}(k;q)(2q+k)^\nu=0. \tag*{(15)} G ν μ ( k ; q ) ( 2 q + k ) ν = 0. (15)
We will sometimes utilize the appropriately raised/lowered G − K G-K G − K tensors, defined by G ν μ = η μ ρ G ρ σ η σ ν G^\mu_{\ \nu}=\eta_{\mu\rho}G^{\rho\sigma}\eta_{\sigma\nu} G ν μ = η μ ρ G ρ σ η σ ν . Let us interpret these equations, in a manner useful to the discussion that follows. The G gluon supplies power suppression in the soft limit because G ν μ ( k ; q ) ( 2 q + k ) ν = 0 G^\mu_{\ \nu}(k;q)(2q+k)^\nu=0 G ν μ ( k ; q ) ( 2 q + k ) ν = 0 only if it attaches to a jet with lightlike momentum q q q . However, when it is chosen to be time-like as we do here, gauge freedom (the freedom to shift reference vectors for gluons) ensures the difference between the power suppressed G gluon with reference vector p p p (jet momentum) and reference vector q q q (arbitrary reference) is a longitudinal degree of freedom (in the soft limit).
The first equation in Eq. (14) will find use in collinear limits when k ∥ p i k\parallel p_i k ∥ p i where p i p_i p i is some light-like vector. This identity suggests that when the G gluon attaches to a p i p_i p i jet, it is automatically power suppressed.
Finally, we will frequently apply the G-K decomposition at both ends of a gluon propagator. Writing the momentum at the opposite end as k 1 = − k k_1=-k k 1 = − k , we have
− i g μ ν k 2 + i 0 = [ G μ α ( k ; q ) + K μ α ( k ; q ) ] − i g α β [ G ν β ( k 1 ; q ) + K ν β ( k 1 ; q ) ] k 2 + i 0 = − i ( G g G ) μ ν + ( G g K ) μ ν + ( K g G ) μ ν + ( K g K ) μ ν k 2 + i 0 . (16) \begin{aligned}
\frac{-ig_{\mu\nu}}{k^2+i0}
=
\frac{\left[G^\alpha_{\ \mu}(k;q)+K^\alpha_{\ \mu}(k;q)\right]-ig_{\alpha\beta}\left[G^\beta_{\ \nu}(k_1;q)+K^\beta_{\ \nu}(k_1;q)\right]}{k^2+i0} \\
&= -i\frac{(GgG)_{\mu\nu}+(GgK)_{\mu\nu}+(KgG)_{\mu\nu}+(KgK)_{\mu\nu}}{k^2+i0}. \tag*{(16)}
\end{aligned} k 2 + i 0 − i g μν = k 2 + i 0 [ G μ α ( k ; q ) + K μ α ( k ; q ) ] − i g α β [ G ν β ( k 1 ; q ) + K ν β ( k 1 ; q ) ] = − i k 2 + i 0 ( G g G ) μν + ( G g K ) μν + ( K g G ) μν + ( K g K ) μν . (16)
\mu\cdots\nu= \mu\mathop{\circled{G}} \cdots\nu+ \mu\mathop{\circled{K}} \cdots\nu
\mu\cdots\nu= \mu\mathop{\circled{G}} \cdots\mathop{\circled{G}} \cdots\nu+ \mu\mathop{\circled{K}} \cdots\mathop{\circled{K}} \cdots\nu+ \mu\mathop{\circled{G}} \cdots\mathop{\circled{K}} \cdots\nu+ \mu\mathop{\circled{K}} \cdots\mathop{\circled{G}} \cdots\nu
The two-sided decomposition separates the gluon propagator into four contributions. Terms containing longitudinal polarizations can be simplified using Ward identities, while the remaining contributions will be finite in all collinear limits. Their soft limits will be treated explicitly with counterterms.
Before ending this section, let us introduce graphical notation for the G-K when applied to a gluon propagator. We will use the diagrammatic identities in Figure 6 as modified Feynman rules throughout this paper. These identities are exact and are in one-to-one correspondence with the one-sided G-K decomposition in Eq. (12) and the two-sided decomposition in Eq. (16) , respectively. The circles labeled G G G and K K K represent the corresponding projectors acting on the Lorentz indices of the gluon propagator.
Figure 6. Modified Feynman rules corresponding to the one-sided G-K decomposition in Eq. (12) (top) and the two-sided G-K decomposition in Eq. (16) (bottom). The circles represent the G G G and K K K projectors, with the propagator denominator and overall factor of − i -i − i understood.
Transient infrared singularities
In this section, we begin our study of the singular graphs with transient singularities. These are all graphs in M I R , p i n c h e d ( 2 ) \mathcal{M}^{(2)}_{\mathrm{IR,pinched}} M IR , pinched ( 2 ) as it appears in Eq. (10) and in Figure 4 .
Let us begin by considering graphs (viii)–(x) of Figure 4 . These graphs have two gluons attached to the massive fermion loop, while all three gluon vertices on the open light-quark line are ordinary quark–gluon vertices. We will label one of the gluons entering the fermion loop with loop momentum k k k and the other by r = ∑ i = 1 n q i − k r = \sum_{i=1}^{n} q_i - k r = ∑ i = 1 n q i − k . Sometimes, we will use the shorthand Q μ = ∑ i = 1 n q i Q^\mu= \sum_{i=1}^{n} q_i Q μ = ∑ i = 1 n q i (to be distinguished from q q q , the reference vector for the G-K decomposition). The strategy here will be to first write their Feynman integrands and then apply the one-sided G-K decomposition. The decomposition will utilize the integrated fermion loop ward identity to make the integrand locally infrared finite. The result of this replacement is shown in Figure 7 .
Figure 7. The ward identity based replacement Feynman rules for graph topologies (viii)–(x). The top and bottom panels are exactly equal after hard loop integration and diagram sum.
We will utilise the concise symbols for vertex factors and propagators
S 0 ( l ) = i l / l 2 + i 0 , D μ ν ( l ) = − i g μ ν l 2 + i 0 , Γ a μ = − i g s T a γ μ , Γ 3 = − i g s T x f 3 ( p 3 ; p 1 ) . (17) S_0(l) = \frac{il\!\!\!/}{l^2+i0}, \qquad D_{\mu\nu}(l) = \frac{-ig_{\mu\nu}}{l^2+i0}, \qquad\Gamma_a^\mu= -ig_s T^a \gamma^\mu, \qquad\Gamma_3 = -ig_s T^x f_3(p_3;p_1). \tag*{(17)} S 0 ( l ) = l 2 + i 0 i l / , D μν ( l ) = l 2 + i 0 − i g μν , Γ a μ = − i g s T a γ μ , Γ 3 = − i g s T x f 3 ( p 3 ; p 1 ) . (17)
Recall that H \mathcal{H} H includes the permutation sum and the massive fermion-loop integration. Let us now study the integrands with the propagators of the two gluons entering H \mathcal{H} H explicitly written,
I v i i i ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ b ρ S 0 ( p 1 − p 3 ) Γ 3 u ( p 1 ) D μ ν ( k ) D ρ σ ( r ) H ν σ a b [ 2 , m ] ( k , r ; { q j } ) , (18) \mathcal{I}_{\mathrm{viii}}(k)=\bar{v}(p_{2})\Gamma_{a}^{\mu}S_{0}(k-p_{2})\Gamma_{b}^{\rho}S_{0}(p_{1}-p_{3})\Gamma_{3}u(p_{1})D_{\mu}^{\nu}(k)D_{\rho}^{\sigma}(r)\mathcal{H}_{\nu\sigma}^{ab}[2,m](k,r;\{q_{j}\}), \tag*{(18)} I viii ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ b ρ S 0 ( p 1 − p 3 ) Γ 3 u ( p 1 ) D μ ν ( k ) D ρ σ ( r ) H ν σ ab [ 2 , m ] ( k , r ; { q j }) , (18)
I i x ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ 3 S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ν ( k ) D ρ σ ( r ) H ν σ a b [ 2 , m ] ( k , r ; { q j } ) , (19) \mathcal{I}_{\mathrm{ix}}(k)=\bar{v}(p_{2})\Gamma_{a}^{\mu}S_{0}(k-p_{2})\Gamma_{3}S_{0}(p_{1}-r)\Gamma_{b}^{\rho}u(p_{1})D_{\mu}^{\nu}(k)D_{\rho}^{\sigma}(r)\mathcal{H}_{\nu\sigma}^{ab}[2,m](k,r;\{q_{j}\}), \tag*{(19)} I ix ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ 3 S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ν ( k ) D ρ σ ( r ) H ν σ ab [ 2 , m ] ( k , r ; { q j }) , (19)
I x ( k ) = v ˉ ( p 2 ) Γ 3 S 0 ( p 3 − p 2 ) Γ a μ S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ν ( k ) D ρ σ ( r ) H ν σ a b [ 2 , m ] ( k , r ; { q j } ) . (20) \mathcal{I}_{\mathrm{x}}(k)=\bar{v}(p_{2})\Gamma_{3}S_{0}(p_{3}-p_{2})\Gamma_{a}^{\mu}S_{0}(p_{1}-r)\Gamma_{b}^{\rho}u(p_{1})D_{\mu}^{\nu}(k)D_{\rho}^{\sigma}(r)\mathcal{H}_{\nu\sigma}^{ab}[2,m](k,r;\{q_{j}\}). \tag*{(20)} I x ( k ) = v ˉ ( p 2 ) Γ 3 S 0 ( p 3 − p 2 ) Γ a μ S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ν ( k ) D ρ σ ( r ) H ν σ ab [ 2 , m ] ( k , r ; { q j }) . (20)
Here the three expressions correspond, respectively, to emission of p 3 p_{3} p 3 next to the p 1 p_{1} p 1 endpoint, between the two gluons going into H \mathcal{H} H , and next to the p 2 p_{2} p 2 endpoint. In Eqs. (18) –(20) the symmetry factors of these graphs have been suppressed since the cancellation only requires the fermion loop ward identity. More precisely, each graph contribution is ∫ d d k 1 ∣ G ∣ I m g r a p h ( k ) / ( 2 π ) d \int\mathrm{d}^{d}k \,\frac{1}{|\mathcal{G}|}\mathcal{I}_{\mathrm{mgraph}}(k)/(2\pi)^{d} ∫ d d k ∣ G ∣ 1 I mgraph ( k ) / ( 2 π ) d .
Because the fermion loop has only two external gluons and all other attachments are color singlets, its integrated Ward identities read
k ν H ν σ a b [ 2 , m ] ( k , r ; { q j } ) = 0 , r σ H ν σ a b [ 2 , m ] ( k , r ; { q j } ) = 0. (21) k^{\nu}\mathcal{H}_{\nu\sigma}^{ab}[2,m](k,r;\{q_{j}\})=0,\qquad r^{\sigma}\mathcal{H}_{\nu\sigma}^{ab}[2,m](k,r;\{q_{j}\})=0. \tag*{(21)} k ν H ν σ ab [ 2 , m ] ( k , r ; { q j }) = 0 , r σ H ν σ ab [ 2 , m ] ( k , r ; { q j }) = 0. (21)
We can therefore replace each of the two gluon propagators by its one-sided G G G term, without changing the result at fixed k , r k,r k , r :
D μ ν ( k ) H ν σ = − i k 2 + i 0 G μ ν ( k ; q ) H ν σ , (22) D_{\mu}^{\nu}(k)\mathcal{H}_{\nu\sigma}=\frac{-i}{k^{2}+i0}G_{\mu}^{\nu}(k;q)\mathcal{H}_{\nu\sigma}, \tag*{(22)} D μ ν ( k ) H ν σ = k 2 + i 0 − i G μ ν ( k ; q ) H ν σ , (22)
D ρ σ ( r ) H ν σ = − i r 2 + i 0 G ρ σ ( r ; q ) H ν σ . (23) D_{\rho}^{\sigma}(r)\mathcal{H}_{\nu\sigma}=\frac{-i}{r^{2}+i0}G_{\rho}^{\sigma}(r;q)\mathcal{H}_{\nu\sigma}. \tag*{(23)} D ρ σ ( r ) H ν σ = r 2 + i 0 − i G ρ σ ( r ; q ) H ν σ . (23)
Here, we have use the fact that the K K K gluonic replacement is
− i k 2 + i 0 K μ ν ( k ; q ) H ν σ = − i k 2 + i 0 ( 2 q + k ) μ k ν H ν σ k 2 + 2 q ⋅ k + i 0 = 0 , (24) \frac{-i}{k^{2}+i0}K_{\mu}^{\nu}(k;q)\mathcal{H}_{\nu\sigma}=\frac{-i}{k^{2}+i0}\frac{(2q+k)_{\mu}k^{\nu}\mathcal{H}_{\nu\sigma}}{k^{2}+2q\cdot k+i0}=0, \tag*{(24)} k 2 + i 0 − i K μ ν ( k ; q ) H ν σ = k 2 + i 0 − i k 2 + 2 q ⋅ k + i 0 ( 2 q + k ) μ k ν H ν σ = 0 , (24)
and similarly for r r r . Throughout this section we use the shorthand
D μ G ν ( k ) ≡ − i k 2 + i 0 G μ ν ( k ; q ) , (25) D^{G\,\nu}_{\mu}(k) \equiv\frac{-i}{k^2+i0}G_{\mu}^{\nu}(k;q), \tag*{(25)} D μ G ν ( k ) ≡ k 2 + i 0 − i G μ ν ( k ; q ) , (25)
which is equal to D μ ν ( k ) D^\nu_\mu(k) D μ ν ( k ) when contracted with H \mathcal{H} H , since the K K K part vanishes by the Ward identity in Eq. (21) . After this replacement, the new integrands are
I v i i i ( G ) ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ b ρ S 0 ( p 1 − p 3 ) Γ 3 u ( p 1 ) D μ G ν ( k ) D ρ σ ( r ) H ν σ a b [ 2 , m ] ( k , r ; { q j } ) , (26) \mathcal{I}^{(G)}_{\mathrm{viii}}(k)=\bar{v}(p_2)\Gamma^\mu_a S_0(k-p_2)\Gamma^\rho_b S_0(p_1-p_3)\Gamma_3 u(p_1)D^{G\,\nu}_{\mu}(k)D^\sigma_{\rho}(r)\mathcal{H}^{ab}_{\nu\sigma}[2,m](k,r;\{q_j\}), \tag*{(26)} I viii ( G ) ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ b ρ S 0 ( p 1 − p 3 ) Γ 3 u ( p 1 ) D μ G ν ( k ) D ρ σ ( r ) H ν σ ab [ 2 , m ] ( k , r ; { q j }) , (26)
I i x ( G ) ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ 3 S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ( G ) ν ( k ) D ρ ( G ) σ ( r ) H ν σ a b [ 2 , m ] ( k , r ; { q j } ) , (27) \mathcal{I}^{(G)}_{\mathrm{ix}}(k)=\bar{v}(p_2)\Gamma^\mu_a S_0(k-p_2)\Gamma_3 S_0(p_1-r)\Gamma^\rho_b u(p_1)D^{(G)\,\nu}_{\mu}(k)D^{(G)\,\sigma}_{\rho}(r)\mathcal{H}^{ab}_{\nu\sigma}[2,m](k,r;\{q_j\}), \tag*{(27)} I ix ( G ) ( k ) = v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) Γ 3 S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ( G ) ν ( k ) D ρ ( G ) σ ( r ) H ν σ ab [ 2 , m ] ( k , r ; { q j }) , (27)
I x ( G ) ( k ) = v ˉ ( p 2 ) Γ 3 S 0 ( p 3 − p 2 ) Γ a μ S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ν ( k ) D ρ ( G ) σ ( r ) H ν σ a b [ 2 , m ] ( k , r ; { q j } ) . (28) \mathcal{I}^{(G)}_{\mathrm{x}}(k)=\bar{v}(p_2)\Gamma_3 S_0(p_3-p_2)\Gamma^\mu_a S_0(p_1-r)\Gamma^\rho_b u(p_1)D^\nu_{\mu}(k)D^{(G)\,\sigma}_{\rho}(r)\mathcal{H}^{ab}_{\nu\sigma}[2,m](k,r;\{q_j\}). \tag*{(28)} I x ( G ) ( k ) = v ˉ ( p 2 ) Γ 3 S 0 ( p 3 − p 2 ) Γ a μ S 0 ( p 1 − r ) Γ b ρ u ( p 1 ) D μ ν ( k ) D ρ ( G ) σ ( r ) H ν σ ab [ 2 , m ] ( k , r ; { q j }) . (28)
Here, we have made an explicit choice in the reference vector that defines the K K K polarized longitudinal gluon. Specifically, we choose the same fixed timelike reference vector P P P in both projectors P 2 > 0 P^2>0 P 2 > 0 for the massive color-singlet final state. Before discussing the behavior of the modified graphs near pinches of the original graph, let us discuss the two limits in which the newly introduced denominator vanishes. First, it vanishes in the soft region where k → 0 k\to0 k → 0 . However, in this region, assuming canonical soft scaling (k μ → λ Q k^\mu\to\lambda Q k μ → λ Q ; λ → 0 \lambda\to0 λ → 0 ), the newly added term is k ν q ⋅ k \frac{k^\nu}{q\cdot k} q ⋅ k k ν which is O ( 1 ) O(1) O ( 1 ) . The denominator can also vanish when k μ k^\mu k μ is space-like, with a potential pinched subgraph with on-shell auxiliary denominator and another denominator which is vanishing. For the graphs occurring in this section, such a pinch can be explicitly excluded when q = P q=P q = P for generic kinematic configurations.
Let us now discuss infrared fineness of these graphs in the IR limits where the original graphs were singular. Consider the graph (viii) which was singular in the k ∥ p 2 k\parallel p_2 k ∥ p 2 limit. The original singularity was logarithmic. Let us check this statement. Decompose
k = α 1 p 2 + α 2 p ˉ 2 + k ⊥ , (29) k=\alpha_1p_2+\alpha_2\bar{p}_2+k_\perp, \tag*{(29)} k = α 1 p 2 + α 2 p ˉ 2 + k ⊥ , (29)
with p ˉ 2 = ( p 2 0 , − p ⃗ 2 ) \bar{p}_2=(p_2^0,-\vec{p}_2) p ˉ 2 = ( p 2 0 , − p 2 ) and observe that
v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) η μ ν k 2 + i 0 = v ˉ ( p 2 ) − i g T a ( α − 1 ) 2 p 2 μ ( k − p 2 ) 2 + i 0 η μ ν k 2 + i 0 + O ( k ⊥ ) , (30) \bar{v}(p_2)\Gamma^\mu_a S_0(k-p_2)\frac{\eta_{\mu\nu}}{k^2+i0}
=\bar{v}(p_2)-igT_a(\alpha-1)\frac{2p_2^\mu}{(k-p_2)^2+i0}\frac{\eta_{\mu\nu}}{k^2+i0}+O(k_\perp), \tag*{(30)} v ˉ ( p 2 ) Γ a μ S 0 ( k − p 2 ) k 2 + i 0 η μν = v ˉ ( p 2 ) − i g T a ( α − 1 ) ( k − p 2 ) 2 + i 0 2 p 2 μ k 2 + i 0 η μν + O ( k ⊥ ) , (30)
up to terms that are power suppressed in the collinear limit. One can check that the other denominators not appearing here are all finite in the collinear limit. As a result, one finds that the assuming collinear pinch scaling k ⊥ → λ Q k_\perp\to\sqrt{\lambda}Q k ⊥ → λ Q , α 2 → λ \alpha_2\to\lambda α 2 → λ , in the uniform λ → 0 \lambda\to0 λ → 0 limit, one finds a logarithmic divergence (both the measure and the denominator in Eq. (30) vanish uniformly as λ 2 \lambda^2 λ 2 ). However, the G gluon replacement that we have made in Eqs. (23) induces additional power suppression since
p 2 μ ( G μ ν ) k 2 + i 0 = 0 + O ( k ⊥ ) . (31) \frac{p_2^\mu(G_{\mu\nu})}{k^2+i0}=0+O(k_\perp). \tag*{(31)} k 2 + i 0 p 2 μ ( G μν ) = 0 + O ( k ⊥ ) . (31)
This proves power counting finiteness of (viii) in the k ∥ p 2 k \parallel p_2 k ∥ p 2 limit. Let us briefly discuss the same mechanism of power suppression near the pinch in the other two graphs. Graph (x) was singular in the r ∥ p 1 r \parallel p_1 r ∥ p 1 limit. The singularity is again logarithmic and arises from the propagators r 2 r^2 r 2 and ( p 1 − r ) 2 (p_1-r)^2 ( p 1 − r ) 2 . However, the replacement of the gluon propagator by its G part gives
p 1 μ G μ ν ( r ; q ) r 2 + i 0 = O ( r ⊥ ) r 2 + i 0 , r ∥ p 1 . (32) \frac{p_1^\mu G_{\mu\nu}(r;q)}{r^2+i0}=\frac{O(r_\perp)}{r^2+i0},\qquad r\parallel p_1. \tag*{(32)} r 2 + i 0 p 1 μ G μν ( r ; q ) = r 2 + i 0 O ( r ⊥ ) , r ∥ p 1 . (32)
Thus, the original logarithmic singularity is power suppressed, and graph (x) is locally finite.
Graph (ix) has both of the collinear limits discussed above, namely k ∥ p 2 k \parallel p_2 k ∥ p 2 and r ∥ p 1 r \parallel p_1 r ∥ p 1 . Since we have made the G replacement on both gluon propagators, we find
p 2 μ G μ ν ( k ; q ) = O ( k ⊥ ) , p 1 μ G μ ν ( r ; q ) = O ( r ⊥ ) . (33) p_2^\mu G_{\mu\nu}(k;q)=O(k_\perp),\qquad p_1^\mu G_{\mu\nu}(r;q)=O(r_\perp). \tag*{(33)} p 2 μ G μν ( k ; q ) = O ( k ⊥ ) , p 1 μ G μν ( r ; q ) = O ( r ⊥ ) . (33)
The original logarithmic singularities are therefore removed in both limits. There is one additional pinch surface, where the two internal quark propagators become collinear to p 3 p_3 p 3 . Writing their momenta as ℓ \ell ℓ and ℓ + p 3 \ell+p_3 ℓ + p 3 , the leading numerator vanishes because
ℓ / ϵ / 3 ( ℓ / + p / 3 ) = O ( ℓ ⊥ ) , ℓ ∥ p 3 , (34) \ell\!\!\!/\,\epsilon\!\!\!/_{3}(\ell\!\!\!/+p\!\!\!/_{3})=O(\ell_\perp),\qquad\ell\parallel p_3, \tag*{(34)} ℓ / ϵ / 3 ( ℓ / + p / 3 ) = O ( ℓ ⊥ ) , ℓ ∥ p 3 , (34)
where we have used ϵ 3 ⋅ p 3 = 0 \epsilon_3\cdot p_3=0 ϵ 3 ⋅ p 3 = 0 . Thus graph (ix) is locally finite in this limit as well.
We now consider graphs (v)–(vii). In all three cases, the only pinch surface is reached when an internal loop momentum is collinear to p 3 p_3 p 3 . We will find that while (v),(vi) are already locally finite, we may once again utilize the one sided G-K decomposition to ensure that (vii) is locally finite. The graphs and the replacement Feynman rules are summarized in Figure 8 .
Figure 8. Graphs (v),(vi), and (vii). While the first two are already locally finite, the last is made finite using the G-K decomposition.
The Feynman rules for the graphs in Figure 8 in the notation of Eq. (17) , at the level of the three integrands are
I v ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) H μ ρ σ a b c [ 3 , m ] ( P , k , − k − p 3 ; { q j } ) × D ρ η ( k ) D σ θ ( k + p 3 ) [ V b c x ] η θ λ ( − k , k + p 3 , − p 3 ) ϵ 3 λ , (35) \mathcal{I}_{\mathrm{v}}(k)=\bar{v}(p_{2})\Gamma^{\alpha}_{a}u(p_{1})D_{\alpha}^{\mu}(P)\mathcal{H}^{abc}_{\mu\rho\sigma}[3,m](P,k,-k-p_{3};\{q_{j}\})
\\
\times D^{\rho\eta}(k)D^{\sigma\theta}(k+p_{3})[V^{bcx}]_{\eta\theta\lambda}(-k,k+p_{3},-p_{3})\epsilon_{3}^{\lambda}, \tag*{(35)} I v ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) H μ ρ σ ab c [ 3 , m ] ( P , k , − k − p 3 ; { q j }) × D ρ η ( k ) D σ θ ( k + p 3 ) [ V b c x ] η θ λ ( − k , k + p 3 , − p 3 ) ϵ 3 λ , (35)
I v i ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) V μ η ζ a b c ( P , k − Q , − k − p 3 ) × D η ν ( Q − k ) D ζ β ( k + p 3 ) [ V c d x ] β γ λ ( k + p 3 , − k , − p 3 ) × D γ σ ( k ) H ν σ b d [ 2 , m ] ( Q − k , k ; { q j } ) ϵ 3 λ . (36) \mathcal{I}_{\mathrm{vi}}(k)=\bar{v}(p_{2})\Gamma^{\alpha}_{a}u(p_{1})D_{\alpha}^{\mu}(P)V^{abc}_{\mu\eta\zeta}(P,k-Q,-k-p_{3})
\\
\times D^{\eta\nu}(Q-k)D^{\zeta\beta}(k+p_{3})[V^{cdx}]_{\beta\gamma\lambda}(k+p_{3},-k,-p_{3})
\\
\times D^{\gamma\sigma}(k)\mathcal{H}^{bd}_{\nu\sigma}[2,m](Q-k,k;\{q_{j}\})\epsilon_{3}^{\lambda}. \tag*{(36)} I vi ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) V μ η ζ ab c ( P , k − Q , − k − p 3 ) × D η ν ( Q − k ) D ζ β ( k + p 3 ) [ V c d x ] β γ λ ( k + p 3 , − k , − p 3 ) × D γ σ ( k ) H ν σ b d [ 2 , m ] ( Q − k , k ; { q j }) ϵ 3 λ . (36)
I v i i ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) V μ η ζ a b c ( P , k − Q , − k − p 3 ) × D η ν ( Q − k ) D ζ β ( k + p 3 ) [ V c x d ] β λ γ ( k + p 3 , − p 3 , − k ) × D γ σ ( k ) H ν σ b d [ 2 , m ] ( Q − k , k ; { q j } ) ϵ 3 λ . (37) \mathcal{I}_{\mathrm{vii}}(k)=\bar{v}(p_{2})\Gamma^{\alpha}_{a}u(p_{1})D_{\alpha}^{\mu}(P)V^{abc}_{\mu\eta\zeta}(P,k-Q,-k-p_{3})
\\
\times D^{\eta\nu}(Q-k)D^{\zeta\beta}(k+p_{3})[V^{cxd}]_{\beta\lambda\gamma}(k+p_{3},-p_{3},-k)
\\
\times D^{\gamma\sigma}(k)\mathcal{H}^{bd}_{\nu\sigma}[2,m](Q-k,k;\{q_{j}\})\epsilon_{3}^{\lambda}. \tag*{(37)} I vii ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) V μ η ζ ab c ( P , k − Q , − k − p 3 ) × D η ν ( Q − k ) D ζ β ( k + p 3 ) [ V c x d ] β λγ ( k + p 3 , − p 3 , − k ) × D γ σ ( k ) H ν σ b d [ 2 , m ] ( Q − k , k ; { q j }) ϵ 3 λ . (37)
Here we have suppressed the numerical symmetry factors, as in Eqs. (18) –(20) . The scalar-scalar-gluon vertices in (v) and (vi) give, respectively,
ϵ 3 λ [ V b c x ] η θ λ ( − k , k + p 3 , − p 3 ) = 2 g s f b c x g η θ ϵ 3 ⋅ k , ϵ 3 λ [ V c d x ] β γ λ ( k + p 3 , − k , − p 3 ) = − 2 g s f c d x g β γ ϵ 3 ⋅ k . (38) \epsilon_{3}^{\lambda}[V^{bcx}]_{\eta\theta\lambda}(-k,k+p_{3},-p_{3})=2g_{s}f^{bcx}g_{\eta\theta}\,\epsilon_{3}\cdot k,
\\
\epsilon_{3}^{\lambda}[V^{cdx}]_{\beta\gamma\lambda}(k+p_{3},-k,-p_{3})=-2g_{s}f^{cdx}g_{\beta\gamma}\,\epsilon_{3}\cdot k. \tag*{(38)} ϵ 3 λ [ V b c x ] η θ λ ( − k , k + p 3 , − p 3 ) = 2 g s f b c x g η θ ϵ 3 ⋅ k , ϵ 3 λ [ V c d x ] β γ λ ( k + p 3 , − k , − p 3 ) = − 2 g s f c d x g β γ ϵ 3 ⋅ k . (38)
Since ϵ 3 ⋅ p 3 = 0 \epsilon_{3}\cdot p_{3}=0 ϵ 3 ⋅ p 3 = 0 , both numerators are O ( k ⊥ ) O(k_{\perp}) O ( k ⊥ ) when k ∥ p 3 k\parallel p_{3} k ∥ p 3 . Thus (v) and (vi) are locally finite without any propagator modification.
For (vii), the two scalar legs instead carry p 3 p_{3} p 3 and k + p 3 k+p_{3} k + p 3 , so
ϵ 3 λ [ V c x d ] β λ γ ( k + p 3 , − p 3 , − k ) = − g s f c x d ϵ 3 β ( 2 p 3 + k ) γ . (39) \epsilon_{3}^{\lambda}[V^{cxd}]_{\beta\lambda\gamma}(k+p_{3},-p_{3},-k)=-g_{s}f^{cxd}\epsilon_{3\beta}(2p_{3}+k)_{\gamma}. \tag*{(39)} ϵ 3 λ [ V c x d ] β λγ ( k + p 3 , − p 3 , − k ) = − g s f c x d ϵ 3 β ( 2 p 3 + k ) γ . (39)
The gluon of momentum k k k enters the complete H [ 2 , m ] \mathcal{H}[2,m] H [ 2 , m ] , and its Ward identity allows us to replace D ( k ) D(k) D ( k ) by D G ( k ; q ) D^{G}(k;q) D G ( k ; q ) , exactly as above. Concretely, the new integrand for (vii) reads
I v i i ( G ) ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) V μ η ζ a b c ( P , k − Q , − k − p 3 ) × D η ν ( Q − k ) D ζ β ( k + p 3 ) [ V c x d ] β λ γ ( k + p 3 , − p 3 , − k ) × D G γ σ ( k ) H ν σ b d [ 2 , m ] ( Q − k , k ; { q j } ) ϵ 3 λ . (40) \mathcal{I}^{(G)}_{\mathrm{vii}}(k)=\bar{v}(p_{2})\Gamma^{\alpha}_{a}u(p_{1})D_{\alpha}^{\mu}(P)V^{abc}_{\mu\eta\zeta}(P,k-Q,-k-p_{3})
\\
\times D^{\eta\nu}(Q-k)D^{\zeta\beta}(k+p_{3})[V^{cxd}]_{\beta\lambda\gamma}(k+p_{3},-p_{3},-k)
\\
\times D^{G\,\gamma\sigma}(k)\mathcal{H}^{bd}_{\nu\sigma}[2,m](Q-k,k;\{q_{j}\})\epsilon_{3}^{\lambda}. \tag*{(40)} I vii ( G ) ( k ) = v ˉ ( p 2 ) Γ a α u ( p 1 ) D α μ ( P ) V μ η ζ ab c ( P , k − Q , − k − p 3 ) × D η ν ( Q − k ) D ζ β ( k + p 3 ) [ V c x d ] β λγ ( k + p 3 , − p 3 , − k ) × D G γ σ ( k ) H ν σ b d [ 2 , m ] ( Q − k , k ; { q j }) ϵ 3 λ . (40)
Since k γ G σ γ ( k ; q ) = 0 k_{\gamma}G^{\gamma}_{\ \sigma}(k;q)=0 k γ G σ γ ( k ; q ) = 0 ,
( 2 p 3 + k ) γ G σ γ ( k ; q ) = 2 p 3 γ G σ γ ( k ; q ) = O ( k ⊥ ) , k ∥ p 3 . (41) (2p_{3}+k)_{\gamma}G^{\gamma}_{\ \sigma}(k;q)=2p_{3\gamma}G^{\gamma}_{\ \sigma}(k;q)=O(k_{\perp}),\qquad k\parallel p_{3}. \tag*{(41)} ( 2 p 3 + k ) γ G σ γ ( k ; q ) = 2 p 3 γ G σ γ ( k ; q ) = O ( k ⊥ ) , k ∥ p 3 . (41)
In all three cases, the graphs are power counting finite.
Let us now turn to the final class of graphs, (i)-(iv). These have potential singularities in all collinear limits as well as the soft limit. For graphs (i),(ii) we may continue using the one sided G-K decomposition to manifest local finiteness. The result of this is shown in Figure 9
Figure 9. One sided G-K replacement rules for graphs (i),(ii). The K remainder is identically zero as before.
We use the momentum routing shown in Figure 9 , where
P = p 1 + p 2 = k + r + p 3 , Q = k + r . (42) P = p_1 + p_2 = k + r + p_3,\qquad Q = k + r. \tag*{(42)} P = p 1 + p 2 = k + r + p 3 , Q = k + r . (42)
The two gluons entering the massive fermion loop carry momenta k , r k,r k , r , and the propagator between the two gluon attachments on the quark line carries momentum k − p 2 k-p_2 k − p 2 . The integrands after the G replacement are
T j ( G ) ( k ) = v ˉ ( p 2 ) Γ c μ S 0 ( k − p 2 ) Γ d α u ( p 1 ) D G μ ν ( k ; q ) D α η ( r + p 3 ) × ϵ 3 λ [ V d e x ] η β λ ( j ) ( r + p 3 , − r , − p 3 ) D j β ρ ( r ; q ) H j ν ρ [ 2 , m ] ( k , r ) , (43) \mathcal{T}^{(G)}_j(k) =
\bar{v}(p_2)\Gamma^\mu_c S_0(k-p_2)\Gamma^\alpha_d u(p_1)D^{G\mu\nu}(k;q)D^{\alpha\eta}(r+p_3)
\times\epsilon_3^\lambda[V^{dex}]^{(j)}_{\eta\beta\lambda}(r+p_3,-r,-p_3)D_j^{\beta\rho}(r;q)\mathcal{H}_j^{\nu\rho}[2,m](k,r), \tag*{(43)} T j ( G ) ( k ) = v ˉ ( p 2 ) Γ c μ S 0 ( k − p 2 ) Γ d α u ( p 1 ) D G μν ( k ; q ) D α η ( r + p 3 ) × ϵ 3 λ [ V d e x ] η β λ ( j ) ( r + p 3 , − r , − p 3 ) D j β ρ ( r ; q ) H j ν ρ [ 2 , m ] ( k , r ) , (43)
where
D i ( r ; q ) = D ( r ) , D i i ( r ; q ) = D G ( r ; q ) . (44) D_i(r;q)=D(r),\qquad D_{ii}(r;q)=D^G(r;q). \tag*{(44)} D i ( r ; q ) = D ( r ) , D ii ( r ; q ) = D G ( r ; q ) . (44)
The Feynman rules give the relations
ϵ 3 λ [ V d e x ] η β λ ( i ) ( r + p 3 , − r , − p 3 ) = − 2 g s f d e x g η β ϵ 3 ⋅ r . (45) \epsilon_3^\lambda[V^{dex}]^{(i)}_{\eta\beta\lambda}(r+p_3,-r,-p_3)
=-2g_s f^{dex}g_{\eta\beta}\epsilon_3\cdot r. \tag*{(45)} ϵ 3 λ [ V d e x ] η β λ ( i ) ( r + p 3 , − r , − p 3 ) = − 2 g s f d e x g η β ϵ 3 ⋅ r . (45)
ϵ 3 λ [ V d e x ] η β λ ( i i ) ( r + p 3 , − r , − p 3 ) = g s f d e x ϵ 3 η ( r + 2 p 3 ) β . (46) \epsilon_3^\lambda[V^{dex}]^{(ii)}_{\eta\beta\lambda}(r+p_3,-r,-p_3)
=g_s f^{dex}\epsilon_{3\eta}(r+2p_3)_\beta. \tag*{(46)} ϵ 3 λ [ V d e x ] η β λ ( ii ) ( r + p 3 , − r , − p 3 ) = g s f d e x ϵ 3 η ( r + 2 p 3 ) β . (46)
Let us first consider the collinear limits. In the k ∥ p 2 k\parallel p_2 k ∥ p 2 limit, both graphs have the same logarithmic singularity arising from the propagators k 2 k^2 k 2 and ( k − p 2 ) 2 (k-p_2)^2 ( k − p 2 ) 2 . As before, the G replacement gives
p 2 μ G μ ν ( k ; q ) = O ( k ⊥ ) , (47) p_2^\mu G_{\mu\nu}(k;q)=O(k_\perp), \tag*{(47)} p 2 μ G μν ( k ; q ) = O ( k ⊥ ) , (47)
which makes the singularity locally integrable.
Next, consider r + p 3 ∥ p 1 r+p_{3} \parallel p_{1} r + p 3 ∥ p 1 . In graph (i), the decomposed numerator satisfies
ϵ 3 ⋅ r = ϵ 3 ⋅ ( r + p 3 ) , (48) \epsilon_{3} \cdot r = \epsilon_{3} \cdot(r+p_{3}), \tag*{(48)} ϵ 3 ⋅ r = ϵ 3 ⋅ ( r + p 3 ) , (48)
which is power suppressed since ϵ 3 ⋅ p 1 = 0 \epsilon_{3} \cdot p_{1}=0 ϵ 3 ⋅ p 1 = 0 . In graph (ii), the leading quark numerator in either the soft or collinear to p 1 p_{1} p 1 limit is proportional to p 1 η p_{1}^{\eta} p 1 η , whose contraction with ϵ 3 η \epsilon_{3\eta} ϵ 3 η vanishes for the same reason.
Finally, in the r ∥ p 3 r \parallel p_{3} r ∥ p 3 limit, the singular propagators are r 2 r^{2} r 2 and ( r + p 3 ) 2 (r+p_{3})^{2} ( r + p 3 ) 2 . The numerator of graph (i) is suppressed by ϵ 3 ⋅ r \epsilon_{3} \cdot r ϵ 3 ⋅ r , while in graph (ii) we have
( r + 2 p 3 ) β G β ρ ( r ; q ) = 2 p 3 β G β ρ ( r ; q ) = O ( r ⊥ ) . (49) (r+2p_{3})^{\beta}G_{\beta\rho}(r;q)=2p_{3}^{\beta}G_{\beta\rho}(r;q)=O(r_{\perp}). \tag*{(49)} ( r + 2 p 3 ) β G β ρ ( r ; q ) = 2 p 3 β G β ρ ( r ; q ) = O ( r ⊥ ) . (49)
Thus both graphs are locally finite in all three collinear limits.
For k → 0 k \to0 k → 0 , only k 2 k^{2} k 2 and ( k − p 2 ) 2 (k-p_{2})^{2} ( k − p 2 ) 2 vanish, scaling as λ 2 \lambda^{2} λ 2 and λ \lambda λ , respectively. Similarly, for r → 0 r \to0 r → 0 , only r 2 r^{2} r 2 and ( r + p 3 ) 2 (r+p_{3})^{2} ( r + p 3 ) 2 vanish, with the same scaling. Both limits are therefore integrable without requiring any additional numerator suppression.
The remaining soft limit is r + p 3 → 0 r+p_{3} \to0 r + p 3 → 0 . Here three denominators vanish: ( r + p 3 ) 2 (r+p_{3})^{2} ( r + p 3 ) 2 , r 2 r^{2} r 2 and ( k − p 2 ) 2 (k-p_{2})^{2} ( k − p 2 ) 2 , scaling as λ 2 , λ , λ \lambda^{2},\lambda,\lambda λ 2 , λ , λ . In graph (i), ϵ 3 ⋅ r = ϵ 3 ⋅ ( r + p 3 ) \epsilon_{3} \cdot r=\epsilon_{3} \cdot(r+p_{3}) ϵ 3 ⋅ r = ϵ 3 ⋅ ( r + p 3 ) supplies one additional power of λ \lambda λ . In graph (ii), the leading quark numerator vanishes because ϵ 3 ⋅ p 1 = 0 \epsilon_{3} \cdot p_{1}=0 ϵ 3 ⋅ p 1 = 0 , giving the same suppression. Hence this limit is also integrable.
We conclude that graphs (i) and (ii) are locally finite in all their infrared limits. In arguing the finiteness of (i),(ii) after appropriate G gluon replacement, we have utilized the relationship ϵ 3 ⋅ p 1 = 0 \epsilon_{3} \cdot p_{1}=0 ϵ 3 ⋅ p 1 = 0 . Although, (iii),(iv) are the class of Feynman graphs generated by switching the roles of p 1 , p 2 p_{1},p_{2} p 1 , p 2 , a symmetric argument fails because ϵ 3 ( p 3 ; p 1 ) ⋅ p 2 ≠ 0 \epsilon_{3}(p_{3};p_{1}) \cdot p_{2} \ne0 ϵ 3 ( p 3 ; p 1 ) ⋅ p 2 = 0 . The strategy in this case will be to study (iii),(iv) with a polarization vector that satisfies ϵ 3 ( p 3 ; p 2 ) ⋅ p 2 = 0 \epsilon_{3}(p_{3};p_{2}) \cdot p_{2}=0 ϵ 3 ( p 3 ; p 2 ) ⋅ p 2 = 0 and study the gauge vector changing K gluon separately. Concretely, we propose to utilize the identity
ϵ 3 μ ( p 3 ; p 1 ) = ϵ 3 μ ( p 3 ; p 2 ) − ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 p 3 ⋅ p 1 p 3 μ . (50) \epsilon_{3}^{\mu}(p_{3};p_{1})=\epsilon_{3}^{\mu}(p_{3};p_{2})-\frac{\epsilon_{3}(p_{3};p_{2}) \cdot p_{1}}{p_{3} \cdot p_{1}}p_{3}^{\mu}. \tag*{(50)} ϵ 3 μ ( p 3 ; p 1 ) = ϵ 3 μ ( p 3 ; p 2 ) − p 3 ⋅ p 1 ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 p 3 μ . (50)
Replacing the polarization vector ϵ 3 μ ( p 3 ; p 1 ) \epsilon_{3}^{\mu}(p_{3};p_{1}) ϵ 3 μ ( p 3 ; p 1 ) by the right hand side of Eq. (50) in graphs (iii),(iv), the first term can now be made locally finite in the same way as before. We will focus on properties of the second term in what follows. We have inserted Eq. (50) and made the same replacements as in Figure 9 , and the remainder in Figure 10 .
Figure 10. Utilizing the identity of Eq. (50) in diagram (iii),(iv). In the second panel we have replaced ϵ 3 ( p 3 ; p 1 ) \epsilon_3(p_3;p_1) ϵ 3 ( p 3 ; p 1 ) by ϵ 3 ( p 3 ; p 2 ) \epsilon_3(p_3;p_2) ϵ 3 ( p 3 ; p 2 ) , and made the G replacements akin to Figure 9 . The last panel contains the remainder.
Following the arguments made previously (iii)a _{a} a ,(iv)a _{a} a are locally finite after the relevant G replacements are made. As usual the expression for I ( i i i ) , ( i v ) \mathcal{I}_{\mathrm{(iii),(iv)}} I ( iii ) , ( iv ) in the notation of Eq. (17) with ℓ = k + p 3 \ell=k+p_{3} ℓ = k + p 3 and r = P − ℓ r=P-\ell r = P − ℓ is
I i i i ( k ; ϵ 3 ) = − g s f c d x v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 ) D ρ μ ( ℓ ) D η τ ( r ) × g μ ν ( 2 k + p 3 ) ⋅ ϵ 3 D ν σ ( k ) H σ τ d b [ 2 , m ] ( k , r ) , (51) \mathcal{I}_{\mathrm{iii}}(k;\epsilon_{3})=-g_{s}f^{cdx}\bar{v}(p_{2})\Gamma_{\rho}^{c}S_{0}(\ell-p_{2})\Gamma_{\eta}^{b}u(p_{1})D^{\rho\mu}(\ell)D^{\eta\tau}(r)
\\
{}\times g_{\mu\nu}(2k+p_{3})\cdot\epsilon_{3}D^{\nu\sigma}(k)\mathcal{H}_{\sigma\tau}^{db}[2,m](k,r), \tag*{(51)} I iii ( k ; ϵ 3 ) = − g s f c d x v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 ) D ρ μ ( ℓ ) D η τ ( r ) × g μν ( 2 k + p 3 ) ⋅ ϵ 3 D ν σ ( k ) H σ τ d b [ 2 , m ] ( k , r ) , (51)
I i v ( k ; ϵ 3 ) = g s f c d x v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 ) D ρ μ ( ℓ ) D η τ ( r ) × ϵ 3 μ ( k + 2 p 3 ) ν D ν σ ( k ) H σ τ d b [ 2 , m ] ( k , r ) . (52) \mathcal{I}_{\mathrm{iv}}(k;\epsilon_{3})=g_{s}f^{cdx}\bar{v}(p_{2})\Gamma_{\rho}^{c}S_{0}(\ell-p_{2})\Gamma_{\eta}^{b}u(p_{1})D^{\rho\mu}(\ell)D^{\eta\tau}(r)
\\
{}\times\epsilon_{3\mu}(k+2p_{3})_{\nu}D^{\nu\sigma}(k)\mathcal{H}_{\sigma\tau}^{db}[2,m](k,r). \tag*{(52)} I iv ( k ; ϵ 3 ) = g s f c d x v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 ) D ρ μ ( ℓ ) D η τ ( r ) × ϵ 3 μ ( k + 2 p 3 ) ν D ν σ ( k ) H σ τ d b [ 2 , m ] ( k , r ) . (52)
The splitting equation described in Figure 10 is given by
I i i i ( k ; ϵ 3 ( p 3 ; p 1 ) ) = I i i i , a ( k ) + I i i i , b ( k ) , I i i i , a ( k ) = I i i i ( k ; ϵ 3 ( p 3 ; p 2 ) ) , I i i i , b ( k ) = − ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 p 3 ⋅ p 1 I i i i ( k ; p 3 ) , (53) \begin{aligned}
\mathcal{I}_{\mathrm{iii}}(k;\epsilon_3(p_3;p_1)) &= \mathcal{I}_{\mathrm{iii},a}(k) + \mathcal{I}_{\mathrm{iii},b}(k),\\
\mathcal{I}_{\mathrm{iii},a}(k) &= \mathcal{I}_{\mathrm{iii}}(k;\epsilon_3(p_3;p_2)),\\
\mathcal{I}_{\mathrm{iii},b}(k) &= -\frac{\epsilon_3(p_3;p_2)\cdot p_1}{p_3\cdot p_1}\mathcal{I}_{\mathrm{iii}}(k;p_3), \tag*{(53)}
\end{aligned} I iii ( k ; ϵ 3 ( p 3 ; p 1 )) I iii , a ( k ) I iii , b ( k ) = I iii , a ( k ) + I iii , b ( k ) , = I iii ( k ; ϵ 3 ( p 3 ; p 2 )) , = − p 3 ⋅ p 1 ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 I iii ( k ; p 3 ) , (53)
and
I i v ( k ; ϵ 3 ( p 3 ; p 1 ) ) = I i v , a ( k ) + I i v , b ( k ) , I i v , a ( k ) = I i v ( k ; ϵ 3 ( p 3 ; p 2 ) ) , I i v , b ( k ) = − ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 p 3 ⋅ p 1 I i v ( k ; p 3 ) . (54) \begin{aligned}
\mathcal{I}_{\mathrm{iv}}(k;\epsilon_3(p_3;p_1)) &= \mathcal{I}_{\mathrm{iv},a}(k) + \mathcal{I}_{\mathrm{iv},b}(k),\\
\mathcal{I}_{\mathrm{iv},a}(k) &= \mathcal{I}_{\mathrm{iv}}(k;\epsilon_3(p_3;p_2)),\\
\mathcal{I}_{\mathrm{iv},b}(k) &= -\frac{\epsilon_3(p_3;p_2)\cdot p_1}{p_3\cdot p_1}\mathcal{I}_{\mathrm{iv}}(k;p_3). \tag*{(54)}
\end{aligned} I iv ( k ; ϵ 3 ( p 3 ; p 1 )) I iv , a ( k ) I iv , b ( k ) = I iv , a ( k ) + I iv , b ( k ) , = I iv ( k ; ϵ 3 ( p 3 ; p 2 )) , = − p 3 ⋅ p 1 ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 I iv ( k ; p 3 ) . (54)
Following Eq. (43) , in i i i a , i v a {\rm iii}_{a},{\rm iv}_{a} iii a , iv a we replace the k , r k,r k , r gluons by their G counterparts and find that both graphs are power counting finite.
Let us turn our attention to ( i i i ) b , ( i v ) b ({\rm iii})_{b},({\rm iv})_{b} ( iii ) b , ( iv ) b . Here, the lines labeled by r , k r,k r , k can be replaced by their one sided G components, since as usual, the K components vanish by the fermion loop Ward identity in Eq. (21) . However, as we will establish presently, the K components of ℓ = k + p 3 \ell=k+p_{3} ℓ = k + p 3 also vanishes after summing over ( i i i ) b , ( i v ) b ({\rm iii})_{b},({\rm iv})_{b} ( iii ) b , ( iv ) b and we may replace ℓ \ell ℓ propagator by its G component.
I i i i , b ( G ) ( k ) = g s f c d x ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 p 3 ⋅ p 1 v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 ) \mathcal{I}^{(G)}_{{\rm iii},b}(k)=g_s f^{cdx}\frac{\epsilon_3(p_3;p_2)\cdot p_1}{p_3\cdot p_1}\bar{v}(p_2)\Gamma^c_{\rho}S_0(\ell-p_2)\Gamma^b_{\eta}u(p_1) I iii , b ( G ) ( k ) = g s f c d x p 3 ⋅ p 1 ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 )
× D G , ρ μ ( ℓ ; q ) D G , η τ ( r ; q ) g μ ν p 3 ⋅ ( 2 k + p 3 ) D G , ν σ ( k ; q ) H σ τ d b [ 2 , m ] ( k , r ) . (55) \qquad{}\times D^{G,\rho\mu}(\ell;q)D^{G,\eta\tau}(r;q)g_{\mu\nu}p_3\cdot(2k+p_3)D^{G,\nu\sigma}(k;q)H^{db}_{\sigma\tau}[2,m](k,r). \tag*{(55)} × D G , ρ μ ( ℓ ; q ) D G , η τ ( r ; q ) g μν p 3 ⋅ ( 2 k + p 3 ) D G , ν σ ( k ; q ) H σ τ d b [ 2 , m ] ( k , r ) . (55)
I i v , b ( G ) ( k ) = − g s f c d x ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 p 3 ⋅ p 1 v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 ) \mathcal{I}^{(G)}_{{\rm iv},b}(k)=-g_s f^{cdx}\frac{\epsilon_3(p_3;p_2)\cdot p_1}{p_3\cdot p_1}\bar{v}(p_2)\Gamma^c_{\rho}S_0(\ell-p_2)\Gamma^b_{\eta}u(p_1) I iv , b ( G ) ( k ) = − g s f c d x p 3 ⋅ p 1 ϵ 3 ( p 3 ; p 2 ) ⋅ p 1 v ˉ ( p 2 ) Γ ρ c S 0 ( ℓ − p 2 ) Γ η b u ( p 1 )
× D G , ρ μ ( ℓ ; q ) D G , η τ ( r ; q ) p 3 μ ( k + 2 p 3 ) ν D G , ν σ ( k ; q ) H σ τ d b [ 2 , m ] ( k , r ) . (56) \qquad{}\times D^{G,\rho\mu}(\ell;q)D^{G,\eta\tau}(r;q)p_{3\mu}(k+2p_3)_{\nu}D^{G,\nu\sigma}(k;q)H^{db}_{\sigma\tau}[2,m](k,r). \tag*{(56)} × D G , ρ μ ( ℓ ; q ) D G , η τ ( r ; q ) p 3 μ ( k + 2 p 3 ) ν D G , ν σ ( k ; q ) H σ τ d b [ 2 , m ] ( k , r ) . (56)
This equality is shown in Figure 11 Using the usual power counting relations, one may argue that that the second panel of Figure 11 is indeed power counting finite.
Figure 11. The G gluonic equality in Eqs. (55) and (56)
Before concluding this section, let us check that the K gluonic component of ℓ \ell ℓ vanishes identically in the polarization changing terms. The K component of the ℓ = k + p 3 \ell=k+p_3 ℓ = k + p 3 propagator contracts the Scalar-Scalar-Gluon vertex with p 3 μ p_3^\mu p 3 μ . Summing the two SSG vertices in Eqs. (51) and (52) , and contracting with the longitudinal polarization p 3 λ p_3^\lambda p 3 λ , we find
ℓ μ p 3 λ [ − g s f c d x g μ ν ( 2 k + p 3 ) λ + g s f c d x g λ μ ( k + 2 p 3 ) ν ] = g s f c d x [ − ( ℓ 2 − k 2 ) ℓ ν + ( ℓ ⋅ p 3 ) ( k + 2 p 3 ) ν ] = − g s f c d x ( k ⋅ p 3 ) k ν , (57) \begin{aligned}
\ell^\mu p_3^\lambda\left[-g_s f^{cdx}g_{\mu\nu}(2k+p_3)_\lambda+g_s f^{cdx}g_{\lambda\mu}(k+2p_3)_\nu\right]
&=g_s f^{cdx}\left[-(\ell^2-k^2)\ell_\nu+(\ell\cdot p_3)(k+2p_3)_\nu\right]\\
&=-g_s f^{cdx}(k\cdot p_3)k_\nu, \tag*{(57)}
\end{aligned} ℓ μ p 3 λ [ − g s f c d x g μν ( 2 k + p 3 ) λ + g s f c d x g λ μ ( k + 2 p 3 ) ν ] = g s f c d x [ − ( ℓ 2 − k 2 ) ℓ ν + ( ℓ ⋅ p 3 ) ( k + 2 p 3 ) ν ] = − g s f c d x ( k ⋅ p 3 ) k ν , (57)
where we have used ℓ = k + p 3 \ell= k + p_3 ℓ = k + p 3 and p 3 2 = 0 p_3^2 = 0 p 3 2 = 0 . The remaining longitudinal momentum k ν k_\nu k ν vanishes upon contraction with the hard loop, since
k ν D ν σ ( k ) H σ τ d b [ 2 , m ] ( k , r ) = − i k 2 + i 0 k σ H σ τ d b [ 2 , m ] ( k , r ) = 0. (58) k_\nu D^{\nu\sigma}(k)H^{db}_{\sigma\tau}[2,m](k,r) = \frac{-i}{k^2+i0}k^\sigma H^{db}_{\sigma\tau}[2,m](k,r) = 0. \tag*{(58)} k ν D ν σ ( k ) H σ τ d b [ 2 , m ] ( k , r ) = k 2 + i 0 − i k σ H σ τ d b [ 2 , m ] ( k , r ) = 0. (58)
Thus, the K component of the ℓ \ell ℓ propagator cancels exactly between (iii)b _b b and (iv)b _b b , and the sum is unchanged upon replacing this propagator by its G component, as shown in Figure 11 .
To summarize, in this section, we have seen that all graphs with transient infrared divergences can be made locally finite by replacing some propagators by appropriately defined G gluonic counterparts. While the exact replacement needed varies from graph to graph, in each case, we utilize a ward identity that is manifest after the hard loop is integrated over and all permutations within the hard loop are summed over. In the next section, we will turn our attention to the genuinely infrared divergent graphs contained in M I R \mathcal{M}_{\mathrm{IR}} M IR .
Infrared divergent graphs
In this section, we study the IR divergent graphs in Figure 3 . In all the graphs of interest, we will apply the G-K decomposition and find that any graph with at least one instance of a K gluon factorizes into a product of lower loop graphs. The G gluonic graphs will be collinear finite while its soft limit will be handled with explicit counterterms.
We begin with graph (IV), decomposing its three-gluon vertex into three terms,
I I V ( k ) = I I V , a ( k ) + I I V , b ( k ) + I I V , c ( k ) . (59) \mathcal{I}_{\mathrm{IV}}(k) = \mathcal{I}_{\mathrm{IV},a}(k) + \mathcal{I}_{\mathrm{IV},b}(k) + \mathcal{I}_{\mathrm{IV},c}(k). \tag*{(59)} I IV ( k ) = I IV , a ( k ) + I IV , b ( k ) + I IV , c ( k ) . (59)
The decomposition is shown in Figure 12
!Scalar-scalar-gluon decomposition diagram labelled “(IV): momentum routing”, with momenta p 1 p_1 p 1 , p 2 p_2 p 2 , P P P , k k k , P − k P-k P − k , and k − p 3 k-p_3 k − p 3 , and H [ 2 , m ] \mathcal{H}[2,m] H [ 2 , m ] ; decomposition into ( I V ) a (\mathrm{IV})_a ( IV ) a , ( I V ) b (\mathrm{IV})_b ( IV ) b , and ( I V ) c (\mathrm{IV})_c ( IV ) c . bbox=0.13,0.14,0.88,0.43
Figure 12. The scalar-scalar-gluon decomposition of graph (IV).
We will show that the last two terms are locally infrared finite.
With the momenta entering the first three-gluon vertex given by ( P , − k , k − P ) (P,-k,k-P) ( P , − k , k − P ) , the sum of the two relevant SSG vertices is
[ V b + c a b c ] μ ν α = − g s f a b c [ g μ ν ( P + k ) α + g ν α ( P − 2 k ) μ ] . (60) [V^{abc}_{b+c}]_{\mu\nu\alpha} = -g_s f^{abc}\left[g_{\mu\nu}(P+k)_\alpha+ g_{\nu\alpha}(P-2k)_\mu\right]. \tag*{(60)} [ V b + c ab c ] μν α = − g s f ab c [ g μν ( P + k ) α + g ν α ( P − 2 k ) μ ] . (60)
The corresponding integrand is
I I V , b + c ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V b + c a b c ] μ ν α D ν β ( k ) × g s f b x d ( 2 p 3 − k ) β ϵ 3 σ D α γ ( P − k ) D σ δ ( k − p 3 ) H γ δ c d [ 2 , m ] ( P − k , k − p 3 ) . (61) \begin{aligned}
\mathcal{I}_{\mathrm{IV},b+c}(k) ={}& \bar{v}(p_2)\Gamma^a_\rho u(p_1)D^{\rho\mu}(P)[V^{abc}_{b+c}]_{\mu\nu\alpha}D^{\nu\beta}(k)\\
&\times g_s f^{bxd}(2p_3-k)_\beta\epsilon_{3\sigma}D^{\alpha\gamma}(P-k)D^{\sigma\delta}(k-p_3)H^{cd}_{\gamma\delta}[2,m](P-k,k-p_3). \tag*{(61)}
\end{aligned} I IV , b + c ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V b + c ab c ] μν α D ν β ( k ) × g s f b x d ( 2 p 3 − k ) β ϵ 3 σ D α γ ( P − k ) D σ δ ( k − p 3 ) H γ δ c d [ 2 , m ] ( P − k , k − p 3 ) . (61)
Let us now check that the K K K gluonic component of the k k k propagator is vanishing. First, contracting k ν k^\nu k ν into the vertex, we get
k ν [ V b + c a b c ] μ ν α = − g s f a b c [ k μ ( P + k ) α + k α ( P − 2 k ) μ ] = − g s f a b c [ P μ k α + k μ ( P − k ) α ] . (62) k^\nu[V^{abc}_{b+c}]_{\mu\nu\alpha} = -g_s f^{abc}\left[k_\mu(P+k)_\alpha+ k_\alpha(P-2k)_\mu\right] = -g_s f^{abc}\left[P_\mu k_\alpha+ k_\mu(P-k)_\alpha\right]. \tag*{(62)} k ν [ V b + c ab c ] μν α = − g s f ab c [ k μ ( P + k ) α + k α ( P − 2 k ) μ ] = − g s f ab c [ P μ k α + k μ ( P − k ) α ] . (62)
This identity of the three gluon vertex is shown in Figure 13
Figure 13. Ward identity of the three gluon vertex
Next, we use the Ward identities,
v ˉ ( p 2 ) Γ μ a u ( p 1 ) P μ = 0 H γ δ c d [ 2 , m ] ( P − k , k − p 3 ) ( P − k ) γ = 0. (63) \begin{aligned}
\bar{v}(p_2)\Gamma^a_\mu u(p_1)P_\mu&= 0 \\
\mathcal{H}^{cd}_{\gamma\delta}[2,m](P-k,k-p_3)(P-k)_\gamma&= 0. \tag*{(63)}
\end{aligned} v ˉ ( p 2 ) Γ μ a u ( p 1 ) P μ H γ δ c d [ 2 , m ] ( P − k , k − p 3 ) ( P − k ) γ = 0 = 0. (63)
The K component of the k k k propagator therefore vanishes in the sum of the two graphs. We may consequently replace
I I V , b + c G ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V b + c a b c ] μ ν α D G , ν β ( k ) × g s f b x d ( 2 p 3 − k ) β ϵ 3 σ D α γ ( P − k ) D σ δ ( k − p 3 ) H γ δ c d [ 2 , m ] ( P − k , k − p 3 ) (64) \mathcal{I}^{G}_{\mathrm{IV},b+c}(k)=\bar{v}(p_2)\Gamma^a_\rho u(p_1)D^{\rho\mu}(P)[V^{abc}_{b+c}]_{\mu\nu\alpha}D^{G,\nu\beta}(k)
\times g_sf^{bxd}(2p_3-k)_\beta\epsilon_{3\sigma}D^{\alpha\gamma}(P-k)D^{\sigma\delta}(k-p_3)\mathcal{H}^{cd}_{\gamma\delta}[2,m](P-k,k-p_3) \tag*{(64)} I IV , b + c G ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V b + c ab c ] μν α D G , ν β ( k ) × g s f b x d ( 2 p 3 − k ) β ϵ 3 σ D α γ ( P − k ) D σ δ ( k − p 3 ) H γ δ c d [ 2 , m ] ( P − k , k − p 3 ) (64)
without changing their sum.
We can now establish local finiteness. The only potentially collinear region is k ∥ p 3 k \parallel p_3 k ∥ p 3 , where both k 2 k^2 k 2 and ( k − p 3 ) 2 (k-p_3)^2 ( k − p 3 ) 2 vanish. In this region,
( 2 p 3 − k ) β G ν β ( k ; q ) = − 2 ( k − p 3 ) β G ν β ( k ; q ) = O ( k ⊥ ) , (65) (2p_3-k)_\beta G^{\beta}_{\nu}(k;q)=-2(k-p_3)_\beta G^{\beta}_{\nu}(k;q)=O(k_\perp), \tag*{(65)} ( 2 p 3 − k ) β G ν β ( k ; q ) = − 2 ( k − p 3 ) β G ν β ( k ; q ) = O ( k ⊥ ) , (65)
where we used k β G ν β ( k ; q ) = 0 k_\beta G^{\beta}_{\nu}(k;q)=0 k β G ν β ( k ; q ) = 0 . Since the two propagator denominators are O ( k ⊥ 2 ) O(k_\perp^2) O ( k ⊥ 2 ) , this additional transverse-momentum suppression renders the collinear integral finite. Continuing with our treatment of divergent graphs, we first consider the three graphs (I)–(III), with the momenta of the upper and lower gluons entering the massless quark as in Figure 14 line taken to be r = P − k r=P-k r = P − k and k k k , respectively.
Figure 14. Graphs (I)–(III), with one momentum routing throughout. The external p 3 p_3 p 3 line is dotted at an SSG vertex and wavy at the ordinary gluonic attachment to the hard loop.
Their integrands are
I I ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D μ α ( r ) D ρ τ ( k ) ϵ 3 λ [ V S S G a c x ] α β λ ( r , p 3 − r , − p 3 ) D β ν ( r − p 3 ) H ν τ c b [ 2 , m ] ( r − p 3 , k ) , (66) \mathcal{I}_{\mathrm{I}}(k)=\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D^{\mu\alpha}(r)D^{\rho\tau}(k)\epsilon^\lambda_3
[V^{acx}_{\mathrm{SSG}}]_{\alpha\beta\lambda}(r,p_3-r,-p_3)D^{\beta\nu}(r-p_3)\mathcal{H}^{cb}_{\nu\tau}[2,m](r-p_3,k), \tag*{(66)} I I ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D μα ( r ) D ρ τ ( k ) ϵ 3 λ [ V SSG a c x ] α β λ ( r , p 3 − r , − p 3 ) D β ν ( r − p 3 ) H ν τ c b [ 2 , m ] ( r − p 3 , k ) , (66)
I I I ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D μ ν ( r ) D ρ α ( k ) ϵ 3 λ [ V S S G b c x ] α β λ ( k , p 3 − k , − p 3 ) D β τ ( k − p 3 ) H ν τ a c [ 2 , m ] ( r , k − p 3 ) , (67) \mathcal{I}_{\mathrm{II}}(k)=\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D^{\mu\nu}(r)D^{\rho\alpha}(k)\epsilon^\lambda_3
[V^{bcx}_{\mathrm{SSG}}]_{\alpha\beta\lambda}(k,p_3-k,-p_3)D^{\beta\tau}(k-p_3)\mathcal{H}^{ac}_{\nu\tau}[2,m](r,k-p_3), \tag*{(67)} I II ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D μν ( r ) D ρ α ( k ) ϵ 3 λ [ V SSG b c x ] α β λ ( k , p 3 − k , − p 3 ) D β τ ( k − p 3 ) H ν τ a c [ 2 , m ] ( r , k − p 3 ) , (67)
I I I I ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D μ ν ( r ) D ρ τ ( k ) ϵ 3 λ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) . (68) \mathcal{I}_{\mathrm{III}}(k)=\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D^{\mu\nu}(r)D^{\rho\tau}(k)\epsilon^\lambda_3\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_3). \tag*{(68)} I III ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D μν ( r ) D ρ τ ( k ) ϵ 3 λ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) . (68)
The two gluons entering H [ 3 , m ] \mathcal{H}[3,m] H [ 3 , m ] are distinctly labelled, and the corresponding two permutations cancel the 1 / 2 ! 1/2! 1/2 ! in the unlabelled graph sum.
With the quark–gluon rule Γ μ a = − i g s T a γ μ \Gamma^{a}_{\mu}=-ig_{s}T^{a}\gamma_{\mu} Γ μ a = − i g s T a γ μ , the Ward identity follows from the telescoping of the fermion loop identity color commutator,
S t ( ℓ ) ( − i g s T a r̸ ) S t ( ℓ + r ) = g s T a [ S t ( ℓ ) − S t ( ℓ + r ) ] , S_{t}(\ell)(-ig_{s}T^{a}\not{r})S_{t}(\ell+r)=g_{s}T^{a}\left[S_{t}(\ell)-S_{t}(\ell+r)\right], S t ( ℓ ) ( − i g s T a r ) S t ( ℓ + r ) = g s T a [ S t ( ℓ ) − S t ( ℓ + r ) ] ,
g s ( T b T a − T a T b ) = − i g s f a b c T c . (69) g_{s}(T^{b}T^{a}-T^{a}T^{b})=-ig_{s}f^{abc}T^{c}. \tag*{(69)} g s ( T b T a − T a T b ) = − i g s f ab c T c . (69)
The corresponding non-Abelian Ward identity for the integrated hard loop is
k 1 μ 1 H μ 1 μ 2 μ 3 a 1 a 2 a 3 [ 3 , m ] ( k 1 , k 2 , k 3 ) = − i g s f a 1 a 2 c H μ 2 μ 3 c a 3 [ 2 , m ] ( k 1 + k 2 , k 3 ) − i g s f a 1 a 3 c H μ 2 μ 3 a 2 c [ 2 , m ] ( k 2 , k 1 + k 3 ) . (70) k_{1}^{\mu_{1}}\mathcal{H}^{a_{1}a_{2}a_{3}}_{\mu_{1}\mu_{2}\mu_{3}}[3,m](k_{1},k_{2},k_{3})=-ig_{s}f^{a_{1}a_{2}c}\mathcal{H}^{ca_{3}}_{\mu_{2}\mu_{3}}[2,m](k_{1}+k_{2},k_{3})
-ig_{s}f^{a_{1}a_{3}c}\mathcal{H}^{a_{2}c}_{\mu_{2}\mu_{3}}[2,m](k_{2},k_{1}+k_{3}). \tag*{(70)} k 1 μ 1 H μ 1 μ 2 μ 3 a 1 a 2 a 3 [ 3 , m ] ( k 1 , k 2 , k 3 ) = − i g s f a 1 a 2 c H μ 2 μ 3 c a 3 [ 2 , m ] ( k 1 + k 2 , k 3 ) − i g s f a 1 a 3 c H μ 2 μ 3 a 2 c [ 2 , m ] ( k 2 , k 1 + k 3 ) . (70)
In particular, the two contractions in (III) are
r ν H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) = − i g s f a b c H τ λ c x [ 2 , m ] ( P , − p 3 ) − i g s f a x c H τ λ b c [ 2 , m ] ( k , r − p 3 ) , r^{\nu}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3})=-ig_{s}f^{abc}\mathcal{H}^{cx}_{\tau\lambda}[2,m](P,-p_{3})-ig_{s}f^{axc}\mathcal{H}^{bc}_{\tau\lambda}[2,m](k,r-p_{3}), r ν H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) = − i g s f ab c H τ λ c x [ 2 , m ] ( P , − p 3 ) − i g s f a x c H τ λ b c [ 2 , m ] ( k , r − p 3 ) ,
k τ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) = − i g s f b a c H ν λ c x [ 2 , m ] ( P , − p 3 ) − i g s f b x c H ν λ a c [ 2 , m ] ( r , k − p 3 ) . (71) k^{\tau}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3})=-ig_{s}f^{bac}\mathcal{H}^{cx}_{\nu\lambda}[2,m](P,-p_{3})-ig_{s}f^{bxc}\mathcal{H}^{ac}_{\nu\lambda}[2,m](r,k-p_{3}). \tag*{(71)} k τ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) = − i g s f ba c H ν λ c x [ 2 , m ] ( P , − p 3 ) − i g s f b x c H ν λ a c [ 2 , m ] ( r , k − p 3 ) . (71)
First decompose the propagators of (III) only at their ends adjacent to H [ 3 , m ] \mathcal{H}[3,m] H [ 3 , m ] . In the notation of Figure 15 ,
Figure 15. The one-sided G–K decomposition of the two gluons entering the three-gluon hard loop of graph (III). The gluon of momentum r = P − k r=P-k r = P − k is drawn above the gluon of momentum k k k .
I I I I ( k ) = I I I I , f i n ( k ) + I I I I , α ( k ) + I I I I , β ( k ) + I I I I , γ ( k ) , \mathcal{I}_{\mathrm{III}}(k)=\mathcal{I}_{\mathrm{III},\mathrm{fin}}(k)+\mathcal{I}_{\mathrm{III},\alpha}(k)+\mathcal{I}_{\mathrm{III},\beta}(k)+\mathcal{I}_{\mathrm{III},\gamma}(k), I III ( k ) = I III , fin ( k ) + I III , α ( k ) + I III , β ( k ) + I III , γ ( k ) ,
I I I I , f i n ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μ ν ( r ; q ) D G ρ τ ( k ; q ) ϵ 3 λ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) , \mathcal{I}_{\mathrm{III},\mathrm{fin}}(k)=\bar{v}(p_{2})\Gamma^{b}_{\rho}S_{0}(k-p_{2})\Gamma^{a}_{\mu}u(p_{1})D^{\mu\nu}_{G}(r;q)D^{\rho\tau}_{G}(k;q)\epsilon_{3}^{\lambda}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3}), I III , fin ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μν ( r ; q ) D G ρ τ ( k ; q ) ϵ 3 λ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) ,
I I I I , α ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μ ν ( r ; q ) D K ρ τ ( k ; q ) ϵ 3 λ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) , \mathcal{I}_{\mathrm{III},\alpha}(k)=\bar{v}(p_{2})\Gamma^{b}_{\rho}S_{0}(k-p_{2})\Gamma^{a}_{\mu}u(p_{1})D^{\mu\nu}_{G}(r;q)D^{\rho\tau}_{K}(k;q)\epsilon_{3}^{\lambda}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3}), I III , α ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μν ( r ; q ) D K ρ τ ( k ; q ) ϵ 3 λ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) ,
I I I I , β ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D K μ ν ( r ; q ) D G ρ τ ( k ; q ) ϵ 3 λ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) , \mathcal{I}_{\mathrm{III},\beta}(k)=\bar{v}(p_{2})\Gamma^{b}_{\rho}S_{0}(k-p_{2})\Gamma^{a}_{\mu}u(p_{1})D^{\mu\nu}_{K}(r;q)D^{\rho\tau}_{G}(k;q)\epsilon_{3}^{\lambda}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3}), I III , β ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D K μν ( r ; q ) D G ρ τ ( k ; q ) ϵ 3 λ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) ,
I I I I , γ ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D K μ ν ( r ; q ) D K ρ τ ( k ; q ) ϵ 3 λ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) . (72) \mathcal{I}_{\mathrm{III},\gamma}(k)=\bar{v}(p_{2})\Gamma^{b}_{\rho}S_{0}(k-p_{2})\Gamma^{a}_{\mu}u(p_{1})D^{\mu\nu}_{K}(r;q)D^{\rho\tau}_{K}(k;q)\epsilon_{3}^{\lambda}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3}). \tag*{(72)} I III , γ ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D K μν ( r ; q ) D K ρ τ ( k ; q ) ϵ 3 λ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) . (72)
The first term is finite in the r ∥ p 1 r\parallel p_{1} r ∥ p 1 and k ∥ p 2 k\parallel p_{2} k ∥ p 2 regions. This follows from the collinear suppression of the G gluons discussed in § 3 . The double-longitudinal contraction ( I γ ) (\mathcal{I}_{\gamma}) ( I γ ) is non zero but factorizes from the hard loop completely:
r ν k τ H ν τ λ a b x [ 3 , m ] ( r , k , − p 3 ) = − i g s f a b c k τ H τ λ c x [ 2 , m ] ( P , − p 3 ) . (73) r^{\nu}k^{\tau}\mathcal{H}^{abx}_{\nu\tau\lambda}[3,m](r,k,-p_{3})=-ig_{s}f^{abc}k^{\tau}\mathcal{H}^{cx}_{\tau\lambda}[2,m](P,-p_{3}). \tag*{(73)} r ν k τ H ν τ λ ab x [ 3 , m ] ( r , k , − p 3 ) = − i g s f ab c k τ H τ λ c x [ 2 , m ] ( P , − p 3 ) . (73)
Finally, applying the non abelian ward identity in Eq. (70) to I I I I , α / β \mathcal{I}_{\mathrm{III},\alpha/\beta} I III , α / β , we get
I I I I , α = I α , n f + I α , f a c t I I I I , β = I β , n f + I β , f a c t I α , n f = − i g s f b x c v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μ ν ( r ; q ) − i ( 2 q + k ) ρ ( k 2 + i 0 ) ( k 2 + 2 q ⋅ k + i 0 ) ϵ 3 λ H λ ν a c [ 2 , m ] ( r , k − p 3 ) , I β , n f = − i g s f a x c v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) − i ( 2 q + r ) μ ( r 2 + i 0 ) ( r 2 + 2 q ⋅ r + i 0 ) D G ρ τ ( k ; q ) ϵ 3 λ H τ λ b c [ 2 , m ] ( k , r − p 3 ) . (74) \begin{aligned}
\mathcal{I}_{\mathrm{III},\alpha} &= \mathcal{I}_{\alpha,\mathrm{nf}}+\mathcal{I}_{\alpha,\mathrm{fact}} \\
\mathcal{I}_{\mathrm{III},\beta} &= \mathcal{I}_{\beta,\mathrm{nf}}+\mathcal{I}_{\beta,\mathrm{fact}} \\
\mathcal{I}_{\alpha,\mathrm{nf}} &= -ig_s f^{bxc}\,\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D_G^{\mu\nu}(r;q) \frac{-i(2q+k)^\rho}{(k^2+i0)(k^2+2q\cdot k+i0)}\epsilon_3^\lambda\mathcal{H}^{ac}_{\lambda\nu}[2,m](r,k-p_3), \\
\mathcal{I}_{\beta,\mathrm{nf}} &= -ig_s f^{axc}\,\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1) \frac{-i(2q+r)^\mu}{(r^2+i0)(r^2+2q\cdot r+i0)}D_G^{\rho\tau}(k;q)\epsilon_3^\lambda\mathcal{H}^{bc}_{\tau\lambda}[2,m](k,r-p_3). \tag*{(74)}
\end{aligned} I III , α I III , β I α , nf I β , nf = I α , nf + I α , fact = I β , nf + I β , fact = − i g s f b x c v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μν ( r ; q ) ( k 2 + i 0 ) ( k 2 + 2 q ⋅ k + i 0 ) − i ( 2 q + k ) ρ ϵ 3 λ H λ ν a c [ 2 , m ] ( r , k − p 3 ) , = − i g s f a x c v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) ( r 2 + i 0 ) ( r 2 + 2 q ⋅ r + i 0 ) − i ( 2 q + r ) μ D G ρ τ ( k ; q ) ϵ 3 λ H τ λ b c [ 2 , m ] ( k , r − p 3 ) . (74)
For graphs (I) and (II), the Abelian Ward identity of H [ 2 , m ] \mathcal{H}[2,m] H [ 2 , m ] first permits D ( k ) → D G ( k ; q ) D(k)\to D_G(k;q) D ( k ) → D G ( k ; q ) in (I), and D ( r ) → D G ( r ; q ) D(r)\to D_G(r;q) D ( r ) → D G ( r ; q ) in (II). We then decompose the remaining propagator at the SSG vertex,
I I ( k ) = I I , δ ( k ) + I I , a ( k ) , I I I ( k ) = I I I , σ ( k ) + I I I , a ( k ) . (75) \begin{aligned}
\mathcal{I}_{\mathrm{I}}(k) &= \mathcal{I}_{\mathrm{I},\delta}(k)+\mathcal{I}_{\mathrm{I},a}(k), \\
\mathcal{I}_{\mathrm{II}}(k) &= \mathcal{I}_{\mathrm{II},\sigma}(k)+\mathcal{I}_{\mathrm{II},a}(k). \tag*{(75)}
\end{aligned} I I ( k ) I II ( k ) = I I , δ ( k ) + I I , a ( k ) , = I II , σ ( k ) + I II , a ( k ) . (75)
Here the labels δ \delta δ and σ \sigma σ designate the longitudinal contributions shown in Figure 16 . Their
Figure 16. One-sided G–K decomposition of graphs (I),(II) at the SSG ends, after replacing the other propagator by its G component using the two-gluon hard-loop Ward identity.
longitudinal contractions are
r α [ V S S G a c x , ( 23 ) ] α β λ ( r , p 3 − r , − p 3 ) = g s f a c x ( r − p 3 ) 2 g β λ , k α [ V S S G b c x , ( 23 ) ] α β λ ( k , p 3 − k , − p 3 ) = g s f b c x ( k − p 3 ) 2 g β λ . (76) \begin{aligned}
r^\alpha[V_{\mathrm{SSG}}^{acx,(23)}]_{\alpha\beta\lambda}(r,p_3-r,-p_3) &= g_s f^{acx}(r-p_3)^2g_{\beta\lambda}, \\
k^\alpha[V_{\mathrm{SSG}}^{bcx,(23)}]_{\alpha\beta\lambda}(k,p_3-k,-p_3) &= g_s f^{bcx}(k-p_3)^2g_{\beta\lambda}. \tag*{(76)}
\end{aligned} r α [ V SSG a c x , ( 23 ) ] α β λ ( r , p 3 − r , − p 3 ) k α [ V SSG b c x , ( 23 ) ] α β λ ( k , p 3 − k , − p 3 ) = g s f a c x ( r − p 3 ) 2 g β λ , = g s f b c x ( k − p 3 ) 2 g β λ . (76)
Inserting these equations into the two longitudinal terms gives
I I , δ ( k ) = − i g s f a c x v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) − i r 2 + i 0 ( 2 q + r ) μ r 2 + 2 q ⋅ r + i 0 × D G ρ τ ( k ; q ) ϵ 3 λ H λ τ c b [ 2 , m ] ( r − p 3 , k ) , (77) \mathcal{I}_{I,\delta}(k)=-ig_sf^{acx}\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)\frac{-i}{r^2+i0}\frac{(2q+r)^\mu}{r^2+2q\cdot r+i0}\times D_G^{\rho\tau}(k;q)\epsilon_3^\lambda\mathcal{H}^{cb}_{\lambda\tau}[2,m](r-p_3,k), \tag*{(77)} I I , δ ( k ) = − i g s f a c x v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) r 2 + i 0 − i r 2 + 2 q ⋅ r + i 0 ( 2 q + r ) μ × D G ρ τ ( k ; q ) ϵ 3 λ H λ τ c b [ 2 , m ] ( r − p 3 , k ) , (77)
I I I , σ ( k ) = − i g s f b c x v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μ ν ( r ; q ) − i k 2 + i 0 ( 2 q + k ) ρ k 2 + 2 q ⋅ k + i 0 × ϵ 3 λ H ν λ a c [ 2 , m ] ( r , k − p 3 ) . (78) \mathcal{I}_{II,\sigma}(k)=-ig_sf^{bcx}\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D_G^{\mu\nu}(r;q)\frac{-i}{k^2+i0}\frac{(2q+k)^\rho}{k^2+2q\cdot k+i0}\times\epsilon_3^\lambda\mathcal{H}^{ac}_{\nu\lambda}[2,m](r,k-p_3). \tag*{(78)} I I I , σ ( k ) = − i g s f b c x v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μν ( r ; q ) k 2 + i 0 − i k 2 + 2 q ⋅ k + i 0 ( 2 q + k ) ρ × ϵ 3 λ H ν λ a c [ 2 , m ] ( r , k − p 3 ) . (78)
The two nonfactorizing terms generated by the hard-loop Ward identity in graph III, given in Eq. (74) cancel these terms at fixed k k k :
I I I I , β , n f ( k ) + I I , δ ( k ) = 0 , \mathcal{I}_{III,\beta,\mathrm{nf}}(k)+\mathcal{I}_{I,\delta}(k)=0, I I I I , β , nf ( k ) + I I , δ ( k ) = 0 ,
I I I I , α , n f ( k ) + I I I , σ ( k ) = 0. (79) \mathcal{I}_{III,\alpha,\mathrm{nf}}(k)+\mathcal{I}_{II,\sigma}(k)=0. \tag*{(79)} I I I I , α , nf ( k ) + I I I , σ ( k ) = 0. (79)
Adding all five longitudinal contributions, the entire sum reduces to three explicitly factorized terms:
I I I I , α ( k ) + I I I I , β ( k ) + I I I I , γ ( k ) + I I , δ ( k ) + I I I , σ ( k ) = i g s f a b c v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) ϵ 3 λ { D G μ ν ( r ; q ) − i k 2 + i 0 ( 2 q + k ) ρ k 2 + 2 q ⋅ k + i 0 H ν λ c x [ 2 , m ] ( P , − p 3 ) − − i r 2 + i 0 ( 2 q + r ) μ r 2 + 2 q ⋅ r + i 0 D G ρ τ ( k ; q ) H τ λ c x [ 2 , m ] ( P , − p 3 ) − ( − i ) 2 ( r 2 + i 0 ) ( k 2 + i 0 ) ( r 2 + 2 q ⋅ r + i 0 ) ( k 2 + 2 q ⋅ k + i 0 ) ( 2 q + r ) μ ( 2 q + k ) ρ k τ H τ λ c x [ 2 , m ] ( P , − p 3 ) } . (80) \begin{aligned}
\mathcal{I}_{III,\alpha}(k)+\mathcal{I}_{III,\beta}(k)+\mathcal{I}_{III,\gamma}(k)+\mathcal{I}_{I,\delta}(k)+\mathcal{I}_{II,\sigma}(k)
={}&ig_sf^{abc}\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)\epsilon_3^\lambda\Bigg\{D_G^{\mu\nu}(r;q)\frac{-i}{k^2+i0}\frac{(2q+k)^\rho}{k^2+2q\cdot k+i0}\mathcal{H}^{cx}_{\nu\lambda}[2,m](P,-p_3)\\
&-\frac{-i}{r^2+i0}\frac{(2q+r)^\mu}{r^2+2q\cdot r+i0}D_G^{\rho\tau}(k;q)\mathcal{H}^{cx}_{\tau\lambda}[2,m](P,-p_3)\\
&-\frac{(-i)^2}{(r^2+i0)(k^2+i0)(r^2+2q\cdot r+i0)(k^2+2q\cdot k+i0)}(2q+r)^\mu(2q+k)^\rho k^\tau\mathcal{H}^{cx}_{\tau\lambda}[2,m](P,-p_3)\Bigg\}. \tag*{(80)}
\end{aligned} I I I I , α ( k ) + I I I I , β ( k ) + I I I I , γ ( k ) + I I , δ ( k ) + I I I , σ ( k ) = i g s f ab c v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) ϵ 3 λ { D G μν ( r ; q ) k 2 + i 0 − i k 2 + 2 q ⋅ k + i 0 ( 2 q + k ) ρ H ν λ c x [ 2 , m ] ( P , − p 3 ) − r 2 + i 0 − i r 2 + 2 q ⋅ r + i 0 ( 2 q + r ) μ D G ρ τ ( k ; q ) H τ λ c x [ 2 , m ] ( P , − p 3 ) − ( r 2 + i 0 ) ( k 2 + i 0 ) ( r 2 + 2 q ⋅ r + i 0 ) ( k 2 + 2 q ⋅ k + i 0 ) ( − i ) 2 ( 2 q + r ) μ ( 2 q + k ) ρ k τ H τ λ c x [ 2 , m ] ( P , − p 3 ) } . (80)
These are fully factorized, exact graphical identities.
The remaining parts of graphs (I) and (II), denoted by I I , a \mathcal{I}_{I,a} I I , a and I I I , a \mathcal{I}_{II,a} I I I , a , contain a one-sided G projector pointing toward the fermion loop as well as the triple gluon vertex for the r r r and k k k propagators, respectively. Explicitly they are
I I , a ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) − i r 2 + i 0 G ξ μ ( r ; q ) g ξ α D G ρ τ ( k ; q ) ϵ 3 λ × [ V S S G a c x , ( 23 ) ] α β λ ( r , p 3 − r , − p 3 ) D β ν ( r − p 3 ) H ν τ c b [ 2 , m ] ( r − p 3 , k ) , (81) \mathcal{I}_{I,a}(k)=\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)\frac{-i}{r^2+i0}G^\mu_\xi(r;q)g^{\xi\alpha}D_G^{\rho\tau}(k;q)\epsilon_3^\lambda
\times[V_{SSG}^{acx,(23)}]_{\alpha\beta\lambda}(r,p_3-r,-p_3)D^{\beta\nu}(r-p_3)\mathcal{H}^{cb}_{\nu\tau}[2,m](r-p_3,k), \tag*{(81)} I I , a ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) r 2 + i 0 − i G ξ μ ( r ; q ) g ξ α D G ρ τ ( k ; q ) ϵ 3 λ × [ V S S G a c x , ( 23 ) ] α β λ ( r , p 3 − r , − p 3 ) D β ν ( r − p 3 ) H ν τ c b [ 2 , m ] ( r − p 3 , k ) , (81)
I I I , a ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μ ν ( r ; q ) − i k 2 + i 0 G ξ ρ ( k ; q ) g ξ α ϵ 3 λ × [ V S S G b c x , ( 23 ) ] α β λ ( k , p 3 − k , − p 3 ) D β τ ( k − p 3 ) H ν τ a c [ 2 , m ] ( r , k − p 3 ) . (82) \mathcal{I}_{II,a}(k)=\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D_G^{\mu\nu}(r;q)\frac{-i}{k^2+i0}G^\rho_\xi(k;q)g^{\xi\alpha}\epsilon_3^\lambda
\times[V_{SSG}^{bcx,(23)}]_{\alpha\beta\lambda}(k,p_3-k,-p_3)D^{\beta\tau}(k-p_3)\mathcal{H}^{ac}_{\nu\tau}[2,m](r,k-p_3). \tag*{(82)} I I I , a ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μν ( r ; q ) k 2 + i 0 − i G ξ ρ ( k ; q ) g ξ α ϵ 3 λ × [ V S S G b c x , ( 23 ) ] α β λ ( k , p 3 − k , − p 3 ) D β τ ( k − p 3 ) H ν τ a c [ 2 , m ] ( r , k − p 3 ) . (82)
In graphs I I , a , I I I , a \mathcal{I}_{I,a},\mathcal{I}_{II,a} I I , a , I I I , a , we now decompose the gluon labeled r r r , k k k respectively in their opposite , quark-current end. For a gluon of momentum u u u , the mixed term in the two sided GK is
− i u 2 + i 0 K ξ μ ( − u ; q ) g ξ ζ G ζ α ( u ; q ) = − i u 2 + i 0 − u μ ( 2 q − u ) ζ G ζ α ( u ; q ) u 2 − 2 q ⋅ u + i 0 . (83) \begin{aligned}
\frac{-i}{u^2+i0}K^\mu_\xi(-u;q)g^{\xi\zeta}G^\alpha_\zeta(u;q) \\
&=\frac{-i}{u^2+i0}\frac{-u^\mu(2q-u)^\zeta G^\alpha_\zeta(u;q)}{u^2-2q\cdot u+i0}. \tag*{(83)}
\end{aligned} u 2 + i 0 − i K ξ μ ( − u ; q ) g ξ ζ G ζ α ( u ; q ) = u 2 + i 0 − i u 2 − 2 q ⋅ u + i 0 − u μ ( 2 q − u ) ζ G ζ α ( u ; q ) . (83)
In particular, the momentum at the current end is exactly u μ u^\mu u μ . The resulting identities, shown in Figure 17 , are
Figure 17. Second-end G G G –K K K decomposition of ( I ) a (\mathrm{I})_a ( I ) a and ( I I ) a (\mathrm{II})_a ( II ) a . The G G G at the SSG vertex stays fixed; J 1 J_1 J 1 and J 2 J_2 J 2 have the K K K insertion at the quark-current end.
I I , a ( k ) = I I , G G ( k ) + J 1 ( k ) , \mathcal{I}_{\mathrm{I},a}(k)=\mathcal{I}_{\mathrm{I},GG}(k)+J_1(k), I I , a ( k ) = I I , GG ( k ) + J 1 ( k ) ,
I I I , a ( k ) = I I I , G G ( k ) + J 2 ( k ) . (84) \mathcal{I}_{\mathrm{II},a}(k)=\mathcal{I}_{\mathrm{II},GG}(k)+J_2(k). \tag*{(84)} I II , a ( k ) = I II , GG ( k ) + J 2 ( k ) . (84)
The terms labeled I , I I G G \mathrm{I},\mathrm{II}_{GG} I , II GG are collinear finite in all collinear limits following the arguments made in § 3 . The two remaining longitudinal integrands are explicitly
J 1 ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G ρ τ ( k ; q ) × − i r 2 + i 0 K ξ μ ( − r ; q ) g ξ ζ G ζ α ( r ; q ) × ϵ 3 a b c [ V S S G a c x , ( 23 ) ] α β λ ( r , − s , − p 3 ) D β ν ( s ) H ν τ c b [ 2 , m ] ( s , k ) , (85) \begin{aligned}
J_1(k)={}&\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D_G^{\rho\tau}(k;q)\\
&\times\frac{-i}{r^2+i0}K^\mu_\xi(-r;q)g^{\xi\zeta}G^\alpha_\zeta(r;q)\\
&\times\epsilon_3^{abc}\bigl[V_{\mathrm{SSG}}^{acx,(23)}\bigr]_{\alpha\beta\lambda}(r,-s,-p_3)D^{\beta\nu}(s)H^{cb}_{\nu\tau}[2,m](s,k), \tag*{(85)}
\end{aligned} J 1 ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G ρ τ ( k ; q ) × r 2 + i 0 − i K ξ μ ( − r ; q ) g ξ ζ G ζ α ( r ; q ) × ϵ 3 ab c [ V SSG a c x , ( 23 ) ] α β λ ( r , − s , − p 3 ) D β ν ( s ) H ν τ c b [ 2 , m ] ( s , k ) , (85)
J 2 ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μ ν ( r ; q ) × − i k 2 + i 0 K ξ ρ ( − k ; q ) g ξ ζ G ζ α ( k ; q ) × ϵ 3 a b c [ V S S G b c x , ( 23 ) ] α β λ ( k , − t , − p 3 ) D β τ ( t ) H ν τ a c [ 2 , m ] ( r , t ) . (86) \begin{aligned}
J_2(k)={}&\bar{v}(p_2)\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu u(p_1)D_G^{\mu\nu}(r;q)\\
&\times\frac{-i}{k^2+i0}K^\rho_\xi(-k;q)g^{\xi\zeta}G^\alpha_\zeta(k;q)\\
&\times\epsilon_3^{abc}\bigl[V_{\mathrm{SSG}}^{bcx,(23)}\bigr]_{\alpha\beta\lambda}(k,-t,-p_3)D^{\beta\tau}(t)H^{ac}_{\nu\tau}[2,m](r,t). \tag*{(86)}
\end{aligned} J 2 ( k ) = v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) Γ μ a u ( p 1 ) D G μν ( r ; q ) × k 2 + i 0 − i K ξ ρ ( − k ; q ) g ξ ζ G ζ α ( k ; q ) × ϵ 3 ab c [ V SSG b c x , ( 23 ) ] α β λ ( k , − t , − p 3 ) D β τ ( t ) H ν τ a c [ 2 , m ] ( r , t ) . (86)
The explicit expressions for I , I G G \mathcal{I},\mathcal{I}_{GG} I , I GG are obtained by replacing K ξ μ ( − r ; q ) K^\mu_\xi(-r;q) K ξ μ ( − r ; q ) , K ξ ρ ( − k ; q ) K^\rho_\xi(-k;q) K ξ ρ ( − k ; q ) by G ξ μ ( − r ; q ) G^\mu_\xi(-r;q) G ξ μ ( − r ; q ) , G ξ ρ ( − k ; q ) G^\rho_\xi(-k;q) G ξ ρ ( − k ; q ) in Eqs. (85) ,(86) respectively. Having made this decomposition, we will find that the graphs J 1 , J 2 J_1,J_2 J 1 , J 2 will cancel a third term coming from graph ( I V ) a (\mathrm{IV})_a ( IV ) a which we now turn to.
For these graphs we use precisely the same routing r = P − k r=P-k r = P − k , t = k − p 3 t=k-p_3 t = k − p 3 . The relevant vertices in ( I V ) a (\mathrm{IV})_a ( IV ) a are
[ V a b c , ( 31 ) ] μ ν α ( P , − k , − r ) = g s f a b c g μ α ( 2 P − k ) ν , [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) = − g s f b d x g γ λ ( 2 p 3 − k ) β . (87) \begin{aligned}
\bigl[V^{abc,(31)}\bigr]_{\mu\nu\alpha}(P,-k,-r)&=g_s f^{abc}g_{\mu\alpha}(2P-k)_\nu,\\
\bigl[V^{bdx,(23)}\bigr]_{\beta\gamma\lambda}(k,-t,-p_3)&=-g_s f^{bdx}g_{\gamma\lambda}(2p_3-k)_\beta. \tag*{(87)}
\end{aligned} [ V ab c , ( 31 ) ] μν α ( P , − k , − r ) [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) = g s f ab c g μα ( 2 P − k ) ν , = − g s f b d x g γ λ ( 2 p 3 − k ) β . (87)
The complete integrands are
I I V , a ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V a b c , ( 31 ) ] μ ν α ( P , − k , − r ) × D ν β ( k ) D α α ′ ( r ) ϵ 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) × D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , (88) \mathcal{I}_{\mathrm{IV},a}(k)=\bar{v}(p_{2})\Gamma^{a}_{\rho}u(p_{1})D^{\rho\mu}(P)[V^{abc,(31)}]_{\mu\nu\alpha}(P,-k,-r)
\times D^{\nu\beta}(k)D^{\alpha\alpha'}(r)\epsilon^{\lambda}_{3}[V^{bdx,(23)}]_{\beta\gamma\lambda}(k,-t,-p_{3})
\times D^{\gamma\gamma'}(t)H^{cd}_{\alpha'\gamma'}[2,m](r,t), \tag*{(88)} I IV , a ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V ab c , ( 31 ) ] μν α ( P , − k , − r ) × D ν β ( k ) D α α ′ ( r ) ϵ 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) × D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , (88)
I V ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) D β β ′ ( k ) D γ γ ′ ( t ) × ϵ 3 λ [ V b d x , ( 23 ) ] β ′ γ λ ( k , − t , − p 3 ) H μ β ′ γ ′ a b d [ 3 , m ] ( P , − k , t ) . (89) \mathcal{I}_{\mathrm{V}}(k)=\bar{v}(p_{2})\Gamma^{a}_{\rho}u(p_{1})D^{\rho\mu}(P)D^{\beta\beta'}(k)D^{\gamma\gamma'}(t)
\times\epsilon^{\lambda}_{3}[V^{bdx,(23)}]_{\beta'\gamma\lambda}(k,-t,-p_{3})H^{abd}_{\mu\beta'\gamma'}[3,m](P,-k,t). \tag*{(89)} I V ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) D β β ′ ( k ) D γ γ ′ ( t ) × ϵ 3 λ [ V b d x , ( 23 ) ] β ′ γ λ ( k , − t , − p 3 ) H μ β ′ γ ′ ab d [ 3 , m ] ( P , − k , t ) . (89)
These topologies and their projector assignments are shown in Figure 18 .
Figure 18. The exact three-sector decomposition of ( I V ) a (\mathrm{IV})_a ( IV ) a and ( V ) (\mathrm{V}) ( V ) . The K − G K-G K − G terms produce J 3 J_3 J 3 , J 4 J_4 J 4 and J 5 J_5 J 5 . The two other sectors in each row are locally infrared finite.
The k k k line is now decomposed first at its scalar-scalar-gluon end and then, only for its G G G component, at the other end in both graphs:
D ν β ( k ) = − i k 2 + i 0 [ K ξ ν ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + G ξ ν ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + g ν ζ K ζ β ( k ; q ) ] D^{\nu\beta}(k)=\frac{-i}{k^{2}+i0}\left[K^{\nu}_{\xi}(-k;q)g^{\xi\zeta}G_{\zeta}^{\ \beta}(k;q)+G^{\nu}_{\xi}(-k;q)g^{\xi\zeta}G_{\zeta}^{\ \beta}(k;q)+g^{\nu\zeta}K_{\zeta}^{\ \beta}(k;q)\right] D ν β ( k ) = k 2 + i 0 − i [ K ξ ν ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + G ξ ν ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + g ν ζ K ζ β ( k ; q ) ]
D β β ′ ( k ) = − i k 2 + i 0 [ K ξ β ′ ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + G ξ β ′ ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + g β ′ ζ K ζ β ( k ; q ) ] . (90) D^{\beta\beta'}(k)=\frac{-i}{k^{2}+i0}\left[K^{\beta'}_{\xi}(-k;q)g^{\xi\zeta}G_{\zeta}^{\ \beta}(k;q)+G_{\xi}^{\ \beta'}(-k;q)g^{\xi\zeta}G_{\zeta}^{\ \beta}(k;q)+g^{\beta'\zeta}K_{\zeta}^{\ \beta}(k;q)\right]. \tag*{(90)} D β β ′ ( k ) = k 2 + i 0 − i [ K ξ β ′ ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + G ξ β ′ ( − k ; q ) g ξ ζ G ζ β ( k ; q ) + g β ′ ζ K ζ β ( k ; q ) ] . (90)
Let us first write down explicit expressions for the locally finite GG, fin terms in Figure 18 .
I I V , a , G G ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V a b c , ( 31 ) ] μ ν α ( P , − k , − r ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , × G ξ ν ( − k ; q ) g ξ ζ G ζ β ( k ; q ) D α α ′ ( r ) ϵ 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) (91) \mathcal{I}_{\mathrm{IV},a,\mathrm{GG}}(k)=\bar{v}(p_{2})\Gamma^{a}_{\rho}u(p_{1})D^{\rho\mu}(P)[V^{abc,(31)}]_{\mu\nu\alpha}(P,-k,-r)D^{\gamma\gamma'}(t)H^{cd}_{\alpha'\gamma'}[2,m](r,t),
\times G^{\nu}_{\xi}(-k;q)g^{\xi\zeta}G_{\zeta}^{\ \beta}(k;q)D^{\alpha\alpha'}(r)\epsilon^{\lambda}_{3}[V^{bdx,(23)}]_{\beta\gamma\lambda}(k,-t,-p_{3}) \tag*{(91)} I IV , a , GG ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V ab c , ( 31 ) ] μν α ( P , − k , − r ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , × G ξ ν ( − k ; q ) g ξ ζ G ζ β ( k ; q ) D α α ′ ( r ) ϵ 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) (91)
I I V , a , f i n ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V a b c , ( 31 ) ] μ ν α ( P , − k , − r ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , × g ν ζ K ζ β ( k ; q ) D α α ′ ( r ) ϵ 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) . (92) \mathcal{I}_{\mathrm{IV},a,\mathrm{fin}}(k)=\bar{v}(p_{2})\Gamma^{a}_{\rho}u(p_{1})D^{\rho\mu}(P)[V^{abc,(31)}]_{\mu\nu\alpha}(P,-k,-r)D^{\gamma\gamma'}(t)H^{cd}_{\alpha'\gamma'}[2,m](r,t),
\times g^{\nu\zeta}K_{\zeta}^{\ \beta}(k;q)D^{\alpha\alpha'}(r)\epsilon^{\lambda}_{3}[V^{bdx,(23)}]_{\beta\gamma\lambda}(k,-t,-p_{3}). \tag*{(92)} I IV , a , fin ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) [ V ab c , ( 31 ) ] μν α ( P , − k , − r ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , × g ν ζ K ζ β ( k ; q ) D α α ′ ( r ) ϵ 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) . (92)
The first GG piece is IR finite following the standard G gluon collinear suppression encountered in § 3 . The second expression is IR finite because the contraction $k^{\beta}[V^{bdx}]_{\beta\gamma\lambda}(k,-tp 3 ) ∝ t 2 p_3) \propto t^2 p 3 ) ∝ t 2 which cancels the t 2 t^2 t 2 denominator. As a result, in the t ∥ p 3 t \parallel p_3 t ∥ p 3 limit, the previously logarithmic divergence is now power counting convergent.
Similarly for V V V , we have
I V , G G ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) D γ γ ′ ( t ) H μ β ′ γ ′ a b d [ 3 , m ] ( P , − k , t ) × k 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) G ξ β ( − k ; q ) g ξ ζ G ζ β ′ ( k ; q ) , (93) \begin{aligned}
I_{V,GG}(k)={}&\bar{v}(p_2)\Gamma^a_\rho u(p_1)D^{\rho\mu}(P)D^{\gamma\gamma'}(t)\mathcal{H}^{abd}_{\mu\beta'\gamma'}[3,m](P,-k,t)\\
&\times k_3^\lambda[V^{bdx,(23)}]_{\beta\gamma\lambda}(k,-t,-p_3)G^{\xi\beta}(-k;q)g_{\xi\zeta}G^{\zeta\beta'}(k;q), \tag*{(93)}
\end{aligned} I V , GG ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) D γ γ ′ ( t ) H μ β ′ γ ′ ab d [ 3 , m ] ( P , − k , t ) × k 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) G ξ β ( − k ; q ) g ξ ζ G ζ β ′ ( k ; q ) , (93)
I V , f i n ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) D γ γ ′ ( t ) H μ β ′ γ ′ a b d [ 3 , m ] ( P , − k , t ) × k 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) g β ′ ξ K ξ β ( k ; q ) , (94) \begin{aligned}
I_{V,\mathrm{fin}}(k)={}&\bar{v}(p_2)\Gamma^a_\rho u(p_1)D^{\rho\mu}(P)D^{\gamma\gamma'}(t)\mathcal{H}^{abd}_{\mu\beta'\gamma'}[3,m](P,-k,t)\\
&\times k_3^\lambda[V^{bdx,(23)}]_{\beta\gamma\lambda}(k,-t,-p_3)g^{\beta'\xi}K_\xi{}^\beta(k;q), \tag*{(94)}
\end{aligned} I V , fin ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) D γ γ ′ ( t ) H μ β ′ γ ′ ab d [ 3 , m ] ( P , − k , t ) × k 3 λ [ V b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) g β ′ ξ K ξ β ( k ; q ) , (94)
In much the same way as I V a I_{V_a} I V a , these pieces are also locally finite because in I V , G G ( k ) I_{V,GG}(k) I V , GG ( k ) the G tensors are power suppressed near the logarithmically divergent pinch k ∥ p 3 k \parallel p_3 k ∥ p 3 and I V , f i n ( k ) I_{V,\mathrm{fin}}(k) I V , fin ( k ) is finite because k β [ V b d x ] β γ λ ( k , − t − p 3 ) ∝ t 2 k^\beta[V^{bdx}]_{\beta\gamma\lambda}(k,-t-p_3)\propto t^2 k β [ V b d x ] β γ λ ( k , − t − p 3 ) ∝ t 2 as before.
Turning to what remains, in the J 3 + J 4 J_3+J_4 J 3 + J 4 terms in Figure 18 , the K G KG K G term admits a vertex factor
k ν [ V a b c , ( 31 ) ] μ ν α ( P , − k , − r ) = g s f a b c ( P 2 − r 2 ) g μ α . (95) k^\nu[V^{abc,(31)}]_{\mu\nu\alpha}(P,-k,-r)=g_sf^{abc}(P^2-r^2)g_{\mu\alpha}. \tag*{(95)} k ν [ V ab c , ( 31 ) ] μν α ( P , − k , − r ) = g s f ab c ( P 2 − r 2 ) g μα . (95)
Since
( P 2 − r 2 ) D ρ μ ( P ) g μ α D α α ′ ( r ) = − i D ρ α ′ ( r ) + i D ρ α ′ ( P ) , (96) (P^2-r^2)D^{\rho\mu}(P)g_{\mu\alpha}D^{\alpha\alpha'}(r)=-iD^{\rho\alpha'}(r)+iD^{\rho\alpha'}(P), \tag*{(96)} ( P 2 − r 2 ) D ρ μ ( P ) g μα D α α ′ ( r ) = − i D ρ α ′ ( r ) + i D ρ α ′ ( P ) , (96)
we obtain J 3 J_3 J 3 and J 4 J_4 J 4 separately:
J 3 ( k ) = − i g s f a b c v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ α ′ ( r ) × − i k 2 + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) k 2 − 2 q ⋅ k + i 0 × k 3 λ [ V S S G b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , (97) \begin{aligned}
J_3(k)={}&-ig_sf^{abc}\bar{v}(p_2)\Gamma^a_\rho u(p_1)D^{\rho\alpha'}(r)\\
&\times\frac{-i}{k^2+i0}\frac{-(2q-k)_\zeta G^{\zeta\beta}(k;q)}{k^2-2q\cdot k+i0}\\
&\times k_3^\lambda[V^{bdx,(23)}_{SSG}]_{\beta\gamma\lambda}(k,-t,-p_3)D^{\gamma\gamma'}(t)\mathcal{H}^{cd}_{\alpha'\gamma'}[2,m](r,t), \tag*{(97)}
\end{aligned} J 3 ( k ) = − i g s f ab c v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ α ′ ( r ) × k 2 + i 0 − i k 2 − 2 q ⋅ k + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) × k 3 λ [ V S S G b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) , (97)
J 4 ( k ) = + i g s f a b c v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ α ′ ( P ) × − i k 2 + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) k 2 − 2 q ⋅ k + i 0 × k 3 λ [ V S S G b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) . (98) \begin{aligned}
J_4(k)={}&+ig_sf^{abc}\bar{v}(p_2)\Gamma^a_\rho u(p_1)D^{\rho\alpha'}(P)\\
&\times\frac{-i}{k^2+i0}\frac{-(2q-k)_\zeta G^{\zeta\beta}(k;q)}{k^2-2q\cdot k+i0}\\
&\times k_3^\lambda[V^{bdx,(23)}_{SSG}]_{\beta\gamma\lambda}(k,-t,-p_3)D^{\gamma\gamma'}(t)\mathcal{H}^{cd}_{\alpha'\gamma'}[2,m](r,t). \tag*{(98)}
\end{aligned} J 4 ( k ) = + i g s f ab c v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ α ′ ( P ) × k 2 + i 0 − i k 2 − 2 q ⋅ k + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) × k 3 λ [ V S S G b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) H α ′ γ ′ c d [ 2 , m ] ( r , t ) . (98)
Let us now argue that J 1 + J 2 + J 3 = 0 J_1+J_2+J_3=0 J 1 + J 2 + J 3 = 0 , as an exact ward identity. First, we utilize the abelian Ward identity to replace in Eq. (97) , D ρ α ′ ( r ) D^{\rho\alpha'}(r) D ρ α ′ ( r ) by D G ρ α ′ ( r ) D_G^{\rho\alpha'}(r) D G ρ α ′ ( r ) . In J 1 J_1 J 1 we shift k ↦ P − k k\mapsto P-k k ↦ P − k , relabel the dummy color indices, and use Bose symmetry of the integrated two-gluon hard loop. Both quark-line attachments then carry the same longitudinal momentum k k k , and
k ν v ˉ ( p 2 ) [ Γ μ a S 0 ( p 1 − k ) Γ ρ b + Γ ρ b S 0 ( k − p 2 ) Γ μ a ] u ( p 1 ) = g s v ˉ ( p 2 ) [ T b , Γ μ a ] u ( p 1 ) = i g s f b a c v ˉ ( p 2 ) Γ μ c u ( p 1 ) . (99) \begin{aligned}
k^\nu\bar{v}(p_2)\left[\Gamma^a_\mu S_0(p_1-k)\Gamma^b_\rho+\Gamma^b_\rho S_0(k-p_2)\Gamma^a_\mu\right]u(p_1)\\
=g_s\bar{v}(p_2)[T^b,\Gamma^a_\mu]u(p_1)=ig_sf^{bac}\bar{v}(p_2)\Gamma^c_\mu u(p_1). \tag*{(99)}
\end{aligned} k ν v ˉ ( p 2 ) [ Γ μ a S 0 ( p 1 − k ) Γ ρ b + Γ ρ b S 0 ( k − p 2 ) Γ μ a ] u ( p 1 ) = g s v ˉ ( p 2 ) [ T b , Γ μ a ] u ( p 1 ) = i g s f ba c v ˉ ( p 2 ) Γ μ c u ( p 1 ) . (99)
Utilizing Eq. (99) in Eqs. (85) , (86) and combing with Eq. (97) , we find
J 1 ( P − k ) + J 2 ( k ) + J 3 ( k ) = 0. (100) J_1(P-k)+J_2(k)+J_3(k)=0. \tag*{(100)} J 1 ( P − k ) + J 2 ( k ) + J 3 ( k ) = 0. (100)
This ward identity is shown in Figure 19 .
Figure 19. The Ward identity at the massless fermion current.
The remaining K G KG K G contribution to (V) is
J 5 ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) × − i k 2 + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) k 2 − 2 q ⋅ k + i 0 × ϵ 3 λ [ V S S G b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) × k β ′ H μ β ′ γ ′ [ 3 , m ] a b d ( P , − k , t ) . (101) \begin{aligned}
J_5(k)={}&\bar{v}(p_2)\Gamma^a_{\rho}u(p_1)D^{\rho\mu}(P)\\
&\times\frac{-i}{k^2+i0}\frac{-(2q-k)_{\zeta}G^{\zeta\beta}(k;q)}{k^2-2q\cdot k+i0}\\
&\times\epsilon_3^{\lambda}\left[V_{\mathrm{SSG}}^{bdx,(23)}\right]_{\beta\gamma\lambda}(k,-t,-p_3)D^{\gamma\gamma'}(t)\\
&\times k^{\beta'}\mathcal{H}^{abd}_{\mu\beta'\gamma'[3,m]}(P,-k,t). \tag*{(101)}
\end{aligned} J 5 ( k ) = v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) × k 2 + i 0 − i k 2 − 2 q ⋅ k + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) × ϵ 3 λ [ V SSG b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) × k β ′ H μ β ′ γ ′ [ 3 , m ] ab d ( P , − k , t ) . (101)
The three-gluon hard-loop Ward identity gives
k β ′ H μ β ′ γ ′ [ 3 , m ] a b d ( P , − k , t ) = − i g s f a b c H μ γ ′ [ 2 , m ] c d ( r , t ) + i g s f b d c H μ γ ′ [ 2 , m ] a c ( P , − p 3 ) . (102) k^{\beta'}\mathcal{H}^{abd}_{\mu\beta'\gamma'[3,m]}(P,-k,t)
=-ig_sf^{abc}\mathcal{H}^{cd}_{\mu\gamma'[2,m]}(r,t)+ig_sf^{bdc}\mathcal{H}^{ac}_{\mu\gamma'[2,m]}(P,-p_3). \tag*{(102)} k β ′ H μ β ′ γ ′ [ 3 , m ] ab d ( P , − k , t ) = − i g s f ab c H μ γ ′ [ 2 , m ] c d ( r , t ) + i g s f b d c H μ γ ′ [ 2 , m ] a c ( P , − p 3 ) . (102)
Its first term equals − J 4 ( k ) -J_4(k) − J 4 ( k ) , with exactly the same SSG projector factor, whereas the second has a hard loop independent of k k k . Thus the nonfactorizing pieces cancel:
J 4 ( k ) + J 5 ( k ) = i g s f b d c v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) × − i k 2 + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) k 2 − 2 q ⋅ k + i 0 × ϵ 3 λ [ V S S G b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) H μ γ ′ [ 2 , m ] a c ( P , − p 3 ) . (103) \begin{aligned}
J_4(k)+J_5(k)={}&ig_sf^{bdc}\bar{v}(p_2)\Gamma^a_{\rho}u(p_1)D^{\rho\mu}(P)\\
&\times\frac{-i}{k^2+i0}\frac{-(2q-k)_{\zeta}G^{\zeta\beta}(k;q)}{k^2-2q\cdot k+i0}\\
&\times\epsilon_3^{\lambda}\left[V_{\mathrm{SSG}}^{bdx,(23)}\right]_{\beta\gamma\lambda}(k,-t,-p_3)D^{\gamma\gamma'}(t)\mathcal{H}^{ac}_{\mu\gamma'[2,m]}(P,-p_3). \tag*{(103)}
\end{aligned} J 4 ( k ) + J 5 ( k ) = i g s f b d c v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) × k 2 + i 0 − i k 2 − 2 q ⋅ k + i 0 − ( 2 q − k ) ζ G ζ β ( k ; q ) × ϵ 3 λ [ V SSG b d x , ( 23 ) ] β γ λ ( k , − t , − p 3 ) D γ γ ′ ( t ) H μ γ ′ [ 2 , m ] a c ( P , − p 3 ) . (103)
Using f b d c f b d x = C A δ c x f^{bdc}f^{bdx}=C_A\delta^{cx} f b d c f b d x = C A δ c x , the explicitly factorized result is
J 4 ( k ) + J 5 ( k ) = − i g s 2 C A ( 2 q − k ) ζ G ζ β ( k ; q ) ( 2 p 3 − k ) β ( k 2 + i 0 ) ( t 2 + i 0 ) ( k 2 − 2 q ⋅ k + i 0 ) (104) J_4(k)+J_5(k)=-ig_s^2C_A\frac{(2q-k)_{\zeta}G^{\zeta\beta}(k;q)(2p_3-k)_{\beta}}{(k^2+i0)(t^2+i0)(k^2-2q\cdot k+i0)} \tag*{(104)} J 4 ( k ) + J 5 ( k ) = − i g s 2 C A ( k 2 + i 0 ) ( t 2 + i 0 ) ( k 2 − 2 q ⋅ k + i 0 ) ( 2 q − k ) ζ G ζ β ( k ; q ) ( 2 p 3 − k ) β (104)
× v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) ϵ 3 γ ′ H μ γ ′ [ 2 , m ] a x ( P , − p 3 ) . (105) \times\bar{v}(p_2)\Gamma^a_{\rho}u(p_1)D^{\rho\mu}(P)\epsilon_3^{\gamma'}\mathcal{H}^{ax}_{\mu\gamma'[2,m]}(P,-p_3). \tag*{(105)} × v ˉ ( p 2 ) Γ ρ a u ( p 1 ) D ρ μ ( P ) ϵ 3 γ ′ H μ γ ′ [ 2 , m ] a x ( P , − p 3 ) . (105)
The hard tensor on the right has no loop-momentum dependence. This gives a genuine factorized correction to the external p 3 p_3 p 3 line.
The G G GG GG contributions of (I) and (II) are collinear finite, but retain their respective genuine soft singularities at r → 0 r\to0 r → 0 and k → 0 k\to0 k → 0 . These are removed by the local soft counterterms which we discuss in conclusion to this section. All longitudinal contributions from (I)–(V) have meanwhile either canceled or reduced to the factorized expressions in Eqs. (80) and (105) .
Let us finally turn to the soft counterterms to the graphs I G G , I I G G I_{GG},II_{GG} I GG , I I GG . To construct these, we simply substitute the relevant soft limit in the integrand.
In the limit r → 0 r \to0 r → 0 , the massless quark propagator and the SSG vertex in graph (I) satisfy
S 0 ( p 1 − r ) Γ μ a u ( p 1 ) = − g s T a − 2 p 1 μ ( p 1 − r ) 2 + i 0 u ( p 1 ) + O ( r 0 ) , S_0(p_1-r)\Gamma^a_\mu u(p_1)=-g_sT^a\frac{-2p_{1\mu}}{(p_1-r)^2+i0}u(p_1)+O(r^0), S 0 ( p 1 − r ) Γ μ a u ( p 1 ) = − g s T a ( p 1 − r ) 2 + i 0 − 2 p 1 μ u ( p 1 ) + O ( r 0 ) ,
ϵ 3 λ [ V S S G a c x , ( 23 ) ] α β λ ( r , − s , − p 3 ) D β ν ( s ) = − i g s f a c x − 2 p 3 α ( p 3 − r ) 2 + i 0 ϵ 3 ν + O ( r 0 ) . (106) \epsilon_3^\lambda[V^{acx,(23)}_{SSG}]_{\alpha\beta\lambda}(r,-s,-p_3)D^{\beta\nu}(s)=-ig_sf^{acx}\frac{-2p_{3\alpha}}{(p_3-r)^2+i0}\epsilon_3^\nu+O(r^0). \tag*{(106)} ϵ 3 λ [ V S S G a c x , ( 23 ) ] α β λ ( r , − s , − p 3 ) D β ν ( s ) = − i g s f a c x ( p 3 − r ) 2 + i 0 − 2 p 3 α ϵ 3 ν + O ( r 0 ) . (106)
Similarly, in the limit k → 0 k \to0 k → 0 , graph (II) satisfies
v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) = g s − p 2 ρ ( p 2 − k ) 2 + i 0 v ˉ ( p 2 ) T b + O ( k 0 ) , \bar v(p_2)\Gamma^b_\rho S_0(k-p_2)=g_s\frac{-p_{2\rho}}{(p_2-k)^2+i0}\bar v(p_2)T^b+O(k^0), v ˉ ( p 2 ) Γ ρ b S 0 ( k − p 2 ) = g s ( p 2 − k ) 2 + i 0 − p 2 ρ v ˉ ( p 2 ) T b + O ( k 0 ) ,
ϵ 3 λ [ V S S G b c x , ( 23 ) ] α β λ ( k , − t , − p 3 ) D β τ ( t ) = − i g s f b c x − 2 p 3 α ( p 3 − k ) 2 + i 0 ϵ 3 τ + O ( k 0 ) . (107) \epsilon_3^\lambda[V^{bcx,(23)}_{SSG}]_{\alpha\beta\lambda}(k,-t,-p_3)D^{\beta\tau}(t)=-ig_sf^{bcx}\frac{-2p_{3\alpha}}{(p_3-k)^2+i0}\epsilon_3^\tau+O(k^0). \tag*{(107)} ϵ 3 λ [ V S S G b c x , ( 23 ) ] α β λ ( k , − t , − p 3 ) D β τ ( t ) = − i g s f b c x ( p 3 − k ) 2 + i 0 − 2 p 3 α ϵ 3 τ + O ( k 0 ) . (107)
In either limit, the massive fermion loop reduces to H [ 2 , m ] ( P , − p 3 ) \mathcal{H}[2,m](P,-p_3) H [ 2 , m ] ( P , − p 3 ) , and the remaining hard gluon propagator can be replaced by its unprojected counterpart using the Abelian Ward identity.
Writing H μ ν a b [ 2 , m ] = δ a b H μ ν [ 2 , m ] H^{ab}_{\mu\nu}[2,m]=\delta^{ab}H_{\mu\nu}[2,m] H μν ab [ 2 , m ] = δ ab H μν [ 2 , m ] , the corresponding one-loop amplitude is
M ( 1 ) = v ˉ ( p 2 ) Γ ρ x u ( p 1 ) D ρ ν ( P ) ϵ 3 λ H ν λ [ 2 , m ] ( P , − p 3 ) . (108) \mathcal{M}^{(1)}=\bar v(p_2)\Gamma^x_\rho u(p_1)D^{\rho\nu}(P)\epsilon_3^\lambda H_{\nu\lambda}[2,m](P,-p_3). \tag*{(108)} M ( 1 ) = v ˉ ( p 2 ) Γ ρ x u ( p 1 ) D ρ ν ( P ) ϵ 3 λ H ν λ [ 2 , m ] ( P , − p 3 ) . (108)
Using the color identities
f a b x T b T a = − i C A 2 T x , f b a x T b T a = i C A 2 T x , (109) f^{abx}T^bT^a=-\frac{iC_A}{2}T^x,\qquad f^{bax}T^bT^a=\frac{iC_A}{2}T^x, \tag*{(109)} f ab x T b T a = − 2 i C A T x , f ba x T b T a = 2 i C A T x , (109)
we obtain the soft counterterms
C I s o f t ( k ) = g s 2 C A 2 − i r 2 + i 0 × ( 2 p 1 − r ) μ G ξ μ ( − r ; q ) g ξ ζ G ζ α ( r ; q ) ( 2 p 3 − r ) α ( ( p 1 − r ) 2 + i 0 ) ( ( p 3 − r ) 2 + i 0 ) M ( 1 ) , (110) \mathcal{C}^{\mathrm{soft}}_{\mathrm{I}}(k)=\frac{g_s^2C_A}{2}\frac{-i}{r^2+i0}
\times\frac{(2p_1-r)_\mu G^{\xi\mu}(-r;q)g_{\xi\zeta}G^\zeta{}_\alpha(r;q)(2p_3-r)^\alpha}{((p_1-r)^2+i0)((p_3-r)^2+i0)}\mathcal{M}^{(1)}, \tag*{(110)} C I soft ( k ) = 2 g s 2 C A r 2 + i 0 − i × (( p 1 − r ) 2 + i 0 ) (( p 3 − r ) 2 + i 0 ) ( 2 p 1 − r ) μ G ξ μ ( − r ; q ) g ξ ζ G ζ α ( r ; q ) ( 2 p 3 − r ) α M ( 1 ) , (110)
C I I s o f t ( k ) = g s 2 C A 2 − i k 2 + i 0 × ( 2 p 2 − k ) ρ G ξ ρ ( − k ; q ) g ξ ζ G ζ α ( k ; q ) ( 2 p 3 − k ) α ( ( p 2 − k ) 2 + i 0 ) ( ( p 3 − k ) 2 + i 0 ) M ( 1 ) . (111) \mathcal{C}^{\mathrm{soft}}_{\mathrm{II}}(k)=\frac{g_s^2C_A}{2}\frac{-i}{k^2+i0}
\times\frac{(2p_2-k)_\rho G^{\xi\rho}(-k;q)g_{\xi\zeta}G^\zeta{}_\alpha(k;q)(2p_3-k)^\alpha}{((p_2-k)^2+i0)((p_3-k)^2+i0)}\mathcal{M}^{(1)}. \tag*{(111)} C II soft ( k ) = 2 g s 2 C A k 2 + i 0 − i × (( p 2 − k ) 2 + i 0 ) (( p 3 − k ) 2 + i 0 ) ( 2 p 2 − k ) ρ G ξ ρ ( − k ; q ) g ξ ζ G ζ α ( k ; q ) ( 2 p 3 − k ) α M ( 1 ) . (111)
We define the subtracted integrands by
I I , G G s u b ( k ) = I I , G G ( k ) − C I s o f t ( k ) , \mathcal{I}^{\mathrm{sub}}_{\mathrm{I,GG}}(k)=\mathcal{I}_{\mathrm{I,GG}}(k)-\mathcal{C}^{\mathrm{soft}}_{\mathrm{I}}(k), I I , GG sub ( k ) = I I , GG ( k ) − C I soft ( k ) ,
I I I , G G s u b ( k ) = I I I , G G ( k ) − C I I s o f t ( k ) . (112) \mathcal{I}^{\mathrm{sub}}_{\mathrm{II,GG}}(k)=\mathcal{I}_{\mathrm{II,GG}}(k)-\mathcal{C}^{\mathrm{soft}}_{\mathrm{II}}(k). \tag*{(112)} I II , GG sub ( k ) = I II , GG ( k ) − C II soft ( k ) . (112)
By construction, these counterterms reproduce the leading soft behavior at r = 0 r=0 r = 0 and k = 0 k=0 k = 0 , respectively. They remove the remaining soft logarithms without altering the previously established collinear finiteness of the G G GG GG contributions. In this section we choose q = P = p 1 + p 2 q=P=p_1+p_2 q = P = p 1 + p 2 . Since P P P is timelike, the auxiliary denominator k 2 + 2 P ⋅ k k^2+2P\cdot k k 2 + 2 P ⋅ k is nonzero on every nonzero collinear momentum k = x p i k=xp_i k = x p i . For the momentum routings in Eqs. (80) –(105) , the only additional pinch allowed by the Landau equations occurs when the two denominators k 2 ± 2 P ⋅ k + i 0 k^2 \pm2P \cdot k + i0 k 2 ± 2 P ⋅ k + i 0 on the same gluon line trap the contour at k = 0 k = 0 k = 0 . In I V a , G G I_{V_{a},GG} I V a , GG , the vertex factor 2 P − k 2P-k 2 P − k is annihilated by G ( − k ; P ) G(-k;P) G ( − k ; P ) , so this term vanishes. In I G G I_{GG} I GG and I I G G I_{IGG} I I GG , contraction with the longitudinal momentum cancels the adjacent propagators, and the soft counterterms reproduce the remaining numerator at zero momentum since J ⋅ P = 0 J \cdot P = 0 J ⋅ P = 0 . In I V G G I_{VGG} I V GG , the same contraction cancels the ( k − p 3 ) 2 (k-p_3)^2 ( k − p 3 ) 2 propagator and leaves an explicit k k k contracted with the hard tensor. Averaging each finite remainder under four-momentum reversal (r → − r r \to-r r → − r in I, k → − k k \to-k k → − k otherwise) therefore leaves a numerator of order ∣ k ∣ 2 |k|^2 ∣ k ∣ 2 at the auxiliary pinch. In the P P P rest frame the scalar residue is of order ∣ k ∣ − 4 |k|^{-4} ∣ k ∣ − 4 , so the remaining three-momentum integral is locally convergent.
In this section, we have studied the graphs I – V. In each graph, we have chosen some propagators to decompose on one or both ends using the identity G + K = 1 G+K=1 G + K = 1 . In each case, we found that the graphs obtained after Ward identity cancellations were either locally finite, factorized into products of one loop graphs or divergent in the soft limit and handled using a soft counterterm. Combined with the results of Sec. 3 every genuine two loop graph that contributes to the amplitude has been rendered locally finite.
Conclusions
In this paper, we have studied local infrared finiteness for the process q + q ˉ → g + n H q+\bar{q}\to g+nH q + q ˉ → g + n H (and by extension its crossed processes). We have classified graphs into topologically factorized, IR finite, transient singular and IR divergent. Some finite graphs were then added to the set of IR divergent graphs to manifest Ward identities.
Here, we have used two tools the scalar-scalar-gluon decomposition as well as the one-sided and two-sided G-K decomposition. For transient singularities, we showed that it is possible to make every graph locally infrared finite by a non-local counter-term (which only integrates to zero after summing over its own gauge invariant subset). This was constructed utilizing the one-sided G-K decomposition and the Ward identity that is only satisfied after the permutation sum on the fermion loop and its loop integral is carried out. The key observation here was that it is possible to replace a gluon by its G counterpart when only two gluons attach to the fermion loop. We did not need shift counterterms that are often used in other approaches to local finiteness.
For IR divergent graphs, we observed that Feynman graphs after the two-sided GK decomposition, satisfy exact Ward identities which factorize longitudinal K gluons into products of one loop graphs. The physical G gluonic propagation is always collinear finite due to the identity in Eq. (14) while soft divergence was identified with gauge invariance and gauge freedom. We constructed explicit soft counter terms which are easy to read off for the processes studied here.
Let us explain the new features of this decomposition. First, we have studied colored particles in both the initial and final state. We have supplied a modification of the Feynman rules for graphs that have genuinely two-loop topologies, when replacing a gluon with its physical polarizations. Unless the polarizations are chosen to be lightlike, such a procedure is bound to leave behind soft singularities. We constructed soft counter-terms for such topologies while we retained time-like reference vectors. This opens the door to include more colored particles, particularly to include incoming and outgoing gluons as well as to systematically generate counter-terms in an automated fashion. It would be interesting to use these tools to attack the three loop form factor which resisted local factorization in [35 ] .
Acknowledgments
I have utilized OpenAI’s ChatGPT 6 to draw the figures used in this paper. I have used it to organize the one loop integrals needed to carry out the checks described here. I have checked in each case that the integral has been correctly transcribed. I have also benefited from Feynrules and Feynarts as described in [38 ] to generate graphs to verify the completeness of the graph list as well as to inspect the symmetry factors and integrands of various graphs.
Checks
We evaluate the four contributions to the infrared poles of M ( 2 ) \mathcal{M}^{(2)} M ( 2 ) . We use d = 4 − 2 ϵ d = 4 - 2\epsilon d = 4 − 2 ϵ and the M S ‾ \overline{\mathrm{MS}} MS loop measure. The scalar integrals we need are
B ( a , b ) = ( μ 2 γ E ) ϵ ∫ d d k i π d / 2 1 ( k 2 − 2 a ⋅ k + i 0 ) ( k 2 − 2 b ⋅ k + i 0 ) , B(a,b) = \left(\mu^2\gamma_E\right)^\epsilon\int\frac{\mathrm{d}^d k}{i\pi^{d/2}} \frac{1}{(k^2 - 2a \cdot k + i0)(k^2 - 2b \cdot k + i0)}, B ( a , b ) = ( μ 2 γ E ) ϵ ∫ i π d /2 d d k ( k 2 − 2 a ⋅ k + i 0 ) ( k 2 − 2 b ⋅ k + i 0 ) 1 ,
T ( a , b ) = ( μ 2 γ E ) ϵ ∫ d d k i π d / 2 1 ( k 2 + i 0 ) ( k 2 − 2 a ⋅ k + i 0 ) ( k 2 − 2 b ⋅ k + i 0 ) . (113) T(a,b) = \left(\mu^2\gamma_E\right)^\epsilon\int\frac{\mathrm{d}^d k}{i\pi^{d/2}} \frac{1}{(k^2+i0)(k^2 - 2a \cdot k + i0)(k^2 - 2b \cdot k + i0)}. \tag*{(113)} T ( a , b ) = ( μ 2 γ E ) ϵ ∫ i π d /2 d d k ( k 2 + i 0 ) ( k 2 − 2 a ⋅ k + i 0 ) ( k 2 − 2 b ⋅ k + i 0 ) 1 . (113)
For lightlike p p p , p ′ p' p ′ and future timelike q q q , their triangle poles are
4 p ⋅ p ′ T ( p , p ′ ) ∣ p o l e s = − 2 ϵ 2 + 2 ϵ ln 2 p ⋅ p ′ − i 0 μ 2 , 4p \cdot p' T(p,p')\big|_{\mathrm{poles}} = -\frac{2}{\epsilon^2} + \frac{2}{\epsilon}\ln\frac{2p \cdot p' - i0}{\mu^2}, 4 p ⋅ p ′ T ( p , p ′ ) poles = − ϵ 2 2 + ϵ 2 ln μ 2 2 p ⋅ p ′ − i 0 ,
4 p ⋅ ( ± q ) T ( p , ± q ) ∣ p o l e s = − 1 ϵ 2 + 2 ϵ ln ± 2 p ⋅ q − i 0 μ q 2 , 4p \cdot(\pm q) T(p,\pm q)\big|_{\mathrm{poles}} = -\frac{1}{\epsilon^2} + \frac{2}{\epsilon}\ln\frac{\pm2p \cdot q - i0}{\mu\sqrt{q^2}}, 4 p ⋅ ( ± q ) T ( p , ± q ) poles = − ϵ 2 1 + ϵ 2 ln μ q 2 ± 2 p ⋅ q − i 0 ,
4 q 2 T ( q , − q ) ∣ p o l e s = − 2 ϵ . (114) 4q^2 T(q,-q)\big|_{\mathrm{poles}} = -\frac{2}{\epsilon}. \tag*{(114)} 4 q 2 T ( q , − q ) poles = − ϵ 2 . (114)
The non-scaleless bubbles below have pole 1 / ϵ 1/\epsilon 1/ ϵ . The on-shell bubble B ( 0 , p ) B(0,p) B ( 0 , p ) vanishes in dimensional regularization. Its UV and IR residues are + 1 +1 + 1 and − 1 -1 − 1 , respectively.
Five longitudinal terms. Consider the three terms on the right of Eq. (80) in the limit k ∥ p 2 k \parallel p_2 k ∥ p 2 . At leading power we set k = x p 2 k = xp_2 k = x p 2 in the fermion numerator and in the hard propagator carrying momentum r = P − k r = P-k r = P − k . Since r 2 = ( 1 − x ) P 2 r^2 = (1-x)P^2 r 2 = ( 1 − x ) P 2 , the Dirac equation gives
v ˉ ( p 2 ) γ ρ ( k̸ − ̸ p 2 ) γ μ u ( p 1 ) r 2 ∣ k = x p 2 = − 2 ( 1 − x ) p 2 ρ v ˉ ( p 2 ) γ μ u ( p 1 ) ( 1 − x ) P 2 = − 2 p 2 ρ v ˉ ( p 2 ) γ μ u ( p 1 ) P 2 . (115) \left.\frac{\bar{v}(p_2)\gamma_\rho(\not{k}-\not{p}_2)\gamma_\mu u(p_1)}{r^2}\right|_{k=xp_2}
= \frac{-2(1-x)p_{2\rho}\bar{v}(p_2)\gamma_\mu u(p_1)}{(1-x)P^2}
= -2p_{2\rho}\frac{\bar{v}(p_2)\gamma_\mu u(p_1)}{P^2}. \tag*{(115)} r 2 v ˉ ( p 2 ) γ ρ ( k − p 2 ) γ μ u ( p 1 ) k = x p 2 = ( 1 − x ) P 2 − 2 ( 1 − x ) p 2 ρ v ˉ ( p 2 ) γ μ u ( p 1 ) = − 2 p 2 ρ P 2 v ˉ ( p 2 ) γ μ u ( p 1 ) . (115)
The second term vanishes at this order because p 2 ρ G ρ τ ( k ; q ) = 0 p_{2\rho}G^{\rho\tau}(k;q)=0 p 2 ρ G ρ τ ( k ; q ) = 0 at k = x p 2 k=xp_2 k = x p 2 . In the first term, expand G μ ν ( r ; q ) G^{\mu\nu}(r;q) G μν ( r ; q ) on the hard line. Its longitudinal part cancels the third term by r ν H ν λ [ 2 , m ] ( P , − p 3 ) = − k ν H ν λ [ 2 , m ] ( P , − p 3 ) r^\nu\mathcal{H}_{\nu\lambda}[2,m](P,-p_3)=-k^\nu\mathcal{H}_{\nu\lambda}[2,m](P,-p_3) r ν H ν λ [ 2 , m ] ( P , − p 3 ) = − k ν H ν λ [ 2 , m ] ( P , − p 3 ) . The metric part of G μ ν ( r ; q ) G^{\mu\nu}(r;q) G μν ( r ; q ) then joins the quark current to the hard loop with the Born propagator 1 / P 2 1/P^2 1/ P 2 , giving M ( 1 ) \mathcal{M}^{(1)} M ( 1 ) of Eq. (108) .
The remaining numerator is − 2 p 2 ⋅ ( 2 q + x p 2 ) = − 4 p 2 ⋅ q -2p_2\cdot(2q+xp_2)=-4p_2\cdot q − 2 p 2 ⋅ ( 2 q + x p 2 ) = − 4 p 2 ⋅ q . Keeping the full k k k dependence of the three remaining denominators, we obtain the scalar triangle
− 4 p 2 ⋅ q T ( p 2 , − q ) = ( μ 2 e γ E ) ϵ ∫ d d k i π d / 2 − 4 p 2 ⋅ q ( k 2 + i 0 ) ( ( k − p 2 ) 2 + i 0 ) ( k 2 + 2 q ⋅ k + i 0 ) . (116) -4p_2\cdot q\,T(p_2,-q)=(\mu^2 e^{\gamma_E})^\epsilon\int\frac{d^d k}{i\pi^{d/2}}\frac{-4p_2\cdot q}{(k^2+i0)((k-p_2)^2+i0)(k^2+2q\cdot k+i0)}. \tag*{(116)} − 4 p 2 ⋅ q T ( p 2 , − q ) = ( μ 2 e γ E ) ϵ ∫ i π d /2 d d k ( k 2 + i 0 ) (( k − p 2 ) 2 + i 0 ) ( k 2 + 2 q ⋅ k + i 0 ) − 4 p 2 ⋅ q . (116)
This triangle also reproduces the leading soft integrand at k = 0 k=0 k = 0 , where r = P r=P r = P . The region r ∥ p 1 r\parallel p_1 r ∥ p 1 gives the corresponding triangle − 4 p 1 ⋅ q T ( p 1 , − q ) -4p_1\cdot q\,T(p_1,-q) − 4 p 1 ⋅ q T ( p 1 , − q ) . Restoring the colour and coupling factors, we find
[ ∫ d d k ( 2 π ) d ( I I I I , α + I I I I , β + I I I I , γ + I I I , δ + I I I I , σ ) ] p o l e s = [ g s 2 C A 32 π 2 M ( 1 ) ∑ i = 1 2 ( − 4 p i ⋅ q ) T ( p i , − q ) ] p o l e s . (117) \left[\int\frac{d^d k}{(2\pi)^d}\left(\mathcal{I}_{\mathrm{III},\alpha}+\mathcal{I}_{\mathrm{III},\beta}+\mathcal{I}_{\mathrm{III},\gamma}+\mathcal{I}_{\mathrm{II},\delta}+\mathcal{I}_{\mathrm{III},\sigma}\right)\right]_{\mathrm{poles}}
=
\left[\frac{g_s^2 C_A}{32\pi^2}\mathcal{M}^{(1)}\sum_{i=1}^{2}(-4p_i\cdot q)T(p_i,-q)\right]_{\mathrm{poles}}. \tag*{(117)} [ ∫ ( 2 π ) d d d k ( I III , α + I III , β + I III , γ + I II , δ + I III , σ ) ] poles = [ 32 π 2 g s 2 C A M ( 1 ) i = 1 ∑ 2 ( − 4 p i ⋅ q ) T ( p i , − q ) ] poles . (117)
Consider next J 4 + J 5 J_4+J_5 J 4 + J 5 in Eq. (105) . The hard gluon carries momentum P P P , and the two momenta entering the fermion loop are P P P and − p 3 -p_3 − p 3 . The quark current, the hard propagator and the fermion loop therefore give precisely M ( 1 ) \mathcal{M}^{(1)} M ( 1 ) of Eq. (108) . This factorization is exact:
J 4 ( k ) + J 5 ( k ) = − i g s 2 C A M ( 1 ) ( 2 q − k ) ζ G ζ β ( k ; q ) ( 2 p 3 − k ) β ( k 2 + i 0 ) ( ( k − p 3 ) 2 + i 0 ) ( k 2 − 2 q ⋅ k + i 0 ) . (118) J_4(k)+J_5(k)=-ig_s^2 C_A\mathcal{M}^{(1)}\frac{(2q-k)_\zeta G^{\zeta\beta}(k;q)(2p_3-k)_\beta}{(k^2+i0)((k-p_3)^2+i0)(k^2-2q\cdot k+i0)}. \tag*{(118)} J 4 ( k ) + J 5 ( k ) = − i g s 2 C A M ( 1 ) ( k 2 + i 0 ) (( k − p 3 ) 2 + i 0 ) ( k 2 − 2 q ⋅ k + i 0 ) ( 2 q − k ) ζ G ζ β ( k ; q ) ( 2 p 3 − k ) β . (118)
It remains to integrate the scalar factor. Expanding G ( k ; q ) G(k;q) G ( k ; q ) and using k ⋅ ( 2 p 3 − k ) = − ( k − p 3 ) 2 k\cdot(2p_3-k)=-(k-p_3)^2 k ⋅ ( 2 p 3 − k ) = − ( k − p 3 ) 2 , we obtain
( 2 q − k ) ζ G ζ β ( k ; q ) ( 2 p 3 − k ) β = 4 q ⋅ p 3 + ( k 2 − 2 q ⋅ k ) + ( k − p 3 ) 2 − k 2 + ( 4 q 2 − k 2 ) ( k − p 3 ) 2 k 2 + 2 q ⋅ k + i 0 . (119) (2q-k)_\zeta G^{\zeta\beta}(k;q)(2p_3-k)_\beta=4q\cdot p_3+(k^2-2q\cdot k)+(k-p_3)^2-k^2+\frac{(4q^2-k^2)(k-p_3)^2}{k^2+2q\cdot k+i0}. \tag*{(119)} ( 2 q − k ) ζ G ζ β ( k ; q ) ( 2 p 3 − k ) β = 4 q ⋅ p 3 + ( k 2 − 2 q ⋅ k ) + ( k − p 3 ) 2 − k 2 + k 2 + 2 q ⋅ k + i 0 ( 4 q 2 − k 2 ) ( k − p 3 ) 2 . (119)
Cancelling the corresponding propagators gives
∫ d d k ( 2 π ) d [ J 4 ( k ) + J 5 ( k ) ] = g s 2 C A 16 π 2 M ( 1 ) [ 4 q ⋅ p 3 T ( p 3 , q ) + B ( 0 , p 3 ) + B ( 0 , q ) − B ( p 3 , q ) + 4 q 2 T ( q , − q ) − B ( q , − q ) ] . (120) \int\frac{d^d k}{(2\pi)^d}[J_4(k)+J_5(k)]=\frac{g_s^2C_A}{16\pi^2}\mathcal{M}^{(1)}[4q\cdot p_3T(p_3,q)+B(0,p_3)+B(0,q)-B(p_3,q)+4q^2T(q,-q)-B(q,-q)]. \tag*{(120)} ∫ ( 2 π ) d d d k [ J 4 ( k ) + J 5 ( k )] = 16 π 2 g s 2 C A M ( 1 ) [ 4 q ⋅ p 3 T ( p 3 , q ) + B ( 0 , p 3 ) + B ( 0 , q ) − B ( p 3 , q ) + 4 q 2 T ( q , − q ) − B ( q , − q )] . (120)
The on-shell bubble B ( 0 , p 3 ) B(0,p_3) B ( 0 , p 3 ) vanishes, and the poles of B ( 0 , q ) B(0,q) B ( 0 , q ) and B ( p 3 , q ) B(p_3,q) B ( p 3 , q ) cancel. The last two terms give − 2 / ϵ -2/\epsilon − 2/ ϵ and − 1 / ϵ -1/\epsilon − 1/ ϵ , respectively. Inserting the pole of T ( p 3 , q ) T(p_3,q) T ( p 3 , q ) from Eq. (A.2), we find
[ ∫ d d k ( 2 π ) d [ J 4 ( k ) + J 5 ( k ) ] ] p o l e s = [ g s 2 C A 16 π 2 M ( 1 ) ( − 1 ϵ 2 + 2 ϵ ln 2 q ⋅ p 3 μ q 2 − 3 ϵ ) ] p o l e s . (121) \left[\int\frac{d^d k}{(2\pi)^d}[J_4(k)+J_5(k)]\right]_{\mathrm{poles}}
=
\left[\frac{g_s^2C_A}{16\pi^2}\mathcal{M}^{(1)}\left(-\frac{1}{\epsilon^2}+\frac{2}{\epsilon}\ln\frac{2q\cdot p_3}{\mu\sqrt{q^2}}-\frac{3}{\epsilon}\right)\right]_{\mathrm{poles}}. \tag*{(121)} [ ∫ ( 2 π ) d d d k [ J 4 ( k ) + J 5 ( k )] ] poles = [ 16 π 2 g s 2 C A M ( 1 ) ( − ϵ 2 1 + ϵ 2 ln μ q 2 2 q ⋅ p 3 − ϵ 3 ) ] poles . (121)
The three topologically factorized vertex graphs give the one-loop q q ˉ g ∗ q\bar qg^* q q ˉ g ∗ vertex with p 1 2 = p 2 2 = 0 p_1^2=p_2^2=0 p 1 2 = p 2 2 = 0 and gluon virtuality P 2 P^2 P 2 . Using the massless vertex in Feynman gauge [39 ] and including the on-shell residues of the two quarks and the gluon, we obtain, after UV renormalization,
M f a c t ( 2 ) ∣ p o l e s , i n c l . l e g s = [ g s 2 16 π 2 M ( 1 ) { ( 2 C F − C A ) ( − 1 ε 2 + 1 ε ln − P 2 − i 0 μ 2 ) − 3 C F ε − 5 C A − 4 T R n f 6 ε } ] p o l e s (122) \left.\mathcal{M}_{\mathrm{fact}}^{(2)}\right|_{\mathrm{poles,\ incl.\ legs}}=\left[\frac{g_s^2}{16\pi^2}\mathcal{M}^{(1)}\left\{(2C_F-C_A)\left(-\frac{1}{\varepsilon^2}+\frac{1}{\varepsilon}\ln\frac{-P^2-\mathrm{i}0}{\mu^2}\right)-\frac{3C_F}{\varepsilon}-\frac{5C_A-4T_Rn_f}{6\varepsilon}\right\}\right]_{\mathrm{poles}} \tag*{(122)} M fact ( 2 ) poles , incl. legs = [ 16 π 2 g s 2 M ( 1 ) { ( 2 C F − C A ) ( − ε 2 1 + ε 1 ln μ 2 − P 2 − i 0 ) − ε 3 C F − 6 ε 5 C A − 4 T R n f } ] poles (122)
Finally, consider C I I s o f t \mathcal{C}_{\mathrm{II}}^{\mathrm{soft}} C II soft in Eq. (111) . We retain both currents as written. Expanding the two G G G tensors cancels adjacent propagators through k ⋅ ( 2 p j − k ) = − ( p j − k ) 2 k\cdot(2p_j-k)=-(p_j-k)^2 k ⋅ ( 2 p j − k ) = − ( p j − k ) 2 , for j = 2 , 3 j=2,3 j = 2 , 3 . The exact reduction is
∫ d d k ( 2 π ) d C I I s o f t ( k ) = g s 2 C A 32 π 2 M ( 1 ) [ 4 p 2 ⋅ p 3 T ( p 2 , p 3 ) − B ( p 2 , p 3 ) \int\frac{\mathrm{d}^d k}{(2\pi)^d}\mathcal{C}_{\mathrm{II}}^{\mathrm{soft}}(k)=\frac{g_s^2C_A}{32\pi^2}\mathcal{M}^{(1)}[4p_2\cdot p_3\,T(p_2,p_3)-B(p_2,p_3) ∫ ( 2 π ) d d d k C II soft ( k ) = 32 π 2 g s 2 C A M ( 1 ) [ 4 p 2 ⋅ p 3 T ( p 2 , p 3 ) − B ( p 2 , p 3 )
+ B ( p 3 , q ) − B ( 0 , q ) − 4 p 3 ⋅ q T ( p 3 , q ) \qquad+B(p_3,q)-B(0,q)-4p_3\cdot q\,T(p_3,q) + B ( p 3 , q ) − B ( 0 , q ) − 4 p 3 ⋅ q T ( p 3 , q )
+ B ( p 2 , − q ) − B ( 0 , − q ) + 4 p 2 ⋅ q T ( p 2 , − q ) \qquad+B(p_2,-q)-B(0,-q)+4p_2\cdot q\,T(p_2,-q) + B ( p 2 , − q ) − B ( 0 , − q ) + 4 p 2 ⋅ q T ( p 2 , − q )
+ B ( q , − q ) − 4 q 2 T ( q , − q ) ] . (123) \qquad+B(q,-q)-4q^2T(q,-q)]. \tag*{(123)} + B ( q , − q ) − 4 q 2 T ( q , − q )] . (123)
The six bubble poles cancel. . The change of variable k → P − k k\to P-k k → P − k gives the same reduction for C I s o f t \mathcal{C}_{\mathrm{I}}^{\mathrm{soft}} C I soft with p 2 → p 1 p_2\to p_1 p 2 → p 1 . Hence
[ ∫ d d k ( 2 π ) d ( C I s o f t + C I I s o f t ) ] p o l e s = [ g s 2 C A 32 π 2 M ( 1 ) ∑ i = 1 2 ( 4 p i ⋅ p 3 T ( p i , p 3 ) + 4 p i ⋅ q T ( p i , − q ) − 4 p 3 ⋅ q T ( p 3 , q ) + 2 ε ) ] p o l e s . (124) \left[\int\frac{\mathrm{d}^d k}{(2\pi)^d}\left(\mathcal{C}_{\mathrm{I}}^{\mathrm{soft}}+\mathcal{C}_{\mathrm{II}}^{\mathrm{soft}}\right)\right]_{\mathrm{poles}}=
\left[\frac{g_s^2C_A}{32\pi^2}\mathcal{M}^{(1)}\sum_{i=1}^{2}\left(4p_i\cdot p_3\,T(p_i,p_3)+4p_i\cdot q\,T(p_i,-q)-4p_3\cdot q\,T(p_3,q)+\frac{2}{\varepsilon}\right)\right]_{\mathrm{poles}}. \tag*{(124)} [ ∫ ( 2 π ) d d d k ( C I soft + C II soft ) ] poles = [ 32 π 2 g s 2 C A M ( 1 ) i = 1 ∑ 2 ( 4 p i ⋅ p 3 T ( p i , p 3 ) + 4 p i ⋅ q T ( p i , − q ) − 4 p 3 ⋅ q T ( p 3 , q ) + ε 2 ) ] poles . (124)
The three triangle double poles cancel within each counterterm.
On adding Eqs. (A.5), (A.9), and (A.12), the coefficients of T ( p i , − q ) T(p_i,-q) T ( p i , − q ) and T ( p 3 , q ) T(p_3,q) T ( p 3 , q ) vanish. The remaining explicit single pole from these three sources is − C A / ε -C_A/\varepsilon − C A / ε . Adding Eq. (A.10) and evaluating the massless triangles gives
M r e n ( 2 ) ∣ p o l e s = [ g s 2 16 π 2 M ( 1 ) { − 2 C F + C A ε 2 + 2 C F − C A ε ln − P 2 − i 0 μ 2 + C A ε ∑ i = 1 2 ln 2 p i ⋅ p 3 μ 2 − 3 C F ε − 11 C A − 4 T R n f 6 ε } ] p o l e s . (125) \left.\mathcal{M}_{\mathrm{ren}}^{(2)}\right|_{\mathrm{poles}}=
\left[\frac{g_s^2}{16\pi^2}\mathcal{M}^{(1)}
\left\{-\frac{2C_F+C_A}{\varepsilon^2}
+\frac{2C_F-C_A}{\varepsilon}\ln\frac{-P^2-\mathrm{i}0}{\mu^2}
+\frac{C_A}{\varepsilon}\sum_{i=1}^{2}\ln\frac{2p_i\cdot p_3}{\mu^2}
-\frac{3C_F}{\varepsilon}
-\frac{11C_A-4T_Rn_f}{6\varepsilon}\right\}\right]_{\mathrm{poles}}. \tag*{(125)} M ren ( 2 ) poles = [ 16 π 2 g s 2 M ( 1 ) { − ε 2 2 C F + C A + ε 2 C F − C A ln μ 2 − P 2 − i 0 + ε C A i = 1 ∑ 2 ln μ 2 2 p i ⋅ p 3 − ε 3 C F − 6 ε 11 C A − 4 T R n f } ] poles . (125)
This agrees with the prediction of poles for this process made in [40 ] for q q ˉ g q\bar{q}g q q ˉ g final state continued to the annihilation channel. Since the tree amplitude vanishes, the insertion multiplies the full d d d -dimensional M ( 1 ) \mathcal{M}^{(1)} M ( 1 ) of Eq. (108) .