1 sec:Introduction↩︎

Thermal field theory describes relativistic matter, from the quark–gluon plasma to the early Universe, and provides a framework for studying thermal dispersion, damping, transport, and particle production [1]–[10]. For perturbative calculations at weak coupling, temperature is usually treated as the dominant scale, which leads to scale separation: hard scales are of order \(T\), and soft scales are of order \(eT\), where \(e\) is the coupling. At a given loop order, the corresponding amplitudes can therefore be decomposed into momentum regions according to whether the internal lines carry hard or soft momenta. At two loops, this decomposition gives the hard–hard, hard–soft, and soft–soft regions. When soft internal lines are present, the bare perturbative expansion contains an infinite series of contributions at the same parametric order. This arises because leading thermal self-energies and vertices, known as Hard Thermal Loop (HTL) insertions, are of the same parametric order as the corresponding bare terms. HTL resummation reorganizes this series into amplitudes with resummed soft propagators and vertices [11]–[16].

For example, the soft one-loop self-energy with HTL-resummed propagators and vertices contains the series \(\Sigma^{*}=\Sigma_S+\Sigma_{HS}+\Sigma_{H^2S}+\cdots\), where \(H\), \(S\), and the star denote a hard subgraph, a soft loop, and the HTL-resummed one-loop contribution, respectively. The zeroth and first terms above correspond to the bare soft one-loop and hard-soft two-loop configurations, while subsequent terms include additional hard subgraphs. In Feynman gauge and for \(\mathbf{q}\to0\), Ref. [17] showed that the bare hard-soft two-loop contribution to the fermion self-energy, which scales as \(e^2T\), agrees with the first term of the HTL-resummed one-loop result. Therefore, the bare hard-soft configuration is already included in the soft one-loop HTL-resummed self-energy and should not be counted separately. The soft-soft region also belongs to the resummed two-loop soft sector. By contrast, in the hard region, HTL insertions are suppressed relative to hard bare inverse propagators, so soft-line resummation is not required. Therefore, strict perturbation theory applies to this region, which is the subject of this study. For the hard-region amplitudes considered here, the external momenta are taken to be soft.

Throughout, the terminology is as follows. Power order denotes the expansion in the soft external momentum at fixed loop order: leading power (LP), next-to-leading power (NLP), and so on. We also use labels such as one-loop LP, one-loop NLP, and two-loop LP. By contrast, leading order (LO), next-to-leading order (NLO), and next-to-next-to-leading order (NNLO) refer to the perturbative hierarchy of a physical quantity, such as a mass shift, effective Lagrangian, or damping rate. For the first two quantities, LO denotes the first nonvanishing contribution. For the damping rate, following Ref. [18], the leading contribution is defined as zero. The term “axial coefficient” refers only to the coefficient of the \(\gamma_\nu\gamma_5\) Dirac structure.

The power-counting analysis of Ref. [19] shows that the NLO fermion damping rate [18], [20], [21] and mass shift [22], [23] arise from the resummed soft one-loop self-energy. For the fermion self-energy, the hard one-loop NLP and hard-hard two-loop LP contributions first appear at order \(e^3T\), which enter the NNLO in the pole quantities. The corresponding gluonic quantities obey the same hierarchy [3], [24], [25], whereas photon-sector quantities follow a different hard/soft hierarchy [19].

The hard contributions to gauge-boson self-energies have been studied up to two-loop order [26]–[31]. These hard corrections to the photon self-energy are gauge independent [30], [31]. By contrast, the one-loop NLP and two-loop LP hard-region contributions to the gluon self-energy are gauge dependent [29], but this does not signal an inconsistency. A physical quantity requires combining all same-order contributions; thus, for the gluon self-energy, the corresponding same-order soft contributions must also be included [29]. Related Feynman-gauge results have also been derived using kinetic-theory methods [32]. On the other hand, the hard-region fermion self-energy has been studied at one-loop LP and NLP orders [2], [27]. At NLP order, gauge-dependent corrections and a local UV divergence appear; the latter is removed by the zero-temperature renormalization counterterm [27]. However, hard two-loop corrections to the fermion self-energy remain unstudied. Similarly, the off-shell fermion–photon vertex has mainly been analyzed at one-loop LP [26], [33]. To our knowledge, higher corrections, including one-loop NLP and hard two-loop contributions, have not been studied.

The Abelian LO HTL effective action gives the one-loop LP photon polarization tensor and the two-fermion–\(N\)-photon vertices, with \(N=0\) corresponding to the fermion self-energy. That of the non-Abelian gauge theory gives the corresponding gluon polarization tensor, pure \(N\)-gluon HTL vertices with \(N\geq 3\), and two-fermion–\(N\)-gluon vertices, again with \(N=0\) corresponding to the fermion self-energy [11]–[16], [26], [34]. At one-loop LP, these HTL amplitudes are organized by recursive WTIs: contraction with an external gauge-boson momentum relates higher-point HTL vertices to lower-point HTL functions. Since the WTI determines only the projection longitudinal to the photon momentum, structures transverse to that momentum remain invisible to the WTI [33], [35]–[37]. This limitation is important beyond LP, where axial coefficients with a different Dirac structure can appear and are themselves transverse to the photon momentum. Thus, higher-point amplitudes cannot, in general, be inferred from lower-point amplitudes by WTIs alone. The NLO photon-sector effective action has been constructed in Refs. [27], [30]. By contrast, completing the construction of the fermionic NLO effective action requires one-loop NLP and two-loop LP inputs from the self-energy and the two-fermion–\(N\)-photon vertex functions. The one-loop NLP self-energy contribution was constructed in Ref. [27]. Gauge consistency of these amplitudes is constrained by the WTI and axial-transversality conditions. So far, WTI checks have been confined to the one-loop LP level [26], [38]–[41].

We address these closely connected issues in two steps. First, we derive nonperturbative symmetry constraints on fermionic amplitudes. These constraints lead to definite momentum-reversal, fermion-leg exchange, and reality properties. Under momentum reversal, the fermion spectral function is even, while the self-energy is odd up to the retarded/advanced prescriptions. Chiral invariance also restricts the two-fermion–\(N\)-photon amplitudes to vector and axial-vector Dirac structures, each of which obeys definite symmetry constraints. These constraints lead to selection rules in the soft-external-momentum expansion. Second, we compute the one-loop NLP and two-loop LP hard-region corrections to the fermion self-energy and off-shell fermion–photon vertex in massless QED in a general covariant \(R_\xi\) gauge. To remove UV subdivergences, we include the zero-temperature counterterm contributions to these amplitudes. The WTI enforces gauge consistency between the self-energy and the Dirac-vector part of the vertex. The axial coefficient is instead transverse to the photon momentum. We impose the discrete-symmetry constraints, the WTI, and axial transversality at the integrand level as consistency checks. In addition, we show that the hard self-energy corrections give no finite contribution to the damping rate. Instead, these hard-region contributions provide the inputs required for complete mass-shift and inclusive-rate analyses and serve as starting points for constructing the corresponding NLO effective Lagrangian. Following the hard/soft matching logic of Refs. [42]–[44], we summarize the self-energy matching analysis and show how all contributions at the same order should be combined to obtain pole quantities.

The paper is organized as follows. Sec. 2 introduces the real-time formalism and conventions, with technical details given in App. 8. Sec. 3 presents the renormalized hard self-energy corrections. Sec. 4 presents the symmetry constraints on the two-fermion–\(N\)-photon vertices and derives the corresponding selection rules. Sec. 5 presents the fermion–photon vertex results. Sec. 6 discusses the matching analysis for the mass shift and damping rate. Sec. 7 summarizes the main findings. The main text contains the conceptual arguments, while Apps. 9–13 provide technical details that may be skipped on a first reading. App. 14 collects lengthy expressions. The ancillary file contains the integrand-level one-loop NLP and two-loop LP results. These expressions are part of the result set of the paper, but several of them are too long to be included even in App. 14. The WTI and axial-transversality checks, along with the other algebraic checks stated in the paper, are also implemented in the ancillary file, and a summary table collects the sector-by-sector verifications.

2 Real-time Formalism and Conventions↩︎

We work in massless QED in a homogeneous, isotropic, equilibrium medium. We adopt the Schwinger–Keldysh real-time formalism [45]–[50], in which the retarded, advanced, and symmetric components of self-energies and propagators are related to their \(r/a\) counterparts as [49], [51] \[\begin{align} \label{3} \begin{pmatrix} \Delta_{rr} & \Delta_{ra}\\ \Delta_{ar} & \Delta_{aa} \end{pmatrix} = \begin{pmatrix} \Delta_s & \Delta_R\\ \Delta_A & 0 \end{pmatrix}, \quad \begin{pmatrix} \Pi_{rr} & \Pi_{ra}\\ \Pi_{ar} & \Pi_{aa} \end{pmatrix} = \begin{pmatrix} 0 & \Pi_A\\ \Pi_R & \Pi_s \end{pmatrix}. \end{align}\tag{1}\] Here \(R\), \(A\), and \(s\) denote retarded, advanced, and symmetric components, respectively. The free fermion and gauge-boson propagators are \[\begin{align} \label{5} S_j(P) &= \cancel{P}\,\Delta_j(P), \qquad D_j^{\mu\nu}(P) = -g^{\mu\nu}\widetilde{\Delta}_j(P) -i(1-\xi)P^\mu P^\nu\widetilde{\Delta}_j^{(2)}(P), \qquad j=R, A,s, \end{align}\tag{2}\] where \(\xi\) is the covariant-gauge parameter. In the loop calculations, the contributions associated with the \(g^{\mu\nu}\) and \((1-\xi)\)-dependent parts are referred to as the metric and longitudinal sectors, respectively. Here, the scalar functions are defined by \[\begin{align} \label{scalar-functions} \Delta_R(P) &= \frac{i}{P^2+ip^0\eta}, \qquad \Delta_A(P) = \frac{i}{P^2-ip^0\eta}, \qquad \Delta_{R/A}^{(n)}(P) \equiv \left[\Delta_{R/A}(P)\right]^n,\nonumber \\ \Delta_d^{(n)}(P) &= \Delta_R^{(n)}(P)-\Delta_A^{(n)}(P), \qquad \Delta_s^{(n)}(P) = N_F(P)\Delta_d^{(n)}(P), \qquad \widetilde{\Delta}_s^{(n)}(P) = N_B(P)\Delta_d^{(n)}(P), \end{align}\tag{3}\] with \(\eta\to0^+\). The retarded, advanced, and difference scalar functions are the same for fermions and bosons; the tilde distinguishes the bosonic symmetric propagator from the fermionic one. The statistical factors are \[\begin{align} N_F(P) = \frac{1}{2}-n_F(p^0), \qquad N_B(P) = \frac{1}{2}+n_B(p^0), \qquad N_{F/B}'(P) \equiv \frac{dN_{F/B}(P)}{dp^0}. \end{align}\] For \(n=1\), the difference propagator is the spectral function, and we write \(\Delta_d^{(1)}\equiv\Delta_d\). Table 1 summarizes the remaining conventions. The \(\overline{\rm MS}\) scheme and dimensional regularization (DR) are used throughout, and the radial and angular integrals needed below are collected in App. 8. Adopting the BMHV scheme [52], [53], we used FeynCalc [54]–[57] for the Dirac algebra and scalar products.

We use the following QED renormalization convention \[\begin{align} \psi_0=Z_2^{1/2}\psi,\qquad A_0^\mu=Z_3^{1/2}A^\mu,\qquad \xi_0=Z_3\xi,\qquad e_0=\bar\Lambda^\epsilon Z_1 Z_2^{-1} Z_3^{-1/2}e . \end{align}\] Using \(Z_i=1+\delta_i\), one writes the bare Lagrangian, expressed in terms of renormalized fields and parameters, as \[\begin{align} \mathcal{L} &= -\frac{1+\delta_3}{4}F_{\mu\nu}F^{\mu\nu} -\frac{1}{2\xi}(\partial_\mu A^\mu)^2 +(1+\delta_2)\bar\psi i\cancel{\partial}\psi -e\,\bar\Lambda^\epsilon(1+\delta_1) \bar\psi\gamma^\mu\psi A_\mu, \end{align}\] with the local counterterm insertions \[\begin{align} \label{Local-counterterms-insertions} -i\mathcal{C}_\psi(P) = i\delta_2\cancel{P},\quad -i\mathcal{C}_A^{\mu\nu}(P) = i\delta_3\left(P^2g^{\mu\nu}-P^\mu P^\nu\right),\quad -ie\mathcal{C}_{\bar\psi A\psi}^{\mu} = -ie\,\bar\Lambda^\epsilon\delta_1\gamma^\mu. \end{align}\tag{4}\] Here, the WTI gives \(Z_1=Z_2\) or \(\delta_1=\delta_2\). For one massless fermion flavor, the one-loop zero-temperature counterterms in DR are \[\begin{align} \label{Local-counterterms} \delta^{(1)}_2 = -\frac{e^2}{16\pi^2}\frac{\xi}{\epsilon}, \quad \delta^{(1)}_3 = -\frac{e^2}{12\pi^2}\frac{1}{\epsilon}, \end{align}\tag{5}\] where the superscript \((1)\) denotes the one-loop correction and \(\epsilon\rightarrow 0\). Additionally, the dressed counterterm propagators are \[\begin{align} \label{ra-ct} S_{j,{\rm CT}}(P) = i\delta_2 P^2\cancel{P}\,\Delta_j^{(2)}(P),\qquad D_{j,{\rm CT}}^{\mu\nu}(P) = i\delta_3 \left(P^2g^{\mu\nu}-P^\mu P^\nu\right) \widetilde{\Delta}_j^{(2)}(P), \qquad j=R, A,s. \end{align}\tag{6}\] Figure 1 shows the free \(r/a\) propagators, where Lorentz indices on gauge-boson lines are suppressed. Throughout the appendices, a cross on an internal fermion or photon line denotes the insertion of these propagators. The ordinary QED Feynman rules [58] are used for each \(r/a\) assignment. Each Feynman diagram is replaced by the sum over its allowed \(r/a\) contributions.

We denote virtual corrections by \(\Sigma^{(l;r)}\) and \(\Gamma^{\mu\nu\cdots(l;r)}\). Lorentz, temporal, and spatial indices are written before the parentheses, while \((l;r)\) labels the loop order and the order in the soft-momentum expansion, with \(l\in\{1,2\}\) and \(r\in\{\mathrm{LP},\mathrm{NLP},\ldots\}\). Before the soft expansion is performed, the power label is omitted. Subscripts distinguish gauge sectors, statistical sectors, and topologies when needed. We use the following gauge sector conventions: the purely metric sector, corresponding to the \(g^{\mu\nu}\) part of the gauge-boson propagators, is denoted by the subscript \(m\); sectors proportional to a single factor of \((1-\xi)\) are denoted by the subscript \(\xi\); and the sector proportional to \((1-\xi)^2\) is denoted by the subscript \(\xi^2\).

In the \(r/a\) representation, retarded and advanced propagators encode causal propagation, while symmetric propagators carry the statistical factors [29]–[31], [49]. Using Eqs. 1 and 2 and Tab. 1, one can construct all non-vanishing \(r/a\) assignments, as discussed in Ref. [49]. In imaginary time, Gaudin’s method [17], [59], [60] provides a topological organization of Matsubara sums, rewriting them as a complete set of tree-like graphs. This method is the diagrammatic analog of the real-time \(r/a\) formalism used below. For an \(L\)-loop diagram, we first choose \(L\) internal lines to carry symmetric propagators. These lines are chosen so that, once separated from the original diagram, the remaining graph is connected and tree-like. The remaining internal lines are then completed with retarded/advanced propagators, with the \(r/a\) indices assigned according to the \(N\)-point rules [18], [50]. In our case, the required rules are given in Eqs. 1 and 2 and Tab. 1. For a graph with \(N_v\) vertices and \(L\) loops, the numbers of retarded/advanced and symmetric propagators are \(N_v-1\) and \(L\), respectively [59], [60]. The complete thermal contribution is obtained by summing over all such choices of symmetric propagators and the retarded/advanced tree completions. A representative assignment is given in Sec. 3.2.

This construction isolates the temperature-dependent sector through the symmetric propagators. At two loops, there are also zero-temperature \(r/a\) contributions containing no symmetric propagator [30]. These can be obtained directly from the rules stated above, but they are excluded from this study.

3 Fermion Self-Energy↩︎

We begin this section with the symmetry constraints on the nonperturbative fermion spectral function and self-energy, with derivations given in App. 9. Chiral and parity invariance restrict the spectral function to \[\rho_F(Q)=a(Q)\,\cancel{Q}+b(Q)\,\cancel{u},\] where \(a(Q)\) and \(b(Q)\) are real scalar functions of the Lorentz invariants \(Q^2\) and \(u\cdot Q\). Their reality follows from the pseudo-Hermiticity relation of the spectral function, Eq. 72 . Charge conjugation further enforces the evenness of the spectral function under momentum reversal: \(\rho_F(-Q)=\rho_F(Q)\). Together with the spectral representation of the fermion propagator, Eq. 75 , this implies \(S_A(-Q)=-S_R(Q)\) and hence \(\Sigma_A(-Q)=-\Sigma_R(Q)\). The difference between the advanced and retarded cases lies in the explicit \(\pm i\eta\) prescriptions, while their common principal-value part is the same. At the integrand level, we suppress these prescriptions before integration and compare the algebraic part. In this algebraic sense, the relation above becomes \[\begin{align} \label{algebraic} \Sigma_{\rm alg}(-Q)=-\Sigma_{\rm alg}(Q), \end{align}\tag{7}\] which gives the oddness of the self-energy up to the retarded/advanced prescriptions.

In the hard sector, we seek to capture the contributions from internal loop momenta of order \(P\sim T\) (hard momenta), where the distribution functions peak. The hard-sector computation is implemented by rescaling \(Q \to a\, Q\), performing a Laurent expansion in the bookkeeping parameter \(a\), and then taking the limit \(a \to 1\) [29], [31]. Following Refs. [30], [31], we first perform the soft expansion at the algebraic level and restore the retarded prescription (\(q^0\to q^0+i\eta\)) for the external momentum in the subsequent integrations, which yields the retarded fermion self-energy. Expanded in the soft external momentum, Eq. 7 becomes, \[\begin{align} \Sigma_{\rm alg}(aQ)=\sum_r a^r \Sigma^{(r)}(Q), \end{align}\] which implies \[\begin{align} \big[(-1)^r+1\big]\Sigma^{(r)}(Q)=0. \end{align}\] Thus, this relation gives a selection rule that forbids even powers in the soft expansion of the self-energy. For the self-energy, LP and NLP correspond to the \(a^{-1}\) and \(a^{1}\) terms in the soft expansion, respectively.

In the plasma rest frame \(u^\mu=(1,\mathbf{0})\), the retarded self-energy takes the form \[\label{self-energy32rest32frame} \Sigma(Q)= \Sigma^{0}(Q)\,\gamma^0+\Sigma^{i}(Q)\,\gamma^i,\tag{8}\] where we have dropped the retarded subscript \(R\). The temporal and spatial components are then \[\label{projected32self-energy} \Sigma^{0}(Q)\equiv\frac{1}{4}\,\mathrm{Tr}\!\left[\gamma^0\Sigma(Q)\right], \qquad \Sigma^{i}(Q)\equiv-\frac{1}{4}\,\mathrm{Tr}\!\left[\gamma^i\Sigma(Q)\right].\tag{9}\]

3.1 One-Loop Results↩︎

At one-loop order, the self-energy contains both metric and longitudinal contributions and takes the form (see Eq. 91 ) \[\begin{align} \label{Eq:32LO32self-energy32general} \Sigma^{(1)}(Q)=\Sigma_{m}^{(1)}+\Sigma_{\xi}^{(1)}, \end{align}\tag{10}\] with \(m\) and \(\xi\) denoting the metric and longitudinal sectors. As computed in App. 10, the retarded self-energy at LP is [2], [26] \[\label{14} \Sigma^{(1;\rm LP)}(Q)\equiv \mathcal{H}(T)=m^2(T)\left[\gamma^0 L(Q)+\frac{\hat{\mathbf{q}}.\boldsymbol{\gamma}}{q}\bigl[1-q^0 L(Q)\bigr] \right],\tag{11}\] where \[L(Q)\equiv \frac{1}{2q}\ln\!\left(\frac{q^0+q+i\epsilon}{q^0-q+i\epsilon}\right) =\frac{1}{q}\mathcal{Q}_0\!\left(\frac{q^0}{q}\right).\] Here, the gauge-dependent sector does not contribute, in agreement with earlier results in the literature [17], [27]. Additionally, \(\mathcal{Q}_0(q^0/q)\) denotes the Legendre function of the second kind, and \(m^2(T)=\frac{e^2T^2}{8}\) is the (QED) fermion thermal mass [26]. This LP expression is consistent with the momentum-reversal property: under \(Q\to -Q\), one has \(\Sigma_{LP,R}(-Q)=-\Sigma_{LP,A}(Q)\) after integration. As shown in Eqs. 95 and 96 , the soft-expanded integrand obeys the same selection rule: only odd powers of the external momentum survive, while the even-in-\(Q\) terms vanish by loop-momentum reversal.

After the HTL expansion, the relevant one-loop integrands have the schematic form \[\begin{align} \int_P \Delta_d(P) N_{B/F}(P)\,F(Q,P). \end{align}\] The contribution \(\Delta_d(P)N_{B/F}(P)\) is even, under \(P\to-P\). Hence, the factor \(F(Q,P)\) must be even under \(P\to -P\) to yield a nonzero contribution. This integrand symmetry is compatible with the oddness selection rule: it retains only the odd-in-\(Q\) terms. If higher difference propagators appear, the same argument applies after using Eq. 57 .

The NLP corrections are obtained by expanding to order \(a^1\), as presented in Eq. 97 . As can be seen, the expanded integrand is odd under \(Q\to -Q\). Using the radial and angular integrations given in App. 8, one obtains the NLP corrections as \[\tag{12} \begin{align} &\Sigma^{(1;\rm NLP)}_{m} =\frac{e^2\, \cancel{Q}}{16\pi^2}\Bigl\{-\frac{1}{\epsilon}-2\ln\left(\frac{e^{\gamma_E}\bar{\Lambda}}{2\pi T}\right)+1-2q^0L(Q)\Bigr\}, \tag{13} \\ &\Sigma^{(1;\rm NLP)}_{\xi}=(1-\xi)\frac{e^2}{16\pi^2}\Bigl\{\cancel{Q}\Bigl[\frac{1}{\epsilon}+2\ln\left(\frac{e^{\gamma_E}\bar{\Lambda}}{2\pi T}\right)-1+2q^0 L(Q)\Bigr]+\frac{Q^2}{q^2}\mathbf{q}.\boldsymbol{\gamma}\Bigl[1-q^0 L(Q)\Bigr]\Bigr\} \tag{14}. \end{align}\] The divergences appearing in the equations above are UV in origin. To see this explicitly, we include the one-loop local counterterm, Eqs. 4 –5 , which yields \[\begin{align} \label{renormalized-NLP} \Sigma^{(1;\rm NLP)}_{\rm Ren}(Q)&=\Sigma^{(1;\rm NLP)}_{m}+\Sigma^{(1;\rm NLP)}_{\xi}-\delta_2^{(1)}\cancel{Q}\nonumber\\ &=\frac{e^2\, }{16\pi^2}\Bigl\{-\xi\, \cancel{Q}\Bigl[2\ln\left(\frac{e^{\gamma_E}\bar{\Lambda}}{2\pi T}\right)-1+2q^0L(Q)\Bigr]+(1-\xi)\frac{Q^2}{q^2}\mathbf{q}.\boldsymbol{\gamma}\Bigl[1-q^0 L(Q)\Bigr]\Bigr\}. \end{align}\tag{15}\] The singularities cancel out, yielding a finite result. The second square bracket cancels in Feynman gauge (\(\xi=1\)), whereas the first square bracket cancels in Landau gauge (\(\xi=0\)). Our result agrees with Ref. [27]. For \(q^0>q\), the functions above are real, so the one-loop NLP gives no damping contribution.

3.2 Two-loop organization before the HTL expansion↩︎

At two loops, three distinct topologies contribute to the fermion self-energy: the rainbow, cross-photon, and bubble diagrams, as shown in Figs. 2 and 6. At this stage, we focus on the rainbow diagram to illustrate the \(r/a\) assignment using the method explained in Sec. 2. We choose two lines to be symmetric propagators such that, after separating them from the main diagram, the remaining graph is connected and tree-like (Fig. [choosing-symmetric32lines]). We then assign the remaining \(r/a\) labels to determine the causality flow (Figs. [tree-like32graph]–[Completed32r47a32diagram]). Momenta are omitted in the \(r/a\) assignment figures since the corresponding Feynman diagram already fixes them. The full set of \(r/a\) diagrams is obtained by summing over all possible choices of symmetric propagators and completing the resulting tree-like graphs with \(r/a\) labels, as represented in Fig. 5. The remaining Feynman diagrams are treated in the same way.

Using Eq. 2 and the propagator conventions of Fig. 1, this \(r/a\) assignment gives \[\begin{align} \label{represetive} -i\Sigma_{{R,{\rm Rep}}}=-ie^4 \int_P\int_L &\Delta_s(PLQ)\Delta_R^{(2)}(PQ) \Bigl\{ \mathcal{S}_{R_{\rm Ph}}\, \tilde{\Delta}_s(P)\Delta_A(L) -(1-\xi)^2 \mathcal{S}_{R_{\xi^2}}\, \tilde{\Delta}_s^{(2)}(P)\Delta_A^{(2)}(L) \nonumber\\ & +i(1-\xi) \Bigl[ \mathcal{S}_{R_{\xi_L}}\tilde{\Delta}_s(P)\Delta_A^{(2)}(L) +\mathcal{S}_{R_{\xi_P}}\tilde{\Delta}_s^{(2)}(P)\Delta_A(L) \Bigr] \Bigr\}, \end{align}\tag{16}\] where \[\begin{align} \label{Dirac32rainbow} &\mathcal{S}_{R_{m}}\equiv \gamma^\alpha(\cancel{P}+\cancel{Q}) \gamma^\nu(\cancel{P}+\cancel{L}+\cancel{Q}) \gamma_\nu(\cancel{P}+\cancel{Q}) \gamma_\alpha ,\quad \mathcal{S}_{R_{\xi^2}}\equiv \cancel{P}(\cancel{P}+\cancel{Q}) \cancel{L}(\cancel{P}+\cancel{L}+\cancel{Q}) \cancel{L}(\cancel{P}+\cancel{Q}) \cancel{P}, \nonumber\\ &\mathcal{S}_{R_{\xi_L}}\equiv \gamma^\alpha(\cancel{P}+\cancel{Q}) \cancel{L}(\cancel{P}+\cancel{L}+\cancel{Q}) \cancel{L}(\cancel{P}+\cancel{Q}) \gamma_\alpha ,\quad \mathcal{S}_{R_{\xi_P}}\equiv \cancel{P}(\cancel{P}+\cancel{Q}) \gamma^\nu(\cancel{P}+\cancel{L}+\cancel{Q}) \gamma_\nu(\cancel{P}+\cancel{Q}) \cancel{P}. \end{align}\tag{17}\] Here, \(R\) denotes the rainbow topology, and the compact momentum notation introduced in Tab. 1 is used. Three gauge sectors contribute in Eq. 16 : the purely metric sector, the mixed metric-longitudinal sector proportional to \((1-\xi)\), and the purely longitudinal sector proportional to \((1-\xi)^2\). The metric sector originates from the \(g^{\mu\nu}\) part of the gauge-boson propagators; the mixed metric–longitudinal sector originates from the multiplication \(g^{\mu\nu}\) and the \((1-\xi)\)-dependent longitudinal structure; the purely longitudinal originates from the multiplication of \((1-\xi)\)-dependent parts of both internal photon propagators. In the mixed sector, there are two inequivalent metric-longitudinal contractions, leading to the two Dirac structures \(\mathcal{S}_{R_{\xi_L}}\) and \(\mathcal{S}_{R_{\xi_P}}\).

The symmetric propagators in the curly brackets of Eq. 16 contain spectral and difference functions, namely \(\Delta_d(P)\) and \(\Delta_d^{(2)}(P)\). The difference propagator can be reduced to a spectral function, together with derivatives acting on the remaining integrand, by using Eq. 58 . The spectral function then sets the corresponding momentum on shell, \(P^2=0\), as in Eq. 59 . For this representative assignment, the other internal momentum can be put on shell, \(L^2=0\), by shifting \(L\to L-P-Q\), so that the other symmetric propagator depends only on \(L\). At two loops, however, the computation is not that simple. Products such as \(\Delta_R(PQ)\Delta_A(PQ)\) give rise to pinch structures in some topologies. These pinch contributions are removed by using the KMS relation, Eq. 60 , after summing the relevant \(r/a\) assignments. Thus, before the HTL expansion, we perform two reorganizations. First, we rewrite the full set of \(r/a\) contributions in a pinch-free form. Second, we shift the loop momenta so that each spectral or difference propagator depends only on one internal momentum, either \(P\) or \(L\). The difference propagators are then reduced to spectral functions using Eq. 58 . This sequence allows us to impose the on-shell condition before the soft expansion. The detailed implementation of these steps is deferred to App. 11; see also Ref. [31] for a more pedagogical discussion.

From Fig. [Rainbow-original], the different possible choices of symmetric propagators correspond to choosing two photon lines, one photon line and one fermion line, or two fermion lines. At two loops, the corresponding statistical factors therefore fall into the three independent products \[\begin{align} \label{main32statistical32sectors} N_F(L)N_B(P),\quad N_B(L)N_B(P),\quad N_F(L)N_F(P). \end{align}\tag{18}\]

The sector decomposition displayed in Eqs. 16 –17 is encoded in a compact \((i,j)\) notation, introduced in App. 11. This notation is used only for bookkeeping of the pre-HTL expressions: it rewrites the metric, mixed metric–longitudinal, and fully longitudinal gauge sectors into a single equation. In this convention, the purely metric sector corresponds to \((i,j)=(1,1)\), the mixed sector to the off-diagonal sum \((1,2)+(2,1)\), and the fully longitudinal sector to \((2,2)\). The representative assignment in Eq. 16 then takes the compact form \[\begin{align} \label{representative32compact} \Sigma_{R,\mathrm{Rep},ij} = e^4 \int_P\int_L \hat{\xi}_{ij} \Delta_s(PLQ)\Delta_R^{(2)}(PQ) \tilde{\Delta}^{(i)}_s(P)\Delta_A^{(j)}(L) \mathcal{S}_{R,ij}, \end{align}\tag{19}\] where \(\mathcal{S}_{R,ij}\) and \(\hat{\xi}_{ij}\) are defined in Eqs. 105 and 103 , respectively.

The final pre-HTL expressions used for the soft expansion are given in Eqs. 110 and 112 . Although the self-energy and vertex pre-HTL expressions in Apps. 11–12 use this compact notation, the HTL-expanded integrands and the integrated results are written in the sector notation introduced in Sec. 2. Thus, the compact notation is auxiliary: it organizes the intermediate algebra, while the final results are read directly in terms of the metric, mixed, and longitudinal gauge sectors. Readers interested only in the final results may safely skip this bookkeeping.

3.3 Two-Loop Results↩︎

We apply the soft expansion to the pinch-free expressions collected in Eqs. 110 and 112 . The LP metric and mixed metric–longitudinal contributions are given in Eqs. 123 and 124 , respectively. These integrand-level results provide the self-energy side of the WTI and allow the identity to be checked separately in each gauge sector and for each independent statistical sector. The metric self-energy further separates into independent \(\mathcal{D}^2\), \(\mathcal{D}^1\), and \(\mathcal{D}^0\) dimensional parts, with each part satisfying the WTI independently. At this order, the fully longitudinal \((1-\xi)^2\) sector vanishes by integrand symmetry, in analogy with the corresponding LP cancellation at one loop. The relevant terms are odd under \(P\to -P\) and/or \(L\to -L\) and therefore integrate to zero. Consequently, the three fully longitudinal statistical sectors vanish at LP, while the six non-vanishing sectors are verified explicitly: three metric sectors and three mixed metric–longitudinal sectors. At NLP, however, the fully longitudinal sector no longer vanishes and contributes to the self-energy. A representative example, together with its Ward–Takahashi verification, is provided in the ancillary file.

As shown in Eq. 123 , the metric sector contains collinear singularities arising from the relative angle between the two hard momenta: \[\begin{align} \frac{1}{L\cdot P} = \frac{1}{p l}\,\frac{1}{1-z}, \qquad z\equiv \hat{\mathbf{p}}\!\cdot\!\hat{\mathbf{l}}, \end{align}\] with the collinear region corresponding to \(z\to1\). The required collinear moments are collected in Eqs. 54 –55 . Using these identities, the coupled angular dependence between the two hard directions reduces to a combination of ordinary HTL angular moments and the collinear moment. For example, \[\int_{v}\int_{v'}\frac{\cancel{v}\,(v^{\prime}\!\cdot Q)}{(v\!\cdot v^{\prime})\,(v\!\cdot Q)^2} = q^0\,\mathcal{A}_0\,\cancel{\mathcal{A}}_2 +(\mathcal{C}_1-\mathcal{A}_0)\,\cancel{\mathcal{A}}_1, \label{eq:example-collinear-reduction}\tag{20}\] where Eqs. 52 and 55 have been used. Here, \(\cancel{\mathcal{A}}_\alpha\equiv \mathcal{A}_\alpha\gamma^0-\mathcal{A}_\alpha^{\,i}\gamma_i\), and \(\mathcal{C}_1\) denotes the collinear moment.

Since the collinear terms originate from the angular integrations, an overlap between radial and collinear singularities could occur. To obtain the integrated results, we use Eq. 50 for the radial integrals, Eqs. 52 and 53 for the non-collinear angular integrals, and Eqs. 54 –55 for the collinear moments. This yields \[\tag{21} \begin{align} &\Sigma^{(2;\rm LP)}_{m}(Q)=\frac{e^4}{4}\mathcal{N}^2\nonumber\\ &\times\biggl\{\mathcal{A}_0\Bigl\{\mathcal{D}^2 (\mathcal{R}_1+\mathcal{R}_2) \Bigl[\mathcal{R}_3 \left(-\mathcal{A}_1\gamma_0+\mathcal{A}_2\cancel{Q}-Q^2 \cancel{\mathcal{A}}_3+\cancel{\mathcal{A}}_1\right)-Q^2 \mathcal{R}_4 \cancel{\mathcal{A}}_3+q^0 \mathcal{R}_4 \cancel{\mathcal{A}}_2-\mathcal{R}_6 \cancel{\mathcal{A}}_1\Bigr]\nonumber\\ &+4 \mathcal{D} \mathcal{R}_2\Bigl[-\mathcal{R}_3 \bigl[-\mathcal{A}_1\gamma_0+\mathcal{A}_2\cancel{Q}-Q^2 \cancel{\mathcal{A}}_3+q^0 \cancel{\mathcal{A}}_2+\cancel{\mathcal{A}}_1\bigr]+Q^2 \mathcal{R}_4 \cancel{\mathcal{A}}_3+\mathcal{R}_5 \cancel{\mathcal{A}}_1\Bigr]\nonumber\\ &+8 \mathcal{R}_2 (\mathcal{R}_3-\mathcal{R}_4) \bigl[-\mathcal{A}_1\gamma_0+\mathcal{A}_2\cancel{Q}-q^0 \cancel{\mathcal{A}}_2+2 \cancel{\mathcal{A}}_1\bigr]\Bigr\}+2\mathcal{D} Q^2 \mathcal{A}_2 \cancel{\mathcal{A}}_1 (\mathcal{R}_1+\mathcal{R}_2) (\mathcal{R}_3+\mathcal{R}_4)\nonumber\\ &\quad+\mathcal{C}_1 \cancel{\mathcal{A}}_1 (\mathcal{R}_4-\mathcal{R}_3) \bigl[\mathcal{D}^2 (\mathcal{R}_1+\mathcal{R}_2)+16 \mathcal{R}_2\bigr]\biggr\},\tag{22}\\ &\Sigma^{(2;\rm LP)}_{\xi}(Q)=-\frac{e^4}{4}(1-\xi)\mathcal{D}\mathcal{N}^2(\mathcal{R}_1 +\mathcal{R}_2 )\nonumber\\ &\times\Bigl\{\mathcal{A}_0 \Bigl[-2 \mathcal{A}_2 \mathcal{R}_4 \cancel{Q}-\mathcal{R}_5 \cancel{\mathcal{A}}_1+\mathcal{R}_3 \bigl[-\mathcal{A}_1 \gamma_0+2 \mathcal{A}_2 \cancel{Q}-2 Q^2 \cancel{\mathcal{A}}_3+q^0 \cancel{\mathcal{A}}_2+\cancel{\mathcal{A}}_1\bigr]\Bigr]+\mathcal{A}_2 Q^2 \mathcal{R}_4 \cancel{\mathcal{A}}_1\nonumber\\ &\quad+\cancel{Q}\Bigl[\mathcal{A}^j_1\mathcal{A}^j_1 (\mathcal{R}_3 -\mathcal{R}_5 )+\mathcal{A}_1^j\mathcal{A}_2^j q^0 \mathcal{R}_3 -\mathcal{A}^j_1\mathcal{A}^j_3 Q^2 (\mathcal{R}_3 +\mathcal{R}_4 )+\mathcal{A}_1 \big[\mathcal{A}_1 \mathcal{R}_5 -\mathcal{A}_2 q^0 \mathcal{R}_3 +\mathcal{A}_3 Q^2 (\mathcal{R}_3 +\mathcal{R}_4 )\big]\Bigr]\Bigr\}.\tag{23} \end{align}\] where \(\mathcal{N}\) is given in Eq. 49 . Radial singularities arise only through \(\mathcal{R}_3\) and \(\mathcal{R}_4\), whose singular parts are identical. Hence, the difference \(\mathcal{R}_4-\mathcal{R}_3\) is free of radial singularities and leaves a finite remainder proportional to \(\log (2)\), with further \(O(\epsilon)\) contributions. In the last term of Eq. 22 , this radial difference multiplies the collinear moment \(\mathcal{C}_1\). Therefore, the resulting contribution contains a collinear pole whose coefficient is proportional to \(\log(2)\). This makes the collinear singularity distinguishable from the purely radial singularities; in particular, this term does not generate a radial–collinear double pole.

We expand both the angular and radial integrals through \(O(\epsilon)\). This is necessary for two reasons. First, in the collinear contributions, the combination \(\mathcal{R}_3-\mathcal{R}_4\) cancels the \(\log(2\pi T)\) term at \(O(\epsilon^0)\); see App. 8. Since the scale dependence must enter through the dimensionless ratio of the renormalization scale \(\bar{\Lambda}\) to the thermal scale \(2\pi T\), the radial expansions must be kept through \(O(\epsilon)\). Only then do they reproduce the correct scale-dependent logarithm, \(\log(e^{\gamma_E}\bar{\Lambda}/2\pi T)\). Second, products of radial and angular integrals can generate finite contributions through \(O(\epsilon)\times 1/\epsilon\). In particular, the poles in \(\mathcal{R}_{3,4}\) and in \(\mathcal{C}_1\) can multiply the \(O(\epsilon)\) parts of the remaining radial integrals or of the angular moments, producing finite terms. Thus, both the radial and angular integrals must be kept through \(O(\epsilon)\) to obtain the complete finite part. We also keep the calculation explicitly \(D\)-dimensional throughout, because \(D\)-dependent prefactors, such as \(\mathcal{D}=D-2=2-2\epsilon\), carry \(O(\epsilon)\) corrections that multiply \(1/\epsilon\) poles and generate finite contributions.

Finally, the integrated results become, \[\label{self-energy-integrated} \begin{align} \Sigma_m^{(2;\rm LP)}(Q) &= \frac{e^2 \mathcal{H}(T)}{2\pi^2} \Bigl[ \mathcal{K}-\frac{7}{3} +\frac{1}{2}\log\Bigl(\!-\frac{q^2}{Q^2}\Bigr)\bigl[1-\frac{7}{3}\log(2)\bigr] -\frac{7\mathcal{K}}{3}\log (2) +\frac{17}{6}\log^2 (2)\Bigr]\nonumber \\ &\quad +\frac{e^4T^2}{16\pi^2} \Bigl\{ \gamma^0 \Bigl[ q_0L(Q)^2 + \frac{q_0}{Q^2}\bigl[ -\frac{1}{6}+\frac{2}{3}\log (2)\bigr] + \frac{\Delta_2(Q)}{4q} \bigl[1-\frac{7}{3}\log(2)\bigr] + L(Q) \bigl[\frac{5}{6}+\frac{7}{3}\log(2)\bigr] \Bigr]\nonumber \\ & + \frac{\boldsymbol{\gamma}.\hat{\mathbf{q}}}{q} \Bigl[ 2 -\frac{1}{2}\log\Bigl(\!-\frac{q^2}{Q^2}\Bigr) + \log(2)\bigl[ -\frac{10}{3} +\frac{7}{6}\log\Bigl(\!-\frac{q^2}{Q^2}\Bigr)\bigr] + \frac{7}{3}\log^2(2)+ \frac{q^2}{Q^2}\bigl[\frac{1}{6}-\frac{2}{3}\log(2)\bigr]\nonumber \\ &-\frac{q_0\Delta_2(Q)}{4q}\bigl[1-\frac{7}{3}\log(2)\bigr]-q_0^2L(Q)^2\Bigr]\Bigr\}, \end{align} \begin{align} \Sigma_\xi^{(2;\rm LP)}(Q) &= -\frac{e^2\mathcal{H}(T)}{8\pi^2}(1-\xi)\Bigl[ \mathcal{K}-\frac{7}{2} +\frac{2}{3}\log(2) +\frac{1}{2}\log\Bigl(\!-\frac{q^2}{Q^2}\Bigr)\Bigr]\nonumber \\ &\quad -\frac{e^4T^2}{64\pi^2}(1-\xi) \Bigl\{ \gamma^0 \Bigl[ \frac{q_0}{q^2} + \frac{\Delta_2(Q)}{4q} + q_0 \bigl[1+\frac{q_0^2}{q^2}\bigr] L(Q)^2 - \frac{2q_0\log(2)}{Q^2} - \frac{2Q^2}{q^2}L(Q) \Bigr]\nonumber \\ & + \frac{\boldsymbol{\gamma}.\hat{\mathbf{q}}}{q} \Bigl[ 3 -\log(2) -\frac{1}{2}\log\Bigl(\!-\frac{q^2}{Q^2}\Bigr) + \frac{2q^2\log 2}{Q^2} - \frac{q_0\Delta_2(Q)}{4q} - 2q_0^2L(Q)^2 \Bigr] \Bigr\}, \end{align}\tag{24}\] where \[\begin{align} \Delta_2(Q)\equiv\text{Li}_2\bigl(\frac{q^0+q}{{q^0}-q}\bigr)-\text{Li}_2\bigl(\frac{q^0-q}{{q^0}+q}\bigr),\quad \mathcal{K}\equiv -\gamma_E+\frac{1}{2\epsilon}+12\log(A)+2\log\!\Bigl(\frac{e^{\gamma_E}\bar{\Lambda}}{2\pi T}\Bigr), \end{align}\] where \(A\) is the Glaisher–Kinkelin constant, and \(\mathcal{H}(T)\) denotes the one-loop LP self-energy in Eq. 11 . All singular terms in the contributions above are proportional to \(\mathcal{H}(T)\) and are collected in \(\mathcal{K}\). The poles proportional to \(\log (2)\) identify the collinear part.

Let us now evaluate the imaginary part of the corrections above on the timelike branch: \(q^0>q\). With the retarded prescription \(q^0\to q^0+i0^+\), the relevant functions have \[\begin{align} \Im\Bigl[ L(Q)\Bigr]=0,\quad \Im\Bigl[\Delta_2(Q)\Bigr]=-2\pi q\,L(Q),\quad \Im\Bigl[\log\!\left(-\frac{q^2}{Q^2}\right)\Bigr]=\pi . \end{align}\] Using these relations, the imaginary parts of the metric and mixed metric–longitudinal sectors cancel separately: \[\begin{align} \Im\Sigma^{(2;{\rm LP})}_{m}(Q) = \Im\Sigma^{(2;{\rm LP})}_{\xi}(Q) =0, \qquad q^0>q . \end{align}\] Thus, on the timelike branch, the hard two-loop LP corrections computed here do not generate a finite damping contribution. This behavior is consistent with the corresponding timelike hard corrections to the photon polarization tensor [31].

At the same parametric order, zero-temperature counterterm insertions must also be included to remove the UV subdivergences. These insertions with the one-loop self-energies are collected in Eq. 118 , with the corresponding LP corrections given in Eq. 119 . The sum of these one-loop LP counterterms then becomes \[\begin{align} \label{self-summing-cts} \Sigma^{(1;LP)}_{{\rm CT}}(Q) = \left(\delta_2^{(1)}+\delta_3^{(1)}\right)\mathcal{H}(T)=\Bigl(-\frac{e^2}{16\pi^2}\frac{\xi}{\epsilon} -\frac{e^2}{12\pi^2}\frac{1}{\epsilon}\Bigr)\mathcal{H}(T). \end{align}\tag{25}\] The renormalized hard-region contribution is therefore \[\begin{align} \label{renormalized-two-loop-LP} \Sigma_{{\rm ren}}^{(2;{\rm LP})}(Q) = \Sigma^{(2;{\rm LP})}(Q) + \Sigma_{{\rm CT}}^{(1;LP)}(Q). \end{align}\tag{26}\] Using Eqs. 24 and 25 , the divergences take the form \[\begin{align} \label{divergences32in32hard32sector} \Sigma_{\rm Div}(Q)=\frac{e^2}{48\pi^2\epsilon}\Bigl(5-28\log(2)\Bigr)\mathcal{H}(T) \end{align}\tag{27}\] The \(\xi\)-dependent pole is removed by the zero-temperature counterterm insertion. After this subtraction, the remaining singularities are gauge-independent collinear and hard-region IR poles. The hard-region IR pole refers to the radial endpoint singularity of the hard expansion, where the expanded hard integrals overlap with softer momentum regions. In the following sections, we explain how these singular structures enter the matching at full parametric order \(e^3T\). The final physical requirement is that, after all contributions at the same order are included, observables contain no uncancelled singularities or gauge-dependent structures.

4 Two-Fermion–N-Photon Vertex↩︎

In this section, we first summarize the symmetry constraints on the amputated two-fermion–\(N\)-photon vertex kernel; the details and derivations are given in App. 9. We suppress the explicit \(r/a\) prescriptions. Chiral invariance restricts the most general Dirac structure of the amputated two-fermion–\(N\)-photon kernel, with fermion momenta \(Q_{1,2}\), to \[\begin{align} \label{general32results32of32n32kernel} \Gamma^{(N)}(Q_2,Q_1;\{K_i\}) &= V^{(N)}(Q_2,Q_1;\{K_i\})\cdot \gamma + A^{(N)}(Q_2,Q_1;\{K_i\})\cdot \gamma\,\gamma_5, \end{align}\tag{28}\] where all photon momenta are taken incoming. The \(N-1\) independent photon momenta are denoted by \(\{K_i\}\equiv \{K_1,\dots,K_{N-1}\}\), while the remaining momentum \(K_N\) is fixed by momentum conservation, \(Q_2 = Q_1 + \sum_{i=1}^N K_i\). For the \(c\)-number tensors \(V^{(N)}\) and \(A^{(N)}\), Lorentz indices are suppressed, and “\(\cdot\gamma\)” denotes contraction with the corresponding Dirac matrices. The explicit index form is given in App. 9.

Under parity invariance, \(V^{(N)}\) and \(A^{(N)}\) transform as a rank-\((N+1)\) tensor and a rank-\((N+1)\) pseudotensor, respectively; see Eq. 81 . The \(P\mathcal{T}\) symmetry¹ determines the reality properties \[\label{reality-constraints} V^{(N)\,*}=V^{(N)},\qquad A^{(N)\,*}=-A^{(N)}.\tag{29}\] Hence, the vector and axial coefficients are real and imaginary, respectively. Together with hermiticity, this implies the fermion-leg exchange properties \[\begin{align} V^{(N)}(Q_2,Q_1;\{K_i\}) &= V^{(N)}(Q_1,Q_2;\{-K_i\}), \nonumber\\ A^{(N)}(Q_2,Q_1;\{K_i\}) &= - A^{(N)}(Q_1,Q_2;\{-K_i\}). \label{eq:exchangeSummary} \end{align}\tag{30}\] Thus, under the exchange \(Q_1\leftrightarrow Q_2\), the axial coefficient receives a minus sign and all independent photon momenta change sign, \(\{K_i\}\to\{-K_i\}\). Charge conjugation further gives \[\begin{align} V^{(N)}(Q_2,Q_1;\{K_i\}) &= (-1)^{N+1} V^{(N)}(-Q_1,-Q_2;\{K_i\}), \nonumber\\ A^{(N)}(Q_2,Q_1;\{K_i\}) &= (-1)^N A^{(N)}(-Q_1,-Q_2;\{K_i\}). \label{eq:Csummary} \end{align}\tag{31}\] Using Eqs. 30 and 31 , one obtains \[\begin{align} \label{combined-exC} X^{(N)}(Q_2,Q_1;\{K_i\}) = (-1)^{N+1} X^{(N)}(-Q_2,-Q_1;\{-K_i\}), \end{align}\tag{32}\] where \(X^{(N)}\) denotes either \(V^{(N)}\) or \(A^{(N)}\). To obtain the corresponding selection rule, we rescale all soft external momenta as \(Q_{1,2}\to a\, Q_{1,2}\) and \(\{K_i\}\to \{a K_i\}\). Hence, \[\begin{align} \label{expanding} X^{(N)}(aQ_2,aQ_1;\{aK_i\}) = \sum_{r\in\mathbb{Z}}a^r X^{(N)}_r(Q_2,Q_1;\{K_i\}). \end{align}\tag{33}\] On the other hand, using the rescaled form of Eq. 32 gives \[\begin{align} \label{combined-expanding} X^{(N)}(aQ_2,aQ_1;\{aK_i\}) &= \sum_{r\in\mathbb{Z}}(-1)^{N+1+r}a^r X^{(N)}_r(Q_2,Q_1;\{K_i\}). \end{align}\tag{34}\] Matching powers of \(a\) leads to \[\begin{align} \left[1-(-1)^{N+1+r}\right]X^{(N)}_r(Q_2,Q_1;\{K_i\})=0. \end{align}\] Therefore, the contribution at order \(a^r\) can be nonzero only if \[\begin{align} r+N+1 \quad \text{is even}. \end{align}\] Thus, for odd \(N\), only even soft powers can be nonzero, while for even \(N\), only odd soft powers can be nonzero. For example, the fermion–photon vertex, \(N=1\), can receive contributions from even powers of the soft expansion: \(\{a^{-2},a^0,\ldots\}\). For \(N=0\), the condition \(r+N+1\) even reduces to \(r+1\) even, so only odd powers contribute, reproducing the self-energy selection rule.

As an example, consider the two-fermion–two-photon vertex \(\Gamma^{\mu\nu}(Q_2,Q_1;K_1)\), where \(K_1\) is an independent incoming photon momentum. The one-loop LP HTL contribution is [26] \[\begin{align} \Gamma^{\mu\nu}(Q_2,Q_1;K_1) &= -m^2(T)\int\frac{d\Omega}{4\pi}\, \frac{{v}^\mu\,{v}^\nu\, \cancel{v}}{\big[(Q_1+K_1)\!\cdot\! v\big]\big[(Q_2-K_1)\!\cdot\! v\big]} \Bigl[\frac{1}{Q_1\!\cdot\!v}+\frac{1}{Q_2\!\cdot\!v}\Bigr], \end{align}\] with \(v^\mu=(1,\hat{v})\) and the integral over \(\hat{v}\). This vector-like result satisfies the reality, exchange, and charge-conjugation constraints given in Eqs. 29 , 30 , and 31 , respectively. The axial coefficient does not contribute at this power order. This expression scales as \(a^{-3}\) under the soft rescaling, consistent with the selection rule for \(N=2\).

Note that Levi–Civita tensor structures are allowed in the axial coefficients. For a fermion–photon vertex, this implies \[\label{Levi-Civita-structure} \epsilon^{\mu\nu\rho\sigma} Q_{1\rho} Q_{2\sigma}\,\gamma_\nu\gamma_5 = -i\Big( \gamma^\mu\,\cancel{Q}_1\,\cancel{Q}_2 - Q_1^\mu\,\cancel{Q}_2 + Q_2^\mu\,\cancel{Q}_1 - (Q_1\!\cdot\!Q_2)\,\gamma^\mu \Big).\tag{35}\] In vacuum, such axial contributions in on-shell matrix elements \(\bar u(Q_2)\Gamma_\mu(Q_2, Q_1)u(Q_1)\) can be reduced to vector structures by using the on-shell Dirac equations (see e.g.Ch. 10 of Ref. [61]) \[\cancel{Q}\,u(Q)=m\,u(Q),\qquad \bar u(Q)\cancel{Q}=m\,\bar u(Q),\] so that for \(m=0\) the right-hand sides vanish. Equivalently, the renormalized on-shell external states satisfy the free Dirac equation (up to an overall wavefunction renormalization), which enforces this on-shell reduction. At finite temperature, however, quasiparticle dispersion relations and spinors are modified [62]–[65], so this reduction need not hold. We therefore keep the full off-shell decomposition to avoid missing contributions relevant for subsequent analysis.

5 Fermion–Photon Vertex↩︎

We use the vertex component \(\Gamma^\mu_{arr}(Q_2,Q_1)\), where the three \(r/a\) indices are ordered, from left to right, as the incoming fermion with momentum \(Q_1\), the photon with momentum \(K=Q_2-Q_1\), and the outgoing fermion with momentum \(Q_2\). Within our conventions, the WTI reads [39] \[\begin{align} \label{WT32identity} K_\mu \Gamma_{arr}^\mu(Q_2,Q_1)=\Sigma_{ar}(Q_1)-\Sigma_{ar}(Q_2), \end{align}\tag{36}\] with the causal prescription imposed at the end, \(Q_{1,2}\rightarrow Q_{1,2}+i\eta_{1,2}\). Other vertex components differ by the advanced/retarded assignment of the external legs (see Tab. 1)  [38]–[41]. Here, we drop the \(r/a\) indices: \(\Gamma^\mu\equiv \Gamma^\mu_{arr}\). Chiral invariance restricts the vertex correction to \[\begin{align} \label{vertex32decomposition} \Gamma^\mu(Q_2,Q_1)= V^{\mu\nu}\gamma_\nu+ A^{\mu\nu}\gamma_{\nu}\gamma_5, \end{align}\tag{37}\] with the coefficients given by \[\begin{align} \label{Eq:vertex32projections} V^{\mu \nu}= \frac{1}{4}\, \mathrm{Tr}\!\left[\gamma^\nu\, \Gamma^\mu(Q_2,Q_1)\right],\quad A^{\mu \nu}= -\frac{1}{4}\, \mathrm{Tr}\!\left[\gamma^\nu\gamma^5\, \Gamma^\mu(Q_2,Q_1)\right]. \end{align}\tag{38}\] The temporal and spatial components are read directly from these covariant expressions. The symmetry constraints follow from the \(N=1\) case of the general constraints discussed above. Contracting Eq. 38 with the photon momentum gives \[\begin{align} K_\mu A^{\mu\nu} &\propto \mathrm{Tr}\!\left[\gamma^\nu\gamma^5\gamma^\rho\right] =0, \end{align}\] where Eqs. 38 , 36 and 8 are used. Thus, \(A^{\mu\nu}\) is transverse to the photon momentum. The WTI fixes the longitudinal projection of the vector coefficient, \(K_\mu V^{\mu\nu}\), via the self-energy difference. Below, we compute the one- and two-loop contributions to the fermion–photon vertex and explicitly verify these constraints.

5.1 One-Loop LP and NLP Corrections↩︎

The one-loop fermion–photon diagram is shown in Fig. 4, and its corresponding expression is given in Eq. 100 . We perform loop-momentum shifts to write the result as \[\label{Eq:one-loop32vertex} \begin{align} &-ie\Gamma^{\mu(1)}_{m}(Q_2,Q_1)=-ie^3\int_P \Delta_d(P)\Bigl[N_F(P)\Delta_A(P-Q_1)\Delta_A(P+Q_2-Q_1)\mathcal{V}_{{{m}(a_v)}}+N_B(P)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2)\mathcal{V}_{m}\nonumber\\ &+N_F(P)\Delta_{R}(P+Q_1-Q_2)\Delta_A(P-Q_2)\mathcal{V}_{m{(m_v)}}\Bigr],\\ &-ie\Gamma_\xi^{\mu(1)}(Q_2,Q_1)=e^3\int_P(1-\xi)\Bigl[\Delta_d(P)N_F(P)\Delta_{A}^{(2)}(P-Q_1)\Delta_A(P+Q_2-Q_1)\mathcal{V}_{{{\xi}}{(a_v)}}\nonumber\\ &+\Delta_d^{(2)}(P)N_B(P)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2)\mathcal{V}_{\xi}\nonumber\\ &+\Delta_d(P)N_F(P)\Delta_{R}(P+Q_1-Q_2)\Delta_{A}^{(2)}(P-Q_2)\mathcal{V}_{{\xi}{(m_v)}}\Big], \end{align}\tag{39}\] where \[\begin{align} \label{Eq:one-loopDirac} \mathcal{V}_{{m}}\equiv\gamma^\nu (\cancel{P}+\cancel{Q}_1) \gamma ^\mu (\cancel{P}+\cancel{Q}_2)\gamma_\nu,\qquad \mathcal{V}_{\xi}\equiv\cancel{P} (\cancel{P}+\cancel{Q}_1) \gamma ^\mu (\cancel{P}+\cancel{Q}_2)\cancel{P}. \end{align}\tag{40}\] Here, the labels \(a_v:P\to P-Q_1\) and \(m_v:P\to P-Q_2\) indicate that the corresponding shift is applied to the Dirac numerator. To perform the power expansion in the soft external momenta, we use \[\begin{align} K =Q_2-Q_1 , \qquad Q_1\to aQ_1,\quad Q_2\to aQ_2,\quad K\to aK , \end{align}\] and rewrite all denominators containing \(Q_2-Q_1\) in terms of \(K\), together with the corresponding numerator factors. For the metric Fermi–Dirac sector, this amounts to the replacement \[\begin{align} N_F(P)\Bigl\{\Delta_A(P-Q_1)\Delta_A(P+K)\left[\gamma^\nu \cancel{P} \gamma^\mu (\cancel{P}+\cancel{K})\gamma_\nu\right]+\Delta_{R}(P-K)\Delta_A(P-Q_2)\left[\gamma^\nu (\cancel{P}-\cancel{K}) \gamma^\mu \cancel{P}\gamma_\nu\right]\Bigr\}. \end{align}\] The leading non-vanishing contribution scales as \(a^{-2}\) and lies in the vector coefficient, while the axial coefficient vanishes by integrand symmetry. Hence, the LP result becomes \[\begin{align} \label{LP-Vertex} -ie\,\Gamma^{\mu(1;\rm LP)}(Q_2,Q_1) &= -\frac{ie^3\mathcal{D}}{4} \int_P \Delta_d(P)\, \bigl[N_B(P)-N_F(P)\bigr]\, \frac{2P^\mu\cancel{P}}{(P\cdot Q_1)(P\cdot Q_2)}, \end{align}\tag{41}\] which satisfies the WTI upon comparison with Eqs. 95 and 96 . We use the same prescription in the remaining cases: denominators involving a single external momentum are left unchanged, whereas those involving linear combinations of external momenta are rewritten in terms of the momentum transfer, with the corresponding numerator factors transformed in the same way; see App. 12 for a two-loop example.²

The NLP corrections arise at order \(a^0\). At this order, both vector and axial coefficients receive contributions from both metric and longitudinal sectors, as given in Eqs. 125 and 128 . The different gauge and statistical sectors satisfy these constraints independently: \[\begin{align} K_\mu \Gamma^{\mu(1;\rm NLP)}_{g,s}(Q_2,Q_1) &= \Sigma^{(1;\rm NLP)}_{g,s}(Q_1) - \Sigma^{(1;\rm NLP)}_{g,s}(Q_2), \quad g\in\{m,\xi\},\quad s\in\{N_B(P),N_F(P)\}, \end{align}\] with \[\begin{align} K_\mu A^{\mu\nu(1;\rm NLP)}_{g,s}(Q_2,Q_1)=0 . \end{align}\] The ancillary file provides the Mathematica expressions together with the corresponding WTI and axial-transversality checks. Notably, the axial coefficients carry an explicit imaginary factor, consistent with their symmetry properties. Using Eq. 35 , however, the corresponding vertex contributions can be rewritten in terms of Dirac-vector structures and products of Dirac slashes. In this representation, the explicit imaginary factor is absorbed, and the rewritten contributions are not purely imaginary. The renormalized vertex is also written as \[\begin{align} \Gamma^{\mu(1;NLP)}_{Ren}=\sum_{g,s}\Gamma_{g,s}^{\mu(1;NLP)}+\bar{\Lambda}^\epsilon \delta_1 \gamma^\mu. \end{align}\]

5.2 Two-Loop Corrections↩︎

At two-loop order, seven distinct topologies contribute to the fermion–photon vertex; see Fig. 7. The pinch-free pre-HTL expressions are collected in Eq. 116 , using the notation of Tab. 3. In these expressions, the spectral and higher-difference functions depend only on the internal momenta \(P\) or \(L\), allowing the on-shell conditions to be imposed before the soft expansion. We then apply the soft expansion introduced above.

The non-planar and self-energy-corrected topologies occur in two orientations, denoted type I and type II. At the level of the full amputated vertex, the type-II contribution is obtained from type I by the exchange \(Q_1\leftrightarrow Q_2\) [66], [67]. This interchange does not generally reproduce the exact intermediate \(r/a\) assignment obtained from a direct type-II construction. For example, a direct treatment of a type-II diagram may yield a propagator structure of the form \[\Delta_R(L-Q_1)\Delta_A(PQ_1)\Delta_A(PQ_2)\Delta_R(PL),\] whereas applying the exchange prescription to the corresponding type-I diagram may instead generate \[\Delta_A(L-Q_1)\Delta_R(PQ_1)\Delta_R(PQ_2)\Delta_A(PL).\] Thus, the two procedures need not coincide at the level of intermediate \(r/a\) assignments. However, they use the same momentum routing and the same assignment of symmetric propagators. Therefore, the resulting integrands are equivalent up to the retarded/advanced prescription. So, we use \(Q_1\leftrightarrow Q_2\) for the integrand and soft expansion. The retarded/advanced prescription is imposed for the integration: \(Q_{1,2}\to Q_{1,2}+i\eta_{1,2}\). After projection onto the vector and axial components in Eq. 37 , the exchange \(Q_1\leftrightarrow Q_2\) acts trivially in the vector sector. In contrast, the axial sector acquires the additional minus sign implied by Eq. 30 . Thus, the type-I to type-II map is implemented by \[Q_1\leftrightarrow Q_2, \qquad K\to -K,\] with the corresponding axial sign included.

At two loops, in contrast to the one-loop LP vertex, the axial coefficient contributes already at LP. Moreover, the mixed metric–longitudinal sector, proportional to \((1-\xi)\), also contributes, in parallel with the self-energy. The fully longitudinal sector, proportional to \((1-\xi)^2\), vanishes at LP by integrand symmetry, but appears at NLP and obeys the WTI and axial-transversality constraints. In the metric sector for the \(N_B(L)N_B(P)\) statistical sector, the LP results are presented in Eqs. 131 –132 . Because of the antisymmetry of the Levi–Civita tensor, the photon-momentum contraction of the axial Dirac structure, \(K_\mu A^{\mu\rho}\gamma_\rho\gamma_5\), vanishes identically for this configuration. In contrast, the vector Dirac structure reproduces the corresponding self-energy difference. The remaining sectors are collected in the ancillary file, and the identities are also checked sector by sector. In the bookkeeping used here, LP terms scale as \(a^{-2}\), NLP terms scale as \(a^0\), and all odd powers vanish, as required by the selection rule for the fermion–photon vertex. Tab. 2 summarizes the explicit self-energy and fermion–photon vertex relations verified in this study. As an additional check, we extended the algorithmic expansion to \(\mathcal{O}(a^{10})\) for the vector sector. At this order, the expansion initially generates \(\sim 4.9\times10^7\) terms, which reduce to roughly \(\mathcal{O}(10^5)\) terms after using integrand symmetries and contracting with \(K_\mu\). The WTI is preserved throughout this reduction: after contraction, all terms separate into \(Q_1\)-only and \(Q_2\)-only structures, and these isolated structures reproduce the corresponding self-energy difference.

Each two-loop vertex sector satisfies \[\begin{align} K_\mu\Gamma_{g,s}^{\mu,(2;r)}(Q_2,Q_1) &= \Sigma^{(2;r)}_{g,s}(Q_1) - \Sigma^{(2;r)}_{g,s}(Q_2),\quad K_\mu A^{\mu\nu,(2;r)}_{g,s}(Q_2,Q_1)=0, \quad s\in\{N_BN_B,\,N_FN_F,\,N_FN_B\}, \end{align}\] with \(g\in\{m,\xi\}\) at LP and \(g\in\{m,\xi,\xi^2\}\) at NLP and higher orders, since the fully longitudinal \((1-\xi)^2\) sector does not contribute at LP. This implies that each distinct sector satisfies the corresponding constraint independently.

The ancillary file associated with the two-loop vertex contains the complete LP vector and axial coefficients, and verifies the WTI and axial transversality sector by sector. It also demonstrates the preservation of the WTI through \(\mathcal{O}(a^{10})\) for the \(N_B(P)N_F(L)\) sector in the \((1-\xi)\) gauge sector.³ For the \(N_B(P)N_F(L)\) sector, it explicitly shows that the fully longitudinal \((1-\xi)^2\) sector vanishes at LP but contributes at NLP, while maintaining the WTI and axial transversality. The remaining statistical sectors follow the same algorithmic procedure. Unlike \(N_B(P)N_F(L)\), the \(N_F(L)N_F(P)\) and \(N_B(L)N_B(P)\) sectors are symmetric under \(L\leftrightarrow P\). For instance, their intermediate Dirac vector expressions can contain mixed \(Q_1\) and \(Q_2\) terms after contraction with \(K_\mu\). However, these terms vanish by integrand symmetries, leaving a final result depending only on \(Q_1\) or only on \(Q_2\), as shown for the LP cases in the ancillary file.

The one-loop counterterm insertions are given in Eq. 120 , and their LP contributions are collected in Eq. 121 . The renormalized LP vertex at two-loop order is therefore \[\begin{align} \Gamma^{\mu(2;LP)}_{\rm Ren}(Q_2,Q_1) = \sum_{g,s}\Gamma^{\mu(2;\mathrm{LP})}_{g,s}(Q_2,Q_1) + \Gamma^{\mu(1;LP)}_{\rm CT}(Q_2,Q_1), \end{align}\] where the sum runs over the two-loop gauge and statistical sectors listed in the ancillary file, and \[\begin{align} \label{CT-vertex-result} \Gamma^{\mu(1;LP)}_{\rm CT}(Q_2,Q_1) = \Bigl(\delta_2^{(1)}+\delta_3^{(1)}\Bigr) \Gamma^{\mu(1;LP)}(Q_2,Q_1). \end{align}\tag{42}\] Comparison with Eq. 25 shows that the counterterm insertions in the self-energy and in the vertex satisfy the WTI independently of the ordinary two-loop contributions. Together with the photon-sector analyses in Refs. [30], [31], the present fermionic self-energy and vertex results provide the hard-region building blocks needed for a gauge-consistent treatment of observables in a general covariant gauge. Taking the Abelian limit of the non-Abelian result gives an explicitly \(\xi\)-independent hard-region polarization tensor [29], confirming the gauge independence of the corresponding photon result.

6 Self-Energy Region Matching↩︎

We now summarize the region analysis for the fermion self-energy with soft external momentum⁴. Schematically, \[\Sigma(Q) = \Sigma^{(eT)}(Q) + \Sigma^{(e^2T)}(Q) + \Sigma^{(e^3T)}(Q) +\cdots . \label{Self-Energy-Hierarchy}\tag{43}\] The first term is the one-loop LP hard contribution, Eq. 11 , which enters the LO pole equation. The second term is the soft one-loop self-energy evaluated with LO HTL-resummed propagators and vertices [18]. This term already contains the leading tower with one soft loop and \(N\) LP hard insertions, with \(N=0,1,2,\ldots\) . At order \(e^3T\), the matched self-energy takes the form \[\begin{align} \Sigma^{(e^3T)}(Q) = \Sigma_{\rm Ren}^{(1,\mathrm{NLP})}(Q) + \Sigma_{\rm Ren}^{(2,\mathrm{LP})}(Q) + \Sigma_{N_BN_B}^{*(2)}(Q) - \Sigma_{\rm overlap}^{(e^3T)}(Q). \label{eq:e3T95self95energy95structure} \end{align}\tag{44}\] The first two terms are the hard one-loop NLP and hard-hard two-loop LP contributions, given in Eqs. 15 and 26 . The third term denotes the two-loop soft self-energy in the \(N_BN_B\) sector, with the internal lines and vertices replaced by their LO HTL-resummed counterparts. The LO HTL fermion–photon vertex is \(O(1)\) at the soft scale, as follows from Eq. 41 . For the representative topology with two symmetric photon propagators and three retarded/advanced fermion propagators, the power counting is \[d^4P\,d^4L\sim (eT)^8, \qquad N_B(P)\sim n_B(P)\sim \frac{T}{e T}\sim \frac{1}{e},\] and therefore \[\widetilde{D}_s^*(P) \sim N_B(P)\rho_B^*(P) \sim \frac{1}{e}\,\frac{1}{(eT)^2} \sim \frac{1}{e^3T^2}, \qquad S^*(P)\sim \frac{1}{eT},\] so the power counting becomes \[\Sigma_{N_BN_B}^{*(2)}(Q) \sim e^4 (eT)^8 \left(\frac{1}{e^3T^2}\right)^2 \left(\frac{1}{eT}\right)^3 \sim e^3T.\] The other soft statistical sectors are suppressed at this order. The overlap term is the asymptotic part of the hard and soft descriptions, \[\Sigma_{\rm overlap}^{(e^3T)} = \left(\Sigma_{\rm hard}\right)_{\rm soft\;limit} = \left(\Sigma_{\rm soft}\right)_{\rm hard\;limit}. \label{eq:overlap95definition}\tag{45}\] It is obtained by matching the IR limit of the hard-region result against the UV limit of the soft-sector calculation. In the present work, the two-loop soft self-energy and the corresponding overlap are not evaluated. Once these pieces are computed, the self-energy at order \(e^3T\) is complete within this matched expansion. Since the soft sector and overlap subtraction are not evaluated here, the precise cancellation channel of the hard-region singularities is not assigned at this stage. Pole observables are obtained from the zeros of the retarded inverse propagator. For a massless fermion self-energy, we write \[\Sigma_R(Q) = a(Q)\gamma^0 + b(Q)\,\hat{\mathbf{q}}\cdot\boldsymbol{\gamma} ,\] where \[\Sigma_s(Q)=a(Q)+s\,b(Q), \qquad D_s(Q)=q^0-sq-\Sigma_s(Q), \qquad s=\pm .\] The leading pole is obtained by \[D_s^{\rm LO}(\omega_s^{\rm LO},q)=0, \qquad D_s^{\rm LO}(Q) = q^0-sq-\Sigma_s^{(eT)}(Q),\] where \(\Sigma_s^{(eT)}(Q)\) is the helicity projection of the one-loop LP self-energy [26]. Expanding the self-energy and pole gives \[\Sigma_s = \Sigma_s^{(eT)} + \Sigma_s^{(e^2T)} + \Sigma_s^{(e^3T)} +\cdots , \qquad \omega_s = \omega_s^{\rm LO} + \delta\omega_s^{(e^2T)} + \delta\omega_s^{(e^3T)} +\cdots ,\] and therefore the NLO and NNLO pole shifts are \[\begin{align} &\delta\omega_s^{(e^2T)} =\frac{\Sigma_s^{(e^2T)}(\omega_s^{\rm LO},q)}{Z_s^{\rm LO}}, \quad Z_s^{\rm LO}=1-\left.\partial_{q^0}\Sigma_s^{(eT)}(q^0,q)\right|_{q^0=\omega_s^{\rm LO}}, \tag{46}\\ &\delta\omega_s^{(e^3T)} =\frac{\Sigma_s^{(e^3T)}+\delta\omega_s^{(e^2T)}\partial_{q^0}\Sigma_s^{(e^2T)} +\frac{1}{2}\left(\delta\omega_s^{(e^2T)}\right)^2\partial_{q^0}^2\Sigma_s^{(eT)}}{Z_s^{\rm LO}}\bigg|_{q^0=\omega_s^{\rm LO}}. \tag{47} \end{align}\] Thus, the nonlinear expansion of the pole equation also generates iteration terms at order \(e^3T\). Similar collinear structures have been reported in the leading hard–soft two-loop self-energy [17], which is already contained in the HTL-resummed one-loop soft self-energy. These structures can provide a natural source for the cancellation of the hard-region collinear divergence. Moreover, gauge-dependent hard-region self-energy contributions must cancel in the complete same-order pole equation, as required by gauge consistency.

For a complex pole, \[\omega_s(q)=\Omega_s(q)-i\gamma_s(q),\] the damping rate is \[\gamma_s^{(n)}(q) = -\Im\,\delta\omega_s^{(n)}(q).\] The imaginary part of the one-loop LP hard self-energy vanishes on the timelike branch, Eq. 11 , so the first nonzero damping contribution comes from the soft HTL-resummed sector [18]. The hard virtual corrections computed here do not generate a finite damping contribution on this branch. Therefore the two-soft-loop \(N_BN_B\) sector, together with the pole-iteration terms, is one of the same-order ingredients needed for the complete damping rate at NNLO.

7 Conclusion↩︎

We have derived the symmetry constraints of massless QED. Chiral invariance restricts the off-shell two-fermion–\(N\)-photon vertex to vector and axial-vector Dirac structures, each of which has definite behavior under momentum reversal and fermion-leg exchange. By omitting the retarded/advanced prescriptions, the vector structures are tensors with real amplitudes, while the axial structures are pseudotensors with purely imaginary amplitudes. The WTI constrains the vector structure of the fermion-photon vertex by relating it to the self-energies. On the other hand, the axial structure is transverse to the photon momentum. The self-energy is odd under momentum reversal up to retarded/advanced prescriptions, while the fermion spectral function is even. These symmetry constraints further lead to the selection rules for the soft expansions: at fixed power \(r\), a contribution can be nonzero only when \(r+N+1\) is even, with \(N=0\) corresponding to the fermion self-energy.

Using DR throughout, we have computed the renormalized one-loop NLP and two-loop LP contributions to the fermion self-energy and the fermion-photon vertex in a general covariant \(R_\xi\) gauge. The insertions of the zero-temperature counterterms remove the UV subdivergences. Axial-vector structures are absent in the one-loop LP vertex, but appear beyond that order, including in the two-loop LP vertex. The results are organized by gauge sector, statistical sector, and soft power. Each sector at fixed soft power is checked to preserve the WTI and axial transversality, as required by gauge consistency. The counterterm insertions also obey the WTI independently. Notably, the fully longitudinal sectors vanish at two-loop LP, consistent with the vanishing of the one-loop LP longitudinal sector. As an additional algorithmic test, the vector sector was extended through \(O(a^{10})\), for which the same selection rule and WTI persist. The ancillary file contains the machine-readable one-loop NLP and two-loop LP coefficients, along with the corresponding WTI and axial-transversality checks. In addition, for the fully longitudinal contribution in the \(N_B(P)N_F(L)\) statistical sector, the ancillary file shows that all LP contributions vanish. In contrast, they appear at NLP, where the vector and axial coefficients satisfy the WTI and transversality constraints, respectively.

For the self-energy, the hard contribution at one-loop NLP is gauge dependent and contains a local UV singularity, which is removed by the zero-temperature fermion counterterm. At two-loop LP, the metric and mixed metric–longitudinal sectors contribute. After the corresponding counterterm subtraction, the remaining divergence is gauge independent, \(\Sigma_{\mathrm{Div}}(Q)=(e^2/48\pi^2\epsilon)(5-28\log(2))\mathcal{H}(T)\), where \(\mathcal{H}(T)\) is the LO self-energy contribution (one-loop LP). The singularity corresponding to \(\log(2)\) is collinear, while the remaining singularity is the hard-region IR singularity. Since the soft sector and overlap subtraction are not evaluated here, the precise cancellation channel of these singularities is not assigned at this stage. The cancellation of the singularities can be checked in the pole equation, after the iteration terms in Eq. 47 are also included. Furthermore, the gauge-dependent contributions that appear in the hard-region self-energy must cancel out in the complete same-order pole equation. On the timelike branch \(q^0>q\), although individual functions at two-loop LP contain imaginary parts, these contributions cancel separately in the metric and mixed metric–longitudinal sectors. Thus, the two-loop LP terms do not generate a finite damping contribution. The one-loop NLP hard contribution is real on the timelike branch and therefore gives no damping contribution.

At the same order, a complete calculation of quasiparticle observables must combine the hard contributions with the HTL-resummed soft sector, the appropriate overlap subtraction, and the iteration terms. Inclusive rates also require the corresponding real-emission and absorption contributions according to the KLN theorem [68], [69]. The present work supplies the hard-region part of this program, along with its symmetry constraints, WTI, and axial-transversality checks. These hard-region amplitudes can also be used for the future construction of the fermionic NLO effective action.

Acknowledgments↩︎

I am grateful to Kaapo Seppänen for helpful correspondence on hard-thermal-loop calculations, especially on higher-order difference propagators and hard-region expansions. I am especially grateful to S. S. Gousheh for helpful discussions, steady encouragement, and support at difficult stages of this work.

8 Dimensional Regularization↩︎

We assume \(D=4-2\epsilon\) where \(\epsilon\rightarrow 0\) is the DR regulator and \(d=D-1\) is the spatial dimension. Considering the spherical symmetry yields [31] \[\begin{align} \int_P\equiv \left(\frac{e^{\gamma_E}\bar{\Lambda}^2}{4\pi}\right)^{\frac{4-D}{2}} \int\frac{d^{D}P}{(2\pi)^{D}} \equiv \mathcal{N}\int_{-\infty}^{+\infty}\frac{d{p^0}}{2\pi}\int_p\int_z, \end{align}\] with \[\begin{align} \label{2} \int_p\equiv\int_0^\infty dp\, p^{d-1},\qquad \int_z\equiv \int_{-1}^{+1}(1-z^2)^{\frac{d-3}{2}} dz,\qquad \mathcal{N}=\frac{4}{(4\pi)^{\frac{d+1}{2}}\Gamma\!\left(\frac{d-1}{2}\right)} \left(\frac{e^{\gamma_E}{\bar{\Lambda}}^{2}}{4\pi}\right)^{\frac{3-d}{2}} . \end{align}\tag{48}\] Here, \(\gamma_E\) and \(\bar{\Lambda}\) are the Euler–Mascheroni constant and the \(\overline{\mathrm{MS}}\) scale. For \(d\to 3-2\epsilon\), \[\begin{align} \label{Eq:N-expansion} \mathcal{N}=\frac{1}{4\pi^2}+\frac{1}{4\pi^2}\ln({\bar{\Lambda}}^2)\,\epsilon+O(\epsilon^2). \end{align}\tag{49}\] We also define the on-shell unit vector \(v^\mu=(1,\hat{\mathbf{p}})\) and [31] \[L(Q) \equiv \frac{1}{2q}\ln\!\left[\frac{q^0 + q + i\eta}{q^0 - q + i\eta}\right].\]

8.1 Radial, Angular, and Collinear Integrals↩︎

The relevant radial integrals used in this work are \[\begin{align} \label{Rdial32integral} \mathcal{R}_1 &\equiv \int_p \frac{1}{p}N_B(p) = \frac{\pi^2T^2}{6}\Big[1+\epsilon\big(24\ln A-2-2\gamma_E+2\ell_T\big)\Big] +\mathcal{O}(\epsilon^2),\nonumber\\ \mathcal{R}_2 &\equiv -\!\int_p \frac{1}{p}N_F(p) = \frac{1}{2}\mathcal{R}_1-\frac{\pi^2T^2}{6}\ln(2)\,\epsilon +\mathcal{O}(\epsilon^2),\nonumber\\ \mathcal{R}_3 &\equiv \int_p \frac{1}{p^3}N_B(p) = \frac{1}{4\epsilon}+\frac{1}{2}\ell_T +\frac{\epsilon}{8}\Big(\pi^2-4\gamma_E^{\,2}+4\ell_T^{\,2}-8\gamma_1\Big) +\mathcal{O}(\epsilon^2),\nonumber\\ \mathcal{R}_4 &\equiv \int_p \frac{1}{p^3}N_F(p) = \mathcal{R}_3+\ln 2\Big[1+\epsilon\big(2\ell_T+\ln 2\big)\Big] +\mathcal{O}(\epsilon^2),\nonumber\\ \mathcal{R}_5 &\equiv \int_p \frac{1}{p^2}\frac{dN_B(p)}{dp} =2\epsilon\,\mathcal{R}_3,\qquad \mathcal{R}_6 \equiv \int_p \frac{1}{p^2}\frac{dN_F(p)}{dp} =2\epsilon\,\mathcal{R}_4. \end{align}\tag{50}\] Here \(\ell_T\equiv \ln(e^{\gamma_E}/2\pi T)\), while \(\gamma_1\) and \(A\) denote the first Stieltjes constant and the Glaisher–Kinkelin constant, respectively. The relations for \(\mathcal{R}_2\) and \(\mathcal{R}_4\) are valid only up to \(\mathcal{O}(\epsilon^2)\). In addition, the last line follows from an integration-by-parts (IBP) identity⁵. The radial integrands also contain vacuum (\(T=0\)) contributions. Indeed, after performing the soft-momentum expansion, the vacuum part reduces to a scaleless power integral of the form \[\label{Eq:scaleless} \Bigl(\text{p-independent factor}\Bigr)\times \int_{0}^{\infty} dp\, p^{\alpha},\tag{51}\] which is set to zero in DR. Therefore, all non-vanishing contributions arise from the finite-temperature part, which is not scaleless due to the presence of the distribution functions \(n_{b/f}\) (Tab. 1).

The angular integrals involving temporal components of \(v^\mu\) are \[\label{Angular32Integrals} \begin{align} \mathcal{A}_0 &\equiv \!\int_z 1 = 2+4\epsilon [1 - \ln 2] + O(\epsilon^2),\\ \mathcal{A}_1 &\equiv \!\int_z \frac{1}{v\!\cdot\! Q} = 2L(Q) +2\epsilon\!\left\{ L(Q)\ln\!\left(-\frac{q^2}{Q^2}\right) - \frac{1}{2q}\!\left[ \operatorname{Li}_2\!\left(\frac{q^0 - q}{q^0 + q}\right) - \operatorname{Li}_2\!\left(\frac{q^0 + q}{q^0 - q}\right) \right]\!\right\}+O(\epsilon^2),\\ \mathcal{A}_2 &\equiv \!\int_z \frac{1}{(v\!\cdot\! Q)^2} = \frac{2}{Q^2} - \frac{4\epsilon}{Q^2}[\ln(2) - q^0 L(Q)] + O(\epsilon^2),\\ \mathcal{A}_3 &\equiv \!\int_z \frac{1}{(v\!\cdot\! Q)^3} = \frac{2q^0}{(Q^2)^2} + \frac{2\epsilon}{(Q^2)^2}\!\left[L(Q)\!\left(q^2+(q^0)^2\right) + q^0(1-2\ln(2))\right] + O(\epsilon^2). \end{align}\tag{52}\] Here \(\operatorname{Li}_2(x)\) denotes the dilogarithm, \[\operatorname{Li}_2(x)=-\int_0^x dt\,\frac{\ln(1-t)}{t}.\] For spatial components, rotational symmetry implies \[\label{Angular32integrals-spatial} \mathcal{A}^i_\alpha \equiv \!\int_z \frac{v^i}{(v\!\cdot\! Q)^\alpha} = \hat{q}^i\!\int_z \frac{z}{(v\!\cdot\! Q)^\alpha} = \frac{q^i}{q^2}\!\left(q^0 \mathcal{A}_\alpha - \mathcal{A}_{\alpha-1}\right).\tag{53}\] The collinear denominators arise from \[\frac{1}{v\!\cdot v'}=\frac{1}{1-z},\qquad z\equiv \hat{v}\!\cdot\hat{v}',\] with the collinear limit \(z\to1\). Holding \(v\) fixed and integrating over \(v'\), DR gives \[\begin{align} \label{collinear32integrals1} \mathcal{C}_1 \equiv \int_{z} \frac{1}{1-z} &= -\frac{1}{\epsilon}+2\ln2+\frac{\epsilon}{6}\bigl(\pi^2-12\ln^2 2\bigr)+O(\epsilon^2),\nonumber\\ \mathcal{C}^{\,i}_1 \equiv \int_{z} \frac{v'^i}{1-z} &= \hat{v}^{\,i}\int_{z} \frac{z}{1-z} = \hat{v}^{\,i}\,(\mathcal{C}_1-\mathcal{A}_0), \end{align}\tag{54}\] where the vector relation follows from axial symmetry about \(\hat{v}\). The remaining integration is \[\begin{align} \label{collinear32integrals3} &\int_{v'}\frac{v'\!\cdot Q}{v\!\cdot v'}=q^0\,\mathcal{A}_0+(\mathcal{C}_1-\mathcal{A}_0)\,v\!\cdot Q. \end{align}\tag{55}\] The slashed notation is also introduced to represent more compact relations \[\begin{align} \label{slashed32angular32notations} \cancel{\mathcal{A}}_\alpha \equiv \mathcal{A}_\alpha\,\gamma^0-\mathcal{A}_\alpha^{\,i}\gamma^i. \end{align}\tag{56}\]

8.2 Difference-Propagator and KMS Identities↩︎

The difference propagators satisfy the following relations⁶ [29], [31] \[\tag{57} \begin{align} &\int_{-\infty}^{+\infty}\frac{d{p^0}}{2\pi} \Delta^{(n)}_d(P)f({p^0},p)=\frac{(-i)^{n+1}}{(n-1)!}\int_{-\infty}^{+\infty}\frac{d{p^0}}{2\pi}\Delta_d(P)\,\Bigl\{2{p^0} \Bigl[\frac{-1}{2{p^0}}\frac{\partial}{\partial {p^0}}\Bigr]^{n-1}\Bigl[\frac{-f({p^0,p})}{2{p^0}}\Bigr]\Bigr\},\tag{58}\\ &\int_P \Delta_d(P)\, f({p^0},\vec{p})=\int_{\vec{p} } \frac{f(p,\vec{p})-f(-p,\vec{p})}{2p}.\tag{59} \end{align}\] The last line vanishes if \(f(p^0,p)\) is even in \(p^0\). In two-loop computations, pinch singularities can arise from products of retarded and advanced propagators whose poles lie on infinitesimally opposite sides of the real axis. These pinch structures are removed before the soft expansion by applying the KMS identity [29], [49] \[\begin{align} \label{KMS32relation} N_F(P_1)N_F(P_2) + N_F(P_2)N_B(P_3) + N_B(P_3)N_F(P_1) + \frac{1}{4} = 0, \end{align}\tag{60}\] with the momenta satisfying \(P_1+P_2+P_3=0\). This relation is sufficient for the purpose of this study, while its generalization is provided in Ref. [49].

9 Symmetry Constraints on Building Blocks↩︎

In massless QED, with no background source that breaks the discrete symmetries and chiral invariance, the Lagrangian and the generating functional are invariant under \(P\), \(\mathcal{T}\), \(C\), and chiral transformation. Invariance under \(\mathcal{T}\) is the invariance under an antiunitary transformation. We adopt the notation \(X\,\Phi\,X^{-1}\equiv \Phi_X\) with \(X\in \{P,\mathcal{T}, C\}\) and \(\Phi\) a generic field. In addition, \[Q^\mathcal{T}=Q^P=(q^0,-\mathbf{q}),\qquad x^P=(t,-\mathbf{x}),\qquad x^\mathcal{T}=(-t,\mathbf{x}),\] with Lorentz indices suppressed. A complete basis of Dirac matrices is [58]: \(\{\mathbb{1},\gamma_5,\gamma_\mu,\gamma_\mu\gamma_5,\sigma_{\mu\nu}\}\). The chiral anticommutation relations are \[\begin{align} \label{chiral} \{\gamma_5,\mathbb{1}\}=2\gamma_5,\quad \{\gamma_5,\gamma_5\}=2,\quad \{\gamma_5,\gamma_\mu\}=0,\quad \{\gamma_5,\gamma_\mu\gamma_5\}=0,\quad \{\gamma_5,\sigma_{\mu\nu}\}=2\,\sigma_{\mu\nu}\gamma_5\neq 0. \end{align}\tag{61}\] The Dirac matrices satisfy \[\begin{align} \label{B2} \gamma_0\gamma_\mu\gamma_0=\Lambda^{\nu}{}_\mu\gamma_\nu,\quad \gamma_0(\gamma_\mu\gamma_5)\gamma_0=-\Lambda^{\nu}{}_\mu\gamma_\nu\gamma_5, \end{align}\tag{62}\] with \(\Lambda^{\nu}{}_\mu=\mathrm{diag}(1,-1,-1,-1)\). The fermion and vector fields transform under parity as [58] \[\begin{align} \label{parity32properties} \psi_P(x)= \gamma_0\,\psi(x^P),\qquad \bar\psi_P(x)= \bar\psi(x^P)\gamma_0,\qquad A^\mu_P(x)=\Lambda^{\mu}{}_\nu A^\nu(x^P). \end{align}\tag{63}\] The overall fermionic phase is omitted since it cancels between \(\psi\) and \(\bar\psi\) in bilinears. Similarly, charge conjugation acts as [58] \[\begin{align} \label{charge32conjugation} &\psi_C(x)= -i\big[\bar{\psi}(x) M\big]^{T},\qquad \bar{\psi}_C(x)=-i\big[M\psi(x)\big]^{T},\qquad A^\mu_C(x)=-A^\mu(x),\nonumber\\ &\psi_C(Q)=-i\big[\bar\psi(-Q)M\big]^T,\qquad \bar\psi_C(Q)=-i\,\big[M\psi(-Q)\big]^T, \end{align}\tag{64}\] with \(M\equiv\gamma_0\gamma_2\) and \(T\) denoting transpose in Dirac space. The matrix \(M\) satisfies \[\begin{align} \label{M32properties} M^{T}(\gamma_\mu)^T M^T=-\gamma_\mu,\quad M^{T}(\gamma_\mu\gamma_5)^T M^T=\gamma_\mu\gamma_5,\quad M^T=-M,\quad M^{-1}=M. \end{align}\tag{65}\] Time reversal acts on fields as [58] \[\label{B6} \psi_{\mathcal{T}}(Q)=B\,\psi(Q^{\mathcal{T}}),\qquad \bar\psi_{\mathcal{T}}(Q)=-\bar\psi(Q^{\mathcal{T}})B,\qquad A^\mu_{\mathcal{T}}(Q) = \Lambda^\mu{}_\nu A^\nu(Q^{\mathcal{T}}),\tag{66}\] with \(B=\gamma_{1}\gamma_{3}\) (up to a phase). The required gamma-matrix identities are \[\label{B77} B\gamma_{\mu}^{*}B=-\Lambda^{\nu}{}_{\mu}\gamma_{\nu}, \qquad B(\gamma_{\mu}\gamma_{5})^{*}B=-\Lambda^{\nu}{}_{\mu}\gamma_{\nu}\gamma_{5}.\tag{67}\] Since \(\mathcal{T}\) is antiunitary, it complex conjugates the \(c\)-number coefficients and the matrix entries of the Dirac structures.

9.1 Spectral Function and Fermion Self-Energy↩︎

We start with the fermionic spectral function defined in position space by \(\rho_F(x,y)=\big\langle \{\psi(x),\bar\psi(y)\}\big\rangle_T\), with Fourier transform \(\rho_F(Q)\). The most general Dirac decomposition gives \[\label{eq:rho-decomp} \rho_F(Q) = \mathcal{S}\,\mathbb{1} + \mathcal{P}\,\gamma_5 + V^\mu\,\gamma_\mu + A^\mu\,\gamma_\mu\gamma_5 + \tfrac{1}{2}\,T^{\mu\nu}\,\sigma_{\mu\nu}.\tag{68}\] Chiral invariance implies \(\mathcal{S}=\mathcal{P}=T^{\mu\nu}=0\), yielding \[\label{eq:rho-VA} \rho_F(Q)= V^\mu(Q)\,\gamma_\mu + A^\mu(Q)\,\gamma_\mu\gamma_5.\tag{69}\] We now impose discrete symmetries. Under parity invariance, the spectral matrix transforms as \[\label{eq:parity-rho} \rho_F(q^0,\mathbf{q}) = \gamma_0\,\rho_F(q^0,-\mathbf{q})\,\gamma_0,\tag{70}\] where we used Eq. 63 . Using Eqs. 62 , 69 and 70 , one finds that \(V^\mu\) transforms as a polar vector, while \(A^\mu\) transforms as a pseudovector. For the latter, no nonzero pseudovector can be constructed from \(Q^\mu\) and \(u^\mu\) alone. Equivalently, any Levi–Civita construction built only from \(Q^\mu\) and \(u^\mu\) vanishes without an additional pseudovector/pseudoscalar background. Hence, setting \(A^\mu(Q)=0\), we obtain \[\label{eq:rho-final} \rho_F(Q)=a(Q)\,Q\!\!\!/+ b(Q)\,u\!\!\!/,\tag{71}\] with real scalar coefficients \(a(Q)\) and \(b(Q)\) depending on the invariants \(Q^2\) and \(u\!\cdot\!Q\). The reality is ensured by Dirac pseudo-Hermiticity \[\begin{align} \label{spectral-pseudo-Hermiticity} \rho_F^\dagger(Q)=\gamma_0\rho_F(Q)\gamma_0. \end{align}\tag{72}\] Next, using the charge-conjugation properties in Eq. 64 and translation invariance, \(\rho_F(x,y)=\rho_F(x-y)\), one obtains⁷ \[\rho_F(Q) = -M^{T}\,\rho_F(-Q)^{T}\,M^{T}.\] Substituting Eq. 71 and using Eq. 65 , one obtains \[a(-Q) = -a(Q), \qquad b(-Q) = b(Q),\] which yields \[\label{spectral32property} \rho_F(-Q)=\rho_F(Q).\tag{74}\] The evenness above, together with the spectral representation [26], yields \[\begin{align} \label{advanced32property} S_A(-Q) &=\int_{-\infty}^\infty\frac{d\omega}{2\pi}\, \frac{\rho_F(\omega,-\mathbf{q})}{-Q_0-\omega-i\eta} = -\int_{-\infty}^\infty\frac{d\omega}{2\pi}\, \frac{\rho_F(-\omega,-\mathbf{q})}{Q_0-\omega+i\eta} = -S_R(Q), \end{align}\tag{75}\] where we used \(\rho_F(-\omega,-\mathbf{q})=\rho_F(\omega,\mathbf{q})\). Using the Dyson relation \(S^{-1}_{R/A}=S^{-1}_{0,R/A}-\Sigma_{R/A}\), one obtains \[\Sigma_A(-Q)=-\Sigma_R(Q).\]

9.2 Two-Fermion–\(N\)-Photon Vertices↩︎

We consider the part of the 1PI effective action containing two fermion fields and \(N\) external photon insertions, \[\mathbf{\Gamma}[\psi,\bar\psi,A]\supset \sum_{N\geq 1}\frac{(-1)^N}{N!} \int d^4y\,d^4z\,d^4x_1\cdots d^4x_N\; \bar\psi(y)\, \Gamma^{\mu_1\cdots\mu_N}(y,z;x_1,\ldots,x_N)\, \psi(z)\, A_{\mu_1}(x_1)\cdots A_{\mu_N}(x_N),\] where \(\Gamma^{\mu_1\cdots\mu_N}(y,z;x_1,\ldots,x_N)\) is the two-fermion–\(N\)-photon vertex function in position space and is a \(4\times4\) matrix in Dirac space. Under our assumptions, the equilibrium generating functional is invariant under the discrete transformations defined above. The corresponding 1PI effective action therefore inherits these symmetries through the Legendre-transform construction [70]: \[\label{effective32action32invariance} \mathbf{\Gamma}[\psi_{X},\bar\psi_{X},A_{X}] =\mathbf{\Gamma}[\psi,\bar\psi,A], \qquad X\in\{C, P, \mathcal{T}\}.\tag{76}\] The corresponding dressed \(N\)-current kernel is defined by \[j^{\mu_1\cdots\mu_N}(x_1,\ldots,x_N) \equiv (-1)^N \frac{\delta^N\mathbf{\Gamma}[\psi,\bar\psi,A]}{\delta A_{\mu_1}(x_1)\cdots\delta A_{\mu_N}(x_N)} = \int d^4y\,d^4z\; \bar\psi(y)\, \Gamma^{\mu_1\cdots\mu_N}(y,z;x_1,\ldots,x_N)\, \psi(z). \label{eq:Ncurrent95kernel}\tag{77}\] In a translationally invariant state, \(\Gamma^{\mu_1\cdots\mu_N}(y,z;x_1,\ldots,x_N)\) depends only on relative coordinates. For example, one may choose \(x_N\) as a reference point and write it as a function of \(y-x_N\), \(z-x_N\), and \(x_i-x_N\) with \(i=1,\ldots,N-1\). Equivalently, its Fourier transform contains the overall momentum-conservation delta function. Therefore, the Fourier-space kernel can be written as \[j^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) = \int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(Q_2)\, \Gamma^{\mu_1\cdots\mu_N}(Q_2,Q_1;\{K_i\})\, \psi(Q_1), \qquad Q_2=Q_1+\sum_{i=1}^N K_i , \label{eq:Ncurrent95momentum}\tag{78}\] where all photon momenta are taken incoming. By momentum conservation, only \(N-1\) photon momenta are independent. We denote them by \(\{K_i\}\equiv\{K_1,\ldots,K_{N-1}\}\), with \(K_N\) fixed by momentum conservation. Chiral invariance implies that the vertex matrix anticommutes with \(\gamma_5\). Therefore, scalar, pseudoscalar, and tensor Dirac structures are forbidden, and the most general decomposition of the current is \[\label{B45} j^{\mu_1\cdots\mu_N}(K_1,\dots,K_N) = \int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(Q_2)\, \Big[ V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\})\,\gamma_\nu + A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\})\,\gamma_\nu\gamma_5 \Big]\, \psi(Q_1),\tag{79}\] where \(V^{\mu_1\cdots\mu_N,\nu}\) and \(A^{\mu_1\cdots\mu_N,\nu}\) are coefficient tensors constructed from the available four-vectors \(Q_{1,2}^\mu\), \(K_i^\mu\), and \(u^\mu\), together with \(g^{\mu\nu}\) and \(\epsilon^{\mu\nu\rho\sigma}\) (see, e.g., the zero-temperature discussion [61], Ch. 10). In our case, \(u^\mu=(1,0,0,0)\).

9.2.0.1 Parity.

Using Eq. 63 , one obtains the parity transform of the current as \[\begin{align} j_{P}^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) &=\int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(Q_2^P)\Big[ V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) \Lambda^{\beta}{}_\nu\,\gamma_\beta- A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) \Lambda^{\beta}{}_\nu\,\gamma_\beta\gamma_5 \Big]\psi(Q_1^P), \end{align}\] with \(Q^P=(q^0,-\mathbf{q})\) and \(\Lambda^\mu{}_\nu=\mathrm{diag}(1,-1,-1,-1)\). Since each photon field is a polar vector, parity invariance of the effective action implies that the current transforms as a rank-\(N\) polar tensor: \[\begin{align} \label{Polar-tensor} j_{P}^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N)=\int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(Q_2^P)\, &\Lambda^{\mu_1}{}_{\rho_1}\cdots \Lambda^{\mu_N}{}_{\rho_N}\nonumber\\ & \times\Big[ V^{\rho_1\cdots\rho_N,\beta}(Q_2^P,Q_1^P;\{K_i^P\})\gamma_\beta + A^{\rho_1\cdots\rho_N,\beta}(Q_2^P,Q_1^P;\{K_i^P\})\gamma_\beta\gamma_5 \Big]\psi(Q_1^P). \end{align}\tag{80}\] Since these expressions agree for arbitrary spinor fields, one obtains \[\begin{align} V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= \Lambda^{\mu_1}{}_{\rho_1}\cdots \Lambda^{\mu_N}{}_{\rho_N}\Lambda^\nu{}_\beta\, V^{\rho_1\cdots\rho_N,\beta} (Q_2^P,Q_1^P;\{K_i^P\}), \nonumber\\ A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= -\Lambda^{\mu_1}{}_{\rho_1}\cdots \Lambda^{\mu_N}{}_{\rho_N}\Lambda^\nu{}_\beta\, A^{\rho_1\cdots\rho_N,\beta} (Q_2^P,Q_1^P;\{K_i^P\}). \label{eq:parity95Nvertex} \end{align}\tag{81}\] Thus \(V^{\mu_1\cdots\mu_N,\nu}\) is a rank-\((N+1)\) tensor, while \(A^{\mu_1\cdots\mu_N,\nu}\) is a rank-\((N+1)\) pseudotensor.

9.2.0.2 Time reversal and \(P\mathcal{T}\).

Using Eqs. 66 and 67 , and accounting for the antiunitarity of \(\mathcal{T}\), one obtains the transformed dressed current as \[\begin{align} \label{eq:T-current-direct-N} j_{\mathcal{T}}^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) &=\int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(Q_2^{\mathcal{T}}) \Big[ V^{*\,\mu_1\cdots\mu_N,\rho}(Q_2,Q_1;\{K_i\}) \Lambda^\nu{}_\rho\,\gamma_\nu+ A^{*\,\mu_1\cdots\mu_N,\rho}(Q_2,Q_1;\{K_i\}) \Lambda^\nu{}_\rho\,\gamma_\nu\gamma_5 \Big] \psi(Q_1^{\mathcal{T}}) . \end{align}\tag{82}\] Note that \(Q^P=Q^{\mathcal{T}}\) and \(K_i^P=K_i^{\mathcal{T}}\). The invariance of the effective action, together with the transformation law of the photon field under \(\mathcal{T}\), implies that the dressed \(N\)-current transforms as a rank-\(N\) polar tensor, as in Eq. 80 with \(P\) replaced by \(\mathcal{T}\). Equating both equations gives \[\begin{align} \label{eq:T-Nvertex} V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= \Lambda^{\mu_1}{}_{\alpha_1}\cdots \Lambda^{\mu_N}{}_{\alpha_N} \Lambda^\nu{}_\beta\, V^{*\,\alpha_1\cdots\alpha_N,\beta} (Q_2^{P},Q_1^{P};\{K_i^{P}\}), \nonumber\\ A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= \Lambda^{\mu_1}{}_{\alpha_1}\cdots \Lambda^{\mu_N}{}_{\alpha_N} \Lambda^\nu{}_\beta\, A^{*\,\alpha_1\cdots\alpha_N,\beta} (Q_2^{P},Q_1^{P};\{K_i^{P}\}) . \end{align}\tag{83}\] Combining Eq. 83 with the parity constraint Eq. 81 , one obtains the \(P\mathcal{T}\) reality constraints \[\begin{align} \label{eq:PT-Nvertex} V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= V^{*\,\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}), \nonumber\\ A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= -\,A^{*\,\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) . \end{align}\tag{84}\] Thus, the vector and axial coefficient tensors are real and imaginary, respectively.

9.2.0.3 Charge conjugation.

Charge conjugation acts on the dressed \(N\)-current as \[\begin{align} \label{eq:C-current-direct-N} j_{C}^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) &= -\int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(-Q_1) \Big[ V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\})\gamma_\nu - A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\})\gamma_\nu\gamma_5 \Big]\psi(-Q_2), \end{align}\tag{85}\] where Eqs. 64 , and 65 , together with the Grassmann identity \(\psi^T \mathcal{X}\,\bar\psi^T=-\bar\psi\,\mathcal{X}^T\psi\), have been used. Here, \(\mathcal{X}\) is an arbitrary matrix in Dirac space. Changing variables gives \[\begin{align} \label{eq:C-current-changed-N} j_{C}^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) &= -\int\!\frac{d^4Q_1}{(2\pi)^4}\; \bar\psi(Q_2) \Big[ V^{\mu_1\cdots\mu_N,\nu}(-Q_1,-Q_2;\{K_i\})\gamma_\nu - A^{\mu_1\cdots\mu_N,\nu}(-Q_1,-Q_2;\{K_i\})\gamma_\nu\gamma_5 \Big]\psi(Q_1). \end{align}\tag{86}\] Since each photon field is odd under charge conjugation, \(C\)-invariance of the effective action implies \[j_C^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) = (-1)^N j^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N).\] Comparing with Eq. 86 , one obtains \[\begin{align} \label{eq:C-Nvertex} V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= (-1)^{N+1} V^{\mu_1\cdots\mu_N,\nu}(-Q_1,-Q_2;\{K_i\}), \nonumber\\ A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= (-1)^N A^{\mu_1\cdots\mu_N,\nu}(-Q_1,-Q_2;\{K_i\}). \end{align}\tag{87}\]

9.2.0.4 Hermiticity and fermion-leg exchange.

The Hermiticity of the dressed \(N\)-current in position space \[j^{\mu_1\cdots\mu_N}(x_1,\ldots,x_N)^\dagger = j^{\mu_1\cdots\mu_N}(x_1,\ldots,x_N),\] implies \[j^{\mu_1\cdots\mu_N}(K_1,\ldots,K_N) = j^{\mu_1\cdots\mu_N}(-K_1,\ldots,-K_N)^\dagger,\] which yields \[\label{eq:Herm-Nvertex} \Gamma^{\mu_1\cdots\mu_N}(Q_2,Q_1;\{K_i\}) = \gamma_0\, \Gamma^{\mu_1\cdots\mu_N}(Q_1,Q_2;\{-K_i\})^\dagger\, \gamma_0 .\tag{88}\] Using the vector–axial decomposition, this gives \[\begin{align} \label{eq:Herm-VA-Nvertex} V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= V^{*\,\mu_1\cdots\mu_N,\nu}(Q_1,Q_2;\{-K_i\}), \nonumber\\ A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= A^{*\,\mu_1\cdots\mu_N,\nu}(Q_1,Q_2;\{-K_i\}). \end{align}\tag{89}\] Combining Eq. 89 with the \(P\mathcal{T}\) reality properties in Eq. 84 , one obtains the exchange properties \[\begin{align} \label{eq:exchange-Nvertex} V^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= V^{\mu_1\cdots\mu_N,\nu}(Q_1,Q_2;\{-K_i\}), \nonumber\\ A^{\mu_1\cdots\mu_N,\nu}(Q_2,Q_1;\{K_i\}) &= - A^{\mu_1\cdots\mu_N,\nu}(Q_1,Q_2;\{-K_i\}). \end{align}\tag{90}\] Thus \(V\) is exchange-symmetric while \(A\) is exchange-antisymmetric under interchange of the fermion legs, with the induced \(\{K_i\}\to\{-K_i\}\) fixed by momentum conservation.

10 One-Loop Order Corrections↩︎

Before giving the one-loop derivation, we collect the notation used in Apps. C–E. We follow the conventions introduced in Sec. 2 and Tab. 3. In this table, the first block applies throughout, while the second block is used only for the two-loop computations discussed in the following appendices.

Table 3: The self-energy and vertex conventions. The first block applies generally, while the second block is used only for the two-loop computations.

Convention	Description
Virtual corrections	\(\Sigma_{X}\): self-energy; \(\Gamma^\mu_X\): vertex; \(X\) denotes topology type, gauge sectors, and statistical sectors.
Dirac structures	\(\mathcal{S}\): self-energy; \(\mathcal{V}\): vertex.
Subscripts on \(\mathcal{S}_{(x_s)}\) and \(\mathcal{V}_{(x_v)}\)	The subscripts denote that the shift momenta are applied to the corresponding Dirac numerator.
Composite operations	\(S_{(a_s g_s)}\): sequential application, first \(a_s\) and then \(g_s\), on \(S\).
Prime notation	\(\mathcal{S}'_{(a_sc_s)}\): \(P \leftrightarrows L\) transformation applied to \(\mathcal{S}{(a_sc_s)}\).
Compact momentum notations	Plus signs in internal propagators are omitted; e.g., \(\Delta_j(PLQ):=\Delta_j(P+L+Q)\) with \(j=R,A,s\).
Metric sector	\(\Sigma_{{X}_{11}}\),\(\Gamma^\mu_{X_{11}}\).
Metric–longitudinal \((1-\xi)\)	\(\Sigma_{X_{12}}+\Sigma_{X_{21}}\),\(\Gamma^\mu_{X_{12}}+\Gamma^\mu_{X_{21}}\).
Fully longitudinal \((1-\xi)^2\)	\(\Sigma_{X_{22}}\),\(\Gamma^\mu_{X_{22}}\) .
Dirac structures in new basis	self-energy: \(\left\{{\mathcal{S}_{11},\mathcal{S}_{12}/\mathcal{S}_{21},\mathcal{S}_{22}}\right\}\),vertex:\(\left\{{\mathcal{V}_{11},\mathcal{V}_{12}/\mathcal{V}_{21},\mathcal{V}_{22}}\right\}\).
Statistical sectors	\(N_B(P)N_B(L)\), \(N_B(P)N_F(L)\), \(N_F(L)N_F(P)\)

10.1 Self-Energy Corrections↩︎

The \(r/a\) assignments for the one-loop fermion self-energy, shown in Fig. 3, are drawn using the propagator conventions of Fig. 1. Together with Eq. 2 and the ordinary QED Feynman rules, they give \[\begin{align} \label{eq:C1} -i\Sigma^{(1)}_{m}(Q) &= e^2\!\int_P \, \mathcal{S}_{m} \Big[\Delta_s(PQ)\Delta_A(P)+\tilde{\Delta}_s(P)\Delta_R(PQ)\Big], \nonumber\\ -i\Sigma^{(1)}_{\xi}(Q) &= i e^2(1-\xi)\!\int_P \, \mathcal{S}_{\xi} \Big[\Delta_s(PQ)\Delta_A^{(2)}(P)+\tilde{\Delta}_s^{(2)}(P)\Delta_R(PQ)\Big], \end{align}\tag{91}\] with Dirac structures \[\label{Dirac-one-loop} \mathcal{S}_{m}=\gamma_\mu(\cancel{P}+\cancel{Q})\gamma^\mu, \qquad \mathcal{S}_{\xi}=\cancel{P}(\cancel{P}+\cancel{Q})\cancel{P}.\tag{92}\] We shift the loop momentum in the first term so that the symmetric propagator depends only on \(P\). This allows us to impose the on-shell condition through \(\Delta_d(P)\): \[\begin{align} \label{basic-self-energy} \Sigma^{(1)}_{m}(Q) &=i e^2\!\int_P \Big[\mathcal{S}_{m(as)}\,\Delta_s(P)\Delta_A(P-Q)+\mathcal{S}_{m}\,\tilde{\Delta}_s(P)\Delta_R(PQ)\Big], \nonumber\\ \Sigma^{(1)}_{\xi}(Q) &=- e^2(1-\xi)\!\int_P \Big[\mathcal{S}_{\xi(as)}\,\Delta_s(P)\Delta_A^{(2)}(P-Q)+\mathcal{S}_{\xi}\,\tilde{\Delta}_s^{(2)}(P)\Delta_R(PQ)\Big]. \end{align}\tag{93}\] For the second term in the last line, we use Eq. 57 : \[\begin{align} \label{C4} \int_P \tilde{\Delta}_s^{(2)}(P) \mathcal{S}_\xi \Delta_R(PQ)=i\int_P \Delta_d(P)\frac{d}{dp^0}\Bigl[\frac{N_B(P)\mathcal{S}_\xi\Delta_R(PQ)}{2p^0}\Bigr]. \end{align}\tag{94}\] In this case, the expression inside the brackets is differentiated with respect to \(p^0\). The on-shell condition is then applied, and the HTL expansion is performed at the end.

10.1.0.1 LP Order

Following the expansion procedure explained in Sec. 3, one obtains the LP contribution as, \[\begin{align} \label{A2} \Sigma_{m}^{0(1;LP)}=\frac{1}{4}\mathrm{Tr}\left[\gamma^0 \Sigma_{m}^{(1;LP)}(Q)\right]&=-\frac{ e^2\mathcal{D}}{4}\int_P \Delta_d(P)\left[N_B(P)-N_F(P)\right]\,\frac{{-2p^0}}{P.Q}\nonumber\\ &=\frac{e^2\mathcal{D}}{2}\mathcal{N}\int_{\vec{p}}\frac{1}{p}\left[N_B(P)-N_F(P)\right]\int_z \frac{1}{v.Q}\nonumber\\ &=\frac{e^2T^2}{8}L(Q)\equiv m^2(T)\, L(Q), \end{align}\tag{95}\] where the functions given in App. 8 has been used. The spatial component is similarly as, \[\begin{align} \label{A3} \Sigma_{m}^{\,i (1;LP)}=-\frac{1}{4}\mathrm{Tr}\left[\gamma^i\Sigma^{(1;LP)}_{m}(Q)\right] &=-\frac{e^2}{2}\mathcal{D}\mathcal{N}(\mathcal{R}_2+\mathcal{R}_1)\mathcal{A}^i_1= m^2(T)\,\frac{\hat{q}^i}{q} \left[1-q^0 L(Q)\right]. \end{align}\tag{96}\] At this power order, the longitudinal-sector integrand is odd under the on-shell momentum reversal and therefore vanishes.

10.1.0.2 NLP Order

At NLP order, we first compute the temporal and spatial projections and then recombine them into Dirac form. The expanded results are \[\tag{97} \begin{align} \Sigma^{(1;NLP)}_{\rm m}(Q) &=\frac{e^2}{4}\mathcal{D}\int_P \Delta_d(P)\biggl\{\left[N_B(P)-N_F(P)\right]\frac{\cancel{P}\,Q^4}{2 (P\!\cdot\! Q)^3} -N_B(P)\frac{\cancel{Q}\,Q^2 }{(P\!\cdot\! Q)^2}\biggr\},\tag{98}\\ \Sigma^{(1;NLP)}_{\xi}(Q) &=\frac{e^2}{4}(1-\xi)\int_P \Delta_d(P)\biggl\{\left[N_B(P)+N_F(P)\right]\Bigl[\frac{2 Q^2 \cancel{Q}}{(P\!\cdot\! Q)^2}-\frac{Q^4 \cancel{P}}{(P\!\cdot\! Q)^3}\Bigr]\nonumber\\ &+N_B(P)\Bigl[\frac{Q^2 \cancel{P}}{({p^0})^2 (P\!\cdot\! Q)}+\frac{Q^2 q^0 \cancel{P}}{{p^0} (P\!\cdot\! Q)^2}-\frac{Q^2 \cancel{Q}}{(P\!\cdot\! Q)^2}-\frac{\gamma^0 Q^2}{{p^0} (P\!\cdot\! Q)}\Bigr] -\frac{Q^2 \cancel{P} }{{p^0} (P\!\cdot\! Q)} N^\prime_B(P)\biggr\}\tag{99}, \end{align}\] with \(\mathcal{D}=D-2\).

10.2 Vertex Corrections↩︎

For the one-loop vertex (Fig. [one-loop32three32vertex]), the corresponding r/a assignments illustrated in Fig. [ra32one-loop32three32vertex] yield the following contributions, respectively \[\label{one-loop32vertex} \begin{align} &-i e\Gamma_{m}^{\mu(1)}(Q_2,Q_1)=-ie^3\int_P \mathcal{V}_{m}\Bigl[\Delta_s(PQ_1)\Delta_A(P)\Delta_A(PQ_2)+\tilde{\Delta}_s(P)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2)+\Delta_s(PQ_2)\Delta_{R}(PQ_1)\Delta_A(P)\Bigr]\\ &-ie\Gamma_{\xi}^{\mu(1)}(Q_2,Q_1)=e^3(1-\xi)\int_P\mathcal{V}_{\xi}\\ &\times\Bigl[\Delta_s(PQ_1)\Delta_{A}^{(2)}(P)\Delta_A(PQ_2)+\tilde{\Delta}^{(2)}_s(P)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2)+\Delta_s(PQ_2)\Delta_{R}(PQ_1)\Delta_{A}^{(2)}(P)\Bigr], \end{align}\tag{100}\] with the Dirac structures given in Eq. 40 . The same momentum-shift, used for the self-energy, is applied to bring each symmetric propagator to the form \(\Delta_s(P)\) or \(\tilde{\Delta}_s(P)\).

11 Two-Loop Fermion Self-Energy: Intermediate Expressions↩︎

We present the intermediate steps for obtaining pinch-free pre-HTL results for the two-loop self-energy, using the conventions defined below. The loop-momentum transformations used in this appendix are \[\begin{align} \label{self-energy32transformation} &a_s: P\rightarrow P-Q,\qquad b_s: L\rightarrow L-Q,\qquad c_s: L\rightarrow L-P,\qquad d_s: P\rightarrow P-L, \nonumber\\ &e_s: L\rightarrow L-P-Q,\qquad f_s: P\rightarrow P-L-Q,\qquad g_s: L\rightarrow L-P+Q . \end{align}\tag{101}\] When a label such as \((a_s)\) is attached to a Dirac numerator, it denotes the corresponding substitution in Eq. 101 applied to that numerator. Composite labels denote sequential substitutions, Tab. 3. A prime on a Dirac structure denotes the exchange \(P\leftrightarrow L\) acting only on the Dirac structure. We use a compact notation to treat the three gauge sectors discussed in Sec. 3.2 simultaneously.

Let the pair \((i,j)\), with \(i\in\{1,2\}\) and \(j\in \{1,2\}\), represent the gauge sectors as \[\begin{align} \label{gauge32sector32map32app} \Sigma_{X_{11}}\equiv \Sigma_{X_{m}},\qquad \Sigma_{X_{12}}+\Sigma_{X_{21}}\equiv \Sigma_{X_{\xi}},\qquad \Sigma_{X_{22}}\equiv \Sigma_{X_{\xi^2}},\quad X\in\{R,B,C\}, \end{align}\tag{102}\] where \(X\) labels the topology: \(R\), \(B\), and \(C\) denote the rainbow, bubble, and cross-photon topologies, respectively. The corresponding gauge factors are then \[\begin{align} \label{gauge32factors32app} \hat{\xi}_{11}\equiv 1,\qquad \hat{\xi}_{12}=\hat{\xi}_{21}\equiv i(1-\xi),\qquad \hat{\xi}_{22}\equiv -(1-\xi)^2 . \end{align}\tag{103}\] To write the Dirac numerators compactly, we introduce \[\begin{align} \label{compact32Dirac} \mathcal{A}_1 \equiv \gamma^\alpha,\quad \mathcal{A}_2 \equiv \cancel{P},\qquad \mathcal{B}_1 \equiv \gamma^\nu,\quad \mathcal{B}_2 \equiv \cancel{L},\qquad \mathcal{C}_1 \equiv \gamma^\nu,\quad \mathcal{C}_2 \equiv \cancel{P},\qquad \mathcal{D}_1 \equiv \gamma^\alpha,\quad \mathcal{D}_2 \equiv \cancel{L}, \end{align}\tag{104}\] with repeated Lorentz indices contracted. For example, the Dirac numerator corresponding to the rainbow self-energy is \[\begin{align} \label{rainbow32compact32dirac32app} \mathcal{S}_{R_{ij}} = \mathcal{A}_i(\cancel{P}+\cancel{Q}) \mathcal{B}_j(\cancel{P}+\cancel{L}+\cancel{Q}) \mathcal{B}_j(\cancel{P}+\cancel{Q}) \mathcal{A}_i, \qquad i,j=1,2, \end{align}\tag{105}\] which, upon comparison with the explicit structures in Eq. 17 , gives \[\begin{align} \label{rainbow32sector32map32app} \mathcal{S}_{R_{11}}\equiv \mathcal{S}_{R_{m}},\qquad \mathcal{S}_{R_{12}}\equiv \mathcal{S}_{R_{\xi_L}},\qquad \mathcal{S}_{R_{21}}\equiv \mathcal{S}_{R_{\xi_P}},\qquad \mathcal{S}_{R_{22}}\equiv \mathcal{S}_{R_{\xi^2}} . \end{align}\tag{106}\] The same compact organization is used throughout for the remaining two-loop self-energy and vertex topologies. These conventions are summarized in Tab. 3.

The different gauge contributions arise from the photon propagators. For the two internal photon lines, one assigns the labels \(i\) and \(j\) to indicate the gauge structure. The Dirac numerator \(\mathcal{S}_{ij}\) must be matched to the corresponding explicit Dirac numerators; see, for instance, Eq. 17 . Afterward, the remaining steps are consistent, and the introduced notation retains all gauge sectors simultaneously.

11.1 The Rainbow Self-Energy↩︎

All allowed \(r/a\) assignments are shown in Fig. 5. Using the notation above, the sum of \(r/a\) assignments gives \[\begin{align} \label{rainbow32self-energy32direct} -i\Sigma^{(2)}_{R_{ij}}(Q) &= e^4 \int_P\int_L \hat{\xi}_{ij}\, \mathcal{S}_{R_{ij}} \Bigl\{ \tilde{\Delta}^{(i)}_s(P)\Delta_R^{(2)}(PQ) \Bigl[ \Delta_s(PLQ)\Delta_A^{(j)}(L) +\tilde{\Delta}^{(j)}_s(L)\Delta_R(PLQ) \Bigr] \nonumber\\ &\quad +\Delta_s(PQ)\tilde{\Delta}^{(j)}_s(L)\Delta_A^{(i)}(P) \Bigl[ \Delta_A(PLQ)\Delta_A(PQ) +\Delta_R(PLQ)\Delta_R(PQ) \Bigr] \nonumber\\ &\quad +\Delta_s(PQ)\Delta_s(PLQ)\Delta_A^{(i)}(P) \Bigl[ \Delta_R^{(j)}(L)\Delta_A(PQ) +\Delta_R(PQ)\Delta_A^{(j)}(L) \Bigr] \nonumber\\ &\quad +\Delta_s(PLQ)\tilde{\Delta}^{(j)}_s(L) \Delta_R(PQ)\Delta_A(PQ)\Delta_A^{(i)}(P) \Bigr\}. \end{align}\tag{107}\] The last line contains the pinch structure \(\Delta_R(PQ)\Delta_A(PQ)\). To remove the pinch structure, we first sum the contributions from the second, third, and fourth lines. We then use the KMS relation, Eq. 60 , to rearrange them into a pinch-free form. In the diagrammatic representation, these terms correspond to the five diagrams shown in the second block of Fig. 5. Afterward, the pinch-free expression becomes \[\begin{align} \label{rainbow32self-energy32pinchfree} -i\Sigma^{(2)}_{R_{ij}}(Q) &= e^4 \int_P\int_L \hat{\xi}_{ij}\, \mathcal{S}_{R_{ij}} \Bigl\{ \tilde{\Delta}^{(i)}_s(P)\Delta_R^{(2)}(PQ) \Bigl[ \Delta_s(PLQ)\Delta_A^{(j)}(L) +\tilde{\Delta}^{(j)}_s(L)\Delta_R(PLQ) \Bigr] \nonumber\\ &\quad +\tilde{\Delta}^{(j)}_s(L)\Delta_A^{(i)}(P)N_F(PQ) \Bigl[ \Delta_R^{(2)}(PQ)\Delta_R(PLQ) -\Delta_A^{(2)}(PQ)\Delta_A(PLQ) \Bigr] \nonumber\\ &\quad +\Delta_s(PLQ)\Delta_A^{(i)}(P)N_F(PQ) \Bigl[ \Delta_R^{(2)}(PQ)\Delta_A^{(j)}(L) -\Delta_A^{(2)}(PQ)\Delta_R^{(j)}(L) \Bigr] \Bigr\}. \end{align}\tag{108}\] We next rewrite the integrand so that every spectral or difference function depends only on \(P\) or only on \(L\). The second and third lines of Eq. 108 contain only one explicit spectral factor. The second on-shell structure is generated by adding and subtracting advanced propagators, equivalently by introducing difference propagators. For example, for the second line we can write \[\begin{align} \label{rainbow32difference32example} X &= \tilde{\Delta}^{(j)}_s(L)\Delta_A^{(i)}(P)N_F(PQ) \Bigl[ \Delta_R^{(2)}(PQ)\Delta_R(PLQ) -\Delta_A^{(2)}(PQ)\Delta_A(PLQ) \Bigr] \nonumber\\ &= \tilde{\Delta}^{(j)}_s(L)\Delta_A^{(i)}(P)N_F(PQ) \Bigl[ \Delta_d^{(2)}(PQ)\Delta_R(PLQ) +\Delta_A^{(2)}(PQ)\Delta_d(PLQ) \Bigr]. \end{align}\tag{109}\] The same rearrangement applies to the third line of Eq. 108 . After the momentum shifts in Eq. 101 are performed, the rainbow contribution becomes \[\begin{align} \label{rainbow32compact} -i\Sigma^{(2)}_{R_{ij}}(Q) &= e^4 \int_P\int_L \hat{\xi}_{ij}\nonumber\\ &\times\Bigl\{ \Delta_d^{(i)}(P)N_B(P)\Delta_R^{(2)}(PQ) \Bigl[ \Delta_d(L)N_F(L)\Delta_A^{(j)}(L-P-Q)\, \mathcal{S}_{R_{ij}(e_s)} +\Delta_d^{(j)}(L)N_B(L)\Delta_R(PLQ)\, \mathcal{S}_{R_{ij}} \Bigr] \nonumber\\ &\quad +\Delta_d^{(2)}(L)N_F(L)\Delta_A^{(i)}(L-Q) \Bigl[ \Delta_d^{(j)}(P)N_B(P)\Delta_R(PL)\, \mathcal{S}'_{R_{ij}(a_s)} +\Delta_d(P)N_F(P)\Delta_A^{(j)}(P-L)\, \mathcal{S}'_{R_{ij}(a_sc_s)} \Bigr] \nonumber\\ &\quad +\Delta_d(L)\Delta_d^{(j)}(P)N_F(L)N_B(P) \Delta_A^{(i)}(L-P-Q)\Delta_A^{(2)}(L-P)\, \mathcal{S}'_{R_{ij}(f_s)} \Bigr\}. \end{align}\tag{110}\] The on-shell condition \(P^2=L^2=0\) is imposed for terms proportional to \(\Delta_d(P)\Delta_d(L)\). Higher difference propagators, such as \(\Delta_d^{(2)}(L)\) or products of the form \(\Delta_d^{(n)}(P)\Delta_d^{(m)}(L)\), are first converted to spectral functions using Eq. 57 . For example, the metric component of the second line in the equation above becomes \[\begin{align} \label{rainbow32derivative32example} i\int_P \Delta_d(P)\int_L \Delta_d(L) \frac{d}{dl^0} \Biggl\{ \frac{\Delta_A(L-Q)N_F(L)}{2l^0} \Bigl[ N_B(P)\Delta_R(PL)\mathcal{S}'_{R_{m}(a_s)} +N_F(P)\Delta_A(P-L)\mathcal{S}'_{R_{m}(a_sc_s)} \Bigr] \Biggr\}. \end{align}\tag{111}\] The derivative acts only on the \(l^0\)-dependent factors inside the curly bracket. Since no derivative acts on \(p^0\), one may first impose \(P^2=0\). The \(l^0\) derivative is then taken, after which \(L^2=0\) is imposed. Finally, the HTL expansion is carried out. Equivalently, one may take the \(l^0\) derivative, and then impose \(P^2=L^2=0\); the result is unchanged. The same prescription applies to the remaining sectors.

11.2 The Cross-Photon and Bubble Topologies↩︎

For the cross-photon and bubble topologies (see Fig. 6), we obtain the final results as \[\tag{112} \begin{align} &-i\Sigma^{(2)}_{C_{ij}}(Q)=e^4 \int_P\int_L\hat{\xi}_{ij} \nonumber\\ &\times\biggl\{ \Delta_d^{(j)}(L)\Delta_d^{(i)}(P)N_B(L)N_B(P)\Delta_{R}(PQ)\Delta_{R}(PLQ)\Delta_{R}(LQ)\mathcal{S}_{C_{ij}}\nonumber\\ &\quad+\Delta_d(L)\Delta_d(P)N_F(L)N_F(P)\Bigl\{\Delta_A^{(i)}(P-Q)\Delta_A^{(j)}(L-Q)\Delta_A(P+L-Q)\mathcal{S}_{C_{ij}(a_sb_s)}\nonumber\\ & +\Delta_{R}(L-P+Q)\left[\Delta_A^{(i)}(P-Q)\Delta_{R}^{(j)}(L-P)\mathcal{S}_{C_{ij}(a_sc_s)}+\Delta_A^{(j)}(P-Q)\Delta_A^{(i)}(L-P)\mathcal{S}^\prime_{C_{ij}(b_sd_s)}\right]\Bigr\}\nonumber\\ &\quad+\Delta_d(L)\Delta_d^{(i)}(P)N_F(L)N_B(P)\Delta_{R}(PQ)\Bigl\{\Delta_A^{(j)}(L-Q)\Delta_{R}(PL)\mathcal{S}_{C_{ij}(b_s)}+\Delta_A^{(j)}(L-P-Q)\Delta_A(L-P)\mathcal{S}_{C_{ij}(e_s)}\Bigr\}\nonumber\\ &\quad+\Delta_d(L)\Delta_d^{(j)}(P)N_F(L)N_B(P)\Delta_{R}(PQ)\Bigl\{\Delta_A^{(i)}(L-Q)\Delta_A(PL)\mathcal{S}^\prime_{C_{ij}(a_s)}+\Delta_A^{(i)}(L-P-Q)\Delta_{R}(L-P)\mathcal{S}^\prime_{C_{ij}(f_s)}\Bigr\}\biggr\},\\ &-i\Sigma^{(2)}_{B_{ij}}(Q)=-e^4 \int_P\int_L\hat{\xi}_{ij}\Delta_d(L) \nonumber\\ &\times\biggl\{ \Delta_d^{(i+j)}(P)N_F(L)N_B(P)\Delta_R(PQ)\left[\Delta_{R}(PL)\mathcal{S}_{B_{ij}}+\Delta_A(L-P)\mathcal{S}_{B_{ij}(c_s)}\right]\nonumber\\ &\quad +\Delta_d(P)N_F(L)N_F(P)\Bigl[\Delta_{A}^{(i+j)}(P-Q)\left[\Delta_{R}(L-P+Q)\mathcal{S}_{B_{ij}(a_sg_s)}+\Delta_A(P+L-Q)\mathcal{S}_{B_{ij}(a_s)}\right]\nonumber\\ &+\Delta_{R}(P-L+Q)\Delta_{A}^{(i+j)}(P-L)\mathcal{S}_{B_{ij}(d_s)} \Bigr]\biggr\},\tag{113} \end{align}\] where the Dirac structures are \[\begin{align} &\mathcal{S}_{C_{ij}}\equiv\mathcal{A}_i(\cancel{P}+\cancel{Q})\mathcal{B}_j(\cancel{P}+\cancel{L}+\cancel{Q})\mathcal{A}_i(\cancel{L}+\cancel{Q})\mathcal{B}_j,\qquad \mathcal{S}_{B_{ij}}\equiv\mathcal{A}_i(\cancel{P}+\cancel{Q})\mathcal{C}_j\,\mathrm{Tr}\left[\mathcal{A}_i\,\cancel{L}\,\mathcal{C}_j(\cancel{P}+\cancel{L})\right] \end{align}\] The bubble contribution also contains pinch structures at an intermediate stage. After using the KMS relation, the final bubble expressions written above are pinch-free.

Now, consider the purely longitudinal sector, \(i=j=2\). In this case, the second line of Eq. 113 can be rewritten in spectral form with derivatives as, \[\begin{align} \label{D15} \frac{(-i)^5}{3!}\,N_F(L)\Delta_d(L)\Delta_d(P)\Bigl\{2p^0\Bigl[\frac{-1}{2p^0}\frac{d}{dp^0}\Bigr]^3 \Bigl[-\frac{N_B(P)\Delta_R(PQ)}{2p^0}\bigl(\Delta_{R}(PL)\mathcal{S}_{B_{ij}}+\Delta_A(L-P)\mathcal{S}_{B_{ij}(c_s)}\bigr)\Bigr]\Bigr\}. \end{align}\tag{114}\] Since \(\Delta_d(L)\) enforces \(L^2=0\), the contribution above vanishes after evaluation. Therefore, it is useful to impose the on-shell condition associated with the spectral function first and then take the derivatives with respect to the other momentum. The same prescription is used for the remaining sectors and for the vertex corrections.

12 Two-Loop Order Pre-HTL Vertex↩︎

This appendix collects the pinch-free pre-HTL expressions for the fermion–photon vertex. The seven two-loop topologies are shown in Fig. 7. As discussed in Sec. 5.2, the type-II non-planar and self-energy-corrected diagrams are generated from the corresponding type-I diagrams by the exchange \(Q_1\leftrightarrow Q_2\). The bubble and self-energy-corrected contributions contain pinch structures at intermediate stages, which are removed using the KMS relation. We use the following momentum transformations: \[\begin{align} \label{vertex32transformations} &a_v: P\rightarrow P-Q_1,\qquad\quad b_v:L\rightarrow L-Q_2,\quad\qquad c_v:L\rightarrow L-P,\qquad\qquad\qquad\quad d_v:P\rightarrow P-L+Q_2-Q_1,\nonumber\\ &e_v:L\rightarrow L-P-Q_1,\quad f_v:P\rightarrow P-L-Q_1,\quad g_v:L\rightarrow L-P-Q_2+Q_1,\qquad h_v:P\rightarrow P-L,\nonumber\\ &j_v: L\rightarrow L-P-Q_2,\quad k_v:P\rightarrow P-L-Q_2,\quad m_v: P\rightarrow P-Q_2,\quad\qquad\qquad\quad n_v:L\rightarrow L-P+Q_2-Q_1,\nonumber\\ &r_v:L\rightarrow L-P+Q_1,\quad s_v:L\rightarrow L-P+Q_2,\quad t_v: L\rightarrow L-Q_1. \end{align}\tag{115}\] When one of these labels is attached to a Dirac numerator, the corresponding substitution is applied to that numerator. The cross-photon, rainbow, bubble, type-I non-planar, and type-I self-energy-corrected contributions are given below, \[\label{vertex32compact} \begin{align} &-ie\Gamma_{C_{ij}}^{\mu(2)}(Q_2,Q_1)=-ie^5\int_P\int_L \hat{\xi}_{ij}\, \nonumber\\ &\times\Bigl\{\Delta^{(j)}_d(L)\Delta^{(i)}_d(P)N_B(L)N_B(P)\Delta_{R}(PQ_1)\Delta_{R}(PLQ_1)\Delta_{R}(PLQ_2)\Delta_{R}(LQ_2)\mathcal{V}_{C_{ij}}^\mu\nonumber\\ &\quad+\Delta_d(L)\Delta_d(P)N_F(L)N_F(P)\Bigl[ \Delta_A^{(i)}(P-Q_1)\Delta_A(P+L-Q_2)\Delta_A(P+L-Q_1)\Delta_A^{(j)}(L-Q_2)\mathcal{V}_{C_{ij}(a_vb_v)}^\mu\nonumber\\ &+\Delta_A^{(i)}(P-Q_1)\Delta_{R}^{(j)}(L-P)\Delta_A(L+Q_2-Q_1)\Delta_{R}(L-P+Q_2)\mathcal{V}_{C_{ij}(a_vc_v)}^\mu\nonumber\\ &+\Delta_{R}(P-L+Q_2)\Delta_A^{(i)}(P-L+Q_2-Q_1)\Delta_A(P+Q_2-Q_1)\Delta_A^{(j)}(L-Q_2)\mathcal{V}_{C_{ij}(b_vd_v)}^\mu\nonumber\\ &+\Delta_A^{(i)}(P-Q_1)\Delta_{R}^{(j)}(L-P-Q_2+Q_1)\Delta_{R}(L-Q_2+Q_1)\Delta_{R}(L-P+Q_1)\mathcal{V}_{C_{ij}(a_vg_v)}^\mu\nonumber\\ &+\Delta_{R}(P-L+Q_1)\Delta_A^{(i)}(P-L)\Delta_{R}(P-Q_2+Q_1)\Delta_A^{(j)}(L-Q_2)\mathcal{V}_{C_{ij}(b_vh_v)}^\mu\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta^{(j)}_d(P)N_F(L)N_B(P)\Bigl[\Delta_A^{(i)}(L-Q_1)\Delta_A(PL)\Delta_A(P+L+Q_2-Q_1)\Delta_{R}(PQ_2){\mathcal{V}^\prime}_{C_{ij}(a_v)}^\mu\nonumber\\ &+\Delta_A^{(i)}(L-P-Q_1)\Delta_{R}(L-P)\Delta_A(L+Q_2-Q_1)\Delta_{R}(PQ_2){\mathcal{V}^\prime}_{C_{ij}(f_v)}^\mu\nonumber\\ &+\Delta_A^{(i)}(L-P-Q_2)\Delta_{R}(L-P-Q_2+Q_1)\Delta_{R}(L-Q_2+Q_1)\Delta_{R}(PQ_2){\mathcal{V}^\prime}_{C_{ij}(k_v)}^\mu\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta^{(i)}_d(P)N_F(L)N_B(P)\Bigl[ \Delta_{R}(PQ_1)\Delta_A^{(j)}(L-Q_2)\Delta_{R}(P+L-Q_2+Q_1)\Delta_{R}(PL)\mathcal{V}_{C_{ij}(b_v)}^\mu\nonumber\\ &+\Delta_{R}(PQ_1)\Delta_A^{(j)}(L-P-Q_1)\Delta_A(L+Q_2-Q_1)\Delta_A(L-P+Q_2-Q_1)\mathcal{V}_{C_{ij}(e_v)}^\mu\nonumber\\ &+\Delta_{R}(PQ_1)\Delta_{R}(L-Q_2+Q_1)\Delta_A^{(j)}(L-P-Q_2)\Delta_A(L-P)\mathcal{V}_{C_{ij}(j_v)}^\mu\Bigr] \Bigr\},\\ &-ie\Gamma_{R_{ij}}^{\mu(2)}(Q_2,Q_1)=-ie^5\int_P\int_L\hat{\xi}_{ij} \nonumber\\ &\times\Bigl\{\Delta_d^{(j)}(L)\Delta_d^{(i)}(P)N_B(L)N_B(P)\Delta_{R}(P Q_1)\Delta_{R}(PLQ_1)\Delta_{R}(PLQ_2)\Delta_{R}(PQ_2)\mathcal{V}_{R_{ij}}^\mu\nonumber\\ &\quad+\Delta_d(L)\Delta_d(P)N_F(L)N_F(P)\Bigl[\Delta_A^{(i)}(P-Q_1)\Delta_{R}^{(j)}(L-P)\Delta_A(L+Q_2-Q_1)\Delta_A(P+Q_2-Q_1)\mathcal{V}_{R_{ij}(a_vc_v)}^\mu\nonumber\\ &+\Delta_A^{(i)}(P-Q_2)\Delta_{R}(P-Q_2+Q_1)\Delta_A^{(j)}(L-P+Q_2-Q_1)\Delta_A(L+Q_2-Q_1)\mathcal{V}^\mu_{R_{ij}(m_vn_v)}\nonumber\\ &+\Delta_A^{(i)}(P-Q_1)\Delta_{R}^{(j)}(L-P-Q_2+Q_1)\Delta_{R}(L-Q_2+Q_1)\Delta_A(P+Q_2-Q_1)\mathcal{V}^\mu_{R_{ij}(a_vg_v)}\nonumber\\ &+\Delta_{R}(P+Q_1-Q_2)\Delta_A^{(i)}(P-Q_2)\Delta_{R}(L+Q_1-Q_2)\Delta_A^{(j)}(L-P)\mathcal{V}^\mu_{R_{ij}(m_vc_v)}\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta_d^{(j)}(P)N_F(L)N_B(P)\Bigl[\Delta_A^{(i)}(L-Q_1)\Delta_A(PL)\Delta_A(P+L+Q_2-Q_1)\Delta_A(L+Q_2-Q_1){\mathcal{V}^\prime}_{R_{ij}(a_v)}^\mu\nonumber\\ &+\Delta_{R}(L-Q_2+Q_1)\Delta_{R}(P+L-Q_2+Q_1)\Delta_{R}(PL)\Delta_A^{(i)}(L-Q_2){\mathcal{V}^\prime}^\mu_{R_{ij}(m_v)}\nonumber\\ &+\Delta_A^{(i)}(L-P-Q_1)\Delta_{R}(L-P)\Delta_A(L+Q_2-Q_1)\Delta_A(L-P+Q_2-Q_1){\mathcal{V}^\prime}^\mu_{R_{ij}(f_v)}\nonumber\\ &+\Delta_A^{(i)}(L-P-Q_2)\Delta_{R}(L-P-Q_2+Q_1)\Delta_{R}(L-Q_2+Q_1)\Delta_A(L-P){\mathcal{V}^\prime}^\mu_{R_{ij}(k_v)}\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta_d^{(i)}(P)N_F(L)N_B(P)\Bigl[\Delta_{R}(PQ_1)\Delta_A^{(j)}(L-P-Q_1)\Delta_A(L+Q_2-Q_1)\Delta_{R}(PQ_2)\mathcal{V}^\mu_{R_{ij}(e_v)}\nonumber\\ &+\Delta_{R}(PQ_1)\Delta_{R}(L-Q_2+Q_1)\Delta_A^{(j)}(L-P-Q_2)\Delta_{R}(PQ_2)\mathcal{V}^\mu_{R_{ij}(j_v)}\Bigr]\Bigr\},\\ &-ie\Gamma_{B_{ij}}^{\mu(2)}(Q_2,Q_1)=+ie^5\int_P\int_L\hat{\xi}_{ij}\Delta_d(L)\nonumber\\ &\times\Bigl\{\Delta^{(i+j)}_d(P)N_F(L)N_B(P)\Delta_{R}(P Q_1)\Delta_{R}(PQ_2)\left[\Delta_{R}(PL)\mathcal{V}^\mu_{B_{ij}}+\Delta_A(L-P)\mathcal{V}^\mu_{B_{ij}(c_v)}\right]\nonumber\\ &+\Delta_d(P) N_F(L)N_F(P)\Big[\Delta_{R}(P-L+Q_1)\Delta_{R}(P-L+Q_2)\Delta_{A}^{(i+j)}(P-L)\mathcal{V}^\mu_{B_{ij}(h_v)}\nonumber\\ &\quad\qquad\qquad\qquad+ \Delta_A(P+Q_2-Q_1)\Delta_{A}^{(i+j)}(P-Q_1)\Big(\Delta_{R}(L-P+Q_1)\mathcal{V}^\mu_{B_{ij}(a_vr_v)}+\Delta_A(P+L-Q_1)\mathcal{V}^\mu_{B_{ij}(a_v)}\Big)\nonumber\\ &\quad\qquad\qquad\qquad+\Delta_{R}(P-Q_2+Q_1)\Delta_{A}^{(i+j)}(P-Q_2)\Big(\Delta_{R}(L-P+Q_2)\mathcal{V}^\mu_{B_{ij}(m_vs_v)}+\Delta_A(P+L-Q_2)\mathcal{V}^\mu_{B_{ij}(m_v)}\Big)\Big]\Bigr\},\\ &-ie\Gamma_{N_{I_{ij}}}^{\mu(2)}(Q_2,Q_1)=-ie^5\int_P\int_L\hat{\xi}_{ij} \nonumber\\ &\times\Bigl\{ \Delta_d^{(i)}(L)\Delta_d^{(j)}(P)N_B(L)N_B(P)\Delta_{R}(LQ_1)\Delta_{R}(PLQ_1)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2)\mathcal{V}^\mu_{N_{I_{ij}}}\nonumber\\ &\quad+\Delta_d(L)\Delta_d(P)N_F(L)N_F(P)\Bigl[\Delta_{R}^{(j)}(P-L)\Delta_A^{(i)}(L-Q_1)\Delta_{R}(P-L+Q_1)\Delta_{R}(P-L+Q_2)\mathcal{V}^\mu_{N_{I_{ij}}(t_vh_v)}\nonumber\\ &+\Delta_A^{(i)}(L-Q_1)\Delta_A(P+L-Q_1)\Delta_A^{(j)}(P-Q_1)\Delta_A(P+Q_2-Q_1)\mathcal{V}^\mu_{N_{I_{ij}}(t_va_v)}\nonumber\\ &+\Delta_A^{(i)}(L-Q_1)\Delta_A(P+L-Q_2)\Delta_A^{(j)}(P-Q_2)\Delta_{R}(P-Q_2+Q_1)\mathcal{V}^\mu_{N_{I_{ij}}(t_vm_v)}\nonumber\\ &+\Delta_A^{(i)}(L-P)\Delta_{R}(L-P+Q_1)\Delta_A^{(j)}(P-Q_1)\Delta_A(P+Q_2-Q_1)\mathcal{V}^\mu_{N_{I_{ij}}(a_vc_v)}\nonumber\\ &+\Delta_{R}(L-P+Q_2)\Delta_A^{(i)}(L-P+Q_2-Q_1)\Delta_{R}(P-Q_2+Q_1)\Delta_A^{(j)}(P-Q_2)\mathcal{V}^\mu_{N_{I_{ij}}(m_vn_v)}\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta_d^{(i)}(P)N_F(L)N_B(P)\Bigl[ \Delta_{R}(PQ_1)\Delta_A^{(j)}(L-P-Q_1)\Delta_A(L-P+Q_2-Q_1)\Delta_A(L-P){\mathcal{V}^\prime}^\mu_{N_{I_{ij}}(f_v)}\nonumber\\ &+\Delta_{R}(PQ_1)\Delta_{R}(PL)\Delta_A^{(j)}(L-Q_1)\Delta_A(L+Q_2-Q_1){\mathcal{V}^\prime}^\mu_{N_{I_{ij}}(a_v)}\nonumber\\ &+\Delta_{R}(PQ_1)\Delta_{R}(P+L-Q_2+Q_1)\Delta_A^{(j)}(L-Q_2)\Delta_{R}(L-Q_2+Q_1){\mathcal{V}^\prime}^\mu_{N_{I_{ij}}(m_v)}\nonumber\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta_d^{(j)}(P)N_F(L)N_B(P)\Bigl[\Delta_A^{(i)}(L-Q_1)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2)\Delta_A(PL)\mathcal{V}^\mu_{N_{I_{ij}}(t_v)}\nonumber\\ &+\Delta_{R}(L-P)\Delta_A^{(i)}(L-P-Q_1)\Delta_{R}(PQ_1)\Delta_{R}(PQ_2){\mathcal{V}}^\mu_{N_{I_{ij}}(e_v)}\Bigr] \Bigr\},\\ &-ie\Gamma_{S_{I_{ij}}}^{\mu(2)}(Q_2,Q_1)=-ie^5\int_P\int_L \hat{\xi}_{ij}\nonumber\\ &\times\Bigl\{\Delta^{(j)}_d(L)\Delta^{(i)}_d(P)N_B(L)N_B(P)\Delta_{R}^{(2)}(PQ_1)\Delta_{R}(PLQ_1)\Delta_{R}(PQ_2){\mathcal{V}}^\mu_{S_{I_{ij}}}\nonumber\\ &\quad+\Delta_d(L)\Delta_d(P)N_F(L)N_F(P)\Delta_A^{(i)}(P-Q_2)\Delta_{R}^{(2)}(P-Q_2+Q_1)\Delta_A^{(j)}(L-P+Q_2-Q_1){\mathcal{V}}^\mu_{S_{I_{ij}}(m_vn_v)}\nonumber\\ &\quad+\Delta_d(L)\Delta^{(2)}_d(P) N_F(L)N_F(P)\Delta_A(P+Q_2-Q_1)\Delta_A^{(i)}(P-Q_1)\Delta_A^{(j)}(L-P)\mathcal{V}^\mu_{S_{I_{ij}}(a_vc_v)}\nonumber\\ &\quad+\Delta_d(L)\Delta^{(j)}_d(P)N_F(L)N_B(P)\Bigl[\Delta_A^{(i)}(L-P-Q_1)\Delta_A(L-P+Q_2-Q_1)\Delta_{A}^{(2)}(L-P){\mathcal{V}^\prime}^\mu_{S_{I_{ij}}(f_v)}\nonumber\\ &\quad+\Delta_A^{(i)}(L-Q_2)\Delta_{R}^{(2)}(L-Q_2+Q_1)\Delta_{R}(P+L-Q_2+Q_1){\mathcal{V}^\prime}^\mu_{S_{I_{ij}}(m_v)}\Bigr]\nonumber\\ &\quad+\Delta_d(L)\Delta^{(i)}_d(P)N_F(L)N_B(P)\Delta_{R}^{(2)}(PQ_1)\Delta_A^{(j)}(L-P-Q_1)\Delta_{R}(PQ_2){\mathcal{V}}^\mu_{S_{I_{ij}}(e_v)}\nonumber\\ &\quad+\Delta^{(2)}_d(L)\Delta^{(j)}_d(P)N_F(L)N_B(P)\Delta_A^{(i)}(L-Q_1)\Delta_A(L+Q_2-Q_1)\Delta_{R}(PL){\mathcal{V}^\prime}^\mu_{S_{I_{ij}}(a_v)} \Bigr\}, \end{align}\tag{116}\] where the Dirac structures are \[\begin{align}\label{vertex32cross-photon32Dirac} &\mathcal{V}_{C_{ij}}^\mu\equiv\mathcal{A}_i(\cancel{P}+\cancel{Q_1})\mathcal{B}_j(\cancel{P}+\cancel{L}+\cancel{Q_1})\gamma^\mu(\cancel{P}+\cancel{L}+\cancel{Q_2})\mathcal{A}_i(\cancel{L}+\cancel{Q_2})\mathcal{B}_j,\\ &\mathcal{V}_{R_{ij}}^\mu\equiv \mathcal{A}_i(\cancel{P}+\cancel{Q_1})\mathcal{B}_j(\cancel{P}+\cancel{L}+\cancel{Q_1})\gamma^\mu(\cancel{P}+\cancel{L}+\cancel{Q_2})\mathcal{B}_j(\cancel{P}+\cancel{Q_2})\mathcal{A}_i,\\ &\mathcal{V}_{B_{ij}}^\mu\equiv\mathcal{A}_i(\cancel{P}+\cancel{Q_1})\gamma^\mu(\cancel{P}+\cancel{Q_2})\mathcal{C}_j\,\mathrm{Tr}\left[\mathcal{A}_i\,\cancel{L}\,\mathcal{C}_j(\cancel{P}+\cancel{L})\right],\\ &\mathcal{V}^\mu_{N_{I_{ij}}}=\mathcal{D}_i(\cancel{L}+\cancel{Q_1})\mathcal{C}_j(\cancel{P}+\cancel{L}+\cancel{Q_1})\mathcal{D}_i(\cancel{P}+\cancel{Q_1})\gamma^\mu(\cancel{P}+\cancel{Q_2})\mathcal{C}_j,\\ &\mathcal{V}^\mu_{S_{I_{ij}}}=\mathcal{A}_i(\cancel{P}+\cancel{Q_1})\mathcal{B}_j(\cancel{P}+\cancel{L}+\cancel{Q_1})\mathcal{B}_j(\cancel{P}+\cancel{Q_1})\gamma^\mu(\cancel{P}+\cancel{Q_2})\mathcal{A}_i. \end{align}\tag{117}\] Higher difference functions are converted to spectral form using Eq. 57 . When products of spectral and difference functions occur, the derivative-free on-shell constraint is imposed first, as in Eq. 114 .

As an explicit example of the soft-expansion bookkeeping, consider the rainbow contribution with \(i=j=2\). Terms whose denominators depend on a single soft momentum, either \(Q_1\) or \(Q_2\), are left unchanged. Terms containing linear combinations of \(Q_1\) and \(Q_2\) are rewritten in terms of the transfer momentum \(K\equiv Q_2-Q_1\). For the second and third terms, this gives \[\begin{align} \Delta_d(P)\Delta_d(L)N_F(L)N_F(P)&\Bigl[\Delta^2_A(P-Q_1)\Delta_R^2(L-P)\Delta_A(L+K)\Delta_A(P+K)\mathcal{V}_{R_{22}(a_vc_v)}\nonumber\\ &+\Delta_A^2(P-Q_2)\Delta_A(P-K)\Delta^2_A(L-P+K)\Delta_A(L+K)\mathcal{V}_{R_{22}(m_vn_v)}\Bigr], \end{align}\] where the numerators are rewritten as \[\begin{align} &\mathcal{V}_{R_{22}(a_vc_v)}=(\cancel{P}-\cancel{Q}_1)\cancel{P}(\cancel{L}-\cancel{P})\cancel{L}\gamma^\mu (\cancel{L}+\cancel{K})(\cancel{L}-\cancel{P})(\cancel{P}+\cancel{K})(\cancel{P}-\cancel{Q}_1),\nonumber\\ &\mathcal{V}_{R_{22}(m_vn_v)}=(\cancel{P}-\cancel{Q}_2)(\cancel{P}-\cancel{K})(\cancel{L}-\cancel{P}+\cancel{K})\cancel{L}\gamma^\mu (\cancel{L}+\cancel{K})(\cancel{L}-\cancel{P}+\cancel{K})\cancel{P}(\cancel{P}-\cancel{Q}_2). \end{align}\] All other vertex expressions follow the same prescription.

13 Zero-temperature Counterterm Insertions↩︎

13.1 Self-Energy↩︎

The corresponding one-loop diagrams, containing the vertex counterterm and the counterterm insertions on the internal fermion and photon lines, are shown in Fig. 8. Using the \(r/a\) representations of the counterterm lines in Eq. 6 , together with the free \(r/a\) propagators in Eq. 2 , one obtains \[\begin{align} \label{mainCT-Self} &-i\Sigma^{(1)}_{V_{m},{\rm CT}} = 2e^2\delta^{(1)}_2\int_P \Bigl[ \Delta_s(P)\Delta_A(P-Q)\mathcal{S}_{m{(a_s)}} + \tilde{\Delta}_s(P)\Delta_R(PQ)\mathcal{S}_{m} \Bigr]=-2i\delta^{(1)}_2 \Sigma^{(1)}_{m}(Q), \nonumber\\ &-i\Sigma^{(1)}_{V_{\xi},{\rm CT}} = 2ie^2\delta^{(1)}_2(1-\xi)\int_P \Bigl[ \Delta_s(P)\Delta_A^{(2)}(P-Q)\mathcal{S}_{\xi{(a_s)}} + \tilde{\Delta}_s^{(2)}(P)\Delta_R(PQ)\mathcal{S}_{\xi} \Bigr]=-2i\delta^{(1)}_2\Sigma_\xi^{(1)}(Q), \nonumber\\ &-i\Sigma^{(1)}_{f_{m},{\rm CT}} = ie^2\delta^{(1)}_2\int_P \Bigl[ \mathcal{S}_{m{(a_s)}}P^2\Delta_s^{(2)}(P)\Delta_A(P-Q) + \mathcal{S}_{m}(P+Q)^2\tilde{\Delta}_s(P)\Delta_R^{(2)}(PQ) \Bigr], \nonumber\\ &-i\Sigma^{(1)}_{f_{\xi},{\rm CT}} = -e^2\delta^{(1)}_2(1-\xi)\int_P \Bigl[ \mathcal{S}_{\xi(a_s)}P^2\Delta_s^{(2)}(P)\Delta_A^{(2)}(P-Q) + \mathcal{S}_{\xi}(P+Q)^2\tilde{\Delta}_s^{(2)}(P)\Delta_R^{(2)}(PQ) \Bigr], \nonumber\\ &-i\Sigma^{(1)}_{\gamma_{m},{\rm CT}} = -ie^2\delta^{(1)}_3\int_P \Bigl\{ \Delta_s(P)\Delta_A^{(2)}(P-Q) \Bigl[ (P-Q)^2\mathcal{S}_{m(a_s)}-\mathcal{S}_{\xi(a_s)} \Bigr] + \tilde{\Delta}_s^{(2)}(P)\Delta_R(PQ) \Bigl[ P^2\mathcal{S}_{m}-\mathcal{S}_{\xi} \Bigr] \Bigr\}. \end{align}\tag{118}\] Here, the subscripts \(V\), \(f\), and \(\gamma\) label the counterterm insertions at the vertex, on the fermion line, and on the photon line, respectively. The factor of two in the vertex-counterterm contribution accounts for its two possible placements, at the left and right vertices. The Dirac structures \(\mathcal{S}{m}\) and \(\mathcal{S}{\xi}\) are defined in Eq. 92 , and the label \(a_s\) denotes the shift \(P\to P-Q\). Since the photon counterterm tensor is transverse and gauge-parameter independent, there is no separate contribution proportional to \((1-\xi)\). The LP contributions are \[\begin{align} \label{LP-CT-Self} \Sigma^{(1;\rm LP)}_{{V}_{m},CT}=2\delta^{(1)}_2 \mathcal{H}(T),\quad \Sigma^{(1;\rm LP)}_{f_{m},CT}=-\delta^{(1)}_2 \mathcal{H}(T),\quad \Sigma^{(1;\rm LP)}_{f_\xi,CT}=\Sigma^{(1;LP)}_{{V}_{\xi},CT}=0,\quad \Sigma^{(1;\rm LP)}_{\gamma_{m},{\rm CT}}=\delta^{(1)}_3\mathcal{H}(T), \end{align}\tag{119}\] where \(\mathcal{H}(T)\) is the bare one-loop LP contribution, Eq. 11 . The nonvanishing terms are collected in Eq. 25 .

13.2 Vertex↩︎

The one-loop diagrams containing the counterterm insertions are shown in Fig. 9, yielding the contributions \[\begin{align} \label{Vertex-CT-one-loop} &-i e\Gamma_{V_{m},CT}^{\mu(1)}(Q_2,Q_1)=-3ie\,\delta^{(1)}_2\Gamma^{\mu(1)}_{m}(Q_2,Q_1),\quad-ie\Gamma_{V_{\xi},CT}^{\mu}(Q_2,Q_1)=-3ie\,\delta^{(1)}_2\Gamma^{\mu(1)}_{\xi}(Q_2,Q_1),\nonumber\\ &-ie\Gamma^{\mu(1)}_{f_{I_{m}},CT}(Q_2,Q_1)=e^3\delta^{(1)}_2\int_P\Bigl\{(P+Q_1)^2 \Delta^{(2)}_R(PQ_1)\Delta_R(PQ_2)\tilde{\Delta}_s(P)\mathcal{V}_{m}\nonumber\\ &+P^2\Delta^{(2)}_s(P)\Delta_A(P+Q_2-Q_1)\Delta_A(P-Q_1)\mathcal{V}_{m(a_v)}\nonumber\\ &+(P+Q_1-Q_2)^2\Delta_s(P)\Delta^{(2)}_R(P+Q_1-Q_2)\Delta_A(P-Q_2)\mathcal{V}_{m(m_v)}\Bigr\},\nonumber\\ &-ie\Gamma^{\mu(1)}_{f_{I_{\xi}},CT}(Q_2,Q_1)=ie^3\delta^{(1)}_2\int_P(1-\xi)\Bigl\{(P+Q_1)^2 \Delta^{(2)}_R(PQ_1)\Delta_R(PQ_2)\tilde{\Delta}^{(2)}_s(P)\mathcal{V}_{\xi}\nonumber\\ &+P^2\Delta^{(2)}_s(P)\Delta_A(P+Q_2-Q_1)\Delta^{(2)}_A(P-Q_1)\mathcal{V}_{\xi(a_v)}\nonumber\\ &+(P+Q_1-Q_2)^2\Delta_s(P)\Delta^{(2)}_R(P+Q_1-Q_2)\Delta^{(2)}_A(P-Q_2)\mathcal{V}_{\xi(m_v)}\Bigr\},\nonumber\\ &-i e\Gamma_{{\gamma}_m,CT}^{\mu(1)}(Q_2,Q_1)=-e^3\delta^{(1)}_3\int_P\Bigl\{\Bigl[P^2 \mathcal{V}_{m}-\mathcal{V}_\xi\Bigr]\tilde{\Delta}_s^{(2)}(P)\Delta_R(PQ_1)\Delta_R(PQ_2)\nonumber\\ &+\Bigl[(P-Q_1)^2\mathcal{V}_{m(a_v)}-\mathcal{V}_{\xi(a_v)}\Bigr]\Delta_s(P)\Delta_A(P+Q_2-Q_1)\Delta^{(2)}_A(P-Q_1)\nonumber\\ &+\Bigl[(P-Q_2)^2\mathcal{V}_{m(m_v)}-\mathcal{V}_{\xi(m_v)}\Bigr]\Delta_s(P)\Delta_R(P+Q_1-Q_2)\Delta^{(2)}_A(P-Q_2)\Bigr\}. \end{align}\tag{120}\] The vertex counterterm can be inserted at any of the three vertices, and the factor of three accounts for these possible placements. The counterterm insertion on the fermion line, for the second diagram in Fig. 9, is given above. The remaining diagram is obtained by the exchange \(Q_1\leftrightarrow Q_2\), with the corresponding vector and axial projections following from the exchange relations in Eq. 30 . The LP corrections then become \[\begin{align} \label{Vertex-CT-one-loop-LP} &\Gamma_{V_{m},CT}^{\mu(1;\rm LP)}(Q_2,Q_1)=3\,\delta^{(1)}_2\Gamma^{\mu(1;\rm LP)}(Q_2,Q_1),\quad \Gamma^{\mu(1;\rm LP)}_{f_{{\rm I}_{m}},CT}(Q_2,Q_1)+\Gamma^{\mu(1;\rm LP)}_{f_{{\rm{II}}_{m}},CT}(Q_2,Q_1)=-2\delta_2^{(1)}\Gamma^{\mu(1;\rm LP)}(Q_2,Q_1),\nonumber\\ &\Gamma_{\gamma_{m},CT}^{\mu(1;\rm LP)}(Q_2,Q_1)=\delta_3^{(1)}\Gamma^{\mu(1;\rm LP)}(Q_2,Q_1),\quad\Gamma^{\mu(1;\rm LP)}_{f_{{\rm I}_{\xi}},CT}(Q_2,Q_1)=\Gamma^{\mu(1;\rm LP)}_{f_{{\rm II}_{\xi}},CT}(Q_2,Q_1)=\Gamma_{V_{\xi},CT}^{\mu(1;\rm LP)}(Q_2,Q_1)=0, \end{align}\tag{121}\] where \(\Gamma^{\mu(1;LP)}(Q_2,Q_1)\) is given in Eq. 41 . The nonvanishing terms are collected in Eq. 42 .

14 Final Results of Vertex and Self-Energy Contributions↩︎

14.1 Two-Loop Self-Energy Results↩︎

\[\tag{122} \begin{align} &-i\Sigma^{(2;\rm LP)}_{m}=-\frac{ie^4}{4} \int_P\int_L \Delta_d(L)\Delta_d(P)\nonumber\\ &\times\Bigl\{N_B(L) N_B(P) \Bigl[\mathcal{D}^2 \Bigl(-\frac{\cancel{L}}{L.P P.Q}-\frac{Q^2 \cancel{P}}{(P.Q)^3}+\frac{\cancel{Q}}{(P.Q)^2}\Bigr)+\frac{2 \mathcal{D} \, Q^2 \cancel{L}}{L.Q (P.Q)^2}\Bigr]+N_F(L)N_F^\prime(P)\Bigl[\frac{\mathcal{D}^2 \cancel{P}}{{p^0} P.Q}\Bigr]\nonumber\\ &\quad+ N_F(L) N_F(P)\Bigl[\mathcal{D}^2 \Bigl(-\frac{\cancel{L}}{{l^0}^2 L.Q}+\frac{\gamma_0}{{l^0} L.Q}-\frac{{q^0} \cancel{L}}{{l^0} (L.Q)^2}-\frac{\cancel{P}}{L.P L.Q}+\frac{Q^2 \cancel{L}}{(L.Q)^3}\Bigr)\nonumber\\ &+\mathcal{D} \Bigl(-\frac{2 Q^2 \cancel{L}}{L.Q (P.Q)^2}-\frac{4 Q^2 \cancel{P}}{(P.Q)^3}\Bigr)-\frac{8 \cancel{L}}{L.P P.Q}-\frac{8 \cancel{P} L.Q}{L.P (P.Q)^2}+\frac{8 \cancel{Q}}{(P.Q)^2}\Bigr]\nonumber\\ &\quad+ N_F(L)N_B(P)\Bigl[\mathcal{D}^2\Bigl( \frac{\cancel{L}}{{l^0}^2 L.Q}+\frac{\cancel{L}{q^0}}{{l^0} (L.Q)^2}+\frac{\cancel{L}}{L.P P.Q}-\frac{\cancel{L}Q^2}{(L.Q)^3}-\frac{\gamma_0}{{l^0} L.Q}+ \frac{\cancel{P}}{L.P L.Q}+\frac{\cancel{P}Q^2}{(P.Q)^3}-\frac{\cancel{Q}}{(P.Q)^2}\Bigr)\nonumber\\ &+\mathcal{D}\Bigl(\frac{2\cancel{P} Q^2}{(L.Q)^2 P.Q}+\frac{4\cancel{P} }{{p^0}^2 P.Q}+\frac{4\cancel{P} {q^0}}{{p^0} (P.Q)^2}-\frac{4\cancel{P} Q^2}{(P.Q)^3}-\frac{2 Q^2 \cancel{L}}{L.Q (P.Q)^2}-\frac{4 \gamma_0}{{p^0} P.Q}+\frac{4 \cancel{Q}}{(P.Q)^2}\Bigr)\nonumber\\ &+\frac{8\cancel{L}}{L.P P.Q}+\frac{8\cancel{P} L.Q}{L.P (P.Q)^2}-\frac{8\cancel{Q}}{(P.Q)^2}\Bigr]\nonumber\\ &\quad-N_B(P)N_F^\prime(L)\Bigl(\frac{\mathcal{D}^2 \cancel{L}}{{l^0} L.Q}\Bigr)-N_F(L)N_B^\prime(P)\Big(\frac{4\mathcal{D} \cancel{P}}{{p^0} P.Q}\Big)\Bigr\},\tag{123}\\ &-i\Sigma^{(2;\rm LP)}_{\xi}=\frac{ie^4\mathcal{D}}{4} \int_P\int_L(1-\xi) \Delta_d(L)\Delta_d(P)\nonumber\\ &\times\Bigl\{N_B(L) N_B(P)\Bigl[\cancel{P}\Bigl(\frac{1}{{p^0}^2 P.Q}+\frac{{q^0}}{{p^0} (P.Q)^2}-\frac{2 Q^2}{(P.Q)^3}\Bigr)-\frac{\gamma^0}{{p^0} P.Q}\nonumber\\ &+\cancel{Q}\Bigl(\frac{{l^0}}{{p^0} L.Q P.Q}-\frac{L.P}{{p^0}^2 L.Q P.Q}-\frac{{q^0} L.P}{{p^0} L.Q (P.Q)^2}+\frac{Q^2 L.P}{L.Q (P.Q)^3}+\frac{2}{(P.Q)^2}\Bigr)\Bigr]\nonumber\\ &\quad+N_B(L)N_B^\prime(P)\Bigl[\frac{\cancel{Q} L.P}{{p^0} L.Q P.Q}-\frac{\cancel{P}}{{p^0} P.Q}\Bigr]+N_F(L)N_F(P)\Bigl[-\frac{Q^2 \cancel{L}}{L.Q (P.Q)^2}-\frac{Q^2 \cancel{Q} L.P}{(L.Q)^3 P.Q}+\frac{2 \cancel{Q}}{(P.Q)^2}\Bigr]\nonumber\\ &\quad+ N_F(L) N_B(P)\Bigl[\cancel{P}\Bigl(\frac{Q^2}{(L.Q)^2 P.Q}-\frac{1}{{p^0}^2 P.Q}-\frac{{q^0}}{{p^0} (P.Q)^2}+\frac{2 Q^2}{(P.Q)^3}\Bigr)+\frac{\gamma^0}{{p^0} P.Q}\nonumber\\ &+\cancel{Q}\Bigl(-\frac{{l^0}}{{p^0} L.Q P.Q}+\frac{L.P}{{p^0}^2 L.Q P.Q}+\frac{{q^0} L.P}{{p^0} L.Q (P.Q)^2}+\frac{Q^2 L.P}{(L.Q)^3 P.Q}-\frac{Q^2 L.P}{L.Q (P.Q)^3}-\frac{2}{(L.Q)^2}-\frac{2}{(P.Q)^2}\Bigr)\Bigr]\nonumber\\ &\quad+N_F(L)N_B^\prime(P)\Bigl[-\frac{\cancel{Q} L.P}{{p^0} L.Q P.Q}+\frac{\cancel{P}}{{p^0} P.Q}\Bigr]\Bigr\}.\tag{124} \end{align}\]

14.2 One-Loop NLP Vertex↩︎

\[\tag{125} \begin{align}\tag{126} &-ie\Bigl[V_{m}^{\mu\rho(1;\rm NLP)}\Bigr]\gamma_\rho=\frac{ie^3\mathcal{D}}{4}\int_P\Delta_d(P)\nonumber\\ &\times\biggl\{N_F(P) \Bigl\{\cancel{P}\Bigl[\frac{K^4 {P^\mu}}{2 (K.P)^2 P.Q_1 P.Q_2}-\frac{K^2 {K^\mu}}{2 (K.P)^2 P.Q_1}+\frac{K^2 {K^\mu}}{2 (K.P)^2 P.Q_2}-\frac{K^2 {P^\mu} Q_1^2}{2 (K.P)^2 (P.Q_1)^2}-\frac{K^2 {P^\mu} Q_2^2}{2 (K.P)^2 (P.Q_2)^2}+\frac{{K^\mu} Q_1^2}{2 K.P (P.Q_1)^2}\nonumber\\ &+\frac{{K^\mu} Q_2^2}{2 K.P (P.Q_2)^2}+\frac{{P^\mu} Q_1^4}{2 K.P (P.Q_1)^3}-\frac{{P^\mu} Q_2^4}{2 K.P (P.Q_2)^3}\Bigr]+{\gamma}^\mu\Bigl[\frac{K^2}{2 K.P P.Q_1}-\frac{K^2}{2 K.P P.Q_2}-\frac{Q_1^2}{2 (P.Q_1)^2}-\frac{Q_2^2}{2 (P.Q_2)^2}\Bigr]\nonumber\\ &+\cancel{K}\Bigl[-\frac{K^2 {P^\mu}}{2 (K.P)^2 P.Q_1}+\frac{K^2 {P^\mu}}{2 (K.P)^2 P.Q_2}+\frac{{P^\mu} Q_1^2}{2 K.P (P.Q_1)^2}+\frac{{P^\mu} Q_2^2}{2 K.P (P.Q_2)^2}\Bigr]\Bigr\}\nonumber\\ &\quad+N_B(P)\Bigl\{\cancel{P}\Bigl[-\frac{{P^\mu} Q_1^4}{2 (P.Q_1)^3 P.Q_2}-\frac{{P^\mu} Q_1^2 Q_2^2}{2 (P.Q_1)^2 (P.Q_2)^2}-\frac{{P^\mu} Q_2^4}{2 P.Q_1 (P.Q_2)^3}+\frac{Q_1^2 {Q_1^\mu}}{2 (P.Q_1)^2 P.Q_2}+\frac{{Q_1^\mu} Q_2^2}{2 P.Q_1 (P.Q_2)^2}+\frac{Q_2^2 {Q_2^\mu}}{2 P.Q_1 (P.Q_2)^2}\nonumber\\ &+\frac{Q_1^2 {Q_2^\mu}}{2 (P.Q_1)^2 P.Q_2}\Bigr]+{\gamma}^\mu\Bigl[-\frac{Q_1^2}{2 P.Q_1 P.Q_2}-\frac{Q_1^2}{2 (P.Q_1)^2}-\frac{Q_2^2}{2 P.Q_1 P.Q_2}+\frac{Q_1.Q_2}{P.Q_1 P.Q_2}-\frac{Q_2^2}{2 (P.Q_2)^2}\Bigr]\nonumber\\ &+\cancel{Q_2}\Bigl[\frac{{P^\mu} Q_1^2}{2 (P.Q_1)^2 P.Q_2}+\frac{{P^\mu} Q_2^2}{2 P.Q_1 (P.Q_2)^2}-\frac{{Q_1^\mu}}{P.Q_1 P.Q_2}\Bigr]+\cancel{Q_1}\Bigl[\frac{{P^\mu} Q_1^2}{2 (P.Q_1)^2 P.Q_2}+\frac{{P^\mu} Q_2^2}{2 P.Q_1 (P.Q_2)^2}-\frac{{Q_2^\mu}}{P.Q_1 P.Q_2}\Bigr]\Bigr\}\biggr\}. \end{align} \begin{align}\tag{127} &-ie\Bigl[A_{m}^{\mu\rho(1;\rm NLP)}\Bigr]\gamma_\rho\gamma_5=\frac{ie^3}{4}\int_P\Delta_d(P)(D-6)\epsilon ^{\mu \alpha \beta \rho } \gamma_\rho\gamma_5\nonumber\\ &\times\biggl\{-iN_F(P)\Bigl[\frac{K^2 K_\alpha P_\beta}{2 (K.P)^2 P.Q_1}+\frac{K^2 K_\alpha P_\beta}{2 (K.P)^2 P.Q_2}-\frac{K_\alpha P_\beta Q_1^2}{2 K.P (P.Q_1)^2}+\frac{K_\alpha P_\beta Q_2^2}{2 K.P (P.Q_2)^2}\Bigr]\nonumber\\ &\quad-i N_B(P) \Bigl[\frac{P_\alpha Q_1^2 K_\beta}{2 (P.Q_1)^2 P.Q_2}+\frac{P_\alpha Q_2^2 K_\beta}{2 P.Q_1 (P.Q_2)^2}-\frac{{Q_1}_\alpha {Q_2}_\beta}{P.Q_1 P.Q_2}\Bigr]\biggr\}. \end{align}\] \[\tag{128} \begin{align}\tag{129} &-ie\Bigl[V_{\xi}^{\mu\rho(1;\rm NLP)}\Bigr]\gamma_\rho=-\frac{ie^3}{4}\int_P(1-\xi)\Delta_d(P)\nonumber\\ &\times\biggl\{N_F(P)\Bigl\{\cancel{P}\Bigl[\frac{K^2 {P^\mu} {Q_1}^2}{2 (K.P)^2 ({P.Q_1})^2}+\frac{K^2 {P^\mu} K.{Q_1}}{2 (K.P)^2 ({P.Q_1})^2}+\frac{K^2 {P^\mu} {Q_2}^2}{2 (K.P)^2 ({P.Q_2})^2}-\frac{K^2 {P^\mu} K.{Q_2}}{2 (K.P)^2 ({P.Q_2})^2}-\frac{K^2 {Q_1^\mu}}{2 K.P ({P.Q_1})^2}+\frac{K^2 {Q_2^\mu}}{2 K.P ({P.Q_2})^2}\nonumber\\ &-\frac{K^\mu {Q_1}^2}{2 K.P ({P.Q_1})^2}-\frac{K^\mu {Q_2}^2}{2 K.P ({P.Q_2})^2}-\frac{{P^\mu} {Q_1}^4}{K.P ({P.Q_1})^3}-\frac{{P^\mu} {Q_1}^2 K.{Q_1}}{K.P ({P.Q_1})^3}+\frac{{P^\mu} {Q_2}^4}{K.P ({P.Q_2})^3}-\frac{{P^\mu} {Q_2}^2 K.{Q_2}}{K.P ({P.Q_2})^3}+\frac{{Q_1}^2 {Q_1^\mu}}{({P.Q_1})^3}\nonumber\\ &+\frac{{Q_2}^2 {Q_2^\mu}}{({P.Q_2})^3}\Bigr]+{\gamma}^\mu\Bigl[\frac{K^2}{2 K.P {P.Q_1}}-\frac{K^2}{2 K.P {P.Q_2}}-\frac{{Q_1}^2}{2 ({P.Q_1})^2}-\frac{{Q_2}^2}{2 ({P.Q_2})^2}\Bigr]\nonumber\\ &+\cancel{Q_2}\Bigl[-\frac{K^2 {P^\mu}}{2 (K.P)^2 {P.Q_1}}-\frac{K^2 {P^\mu}}{2 (K.P)^2 {P.Q_2}}+\frac{K^\mu}{K.P {P.Q_2}}+\frac{{P^\mu} {Q_1}^2}{2 K.P ({P.Q_1})^2}-\frac{3 {P^\mu} {Q_2}^2}{2 K.P ({P.Q_2})^2}+\frac{{P^\mu} K.{Q_2}}{K.P ({P.Q_2})^2}-\frac{{Q_2^\mu}}{({P.Q_2})^2}\Bigr]\nonumber\\ &+\cancel{Q_1}\Bigl[-\frac{K^2 {P^\mu}}{2 (K.P)^2 {P.Q_1}}-\frac{K^2 {P^\mu}}{2 (K.P)^2 {P.Q_2}}+\frac{K^\mu}{K.P {P.Q_1}}+\frac{3 {P^\mu} {Q_1}^2}{2 K.P ({P.Q_1})^2}+\frac{{P^\mu} K.{Q_1}}{K.P ({P.Q_1})^2}-\frac{{P^\mu} {Q_2}^2}{2 K.P ({P.Q_2})^2}-\frac{{Q_1^\mu}}{({P.Q_1})^2}\Bigr]\Bigr\}\nonumber\\ &\quad +N_B(P)\Bigl\{\cancel{P}\Bigl[-\frac{{Q_1^\mu}}{{p^0}^2 P.Q_1}-\frac{{Q_2^\mu}}{{p^0}^2 P.Q_2}-\frac{{q_1^0} {Q_1^\mu}}{{p^0} (P.Q_1)^2}-\frac{{q_2^0} {Q_2^\mu}}{{p^0} (P.Q_2)^2}+\frac{Q_1^2 {Q_1^\mu}}{(P.Q_1)^3}+\frac{Q_1^2 {Q_2^\mu}}{2 P.Q_1 (P.Q_2)^2}+\frac{{Q_1^\mu} Q_2^2}{2 P.Q_1 (P.Q_2)^2}+\frac{Q_2^2 {Q_2^\mu}}{(P.Q_2)^3}\nonumber\\ &+\frac{{P^\mu} {q_2^0} Q_1.Q_2}{{p^0} P.Q_1 (P.Q_2)^2}-\frac{{P^\mu} Q_1^2 Q_1.Q_2}{2 (P.Q_1)^2 (P.Q_2)^2}-\frac{{P^\mu} Q_2^2 Q_1.Q_2}{2 (P.Q_1)^2 (P.Q_2)^2}-\frac{{P^\mu} Q_2^2 Q_1.Q_2}{P.Q_1 (P.Q_2)^3}+\frac{Q_1^2 {Q_2^\mu}}{2 (P.Q_1)^2 P.Q_2}+\frac{{Q_1^\mu} Q_2^2}{2 (P.Q_1)^2 P.Q_2}\nonumber\\ &-\frac{g^{0\mu} Q_1.Q_2}{{p^0} P.Q_1 P.Q_2}+\frac{{P^\mu} Q_1.Q_2}{{p^0}^2 P.Q_1 P.Q_2}+\frac{{P^\mu} {q_1^0} Q_1.Q_2}{{p^0} (P.Q_1)^2 P.Q_2}-\frac{{P^\mu} Q_1^2 Q_1.Q_2}{(P.Q_1)^3 P.Q_2}\Bigr]\nonumber\\ &+\gamma^0\Bigl[-\frac{{P^\mu} Q_1.Q_2}{{p^0} P.Q_1 P.Q_2}+\frac{{Q_1^\mu}}{{p^0} P.Q_1}+\frac{{Q_2^\mu}}{{p^0} P.Q_2}\Bigr]+{\gamma}^\mu\Bigl[-\frac{Q_1^2}{2 P.Q_1 P.Q_2}-\frac{Q_1^2}{2 (P.Q_1)^2}-\frac{Q_2^2}{2 P.Q_1 P.Q_2}+\frac{Q_1.Q_2}{P.Q_1 P.Q_2}-\frac{Q_2^2}{2 (P.Q_2)^2}\Bigr]\nonumber\\ &+\cancel{Q_2}\Bigl[\frac{{P^\mu} Q_1^2}{2 (P.Q_1)^2 P.Q_2}+\frac{{P^\mu} Q_2^2}{2 P.Q_1 (P.Q_2)^2}-\frac{{Q_1^\mu}}{P.Q_1 P.Q_2}\Bigr]+\cancel{Q_1}\Bigl[\frac{{P^\mu} Q_1^2}{2 (P.Q_1)^2 P.Q_2}+\frac{{P^\mu} Q_2^2}{2 P.Q_1 (P.Q_2)^2}-\frac{{Q_2^\mu}}{P.Q_1 P.Q_2}\Bigr]\Bigr\}\nonumber\\ &\quad+N_B^\prime(P)\cancel{P}\Bigl[-\frac{{P^\mu} Q_1.Q_2}{{p^0} P.Q_1 P.Q_2}+\frac{{Q_1^\mu}}{{p^0} P.Q_1}+\frac{{Q_2^\mu}}{{p^0} P.Q_2}\Bigr]\biggr\}. \end{align} \begin{align}\tag{130} &-ie\Bigl[A_{\xi}^{\mu\rho(1;\rm NLP)}\Bigr]\gamma_\rho\gamma_5=-\frac{ie^3}{4}\int_P(1-\xi)\Delta_d(P)(D-6)\gamma_\rho\gamma_5\nonumber\\ &\times\biggl\{-i N_F(P)\Bigl\{\epsilon ^{\mu \alpha \beta \rho } \Bigl[\frac{K^2 P^\alpha Q_1^\beta}{2 K.P (P.Q_1)^2}+\frac{K^2 P^\alpha Q_2^\beta}{2 K.P (P.Q_2)^2}+\frac{K^\alpha P^\beta Q_1^2}{2 K.P (P.Q_1)^2}-\frac{K^\alpha P^\beta Q_2^2}{2 K.P (P.Q_2)^2}-\frac{K^\alpha Q_1^\beta}{K.P P.Q_1}+\frac{K^\alpha Q_2^\beta}{K.P P.Q_2}+\frac{P^\alpha Q_1^\beta K.Q_1}{K.P (P.Q_1)^2}\nonumber\\ &-\frac{P^\alpha Q_2^\beta K.Q_2}{K.P (P.Q_2)^2}-\frac{P^\alpha Q_1^2 Q_1^\beta}{(P.Q_1)^3}+\frac{P^\alpha Q_2^2 Q_2^\beta}{(P.Q_2)^3}\Bigr]\nonumber\\ &+K^\alpha P^\beta \epsilon ^{\alpha \beta \nu \rho }\Bigl[-\frac{K^2 P^\mu Q_1^\nu}{2 (K.P)^2 (P.Q_1)^2}-\frac{K^2 P^\mu Q_2^\nu}{2 (K.P)^2 (P.Q_2)^2}+\frac{P^\mu Q_1^2 Q_1^\nu}{K.P (P.Q_1)^3}-\frac{P^\mu Q_2^2 Q_2^\nu}{K.P (P.Q_2)^3}-\frac{Q_1^\mu Q_1^\nu}{K.P (P.Q_1)^2}+\frac{Q_2^\mu Q_2^\nu}{K.P (P.Q_2)^2}\Bigr]\Bigr\}\nonumber\\ &-i N_B(P)\Bigl[\epsilon ^{\alpha \beta \nu \rho } \Bigl(\frac{g^{0\alpha}P^\mu Q_1^\beta Q_2^\nu}{{p^0}P.Q_1 P.Q_2}+\frac{g^{0\mu} P^\alpha Q_1^\beta Q_2^\nu}{{p^0}P.Q_1 P.Q_2}-\frac{P^\alpha P^\mu Q_1^\beta Q_2^\nu}{(p^0)^2 P.Q_1 P.Q_2}-\frac{P^\alpha P^\mu q_1^0Q_1^\beta Q_2^\nu}{{p^0}(P.Q_1)^2 P.Q_2}-\frac{P^\alpha P^\mu Q_1^\beta q_2^0Q_2^\nu}{{p^0}P.Q_1 (P.Q_2)^2}\Bigr)\nonumber\\ &+\epsilon ^{\mu \alpha \beta \rho } \Bigl(-\frac{g^{0\alpha}Q_1^\beta}{{p^0}P.Q_1}+\frac{g^{0\alpha}Q_2^\beta}{{p^0}P.Q_2}+\frac{P^\alpha Q_1^\beta}{({p^0})^2 P.Q_1}-\frac{P^\alpha Q_2^\beta}{({p^0})^2 P.Q_2}+\frac{P^\alpha q_1^0Q_1^\beta}{{p^0}(P.Q_1)^2}-\frac{P^\alpha q_2^0Q_2^\beta}{{p^0}(P.Q_2)^2}\Bigr)\Bigr]\nonumber\\ &-iN_B'(P)\Bigl[\frac{\gamma^\rho P^\alpha P^\mu Q_1^\beta Q_2^\nu \epsilon ^{\alpha \beta \nu \rho }}{{p^0}P.Q_1 P.Q_2}+\epsilon ^{\mu \alpha \beta \rho } \Bigl(\frac{P^\alpha Q_2^\beta}{{p^0}P.Q_2}-\frac{P^\alpha Q_1^\beta}{{p^0}P.Q_1}\Bigr)\Bigr]\biggr\}. \end{align}\]

14.3 Two-Loop LP Vertex↩︎

\[\begin{align} \label{NBNB-metricsector-vector-vertex} &-ie\Bigl[V^{\mu\rho(2;\rm LP)}_{m,N_BN_B}\Bigr]\gamma_\rho=-\frac{ie^5}{4}\int_P\int_L\Delta_d(P)\Delta_d(L)N_B(L)N_B(P)\nonumber\\ &\times\biggl\{\mathcal{D}^2\biggl[\cancel{P}\Bigl[-\frac{P^\mu \text{Q1}^2}{(P.Q_1)^3 P.Q_2}-\frac{P^\mu \text{Q1}^2}{2 (P.Q_1)^2 (P.Q_2)^2}-\frac{P^\mu \text{Q2}^2}{2 (P.Q_1)^2 (P.Q_2)^2}-\frac{P^\mu \text{Q2}^2}{P.Q_1 (P.Q_2)^3}+\frac{Q_1^\mu}{2 (P.Q_1)^2 P.Q_2}+\frac{Q_1^\mu}{2 P.Q_1 (P.Q_2)^2}\nonumber\\ &+\frac{Q_2^\mu}{2 (P.Q_1)^2 P.Q_2}+\frac{Q_2^\mu}{2 P.Q_1 (P.Q_2)^2}\Bigr]-\frac{P^\mu \cancel{L}}{L.P P.Q_1 P.Q_2}\nonumber\\ &\qquad\quad+\frac{P^\mu \cancel{Q}_1}{2 (P.Q_1)^2 P.Q_2}+\frac{P^\mu \cancel{Q}_1}{2 P.Q_1 (P.Q_2)^2}+\frac{P^\mu \cancel{Q}_2}{2 (P.Q_1)^2 P.Q_2}+\frac{P^\mu \cancel{Q}_2}{2 P.Q_1 (P.Q_2)^2}+\gamma^\mu \Bigl[-\frac{1}{2 (P.Q_1)^2}-\frac{1}{2 (P.Q_2)^2}\Bigr]\biggr]\nonumber\\ &\quad+\mathcal{D}\biggl[\cancel{L} \Bigl[-\frac{L^\mu}{L.P L.Q_2 P.Q_1}+\frac{L^\mu}{L.P L.Q_1 P.Q_2}+\frac{L^\mu}{L.P L.Q_1 P.Q_1}+\frac{L^\mu}{L.P L.Q_2 P.Q_2}+\frac{P^\mu \text{Q1}^2}{L.Q_1 (P.Q_1)^2 P.Q_2}+\frac{P^\mu \text{Q1}^2}{L.Q_2 (P.Q_1)^2 P.Q_2}\nonumber\\ &\qquad\qquad+\frac{P^\mu \text{Q2}^2}{L.Q_1 P.Q_1 (P.Q_2)^2}+\frac{P^\mu \text{Q2}^2}{L.Q_2 P.Q_1 (P.Q_2)^2}+\frac{2 P^\mu}{L.P P.Q_1 P.Q_2}-\frac{P^\mu L.Q_2}{L.P L.Q_1 P.Q_1 P.Q_2}-\frac{P^\mu L.Q_1}{L.P L.Q_2 P.Q_1 P.Q_2}\nonumber\\ &\qquad\qquad-\frac{Q_1^\mu}{L.Q_1 P.Q_1 P.Q_2}-\frac{Q_1^\mu}{L.Q_2 P.Q_1 P.Q_2}-\frac{Q_2^\mu}{L.Q_1 P.Q_1 P.Q_2}-\frac{Q_2^\mu}{L.Q_2 P.Q_1 P.Q_2}\Bigr]\nonumber\\ &\qquad\quad+\cancel{P}\Bigl[\frac{P^\mu \text{Q1}^2}{(L.Q_1)^2 P.Q_1 P.Q_2}+\frac{P^\mu \text{Q2}^2}{(L.Q_2)^2 P.Q_1 P.Q_2}-\frac{2 P^\mu}{L.P L.Q_2 P.Q_1}\Bigr]\biggr]\biggr\}. \end{align}\tag{131}\] \[\begin{align} \label{NBNB-metricsector-axial-vertex} &-ie\Bigl[A^{\mu\rho(2;\rm LP)}_{m,N_BN_B}\Bigr]\gamma_\rho\gamma_5=-\frac{ie^5}{4}\int_P\int_L\Delta_d(P)\Delta_d(L)(-i)N_B(L)N_B(P)\gamma_\rho\gamma_5\nonumber\\ &\times\Bigl\{\mathcal{D}^2 \epsilon ^{\mu \rho \alpha \beta } \Bigl[\frac{K_\beta P_\alpha}{2 (P.Q_1)^2 P.Q_2}+\frac{K_\beta P_\alpha}{2 P.Q_1 (P.Q_2)^2}\Bigr]+\mathcal{D}\,\epsilon ^{\rho \alpha \beta \nu }\Bigl[\frac{K_\nu L_\alpha L^\mu P_\beta}{L.P L.Q_1 P.Q_1 P.Q_2}+\frac{K_\nu L_\alpha L^\mu P_\beta}{L.P L.Q_2 P.Q_1 P.Q_2}\Bigr]\nonumber\\ &+\mathcal{D}\,\epsilon ^{\mu \rho \alpha \beta } \Bigl[-\frac{K_\beta L_\alpha}{L.Q_1 P.Q_1 P.Q_2}-\frac{K_\beta L_\alpha}{L.Q_2 P.Q_1 P.Q_2}-\frac{2 K_\beta P_\alpha}{(P.Q_1)^2 P.Q_2}-\frac{2 K_\beta P_\alpha}{P.Q_1 (P.Q_2)^2}+\frac{L_\alpha P_\beta L.Q_2}{L.P L.Q_1 P.Q_1 P.Q_2}-\frac{L_\alpha P_\beta L.Q_1}{L.P L.Q_2 P.Q_1 P.Q_2}\Bigr]\Bigr\}. \end{align}\tag{132}\]

References↩︎