Method
Every process follows the same sequence of steps, one notebook each; the processes differ in the integral families and counterterms they need. The squaring is done three times, independently, by FeynCalc, FORM and Tracer; from the integrals onward the same reduction and assembly act on each workflow's squares, and the results are required to agree.
| Step | Notebook | Tools | Output |
|---|---|---|---|
| 1 | Diagrams and amplitudes | FeynArts | plain records: coefficient, open fermion chains, closed loops |
| 2 | Spin and color algebra | FeynCalc | FORM + color.h | Tracer | Born table, virtual (one-loop) interference, real-emission squares |
| 3 | Integrals | reverse unitarity; Kira (IBP reduction); SubTropica (master integrals) | real and virtual contributions as Laurent series in $\epsilon$ |
| 4–5 | Assembly | Mathematica | distributions, MSbar renormalization and factorization, finite coefficients, comparison with the references |
1. Diagrams
FeynArts generates the Born, virtual (one-loop) and real-emission diagrams of each process (closed quark loops counted by $N_f$). They are written once as plain records, a commuting coefficient times open fermion chains and closed fermion loops, from which each algebra workflow computes its squares on its own; in the proton–proton processes the FORM and Tracer notebooks read the FeynCalc squares only at their end, to compare with them.
2. Three independent algebra workflows
| Workflow | Dirac algebra | Color |
|---|---|---|
| FeynCalc | traces and contractions in Mathematica, each amplitude simplified once | SU(N) algebra of FeynCalc, once per color network |
| FORM | one FORM job per product $M_i M_j^*$; Mathematica writes the diagrams as color structures times Lorentz parts | the FORM package color.h, before the Dirac algebra |
| Tracer | the Tracer package, in a kernel without FeynCalc | an explicit SU(N) reduction written in the notebook (traces, Fierz identity, closed loops) |
Spin densities, for an incoming quark of momentum $k$ and an outgoing quark of momentum $p$: $\tfrac12\not k$ for an unpolarized incoming quark and $\not p$ for the spin sum of an outgoing one, $\tfrac12\not k\gamma_5$ for the helicity difference, $\tfrac12\not k\gamma_5\not s_T$ for a transversely polarized incoming quark, and the projection $\not p\gamma_5\not s_T'$ on an observed transversely polarized quark (antiquarks with the charge-conjugated matrices). Gluons have physical polarization sums with a reference vector $n$, $-g^{\mu\nu}+(k^\mu n^\nu+n^\mu k^\nu)/(k\cdot n)$, averaged with $1/(D-2)$ for an unpolarized incoming gluon, and $i\epsilon^{\mu\nu\alpha\beta}k_\alpha n_\beta/(2k\cdot n)$, with $n$ lightlike, for the helicity difference of an incoming gluon; for selected processes gauge invariance is checked by changing the reference vectors. $\gamma_5$ follows the HVBM scheme ('t Hooft and Veltman; Breitenlohner and Maison): it anticommutes with the four-dimensional and commutes with the $(D-4)$-dimensional Dirac matrices, and the hatted components of loop and unobserved momenta are kept.
In the proton–proton processes, at the end of step 2 each workflow's squares are compared with the FeynCalc ones in exact rational arithmetic at two random points of all variables, with a separate random value assigned to each scalar integral. These are sampled comparisons of linear combinations of integral coefficients, not coefficientwise symbolic identities. In every process the finite coefficients of the three workflows are compared at the end: symbolically for the lepton–proton and TMD calculations, at test points for proton–proton collisions.
3. Integrals
Two programs do the integration. Kira uses integration-by-parts identities to reduce both virtual loop integrals and, through reverse unitarity, real-emission phase-space integrals to a set of master integrals per family. SubTropica then evaluates each master analytically from its Feynman-parameter (Euler-integral) representation, with the integrator HyperIntica, and returns its Laurent series in $\epsilon$ to the order needed. In the collinear processes the masters that are Gamma or hypergeometric functions are also compared with their closed forms.
Real emission: reverse unitarity
At fixed observed momentum $p$, the real emission is integrated over the unobserved massless pair with total momentum $P_X$ and invariant mass $P_X^2=s_4=sv(1-w)$, where $P_X=k_a+k_b-p$ for parton–parton and $P_X=l+k-p$ for lepton–parton scattering: $$\int d\Phi_2^{(D)}\,F(r)=(2\pi)^{2-D}\int d^Dr\,\delta_+(r^2)\,\delta_+\big((P_X-r)^2\big)\,F(r),\qquad \delta_+(k^2)=\theta(k^0)\,\delta(k^2).$$ Reverse unitarity writes each on-shell constraint as a cut propagator, $$\delta(X)=\frac{1}{2\pi i}\Big(\frac{1}{X-i0}-\frac{1}{X+i0}\Big),$$ so that the phase-space integrals become cut integrals, reduced by integration-by-parts identities like loop integrals; the positive-energy conditions are kept in the definition of the cut masters. Partial fractions rewrite the integrand as a sum of terms with at most two linearly independent ordinary denominators, so each integral family has the two cut propagators and two ordinary denominators (six families in proton–proton collisions, three in lepton–quark scattering); integrals with a nonpositive power of a cut propagator vanish. Kira generates the reduction identities at a symbolic integration dimension $d$, evaluated at $d=D$ for ordinary integrals and at the shifted dimensions below for integrals with evanescent numerators. The masters are the two-body phase-space volume and angular integrals with both ordinary denominators; up to prefactors, the angular masters are values of one dimension-dependent function at different cross ratios. Their singular terms take the form $(1-w)^{-1-\lambda\epsilon}B_\lambda(w,\epsilon)$ with $B_\lambda$ regular at $w=1$, and are expanded in distributions in step 4.
Virtual corrections
The virtual integrals reduce to an off-shell massless bubble in lepton–quark scattering, and to massless bubbles and boxes in the hadronic processes, written in the three box families (s,t), (s,u), (t,u). The bubbles are Euler integrals; the boxes are evaluated with SubTropica from their Feynman-parameter representation in the Euclidean region and continued with the Feynman prescription to the physical region $s>0$ before the real part of the interference is taken. In the hadronic processes the loop normalization $2/(4\pi)^{D/2}$ is derived from the FeynArts loop measure and checked twice: the coefficient of $1/\epsilon^2$ must equal the universal $-\frac{\alpha_s}{2\pi}\sum_i C_i$ times the Born term, with $C_i=C_F$ for a quark and $C_A$ for a gluon summed over the external partons $i$, and the poles must cancel against the real emission.
HVBM: the dimension shift
In the HVBM scheme a $D$-dimensional momentum $q$ is split into a four-dimensional part and a $(D-4)$-dimensional part $\widehat q$; the external momenta are four-dimensional. After the traces, the squared matrix elements therefore contain powers of $\widehat q^{\,2}$ for the loop momentum and for the unobserved momenta of the real emission, in addition to the ordinary propagators. Such evanescent numerators are removed by raising the dimension of the integral. For a scalar integrand $F(q)$ with four-dimensional external momenta, and with $\mu_q^2=-\widehat q^{\,2}$ in the mostly-minus metric, the normalized one-loop identity is
$$\int\frac{d^Dq}{i\pi^{D/2}}\,(\mu_q^2)^n F(q)=(-\epsilon)_n\int\frac{d^{D+2n}q}{i\pi^{(D+2n)/2}}\,F(q), \qquad D=4-2\epsilon,$$where $n$ is a nonnegative integer and $(-\epsilon)_n=\Gamma(n-\epsilon)/\Gamma(-\epsilon)$. For an insertion $(\widehat q^{\,2})^n$ the coefficient of the shifted integral is $(-1)^n(-\epsilon)_n$, equal to $+\epsilon$ for $n=1$. For the two-body phase-space measure of the real emission above, the corresponding factor is $(4\pi)^n(-1)^n(-\epsilon)_n$.
Only the integral moves to $D+2n$ dimensions: its IBP coefficients and master integrals are evaluated at $d=D+2n$, while $D=4-2\epsilon$ is kept in the trace coefficients, spin projectors and spin averages. Shifting those as well would change, for instance, the $\delta(1-w)$ coefficient of the helicity difference by a rational term proportional to the Born. The evanescent numerator factors are retained because their products with poles can remain finite.
Transverse spins
The momenta $k_a$, $k_b$ of the two incoming partons and $p$ of the observed parton span the scattering plane (the code calls the incoming momenta $p_a$, $p_b$). Spin vectors are decomposed on in-plane axes $X$ and on the normal $N\propto\epsilon^{\mu k_a k_b p}$, with $N^2=-1$; the squares are computed as the components XX, NN, XN, NX, and the mixed ones must vanish.
The normal components of the unobserved momenta and of the loop momentum enter only through $(N\cdot q)^2$, where $q$ is the momentum in question. Write $q=q_\parallel+c_N N+\widehat q$, with $q_\parallel$ in the scattering plane, $c_N$ a number and $\widehat q$ the $(D-4)$-dimensional part introduced above. Since $N$ is orthogonal to the plane and to the $(D-4)$-dimensional directions, $q^2=q_\parallel^2-c_N^2+\widehat q^{\,2}$, so that
$$(N\cdot q)^2=c_N^2=q_\parallel^2-q^2+\widehat q^{\,2}.$$The in-plane part is fixed by the three scalar products of $q$ with $k_a$, $k_b$ and $p$. Writing $p_1=k_a$, $p_2=k_b$, $p_3=p$ and $G_{ij}=p_i\cdot p_j$ for the $3\times3$ table of their scalar products (the Gram matrix),
$$q_\parallel^2=\sum_{i,j=1}^{3}(q\cdot p_i)\,\big(G^{-1}\big)_{ij}\,(q\cdot p_j).$$For the unobserved partons of the real emission $q^2=0$; for the loop momentum, $q^2$ and the products $q\cdot p_i$ are combinations of the propagators, and $\widehat q^{\,2}$ is handled by the dimension shift above. Odd powers of $N\cdot q$ vanish: in the phase-space integral by reflection through the scattering plane, and in the loop because its tensor integrals can only be built from $k_a$, $k_b$ and $p$. No new integrals appear.
4. Distributions and counterterms
Near threshold the real emission is expanded in distributions, $$(1-w)^{-1-\lambda\epsilon} = -\frac{\delta(1-w)}{\lambda\epsilon} + \Big[\frac{1}{1-w}\Big]_+ - \lambda\epsilon\Big[\frac{\ln(1-w)}{1-w}\Big]_+ + \dots,$$ with $\lambda$ the exponent of the singular term, the virtual correction is renormalized in MSbar, and the collinear singularities are removed with the $D$-dimensional Born table (its order-$\epsilon$ terms included) and the unpolarized, helicity or transversity kernels, e.g. $P^T_{qq}=C_F\big[\frac{2z}{(1-z)_+}+\frac32\delta(1-z)\big]$ ($\delta P_{qq}$ in the code). For the helicity difference the finite term $-(\alpha_s/2\pi)\,\Delta f_{qq}\otimes\sigma_{LL}^{(0)}$, with $\Delta f_{qq}(z)=-4C_F(1-z)$, is added on each longitudinally polarized incoming quark or antiquark; it restores the helicity convention of the polarized MSbar parton distributions. Transversity needs no such term. Double and single poles must cancel exactly.
5. Checks
- the three workflows: equal finite coefficients (with separate reduction and finite assembly for each route), and for proton–proton collisions equal squares (exact arithmetic at random rational points);
- for selected processes, gauge invariance of the real-emission squares under a change of the gluon reference vectors, and Ward identities of the TMD currents;
- the universal double pole of the virtual correction, and the cancellation of all poles;
- the scale dependence: the derivatives with respect to the factorization scales equal the Born terms convolved with the leading-order splitting kernels;
- helicity conservation in the pure-annihilation channel $q_i\bar q_i\to q_k+X$, $k\ne i$: $\sigma_{LL}=-\sigma$ in each distribution sector;
- in the TMD calculation, the cancellation of the rapidity-scale dependence between the beam and soft functions;
- independent results: the Fortran codes of Aversa et al. (UU), Jäger, Schäfer, Stratmann, Vogelsang (LL) and Mukherjee, Stratmann, Vogelsang (TT), and the published unpolarized and longitudinal lepton–proton and TMD results;
- the generalized crossing symmetry between the double-transverse cross sections and the transverse spin transfer.
Every check is a Require in the notebook that performs it: a run that does not pass it stops.