qT Subtraction Formalism Overview

• q_T subtraction formalism is a non-local method that extracts singular behavior at small transverse momentum by leveraging the fixed-order expansion of resummed spectra.
• It decomposes higher-order cross sections into a hard Born-level component and a finite q_T+jet remainder, ensuring robust predictions for both colourless and coloured processes.
• The approach has been extended to incorporate electroweak effects, mixed QCD×QED corrections, and even N³LO calculations, showcasing its versatility in high-energy physics.

The $q_T$ subtraction formalism is a non-local subtraction method for higher-order perturbative calculations in hadron collisions. It is designed for processes in which the observed hard system $F$ is produced with small transverse momentum $q_T$ at Born level, so that the singular structure of the $q_T\to 0$ limit can be extracted from transverse-momentum resummation and used as a fixed-order counterterm. In its standard form, the method was formulated for high-mass colourless final states and later generalized to heavier and more elaborate settings, including heavy-quark production, electroweak corrections, and mixed QCD$\otimes$QED contributions. Its practical power lies in the decomposition of a higher-order cross section into a hard contribution at Born kinematics and a finite remainder obtained by subtracting from the $F+$jet contribution a counterterm that reproduces the full small-$q_T$ singular behaviour (Cieri et al., 2018, Buonocore, 2019).

1. Formal definition and master decomposition

The basic setup considers reactions of the form

$h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$

where $F$ is the observed high-mass system, $q=\sum_i q_i$ is its total momentum, $F$0 its invariant mass, and $F$1 its transverse momentum with respect to the beam axis. For colour-singlet production, $F$2 at Born level. At any perturbative order $F$3 with $F$4, the region $F$5 is identified with the production of $F$6 together with QCD radiation, so the only genuinely new singular behaviour is localized in the limit $F$7 (Cieri et al., 2018).

The standard master formula is

$F$8

Here $F$9 is the Born cross section, $q_T$0 is the hard-collinear coefficient, $q_T$1 is the real-radiation contribution at one order lower, and $q_T$2 is the subtraction counterterm derived from the fixed-order expansion of the resummed small-$q_T$3 spectrum. The square bracket is finite as $q_T$4, although its two terms are separately divergent (Buonocore, 2019).

This decomposition is the defining feature of the formalism. The counterterm is not introduced by a process-by-process local approximation in the full radiation phase space; it is inherited from the universal logarithmic structure of transverse-momentum resummation. In that sense, $q_T$5 subtraction is a subtraction method built directly from resummation theory rather than an independent local partition of unresolved limits (Cieri et al., 2018).

2. Small-$q_T$6 factorization and universal ingredients

The subtraction counterterm is generated by the small-$q_T$7 factorization formula in impact-parameter space. In the notation used for Higgs production, the singular component of the spectrum is obtained from a Bessel transform involving the Sudakov form factor

$q_T$8

with perturbative coefficients $q_T$9 and $q_T\to 0$0. The singular kernel $q_T\to 0$1 embodies all terms of the form $q_T\to 0$2 that are divergent as $q_T\to 0$3, whereas $q_T\to 0$4 collects the $q_T\to 0$5 terms and the finite hard-virtual contribution (Cieri et al., 2018).

For gluon-initiated Higgs production, the resummed hard-collinear structure is organized in terms of a process-dependent hard factor $q_T\to 0$6, universal hard-collinear coefficient functions $q_T\to 0$7, and gluon helicity-flip functions $q_T\to 0$8. Their perturbative combination defines the full hard coefficient $q_T\to 0$9. This split makes explicit which ingredients are universal for a class of initial states and which remain process dependent (Cieri et al., 2018).

The same architecture persists in the mixed QCD$\otimes$0QED extension. At $\otimes$1, the counterterm contains basis functions $\otimes$2 with coefficients assembled from mixed Sudakov coefficients $\otimes$3, $\otimes$4, mixed splitting kernels $\otimes$5, mixed collinear functions $\otimes$6, and the process-dependent hard coefficient $\otimes$7. The highest mixed logarithm is controlled by

$\otimes$8

which makes the overlap of one-loop QCD and one-loop QED Sudakov structure explicit (Cieri et al., 2020).

A central structural point is that the counterterm is scheme independent order by order even though the individual resummation ingredients are not. The formalism therefore organizes infrared singular behaviour through universal building blocks while isolating the genuinely process-dependent finite virtual information in the hard function (Cieri et al., 2018, Cieri et al., 2020).

3. Domain of applicability and methodological extensions

The original domain of the formalism is the production of high-mass colourless systems, such as vector bosons, Higgs bosons, dilepton systems, dibosons, and diphotons. Within that class, the method became a standard framework for fully differential predictions through NNLO and beyond (Buonocore, 2019, Cieri et al., 2018).

A major extension was the application to a coloured high-mass final state, namely top-quark pair production. In that case the observed system is $\otimes$9, and the counterterm requires genuinely new ingredients absent in colourless production: soft anomalous dimension matrices in colour space, nontrivial dependence on the heavy-quark kinematics, and azimuthal correlations of the heavy-quark pair. The NLO counterterm contains an additional soft term proportional to the matrix element of the first-order soft anomalous dimension $F+$0 acting on the Born amplitude. This extension yielded a fully differential NLO calculation and the flavour off-diagonal NNLO channels, with numerical validation against MCFM and Top++ (Bonciani et al., 2015).

A second line of development concerns electroweak corrections. For neutral-current Drell–Yan production of a massive lepton pair, the small-$F+$1 singular structure relevant to NLO EW corrections can be obtained by abelianisation of the known QCD formulas. In this setting, photon radiation from final-state massive leptons is not collinearly singular, because the lepton mass regulates the collinear divergence, but it still produces a nontrivial soft structure that modifies the power-suppressed behaviour of the subtraction method. The implementation was validated against an independent Catani–Seymour subtraction calculation using Recola for virtual amplitudes (Buonocore, 2019).

The mixed QCD$F+$2QED generalization extends the same logic to the first mixed order $F+$3. The explicit application studied an off-shell $F+$4 boson decaying into neutrinos, which avoids final-state QED radiation and isolates the initial-state mixed structure. At this order, “jet” includes quarks, antiquarks, gluons, and photons, and the formalism requires mixed anomalous dimensions, mixed splitting kernels, and mixed collinear functions. The corresponding implementation was validated by comparison with analytic inclusive results, with channel decompositions including $F+$5, $F+$6, $F+$7, and $F+$8 initial states (Cieri et al., 2020).

These developments establish that the formalism is not limited to its original colour-singlet QCD setting, although each extension demands a corresponding generalization of the small-$F+$9 resummation structure from which the subtraction counterterm is built (Bonciani et al., 2015, Buonocore, 2019, Cieri et al., 2020).

4. Technical slicing and power-suppressed terms

Operationally, the method is implemented with a technical slicing parameter. Defining

$q_T$0

or, in some applications, $q_T$1, one introduces a cutoff $q_T$2 and evaluates the subtracted real contribution only above that threshold: $q_T$3 The exact result is recovered only in the limit $q_T$4. At finite $q_T$5, the cross section misses power-suppressed terms from the region below the cut (Buonocore, 2019, Buonocore et al., 2021).

For generic colour-singlet processes and sufficiently inclusive observables, the residual cutoff dependence is expected to be quadratic,

$q_T$6

possibly with logarithmic enhancements at higher orders. This quadratic behaviour is the basis for the usual numerical robustness of the method (Buonocore et al., 2021).

That expectation is not universal. In NLO EW Drell–Yan with massive final-state leptons, a detailed analytical study showed that the dominant linear behaviour originates from final-state radiation rather than initial-state radiation. For the inclusive partonic cross section, the final-state-radiation contribution contains a next-to-leading-power term linear in $q_T$7, whereas the initial-state-radiation contribution remains quadratic up to logarithmic enhancements. The linear term arises because the angular-integrated final-state-radiation matrix element behaves as $q_T$8 in the soft limit, while the corresponding initial-state-radiation contribution is finite as $q_T$9 (Buonocore, 2019).

This distinction is practically important. Poor $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$0 convergence can reflect a structural property of the observable and of the unresolved radiation, not merely a numerical deficiency of the implementation. The NLO EW study therefore added and subtracted a local soft term below the cutoff, using the mapping of Buonocore–Nason–Tramontano, so that the final-state-radiation-induced linear term is cancelled and only quadratic residual dependence remains (Buonocore, 2019).

The same logic clarifies a common misconception. A logarithmically unstable fixed-order prediction for observables directly sensitive to $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$1, such as the pair transverse momentum itself or small $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$2, is not a failure of subtraction; it indicates the need for transverse-momentum resummation (Cieri et al., 2020).

5. Fiducial cuts, recoil sensitivity, and linear power corrections

A separate and especially important source of linear power corrections arises from fiducial cuts on two-body final states. Transverse-momentum cuts on undistinguished particles in two-body kinematics induce enhanced sensitivity to low momentum scales because, at Born level, the two particles are back-to-back in the transverse plane and have equal transverse momenta. If the cuts are imposed symmetrically on both particles, or asymmetrically on the leading and subleading particle ordered by $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$3, an arbitrarily small recoil can change whether an event passes or fails the cuts. The acceptance then becomes non-analytic in the soft recoil scale, and the residual $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$4 dependence is degraded from quadratic to linear (Buonocore et al., 2021).

Within $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$5 subtraction, the concrete origin of this effect is a mismatch in the evaluation of the fiducial cuts below the slicing boundary. The standard counterterm is evaluated on Born kinematics $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$6, whereas the exact real-radiation contribution would populate a recoiling configuration in which the colour-singlet system carries transverse momentum $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$7. If the cuts are recoil sensitive, that difference matters already at linear order. One remedy is to add the missing linear power correction explicitly through

$h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$8

where $h_1(p_1)+h_2(p_2)\to F(\{q_i\})+X,$9 is a recoiled Born phase space obtained by boosting the Born-level final state from the Collins–Soper rest frame to a frame in which the colour-singlet system carries transverse momentum $F$0. After this correction, the residual cutoff dependence is restored to $F$1 when the linear term is of purely kinematic origin (Buonocore et al., 2021).

A closely related formulation introduces a fiducial-power-correction term

$F$2

so that the fiducial cross section becomes the standard $F$3-subtraction result above the cut plus an unresolved recoil correction below the cut. In the Drell–Yan implementation in DYTurbo, the recoil prescription assigns transverse momentum $F$4 to each incoming parton, and the resulting predictions agree at the per mille level with local subtraction formalisms up to NNLO (Camarda et al., 2021).

Phenomenologically, the mechanism is sharp. In neutral-current Drell–Yan with symmetric lepton cuts $F$5 and $F$6, the standard fiducial cross section shows a clear linear dependence on $F$7, while the recoil-corrected result becomes quadratic or smaller and residual effects fall below the per-mille level. At NNLO, channel-by-channel comparisons of the $F$8, $F$9, $q=\sum_i q_i$0, and $q=\sum_i q_i$1 contributions show the same pattern, and the extrapolated results agree well with NNLOJET antenna-subtraction results (Buonocore et al., 2021).

The phenomenon is not limited to the equal-cut case. Leading/subleading asymmetric cuts are still imposed on “undistinguished” particles, since a tiny recoil can swap which particle is leading, so their $q=\sum_i q_i$2 dependence remains linearly sensitive to $q=\sum_i q_i$3. By contrast, staggered cuts on physically identified particles, such as separate thresholds for the positively charged and negatively charged leptons, eliminate the problem. A threshold difference $q=\sum_i q_i$4 is already sufficient to remove the linear dependence in the relevant range, consistent with explicit analyses showing that linear power corrections are absent for most of phase space as long as $q=\sum_i q_i$5 (Buonocore et al., 2021).

The same recoil logic applies successfully to on-shell $q=\sum_i q_i$6 production with symmetric cuts on the two vector bosons, but it does not provide a universal cure for diphoton production. There, smooth-cone isolation introduces additional linear power corrections independent of the recoil-driven ones, and in realistic $q=\sum_i q_i$7 production the isolation effect dominates. The $q=\sum_i q_i$8 channel is an interesting exception, for which the recoil effects account for the observed linear term both at NLO and NNLO, but the general diphoton problem remains controlled by isolation rather than recoil (Buonocore et al., 2021).

Taken together, these results support a key conceptual distinction: not all linear power corrections in $q=\sum_i q_i$9 subtraction are alike. Recoil-driven linear terms induced by fiducial cuts on undistinguished particles can be repaired by a suitable kinematic mapping; isolation-driven terms cannot (Buonocore et al., 2021, Camarda et al., 2021).

6. Higher orders and beyond-leading-power structure

The formalism has been pushed to N$F$00LO QCD for Higgs production in gluon fusion in the large-top-mass limit. In that extension, the $F$01 region is supplied by the NNLO Higgs-plus-jet calculation, while the singular counterterm requires the fixed-order expansion of the resummation formula through third order. The relevant ingredients include $F$02, $F$03, the third-order hard-collinear functions $F$04, the second-order helicity-flip functions $F$05, and the third-order hard-virtual coefficient $F$06. In the 2018 implementation, $F$07, $F$08, and part of $F$09 were not known analytically, so the unknown $F$10 contribution was approximated by a constant $F$11, fixed from the known inclusive N$F$12LO total cross section through the unitarity relation of the $F$13 spectrum (Cieri et al., 2018).

This approximation affects only the $F$14 Born-like contribution, not the finite-$F$15 counterterm. For the Higgs rapidity distribution, the procedure was tested at NNLO through an analogous $F$16 approximation and found to reproduce the exact result to better than $F$17 over $F$18. In the N$F$19LO application, $F$20 values of $F$21, $F$22, and $F$23 GeV were used, with residual dependence at the per-mille level for the total cross section and negligible dependence in the central rapidity region of the rapidity distribution (Cieri et al., 2018).

A complementary development concerns the systematic construction of the small-$F$24 spectrum beyond leading power. An algorithm was introduced to derive the NLO $F$25 distribution of colour-singlet production to arbitrary power in $F$26, including suitable zero-bin subtractions for a generic class of rapidity regulators. The explicit Higgs analysis reached NNLP accuracy and showed that boundary contributions from the phase-space endpoints are indispensable: without them, any claimed subleading-power accuracy is lost as $F$27 (Ferrera et al., 2023).

The resulting asymptotic expansion has the structure

$F$28

At LP, the overlap subtraction reduces to the ordinary soft limit. Beyond LP, however, the correct zero-bin subtraction is not in general the same-power soft limit; it requires a dual-scaling construction that keeps the appropriate soft expansion after the collinear expansion. This refines the usual leading-power regions analysis and makes the beyond-leading-power $F$29 expansion regulator generic (Ferrera et al., 2023).

Numerically, the NNLP-accurate Higgs spectra reproduce the exact NLO $F$30 distribution much better than previously available results, and the NNLP approximation can approximate the exact spectrum up to $F$31 at the percent level for rapidities $F$32. This suggests that higher-power information can be organized into improved counterterms or improved unresolved-bin corrections, thereby reducing finite-cutoff artifacts in $F$33-subtraction calculations (Ferrera et al., 2023).

Across these higher-order and beyond-leading-power developments, the formalism retains the same basic identity: the infrared problem is reduced to a universal small-$F$34 structure plus a finite hard contribution at $F$35. What changes at higher precision is the amount of control required over the structure of power corrections, recoil effects, and endpoint contributions (Cieri et al., 2018, Ferrera et al., 2023).