S-MC@NLO: Sherpa Matching in NLO QCD

• S-MC@NLO is a matching algorithm that combines fixed-order NLO QCD calculations with parton shower resummation to accurately simulate hard emissions.
• It employs a modified Catani–Seymour subtraction method to remove double counting between NLO real emissions and shower approximations, ensuring precise normalization.
• Its application in W+n-jet production demonstrates robust phenomenology, effectively handling negative event weights and validating predictions against experimental data.

S-MC@NLO is the Sherpa implementation of MC@NLO-style matching of next-to-leading-order QCD calculations to a parton shower, introduced in the study of $W+n$-jet production with $n=0,1,2,3$. Its purpose is to combine exact NLO normalization and the first hard emission with parton-shower resummation of subsequent soft and collinear radiation, while preventing the double counting that would arise if the NLO real-emission term and the shower were added naively (Siegert et al., 2012).

1. Physical motivation and scope

The algorithm addresses the mismatch between two standard calculational tools. Fixed-order NLO calculations contain the Born contribution $\mathcal{B}$, virtual correction $\mathcal{V}$, and real-emission correction $\mathcal{R}$, and are exact at a given perturbative order, but they become unreliable in regions with large logarithms, such as soft or collinear emission. Parton showers resum leading soft and collinear logarithms to all orders, but only in approximated form, using the collinear limit, the large-$N_C$ approximation, and approximate branching probabilities (Siegert et al., 2012).

In the shower picture, the real-emission matrix element factorizes in the collinear limit into a Born term times a shower splitting kernel, and this factorized structure is used to define the no-branching probability, or Sudakov factor, $\Delta^{\rm(PS)}(t)$. The unresolved contribution corresponds to no emission above the shower cutoff, while the resolved contribution corresponds to one shower emission. This suggests the basic logic of S-MC@NLO: exact fixed-order control of normalization and first hard radiation must be combined with a Sudakov-resummed description of subsequent radiation (Siegert et al., 2012).

A complementary formulation emphasizes that the shower approximation itself is an $\mathcal O(\alpha_s)$ contribution that must be removed from the NLO real matrix element before showering. In that language, the MC counterterm is the shower’s approximation to real emission, and MC@NLO-type matching modifies the NLO subtraction structure so that inclusive observables remain NLO accurate after showering (Nason et al., 2012).

2. Matched cross section and event structure

The formal starting point is the standard NLO decomposition of an infrared-safe observable into an underlying-Born contribution and a real-emission contribution, with local subtraction counterterms making the real and virtual pieces finite separately. MC@NLO modifies this construction by replacing the abstract subtraction with a shower-aware one. In schematic form, the modified cross section is written as

$$\sigma_{\rm mod} = \left[ B(\Phi_B)+\hat V(\Phi_B)+\int d\Phi_{\rm rad}\,(R)_{\rm MC} \right] + \left[ R(\Phi_R)-(R)_{\rm MC} \right],$$

where $(R)_{\rm MC}$ is the shower approximation to the real-emission matrix element (Nason et al., 2012).

This induces the standard separation into Born-like and real-emission event classes. In the review terminology, the first bracket defines the $n=0,1,2,3$0 events, while the second defines the $n=0,1,2,3$1 events. The $n=0,1,2,3$2 class carries the Born term, virtual correction, and integrated shower subtraction; the $n=0,1,2,3$3 class carries the hard real-emission remainder. Because the subtraction is performed at the event-weight level, both classes may have negative weights (Nason et al., 2012).

In the Sherpa formulation, the double-counting problem is addressed by introducing modified subtraction terms $n=0,1,2,3$4, chosen to match the shower approximation used in event generation. The matched expectation value is organized in terms of a modified Born weight $n=0,1,2,3$5, a Sudakov factor $n=0,1,2,3$6, and a hard remainder $n=0,1,2,3$7. In compact notation,

$n=0,1,2,3$8

This construction guarantees that the NLO expansion is reproduced to $n=0,1,2,3$9, while the shower adds higher-order logarithmic resummation (Siegert et al., 2012).

3. Sherpa realization: Catani–Seymour-based exponentiation

In the 2012 Sherpa implementation, S-MC@NLO denotes a specific realization of MC@NLO in which the key algorithmic choice is the definition of $\mathcal{B}$0. Two options are discussed. One is to identify $\mathcal{B}$1 with the shower kernels $\mathcal{B}$2, which naturally limits exponentiation by the factorization scale $\mathcal{B}$3, but does not fully cover soft singularities. The other is to identify $\mathcal{B}$4 with the full Catani–Seymour dipoles $\mathcal{B}$5, which is the option emphasized for Sherpa 0.8 / S-MC@NLO (Siegert et al., 2012).

The Catani–Seymour-based choice has three consequences that define the Sherpa algorithmic identity. First, $\mathcal{B}$6 simplifies. Second, exact NLO accuracy is retained. Third, subleading-colour configurations are treated exactly by exponentiating Catani–Seymour subtraction terms. The price is that dipoles can become negative, so $\mathcal{B}$7 may exceed 1. Sherpa handles this with a weighted one-step shower based on the subtraction terms (Siegert et al., 2012).

The exponentiation region is not left unrestricted. The phase-space boundary for exponentiation must be restricted “artificially” with cuts; in the reported results an $\mathcal{B}$8 cut is used. This is a practical feature of the Sherpa realization rather than a generic requirement of all MC@NLO-type schemes (Siegert et al., 2012).

4. Event-generation workflow and shower dependence

The practical event-generation logic in Sherpa is explicit. The procedure is:

1. Compute NLO ingredients: $\mathcal{B}$9, $\mathcal{V}$0, $\mathcal{V}$1, and subtraction terms $\mathcal{V}$2.
2. Define matching subtraction terms $\mathcal{V}$3: either shower kernels or full dipoles; in the Sherpa implementation discussed here, the dipole-based choice is used.
3. Construct the modified Born weight: $\mathcal{V}$4.
4. Generate events: with probability proportional to $\mathcal{V}$5 for Born-like kinematics, or with weight $\mathcal{V}$6 for real-emission kinematics.
5. Apply one-step shower from $\mathcal{V}$7 events: generate zero or one hardest emission using $\mathcal{V}$8, with emission probability governed by $\mathcal{V}$9.
6. Let the full parton shower continue: subsequent emissions are generated by the standard shower.

The result is a fully exclusive hadron-level event sample (Siegert et al., 2012).

A recurrent theme in the broader MC@NLO literature is that such matching is shower-specific. In one Herwig++ discussion, the “S” is explicitly interpreted as “shower,” meaning that the subtraction terms are tailored to the specific shower algorithm and variables of the Monte Carlo. The same structural point appears in MC@NLO implementations for Herwig++, PYTHIA, and the Nagy–Soper shower: the shower approximation used in the subtraction must be the one actually realized by the generator, otherwise local cancellation and first-emission consistency are lost (Frixione et al., 2010, Torrielli et al., 2010, Czakon et al., 2016).

5. $\mathcal{R}$0-jet realization and phenomenological performance

The canonical application of the Sherpa algorithm is $\mathcal{R}$1-jet production at the LHC for $\mathcal{R}$2. For $\mathcal{R}$3 jets, the virtual matrix element is included in a leading-colour approximation. For $\mathcal{R}$4, events require $\mathcal{R}$5 jets with $\mathcal{R}$6 GeV, and the exponentiation region is restricted with an $\mathcal{R}$7 cut in the dipole terms (Siegert et al., 2012).

The comparison is performed at three levels: fixed-order NLO, “MC@NLO 1em” for hardest-emission only, and “MC@NLO PL” for the full parton shower. The observables studied are the transverse momenta of the first, second, and third jets, and the angular correlations of the two leading jets in $\mathcal{R}$8-jet production. These observables probe hard jet spectra, the multi-jet radiation pattern, and angular structure sensitive to higher-order QCD effects (Siegert et al., 2012).

The reported phenomenology is that the NLO+PS predictions describe the ATLAS data very well, are stable across the observables studied, and provide the first NLO+PS predictions for $\mathcal{R}$9 jets in this setup. The overall conclusion is that the Sherpa S-MC@NLO implementation yields phenomenologically successful, exclusive hadron-level predictions at NLO accuracy, with good agreement with measured jet $N_C$0 spectra and jet angular correlations (Siegert et al., 2012).

6. Terminology, extensions, and recurrent issues

The literature uses the label “S-MC@NLO” in more than one sense. In the Sherpa paper it denotes Sherpa’s implementation, using Catani–Seymour-based subtraction and exponentiation. In one Herwig++ treatment the label is tied to shower-aware counterterms, with the “S” denoting the shower. In the 0.8/Alaric framework it denotes a matching algorithm that “accounts for both non-trivial color correlations and non-trivial spin correlations across the hard process and the first splitting by reweighting the parton shower with the relevant insertion operators on a point-by-point basis,” and is combined with an NLL-preserving kinematics mapping (Siegert et al., 2012, Frixione et al., 2010, Höche et al., 30 Jul 2025).

Source	Use of “S-MC@NLO”	Distinctive feature
Sherpa study (Siegert et al., 2012)	Sherpa implementation of MC@NLO	Catani–Seymour-based subtraction/exponentiation
Herwig++ discussion (Frixione et al., 2010)	Shower-aware MC@NLO realization	Subtraction terms tailored to Herwig++ shower variables
0.8/Alaric study (Höche et al., 30 Jul 2025)	Matching and multi-jet merging algorithm	NLL-preserving kinematics mapping with color/spin reweighting

A persistent misconception is to treat S-MC@NLO as merely “NLO + shower.” In the sources, it is instead a carefully subtracted matching scheme in which the shower approximation to the first emission is removed from the fixed-order real contribution and restored through the shower evolution. Negative weights are therefore not an accidental implementation artifact but a structural consequence of subtraction-based matching. In the Sherpa variant, negativity also appears because Catani–Seymour dipoles can become negative and thus lead to $N_C$1 (Siegert et al., 2012, Nason et al., 2012).

Later work has focused on this issue without redefining the basic matching principle. “Born spreading” was introduced as a way of reducing negative-weight $N_C$2 events in standard MC@NLO by redistributing the Born contribution over radiative variables; the paper is explicit that it does not present Born spreading as an S-MC@NLO algorithm. MC@NLO-$N_C$3 modifies the matching by inserting a Sudakov-like factor $N_C$4, and the reported predictions are consistent with standard MC@NLO within the typical matching systematics while significantly reducing negative-weight fractions (Frederix et al., 2023, Frederix et al., 2020).

The same period also produced systematic extensions from matching to merging. “Merging meets matching in MC@NLO” shows that simultaneous prediction of observables exclusive in different light-jet multiplicities cannot simply be obtained by summing standalone MC@NLO results; a suitable merging procedure must be defined. More recently, the Alaric study reports the first NLO matched and multi-jet merged predictions based on that shower, validated for $N_C$5 up to 5 jets and obtained with an evolution algorithm with NLL-preserving kinematics mapping (Frederix et al., 2012, Höche et al., 30 Jul 2025).