Les Houches Tune (LH Tune) for Hadronization

• Les Houches Tune (LH Tune) is a hybrid setup that merges Herwig 7’s Angular Ordered Parton Shower with Pythia8’s Lund string hadronization for controlled model comparisons.
• It employs a systematic four-stage tuning workflow using Rivet and Professor to optimize fragmentation, flavour, ISR, and MPI parameters for improved collider event simulation.
• The LH Tune demonstrates competitive phenomenology across diverse observables while isolating non-perturbative uncertainties and guiding future refinements in Monte Carlo tuning.

The Les Houches Tune, or LH Tune, is a general-purpose tune for Herwig 7 in which the generator’s default cluster hadronization model is replaced by the Lund string hadronization model from Pythia 8, accessed through TheP8I, while keeping Herwig 7.3’s Angular Ordered Parton Shower (AOPS) as the perturbative baseline. Its defining purpose is to enable a controlled comparison of hadronization models inside a common event-generator framework, so that Herwig 7 + AOPS + cluster hadronization can be compared directly with Herwig 7 + AOPS + Lund string hadronization without simultaneously changing the shower algorithm (Divisova et al., 2 Sep 2025). The tune was finished during the Les Houches PhysTeV workshop, which is the origin of the name, and it is intended for both lepton and hadron collider physics; the paper states that it will be included in the Herwig 7.4 release (Divisova et al., 2 Sep 2025).

1. Definition and historical placement

In the 2025 study “Herwig 7 with the Lund String Model: Tuning and Comparative Hadronization Studies”, the LH Tune is introduced as a new tune for a hybrid setup: Herwig 7 provides the hard-process and shower environment, while Pythia 8 provides the Lund string hadronization model and the QCD-based colour reconnection (CR) model through TheP8I (Divisova et al., 2 Sep 2025). The tune is explicitly presented as a tool for comparative hadronization studies rather than as a purely phenomenological replacement for existing defaults.

A central point of clarification is that the phrase “Les Houches Tune” is not, in the accessible Les Houches 2015 and 2017 Standard Model Working Group reports, the name of an explicit benchmark tune. In the visible material of the 2015 report, there is no explicit “Les Houches Tune”, “LH Tune”, “benchmark tune”, or “recommended tune”, although the report identifies “new developments in Monte Carlo event generators” and a chapter titled “MC uncertainties and output formats” as relevant themes (Badger et al., 2016). Likewise, the visible material of the 2017 report does not provide a self-contained tune card called “Les Houches Tune”, but it does discuss the rationale for common generator setups by emphasizing parton-shower comparisons, tune/retune issues, and the correlation of PDF choices with underlying-event modeling via multiparton interactions (MPI) (Bendavid et al., 2018). This establishes that the modern LH Tune is a specific 2025 deliverable rather than a long-standing tune label from the earlier workshop reports.

The tune also belongs to a longer lineage of systematic Monte Carlo tuning. The 2009 Professor paper does not mention Les Houches, but it introduced a reproducible Professor + Rivet framework and produced recommended base tunes for LHC experiments in Pythia 6, thereby contributing to the methodological background from which later community-style reference tunes emerged (0907.2973).

2. Generator architecture and physics rationale

The LH Tune is defined by a deliberate division of responsibilities between Herwig 7 and Pythia 8. The perturbative environment is fixed to Herwig’s Angular Ordered Parton Shower, and only the non-perturbative hadronization and CR sector is swapped. This design is the basis for the paper’s claim that hadronization-model systematics can be isolated more cleanly than in a standard Herwig-versus-Pythia comparison, where both shower and hadronization differ simultaneously (Divisova et al., 2 Sep 2025).

Subsystem	Provider	Role in the LH Tune
LO matrix elements, AOPS, MPI/UE, PDFs, event generation infrastructure	Herwig 7 / ThePEG	Perturbative and hadron-collision baseline
Lund string hadronization	Pythia 8 via TheP8I	Non-perturbative string fragmentation
QCD-based colour reconnection and associated beam remnants	Pythia 8 via TheP8I	CR sector used with the string model

In this setup, Herwig 7 supplies LO matrix elements, AOPS for both FSR and ISR, the hadron-collision framework including the MPI / Underlying Event model, CT14LO PDFs via LHAPDF, and the overall event-generation infrastructure through ThePEG (Divisova et al., 2 Sep 2025). Through TheP8I, Pythia 8 contributes the Lund string hadronization model, the QCD-based colour reconnection model, and the associated beam-remnant treatment required by that CR model (Divisova et al., 2 Sep 2025).

The theoretical basis of the hadronization model is described in the paper through the linear confinement potential

$V(r)=\kappa r,$

with $\kappa$ the string tension, and through the Lund/Bowler fragmentation form

$$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$

where $a$ and $b$ are the Lund longitudinal-fragmentation parameters, $\sigma$ controls the Gaussian transverse-momentum kicks at string breaking, and the heavy-quark Bowler modification is governed through rFactC and rFactB (Divisova et al., 2 Sep 2025).

The Angular Ordered Parton Shower serves two roles. It is, first, the perturbative backbone of the generator setup, and, second, the fixed reference point against which hadronization can be varied. This allows two complementary comparisons: Herwig string vs Herwig cluster, which isolates hadronization-model differences at fixed shower, and Herwig string vs Pythia string, which exposes shower/framework differences at fixed hadronization model (Divisova et al., 2 Sep 2025).

3. Tuning workflow and methodology

The LH Tune is obtained through a four-stage tuning strategy implemented with Rivet for observable evaluation and Professor for interpolation-based optimization (Divisova et al., 2 Sep 2025). The stages are:

1. String fragmentation + FSR parameters tuned to LEP data.
2. String flavour parameters tuned to identified-particle multiplicity and flavour observables at LEP.
3. ISR $\alpha_s$ and primordial $k_T$ tuned to LHC Drell–Yan $Z$-boson data.
4. MPI + colour reconnection parameters tuned to Minimum Bias and Underlying Event data from hadron collisions (Divisova et al., 2 Sep 2025).

This decomposition follows the assumption, also used in earlier Pythia tuning practice, that fragmentation and flavour sectors can be approximately decoupled enough to be tuned in stages without unacceptable correlations (Divisova et al., 2 Sep 2025). That factorized logic closely matches the methodology described in the Professor framework paper, which advocated systematic tuning through per-bin response parameterization rather than manual or brute-force optimization (0907.2973).

The objective function is the standard weighted Professor goodness-of-fit,

$$\chi^2(\boldsymbol{p})= \sum_{\mathcal O} w_{\mathcal O} \sum_{b\in\mathcal O} \frac{\left(f^{(b)}(\boldsymbol p)-\mathcal R_b\right)^2}{\Delta_b^2},$$

where $\kappa$0 is the observable weight, $\kappa$1 is the Professor response in bin $\kappa$2, $\kappa$3 is the reference value, and $\kappa$4 is the total uncertainty (Divisova et al., 2 Sep 2025). The use of weighted observables is consistent with the Professor methodology, which formalized a weighted $\kappa$5 over observable bins and emphasized that weighting is subjective but unavoidable in practical tune construction (0907.2973).

The first LEP stage tunes 7 string-fragmentation parameters and 2 final-state shower parameters using $\kappa$6 data at $\kappa$7 GeV. The observables include event shapes, momentum spectra, mean charged multiplicities, and flavour-specific mean charged multiplicities, with input from DELPHI 1996 and OPAL 1998. The sampling setup is a 9-dimensional hypercube, 575 sampled parameter points, 1 million events per run, and 40 different interpolations, each built from 400 generator runs chosen randomly from the full set. A sequential parameter-fixing procedure is then applied to parameters whose optimized values are stable across interpolations (Divisova et al., 2 Sep 2025).

The second LEP stage tunes 11 flavour parameters in an 11-dimensional hypercube using 950 sampled points, 1 million events per run, and 40 interpolations, each built from 500 runs. The data include the $\kappa$8-quark fragmentation function from DELPHI, $\kappa$9 and $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$0 multiplicities from a PDG compilation, numerous identified-hadron multiplicity ratios with respect to $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$1, and reduced-weight OPAL flavour-tagged charged multiplicities (Divisova et al., 2 Sep 2025).

The hadron-collision ISR stage tunes $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$2 and primordial $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$3 to 7 TeV Drell–Yan $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$4-boson production, using ATLAS 2012 $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$5, ATLAS 2014 $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$6 $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$7, and CMS 2012 $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$8 $$f(z)\propto \frac{1}{z^{1+r_Q b m_Q^2}}(1-z)^a e^{-b m_T^2/z}, \qquad m_T^2=m^2+p_T^2,$$9 data. Following a previous Herwig strategy, the paper uses muon-channel $a$0 $a$1 distributions and electron-channel angular observables to reduce correlations (Divisova et al., 2 Sep 2025).

The final hadron-collision stage tunes MPI + CR against MB and UE data at 0.9 TeV, 1.8 TeV, 7 TeV, and 13 TeV. The original sample contains 1000 generator runs, but only 265 valid runs are retained for interpolation because the multi-energy setup reduces the number of usable runs (Divisova et al., 2 Sep 2025). The stage also studies the energy evolution of the MPI cutoff through

$a$2

with reference energy fixed to $a$3 TeV (Divisova et al., 2 Sep 2025).

4. Parameter content of the LH Tune

The LH Tune specifies a complete parameter set across fragmentation, flavour, ISR, MPI, and colour reconnection (Divisova et al., 2 Sep 2025).

Final-state shower and string-fragmentation sector: the tuned values are $a$4, $a$5 GeV, $a$6, $a$7 GeV$a$8, $a$9 GeV, aExtraSQuark $b$0, aExtraDiquark $b$1, rFactC $b$2, and rFactB $b$3 (Divisova et al., 2 Sep 2025).

String flavour sector: the tuned flavour parameters are probStoUD $b$4, probQQtoQ $b$5, probSQtoQQ $b$6, probQQ1toQQ0 $b$7, etaSup $b$8, etaPrimeSup $b$9, popcornRate $\sigma$0, mesonUDvector $\sigma$1, mesonSvector $\sigma$2, mesonCvector $\sigma$3, and mesonBvector $\sigma$4 (Divisova et al., 2 Sep 2025). The paper describes these parameters as controlling, respectively, strangeness suppression, diquark suppression, strange-diquark suppression, spin-1 diquark suppression, $\sigma$5 and $\sigma$6 suppression, baryon–meson–antibaryon popcorn production, and vector-to-pseudoscalar ratios in the light, strange, charm, and bottom sectors (Divisova et al., 2 Sep 2025).

Hadron-collision ISR sector: the tuned values are $\sigma$7 and primordial $\sigma$8 GeV (Divisova et al., 2 Sep 2025).

MPI and colour-reconnection sector: in the energy-extrapolated LH Tune, the parameters are Power $\sigma$9 $\alpha_s$0, $\alpha_s$1 GeV, Offset $\alpha_s$2 $\alpha_s$3 GeV, $\alpha_s$4 GeV$\alpha_s$5, ladderMult $\alpha_s$6, ladderbFactor $\alpha_s$7, $\alpha_s$8, m0 $\alpha_s$9 GeV, and junctionCorrection $k_T$0 (Divisova et al., 2 Sep 2025).

The CR model is Pythia8’s QCD-based colour reconnection scheme with the new beam remnant model, including junction strings, but the implementation is intentionally simplified. Several optional features are disabled, including double-junction remnants, time causality to string decays, and special heavy-quark parity treatment, so that CR is effectively controlled mainly by m0 and junctionCorrection (Divisova et al., 2 Sep 2025). In the paper’s terminology, m0 is a lower bound on the mass of the colour-reconnected system, while junctionCorrection rescales this threshold for junction strings and therefore governs how readily junctions form (Divisova et al., 2 Sep 2025).

For practical use, the paper states that the tune will be included in the Herwig 7.4 release and can also be used with Herwig 7.3, provided a patch is applied to the herwig-bootstrap installation script. The relevant TheP8I modifications expose Pythia8 CR controls such as CRreconnect, CRmode, BRremnantMode, CRm0, and CRjunctionCorrection (Divisova et al., 2 Sep 2025).

5. Phenomenology and validation

The paper’s overall conclusion is that the LH Tune gives a good and competitive description across a broad range of observables in both lepton and hadron collisions, while remaining most valuable as a controlled platform for hadronization comparisons (Divisova et al., 2 Sep 2025).

For LEP observables, the tune describes fragmentation and flavour observables accurately. On many event shapes and momentum spectra it lies between the Herwig cluster tune and the Pythia Monash string tune, and in hadronization-sensitive regions it often follows Pythia’s string tune closely, which the paper attributes naturally to the shared string model and similar tuning philosophy (Divisova et al., 2 Sep 2025). Where observables are more sensitive to the shower, the differences with Pythia grow, showing the effect of AOPS relative to Pythia’s $k_T$1-ordered shower (Divisova et al., 2 Sep 2025).

The paper records several more specific LEP-level observations. $k_T$2 and $k_T$3 multiplicities are modeled well by the Herwig string tunes; meson multiplicities are generally reasonable for all tunes; all tunes produce too many $k_T$4 mesons; and baryon multiplicities are harder to model, with $k_T$5 relatively well described (Divisova et al., 2 Sep 2025). Some event shapes are within approximately $k_T$6 over much of the distribution, but tails of thrust, thrust minor, aplanarity, and the $k_T$7-parameter remain difficult, largely because of missing higher perturbative orders (Divisova et al., 2 Sep 2025). The tune also generalises reasonably well to observables not included in the fit, which the authors interpret as evidence against severe overtuning (Divisova et al., 2 Sep 2025).

For the Drell–Yan ISR tuning stage, the LH Tune describes small and intermediate $k_T$8 and $k_T$9 very well, but at large $Z$0 it undershoots the data. The paper identifies the reason as the use of LO matrix elements without higher-order corrections (Divisova et al., 2 Sep 2025).

For Underlying Event measurements at 0.9, 1.8, 7, and 13 TeV, the tune is generally within the envelope formed by the Herwig cluster and Pythia string tunes (Divisova et al., 2 Sep 2025). It often follows the Herwig cluster tune in shape, but with slightly better data agreement, except close to the leading-particle direction, where it moves toward the Pythia result (Divisova et al., 2 Sep 2025). In high-multiplicity regions, especially for UE observables sensitive to colour reconnection and hadronization, the LH Tune can improve on the cluster tune (Divisova et al., 2 Sep 2025). For validation observables not used in the fit, trans-max and trans-min are reasonably described, while trans-diff is overestimated by all tunes by roughly 10–20%, with the Herwig cluster tune slightly better (Divisova et al., 2 Sep 2025).

For Minimum Bias observables, the spread between tunes is larger than for UE, with differences up to 30–40% in some regions; the paper notes that this is expected because MB is more sensitive to soft physics and is not anchored by a hard trigger (Divisova et al., 2 Sep 2025). At low $Z$1, the LH Tune often follows the Herwig cluster tune, plausibly reflecting the same diffractive model; at high $Z$2, it tends to follow the Pythia string tune and often performs better than the cluster tune, plausibly reflecting different CR behavior (Divisova et al., 2 Sep 2025). Charged-particle pseudorapidity remains problematic; in some 100 MeV-track selections the LH Tune is furthest from the data among the compared tunes, although the inter-tune spread is only about 10% (Divisova et al., 2 Sep 2025).

A major caveat concerns identified-particle spectra at the LHC. The paper states that none of the tunes describe identified-particle spectra at the LHC well; for $Z$3, $Z$4, $Z$5, $Z$6, $Z$7, $Z$8, $Z$9, and related observables, discrepancies between models can reach up to 50% (Divisova et al., 2 Sep 2025). The authors therefore conclude that tuning flavour parameters alone is unlikely to remove all tensions and that genuine model improvements are needed (Divisova et al., 2 Sep 2025).

The tune was not explicitly tuned to diffraction-sensitive observables, but it nonetheless gives a reasonable description of forward rapidity gaps. Agreement with Herwig cluster is within about 20%; the non-diffractive small-gap region is better modeled by Pythia, the diffractive large-gap region is better modeled by Herwig cluster, and the Herwig string setup tends to underestimate the diffraction-dominated region (Divisova et al., 2 Sep 2025).

6. Significance, limitations, and relation to broader tuning practice

The principal significance of the LH Tune is methodological. The paper argues not that one model wins universally, but that the tune creates a more meaningful basis for assessing non-perturbative uncertainties by permitting controlled string-versus-cluster comparisons at fixed shower (Divisova et al., 2 Sep 2025). This yields four broader lessons explicitly highlighted in the study: hadronization model dependence matters; shower and hadronization are intertwined; universality is only partial; and the LH Tune is valuable as a systematic tool in addition to being a phenomenological tune (Divisova et al., 2 Sep 2025).

This positioning aligns with earlier Les Houches discussions of Monte Carlo systematics. The visible 2017 report emphasizes that a sensible description of data sensitive to non-perturbative effects may require retuning the event generator when varying renormalization scales, and it stresses the correlation between PDF choices and underlying-event modeling via multiparton interactions (Bendavid et al., 2018). That point is conceptually important for interpreting the LH Tune: tune parameters, shower choices, PDF choices, and non-perturbative modeling are not independent knobs in any physically faithful uncertainty assessment. A plausible implication is that the LH Tune is best understood not as an isolated parameter card but as a coherent generator configuration.

The tune also sits naturally in the tradition of systematic community tuning formalized by Professor + Rivet. The 2009 Professor paper described the now-standard practice of sampling a parameter hypercube, building bin-by-bin polynomial response functions, and minimizing a weighted global goodness-of-fit to construct recommended LHC baseline tunes (0907.2973). The LH Tune adopts the same infrastructure and the same broad philosophy of staged, data-driven calibration, but applies it to a hybrid Herwig + Pythia8-string setup whose scientific purpose is comparative hadronization studies rather than only production of a new default (Divisova et al., 2 Sep 2025).

The paper’s limitations are stated directly. Large-$$\chi^2(\boldsymbol{p})= \sum_{\mathcal O} w_{\mathcal O} \sum_{b\in\mathcal O} \frac{\left(f^{(b)}(\boldsymbol p)-\mathcal R_b\right)^2}{\Delta_b^2},$$0 Drell–Yan tails are not expected to be perfect because the setup uses LO matrix elements without higher-order corrections; some event-shape tails are perturbatively limited rather than hadronization limited; diffraction was not directly tuned; identified-flavour spectra remain poorly described by all available tunes; and the implemented CR model is a simplified subset of Pythia’s full QCD-based CR model (Divisova et al., 2 Sep 2025). The authors therefore recommend the LH Tune as a useful basis for comparative uncertainty studies and suggest future directions such as repeating the analysis with Herwig’s dipole shower, providing systematic variations/eigentunes, and incorporating additional LHC hadronization-sensitive observables in future tuning campaigns (Divisova et al., 2 Sep 2025).