Pith. sign in

REVIEW 3 major objections 6 minor 62 references

Optimal transport beats Euclidean metrics for cleaning negative MC weights

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

2026-07-10 02:14 UTC pith:ST2F2UUW

load-bearing objection OT-based cell resampling is a solid practical advance, but the key comparison against the prior Euclidean metric has a structural bias in its favor that the authors don't address. the 3 major comments →

arxiv 2607.08723 v1 pith:ST2F2UUW submitted 2026-07-09 hep-ph physics.comp-phphysics.data-an

Optimal-Transport-Based Cell Resampling for Negative and Pathological Event Weights

Regan Doherty , Lauren Hay , Rishabh Jain , Matt LeBlanc , Julia Marrinan , Camille Mauceri , Jennifer Roloff This is my paper
classification hep-ph physics.comp-phphysics.data-an
keywords optimal transportcell resamplingnegative event weightsEnergy Mover's DistanceMonte Carlo event generationinfrared collinear safetyCross-Section Mover's DistanceNLO QCD
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that cell resampling algorithms — which locally redistribute negative and pathologically large event weights among nearby events in a metric space to make all weights positive — perform better when the notion of 'nearby' is defined by optimal-transport distances rather than by the Euclidean object-based metrics used in prior work. The specific metrics studied are the Energy Mover's Distance (EMD), which computes the minimum work needed to rearrange one event's radiation pattern into another's, and a spectral variant (sEMD) that operates on a one-dimensional energy-weighted angular representation. Because these metrics are infrared- and collinear-safe by construction, the resampling can be applied directly to particles at any stage of Monte Carlo event generation — after the hard-scattering matrix element, after the parton shower, or after hadronization — without first clustering particles into jets, which prior Euclidean approaches required and which introduced an unwanted dependence on jet algorithm parameters. Applied to next-to-leading-order Z+jets and top-antitop samples, the EMD with angular exponent β=1 applied to fully hadronized events yields the lowest bias across kinematic distributions and a holistic sample-level distance metric the authors call the Cross-Section Mover's Distance (ΣMD), outperforming the Euclidean metric particularly at high reweighting fractions.

Core claim

The central discovery is that optimal-transport-based distances — specifically the EMD with β=1 on hadronized events — provide an infrared-safe, jet-definition-independent metric for cell resampling that introduces less bias into kinematic distributions than the Euclidean object-based metric from prior literature, while also enabling reweighting at any stage of event generation. The authors further introduce the ΣMD, an unbinned optimal-transport distance between entire event samples, as a general-purpose figure of merit for quantifying reweighting bias that does not depend on choosing specific observables or binning.

What carries the argument

The EMD between two collider events is defined as the minimum transport cost to rearrange one event's energy flow into the other's, computed over the two-dimensional rapidity-azimuth plane with a ground metric of Euclidean angular distance, an angular exponent β controlling sensitivity to different angular scales, and a parameter R setting the relative importance of transport cost versus total-energy difference. Cell resampling uses this metric to define hyperspherical neighborhoods (cells) around negatively-weighted seed events; when the total weight in a cell is positive, weights are redistributed via the transformation w_i → (Σ w_j / Σ |w_j|) |w_i|, making all cell weights positive while保

Load-bearing premise

The performance advantage of the optimal-transport metrics over the Euclidean object-based metric is established on samples of only 100,000 events, whereas real LHC productions use millions. At larger sample sizes, the metric space becomes denser, cell radii shrink, and the Euclidean metric's superior computational scaling may erode or reverse the bias advantage demonstrated here. The authors themselves flag this concern.

What would settle it

If, when applied to million-event samples comparable to those used in actual LHC productions, the EMD-based cell resampling does not produce measurably lower bias than the Euclidean object-based metric (or if its computational cost makes it impractical at that scale), the central claim of superiority would not hold in the regime where it matters for experiments.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the EMD-based cell resampling scales to the millions-of-events samples used in real LHC productions, it could reduce the effective sample sizes needed for NLO simulations by factors of 6–16 (depending on the negative-weight fraction), directly alleviating the storage and computing bottleneck projected for the High-Luminosity LHC era.
  • The IRC-safety of the EMD means cell resampling can be applied as a post-hoc afterburner at any stage of the simulation pipeline without modifying the generator itself, making it compatible with existing production workflows.
  • The ΣMD provides a standardized, binning-independent benchmark that any group developing reweighting or sample-correction algorithms could adopt to report bias in a comparable way.
  • The constant-offset procedure for handling negative weights in optimal transport — adding a uniform shift to make all weights non-negative, exploiting the cancellation property of generalized Wasserstein distances — is a general technique applicable to any signed-measure optimal transport problem, not just cell resampling.
  • If the approach generalizes beyond Z+jets and tt̄ to more complex final states (e.g., multi-leg merged samples with higher negative-weight fractions), it could become a standard post-processing step in Monte Carlo production chains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The comparison against the Euclidean metric was conducted on samples of 10^5 events, far smaller than the millions used in experimental productions. The authors acknowledge that the Euclidean metric's N log(N) scaling may give it a practical advantage at realistic sample sizes, so the demonstrated superiority of EMD may not hold in the regime where LHC experiments actually operate.
  • The O(N^3 log N) computational cost of exact EMD computation for high-multiplicity events (e.g., tt̄ after hadronization) could become a bottleneck when applied to millions of events, even though the bias reduction is demonstrated convincingly at the smaller scale studied.
  • The constant-offset trick for negative weights in the ΣMD works because cell resampling preserves event support (no events added or removed), but extending it to samples with genuinely different supports — as in generative model evaluation — would require ghost events or a different approach.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. This manuscript proposes using optimal-transport (OT) based metrics — the Energy Mover's Distance (EMD) and the Spectral Energy Mover's Distance (sEMD) — as distance measures for cell resampling algorithms that eliminate negative Monte Carlo event weights in NLO QCD samples. Because these metrics are IRC-safe by construction, the resampling can be applied directly to particle-level information without intermediate jet clustering, and at any stage of event generation (hard scatter, parton shower, or hadronization). The authors also introduce the Cross-Section Mover's Distance (ΣMD) as a holistic, unbinned figure of merit for quantifying reweighting bias. The approach is validated on Z+jets and tt̄ samples generated with MadGraph5_aMC@NLO + Pythia8, and the OT-based metrics are shown to reduce bias relative to the Euclidean object-based metric of Andersen et al. [13].

Significance. The problem addressed — negative and pathological weights in NLO+shower MC samples — is timely and practically important for the HL-LHC computing program. The key methodological contribution is the use of IRC-safe OT-based metrics that eliminate the jet-definition dependence of prior cell resampling approaches. The introduction of the ΣMD as an observable-independent, unbinned figure of merit is a useful and broadly applicable tool. The study of reweighting at different generation stages (HS, PS, HAD) is a genuine advantage of IRC-safe metrics and is explored systematically. The constant-offset procedure for handling negative weights in the ΣMD (Sec. 4.2.1) is a clean, well-justified practical contribution grounded in the generalized Wasserstein cancellation property. The paper provides falsifiable, quantitative comparisons across two distinct final-state topologies.

major comments (3)
  1. Sec. 4.2, Eq. (4.2) and Fig. 14: The ΣMD uses EMD as its ground metric (Θ_ij = EMD_{β,R}(E_i, E'_j)), which means the ΣMD measures sample-level distortion in the same metric space that the EMD-based reweighting uses to define event locality. This creates a structural preference for EMD-based reweighting over the Andersen et al. Euclidean metric: reweightings that are locally well-behaved in EMD space will naturally score better in the ΣMD, while reweightings that are locally well-behaved in a different metric space may appear more distorting. The authors do not discuss this potential circularity. The 1D kinematic comparisons (Figs. 12–13) are independent of this concern, but at f_rw = 0.25 the differences between all methods are mostly within statistical uncertainties, and at f_rw = 0.75 the EMD advantage is visible primarily in H_T and N_jets but not in ΔR(j1,j2) or pT,j1/pT,j2. The ΣMD
  2. Sec. 3 and Sec. 5.3: All comparisons use samples of 10^5 events, far smaller than the millions of events used in experimental productions. The authors acknowledge that 'some aspects of [the Andersen et al. metric's] performance may be due to its application to a much sparser metric space' (Sec. 5.3). This caveat is important because the Euclidean metric of Ref. [13] was developed and validated on much larger samples. If the relative performance advantage of EMD diminishes at realistic sample sizes — where the Euclidean metric's N log(N) scaling gives it a practical computational edge — the central comparative claim would not hold in the regime where it matters most for LHC experiments. The authors should either (a) provide at least one comparison at a larger sample size to test whether the ranking is stable, or (b) more clearly scope the comparative claim to the sparse-sample regime and
  3. Sec. 2.1: The EMD computation scales as O(N^3 log N) per event pair (or O(N^2) with approximate solvers), and the authors note that 'further improvements related to runtime efficiency have been left to future studies.' However, no timing benchmarks are provided, making it difficult to assess whether the method is practical for production-scale samples. A table of typical per-event-pair computation times for the Z+jets and tt̄ samples (which have different particle multiplicities) would substantially strengthen the paper's claim of practical applicability.
minor comments (6)
  1. Sec. 2.1: The statement that R must be at least half of the maximum possible ground space distance when β > 1 is mentioned but the specific value R = 11.64 used throughout is only introduced later in the section. Consider stating the chosen R value when it is first needed.
  2. Sec. 5.2, Fig. 8: The caption states the two events had H_T values of 187 GeV and 141 GeV, but the figure y-axis label reads 'max(H_T, H_T')' and '|H_T - H_T'|' without clear indication of which curve corresponds to which limit. Please clarify in the figure legend or caption.
  3. Sec. 5.1, Fig. 3b: The x-axis is labeled 'Radius [GeV]' but the EMD ground metric is defined in the (η, φ) plane, which is dimensionless. Please clarify the units.
  4. Sec. 4.1: The observable m_{b,l} is described as having a kinematic endpoint at √(m_t² - m_W²) ≈ 153 GeV, but this endpoint applies only to the specific decay topology where the b-jet and lepton come from the same top. In dileptonic tt̄ events, the endpoint differs. Please specify the decay channel or clarify.
  5. Sec. 5.3: The sEMD is described as showing 'similar but consistently worse performance' than the EMD, but no explanation is offered for why this is the case. A brief discussion of whether the spectral representation's loss of geometric information (mentioned in Sec. 2.2) is responsible would strengthen the analysis.
  6. The reference list includes Refs. [62, 63] which appear to be from 2025–2026; please verify these citations are complete and correctly attributed.

Circularity Check

1 steps flagged

ΣMD evaluation metric shares its ground metric (EMD) with one of the methods being evaluated, creating structural affinity in the key comparative claim — but this is evaluation bias, not strict derivation circularity, and independent kinematic evidence partially corroborates.

specific steps
  1. self definitional [Eq. 4.2 and Figure 14 (Sec. 5.3)]
    "The ground metric Θ_ij is an event-level EMD, Θ_ij = EMD_{β,R}(E_i, E'_j), (4.2) so that the cost of transporting cross section between two events is determined by how distant those events are in the ground space."

    The ΣMD measures sample-level distortion using EMD as its ground metric (Eq. 4.2). The EMD-based cell resampling defines event 'nearby' using the same EMD. When EMD-based reweighting redistributes weights among nearby events in EMD space, the resulting cross-section transport measured by ΣMD will involve short-distance moves in EMD space, yielding a small ΣMD by construction of the locality criterion. The Andersen et al. reweighting defines locality in a different (Euclidean-jet) metric space; its weight redistributions may be locally well-behaved in that space but appear as long-distance transport in EMD space, inflating the ΣMD. Thus the ΣMD comparison in Figure 14 — the paper's clearest evidence that 'OT-based metrics reduce the observed bias relative to other cell resampling techniques

full rationale

The structural affinity between the ΣMD's ground metric (EMD) and the EMD-based reweighting method is a genuine methodological concern: the evaluation metric is defined using the same distance as one evaluated method, which could systematically favor that method in the comparison. However, this does not constitute strict circularity in the derivation-chain sense. The ΣMD is not trivially zero for EMD-reweighted samples — the reweighting changes event weights, not event positions, so the ΣMD remains a nontrivial measurement. The result is not 'equivalent to its inputs by construction.' Furthermore, the EMD and ΣMD are drawn from external literature (Komiske, Metodiev, Thaler; Refs. [23, 27]) — not self-citations by the present authors. The paper also provides independent evidence through 1D kinematic distributions (Figs. 12–13), which do not depend on the EMD metric. At f_rw=0.75, these show EMD advantages in H_T and N_jets, though not in ΔR(j1,j2) or pT,j1/pT,j2, and at f_rw=0.25 differences are mostly within statistical uncertainties. The paper acknowledges that 'all figures of merit should be considered together' (Sec. 4.2) but does not discuss the specific metric-affinity concern. This is a real evaluation-bias risk worth flagging, but it falls short of the strict circularity patterns (fitted input renamed as prediction, self-citation chain forcing the result, or result = input by definition) that would warrant a score of 6+.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities, particles, or forces. The EMD, sEMD, and ΣMD are mathematical constructions from prior literature. The free parameters (β, R, γ, S, τ) are hyperparameters of the distance metrics, chosen by domain reasoning or performance scanning rather than fitted to physical data. The constant-offset procedure for negative weights is a new algorithmic technique but not a new entity.

free parameters (5)
  • β (angular exponent) = 1
    Studied at values 0, 0.5, 1, 2, ∞; β=1 selected as optimal trade-off of accuracy and computational efficiency (Sec. 5.2). Not fitted to data per se, but chosen by scanning performance.
  • R (EMD distance scale) = 11.64
    Set to the maximum possible ground space distance given the ATLAS/CMS pseudorapidity acceptance (Sec. 2.1). Not a fitted parameter but a domain-specific choice.
  • τ (Euclidean metric pT weighting) = 0
    Left at default value of 0 for the Andersen et al. metric comparison (Sec. 1.1).
  • γ (ΣMD angular exponent) = 1
    Set to 1 for the ΣMD computation (Sec. 4.2).
  • S (ΣMD distance scale) = max distance in EMD matrix
    Set to normalize ΣMD values to [0,1] range (Sec. 4.2).
axioms (4)
  • standard math The EMD and sEMD are IRC-safe metrics on the space of collider events.
    Established in Refs. [23–25]; invoked throughout Sec. 2 and used as the foundational justification for applying these metrics directly to particles without jet clustering.
  • domain assumption Cell resampling with sufficiently small cell radius R introduces undetectable bias on physical observables.
    Stated in Sec. 1.1: 'When R becomes smaller than what is experimentally resolvable, the effect of this reweighting on physical observables becomes undetectable.' This is the core assumption enabling the entire approach.
  • standard math The generalized Wasserstein distance satisfies W(μ+η, ν+η) = W(μ,ν), justifying the constant-offset procedure for negative weights.
    Invoked in Sec. 4.2.1, proven in Refs. [55, 56]. Used to handle negative weights in the ΣMD computation.
  • domain assumption The MC@NLO matching procedure produces a stable fraction of negative weights that can be locally redistributed without biasing the inclusive cross section.
    Implicit in the entire approach; the cell resampling preserves total cell weight (Eq. 1.1) but assumes local phase-space density is sufficient for small-R reweighting.

pith-pipeline@v1.1.0-glm · 24289 in / 2942 out tokens · 455917 ms · 2026-07-10T02:14:21.523839+00:00 · methodology

0 comments
read the original abstract

Negative and pathologically large Monte Carlo event weights strain the computing budgets of experiments at the Large Hadron Collider. Cell resampling algorithms locally redistribute event weights among nearby events in a metric space. We study the performance of metrics defined in terms of Optimal Transport, namely the Energy Mover's Distance and a spectral variant, in the context of such algorithms. As these metrics are insensitive to the addition of soft and collinear radiation, they may be applied directly to particles at any stage of event generation. When applied to samples simulated at next-to-leading-order in quantum chromodynamics, this approach reduces the observed bias relative to other cell resampling techniques presented in the literature. We also study the Cross-Section Mover's Distance as an unbinned, broadly-applicable figure of merit for quantifying the bias introduced by any full-phase-space reweighting.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages · 35 internal anchors

  1. [1]

    Evans and P

    L. Evans and P. Bryant,LHC Machine,JINST3(2008) S08001

    work page 2008

  2. [2]

    Aberle, I

    O. Aberle, I. B´ ejar Alonso, O. Br¨ uning, P. Fessia, L. Rossi, L. Tavian et al.,High-Luminosity Large Hadron Collider (HL-LHC): Technical design report, CERN Yellow Reports: Monographs, CERN, Geneva (2020), 10.23731/CYRM-2020-0010

    work page doi:10.23731/cyrm-2020-0010 2020

  3. [3]

    ATLAS Collaboration,ATLAS Software and Computing HL-LHC Roadmap, Tech. Rep. CERN-LHCC-2022-005, CERN, Geneva (2022)

    work page 2022

  4. [4]

    CMS Offline Software and Computing,CMS Phase-2 Computing Model: Update Document, Tech. Rep. CERN-CMS-NOTE-2022-008, CERN, Geneva (2022)

    work page 2022

  5. [5]

    ATLAS Collaboration,Modelling and computational improvements to the simulation of single vector-boson plus jet processes for the ATLAS experiment,JHEP08(2022) 089 [2112.09588]

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  6. [6]

    A General Algorithm for Calculating Jet Cross Sections in NLO QCD

    S. Catani and M.H. Seymour,A General algorithm for calculating jet cross-sections in NLO QCD,Nucl. Phys. B485(1997) 291 [hep-ph/9605323]. – 28 –

    work page internal anchor Pith review Pith/arXiv arXiv 1997

  7. [7]

    Three-jet cross sections to next-to-leading order

    S. Frixione, Z. Kunszt and A. Signer,Three jet cross-sections to next-to-leading order,Nucl. Phys. B467(1996) 399 [hep-ph/9512328]

    work page internal anchor Pith review Pith/arXiv arXiv 1996

  8. [8]

    Matching NLO QCD computations and parton shower simulations

    S. Frixione and B.R. Webber,Matching NLO QCD computations and parton shower simulations,JHEP06(2002) 029 [hep-ph/0204244]

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  9. [9]

    Merging meets matching in MC@NLO

    R. Frederix and S. Frixione,Merging meets matching in MC@NLO,JHEP12(2012) 061 [1209.6215]

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  10. [10]

    Correcting the Colour-Dipole Cascade Model with Fixed Order Matrix Elements

    L. L¨ onnblad,Correcting the Colour-Dipole Cascade Model with Fixed Order Matrix Elements,JHEP05(2002) 046 [hep-ph/0112284]

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  11. [11]

    W+jets Matrix Elements and the Dipole Cascade

    N. Lavesson and L. Lonnblad,W+jets matrix elements and the dipole cascade,JHEP07 (2005) 054 [hep-ph/0503293]

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  12. [12]

    Matching Tree-Level Matrix Elements with Interleaved Showers

    L. L¨ onnblad and S. Prestel,Matching tree-level matrix elements with interleaved showers, JHEP03(2012) 019 [1109.4829]

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  13. [13]

    Unbiased Elimination of Negative Weights in Monte Carlo Samples

    J.R. Andersen and A. Maier,Unbiased elimination of negative weights in Monte Carlo samples,Eur. Phys. J. C82(2022) 433 [2109.07851]

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  14. [14]

    Efficient negative-weight elimination in large high-multiplicity Monte Carlo event samples

    J.R. Andersen, A. Maier and D. Maˆ ıtre,Efficient negative-weight elimination in large high-multiplicity Monte Carlo event samples,Eur. Phys. J. C83(2023) 835 [2303.15246]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  15. [15]

    Precision Cell Resampling with a Relative and Resonant Aware Metric

    J.R. Andersen, E. Cole and A. Maier,Precision Cell Resampling with a Relative and Resonant Aware Metric,2605.13237

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    Pele and B

    O. Pele and B. Taskar,The tangent earth mover’s distance, inGeometric Science of Information, F. Nielsen and F. Barbaresco, eds., (Berlin, Heidelberg), pp. 397–404, Springer Berlin Heidelberg, 2013, https://doi.org/10.1007/978-3-642-40020-9 43

    work page doi:10.1007/978-3-642-40020-9 2013

  17. [17]

    Peleg, M

    S. Peleg, M. Werman and H. Rom,A unified approach to the change of resolution: space and gray-level,IEEE Transactions on Pattern Analysis and Machine Intelligence11(1989) 739

    work page 1989

  18. [18]

    Rubner, C

    Y. Rubner, C. Tomasi and L.J. Guibas,A metric for distributions with applications to image databases, inProceedings of the Sixth International Conference on Computer Vision, ICCV ’98, (Washington, DC, USA), pp. 59–, IEEE Computer Society, 1998, https://doi.org/10.1109/ICCV.1998.710701

    work page doi:10.1109/iccv.1998.710701 1998

  19. [19]

    Rubner, C

    Y. Rubner, C. Tomasi and L.J. Guibas,The earth mover’s distance as a metric for image retrieval,International Journal of Computer Vision40(2000) 99

    work page 2000

  20. [20]

    Pele and M

    O. Pele and M. Werman,A linear time histogram metric for improved sift matching, in Computer Vision – ECCV 2008, D. Forsyth, P. Torr and A. Zisserman, eds., (Berlin, Heidelberg), pp. 495–508, Springer Berlin Heidelberg, 2008, https://doi.org/10.1007/978-3-540-88690-7 37

    work page doi:10.1007/978-3-540-88690-7 2008

  21. [21]

    Wasserstein,Markov processes over denumerable products of spaces describing large systems of automata,Problems of Information Transmission5(1969) 47

    L.N. Wasserstein,Markov processes over denumerable products of spaces describing large systems of automata,Problems of Information Transmission5(1969) 47

    work page 1969

  22. [22]

    Dobrushin,Prescribing a system of random variables by conditional distributions, Theory of Probability & Its Applications15(1970) 458

    R.L. Dobrushin,Prescribing a system of random variables by conditional distributions, Theory of Probability & Its Applications15(1970) 458

    work page 1970

  23. [23]

    The Metric Space of Collider Events

    P.T. Komiske, E.M. Metodiev and J. Thaler,Metric Space of Collider Events,Phys. Rev. Lett.123(2019) 041801 [1902.02346]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  24. [24]

    A Spectral Metric for Collider Geometry

    A.J. Larkoski and J. Thaler,A spectral metric for collider geometry,JHEP08(2023) 107 [2305.03751]. – 29 –

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    SPECTER: Efficient Evaluation of the Spectral EMD

    R. Gambhir, A.J. Larkoski and J. Thaler,SPECTER: efficient evaluation of the spectral EMD,JHEP12(2025) 219 [2410.05379]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  26. [26]

    Flamary and N

    R. Flamary and N. Courty,POT python optimal transport library, 2017

    work page 2017

  27. [27]

    The Hidden Geometry of Particle Collisions

    P.T. Komiske, E.M. Metodiev and J. Thaler,The Hidden Geometry of Particle Collisions, JHEP07(2020) 006 [2004.04159]

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  28. [28]

    Torgerson,Multidimensional scaling: I

    W.S. Torgerson,Multidimensional scaling: I. theory and method,Psychometrika17(1952) 401–419

    work page 1952

  29. [29]

    A Cell Resampler study of Negative Weights in Multi-jet Merged Samples

    J.R. Andersen, A. Cueto, S.P. Jones and A. Maier,A Cell Resampler study of Negative Weights in Multi-jet Merged Samples,2411.11651

    work page internal anchor Pith review Pith/arXiv arXiv

  30. [30]

    Towards Jetography

    G.P. Salam,Towards Jetography,Eur. Phys. J. C67(2010) 637 [0906.1833]

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  31. [31]

    ATLAS Collaboration,Measurements of multijet event isotropies using optimal transport with the ATLAS detector,JHEP10(2023) 060 [2305.16930]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  32. [32]

    Cesarotti and M

    C. Cesarotti and M. LeBlanc,A field guide to event-shape observables using optimal transport,JHEP25(2020) 014 [2409.13150]

    work page arXiv 2020

  33. [33]

    Jet Substructure at the Large Hadron Collider: Experimental Review

    R. Kogler et al.,Jet Substructure at the Large Hadron Collider: Experimental Review,Rev. Mod. Phys.91(2019) 045003 [1803.06991]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  34. [34]

    The Monte Carlo Ecosystem in High-Energy Physics: A Primer

    M. van Beekveld, E. Bothmann, A. Buckley, C. G¨ utschow, P. Skands and R. Winterhalder, The Monte Carlo Ecosystem in High-Energy Physics: A Primer,2605.16036

    work page internal anchor Pith review Pith/arXiv arXiv

  35. [35]

    The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations

    J. Alwall, R. Frederix, S. Frixione, V. Hirschi, F. Maltoni, O. Mattelaer et al.,The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations,JHEP07(2014) 079 [1405.0301]

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  36. [36]

    Parton distributions with LHC data

    NNPDF Collaboration, R.D. Ball et al.,Parton distributions with LHC data,Nucl. Phys. B 867(2013) 244 [1207.1303]

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  37. [37]

    A comprehensive guide to the physics and usage of PYTHIA 8.3

    C. Bierlich et al.,A comprehensive guide to the physics and usage of PYTHIA 8.3,SciPost Phys. Codeb.2022(2022) 8 [2203.11601]

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  38. [38]

    Tuning PYTHIA 8.1: the Monash 2013 Tune

    P. Skands, S. Carrazza and J. Rojo,Tuning PYTHIA 8.1: the Monash 2013 Tune,Eur. Phys. J. C74(2014) 3024 [1404.5630]

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  39. [39]

    Andersson, G

    B. Andersson, G. Gustafson, G. Ingelman and T. Sj¨ ostrand,Parton fragmentation and string dynamics,Phys. Rept.97(1983) 31

    work page 1983

  40. [40]

    Sj¨ ostrand,Jet fragmentation of multiparton configurations in a string framework,Nucl

    T. Sj¨ ostrand,Jet fragmentation of multiparton configurations in a string framework,Nucl. Phys. B248(1984) 469

    work page 1984

  41. [41]

    On the reduction of negative weights in MC@NLO-type matching procedures

    R. Frederix, S. Frixione, S. Prestel and P. Torrielli,On the reduction of negative weights in MC@NLO-type matching procedures,JHEP07(2020) 238 [2002.12716]

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  42. [42]

    Catani, Y.L

    S. Catani, Y.L. Dokshitzer, M.H. Seymour and B.R. Webber,Longitudinally invariantK t clustering algorithms for hadron hadron collisions,Nucl. Phys. B406(1993) 187

    work page 1993

  43. [43]

    The anti-k_t jet clustering algorithm

    M. Cacciari, G.P. Salam and G. Soyez,The anti-k t jet clustering algorithm,JHEP04(2008) 063 [0802.1189]

    work page internal anchor Pith review Pith/arXiv arXiv 2008

  44. [44]

    FastJet user manual

    M. Cacciari, G.P. Salam and G. Soyez,FastJet User Manual,Eur. Phys. J. C72(2012) 1896 [1111.6097]. – 30 –

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  45. [45]

    CMS Collaboration,Measurement of the top quark mass in the dileptonict ¯tdecay channel using the mass observablesM bℓ,M T2 , andM bℓν in pp collisions at √s= 8TeV,Phys. Rev. D96(2017) 032002 [1704.06142]. [46]ATLAScollaboration,Measurement of the top quark mass in thet ¯t→dilepton channel from√s= 8TeV ATLAS data,Phys. Lett. B761(2016) 350 [1606.02179]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  46. [46]

    M¨ uller,Integral probability metrics and their generating classes of functions,Advances in Applied Probability29(1997) 429–443

    A. M¨ uller,Integral probability metrics and their generating classes of functions,Advances in Applied Probability29(1997) 429–443

    work page 1997

  47. [47]

    Kullback and R.A

    S. Kullback and R.A. Leibler,On Information and Sufficiency,The Annals of Mathematical Statistics22(1951) 79

    work page 1951

  48. [48]

    Lin,Divergence measures based on the shannon entropy,IEEE Transactions on Information Theory37(1991) 145

    J. Lin,Divergence measures based on the shannon entropy,IEEE Transactions on Information Theory37(1991) 145

    work page 1991

  49. [49]

    Wong and M

    A.K.C. Wong and M. You,Entropy and distance of random graphs with application to structural pattern recognition,IEEE Transactions on Pattern Analysis and Machine IntelligencePAMI-7(1985) 599

    work page 1985

  50. [50]

    Pearson,X

    K. Pearson,X. on the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling,The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science50(1900) 157

    work page 1900

  51. [51]

    CaloFlow: Fast and Accurate Generation of Calorimeter Showers with Normalizing Flows

    C. Krause and D. Shih,Fast and accurate simulations of calorimeter showers with normalizing flows,Phys. Rev. D107(2023) 113003 [2106.05285]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  52. [52]

    Neural Quasiprobabilistic Likelihood Ratio Estimation with Negatively Weighted Data

    M. Drnevich, S. Jiggins, J. Katzy and K. Cranmer,Neural quasiprobabilistic likelihood ratio estimation with negatively weighted data,Mach. Learn. Sci. Tech.6(2025) 045023 [2410.10216]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  53. [53]

    Evaluating generative models in high energy physics

    R. Kansal, A. Li, J. Duarte, N. Chernyavskaya, M. Pierini, B. Orzari et al.,Evaluating generative models in high energy physics,Phys. Rev. D107(2023) 076017 [2211.10295]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  54. [54]

    A Wasserstein norm for signed measures, with application to nonlocal transport equation with source term

    B. Piccoli, F. Rossi and M. Tournus,A Wasserstein norm for signed measures, with application to nonlocal transport equation with source term,1910.05105

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  55. [55]

    On properties of the Generalized Wasserstein distance

    B. Piccoli and F. Rossi,On properties of the generalized Wasserstein distance,Archive for Rational Mechanics and Analysis222(2016) 1339 [1304.7014]

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  56. [56]

    Drnevich, S

    M. Drnevich, S. Jiggins and K. Cranmer,Quasiprobabilistic Density Ratio Estimation with a Reverse Engineered Classification Loss Function,2512.19913

    work page arXiv

  57. [57]

    Richards,A flexible growth function for empirical use,Journal of Experimental Botany 10(1959) 290

    F.J. Richards,A flexible growth function for empirical use,Journal of Experimental Botany 10(1959) 290

    work page 1959

  58. [58]

    Wiegand,Kish, l.: Survey sampling

    H. Wiegand,Kish, l.: Survey sampling. john wiley & sons, inc., new york, london 1965, ix + 643 s., 31 abb., 56 tab., preis 83 s.,Biometrische Zeitschrift10(1968) 88

    work page 1965

  59. [59]

    Farkh, R

    S.E. Farkh, R. Frederix and M. Gouighri,MC@NLO event generation by reweighting unweighted Born events,2602.18124

    work page arXiv

  60. [60]

    Reducing negative weights in Monte Carlo event generation with Sherpa

    K. Danziger, S. H¨ oche and F. Siegert,Reducing negative weights in Monte Carlo event generation with Sherpa,2110.15211

    work page internal anchor Pith review Pith/arXiv arXiv

  61. [61]

    Stay Positive: Neural Refinement of Sample Weights

    B. Nachman and D. Noll,Neural refinement of sample weights,Phys. Rev. D112(2025) 096009 [2505.03724]. – 31 –

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  62. [62]

    Palmer and B

    C. Palmer and B. Kronheim,Improving statistical precision in Monte Carlo samples with negative weights via reweighting and uncertainty quantification,Phys. Rev. D113(2026) 012003 [2510.16217]. – 32 –

    work page arXiv 2026