pith. sign in

arxiv: 2606.28492 · v1 · pith:ZOHIGCQ5new · submitted 2026-06-26 · ✦ hep-ph

Propagating data noise through the fit: the Monte Carlo replica distribution

Mark N. Costantini This is my paper

Pith reviewed 2026-06-30 01:11 UTC · model grok-4.3

classification ✦ hep-ph
keywords Monte Carlo replica methodparton distribution functionsSMEFTBayesian posteriorparameter uncertaintiesnonlinear modelsHessian matrixdata noise propagation
0
0 comments X

The pith

The Monte Carlo replica method for uncertainty estimation produces a distribution that differs from the Bayesian posterior by the residual-weighted Hessian of the model at the best fit.

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

The paper derives the leading-order distribution generated by the Monte Carlo replica method when fitting models to noise-perturbed data copies. This distribution matches the Laplace approximation to the Bayesian posterior only for linear models and otherwise differs by one matrix whose elements are weighted by the fit residuals. The sign and size of that matrix determine whether the method overestimates or underestimates the parameter uncertainties. A reader would care because the expression gives a closed-form diagnostic for the size of the departure without needing a full Bayesian calculation.

Core claim

The Monte Carlo replica method fits a model to many noise-perturbed copies of the data and treats the empirical distribution of the resulting best-fit parameters as the uncertainty distribution. When the model is nonlinear this empirical distribution departs from the Bayesian posterior. At leading order the departure is given by a single matrix, the residual-weighted Hessian of the model evaluated at the best-fit point. The sign and magnitude of this matrix fix whether the Monte Carlo uncertainties are larger or smaller than the Bayesian ones. The same closed-form relation is verified on two single-parameter models that can be solved exactly.

What carries the argument

the residual-weighted Hessian of the model at the best fit, which supplies the leading-order difference between the Monte Carlo replica distribution and the Laplace-approximated Bayesian posterior

If this is right

  • The Monte Carlo method reproduces the Bayesian posterior exactly only when the model is linear in its parameters.
  • The difference between the two distributions is fixed by one computable matrix.
  • The sign and magnitude of the matrix determine over- or under-estimation of uncertainties.
  • The same matrix can be evaluated in full global PDF and SMEFT fits to quantify the bias.

Where Pith is reading between the lines

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

  • Fitting groups could compute the matrix once after the central fit and use it to rescale reported uncertainties without rerunning Bayesian sampling.
  • The same diagnostic applies to any least-squares fit that uses replica methods, not only PDFs and SMEFT coefficients.
  • When the matrix is large, higher-order terms in the nonlinearity expansion may need separate study to keep the correction accurate.

Load-bearing premise

The derivation assumes that a leading-order expansion in the model's nonlinearity is sufficient and that higher-order curvature terms can be neglected.

What would settle it

Compute the residual-weighted Hessian for a concrete nonlinear model, generate the Monte Carlo replica distribution numerically, and check whether its spread matches the Bayesian posterior spread shifted by exactly that matrix.

read the original abstract

The Monte Carlo (MC) replica method quantifies parameter uncertainties in global fits of parton distribution functions (PDFs) and Standard Model Effective Field Theory (SMEFT) Wilson coefficients by fitting a model to many noise-perturbed copies of the data and taking the empirical distribution of the best-fit parameters as the uncertainty. The method reproduces the Bayesian posterior exactly only when the model is linear in its parameters, and departs from it in the nonlinear case. We derive the leading-order distribution the method produces and compare it with the Laplace approximation of the Bayesian posterior: the two differ by a single computable matrix, the residual-weighted Hessian of the model at the best fit, whose sign and magnitude set the over- or under-estimation of the parameter uncertainties. This closed-form expression quantifies when and by how much the MC method departs from Bayesian inference. We illustrate it on two single-parameter examples solvable in closed form and point to its evaluation in full PDF and SMEFT fits as a natural next step.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript derives the leading-order distribution of best-fit parameters obtained via the Monte Carlo replica method when propagating data noise through a nonlinear fit. It compares this distribution to the Laplace approximation of the Bayesian posterior, finding that they differ by a single computable matrix—the residual-weighted Hessian of the model evaluated at the best-fit point—whose properties determine whether the MC method over- or under-estimates parameter uncertainties. The result is illustrated on two exactly solvable single-parameter models, with extension to full PDF and SMEFT fits proposed as future work.

Significance. If the leading-order derivation holds under controlled conditions, the closed-form expression supplies a practical diagnostic for the accuracy of the MC replica method relative to Bayesian inference in global fits. This is relevant for PDF and SMEFT analyses where nonlinearity is common. The explicit single-parameter examples, where the difference matrix can be evaluated exactly, constitute a clear strength, as does the reduction of the discrepancy to one identifiable matrix.

major comments (1)
  1. [Abstract] Abstract: The central derivation is performed at leading order in an expansion in the model's nonlinearity (departure from linearity in parameters). However, no bound, scaling criterion, or regime of validity is supplied for when higher-order terms remain negligible compared with the leading correction. This is load-bearing for the claim that the result 'quantifies when and by how much the MC method departs from Bayesian inference' in general PDF and SMEFT fits, where model curvature may be appreciable.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for their careful reading of the manuscript and for the positive assessment of its significance. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central derivation is performed at leading order in an expansion in the model's nonlinearity (departure from linearity in parameters). However, no bound, scaling criterion, or regime of validity is supplied for when higher-order terms remain negligible compared with the leading correction. This is load-bearing for the claim that the result 'quantifies when and by how much the MC method departs from Bayesian inference' in general PDF and SMEFT fits, where model curvature may be appreciable.

    Authors: We agree that the derivation is performed at leading order in the nonlinearity parameter. The closed-form result supplies the exact leading correction to the MC replica distribution relative to the Laplace approximation; its magnitude is set by the residual-weighted Hessian, which can be evaluated directly at the best-fit point. In the two single-parameter models we solve exactly, this leading term accounts for the observed discrepancy, with higher-order contributions visible but sub-dominant for moderate nonlinearity. We will revise the abstract to state explicitly that the quantification applies at leading order in the nonlinearity and that the matrix provides a practical diagnostic for the size of the leading departure. A general bound on the remainder would require a higher-order expansion, which lies beyond the present scope. revision: yes

standing simulated objections not resolved
  • A general bound, scaling criterion, or regime of validity quantifying when higher-order terms in the nonlinearity expansion become negligible is not supplied and cannot be derived without extending the analysis to higher orders.

Circularity Check

0 steps flagged

No significant circularity; analytic derivation is self-contained

full rationale

The paper derives the leading-order MC replica distribution analytically from the properties of the replica method applied to a nonlinear model and compares it to the Laplace Bayesian posterior, expressing the difference via the residual-weighted Hessian at the best fit. This is a first-principles expansion in model nonlinearity with no reduction of the claimed result to a fitted input, self-definition, or load-bearing self-citation. The provided text contains no equations or claims that equate the output distribution to its inputs by construction, and the derivation is presented as independent of external fitted quantities beyond the model itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated or can be identified from the provided text.

pith-pipeline@v0.9.1-grok · 5697 in / 1192 out tokens · 41663 ms · 2026-06-30T01:11:08.313777+00:00 · methodology

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

25 extracted references · 23 canonical work pages · 8 internal anchors

  1. [1]

    Neural Network Parametrization of Deep-Inelastic Structure Functions

    S. Forte, L. Garrido, J. I. Latorre, and A. Piccione,Neural network parametrization of deep inelastic structure functions,JHEP05(2002) 062, [hep-ph/0204232]. [2]NNPDFCollaboration, R. D. Ball, L. Del Debbio, S. Forte, A. Guffanti, J. I. Latorre, A. Piccione, J. Rojo, and M. Ubiali,A Determination of parton distributions with faithful uncertainty estimatio...

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  2. [2]

    R. D. Ball, L. Del Debbio, S. Forte, A. Guffanti, J. I. Latorre, J. Rojo, and M. Ubiali,A first unbiased global NLO determination of parton distributions and their uncertainties,Nucl. Phys. B838(2010) 136–206, [arXiv:1002.4407]. [4]NNPDFCollaboration, R. D. Ball, V. Bertone, F. Cerutti, L. Del Debbio, S. Forte, A. Guffanti, J. I. Latorre, J. Rojo, and M. ...

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  3. [3]

    R. D. Ball et al.,Parton distributions with LHC data,Nucl. Phys. B867(2013) 244–289, [arXiv:1207.1303]

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  4. [4]

    NNPDFCollaboration, R. D. Ball et al.,Parton distributions for the LHC Run II,JHEP04 (2015) 040, [arXiv:1410.8849]. [7]NNPDFCollaboration, R. D. Ball et al.,Parton distributions from high-precision collider data,Eur. Phys. J. C77(2017), no. 10 663, [arXiv:1706.00428]. [8]NNPDFCollaboration, R. D. Ball et al.,The path to proton structure at 1% accuracy,Eur...

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  5. [5]

    JAMCollaboration, N. T. Hunt-Smith, C. Cocuzza, W. Melnitchouk, N. Sato, A. W. Thomas, and M. J. White,New Data-Driven Constraints on the Sign of Gluon Polarization in the Proton,Phys. Rev. Lett.133(2024), no. 16 161901, [arXiv:2403.08117]. [12]NNPDFCollaboration, R. D. Ball, J. Cruz-Martinez, L. Del Debbio, S. Forte, Z. Kassabov, E. R. Nocera, J. Rojo, R...

    work page arXiv 2024

  6. [6]

    Courtoy, J

    A. Courtoy, J. Huston, P. Nadolsky, K. Xie, M. Yan, and C. P. Yuan,Parton distributions need representative sampling,Phys. Rev. D107(2023), no. 3 034008, [arXiv:2205.10444]. – 14 –

    work page arXiv 2023

  7. [7]

    Giani, G

    T. Giani, G. Magni, and J. Rojo,SMEFiT: a flexible toolbox for global interpretations of particle physics data with effective field theories,Eur. Phys. J. C83(2023), no. 5 393, [arXiv:2302.06660]. [15]SMEFiTCollaboration, J. J. Ethier, G. Magni, F. Maltoni, L. Mantani, E. R. Nocera, J. Rojo, E. Slade, E. Vryonidou, and C. Zhang,Combined SMEFT interpretati...

    work page arXiv 2023

  8. [8]

    J. J. Ethier, R. Gomez-Ambrosio, G. Magni, and J. Rojo,SMEFT analysis of vector boson scattering and diboson data from the LHC Run II,Eur. Phys. J. C81(2021), no. 6 560, [arXiv:2101.03180]

    work page arXiv 2021

  9. [9]

    N. P. Hartland, F. Maltoni, E. R. Nocera, J. Rojo, E. Slade, E. Vryonidou, and C. Zhang,A Monte Carlo global analysis of the Standard Model Effective Field Theory: the top quark sector, JHEP04(2019) 100, [arXiv:1901.05965]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  10. [10]

    The Gauge-Higgs Legacy of the LHC Run II

    A. Biekötter, T. Corbett, and T. Plehn,The Gauge-Higgs Legacy of the LHC Run II,SciPost Phys.6(2019), no. 6 064, [arXiv:1812.07587]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  11. [11]

    Carrazza, C

    S. Carrazza, C. Degrande, S. Iranipour, J. Rojo, and M. Ubiali,Can New Physics hide inside the proton?,Phys. Rev. Lett.123(2019), no. 13 132001, [arXiv:1905.05215]

    work page arXiv 2019

  12. [12]

    Greljo, S

    A. Greljo, S. Iranipour, Z. Kassabov, M. Madigan, J. Moore, J. Rojo, M. Ubiali, and C. Voisey, Parton distributions in the SMEFT from high-energy Drell-Yan tails,JHEP07(2021) 122, [arXiv:2104.02723]

    work page arXiv 2021

  13. [13]

    Iranipour and M

    S. Iranipour and M. Ubiali,A new generation of simultaneous fits to LHC data using deep learning,JHEP05(2022) 032, [arXiv:2201.07240]

    work page arXiv 2022

  14. [14]

    Kassabov, M

    Z. Kassabov, M. Madigan, L. Mantani, J. Moore, M. Morales Alvarado, J. Rojo, and M. Ubiali,The top quark legacy of the LHC Run II for PDF and SMEFT analyses,JHEP05 (2023) 205, [arXiv:2303.06159]

    work page arXiv 2023

  15. [15]

    Hammou, Z

    E. Hammou, Z. Kassabov, M. Madigan, M. L. Mangano, L. Mantani, J. Moore, M. M. Alvarado, and M. Ubiali,Hide and seek: how PDFs can conceal new physics,JHEP11(2023) 090, [arXiv:2307.10370]

    work page arXiv 2023

  16. [16]

    PBSPCollaboration, M. N. Costantini, E. Hammou, Z. Kassabov, M. Madigan, L. Mantani, M. Morales Alvarado, J. M. Moore, and M. Ubiali,SIMUnet: an open-source tool for simultaneous global fits of EFT Wilson coefficients and PDFs,Eur. Phys. J. C84(2024), no. 8 805, [arXiv:2402.03308]

    work page arXiv 2024

  17. [17]

    E. Cole, M. N. Costantini, E. Hammou, L. Mantani, F. Merlotti, M. Morales-Alvarado, and M. Ubiali,Tailored PDFs for New Physics searches,arXiv:2602.20235

    work page arXiv

  18. [18]

    J. M. Bardsley, A. Solonen, H. Haario, and M. Laine,Randomize-then-optimize: A method for sampling from posterior distributions in nonlinear inverse problems,SIAM Journal on Scientific Computing36(2014), no. 4 A1895–A1910

    2014

  19. [19]

    Efron,Bayesian inference and the parametric bootstrap,The Annals of Applied Statistics6 (Dec., 2012)

    B. Efron,Bayesian inference and the parametric bootstrap,The Annals of Applied Statistics6 (Dec., 2012)

    2012

  20. [20]

    Randomized Prior Functions for Deep Reinforcement Learning

    I. Osband, J. Aslanides, and A. Cassirer,Randomized Prior Functions for Deep Reinforcement Learning, inAdvances in Neural Information Processing Systems (NeurIPS), vol. 31, 2018. arXiv:1806.03335

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  21. [21]

    M. N. Costantini, M. Madigan, L. Mantani, and J. M. Moore,A critical study of the Monte Carlo replica method,JHEP12(2024) 064, [arXiv:2404.10056]. – 15 –

    work page arXiv 2024

  22. [22]

    Del Debbio, T

    L. Del Debbio, T. Giani, and M. Wilson,Bayesian approach to inverse problems: an application to NNPDF closure testing,Eur. Phys. J. C82(2022), no. 4 330, [arXiv:2111.05787]

    work page arXiv 2022

  23. [23]

    Candido, L

    A. Candido, L. Del Debbio, T. Giani, and G. Petrillo,Bayesian inference with Gaussian processes for the determination of parton distribution functions,Eur. Phys. J. C84(2024), no. 7 716, [arXiv:2404.07573]

    work page arXiv 2024

  24. [24]

    M. N. Costantini, L. Mantani, J. M. Moore, and M. Ubiali,A linear PDF model for Bayesian inference,JHEP04(2026) 068, [arXiv:2507.16913]

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  25. [25]

    M. N. Costantini, L. Mantani, J. M. Moore, V. S. Sánchez, and M. Ubiali,Colibri: A new tool for fast-flying PDF fits,Eur. Phys. J. C86(2026), no. 1 22, [arXiv:2510.03391]. – 16 –

    work page arXiv 2026