pith. sign in

arxiv: 2606.25967 · v1 · pith:R4JD4QEUnew · submitted 2026-06-24 · 📊 stat.ME · math.ST· stat.CO· stat.TH

Slice Monte Carlo Integration

Johannes K. Krondorfer , Christian W. Binder , Matthias Neumann , Wolfgang von der Linden This is my paper

Pith reviewed 2026-06-25 19:04 UTC · model grok-4.3

classification 📊 stat.ME math.STstat.COstat.TH
keywords Slice Monte Carlo integrationsurrogate modelstratified samplingnested samplingnumerical integrationMonte Carlo methodsBayesian inference
0
0 comments X

The pith

SℓMC uses a surrogate to slice the space and stratify prior samples for low-variance integration of expensive targets.

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

The paper presents Slice Monte Carlo integration as a way to handle numerical integrals when the target function is costly to evaluate. A cheap surrogate model undergoes a Nested Sampling-like process that divides the space into slices and draws prior samples inside each slice. The expensive target is then evaluated only on those samples, but now in a stratified manner across the slices. Because slice volumes come from the surrogate alone, the method can allocate target evaluations adaptively according to observed variance. The approach is demonstrated on benchmark problems and is also shown to support posterior sampling.

Core claim

Slice Monte Carlo integration (SℓMC) leverages a Nested Sampling-like procedure on the surrogate to partition the space into informative strata, or slices, while generating samples in the parameter space drawn from the prior within each slice. This enables stratified Monte Carlo integration of the expensive target function over the surrogate-induced partition, yielding an efficient estimate of the target integral. The decoupling of slice volume estimation from target function evaluation allows adaptive, variance-aware allocation of computational effort.

What carries the argument

The surrogate-induced slices created by a Nested Sampling-like procedure, which partition the domain so that prior samples can be grouped and the target integral estimated by stratified Monte Carlo.

If this is right

  • Target evaluations can be allocated slice by slice according to measured variance instead of uniformly.
  • Posterior samples can be generated directly from the stratified draws without additional reweighting steps.
  • The total number of expensive evaluations needed for a given precision is reduced when the surrogate partition is informative.
  • Volume estimation for each slice is performed once on the surrogate and reused for any number of target evaluations.

Where Pith is reading between the lines

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

  • The same slicing idea could be tested on integration tasks outside Bayesian settings, such as expectation values in physics simulations.
  • If the surrogate is a neural network or Gaussian process, one could measure how surrogate accuracy affects the final variance reduction.
  • Extending the adaptive allocation rule to decide when to stop sampling a slice would be a direct next step.

Load-bearing premise

The slices produced from the surrogate must actually lower the variance of the target integral estimate compared with ordinary Monte Carlo sampling.

What would settle it

Run the method and standard Monte Carlo on the same benchmark integral with the same number of target evaluations; if the empirical variance of the SℓMC estimator is not smaller, the efficiency claim is false.

Figures

Figures reproduced from arXiv: 2606.25967 by Christian W. Binder, Johannes K. Krondorfer, Matthias Neumann, Wolfgang von der Linden.

Figure 1
Figure 1. Figure 1: FIG. 1: Relative error [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Required target-evaluation cost [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
read the original abstract

Numerical integration involving expensive target functions is a common bottleneck in Bayesian inference and simulation. When a cheap surrogate is available, standard approaches such as reweighting or importance sampling often suffer from high variance and inefficient use of function evaluations. We introduce Slice Monte Carlo integration (S$\ell$MC), a method that leverages a Nested Sampling-like procedure on the surrogate to partition the space into informative strata, or $\textit{slices}$, while generating samples in the parameter space drawn from the prior within each slice. This enables stratified Monte Carlo integration of the expensive target function over the surrogate-induced partition, yielding an efficient estimate of the target integral. A key advantage of S$\ell$MC is the decoupling of slice volume estimation from target function evaluation, which allows for adaptive, variance-aware allocation of computational effort. We investigate the properties of S$\ell$MC, demonstrate how to efficiently generate posterior samples, and validate the method on simple benchmark problems.

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

2 major / 1 minor

Summary. The paper introduces Slice Monte Carlo Integration (SℓMC) for numerical integration of expensive target functions. It uses a Nested Sampling-like procedure on a surrogate to partition the parameter space into slices, estimates slice volumes from surrogate prior-mass shrinkage (decoupled from target evaluations), draws prior samples within slices, and applies stratified Monte Carlo to the target with adaptive, variance-aware allocation of evaluations. The manuscript investigates the method's properties, shows how to generate posterior samples, and validates it on benchmark problems.

Significance. If the variance reduction holds, SℓMC could offer a practical advantage over reweighting or importance sampling when surrogates are available, by enabling adaptive allocation without target-dependent volume estimates. The benchmark validation provides some empirical support for utility in Bayesian settings, but the lack of any derived bound on the variance reduction factor restricts the assessed generality.

major comments (2)
  1. [Method section on slice definition and partitioning] Method section on slice definition and partitioning: the central efficiency claim requires that surrogate-induced strata satisfy Var(target | slice) ≪ Var(target), yet no theorem, inequality, or condition is stated that relates surrogate-target discrepancy to the achieved variance reduction or to the reliability of the adaptive allocation step.
  2. [Properties investigation] Properties investigation: no worst-case analysis or general bound is given on the variance reduction factor when the surrogate is only moderately informative, which is load-bearing for the claim of yielding an efficient estimate relative to standard Monte Carlo.
minor comments (1)
  1. [Abstract] Abstract: the notation SℓMC and the phrase 'Nested Sampling-like procedure' could be defined on first use for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below, acknowledging where the manuscript can be strengthened through clarification or added discussion.

read point-by-point responses

  1. Referee: Method section on slice definition and partitioning: the central efficiency claim requires that surrogate-induced strata satisfy Var(target | slice) ≪ Var(target), yet no theorem, inequality, or condition is stated that relates surrogate-target discrepancy to the achieved variance reduction or to the reliability of the adaptive allocation step.

    Authors: The manuscript does not contain a formal theorem or inequality linking surrogate-target discrepancy to variance reduction, as the presentation emphasizes the algorithmic structure (Nested Sampling-like partitioning on the surrogate, decoupled volume estimation, and variance-aware allocation) together with empirical validation. The efficiency claim rests on the practical observation that informative surrogates induce strata with reduced conditional variance, which is standard in surrogate-assisted integration methods. We will revise the method section to include an explicit discussion of the conditions under which Var(target | slice) ≪ Var(target) is expected, namely when the surrogate and target exhibit positive correlation over the prior support. revision: yes

  2. Referee: Properties investigation: no worst-case analysis or general bound is given on the variance reduction factor when the surrogate is only moderately informative, which is load-bearing for the claim of yielding an efficient estimate relative to standard Monte Carlo.

    Authors: The properties investigation in the manuscript consists of analysis of the decoupling between volume estimation and target evaluations plus benchmark experiments; no worst-case bound on the variance reduction factor is derived. We agree that such a bound would provide additional context for moderately informative surrogates. Deriving a general, non-vacuous bound requires assumptions on surrogate error that lie outside the current scope. We will add a limitations paragraph in the properties section that qualitatively describes expected performance when the surrogate is only moderately informative and note that the method’s advantage is most pronounced for high-quality surrogates. revision: partial

Circularity Check

0 steps flagged

No significant circularity; method presented as independent procedure

full rationale

The paper defines SℓMC as a new combination of NS-like surrogate partitioning into slices followed by stratified MC on the target, with volume estimation decoupled from target evaluations. No equations, derivations, or self-citations are shown that reduce the efficiency claim or variance reduction to a quantity fitted or defined by the method itself. The central procedure is self-contained against external benchmarks and does not invoke load-bearing self-citations, uniqueness theorems from the authors, or ansatzes smuggled via prior work. Absence of a general bound on variance reduction is a potential correctness gap, not a circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method rests on the domain assumption that a surrogate exists and can be used to create informative strata; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • domain assumption A cheap surrogate is available that can be used to partition the space into informative strata via a nested sampling-like procedure.
    This premise is required for the slice construction step described in the abstract.

pith-pipeline@v0.9.1-grok · 5696 in / 1213 out tokens · 26126 ms · 2026-06-25T19:04:50.532509+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

58 extracted references · 15 canonical work pages

  1. [1]

    and Dose, V

    von der Linden, W. and Dose, V. and von Toussaint, U. , title =. 2014 , isbn =

    2014

  2. [2]

    An introduction to

    Andrieu, Christophe and de Freitas, Nando and Doucet, Arnaud and Jordan, Michael I , journal=. An introduction to. 2003 , doi=

    2003

  3. [3]

    The Journal of the Acoustical Society of America , volume=

    Nested sampling applied in Bayesian room-acoustics decay analysis , author=. The Journal of the Acoustical Society of America , volume=. 2012 , doi=

    2012

  4. [4]

    Parameter estimation for compact binaries with ground-based gravitational-wave observations using the LALInference software library , author =. Phys. Rev. D , volume =. 2015 , month =

    2015

  5. [5]

    Properties of the Binary Neutron Star Merger GW170817 , author =. Phys. Rev. X , volume =. 2019 , month =

    2019

  6. [6]

    Statistics and Computing , volume=

    Understanding predictive information criteria for Bayesian models , author=. Statistics and Computing , volume=. 2014 , doi=

    2014

  7. [7]

    Goggans, Paul and Henderson, R. W. and Xiang, Ning , title =. Proceedings of Meetings on Acoustics , volume =. 2013 , month =

    2013

  8. [8]

    Uncertainty Quantification and Polynomial Chaos Techniques in Computational Fluid Dynamics

    Najm, Habib N. Uncertainty Quantification and Polynomial Chaos Techniques in Computational Fluid Dynamics. Annual Review of Fluid Mechanics. 2009. doi:10.1146/annurev.fluid.010908.165248

    work page doi:10.1146/annurev.fluid.010908.165248 2009

  9. [9]

    2024 , edition =

    Uncertainty Quantification: Theory, Implementation, and Applications , author =. 2024 , edition =. doi:10.1137/1.9781611977844 , url =

    work page doi:10.1137/1.9781611977844 2024

  10. [10]

    Uncertainty Quantification in Climate Modeling and Projection

    Qian, Yun and Jackson, Charles and Giorgi, Filippo and Booth, Ben and Duan, Qingyun and Forest, Chris and others. Uncertainty Quantification in Climate Modeling and Projection. Bulletin of the American Meteorological Society. 2016. doi:10.1175/BAMS-D-15-00297.1

    work page doi:10.1175/bams-d-15-00297.1 2016

  11. [11]

    Physical Review E , volume=

    Nested sampling for materials: The case of hard spheres , author=. Physical Review E , volume=. 2014 , publisher=

    2014

  12. [12]

    Journal of Chemical Theory and Computation , volume=

    Direct computation of the quantum partition function by path-integral nested sampling , author=. Journal of Chemical Theory and Computation , volume=. 2018 , publisher=

    2018

  13. [13]

    and Krondorfer, Johannes K

    Binder, Christian W. and Krondorfer, Johannes K. and Hauser, Andreas W. The ridge integration method and its application to molecular sieving , demonstrated for gas purification via graphdiyne membranes. Mol. Syst. Des. Eng. 2022. doi:10.1039/D2ME00120A

    work page doi:10.1039/d2me00120a 2022

  14. [14]

    2010 , publisher=

    Quantum mechanics and path integrals , author=. 2010 , publisher=

    2010

  15. [15]

    Reviews of modern physics , volume=

    Space-time approach to non-relativistic quantum mechanics , author=. Reviews of modern physics , volume=. 1948 , publisher=

    1948

  16. [16]

    Herman, M. F. and Bruskin, E. J. and Berne, B. J. , title =. The Journal of Chemical Physics , volume =. 1982 , month =

    1982

  17. [17]

    The European Physical Journal B , author =

    Nested sampling for materials , volume =. The European Physical Journal B , author =. 2021 , pages =. doi:10.1140/epjb/s10051-021-00172-1 , number =

    work page doi:10.1140/epjb/s10051-021-00172-1 2021

  18. [18]

    Nested Transition Path Sampling , author =. Phys. Rev. Lett. , volume =. 2018 , month =

    2018

  19. [19]

    and Binder, Christian W

    Krondorfer, Johannes K. and Binder, Christian W. and Hauser, Andreas W. , title =. The Journal of Chemical Physics , volume =. 2023 , month =

    2023

  20. [20]

    and Ricciuto, D

    Lu, D. and Ricciuto, D. , TITLE =. Geoscientific Model Development , VOLUME =. 2019 , NUMBER =

    2019

  21. [21]

    Journal of the American Chemical Society , volume=

    MACE-OFF: Short-Range Transferable Machine Learning Force Fields for Organic Molecules , author=. Journal of the American Chemical Society , volume=. 2025 , doi=

    2025

  22. [22]

    and Finney, Jacob M

    Lu, Fenris and Cheng, Lixue and DiRisio, Ryan J. and Finney, Jacob M. and Boyer, Mark A. and Moonkaen, Pattarapon and others , title =. The Journal of Physical Chemistry A , volume =. 2022 , doi =

    2022

  23. [23]

    and Wilkinson, Richard D

    Uteva, Elena and Graham, Richard S. and Wilkinson, Richard D. and Wheatley, Richard J. , title =. The Journal of Chemical Physics , volume =. 2018 , doi =

    2018

  24. [24]

    Dynamic Nested Sampling: An Improved Algorithm for Parameter Estimation and Evidence Calculation , shorttitle =

    Edward Higson and William Handley and Michael Hobson and Anthony Lasenby , title =. Statistics and Computing , year =. doi:10.1007/s11222-018-9844-0 , publisher =

    work page doi:10.1007/s11222-018-9844-0

  25. [25]

    Monthly Notices of the Royal Astronomical Society , volume =

    Speagle, Joshua S , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2020 , month =

    2020

  26. [26]

    and Schön, Thomas B

    Wills, Adrian G. and Schön, Thomas B. Sequential Monte Carlo: A Unified Review. Annual Review of Control, Robotics, and Autonomous Systems. 2023. doi:10.1146/annurev-control-042920-015119

    work page doi:10.1146/annurev-control-042920-015119 2023

  27. [27]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

    Salomone, Robert and South, Leah F and Drovandi, Christopher and Kroese, Dirk P and Johansen, Adam M , title =. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =. 2025 , month =

    2025

  28. [28]

    and Kass, Robert E

    Tokdar, Surya T. and Kass, Robert E. , title =. WIREs Computational Statistics , volume =. doi:10.1002/wics.56 , year =

    work page doi:10.1002/wics.56

  29. [29]

    Giles, Michael B. , year=. Multilevel Monte Carlo methods , volume=. doi:10.1017/S096249291500001X , journal=

    work page doi:10.1017/s096249291500001x

  30. [30]

    SIAM Review , volume =

    Peherstorfer, Benjamin and Willcox, Karen and Gunzburger, Max , title =. SIAM Review , volume =. 2018 , doi =

    2018

  31. [31]

    SIAM Journal on Scientific Computing , volume =

    Peherstorfer, Benjamin and Willcox, Karen and Gunzburger, Max , title =. SIAM Journal on Scientific Computing , volume =. 2016 , doi =

    2016

  32. [32]

    AIP Conference Proceedings , volume =

    Feroz, Farhan and Skilling, John , title =. AIP Conference Proceedings , volume =. 2013 , month =

    2013

  33. [33]

    AIP Conference Proceedings , volume =

    Skilling, John , title =. AIP Conference Proceedings , volume =. 2012 , month =

    2012

  34. [34]

    AIP Conference Proceedings , volume =

    Skilling, John , title =. AIP Conference Proceedings , volume =. 2009 , doi =

    2009

  35. [35]

    Proceedings , VOLUME =

    Skilling, John , TITLE =. Proceedings , VOLUME =. 2019 , NUMBER =

    2019

  36. [36]

    2021 , eprint=

    Nested Sampling And Likelihood Plateaus , author=. 2021 , eprint=

    2021

  37. [37]

    MNRAS503(1), 1199–1205 (2021) https://doi.org/10.1093/mnras/stab590 arXiv:2010.13884 [stat.CO]

    Fowlie, Andrew and Handley, Will and Su, Liangliang , year=. Nested sampling with plateaus , volume=. Monthly Notices of the Royal Astronomical Society , publisher=. doi:10.1093/mnras/stab590 , number=

    work page doi:10.1093/mnras/stab590

  38. [38]

    , journal =

    Chopin, Nicolas and Robert, Christian P. , journal =. Properties of nested sampling , volume =. 2010 , doi =

    2010

  39. [39]

    Peter Lepage

    A new algorithm for adaptive multidimensional integration , journal =. 1978 , issn =. doi:10.1016/0021-9991(78)90004-9 , url =

    work page doi:10.1016/0021-9991(78)90004-9 1978

  40. [40]

    and Farrar, Glennys R

    Press, William H. and Farrar, Glennys R. , title =. 1990 , issue_date =. doi:10.1063/1.4822899 , journal =

    work page doi:10.1063/1.4822899 1990

  41. [41]

    Caflisch, Russel E. , year=. Monte Carlo and quasi-Monte Carlo methods , volume=. doi:10.1017/S0962492900002804 , journal=

    work page doi:10.1017/s0962492900002804

  42. [42]

    The Annals of Statistics , volume=

    Slice Sampling , author=. The Annals of Statistics , volume=. 2003 , doi=

    2003

  43. [43]

    Monthly Notices of the Royal Astronomical Society , volume=

    MultiNest: an efficient and robust Bayesian inference tool for cosmology and particle physics , author=. Monthly Notices of the Royal Astronomical Society , volume=. 2009 , doi=

    2009

  44. [44]

    Bayesian Analysis , number =

    John Skilling , title =. Bayesian Analysis , number =. 2006 , doi =

    2006

  45. [45]

    AIP Conference Proceedings , volume =

    Skilling, John , title =. AIP Conference Proceedings , volume =. 2004 , month =

    2004

  46. [46]

    Nature Reviews Methods Primers , author =

    Nested sampling for physical scientists , volume =. Nature Reviews Methods Primers , author =. 2022 , pages =. doi:10.1038/s43586-022-00121-x , abstract =

    work page doi:10.1038/s43586-022-00121-x 2022

  47. [47]

    , title =

    Keeton, Charles R. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2011 , month =

    2011

  48. [48]

    Statistics Surveys , number =

    Johannes Buchner , title =. Statistics Surveys , number =. 2023 , doi =

    2023

  49. [49]

    Statistics and Computing , volume=

    Point process-based Monte Carlo estimation , author=. Statistics and Computing , volume=. 2017 , publisher=

    2017

  50. [50]

    Statistical Science , volume=

    Recursive pathways to marginal likelihood estimation with prior-sensitivity analysis , author=. Statistical Science , volume=. 2014 , publisher=. doi:10.1214/13-STS465 , url=

    work page doi:10.1214/13-sts465 2014

  51. [51]

    Bayesian Analysis , number =

    Edward Higson and Will Handley and Mike Hobson and Anthony Lasenby , title =. Bayesian Analysis , number =. 2018 , doi =

    2018

  52. [52]

    The Open Journal of Astrophysics , volume=

    Importance Nested Sampling and the MultiNest Algorithm , author=. The Open Journal of Astrophysics , volume=. 2019 , doi=

    2019

  53. [53]

    , title =

    Bishop, Christopher M. , title =. 2006 , isbn =

    2006

  54. [54]

    Practical

    Snoek, Jasper and Larochelle, Hugo and Adams, Ryan P , editor =. Practical. Advances in

  55. [55]

    Handley, W. J. and Hobson, M. P. and Lasenby, A. N. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2015 , doi =

    2015

  56. [56]

    Journal of Open Source Software , volume =

    Buchner, Johannes , title =. Journal of Open Source Software , volume =. 2021 , doi =

    2021

  57. [57]

    Peter , title =

    Lepage, G. Peter , title =. Journal of Computational Physics , volume =. 2021 , doi =

    2021

  58. [58]

    and Binder, Christian W

    Krondorfer, Johannes K. and Binder, Christian W. and Neumann, Matthias and von der Linden, Wolfgang , title =. 2026 , note =

    2026