pith. machine review for the scientific record. sign in

arxiv: 1306.2144 · v3 · pith:3HMGV22Qnew · submitted 2013-06-10 · 🌌 astro-ph.IM · physics.data-an· stat.CO

Importance Nested Sampling and the MultiNest Algorithm

F. Feroz , M.P. Hobson , E. Cameron , A.N. Pettitt This is my paper

Pith reviewed 2026-05-17 20:34 UTC · model grok-4.3

classification 🌌 astro-ph.IM physics.data-anstat.CO
keywords importance nested samplingMultiNestBayesian evidencenested samplingmodel selectionmulti-modal posteriorsMonte Carlo integrationastrophysical inference
0
0 comments X

The pith

Importance nested sampling reuses all MultiNest points to estimate Bayesian evidence up to ten times more accurately than standard nested sampling.

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

The paper presents importance nested sampling as an alternative way to sum the points already generated by the MultiNest algorithm. Instead of discarding points that fall below the current likelihood threshold during the nested sampling process, it treats the entire collection of points as a pseudo-importance sample for the evidence integral. This produces more accurate model evidence values without requiring any change to how MultiNest explores the parameter space. The improvement matters for Bayesian model comparison in multi-modal problems common in astrophysics, where small errors in evidence can affect which model is preferred.

Core claim

The paper claims that treating the full set of MultiNest draws, including those previously discarded under constrained likelihood sampling, as a pseudo-importance sample allows calculation of the Bayesian evidence at up to an order of magnitude higher accuracy than vanilla nested sampling while leaving the exploration procedure unchanged.

What carries the argument

importance nested sampling (INS) as an alternative summation that reuses the complete set of MultiNest points as a pseudo-importance sample for the evidence integral

If this is right

  • INS can be applied after any existing MultiNest run without additional sampling or parameter-space exploration.
  • Higher evidence accuracy improves the reliability of Bayesian model selection for multimodal posteriors.
  • The method requires no modification to the core MultiNest algorithm or its constrained sampling steps.
  • Test results on challenging problems show the accuracy gain holds across different dimensionalities and modalities.

Where Pith is reading between the lines

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

  • The same reuse principle could be tested on other nested sampling implementations to see whether the accuracy gain is specific to MultiNest or general.
  • If the unbiasedness assumption holds, repeated independent MultiNest runs might become less necessary for reaching target precision.
  • The approach suggests a broader way to recycle intermediate samples in any constrained Monte Carlo scheme for integration tasks.

Load-bearing premise

The complete collection of points gathered by MultiNest, including previously discarded ones, forms an unbiased pseudo-importance sample from which the evidence can be summed directly.

What would settle it

Run both vanilla nested sampling and INS on a test problem whose true evidence value is known exactly by direct integration, then check whether the INS error bars are consistently smaller by roughly a factor of ten for the same number of likelihood evaluations.

read the original abstract

Bayesian inference involves two main computational challenges. First, in estimating the parameters of some model for the data, the posterior distribution may well be highly multi-modal: a regime in which the convergence to stationarity of traditional Markov Chain Monte Carlo (MCMC) techniques becomes incredibly slow. Second, in selecting between a set of competing models the necessary estimation of the Bayesian evidence for each is, by definition, a (possibly high-dimensional) integration over the entire parameter space; again this can be a daunting computational task, although new Monte Carlo (MC) integration algorithms offer solutions of ever increasing efficiency. Nested sampling (NS) is one such contemporary MC strategy targeted at calculation of the Bayesian evidence, but which also enables posterior inference as a by-product, thereby allowing simultaneous parameter estimation and model selection. The widely-used MultiNest algorithm presents a particularly efficient implementation of the NS technique for multi-modal posteriors. In this paper we discuss importance nested sampling (INS), an alternative summation of the MultiNest draws, which can calculate the Bayesian evidence at up to an order of magnitude higher accuracy than `vanilla' NS with no change in the way MultiNest explores the parameter space. This is accomplished by treating as a (pseudo-)importance sample the totality of points collected by MultiNest, including those previously discarded under the constrained likelihood sampling of the NS algorithm. We apply this technique to several challenging test problems and compare the accuracy of Bayesian evidences obtained with INS against those from vanilla NS.

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 / 2 minor

Summary. The manuscript introduces importance nested sampling (INS) as an alternative summation technique for the points generated by the MultiNest algorithm. By treating the complete set of live and discarded points as a pseudo-importance sample, the authors claim that the Bayesian evidence can be estimated with up to an order of magnitude higher accuracy than standard nested sampling without altering the ellipsoidal exploration strategy. The approach is demonstrated on several challenging test problems with comparisons to vanilla NS.

Significance. If the central claim is substantiated, INS would improve the efficiency of evidence estimation for multi-modal posteriors in astrophysical applications by better utilizing all samples already generated during a MultiNest run. The empirical tests on test problems provide initial support for reduced variance in the evidence estimator.

major comments (2)
  1. [§3] The derivation of the importance weights (likely in §3) does not explicitly show that the weights equal the ratio of the target prior measure to the effective sampling density induced by the sequence of evolving likelihood constraints and ellipsoidal proposals; without this step the estimator is not guaranteed to be unbiased rather than merely lower-variance.
  2. [§4] Table or figure reporting the accuracy comparisons (likely §4 or §5) shows only point estimates of evidence error; no repeated-run variance, effective sample size, or statistical test is provided to confirm that the reported order-of-magnitude improvement is robust and not due to post-hoc selection of favorable realizations.
minor comments (2)
  1. [Abstract] The abstract states 'up to an order of magnitude higher accuracy' without specifying the error metric (relative error on Z, variance of log Z, etc.) or naming the test problems.
  2. [Notation] Notation for the constrained prior volumes X(L > L_k) should be aligned with standard NS literature to avoid confusion when the weights are introduced.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and their constructive comments. We address each of the major comments below.

read point-by-point responses

  1. Referee: [§3] The derivation of the importance weights (likely in §3) does not explicitly show that the weights equal the ratio of the target prior measure to the effective sampling density induced by the sequence of evolving likelihood constraints and ellipsoidal proposals; without this step the estimator is not guaranteed to be unbiased rather than merely lower-variance.

    Authors: We agree with the referee that an explicit demonstration is needed to establish that the importance weights correspond to the ratio of the target prior measure to the effective sampling density induced by the sequence of likelihood constraints and ellipsoidal proposals. This step is required to rigorously confirm that the estimator is unbiased. In the revised manuscript we will expand the derivation in §3 to include this explicit calculation. revision: yes

  2. Referee: [§4] Table or figure reporting the accuracy comparisons (likely §4 or §5) shows only point estimates of evidence error; no repeated-run variance, effective sample size, or statistical test is provided to confirm that the reported order-of-magnitude improvement is robust and not due to post-hoc selection of favorable realizations.

    Authors: The referee correctly notes that the current comparisons rely on single-run point estimates. To demonstrate robustness we will add results from multiple independent runs on the test problems, including estimates of variance, effective sample size, and statistical comparisons between the INS and vanilla NS estimators. These will be incorporated into the relevant section of the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: INS is an independent re-summation of existing MultiNest draws

full rationale

The paper introduces importance nested sampling as an alternative summation applied to the complete collection of points already generated by the MultiNest algorithm. The central step treats these points (live and previously discarded) as a pseudo-importance sample whose weighted sum estimates the evidence. This summation uses the points as given inputs and does not introduce any fitted parameters, self-referential definitions, or load-bearing self-citations that reduce the claimed accuracy gain to a quantity defined by the paper's own equations. Accuracy is assessed via direct comparison on external test problems rather than by algebraic identity with the input draws. The derivation chain therefore remains self-contained against the collected samples and does not collapse by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper introduces no new free parameters, axioms, or invented entities; it relies on the existing MultiNest sampling procedure and standard nested sampling assumptions.

pith-pipeline@v0.9.0 · 5574 in / 1132 out tokens · 60868 ms · 2026-05-17T20:34:47.957823+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 17 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Peering down the barrel with DESI DR2: 10 000+ inflows at $z$ < 0.6 reveal how galaxies accrete cold gas

    astro-ph.GA 2026-05 unverdicted novelty 7.0

    A large DESI sample reveals thousands of infalling cold gas absorbers at low redshift, with velocity distributions indicating multiple accretion pathways including radial inflows and satellite accretion.

  2. Direct Images of CO2 Absorption in the Atmosphere of a Super-Jupiter: Enhanced Metallicity Suggestive of Formation in a Disk

    astro-ph.EP 2026-04 unverdicted novelty 7.0

    Direct imaging reveals enhanced metallicity and aligned orbit in 29 Cyg b, supporting protoplanetary disk formation near the deuterium burning limit.

  3. High-dimensional inference for the $\gamma$-ray sky with differentiable programming

    astro-ph.HE 2026-04 unverdicted novelty 7.0

    A differentiable forward model and likelihood enable probabilistic inference over many spatial morphologies for the Galactic Center gamma-ray Excess using variational methods on GPUs.

  4. Rapid and robust simulation-based inference for kilonovae

    astro-ph.IM 2026-05 unverdicted novelty 6.0

    Simulation-based inference with a Gaussian process emulator trained on ~1300 POSSIS simulations enables rapid, robust kilonova parameter estimation that avoids MCMC biases from likelihood misspecification.

  5. The Range of Cumulative XUV Flux on GJ 1132 b

    astro-ph.EP 2026-05 unverdicted novelty 6.0

    GJ 1132 b is estimated to have received at least 50 times the cumulative XUV flux of modern Earth with over 95% probability across models, supporting its classification as an atmosphere-free world.

  6. Analytical and Machine Learning Methods for Model Discernment at CE$\nu$NS Experiments

    hep-ph 2026-04 unverdicted novelty 6.0

    Shape correlations in CEνNS allow likelihood and CNN analyses to discriminate sterile neutrinos from NSI and approximately localize sterile parameters in favorable regions.

  7. dynesty: A Dynamic Nested Sampling Package for Estimating Bayesian Posteriors and Evidences

    astro-ph.IM 2019-04 accept novelty 6.0

    dynesty is an open-source Python package for dynamic nested sampling that improves efficiency in Bayesian posterior and evidence estimation compared to MCMC on certain problems.

  8. Inferring the star-formation histories of massive quiescent galaxies with BAGPIPES: Evidence for multiple quenching mechanisms

    astro-ph.GA 2017-12 unverdicted novelty 6.0

    BAGPIPES fitting of 9289 massive quiescent galaxies shows most SFHs rise gradually then quench in 1-2 Gyr, with faster quenching at z>1 and slower at z<1, interpreted as multiple AGN feedback and gas-supply mechanisms.

  9. Morphological and Star Formation Properties of Cosmic Noon Massive Quiescent Galaxies

    astro-ph.GA 2026-05 unverdicted novelty 5.0

    Massive quiescent galaxies at cosmic noon are compact and bulge-dominated with inside-out quenching, where inner regions formed stars ~0.5 Gyr earlier and quenched faster than outskirts.

  10. Sample continuation in Bayesian hierarchical model via variational inference

    stat.ME 2026-04 unverdicted novelty 5.0

    Augmented SVGD with birth-death sampling tracks posterior sample changes as prior shape parameters vary in non-conjugate hierarchical models, enabling mode discovery during transitions from unimodal to multimodal posteriors.

  11. Constraints on Primordial Black Holes from Galactic Diffuse Synchrotron Emissions

    hep-ph 2026-01 conditional novelty 5.0

    Galactic synchrotron emissions above 20 MHz can set tighter upper limits on the abundance of primordial black holes with masses above 10^16 grams than previous cosmic-ray electron data.

  12. Symbolic Emulators for Cosmology: Accelerating Cosmological Analyses Without Sacrificing Precision

    astro-ph.CO 2025-10 unverdicted novelty 5.0

    Symbolic emulators approximate key Lambda CDM functions to 0.001-0.05% accuracy across relevant redshifts and Omega_m values, enabling faster 3x2pt inference with consistent results.

  13. Tests of General Relativity with Binary Black Holes from the second LIGO-Virgo Gravitational-Wave Transient Catalog

    gr-qc 2020-10 accept novelty 5.0

    No evidence for deviations from general relativity is found in LIGO-Virgo binary black hole events, with improved constraints on waveform parameters, graviton mass, and ringdown properties.

  14. Bilby: A user-friendly Bayesian inference library for gravitational-wave astronomy

    astro-ph.IM 2018-11 accept novelty 5.0

    Bilby introduces a user-friendly Python library for accurate Bayesian inference on gravitational-wave signals from compact binaries and other sources, including hierarchical population modeling.

  15. KiDS+VIKING-450 cosmology with Bayesian hierarchical model redshift distributions

    astro-ph.CO 2026-05 conditional novelty 4.0

    Bayesian hierarchical modeling of photometric redshifts in KiDS+VIKING-450 raises S8 to 0.756 ± 0.039 and reduces Planck tension to 1.9σ.

  16. New $H(z)$ measurement at Redshift = 0.12 with DESI Data Release 1

    astro-ph.CO 2026-01 unverdicted novelty 4.0

    New measurement of the Hubble parameter H(z=0.12) = 71.33 ± 4.20 km s^{-1} Mpc^{-1} obtained from cosmology-independent stellar ages of passive galaxies in DESI Data Release 1.

  17. Tests of General Relativity with GWTC-3

    gr-qc 2021-12 accept novelty 3.0

    No evidence for physics beyond general relativity is found in the analysis of 15 GW events from GWTC-3, with consistency in residuals, PN parameters, and remnant properties.

Reference graph

Works this paper leans on

7 extracted references · 7 canonical work pages · cited by 17 Pith papers · 4 internal anchors

  1. [1]

    Initializing adaptive importance sampling with Markov chains

    Allanach, B. C., and Lester, C. G. (2008), “Sampling using a bank of clues,” Computer Physics Communications, 179, 256–266. Beaujean, F., and Caldwell, A. (2013), “Initializing adaptive importance sampling with Markov chains,” ArXiv e-prints [arXiv:1304.7808], . Billingsley, P. (1995), “Probability and measure.,” John Wiley & Sons, New York, . Bridges, M....

    work page internal anchor Pith review Pith/arXiv arXiv 2008

  2. [2]

    Cluster detection in weak lensing surveys

    Cornuet, J., Marin, J.-M., Mira, A., and Robert, C. P. (2012), “Adaptive multiple importance sampling,” Scandinavian Journal of Statistics, 39(4), 798–812. Corsaro, E., and De Ridder, J. (2014), “DIAMONDS: A new Bayesian nested sampling tool,” A&A, 571, A71. Feroz, F., Balan, S. T., and Hobson, M. P. (2011a), “Bayesian evidence for two companions orbiting...

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  3. [3]

    Consistency of the Adaptive Multiple Importance Sampling

    Kong, A., McCullagh, P., Meng, X.-L., Nicolae, D., and Tan, Z. (2003), “A theory of statistical models for Monte Carlo integration,” Journal of the Royal Statistical Society: Series B (Statistical Methodology), 65(3), 585–604. Liu, J. (2008), Monte Carlo Strategies in Scientific Computing, Springer Series in Statistics Series. Mackay, D. J. C. (2003), Info...

    work page internal anchor Pith review Pith/arXiv arXiv 2003

  4. [4]

    Theory of binless multi-state free energy estimation with applications to protein-ligand binding,

    Tan, Z., Gallicchio, E., Lapelosa, M., and Levy, R. M. (2012), “Theory of binless multi-state free energy estimation with applications to protein-ligand binding,” The Journal of Chemical Physics, 136, 144102. Teachey, A., and Kipping, D. (2018), “Evidence for a large exomoon orbiting Kepler-1625b,” Science Advances, 4,

    work page 2012

  5. [5]

    Bayes in the sky: Bayesian inference and model selection in cosmology,

    Trotta, R. (2008), “Bayes in the sky: Bayesian inference and model selection in cosmology,” Contemporary Physics, 49, 71–104. Vardi, Y . (1985), “Empirical distributions in selection bias models,”The Annals of Statistics, pp. 178–203. Veach, E., and Guibas, L. (1995), “Bidirectional estimators for light transport,” in Photorealistic Rendering TechniquesSp...

    work page 2008

  6. [6]

    A remarkably simple and accurate method for computing the Bayes Factor from a Markov chain Monte Carlo Simulation of the Posterior Distribution in high dimension

    Weinberg, M. D., Yoon, I., and Katz, N. (2013), “A remarkably simple and accurate method for computing the Bayes Factor from a Markov chain Monte Carlo Simulation of the Posterior Distribution in high dimension,” ArXiv e-prints [arXiv:1301.3156], . White, M. J., and Feroz, F. (2010), “MSSM dark matter measurements at the LHC without squarks and sleptons,”...

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  7. [7]

    On the convergence properties of the EM algorithm,

    Wu, C. (1983), “On the convergence properties of the EM algorithm,” The Annals of Statistics, 11(1), 95–103. This paper was built using the Open Journal of Astrophysics L ATEX template. The OJA is a journal which provides fast and easy peer review for new papers in the astro-ph section of the arXiv, making the reviewing process simpler for authors and ref...

    work page 1983