Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Intrepid MCMC: Metropolis-Hastings with Exploration

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Intrepid MCMC claims a small dose of exploratory jumps lets random-walk Metropolis find every disconnected mode while keeping detailed-balance guarantees.

desk verdict Sound new MH variant with a correct core derivation, but the broad-applicability claim rests on benchmarks where the RTF is trivial, and the recommended truncated-normal proposal has a real symmetry error. read the letter →

arxiv 2411.17639 v1 pith:PPAONMMN submitted 2024-11-26 stat.ME

Promit Chakroborty , Michael D. Shields This is my paper · ORCID
classification stat.ME MSC 60J2265C0562F15
keywords MarkovchainMonteCarloMetropolis-HastingsmultimodalsamplingmodediscoveryradialtransformationfunctionmixturetransitionkernelBayesianinferencerandom-walkexploration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Intrepid MCMC claims that random-walk Metropolis chains can be made to discover and sample all disconnected modes of a multimodal target by mixing in a small fraction of deliberately exploratory jumps. The paper builds a two-component transition kernel: a local component-wise Metropolis step that fills in an already-found mode, and an 'Intrepid' proposal that jumps along approximate contours of a parent density to sweep the parameter space. With as little as 1 to 10 percent of steps devoted to exploration, a single chain finds every mode on nine two-dimensional benchmark targets and on a Bayesian oscillator inference problem, while vanilla Metropolis gets stuck. The method needs no gradient, no tempering, and no prior optimization, and it keeps the same Metropolis-Hastings acceptance-correction logic. If true, this gives engineers a cheap way to get representative samples from expensive, implicitly defined multimodal posteriors.

What carries the argument

The load-bearing object is the mixture transition kernel $K(x,dy)=(1-\beta)K_L(x,dy)+\beta K_I(x,dy)$, where $K_L$ is component-wise Metropolis-Hastings and $K_I$ is the Intrepid proposal. The Intrepid proposal expresses the state in hyperspherical coordinates around a fixed anchor, draws new angles by perturbing the current angles, and draws a radial factor $\gamma$; when a Radial Transformation Function exists, the radius is mapped so the candidate lies on the same parent-density contour as the current state, with the acceptance ratio containing the Jacobian factors that make the kernel reversible. The RTF is the identity for radially symmetric parents such as the Gaussian used in all demonstrations, which is why the reference-direction construction was not needed in the numerical studies.

What would settle it

Run the same nine-target protocol with a parent distribution that is itself bimodal or has non-convex contours, and measure the fraction of 100 chains that visit every mode; if Intrepid with the fallback $r_c = \gamma r_s$ finds modes no more often than component-wise Metropolis, then the contour-preserving mechanism, not the mixture alone, is the reason for the improvement.

Watch

Extended reading notes

Core claim

The paper's central claim is that the convergence failure of random-walk Metropolis on multimodal targets is not inherent: injecting a fraction $\beta$ of exploratory steps built from the parent-distribution geometry creates a Markov chain whose invariant distribution is still the target $\pi$, but whose practical mobility between modes is much better. The Intrepid proposal works in hyperspherical coordinates anchored at a central point; it perturbs the angular coordinates of the current state and rescales the radial coordinate along the parent-density contour, so a candidate can land far away in space while staying on an equal-probability contour of the parent $p(x)$. Detailed balance is restored through a Metropolis acceptance ratio that accounts for the change of variables, and irreducibility is guaranteed by the local component-wise kernel in the mixture. On nine multimodal targets and a two-degree-of-freedom oscillator Bayesian calibration, chains with $\beta \approx 0.1$ converge in distribution and in mean with small variance across trials, whereas component-wise Metropolis alone either misses modes or finds different modes in different runs.

Load-bearing premise

The exploration jump only stays on the intended parent-density contour when the parent distribution has 'probabilistic radial equivalence' around the chosen anchor; all numerical results use a Gaussian parent, where the required transformation is the identity, so the method's advantage for the non-convex and multimodal parents that motivate it is not demonstrated.

Editorial extensions

If this is right

  • A single chain with $\beta=0.1$ finds and samples all modes on all nine two-dimensional benchmark targets, so multi-chain restarts are not needed for those shapes.
  • Because the mixture kernel preserves $\pi$ as the invariant distribution, replacing component-wise Metropolis with Intrepid at $\beta=0.1$ changes the path to the target, not the limiting distribution.
  • The acceptance rate falls only modestly for $\beta \leq 0.1$, so exploration can be bought cheaply; above roughly $\beta=0.3$ the chain wastes samples on exploration and acceptance drops sharply.
  • The method's convergence advantage over component-wise Metropolis shrinks with dimension, becoming comparable around 10 to 50 dimensions, but it does not perform worse in the tested cases.
  • Any target can be cast into the required product form by choosing an arbitrary parent $p$ and setting $T=\pi/p$, extending the method beyond problems that naturally factor.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next test is a parent distribution that is itself multimodal or non-convex, where the fallback $r_c = \gamma r_s$ no longer follows equal-probability contours; the paper's current evidence leaves that regime unverified.
  • Applying the exploratory step component-wise rather than radially in full dimension might reduce the observed degradation above roughly five dimensions; the paper leaves this as future work, and it is a direct extension of the same contour logic.
  • The exploration ratio $\beta$ is chosen empirically at 0.1; an adaptive scheme that tunes $\beta$ per target, or reduces it once modes are discovered, would be a useful and untested extension.
  • The contour-hugging idea could be combined with delayed rejection, adaptive covariance tuning, or ensembles of Intrepid chains to lower the cost of switching modes; the authors mention these combinations as future work rather than demonstrated results.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces Intrepid MCMC, a single-chain Metropolis-Hastings variant that alternates between a component-wise local proposal and a globally explorative proposal in a hyperspherical coordinate system anchored at a fixed point x_a. The explorative proposal perturbs angular coordinates and selects a radial coordinate so that, when a Radial Transformation Function (RTF) exists for the parent density p(x), candidates are generated near equal-probability contours of p; when no RTF exists, the construction falls back to rc = γ rs. The authors derive the MH acceptance ratio for both cases in Appendix A, state irreducibility and aperiodicity via the mixture-kernel construction, provide practical tuning guidance in Section 3.4, and evaluate the algorithm on nine two-dimensional multimodal targets in Section 4.1, a dimension-scaling study in Section 4.2, a mixing study in Section 4.3, and a Bayesian oscillator inference problem in Section 4.4. The central empirical claim is that injecting a small fraction β ≈ 0.1 of exploration steps markedly improves convergence and mode discovery relative to component-wise MH.

Significance. Intrepid MCMC is potentially a useful addition to the random-walk MH toolbox: it preserves much of the simplicity of MH, comes with explicit pseudocode, and the detailed-balance derivation in Appendix A is careful and, as far as I can verify, correct. The empirical study is extensive, with 100 repeated chains per configuration and comparisons across several multimodal targets. The main value, if the scope is properly established, would be a simple single-chain method that can discover disconnected modes that trap ordinary MH. However, the significance as currently presented is limited by a gap between the advertised generality and the evidence: the mode-finding mechanism is only guaranteed when the parent distribution admits a Radial Transformation Function, and all nine analytical examples in Section 4.1 use a radially symmetric Gaussian parent for which the RTF is the identity. The paper's own Section 3.2 states that for multimodal parents the RTF often does not exist, which is precisely the regime the motivating engineering examples target.

major comments (4)
  1. [Section 3.2 and Section 4.1] The central mode-finding mechanism is contour-preserving exploration, which is available only when the parent distribution admits a Radial Transformation Function. Section 3.2 explicitly states that for multimodal parent distributions the RTF often does not exist and that direction-dependent radial proposals are beyond the scope of this work. All nine analytical targets in Section 4.1 use f1, the radially symmetric Gaussian, as the parent, for which the RTF is the identity and the proposal in Eq. (12) reduces to a pure angular move on a circle. The numerical section therefore exercises only the special case where the contour-sweeping property is guaranteed. When the RTF is absent, the fallback rc = γ rs in Eq. (12) is a radial scaling from the anchor, and although the acceptance ratio in Eq. (52) remains valid, there is no mechanism that preserves p-contours and no experimental evidence that this fallback alone discovers disconnected modes. This gap directly affects the abstract's claim of applicability to 'a wide variety of target distribution shapes' and should be addressed by either narrowing the claims or adding experiments with non-Gaussian, non-RTF-admitting parents.
  2. [Section 4.4] The Bayesian oscillator example is the only demonstration with a non-Gaussian parent: the prior is a product of two independent Lognormal distributions, which is not radially symmetric about any point. The paper does not state whether an RTF exists for this parent, whether one was constructed numerically, how the anchor was chosen, or whether the no-RTF fallback was active. Without this information the reader cannot determine whether the reported improvement over CMH comes from the contour-preserving exploration that is the paper's core idea or from the generic radial-scaling fallback. This is load-bearing because Section 4.4 is the paper's principal evidence for real-world applicability, and the missing implementation details should be supplied.
  3. [Section 3.2, Eq. (15)] The claim that for uniform distributions the RTF exists for any anchor, with the formula R1,2(r) = r(λ2/λ1) and λi the maximum radial extent in direction θi, is not correct for non-convex supports. If the support of the uniform distribution is disconnected or non-convex, a ray from the anchor can enter and exit the support multiple times, so the maximum radial extent does not define an order-preserving bijection between the radial coordinates in two directions, and Definition 1 in Appendix B is not satisfied. This is an internal inconsistency in one of the three 'cases of practical interest' listed in Section 3.2, and it should be corrected or qualified by restricting Eq. (15) to star-shaped or convex supports.
  4. [Section 4.2] The dimension-scaling experiment uses the d-dimensional Gauss-Planes target, whose two modes are disconnected half-spaces rather than compact isolated modes. The conclusion that Intrepid MCMC degrades to CMH-like performance for d ≥ 10 may depend on this specific geometry, where the explorative angular proposals must navigate a large angular volume to cross between two unbounded regions. The paper should either acknowledge this limitation or include a compact-mode example in higher dimensions before drawing general conclusions about dimensionality.
minor comments (5)
  1. [Section 3.1] The notation θs = (xs − xa)/rs is inconsistent with the angular-coordinate definition in Eq. (10), since the right-hand side is a unit vector in Cartesian coordinates while the left-hand side is a vector of hyperspherical angles; please clarify the conversion or use separate symbols.
  2. [Eq. (14)] In the RTF case of Eq. (14), the notation R′0,c(·) and Rs,0(rs) is introduced without a compact definition; a short notation table or a sentence explaining the subscript convention would improve readability.
  3. [Figures 3–8] The violin plots are width-normalized to the number of samples and truncated at the measured extremes, which visually exaggerates multi-modality; please state the number of samples used for each violin and the bandwidth details in the captions.
  4. [Section 4.1] The reference empirical distributions are estimated from 50 million rejection-sampling draws, but the effective number of accepted samples per target is not reported; since the TVD is computed against these empirical references, the Monte Carlo error in the reference should be quantified or at least bounded.
  5. [Algorithm 2] Step 8 says to draw γ from qr(γ|θc) or qr(γ) 'as appropriate', but the decision depends on RTF existence, which is only checked in Step 9; reorder the steps so that the RTF existence is determined before the radial draw, and state explicitly which form of qr is used in each branch.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Intrepid proposal and acceptance ratio are derived from first principles, and the acknowledged RTF scope limitation is not a circular reduction.

full rationale

The paper's central derivation is self-contained. The Intrepid proposal q_I and acceptance ratio α_I are obtained from the standard Metropolis-Hastings detailed-balance condition through an explicit change of variables into hyperspherical coordinates (Appendix A, Eqs. 39-52); no equation sets the target result equal to its own input by construction. The Radial Transformation Function existence results (Appendix B, Theorems 1-5 and supporting lemmas) are proved in-paper from definitions of level sets and convexity, and are not imported from the authors' prior work. The irreducibility/aperiodicity argument relies on the Component-wise Metropolis-Hastings component via an external theorem of Tierney and the CMH construction of Au and Beck—not on a self-citation. The self-citations in the reference list (e.g., [7], [14]) appear only in background statements about Bayesian inference and reliability applications and are not load-bearing for the derivation. The tuning constants β, γ0, and angular proposal widths are chosen empirically from the benchmark convergence/acceptance plots and are not presented as predictions; the theoretical invariance of the chain holds for any β in (0,1) and any valid proposal densities. The limitation that the RTF may not exist for multimodal or non-convex parents (Section 3.2) narrows the demonstrated applicability—all nine analytical examples use a Gaussian parent for which the RTF is the identity—but this is an acknowledged scope boundary, not a circular reduction. The fallback rc = γrs in Eq. (12) is paired with a valid MH acceptance ratio in Eq. (52), so the method remains a correct MCMC scheme even when the contour-preserving exploration property is lost. No fitted parameter is renamed as a prediction, and no uniqueness claim from the authors' prior work is used to force a choice.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The derivation leans on standard MCMC theory (detailed balance, mixture-kernel irreducibility) and on a domain assumption that a known parent distribution is available with contour structure matching the anchor choice. No new physical entities are posited. The user-facing tuning parameters are beta, the radial proposal scale gamma_0, the angular proposal scales, and the anchor point; the general-problem trick in Section 3.5 additionally introduces an arbitrary parent choice.

free parameters (5)
  • Exploration ratio beta = 0.1
    Proportion of exploration steps in the mixture kernel (Eq. 6); selected empirically in Section 4.1 because beta=0.1 gave good TVD on the nine test distributions, then adopted in later examples.
  • Radial proposal scale gamma_0 = 2
    Parameter of qr(gamma) = Uniform(1/gamma_0, gamma_0) in Section 3.4; called 'promising' and used for all experiments without sensitivity analysis.
  • Truncated-normal angular scales sigma_j = sigma_{d-1}=pi, sigma_j=pi/2 for j<d-1
    Anecdotal recommendations in Section 3.4 for the alternative TruncNorm angular proposals; not used in experiments, which used uniform angular proposals.
  • Anchor point x_a = mean of parent distribution (origin) in all experiments
    Anchor selection (Section 3.2) determines the coordinate frame for exploration; all experiments use Gaussian parents centered at the anchor, so the RTF is identity.
  • Parent distribution p(x) when decomposition is not natural = arbitrary, e.g., Gaussian
    Section 3.5 permits arbitrary p with T = pi/p; performance depends strongly on this choice, but no concrete guidance beyond examples is given.
assumptions (6)
  • standard math Detailed balance is sufficient for the target pi to be invariant (Chib and Greenberg, Eq. 3).
    Invoked in Section 2 to construct MH kernels; standard result.
  • standard math A mixture of two kernels is irreducible and aperiodic if at least one component kernel is irreducible and aperiodic (Tierney 1994).
    Used in Section 3 to guarantee global chain properties via the CMH component.
  • domain assumption Component-wise Metropolis-Hastings with positive-density full-support Gaussian proposals yields an irreducible, aperiodic chain.
    Invoked in Section 3; holds for the standard Gaussian proposal used in the experiments.
  • domain assumption The target admits a decomposition pi(x) proportional to T(x) p(x) with p known and cheap to evaluate or sample.
    The Intrepid proposal is built on p's contours; if p is unavailable the user must introduce an arbitrary one (Eq. 32).
  • ad hoc to paper For contour-preserving exploration, p must have probabilistic radial equivalence (RTF existence) with the chosen anchor, per Theorems 1-5 in Appendix B.
    This is the paper's core new mathematical premise; it holds for convex-contoured parents such as Gaussians but fails for multimodal or non-convex parents.
  • standard math Change-of-variables Jacobian for hyperspherical coordinates is correctly applied in Appendix A.
    Needed to derive qI; standard calculus.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Intrepid MCMC: Metropolis-Hastings with Exploration." pith.science (2026). https://pith.science/paper/PPAONMMN

@misc{pith2026241117639,
  author       = {Pith},
  title        = {Pith review of: Intrepid MCMC: Metropolis-Hastings with Exploration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PPAONMMN}},
  note         = {Machine review of arXiv:2411.17639}
}
read the original abstract

In engineering examples, one often encounters the need to sample from unnormalized distributions with complex shapes that may also be implicitly defined through a physical or numerical simulation model, making it computationally expensive to evaluate the associated density function. For such cases, MCMC has proven to be an invaluable tool. Random-walk Metropolis Methods (also known as Metropolis-Hastings (MH)), in particular, are highly popular for their simplicity, flexibility, and ease of implementation. However, most MH algorithms suffer from significant limitations when attempting to sample from distributions with multiple modes (particularly disconnected ones). In this paper, we present Intrepid MCMC - a novel MH scheme that utilizes a simple coordinate transformation to significantly improve the mode-finding ability and convergence rate to the target distribution of random-walk Markov chains while retaining most of the simplicity of the vanilla MH paradigm. Through multiple examples, we showcase the improvement in the performance of Intrepid MCMC over vanilla MH for a wide variety of target distribution shapes. We also provide an analysis of the mixing behavior of the Intrepid Markov chain, as well as the efficiency of our algorithm for increasing dimensions. A thorough discussion is presented on the practical implementation of the Intrepid MCMC algorithm. Finally, its utility is highlighted through a Bayesian parameter inference problem for a two-degree-of-freedom oscillator under free vibration.

Figures

Figures reproduced from arXiv: 2411.17639 by the authors.

Figure 1
Figure 1. Visualization of (a) a standard Metropolis-Hastings proposal, and (b) the Intrepid proposal. Both figures [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Target distributions used in Section 4.1. (a) Case 1 (Gauss-Ring), (b) Case 2 (Gauss-Planes), (c) Case 3 [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Violin plots of total variation distance from 100 repeated trials for Intrepid MCMC with varying exploration [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Acceptance rate for varying exploration ratios for the nine distributions considered in Section 4.1. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Violin plots showing convergence of the total variation distance from 100 repeated trials for Intrepid MCMC [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Violin plots showing convergence of the mean from 100 repeated trials for Intrepid MCMC with [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Convergence in mean plot for Section 4.2. The plotted quantity is the Euclidean distance between the true [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Convergence in covariance plot for Section 4.2. The plotted quantity is the Frobenius norm between the true [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: Results for Section 4.3. (a) Plot of total variation distance of the ensemble distribution of the [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Results for the Bayesian inference problem (Section 4.4). (a) Plot of Total Variation Distance vs. chain [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Representative scatterplot of samples generated by (a) CMH and (b) Intrepid MCMC ( [PITH_FULL_IMAGE:figures/full_fig_p020_11.png]
Figure 12
Figure 12. Figure 12: Visualization of (a) a unimodal distribution with convex contours, (b) a unimodal distribution with non [PITH_FULL_IMAGE:figures/full_fig_p023_12.png]
Figure 13
Figure 13. Figure 13: Scatterplot of samples produced by a single Intrepid Markov chain ( [PITH_FULL_IMAGE:figures/full_fig_p033_13.png]
Figure 14
Figure 14. Figure 14: Scatterplot of samples produced by a single CMH Markov chain for all cases from Section 4.1. (a) Case 1 [PITH_FULL_IMAGE:figures/full_fig_p034_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. The tail wags the distribution: Only sample the tails for efficient reliability analysis

    stat.ME 2025-05 conditional novelty 5.0 of 10

    TSS stratifies the input distribution into nested tail sets with geometrically decreasing probabilities, yielding failure-probability estimates with exponentially decaying bias and variance that does not grow as the f...

Reference graph

Works this paper leans on

49 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    Rosenbluth, Marshall N

    Nicholas Metropolis, Arianna W. Rosenbluth, Marshall N. Rosenbluth, and Augusta H. Teller. Equations of state calculations by fast computing machines. The Journal of Chemical Physics, 21(6), 6 1953

    work page 1953

  2. [2]

    Understanding the metropolis-hastings algorithm

    Siddhartha Chib and Edward Greenberg. Understanding the metropolis-hastings algorithm. The American Statistician, 49(4):327–335, 11 1995

    work page 1995

  3. [3]

    Updating models and their uncertainties

    James L Beck and Lambros S Katafygiotis. Updating models and their uncertainties. i: Bayesian statistical framework. Journal of Engineering Mechanics, 124(4):455–461, 1998

    work page 1998

  4. [4]

    A computational framework for bayesian inference in plasticity models characterisation

    Erfan Asaadi and P Stephan Heyns. A computational framework for bayesian inference in plasticity models characterisation. Computer Methods in Applied Mechanics and Engineering, 321:455–481, 2017

    work page 2017

  5. [5]

    Bayesian calibration of hysteretic reduced order structural models for earthquake engineering applications

    Dimitrios Patsialis, Aikaterini P Kyprioti, and Alexandros A Taflanidis. Bayesian calibration of hysteretic reduced order structural models for earthquake engineering applications. Engineering Structures, 224:111204, 2020

    work page 2020

  6. [6]

    Bayesian updating and model class selection with subset simulation

    FA DiazDelaO, A Garbuno-Inigo, SK Au, and I Yoshida. Bayesian updating and model class selection with subset simulation. Computer Methods in Applied Mechanics and Engineering, 317:1102–1121, 2017

    work page 2017

  7. [7]

    The effect of prior probabilities on quantification and propagation of imprecise probabilities resulting from small datasets

    Jiaxin Zhang and Michael D Shields. The effect of prior probabilities on quantification and propagation of imprecise probabilities resulting from small datasets. Computer Methods in Applied Mechanics and Engineering, 334:483–506, 2018

    work page 2018

  8. [8]

    Bayesian system identification based on probability logic

    James L Beck. Bayesian system identification based on probability logic. Structural Control and Health Monitoring, 17(7):825–847, 2010

    work page 2010

Show all 49 references
  1. [9]

    Bayesian inference with subset simulation: strategies and improvements

    Wolfgang Betz, Iason Papaioannou, James L Beck, and Daniel Straub. Bayesian inference with subset simulation: strategies and improvements. Computer Methods in Applied Mechanics and Engineering, 331:72–93, 2018

    2018

  2. [10]

    Adaptive approach for estimation of pipeline corrosion defects via bayesian inference

    Kyeongsu Kim, Gunhak Lee, Keonhee Park, Seongho Park, and Won Bo Lee. Adaptive approach for estimation of pipeline corrosion defects via bayesian inference. Reliability Engineering & System Safety, 216:107998, 2021

    2021

  3. [11]

    Bayesian operator inference for data-driven reduced- order modeling

    Mengwu Guo, Shane A McQuarrie, and Karen E Willcox. Bayesian operator inference for data-driven reduced- order modeling. Computer Methods in Applied Mechanics and Engineering, 402:115336, 2022

    2022

  4. [12]

    Realistic and efficient reliability estimation for aerospace structures

    HJ Pradlwarter, MF Pellissetti, CA Schenk, GI Schueller, A Kreis, S Fransen, A Calvi, and M Klein. Realistic and efficient reliability estimation for aerospace structures. Computer Methods in Applied Mechanics and Engineering, 194(12-16):1597–1617, 2005. 30 A PREPRINT - NOVEMB...

    2005

  5. [13]

    Reliability assessment in structural dynamics

    B Goller, HJ Pradlwarter, and GI Schuëller. Reliability assessment in structural dynamics. Journal of Sound and Vibration, 332(10):2488–2499, 2013

    2013

  6. [14]

    Reliability estimation of an advanced nuclear fuel using coupled active learning, multifidelity modeling, and subset simulation

    Somayajulu LN Dhulipala, Michael D Shields, Promit Chakroborty, Wen Jiang, Benjamin W Spencer, Jason D Hales, Vincent M Laboure, Zachary M Prince, Chandrakanth Bolisetti, and Yifeng Che. Reliability estimation of an advanced nuclear fuel using coupled active learning, multifid...

    2022

  7. [15]

    Accelerated subset simulation with neural networks for reliability analysis.Computer Methods in Applied Mechanics and Engineering, 223:70–80, 2012

    Vissarion Papadopoulos, Dimitris G Giovanis, Nikos D Lagaros, and Manolis Papadrakakis. Accelerated subset simulation with neural networks for reliability analysis.Computer Methods in Applied Mechanics and Engineering, 223:70–80, 2012

    2012

  8. [16]

    Mcmc algorithms for subset simulation

    Iason Papaioannou, Wolfgang Betz, Kilian Zwirglmaier, and Daniel Straub. Mcmc algorithms for subset simulation. Probabilistic Engineering Mechanics, 41:89–103, 2015

    2015

  9. [17]

    Sequential importance sampling for structural reliability analysis

    Iason Papaioannou, Costas Papadimitriou, and Daniel Straub. Sequential importance sampling for structural reliability analysis. Structural safety, 62:66–75, 2016

    2016

  10. [18]

    Active learning for structural reliability: Survey, general framework and benchmark

    Maliki Moustapha, Stefano Marelli, and Bruno Sudret. Active learning for structural reliability: Survey, general framework and benchmark. Structural Safety, 96:102174, 2022

    2022

  11. [19]

    Markov chains for exploring posterior distributions

    Luke Tierney. Markov chains for exploring posterior distributions. the Annals of Statistics, pages 1701–1728, 1994

    1994

  12. [20]

    W. K. Hastings. Monte carlo sampling methods using markov chains and their applications. Biometrika, 57(1):97– 109, 1970

    1970

  13. [21]

    Carlin, Hal S

    Andrew Gelman, John B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin.Bayesian Data Analysis. Chapman and Hall, 3rd edition, 2013

    2013

  14. [22]

    Roberts and Jeffrey S

    Gareth O. Roberts and Jeffrey S. Rosenthal. Optimal scaling for various Metropolis-Hastings algorithms.Statistical Science, 16(4):351 – 367, 2001

    2001

  15. [23]

    Adaptive proposal distribution for random walk Metropolis algorithm

    Heikki Haario, Eero Saksman, and Johanna Tamminen. Adaptive proposal distribution for random walk Metropolis algorithm. Computational Statistics, 14:375 – 395, 1999

    1999

  16. [24]

    An adaptive Metropolis algorithm

    Heikki Haario, Eero Saksman, and Johanna Tamminen. An adaptive Metropolis algorithm. Bernoulli, 7(2):223 – 242, 2001

    2001

  17. [25]

    Some adaptive monte carlo methods for bayesian inference

    Luke Tierney and Antonietta Mira. Some adaptive monte carlo methods for bayesian inference. Statistics in Medicine, 18(17-18):2507–2515, 1999

    1999

  18. [26]

    On metropolis-hastings algorithms with delayed rejection

    Antonietta Mira et al. On metropolis-hastings algorithms with delayed rejection. Metron, 59(3-4):231–241, 2001

    2001

  19. [27]

    Delayed rejection in reversible jump metropolis–hastings

    Peter J Green and Antonietta Mira. Delayed rejection in reversible jump metropolis–hastings. Biometrika, 88(4):1035–1053, 2001

    2001

  20. [28]

    Dram: efficient adaptive mcmc

    Heikki Haario, Marko Laine, Antonietta Mira, and Eero Saksman. Dram: efficient adaptive mcmc. Statistics and computing, 16:339–354, 2006

    2006

  21. [29]

    Ensemble samplers with affine invariance

    Jonathan Goodman and Jonathan Weare. Ensemble samplers with affine invariance. Communications in applied mathematics and computational science, 5(1):65–80, 2010

    2010

  22. [30]

    A markov chain monte carlo version of the genetic algorithm differential evolution: easy bayesian computing for real parameter spaces

    Cajo JF Ter Braak. A markov chain monte carlo version of the genetic algorithm differential evolution: easy bayesian computing for real parameter spaces. Statistics and Computing, 16:239–249, 2006

    2006

  23. [31]

    Differential evolution markov chain with snooker updater and fewer chains

    Cajo JF Ter Braak and Jasper A Vrugt. Differential evolution markov chain with snooker updater and fewer chains. Statistics and Computing, 18:435–446, 2008

    2008

  24. [32]

    Accelerating markov chain monte carlo simulation by differential evolution with self-adaptive randomized subspace sampling

    Jasper A Vrugt, Cajo JF ter Braak, Cees GH Diks, Bruce A Robinson, James M Hyman, and Dave Higdon. Accelerating markov chain monte carlo simulation by differential evolution with self-adaptive randomized subspace sampling. International journal of nonlinear sciences and numeri...

    2009

  25. [33]

    Markov chain monte carlo simulation using the dream software package: Theory, concepts, and matlab implementation

    Jasper A Vrugt. Markov chain monte carlo simulation using the dream software package: Theory, concepts, and matlab implementation. Environmental Modelling & Software, 75:273–316, 2016

    2016

  26. [34]

    A repelling–attracting metropolis algorithm for multimodal- ity

    Hyungsuk Tak, Xiao-Li Meng, and David A van Dyk. A repelling–attracting metropolis algorithm for multimodal- ity. Journal of Computational and Graphical Statistics, 27(3):479–490, 2018

    2018

  27. [35]

    Fully adaptive gaussian mixture metropolis-hastings algorithm

    David Luengo and Luca Martino. Fully adaptive gaussian mixture metropolis-hastings algorithm. In 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, pages 6148–6152. IEEE, 2013

    2013

  28. [36]

    Strategies for improving mcmc

    Walter R Gilks and Gareth O Roberts. Strategies for improving mcmc. In Markov chain Monte Carlo in practice, chapter 6, pages 89–114. Chapman & Hall/CRC, 1996. 31 A PREPRINT - NOVEMBER 27, 2024

    1996

  29. [37]

    Markov chain maximum likelihood

    CJ Geyer. Markov chain maximum likelihood. In Computing Science and Statistics: The 23rd Symposium on the Interface, pages 156–163, 1991

    1991

  30. [38]

    An adaptive parallel tempering algorithm

    Bła˙zej Miasojedow, Eric Moulines, and Matti Vihola. An adaptive parallel tempering algorithm. Journal of Computational and Graphical Statistics, 22(3):649–664, 2013

    2013

  31. [39]

    Simulated tempering: a new monte carlo scheme.Europhysics letters, 19(6):451, 1992

    Enzo Marinari and Giorgio Parisi. Simulated tempering: a new monte carlo scheme.Europhysics letters, 19(6):451, 1992

    1992

  32. [40]

    Adwait Sharma and C.S. Manohar. Modified replica exchange-based mcmc algorithm for estimation of structural reliability based on particle splitting method. Probabilistic Engineering Mechanics, 72:103448, 2023

    2023

  33. [41]

    Sufficient Conditions for Torpid Mixing of Parallel and Simulated Tempering

    Dawn Woodard, Scott Schmidler, and Mark Huber. Sufficient Conditions for Torpid Mixing of Parallel and Simulated Tempering. Electronic Journal of Probability, 14(none):780 – 804, 2009

    2009

  34. [42]

    Smart darting monte carlo.The Journal of Chemical Physics, 114(16):6994–7000, 2001

    Ioan Andricioaei, John E Straub, and Arthur F V oter. Smart darting monte carlo.The Journal of Chemical Physics, 114(16):6994–7000, 2001

    2001

  35. [43]

    Generalized darting monte carlo

    Cristian Sminchisescu and Max Welling. Generalized darting monte carlo. In Artificial Intelligence and Statistics, pages 516–523. PMLR, 2007

    2007

  36. [44]

    Distributed and adaptive darting monte carlo through regenerations

    Sungjin Ahn, Yutian Chen, and Max Welling. Distributed and adaptive darting monte carlo through regenerations. In Artificial Intelligence and Statistics, pages 108–116. PMLR, 2013

    2013

  37. [45]

    A framework for adaptive MCMC targeting multimodal distributions

    Emilia Pompe, Chris Holmes, and Krzysztof Łatuszy´nski. A framework for adaptive MCMC targeting multimodal distributions. The Annals of Statistics, 48(5):2930 – 2952, 2020

    2020

  38. [46]

    Estimation of small failure probabilities in high dimensions by subset simulation

    Siu-Kui Au and James L Beck. Estimation of small failure probabilities in high dimensions by subset simulation. Probabilistic engineering mechanics, 16(4):263–277, 2001

    2001

  39. [47]

    Bayesian updating of structural models and reliability using markov chain monte carlo simulation

    James L Beck and Siu-Kui Au. Bayesian updating of structural models and reliability using markov chain monte carlo simulation. Journal of engineering mechanics, 128(4):380–391, 2002

    2002

  40. [48]

    Bayesian updating with structural reliability methods

    Daniel Straub and Iason Papaioannou. Bayesian updating with structural reliability methods. Journal of Engineering Mechanics, 141(3):04014134, 2015

    2015

  41. [49]

    Convex bodies: the Brunn–Minkowski theory, volume 151

    Rolf Schneider. Convex bodies: the Brunn–Minkowski theory, volume 151. Cambridge university press, 2013. 32 A PREPRINT - NOVEMBER 27, 2024 (a) (b) (c) (d) (e) (f) (g) (h) (i) Figure 13: Scatterplot of samples produced by a single Intrepid Markov chain (β = 0.1) for all cases f...

    2013

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.