pith. sign in

arxiv: 2605.15356 · v1 · pith:WVTAAZQ5new · submitted 2026-05-14 · 🧮 math.NA · cs.NA· stat.ML

Proposal-Guided Greedy Surrogate Refinement for PDE-Driven High-Dimensional Rare-Event Estimation

Zhiwei Gao , George Karniadakis This is my paper

Pith reviewed 2026-05-19 15:39 UTC · model grok-4.3

classification 🧮 math.NA cs.NAstat.ML
keywords rare-event simulationadaptive importance samplingsurrogate modelinghigh-dimensional PDEslatent spacegreedy selectionprobability of failure
0
0 comments X p. Extension
pith:WVTAAZQ5 Add to your LaTeX paper What is a Pith Number? 
\usepackage{pith}
\pithnumber{WVTAAZQ5}

Prints a linked pith:WVTAAZQ5 badge after your title and writes the identifier into PDF metadata. Compiles on arXiv with no extra files. Learn more

The pith

A surrogate model refined locally along an evolving adaptive proposal estimates rare-event probabilities for high-dimensional PDEs with far fewer expensive evaluations.

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

The paper develops a surrogate-assisted version of adaptive importance sampling in which the surrogate is updated only in the failure-relevant region identified by the current proposal distribution. An encoder maps inputs to a latent space where a greedy rule selects a balanced set of candidates near the estimated failure boundary for high-fidelity PDE evaluation. One-step stability bounds are proved that relate local surrogate errors to the quality of the subsequent proposal update, together with bounds on misclassification and finite-sample estimation error. Experiments on multimodal test functions and PDE-driven rare-event problems up to 100 dimensions show probability estimates that match those obtained when the true model drives the adaptation, yet at substantially lower cost in high-fidelity solves.

Core claim

By coupling surrogate refinement to the evolving proposal of cross-entropy adaptive importance sampling and using a greedy latent-space selection rule that trades off boundary proximity against diversity, accurate rare-event probability estimates are obtained for expensive PDE models in high dimensions while keeping the number of high-fidelity evaluations far below what a globally accurate surrogate or a true-model adaptive sampler would require.

What carries the argument

The greedy latent-space selection rule that chooses which proposal samples to evaluate with the high-fidelity PDE model by balancing proximity to the current estimate of the failure boundary and sample diversity.

If this is right

  • Rare-event probabilities for PDE models that are too costly to evaluate at every iteration become computable in dimensions up to at least 100.
  • The total number of high-fidelity PDE solves needed for a target accuracy drops substantially compared with both global surrogate construction and true-model adaptive importance sampling.
  • Error guarantees for the final probability estimate can be stated in terms of local rather than global surrogate accuracy.
  • The same local-refinement idea extends directly to other adaptive sampling schemes that maintain an evolving proposal.

Where Pith is reading between the lines

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

  • The latent-space representation may prove useful for problems whose failure boundaries move with time or with additional parameters.
  • Combining the greedy selection with residual-based physics-informed losses could further lower the required number of high-fidelity solves.
  • The stability analysis supplies a quantitative criterion for deciding when a surrogate update is safe enough to trigger the next proposal adaptation.

Load-bearing premise

The local surrogate errors produced by the greedy selection rule remain small enough that the one-step proposal stability bounds continue to hold and the adaptive updates do not drift.

What would settle it

A controlled 50-dimensional PDE experiment in which the surrogate is deliberately allowed to misclassify a growing fraction of the failure boundary; if the estimated failure probability then deviates by more than the derived error bound from the true-model reference, the stability claim is falsified.

read the original abstract

Accurate surrogate construction for PDE-driven high-dimensional rare-event simulation is challenging when performance evaluations are expensive. Since a globally accurate surrogate may require many high-fidelity evaluations, adaptive importance sampling provides a natural localization tool: its evolving proposal distribution progressively identifies the failure-relevant region. Motivated by this observation, we propose a surrogate-assisted adaptive importance sampling framework that refines the surrogate locally along the evolving proposal, rather than over the entire input space. The surrogate combines an encoder with a neural network, providing a low-dimensional latent representation for both prediction and sample selection. At each adaptive iteration, candidates drawn from the current proposal are selected by a greedy latent-space rule balancing proximity to the estimated failure boundary and sample diversity. The selected samples are evaluated by the high-fidelity model and used to refine the surrogate, which then guides the subsequent cross-entropy-type adaptive proposal update. We establish one-step proposal stability bounds under local surrogate errors, together with surrogate-induced misclassification and finite-sample estimation error bounds. Numerical experiments on multimodal benchmarks and PDE-driven rare-event problems up to 100 dimensions show that the proposed method achieves accuracy comparable to true-model adaptive importance sampling while requiring substantially fewer high-fidelity evaluations.

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 proposes a surrogate-assisted adaptive importance sampling framework for PDE-driven high-dimensional rare-event estimation. It combines an encoder-neural network surrogate with a greedy latent-space selection rule to refine the surrogate locally along the evolving proposal distribution. The method performs repeated cross-entropy proposal updates guided by the current surrogate. Theoretical results include one-step proposal stability bounds under local surrogate errors, surrogate-induced misclassification bounds, and finite-sample estimation error bounds. Experiments on multimodal benchmarks and PDE-driven problems up to 100 dimensions report accuracy comparable to true-model adaptive importance sampling while using substantially fewer high-fidelity evaluations.

Significance. If the central claims hold, the work offers a practical reduction in high-fidelity PDE evaluations for rare-event problems in high dimensions by localizing surrogate refinement via adaptive proposals. The provision of one-step stability and misclassification bounds supplies theoretical support beyond pure heuristics, and the scalability demonstrations to 100 dimensions are a positive indicator for applicability in reliability analysis and uncertainty quantification. The integration of latent-space greedy selection with cross-entropy adaptation is a clear strength.

major comments (2)
  1. [§3] §3 (one-step proposal stability results): The one-step stability bounds under local surrogate errors from the greedy latent-space rule do not automatically control accumulated drift across the sequence of cross-entropy proposal updates. An explicit multi-step contraction, total-variation, or KL-drift bound is required to ensure the final estimator bias and variance remain controlled when persistent boundary misclassification occurs.
  2. [§5] §5 (numerical experiments, 100-dimensional PDE cases): The reported accuracy gains relative to true-model AIS rest on finite-sample performance; without explicit verification that the observed misclassification rates stay inside the regime where the one-step bounds remain valid, it is unclear whether the finite-sample error bounds are tight or whether the gains depend on post-hoc choice of latent dimension and selection threshold.
minor comments (2)
  1. [Abstract] The abstract states that stability, misclassification, and finite-sample bounds are established; the precise statements (e.g., in probability or almost-sure) and the section numbers should be referenced explicitly for reader convenience.
  2. [Notation and Algorithm] Notation for the latent dimension and the greedy selection weighting/threshold should be introduced once and used consistently in both the algorithm description and the bound statements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each of the major comments in detail below, indicating the revisions we intend to make to strengthen the paper.

read point-by-point responses

  1. Referee: [§3] §3 (one-step proposal stability results): The one-step stability bounds under local surrogate errors from the greedy latent-space rule do not automatically control accumulated drift across the sequence of cross-entropy proposal updates. An explicit multi-step contraction, total-variation, or KL-drift bound is required to ensure the final estimator bias and variance remain controlled when persistent boundary misclassification occurs.

    Authors: We agree that controlling the accumulated drift over multiple iterations is important for rigorously bounding the final estimator's bias and variance. The current one-step bounds provide local control at each cross-entropy update. In the revised manuscript, we will extend this to a multi-step analysis. Specifically, we will derive a recursive bound on the total variation distance between the surrogate-guided proposal and the true-model proposal, assuming that the local surrogate error decreases monotonically with the number of refinements. This will yield a contraction factor that ensures the accumulated error remains bounded over the finite number of adaptive iterations used in practice. We will add this analysis to Section 3. revision: yes

  2. Referee: [§5] §5 (numerical experiments, 100-dimensional PDE cases): The reported accuracy gains relative to true-model AIS rest on finite-sample performance; without explicit verification that the observed misclassification rates stay inside the regime where the one-step bounds remain valid, it is unclear whether the finite-sample error bounds are tight or whether the gains depend on post-hoc choice of latent dimension and selection threshold.

    Authors: Thank you for highlighting this aspect of the experimental validation. To address this, we will augment the numerical experiments section with additional diagnostics. Specifically, we will report the empirical misclassification rates for the surrogate in the 100-dimensional PDE examples and confirm that they lie within the error regime assumed by the one-step bounds. Furthermore, we will include a sensitivity study varying the latent dimension and the greedy selection threshold, demonstrating that the observed accuracy improvements are consistent across reasonable choices of these hyperparameters and not reliant on specific post-hoc tuning. These additions will be incorporated into Section 5. revision: yes

Circularity Check

0 steps flagged

No circularity: bounds and experiments are independent of fitted surrogate inputs

full rationale

The paper's derivation establishes one-step proposal stability bounds under local surrogate errors together with misclassification and finite-sample error bounds, then validates the overall procedure via numerical experiments on multimodal and PDE-driven problems up to 100 dimensions. These elements do not reduce any reported accuracy or guarantee to a quantity defined in terms of the fitted surrogate parameters themselves; the adaptive proposal updates and greedy selection rule are analyzed with explicit error terms rather than by construction. No self-citation is invoked as a load-bearing uniqueness theorem, and the central performance claim rests on external high-fidelity comparisons rather than tautological renaming or fitted-input predictions. The derivation chain therefore remains self-contained against the stated assumptions and benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The framework assumes that a low-dimensional latent representation suffices to capture proximity to the failure boundary and that local surrogate errors remain small enough for the one-step stability bounds to apply; these are domain assumptions rather than derived results.

free parameters (2)
  • latent dimension
    Chosen to balance representation power and sample selection; value not stated in abstract but required for the encoder-NN surrogate.
  • greedy selection threshold or weighting
    Balances proximity to estimated failure boundary against sample diversity; appears as a tunable hyper-parameter in the selection rule.
axioms (2)
  • domain assumption The evolving proposal distribution progressively identifies the failure-relevant region.
    Invoked to justify local rather than global surrogate refinement.
  • domain assumption Local surrogate errors permit one-step proposal stability.
    Required for the established stability bounds to remain valid.

pith-pipeline@v0.9.0 · 5746 in / 1559 out tokens · 36336 ms · 2026-05-19T15:39:19.719850+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · 1 internal anchor

  1. [1]

    Introduction to rare event simulation , volume 5

    James Antonio Bucklew and J Bucklew. Introduction to rare event simulation , volume 5. Springer, 2004

    work page 2004

  2. [2]

    Rare event simulation using Monte Carlo methods , volume 73

    Gerardo Rubino, Bruno Tuffin, et al. Rare event simulation using Monte Carlo methods , volume 73. Wiley Online Library, 2009

    work page 2009

  3. [3]

    Cross-entropy-based adaptive importance sampling using gaus- sian mixture

    Nolan Kurtz and Junho Song. Cross-entropy-based adaptive importance sampling using gaus- sian mixture. Structural Safety, 42:35–44, 2013

    work page 2013

  4. [4]

    Improved cross entropy-based impor- tance sampling with a flexible mixture model

    Iason Papaioannou, Sebastian Geyer, and Daniel Straub. Improved cross entropy-based impor- tance sampling with a flexible mixture model. Reliability Engineering & System Safety , 191:106564, 2019

    work page 2019

  5. [5]

    Safe cross-entropy-based importance sampling for rare event simulations

    Zhiwei Gao and George Karniadakis. Safe cross-entropy-based importance sampling for rare event simulations. arXiv preprint arXiv:2509.07160 , 2025

    work page arXiv 2025

  6. [6]

    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

    work page 2001

  7. [7]

    Bayesian subset simulation

    Julien Bect, Ling Li, and Emmanuel Vazquez. Bayesian subset simulation. SIAM/ASA Journal on Uncertainty Quantification, 5(1):762–786, 2017

    work page 2017

  8. [8]

    Sequential monte carlo for rare event estimation

    Fr´ ed´ eric C´ erou, Pierre Del Moral, Teddy Furon, and Arnaud Guyader. Sequential monte carlo for rare event estimation. Statistics and computing , 22(3):795–808, 2012

    work page 2012

  9. [9]

    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

    work page 2016

  10. [10]

    Large deviation theory-based adaptive importance sampling for rare events in high dimensions

    Shanyin Tong and Georg Stadler. Large deviation theory-based adaptive importance sampling for rare events in high dimensions. SIAM/ASA Journal on Uncertainty Quantification , 11(3):788–813, 2023

    work page 2023

  11. [11]

    Multilevel sequential importance sampling for rare event estimation

    Fabian Wagner, Jonas Latz, Iason Papaioannou, and Elisabeth Ullmann. Multilevel sequential importance sampling for rare event estimation. SIAM Journal on Scientific Computing , 42(4):A2062–A2087, 2020

    work page 2020

  12. [12]

    Multilevel estimation of rare events

    Elisabeth Ullmann and Iason Papaioannou. Multilevel estimation of rare events. SIAM/ASA Journal on Uncertainty Quantification , 3(1):922–953, 2015

    work page 2015

  13. [13]

    Efficient rare event simulation for failure problems in random media

    Jingchen Liu, Jianfeng Lu, and Xiang Zhou. Efficient rare event simulation for failure problems in random media. SIAM Journal on Scientific Computing , 37(2):A609–A624, 2015

    work page 2015

  14. [14]

    Response-surface approach for reliability analysis

    Lucia Faravelli. Response-surface approach for reliability analysis. Journal of engineering mechanics, 115(12):2763–2781, 1989

    work page 1989

  15. [15]

    A new look at the response surface approach for reliability analysis

    Malur R Rajashekhar and Bruce R Ellingwood. A new look at the response surface approach for reliability analysis. Structural safety, 12(3):205–220, 1993

    work page 1993

  16. [16]

    An adaptive polynomial chaos expansion for high- dimensional reliability analysis

    Wanxin He, Yan Zeng, and Gang Li. An adaptive polynomial chaos expansion for high- dimensional reliability analysis. Structural and Multidisciplinary Optimization, 62(4):2051– 2067, 2020

    work page 2051

  17. [17]

    Sequential active learning of low-dimensional model representations for reliability analysis

    Max Ehre, Iason Papaioannou, Bruno Sudret, and Daniel Straub. Sequential active learning of low-dimensional model representations for reliability analysis. SIAM Journal on Scientific Computing, 44(3):B558–B584, 2022

    work page 2022

  18. [18]

    Kriging-based adaptive importance sampling algorithms for rare event estimation

    Mathieu Balesdent, Jerome Morio, and Julien Marzat. Kriging-based adaptive importance sampling algorithms for rare event estimation. Structural Safety, 44:1–10, 2013

    work page 2013

  19. [19]

    Ak-mcs: an active learning reliability method combining kriging and monte carlo simulation

    Benjamin Echard, Nicolas Gayton, and Maurice Lemaire. Ak-mcs: an active learning reliability method combining kriging and monte carlo simulation. Structural safety, 33(2):145–154, 2011

    work page 2011

  20. [20]

    Evaluation of failure probability via surrogate models

    Jing Li and Dongbin Xiu. Evaluation of failure probability via surrogate models. Journal of Computational Physics, 229(23):8966–8980, 2010

    work page 2010

  21. [21]

    An efficient surrogate-based method for computing rare failure probability

    Jing Li, Jinglai Li, and Dongbin Xiu. An efficient surrogate-based method for computing rare failure probability. Journal of Computational Physics , 230(24):8683–8697, 2011

    work page 2011

  22. [22]

    Accurate and efficient evaluation of failure probability for partial differential equations with random input data

    Peng Chen and Alfio Quarteroni. Accurate and efficient evaluation of failure probability for partial differential equations with random input data. Computer Methods in Applied Me- chanics and Engineering , 267:233–260, 2013

    work page 2013

  23. [23]

    Multifidelity importance sampling

    Benjamin Peherstorfer, Tiangang Cui, Youssef Marzouk, and Karen Willcox. Multifidelity importance sampling. Computer Methods in Applied Mechanics and Engineering, 300:490– 509, 2016

    work page 2016

  24. [24]

    Active learning relevant vector machine for reliability analysis

    TZ Li, Q Pan, and D Dias. Active learning relevant vector machine for reliability analysis. Applied Mathematical Modelling, 89:381–399, 2021

    work page 2021

  25. [25]

    An active learning method combining deep neural network and weighted sampling for structural reliability analysis

    Zhengliang Xiang, Jiahui Chen, Yuequan Bao, and Hui Li. An active learning method combining deep neural network and weighted sampling for structural reliability analysis. Mechanical Systems and Signal Processing , 140:106684, 2020

    work page 2020

  26. [26]

    A new active learning PGGR 25 method based on the learning function u of the ak-mcs reliability analysis method

    Zheng Peijuan, Wang Chien Ming, Zong Zhouhong, and Wang Liqi. A new active learning PGGR 25 method based on the learning function u of the ak-mcs reliability analysis method. Engi- neering structures, 148:185–194, 2017

    work page 2017

  27. [27]

    Deep uq: Learning deep neural network surrogate models for high dimensional uncertainty quantification.Journal of computational physics, 375:565– 588, 2018

    Rohit K Tripathy and Ilias Bilionis. Deep uq: Learning deep neural network surrogate models for high dimensional uncertainty quantification.Journal of computational physics, 375:565– 588, 2018

    work page 2018

  28. [28]

    David J. C. MacKay. A practical bayesian framework for backpropagation networks. Neural Computation, 4(3):448–472, 1992

    work page 1992

  29. [29]

    Dropout: A simple way to prevent neural networks from overfitting

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(56):1929–1958, 2014

    work page 1929

  30. [30]

    Adaptive operator learning for infinite- dimensional bayesian inverse problems

    Zhiwei Gao, Liang Yan, and Tao Zhou. Adaptive operator learning for infinite- dimensional bayesian inverse problems. SIAM/ASA Journal on Uncertainty Quantifi- cation, 12(4):1389–1423, 2024

    work page 2024

  31. [31]

    Stein variational gradient descent with local approximations

    Liang Yan and Tao Zhou. Stein variational gradient descent with local approximations. Com- puter Methods in Applied Mechanics and Engineering , 386:114087, 2021

    work page 2021

  32. [32]

    Relaxation-based importance sampling for structural reliability analysis

    Jianhua Xian and Ziqi Wang. Relaxation-based importance sampling for structural reliability analysis. Structural Safety, 106:102393, 2024

    work page 2024

  33. [33]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  34. [34]

    iCE-NGM: Improved cross-entropy importance sampling with non-parametric adaptive Gaussian mixtures and budget-informed stopping criterion

    Tianyu Zhang and Jize Zhang. iCE-NGM: Improved cross-entropy importance sampling with non-parametric adaptive Gaussian mixtures and budget-informed stopping criterion. Reli- ability Engineering & System Safety , 264, 2025. Article 111322

    work page 2025

  35. [35]

    The ensemble kalman filter for rare event estimation

    Fabian Wagner, Iason Papaioannou, and Elisabeth Ullmann. The ensemble kalman filter for rare event estimation. SIAM/ASA Journal on Uncertainty Quantification , 10(1):317–349, 2022

    work page 2022

  36. [36]

    Numerical methods for the discretiza- tion of random fields by means of the Karhunen–Lo` eve expansion

    Wolfgang Betz, Iason Papaioannou, and Daniel Straub. Numerical methods for the discretiza- tion of random fields by means of the Karhunen–Lo` eve expansion. Computer Methods in Applied Mechanics and Engineering , 271:109–129, 2014

    work page 2014