pith. sign in

arxiv: 2604.18776 · v1 · submitted 2026-04-20 · 💻 cs.CE

Multiscale Structural Reliability Analysis in high dimensions with Tensor Trains and Physics-Augmented Neural Networks

Aryan Tyagi , Alex de Beer , Tiangang Cui , Jan N. Fuhg This is my paper

Pith reviewed 2026-05-10 03:02 UTC · model grok-4.3

classification 💻 cs.CE
keywords structural reliabilitymultiscale analysistensor trainsphysics-augmented neural networksfailure probabilityhigh-dimensional uncertaintycompositesimportance sampling
0
0 comments X

The pith

Coupling a physics-augmented neural network with tensor-train importance sampling produces low-variance estimates of structural failure probability in dimensions up to 150.

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

The paper develops a computational framework for evaluating the probability that a composite structure fails when microscale material properties vary randomly and propagate through multiple length scales. It pairs a neural network that instantly returns a physically valid homogenized stiffness tensor with a sampling technique that represents the best importance distribution as a chain of tensor-train functions. This combination avoids the repeated solution of coupled finite-element problems across scales that renders standard methods unusable once the number of uncertain parameters grows large. A reader would care because many engineering materials are composites whose safety margins depend on exactly this kind of multiscale uncertainty, yet conventional risk calculations become impossible beyond modest dimensions.

Core claim

The authors show that the VRNN-DIRT framework supplies low-variance failure-probability estimates for a three-dimensional heterogeneous benchmark in which microscale stiffness is described by a Bayesian posterior inferred from limited strain data; the VRNN supplies the homogenized stiffness tensor in near-constant time while guaranteeing symmetry, positive-definiteness and strict placement inside the Voigt-Reuss bounds, and the DIRT construction assembles a sequence of functional tensor-train surrogates that approximate the optimal importance distribution arising from the Karhunen-Loève expansion of the random fields.

What carries the argument

The central mechanism is the pairing of the Voigt-Reuss Neural Network, which maps local strain and material parameters to a symmetric positive-definite stiffness tensor lying strictly inside the Voigt-Reuss bounds, with the Deep Inverse Rosenblatt Transport procedure that builds successive tensor-train approximations to the high-dimensional optimal importance distribution.

If this is right

  • The VRNN replaces every expensive FE² micro-macro solve with a single forward pass, making repeated sampling feasible.
  • The tensor-train representation of the importance distribution removes the exponential cost growth that normally accompanies Karhunen-Loève expansions beyond a few dozen modes.
  • The same posterior-to-failure pipeline works directly with material parameters that have been conditioned on sparse experimental strain measurements.
  • The method has already been exercised on a realistic three-dimensional heterogeneous specimen up to dimension 150.

Where Pith is reading between the lines

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

  • If the separation-of-scales premise remains valid for more complex geometries or loading conditions, the same neural-network surrogate could be reused across families of macroscopic designs without retraining.
  • Replacing the linear-elastic VRNN with a network that respects analogous thermodynamic bounds for finite-strain or viscoelastic response would extend the framework while preserving the physical guarantees.
  • The tensor-train construction of an importance distribution may transfer to other high-dimensional reliability tasks that currently rely on Markov-chain Monte Carlo.
  • Direct comparison of the framework's predictions against physical experiments on manufactured composite specimens would test whether the reported numerical accuracy survives model-form error.

Load-bearing premise

The entire construction rests on the material obeying separation of scales together with anisotropic linear elasticity.

What would settle it

A high-fidelity Monte Carlo reference computation performed on the same 150-dimensional benchmark problem that yields a failure-probability estimate lying outside the reported confidence interval of the VRNN-DIRT result would disprove the low-variance claim.

Figures

Figures reproduced from arXiv: 2604.18776 by Alex de Beer, Aryan Tyagi, Jan N. Fuhg, Tiangang Cui.

Figure 1
Figure 1. Figure 1: Workflow of the VRNN used for homogenization. The input RVE provides microscale descriptors [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Multiscale reliability analysis framework with DIRT and VRNN. [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Microscale and macroscale geometries used in the numerical experiment. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of homogenized stiffness tensor components predicted by VRNN with reference finite element [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Training and validation loss evolution for VRNN. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Realizations of the random fields 0.00 0.05 0.10 0.15 0.20 x1 [m] 0.00 0.05 0.10 0.15 0.20 x2 [m] 1 2 3 4 5 6 7 8 9 10 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Finite element mesh of the plate with hole showing the ten locations (red markers) used for strain measurements [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of the true, reconstructed, and relative error fields for [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Running estimate of Pˆ(F | y)MC with increasing Nacc. For visualization, values of Pˆ(F | y)MC = 0 are displayed as 10−15 in the plot. 12 Dimensional Monte Carlo reference Before scaling the DIRT algorithm to higher dimensions, we computed a reference posterior failure probability Pˆ(F | y)MC for d = 12. For obtaining samples from the posterior, we generated Nprop = 9.6 × 108 samples from the prior with 20… view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of the macroscale fields predicted using classical FE [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
read the original abstract

Structural reliability evaluation for composites constitutes a fundamentally high-dimensional multiscale problem, as microscale material uncertainties must propagate to the macroscale and can be quantified as high-dimensional random fields. Conventional approaches are computationally intractable, as they rely on repeatedly solving coupled partial differential equation systems across scales while contending with the exponential complexity inherent in high-dimensional uncertainty quantification. This work introduces a scalable and physically consistent framework that addresses both bottlenecks simultaneously in the case of separation of scales and (anisotropic) linear elasticity. In particular, we couple a physics-augmented Voigt--Reuss Neural Network (VRNN) with the Deep Inverse Rosenblatt Transport (DIRT) method to estimate the posterior probability of structural failure. The VRNN is used to resolve the computationally expensive FE$^2$ scheme by providing a near-instantaneous evaluation of the homogenized stiffness tensor that is guaranteed to be symmetric, positive-definite, and strictly bounded within the Voigt--Reuss limits, enabling fast evaluation of the homogenized responses. The DIRT method constructs a sequence of functional tensor train approximations to efficiently store an approximation of the high-dimensional optimal importance sampling distribution for estimating the probability of failure. This mitigates the curse of dimensionality arising from the Karhunen--Lo\`eve expansion of the random fields. The framework is demonstrated on a three-dimensional heterogeneous benchmark problem, where the uncertainty in the microscale material properties is characterized by a Bayesian posterior distribution obtained from limited strain observations. Our results show that the proposed framework can provide low-variance estimates of failure probabilities in dimensions up to 150.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 1 minor

Summary. The manuscript introduces a framework for high-dimensional multiscale structural reliability analysis of composites by coupling a physics-augmented Voigt-Reuss Neural Network (VRNN) for fast, constraint-satisfying evaluation of homogenized stiffness tensors with the Deep Inverse Rosenblatt Transport (DIRT) method, which builds a sequence of functional tensor-train approximations to the optimal importance sampling distribution. Under assumptions of scale separation and anisotropic linear elasticity, the approach is applied to a three-dimensional heterogeneous benchmark where microscale uncertainties are represented via a Bayesian posterior from limited strain observations, with the central claim being that low-variance estimates of failure probabilities can be obtained in dimensions up to 150.

Significance. If the numerical performance claims are substantiated with appropriate accuracy metrics, the work would offer a meaningful advance in computational mechanics by simultaneously tackling the expense of multiscale FE² evaluations and the curse of dimensionality in uncertainty propagation for reliability analysis. The physics-augmented neural network and tensor-train importance sampling combination could enable practical probabilistic assessments in engineering settings where conventional Monte Carlo sampling is prohibitive.

major comments (1)
  1. [Numerical results / abstract] The central claim of low-variance failure-probability estimates up to dimension 150 rests on DIRT constructing accurate functional tensor-train approximations to the optimal importance sampling density derived from the posterior over KL coefficients. The manuscript reports no tensor-train ranks, cross-validation residuals, effective sample sizes, or direct variance comparisons against crude Monte Carlo for the highest-dimensional case (see abstract and numerical results). If the approximation error is non-negligible, the importance weights become biased and the claimed variance reduction does not hold, even with fast VRNN evaluations.
minor comments (1)
  1. [Abstract] The abstract and introduction would benefit from explicitly stating the form of the limit-state function and the specific dimensions of the benchmark problem to allow readers to assess the scope of the 150-dimensional demonstration.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thorough review and valuable feedback on our manuscript. The concern regarding the substantiation of the DIRT approximations and variance reduction claims is well-taken, and we address it directly below. We will incorporate the requested details to strengthen the numerical evidence.

read point-by-point responses

  1. Referee: The central claim of low-variance failure-probability estimates up to dimension 150 rests on DIRT constructing accurate functional tensor-train approximations to the optimal importance sampling density derived from the posterior over KL coefficients. The manuscript reports no tensor-train ranks, cross-validation residuals, effective sample sizes, or direct variance comparisons against crude Monte Carlo for the highest-dimensional case (see abstract and numerical results). If the approximation error is non-negligible, the importance weights become biased and the claimed variance reduction does not hold, even with fast VRNN evaluations.

    Authors: We agree that explicit reporting of these diagnostics is necessary to fully substantiate the central claim. In the revised manuscript we will add a new subsection (or expanded table) in the numerical results section that documents: (i) the tensor-train ranks selected for each DIRT layer at dimension 150, (ii) the cross-validation residuals obtained during construction of the functional approximations, (iii) the effective sample sizes realized by the resulting importance sampler, and (iv) variance comparisons against crude Monte Carlo for all dimensions where the latter remains computationally tractable (e.g., up to d=50). For d=150 we will additionally supply a posteriori error bounds on the tensor-train approximation together with a sensitivity study showing that the observed variance reduction is robust to modest rank truncation. These additions will confirm that the approximation error remains sufficiently small for the importance weights to stay unbiased and for the reported low-variance estimates to hold. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The framework couples a physics-augmented VRNN (enforcing symmetry, positive-definiteness and Voigt-Reuss bounds by architectural constraints) with DIRT tensor-train approximations to the optimal importance density. The VRNN is trained on microscale homogenization data independent of the macroscale failure indicator; DIRT builds a sequence of functional approximations from posterior samples and the limit-state function without requiring the failure probability as an input. The reported low-variance estimates in dimensions up to 150 are numerical outcomes of the resulting importance sampler, not a definitional reduction or self-fit. No load-bearing step collapses to a prior self-citation, ansatz smuggling, or renaming of the target quantity. The central claim therefore rests on independent physical constraints and sampling theory rather than circular construction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The framework rests on the separation-of-scales assumption and linear elasticity; the VRNN introduces learned weights that are constrained but still fitted; DIRT introduces tensor-train ranks and a sequence of transport maps that are constructed numerically.

free parameters (2)
  • VRNN network weights
    Trained parameters of the physics-augmented neural network that map microscale fields to homogenized stiffness; fitted to satisfy Voigt-Reuss bounds.
  • DIRT tensor-train ranks and transport maps
    Numerical ranks and coefficients of the sequence of functional tensor trains approximating the optimal importance distribution.
axioms (2)
  • domain assumption Separation of scales holds and the problem remains in the linear-elastic regime.
    Stated explicitly in the abstract as the setting in which the VRNN and DIRT coupling is valid.
  • domain assumption The Karhunen-Loève expansion of the random fields adequately represents the microscale uncertainty.
    Implicit in the claim that DIRT mitigates the curse of dimensionality arising from the KL expansion.

pith-pipeline@v0.9.0 · 5594 in / 1480 out tokens · 33432 ms · 2026-05-10T03:02:53.219142+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

70 extracted references · 70 canonical work pages

  1. [1]

    A review and assessment of importance sampling methods for reliability analysis.Structural Safety, 97:102216, 2022

    Armin Tabandeh, Gaofeng Jia, and Paolo Gardoni. A review and assessment of importance sampling methods for reliability analysis.Structural Safety, 97:102216, 2022

    work page 2022

  2. [2]

    Limit state function identification using support vector machines for discontinuous responses and disjoint failure domains.Probabilistic Engineering Mechanics, 23(1):1–11, 2008

    Anirban Basudhar, Samy Missoum, and Antonio Harrison Sanchez. Limit state function identification using support vector machines for discontinuous responses and disjoint failure domains.Probabilistic Engineering Mechanics, 23(1):1–11, 2008

    work page 2008

  3. [3]

    Gavin and Siu Chung Yau

    Henri P. Gavin and Siu Chung Yau. High-order limit state functions in the response surface method for structural reliability analysis.Structural Safety, 30(2):162–179, 2008

    work page 2008

  4. [4]

    Estimation of small failure probabilities in high dimensions by subset simulation

    Siu-Kui Au and James Beck. Estimation of small failure probabilities in high dimensions by subset simulation. Probabilistic Engineering Mechanics, 16:263–277, 10 2001

    work page 2001

  5. [5]

    Failure probability estimation through high-dimensional elliptical distribution modeling with multiple importance sampling.Reliability Engineering & System Safety, 235:109238, 2023

    Marie Chiron, Christian Genest, Jérôme Morio, and Sylvain Dubreuil. Failure probability estimation through high-dimensional elliptical distribution modeling with multiple importance sampling.Reliability Engineering & System Safety, 235:109238, 2023

    work page 2023

  6. [6]

    Echard, N

    B. Echard, N. Gayton, M. Lemaire, and N. Relun. A combined importance sampling and kriging reliability method for small failure probabilities with time-demanding numerical models.Reliability Engineering & System Safety, 111:232–240, 2013

    work page 2013

  7. [7]

    An exact and invariant first order reliability format.Journal of Engineering Mechanics, 100, 01 1974

    Hasofer AM and Niels Lind. An exact and invariant first order reliability format.Journal of Engineering Mechanics, 100, 01 1974

    work page 1974

  8. [8]

    Asymptotic approximations for multinormal integrals.Journal of Engineering Mechanics, 110(3):357–366, 1984

    Karl Breitung. Asymptotic approximations for multinormal integrals.Journal of Engineering Mechanics, 110(3):357–366, 1984

    work page 1984

  9. [9]

    Global reliability sensitivity estimation based on failure samples

    Luyi Li, Iason Papaioannou, and Daniel Straub. Global reliability sensitivity estimation based on failure samples. Structural Safety, 81:101871, 11 2019

    work page 2019

  10. [10]

    Sequential importance sampling for structural reliability analysis.Structural Safety, 62:66–75, 09 2016

    Iason Papaioannou, Costas Papadimitriou, and Daniel Straub. Sequential importance sampling for structural reliability analysis.Structural Safety, 62:66–75, 09 2016

    work page 2016

  11. [11]

    Improved cross entropy-based importance sampling with a flexible mixture model.Reliability Engineering & System Safety, 191:106564, 07 2019

    Iason Papaioannou, Sebastian Geyer, and Daniel Straub. Improved cross entropy-based importance sampling with a flexible mixture model.Reliability Engineering & System Safety, 191:106564, 07 2019

    work page 2019

  12. [12]

    A method of combined metamodel and subset simulation for reliability analysis of rare events.Advances in Engineering Software, 195:103693, 2024

    Yuming Zhang and Juan Ma. A method of combined metamodel and subset simulation for reliability analysis of rare events.Advances in Engineering Software, 195:103693, 2024

    work page 2024

  13. [13]

    Error analysis for probabilities of rare events with approximate models.SIAM Journal on Numerical Analysis, 59(4):1948–1975, 2021

    Fabian Wagner, Jonas Latz, Iason Papaioannou, and Elisabeth Ullmann. Error analysis for probabilities of rare events with approximate models.SIAM Journal on Numerical Analysis, 59(4):1948–1975, 2021

    work page 1948

  14. [14]

    Deep inverse Rosenblatt transport for structural reliability analysis

    Aryan Tyagi and Jan Fuhg. Deep inverse Rosenblatt transport for structural reliability analysis. 09 2025

    work page 2025

  15. [15]

    Bayesian inference of random fields represented with the karhunen–loève expansion.Computer Methods in Applied Mechanics and Engineering, 358:112632, 2020

    Felipe Uribe, Iason Papaioannou, Wolfgang Betz, and Daniel Straub. Bayesian inference of random fields represented with the karhunen–loève expansion.Computer Methods in Applied Mechanics and Engineering, 358:112632, 2020

    work page 2020

  16. [16]

    Felipe Uribe, Iason Papaioannou, Jonas Latz, Wolfgang Betz, Elisabeth Ullmann, and Daniel Straub. Bayesian inference with subset simulation in varying dimensions applied to the karhunen–loève expansion.International Journal for Numerical Methods in Engineering, 122:5100–5127, 06 2021

    work page 2021

  17. [17]

    Valdebenito, and Matthias G.R

    Pei-Pei Li, Yan-Gang Zhao, Chao Dang, Matteo Broggi, Marcos A. Valdebenito, and Matthias G.R. Faes. An efficient bayesian updating framework for characterizing the posterior failure probability.Mechanical Systems and Signal Processing, 222:111768, 2025

    work page 2025

  18. [18]

    Cross-entropy-based importance sampling with failure-informed dimension reduction for rare event simulation.SIAM/ASA Journal on Uncertainty Quantification, 9(2):818–847, 2021

    Felipe Uribe, Iason Papaioannou, Youssef M Marzouk, and Daniel Straub. Cross-entropy-based importance sampling with failure-informed dimension reduction for rare event simulation.SIAM/ASA Journal on Uncertainty Quantification, 9(2):818–847, 2021

    work page 2021

  19. [19]

    Material spatial randomness: From statistical to representative volume element

    Martin Ostoja-Starzewski. Material spatial randomness: From statistical to representative volume element. Probabilistic engineering mechanics, 21(2):112–132, 2006

    work page 2006

  20. [20]

    Concurrent multiscale simulations of nonlinear random materials using probabilistic learning.Computer Methods in Applied Mechanics and Engineering, 422:116837, 2024

    Peiyi Chen, Johann Guilleminot, and Christian Soize. Concurrent multiscale simulations of nonlinear random materials using probabilistic learning.Computer Methods in Applied Mechanics and Engineering, 422:116837, 2024

    work page 2024

  21. [21]

    Marc GD Geers, Varvara G Kouznetsova, and W AM Brekelmans. Multiscale first-order and second-order compu- tational homogenization of microstructures towards continua.International Journal for Multiscale Computational Engineering, 1(4), 2003. 19 APREPRINT- APRIL22, 2026

    work page 2003

  22. [22]

    Frédéric Feyel and Jean-Louis Chaboche. Fe2 multiscale approach for modelling the elastoviscoplastic behaviour of long fibre sic/ti composite materials.Computer Methods in Applied Mechanics and Engineering, 183(3):309– 330, 2000

    work page 2000

  23. [23]

    Smit, W.A.M

    R.J.M. Smit, W.A.M. Brekelmans, and H.E.H. Meijer. Prediction of the mechanical behavior of nonlinear heterogeneous systems by multi-level finite element modeling.Computer Methods in Applied Mechanics and Engineering, 155(1):181–192, 1998

    work page 1998

  24. [24]

    Kouznetsova, W.A.M

    V . Kouznetsova, W.A.M. Brekelmans, and F.P.T. Baaijens. Approach to micro-macro modeling of heterogeneous materials.Computational Mechanics, 27(1):37 – 48, 2001

    work page 2001

  25. [25]

    Miehe and A

    C. Miehe and A. Koch. Computational micro-to-macro transitions of discretized microstructures undergoing small strains.Archive of Applied Mechanics (Ingenieur Archiv), 72:300–317, 07 2002

    work page 2002

  26. [26]

    Fast numerical method for computing the linear and nonlinear properties of composites.Comptes Rendus de l’Académie des Sciences

    Hervé Moulinec and Pierre Suquet. Fast numerical method for computing the linear and nonlinear properties of composites.Comptes Rendus de l’Académie des Sciences. Série II, 318, 04 1994

    work page 1994

  27. [27]

    Moulinec and P

    H. Moulinec and P. Suquet. A numerical method for computing the overall response of nonlinear composites with complex microstructure.Computer Methods in Applied Mechanics and Engineering, 157(1):69–94, 1998

    work page 1998

  28. [28]

    An imprecise multiscale uncertainty quantification framework for fiber reinforced composites.Probabilistic Engineering Mechanics, 78:103686, 2024

    Haodong Zhao and Changcong Zhou. An imprecise multiscale uncertainty quantification framework for fiber reinforced composites.Probabilistic Engineering Mechanics, 78:103686, 2024

    work page 2024

  29. [29]

    Balasubramani, B

    N.K. Balasubramani, B. Zhang, N.T. Chowdhury, A. Mukkavilli, M. Suter, and G.M. Pearce. Micro-mechanical analysis on random rve size and shape in multiscale finite element modelling of unidirectional frp composites. Composite Structures, 282:115081, 2022

    work page 2022

  30. [30]

    Proper orthogonal decomposition for model updating of non-linear mechanical systems.Mechanical Systems and Signal Processing, 15(1):31–43, 2001

    V Lenaerts, Gaëtan Kerschen, and Jean-Claude Golinval. Proper orthogonal decomposition for model updating of non-linear mechanical systems.Mechanical Systems and Signal Processing, 15(1):31–43, 2001

    work page 2001

  31. [31]

    A comparison of projection-based model reduction concepts in the context of nonlinear biomechanics.Archive of Applied Mechanics, 83(8):1193–1213, 2013

    Annika Radermacher and Stefanie Reese. A comparison of projection-based model reduction concepts in the context of nonlinear biomechanics.Archive of Applied Mechanics, 83(8):1193–1213, 2013

    work page 2013

  32. [32]

    Pod-based model reduction with empirical interpolation applied to nonlinear elasticity.International Journal for Numerical Methods in Engineering, 107(6):477–495, 2016

    Annika Radermacher and Stefanie Reese. Pod-based model reduction with empirical interpolation applied to nonlinear elasticity.International Journal for Numerical Methods in Engineering, 107(6):477–495, 2016

    work page 2016

  33. [33]

    Multiscale computation on feedforward neural network and recurrent neural network

    Bin Li and Xiaoying Zhuang. Multiscale computation on feedforward neural network and recurrent neural network. Frontiers of Structural and Civil Engineering, 14(6):1285–1298, 2020

    work page 2020

  34. [34]

    Deep neural networks for parameterized homogenization in concurrent multiscale structural optimization.Structural and Multidisciplinary Optimization, 66(1):20, 2023

    Nolan Black and Ahmad R Najafi. Deep neural networks for parameterized homogenization in concurrent multiscale structural optimization.Structural and Multidisciplinary Optimization, 66(1):20, 2023

    work page 2023

  35. [35]

    Geometric deep learning for computational mechanics part i: Anisotropic hyperelasticity.Computer Methods in Applied Mechanics and Engineering, 371:113299, 2020

    Nikolaos N Vlassis, Ran Ma, and WaiChing Sun. Geometric deep learning for computational mechanics part i: Anisotropic hyperelasticity.Computer Methods in Applied Mechanics and Engineering, 371:113299, 2020

    work page 2020

  36. [36]

    Jan N Fuhg, Michele Marino, and Nikolaos Bouklas. Local approximate gaussian process regression for data- driven constitutive models: development and comparison with neural networks.Computer Methods in Applied Mechanics and Engineering, 388:114217, 2022

    work page 2022

  37. [37]

    Jadoon, Karl A

    Asghar A. Jadoon, Karl A. Kalina, Manuel K. Rausch, Reese Jones, and Jan Niklas Fuhg. Inverse design of anisotropic microstructures using physics-augmented neural networks.Journal of the Mechanics and Physics of Solids, 203:106161, 2025

    work page 2025

  38. [38]

    Jadoon, D

    Asghar A. Jadoon, D. Thomas Seidl, Reese E. Jones, and Jan N. Fuhg. Input specific neural networks, 2025

    work page 2025

  39. [39]

    Kalina, Jörg Brummund, WaiChing Sun, and Markus Kästner

    Karl A. Kalina, Jörg Brummund, WaiChing Sun, and Markus Kästner. Neural networks meet anisotropic hyperelasticity: A framework based on generalized structure tensors and isotropic tensor functions.Computer Methods in Applied Mechanics and Engineering, 437:117725, 2025

    work page 2025

  40. [40]

    Wu, and M

    Zeliang Liu, C.T. Wu, and M. Koishi. A deep material network for multiscale topology learning and accelerated nonlinear modeling of heterogeneous materials.Computer Methods in Applied Mechanics and Engineering, 345:1138–1168, 2019

    work page 2019

  41. [41]

    Mauricio Fernández, Mostafa Jamshidian, Thomas Böhlke, Kristian Kersting, and Oliver Weeger. Anisotropic hyperelastic constitutive models for finite deformations combining material theory and data-driven approaches with application to cubic lattice metamaterials.Computational Mechanics, 67:1–25, 02 2021

    work page 2021

  42. [42]

    A review on data-driven constitutive laws for solids: Jn fuhg et al.Archives of Computational Methods in Engineering, 32(3):1841–1883, 2025

    Jan N Fuhg, Govinda Anantha Padmanabha, Nikolaos Bouklas, Bahador Bahmani, WaiChing Sun, Nikolaos N Vlassis, Moritz Flaschel, Pietro Carrara, and Laura De Lorenzis. A review on data-driven constitutive laws for solids: Jn fuhg et al.Archives of Computational Methods in Engineering, 32(3):1841–1883, 2025

    work page 2025

  43. [43]

    Spectral normalization and voigt-reuss net: A universal approach to microstructure-property forecasting with physical guarantees.GAMM-Mitteilungen, 48, 06 2025

    Sanath Keshav, Julius Herb, and Felix Fritzen. Spectral normalization and voigt-reuss net: A universal approach to microstructure-property forecasting with physical guarantees.GAMM-Mitteilungen, 48, 06 2025. 20 APREPRINT- APRIL22, 2026

    work page 2025

  44. [44]

    An adaptive scheme for random field discretization using kl expansion.Engineering with Computers, 38:2937 – 2954, 2021

    Kamaljyoti Nath, Anjan Dutta, and Budhaditya Hazra. An adaptive scheme for random field discretization using kl expansion.Engineering with Computers, 38:2937 – 2954, 2021

    work page 2021

  45. [45]

    Dannert, Ammar A

    Esther V oelsen, Mona M. Dannert, Ammar A. Basmaji, Fynn Bensel, and Udo Nackenhorst. Sparse poly- nomial chaos expansion for nonlinear finite element simulations with random material properties.PAMM, 23(1):e202200131, 2023

    work page 2023

  46. [46]

    Probabilistic structural analysis of a composite beam considering random field of interface.Structures, 57:105079, 2023

    Yan Yang, Xiang Liu, and Xiao-Yi Zhang. Probabilistic structural analysis of a composite beam considering random field of interface.Structures, 57:105079, 2023

    work page 2023

  47. [47]

    Prior information, sampling distributions, and the curse of dimensionality

    Andrew Curtis and Anthony Lomax. Prior information, sampling distributions, and the curse of dimensionality. Geophysics, 66(2):372–378, 2001

    work page 2001

  48. [48]

    Springer, 2004

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

    work page 2004

  49. [49]

    Approximation and sampling of multivariate probability distributions in the tensor train decomposition.Statistics and Computing, 30, 05 2020

    Sergey Dolgov, Karim Anaya-Izquierdo, Colin Fox, and Robert Scheichl. Approximation and sampling of multivariate probability distributions in the tensor train decomposition.Statistics and Computing, 30, 05 2020

    work page 2020

  50. [50]

    Deep importance sampling using tensor trains with application to a priori and a posteriori rare events.SIAM Journal on Scientific Computing, 46(1):C1–C29, 2024

    Tiangang Cui, Sergey Dolgov, and Robert Scheichl. Deep importance sampling using tensor trains with application to a priori and a posteriori rare events.SIAM Journal on Scientific Computing, 46(1):C1–C29, 2024

    work page 2024

  51. [51]

    Deep composition of tensor-trains using squared inverse Rosenblatt transports

    Tiangang Cui and Sergey Dolgov. Deep composition of tensor-trains using squared inverse Rosenblatt transports. Foundations of Computational Mathematics, 22(6):1863–1922, 2022

    work page 1922

  52. [52]

    Homogenization and two-scale convergence.SIAM Journal on Mathematical Analysis, 23(6):1482–1518, 1992

    Grégoire Allaire. Homogenization and two-scale convergence.SIAM Journal on Mathematical Analysis, 23(6):1482–1518, 1992

    work page 1992

  53. [53]

    Beck and Konstantin M

    James L. Beck and Konstantin M. Zuev. Rare event simulation, 2015

    work page 2015

  54. [54]

    Springer Publishing Company, Incorporated, 1st edition, 2010

    James Bucklew.Introduction to Rare Event Simulation. Springer Publishing Company, Incorporated, 1st edition, 2010

    work page 2010

  55. [55]

    Owen.Monte Carlo theory, methods and examples.https://artowen.su.domains/mc/, 2013

    Art B. Owen.Monte Carlo theory, methods and examples.https://artowen.su.domains/mc/, 2013

    work page 2013

  56. [56]

    A friendly introduction to triangular transport, 2025

    Maximilian Ramgraber, Daniel Sharp, Mathieu Le Provost, and Youssef Marzouk. A friendly introduction to triangular transport, 2025

    work page 2025

  57. [57]

    Spectral tensor-train decomposition.SIAM Journal on Scientific Computing, 38(4):A2405–A2439, 2016

    Daniele Bigoni, Allan P Engsig-Karup, and Youssef M Marzouk. Spectral tensor-train decomposition.SIAM Journal on Scientific Computing, 38(4):A2405–A2439, 2016

    work page 2016

  58. [58]

    A continuous analogue of the tensor-train decomposition

    Alex Gorodetsky, Sertac Karaman, and Youssef Marzouk. A continuous analogue of the tensor-train decomposition. Computer Methods in Applied Mechanics and Engineering, 347:59–84, 2019

    work page 2019

  59. [59]

    TT-cross approximation for multidimensional arrays.Linear Algebra and its Applications, 432(1):70–88, 2010

    Ivan Oseledets and Eugene Tyrtyshnikov. TT-cross approximation for multidimensional arrays.Linear Algebra and its Applications, 432(1):70–88, 2010

    work page 2010

  60. [60]

    Self-reinforced polynomial approximation methods for concentrated probability densities, 2023

    Tiangang Cui, Sergey Dolgov, and Olivier Zahm. Self-reinforced polynomial approximation methods for concentrated probability densities, 2023

    work page 2023

  61. [61]

    Contributions to the theory of convex bodies.Michigan Mathematical Journal, 4(1):39–52, 1957

    Herbert Knothe. Contributions to the theory of convex bodies.Michigan Mathematical Journal, 4(1):39–52, 1957

    work page 1957

  62. [62]

    Remarks on a multivariate transformation.The Annals of Mathematical Statistics, 23(3):470– 472, 1952

    Murray Rosenblatt. Remarks on a multivariate transformation.The Annals of Mathematical Statistics, 23(3):470– 472, 1952

    work page 1952

  63. [63]

    Sequential Monte Carlo samplers.Journal of the Royal Statistical Society Series B: Statistical Methodology, 68(3):411–436, 2006

    Pierre Del Moral, Arnaud Doucet, and Ajay Jasra. Sequential Monte Carlo samplers.Journal of the Royal Statistical Society Series B: Statistical Methodology, 68(3):411–436, 2006

    work page 2006

  64. [64]

    Multiscale finite element calculations in python using sfepy.Advances in Computational Mathematics, 2019

    Robert Cimrman, Vladimír Lukeš, and Eduard Rohan. Multiscale finite element calculations in python using sfepy.Advances in Computational Mathematics, 2019

    work page 2019

  65. [65]

    Scalable conditional deep inverse rosenblatt transports using tensor trains and gradient-based dimension reduction.Journal of Computational Physics, 485:112103, 2023

    Tiangang Cui, Sergey Dolgov, and Olivier Zahm. Scalable conditional deep inverse rosenblatt transports using tensor trains and gradient-based dimension reduction.Journal of Computational Physics, 485:112103, 2023

    work page 2023

  66. [66]

    Predicting stress, strain and deformation fields in materials and structures with graph neural networks.Scientific Reports, 12(1):21834, Dec 2022

    Marco Maurizi, Chao Gao, and Filippo Berto. Predicting stress, strain and deformation fields in materials and structures with graph neural networks.Scientific Reports, 12(1):21834, Dec 2022

    work page 2022

  67. [67]

    Certified dimension reduction in nonlinear Bayesian inverse problems.Mathematics of Computation, 91(336):1789–1835, 2022

    Olivier Zahm, Tiangang Cui, Kody Law, Alessio Spantini, and Youssef Marzouk. Certified dimension reduction in nonlinear Bayesian inverse problems.Mathematics of Computation, 91(336):1789–1835, 2022

    work page 2022

  68. [68]

    Shanyin Tong, Eric Vanden-Eijnden, and Georg Stadler. Extreme event probability estimation using PDE- constrained optimization and large deviation theory, with application to tsunamis.Communications in Applied Mathematics and Computational Science, 16(2):181–225, 2021

    work page 2021

  69. [69]

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

    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. 21 APREPRINT- APRIL22, 2026

    work page 2023

  70. [70]

    Survey of multifidelity methods in uncertainty propagation, inference, and optimization.SIAM Review, 60(3):550–591, 2018

    Benjamin Peherstorfer, Karen Willcox, and Max Gunzburger. Survey of multifidelity methods in uncertainty propagation, inference, and optimization.SIAM Review, 60(3):550–591, 2018. 22

    work page 2018