Pith. sign in

REVIEW 2 major objections 6 minor 68 references

A GPU-differentiable finite-element code recovers anisotropic metal plasticity parameters from one full-field displacement test, with multi-fold forward speedups.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 17:47 UTC pith:7NJ7BEAY

load-bearing objection Solid systems paper: first end-to-end JAX GPU+AD finite-strain anisotropic plasticity (Hill + Yld2004-18p) with real Abaqus speedups and high-dimensional synthetic FEMU that works. the 2 major comments →

arxiv 2606.17390 v2 pith:7NJ7BEAY submitted 2026-06-16 cs.CE

JAX-FEM-ANISO: Differentiable GPU-Accelerated Finite Element Framework for Inverse Identification of Finite-Strain Anisotropic Plasticity

Deepak Sharma , Itzel Salgado , Lu Huang , Hui-Ping Wang , Jian Cao This is my paper
classification cs.CE
keywords inverse material characterizationfinite-strain anisotropic plasticitydifferentiable simulationsJAX-FEMPDE-constrained optimizationGPU accelerationautomatic differentiationfull-field identification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper claims that finite-strain anisotropic plasticity can be simulated and calibrated end-to-end on GPUs by making the entire finite-element workflow differentiable. Automatic differentiation supplies consistent material tangents and adjoint gradients of full-field displacement mismatch, so high-dimensional yield and hardening parameters—including spatially varying fields and Barlat-type coefficients—can be recovered by gradient optimization from a single information-rich specimen rather than many conventional tests. On a three-million-degree-of-freedom problem the same framework runs up to roughly nine times faster than a 24-core commercial baseline. A sympathetic reader cares because metal forming, welding, and additive manufacturing need both fast large-deformation solvers and reliable calibration of anisotropic models; the work argues that differentiability plus GPU throughput removes the two main bottlenecks that have kept one-shot, high-fidelity characterization impractical.

Core claim

A fully differentiable, GPU-accelerated finite-element implementation of finite-strain anisotropic elasto-plasticity (Hill–48 and Barlat Yld2004-18p) recovers anisotropic yield and hardening parameters—including spatially varying fields—from single information-rich full-field displacement tests via adjoint automatic-differentiation gradients, while delivering up to about 9.4× forward speed-up versus a 24-core commercial CPU baseline at roughly three million degrees of freedom.

What carries the argument

End-to-end automatic differentiation through the constitutive return map, global Newton solver, and discrete adjoint of the time-stepping residual: custom Jacobian-vector products give consistent local tangents without manual derivation, while reverse-mode adjoints supply objective gradients whose cost is nearly independent of parameter dimension.

Load-bearing premise

Every inverse-identification result uses synthetic displacement data generated by the same constitutive family and finite-element model (with optional added noise), not real full-field measurements with imperfect boundaries and model error.

What would settle it

Apply the same adjoint pipeline to real digital-image-correlation displacements from a physical information-rich specimen and check whether recovered anisotropic parameters predict independent validation tests within engineering tolerance; failure of that transfer would overturn the single-test claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • High-dimensional anisotropic yield surfaces become calibratable from one carefully designed full-field test instead of multi-test campaigns.
  • Spatially heterogeneous plasticity parameters induced by manufacturing can be identified by the same adjoint workflow.
  • Forward large-deformation anisotropic simulations at multi-million DOF become practical on a single high-end GPU.
  • Finite-difference sensitivity analysis is no longer required for gradient-based FEMU of complex plasticity models.

Where Pith is reading between the lines

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

  • If synthetic-to-real transfer holds, the same pipeline could co-optimize specimen topology and loading paths for maximum parameter identifiability.
  • Adjoint gradients of full-field mismatch open a path to joint geometry-and-material design under manufacturing-induced heterogeneity.
  • The framework’s differentiability would also support uncertainty quantification over plasticity parameters once experimental noise models are included.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The manuscript presents JAX-FEM-ANISO, a fully differentiable GPU-accelerated finite-element framework built on JAX-FEM for forward simulation and inverse identification of finite-strain anisotropic plasticity (Hill–48 and Barlat Yld2004-18p). It implements Miehe’s modular logarithmic-strain formulation with F-bar kinematics, AD-derived consistent tangents and custom JVPs for local return mapping and spectral maps, GPU-resident residual/stiffness evaluation, sparse assembly, and linear solves (AMGX/CuDSS), plus discrete adjoint gradients for FEMU. Forward results are verified against Abaqus (flange drawing; single-element Barlat path). On ~3M DOF, a single H100 yields up to 9.4× wall-clock speedup versus a 24-core Abaqus CPU baseline. Adjoint gradients match central finite differences to <0.2% at h=10⁻⁶. Inverse studies on synthetic full-field displacements recover anisotropic yield/hardening parameters for information-rich tensile and cruciform specimens (with noise), two-region spatial heterogeneity, and a high-dimensional Barlat parameter set from a topology-optimized specimen.

Significance. If the reported accuracy and performance hold, this is a substantial systems and methods contribution for computational solid mechanics and Material Testing 2.0: it removes manual consistent-tangent derivation for advanced anisotropic models, makes high-dimensional FEMU gradients practical via reverse-mode AD, and demonstrates GPU scaling on million-DOF nonlinear problems. Strengths include careful Abaqus verification, explicit FD–AD gradient checks with the expected step-size U-curve, transparent reporting of Voce (Q,b) non-uniqueness while matching hardening curves, and progressive inverse demos up to Barlat Yld2004-18p and multi-material fields. These establish differentiable GPU FEM as a credible engine for simulation and synthetic-data inverse design; real DIC transfer remains future work as stated in §4.

major comments (2)
  1. Abstract and §4 claim a “practical high-throughput engine for … characterization” and that information-rich single tests can replace many conventional experiments. All inverse results (§3.3–3.6) use same-model synthetic displacements (optional i.i.d. Gaussian noise), not real DIC with imperfect BCs, incomplete observations, or model discrepancy—explicitly deferred to future work in §4. The algorithmic claims (AD gradients, GPU speedups, synthetic recovery) are supported; the characterization-engine phrasing overreaches the evidence. Please qualify abstract/conclusions/keywords so “single-test characterization” is framed as demonstrated on synthetic full-field data and as a pathway toward DIC-based MT2.0, not as an established experimental replacement.
  2. §3.5 titles and abstract language refer to “spatially varying material properties,” but the demonstration is two discrete material regions with piecewise-constant parameters (12 scalars), not a continuous field (e.g., nodal or element-wise parameter map). That is still a useful multi-material test and shows AD scaling with parameter count, but it does not yet establish continuous spatial-field identification. Clarify the claim (two-region heterogeneity vs. field reconstruction) and, if continuous fields are intended as a contribution, either add a small field example or soften the wording.
minor comments (6)
  1. Typos/grammar: “largly unexplored” (Introduction); “memeory demands,” “stiffness materix,” “uisng JAX” (§2.1); “V oce” spacing throughout (should be Voce); “non-finite solutions” is fine but check consistency of “finite-difference” hyphenation.
  2. Table 4: Abaqus 24-core reference is on Windows 11 while other Abaqus runs are on RHEL; note this more prominently when interpreting the 9.4× figure so readers do not over-interpret cross-platform wall-clock ratios.
  3. §3.2 gradient check: direction vector Di=0.1 is somewhat arbitrary; a brief note that relative component-wise errors (Table 6) are the primary verification is enough.
  4. §3.6 Barlat inverse: recovered coefficients differ from ground truth while yield surfaces match—good—but state explicitly in the table caption that coefficient non-uniqueness is expected and that surface/hardening agreement is the success metric.
  5. Figures 5, 14, 18, 26: ensure colorbars and deformed-mesh scales are readable in print; some multi-panel layouts are dense.
  6. Related work: briefly position against other differentiable FEM/plasticity efforts beyond JAX-FEM/JAX-CPFEM (e.g., other AD-enabled continuum codes) so novelty of the anisotropic finite-strain + inverse stack is sharper.

Circularity Check

1 steps flagged

No load-bearing circularity; inverse recoveries are standard same-model synthetic verification, gradients cross-checked by independent FD, and JAX-FEM base is an engineering substrate not a uniqueness claim.

specific steps
  1. self citation load bearing [Abstract / §1 / §2.2.5 (JAX-FEM base)]
    "Built on JAX-FEM, the framework exploits modern accelerator architectures by parallelizing the three major computational bottlenecks in nonlinear FEM... JAX-FEM provides a general-purpose, GPU-accelerated differentiable finite element framework..."

    The implementation re-uses the authors’ own prior JAX-FEM library for elemental residual/stiffness kernels, sparse assembly and linear solves. This is ordinary engineering reuse of a software substrate rather than a load-bearing uniqueness or ansatz claim; the anisotropic constitutive update, custom JVPs, adjoint formulation and inverse results are independently derived and externally verified against Abaqus and finite differences. Flagged only as a minor self-citation that does not force any scientific conclusion.

full rationale

The paper implements Miehe’s modular logarithmic-strain finite-strain plasticity (Hill–48 and Yld2004-18p) inside an existing differentiable GPU FEM library, supplies consistent tangents and adjoints via custom JVPs and reverse-mode AD, and verifies forward solutions against Abaqus and objective gradients against central finite differences (absolute differences <0.2 % at h=10^{-6}). Inverse examples recover the very parameters used to generate the synthetic full-field displacements (with optional additive Gaussian noise); this is ordinary method verification, not a derivation that forces the result by construction, and the paper itself reports the well-known non-uniqueness of the (Q,b) hardening pair. The sole minor self-reference is the reuse of the authors’ prior JAX-FEM infrastructure for residual/stiffness evaluation and assembly; that infrastructure is an engineering substrate, not a uniqueness theorem or ansatz that closes the logical loop. Real DIC transfer is explicitly left to future work. Consequently the derivation chain is self-contained against external benchmarks and exhibits no circular reduction of the kind enumerated in the analyzer rules.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claims rest on standard continuum mechanics and FEM practice plus engineering modeling choices (additive logarithmic-strain modular plasticity, F-bar, Voce hardening, associative flow, synthetic DIC-like data). No new physical entities are postulated. Free parameters are the constitutive coefficients being identified and a few algorithmic tolerances/step sizes; they are either optimization unknowns or standard numerical controls, not hidden fits that manufacture the forward verification.

free parameters (4)
  • Constitutive parameter vector θ (Hill r_ij, Voce σ0/Q/b, Barlat c′/c′′ coefficients, m)
    These are the inverse unknowns; ground-truth values are chosen for synthetic experiments and recovered by optimization. Claims of accurate recovery depend on these synthetic targets and bounds (±50% for Barlat demo).
  • Finite-difference step size h and AD regularization δ for spectral map
    h is used only for gradient verification; spectral perturbation δ≪1 is an implementation choice to make eigh differentiable. Both affect numerical gradient quality if poorly chosen.
  • L-BFGS-B convergence tolerances, parameter normalization bounds, noise scale δ_noise
    Optimizer and synthetic-noise settings (Tables 9–12) influence reported iteration counts and recovered errors; not physics free parameters but they shape the inverse results shown.
  • Adaptive time-step bounds (Δt_initial, Δt_min, Δt_max) and line-search β, η
    Numerical controls for Newton robustness; affect forward convergence and thus inverse feasibility, especially for Barlat.
axioms (5)
  • domain assumption Finite-strain anisotropic plasticity is adequately represented by Miehe’s additive logarithmic-strain modular framework with associative flow and isotropic Voce hardening.
    §2.2 adopts this modular preprocessor/constitutive/postprocessor split; inverse claims assume model form is correct for the material of interest.
  • domain assumption Quasi-static equilibrium without body forces/inertia; F-bar kinematics control volumetric locking.
    Governing equations (1)–(2) and F-bar §2.2.4; standard solid-mechanics modeling choices.
  • standard math Discrete adjoint / reverse-mode AD through the time-discrete residual yields correct dJ/dθ for path-dependent plasticity when custom VJPs use the implicit-function theorem at local and global Newton levels.
    §2.3 and Appendix A; standard adjoint FEMU theory plus JAX custom_vjp practice.
  • ad hoc to paper Synthetic full-field displacements (optionally with i.i.d. Gaussian noise) are a sufficient proxy to demonstrate inverse identification performance relevant to DIC-based Material Testing 2.0.
    All inverse sections §3.3–3.6 use synthetic data; conclusions defer real DIC. This is the load-bearing experimental-transfer assumption.
  • domain assumption Automatic differentiation through spectral logarithmic strain can be made well-defined via small diagonal regularization or analytic Seth–Hill JVPs.
    §2.2.5; required for end-to-end differentiability of the geometric pre/postprocessors.

pith-pipeline@v1.1.0-grok45 · 37011 in / 3356 out tokens · 32931 ms · 2026-07-14T17:47:00.940074+00:00 · methodology

0 comments
read the original abstract

We present a fully differentiable, GPU-accelerated finite element framework JAX-FEM-ANISO, for forward simulation and inverse parameter identification of finite-strain anisotropic plasticity. Built on JAX-FEM, the framework exploits modern accelerator architectures by parallelizing the three major computational bottlenecks in nonlinear FEM: elemental weak-form and tangent-stiffness evaluation, global sparse matrix assembly, and sparse linear solution. For a large-scale forward problem with 3 million degrees of freedom, JAX-FEM-ANISO on a single NVIDIA H100 GPU achieves up to 9.4$\times$ speed-up over a 24-core CPU Abaqus baseline. Automatic differentiation is applied through the constitutive update and solver workflow, providing consistent Jacobians for complex constitutive models without manual derivation and accurate gradients for PDE-constrained inverse analysis. Compared with finite differences, the JAX-AD gradients avoid step-size sensitivity and provide the required sensitivities at substantially lower computational cost. For inverse characterization, we combine information-rich, topology-optimized heterogeneous specimens with full-field displacement data to identify advanced constitutive model parameters from a single test, replacing what would otherwise require many conventional experiments. We demonstrate accurate recovery of anisotropic yield and hardening parameters in progressively challenging settings, including uniform and spatially varying material properties. The resulting AD-based formulation enables efficient optimization in high-dimensional parameter spaces where finite-difference approaches are computationally infeasible. These results establish differentiable, GPU-accelerated FEM as a practical high-throughput engine for simulation, characterization, and optimization workflows in advanced manufacturing.

Figures

Figures reproduced from arXiv: 2606.17390 by Deepak Sharma, Hui-Ping Wang, Itzel Salgado, Jian Cao, Lu Huang.

Figure 1
Figure 1. Figure 1: Overview of the proposed differentiable GPU-accelerated framework for gradient-based parameter identification of finite-strain anisotropic plasticity: (a) parameterization of the constitutive model (M) by a set of material parameters (θ), (b) forward simulation using differentiable FEM, (c) objective function evaluation by comparing the mismatch of the simulated and ground truth displacement fields, (d) gr… view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of an elastoplastic solid body with displacement and traction boundary conditions. 2.2.2 Additive Lagrangian Modular Approach Finite-strain plasticity with material anisotropy is essential for modeling sheet metals subjected to complex loading paths. Several algorithmic challenges arise at large strains, including the choice of strain measure, objective stress mapping, robust return mapping for p… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the differentiable GPU-accelerated forward-solve implementation. while CuDSS is employed as a fallback when the iterative solve does not converge within a prescribed number of iterations. A comparative assessment of these solver choices in terms of computational performance is provided in Appendix D. Automatic time-stepping scheme. To improve robustness in strongly nonlinear loading regimes, th… view at source ↗
Figure 4
Figure 4. Figure 4: Schematic diagram of the thin circular flange considered for the drawing simulations. All dimensions are in mm [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of equivalent plastic strain in flange drawing simulations between JAX-FEM and Abaqus. The deformed meshes are shown at radial displacements of u = 25, 50, and 75 mm. Results correspond to anisotropic responses characterized by (a) δ = 3.4641 and (b) δ = 0.86603 [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Verification setup for the Barlat Yld2004-18p model. (a) Unit-cube single-element benchmark with the applied boundary conditions. (b) Prescribed displacement history imposed on the highlighted nodes [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of the stress plots obtained using JAX-FEM and Abaqus for the Barlat Yld2004-18p verification test. 3.1.3 Computational Performance Benchmarks To assess the computational performance and scalability of JAX-FEM, we consider the unit-cube geometry shown in [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Schematic of the unit-cube example with loading and boundary conditions used for the forward-simulation benchmarks [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Speed-up comparison between JAX-FEM and Abaqus for the unit-cube benchmark with 100 × 100 × 100 elements, using the Abaqus CPU (24-core) case as the reference configuration. The Abaqus simulations are performed on different CPU configurations ranging from 1 core to 128 cores, together with an Abaqus GPU-enabled run [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Speed-up comparison between JAX-FEM and Abaqus for the larger unit-cube benchmark with approximately 11.2 million degrees of freedom, using the Abaqus CPU (24-core) case as the reference configuration [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Loading and boundary conditions for the unit-cube domain considered for gradient verification [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Finite-difference error check EFD,check as a function of step size h [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Information-rich tensile specimen geometry adapted from Zhang et al. [19]. (a) (b) [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: (a) Distribution of equivalent plastic strain and (b) stress-state distribution for integration points subjected to plastic deformation. Both axes are normalized by the equivalent Hill stress; σ11 and σ22 are aligned with the rolling direction (RD) and transverse direction (TD), respectively [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Objective-function histories until convergence of the optimization [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Comparison of hardening curves for different material-parameter initializations [PITH_FULL_IMAGE:figures/full_fig_p024_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Cruciform specimen geometry with the considered loading and boundary conditions [PITH_FULL_IMAGE:figures/full_fig_p026_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Plastic-strain distribution at the final load step [PITH_FULL_IMAGE:figures/full_fig_p027_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Visualization of noisy synthetic data used for inverse identification at different noise levels [PITH_FULL_IMAGE:figures/full_fig_p027_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Convergence history for inverse identification using noise-free synthetic data. (a) (b) [PITH_FULL_IMAGE:figures/full_fig_p028_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Convergence histories for inverse identification using noisy synthetic data: (a) objective function values and (b) reduction in objective function values measured as the difference between the initial and current objective function values during optimization. 3.5 Inverse Identification of Spatially Varying Material Properties Engineering structures often exhibit spatial heterogeneity in material propertie… view at source ↗
Figure 22
Figure 22. Figure 22: (a) Cruciform geometry with spatially varying material properties; (b) distribution of plastic strain at the final loading step [PITH_FULL_IMAGE:figures/full_fig_p029_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Convergence history for identifying spatially varying material properties [PITH_FULL_IMAGE:figures/full_fig_p030_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Comparison of inverse-identified hardening curves with ground-truth results [PITH_FULL_IMAGE:figures/full_fig_p030_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Topology-optimized specimen geometry used for inverse identification of Barlat’s Yld2004-18p model, including the applied boundary conditions and loading direction. The specimen thickness is set to 1.0 mm. All dimensions are in mm [PITH_FULL_IMAGE:figures/full_fig_p031_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: (a) Equivalent plastic-strain distribution with the deformed geometry at the final time step, (b) principal strain-state diagram at the final time step, and (c) the corresponding principal stress-state diagram with yield surfaces overlaid. Only material points undergoing plastic deformation are shown in the state-space plots. Because negligible out-of-plane shear stresses are expected for the current spec… view at source ↗
Figure 27
Figure 27. Figure 27: Inverse identification of Barlat’s model: (a) comparison of reference and identified yield surfaces in the principal-stress plane for two hardening states (α = 0 and α = 0.2), and (b) comparison of reference and identified hardening curves [PITH_FULL_IMAGE:figures/full_fig_p033_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: Inverse optimization for Barlat material parameter identification: (a) convergence history of the optimization, and (b) comparison of forward-pass simulation times with and without warm-start initialization. 4 Conclusions This work presented an end-to-end differentiable, GPU-accelerated finite element framework for inverse characterization of finite￾strain anisotropic plasticity. The formulation combines … view at source ↗

Review history (2 revisions) →

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

68 extracted references · 1 linked inside Pith

  1. [1]

    J. C. Simo and T. J. R. Hughes.Computational Inelasticity. Springer, New York, NY , 1998

    1998

  2. [2]

    Belytschko, W

    T. Belytschko, W. K. Liu, and B. Moran.Nonlinear Finite Elements for Continua and Structures. John Wiley & Sons, Chichester, UK, 2000

    2000

  3. [3]

    R. Hill. A theory of the yielding and plastic flow of anisotropic metals.Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences, 193(1033):281–297, 1948

    1948

  4. [4]

    Barlat, J

    F. Barlat, J. C. Brem, J. W. Yoon, K. Chung, R. E. Dick, D. J. Lege, F. Pourboghrat, S.-H. Choi, and E. Chu. Plane stress yield function for aluminum alloy sheets—part 1: theory.International Journal of Plasticity, 19(9):1297–1319, 2003

    2003

  5. [5]

    Barlat, H

    F. Barlat, H. Aretz, J. W. Yoon, M. E. Karabin, J. C. Brem, and R. E. Dick. Linear transformation-based anisotropic yield functions.International Journal of Plasticity, 21(5):1009–1039, 2005

    2005

  6. [6]

    E. A. de Souza Neto, D. Peri´c, and D. R. J. Owen.Computational Methods for Plasticity: Theory and Applications. John Wiley & Sons, Chichester, UK, 2008

    2008

  7. [7]

    Michaleris, D

    P. Michaleris, D. A. Tortorelli, and C. A. Vidal. Tangent operators and design sensitivity formulations for transient non- linear coupled problems with applications to elastoplasticity.International Journal for Numerical Methods in Engineering, 37(14):2471–2499, 1994

    1994

  8. [8]

    Miehe, N

    C. Miehe, N. Apel, and M. Lambrecht. Anisotropic additive plasticity in the logarithmic strain space: modular kinematic formulation and implementation based on incremental minimization principles for standard materials.Computer Methods in Applied Mechanics and Engineering, 191:5383–5425, 2002

    2002

  9. [9]

    N. Aravas. On the numerical integration of a class of pressure-dependent plasticity models.International Journal for Numerical Methods in Engineering, 24(7):1395–1416, 1987

    1987

  10. [10]

    Güner, C

    A. Güner, C. Soyarslan, A. Brosius, and A. E. Tekkaya. Characterization of anisotropy of sheet metals employing inhomogeneous strain fields for Yld2000-2D yield function.International Journal of Solids and Structures, 49(25):3517– 3527, 2012

    2012

  11. [11]

    J. H. Kim, F. Barlat, F. Pierron, and M. G. Lee. Determination of anisotropic plastic constitutive parameters using the virtual fields method.Experimental Mechanics, 54(7):1189–1204, 2014

    2014

  12. [12]

    M. A. Sutton, J.-J. Orteu, and H. Schreier.Image Correlation for Shape, Motion and Deformation Measurements: Basic Concepts, Theory and Applications. Springer, New York, NY , 2009

    2009

  13. [13]

    Bertin, F

    M. Bertin, F. Hild, and S. Roux. On the identifiability of Hill-1948 plasticity model with a single biaxial test on very thin sheet.Strain, 53(5):e12233, 2017

    1948

  14. [14]

    Coppieters, T

    S. Coppieters, T. Hakoyama, D. Debruyne, S. Takahashi, and T. Kuwabara. Inverse yield locus identification of sheet metal using a complex cruciform in biaxial tension and digital image correlation.Proceedings, 60(2):382, 2018

    2018

  15. [15]

    Zhang, M

    Y . Zhang, M. Rossi, R. Pettermann, and C. Sommitsch. Inverse identification of plastic anisotropy through multiple non-conventional mechanical experiments.International Journal of Solids and Structures, 285:112534, 2023

    2023

  16. [16]

    Pierron and M

    F. Pierron and M. Grédiac. Towards material testing 2.0. a review of test design for identification of constitutive parameters from full-field measurements.Strain, 57(1):e12370, 2021

    2021

  17. [17]

    J. M. P. Martins, A. Andrade-Campos, and S. Thuillier. Comparison of inverse identification strategies for constitutive mechanical models using full-field measurements.International Journal of Mechanical Sciences, 145:330–345, 2018

    2018

  18. [18]

    Marek, F

    A. Marek, F. M. Davis, M. Rossi, and F. Pierron. Extension of the sensitivity-based virtual fields to large deformation anisotropic plasticity.International Journal of Material Forming, 12(3):457–476, 2019

    2019

  19. [19]

    Zhang, M

    Y . Zhang, M. Rossi, J. Reisch, R. Pettermann, and C. Sommitsch. Enhancing the information-richness of sheet metal specimens for inverse identification of plastic anisotropy through strain fields.International Journal of Mechanical Sciences, 214:106891, 2022

    2022

  20. [20]

    Conde, Y

    M. Conde, Y . Zhang, J. Henriques, S. Coppieters, and A. Andrade-Campos. Design and validation of a heterogeneous interior notched specimen for inverse material parameter identification.Finite Elements in Analysis and Design, 214:103866, 2023

    2023

  21. [21]

    Gonçalves, S

    M. Gonçalves, S. Thuillier, and A. Andrade-Campos. Inverse identification of anisotropic plasticity model parameters using FEMU and a heterogeneous test. InESAFORM 2025, volume 54, pages 1548–1557, 2025

    2025

  22. [22]

    D. T. Seidl and B. N. Granzow. Calibration of elastoplastic constitutive model parameters from full-field data with automatic differentiation-based sensitivities.International Journal for Numerical Methods in Engineering, 123(1):69–100, 2022

    2022

  23. [23]

    Kumar, D

    S. Kumar, D. T. Seidl, B. N. Granzow, J. Yang, and J. N. Fuhg. A comparative study of calibration techniques for finite strain elastoplasticity: Numerically-exact sensitivities for FEMU and VFM.Computer Methods in Applied Mechanics and Engineering, 444:118159, 2025

    2025

  24. [24]

    C. C. Margossian. A review of automatic differentiation and its efficient implementation.Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 9(4):e1305, 2019

    2019

  25. [25]

    Rothe and S

    S. Rothe and S. Hartmann. Automatic differentiation for stress and consistent tangent computation.Archive of Applied Mechanics, 85(8):1103–1125, 2015

    2015

  26. [26]

    Dummer, M

    A. Dummer, M. Neuner, P. Gamnitzer, and G. Hofstetter. Robust and efficient implementation of finite strain generalized continuum models for material failure: Analytical, numerical, and automatic differentiation with hyper-dual numbers. Computer Methods in Applied Mechanics and Engineering, 426:116987, 2024

    2024

  27. [27]

    Y . Jia, W. Li, and X. S. Zhang. Multimaterial topology optimization of elastoplastic composite structures.Journal of the Mechanics and Physics of Solids, 196:106018, 2025

    2025

  28. [28]

    Jia and X

    Y . Jia and X. S. Zhang. Multimaterial topology optimization for finite strain elastoplasticity: theory, methods, and applications. Computer Methods in Applied Mechanics and Engineering, 449:118445, 2026

    2026

  29. [29]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. Vander- Plas, S. Wanderman-Milne, and Q. Zhang. JAX: composable transformations of Python+NumPy programs. http: //github.com/google/jax, 2018

    2018

  30. [30]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, et al. PyTorch: An imperative style, high-performance deep learning library.Advances in Neural Information Processing Systems, 32, 2019

    2019

  31. [31]

    M. Macklin. Warp: A high-performance python framework for GPU simulation and graphics. InNVIDIA GPU Technology Conference (GTC), volume 3, 2022

    2022

  32. [32]

    NVIDIA H100 Tensor Core GPU datasheet

    NVIDIA Corporation. NVIDIA H100 Tensor Core GPU datasheet. https://resources.nvidia.com/ en-us-hopper-architecture/nvidia-tensor-core-gpu-datasheet, 2023

    2023

  33. [33]

    Intel Xeon 6952P Processor specifications

    Intel Corporation. Intel Xeon 6952P Processor specifications. https://www.intel.com/content/www/us/en/ products/sku/241643/intel-xeon-6952p-processor-480m-cache-2-10-ghz/specifications.html, 2026

    2026

  34. [34]

    Williams, A

    S. Williams, A. Waterman, and D. Patterson. Roofline: an insightful visual performance model for multicore architectures. Communications of the ACM, 52(4):65–76, 2009

    2009

  35. [35]

    Martínez-Frutos, P

    J. Martínez-Frutos, P. J. Martínez-Castejón, and D. Herrero-Pérez. Fine-grained GPU implementation of assembly-free iterative solver for finite element problems.Computers&Structures, 157:9–18, 2015

    2015

  36. [36]

    Aissa, T

    M. Aissa, T. Verstraete, and C. Vuik. Toward a GPU-aware comparison of explicit and implicit CFD simulations on structured meshes.Computers&Mathematics with Applications, 74(1):201–217, 2017

    2017

  37. [37]

    Stavroulakis, D

    G. Stavroulakis, D. G. Giovanis, V . Papadopoulos, and M. Papadrakakis. A GPU domain decomposition solution for spectral stochastic finite element method.Computer Methods in Applied Mechanics and Engineering, 327:392–410, 2017

    2017

  38. [38]

    Ghysels and R

    P. Ghysels and R. Synk. High performance sparse multifrontal solvers on modern GPUs.Parallel Computing, 110:102897, 2022

    2022

  39. [39]

    F. Li, F. Zou, and J. Rao. A multi-GPU and CUDA-aware MPI-based spectral element formulation for ultrasonic wave propagation in solid media.Ultrasonics, 134:107049, 2023

    2023

  40. [40]

    Kiran, D

    U. Kiran, D. Sharma, and S. S. Gautam. Development of GPU-based matrix-free strategies for large-scale elastoplasticity analysis using conjugate gradient solver.International Journal for Numerical Methods in Engineering, 125(7):e7421, 2024

    2024

  41. [41]

    S. Hong, G. Jang, and W. K. Jeong. MG-FIM: A multi-GPU fast iterative method using adaptive domain decomposition. SIAM Journal on Scientific Computing, 44(1):C54–C76, 2022

    2022

  42. [42]

    S. Liao, A. Golgoon, M. Mozaffar, and J. Cao. Efficient GPU-accelerated thermomechanical solver for residual stress prediction in additive manufacturing.Computational Mechanics, 71(5):879–893, 2023

    2023

  43. [43]

    E. A. Träff, A. Rydahl, S. Karlsson, O. Sigmund, and N. Aage. Simple and efficient GPU accelerated topology optimisation: Codes and applications.Computer Methods in Applied Mechanics and Engineering, 410:116043, 2023

    2023

  44. [44]

    Kiran, D

    U. Kiran, D. Sharma, and S. S. Gautam. An efficient framework for matrix-free SpMV computation on GPU for elastoplastic problems.Mathematics and Computers in Simulation, 216:318–346, 2024

    2024

  45. [45]

    Karatarakis, P

    A. Karatarakis, P. Karakitsios, and M. Papadrakakis. GPU accelerated computation of the isogeometric analysis stiffness matrix.Computer Methods in Applied Mechanics and Engineering, 269:334–355, 2014

    2014

  46. [46]

    T. Xue, S. Liao, Z. Gan, C. Park, X. Xie, W. K. Liu, and J. Cao. JAX-FEM: A differentiable GPU-accelerated 3d finite element solver for automatic inverse design and mechanistic data science.Computer Physics Communications, 291:108802, 2023

    2023

  47. [47]

    F. Hu, S. Niezgoda, T. Xue, and J. Cao. Efficient GPU-computing simulation platform JAX-CPFEM for differentiable crystal plasticity finite element method.npj Computational Materials, 11(1):46, 2025

    2025

  48. [48]

    F. Hu, R. Zhou, K. Ryou, R. Zha, S. Niezgoda, T. Xue, and J. Cao. An efficient graphical processing unit-accelerated calibration of crystal plasticity model parameters by multi-objective optimization with automatic differentiation-based sensitivities.Journal of Applied Mechanics, 93(2):021006, 2026

    2026

  49. [49]

    B. P. Ferreira and M. A. Bessa. Automatically differentiable model updating (ADiMU): conventional, hybrid, and neural network material model discovery including history-dependency.arXiv preprint arXiv:2505.07801, 2025

    Pith/arXiv arXiv 2025

  50. [50]

    Knapik, S

    S. Knapik, S. Deng, L. Wang, A. Chandrasekhar, W. K. Liu, J. Cao, and W. Chen. Co-design of geometry and thermal-elastic gradient alloy distribution with temperature-dependent material properties.Structural and Multidisciplinary Optimization, 68(7):127, 2025

    2025

  51. [51]

    J. P. Leonor and G. J. Wagner. GO-MELT: GPU-optimized multilevel execution of LPBF thermal simulations.Computer Methods in Applied Mechanics and Engineering, 426:116977, 2024

    2024

  52. [52]

    AmgX: GPU accelerated linear solvers.https://developer.nvidia.com/amgx, 2026

    NVIDIA Corporation. AmgX: GPU accelerated linear solvers.https://developer.nvidia.com/amgx, 2026

    2026

  53. [53]

    E. A. de Souza Neto, D. Peric, and D. R. Owen.Computational Methods for Plasticity: Theory and Applications. John Wiley & Sons, Chichester, UK, 2011

    2011

  54. [54]

    W. M. Scherzinger. A return mapping algorithm for isotropic and anisotropic plasticity models using a line search method. Computer Methods in Applied Mechanics and Engineering, 317:526–553, 2017

    2017

  55. [55]

    Scalet and F

    G. Scalet and F. Auricchio. Computational methods for elastoplasticity: An overview of conventional and less-conventional approaches.Archives of Computational Methods in Engineering, 25(3):545–589, 2018

    2018

  56. [56]

    Alfano, L

    G. Alfano, L. Rosati, and N. Valoroso. A tangent–secant approach to rate-independent elastoplasticity: formulations and computational issues.Computer Methods in Applied Mechanics and Engineering, 179(3–4):379–405, 1999

    1999

  57. [57]

    Miehe and M

    C. Miehe and M. Lambrecht. Algorithms for computation of stresses and elasticity moduli in terms of Seth–Hill’s family of generalized strain tensors.Communications in Numerical Methods in Engineering, 17(5):337–353, 2001

    2001

  58. [58]

    Z. Fu, T. J. Lewis, R. M. Kirby, and R. T. Whitaker. Architecting the finite element method pipeline for the GPU.Journal of Computational and Applied Mathematics, 257:195–211, 2014

    2014

  59. [59]

    Meng, B.-L

    H.-T. Meng, B.-L. Nie, S. Wong, C. Macon, and J.-M. Jin. GPU accelerated finite-element computation for electromagnetic analysis.IEEE Antennas and Propagation Magazine, 56(2), 2014

    2014

  60. [60]

    Kiran, D

    U. Kiran, D. Sharma, and S. S. Gautam. GPU-warp based finite element matrices generation and assembly using coloring method.Journal of Computational Design and Engineering, 6(5):705–718, 2019

    2019

  61. [61]

    M. B. Giles and N. A. Pierce. An introduction to the adjoint approach to design.Flow, Turbulence and Combustion, 65:393–415, 2000

    2000

  62. [62]

    Xu and E

    K. Xu and E. Darve. Physics constrained learning for data-driven inverse modeling from sparse observations.Journal of Computational Physics, 453:110938, 2022

    2022

  63. [63]

    Papadopoulos and J

    P. Papadopoulos and J. Lu. A general framework for the numerical solution of problems in finite elasto-plasticity.Computer Methods in Applied Mechanics and Engineering, 159(1–2):1–18, 1998

    1998

  64. [64]

    unified material model driver for plasticity (UMMDp)

    H. Takizawa, K. Oide, K. Suzuki, T. Yamanashi, T. Inoue, T. Ida, T. Nagai, and T. Kuwabara. Development of the user subroutine library “unified material model driver for plasticity (UMMDp)” for various anisotropic yield functions.Journal of Physics: Conference Series, 1063:012099, 2018

    2018

  65. [65]

    Tammas-Williams and I

    S. Tammas-Williams and I. Todd. Design for additive manufacturing with site-specific properties in metals and alloys. Scripta Materialia, 135:105–110, 2017

    2017

  66. [66]

    Gonçalves, A

    M. Gonçalves, A. Andrade-Campos, and B. Barroqueiro. On the design of mechanical heterogeneous specimens using multilevel topology optimization.Advances in Engineering Software, 175:103314, 2023

    2023

  67. [67]

    Gonçalves, M

    M. Gonçalves, M. G. Oliveira, S. Thuillier, and A. Andrade-Campos. Key performance indicators for heterogeneous mechanical tests.International Journal of Mechanical Sciences, 264:108821, 2024

    2024

  68. [68]

    Efficient and modular implicit differentiation.Advances in neural information processing systems, 35:5230–5242, 2022

    Mathieu Blondel, Quentin Berthet, Marco Cuturi, Roy Frostig, Stephan Hoyer, Felipe Llinares-López, Fabian Pedregosa, and Jean-Philippe Vert. Efficient and modular implicit differentiation.Advances in neural information processing systems, 35:5230–5242, 2022

    2022