arxiv: 2605.09307 · v1 · submitted 2026-05-10 · ⚛️ physics.comp-ph

Recognition: 2 theorem links

· Lean Theorem

Constitutive Priors for Inverse Design

Bahador Bahmani, Jinkyo Han

Pith reviewed 2026-05-12 02:04 UTC · model grok-4.3

classification ⚛️ physics.comp-ph

keywords inverse designconstitutive modelselastic networkslatent representationPDE-constrained optimizationthermodynamic consistencyhomotopy continuationChamfer distance

0 comments

The pith

Inverse design of elastic networks proceeds by optimizing directly over a latent manifold of thermodynamically consistent constitutive behaviors learned from noisy stress-strain data.

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

The paper builds a constitutive prior as a latent representation that maps noisy experimental stress-strain curves onto a manifold of physically admissible material laws. This prior becomes the search space for an inverse design task formulated as PDE-constrained optimization, where the variables parameterize spatially varying elasticity. A homotopy continuation strategy deforms an initial registered point cloud toward the target geometry, and geometry comparison uses the Chamfer distance so that no mesh correspondence is required. An additional neural smoothness prior penalizes abrupt spatial jumps to respect fabrication limits. The result is a complete pipeline that converts noisy material data into manufacturable network designs without intermediate constitutive model fitting.

Core claim

A latent representation learned from noisy stress-strain data can be used to define a manifold of admissible elastic material laws that automatically enforces thermodynamic consistency; the inverse design problem is then solved as PDE-constrained optimization over these latent variables, with homotopy continuation on affinely registered point clouds and a neural smoothness prior providing robustness and manufacturability.

What carries the argument

The constitutive prior, a latent manifold of admissible material laws constructed from noisy stress-strain data while enforcing thermodynamic consistency.

If this is right

The inverse problem becomes well-posed over latent constitutive variables that parameterize arbitrary spatial variation in material response.
Homotopy continuation on intermediate affinely registered point clouds improves convergence of the nonconvex PDE-constrained problem.
Chamfer distance allows geometry matching without requiring identical mesh topology between reference and target configurations.
A neural smoothness prior together with a graph-based metric enforces gradual spatial changes consistent with manufacturing constraints.

Where Pith is reading between the lines

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

The same latent-manifold construction could be applied to other continuum systems such as plasticity or viscoelasticity by replacing the elastic energy functional.
Because the prior is built directly from raw data, the approach may tolerate higher measurement noise than methods that first fit explicit constitutive equations.
The framework suggests a route to co-design of material distribution and network topology when the latent space is further coupled to a topology optimization loop.

Load-bearing premise

The latent representation learned from noisy stress-strain data successfully defines a manifold that contains only thermodynamically consistent material laws and that the homotopy plus smoothness terms make the nonconvex optimization tractable.

What would settle it

An optimized network whose recovered stress-strain response under a new loading path violates the Clausius-Duhem inequality or whose deformed geometry deviates from the target by more than the measurement noise level.

Figures

Figures reproduced from arXiv: 2605.09307 by Bahador Bahmani, Jinkyo Han.

**Figure 1.** Figure 1: Schematic illustration of the adjoint-based optimization for heterogeneous material distributions in elastic [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗

**Figure 2.** Figure 2: Training of the material surrogate with trainable material parameter [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Figure 3: Results of training the material surrogate with “fixed” (non trainable) latent variables [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

**Figure 4.** Figure 4: A one-dimensional bar with length L = 1 m. The left node is fixed (red dot), and a tensile nodal force f = 2 N (blue arrow) is applied at the right node, where the cross sectional area of the bar is set to A = 1 mm2 . The bar shown in [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗

**Figure 5.** Figure 5: Results of the inverse problem for the uniaxial extension of a bar, converged to a loss threshold for the [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

**Figure 6.** Figure 6: The comparison between the two surrogates in the inverse identification task. The prescribed threshold is [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗

**Figure 7.** Figure 7: Problem setting of the airfoil-shaped truss system with [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

**Figure 8.** Figure 8: Inverse design results for the airfoil-shaped truss system under the arbitrary wind-gust loading profile. The [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗

**Figure 9.** Figure 9: Problem setting of the edge crack lattice example. ( [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗

**Figure 10.** Figure 10: Inverse identification of heterogeneous stiffness of the truss system. ( [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗

**Figure 11.** Figure 11: A reference configuration of the truss system with [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗

**Figure 12.** Figure 12: Inverse design of the distribution α in the Neo-Hookean truss system. (a) The optimized truss system using SIREN, and (b) the element-wise parametrization of α [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗

**Figure 13.** Figure 13: Inverse design of the distribution of α in the thermoelastic truss system under a continuous representation, using the MMA as the optimizer. (a) Obtained distribution of the thermal expansion coefficient α, parameterized by SIREN. (b) Corresponding loss history, shown together with the loss histories obtained by the proposed algorithm in [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗

**Figure 14.** Figure 14: A reference configuration of the truss system with [PITH_FULL_IMAGE:figures/full_fig_p017_14.png] view at source ↗

**Figure 15.** Figure 15: Inverse design of the material distribution in the truss system. ( [PITH_FULL_IMAGE:figures/full_fig_p017_15.png] view at source ↗

**Figure 16.** Figure 16: Inverse design of the material distribution in the truss system, resulted without the proposed homotopy [PITH_FULL_IMAGE:figures/full_fig_p018_16.png] view at source ↗

**Figure 17.** Figure 17: Histories of (a) Chamfer distance and (b) residual norm for the settings with and without homotopy-based continuation. The tolerances for the Chamfer distance and the residual norm are shown as black dashed lines. The residual norm is reported in the ℓ 2 norm. Similar to the study in Section 3.2.1, we examine the importance of a smooth constitutive prior in function space during the inversion process. We … view at source ↗

**Figure 18.** Figure 18: Loss histories of the inverse identification using constitutive prior A and constitutive prior B. The threshold [PITH_FULL_IMAGE:figures/full_fig_p019_18.png] view at source ↗

**Figure 19.** Figure 19: Loss histories over multiple trials of the problem in Section [PITH_FULL_IMAGE:figures/full_fig_p025_19.png] view at source ↗

**Figure 20.** Figure 20: Current configurations at the final load step, obtained from three different initializations of the truss-member [PITH_FULL_IMAGE:figures/full_fig_p025_20.png] view at source ↗

**Figure 21.** Figure 21: Loss histories over multiple trials of the problem in Section [PITH_FULL_IMAGE:figures/full_fig_p026_21.png] view at source ↗

**Figure 22.** Figure 22: Additional multiple trials of the problem in Section [PITH_FULL_IMAGE:figures/full_fig_p026_22.png] view at source ↗

read the original abstract

This work introduces an end-to-end framework for inverse design of elastic networks directly in the space of constitutive behaviors. A constitutive prior is constructed from noisy stress-strain data using a latent representation that defines a manifold of admissible material laws while enforcing thermodynamic consistency. The inverse problem is formulated as a PDE-constrained optimization problem over latent constitutive variables that parameterize spatially varying material behavior. To improve robustness in the resulting nonconvex optimization, a homotopy-based continuation strategy is introduced using intermediate target point clouds generated through affine registration. Geometry matching is performed using the Chamfer distance, enabling optimization without requiring mesh correspondence between the target and reference configurations. To account for manufacturing constraints limiting abrupt spatial variation in material properties, the framework additionally incorporates a neural-network-based smoothness prior together with a graph-based smoothness metric. The proposed approach is demonstrated on several inverse design problems for elastic networks and compared against alternative optimization strategies.

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.

Desk Editor's Note private letter to a colleague

The paper offers a coherent framework for inverse design by learning a latent manifold of valid constitutive laws from data and optimizing inside it with homotopy and smoothness tools, and the pieces fit without obvious contradictions.

read the letter

The main thing here is a shift to doing inverse design for elastic networks directly in constitutive space. They learn a prior from noisy stress-strain curves via a latent representation that builds a manifold of admissible material laws and keeps thermodynamic consistency. The optimization then runs over those latent variables for spatially varying properties, with a homotopy continuation that generates intermediate targets via affine registration and matches via Chamfer distance so meshes do not have to line up exactly. A neural smoothness prior plus a graph metric is added to respect manufacturing limits on abrupt property changes. This is new in the specific combination: the prior supplies the feasible set, the continuation aids convergence on the nonconvex problem, and the smoothness term handles real constraints without post-hoc fixes. It does well at keeping the search inside physically reasonable behaviors from the start rather than hoping the optimizer lands there. The demonstrations on elastic network problems are mentioned, and the logic of the pipeline is internally consistent as described. The soft spots are mostly in the level of detail. The abstract and framework stay high-level on how the latent manifold is trained and validated against the noisy data, so it is not clear how much leakage of inconsistent laws occurs or how sensitive the results are to the choice of latent dimension. The Chamfer distance is reasonable for geometry but could under-weight mechanical fidelity if not balanced with other terms, and the neural smoothness prior's training and generalization are not spelled out enough to judge robustness. No major circularity or load-bearing assumption seems to break on its own terms. This is for researchers in computational mechanics who already work on PDE-constrained inverse problems and want concrete ways to inject data-driven priors and continuation tricks. A reader who needs methods for heterogeneous elastic design would find usable ideas even if the gains are incremental. It deserves a serious referee because the approach is grounded and addresses practical pain points, though revisions would likely need more quantitative ablations and comparisons to show the improvement clearly.

Referee Report

2 major / 2 minor

Summary. The paper introduces an end-to-end framework for inverse design of elastic networks directly in constitutive behavior space. A constitutive prior is learned from noisy stress-strain data via a latent representation that parametrizes a manifold of thermodynamically consistent material laws. Inverse design is posed as PDE-constrained optimization over these latent variables; robustness is addressed via homotopy continuation on affine-registered intermediate point clouds, Chamfer-distance geometry matching, and a neural-network smoothness prior to respect manufacturing constraints on spatial variation. The method is demonstrated on several elastic-network inverse-design problems and compared to alternative optimization approaches.

Significance. If the central claims are substantiated by quantitative validation, the work could advance inverse design in computational mechanics by shifting optimization into a data-driven but physically admissible constitutive manifold. The combination of latent-variable priors, homotopy continuation, Chamfer matching, and graph/neural smoothness penalties offers a principled route to non-convex PDE-constrained problems that arise in architected materials. Reproducibility would be strengthened by the explicit use of latent representations and continuation strategies.

major comments (2)

[Abstract and §4] Abstract and §4 (Results): the manuscript states that the approach is 'demonstrated on several inverse design problems' and 'compared against alternative optimization strategies,' yet no quantitative metrics (e.g., relative L2 error on recovered stress-strain curves, success rate over random initializations, or ablation of the homotopy/Chamfer components) are reported. Without these numbers the claim that the latent prior plus continuation improves robustness cannot be evaluated and is load-bearing for the central contribution.
[§3] §3 (Methods): the assertion that the latent representation 'defines a manifold of admissible material laws while enforcing thermodynamic consistency' is central, but the text provides only a high-level description of the auto-encoder and dissipation penalty. It is unclear whether the learned manifold strictly satisfies the dissipation inequality for all latent points (including extrapolations) or whether violations are merely penalized in expectation; an explicit verification (e.g., worst-case residual on the Clausius-Duhem inequality across the test set) is required.

minor comments (2)

[§2] Notation for the latent constitutive variables (denoted variously as z or θ in the abstract) should be unified and defined once in §2 before use in the optimization formulation.
[Figures 3-5] Figure captions for the elastic-network examples should include the specific values of the Chamfer-distance weight and the smoothness-regularization coefficient used in each run.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review. The comments identify key areas where additional quantitative evidence and clarification will strengthen the manuscript. We address each major comment below and will revise the paper accordingly.

read point-by-point responses

Referee: [Abstract and §4] Abstract and §4 (Results): the manuscript states that the approach is 'demonstrated on several inverse design problems' and 'compared against alternative optimization strategies,' yet no quantitative metrics (e.g., relative L2 error on recovered stress-strain curves, success rate over random initializations, or ablation of the homotopy/Chamfer components) are reported. Without these numbers the claim that the latent prior plus continuation improves robustness cannot be evaluated and is load-bearing for the central contribution.

Authors: We agree that quantitative metrics are necessary to substantiate the robustness claims. In the revised manuscript we will expand §4 with relative L2 errors on recovered stress-strain curves, success rates computed over multiple random initializations, and ablation studies that remove the homotopy continuation and Chamfer-distance components individually. These additions will allow direct evaluation of the contribution of the latent prior and continuation strategy. revision: yes
Referee: [§3] §3 (Methods): the assertion that the latent representation 'defines a manifold of admissible material laws while enforcing thermodynamic consistency' is central, but the text provides only a high-level description of the auto-encoder and dissipation penalty. It is unclear whether the learned manifold strictly satisfies the dissipation inequality for all latent points (including extrapolations) or whether violations are merely penalized in expectation; an explicit verification (e.g., worst-case residual on the Clausius-Duhem inequality across the test set) is required.

Authors: The auto-encoder is trained with an explicit dissipation penalty to encourage thermodynamic consistency. We acknowledge that the current text does not provide a quantitative verification of strict satisfaction versus penalization. In the revised §3 we will add an explicit check reporting the worst-case residual of the Clausius-Duhem inequality over the test set and for extrapolated latent points, thereby clarifying the degree of enforcement achieved by the learned manifold. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The provided abstract and framework description outline a data-driven construction of a constitutive prior via latent manifold from noisy stress-strain data, followed by separate PDE-constrained optimization over latent variables with standard techniques (homotopy continuation, Chamfer distance, neural smoothness prior). No equations, self-definitions, fitted inputs renamed as predictions, or load-bearing self-citations are present in the text. The prior supplies the admissible set independently of the optimization step, and the overall chain does not reduce any claimed result to its own inputs by construction. This is the expected non-circular outcome for a methods paper introducing a composite framework.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the ability to learn a valid constitutive prior from noisy data that enforces thermodynamic consistency and on the effectiveness of the described optimization components.

axioms (1)

domain assumption Thermodynamic consistency of material laws
Enforced via the latent representation constructed from noisy stress-strain data.

invented entities (2)

Constitutive prior no independent evidence
purpose: Defines manifold of admissible material laws from data
Constructed using latent representation to enforce consistency and enable optimization.
Latent constitutive variables no independent evidence
purpose: Parameterize spatially varying material behavior in PDE-constrained optimization
Central to formulating the inverse problem.

pith-pipeline@v0.9.0 · 5439 in / 1438 out tokens · 75226 ms · 2026-05-12T02:04:52.299035+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel (J uniqueness and convexity) echoes
We construct a data-driven prior over admissible constitutive behaviors, represented as a parameterized family of energy-based material models... By constructing the prior in terms of an energy potential, thermodynamic consistency is enforced by design... we enforce convexity of the learned energy functional in the stretch variable, while allowing flexible dependence on the material latent variable z through a PICNN.
IndisputableMonolith/Cost.lean Jcost convexity and positivity off-identity echoes
To ensure the existence of minimizers of the total energy functional, appropriate convexity conditions must be imposed on the constitutive model... In this work, we focus on lattice-based systems... well-posedness is ensured by convexity of the energy with respect to the elastic stretch.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 1 internal anchor

[1]

Input convex neural networks

Brandon Amos, Lei Xu, and J Zico Kolter. Input convex neural networks. InInternational conference on machine learning, pages 146–155. PMLR, 2017

work page 2017
[2]

A mechanics-informed artificial neural network approach in data-driven constitutive modeling.International Journal for Numerical Methods in Engineering, 123(12):2738– 2759, 2022

Faisal As’ ad, Philip Avery, and Charbel Farhat. A mechanics-informed artificial neural network approach in data-driven constitutive modeling.International Journal for Numerical Methods in Engineering, 123(12):2738– 2759, 2022

work page 2022
[3]

Conformal quantile regression for neural probabilistic constitutive modeling.Computer Methods in Applied Mechanics and Engineering, 457:118981, 2026

Bahador Bahmani. Conformal quantile regression for neural probabilistic constitutive modeling.Computer Methods in Applied Mechanics and Engineering, 457:118981, 2026

work page 2026
[4]

Physics-constrained symbolic model discovery for polyconvex incom- pressible hyperelastic materials.International Journal for Numerical Methods in Engineering, 125(15):e7473, 2024

Bahador Bahmani and WaiChing Sun. Physics-constrained symbolic model discovery for polyconvex incom- pressible hyperelastic materials.International Journal for Numerical Methods in Engineering, 125(15):e7473, 2024

work page 2024
[5]

Convexity conditions and existence theorems in nonlinear elasticity.Archive for rational mechanics and Analysis, 63(4):337–403, 1976

John M Ball. Convexity conditions and existence theorems in nonlinear elasticity.Archive for rational mechanics and Analysis, 63(4):337–403, 1976. 20 Constitutive PriorA PREPRINT

work page 1976
[6]

Springer Science & Business Media, 2013

Martin Philip Bendsoe and Ole Sigmund.Topology optimization: theory, methods, and applications. Springer Science & Business Media, 2013

work page 2013
[7]

Flexible mechanical metamaterials

Katia Bertoldi, Vincenzo Vitelli, Johan Christensen, and Martin Van Hecke. Flexible mechanical metamaterials. Nature Reviews Materials, 2(11):1–11, 2017

work page 2017
[8]

A framework for data-driven analysis of materials under uncertainty: Countering the curse of dimensionality.Computer Methods in Applied Mechanics and Engineering, 320:633–667, 2017

Miguel A Bessa, Ramin Bostanabad, Zeliang Liu, Anqi Hu, Daniel W Apley, Catherine Brinson, Wei Chen, and Wing Kam Liu. A framework for data-driven analysis of materials under uncertainty: Countering the curse of dimensionality.Computer Methods in Applied Mechanics and Engineering, 320:633–667, 2017

work page 2017
[9]

The elements of differentiable programming.arXiv preprint arXiv:2403.14606, 2024

Mathieu Blondel and Vincent Roulet. The elements of differentiable programming.arXiv preprint arXiv:2403.14606, 2024

work page arXiv 2024
[10]

Shape-shifting structured lattices via multimaterial 4d printing.Proceedings of the National Academy of Sciences, 116(42):20856–20862, 2019

J William Boley, Wim M Van Rees, Charles Lissandrello, Mark N Horenstein, Ryan L Truby, Arda Kotikian, Jennifer A Lewis, and L Mahadevan. Shape-shifting structured lattices via multimaterial 4d printing.Proceedings of the National Academy of Sciences, 116(42):20856–20862, 2019

work page 2019
[11]

Topology optimization for additive manufacturing

David Brackett, Ian Ashcroft, and R Hague. Topology optimization for additive manufacturing. 2011

work page 2011
[12]

Physical learning in reprogrammable metamaterials for adaptation to unknown environments.arXiv preprint arXiv:2510.06442, 2025

Kai Jun Chen, Catherine Catrambone, Christopher Sowinski, Jacob Mukobi, Enzo Andreacchio, Enquan Chew, Alexandre Morland, and Maria Sakovsky. Physical learning in reprogrammable metamaterials for adaptation to unknown environments.arXiv preprint arXiv:2510.06442, 2025

work page arXiv 2025
[13]

Topology optimized architectures with programmable poisson’s ratio over large deformations.Advanced Materials, 27 (37):5523–5527, 2015

Anders Clausen, Fengwen Wang, Jakob Søndergaard Jensen, Ole Sigmund, and Jennifer A Lewis. Topology optimized architectures with programmable poisson’s ratio over large deformations.Advanced Materials, 27 (37):5523–5527, 2015

work page 2015
[14]

Gustavo Assis Da Silva, Andre Teofilo Beck, and Ole Sigmund. Topology optimization of compliant mechanisms considering stress constraints, manufacturing uncertainty and geometric nonlinearity.Computer Methods in Applied Mechanics and Engineering, 365:112972, 2020

work page 2020
[15]

A differentiable physics engine for deep learning in robotics.Frontiers in neurorobotics, 13:406386, 2019

Jonas Degrave, Michiel Hermans, Joni Dambre, et al. A differentiable physics engine for deep learning in robotics.Frontiers in neurorobotics, 13:406386, 2019

work page 2019
[16]

Automated derivation of the adjoint of high-level transient finite element programs.SIAM Journal on Scientific Computing, 35(4):C369–C393, 2013

Patrick E Farrell, David A Ham, Simon W Funke, and Marie E Rognes. Automated derivation of the adjoint of high-level transient finite element programs.SIAM Journal on Scientific Computing, 35(4):C369–C393, 2013

work page 2013
[17]

Unsupervised discovery of interpretable hyperelastic constitutive laws.Computer Methods in Applied Mechanics and Engineering, 381:113852, 2021

Moritz Flaschel, Siddhant Kumar, and Laura De Lorenzis. Unsupervised discovery of interpretable hyperelastic constitutive laws.Computer Methods in Applied Mechanics and Engineering, 381:113852, 2021

work page 2021
[18]

Overhang constraint for topology optimization of self-supported compliant mechanisms considering additive manufacturing.Computer-Aided De- sign, 109:33–48, 2019

Alain Garaigordobil, Rubén Ansola, Estrella Veguería, and Igor Fernandez. Overhang constraint for topology optimization of self-supported compliant mechanisms considering additive manufacturing.Computer-Aided De- sign, 109:33–48, 2019

work page 2019
[19]

Gmsh: A 3-d finite element mesh generator with built-in pre- and post-processing facilities.International journal for numerical methods in engineering, 79(11):1309–1331, 2009

Christophe Geuzaine and Jean-François Remacle. Gmsh: A 3-d finite element mesh generator with built-in pre- and post-processing facilities.International journal for numerical methods in engineering, 79(11):1309–1331, 2009

work page 2009
[20]

naca: Generate 4 and 5 digit naca airfoil sections

Dirk Gorissen. naca: Generate 4 and 5 digit naca airfoil sections. GitHub repository, 2026. URLhttps: //github.com/dgorissen/naca. Accessed: 2026-01-29

work page 2026
[21]

Self-supporting structure de- sign in additive manufacturing through explicit topology optimization.Computer Methods in Applied Mechanics and Engineering, 323:27–63, 2017

Xu Guo, Jianhua Zhou, Weisheng Zhang, Zongliang Du, Chang Liu, and Ying Liu. Self-supporting structure de- sign in additive manufacturing through explicit topology optimization.Computer Methods in Applied Mechanics and Engineering, 323:27–63, 2017

work page 2017
[22]

Programmed shape-morphing into complex target shapes using architected dielectric elastomer actuators.Science advances, 8(28):eabn9198, 2022

Ehsan Hajiesmaili, Natalie M Larson, Jennifer A Lewis, and David R Clarke. Programmed shape-morphing into complex target shapes using architected dielectric elastomer actuators.Science advances, 8(28):eabn9198, 2022

work page 2022
[23]

Printing ferromagnetic domains for untethered fast-transforming soft materials.Nature, 558(7709):274–279, 2018

Yoonho Kim, Hyunwoo Yuk, Ruike Zhao, Shawn A Chester, and Xuanhe Zhao. Printing ferromagnetic domains for untethered fast-transforming soft materials.Nature, 558(7709):274–279, 2018

work page 2018
[24]

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
[25]

Polyconvex anisotropic hyperelasticity with neural networks.Journal of the Mechanics and Physics of Solids, 159:104703, 2022

Dominik K Klein, Mauricio Fernández, Robert J Martin, Patrizio Neff, and Oliver Weeger. Polyconvex anisotropic hyperelasticity with neural networks.Journal of the Mechanics and Physics of Solids, 159:104703, 2022

work page 2022
[26]

Liquid crystal elastomer lattices with thermally programmable deformation via multi-material 3d printing.Advanced Materials, 36(34):2310743, 2024

Arda Kotikian, Audrey A Watkins, Giovanni Bordiga, Andrew Spielberg, Zoey S Davidson, Katia Bertoldi, and Jennifer A Lewis. Liquid crystal elastomer lattices with thermally programmable deformation via multi-material 3d printing.Advanced Materials, 36(34):2310743, 2024. 21 Constitutive PriorA PREPRINT

work page 2024
[27]

Advances in 4d printing: materials and applications.Advanced Functional Materials, 29(2): 1805290, 2019

Xiao Kuang, Devin J Roach, Jiangtao Wu, Craig M Hamel, Zhen Ding, Tiejun Wang, Martin L Dunn, and Hang Jerry Qi. Advances in 4d printing: materials and applications.Advanced Functional Materials, 29(2): 1805290, 2019

work page 2019
[28]

Topology optimization for concurrent design of structures with multi-patch microstructures by level sets.Computer Methods in Applied Mechanics and Engineering, 331: 536–561, 2018

Hao Li, Zhen Luo, Liang Gao, and Qinghua Qin. Topology optimization for concurrent design of structures with multi-patch microstructures by level sets.Computer Methods in Applied Mechanics and Engineering, 331: 536–561, 2018

work page 2018
[29]

Neural networks meet hyperelasticity: A guide to enforcing physics.Journal of the Mechanics and Physics of Solids, 179:105363, 2023

Lennart Linden, Dominik K Klein, Karl A Kalina, Jörg Brummund, Oliver Weeger, and Markus Kästner. Neural networks meet hyperelasticity: A guide to enforcing physics.Journal of the Mechanics and Physics of Solids, 179:105363, 2023

work page 2023
[30]

A new family of constitutive artificial neural networks towards automated model discovery.Computer Methods in Applied Mechanics and Engineering, 403:115731, 2023

Kevin Linka and Ellen Kuhl. A new family of constitutive artificial neural networks towards automated model discovery.Computer Methods in Applied Mechanics and Engineering, 403:115731, 2023

work page 2023
[31]

Self-consistent clustering analysis: an efficient multi-scale scheme for inelastic heterogeneous materials.Computer methods in applied mechanics and engineering, 306:319–341, 2016

Zeliang Liu, MA Bessa, and Wing Kam Liu. Self-consistent clustering analysis: an efficient multi-scale scheme for inelastic heterogeneous materials.Computer methods in applied mechanics and engineering, 306:319–341, 2016

work page 2016
[32]

Springer, 2006

Jorge Nocedal and Stephen J Wright.Numerical optimization. Springer, 2006

work page 2006
[33]

Topology optimization for architected materials design.Annual Review of Materials Research, 46(1):211–233, 2016

Mikhail Osanov and James K Guest. Topology optimization for architected materials design.Annual Review of Materials Research, 46(1):211–233, 2016

work page 2016
[34]

Deepsdf: Learning continuous signed distance functions for shape representation

Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning continuous signed distance functions for shape representation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 165–174, 2019

work page 2019
[35]

Kirigami skins make a simple soft actuator crawl.Science Robotics, 3(15):eaar7555, 2018

Ahmad Rafsanjani, Yuerou Zhang, Bangyuan Liu, Shmuel M Rubinstein, and Katia Bertoldi. Kirigami skins make a simple soft actuator crawl.Science Robotics, 3(15):eaar7555, 2018

work page 2018
[36]

Adeildo S Ramos Jr and Glaucio H Paulino. Convex topology optimization for hyperelastic trusses based on the ground-structure approach: Convex topology optimization for hyperelastic trusses based.Structural and Multidisciplinary Optimization, 51(2):287–304, 2015

work page 2015
[37]

Neural topology optimization: the good, the bad, and the ugly: Sm sanu et al.Structural and Multidisciplinary Optimization, 68(10):213, 2025

Suryanarayanan Manoj Sanu, Alejandro M Aragon, and Miguel A Bessa. Neural topology optimization: the good, the bad, and the ugly: Sm sanu et al.Structural and Multidisciplinary Optimization, 68(10):213, 2025

work page 2025
[38]

Jax md: a framework for differentiable physics.Advances in Neural Information Processing Systems, 33:11428–11441, 2020

Samuel Schoenholz and Ekin Dogus Cubuk. Jax md: a framework for differentiable physics.Advances in Neural Information Processing Systems, 33:11428–11441, 2020

work page 2020
[39]

Invariant formulation of hyperelastic transverse isotropy based on polyconvex free energy functions.International journal of solids and structures, 40(2):401–445, 2003

Jörg Schröder and Patrizio Neff. Invariant formulation of hyperelastic transverse isotropy based on polyconvex free energy functions.International journal of solids and structures, 40(2):401–445, 2003

work page 2003
[40]

Materials with prescribed constitutive parameters: an inverse homogenization problem.Interna- tional Journal of Solids and Structures, 31(17):2313–2329, 1994

Ole Sigmund. Materials with prescribed constitutive parameters: an inverse homogenization problem.Interna- tional Journal of Solids and Structures, 31(17):2313–2329, 1994

work page 1994
[41]

Tailoring materials with prescribed elastic properties.Mechanics of materials, 20(4):351–368, 1995

Ole Sigmund. Tailoring materials with prescribed elastic properties.Mechanics of materials, 20(4):351–368, 1995

work page 1995
[42]

Implicit neural representations with periodic activation functions.Advances in neural information processing systems, 33:7462– 7473, 2020

Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions.Advances in neural information processing systems, 33:7462– 7473, 2020

work page 2020
[43]

Concepts for structurally robust materials that combine low thermal expansion with high stiffness.Journal of the Mechanics and Physics of Solids, 55(9):1803–1822, 2007

Craig A Steeves, Sergio L dos Santos e Lucato, Ming He, Emilio Antinucci, John W Hutchinson, and Anthony G Evans. Concepts for structurally robust materials that combine low thermal expansion with high stiffness.Journal of the Mechanics and Physics of Solids, 55(9):1803–1822, 2007

work page 2007
[44]

The method of moving asymptotes—a new method for structural optimization.International journal for numerical methods in engineering, 24(2):359–373, 1987

Krister Svanberg. The method of moving asymptotes—a new method for structural optimization.International journal for numerical methods in engineering, 24(2):359–373, 1987

work page 1987
[45]

Rasterizing algebraic curves and surfaces.IEEE Computer graphics and applications, 14(2): 14–23, 1994

Gabriel Taubin. Rasterizing algebraic curves and surfaces.IEEE Computer graphics and applications, 14(2): 14–23, 1994

work page 1994
[46]

Deep image prior

Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Deep image prior. InProceedings of the IEEE confer- ence on computer vision and pattern recognition, pages 9446–9454, 2018

work page 2018
[47]

Consistent machine learning for topology optimization with microstructure-dependent neural network material models.Journal of the Mechanics and Physics of Solids, 196:106015, 2025

Harikrishnan Vijayakumaran, Jonathan B Russ, Glaucio H Paulino, and Miguel A Bessa. Consistent machine learning for topology optimization with microstructure-dependent neural network material models.Journal of the Mechanics and Physics of Solids, 196:106015, 2025

work page 2025
[48]

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. 22 Constitutive PriorA PREPRINT

work page 2020
[49]

Inverse design of magneto-active metasurfaces and robots: Theory, computation, and experimental validation.Computer Methods in Applied Mechanics and Engineering, 413:116065, 2023

Chao Wang, Zhi Zhao, and Xiaojia Shelly Zhang. Inverse design of magneto-active metasurfaces and robots: Theory, computation, and experimental validation.Computer Methods in Applied Mechanics and Engineering, 413:116065, 2023

work page 2023
[50]

Autonomous codesign and fabrication of multistimuli-responsive material systems.Science Advances, 11(37):eadx4409, 2025

Liwei Wang, Alexander L Evenchik, Jared M Yang, Ryan L Truby, and Wei Chen. Autonomous codesign and fabrication of multistimuli-responsive material systems.Science Advances, 11(37):eadx4409, 2025

work page 2025
[51]

Jax- fem: A differentiable gpu-accelerated 3d finite element solver for automatic inverse design and mechanistic data science.Computer Physics Communications, 291:108802, 2023

Tianju Xue, Shuheng Liao, Zhengtao Gan, Chanwook Park, Xiaoyu Xie, Wing Kam Liu, and Jian 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

work page 2023
[52]

Ultralight, ultrastiff mechanical metamaterials.Science, 344 (6190):1373–1377, 2014

Xiaoyu Zheng, Howon Lee, Todd H Weisgraber, Maxim Shusteff, Joshua DeOtte, Eric B Duoss, Joshua D Kuntz, Monika M Biener, Qi Ge, Julie A Jackson, et al. Ultralight, ultrastiff mechanical metamaterials.Science, 344 (6190):1373–1377, 2014

work page 2014
[53]

Learning with local and global consistency.Advances in neural information processing systems, 16, 2003

Dengyong Zhou, Olivier Bousquet, Thomas Lal, Jason Weston, and Bernhard Schölkopf. Learning with local and global consistency.Advances in neural information processing systems, 16, 2003. Appendix A Partially Input-Convex Neural Networks In this section, we briefly review the partially input-convex neural networks (PICNNs) architecture introduced in [1]. T...

work page 2003