arxiv: 2604.03522 · v1 · submitted 2026-04-04 · 💻 cs.CE

Recognition: no theorem link

Physics-Informed Transformer for Real-Time High-Fidelity Topology Optimization

Aaron Lutheran , Srijan Das , Alireza Tabarraei

Authors on Pith no claims yet

Pith reviewed 2026-05-13 17:15 UTC · model grok-4.3

classification 💻 cs.CE

keywords topology optimizationphysics-informed transformerreal-time structural designoperator learningfinite element analysismachine learning in engineeringstructural connectivity

0 comments

The pith

A physics-informed transformer learns a direct non-iterative mapping from boundary conditions and loads to optimized structural topologies.

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

The paper reframes topology optimization as an operator-learning task solved by a single forward pass through a transformer instead of repeated finite-element iterations. Traditional methods are limited by their iterative nature, which prevents real-time use and large-scale exploration of designs. The model encodes global parameters via conditioning tokens and physical fields as patch tokens, then applies auxiliary losses to enforce volume, load adherence, and connectivity constraints. Global self-attention captures long-range mechanical interactions that local convolutional models miss. If the approach holds, high-fidelity, manufacturable topologies become available instantly across static and dynamic loading cases.

Core claim

The proposed physics-informed transformer directly maps boundary conditions, loading configurations, and derived stress and strain energy fields to optimized topologies, achieving higher fidelity than diffusion models while using only one forward pass and auxiliary differentiable losses to satisfy volume, load, and connectivity requirements.

What carries the argument

Vision Transformer with conditioning tokens for global parameters and spatially distributed physical-field patch tokens, plus auxiliary losses enforcing physical and manufacturability constraints.

If this is right

Topology optimization becomes feasible as a real-time operation without iterative analysis.
Large-scale design exploration and dynamic loading cases are enabled through frequency-domain encoding and transfer learning.
Single-pass inference removes the computational bottleneck of repeated finite-element solves.
Structural connectivity and load adherence are maintained directly in the output without separate post-processing.

Where Pith is reading between the lines

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

The model could support interactive engineering tools in which loads are adjusted live and updated topologies appear immediately.
Patch-based encoding may scale to three-dimensional problems if training data and memory permit.
Routine optimization tasks could shift away from dedicated finite-element packages toward learned operators.

Load-bearing premise

The learned non-iterative mapping generalizes accurately to unseen boundary conditions, loading configurations, and problem sizes while the auxiliary losses enforce constraints without degrading solution quality or introducing artifacts.

What would settle it

Verification on a held-out set of novel boundary conditions showing that the generated topology violates equilibrium or connectivity when re-analyzed with a conventional finite-element solver.

Figures

Figures reproduced from arXiv: 2604.03522 by Aaron Lutheran, Alireza Tabarraei, Srijan Das.

**Figure 1.** Figure 1: Processing for the inputs of the structural optimization problem for a given problem with boundary conditions [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

**Figure 2.** Figure 2: Transformer block architecture, which maps a set of input [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Full proposed transformer architecture, including a separate treatment of the loading conditions and volume fraction from the stress and [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 4.** Figure 4: Plot of the time sequence (a) and FFT amplitude transfor [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Validation samples from the ViT Small model. From left to [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

**Figure 6.** Figure 6: Stress and strain over an optimized sample after loading. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

**Figure 7.** Figure 7: Peak stress and strain scatter plots for model topology and [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗

**Figure 8.** Figure 8: Select samples from the post-processing results for the ViT [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗

**Figure 9.** Figure 9: Within the training range (0.3–0.5), the model exhibits strong performance, with volume fraction errors remaining within 1–3% and compliance errors generally below 8%. This indicates that the model accurately tracks changes in prescribed material usage and produces structurally consistent designs when operating within the distribution it has learned. Outside the training range, however, systematic devi… view at source ↗

**Figure 9.** Figure 9: Compliance and volume fraction error results for the static [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

**Figure 10.** Figure 10: Example topology with an out of distribution load condi [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗

**Figure 11.** Figure 11: Validation samples from the ViT Small model finetuned [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗

read the original abstract

Topology optimization is used for the design of high-performance structures but remains fundamentally limited by its iterative nature, requiring repeated finite element analyses that prevent real-time deployment and large-scale design exploration. In this work, we introduce a physics-informed transformer architecture that directly learns a non-iterative mapping from boundary conditions, loading configurations, and derived physical fields to optimized structural topologies. By leveraging global self-attention, the proposed model captures long-range mechanical interactions that govern structural response, overcoming the locality limitations of convolutional architectures. A conditioning-token mechanism embeds global problem parameters, while spatially distributed stress and strain energy fields are encoded as patch tokens within a Vision Transformer framework. To ensure physical realism and manufacturability, we incorporate auxiliary loss functions that enforce volume constraints, load adherence, and structural connectivity through a differentiable formulation. The framework is further extended to dynamic loading scenarios using frequency-domain encoding and transfer learning, enabling efficient generalization from static to time-dependent problems. Comprehensive benchmarking demonstrates that the proposed model achieves fidelity beyond that of diffusion models, while requiring only a single forward pass, thereby eliminating iterative inference entirely. This establishes topology optimization as a real-time operator-learning problem, enabling high-fidelity structural design with significant reductions in computational cost.

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

This paper's conditioned Vision Transformer for single-pass topology optimization has a coherent architecture but the performance claims rest on missing quantitative evidence.

read the letter

The main thing here is an attempt to turn topology optimization into a learned single-forward-pass operator using a Vision Transformer that takes boundary conditions, loads, and derived stress/strain fields as input. The architecture adds global conditioning tokens and patch tokens for the physical fields, then layers on differentiable auxiliary losses for volume, load adherence, and connectivity. It also sketches an extension to dynamic problems via frequency encoding and transfer learning from static cases. That combination is the concrete new piece relative to prior physics-informed networks or plain diffusion approaches for the same task. The framing around long-range interactions via self-attention is reasonable and directly addresses a known limitation of convolutional baselines in mechanics problems. The loss formulation is a practical way to keep outputs manufacturable without separate post-processing steps. The soft spots sit in the validation. The abstract asserts higher fidelity than diffusion models and complete removal of iterative inference, yet supplies no error metrics, no ablation on the auxiliary terms, no scaling behavior with domain size, and no explicit checks on unseen load patterns or boundary conditions. If the auxiliary losses trade off against compliance or introduce artifacts outside the training distribution, the claimed real-time advantage does not hold. The stress-test concern about generalization therefore still stands on the evidence given. This is aimed at computational mechanics groups already using topology optimization who want faster design loops, or at scientific ML researchers looking for operator-learning examples in engineering. A reader with background in both areas can pull the tokenization and loss ideas even if the results section needs more work. I would bring it to a reading group to discuss the conditioning mechanism and loss weighting choices. It deserves peer review because the problem is genuine and the architecture is a coherent attempt, though any referee will almost certainly ask for the missing quantitative benchmarks and extrapolation tests.

Referee Report

3 major / 2 minor

Summary. The paper introduces a physics-informed transformer that learns a direct, non-iterative mapping from boundary conditions, loads, and derived stress/strain fields to optimized topologies. It uses global self-attention, conditioning tokens, patch-encoded physical fields in a ViT-style backbone, and auxiliary differentiable losses for volume, load adherence, and connectivity. The method is extended to dynamic problems via frequency-domain encoding and transfer learning, with the central claim that it delivers higher fidelity than diffusion models in a single forward pass.

Significance. If the performance claims are substantiated, the work would be significant for computational mechanics: it reframes topology optimization as a real-time operator-learning task, potentially enabling interactive design loops and large-scale exploration that are currently blocked by iterative FEA. The combination of global attention with physics-informed auxiliary losses offers a concrete route to enforce manufacturability without post-processing, and the transfer-learning extension to dynamics broadens applicability.

major comments (3)

[Abstract and §4] Abstract and §4 (Benchmarking): the assertion that the model 'achieves fidelity beyond that of diffusion models' is unsupported by any quantitative metrics (compliance error, volume deviation, structural similarity index, or runtime comparisons) or error bars; without these numbers the headline superiority claim cannot be evaluated.
[§3.3 and §4.2] §3.3 (Auxiliary Losses) and §4.2 (Generalization Tests): no ablation is presented that isolates the effect of the volume/load/connectivity losses versus a pure data-driven baseline, nor are extrapolation results shown for domain sizes, load patterns, or boundary conditions outside the training distribution; this directly bears on whether the single-pass advantage survives on unseen instances.
[§4.1] §4.1 (Training Details): the manuscript provides no dataset size, hyperparameter values for the loss weights, number of training epochs, or validation curves, leaving open the possibility that reported performance reflects overfitting or post-hoc selection rather than robust generalization.

minor comments (2)

[§3.2] §3.2: the precise definition and dimensionality of the global conditioning token relative to the patch tokens is not stated explicitly, which complicates reproduction of the architecture.
[Figure 3] Figure 3 caption: axis labels and color scales for the stress-field patches are missing, reducing clarity of the input encoding.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed review. The comments highlight important areas for clarification and strengthening of the manuscript. We address each major comment below and will make the corresponding revisions.

read point-by-point responses

Referee: [Abstract and §4] Abstract and §4 (Benchmarking): the assertion that the model 'achieves fidelity beyond that of diffusion models' is unsupported by any quantitative metrics (compliance error, volume deviation, structural similarity index, or runtime comparisons) or error bars; without these numbers the headline superiority claim cannot be evaluated.

Authors: We agree that the superiority claim requires quantitative support. In the revised manuscript we will add a dedicated comparison table in §4 reporting compliance error, volume deviation, SSIM, and runtime metrics versus diffusion models, together with mean values and standard deviations computed over the test set. revision: yes
Referee: [§3.3 and §4.2] §3.3 (Auxiliary Losses) and §4.2 (Generalization Tests): no ablation is presented that isolates the effect of the volume/load/connectivity losses versus a pure data-driven baseline, nor are extrapolation results shown for domain sizes, load patterns, or boundary conditions outside the training distribution; this directly bears on whether the single-pass advantage survives on unseen instances.

Authors: We acknowledge that isolating the contribution of the auxiliary losses and testing extrapolation are necessary. We will add an ablation study in §4.2 that compares the full model against a pure data-driven baseline (identical architecture without the physics losses). We will also include new experiments on out-of-distribution cases (larger domains, unseen load patterns, and altered boundary conditions) with quantitative performance metrics. revision: yes
Referee: [§4.1] §4.1 (Training Details): the manuscript provides no dataset size, hyperparameter values for the loss weights, number of training epochs, or validation curves, leaving open the possibility that reported performance reflects overfitting or post-hoc selection rather than robust generalization.

Authors: We will expand §4.1 to report the exact training dataset size, the numerical values of all loss-weight hyperparameters, the number of training epochs, and will add validation-loss curves (either as a new figure or in the supplementary material) to document convergence behavior. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents an empirical machine-learning approach: a transformer is trained to learn a direct mapping from boundary conditions, loads, and physical fields to topologies, using auxiliary losses for constraints. No mathematical derivation chain exists that reduces any claimed prediction or result to its inputs by construction. There are no self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations that force the central claims. Benchmarking results are presented as evidence rather than tautological outputs. The architecture and losses are standard empirical techniques without circular reduction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The framework assumes standard supervised learning on data generated by conventional topology optimizers plus the existence of a learnable mapping from given fields to optimal topologies; no new physical entities are introduced.

free parameters (2)

loss weighting coefficients
Relative weights on fidelity, volume, load-adherence, and connectivity terms must be chosen or tuned to balance the objectives.
transformer hyperparameters
Number of layers, attention heads, patch size, and embedding dimension are selected to fit the training regime.

axioms (2)

domain assumption Precomputed stress and strain-energy fields plus boundary conditions contain sufficient information to determine the optimal topology.
Core premise enabling the direct mapping learned by the model.
domain assumption Differentiable formulations of volume, load adherence, and connectivity constraints can be enforced via auxiliary losses without compromising solution fidelity.
Assumption required for the physics-informed training to produce manufacturable designs.

pith-pipeline@v0.9.0 · 5513 in / 1329 out tokens · 42671 ms · 2026-05-13T17:15:05.334102+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

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

Diffusion Transformers with Hybrid Conditioning for Structural Optimization
cs.CE 2026-05 unverdicted novelty 6.0

A hybrid-conditioned diffusion transformer generates 2D topologies matching SIMP solutions within 1% compliance error using only five denoising steps.

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages · cited by 1 Pith paper · 3 internal anchors

[1]

M. P. Bendsøe, O. Sigmund, Topology Opti- mization: Theory, Methods, and Applications, Springer, 2003

work page 2003
[2]

Sigmund, K

O. Sigmund, K. Maute, Topology optimization approaches: A comparative review, Structural and multidisciplinary optimization 48 (6) (2013) 1031–1055

work page 2013
[3]

J. D. Deaton, R. V . Grandhi, A survey of structural and multidisciplinary continuum topology opti- mization: post 2000, Structural and Multidisci- plinary Optimization 49 (1) (2014) 1–38

work page 2000
[4]

J. Zhu, W. Zhang, A review of topology opti- mization for additive manufacturing: Status and challenges, Chinese Journal of Aeronautics 34 (1) (2021) 91–110

work page 2021
[5]

L. Meng, W. Zhang, D. Quan, et al., From topol- ogy optimization design to additive manufactur- ing: Today’s success and tomorrow’s roadmap, Archives of Computational Methods in Engineer- ing 27 (2019) 805–830

work page 2019
[6]

M. P. Bendsøe, Optimal shape design as a material distribution problem, Structural Optimization 1 (4) (1989) 193–202

work page 1989
[7]

Yang, C.-J

R.-J. Yang, C.-J. Chen, Stress-based topology op- timization, Structural optimization 12 (2) (1996) 98–105

work page 1996
[8]

Allaire, F

G. Allaire, F. Jouve, A.-M. Toader, Structural opti- mization with freefem++, Structural and Multidis- ciplinary Optimization 28 (2-3) (2004) 187–213. 17

work page 2004
[9]

M. Y . Wang, X. Wang, D. Guo, Level set method for structural topology optimization, Computer Methods in Applied Mechanics and Engineering 192 (1–2) (2003) 227–246. doi:10.1016/S0045- 7825(02)00559-5

work page doi:10.1016/s0045- 2003
[10]

Svanberg, The method of moving asymp- totes—a new method for structural optimization, International Journal for Numerical Methods in Engineering 24 (2) (1987) 359–373

K. Svanberg, The method of moving asymp- totes—a new method for structural optimization, International Journal for Numerical Methods in Engineering 24 (2) (1987) 359–373

work page 1987
[11]

Mukherjee, D

S. Mukherjee, D. Lu, B. Raghavan, P. Breitkopf, S. Dutta, M. Xiao, W. Zhang, Accelerating large- scale topology optimization: state-of-the-art and challenges, Archives of Computational Methods in Engineering 28 (7) (2021) 4549–4571

work page 2021
[12]

M. M. Behzadi, H. T. Ilie¸ s, Real-time topol- ogy optimization in 3d via deep transfer learning, Computer-Aided Design 135 (2021) 103014

work page 2021
[13]

Borrvall, J

T. Borrvall, J. Petersson, Large-scale topology op- timization in 3d using parallel computing, Com- puter methods in applied mechanics and engineer- ing 190 (46-47) (2001) 6201–6229

work page 2001
[14]

J. K. Guest, L. C. Smith Genut, Reducing di- mensionality in topology optimization using adap- tive design variable fields, International journal for numerical methods in engineering 81 (8) (2010) 1019–1045

work page 2010
[15]

Filomeno Coelho, P

R. Filomeno Coelho, P. Breitkopf, C. Knopf- Lenoir, Model reduction for multidisciplinary optimization-application to a 2d wing, Structural and multidisciplinary optimization 37 (1) (2008) 29–48

work page 2008
[16]

X. Guo, W. Zhang, W. Zhong, Doing topol- ogy optimization explicitly and geometrically—a new moving morphable components based frame- work, Journal of Applied Mechanics 81 (8) (2014) 081009

work page 2014
[17]

Chandrasekhar, K

A. Chandrasekhar, K. Suresh, TOuNN: topol- ogy optimization using neural networks, Structural and Multidisciplinary Optimization 63 (3) (2021) 1135–1149

work page 2021
[18]

Jeong, J

H. Jeong, J. Bai, C. P. Batuwatta-Gamage, C. Rathnayaka, Y . Zhou, Y . Gu, A physics- informed neural network-based topology opti- mization (PINNTO) framework for structural op- timization, Engineering Structures 278 (2023) 115484

work page 2023
[19]

M. I. R. Shishir, A. Tabarraei, Multi–materials topology optimization using deep neural network for coupled thermo–mechanical problems, Com- puters & Structures 291 (2024) 107218

work page 2024
[20]

Tabarraei, S

A. Tabarraei, S. A. Bhuiyan, Graph neural network-based topology optimization for self- supporting structures in additive manufacturing, arXiv preprint arXiv:2508.19169 (2025)

work page arXiv 2025
[21]

S. Oh, Y . Jung, S. Kim, I. Lee, N. Kang, Deep gen- erative design: integration of topology optimiza- tion and generative models, Journal of Mechanical Design 141 (11) (2019) 111405

work page 2019
[22]

F. V . Senhora, H. Chi, Y . Zhang, L. Mirabella, T. L. E. Tang, G. H. Paulino, Machine learning for topology optimization: Physics-based learning through an independent training strategy, Com- puter Methods in Applied Mechanics and Engi- neering 398 (2022) 115116

work page 2022
[23]

Zhang, L

W. Zhang, L. Su, X. Wang, Real-time topology optimization based on multi-scale convolutional attention mechanism, Engineering Optimization 57 (6) (2025) 1506–1525

work page 2025
[24]

A. H. Nobari, G. Giannone, L. Regenwet- ter, F. Ahmed, Nito: Neural implicit fields for resolution-free topology optimization, arXiv preprint arXiv:2402.05073 (2024)

work page arXiv 2024
[25]

Tabarraei, Variational quantum latent encoding for topology optimization, Engineering with Com- puters 41 (6) (2025) 4549–4573

A. Tabarraei, Variational quantum latent encoding for topology optimization, Engineering with Com- puters 41 (6) (2025) 4549–4573

work page 2025
[26]

Z. Nie, T. Lin, H. Jiang, L. B. Kara, Topologygan: Topology optimization using generative adversar- ial networks based on physical fields over the ini- tial domain, Journal of Mechanical Design 143 (3) (2021) 031715

work page 2021
[27]

F. Mazé, F. Ahmed, Diffusion models beat GANs on topology optimization (2023)

work page 2023
[28]

Giannone, F

G. Giannone, F. Ahmed, Diffusing the optimal topology: A generative optimization approach (2023)

work page 2023
[29]

Zhang, G

W. Zhang, G. Zhao, L. Su, Research on multi- stage topology optimization method based on la- tent diffusion model, Advanced Engineering Infor- matics 63 (2025) 102966. 18

work page 2025
[30]

Lutheran, S

A. Lutheran, S. Das, A. Tabarraei, Latent space diffusion for topology optimization, arXiv preprint arXiv:2508.05624 (2025)

work page arXiv 2025
[31]

Attention Is All You Need

A. Vaswani, N. Shazeer, N. Parmar, J. Uszko- reit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, Attention is all you need (2023). arXiv:1706.03762

work page internal anchor Pith review Pith/arXiv arXiv 2023
[32]

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weis- senborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, et al., An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv:2010.11929 (2020)

work page internal anchor Pith review Pith/arXiv arXiv 2010
[33]

X. Zhai, A. Kolesnikov, N. Houlsby, L. Beyer, Scaling vision transformers (2022). arXiv:2106.04560

work page arXiv 2022
[34]

S. Cao, H. Dong, N. Goodman, Choose a trans- former: Fourier or galerkin, in: Advances in Neu- ral Information Processing Systems, V ol. 34, 2021, pp. 24961–24973

work page 2021
[35]

Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anandkumar, Fourier neural operator for parametric partial dif- ferential equations, International Conference on Learning Representations (ICLR) (2021)

work page 2021
[36]

J. Lee, M. Cho, Efficient design optimization strat- egy for structural dynamic systems using a re- duced basis method combined with an equivalent static load, Structural and Multidisciplinary Opti- mization 58 (4) (2018) 1489–1504

work page 2018
[37]

J. Rong, Y . Xie, X. Yang, Q. Liang, Topol- ogy optimization of structures under dynamic re- sponse constraints, Journal of Sound and Vibration 234 (2) (2000) 177–189

work page 2000
[38]

J. Zhao, C. Wang, Dynamic response topology op- timization in the time domain using model reduc- tion method, Structural and Multidisciplinary Op- timization 53 (1) (2016) 101–114

work page 2016
[39]

Lee, G.-J

H.-A. Lee, G.-J. Park, Nonlinear dynamic re- sponse topology optimization using the equivalent static loads method, Computer Methods in Ap- plied Mechanics and Engineering 283 (2015) 956– 970

work page 2015
[40]

K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition (2015). arXiv:1512.03385

work page internal anchor Pith review Pith/arXiv arXiv 2015
[41]

K. He, X. Chen, S. Xie, Y . Li, P. Dollár, R. Gir- shick, Masked Autoencoders Are Scalable Vi- sion Learners, arXiv:2111.06377 [cs] (Dec. 2021). doi:10.48550/arXiv.2111.06377. URLhttp://arxiv.org/abs/2111.06377

work page doi:10.48550/arxiv.2111.06377 2021
[42]

Q. Q. Liang, G. P. Steven, A performance-based optimization method for topology design of con- tinuum structures with mean compliance con- straints, Computer methods in applied mechanics and engineering 191 (13-14) (2002) 1471–1489

work page 2002
[43]

Andreassen, A

E. Andreassen, A. Clausen, M. Schevenels, B. S. Lazarov, O. Sigmund, Efficient topology optimiza- tion in matlab using 88 lines of code, Structural and Multidisciplinary Optimization 43 (1) (2011) 1–16

work page 2011
[44]

J. Liu, A. T. Gaynor, S. Chen, Z. Kang, K. Suresh, A. Takezawa, L. Li, J. Kato, J. Tang, C. C. Wang, et al., Current and future trends in topology op- timization for additive manufacturing, Structural and multidisciplinary optimization 57 (6) (2018) 2457–2483

work page 2018
[45]

J. Yin, Z. Wen, S. Li, Y . Zhang, H. Wang, Dynami- cally configured physics-informed neural network in topology optimization applications, Computer Methods in Applied Mechanics and Engineering 426 (2024) 117004

work page 2024
[46]

Allaire, F

G. Allaire, F. Jouve, A.-M. Toader, Structural op- timization using sensitivity analysis and a level-set method, Journal of computational physics 194 (1) (2004) 363–393

work page 2004
[47]

Banga, H

S. Banga, H. Gehani, S. Bhilare, S. Patel, L. Kara, 3d topology optimization using convolutional neu- ral networks, arXiv preprint arXiv:1808.07440 (2018)

work page arXiv 2018
[48]

Kirsch, P

U. Kirsch, P. Y . Papalambros, Structural reanal- ysis for topological modifications–a unified ap- proach, Structural and multidisciplinary Optimiza- tion 21 (5) (2001) 333–344

work page 2001
[49]

E. Riba, D. Mishkin, D. Ponsa, E. Rublee, G. Bradski, Kornia: an open source differentiable computer vision library for pytorch, in: Proceed- ings of the IEEE/CVF Winter Conference on Ap- plications of Computer Vision, 2020, pp. 3674– 3683

work page 2020
[50]

J. Rade, A. Jignasu, E. Herron, A. Corpuz, B. Ganapathysubramanian, S. Sarkar, A. Balu, 19 A. Krishnamurthy, Deep learning-based 3d multi- grid topology optimization of manufacturable de- signs, Engineering Applications of Artificial Intel- ligence 126 (2023) 107033

work page 2023
[51]

H. Chi, Y . Zhang, T. L. E. Tang, L. Mirabella, L. Dalloro, L. Song, G. H. Paulino, Universal ma- chine learning for topology optimization, Com- puter Methods in Applied Mechanics and Engi- neering 375 (2021) 112739

work page 2021
[52]

S. Shin, D. Shin, N. Kang, Topology optimization via machine learning and deep learning: a review, Journal of Computational Design and Engineering 10 (4) (2023) 1736–1766

work page 2023
[53]

Preskill, Quantum computing in the nisq era and beyond, Quantum 2 (2018) 79

J. Preskill, Quantum computing in the nisq era and beyond, Quantum 2 (2018) 79

work page 2018
[54]

Cerezo, A

M. Cerezo, A. Arrasmith, R. Babbush, S. C. Ben- jamin, S. Endo, K. Fujii, J. R. McClean, K. Mi- tarai, X. Yuan, L. Cincio, P. J. Coles, Variational quantum algorithms, Nature Reviews Physics 3 (9) (2021) 625–644

work page 2021
[55]

Z. Ye, X. Qian, W. Pan, Quantum topology op- timization via quantum annealing, IEEE Transac- tions on Quantum Engineering 4 (2023) 1–15

work page 2023
[56]

X. Wang, Z. Wang, B. Ni, Mapping structural topology optimization problems to quantum an- nealing, Structural and Multidisciplinary Opti- mization 67 (5) (2024) 74

work page 2024
[57]

Ohzeki, S

M. Ohzeki, S. Taguchi, S. Tanaka, Quantum an- nealing and deep learning for topology optimiza- tion, in: 2022 International Conference on Quan- tum Computing and Engineering (QCE), IEEE, 2022, pp. 343–349

work page 2022
[58]

Okorie, et al., Topology optimization of an aerospace bracket: Numerical and experimen- tal verification, Applied Sciences 13 (24) (2023) 13218

O. Okorie, et al., Topology optimization of an aerospace bracket: Numerical and experimen- tal verification, Applied Sciences 13 (24) (2023) 13218. 20

work page 2023