Pith. sign in

REVIEW 2 major objections 1 minor 37 references

Explainable quantum neural networks for multi-material topology optimization

T0 review · 2 major / 1 minor · reviewed 2026-07-02 · grok-4.3

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

Pith's one-line read A 10-qubit quantum circuit trained only on 2D fixed-mesh topology histories assigns material types for out-of-distribution loads, refined meshes, and 3D voxel problems without retraining.

desk verdict XQNN trains a 10-qubit circuit on 2D fixed-mesh descriptors and claims it generalizes to new loads, finer meshes, and 3D without retraining. read the letter →

arxiv 2607.00438 v1 pith:RT47KLGF submitted 2026-07-01 cs.CE

Dahyun Joo , Naruethep Sukulthanasorn , Kenjiro Terada , Do-Nyun Kim This is my paper · ORCID
classification cs.CE
keywords explainablequantumneuralnetworkmulti-materialtopologyoptimizationstrainenergydescriptorsSobelboundaryqubit-wiseobservablesgeneralizationacrossdimensionsmaterialassignment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper presents XQNN, which converts topology optimization histories into element-wise strain energy, sensitivity, density, and Sobel boundary descriptors, encodes them into a 10-qubit circuit, and maps qubit-wise Z observables directly to material labels. Trained exclusively on two-dimensional cases at one mesh resolution, the network produces structural layouts and material assignments for new boundary conditions, higher-resolution grids, and full three-dimensional problems. A sympathetic reader would care because the method removes the need to regenerate training data or retrain when the problem scale or dimension changes. The approach also keeps individual qubit observables intact so they can be inspected for links to load paths and material interfaces.

What carries the argument

Ten-qubit quantum circuit that encodes element-wise mechanical descriptors and maps qubit-wise Z observables onto material type labels.

What would settle it

Apply the trained model to a 3D voxel cantilever beam under a new load case with a mesh twice as fine as the training resolution and check whether the predicted material distribution matches the result of a conventional multi-material optimizer to within a few percent.

Watch

Extended reading notes

Core claim

Trained only on two-dimensional topology optimization histories obtained with a fixed mesh resolution, XQNN can be generalized to handle out-of-distribution boundary/loading conditions, progressively refined high-resolution meshes, and voxel-wise three-dimensional problems without additional training. Intermediate solution histories are first converted into element-wise strain energy, sensitivity, density, and Sobel boundary descriptors. Then, they are encoded in a ten-qubit circuit and qubit-wise Z observables are mapped onto material type labels. It is important to preserve qubit-wise observables and add boundary information for improving the optimization accuracy, and certain observables

Load-bearing premise

The element-wise strain energy, sensitivity, density, and Sobel boundary descriptors taken from 2D fixed-mesh histories contain enough information for a 10-qubit circuit to label materials correctly on 3D and refined-mesh cases never seen in training.

Editorial extensions

If this is right

  • Preserving qubit-wise observables improves optimization accuracy over aggregated measurements.
  • Adding Sobel boundary descriptors is required for reliable interface placement.
  • Individual observables maintain consistent associations with load paths, material regions, and interfaces across problem types.
  • The same trained circuit produces usable material assignments on out-of-distribution 2D loads, refined 2D meshes, and full 3D voxel problems.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the transfer works, the same descriptor set may support quantum circuits on other dimension-changing mechanics problems such as heat conduction or fluid topology optimization.
  • The direct mapping from observables to physical quantities could let engineers audit or adjust the circuit outputs using conventional finite-element post-processing.
  • Generalization without retraining implies that the chosen descriptors already embed scale- and dimension-invariant mechanics relations that future hybrid quantum-classical solvers might exploit directly.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 / 1 minor

Summary. The manuscript proposes an explainable quantum neural network (XQNN) for multi-material topology optimization. Intermediate solution histories from 2D fixed-mesh topology optimization are converted into element-wise strain energy, sensitivity, density, and Sobel boundary descriptors. These are encoded into a ten-qubit quantum circuit, with qubit-wise Z observables mapped to material type labels. The key claim is that, when trained only on 2D fixed-resolution histories, the model generalizes to out-of-distribution boundary and loading conditions, progressively refined high-resolution meshes, and voxel-wise 3D problems without additional training. The work also highlights the importance of qubit-wise observables and boundary information for accuracy and their consistent links to mechanical features like load paths and interfaces.

Significance. If the generalization claims hold with supporting evidence, the work would be significant for showing that a fixed quantum circuit can transfer across problem dimensions and resolutions in topology optimization, potentially lowering retraining costs for 3D and multi-scale design tasks. The explicit mapping of observables to physical quantities offers a route to interpretable quantum models in computational mechanics.

major comments (2)
  1. [Abstract] Abstract: the generalization claim to out-of-distribution loads, refined meshes, and 3D voxel problems without retraining is stated but unsupported by any quantitative accuracy metrics, baseline comparisons, error bars, or details on circuit construction/training; this absence is load-bearing for the central result.
  2. [Method] Method (descriptor extraction and encoding): the assumption that 2D-derived element-wise strain energy, sensitivity, density, and Sobel descriptors suffice for 3D voxels is unexamined; the 3D extension of Sobel operators and the effect of altered connectivity on feature distributions are not addressed, directly undermining the no-retraining transfer claim.
minor comments (1)
  1. The abstract asserts that 'certain observables have consistent links to load paths' but provides no concrete examples, figures, or quantitative correlations to support the explainability claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The two major comments identify important gaps in the presentation of evidence and the justification for cross-dimensional transfer. We respond to each below and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the generalization claim to out-of-distribution loads, refined meshes, and 3D voxel problems without retraining is stated but unsupported by any quantitative accuracy metrics, baseline comparisons, error bars, or details on circuit construction/training; this absence is load-bearing for the central result.

    Authors: We agree that the abstract as written states the generalization claims without supporting numbers. The body of the manuscript contains quantitative results (accuracy on OOD loads, mesh-refinement sequences, and 3D voxels) together with circuit specifications and training details; however, these are not summarized in the abstract. We will revise the abstract to include representative accuracy figures, baseline comparisons, error bars, and a concise statement of the ten-qubit circuit and training protocol so that the central claim is quantitatively supported at the point of first reading. revision: yes

  2. Referee: [Method] Method (descriptor extraction and encoding): the assumption that 2D-derived element-wise strain energy, sensitivity, density, and Sobel descriptors suffice for 3D voxels is unexamined; the 3D extension of Sobel operators and the effect of altered connectivity on feature distributions are not addressed, directly undermining the no-retraining transfer claim.

    Authors: The manuscript demonstrates empirical transfer to 3D voxels without retraining, but we acknowledge that the justification for descriptor portability is limited to a brief statement. A detailed description of the 3D Sobel operator (3 imes3 imes3 kernels applied along each axis) and an analysis of how voxel connectivity alters the distribution of the four descriptors relative to the 2D training set are absent. We will add a short subsection that (i) specifies the 3D Sobel implementation, (ii) reports the observed shifts in descriptor statistics between 2D and 3D test cases, and (iii) discusses why the learned mapping remains effective despite these shifts. This addition will make the transfer claim more robust. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; generalization is an empirical claim about model transfer, not a definitional reduction.

full rationale

The paper trains a 10-qubit quantum circuit on element-wise descriptors (strain energy, sensitivity, density, Sobel) extracted from fixed-mesh 2D topology optimization histories, then maps Z observables to material labels. The central claim is that this trained model produces accurate labels on out-of-distribution loads, refined 2D meshes, and 3D voxels without retraining. This is presented as an empirical generalization result rather than a quantity defined by construction from the training data. No equations, fitted parameters renamed as predictions, or self-citation chains are invoked that would make the reported performance equivalent to the inputs. The architecture choices (qubit encoding, observable mapping, boundary descriptors) are independent design decisions whose transferability is tested, not presupposed. The derivation chain is therefore self-contained against external benchmarks.

Assumptions & free parameters 1 free parameters · 1 assumptions · 0 invented entities

The method rests on the domain assumption that classical optimization histories can be faithfully encoded into a fixed-size quantum circuit whose measurements yield material labels, plus the engineering choice of exactly ten qubits and the four listed descriptors; no new physical entities are postulated.

free parameters (1)
  • number of qubits = 10
    Fixed at ten for the circuit; chosen by the authors rather than derived.
assumptions (1)
  • domain assumption A quantum circuit with qubit-wise Z observables can map classical per-element descriptors to discrete material-type labels with useful accuracy.
    Invoked when the paper states that qubit-wise Z observables are mapped onto material type labels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Explainable quantum neural networks for multi-material topology optimization." pith.science (2026). https://pith.science/paper/RT47KLGF

@misc{pith2026260700438,
  author       = {Pith},
  title        = {Pith review of: Explainable quantum neural networks for multi-material topology optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RT47KLGF}},
  note         = {Machine review of arXiv:2607.00438}
}
abstract

We propose an explainable quantum neural network for multi-material topology optimization, XQNN, that determines both load-carrying structural layout and material type assignment for given boundary/loading conditions. Intermediate solution histories are first converted into element-wise strain energy, sensitivity, density, and Sobel boundary descriptors. Then, they are encoded in a ten-qubit circuit and qubit-wise $Z$ observables are mapped onto material type labels. Trained only on two-dimensional topology optimization histories obtained with a fixed mesh resolution, XQNN can be generalized to handle out-of-distribution boundary/loading conditions, progressively refined high-resolution meshes, and voxel-wise three-dimensional problems without additional training. We find that it is important to preserve qubit-wise observables and add boundary information for improving the optimization accuracy, and certain observables have consistent links to load paths, material type regions, and interfaces, demonstrating their usability as auditable mechanics-facing variables.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    Bendsøe, M. P. & Kikuchi, N. Generating optimal topologies in structural design using a homogenization method.Computer Methods in Applied Mechanics and Engineering71, 197–224 (1988)

    work page 1988

  2. [2]

    Bendsøe, M. P. & Sigmund, O.Topology Optimization: Theory, Methods, and Applications(Springer, Berlin, 2003)

    work page 2003

  3. [3]

    A 99 line topology optimization code written in MATLAB

    Sigmund, O. A 99 line topology optimization code written in MATLAB. Structural and Multidisciplinary Optimization21, 120–127 (2001)

    work page 2001

  4. [4]

    & Maute, K

    Sigmund, O. & Maute, K. Topology optimization approaches: A comparative review.Structural and Multidisciplinary Optimization48, 1031–1055 (2013)

    work page 2013

  5. [5]

    Deaton, J. D. & Grandhi, R. V. A survey of structural and multidisciplinary continuum topology optimization: Post 2000.Structural and Multidisciplinary Optimization49, 1–38 (2014)

    work page 2000

  6. [6]

    P., Maute, K., Langelaar, M

    van Dijk, N. P., Maute, K., Langelaar, M. & van Keulen, F. Level-set methods for structural topology optimization: A review.Structural and Multidisciplinary Optimization48, 437–472 (2013)

    work page 2013

  7. [7]

    & Toader, A.-M

    Allaire, G., Jouve, F. & Toader, A.-M. Structural optimization using sensitivity analysis and a level-set method.Journal of Computational Physics194, 363–393 (2004)

    work page 2004

  8. [8]

    & Kim, D.-N

    Jung, K. & Kim, D.-N. Density-based topology optimization using a deformable mesh.Computers & Structures316, 107879 (2025)

    work page 2025

Show all 37 references
  1. [9]

    & Mohseni, S

    Tavakoli, R. & Mohseni, S. M. Alternating active-phase algorithm for multi- material topology optimization problems: A 115-line MATLAB implementation. Structural and Multidisciplinary Optimization49, 621–642 (2014)

    2014

  2. [10]

    & Saitou, K

    Zuo, W. & Saitou, K. Multi-material topology optimization using ordered SIMP interpolation.Structural and Multidisciplinary Optimization55, 477–491 (2017)

    2017

  3. [11]

    Vermaak, N.et al.Material interface effects on the topology optimization of multi-phase structures using a level set method.Structural and Multidisciplinary Optimization50, 623–644 (2014)

    2014

  4. [12]

    & Zhang, N

    Wang, Y., Luo, Z., Kang, Z. & Zhang, N. A multi-material level set-based topol- ogy and shape optimization method.Computer Methods in Applied Mechanics and Engineering283, 1570–1586 (2015). 13

    2015

  5. [13]

    Mirzendehdel, A. M. & Suresh, K. A pareto-optimal approach to multimaterial topology optimization.Journal of Mechanical Design137, 101701 (2015)

    2015

  6. [14]

    & Kim, I

    Li, D. & Kim, I. Y. Multi-material topology optimization for practical lightweight design.Structural and Multidisciplinary Optimization58, 1081–1094 (2018)

    2018

  7. [15]

    & Yoon, G.-H

    Zheng, R., Yi, B., Peng, X. & Yoon, G.-H. An efficient code for the multi-material topology optimization of 2D/3D continuum structures written in Matlab.Applied Sciences14, 657 (2024)

    2024

  8. [16]

    V., Aage, N., Bærentzen, J

    Woldseth, R. V., Aage, N., Bærentzen, J. A. & Sigmund, O. On the use of arti- ficial neural networks in topology optimisation.Structural and Multidisciplinary Optimization65, 294 (2022)

    2022

  9. [17]

    & Jang, I.-H

    Yu, Y., Hur, T., Jung, J. & Jang, I.-H. Deep learning for determining a near- optimal topological design without any iteration.Structural and Multidisciplinary Optimization59, 787–799 (2019)

    2019

  10. [18]

    & Kara, L

    Nie, Z., Lin, T., Jiang, H. & Kara, L. B. TopologyGAN: Topology optimization using generative adversarial networks based on physical fields over the initial domain.Journal of Mechanical Design143, 031715 (2021)

    2021

  11. [19]

    & Dargush, G

    Bielecki, D., Patel, D., Rai, R. & Dargush, G. F. Multi-stage deep neural network accelerated topology optimization.Structural and Multidisciplinary Optimization 64, 3473–3487 (2021)

    2021

  12. [20]

    & Wang, H

    Xue, L., Liu, J., Wen, G. & Wang, H. Efficient, high-resolution topology optimiza- tion method based on convolutional neural networks.Frontiers of Mechanical Engineering16, 80–96 (2021)

    2021

  13. [21]

    K., Garland, A

    Brown, N. K., Garland, A. P., Fadel, G. M. & Li, G. Deep reinforcement learning for engineering design through topology optimization of elementally discretized design domains.Materials & Design218, 110672 (2022)

    2022

  14. [22]

    Gao, Y., Zhou, S. & Li, M. Structural topology optimization based on dif- fusion generative adversarial networks.Engineering Applications of Artificial Intelligence138, 109444 (2024)

    2024

  15. [23]

    & Kim, D.-N

    Lim, J., Jung, K., Jung, Y. & Kim, D.-N. Accelerating topology optimization using deep learning-based image super-resolution.Engineering Applications of Artificial Intelligence133, 108370 (2024)

    2024

  16. [24]

    Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nature Machine Intelligence1, 206–215 (2019)

    Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nature Machine Intelligence1, 206–215 (2019). 14

    2019

  17. [25]

    Roscher, R., Bohn, B., Duarte, M. F. & Garcke, J. Explainable machine learning for scientific insights and discoveries.IEEE Access8, 42200–42216 (2020)

    2020

  18. [26]

    B.et al.Explainable artificial intelligence (XAI): Concepts, tax- onomies, opportunities and challenges toward responsible AI.Information Fusion 58, 82–115 (2020)

    Arrieta, A. B.et al.Explainable artificial intelligence (XAI): Concepts, tax- onomies, opportunities and challenges toward responsible AI.Information Fusion 58, 82–115 (2020)

    2020

  19. [27]

    E.et al.Physics-informed machine learning.Nature Reviews Physics3, 422–440 (2021)

    Karniadakis, G. E.et al.Physics-informed machine learning.Nature Reviews Physics3, 422–440 (2021)

    2021

  20. [28]

    Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems30(2017)

    2017

  21. [29]

    R.et al.Grad-CAM: Visual explanations from deep networks via gradient-based localization.Proceedings of the IEEE International Conference on Computer Vision618–626 (2017)

    Selvaraju, R. R.et al.Grad-CAM: Visual explanations from deep networks via gradient-based localization.Proceedings of the IEEE International Conference on Computer Vision618–626 (2017)

    2017

  22. [30]

    & Terada, K

    Sukulthanasorn, N. & Terada, K. Quantum neural network-assisted topology optimization: Concept and implementation with parameterized quantum circuits. Computer Methods in Applied Mechanics and Engineering448, 118411 (2026)

    2026

  23. [31]

    Variational quantum latent encoding for topology optimization

    Tabarraei, A. Variational quantum latent encoding for topology optimization. Engineering with Computers41, 4549–4573 (2025)

    2025

  24. [32]

    & Coles, P

    Cerezo, M., Verdon, G., Huang, H.-Y., Cincio, L. & Coles, P. J. Challenges and opportunities in quantum machine learning.Nature Computational Science2, 567–576 (2022). 15 a b d e MMA Method (50x25) 𝑘𝑒𝑚 𝑘𝑒𝑚 + 5 𝑘𝑝𝑑 𝒯𝑡𝑟: Element-wise Random Sampling 1 2 3 Label mat0 mat1 mat2 0 ...

    2022

  25. [33]

    The method of moving asymptotes—a new method for structural optimization.International Journal for Numerical Methods in Engineering24, 359–373 (1987)

    Svanberg, K. The method of moving asymptotes—a new method for structural optimization.International Journal for Numerical Methods in Engineering24, 359–373 (1987)

    1987

  26. [34]

    & Yoon, G.-H

    Zheng, R., Yi, B., Peng, X. & Yoon, G.-H. An efficient code for the multi-material topology optimization of two-dimensional and three-dimensional continuum struc- tures written in Matlab.Applied Sciences14, 657 (2024)

    2024

  27. [35]

    https://doi.org/10.48550/arXiv.2405.08810

    Javadi-Abhari, A.et al.Quantum computing with Qiskit.arXiv:2405.08810 (2024). https://doi.org/10.48550/arXiv.2405.08810

    2024 doi arXiv

  28. [36]

    Kingma, D. P. & Ba, J. Adam: A method for stochastic optimization. In International Conference on Learning Representations(2015)

    2015

  29. [37]

    & Terada, K

    Sukulthanasorn, N. & Terada, K. Quantum neural network-assisted topology optimization: Concept and implementation with parameterized quantum circuits. Computer Methods in Applied Mechanics and Engineering448, 118411 (2026). 19

    2026

Pith tools

Reviewed July 2, 2026 · model on record in the stance chip above.