pith. sign in

arxiv: 2607.01812 · v1 · pith:QD4W5HAFnew · submitted 2026-07-02 · 💻 cs.CE

TO-Master: an LLM-agent framework for automated topology optimization

Haoju Lin , Wenchang Zhang , Weipeng Xu , Xiang Li , Tian Xu , Tianju Xue This is my paper

Pith reviewed 2026-07-03 03:25 UTC · model grok-4.3

classification 💻 cs.CE
keywords topology optimizationLLM agentfinite element methodautomated modelingconversational interfacestructural compliancethermal optimizationstress constraints
0
0 comments X

The pith

An LLM agent turns topology optimization into a conversational workflow that builds and runs finite-element models from natural language.

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

The paper introduces TO-Master, an LLM-agent system that accepts natural-language instructions plus optional meshes or images and then selects, configures, and executes the full sequence of finite-element topology optimization steps. The agent handles geometry preparation, meshing, boundary conditions, solver arguments, and post-processing while returning results, fields, and histories without any user code. A supporting ablation shows that explicit tool-usage rules, reasoning guidance, and few-shot examples are required for the agent to interpret ambiguous inputs reliably. If the framework works as described, non-experts gain access to standard and engineering TO problems through ordinary conversation while the underlying deterministic solvers remain unchanged.

Core claim

TO-Master converts finite-element-based topology optimization into a tool-orchestrated conversational workflow: from natural-language instructions and optional mesh, geometry, or image inputs the agent selects computational tools, constructs models, validates meshes and boundary conditions, and launches sensitivity-based optimization, returning optimized designs, field distributions, convergence histories, and interactive artifacts without user-written code.

What carries the argument

The LLM agent that selects and configures finite-element TO tools according to supplied tool-usage rules, internal reasoning guidance, and few-shot examples.

If this is right

  • The same agent can handle 2D/3D compliance minimization, thermal conduction, multiple load cases, and stress-constrained problems.
  • Users can supply generated meshes, uploaded files, or image-to-mesh conversions and still receive complete optimization runs.
  • Ablation results indicate that removing tool-usage rules or few-shot examples degrades performance on ambiguous inputs.
  • The workflow preserves numerical reliability by delegating all analysis and optimization to deterministic finite-element solvers.

Where Pith is reading between the lines

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

  • The same orchestration pattern could be applied to other simulation-driven design tasks that currently require manual FE setup.
  • Extending the agent to accept iterative refinement instructions after an initial run would allow conversational adjustment of designs.
  • Because the numerical core stays deterministic, the framework can be audited by checking the generated input files against the final results.

Load-bearing premise

When given rules, guidance, and examples the LLM agent can correctly interpret ambiguous natural-language instructions and choose the right finite-element tools and boundary conditions for the requested problem.

What would settle it

A test set of ambiguous user instructions for known TO benchmarks in which the agent produces models whose physics or boundary conditions do not match the intended problem statement.

read the original abstract

Topology optimization (TO) has become a mature computational design method, but using it still requires substantial manual effort in geometry preparation, mesh generation, boundary-condition assignment, solver setup, and postprocessing. This implementation barrier limits the use of TO outside expert workflows, even when differentiable finite element solvers are available. This work introduces TO-Master, a large language model (LLM) agent framework that turns finite-element-based TO into a conversational, tool-orchestrated workflow. From natural language instructions and optional mesh, geometry, or image inputs, the agent selects computational tools, constructs finite element TO models, checks meshes and boundary conditions, and launches sensitivity-based optimization with typed solver arguments. The framework supports generated and uploaded meshes, image-to-mesh conversion, 2D and 3D structural compliance minimization, thermal conduction, multiple load cases, stress-constrained optimization, and engineering geometries. Numerical experiments show that TO-Master can reproduce standard benchmark results and solve more complex engineering examples while returning optimized results, field distributions, convergence histories, and interactive artifacts without user-written code. An instruction ablation study further shows that tool-usage rules, internal reasoning guidance, and few-shot examples are critical for robust formulation under ambiguous user input. By combining LLM-agent orchestration with deterministic finite element and optimization tools, TO-Master removes the burden of trivial setup and routine model construction, lowers the modeling barrier of TO, and preserves a reliable numerical workflow. The TO-Master platform is available online at https://www.bohrium.com/en/apps/to-master.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 0 minor

Summary. The paper introduces TO-Master, an LLM-agent framework that automates finite-element-based topology optimization from natural-language instructions (with optional mesh, geometry, or image inputs). The agent selects and configures tools for mesh generation, boundary conditions, sensitivity-based optimization, and post-processing for 2D/3D structural compliance, thermal, multi-load, and stress-constrained problems. Numerical experiments claim reproduction of standard benchmarks and solution of complex engineering cases, returning optimized designs, fields, histories, and artifacts without user code; an ablation study tests the necessity of tool-usage rules, reasoning guidance, and few-shot examples.

Significance. If the reliability claims hold, the work demonstrates a practical LLM-orchestrated workflow that delegates numerical work to deterministic solvers, lowering the manual setup barrier for TO while preserving correctness. The ablation study supplies direct evidence on prompting components, which is a strength for a framework paper.

major comments (1)
  1. [Numerical experiments] Numerical experiments section: the claim that TO-Master 'reproduces standard benchmark results' is not supported by quantitative metrics (compliance values, iteration counts, error to reference solutions, or success/failure rates across trials). Without these, the evidence for correct tool selection and model construction under the weakest assumption remains incomplete.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation of minor revision. We address the single major comment below.

read point-by-point responses

  1. Referee: Numerical experiments section: the claim that TO-Master 'reproduces standard benchmark results' is not supported by quantitative metrics (compliance values, iteration counts, error to reference solutions, or success/failure rates across trials). Without these, the evidence for correct tool selection and model construction under the weakest assumption remains incomplete.

    Authors: We agree that the current numerical experiments section would be strengthened by explicit quantitative metrics. The manuscript presents visual agreement with benchmark topologies and convergence behavior, but does not tabulate objective values, iteration counts, or error measures against published reference solutions. In the revised version we will add a dedicated table (or subsection) that reports, for each standard benchmark: (i) final compliance (or equivalent objective) value, (ii) number of optimization iterations, (iii) relative error with respect to the literature reference, and (iv) success rate across repeated trials when stochastic elements are present. This addition will directly substantiate the claim of correct tool selection and model construction. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents a software framework (TO-Master) that orchestrates existing deterministic finite-element and topology optimization solvers via an LLM agent. No derivation chain, equations, fitted parameters, or first-principles predictions exist in the manuscript. All claims rest on empirical demonstration of tool orchestration and ablation studies on prompting components, with no self-referential reductions or load-bearing self-citations that collapse the central result to its own inputs. The work is self-contained as an engineering description.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The contribution consists of the agent orchestration layer; it rests on standard assumptions from computational mechanics and LLM tool-use reliability rather than new physical postulates or fitted constants.

axioms (1)
  • domain assumption Existing finite-element solvers and sensitivity-based optimizers produce accurate results for the supported problem classes when correctly configured.
    The framework invokes these solvers without modification or re-derivation.

pith-pipeline@v0.9.1-grok · 5820 in / 1262 out tokens · 35427 ms · 2026-07-03T03:25:44.000599+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 26 canonical work pages · 6 internal anchors

  1. [1]

    Computer Methods in Applied Mechanics and Engineering 71(2), 197–224 (1988) https://doi.org/10.1016/0045-7825(88)90086-2

    Bendsøe, M.P., Kikuchi, N.: Generating optimal topologies in structural design using a homogenization method. Computer Methods in Applied Mechanics and Engineering 71(2), 197–224 (1988) https://doi.org/10.1016/0045-7825(88)90086-2

    work page doi:10.1016/0045-7825(88)90086-2 1988

  2. [2]

    Computer Methods in Applied Mechanics and Engi- neering89(1), 309–336 (1991) https://doi.org/10.1016/0045-7825(91)90046-9

    Zhou, M., Rozvany, G.I.N.: The coc algorithm, part ii: Topological, geometrical and generalized shape optimization. Computer Methods in Applied Mechanics and Engi- neering89(1), 309–336 (1991) https://doi.org/10.1016/0045-7825(91)90046-9 . Second World Congress on Computational Mechanics

    work page doi:10.1016/0045-7825(91)90046-9 1991

  3. [3]

    Structural optimization4(3), 250–252 (1992) https://doi.org/10.1007/ BF01742754

    Rozvany, G.I.N., Zhou, M., Birker, T.: Generalized shape optimization without homogenization. Structural optimization4(3), 250–252 (1992) https://doi.org/10.1007/ BF01742754

    1992

  4. [4]

    Archive of Applied Mechanics69(9), 635–654 (1999) https://doi.org/10.1007/ s004190050248

    Bendsøe, M.P., Sigmund, O.: Material interpolation schemes in topology optimiza- tion. Archive of Applied Mechanics69(9), 635–654 (1999) https://doi.org/10.1007/ s004190050248

    1999

  5. [5]

    Computer Methods in Applied Mechanics and Engineering192(1), 227–246 (2003) https://doi.org/10.1016/S0045-7825(02)00559-5 27

    Wang, M.Y., Wang, X., Guo, D.: A level set method for structural topology optimiza- tion. Computer Methods in Applied Mechanics and Engineering192(1), 227–246 (2003) https://doi.org/10.1016/S0045-7825(02)00559-5 27

    work page doi:10.1016/s0045-7825(02)00559-5 2003

  6. [6]

    Journal of Computational Physics194(1), 363–393 (2004) https: //doi.org/10.1016/j.jcp.2003.09.032

    Allaire, G., Jouve, F., Toader, A.-M.: Structural optimization using sensitivity analysis and a level-set method. Journal of Computational Physics194(1), 363–393 (2004) https: //doi.org/10.1016/j.jcp.2003.09.032

    work page doi:10.1016/j.jcp.2003.09.032 2004

  7. [7]

    Computers & Structures49(5), 885–896 (1993) https://doi.org/10.1016/0045-7949(93) 90035-C

    Xie, Y.M., Steven, G.P.: A simple evolutionary procedure for structural optimization. Computers & Structures49(5), 885–896 (1993) https://doi.org/10.1016/0045-7949(93) 90035-C

    work page doi:10.1016/0045-7949(93 1993

  8. [8]

    AIAA Journal37(11), 1483–1488 (1999) https://doi.org/10

    Yang, X.Y., Xie, Y.M., Steven, G.P., Querin, O.M.: Bidirectional evolutionary method for stiffness optimization. AIAA Journal37(11), 1483–1488 (1999) https://doi.org/10. 2514/2.626

    1999

  9. [9]

    arXiv preprint arXiv:2104.01965 (2021)

    Chandrasekhar, A., Sridhara, S., Suresh, K.: Auto: A framework for automatic differentiation in topology optimization. arXiv preprint arXiv:2104.01965 (2021)

    work page arXiv 2021

  10. [10]

    Com- puter Physics Communications292, 108802 (2023) https://doi.org/10.1016/j.cpc.2023

    Xue, T., Hu, Z., Chen, J., Tong, Y., Hu, W.: Jax-fem: A differentiable gpu-accelerated 3d finite element solver for automatic inverse design and mechanistic data science. Com- puter Physics Communications292, 108802 (2023) https://doi.org/10.1016/j.cpc.2023. 108802

    work page doi:10.1016/j.cpc.2023 2023

  11. [11]

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E.H., Le, Q.V., Zhou, D.: Chain-of-thought prompting elicits reasoning in large language models. In: Advances in Neural Information Processing Systems, vol. 35, pp. 24824–24837 (2022). https://arxiv.org/abs/2201.11903

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  12. [12]

    ReAct: Synergizing Reasoning and Acting in Language Models

    Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., Cao, Y.: React: Synergiz- ing reasoning and acting in language models. In: International Conference on Learning Representations (2023).https://arxiv.org/abs/2210.03629

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  13. [13]

    Toolformer: Language Models Can Teach Themselves to Use Tools

    Schick, T., Dwivedi-Yu, J., Dess` ı, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., Scialom, T.: Toolformer: Language models can teach themselves to use tools. In: Advances in Neural Information Processing Systems, vol. 36 (2023). https://arxiv.org/abs/2302.04761

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  14. [14]

    Computer Methods in Applied Mechanics and Engineering (2025)

    Guo, J., Park, C., Qian, D., et al.: Large language model-empowered next-generation computer-aided engineering. Computer Methods in Applied Mechanics and Engineering (2025). in press; also available as arXiv:2509.11447

    work page arXiv 2025

  15. [15]

    Engineering Structures (2024)

    Zhang, T., et al.: Integrating large language models for automated structural analysis. Engineering Structures (2024). arXiv:2407.xxxx, to appear

    2024

  16. [16]

    arXiv preprint arXiv:2509.05969 (2025)

    Fan, E., Hu, K., Wu, Z., Ge, J., Miao, J., Zhang, Y., Sun, H., Wang, W., Zhang, T.: Chatcfd: An llm-driven agent for end-to-end cfd automation with domain-specific structured reasoning. arXiv preprint arXiv:2509.05969 (2025)

    work page arXiv 2025

  17. [17]

    arXiv preprint arXiv:2501.01727 (2025)

    Pandey, S., Kopparthi, A., Sankar, A., Mittal, S.: Openfoamgpt: A rag-augmented llm agent for openfoam-based computational fluid dynamics. arXiv preprint arXiv:2501.01727 (2025)

    work page arXiv 2025

  18. [18]

    Computer- Aided Design (2024)

    Zhang, Y., et al.: Large language models for computer-aided design: A survey. Computer- Aided Design (2024). arXiv:2401.10536

    work page arXiv 2024

  19. [19]

    In: Proceedings of the Design Society (2024)

    Wu, R.,et al.: From text to design: A framework to leverage llm agents for automated cad generation. In: Proceedings of the Design Society (2024)

    2024

  20. [20]

    Large Language Models as Optimization Controllers: Adaptive Continuation for SIMP Topology Optimization

    Yang, S., Wang, J., Wang, Y.: Large language models as optimization controllers: Adaptive continuation for simp topology optimization. arXiv preprint arXiv:2603.25099 (2026) 28

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  21. [21]

    arXiv preprint arXiv:2603.27000 (2026)

    Yang, S., Wang, J., Wang, Y.: Autosimp: Autonomous topology optimization from nat- ural language via llm-driven problem configuration and adaptive solver control. arXiv preprint arXiv:2603.27000 (2026)

    work page arXiv 2026

  22. [22]

    TO-Agents: A Multi-Agent AI Pipeline for Preference-Guided Topology Optimization

    Stewart, I.A., Chen, H., Ahmed, F.: To-agents: A multi-agent ai pipeline for preference- guided topology optimization. arXiv preprint arXiv:2605.21622 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [23]

    Advanced Engineering Informatics68, 103795 (2025)

    Liang, Z., Zhang, Y.-F., Wang, Y., Li, W.: Integrating large models with topology optimization for conceptual design realization. Advanced Engineering Informatics68, 103795 (2025)

    2025

  24. [24]

    Springer, Berlin, Heidelberg (2003)

    Bendsøe, M.P., Sigmund, O.: Topology Optimization: Theory, Methods, and Applica- tions. Springer, Berlin, Heidelberg (2003). https://doi.org/10.1007/978-3-662-05086-6

    work page doi:10.1007/978-3-662-05086-6 2003

  25. [25]

    International Journal for Numerical Methods in Engineering50(9), 2143–2158 (2001) https://doi.org/10.1002/nme.116

    Bourdin, B.: Filters in topology optimization. International Journal for Numerical Methods in Engineering50(9), 2143–2158 (2001) https://doi.org/10.1002/nme.116

    work page doi:10.1002/nme.116 2001

  26. [26]

    International Journal for Numerical Methods in Engineering61(2), 238–254 (2004) https://doi.org/ 10.1002/nme.1064

    Guest, J.K., Pr´ evost, J.H., Belytschko, T.: Achieving minimum length scale in topol- ogy optimization using nodal design variables and projection functions. International Journal for Numerical Methods in Engineering61(2), 238–254 (2004) https://doi.org/ 10.1002/nme.1064

    work page doi:10.1002/nme.1064 2004

  27. [27]

    Structural and Multidisciplinary Optimization 43(6), 767–784 (2011) https://doi.org/10.1007/s00158-010-0602-y

    Wang, F., Lazarov, B.S., Sigmund, O.: On projection methods, convergence and robust formulations in topology optimization. Structural and Multidisciplinary Optimization 43(6), 767–784 (2011) https://doi.org/10.1007/s00158-010-0602-y

    work page doi:10.1007/s00158-010-0602-y 2011

  28. [28]

    Computers & Structures198, 23–39 (2018) https://doi.org/ 10.1016/j.compstruc.2018.01.008

    Yang, D., Liu, H., Zhang, W., Li, S.: Stress-constrained topology optimization based on maximum stress measures. Computers & Structures198, 23–39 (2018) https://doi.org/ 10.1016/j.compstruc.2018.01.008

    work page doi:10.1016/j.compstruc.2018.01.008 2018

  29. [29]

    Structural and Multidisciplinary Optimization31(4), 251–259 (2006) https://doi.org/10.1007/s00158-005-0584-3

    Gersborg-Hansen, A., Bendsøe, M.P., Sigmund, O.: Topology optimization of heat con- duction problems using the finite volume method. Structural and Multidisciplinary Optimization31(4), 251–259 (2006) https://doi.org/10.1007/s00158-005-0584-3

    work page doi:10.1007/s00158-005-0584-3 2006

  30. [30]

    International Journal for Numerical Methods in Engineering24(2), 359–373 (1987) https://doi.org/10.1002/nme.1620240207

    Svanberg, K.: The method of moving asymptotes – a new method for structural opti- mization. International Journal for Numerical Methods in Engineering24(2), 359–373 (1987) https://doi.org/10.1002/nme.1620240207

    work page doi:10.1002/nme.1620240207 1987

  31. [31]

    arXiv preprint arXiv:2512.20469 (2025)

    Zhang, L., Chen, S., Cai, Y., Chai, J., Chang, J., Chen, K., Chen, Z.X., Ding, Z., Du, Y., Gao, Y., et al.: Bohrium+ scimaster: Building the infrastructure and ecosystem for agentic science at scale. arXiv preprint arXiv:2512.20469 (2025)

    work page arXiv 2025

  32. [32]

    Accessed 24 June 2026 (2026)

    Google: Agent Development Kit (ADK). Accessed 24 June 2026 (2026). https://adk. dev/

    2026

  33. [33]

    Accessed 24 June 2026 (2026)

    Model Context Protocol Contributors: Model Context Protocol Documentation. Accessed 24 June 2026 (2026). https://modelcontextprotocol.io/docs/getting-started/ intro

    2026

  34. [34]

    International Journal for Numerical Methods in Engineering79(11), 1309–1331 (2009)

    Geuzaine, C., Remacle, J.-F.: Gmsh: A 3-d finite element mesh generator with built- in pre- and post-processing facilities. International Journal for Numerical Methods in Engineering79(11), 1309–1331 (2009)

    2009

  35. [35]

    TripoSR: Fast 3D Object Reconstruction from a Single Image

    Tochilkin, D., Pankratz, D., Liu, Z., Huang, Z., , Letts, A., Li, Y., Liang, D., Laforte, C., Jampani, V., Cao, Y.-P.: Triposr: Fast 3d object reconstruction from a single image. arXiv preprint arXiv:2403.02151 (2024) 29

    work page internal anchor Pith review Pith/arXiv arXiv 2024