LLM-driven design of physics-constrained constitutive models: Two agents are better than one
Marius Tacke, Matthias Busch, Kian Abdolazizi, Jonas Eichinger, Kevin Linka, Roland Aydin, Christian Cyron · 2026
Read LLM-driven design of physics-constrained constitutive models: Two agents are better than one by Marius Tacke, Matthias Busch, Kian Abdolazizi, Jonas Eichinger, Kevin Linka, Roland Aydin, Christian Cyron (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13630814.txt.
Description
Marius Tacke and six colleagues paired two AI agents to write models of how materials deform. The models must respect the laws of physics. Building such models normally takes years of specialist work. Single language models can draft them quickly. Yet they sometimes produce equations that break basic physical laws.
In their system a Creator agent drafts a neural network model for a material. An Inspector agent audits it against nine physical constraints. It sends violations back for repair. The pair was tested on brain tissue, rubber, and pig skin with two model backbones. Adding the inspector raised the share of fully lawful models. The survivors matched or beat expert-built models. They held up even on loading patterns never seen in training. The authors pitch the split as a general trust upgrade.
Catalog record
| CREATOR | Marius Tacke, Matthias Busch, Kian Abdolazizi, Jonas Eichinger, Kevin Linka, Roland Aydin, Christian Cyron |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 178 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |