LLM Agents Meet the Finite Element Method

Participate via Teams webinar

Setting up a finite element simulation still demands expert hands: choosing material properties, boundary conditions, meshes, and solver settings, then debugging until it converges. PDE-Agents closes the gap between a plain-language request and a running, verified simulation. Three specialist large language model agents (Simulation, Analytics, and Database), orchestrated through a LangGraph supervisor and running fully locally on open-source models, autonomously configure, execute, analyse, and iteratively improve FEM simulations. A GraphRAG knowledge graph grounds their reasoning in curated material data, known failure modes, and past runs.

This talk presents the architecture, a formal verification study, and a controlled ablation showing that how knowledge is integrated, not just what it contains, decides whether retrieval helps or hinders an agent. We close with lessons for building reliable autonomous simulation assistants in industry.

Presented by Sayan Adhikari, PhD, Institute for Energy Technology (IFE).

Reference: PDE-Agents: An LLM-Orchestrated Multi-Agent Framework for Automated Finite Element Simulations with Knowledge Graph-Augmented Reasoning, arXiv:2606.07850