Reproduce “MetaOpenFOAM 2.0: Large Language Model Driven Chain of Thought for Automating CFD Simulation and Post-Processing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuxuan Chen, Xu Zhu, Hua Zhou, Zhuyin Ren
MetaOpenFOAM 2.0, developed at Tsinghua University, automates the full Computational Fluid Dynamics (CFD) simulation and post-processing workflow using a GPT-4o driven multi-agent system, achieving an 86.9% pass rate and average executability of 6.3/7, representing a substantial improvement over previous LLM-driven CFD tools.
Key insights
- The integration of two-stage Question Decomposition COT (QDCOT) and multi-level Iterative Verification and Refinement (ICOT) is crucial for managing the complexity and achieving high success rates in full CFD workflows.
- A positive correlation exists between the number of COT steps (decomposition and iterative refinement) and both system executability and token consumption, aligning with 'post-trained scaling laws' for LLMs.
- An LLM-assisted verification module significantly enhances the physical and numerical accuracy of results by identifying and rectifying errors, demonstrating the value of deep verification within scientific pipelines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.