RSI-PDEbench: A Benchmark for Self-Improving Agents in PDE Solving

Researchers have introduced RSI-PDEbench, a benchmark comprising 1,001 cases across 42 equation templates, to evaluate how large language model (LLM) agents can automate the process of identifying and solving partial differential equations (PDEs). They also developed RSI-PDE, a recursively self-improving agent that significantly enhances performance and efficiency compared to existing methods. This work demonstrates the potential of validated recursive self-improvement in LLM agents for scientific computing.