Reproduce “RSI-PDEbench: A Benchmark for Self-Improving Agents in PDE Solving” 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
Binghang Lu, Lang Cao, Changhong Mou, Yige Sun, Cindy Xiangrui Kong, Zheyuan Deng, Tian Yuan, Bing Hu and 3 more
RSI-PDEbench tests agents identifying hidden equation families and coefficients from sparse samples and constructing numerical solvers. Its harness ranks candidate equations by residual, uses public solution-error feedback to refine solvers, and retains validated updates. Across four models, noisy-case mean pass rate is 63.8% versus 53.3% for native harnesses; the selected cases are not a fresh secret holdout.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.