AutoPDE: Reliable Agentic PDE Solving via Explicitly Represented Solver Strategies

In three linesAutoPDE is an AI agent that solves partial differential equations (PDEs) by maintaining the numerical solver strategy as an explicitly represented, inspectable object. The approach combines PDE analysis, numerical method selection, and adaptive tuning. On PDE Agent Bench, AutoPDE achieves 54.5% pass rate, improving over the strongest baseline by 14.2 percentage points.

Summary generated by Claude — human-verified