PDEAgent-Bench: First Benchmark for PDE Solver Code Generation
Researchers have introduced PDEAgent-Bench, the first multi-metric, multi-library benchmark specifically designed for PDE-to-solver code generation. This task involves automatically synthesizing executable numerical solvers from partial differential equation specifications, requiring understanding of mathematical structure, discretization schemes, and solver configurations. Existing code generation benchmarks focus on syntactic correctness or success on predefined test cases, but do not address the unique challenges of numerical PDE solution, such as accuracy, efficiency, and compatibility with professional finite-element method libraries. PDEAgent-Bench aims to fill this gap by providing a comprehensive evaluation framework.
Key facts
- PDEAgent-Bench is the first multi-metric, multi-library benchmark for PDE-to-solver code generation.
- The benchmark addresses challenges like solver accuracy, efficiency, and compatibility with FEM libraries.
- Existing benchmarks do not capture the unique challenges of numerical PDE solution.
- The task requires understanding PDE structure, discretization schemes, and solver configurations.
- The benchmark is introduced in arXiv paper 2605.09636.
Entities
Institutions
- arXiv