Benchmarks

The benchmarks/ directory of the repository holds three benchmarks. Each writes its results to benchmarks/results/ (or benchmarks/agents/results/), one directory per run, with the environment it ran in. Timings depend on the machine and its load; the accuracy figures do not.

Benchmark	Script	What it measures
Cross-method benchmark	benchmarks/run.py	Accuracy against a reference and cost (wall time, peak memory) of every installed backend, on CPU and GPU, across a ladder of resolutions. Problem suites are generated by benchmarks/problems.py.
BEM convergence	benchmarks/bem_convergence.py	Error of BEM, DDA and FDTD against Mie for a gold sphere, versus resolution and cost. See BEM.
Agent benchmark	benchmarks/agents/run.py	Whether an AI agent answers photonics questions correctly through the MCP server.

benchmarks/README.md describes how to reproduce the cross-method runs and every option of run.py.