A DFT benchmark on water including more than 50 functionals from GGA to double-hybrid levels is reported. The main metric is the accuracy of forces, allowing better structural coverage, higher statistical confidence, and fewer error sources compared to conventional benchmarks. The input structures include water clusters of 4~128 molecules, with highly varied yet realistic configurations of water, ice, and their mixtures. The B97M-V, {\omega}B97M-V, and revDOD-PBEP86-D4 functionals are found to be best on the mGGA, hybrid, and double-hybrid levels, respectively. No satisfying GGA functionals are found, but it can be obtained by combining BLYP-D4 with a simple correction to O-H bonds, achieving accuracy comparable to B97M-V.
@article{arxiv.2512.19375,
title = {Finding the best density functionals for water: benchmarking the forces},
author = {Shifan Cui},
journal= {arXiv preprint arXiv:2512.19375},
year = {2025}
}