AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-20
Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials
The reliability of universal machine-learning interatomic potentials under directional deformation is unclear. The authors compared compression and extension along paired deformation paths from the same parent structures for eleven pretrained potentials, using DFT references. At ±10% strain, every model had a larger median force error under compression. Decomposing the error growth showed that force-scale matching reduces the asymmetry, though a compression-side excess remains in one dataset.