AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-19
Multitask learning with semiempirical orbital charges enables sample-efficient MLIPs
Training MLIPs requires large, expensive datasets, and training on full Hamiltonian matrices is intractable at scale. The authors add orbital charges from the inexpensive GFN1-xTB method as a multitask training target, using a specialized equivariant model to predict them. The model cuts energy error by 58% relative to energy-only baselines and needs five times less data to match them, and it still reduces energy error by 12% against joint energy-and-force training.