AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-23
Unconstrained learning of equivariant atomistic properties
Graph neural network potentials built around O(3) equivariance become costly for high-order atomic properties and can fail to represent them when truncated. The authors used the Point-Edge Transformer, which learns equivariance from data, to predict dipoles, polarizabilities, NMR shielding tensors and electron-density basis expansions. It matched or beat reference models at lower inference cost and memory, and gave accurate IR and Raman spectra.