Topic
Learned models used in place of classical potentials to reproduce material dynamics with higher accuracy.
Hessian Interatomic Potentials predict molecular Hessians directly and run one to two orders of magnitude faster, their authors report. A cheaper Hessian would make second-order chemistry workflows easier to run routinely, if the reported gains survive independent tests.
Reality
Grandel, Benner and George cut phonon errors across 53 materials by fine-tuning a pre-trained interatomic potential on as few as 10 structures each. Models with equal phonon accuracy still disagreed on instabilities, so the energy surface needs its own check.
Reality
Michele Simoncelli and colleagues at Columbia and Cambridge scored machine-learning potentials against quantum calculations on more than 100 crystals. Two models can match on energy and part ways on heat.
Reality
A machine-learning potential for Co-Ni-V-H finds V-centered short-range order raises average hydrogen solution energies and thins out strong binding sites versus a random alloy.
Reality
A billion-atom simulation from Helsinki finds defect counts in tungsten scaling superlinearly with recoil energy, a regime the standard damage models do not contain.
Reality