SPaiNN for machine learning driven excited-state dynamics

Meet SPaiNN, our open-source Python toolkit for machine learning driven excited-state dynamics!

By combining equivariant neural networks (from SchNetPack) with SHARC’s surface hopping engine, SPaiNN makes non-adiabatic molecular dynamics faster and more accessible — without sacrificing quantum accuracy. Equivariant models shine in our tests on methyleneimmonium and alkene systems, outperforming invariant ones in accuracy and generalization.

Check out this collaborative work from the Julia Westermayr (Uni Leipzig) and our lab:

Article Code Documentation