Credit: Image courtesy of Kipton Barros, Los Alamos National Laboratory.
The model’s structure. A neural network processes a molecular geometry to predict a semi-empirical quantum Hamiltonian, which is then solved self-consistently to predict a variety of chemical properties.