Atomistic learning in the electronically grand-canonical ensemble

Abstract A strategy is presented for the machine-learning emulation of electronic structure calculations carried out in electronically grand-canonical ensemble. The approach relies upon a dual-learning scheme, where both system charge and energy are predicted each image. scheme shown to be capable emulating basic electrochemical reactions at range potentials, coupling it with bootstrap-ensemble gives reasonable estimates prediction uncertainty. method also demonstrated accelerate saddle-point searches, extrapolate systems one five water layers. We anticipate that this will allow larger length- time-scale simulations necessary simulations.