Towards ex-machina computations of transport and transformations in complex materials

Harnessing the accuracy of quantum mechanics to design complex materials requires a series of approximations to reach the desired length and time scales. In the context of thermoelectric materials, new automatable computational methods for describing electron-phonon dynamics enabled us to discover new leading-efficiency alloy compositions. In the context of batteries, understanding of ionic transport mechanisms is important for enabling next-generation solid-state electrolytes. To advance the ability to capture complexity of ionic, thermal and reaction dynamics, we are pursuing the paradigm of “ex-machina” computations where data-driven approximations are automatically developed using machine learning algorithms. To accelerate molecular dynamics calculations, we developed the Neural equivariant interatomic potential model (NequIP) based on tensor-valued symmetry-preserving neural network architectures and used them to achieve state-of-the-art accuracy and training efficiency for molecules, liquids, heterogeneous catalysts, and ionic conductors. In order to enable autonomous selection of the training set for reactive systems, we developed the FLARE adaptive closed-loop algorithm that automatically constructs accurate non-parametric interatomic force fields on-the-fly from a molecular dynamics simulation. We demonstrate the performance of ML-accelerated MD simulations by studying 2D-to-3D transformations of quantum materials, shape memory effect in alloys and thermal transport in semiconductors.