Integrating AI/ML and Extreme-scale computing workflows to accelerate and control materials dynamics and synthesis

Traditional approaches to bridge atomic scale dynamics with experimental observations at the microstructural level often rely on phenomenological models or force-fields of the underlying physics, whose free parameters are in turn fitted to a small number of intuition-driven atomic scale simulations under limited number of thermodynamical and kinetic drivers. This approach becomes particularly cumbersome to study synthesis and characterization of materials with complex dependencies on local environment. In this talk, we will demonstrate extremely scalable autonomous workflows that couple automated high-throughput ab initio and large-scale classical atomistic simulations. Applying a wide range of uncertainty quantification-driven parallel Bayesian ensemble sampling algorithms for on-the-fly forcefield generation and then discovering and controlling directed transformation pathways which connect amorphous to crystalline phases in large-scale molecular dynamics, we will discuss how targeted synthesis of quantum materials (e.g. twisted 2D van der Waals MoS2, polymorphs of Bi2Se3 and spiral phases of WSe2) can be achieved and integrated with multimodal experimental synthesis (or manipulation) platforms.