- Danny Perez, Los Alamos National Laboratory
- Thomas Swinburne, CNRS
Atomistic simulations play a fundamental role in theoretical materials science, allowing unique insight into the mechanisms controlling structural and functional properties. Whilst the dawn of the exascale era has brought an explosive increase in available computing power and rate of simulation throughput, the combinatorially vast size of chemical and configurational space still resists exhaustive exploration. Developing accurate-yet-efficient material models, e.g., using modern machine learning and data science approaches, therefore requires judiciously selecting the atomic scale simulations that will provide the most information at the most affordable computational cost. This mini-symposium focuses on the crucial decision-making steps involved in the creation of such models. Contributions will be sought along three main axes: 1) the creation of datasets for machine learning models of atomistic interactions, 2) the efficient exploration of the chemical space of materials, and 3) high-throughput sampling or dynamical approaches to thermodynamic and kinetic properties of materials. In all of these cases, selecting the most information-rich calculations to execute is crucial to the success of the overall approach. Speakers from academia, government laboratories, and industry will share their perspectives on this emerging and important problem. The mini-symposium aims at sharing novel ideas from diverse backgrounds, that can be adapted and transferred across a broad space of applications.