Molecular Dynamics of High Pressure Tin Phases II: Machine Learned Interatomic Potential Development

Atomistic modeling of materials under high pressure shock and ramp conditions can provide information on deformation mechanisms under these extreme conditions. Molecular dynamics modeling of tin under these conditions is of particular interest in order to study the effects of phase transformation on high strain rate plasticity.



In this second talk, we will describe the development of new machine learned interatomic potentials (ML-IAPs) for high-pressure tin phases using two different methodologies and training sets. We compare and contrast the performance of three different ML-IAP frameworks: Spectral Neighbor Analysis Potential (SNAP),Accurate Neural Network Engine for Molecular Energies (ANI), and Hierarchical Interacting Particle Neural Network (HIPNN). Each framework will be trained to two different datasets, each generated with a different collection paradigm. One dataset relies on common structures inferred from equilibrium and defect properties, while the other is generated using an automated methodology which attempts to maximize training set value. The resulting potentials were then compared on a variety of tin material properties such as cold curves, elastic constants, stacking fault energies, etc. Results of the comparison will be presented along with a discussion on how the generation of the training data set impacts the potential performance.



Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA-0003525.