Massively parallel fitting of Gaussian approximation potentials

Abstract We present a data-parallel software package for fitting Gaussian approximation potentials (GAPs) on multiple nodes using the ScaLAPACK library with MPI and OpenMP. Until now maximum training set size GAP models has been limited by available memory single compute node. In our new implementation, descriptor evaluation is carried out in parallel no communication requirement. The subsequent linear solve required to determine model coefficients parallelised ScaLAPACK. Our approach scales thousands of cores, lifting limitation also delivering substantial speedups. This development expands applicability more complex systems as well opening up opportunities efficiently embedding within higher-level workflows such committee or hyperparameter optimisation.