M. Sipka, A. Erlebach, C.J. Heard, P. Nachtigall, L. Grajciar
Keywords: Neural network potentials, Aluminosilicates,
Summary:Machine learning interatomic potentials (MLPs) are becoming increasingly popular in multiple areas of material science. The ability of MLPs to reach the ab initio accuracy at the cost of the standard reactive force fields relies on the use of established and robust training and data curation procedures. EigenSpace, in collaboration with Charles University in Prague, developed a robust learning pipeline to obtain industry level MLPs. The potentials are well tested and verified in aluminosilicates, yet the training process is generally applicable to almost any other system of scientific interest.