This work presents the training of reactive SchNet neural network potentials (NNPs) for accurate and general potential energy surface (PES) modeling of silica, encompassing the structural diversity of zeolites across a wide density range. The trained NNPs facilitate the reoptimization of the Deem and IZA databases, enabling the prediction of structure and vibrational properties, and simulations of reactive phase transformations like glass melting and zeolite amorphization. The NNPs achieve significantly improved accuracy compared to other PES approximations.
Publisher
npj Computational Materials
Published On
Jan 31, 2022
Authors
A. Erlebach
Tags
reactive SchNet
neural network potentials
potential energy surface
silica
zeolites
glass melting
vibrational properties
Related Publications
Explore these studies to deepen your understanding of the subject.
