Bayesian Committee Machine Potential for Isothermal-Isobaric Molecular Dynamics Simulations

In recent advancements in material simulations, the utilization of Sparse Gaussian Process Regression (SGPR)-based machine learning potentials (MLPs) has proven to be highly successful in diverse applications such as catalysis, batteries, and solar cells. In the context of isothermal and isobaric molecular dynamics simulations, achieving precise pressure estimates is crucial for an accurate understanding of the system behavior under constant pressure conditions. In this study, we introduce a novel kernel function designed for estimating the virial term, a critical component for pressure calculations in materials simulations. Our study reveals that the inclusion of a virial prediction in the kernel function leads to significantly improved accuracy in calculating the stress of a system. We present a kernel-based ML potential that can be estimated via the Bayesian Committee Machine without the need for additional training. This improvement allows us to calculate the melting temperature of ice through Isobaric-Isenthalpy $NpH$ molecular dynamics simulations of ice-liquid coexisting phases, employing the BCM potential.