Just another WordPress site - Ruhr-Universität Bochum

Abstract

First principles effective lattice models are potent for studying mesoscopic processes in ferroelectric, magnetic, and multiferroic crystals. The tools for the dynamical simulation of these lattice models are often developed separately in these fields due to different interaction terms and different equations of motion. We introduce the open-source Python package “OpenFerro”, a numerical framework for simulating general on-lattice models. So far, “OpenFerro” supports the simulation of any number of interacting R^3 fields (dipoles, strains, etc.) and SO(3) vector fields (atomistic spins) on Bravais lattice, by integrating generic isothermal-isobaric molecular dynamics and stochastic Landau-Lifshitz-Gilbert equations. OpenFerro is built on JAX’s automatic differentiation engine. Users can easily develop custom interaction terms, in addition to the built-in library of interaction (such as long-range dipolar interaction), to the Hamiltonian. OpenFerro enables routine single-GPU simulation on (~100nm, ~10ns) scales that used to be reigned by phenomenological continuum models. To reach similar scales, machine-learned potential typically needs tremendous GPU resources.