OpenFerro: a GPU-accelerated, auto-differentiable, universal framework for on-lattice atomistic dynamics simulation

First principle-based effective Hamiltonian models are potent for studying equilibrium and dynamical properties of materials with on-lattice order parameters, such as ferroelectric crystals, magnetic crystals, and multiferroic crystals [1,2,3]. Dynamical simulation of these on-lattice models is often developed separately in these fields due to different interaction terms and equations of motion. Here, we introduce OpenFerro, an open-source Python package for on-lattice atomistic simulation where the interaction and evolution of multiple order parameters (e.g. local strain, local dipole, atomistic spin, ...) can be treated under a universal framework. Finite-temperature simulation of generic molecular dynamics and Landau-Lifshitz dynamics can run simultaneously, both constrained by the fluctuation-dissipation theorem. The force engine in OpenFerro is based on auto-differentiation, allowing flexible user-defined Hamiltonian and enhanced sampling. OpenFerro also supports multiple-GPU parallelism. Simulation of mesoscopic processes is possible on the micrometer scale.

[1] Zhong, W., David Vanderbilt, and K. M. Rabe. "First-principles theory of ferroelectric phase transitions for perovskites: The case of BaTiO3." PRB 52.9 (1995): 6301.

[2] Eriksson, Olle, et al. Atomistic spin dynamics: foundations and applications. Oxford University Press, 2017.

[3] Kornev, Igor A., et al. "Finite-temperature properties of multiferroic BiFeO3." PRL 99.22 (2007): 227602.