Evolutionary multi-objective optimization and Pareto-frontal uncertainty quantification of interatomic forcefields for thermal conductivity simulations

Abstract Predictive Molecular Dynamics simulations of thermal transport require forcefields that can simultaneously reproduce several structural, thermodynamic and vibrational properties of materials like lattice constants, phonon density of states, and specific heat. This requires a multi-objective optimization approach for forcefield parameterization. Existing methodologies for forcefield parameterization use ad-hoc and empirical weighting schemes to convert this into a single-objective optimization problem. Here, we provide and describe software to perform multi-objective optimization of Stillinger–Weber forcefields (SWFF) for two-dimensional layered materials using the recently developed 3rd generation non-dominated sorting genetic algorithm (NSGA-III). NSGA-III converges to the set of optimal forcefields lying on the Pareto front in the multi-dimensional objective space. This set of forcefields is used for uncertainty quantification of computed thermal conductivity due to variability in the forcefield parameters. We demonstrate this new optimization scheme by constructing a SWFF for a representative two-dimensional material, 2H-MoSe2 and quantifying the uncertainty in their computed thermal conductivity. Program summary Program Title: MOGA-NSGA3 Program Files doi: http://dx.doi.org/10.17632/pbc6nb7hp9.1 Licensing Provisions: GNU General Public License 3 Programming Language: C++ Nature of problem: Interatomic forcefields used for molecular dynamics simulations of thermal conductivity must be parameterized to accurately capture structural and vibrational properties of the material system being modeled. Therefore, these forcefields must be simultaneously optimized against several (n ≥ 5) material properties. However, such parameterization is difficult using existing forcefield parameterization schemes, which are limited to optimization of a single or few (n 3) objectives. Solution method: We present software to perform evolutionary optimization of forcefields for thermal conductivity simulations using the recently developed 3rd generation non-dominated sorting genetic algorithm (NSGA-III). The algorithm’s unique reference-point-based niching and non-dominated sorting schemes enable efficient exploration of higher-dimensional objective spaces while preserving diversity among forcefield populations. The best set of forcefields on the Pareto front are used for estimating uncertainty in computed thermal conductivity due to forcefield parameterization.