Utilizing Data-Driven Optimization to Automate the Parametrization of Kinetic Monte Carlo Models

Kinetic Monte Carlo (kMC) simulations are a popular toolto investigatethe dynamic behavior of stochastic systems. However, one major limitationis their relatively high computational costs. In the last three decades,significant effort has been put into developing methodologies to makekMC more efficient, resulting in an enhanced runtime efficiency. Nevertheless,kMC models remain computationally expensive. This is in particularan issue in complex systems with several unknown input parameterswhere often most of the simulation time is required for finding asuitable parametrization. A potential route for automating the parametrizationof kinetic Monte Carlo models arises from coupling kMC with a data-drivenapproach. In this work, we equip kinetic Monte Carlo simulations witha feedback loop consisting of Gaussian Processes (GPs) and Bayesianoptimization (BO) to enable a systematic and data-efficient inputparametrization. We utilize the results from fast-converging kMC simulationsto construct a database for training a cheap-to-evaluate surrogatemodel based on Gaussian processes. Combining the surrogate model witha system-specific acquisition function enables us to apply Bayesianoptimization for the guided prediction of suitable input parameters.Thus, the amount of trial simulation runs can be considerably reducedfacilitating an efficient utilization of arbitrary kMC models. Weshowcase the effectiveness of our methodology for a physical processof growing industrial relevance: the space-charge layer formationin solid-state electrolytes as it occurs in all-solid-state batteries.Our data-driven approach requires only 1-2 iterations to reconstructthe input parameters from different baseline simulations within thetraining data set. Moreover, we show that the methodology is evencapable of accurately extrapolating into regions outside the trainingdata set which are computationally expensive for direct kMC simulation.Concluding, we demonstrate the high accuracy of the underlying surrogatemodel via a full parameter space investigation eventually making theoriginal kMC simulation obsolete.