Uncertainty-aware molecular dynamics from Bayesian active learning for phase transformations and thermal transport in SiC

Abstract

Machine learning interatomic force fields are promising for combining high computational efficiency and accuracy in modeling quantum interactions and simulating atomistic dynamics. Active learning methods have been recently developed to train force fields efficiently and automatically. Among them, Bayesian active learning utilizes principled uncertainty quantification to make data acquisition decisions. In this work, we present a general Bayesian active learning workflow, where the force field is constructed from a sparse Gaussian process regression model based on atomic cluster expansion descriptors. To circumvent the high computational cost of the sparse Gaussian process uncertainty calculation, we formulate a high-performance approximate mapping of the uncertainty and demonstrate a speedup of several orders of magnitude. We demonstrate the autonomous active learning workflow by training a Bayesian force field model for silicon carbide (SiC) polymorphs in only a few days of computer time and show that pressure-induced phase transformations are accurately captured. The resulting model exhibits close agreement with both ab initio calculations and experimental measurements, and outperforms existing empirical models on vibrational and thermal properties. The active learning workflow readily generalizes to a wide range of material systems and accelerates their computational understanding.

Predictions of Boron Phase Stability Using an Efficient Bayesian Machine Learning Interatomic Potential

“…This practice bypasses an enormous data set, while the representation of the targeted systems is still fulfilled. The reported SGP applications include heterogeneous catalysis, 20,40 phase transition of SiC, 39 and crystallization of GaN. 41 This paper reports the training procedure and MD results based on the SGP potential for icosahedral boron crystals.…”

“…Recently, a Bayesian-model-driven Gaussian process (GP) has shown outstanding performance in MLP-based MD simulations. Most notably, the sparse Gaussian process (SGP) model developed by Kozinsky and co-workers , enables a significant reduction of the training data size. Furthermore, by incorporation of an active learning scheme, the most relevant structures are produced on the fly, accompanied by MD simulations.…”

“…In this work, the SGP training was conducted using the FLARE program , (version 1.3.3). Detailed information on SGP methodology can be found in papers published by FLARE developers. , …”

“…Detailed information on SGP methodology can be found in papers published by FLARE developers. 20,39 The OTF MD simulations also enable the update of SGP potentials. Once the configuration is updated by a new MD step based on the current potential, the algorithm will estimate the uncertainty of the new structure based on the inherent quantification mechanism of SGP and then determine if the predetermined uncertainty threshold (0.01 eV/Å) is exceeded.…”

“…This practice bypasses an enormous data set, while the representation of the targeted systems is still fulfilled. The reported SGP applications include heterogeneous catalysis, , phase transition of SiC, and crystallization of GaN …”