On-the-fly active learning of interpretable Bayesian force fields for atomistic rare events

Abstract

Machine learned force fields typically require manual construction of training sets consisting of thousands of first principles calculations, which can result in low training efficiency and unpredictable errors when applied to structures not represented in the training set of the model. This severely limits the practical application of these models in systems with dynamics governed by important rare events, such as chemical reactions and diffusion. We present an adaptive Bayesian inference method for automating the training of interpretable, low-dimensional, and multi-element interatomic force fields using structures drawn on the fly from molecular dynamics simulations.Within an active learning framework, the internal uncertainty of a Gaussian process regression model is used to decide whether to accept the model prediction or to perform a first principles calculation to augment the training set of the model. The method is applied to a range of single-and multi-element systems and shown to achieve a favorable balance of accuracy and computational efficiency, while requiring a minimal amount of ab initio training data. We provide a fully open-source implementation of our method, as well as a procedure to map trained models to computationally efficient tabulated force fields.

Evolution of Metastable Structures at Bimetallic Surfaces from Microscopy and Machine-Learning Molecular Dynamics

“…21 All of our interactive Jupyter Notebook files are publicly available. 61 Density functional theory (DFT) data is used to train a Gaussian process (GP) machine-learning force field, 23 which allows us to perform large-scale and long-timescale molecular dynamics (MD) at first-principles accuracy (Sec. 1, SI).…”

“…1; see Experimental Section). Fast and large-scale machine-learning molecular dynamics (MD) is performed using our recently developed Gaussian process (GP) force field, 23 spanning microseconds at first-principles accuracy. Furthermore, a new automated analysis method is used to discover and characterize key surface restructuring mechanisms (Fig.…”

“…(a) Gaussian process (GP) machine-learning force field is trained on DFT data. 23 (b) The GP model is used to perform fast, large-scale, and long-timescale MD simulation of surfaces. (c) The raw MD trajectory is first "clamped" by assigning all atoms onto discrete FCC lattice sites.…”

Machine learning for multi-fidelity scale bridging and dynamical simulations of materials

“…By tracking model uncertainty during the MD simulation, a call to the high-fidelity (but expensive) DFT calculation can made when the system drifts to a configuration where the model is uncertain in the energy/force prediction beyond a certain threshold [102,121,122]. Vandermause et al [123] sampled structures on-the-fly from AIMD and used an adaptive Bayesian inference method to automate the training of low-dimensional multiple element interatomic force fields. Their AL framework uses internal uncertainty of a GPR model to decide acceptance of model prediction or the need to augment training data.…”

“…Crossvalidation, sensitivity analysis and uncertainty quantification will be critical to test the robustness of their developed models. Preliminary work on the use of uncertainty quantification has also lead to autonomous MD workflows such as FLARE [123] that provide a merger between high-fidelity quantum mechanical and low-fidelity machine learning potentials.…”