Accelerated Bayesian Inference for Molecular Simulations using Local Gaussian Process Surrogate Models

Abstract

While Bayesian inference is the gold standard for uncertainty quantification and propagation, its use within physical chemistry encounters formidable computational barriers. These bottlenecks are magnified for modeling data with many independent variables, such as X-ray/neutron scattering patterns and electromagnetic spectra. To address this challenge, we apply a Bayesian framework accelerated via local Gaussian process (LGP) surrogate models. We show that the time-complexity of LGPs scales linearly in the number of independent variables, in stark contrast to the computationally expensive cubic scaling of conventional Gaussian processes. To illustrate the method, we trained a LGP surrogate model on the experimental radial distribution function of liquid neon, and observed a remarkable 288,000-fold speed-up compared to molecular dynamics with insignificant loss in predictive accuracy. We conclude that LGPs are robust and efficient surrogate models, poised to expand the application of Bayesian inference in molecular simulations to a broad spectrum of ever-advancing experimental data.

References in corpus (2)

• Bayesian machine learning for quantum molecular dynamics
• Well-posedness of the Iterative Boltzmann Inversion

Cited by in corpus (3)

• Bayesian Analysis Reveals the Key to Extracting Pair Potentials from Neutron Scattering Data
• Physics-Informed Gaussian Process Inference of Liquid Structure from Scattering Data
• Experimental Evidence of Quantum Drude Oscillator Behavior in Liquids Revealed with Probabilistic Iterative Boltzmann Inversion