A hierarchical Bayesian framework for force field selection in molecular dynamics simulations.

Area of Science:

• Computational chemistry
• Statistical mechanics
• Materials science

Background:

• Selecting accurate force fields for molecular dynamics (MD) simulations is crucial for predicting material properties.
• Existing methods often struggle to account for parameter variability across different experimental conditions and quantify predictive uncertainty.

Purpose of the Study:

• To develop a hierarchical Bayesian framework for robust force field selection in MD simulations.
• To reduce the computational cost of Bayesian inference for complex systems.
• To improve the accuracy of parameter estimation and prediction of future experimental outcomes.

Main Methods:

• Developed a hierarchical Bayesian framework integrating experimental data variability with force field parameter uncertainty.
• Implemented a parallelized Transitional Markov Chain Monte Carlo (TMCMC) method with Laplace Asymptotic Approximation to accelerate computation.
• Validated the framework using MD simulations to predict transport coefficients under varying conditions.

Main Results:

• Demonstrated accurate quantification of uncertainty across multiple experiments.
• Showed significant improvement in posterior probability density function estimation for force field parameters.
• Successfully identified the most plausible force field for a given dataset, enhancing predictive power for future experiments.

Conclusions:

• The hierarchical Bayesian framework provides an efficient and accurate method for selecting molecular dynamics force fields.
• The approach effectively handles hierarchical experimental data, improving parameter estimation and predictive capabilities.
• Applicable to diverse nanoscale simulations requiring integration with structured experimental data.