kinisi: Bayesian analysis of mass transport frommolecular dynamics simulations
Andrew R. McCluskey, Alexander G. Squires, Josh Dunn, Samuel W. Coles, Benjamin J. Morgan
Abstract
kinisi is a Python package for estimating transport coefficients—e.g., self-diffusion coefficients, ∗—and their corresponding uncertainties from molecular dynamics simulation data. It includes an implementation of the approximate Bayesian regression scheme described in McCluskey etal. (2023), wherein the mean-squared displacement (MSD) of mobile atoms is modelled as a multivariate normal distribution that is parametrised from the input simulation data. kinisi uses Markov-chain Monte Carlo (Foreman-Mackey et al., 2019; Goodman & Weare, 2010) to sample this model multivariate normal distribution to give a posterior distribution of linear model ensemble MSDs that are compatible with the observed simulation data. For each linear ensemble MSD, x(), a corresponding estimate of the diffusion coefficient, ̂∗ is given via the Einstein relation, ̂∗ =1d x() / 6 d where is time. The posterior distribution of compatible model ensemble MSDs calculated by kinisi gives a point estimate for the most probable value of ∗ , given the observed simulation data, and an estimate of the corresponding uncertainty in ̂∗. kinisi also provides equivalent functionality for estimating collective transport <br/> coefficients, i.e., jump-diffusion coefficients and ionic conductivities<br/><br/>