Accurate Estimation of Diffusion Coefficients and their Uncertainties#
kinisi is an open-source Python package that can accurately estimate diffusion processes in atomic and molecular systems and determine an accurate estimate of the uncertainty in these processes.
This is achieved by modelling the diffusion process as a multivariate normal distribution, based on that for a random walker.
This ensures and accurate estimation of the diffusion coefficient and it’s uncertainty.
More information about the approach kinisi uses can be found in the methodology article, which is also introduced in this poster.
kinisi; showing the determined mean-squared displacements (solid line),the estimated Einstein diffusion relationship (blue regions representing descreasing credible intervals).
kinisi is built using the scipp library, we recommend familiarising yourself with some of the scipp data structures if you want to start doing more complex things with kinisi.
kinisi can handle simulation trajectories from many common molecular dynamics packages; if your trajectory can be read by MDAnalysis, ASE, or pymatgen, then you can use kinisi.
The Markov chain Monte Carlo algorithm uses the emcee package.
Examples of some of the analyses kinisi can perform are shown in the notebooks.
