Manuscript for first EAGER grant paper on Bayesian inference driven force field parameterization using a multi-fidelity likelihood calculation algorithm
Test container to display in-progress machinery for using amortized inference to refit the 4-D C-C & C-H LJ parameters (to scale up) from the smirnoff99Frosst forcefield. We are training solely on molar volume and heat of vaporization from cyclohexane.
/scripts: contains all python scripts and ipython notebooks demonstrating making estimates of properties usingpymbar, constructing Gaussian process regressions withgpyand implementing MCMC usingemceein order to sample from a posterior distribution of forcefield parameters./simulated_estimates: contains observables calculated by simulation at ~120 different parameter states in order to compare to MBAR./MBAR_estimates: contains statistically robust observables calculated by MBAR used to construct Gaussian Process regressions./figures: contains figures constructed to visually debug simulated energy distributions.