(657e) Automated Calculation of Coarse-Grained Potentials Using the Iterative Boltzmann Inversion (IBI) Method

Computation at mesoscopic length scales between the atomistic and continuum for polymeric and similar soft materials requires coarse-graining (CG) techniques. These techniques enable computational access to the greater length and time scales inherent to these materials by subsuming atoms associated with monomers or other chemical moieties into “coarse-grained” sites and then describing the physics by means of a potential parametrized for the coarse-grained description of the material. The development of methods to properly parametrize coarse-grained potentials is an active area of research compromising many different approaches across the soft matter community. Systematic, "bottom-up" methods, in which the coarse-grained (CG) force-field is directly derived from all-atom (AA) simulation data, have the advantage of retaining a high degree of chemical specificity, and thus, can typically capture equilibrium structure quite well.

A widely used systematic method is Iterative Boltzmann inversion (IBI) which retains chemical structure via the radial distribution function (RDF), and hence, thermodynamic properties. In IBI, an initial guess for the CG bonded and non-bonded potential are iteratively refined by means of a correction proportional to the ratio between the atomistic (target) and coarse-grained sampling distributions. We report here on a software code which automates the development of coarse-grained potentials using IBI. The code base to support this consists of a Python driver, several Python modules with core library functions, an external C++ code for calculating distributions, and the use of LAMMPS as a molecular dynamics (MD) engine. Two major problems make automation difficult: 1) noisy distributions derived from sampling; 2) low sampling regions which introduces discontinuities in the sampling. Both of these make differentiation of the energy to calculate forces difficult and error prone. Our approach addresses these problems by the construction of library functions which combine data smoothing, fitting to Gaussian functionals (bonded potentials), and flexible extrapolation schemes to handle low sampling regions without discontinuity or data distortion. The problem of overfitting is handled in part by error handling to keep the procedure running smoothly. We also introduce a library to automate the conversion of AA to CG representations and have built in an input-output framework which aims at providing reproducibility. Several use cases will be discussed aimed at providing guidance on usage.

References

1. S. Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, J Comp Phys, 117, 1-19 (1995).