(63g) A Simulation Database for Applications of Mbar to Force Field Development | AIChE

(63g) A Simulation Database for Applications of Mbar to Force Field Development


Razavi, S. M. - Presenter, The University of Akron
Elliott, J. R. Jr., The University of Akron
The method of modified Bennett acceptance ratio (MBAR) has been previously demonstrated to provide a basis for force field development when potential parameters are systematically varied over a range of simulations. The problem that arises is how to store the necessary information for a large molecular database such that global force field optimization becomes a viable option in the context of multiple specifications. Examples of specifications include tail corrected, truncated, cut and shifted, and switched. Sample force fields can be characterized by Mie n-6 united atom potentials where n varies from 12 to 16 for sites that include CH3-, >CH2, >CH-, =CH-, >C=, >C<, =ACH-, =AC< to represent paraffinic, olefinic, naphthenic, and aromatic hydrocarbons. The grid of simulations suffices to simultaneously optimize seven key site types for 35 prototypical molecules. Distinctions between naphthenic vs paraffinic CHx and aromatic vs olefinic =CHx- are resolved through appropriate F-tests.

Simulations are conducted in the NVT ensemble with rigid bond lengths according to the isothermal-isochoric (ITIC) grid of state points. Results at ρ ≤ 0.1 g/cc are characterized in terms of second and third virial coefficients. 1000 configurations are sampled and stored at each of the higher densities using an efficient hexadecimal coordinate specification. The configurations permit recalculation with appropriate reweighting within the MBAR formalism, including derivative properties through the Lustig method.

The methodology is demonstrated by developing globally optimized n-6 force fields for tail corrected, cut and shifted, and switched force potentials from the same database. The globally optimized force fields are shown to provide better performance relative to globally defined objective functions and smaller uncertainties in the characterized potential parameters.