(272h) Accelerating Screening of Simulation Parameter Spaces By Orders of Magnitude Using Reweighting and Configuration Space Mapping Algorithms

Classical force field parameterization is computationally intensive. The fitting procedure involves computationally expensive molecular simulations to estimate observables at each iteration. The new set of parameters is either guessed based on a trial and error approach or is generated by symplectic or gradient based optimization. The time spent in either case is dominated by evaluation of the objective function which requires fresh generation of samples for a new guess of parameter set. We can accelerate the force field parameterization process if we can reduce the time spent in sampling at each iteration. In this work we present a parameterization scheme. We gain speed by skipping several steps of molecular simulation by replacing these runs using multistate reweighting techniques to estimate the observables.  The multistate reweighting formalism requires re-evaluation of the energies with the new set of parameters using the configurations generated in simulations run with the initial guess set of parameters. The time consumption for estimating observables using re-evaluated energies is orders of magnitude less compared to a fresh simulation.  By incorporating configuration space transformations, we can reweight over changes in geometric parameters, even for rigid molecular models, as well as energetic parameters. We demonstrate the utility and efficiency of this parameterization scheme by optimizing over the space of noncovalent simulation parameters in free energy calculations, as well as exploring the parameter space of liquid water models.