gmak

gmak performs parameter-space mapping to calibrate force-field parameters for classical molecular simulations, improving agreement between simulation-derived properties and reference data.

Key Features:

• Parameter-Space Mapping (PSM): Implements a PSM workflow to explore and refine force-field parameters.
• Regular-Grid Discretization: Discretizes the parameter search space into a regular grid to structure exploration and sampling.
• Partial Sampling: Samples only a subset of grid points to limit direct simulations while covering the parameter space.
• Surrogate Modeling: Uses surrogate models to predict outcomes for unsampled parameter configurations.
• Multiobjective Optimization: Applies multiobjective optimization concepts to identify local minima and Pareto-efficient points within the parameter space.
• Statistical Error Attenuation: Empirically extends simulation durations to reduce statistical error in sampled results.
• Iterative Search-Space Translation: Iteratively adjusts the search space based on a user-defined scalar objective function.
• Scalar Objective Metric: Supports objective functions such as the weighted root-mean-square deviation of target properties relative to reference data.
• Implementation: Implemented in Python.

Scientific Applications:

• OPC3 water model Lennard-Jones recalibration: Demonstrated recalibration of Lennard-Jones parameters for the 3-point OPC3 water model to align with the GROMOS treatment of nonbonded interactions, reproducing typical pure-liquid properties comparable to OPC3 and outperforming the SPC model.

Methodology:

Regular-grid discretization of the parameter space; partial sampling of the grid with surrogate models to predict unsampled configurations; multiobjective optimization to locate local minima and Pareto-efficient solutions; empirical extension of simulation durations to attenuate statistical error; iterative translation of the search space guided by a user-defined scalar objective (e.g., weighted root-mean-square deviation of target properties to reference data).