UC San Diego

Force fields are widely used computational tools in molecular modeling, providing predictive powers in studying molecular structures, properties, and dynamics. These mathematical functional forms are simplified but effective methods for calculating the interactions between atoms in a molecular system. Traditional additive force fields describe the potential energy of a system by considering bonded interactions and nonbonded interactions. The bonded interactions include bond stretching, angle bending, and torsional rotation, and the nonbonded interactions include van der Waals interactions and electrostatic interactions. Force Fields enable molecular dynamics simulations to study the behaviors of molecules, providing insights into drug design, materials science, and biochemistry. The development of force fields has progressed to address increasing complexity in molecular systems, incorporating electronic polarization, automatic parameterization, bespoke fitting, and enhanced computational efficiency. In molecular design, force fields allow for the exploration of large chemical space, optimization of molecular conformations, and fast prediction of molecular properties, facilitating the design of novel compounds with desired characteristics. In this dissertation, I aim to develop new methods to improve force field accuracy and advance the balance between computational efficiency and accuracy. I examine several key aspects that influence the accuracy of force fields, focusing on the training data, parameterization methods, and explicit treatment of electronic polarization. First, I show a fast and efficient polarizable electrostatics model for molecular dynamics simulation. Second, I propose a new polarizable water model for improved solvent effects with moderate computational cost. Third, I develop non-bonded force field parameters accompanying the electrostatics model for simulations for physical properties. Furthermore, I explore methods to improve force field accuracy for physical property predictions in metal-organic frameworks, proposing a data-driven parameterization approach to guide the design of functional framework structures. In summary, these advancements aim to improve the accuracy of force fields for molecular modeling that effectively guides molecular design.