A Dual Potential Approach for Modeling Structure and Energetics with Coarse-Grained Models
Open Access
- Author:
- Lebold, Kathryn
- Graduate Program:
- Chemistry
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 28, 2019
- Committee Members:
- Will Noid, Dissertation Advisor/Co-Advisor
Will Noid, Committee Chair/Co-Chair
Lasse Jensen, Committee Member
Mark Maroncelli, Committee Member
Adrianus C Van Duin, Outside Member - Keywords:
- coarse-graining
dual-potential
transferability
representability
thermodynamics - Abstract:
- All-atom (AA) computer simulations provide powerful insight, as they describe molecular structure and interactions at atomic resolution without the spatial or temporal averaging that are inherent to many experimental methods. Unfortunately, the computational expense of AA models severely limits their utility for modeling many phenomena of interest. Consequently, simulation studies of soft materials often adopt coarse-grained (CG) models that “integrate out” AA degrees of freedom that are not essential for describing the phenomena of interest. Because CG models represent the system with reduced detail, each simulation time step requires less computational effort. Moreover, because CG models eliminate high-frequency vibrations and the “friction” associated with atomic interactions, they allow larger time steps when propagating dynamics on the softened free energy surface. However, the exact CG potential, i.e., the potential of mean force (PMF), which describes all of the structural and thermodynamic information about the system that is observable at the CG resolution, depends on the temperature and density. Therefore, approximate CG potentials are generally not transferable to thermodynamic state points other than the state point for which they were parameterized. This is commonly known as the transferability problem. The state point dependence of CG models also complicates the treatment of thermodynamic properties, which is known as the representability problem. Indeed, CG models often provide a poor description of the thermodynamic pressure and energy. We propose to simultaneously address the representability and transferability problems by precisely addressing energetic and entropic contributions to the CG PMF. Specifically, we propose to approximate the PMF via conventional methods (such as iterative Boltzmann inversion and force-matching methods). We propose to sample configuration space with the resulting structure-based potential. Conversely, we propose to approximate the energetic contribution via an “energy-matching” method. The resulting energy function should accurately reproduce the atomic energetics at the resolution of the CG model, i.e., it should approximate averages of the atomistic potential energy conditioned upon the CG configuration. Moreover, by comparing the two approximate potentials, we propose to infer an approximation for the entropic contribution to the PMF, which provides a predictive estimate for the temperature-dependence of the structure-based potential. Therefore, via this “dual-potential” approach, we propose to construct CG models that accurately describe the structure and energetics of the AA model. In this dissertation, I first present numerical studies of the temperature- and density-dependence of conventional structure-based potentials. I describe the theory and first implementation of the dual potential approach for both the constant NVT and constant NPT ensembles. Next, I apploy the dual-potential approach to investigate the resolution-dependence of, as well as energetic and entropic contributions to, effective potentials for implicit solvent models. Finally, I present closing comments and outline potential future directions.