| dc.description.abstract | Phase transitions (i.e. transitions from one state to another) play a crucial role in most natural processes, as well as industrial operations such as evaporation. Molecular simulation has emerged as an important tool for examining phase transitions, providing a means to directly probe the transitions, in particular by means of calculating the free energy landscape. However, calculation of the free energy remains a challenge for traditional molecular dynamics (MD) or Monte Carlo (MC) approaches. This thesis focuses on the application and development of advanced sampling methods based on direct calculation of the density of states (DOS), from which thermodynamic properties such as free energy, heat capacity, and thus phase transition behavior, can be determined.
First, we examine the phase transition behavior of self-assembling lipids, using the Wang-Landau (WL) MC method. We find that the WL method can provide a complete view of phase behavior of self-assembly lipids as a function of temperature, while the traditional Metropolis MC tends to miss intermediate phase transitions, providing considerable insight into self-assembly process. Next, we implement the statistical temperature MD (STMD) method, an extended version of WL sampling combined with MD movement, for massively multicore GPU architectures to enable rapid calculation of the DOS. We demonstrate near identical results to WL, fully resolving the lipid phase behavior, but at a fraction of the computational cost. With the STMD method, phase transitions of nano-confined fluids are further determined as a function of pore height and wall-fluid interactions, providing a more clear understanding of the transition from a fluid state to a non-lubricating solid phase.
Finally, we present development of a novel algorithm for 2-dimensional DOS calculation, by creating a hybrid of the WL and STMD methods, verified by examination of binary Lennard-Jones fluids. This novel approach enables STMD to be used with other thermodynamic ensembles beyond the canonical ensemble via WL moves, while still retaining the efficiency of STMD and ability to examine large system sizes. | |
