Non-Equilibrium Umbrella Sampling on Blue Gene/P using One-Sided Communication
SESSION: Research Poster Reception
EVENT TYPE: Poster
TIME: 5:15PM - 7:00PM
AUTHOR(S):Alex Dickson, Jeff R. Hammond, Aaron R. Dinner
ABSTRACT: In molecular simulations, important dynamics are often dominated by rare events. In such cases, reaction rates and transition states can be obtained by parallel sampling of phase space by regions. Nonequilibrium umbrella sampling (NEUS) is appropriate for systems driven far from equilibrium.
In this poster we describe the first massively-parallel implementation of NEUS using Global Arrays. Efficient strong-scaling to 8192 processors of BlueGene/P is demonstrated for a coarse-grained, 262-residue RNA segment under explicit flow modeled with stochastic rotation dynamics. This approach can be used to scale explicit-atom molecular dynamics rare event simulations to many petaflops.
Preliminary results show that the flow-extended RNA molecule can visit different meta-stable states, whose relative probabilities depend strongly on the pressure that drives the flow of the solvent. These developments suggest that the parallel version of NEUS will be a powerful tool to examine rare events in large nonequilibrium systems.