Title: A New Paradigm for Petascale Monte Carlo Simulation*
Speaker: David P. Landau (Center for Simulational Physics The University of Georgia, Athens, GA 30602, USA)
Date/Time: September 16, 2015, Wednesday at 13:40
Place: Sabancı University, FENS G035
While novel Monte Carlo algorithms are revolutionizing our ability to determine the properties of diverse models in statistical physics, the move to massively parallel computers is introducing new challenges to the researcher. After reviewing the powerful Wang-Landau sampling method that iteratively determines the density of states for a system by performing a random walk in energy space, we introduce a simple, generic, Replica Exchange Wang-Landau (REWL) algorithm. The REWL algorithm is naturally suited to implementation on massively parallel, petaflop supercomputers but is also efficient on multicore laptops. This new approach introduces a replica-exchange framework involving densities of states that are determined iteratively for overlapping windows in energy space, each via traditional Wang-Landau sampling. The advantages and general applicability of the method are demonstrated using thousands of cores for several quite different systems (possessing either discrete or continuous degrees of freedom) including those with complex free energy landscapes and topological constraints.
* Research was supported by the National Science Foundation