Nonequilibrium Interface Dynamics:
Theory and Simulation from Atomistic to Continuum Scales

CSIC Building (#406), Seminar Room 4122.
Directions: home.cscamm.umd.edu/directions

Fast Methods for Sampling from a Dynamic pdf
Part II: Very Fast Methods, Additional up-front cost

Dr. Francis Sullivan

Center for Computing Sciences

Abstract:   This pair of talks will cover some basic algorithms for accelerating dynamic Monte Carlo, meaning Monte Carlo for cases in which the probability distribution sampled changes as a consequence of sampling.

We explain how to use a slightly more elaborate tree and a more sophisticated strategy for grouping of the r_i to reduce the cost per step to O(log₂^* M). Here log^* X denotes the number of iterations of X=log X required before X becomes negative. As a point of reference, if D denotes the US national debt in dollars, then log₁₀^* D = 3. In this case, the one-time costs are slightly greater and the cost per step is greatly reduced.

[PRESENTATION SLIDES]