Date:

Monday , September 17, 2001

Speaker:

Mark Fitzgerald

Affiliation:

Mathematics Department, University of Colorado, Denver

e-mail:

fitz@math.cudenver.edu

Title:

Accelerated Convergence in Particle Dispersion Codes

Abstract:

Atmospheric scientists use random-walk models to simulate the dispersion of aerosols under varying meteorological conditions. When these simulation codes are used to estimate rare events, such as aerosol concentration at a certain point far from the primary plume, the simulation is highly inefficient, since much time is wasted simulating particles that are not of interest. Adaptive importance sampling can be used to generate "biased" particle paths that lead to increased efficiency in estimation of rare events. In this talk, I'll introduce the concept of importance sampling, show its extension to stochastic processes, and demonstrate an adaptive learning algorithm in a particle dispersion model.