Probabilistic Physical Modeling on Distributed Computing Systems

A method is proposed which facilitates express analysis of the results of large distributed simulations of physical systems. The method uses the domain decomposition technique based on the idea of transition probabilities. The proposed approach opens the possibility of conducting parallel simulations of strongly coupled systems on loosely coupled distributed computer platforms, such as grid computing environments. The results of the simulations can then be retrieved in a timely manner for express risk analysis. The case of aerosol dispersion in urban environments was considered, and a prototype solver was developed and tested. Fast response of the system is achieved by replacing complex 3D simulations with information retrieval from aerosol dispersion database. The database was created by extensive simulations of different aerosol dispersion scenarios in a grid computing environment. The reduction of system response from a few days to several seconds can be achieved compared to real-time 3D simulations.