By Dr. John M. Chambers (Bell Labs, Lucent Technology)
Thursday, June 17
Abstract: The S language, via the R and S-Plus systems, is currently the medium for implementing a large fraction of new methods for statistics and data analysis. This talk examines programming with S for today’s applications, with the emphasis on a few key ideas—about S, about computing, and about scientific data—that are basic to computing with data. Understanding some of these ideas should make programming for statistical applications more natural and productive.
Domain Decomposition Methods with Convergence Rates Faster than Multigrid
By Dr. Martin J. Gander (McGill University)
Friday, April 16
Abstract: Today most large scale simulations, from aircraft carriers to Jumbo-Jets, are only possible with parallel computers. Often codes are available for partial problems, like the wing or the engine of a plane. A natural paradigm to combine such codes to simulate the entire aircraft is to use domain decomposition techniques. But the performance of these techniques depends very much on the strength of the physical coupling between the pieces of the model. Optimized Schwarz methods are especially designed to weaken this coupling using transmission conditions between the sub-domains which take the physics of the underlying problem into account. They converge often with an order of magnitude less iterations than classical Schwarz methods at the same cost per iteration, and attain contraction rates which are comparable to multigrid methods for Poisson type problems.