Titel: Next generation time integration methods
Zeit: Donnerstag, 7.2.2019, 14:00 Uhr
Ort: Seminarraum 0.02-142 am ZISC (Casa Huber), University of Erlangen
Vortragender: Martin Schreiber
(Informatik I10 : Computer Architecture and Parallel Systems,)
Abstract: Scientific computing faces new challenges due to changes in computer architectures caused by physical limitations. This led to research on partly disruptive mathematical and algorithmic reformulations of simulations, e.g. using parallelism in the time dimension.
This presentation provides an overview and introduction to the variety of recently developed and evaluated time integration methods with the focus on atmospheric models, all aimed at improving the ratio of wall clock time to error: Rational approximations of exponential integrator methods in their various forms, Semi-Lagrangian (SL) methods in combination with the Parareal parallel-in-time algorithm, multi-level time integration of spectral deferred correction (ML-SDC) and, based on this, the “Parallel Full Approximation Scheme in Space and Time” (PFASST).
All studies were mainly conducted with the shallow water equations (SWE) on the f-plane and the rotating sphere. These equations are commonly used to investigate horizontal numerical aspects of dynamical cores for weather and climate simulation. Overall, our results motivate further investigation and combination of these methods not only for operational weather/climate systems, but for other types of applications from the area of scientific computing.
(With contributions and more from Jed Brown, François P. Hamon, Richard Loft, Matthew Normile, Michael Minion, Nathanaël Schaeffer, Andreas Schmitt, Pedro S. Peixoto)
Kontaktperson: Martin Schreiber