Multiscale Geometric Integration of
Deterministic and Stochastic Systems
Molei Tao, CIMS
Abstract:
As part of a continuous effort in developing upscaling methods
for accelerated numerical simulations, we develop multiscale geometric
integrators. These integrators employ coarse steps that do not resolve
the fast scale in the system; nevertheless, they capture the effective
contribution of the fast dynamics --- in fact, accuracies are
demonstrated in a sense called two-scale flow convergence. These
integrators works for a broad class of systems, including stiff ODEs,
SDEs and PDEs, by not requiring an identification of underlying slow
variables or processes. They also numerically preserve intrinsic
geometric structures (e.g., symplecticity, conservation laws, and
invariant distribution), which not only lead to improved long time
accuracy, but also a possibility to sample statistical distributions
via dynamics. These new properties are due to a new philosophy based on
averaging flow maps instead of vector fields.