Title: Properties of discrete adjoints for adaptive models

Authors: Adrian Sandu (Virginia Tech)

In this talk we discuss the properties of discrete adjoints for models that use time step and grid adaptation to control forward numerical errors.

We show how the differentiation of the time step control mechanism can destroy the consistency of the discrete adjoint, and propose ways to overcome this.

For adaptive grids, we show that discontinuous Galerkin provides a flexible framework for both h- and p-refinement, while preserving dual consistency. Apriori and aposteriori error estimates for the solution of inverse problems treated in a fully discrete setting are discussed.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: May 27 2011