University at Buffalo - The State University of New York
Skip to Content

Varis Carey

Varis Carey

Adjoint Methods for Uncertainty Quantification Workshop

Wednesday, April 12th
9:00 AM - 11:00 AM
Student Union, Room 210

Abstract:  We present an interactive introduction to adjoint methods for uncertainty quantification(UQ), using a mixture of jupyter notebooks and virtual machines.  Starting from basic concepts of goal-based error estimation and sensitivity analysis for linear PDEs, we examine extensions to adaptivity, coarse-scale adjoints, and nonlinear problems In addition, we discuss differences between continuous and discrete adjoint fomulations, especially within the context of multiscale/multiphysics problems.  We further discuss emerging methods in UQ that leverage adjoint information, including active subspaces and gradient-enhanced surrogate models.  We also discuss complications and limitations, such as adjoints for turbulent simulations, and dealing with time reversal in time-dependent PDE varis.carey@ucdenver.edu