 Goal error (left), effectivity indices (middle), and number of necessary algebraic solver iterations (right), 10e−8 relative residual precision versus adaptive stopping criteria, 3D vertex singularity |
Main results:
- quantity of interest given by a volumetric (source) or a surface (flux) term;
- same discretization of both primal and dual problems;
- inexact algebraic solvers for both primal and dual problems;
- guaranteed upper bound on the goal error;
- unified framework covering basically any numerical method (conforming, nonconforming, and discontinuous Galerkin finite elements, finite volumes).
Details: paper (preprint) with Gouranga Mallik and Soleiman Yousef.
