Interplay between discretization and algebraic computation in the numerical solution of elliptic PDE problems

The finite element method (FEM) is widely used to discretize variational formulations of partial differential equations. It approximates the exact solution as a linear combination of locally supported basis functions, leading to sparse linear algebraic problems. Solving the linear algebraic problem then ensures the global approximation property of the FEM. The quality of the FEM solution therefore depends not only on the discretization but also on how accurately the algebraic problem is solved. In practice, computational constraints or poor conditioning prevent exact—or even highly accurate—solutions of the algebraic problems. This introduces algebraic errors whose influence must be understood and controlled. Key challenges include error estimation, stopping criteria, and convergence analysis for iterative solvers. In the adaptive FEM, one must additionally consider how the algebraic error is distributed locally in the computational domain. This talk will survey the interplay between discretization and algebraic computation, highlighting open problems and directions for future research.