Randomized matrix approximation in mixed precision

  • Date:
  • Time: 14:00 - 15:30
  • Address:
    Sokolovská 83, Praha
  • Room: K3
  • Speaker: Ieva Daužickaité

The task of approximating a matrix arrises in various settings, including constructing preconditioners for iterative solvers. Throughout the past decade a lot of effort went into developing randomized techniques that can achieve this goal efficiently. Another rising trend in scientific computing is the use of mixed precision. We thus ask a natural question if we can compute the randomized matrix approximations in mixed precision and obtain high performance and suitable accuracy. In this talk, we consider randomized mixed precision algorithms for approximating a symmetric positive definite matrix, and computing a QR decomposition, and discuss how these can be used in preconditioning Krylov subspace solvers.