Fast or Accurate Algorithms for Eigenvalue Problems

Professor James Demmel (UC Berkeley)

We discuss the fastest algorithms now available for computing the symmetric eigendecomposition or SVD, as well as faster algorithms that we expect to be available soon. We discuss other algorithms designed to be much more accurate for computing tiny eigenvalues, which are needed in some quantum mechanics and finite element problems. We use tools ranging from the fast multipole method to the combinatorial structure of totally positive matrices.