Additive preconditioning for matrix computations

Our randomized additive preconditioners are readily available and regularly facilitate the solution of linear systems of equations and eigen-solving for a very large class of input matrices. We study the generation of such preconditioners and their impact on the rank and the condition number of a matrix. We also propose some techniques for their refinement and two alternative versions of randomized preprocessing. Our analysis and experiments show the power of our approach even where we employ weak randomization, that is generate sparse and structured preconditioners, defined by a small number of random parameters.