Entry Date:
June 5, 2013

Matrix Probing: Randomized Fitting for the Wave-Equation Hessian

Principal Investigator Laurent Demanet

Co-investigators Robert van der Hilst , Alison Malcolm


What can be determined about the pseudoinverse pinv(A) of a matrix A from one application of A to a vector of random entries? A surprising lot, provided pinv(A) is known to belong to a "good" vector subspace. This approach offers a compelling preconditioner for the wave-equation Hessian (normal operator) in seismic imaging. The performance guarantees follow from large deviation estimates of independent interest.