Matrix probing and simultaneous sources: a new approach for preconditioning the Hessian

Recent advances based on the mathematical understanding of the Hessian as, under certain conditions, a pseudo-differential operator have resulted in a new preconditioner by L. Demanet et al. Basing their approach on a suitable basis expansion for the Hessian, by suitably 'probing' the Hessian, i.e. applying the Hessian to a small number of randomized model perturbations, one can obtain an approximation to the inverse Hessian in an efficient manner. Building upon this approach, we consider this preconditioner in the context of least-squares migration and Full Waveform Inversion and specifically dimensionality reduction techniques in these domains. By utilizing previous work in simultaneous sources, we are able to develop an efficient least-squares migration scheme which recovers higher quality images and hence higher quality search directions in the context of a Gauss-Newton method for Full Waveform Inversion while simultaneously avoiding inordinate amounts of additional work.

https://slim.gatech.edu/Publications/Public/Conferences/EAGE/2012/dasilv...