Approximate message passing meets exploration seismology

Data collection, data processing, and imaging in exploration seismology increasingly hinge on large-scale sparsity promoting solvers to remove artifacts caused by efforts to reduce costs. We show how the inclusion of a “message term” in the calculation of the residuals improves the convergence of these iterative solvers by breaking correlations that develop between the model iterate and the linear system that needs to be inverted. We compare this message-passing scheme to state-of-the-art solvers for problems in missing-trace interpolation and in dimensionality-reduced imaging with phase encoding.