Fast \(l_ p\) solution of large, sparse, linear systems: Application to seismic travel time tomography (Q1098560)

This is an interesting, well written paper, on some nontrivial numerical experiments for solving large, sparse, rectangle linear systems. The systems considered are those that appear in computerized seismic tomography for imaging complex velocity structures using travel times picks of common shot, pre-stacked vertical seismic profiling or reflection data. This type of application is plagued by a number of difficulties, in addition to the obvious ones due to their size, specially when traditional least squares technique are used. The authors explore in detail the use of conjugate gradient methods as an outer iteration, for the so called iteratively reweighted least squares algorithm for computing \(\ell_ 1\) solutions. This has been shown by several authors to be a good remedy to the effect of bad observations (outliers), and these authors show convincingly its stabilizing effect on this problem, as compared to more traditional least squares techniques. The main advantage over similar remedies, like damped least squares, is that no special parameters need be adjusted in order to obtain a stable solution.