New insights into one-norm solvers from the Pareto curve

Several geophysical ill-posed inverse problems are successfully solved by promoting sparsity using one-norm regularization. The practicality of this approach depends on the effectiveness of the one-norm solver used and on its robustness under limited number of iterations. We propose an approach to understand the behavior and evaluate the performance of one-norm solvers. The technique consists of tracking on a graph the data misfit versus the one norm of successive iterates. By comparing the solution paths to the Pareto curve, we are able to assess the performance of the solvers and the quality of the solutions. Such an assessment is particularly relevant given the renewed interest in one-norm regularization.