Fermat's principle and nonlinear traveltime tomography.

Fermat's principle shows that a deenite convex set of feasible slowness models { depending only on the traveltime data { exists for the fully nonlinear traveltime inversion problem. In a new iterative reconstruction algorithm, the minimum number of nonfeasible ray paths is used as a gure of merit to determine the optimum size of the model correction at each step. The numerical results show that the new algorithm is robust, stable, and produces very good reconstructions even for high contrast materials where standard methods tend to diverge.