SEISMIC INVERSION WITH POOR INITIAL GUESS

We propose a general inversion procedure based on the Taylor series ex pansion of the data vector with respect to the model parameters. If it is possible to make a good initial guess, then the linearized iterati ve least-squares optimization methods can be applied. Otherwise, the h igh-order perturbations are incorporated into such methods without the ir special modification. As additional constraints, the results of the factorized inversion based on the sensitivity analysis without SVD ar e employed. Such an analysis is used to solve the ambiguity and singul arity problems as well. Simple numerical examples show that this non-l inear inversion is much better than the first-order approximation for the same number of iterations and that it is not time-consuming becaus e of the fast convergence of the iteration process. The proposed itera tive scheme does not break down when the initial guess is ''far'' from the solution.