COMPRESSED FWI INVERSION WITH NON-MONOTONE LINE SEARCH LBFGS

Full waveform inversion (FWI) is a high quality seismic imaging method. It is a nonlinear inversion problem which usually needs the monotone line search method to be solved. However, the speed of convergence for such a simple search technique is relatively slow. In this paper, we combine the non-monotone line search technique with the LBFGS method and apply them to the frequency-domain FWI. We test this new method on a two-dimensional Marmousi model. The results show that the method is robust. Comparing with the monotone line search method, the new method could improve the convergence rate of FWI. We also test the new method with a two-dimensional conventional streamer data set and the results show some improvements compared with the conventional FWI method.