A linearized Bregman method for compressive waveform inversion

We present an implementation of a recently developed and relatively simple linearized Bregman method to solve the large-scale $\ell_1$-norm sparsity-promoting Gauss-Newton (GN) full-waveform inversion (FWI) problem. Numerical experiments demonstrate that: the much simpler linearized Bregman method does a comparable and even superior job compared to a state-of-the-art $\ell_1$-norm solver SPGL1, which is previously used in the modified Gauss-Newton FWI; the linearized Bregman method is also more efficient (faster) than SPGL1; the FWI result with the linearized Bregman method solving $\ell_1$-norm sparsity-promoting problems to get the model updates is better than that obtained by solving $\ell_2$-norm least-squares problems to get the model updates.

https://slim.gatech.edu/Publications/Public/Conferences/SEG/2016/chai201...