2021/06/26 by K. Aghazade, Ali Gholami, Aghazade, Kamal +5 · 1 citation
Earth and Planetary Sciences · Engineering · #FOS: Physical sciences #Geophysical Methods and Applications #Geophysics (physics.geo-ph) #Seismic Imaging and Inversion Techniques #Seismic Waves and Analysis
paper · pdf · doi:10.48550/arxiv.2106.14065
openalex publication_date 2021/06/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The augmented Lagrangian (AL) method provides a flexible and efficient\nframework for solving extended-space full-waveform inversion (FWI), a\nconstrained nonlinear optimization problem whereby we seek model parameters and\nwavefields that minimize the data residuals and satisfy the wave equation\nconstraint. The AL-based wavefield reconstruction inversion, also known as\niteratively refined wavefield reconstruction inversion, extends the search\nspace of FWI in the source dimension and decreases sensitivity of the inversion\nto the initial model accuracy. Furthermore, it benefits from the advantages of\nthe alternating direction method of multipliers (ADMM), such as generality and\ndecomposability for dealing with non-differentiable regularizers, e.g., total\nvariation regularization, and large scale problems, respectively. In practice\nany extension of the method aiming at improving its convergence and decreasing\nthe number of wave-equation solves would have a great importance. To achieve\nthis goal, we recast the method as a general fixed-point iteration problem,\nwhich enables us to apply sophisticated acceleration strategies like Anderson\nacceleration. The accelerated algorithm stores a predefined number of previous\niterates and uses their linear combination together with the current iteration\nto predict the next iteration. We investigate the performance of the proposed\naccelerated algorithm on a simple checkerboard model and the benchmark Marmousi\nII and 2004 BP salt models through numerical examples. These numerical results\nconfirm the effectiveness of the proposed algorithm in terms of convergence\nrate and the quality of the final estimated model.\n