(c) Vladimir Kazei and Oleg Ovcharenko (KAUST, 2019)
Helmholtz solver is a submodule, so clone with: git clone --recursive https://github.com/vkazei/fastFWI.git
Standard L2 full-waveform inversion
||D(m)-D_obs||^2 + F_reg(m_0, m) is minimized
where
D_obs is the observed data
D(m) is the data modeled in the model m
m is current model
m_0 is the starting model
F_reg is the regularizing functional
generic FWI implementation (misfit functional and gradient) is based on SimpleFWI by Tristan van Leeuwen LBFGS-B-C optimization implementation by Stephen Becker
We add regularizing functional equivalent to Sobolev space norm for
F_reg = ||m||W1p.
Kazei et al., 2017
It can also be applied as a regularizing
We demonstrate an application of FWI to Marmousi II benchmark velocity model. We initiate inversion from a linear velocity model and search for updates of model parameters by L-BFGS.