We performed one WEMVA pass with 10 conjugate-gradient solver iterations on the ``WEMVA breaks'' dataset. In this instance the phase differences caused by the velocity anomalies are too large to be satisfactorily approximated in a Born manner. The velocity update thus obtained is presented in Figure . It does not converge toward the true anomalies like the corresponding update in the upper panel of Figure . Since this occurs when using the ideal image perturbation, inaccuracies in extracting the perturbation would only amplify the trend.
There are several possible avenues for avoiding either the Born approximation limitations or the Born approximation altogether: 1. obtaining a very good velocity starting guess with other velocity analysis methods; 2. creating the image perturbation in connection with residual migration Sava and Biondi (2001); 3. employing linearizations other than Born Sava and Fomel (2002); 4. inverting for the lower frequencies first Pratt (1999); 5. using DSO WEMVA Sava and Symes (2002) where the image perturbation is extracted directly from the image using a specific operator during the inversion, without the need to find a corrected image.