F = Born Modeling Operator
A = 2-D PEF (Prediction error filter)
Three images:
- Marmousi true model
- Marmousi estimated without PEF: L2 norm minimization of r(m) = Fm-d.
- Marmousi estimated with 2-D PEF: L2 norm minimization of r(m)=A(Fm-d).
Optionally, click on image to enlarge.
While iteration brings the residual downward, it brings the estimated model upward. Without the PEF only 18% of the model is recovered. The PEF brings it up to 31% of the true model. Why not 100%? In principle, the number of iterations required is 121x369=44,649.
Antoine Guitton did all the computing from August 14 to September 14, 2018
Jon Claerbout struggled over the theory and presentation since finishing his 2014 textbook (GIEE).