 |
 |
 |
 | Tomographic full waveform inversion and linear modeling of multiple scattering |  |
![[pdf]](icons/pdf.png) |
Next: Problems with FWI
Up: Conventional Full Waveform Inversion
Previous: Conventional Full Waveform Inversion
The efficient solution of the optimization problem expressed
in equation 1 is performed by gradient based methods,
and thus requires the evaluation of the linear operator
,
which is the linearization of
with respect to velocity perturbations
.
This linear operator can be derived by
perturbing equation 3 as follows
![$\displaystyle \left[ {\bf D_2}- \left({\bf {v}_o}^2 + \delta{{\bf {v}}}^2\right)\nabla^2 \right] \left({\bf P_o}+{\bf {\delta P_o}}\right) ={\bf f},$](img15.png) |
(4) |
where
and
are the background wavefield
and velocity, respectively,
and
is the scattered wavefield.
Equation 4 can be rewritten as the following
two equations:
which represents a nonlinear relationship between
velocity perturbations and scattered wavefield.
To linearize this relationship
we drop the term multiplying the perturbations with each other;
that is, we drop the scattered wavefield
from the right hand side of equation 6
and obtain the following coupled equations:
The linear operator
used to compute
the gradient of the FWI objective function 2
is evaluated by
recursively propagating the background wavefield
and the scattered wavefield
by solving equations 7-8.
The scattered wavefield
is a linear function of the velocity perturbations
because equation 8 takes into account only
fist order scattering.
Notice that the linear operator
is itself a non linear function
of the background velocity, both directly by determining the
propagation speed of the scattered wavefield
(left hand side in equation 8),
and indirectly through the background wavefield
(right hand side in equation 8).
 |
 |
 |
 | Tomographic full waveform inversion and linear modeling of multiple scattering |  |
![[pdf]](icons/pdf.png) |
Next: Problems with FWI
Up: Conventional Full Waveform Inversion
Previous: Conventional Full Waveform Inversion
2012-10-29