[CIG-SEISMO] SPECFEM3D inversion workflow

[Carl Tape]

Dear Alan,

You are looking in the right place on cig/svn. Within ADJOINT_TOMO the
components for making the model update are there; however, there are no
explicit rules for computing the model update, just as with any optimization
problem. There are options for computing a conjugate gradient step, as well
as a source subspace projection step. The basic tools of loading slices from
kernels and performing basic manipulations are there. First look under
ADJOINT_TOMO/iterate_adj/ (and the README). Just now I renamed the
subdirectory "pangu" to "cluster" to be more clear about what is inside (so
"svn update"). Inside "cluster" there are several operations that might be
needed for computing the update. These were used for the study of
Tape-Liu-Maggi-Tromp (2009,2010), and have also been used by Daniel Peter
(middle east) and Hejun Zhu (Europe).

You are right, though -- the workflow could use substantial "stitching
together." I/We would welcome the kinds of improvements that you are
proposing. Perhaps check out what components are there, then move forward.
Let me/us know if you have questions.

The best single file to read is probably:

ADJOINT_TOMO/iterate_adj/cluster/README

Carl

---------------------------------------------------------------
Carl Tape
Assistant Professor
Geophysical Institute
University of Alaska Fairbanks
Phone: 907-474-5456
Email: carltape at gi.alaska.edu
Web: http://www.giseis.alaska.edu/input/carl/
---------------------------------------------------------------


On Thu, Aug 11, 2011 at 7:43 AM, Alan Schiemenz <
schiemenz at geophysik.uni-muenchen.de> wrote:

> Hello,
>
> I am using SPECFEM3D as the central component of a workflow for a FWI of
> an industrial data set.  I understand well to the point of computing
> sensitivity kernels.  From there I must compute model updates from the
> computed kernels, but this procedure is not so clear.  I see
> the /seismo/3D/ADJOINT_TOMO folder in the svn, so gather there are
> routines there which can help.  Are there examples, or instructions
> available for how to compute a model update?
>
>
> Moreover I would be interested in having more direct control of the
> workflow, and so would like to write my own Python routines to do the
> necessary linear algebra.  Are there examples available on loading and
> manipulating the binary kernel computations (either from
> DATABASES_MPI/*kernel.bin or else OUTPUT_FILES/*kernel.mesh from the
> simulation directories) ?
>
>
> Thank you for helping,
> Alan Schiemenz
>
>
> _______________________________________________
> CIG-SEISMO mailing list
> CIG-SEISMO at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://geodynamics.org/pipermail/cig-seismo/attachments/20110811/0f75cb04/attachment.htm