[CIG-SEISMO] time reversed computation in specfem2d
Chris Morency
cmorency at Princeton.EDU
Wed Apr 29 07:53:37 PDT 2009
Hi Dimitri and Jörg,
Yes, it being done in the weeks to come is definitively my priority.
Running a type of benchmark with Jörg's code sounds great. I will keep
in touch.
Thanks,
Christina
Dimitri Komatitsch wrote:
>
> Hi Christina,
>
> No problem at all, doing it in the next few weeks or months
> is fine! Jörg can probably help for the tests since he had planned to
> develop an adjoint code from scratch anyway. Running a few tests with
> an existing code will be much faster.
>
> Thanks,
> Dimitri.
>
> Chris Morency wrote:
>> Hello Dimitri,
>>
>> I feel it is actually my bad ! As you know, we had discussed the
>> merging of the 2D version we have been developing including
>> poroelasticity. This version does have the adjoint capabilities for
>> acoustic, elastic, and poroelastic media. It is still on a branch....
>> I am finishing dealing with some domain decomposition issues when
>> using irregular (CUBIT generated) meshes. It shouldn't be too long
>> and then I should be able to fully merge this new version in place of
>> the present specfem2D.
>> Really sorry about that. I am presently working on it.
>>
>> Christina
>>
>> Dimitri Komatitsch wrote:
>>>
>>> Dear Jörg and Carl,
>>>
>>> This has been done already, by Carl Tape et al. from Caltech (USA).
>>> See for instance their papers below and in PDF at
>>> http://www.gps.caltech.edu/~carltape/research/CV.html#Pubs
>>>
>>> I thought the kernel capability and adjoint runs had been added to
>>> SPECFEM2D already (on the SVN server). Carl, did you do it?
>>> We talked about it (merging your kernels in the SVN code) two years ago
>>> and I thought that it had been done since then. If not, would it be
>>> easy to do? It would clearly be very useful. As always, working on a
>>> local version not under SVN control was a bad idea to begin with...
>>> because now it is a dead branch... could somebody (maybe Jörg) cut
>>> and paste your adjoint routines in the SVN code? (if you sent him a
>>> copy of your old 2D code)
>>>
>>> In the future let us all use SVN all the time from the beginning of
>>> any new SPECFEM project.
>>>
>>> Thanks,
>>> Cheers,
>>>
>>> Dimitri.
>>>
>>> Tape, C., Liu, Q., and J. Tromp, 2007, Finite-frequency tomography
>>> using adjoint methods---Methodology and examples using membrane
>>> surface waves, Geophysical Journal International, v. 168, 1105-1129.
>>>
>>>
>>> Tromp, J., Tape, C., and Q. Liu, 2005, Seismic tomography, adjoint
>>> methods, time reversal, and banana-doughnut kernels, Geophysical
>>> Journal International, v. 160, p. 195-216.
>>>
>>>
>>> pge05bqw at studserv.uni-leipzig.de wrote:
>>>> Hello,
>>>> i'am trying to add a time reversed computation to specfem2d. First
>>>> I thought it would be enough to change sign of deltat(and
>>>> deltatcube...) after some time iteration steps(e.g. half of
>>>> NSTEP). But it doesn't seem to be that easy, because it leads to a
>>>> extremly big maximum norm of vector field, which aborts the program.
>>>> Does anyone have an idea how this could work?
>>>> Thanks,
>>>> Jörg Buchwald
>>>> Student
>>>> University of Leipzig
>>>>
>>>> ----------------------------------------------------------------
>>>> This message was sent using IMP, the Internet Messaging Program.
>>>>
>>>> _______________________________________________
>>>> CIG-SEISMO mailing list
>>>> CIG-SEISMO at geodynamics.org
>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo
>>>>
>>>
>>
>
--
Princeton University
Department of Geosciences
Room 311, 114 Guyot Hall
Princeton, New Jersey 08544-1003
Phone: 609-258-2393
Fax: 609-258-2593
Web: http://www.princeton.edu/~cmorency/
More information about the CIG-SEISMO
mailing list