[CIG-SHORT] A Question About Pylith

Fri Nov 18 05:40:02 PST 2016

On Fri, Nov 18, 2016 at 5:32 AM, zeynep yılmaz <ylmz.zeynep at gmail.com>
wrote:

> Thank you very much for taking time to respond my questions and for your
> detailed explanation. I am new at Pylith, your response helped me so much.
>
> I have implemented a model with the dimensions of 150 km for a 20x10 km
> fault plane as you have suggested. I also increased the resolution of the
> mesh. However, the linear solve converged after 1700 iterations. I would
> also like to ask you how I can choose the number of iterations after which
> Gram-Schmidt orthogonalization is restarted (ksp_gmres_restart).
>

--petsc.ksp_gmres_restart=500

You can also look at the solver slides

  https://wiki.geodynamics.org/software:pylith:cdm2016    at the bottom

 to try and improve that number I would hope you can keep it under 100.

  Thanks,

     Matt

> Thanks a lot again.
>
> Best wishes and kind regards,
> Zeynep
>
> On Mon, Nov 14, 2016 at 11:27 PM, Charles Williams <willic3 at gmail.com>
> wrote:
>
>> Dear Zeynep,
>>
>> I am also sending my response to the cig-short mailing list.  For future
>> requests, could you please make use of the mailing list?  This helps both
>> by reducing the amount of e-mails we need to reply to, and it also helps
>> everyone on the mailing list by answering questions that others might
>> have.  You can join this mailing list at:  https://geodynamics.org/cig/a
>> bout/mailing-lists/.
>>
>> In response to your question, there are several issues involved when
>> comparing Okada’s analytical solution to finite element results:
>>
>> 1.  The finite element solution and the analytical solution are modeling
>> two different things, so the solutions will never agree completely.  The
>> Okada solution represents displacements due to slip on rectangular patches
>> with constant slip on each one, while the finite element solution
>> represents slip at vertices that decreases either linearly or bilinearly
>> (depending on element type) to zero at the surrounding vertices.
>> Obviously, the finite element solution is not capable of representing slip
>> that varies over an infinitesimal distance (and this is actually not
>> realistic in any case).
>>
>> 2.  If you want to accurately represent an analytical solution, you must
>> make sure that you have sufficient mesh resolution, particularly in regions
>> where the solution is changing most rapidly.  Looking at the meshes you
>> provided, this does not seem likely to be the case.  Your meshes are
>> somewhat graded, but do not appear to have sufficient resolution in the
>> vicinity of the fault.  Furthermore, in relation to point #1 above, if you
>> want to accurately represent a constant-slip rectangular patch, you would
>> need extremely high resolution around the fault edges.
>>
>> 3.  Another issue when representing an analytical solution is insuring
>> that the mesh boundaries are sufficiently far from the deformation source
>> that they are not influencing the solution.  Again, this does not seem to
>> be the case for your mesh.  As a general rule of thumb, the distance to the
>> mesh boundaries should be at least a factor of 3 larger than the dimensions
>> of the deformation source.
>>
>> A final question is to determine why it is important to compare the
>> results against the analytical solution.  If it is important to do this,
>> then you probably want to evaluate both the analytical solution and
>> numerical solution at an identical set of points and integrate the error
>> over the volume.  Some of these questions were explored quite some time
>> ago.  You can find some old benchmarks here:
>>
>> https://geodynamics.org/cig/working-groups/short-term-crusta
>> l-dynamics/benchmarks/
>>
>> Unfortunately, these benchmarks are extremely old and have not been
>> updated for current versions of PyLith.
>>
>> Good luck with your modeling, and please contact the cig-short mailing
>> list if you need more help.
>>
>> Cheers,
>> Charles
>>
>>
>>
>> On 15/11/2016, at 1:55 AM, zeynep yılmaz <ylmz.zeynep at gmail.com> wrote:
>>
>> Dear Dr. Charles Williams,
>>
>> I am working on calculation of static displacements for a reverse buried
>> fault (with a dip of 70 degrees) by using Pylith Software. I would like to
>> ask you how I can compare the accuracy of the FEM models (different size
>> models attached below) with the Okada's analytical solution. Is there an
>> easy way to compare the solutions by using Pylith? I have used Coulomb 3
>> software for Okada's solution, however it takes time to adapt. I have
>> also downloaded Cigma package to calculate the erors. However I did not
>> understand well if it is comparing the results with Okada's solution.
>>
>> If you could have time to answer my question, I would be so glad. Thanks
>> in advance.
>>
>> Best Regards,
>> Zeynep Yilmaz
>>
>> <model1_40x40x50.vtk><model2_80x80x100.vtk>
>>
>>
>>
>> *Charles Williams I Geodynamic ModelerGNS Science **I** Te Pῡ Ao*
>> 1 Fairway Drive, Avalon 5010, PO Box 30368, Lower Hutt 5040, New Zealand
>> *Ph* 0064-4-570-4566 I *Mob* 0064-22-350-7326 I *Fax* 0064-4-570-4600
>> *http://www.gns.cri.nz/* <http://www.gns.cri.nz/> *I* *Email: *
>> *C.Williams at gns.cri.nz* <your.email at gns.cri.nz>
>>
>>
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/cig-short/attachments/20161118/e477d77c/attachment-0001.html>