[CIG-SHORT] PyLith: Stresses not transmitted throughout the domain from Drichlet BCs on CUBIT Mesh
Matthew Knepley
knepley at mcs.anl.gov
Tue Oct 20 11:14:34 PDT 2015
On Tue, Oct 20, 2015 at 12:29 PM, Mamun Miah <mmiah at go.olemiss.edu> wrote:
> Hello Brad et al,
>
> Following your tips I was able to run my simulation smoothly for 10 years.
> Now when I increased the same problem for a simulation time of 100 years, I
> am getting SNESSolve convergence error: 'Nonlinear solve did not converge
> due to DIVERGED_MAX_IT iterations 100.' After searching the log file I have
> found the following; not sure if that's the root of this issue.
>
> "Line search: Cubic step no good, shrinking lambda, current gnorm
> 9.815191568334e-05 lambda=1.3655042609323631e-01"
>
For any convergence question I need to see the output from
-snes_view -snes_monitor -ksp_monitor_true_residual
-snes_converged_reason -ksp_converged_reason -snes_linesearch_monitor
Thanks,
Matt
> I have tried your tips in the following threads:
> http://lists.geodynamics.org/pipermail/cig-short/2012-September/001129.html
> http://lists.geodynamics.org/pipermail/cig-short/2014-October/001881.html
> http://lists.geodynamics.org/pipermail/cig-short/2015-July/002174.html
>
> Also, like Matt suggested in this link:
> http://lists.geodynamics.org/pipermail/cig-short/2013-December/001555.html,
> that using a
>
> fs_fieldsplit_0_pc_factor_shift_type nonzero
>
> for Lagrange multipliers weakens the preconditioner. Could it be an issue?
>
> Finally, I have tried various PETSc options from 2015 tutorial but in
> vain. I end up with the same SNESSolve error. I would really appreciate any
> hints or suggestions in this regard.
>
> Attached are the relevant files.
>
> On Thu, Oct 15, 2015 at 5:14 PM, Mamun Miah <mmiah at go.olemiss.edu> wrote:
>
>> Thanks, Matt for the clarification of the zero pivots related to fault.
>> Yep, setting fieldslpit option along with pc_type = lu results in the
>> convergence of the solution. Will make use of the 2015 tutorial like you
>> suggested. Really appreciate Charles's, Matt's, and your help.
>>
>> On Thu, Oct 15, 2015 at 4:48 PM, Brad Aagaard <baagaard at usgs.gov> wrote:
>>
>>> If you tried to use lu with a fault, you will get zero pivots because
>>> the Lagrange multiplier constraints have zero entries on the diagonal.
>>> Using the fieldsplit settings (see examples/3d/tet4/step04.cfg) will remedy
>>> this problem.
>>>
>>> I strongly recommend that you make use of the PyLith tutorial
>>> information posted at https://wiki.geodynamics.org/software:pylith:start.
>>> We have a lot of hints on how to check your solution and debug. From the
>>> 2015 tutorial, you will likely find the slides from Session I and the
>>> presentations for Sessions III and IV most useful.
>>>
>>> Regards,
>>> Brad
>>>
>>>
>>>
>>> On 10/15/2015 04:35 PM, Matthew Knepley wrote:
>>>
>>>> On Thu, Oct 15, 2015 at 6:07 PM, Mamun Miah <mmiah at go.olemiss.edu>
>>>> wrote:
>>>>
>>>> Dear Matt,
>>>>>
>>>>> I have tried with pc_type = lu, which shows zero pivots in the
>>>>> matrices.
>>>>> Did not realize that before setting pc_type = lu. Thanks a lot for
>>>>> telling
>>>>> me that. Then I have tried with pc_type = cholesky, which seems to be
>>>>> working now resulting in convergence of my problem.
>>>>> Thanks to you and Charles again for guiding me in the right direction
>>>>> to
>>>>> investigate the problem further.
>>>>>
>>>>>
>>>> If you have zero pivots, that means your problem is under-constrained,
>>>> and
>>>> you need to fix the model.
>>>>
>>>> Thanks,
>>>>
>>>> Matt
>>>>
>>>>
>>>> On Thu, Oct 15, 2015 at 3:11 PM, Matthew Knepley <knepley at mcs.anl.gov>
>>>>> wrote:
>>>>>
>>>>> On Thu, Oct 15, 2015 at 4:26 PM, Mamun Miah <mmiah at go.olemiss.edu>
>>>>>> wrote:
>>>>>>
>>>>>> Charles,
>>>>>>>
>>>>>>> Thanks again for your suggestion. I have reproduced the geometry and
>>>>>>> improved the mesh quality by checking with 'Allmetrics' in Cubit
>>>>>>> (attached
>>>>>>> below). Condition number is below 2 (1.606) and maximum aspect ratio
>>>>>>> is 3.
>>>>>>> Still I am getting the same convergence errors. I have also tried
>>>>>>> the tips
>>>>>>> found in the threads that Brad answered in the CIG-SHORT:
>>>>>>>
>>>>>>>
>>>>>> Run with LU until you understand the system: --petsc.pc_type=lu
>>>>>>
>>>>>> There is an entire tutorial on the solvers which tells you to do this
>>>>>> first.
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Matt
>>>>>>
>>>>>>
>>>>>>
>>>>>>> http://lists.geodynamics.org/pipermail/cig-short/2014-October/001885.html
>>>>>>>
>>>>>>>
>>>>>>> http://lists.geodynamics.org/pipermail/cig-short/2012-September/001129.html
>>>>>>>
>>>>>>>
>>>>>>> Hex quality, 810 elements:
>>>>>>>
>>>>>>> ------------------------------------
>>>>>>>
>>>>>>> Function Name Average Std Dev Minimum (id) Maximum (id)
>>>>>>>
>>>>>>> ----------------- --------- --------- --------------
>>>>>>> -------------
>>>>>>>
>>>>>>> Aspect Ratio 1.222e+00 6.289e-01 1.000e+00 (1) 3.000e+00 (724)
>>>>>>>
>>>>>>> Skew 8.104e-16 8.536e-16 0.000e+00 (10) 6.821e-15 (733)
>>>>>>>
>>>>>>> Taper 8.561e-16 8.748e-16 0.000e+00 (3) 6.845e-15 (736)
>>>>>>>
>>>>>>> Element Volume 2.500e+07 5.660e+06 9.000e+06 (733) 2.700e+07 (505)
>>>>>>>
>>>>>>> Stretch 9.330e-01 1.895e-01 3.974e-01 (782) 1.000e+00 (10)
>>>>>>>
>>>>>>> Diagonal Ratio 1.000e+00 4.591e-08 1.000e+00 (372) 1.000e+00 (10)
>>>>>>>
>>>>>>> Dimension 1.640e+02 2.602e+01 9.045e+01 (809) 1.732e+02 (315)
>>>>>>>
>>>>>>> Condition No. 1.067e+00 1.907e-01 1.000e+00 (3) 1.606e+00 (737)
>>>>>>>
>>>>>>> Jacobian 2.500e+07 5.660e+06 9.000e+06 (763) 2.700e+07 (315)
>>>>>>>
>>>>>>> Scaled Jacobian 1.000e+00 0.000e+00 1.000e+00 (248) 1.000e+00 (1)
>>>>>>>
>>>>>>> Shear 1.000e+00 0.000e+00 1.000e+00 (248) 1.000e+00 (1)
>>>>>>>
>>>>>>> Shape 9.648e-01 9.963e-02 6.832e-01 (737) 1.000e+00 (1)
>>>>>>>
>>>>>>> Relative Size 7.765e-01 2.288e-01 1.296e-01 (733) 8.573e-01 (487)
>>>>>>>
>>>>>>> Shear And Size 7.765e-01 2.288e-01 1.296e-01 (733) 8.573e-01 (487)
>>>>>>>
>>>>>>> Shape And Size 7.719e-01 2.418e-01 8.854e-02 (733) 8.573e-01 (487)
>>>>>>>
>>>>>>> Distortion 1.000e+00 5.167e-08 1.000e+00 (367) 1.000e+00 (11)
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Not sure what else I should try next. Please also see the attached
>>>>>>> files.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Mamun
>>>>>>>
>>>>>>>
>>>>>>> On Wed, Oct 14, 2015 at 11:11 AM, Charles Williams <
>>>>>>> willic3 at gmail.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>> Dear Mamun,
>>>>>>>>
>>>>>>>> Your linear convergence looks very slow. You could increase the
>>>>>>>> number
>>>>>>>> of iterations, but I’m not sure if that would solve your problem.
>>>>>>>> I would
>>>>>>>> start with a higher quality mesh. Ideally, you would like to have
>>>>>>>> a mesh
>>>>>>>> where the condition number quality measure is less than 2 for the
>>>>>>>> entire
>>>>>>>> mesh, particularly for problems involving fault friction. If this
>>>>>>>> is too
>>>>>>>> difficult to achieve using hexahedral elements, you might need to
>>>>>>>> try
>>>>>>>> tetrahedral elements.
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>> Charles
>>>>>>>>
>>>>>>>>
>>>>>>>> On 15/10/2015, at 6:09 am, Mamun Miah <mmiah at go.olemiss.edu> wrote:
>>>>>>>>
>>>>>>>> Thanks, Charles. Your comments are very helpful. I have been going
>>>>>>>> through each of your comments and discovered that both KSPSolve and
>>>>>>>> SNESSolve do not converge when I uncommented corresponding flag
>>>>>>>> options. So
>>>>>>>> I figured the problem is with the mesh creation. Also set the the
>>>>>>>> normalizer according to the smallest mesh dimension. Attached here
>>>>>>>> is the
>>>>>>>> log file in case you want to see.
>>>>>>>>
>>>>>>>> Thanks again,
>>>>>>>> Mamun
>>>>>>>>
>>>>>>>> On Tue, Oct 13, 2015 at 12:37 PM, Charles Williams <
>>>>>>>> willic3 at gmail.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>> Dear Mamun,
>>>>>>>>>
>>>>>>>>> It is a bit difficult to diagnose your problem without more
>>>>>>>>> information. However, here are several potential problems:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> 1. First of all, the mesh you’re using is not of very high
>>>>>>>>> quality.
>>>>>>>>> The method you’ve used for refinement is pretty crude, and will
>>>>>>>>> never
>>>>>>>>> result in high quality elements. Trelis/Cubit has a number of more
>>>>>>>>> powerful
>>>>>>>>> methods for creating a high quality mesh, and I would take
>>>>>>>>> advantage of
>>>>>>>>> those.
>>>>>>>>> 2. It is unclear why you have set the normalization length scale
>>>>>>>>> to 1
>>>>>>>>> m, since I believe this is much smaller than your smallest element.
>>>>>>>>> 3. There is no indication of whether the linear and nonlinear
>>>>>>>>> solutions are converging, since you have turned off all the
>>>>>>>>> solution
>>>>>>>>> monitoring options, including the journal options. You should
>>>>>>>>> never do
>>>>>>>>> this, as there is then no way to tell what has happened with your
>>>>>>>>> solution.
>>>>>>>>>
>>>>>>>>> You should try running your problem again with all the solution
>>>>>>>>> monitoring options turned back on, and save the results to a log
>>>>>>>>> file.
>>>>>>>>> This will help us diagnose your problem further. I would also try
>>>>>>>>> improving your mesh quality and selecting a normalization length
>>>>>>>>> scale that
>>>>>>>>> is appropriate for your mesh (~equal to the smallest cell
>>>>>>>>> dimension).
>>>>>>>>>
>>>>>>>>> Cheers,
>>>>>>>>> Charles
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 14/10/2015, at 6:26 am, Mamun Miah <mmiah at go.olemiss.edu>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>> Hello Developers,
>>>>>>>>>
>>>>>>>>> I have created the attached geometry (5kmx5kmx4.1km) in CUBIT. It
>>>>>>>>> has
>>>>>>>>> three layer of materials (2000m+100m+200m) and a vertical fault
>>>>>>>>> (along Z
>>>>>>>>> axis) 1000m to the left from the positive X-plane boundary. The
>>>>>>>>> exported
>>>>>>>>> Exodus mesh is then used to run a PyLith simulation with Dirichlet
>>>>>>>>> displacement and displacement rate BCs (e.g. PyLith Tutorial
>>>>>>>>> 'step12.cfg')
>>>>>>>>> on +x and -x sides. The simulation runs fine but even after
>>>>>>>>> running for 100
>>>>>>>>> years I don't see any stress being transmitted to the interior
>>>>>>>>> domain. All
>>>>>>>>> I can see is the stress changes in the boundary elements where the
>>>>>>>>> Dirichlet BCs are applied. I am not sure if it is caused by
>>>>>>>>> improper
>>>>>>>>> meshing or boundary conditions.
>>>>>>>>>
>>>>>>>>> Attached also are the .cfg files, geometry journal file, and stress
>>>>>>>>> output file from simulation. I would highly appreciate any help.
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Thanks & Best Regards,
>>>>>>>>>
>>>>>>>>> Mamun Miah
>>>>>>>>> Graduate Research Assistant
>>>>>>>>> Department of Civil Engineering
>>>>>>>>> The University of Mississippi
>>>>>>>>> Oxford, MS 38655
>>>>>>>>> Phone: (662)-801-1904
>>>>>>>>> Website: http://www.linkedin.com/pub/mamun-miah-eit/21/53/339
>>>>>>>>> <geometry.jpg><application.cfg><pylithapp.cfg><visit0000.png>
>>>>>>>>> <application.jou>_______________________________________________
>>>>>>>>> CIG-SHORT mailing list
>>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Charles A. Williams
>>>>>>>>> Scientist
>>>>>>>>> GNS Science
>>>>>>>>> 1 Fairway Drive, Avalon
>>>>>>>>> PO Box 30368
>>>>>>>>> Lower Hutt 5040
>>>>>>>>> New Zealand
>>>>>>>>> ph (office): 0064-4570-4566
>>>>>>>>> fax (office): 0064-4570-4600
>>>>>>>>> C.Williams at gns.cri.nz
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> CIG-SHORT mailing list
>>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Thanks & Best Regards,
>>>>>>>>
>>>>>>>> Mamun Miah
>>>>>>>> Graduate Research Assistant
>>>>>>>> Department of Civil Engineering
>>>>>>>> The University of Mississippi
>>>>>>>> Oxford, MS 38655
>>>>>>>> Phone: (662)-801-1904
>>>>>>>> Website: http://www.linkedin.com/pub/mamun-miah-eit/21/53/339
>>>>>>>> <log>_______________________________________________
>>>>>>>> CIG-SHORT mailing list
>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>
>>>>>>>>
>>>>>>>> Charles A. Williams
>>>>>>>> Scientist
>>>>>>>> GNS Science
>>>>>>>> 1 Fairway Drive, Avalon
>>>>>>>> PO Box 30368
>>>>>>>> Lower Hutt 5040
>>>>>>>> New Zealand
>>>>>>>> ph (office): 0064-4570-4566
>>>>>>>> fax (office): 0064-4570-4600
>>>>>>>> C.Williams at gns.cri.nz
>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> CIG-SHORT mailing list
>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Thanks & Best Regards,
>>>>>>>
>>>>>>> Mamun Miah
>>>>>>> Graduate Research Assistant
>>>>>>> Department of Civil Engineering
>>>>>>> The University of Mississippi
>>>>>>> Oxford, MS 38655
>>>>>>> Phone: (662)-801-1904
>>>>>>> Website: http://www.linkedin.com/pub/mamun-miah-eit/21/53/339
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> CIG-SHORT mailing list
>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> What most experimenters take for granted before they begin their
>>>>>> experiments is infinitely more interesting than any results to which
>>>>>> their
>>>>>> experiments lead.
>>>>>> -- Norbert Wiener
>>>>>>
>>>>>> _______________________________________________
>>>>>> CIG-SHORT mailing list
>>>>>> CIG-SHORT at geodynamics.org
>>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Thanks & Best Regards,
>>>>>
>>>>> Mamun Miah
>>>>> Graduate Research Assistant
>>>>> Department of Civil Engineering
>>>>> The University of Mississippi
>>>>> Oxford, MS 38655
>>>>> Phone: (662)-801-1904
>>>>> Website: http://www.linkedin.com/pub/mamun-miah-eit/21/53/339
>>>>>
>>>>> _______________________________________________
>>>>> CIG-SHORT mailing list
>>>>> CIG-SHORT at geodynamics.org
>>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> CIG-SHORT mailing list
>>>> CIG-SHORT at geodynamics.org
>>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>
>>>>
>>> _______________________________________________
>>> CIG-SHORT mailing list
>>> CIG-SHORT at geodynamics.org
>>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>
>>
>>
>>
>> --
>> Thanks & Best Regards,
>>
>> Mamun Miah
>> Graduate Research Assistant
>> Department of Civil Engineering
>> The University of Mississippi
>> Oxford, MS 38655
>> Phone: (662)-801-1904
>> Website: http://www.linkedin.com/pub/mamun-miah-eit/21/53/339
>>
>
>
>
> --
> Thanks & Best Regards,
>
> Mamun Miah
> Graduate Research Assistant
> Department of Civil Engineering
> The University of Mississippi
> Oxford, MS 38655
> Phone: (662)-801-1904
> Website: http://www.linkedin.com/pub/mamun-miah-eit/21/53/339
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>
--
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/cig-short/attachments/20151020/f05a6ab8/attachment-0001.html>
More information about the CIG-SHORT
mailing list