[CIG-SHORT] Pylith "zero-pivot" error
Ravi Kanda
rkanda at gps.caltech.edu
Tue Jan 29 13:36:31 PST 2008
Matt,
Thanks for the suggestions on PETSc options usage! Option (2) below did the
trick, and Pylith ran without any PETSc errors this time. I did a quick check
with Paraview and the output now seems to be making sense.
Regards,
Ravi.
-------------------------------------------------------------------------------
Matthew Knepley wrote:
> On Jan 28, 2008 9:49 PM, Ravi Kanda <rkanda at gps.caltech.edu> wrote:
>> Brad,
>>
>> Thanks, the debugging suggestion was very helpful. Pylith now "understands" my
>> CFG files.
>>
>> But now, the run fails with the "Detected zero pivot in LU factorization" error
>> (see attached screen output log). After checking the PETSc documentation page,
>> I tried running with the following in my PYLITHAPP.CFG file:
>> ---------------------------------------------
>> pc_factor_shift_nonzero = true
>> #pc_factor_shift_positive_definite = true
>> ---------------------------------------------
>>
>> A. I get the same error with either/BOTH flags.
>
> So this is about PETSc usage:
>
> 1) For any solve, the first thing to do is see what was used
>
> --petsc.ksp_monitor
>
> 2) I bet it was block Jacobi, so
>
> --petsc.sub_pc_factor_shift_positive_definite
>
> 3) However I would change preconditioners since this is a saddle point
>
> --petsc.sub_pc_type jacobi
>
> --petsc.pc_type lu --petsc.ksp_type preonly --petsc.mat_type aijmumps
>
>> B. I then ran pylith with the mesh debug flag, "debug = 1", and it seems like
>> the problem is with element 2982 (?). On further investigation of the screen
>> output, this turns out to be a cohesive element. So, I am trying to figure out
>> whether there is a problem of mesh-quality or if it has something to do with
>> cohesive element generation.
>
> Yes, they create a saddle point since we are using Lagrange multipliers.
>
>> C. I used Cubit to generate my mesh, and have visually checked the mesh quality
>> to make sure there are only a small number of elements with a poor aspect ratio
>> (and no negative Jacobians). The imported mesh has 1265 elements (and only 4 of
>> those elements have "bad" aspect or edge ratios - i.e. > 4.0). I am not sure
>> if/how Pylith re-numbers the imported elements, especially after the cohesive
>> elements are added to the system - so, I don't know where this element is in the
>> mesh. Is it possible to extract this from the pylith debug information? I
>> couldn't find any coordinate info listed in the screen output.
>
> No, I don't think so.
>
> Matt
>
>> D. Also, there are two more error "blocks" starting at lines 134122 and 134167
>> of the attached log file. Are these associated with the zero pivot error above
>> (line 134077)?
>>
>> Thanks again!
>> Ravi.
>>
>> ----------------------------------------------------------------------------------
>>
>> Brad Aagaard wrote:
>>> Ravi-
>>>
>>> You have an error when specifying the materials bin. Delete line 38 of
>>> SLAB2D_p2_nx101_Th30.cfg.
>>>
>>> What you have translates to:
>>> pylith.timedependent.materials.materials = slab2D_3M
>>> what you mean is:
>>> pylith.timedependent.materials = slab2D_3M
>>>
>>> The easiest way to track down these problems is to use
>>> the --help, --help-properties, and --help-components arguments to pylith. For
>>> example, if you ran PyLith with
>>> pylith YOUR_CFG_FILES_HERE --timedependent.help-components
>>> you would find out that PyLith is still using the default materials bin
>>> (homogeneous). Once you fix your error you should see that PyLith is picking
>>> up your setting on line XX of your cfg file.
>>>
>>> Brad
>>>
>>>
>>> On Tuesday 22 January 2008, Ravi Kanda wrote:
>>>> I am building a 2D elastic model that contains three materials, two fault
>>>> interfaces, and 8 edge BCs. Pylith exits with the following errors:
>>>> " ...
>>>> -- pyre.inventory(error)
>>>> -- timedependent.homogeneous.materials <- 'slab2D_3M'
>>>> -- unrecognized property 'timedependent.homogeneous.materials'
>>>> ...
>>>> "
>>>> So, every other error is of the form "-- unknown component
>>>> timedependent.homogeneous.x.y.z", which is probably due to the first error
>>>> shown above.
>>>>
>>>> 1) I would have thought the order would be
>>>> "timedependent.materials.homogeneous...."
>>>>
>>>> 2) Does it matter where the materials and time information are defined
>>>> (i.e., whether they are defined in the PYLITHAPP.cfg file or the
>>>> <RUN-PREFIX>.cfg file? I am defining both in the latter.
>>>>
>>>> 3) I also want to know if "debug" option is available for Materials, BCs,
>>>> and Faults, or if it is restricted to Mesh importing.
>>>>
>>>> Note on my workflow:
>>>> -----------------------
>>>> I am using python scripts to generate the mesh in Cubit, and export it in
>>>> EXODUS format. The SpatialDB files for each of the above components, as
>>>> well as the CFG files are automatically generated from the problem data
>>>> exported from the Cubit python scripts as well as separately specified
>>>> run-data, to minimize manual errors. I checked using Paraview that the
>>>> mesh, nodesets and material blocks are exported correctly from cubit.
>>>>
>>>> For these initial runs, I have manually (and also by using test scripts I
>>>> wrote) checked that the spatialDB files generated are consistent w.r.t the
>>>> mesh-domain, units, and property values. I also checked that the required
>>>> module definitions of the form: '[pylith.timedependent.x.y.z...]' occur in
>>>> the CFG files, as indicated in the pylith examples, and that the mesh
>>>> nodeset and block labels are correctly identified in these files.
>>>>
>>>> I have been able to run and visualize output from all the pylith examples
>>>> without problem, and used the HEX8 and TWOQUAD4 examples to guide the
>>>> generation my ODB, SPATIALDB, and CFG files.
>>>>
>>>> ATTACHMENTS & Specifics:
>>>> -----------------------
>>>> I am attaching the screen-output for the run, along with the CFG files, and
>>>> the material ODB file. All the required files are in the same run
>>>> directory (*.cfg, *.odb, *.spatialdb). I generated containers
>>>> (slab2D_3M.odb, slab2D_2F.odb, slab2D_8BC.odb) for the above three
>>>> components using the corresponding templates provided in the Pylith binary
>>>> distribution. I also tried variations of the ODB file where I changed the
>>>> object name at the top to the defaults (MyMatBin, MyBC, or MyFaults),
>>>> instead of customized names... but I get the same errors.
>>>>
>>>>
>>>> Thanks for your help!
>>>> Ravi.
>>>>
>>>> ---------------------------------------------------------------------
>>>> Ravi Kanda
>>>> Seismological Laboratory, MC 252-21
>>>> Division of Geological and Planetary Sciences
>>>> California Institute of Technology
>>>>
>>>> 1200 E. California Blvd., Pasadena, CA 91125
>>>> Phone: 626-395-6971, Fax: 626-564-0715
>>>> Web Page: http://www.gps.caltech.edu/~rkanda
>>>>
>>>> ----------------------------------------------------------------------
>>>>
>>>> For a human being, the unexamined life is not worth living - SOCRATES
>>>>
>>>> ----------------------------------------------------------------------
>>>
>> --
>> ---------------------------------------------------------------------
>> Ravi Kanda
>> Seismological Laboratory, MC 252-21
>> Division of Geological and Planetary Sciences
>> California Institute of Technology
>>
>> 1200 E. California Blvd., Pasadena, CA 91125
>> Phone: 626-395-6971, Fax: 626-564-0715
>> Web Page: http://www.gps.caltech.edu/~rkanda
>>
>> ----------------------------------------------------------------------
>>
>> For a human being, the unexamined life is not worth living - SOCRATES
>>
>> ----------------------------------------------------------------------
>>
>> _______________________________________________
>> CIG-SHORT mailing list
>> CIG-SHORT at geodynamics.org
>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>
>>
>
>
>
--
---------------------------------------------------------------------
Ravi Kanda
Seismological Laboratory, MC 252-21
Division of Geological and Planetary Sciences
California Institute of Technology
1200 E. California Blvd., Pasadena, CA 91125
Phone: 626-395-6971, Fax: 626-564-0715
Web Page: http://www.gps.caltech.edu/~rkanda
----------------------------------------------------------------------
For a human being, the unexamined life is not worth living - SOCRATES
----------------------------------------------------------------------
More information about the CIG-SHORT
mailing list