[CIG-SHORT] Smoothing Issues

[Brad Aagaard]

Nick,

It looks like you have some severely distorted cells, which will cause 
very slow convergence with the iterative solver. In LaGriT you will want 
aspect ratios close to 1.0 (above 0.2 or 0.5 would be good); note that 
LaGriT uses an aspect ratio in the range of 0-1, whereas CUBIT/Trelis 
and ParaView use aspect ratio metrics in the range of 1-infinity.

See share/solver_fault_fieldsplit.cfg for a better preconditioner than ASM.

Brad

On 11/07/2016 02:08 PM, Nick Wogan wrote:
> Brad and Charles,
>
> Looks like Cubit isn't in the budget unfortunately, but I am slowly
> learning LaGriT.
>
> Things are looking better. Attached is an image of one mesh used, and
> the resulting stresses which seem to be much smoother. I have run into
> another problem- the iterative solver is not converging for this new
> mesh. For all of the previously run simulations I have used the
> following PETSC settings.
>
> # Set the solver options.
> [pylithapp.petsc]
> malloc_dump =
> pc_type = asm
> sub_pc_factor_shift_type = nonzero
> ksp_rtol = 1.0e-8
> ksp_atol = 1.0e-12
> ksp_max_it = 5000
> ksp_gmres_restart = 50
> ksp_monitor = true
> ksp_view = true
> ksp_converged_reason = true
> ksp_error_if_not_converged = true
> log_summary = true
>
> I changed the settings to a direct solver and it produced a solution:
> [pylithapp.petsc]
> snes_view = true
> pc_type = lu
>
> Now I am trying to fine the mesh further to hopefully produce a smoother
> solution, but the neither solver is working. The iterative solver
> doesn't converging to a reasonable value. The direct solver runs out of
> memory. Is there a better iterative solver that would work better for
> this problem? (Note I reduced poison's ratio of lower crust to 0.45)
>
> Less importantly: the material spacialdb assigned to a specific material
> ID in the .cfg file is not working properly. It seems to assign given
> material properties to any region somewhat arbitrarily. According to
> Lagrit the material id's should be...
>
> Material id 1: Upper crust with square "hole" in it
> Material id 2: Upper crust density anomaly from z=0 to z=-4000
> Material id 3: Upper crust under density anomaly from z = -4000 to z=-8000
> Material id 4: Lower crust
>
> Here is zip containing cfg that produced attached images (wallowa01.cfg).​
>  Wallowa1.zip
> <https://drive.google.com/file/d/0BxnRdwFIDyn5QWRhaFZtMHVBYzA/view?usp=drive_web>
> ​
> Lots of questions here and i totally understand if you can't answer them
> fully. I thank you for any amount of feedback!
>
> Nick
>
>
>
> On Tue, Nov 1, 2016 at 1:54 PM, Nick Wogan <nicholaswogan at gmail.com
> <mailto:nicholaswogan at gmail.com>> wrote:
>
>     Hi Charles,
>
>     I really am appreciating your feedback. For the past few days I have
>     been trying very hard to make an 'irregular' mesh with LaGriT, where
>     there are much finer elements where the density anomaly is. LaGriT
>     is proving to be somewhat difficult to use. I do not have access to
>     Cubit or Trelis, because (as far as i know) those programs are only
>     freely distributed to government organizations such as USGS, and I
>     work for the University of Oregon.
>
>     I just wanted to say I really appreciate your help! I will report
>     back once I have managed to make a mesh using LaGriT.
>
>     Cheers,
>     Nick
>
>     On Tue, Oct 25, 2016 at 2:10 PM, Nick Wogan <nicholaswogan at gmail.com
>     <mailto:nicholaswogan at gmail.com>> wrote:
>
>         Thank you very much for the feedback,
>
>         Attached is config/mesh/spacialdb file, and an image that
>         provides a cross-section of the model giving material
>         properties. I am using unusual material properties. Large parts
>         of the model are youngs mod = 0.1 and Poisson's Ratio = 0.49
>
>         The problem is as follows. There is an upper crust that has
>         typical elastic material properties, and a lower crust that has
>         young's mod = 0.1 and Poisson's Ratio = 0.49. In the center top
>         of the upper crust there is a small chunk of low density. The
>         displacement output looks great, only the stress output looks
>         checkered in certain areas.
>
>         I checked the quality of the mesh. Here is the Lagrit output:
>
>         epsilon1, epsilonaspect: 4.6113656E-11  9.8059273E-32
>         elements with aspect ratio < .01:                      0
>         elements with aspect ratio b/w .01 and .02:      0
>         ...
>         elements with aspect ration b/w .2 and .5 :       0
>         elements with aspect ratio b/w .5 and 1. :         314844
>         min aspect ratio = 0.3795E+00     max aspect ratio = 0.7107E+00
>
>         Thanks you again,
>         Nick
>
>
>         ​
>          Checkered.zip
>         <https://drive.google.com/file/d/0BxnRdwFIDyn5U01aZEYzUURrSzg/view?usp=drive_web>
>         ​
>
>         On Wed, Oct 19, 2016 at 4:40 PM, Nick Wogan
>         <nicholaswogan at gmail.com <mailto:nicholaswogan at gmail.com>> wrote:
>
>             Hi Pylith,
>
>             I am having issues with the Pylith output lacking
>             smoothness. It is very clear in the attached photo plotting
>             stress_zz that there are large jumps in stress that are
>             unexpected. Is this "checker-boarding" something that can be
>             fixed?
>
>             Let me know if I should send the mesh, cfg, spacialdb files.
>
>             Thanks!
>
>             Nick Wogan
>             Research Assistant
>             University of Oregon
>
>
>
>
>
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>