[CIG-SEISMO] SPECFEM3D - test case for meshfem3D "simple model"

[daniel peter]

Salut Camille,

thanks for giving SPECFEM3D a try :) i was reading your mail below and  
it sounded like the example didn't run for you. however, the output  
you attached is the correct one. it seems that you are actually doing  
everything all right and the xmeshfem3D is working just fine for the  
example as well.

maybe you are just misinterpreting the skewness table:
...
histogram of skewness (0. good - 1. bad):

   0.0000000E+00  -   5.0000001E-02       27648     81.81818      %
   5.0000001E-02  -   0.1000000               0    0.0000000E+00  %
...


the first line means that you have 27648 elements with a skewness  
value between 0 and 0.05 (which means the element is basically not  
skewed, just plain regular hexahedral element). the total number of  
elements you have in this mesh is (see in the output_mesher.txt file a  
bit further down):
...
  total number of elements in entire mesh:        33792
...

which gives you that:  27648 / 33792 ~ 81.8 % of all elements are not  
skewed, i.e. regular elements. a fantastic value :)

the histogram lists for this mesh also some stronger skewed elements,  
for example the worst ones belong to:
...
0.6000000      -   0.6500000            2048     6.060606      %
...

about 6 % of all elements have distortions with a skewness value  
between 0.6 and 0.65. the skewness values give you a hint of how good  
your mesh is. in an ideal world, you would want to have no  
distortions, just like the 81% from above. those elements give you the  
best approximate values by the GLL quadrature used in the spectral- 
element method. however, having weakly distorted elements is still  
fine and the solutions are still accurate enough. so empirically,  
values up to around 0.7 are tolerable, above that you should consider  
remeshing...

to give you an idea why some of the elements are distorted, i'll  
attach here a figure of the mesh you obtain in this example. you will  
see that the mesh contains a doubling layer, where we stitch elements  
together such that the size of two elements will transition to the  
size of one element (very useful to keep the ratio of wavespeed /  
element_size about constant). those elements in this doubling layer  
have higher skewness values and make up those 6 % in the histogram.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: mesh_vp.jpg
Type: image/jpeg
Size: 184175 bytes
Desc: not available
Url : http://geodynamics.org/pipermail/cig-seismo/attachments/20110309/5027e64b/attachment-0001.jpg 
-------------- next part --------------



about your concern, just try to continue the example (you might want  
to run >svn update to get the latest version of SPECFEM3D with an  
updated README for this simple_model/ example). then look at the mesh  
files generated in in_out_files/DATABASES_MPI/ after you run the  
xgenerate_databases executable (step 3 in the README, make sure you  
have set SAVE_MESH_FILES to .true. in the Par_file as it comes from  
the example). then you could use paraview (www.paraview.org) to load  
the proc****_vp.vtk files and compare them to the figure attached  
here. they should look the same.

i found this example better than the the second one for the internal  
mesher. the many_interfaces/ example is using a setup with many cpus  
(81), generates a huge mesh (~415 GB) and the forward simulation takes  
about 5h without too much of nice looking seismograms. i think, we  
will have to change that second example in future to make it somewhat  
easier to run...

maybe, in your case the example waterlayered_halfspace/ might be  
interesting as well. it explains how to set up a model with acoustic  
and elastic parts. however, that example works together with CUBIT (http://cubit.sandia.gov/ 
), an external mesher which needs to be purchased (?~$200). especially  
for more complicated setups, accounting for complex bathymetry and  
subsurface structure, meshing with CUBIT becomes more and more powerful.

anyway, in case you encounter some more problems with the example,  
just let me know.

best wishes from Princeton,
daniel


ps. and if you want to contribute your own setup as a new future  
example, please let me know as well. it's always great to provide more  
useful examples :)


--
Daniel Peter
Department of Geosciences
318 Guyot Hall
Princeton University
Princeton, NJ 08540, USA

Tel: +1 (609) 258-5031
url: http://www.princeton.edu/~dpeter



On Mar 8, 2011, at 6:26 PM, Camille Mazoyer wrote:

> Hello,
>
> I'm an engineer in scientific computing at Institut Universitaire  
> Europ?en de la Mer (IUEM), Brest, France. One of the team with whom  
> I work with would like to use SPECFEM3D in order to study submarine  
> seisms and their  Hydroacoustic propagation.
>
> Therefore, I'm learning how to use your software for helping this  
> team. I started with the example of mesh generation called "simple  
> model" found in examples/meshfem3D_examples/ 
> simple_model.Unfortunately, I didn't manage to make this example  
> work. I probably forgot to do something but I can't see where: I  
> just copied the files in in_data_files/meshfem3D_files as it is  
> explained in the README of the example. And I didn't modify any of  
> the files.
> After executing xmeshfem3D, I've got elements with skewness of 81%,  
> which is not correct. I send you in an attached file the  
> output_mesher.txt. I hope this file will help you find what I missed  
> in my simulation.
>
> Thanks a lot for your help.
> I am looking forward to work with this software which seems very  
> accomplished.
>
> Camille Mazoyer
> <output_mesher.txt>_______________________________________________
> CIG-SEISMO mailing list
> CIG-SEISMO at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo