[CIG-SEISMO] Laplacian error in xmeshfem3D

[Dimitri Komatitsch]

Dear Steve,

Somebody (not me) changed the default DATA/Par_file
and committed to SVN a Par_file that does not work
(it leads to a negative Jacobian in the mesh
because NEX_XI is too small compared to the horizontal
size of the model and therefore the elements in the doubling
layers are too flat and very distorted).

To make it work, use this:

# coordinates of mesh block in latitude/longitude and depth in km
LATITUDE_MIN                    = 33.8d0
LATITUDE_MAX                    = 34.1d0
LONGITUDE_MIN                   = -118.4d0
LONGITUDE_MAX                   = -118.1d0

I will fix the problem on the SVN server.

To other developers: please do not commit files that do not work to the 
stable SVN branch. This makes the whole package unusable.
Now we are going to have to release a new version...

Thanks,
Dimitri.

Steve Smith wrote:
> Dr. Komatitsch,
> 
> I've built the SPECFEM3D package up through the meshing stage on my 32 
> bit Fedora Core 3 cluster with IFORT v9.0 and MPICH-1.2.7 (vintage 2005).
> 
> As you know, I am attempting to run xmeshfem3D to generate a sample mesh 
> file. I need to know the mesh format so that I can convert COMSOL/ELMER 
> meshes or geometry files to be run in SPECFEM3D. I have built/run the 
> mesher for use with the default haukson_model files as given in 
> DATA/Par_file.
> 
> On my multicore desktop, MPI xmeshfem3D with the included Par_file and 
> go_mesher script failed with 3D Laplacian errors followed by MPI_Abort 
> communicator errors. The same error occurred on the cluster.
> 
> 
> Therefore, I compiled for serial execution on a single cluster node 
> using the --without-mpi switch. Changes I made to the included files were:
> 
> 1) the "mymachines" file lists my cluster nodes (standard),
> 
> 2) the LOCAL_PATH variable in the Par_file points to /scratch/<my 
> username>/DATABASES_MPI instead of /scratch/komatits/DATABASES_MPI
> 
> 3) NPROC_XI and NPROC_ETA were set to 1
> 
> The OUTPUT_FILES/error_message00000.txt for serial execution reads:
> --------------------------------------------------
> 3D Jacobian undefined
>  Error detected, aborting MPI... proc            0
> --------------------------------------------------
> 
> Computation of the 3D Jacobian within the mesher seems to be the 
> problem. This error preceeds the stdio MPI_Abort(MPI_COMM_WORLD, 30) 
> errors when the mesher runs with MPICH1 and MPICH2 on the cluster or my 
> multicore desktop using NPROC_XI=NPROC_ETA=2
> 
> --------------------------------------------------
>  3D Jacobian undefined
>  Error detected, aborting MPI... proc            2
> --------------------------------------------------
> 
> I am unable to diagnose this any further. What files/output would be 
> helpful in this case? I will cc this to the CIG-SEISMO list.
> 
> Again, thank you for your help.
> 
> Respectfully,
> 
> -Steve Smith
> Geophysics Ph.D. Student
> Center for Wave Phenomena
> Colorado School of Mines
> stesmith   AT   mines.edu
> 

-- 
Dimitri Komatitsch - dimitri.komatitsch aT univ-pau.fr
Professor, University of Pau, Institut universitaire de France,
CNRS and INRIA Magique3D, France   http://www.univ-pau.fr/~dkomati1