[CIG-SEISMO] Negative Jacobian in specfem2d
Dimitri Komatitsch
komatitsch at lma.cnrs-mrs.fr
Thu Aug 4 02:52:08 PDT 2016
Hi Jami,
Thanks! No, there is no resolution limit (we routinely use that code for
ultrasounds for instance).
Unless the mesh point detection routine fails (geometrical tolerance
issue? unlikely, since the tolerance is computed relative to the total
size of the mesh, and in double precision) there is no known limit.
You can type:
./configure --enable-debug
make realclean
make clean
make all
and try again, to see if you get any useful debugging message.
Cheers,
Dimitri.
On 04/08/2016 03:58, Jami Johnson wrote:
> Hi Dimitri and Paul,
>
> Thank you for your replies! I also have run the mesh and received no
> errors when I run xcheck_quality_external_mesh. After another test, I
> found that when I run SPECFEM2D with a mesh size of 0.00013, it runs
> without error, but when I decrease the element size to 0.00003, I get
> the negative Jacobian error (even when xcheck_quality_external_mesh
> shows no error). Is there a resolution limit? In my parameter file, I
> use deltat=5d-10 and velocity of 1500 m/s.
>
> Thanks again for your help.
>
> Cheers,
> Jami
>
> On Thu, Aug 4, 2016 at 3:14 AM cristini <cristini at lma.cnrs-mrs.fr
> <mailto:cristini at lma.cnrs-mrs.fr>> wrote:
>
> Hi Jami,
>
> I did a quick test and did not find any problem. This is probably a Gmsh
> issue that you should fix easily by upgrading to a more recent version.
>
> I took you .geo file and got a very good mesh without any negative
> jacobian.
>
> Thanks,
>
> Best wishes
>
> Paul
>
> PS: the histogram I got
>
> histogram of skewness (0. good - 1. bad):
>
> 0.0000000E+00 - 5.0000001E-02 984857 95.14644 %
> 5.0000001E-02 - 0.1000000 40000 3.864376 %
> 0.1000000 - 0.1500000 2983 0.2881858 %
> 0.1500000 - 0.2000000 2547 0.2460641 %
> 0.2000000 - 0.2500000 1288 0.1244329 %
> 0.2500000 - 0.3000000 1358 0.1311956 %
> 0.3000000 - 0.3500000 597 5.7675809E-02 %
> 0.3500000 - 0.4000000 595 5.7482593E-02 %
> 0.4000000 - 0.4500000 619 5.9801217E-02 %
> 0.4500000 - 0.5000000 189 1.8259175E-02 %
> 0.5000000 - 0.5500000 61 5.8931732E-03 %
> 0.5500000 - 0.6000000 1 9.6609394E-05 %
> 0.6000000 - 0.6500000 0 0.0000000E+00 %
> 0.6500000 - 0.7000000 1 9.6609394E-05 %
> 0.7000000 - 0.7500000 0 0.0000000E+00 %
> 0.7500000 - 0.8000000 0 0.0000000E+00 %
> 0.8000000 - 0.8500000 0 0.0000000E+00 %
> 0.8500000 - 0.9000000 0 0.0000000E+00 %
> 0.9000000 - 0.9500000 0 0.0000000E+00 %
> 0.9500000 - 1.000000 0 0.0000000E+00 %
>
> total number of elements = 1035096
>
>
>
>
> On 03/08/2016 15:43, Dimitri Komatitsch wrote:
> >
> > Hi Jami,
> >
> > Thanks! Probably mesh quality that is not good enough, two elements
> > that are almost flat. Try to see the quality histogram in Gmsh to see
> > if these two elements are poor (in which case even if you manage to
> > run that mesh in SPECFEM2D it will likely blow up). There is also a
> > tool for that in SPECFEM2D:
> >
> > src/auxiliaries/check_quality_external_mesh.f90
> >
> > If it turns out to be a bug in the code (unlikely, but worth checking)
> > please let us know.
> >
> > Thanks,
> > Best wishes,
> >
> > Dimitri.
> >
> > On 03/08/2016 10:01, Jami Johnson wrote:
> >> Hello,
> >> I have created a very dense mesh with Gmsh, and when I check for the
> >> values of the Jacobian in Gmsh they are all positive. However,
> when I
> >> run it in specfem2d, an error is thrown due to a negative
> Jacobian. It
> >> seems to be a very small percentage of my elements causing a
> problem. I
> >> believe all of my line loops are oriented correctly. Are there any
> >> other potential causes of a negative Jacobian?
> >>
> >> Error in specfem2d:
> >> --------------------------
> >> 2 elements have a negative Jacobian, out of 1313871
> >>
> >> geo file:
> >> -----------
> >> lc=0.00003;
> >> xmax=0.015;
> >> xmin=0.0;
> >> ymin=-0.015;
> >> ymax=0.0;
> >> Point(1) = {xmax,ymax , 0, lc};
> >> Point(2) = {xmax, ymin, 0, lc};
> >> Point(3) = {xmin, ymin, 0, lc};
> >> Point(4) = {xmin, ymax, 0, lc};
> >> Line(1) = {1, 4};
> >> Line(2) = {4, 3};
> >> Line(3) = {3, 2};
> >> Line(4) = {2, 1};
> >> R=0.00025;
> >> Centx=0.0075;
> >> Centy=-0.005;
> >> Point(5) = {Centx, R+Centy, 0, lc};
> >> Point(6) = {R+Centx, Centy, 0, lc};
> >> Point(7) = {-R+Centx, Centy, 0, lc};
> >> Point(8) = {Centx, -R+Centy, 0, lc};
> >> Point(9) = {Centx, Centy, 0, lc};
> >> Circle(5) = {5, 9, 7};
> >> Circle(6) = {7, 9, 8};
> >> Circle(7) = {8, 9, 6};
> >> Circle(8) = {6, 9, 5};
> >> Line Loop(9) = {5, 6, 7, 8};
> >> Line Loop(10) = {1, 2, 3, 4};
> >>
> >> Plane Surface(10) = {10, 9};
> >> Plane Surface(11) = {9};
> >>
> >> Recombine Surface{10,11};
> >> Mesh.SubdivisionAlgorithm=1;
> >> Physical Line("Top") = {1};
> >> Physical Line("Left") = {2};
> >> Physical Line("Bottom") = {3};
> >> Physical Line("Right") = {4};
> >> Physical Surface("M1") = {10};
> >> Physical Surface("M2") = {11};
> >>
> >>
> >>
> >>
> >> _______________________________________________
> >> CIG-SEISMO mailing list
> >> CIG-SEISMO at geodynamics.org <mailto:CIG-SEISMO at geodynamics.org>
> >> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo
> >>
> >
>
> --
> *--------------------------------------------------------------------------*
> Warning !!! New address !! Attention !!! nouvelle
> adresse !!
> *--------------------------------------------------------------------------*
> Paul CRISTINI Charge de Recherche CNRS (HDR) / CNRS Research
> Scientist
> *--------------------------------------------------------------------------*
> CNRS - Laboratoire de Mecanique et d'Acoustique (UPR 7051)
> Bureau 119
> 4 Impasse Nikola Tesla, CS 40006, F-13453 Marseille Cedex 13 - France
> mailto: cristini at lma.cnrs-mrs.fr <mailto:cristini at lma.cnrs-mrs.fr>
> phone number: +33 (0)4 84 52 42 51
> http://www.lma.cnrs-mrs.fr/
> *--------------------------------------------------------------------------*
>
--
Dimitri Komatitsch
CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics,
UPR 7051, Marseille, France http://komatitsch.free.fr
More information about the CIG-SEISMO
mailing list