Hello SfePy,
During the problem solution I got the error:
sfepy: nls: iter: 0, residual: 5.088014e+00 (rel: 1.000000e+00) warning: singular matrix zeroing nan and inf entries... /usr/lib/python2.7/dist-packages/scipy/sparse/linalg/dsolve/umfpack/umfpack.py:556: RuntimeWarning: divide by zero encountered in double_scalars econd = 1.0 / self.info[UMFPACK_RCOND] warning: (almost) singular matrix! (estimated cond. number: inf) sfepy: rezidual: 0.01 [s]
I attached the problem description and the mesh for this problem. What is the reason of the error? May be the mesh is not very detailed?
Sincerely, Alexander Kalinin
Hello SfePy,
I have found the reason of error! The mesh file was generated in Gmsh tool and contains the nodes that were not a part of any tetra. So these standalone nodes were the reason of the singular matrix.
I think, there are two ways to avoid this error:
- Write a simple python script that preproceses mesh file and remove standalone nodes.
- Write a simple python script that preproceses mesh file and add proper group number to the nodes. Then in problem description file we can use the node selection code like 'nodes of group 1'.
Sincerely, Alec.
On Tuesday, September 25, 2012 6:11:51 PM UTC+4, Alec Kalinin wrote:
Hello SfePy,
During the problem solution I got the error:
sfepy: nls: iter: 0, residual: 5.088014e+00 (rel: 1.000000e+00) warning: singular matrix zeroing nan and inf entries... /usr/lib/python2.7/dist-packages/scipy/sparse/linalg/dsolve/umfpack/umfpack.py:556: RuntimeWarning: divide by zero encountered in double_scalars econd = 1.0 / self.info[UMFPACK_RCOND] warning: (almost) singular matrix! (estimated cond. number: inf) sfepy: rezidual: 0.01 [s]
I attached the problem description and the mesh for this problem. What is the reason of the error? May be the mesh is not very detailed?
Sincerely, Alexander Kalinin
Hi Alec,
On 09/25/2012 08:23 PM, Alec Kalinin wrote:
Hello SfePy,
I have found the reason of error! The mesh file was generated in Gmsh tool and contains the nodes that were not a part of any tetra. So these standalone nodes were the reason of the singular matrix.
It's good you found the reason.
I think, there are two ways to avoid this error:
- Write a simple python script that preproceses mesh file and remove standalone nodes.
- Write a simple python script that preproceses mesh file and add proper group number to the nodes. Then in problem description file we can use the node selection code like 'nodes of group 1'.
It could be made so that 1. is done in sfepy sutomatically, also giving a warning when such a situation occurs. I will try to fix this soon.
Thanks. r.
Sincerely, Alec.
On Tuesday, September 25, 2012 6:11:51 PM UTC+4, Alec Kalinin wrote:
Hello SfePy,
During the problem solution I got the error:
sfepy: nls: iter: 0, residual: 5.088014e+00 (rel: 1.000000e+00) warning: singular matrix zeroing nan and inf entries... /usr/lib/python2.7/dist-packages/scipy/sparse/linalg/dsolve/umfpack/umfpack.py:556: RuntimeWarning: divide by zero encountered in double_scalars econd = 1.0 / self.info[UMFPACK_RCOND] warning: (almost) singular matrix! (estimated cond. number: inf) sfepy: rezidual: 0.01 [s]
I attached the problem description and the mesh for this problem. What is the reason of the error? May be the mesh is not very detailed?
Sincerely, Alexander Kalinin
On 09/25/2012 10:20 PM, Robert Cimrman wrote:
Hi Alec,
On 09/25/2012 08:23 PM, Alec Kalinin wrote:
Hello SfePy,
I have found the reason of error! The mesh file was generated in Gmsh tool and contains the nodes that were not a part of any tetra. So these standalone nodes were the reason of the singular matrix.
It's good you found the reason.
I think, there are two ways to avoid this error:
- Write a simple python script that preproceses mesh file and remove standalone nodes.
- Write a simple python script that preproceses mesh file and add proper group number to the nodes. Then in problem description file we can use the node selection code like 'nodes of group 1'.
It could be made so that 1. is done in sfepy sutomatically, also giving a warning when such a situation occurs. I will try to fix this soon.
The current master should work with such meshes now.
r.
Thanks. r.
Sincerely, Alec.
On Tuesday, September 25, 2012 6:11:51 PM UTC+4, Alec Kalinin wrote:
Hello SfePy,
During the problem solution I got the error:
sfepy: nls: iter: 0, residual: 5.088014e+00 (rel: 1.000000e+00) warning: singular matrix zeroing nan and inf entries... /usr/lib/python2.7/dist-packages/scipy/sparse/linalg/dsolve/umfpack/umfpack.py:556:
RuntimeWarning: divide by zero encountered in double_scalars econd = 1.0 / self.info[UMFPACK_RCOND] warning: (almost) singular matrix! (estimated cond. number: inf) sfepy: rezidual: 0.01 [s]
I attached the problem description and the mesh for this problem. What is the reason of the error? May be the mesh is not very detailed?
Sincerely, Alexander Kalinin
Hi Robert,
Yes, the current master now works perfect with such meshes. Thank you!
Just for you information. The current master fails two tests: tests/test_poly_spaces.py --- test_continuity: failed! --- test_gradients: failed!
Sincerely, Alexander.
On Wednesday, September 26, 2012 1:48:47 PM UTC+4, Robert Cimrman wrote:
Hi Alec,
On 09/25/2012 08:23 PM, Alec Kalinin wrote:
Hello SfePy,
I have found the reason of error! The mesh file was generated in Gmsh tool and contains the nodes that were not a part of any tetra. So these standalone nodes were the reason of the singular matrix.
It's good you found the reason.
I think, there are two ways to avoid this error:
- Write a simple python script that preproceses mesh file and remove standalone nodes.
- Write a simple python script that preproceses mesh file and add
group number to the nodes. Then in problem description file we can use
On 09/25/2012 10:20 PM, Robert Cimrman wrote: proper the
node selection code like 'nodes of group 1'.
It could be made so that 1. is done in sfepy sutomatically, also giving a warning when such a situation occurs. I will try to fix this soon.
The current master should work with such meshes now.
r.
On 09/27/2012 08:06 AM, Alec Kalinin wrote:
Hi Robert,
Yes, the current master now works perfect with such meshes. Thank you!
Thanks for reporting that problem.
Just for you information. The current master fails two tests: tests/test_poly_spaces.py --- test_continuity: failed! --- test_gradients: failed!
That's "normal", see my other e-mail about updates.
r.
Sincerely, Alexander.
participants (2)
-
Alec Kalinin
-
Robert Cimrman