Hi Robert,

yes, currently I am using the direct solver.

Changing to another direct solver was working. I tried ScipyIterative and PyAMGSolver, both helped.

Is there a known limitation in the size of the matrix for these solvers. The sfepy output above shows my current matrix shape (313416, 313416). It might increase by factor 10. Is that a problem from your point of view?

Thank you and regards,

2017-10-06 14:40 GMT+02:00 Robert Cimrman <cimrman3@ntc.zcu.cz>:
Hi Kathrin,

you are using a direct linear solver from scipy (ls.scipy_direct), right? This error seems to be caused by the SuperLU solver running out of memory.

To be able to solver large problems, you will need to use a preconditioned iterative solver (preferably a multigrid, for example PyAMG, or some iterative solver from petsc). It is not well documented in sfepy, but the latest version has a decent support for such things, see [1]. Unfortunately, there is no example for linear elasticity only yet (I struggle with choosing correct preconditioning options myself).


[1] http://sfepy.org/doc-devel/examples/multi_physics/biot_short_syntax.html

On 10/06/2017 01:32 PM, Kathrin Sobe wrote:
Hi Robert,

I am running a linear elastic simulation, as in my examples described some weeks ago.

The current problem is as follows: After running the simulation with some small test meshes (3D with tetrahedrons) I changed to bigger meshes.

The first example, with 71759 nodes and 310106 tetrahedrons, worked fine. Now I switched to a mesh with 105579 nodes and 460442 tetrahedrons and get the following output:

sfepy: saving regions as groups...
sfepy:   Omega
sfepy:   Surface
sfepy:   Bottom
sfepy: ...done
sfepy: updating variables...
sfepy: ...done
sfepy: setting up dof connectivities...
sfepy: ...done in 0.83 s
sfepy: matrix shape: (313416, 313416)
sfepy: assembling matrix graph...
sfepy: ...done in 3.52 s
sfepy: matrix structural nonzeros: 11941254 (1.22e-04% fill)
sfepy: updating materials...
sfepy:     solid
sfepy:     f
sfepy:     load
sfepy: ...done in 0.80 s
sfepy: nls: iter: 0, residual: 2.878127e+07 (rel: 1.000000e+00)
Can't expand MemType 1: jcol 311926

Do you have a suggestion what the problem could be?

Thanks for your help.
SfePy mailing list

SfePy mailing list