dolfinx
dolfinx copied to clipboard
FEniCS
demo_stokes.py direct solver fails with nan
The direct solver ("blocked") in the python demo demo_stokes.py for dolfinx 0.4.1 is returning nan on Debian Linux systems. Full test log can be found at https://ci.debian.net/data/autopkgtest/unstable/amd64/f/fenics-dolfinx/22703381/log.gz, the end of the log reports
KSP Object: (fieldsplit_p_) 1 MPI processes
type: preonly
maximum iterations=10000, initial guess is zero
tolerances: relative=1e-05, absolute=1e-50, divergence=10000.
left preconditioning
using NONE norm type for convergence test
PC Object: (fieldsplit_p_) 1 MPI processes
type: jacobi
type DIAGONAL
linear system matrix = precond matrix:
Mat Object: (fieldsplit_p_) 1 MPI processes
type: seqaij
rows=1089, cols=1089
total: nonzeros=7361, allocated nonzeros=7361
total number of mallocs used during MatSetValues calls=0
not using I-node routines
linear system matrix followed by preconditioner matrix:
Mat Object: 1 MPI processes
type: seqaij
rows=9539, cols=9539
total: nonzeros=269576, allocated nonzeros=269576
total number of mallocs used during MatSetValues calls=0
has attached null space
using I-node routines: found 5312 nodes, limit used is 5
Mat Object: 1 MPI processes
type: seqaij
rows=9539, cols=9539
total: nonzeros=197829, allocated nonzeros=197829
total number of mallocs used during MatSetValues calls=0
using I-node routines: found 5312 nodes, limit used is 5
(A) Norm of velocity coefficient vector (nested, iterative): 17.478370152447766
(A) Norm of pressure coefficient vector (nested, iterative): 311.44253800229257
(B) Norm of velocity coefficient vector (blocked, iterative): 17.478370158383505
(B) Norm of pressure coefficient vector (blocked, interative): 311.44253801601417
(C) Norm of velocity coefficient vector (blocked, direct): nan
(C) Norm of pressure coefficient vector (blocked, direct): nan
----------------------------- Captured stderr call -----------------------------
Traceback (most recent call last):
File "/tmp/autopkgtest-lxc.lhvvm1tb/downtmp/build.KH9/src/python/demo/demo_stokes.py", line 406, in <module>
assert np.isclose(norm_u_2, norm_u_0)
AssertionError
Do we know what would cause the Stokes direct solver to return nan like this?
This is likely fixed by #2512, which adds proper handling of the pressure nullspace when using a direct solver.