[petsc-users] problem with nested fieldsplits

[Jean-Arthur Louis Olive]

Hi all,
I am using PETSc (dev version) to solve the Stokes + temperature equations. My DM has fields (vx, vy, p, T).

I would like to use nested fieldsplits to separate the T part from the Stokes part, and apply a Schur complement approach to the Stokes block.
Unfortunately, I keep getting this error message:
[1]PETSC ERROR: DMCreateFieldDecomposition() line 1274 in /home/jolive/petsc/src/dm/interface/dm.c Decomposition defined only after DMSetUp

Here are the command line options I tried:

-snes_type ksponly \
-ksp_type fgmres \
# define 2 fields: [vx vy p] and [T]
-pc_type fieldsplit -pc_fieldsplit_0_fields 0,1,2 -pc_fieldsplit_1_fields 3 \
# split [vx vy p] into 2 fields: [vx vy] and [p]
-fieldsplit_0_pc_type fieldsplit \
-pc_fieldsplit_0_fieldsplit_0_fields 0,1 -pc_fieldsplit_0_fieldsplit_1_fields 2 \

# apply schur complement to [vx vy p]
-fieldsplit_0_pc_fieldsplit_type schur \
-fieldsplit_0_pc_fieldsplit_schur_factorization_type upper \

# solve everything with lu, just for testing
-fieldsplit_0_fieldsplit_0_ksp_type preonly \
-fieldsplit_0_fieldsplit_0_pc_type lu -fieldsplit_0_fieldsplit_0_pc_factor_mat_solver_package superlu_dist \
-fieldsplit_0_fieldsplit_1_ksp_type preonly \
-fieldsplit_0_fieldsplit_1_pc_type lu -fieldsplit_0_fieldsplit_1_pc_factor_mat_solver_package superlu_dist \
-fieldsplit_1_ksp_type preonly \
-fieldsplit_1_pc_type lu -fieldsplit_1_pc_factor_mat_solver_package superlu_dist \

Any idea what could be causing this?
Thanks a lot,
Arthur
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140923/92b9374f/attachment.html>