← Back to team overview

dolfin team mailing list archive

Re: Periodic bc

 



Johan Hake wrote:
On Friday 13 March 2009 10:14:08 Garth N. Wells wrote:
Johan Hake wrote:
On Thursday 12 March 2009 23:16:11 Anders Logg wrote:
On Thu, Mar 12, 2009 at 10:07:13PM +0000, Garth N. Wells wrote:
Kristian Valen-Sendstad wrote:
On Thu, March 12, 2009 14:25, Kristian Valen-Sendstad wrote:
Hi,
There seems to be a bug in the periodic boundary conditions demo
located in
/demo/pde/periodic/.

It does not do what one would expect, and there are lots of "PETSC
ERROR"s
in both python and cpp.
Do you use the latest development version? If so, I think you have
the same problems as the buildbot.

Johannes
Yes, I do.
However, Johan Elon Hake came to the rescue and set the linear algebra
backend to uBLAS, which solved the problem.
It would be nice to figure out what the cause of the problem is with
PETSc. What are the first few lines of the error message?

Garth
I get the same thing on my machine and I guess everyone does:

  Applying periodic boundary conditions to linear system.
  [0]PETSC ERROR: --------------------- Error Message
  ------------------------------------
  [0]PETSC ERROR: Object is in wrong state!
  [0]PETSC ERROR: Not for unassembled matrix!

The problem is most likely a mix of calling A.ident(), getrow() and
not appropriately calling apply(). It should be fairly easy to fix but
I haven't had time to track it down.
The problem started after Olas last push. Strange that a changeset that
only affect XML files should have any to say for the PETSc backend?
apply() doesn't do anything for the uBLAS backend as there is no need to
'assemble' internally a sparse matrix. All the other backends require
this. This would explain why uBLAS works.

Yes I know, and I think it is weird that this problem pops up after a push only touching files that handles xml files.


This is fixed now, but it's not very efficient. PeriodicBC needs to be looked at at some point.

Garth

Johan




References