export function initialized by vector to PVD file in parallel

Sorry, dumb question. Although the syntax above (taken from the eigenvalue demo) works only in
serial, the syntax

  u = Function(V)
  u.vector()[:] = rx
  File("u.pvd") << u

works in serial and parallel, so solves my problem.

It seems I was wrong. From what I can tell

  u = Function(V, rx) [BAD]

never works in parallel.

  u = Function(V) [BAD]
  u.vector()[:] = rx

sometimes works in parallel, but sometimes generates PETSc errors about ghost variables.

However a third possibility seems to work reliably.

  u = Function(V) [GOOD]
  u.vector().set_local(rx.array())

Can anyone more knowledgeable confirm that this is the right way to attach a vector to a function?

On Wed, Oct 3, 2012 at 2:01 AM, Doug Arnold
<email address hidden> wrote:
> Question #210114 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/210114
>
> Status: Solved => Open
>
> Doug Arnold is still having a problem:
> It seems I was wrong. From what I can tell
>
> u = Function(V, rx) [BAD]
>
> never works in parallel.
>

The above can work if rx is a Vector and the dof map and the vector
have the same parallel partitioning.

> u = Function(V) [BAD]
> u.vector()[:] = rx
>
> sometimes works in parallel, but sometimes generates PETSc errors about
> ghost variables.
>

Can work if the Vectors u.vector() and rx have the same parallel layout.

> However a third possibility seems to work reliably.
>
> u = Function(V) [GOOD]
> u.vector().set_local(rx.array())
>

This sets the local (to the process) values of u equal to rx. The size
of rx should be equal to the size of u on the process, in which case
it will always work.

> Can anyone more knowledgeable confirm that this is the right way to
> attach a vector to a function?
>

The most powerful approach is

   u.add(rx, rows)
   u.apply("add")

where rx are values and rows is a list of positions in u that the rx
values should be added to. This also works across processes.

Garth

> --
> You received this question notification because you are a member of
> DOLFIN Team, which is an answer contact for DOLFIN.