Qing,
I meet a MPI problem when I simulate a three point bending test. I try
to compute the load force traction on the top boundary. The results are
correct
with the single process. But When I switched to a multi-process, the
results changed.
Let's say, if we wan to get the load force in y direction, the results
are like this:
* 1 core: -0.83
* 2 cores: -0.83
* 3 cores: -0.92
* 4 cores: -0.415
* 5 cores: -0.002
I already use `sum` collective operation after computing locally owned
load force value.
In order to double check it, I try to output each processes' local
value, and get the results
like this:
* 1 core: locally load force y: -0.831775
* 2 core: locally load force y: (0)-0.415888 (1)-0.415888
* 4 core: locally force y: (0)0 (1)-0.408131 (2)-0.00237332 (3)-0.004908
I do not know why I get smaller values with more cores.
I also have checked the boundary conditions and mesh size. In fact, the
displacement results
are correct and same under different cores. So the problem should be
just in this load force
computation.
There is not enough information in your email to say what may be wrong.
Do I read your description correctly to say that you have verified that
the *solution* is the same independent of the number of processors? If
that is so, then the problem must indeed be in the postprocessing step
where you compute the resulting forces. But without seeing your code
that does this computation, it's impossible to say what's wrong.
Best
W.
--
------------------------------------------------------------------------
Wolfgang Bangerth email: bange...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to dealii+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
