[aspect-devel] Not all processors agree whether the last operation on this vector was an addition or a set operation.

[Timo Heister]

Gerry,

this error message happens when parallel linear algebra code is
written in an incorrect manner and is likely to happen if one of the
processor ends up without any DoFs.

A few questions:
1. Is this reproducible? I would hope so in this case.
2. Can you try if the error happens earlier if you run on a higher
number of cores?
3. Does this happen after an (adaptive) refinement step?
4. Does it also happen if you turn off the error postprocessor that
you seem to be running?

Best,
Timo



On Fri, Nov 24, 2017 at 6:50 PM, Elbridge Gerry Puckett
<egp at math.ucdavis.edu> wrote:
> Hi All,
>
> I'm running a benchmark an 2D annulus and after 2223 time steps I get this
> error and a whole string of errors one after the other like this one,
> without taking another time step.
>
> Any ideas,  suggestions?
>
> Thanks!
>
> - Gerry
>
>    Postprocessing:
>      Errors u_L1, p_L1, u_L2, p_L2, rho_L1, rho_L2, rho_Linf: 4.436822e+02,
> 2.420235e+03, 1.243902e+02, 9.358409e+02, 5.442952e+03, 2.203707e+03,
> 1.364592e+03
>      Number of advected particles:                            3072
>
> *** Timestep 2223:  t=0.940216 seconds
>    Skipping temperature solve because RHS is zero.
>
> --------------------------------------------------------
> An error occurred in line <223> of file
> </home/egpuckett/deal.ii-candi/tmp/unpack/deal.II-v8.5.1/source/lac/trilinos_vector_base.cc>
> in function
>     void
> dealii::TrilinosWrappers::VectorBase::compress(dealii::VectorOperation::values)
> The violated condition was:
>     result.max-result.min<1e-5
> Additional information:
>     Not all processors agree whether the last operation on this vector was
> an addition or a set operation. This will prevent the compress() operation
> from succeeding.
>
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel