Thanks. Restarting works, but is annoying.

To answer my own question (in case somebody runs into the same problem):

After a few more tests, the error is unrelated to the "Density Matrix sparsity pattern" message, but turned out to be linked to too many open files, and is somehow connected to NetCDF. It went away after compiling without NetCDF.

For the record: This is Siesta 3.1 on ifort 12.1 and Intel MPI 4.0.3.

  Herbert

On 05/05/12 08:26, kuilin lu wrote:
Dear Herbert,

     Which siesta version did you use in computing? Some versions have
bugs(feature?) in reinitialize DM, see Docs/CHANGES, you could find
some hints about this problem.  Some version disabled "Extrapolating
Density Matrix" by default such as siesta-trunk-367 version.

     I vaguely remember this problem could be solved just by restart
computing, surely continuing, not restart from zero.  If you still
meet the problem, maybe you could considering tuning
DM.AllowExtrapolation option, lower DM.MixingWeight should be more
stable, but it will (generally) slower the convergence.

    This is just my experience, advanced user might have more
appropriate opinion.

Best Wishes,
Kuilin

On 5/5/12, Herbert Fruchtl<[email protected]>  wrote:
Folks,

I think I encountered this before... If an optimisation or dynamics
calculation
runs for 500 steps, it often (always?) fails with the following lines:

=====================
   Folding of H and S is implicitly performed
New_DM. Step:   499
Re-using DM from previous geometry...
Extrapolating Density Matrix...
Density Matrix sparsity pattern changed.
rank 24 in job 1  wardlaw63_48676   caused collective abort of all ranks
=====================

Is something reset or recalculated in the 500th step that breaks if the
geometry
has changed? Can I prevent that?

Cheers,

    Herbert
--
Herbert Fruchtl
Senior Scientific Computing Officer
School of Chemistry, School of Mathematics and Statistics
University of St Andrews
--
The University of St Andrews is a charity registered in Scotland:
No SC013532


--
Herbert Fruchtl
Senior Scientific Computing Officer
School of Chemistry, School of Mathematics and Statistics
University of St Andrews
--
The University of St Andrews is a charity registered in Scotland:
No SC013532

Responder a