[CPMD-list] Wrong kinetic energy when running with LAM (but not for serial exec.)

[Axel Kohlmeyer]

On Thu, 22 Nov 2007, Henryk Wicke wrote:

HW> Dear Axel,

dear henryk,

thanks for being careful and persistent. i could reproduce
your observations with a several machines and thus you have
stumbled across a real bug. it took me a couple of hours to
track it down and the reason for the inconsistent value of
EKIN is the fact, that the input file, that you were testing
was asking to also dump the total density and the electrostatic
potential. this in turn triggers the recomputation of the
kinetic electron energy. in a parallel run the density is
distributed across the nodes and thus you get this behavior
of ekin being (almost) the serial value divided by the number
of nodes (it is not exactly, because the contribution of
each node is a bit different). 

the good news is, that your calculation is working fine and
the overwriting of the ekin term happens only _after_ the
calculation is done. also the energy term that matters - 
the total energy - is not affected and summed up over all
nodes long before. if you remove the lines in the test input
to referring to RHOOUT, ELECTROSTATIC POTENTIAL and ELF,
you should recover the proper value for EKIN.

since i know what is happening, i can come up with a bugfix.
even though this is purely cosmetic, i should be fixed.

HW> 
HW> first of all, thank you very much for your help. It's very much appreciated!

you're welcome. 

[...]

HW> Further comments would be very much appreciated.

i'd like to thank you again for your report and for
being one of the few people that actually _check_
compilations for correctness. this is very commendable
and despite its widespread use CPMD (still) has a lot
of inconsistencies and needs careful people to try it
out and report them, so that they can be fixed. i wish
there were more people paying attention to these details
_and_ reporting them.

cheers,
   axel.

HW> Best regards,
HW> Henryk
HW> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.