[CPMD-list] Quench Bo Electrons

[Axel Kohlmeyer]

On Fri, 20 Jul 2007, Samir H. Mushrif wrote:

SM> Hello CPMD users,

hello samir!

SM>                   Thank you Axel for your suggestions on my previous 
SM> mail regarding the monoclinic cell symmetry problem and the unit cell 
SM> size problem (CPMD list digest - July 10). As per your suggestions, I 
SM> did several single point calculations and then used the Murnaghan 
SM> equation of state to find the optimum lattice constant. It is working 
SM> well and is solving my purpose.

ok. great.

SM>      The question I have this time is regarding the MD. As per the 
SM> instructions given on your website (for hydronium ion in bulk water)
SM> 
SM> http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-intro.html#relax

please note, that the number of timesteps for the 'initial'
NVE calculation of this example is the absolute minimum for
a configuration that has been well pre-equilibrated. most
systems need more (a couple of thousand MD steps or more)
until the system is reasonably equilibrated. 

SM> I did the initial MD (after optimizing the wavefunction) without using 
SM> thermostat (using TEMPCONTROL IONS) to find out the EKINC value (which 
SM> comes to ~0.17 for my system) and then I tried to do QUENCH BO 
SM> ELECTRONS. But my output gives error as

hmmm, 0.17 looks a little high for typical systems. for how long
did you run with NVE and how did EKS behave during that time?
did it fluctuate around a simple average value or did you still
see a drift.

SM> PROGRAM STOPS IN SUBROUTINE MEMORY. TOO BIG VALUE.

what platform are you running on? 
32-bit/64-bit, OS, compiler, parallel/serial, etc...

can you provide the few lines of output before and after
this message (so we can see where it comes from)?

quench bo needs a little more (temporary) memory than 
the regular code, and if your malloc uses a 32-bit integer
for the size, if may not be able to handle this allocation.
note, this error does not come from the fact, that you don't
have any free memory, but that you are requesting, too large
a block at a time (which is strange, too, but can happen).

SM> For my system, the MD run (as well as thew initial wavefunction 
SM> optimization step) takes approximately 6.5 GB of memory and the total 
SM> memory of my machine is 24 GB. So why should this problem come when I 
SM> am trying to do the QUENCH BO option? Does it need more memory than a 
SM> typical wavefunction optimization? If yes, what is the other way to 
SM> proceed without using the QUENCH BO option?

the easiest way to reduce the size of memory allocations in cpmd would
be to run across more nodes in parallel. if you have enough memory in
the machine, you may want to try using twice as many cpus than you have
physically. there will be some overhead, but this should work around
the large allocation issue.

cheers,
   axel.

SM> 
SM> Thanks for all your help and suggestions.
SM> 
SM> Best Regards
SM> Samir
SM>   _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
SM> 
SM> Samir H. Mushrif
SM> Department of Chemical Engineering
SM> McGill University
SM> 3610 University Street
SM> Montreal, Quebec H3A2B2
SM> Canada
SM> Ph: (Lab)514-398-5170
SM>     (Off)514-398-4169
SM> 
SM> _______________________________________________
SM> CPMD-list mailing list
SM> CPMD-list at cpmd.org
SM> http://cpmd.org/mailman/listinfo/cpmd-list
SM> 

-- 
=======================================================================
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.