[CPMD-list] problem with path integrals restart

[Axel Kohlmeyer]

On Wed, 2 Apr 2008, Marek Naryniecki wrote:

MN> Dear  Axel and Serge

dear marek,

MN> I put 2 replicas for fast check of the problem on varius machines and
MN> pseudos. This problem also occure with 8 replicas in 120K. I know it is
MN> small number but more replicas is to expensive for me. I'll try some short

apropos accuracy vs. speed. i noticed that you are using goedecker
pseudopotentials, those usually need rather high plane wave cutoffs.
in particular, i'd be concerned that with oxygen atoms you don't 
have converged forces at 100ry. from my experience, you may need 
120ry to 150ry, whereas equivalent MT potentials for your system 
should have reasonably converged forces at 85 to 100ry.

and please rethink your argument of "i cannot afford more". this may
well be true, but what use is a calculation that makes - for reasons
of computational efficiency - overly large approximations? 

there is always some reasonable range of "what is converged enough?"
and you have to consider how much the level of convergence will affect 
your results when interpreting them. but if you are outside of the 
reasonable range you are just running a very expensive random number 
generator. a popular question in quantum chemistry classes is:
how useful is a coupled cluster calculation with an STO-3G basis set?

cheers,
    axel.


MN> calculation with 32 replicas.
MN> 

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