[Pw_forum] A few questions about MD simulations
marchywka at hotmail.com
Sat Jun 1 16:06:41 CEST 2013
> From: paolo.giannozzi at uniud.it
> To: pw_forum at pwscf.org
> Date: Sat, 1 Jun 2013 15:49:34 +0200
> Subject: Re: [Pw_forum] A few questions about MD simulations
> On Fri, 2013-05-31 at 12:49 -0400, Robert Hembree wrote:
>> I noticed that upon completion of the simulation that the
>> wavefunction files seem to have disappeared.
> wavefunctions do not "disappear", but only those at the last step
> (or the last two, or three, if wavefunctionn interpolation is used)
> are kept
>> Is there an easy way to access this information? Or would it require
>> large changes to the code base?
> it would require a lot of disk space. It should be relatively easy
> to modify the code in such a way that it saves all wavefunctions.
> Please look into the code that performs wavefunction interpolation,
> "update_wfc", for inspiration. Or, you can do the simulation one
> step at the time, save wavefunction files under a different name.
I've run into this issue on my lap top with JDFTX. Don't ignore the posibility
of adding compression code, something like gzip that could use the statistical
properties of the wavefunction values. Offhand I don't know if adding gzip to
the code would help but the time and effort spent on reducing size could be paid back
in a few disk transfers. It also may be interesting to try to write a lossless ( exactly invertible ) compression
algorithm for wavefunctions, or potentials for that matter, as compression relies on
statistics or properties of the data. In theory of course, your PW input file is the most concise
way to describe your wavefunction but "decompression" relies on starting the simulation from
scratch again LOL. The point being that it may be an interesting problem to consider for a number
>> Third: In the source files where does the MD simulation start? For
>> example, an scf cycle is handled in the file electrons.f90.
> look into "move_ions.f90". The MD code is unfortunately very confusing.
>> is the correct way to access the planewave expansion coefficients
>> through the
>> USE wavefunctions_module, ONLY : psic
> no. Wavefunctions are contained in variable evc(i,j), i=plane-wave
> index, j=band index
> Paolo Giannozzi, Dept. Chemistry&Physics&Environment,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
> Pw_forum mailing list
> Pw_forum at pwscf.org
note new address
Mike Marchywka 2295 Collinworth Drive Marietta GA 30062.
More information about the users