[CP2K:767] Re: nose thermostat

[Teodoro Laino]

Ciao Nick,

On 10 Mar 2008, at 22:28, Nichols A. Romero wrote:

> To elaborate on what Teo said, (he can correct me if I am wrong)
>
> In NVT, the total energy of the *real sytem + thermostat* should be  
> conserved.
> (I don't remember if CP2K outputs this quantity, but most codes  
> should).
Exactly.. and this quantity is what we call CONSERVED QNTY or you can  
find the same number in the 6th column of the .ener file.

> The total energy of just the real system (atoms,molecules,etc.)  
> should have a mean value (after equilibration) with fluctuations  
> proportional to 1/sqrt(N) (or something like that).
>
> If the total energy of the real system increases or decreases  
> without bounds, then there is something wrong.
> Or maybe you are near some phase transition.
Absolutely correct. In this case I think that the forces computed  
have something wrong..
Let me explain better once for all. The SCF module is still not fully  
completed. The most important part missing is
the ewald for the coulomb part (we have a project going on for that,  
so it should be hopefully fixed one day ;-) ).
This means that at the moment the coulomb is evaluated through a  
neighbor lists.
This creates ENORMOUS problem if you condensed phases (same problem  
of using a cutoff to truncate the electrostatic interactions
in force field based calculations) but also if you do big molecules  
(and this is her case) and don't set properly
the parameter I wan mentioning in my last e-mail.
What happens here is that during the MD you've interactions turned on  
and off at the boundary and the contribution is such that
you observe a drift.

Of course I could be wrong.. but since the nose-hoover thermostat has  
been so extensively tested i would rarely look for a bug in it ;-)

Teo

>
>
> On Mon, Mar 10, 2008 at 4:37 PM, Teodoro Laino  
> <teodor... at gmail.com> wrote:
>
> Ciao Diederica,
>
> >
> > These are added, as an attachment this time
>
> Thanks.. will look into this problem ASAP.
>
> > I'm modelling an isolated molecule, it should fit in the box. Using
> > NVE without temperature rescaling, with an analogue structure, an
> > energy drift was observed, that got worse after about 20 ps, though
> > no non-convergence was observed. The input file and an energy-plot
> > are added.
> If you have a drift with NVE this means that energy and forces are
> not consistent.. or (what I assume) that the cutoff for the coulomb
> interactions is not large enough..
>
> Please put all of the  RC_COULOMB and RC_INTERACTION and RC_RANGE to
> 50.0.
> You've the problem that in your cell the standard range (24 bohr)
> does not fully cover the extension of your molecule (more than 26  
> bohr).
> Use a larger box (30.0 Angstrom) and set the above keywords to 50.0
> bohr.
> You should not see any jump in the energy (or drift)..
>
> > The potential and kinetic energy are conserved indeed, but the
> > conserved quantity is not
> Sorry to disprove you, but it is not possible that in an NVT you have
> kinetic and potential conserved, since the thermostat
> is continuously exchanging energy with the system.  I't s a matter of
> concept.. we may discuss about the magnitude of the
> thermostat energy.. but that's another story..
>
> Cheers
> teo
>
>
>
> >
> >
> >
> > For the next time, can you please put tables and input files as
> > attachment (maybe .tgz) so that the formatting of mail readers are
> > not changed and the input files not digested ? Thanks..
> >
> > Ciao,
> > teo
> >
> > Thanks for helping!
> >
> > Greetz,
> > Diederica
> >
> >
> >
> >
> > >
> > <mailcp2k.tgz>
>
>
>
>
>
>
> -- 
> Nichols A. Romero, Ph.D.
> DoD User Productivity Enhancement and Technology Transfer (PET) Group
> High Performance Technologies, Inc.
> Reston, VA
> 443-567-8328 (C)
> 410-278-2692 (O)
> >

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20080310/c86989bc/attachment.htm>