[gmx-users] a question about ensemble

[baptista at itqb.unl.pt]

On Sat, 5 May 2012, Dommert Florian wrote:

> On Sat, 2012-05-05 at 13:19 +0200, Patrick Fuchs wrote:
>>>> Interesting discussion indeed. I'm just thinking that there might be
>>>> no fundamental difference to other thermostats. There's nothing in the
>>>> way that causes the friction, but then again, there's no physical
>>>> particle that causes the stochastic term in v-rescale, and the
>>>> Nosé-Hover particle is not physically "touching" the atoms either.  In
>>>> all cases the surroundings couples to the atoms in a way that can't be
>>>> seen in e.g. a test tube.
>>>>
>>>>
>>>> Erik
>>>>
>>>
>>> Hi Erik,
>>>
>>> yes, that is a good point, which I haven't really thought about.
>> Indeed, in a macroscopic system heat will have to diffuse from the edges
>> of the container to the inside of the system. In the simulations, the
>> rescaling of velocities is done on all the atoms at the same time,
>> either on a local or global basis. For the use of SD in explicit
>> systems, the friction and stochastic terms (with the carefully chosen
>> friction coeffictient) can be seen as applying one thermostat per degree
>> of freedom. So Florian, I think you should change your view about the
>> random kicks and friction as a way to modify velocities for
>> thermostating, which is different from mimicking solvant effects in
>> implicit simulations (even if it uses also the friction and random
>> forces).
>
> Yes, I know and already did so, as mentioned yesterday.
>
>
>> What we want is that the thermostat samples the proper
>> canonical ensemble. Then if it does, what also matters is to be aware
>> whether it affects other properties, notably dynamics (because rescaling
>> velocities will in general pertubs dynamics). There has been a nice
>> discussion some years ago in the GROMACS mailing list about that:
>> http://lists.gromacs.org/pipermail/gmx-users/2008-July/035302.html. Also
>> Bussi and Parrinello discuss this aspect in that paper:
>> http://dx.doi.org/10.1016/j.cpc.2008.01.006.
>
> Yes, I am aware of this, too, that is why a careful choice of the
> parameters is inevitable. Especially, the more you get into detail by
> reading various literature, the more you realize, that it is really not
> a simple task to sample an exact canonical ensemble.
>
> However, I thank you all again for this discussion, because thermostats
> and barostats are such an important topic.

Concerning the thermostat realism, there is an instructive experiment that 
can be done for a homogeneous system. If you define a very large 
simulation box where a stochastic (more collision-like) thermostat is 
applied only within a layer close to the box "walls", you will see a 
behaviour that that roughly mimics the exchange of momentum with the walls 
of a real container. I did that a long time ago when I started doing 
simulations, using the Andersen thermostat within a thin layer of an 
otherwise NVE Lennard-Jones system. I remember that the directional 
heating/cooling of a cold/hot core could be easily observed for a 
dense-enough system (where atoms don't diffuse too quickly between the 
core and the "walls"). Obviously, this is just a rough approximation of 
the real thing, because of the periodic boundary conditions and the 
non-macroscopic box size, but it was really cool to see things proceeding 
"almost as they should" in terms of dynamics... :)

An interesting approach to minimize the perturbation of the Newtonian 
trajectories is to use Gauss' principle of least constraint, as done by 
Evans in the 1980's to derive an isokinetic thermostat. Although the 
resulting isokinetic ensemble has a non-canonical distribution of momenta, 
its distribution of coordinates is rigorously canonical, which makes it 
perfectly suitable for most applications. In principle, it should be 
possible to play with other constraints while keeping the trajectories as 
unperturbed as possible, but I don't know if other attempts have been done 
more recently along this line.

Best,
Antonio

>
> /Flo
>
>
>> Ciao,
>>
>> Patrick
>>
>> --
>> _______________________________________________________________________
>> Patrick FUCHS
>> Dynamique des Structures et Interactions des Macromolécules Biologiques
>> INTS, INSERM UMR-S665, Université Paris Diderot,
>> 6 rue Alexandre Cabanel, 75015 Paris
>> Tel : +33 (0)1-44-49-30-57 - Fax : +33 (0)1-43-06-50-19
>> E-mail address: patrick.fuchs at univ-paris-diderot.fr
>> Web Site: http://www.dsimb.inserm.fr/~fuchs
>
> -- 
> Florian Dommert
> Dipl. - Phys.
>
> Institute for Computational Physics
> University Stuttgart
>
> Pfaffenwaldring 27
> 70569 Stuttgart
>
> EMail: dommert at icp.uni-stuttgart.de
> Homepage: http://www.icp.uni-stuttgart.de/~icp/Florian_Dommert
>
> Tel.: +49 - (0)711 - 68563613
> Fax.: +49 - (0)711 - 68563658
>
--
Antonio M. Baptista
Instituto de Tecnologia Quimica e Biologica, Universidade Nova de Lisboa
Av. da Republica - EAN, 2780-157 Oeiras, Portugal
phone: +351-214469619         email: baptista at itqb.unl.pt
fax:   +351-214411277         WWW:   http://www.itqb.unl.pt/~baptista
--------------------------------------------------------------------------