[gmx-users] zero temperature II, found feature

[Michael Brunsteiner]

--- Berk Hess <gmx3 at hotmail.com> wrote:
> >dear all,
> >In a previous mail I mentioned a problem
> >with zero temperature output. I seem to have found what the problem
> is and 
> >this looks like
> >a bug, or at least an undocumented limitation.
> >When I do md with a system containing only frozen atoms and
> (not-frozen) 
> >water both ener.edr and md.log say that the kinetic energy and the 
> >temperature are
> >zero. The thermostate seems to work though, because
> >my system does not "explode".
> >If I "un-freeze" the frozen atoms the
> >output is back to normal. The fact that frozen atoms
> >are formally provided with an infinite mass
> >might somehow cause this problem.
> >Any ideas for a work-around ?
> >Where in the code is E_kin calculated ?
> >cheers,
> >Michael
> 
> I have tested this some time ago (but I don't remember if I mailed it
> to the list) and for me it works fine. Both with berendsen and 
> nose-hoover coupling
> I get Ekin and temperatures that are normal. [...]
> I don't know which features of your system might cause the problem.

found it ... strange though, what happens is:
if I have two groups of atoms, one frozen and one not frozen 
("gra" and "HOH" in my case, frozen graphite and a water slab) 
and if I put the directive "comm_grps HOH"
in my md.mdp then the temperature output is ZERO.
If I use: "comm_grps HOH gra" the temperature output is also ZERO.
If I use: "comm_grps System" the temperature output is back to normal.
If I replace one of the water molecules by a single Argon atom
and combine this with "comm_grps HOH", then the temperature output 
is also back to normal.

This is strange since increasing the number of comm_grps
should reduce the number of degrees of freedom only by 3
per group (in the linear case), but there seems to happen 
something more drastic here ...
In fact I was always a bit worried about the temperature control
in combination with frozen atoms ... it'd be great if somebody
from the developers could check once more whether comm_grps and 
frozen atoms are properly accounted for when calculating the 
degrees of freedom, and whether the output temperature is REALLY
the temperature of the non-frozen part of the system.

a related issue: I seem to recall that at some point it was announced
(can't remember by whom) that in 3.2 there'd be the possibility
to thermostate only a part of the system (i.e.: SUM(tc_grps) != System)
... doesn't seem to work for me, though.

cheers,
Michael


__________________________________
Do you Yahoo!?
Yahoo! Small Business $15K Web Design Giveaway 
http://promotions.yahoo.com/design_giveaway/