[gmx-users] RE : long-term drift and crash

[Stéphane Téletchéa]

BON Michael a écrit :
> Hi,
> 
> Sorry fot no answering accurately.
> Here is new insight for understanding my problem.
> 
> In fact, even simulations of pure water (tip4p and tip4p_ew) don't work !
> And those simulations crash with a strange periodic pattern : still using g_msd, I observe
> on the msd plot sudden jumps that occur exactly every 10.000 ps ! By "jump" I mean a violent
> increase of the slope of the curve that lasts a short time, which is exactly 0.1 ps each time in my case. The curve is linear with the correct diffusion constant otherwise. I observe such jumps between 10.0 and 10.1 ps, 20 and 20.1 ps 30 and 30.1 ps, and so on, and each time the jumps have a higher magnitude than the previous one, till the simulation crashes.
> 
> I don't see anything in my parameters that could explain such a periodicity of 10.000 ps. I tried to half the timestep, and it halved the duration of the jump, which thus became 0.05, but it didn't changed the periodicity of 10 ps. Changing tau_t had no consequence.
> 
> Here are again my parameters for water simulation : 
> constraints = all-bonds

> dt = 0.002

I think you should either use
- constraints = h-bonds,   dt = 0.002
or
- constraints = all-bonds, dt = 0.004
but not a combination of both.

Constraining all-bonds is *merely only* for getting a bigger time step 
for md, but for regular simulations, i would definitely only constrain 
h-bonds and use dt=0.002.

Cheers,
Stéphane

-- 
Stéphane Téletchéa, PhD.                  http://www.steletch.org
Unité Mathématique Informatique et Génome http://migale.jouy.inra.fr/mig
INRA, Domaine de Vilvert                  Tél : (33) 134 652 891
78352 Jouy-en-Josas cedex, France         Fax : (33) 134 652 901