On Fri, Jun 25, 2010 at 8:06 AM, Axel Arnold
<address@hidden> wrote:
On Thursday 24 June 2010 22:14:29 Mikheil Azatov wrote:
> setmd time_step 0.0129; setmd skin 0.5
> set temp 1; set gamma 156
> thermostat langevin $temp $gamma
Just to illustrate what Burkhard was talking about:
For the discretized velocity-Verlet Langevin equation, the half-step "drag
force" is F= -gamma*mass*v/2, the change of velocity therefore dv = F/mass*dt
= -gamma*dt*v/2. Therefore, your simulation can only be stable if gamma*dt/2
is smaller than 1.
In your case with gamma=156, for dt=0.0128, it is 1.0062, for dt=0.0129,
0.9984, so it should be stable for and only for the latter time step, as you
observe. If you play around, you will see that even being 10^-6 above the
critical time step of 2./156 is enough to destabilize the system. As Burkhard
said, the transition is pretty sharp...
For practical applications, I would suggest to stay at least an order of
magnitude below this limit, since otherwise you are quite far from a proper
discretization of the Langevin equation, in which you assume that the velocity
change per time step is small.
Axel
--
JP Dr. Axel Arnold Tel: +49 711 685 67609
ICP, Universität Stuttgart Email: address@hidden
Pfaffenwaldring 27
70569 Stuttgart, Germany