[gmx-users] Compressibility issue - real concern

[Erik Lindahl]

On May 3, 2007, at 5:36 PM, Jones de Andrade wrote:

> Hi Erik.
>
> I was thinking something about it exactly now. I mean, Berk quoted  
> that the scalling goes with beta/tau in Berendsen, and with beta/ 
> tay^2 in Parrinello. I mean, in a similar implementation, how  
> should it go for a nosè implementation?
>
> Also, what would be a good value for tau_p so? I was looking in,  
> for exemple, the "spider tutorial", and there they use a tau_p  
> value of .5, not so different from my value of .4 (and not .1).

There is no widely accepted definition of coupling time for Nose or  
Parrinello barostats. Formally, the entity that enters into the  
equation is the mass of the external heat bath, but since this is an  
extremely awkward concept we have decided to use a "coupling time"  
that corresponds to the period of oscillations at equilibrium.

I would use a tau_p value in the range 1-10ps to avoid interfering  
with the atomic dynamics, and for Parrinello-Rahman even 1ps is  
probably in the lower range since you should compare an entire  
fluctuation period with the simple exponential relaxation of  
Berendsen barostats.

Second, you are asking for problems when you apply any extended- 
ensemble thermo/barostat to a system far from equilibrium since  
you'll get lots of ringing oscillations. Start with Berendsen scaling  
to get the system where you want, and then switch to extended  
ensemble coupling.

Cheers,

Erik