AMBER Archive (2007)

Subject: RE: AMBER: ntt=1 or ntt= 3?

From: Hu, Shaowen \(JSC-SK\)[USRA] (Shaowen.Hu-1_at_nasa.gov)
Date: Thu Mar 22 2007 - 15:24:10 CST

Hi Dr. Case,

Is this possible to use ntt=3 for simulation annealing? It seems that
all people use ntt=1 for SA.

Thanks,
Shaowen

-----Original Message-----
From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf
Of David A. Case
Sent: Wednesday, March 21, 2007 2:30 PM
To: amber_at_scripps.edu
Subject: Re: AMBER: ntt=1 or ntt= 3?

On Wed, Mar 21, 2007, Therese Malliavin wrote:
>
> So, I decided to switch from ntt=1 to ntt=3 for running an usual MD
> simulation in the NTP ensemble (without QMMM). Before that, I was
> heating the system using ntt=1 and a constant volume ensemble.
>
> But, the equilibration simulation has a problem with the water density

> which is about 0.84 in place of slightly larger than 1.

I think we need more information. I've equilibrated many systems using
ntt=3 and gotten correct densities. So, I don't think there is anything
intrinsically wrong with that option. You could equilibrate with ntt=1,
getting a proper density, then continue with ntt=3, and the temperature
and density should continue to be appropriate.

> mean pressure calculated over
> 20 ps is about -100 atm, whereas I always found it around 0 in the
> past when I was using ntt=1).

This part sounds correct: if the density is too low, the pressure should
be negative (saying that the system wants to contract). But it's not
clear why the system doesn't contract, leading to a higher density. What
value of taup and ntp are you using? How long was the equilibration?
Did you "start over"
in equilibration, or continue an existing run where ntt=1 had been used?

Prof. Duan is correct in saying that ntt=1 with a value of tautp of 1 or
2 doesn't show obvious problems (for explicit solvent simulations). I'm
not sure what he means by saying "for some reason, tautp=0.2 remained on
the manual." (I don't see that myself: the default value is 1.0). The
use of
ntt=1 can become problematic for implicit solvent simulations, where
there are relatively few degrees of freedom. And, the Berendsen
algorithm is fragile, and can lead to a non-uniform distribution of
temperature inside a simulation even when the overall temperature looks
OK.

On the other hand, Langevin (ntt=3) simulations can also exhibit funny
behavior, especially if the same random number seed is used for repeated
simulations. This problem has a long history, but a good recent
overview is
here:

%A B.P. Uberuaga
%A M. Anghel
%A A.F. Voter
%T Synchronization of trajectories in canonical molecular-dynamics
simulations: Observation, explanation, and exploitation %J J. Chem.
Phys.
%V 120
%P 6363-6374
%D 2004

A recommendation is that you should explicitly set the random number
seed
("ig") to new values at each restart of an ntt=3 simulation.

The bottom line is that all methods of constant T simulation have
idiosyncracies (including Nose-Hoover thermostats, not discussed here),
and one needs to take care. But the massive problems reported by
Therese probably have some other origin.

...dac

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu