Re: AMBER: Temperature fluctuation BIG with ntt=3

From: David A. Case <>
Date: Thu, 21 Sep 2006 18:39:46 -0700

On Thu, Sep 21, 2006, Mingfeng Yang wrote:
> In Amber manual, it's recommended that Langevin dynamics (ntt=3),
> instead of Berendsen coupling (ntt=1), should be used to control the
> temperature, especially for GB simulation. However, it seems the
> temperature fluctuation (20-40) is pretty significant in the case of
> ntt=3, while it's only ~6 when ntt=1.
> For example, in my current system (a deca-peptide) simulated with GB
> model at 300K. The temperature can sometimes jump to >400K. Is it a big
> deal?

This subject was just discussed recently on this mail-list:

If you use the formula given there, and assume a molecular mass of 1 kD
(roughly correct for a 10 residue peptide), then the predicted
root-mean-square fluctuation in temperature is about 24 degrees. Since you
say you are getting a temperature fluctuation of "20-40", that sounds about
right to me. With this large an rms deviation, seeing an occasional spike
of 100 degrees is not out of the question.

You might try a gamma_ln of 5-10, just to see if that prevents the really big
spikes (although you still want the rms fluctuation to be something like the
number quoted above.)

As discussed in the earlier mail, fluctuations are inversely proportion to the
size of the system, and can be (and should be) quite large when you have only
a single, relatively small peptide in your simulation. The smaller results
you quote for ntt=1 are a reflection that you are *not* getting a canonical
ensemble with that thermostat.

...hope this helps....dac

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to
Received on Sun Sep 24 2006 - 06:07:14 PDT
Custom Search