AMBER: Langevin dynamics in NAB

From: David A. Case <>
Date: Sun, 20 May 2007 18:39:42 -0700

On Fri, May 18, 2007, wrote:

> >>Hi, in my MD simulation (freezing 1141 atoms, moving 255) using NAB
> >>and options shown below, the temperature drifts from the set 300K to
> >>an average of 290.5K.

My previous reply to this was not correct. I had forgotten that Amber and NAB
treat Langevin dynamics a little differently.

There is an ambiguity about the definition of velocity in Langevin
integrators; for a discussion see Pastor, Brooks & Szabo, Mol. Phys. 65,
1409-1419 (1988). For a "straightforward" implemtation of the integrator we
use, and defining velocities in the "verlet" way (i.e. by Eq. 2.8b of this
paper), one should expect the apparent temperature of the system to be too low
by a factor of (1 + gamma_ln * dt/2). (See Eq. 3.5c of the above paper.) For
gamma_ln = 50 ps**-1 and dt = 0.001 ps, this is an error of 2.5%, or about 8
degrees at 300 K. This is what NAB does, and why you are seeing a reported
temperature that is too low.

We usually run at much lower gamma_ln (say about 5 ps**-1 or so), so that
the apparent error is less than 1 K. Also, Amber (but not NAB) adjusts the
velocities to account for the above "error". [Amber follows Eq. 11 of
Loncharich, Brooks and Pastor, Biopolymers 32:523-535 (1992).] The combination
of these two things led me to not recognize the symptoms you reported.

Future versions of NAB will probably be more like Amber in this respect,
although no simple modification is "correct" in all respects. The bottom
line is that what you are seeing is expected. You might consider if you
really need such a large collision frequency; but most likely you can just
ignore it (or mentally mutliply the reported temperatures by 1.025).

Thanks to Adreas Svrcek-Seiler for reminding me about the underlying problem


The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to
Received on Wed May 23 2007 - 06:07:12 PDT
Custom Search