AMBER Archive (2008)

Subject: Re: AMBER: NVE energy and temperature drift

From: David A. Case (
Date: Fri Mar 07 2008 - 10:32:51 CST

On Fri, Mar 07, 2008, Lars Skjærven wrote:
> This is a follow up question to a thread in 2007 regarding NVE simulations (
> I had
> the same problem with energy drifting. Thus, I followed the recommendations
> from Case and Cheatham; lower shake tolerance, increase the accuracy of the
> PME by reducing DSUM_TOL by an order of magnitude, and dt=0.001. The
> simulation is much "smoother" now in terms of energy and temperature. The
> shift has now been reduced to 5 kcal/mol (-75489 to -75484, first to last
> step) during a 10ns simulation. The energy does not yet seem to converge.
> Should I be concerned about this energy drift, or is it small enough such
> that I can accept it? My apologies if this has been answered earlier.

This depends somewhat on the size of your system, since the absolute drift
will depend on how many atoms you have. In my experience, the total energy
will never converge: it will just keep rising approximately linearly in time
(at least up to 200 nsec, the longest single simulation I have).

Here's the argument I often make: check the average temperature near the
begining and the end (say over the first and last nanoseconds). If that is
constant to within about 0.1 (or maybe 0.2) K, then the heating of the system
is not likely to affect your results very much. Experimental temperature
control in most protein experiments is no better than this anyway. Of course,
this is only a heuristic argument, not a rigorous one.


The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to