AMBER Archive (2009)

Subject: Re: [AMBER] default integration scheme in 7 vs. 10

From: case (
Date: Thu Jul 02 2009 - 06:37:17 CDT

On Wed, Jul 01, 2009, Paul Brandt wrote:
> I know this question is asked often, but I'd like to confirm what I've read
> in the amber mail archives. Which integration scheme is used by Sander in
> Amber 7 and by PMEMD in Amber 10?
> If I read correctly in the archive, in 7, leap frog integration was
> employed, and in Sander and PMEMD in 10, velocity Verlet is used. Is this
> right?

Although it may sound like a simple question, it's actually rather complex.
Here are several comments that may help.

1. Algebraically (i.e. ignoring effects of finite precision floating point
calculations) these methods give equivalent trajectories. Taking finite
precision into account, the Amber scheme is closest to a leapfrog scheme.

2. One big difference is how the kinetic energy is estimated. That has been
updated in Amber, so that current versions will give the same sequence of
coordinates as previous ones, but slightly different estimates of the kinetic
energy, and hence of apparent energy conservation.

3. Most practical simulations use SHAKE and/or a thermostat or barostat.
These add a lot of complexity, and defeat attempts to use simple phrases like
"velocity Verlet" or "leapfrog" to describe what is done.

4. For historical reasons, Amber restart files are "leapfrogish": the
velocities are written to the file at a different time point than the
coordinates. A velocity Verlet scheme would more naturally save positions and
velocities at the same time point.

5. Given these complexities, the code itself is the only definitive source
of what is actually done.

...hope this helps....dac

AMBER mailing list