Re: increase in energy over time

From: Edward Patrick Obrien (
Date: Mon Jan 03 2005 - 18:29:46 CST

Hi Peter,
   In the microconanical ensemble (NVE) energy drift (change in total
energy as a function of time) occurs due to finite integration errors
associated with discretizing newton's equations of motion.
Typically the total kinetic energy becomes more positive correlating with
an increase in temperature. You should check these two quantities, and see
if they are increasing.

   If you are running in the canonical (NVT) or Gibbs (NPT) ensemble your
total energy should fluctuate.

   You want to avoid energy drift in NVE because it is non-physical, and
can mess up your dynamics.

   To minimize energy drift make sure your system is well minimized
(Gradient tolerance < 1.0 or less), heat your sytem in steps (I typically
due it over the course of 300 ps) to the
temperature desired, and use a reasonable time integration step
(conventionally 1 fs at 300K).

   I don't believe you are seeing entropic effect, as the energy you are
probably looking at is the energy of the system (which neglects the
entropic term) and not the free energy.

   The minimization step is equivalent to carrying out a simulation at
temperature equal to 0 K, because the motion is determine only by the
gradient of the potential energy, and there is no kinetic term. Thus, if
you did not heat your system up properly, I believe jumping the
temperature from 0K up to 300 K could lead to bad van der Walls
interactions, which would cause your energy to drift.

Hope that helps,

On Mon, 3 Jan 2005, Peter Bazeley wrote:

> Hello
> I am simulating a protein with NAMD, using a basic set of parameters.
> I minimize my protein for 1000 timesteps, then run for 50000
> timesteps. I've found that the TOTAL energy of the system increases
> over time. I guess I'm trying to figure out the reason behind this.
> Could it be that I am seeing the entropic effects on the protein over
> time? Or perhaps the minimization created a structure that is lower in
> energy than would be expected in vivo, and the simulation brings the
> structure back to a more normal state?
> Thanks for any input,
> Peter Bazeley

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:27 CST