[lammps-users] Energy not conserved in an NVE sim .. Included Cu simulation results too..


It is important to remember that all molecular dynamics simulations are
approximate. They *never* conserve energy perfectly. That said, some
simulations are better than others, for the reasons given below. The energy
conservation you are getting in RDX with ReaxFF is about as good as I would
expect. Here is a response I posted on this very subject last year:


Date: Thu, 14 May 2009 09:57:33 -0600
From: "Thompson, Aidan" <[email protected]>
Subject: Re: [lammps-users] ReaxFF: Energy conservation problem

First, some general comments. Energy is never perfectly conserved in NVE
dynamics. The errors are not symmetrically distributed: positive errors
on average greater than negative errors, and the total energy tends to
upwards. The rate of drift depends on:

1. the timestep size
2. the shape of the potential energy surface
3. the temperature.

In addtion, any discontinuities or other errors arising from the computed
potential energy surface and its higher derivative have a strong effect


Dear Aidan, Zhun-Yong, Steve,

Thank you for your replies. I expected some fluctuations in total
energy, but not so much.

I am yet to get an intuition on how much error to expect in NVE
simulations. I expected timestep size and number of atoms simulated to
play a part. As Aidan points out temperature also will play a part for
a given timestep.

The shape of the PES contributing to energy conservation was a
surprise because I assumed that the small time-step took care of not
moving much within the 3N dimensional energy landscape.

Thank you once again.
Best Regards