Re: [AMBER] ETOT drift and nscm in NVE for calculation of diffusion

From: Jason Swails <>
Date: Tue, 21 Jul 2015 18:36:12 -0400

On Tue, Jul 14, 2015 at 2:13 PM, CHAMI F. <> wrote:

> Dear all,
> for calculating the diffusion it is recommended to avoid thermoast,
> barostat and wrapping so the use of NVE ensemble is necess
> ​​
> ary in order to compare to experimental values

​This is because diffusion is a temporal property and thermostats and
barostats do not preserve the meaning of time in classical Newtonian

> However, in NVE there is the energy drift issue, which can be due to
> numerical rounding, the choice of few control inputs :
> dsum_tol , tol and nscm flag for the removal of trans/rot of COM

​dsum_tol and tol certainly play into it, but equally important is the step
size (dt). If your drift is too high, you likely have to decrease your
step size.

> ​​
> my question is what is acceptable drift in energy for the calculated
> diffusion to be correct

​I'm not sure what the typical target is for energy conservation, but the
energy drift I believe scales with the number of particles in your system
(not sure of the exact relationship). If you want to measure diffusion
over, say, 10 ns, you will obviously need to make sure that the energy
drift is not significant over the 10 ns of your simulation (starting with
the simulation not blowing up).

I would suggest doing literature searches on various diffusion (or
rotational diffusion) studies and see what settings they use.

Another common approach is to use the Berendsen thermostat with a very long
coupling constant so that the dynamics is *effectively* NVE, but the energy
drift is dampened by the (very) weak heat bath coupling. Values like
tautp=20 and greater should suffice.

> ​​
> ps. there is also other output frequency and system size
​​output frequency has no effect.


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Tue Jul 21 2015 - 16:00:02 PDT
Custom Search