AMBER Archive (2008)

Subject: Re: AMBER: SPCFW temperature oscillations

From: David A. Case (
Date: Tue Nov 18 2008 - 06:13:16 CST

On Mon, Nov 17, 2008, Naser Alijabbari wrote:

> Using SPCFW, I have only been able to run a 'production run' with an
> integration time step of 2fs. Lower times (1fs or .25fs) result in wild
> fluctuations of temperature and energy. Here is what I have tried:
> After 100ps of adjusting my density using this input file:
> &cntrl
> ntt = 3, gamma_ln = 1.0,
> nstlim = 50000, dt = 0.002,
> &end
> Then I tried running a 10ps test production run using Berendsen algorithm
> and 1fs integration time but my energy and temperature fluctuated from 300K
> to ~330K. I thought my time step was too large so I tried .25fs integration
> time:
> &cntrl
> imin = 0, irest = 1, ntx = 5,
> ntb = 2, pres0=1.0, ntp=1,
> ntt = 1, nscm=0,
> nstlim = 40000, dt = 0.00025,
> ntpr = 1, ntwx = 6, ntwr = 1000,
> &end
> &ewald
> ew_type=0, dsum_tol=0.000001
> &end

Sander (and also pmemd, I think) does weird things when you "restart" a
simulation with a different time step -- basically, it does not know that the
input velocities were created with a different time step, so you usually have
to re-equilibrate at the new time step, at least for a short while.

My guess is that the 2fs time step with SPCFW was actually unstable, but that
the Langevin thermostat was hiding that fact. When you then switched to a
very weak coupling thermostat and changed the time step, the instability

I've never used SPCFW myself, but my guess is that something like a 0.5fs time
step is required. Try equilibrating (say with ntt=3) using that time step,
then changing to NVE or ntt=1.

[If you have a solute, like a peptide or protein, you should also be aware
that protein-water interactions using SPCFW have not been tested nearly as
much as for rigid water models. So, be sure to proceed with caution.]


The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)