AMBER Archive (2004)

Subject: Re: AMBER: Fwd: NVE & polarizable force field]

From: darden (
Date: Sat Jan 24 2004 - 15:14:55 CST

Dear Martina
Dr. Kim is basically right on both accounts. The "Car-Parrinello"
propagator should give you good energy conservation over the 50ps time
frame i.e. it won't look dramatic compared to fluctuations. But over a
nanosecond run you will definitely notice it. Usually I saw gentle heating
instead of cooling but, with iterative methods that don't fully converge
(that is to ~10^-6 debye rms diff between subsequent iterations) you can
get either heating or cooling, so I'm not that surprised. The drift with
Car-Parrinello shouldn't be too great at 1ns compared to non-polarizable
runs however (I got about 10x more than with spc).
Also, Dr. Kim is right about restarts. The energy drops
during iterations so that if you are not fully converged you are running
at a higher energy---the Car-Parrinello method in my experience runs at an
estimated error of ~0.05 debye compared to fully converged. However, at
step 1 you always do a full convergence to 10^-6 debye.
In practice, I think NVT with a weak bath (1-5ps for diptau as well as
tautp) works well for long runs.
If your cooling is more than about 10x the drift in nve with
non-polarizable potential water models like spc let me know. I never
tested with slab conditions so maybe there's something about the interface
making it (Car-Parrinello) not work as well
Tom D
On Fri, 23 Jan 2004, Martina
Roeselova wrote:

> Hi,
> I am not sure if my email got through. I have't received any answer and
> it haven't found it in the Amber reflector archive either. This time, I
> am trying with no figure attached.
> M.R.
> -------- Original Message --------
> Subject: NVE & polarizable force field
> Date: Thu, 22 Jan 2004 11:05:46 -0800
> From: Martina Roeselova <>
> To:
> Hi AMBER users,
> I run simulation of a water slab using polarizable potential (POL3).
> I use 3D periodic boundary conditions to construct the slab geometry. My
> simulation box is 30-by-30-by-100 Angstroms and contains 864 waters. The
> cutoff is 12 Angstroms. I use Ewald. (see the imput file below)
> I noticed few things that I find disturbing:
> (1) Total energy drift during NVE run: Etot (as printed in mdout file)
> goes down steadily. The NVE simulation was started from a restart file
> produced by a previous 0.5 ns NVT run.
> (2) Discontinuity after restart: After restart of NVE run, there is an
> abrupt drop in total energy Etot (mainly due to the drop in potential
> energy EPtot) compared to the values corresponding to the last step of
> the previous NVE run that was used for restart. In the first few ps, the
> total energy goes up, ABOVE the Etot value of the final step of the
> previous run, and then starts going down again (see attached figure -
> please note the difference in the energy scales for total energy as
> opposed to kinetic and potential energy).
> I see the same discontinuity when restarting a NVT run. It is of the
> same size as in the NVE case, however, it is still much smaller than the
> size of the energy fluctuations during the NVT run - and there does not
> seem to be any drift.
> Is it possible that (1) and/or (2) are somehow related to the use of
> polarizable potential, i.e. the way the dipols are propagated?
> Thanks for any hints/comments/explanations.
> Martina Roeselova
> here is the input file:
> &cntrl
> imin = 0
> irest = 1, ntx = 7
> nstlim= 1000000, dt=0.001
> ntb=1, ipol=1
> ntxo = 1, iwrap=0
> ntpr=1000, ntwx=1000, ntwv = 1000, ntwe=0, ntwr=1000
> ndfmin = 0,ntcm = 0,nscm = 9999999
> ntt = 0, temp0 = 300.00, tautp = 0.2, dtemp = 999.
> cut =12.
> ntc=3, ntf = 3
> maxcyc = 9999,ncyc = 250,ntmin = 1,dx0 = 0.010
> dxm = 0.02,dele = 0.01,drms = 0.00010
> &end
> &ewald
> a=30., b=30., c=100.
> &end
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to
> To unsubscribe, send "unsubscribe amber" to

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to