> &cntrl
> nstlim=500000, dt=0.001, ntx=5, irest=1,
> ntpr=500, ntwr=500, ntwx=500, ntwv=500,ntwe=100
>
> tempi=300.0, ntt=0, tautp=2.0,
>
> ntb=1, ntp=0,
>
> ntc=2, ntf=2,
>
> nrespa=1,
> &end
>
> I am experiencing a DOWN-ward drift in my total energy. I know that with a
> similarly equilibrated system, with the same input file except for dt=0.002 I
> experience the "normal" less than optimal upward drift in Etot. Has anyone had
Normally I would not expect an "upward" drift in Etot with DT=0.002;
typically I see a slight downward drift (assuming Ewald or PME are being
performed with appropriate parameters). A Harvey et al. JCC article 19,
726 (1998) and a more recent Chiu et al. (Jacobson) JCC article provide
more details.
Constant pressure and shake tolerances that are too low can lead to energy
loss, albeit at a very slow rate. Set the shake tolerance to be really
low, TOL=0.00000001 and then set the constant pressure coupling time to be
long (tautp=10.0 or larger noting that the larger you make the constant,
the closer you get to constant V but with pressure information reported!).
Also, make sure that you are reading BOX information (if using versions
before AMBER 9.0, set ntx=7 rather than ntx=5 on restart). Finally, make
sure you are removing center of mass translation (if periodic) and center
of mass rotation/translation (if non-periodic), i.e. set NSCM=1 or some
small number of steps that is a multiple of the restart file dumping
frequency during the equilibration phase. If energy is still dropping,
then likely the system is not fully equilibrated or I am forgetting
something obvious (--I assume you are wrapping coordinates if periodic,
IWRAP=1 and/or not restarting too frequently?--). In my experience, AMBER
will (after equilibration) easily maintain constant E with stringent shake
tolerances, no RESPA, and very weak pressure coupling for *normal*
systems. Of course if you are running with a modified force field of any
sort that has large bond/angle force constants, or with restraints (with a
high force constant), a smaller integration time step may be required.
If the above doesn't solve it, to dig deeper we'll need more
information...
--tec3
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Thu Jul 13 2006 - 16:34:52 PDT
