AMBER Archive (2003)

Subject: Re: Etot is not constant in NVE ensemble

From: Thomas Cheatham (cheatham_at_chpc.utah.edu)
Date: Fri May 09 2003 - 10:49:18 CDT

> Tom Cheatham (he can agree/disagree) had a similar problem and found that
> by increasing the default ewald coefficient a bit he got rid of the
> drift. It might be that for highly charged systems the discontinuity in
> direct sum energy/forces at the cutoff [i.e. erfc(beta*r_cutoff)/r_cutoff
> > order = 4, dsum_tol = 0.00001, eedmeth = 1,

As the eminent Dr. Darden mentioned, if you just decrease dsum_tol by 1
order of magnitude, dsum_tol = 0.000001 you should avoid this drift.
Although Professor Duan points out that errors are expected, we can run
under appropriate conditions true NVE for long simulations such as these
without the drift you are seeing... However, I do not think this drift
will significantly impact your structural/dynamic results; if you re-ran
this simulation with better energy conservation, what you will interpret
out of your DNA simulation will very likely not change (or at least not
change due to energy conservation; the changes you will see will be due to
sampling a different ensemble and as you may find out, rather long
simulations are necessary to fully "equilibrate" DNA properties).

Good luck,

\ Thomas E. Cheatham, III (Assistant Professor) College of Pharmacy
| Departments of Medicinal Chemistry and of University of Utah
| Pharmaceutics and Pharmaceutical Chemistry 30 South 2000 East, Room 201
| & Center for High Performance Computing Salt Lake City, Utah 84112
|
| e-mail: tec3_at_utah.edu phone: (801) 587-9652 FAX: (801) 585-9119
\ http://www.chpc.utah.edu/~cheatham Offices: BPRP295A / INSCC 418