[gmx-users] VDW Cut-off vs Switch with CHARMM force field
jalemkul at vt.edu
Sun Aug 25 02:28:16 CEST 2013
On 8/24/13 8:19 PM, Gianluca Interlandi wrote:
> Thanks James!
> I was also wondering whether there is a significant penalty in performance if I
> use rlist=1.2 instead of rlist=1.0
I doubt one would consider it "significant," but of course any time you decrease
rlist, you potentially speed up the simulation a bit. Whether or not doing so
is appropriate is an entirely separate issue.
> There has already been a discussion in the NAMD mailing list about the VDW
> cutoff issue:
> It is apparently common among NAMD users to use switch for VDW between 0.8 and
> 1.0 nm. I have often received negative criticism from supercomputer grant
> reviewers whenever I suggested using a 1.0 - 1.2 switching distance for VDW and
> 1.2 nm for Coulomb as in their eyes a 1.0 cutoff distance is enough (when PME is
> in use).
It is important to remember that all of these settings are interconnected. For
PME, rlist must equal rcoulomb. It is true that the value of rcoulomb is
somewhat flexible when using PME, but you must be sure that the changes you are
making to other settings are not detrimental. If you decrease rcoulomb, then
you must also decrease rlist, which then also has implications for van der Waals
interactions, even if the change was originally only motivated by using PME. To
reproduce the typical CHARMM usage, one would set:
vdwtype = switch
rlist = 1.2
rlistlong = 1.4
rvdw = 1.2
rvdw-switch = 1.0
rcoulomb = 1.2
Unless you can demonstrate that your changes in cutoffs do not produce artifacts
or lead to other inaccuracies using robust test systems, I wouldn't recommend
any deviations. Note that the response you received on the NAMD list also
suggested this to be true.
Justin A. Lemkul, Ph.D.
Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 601
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201
jalemkul at outerbanks.umaryland.edu | (410) 706-7441
More information about the gromacs.org_gmx-users