[gmx-users] Re: rvdw parameters (and other mdp stuff)

[David van der Spoel]

On Thu, 2002-08-15 at 00:13, Graham Smith wrote:
> 
> > Especially the short-range cutoff (0.8) *SHOULD* *NEVER* *BE*
> > *CHANGED*.  For the long range it is not so critical, but you would
> > rather use 0.8/PME in stead of increasing the 1.4 long range
> > cut-off. 
> 
> Just to make sure I've got this clear, 
> with ffG43*, 
> 
> rlist                    = 0.8
> coulombtype              = PME
> rcoulomb-switch          = 0
> rcoulomb                 = 0.8 
> 
> vdwtype                  = Cut-off
> rvdw-switch              = 0
> rvdw                     = 1.4
> 
> (...and not rlist=rcoulomb=0.9, rvdw=1.4, which I started using a
> couple of months ago when I switched to PME. I got this by
> misunderstanding the manual (section 4.6.2) and mdp_opt.html; perhaps
> a note could be added to say that it applies only to ffgmx? Or is
> increasing rlist not as serious as decreasing it?) 
See Berk's notes. increasing rlist does *not* affect Coulomb when using
PME, but it does make the Van der Waals slightly more accurate. Note
that you will not get the exact same energies as Van Gunsteren (which
you won't anyway when using PME).


> What parameters should I use with the new opls ff? Kaminski et al, J
> Phys Chem B 201 105 6474, state what cutoffs they've used only for
> their simulations of methanethiol and ethanethiol; 
> unless I'm missing it, they don't mention the conditions for the 
> simulations of peptides. They say 
> 
> intermolecular interactions were truncated at 11A, with the standard
> correction for the interactions beyond that radius [ref to 1984 JACS
> paper]. The ES interactions were quadratically feathered to zero over
> the last 0.5A before the cutoff. 
You can use shift function for Coulomb iso PME, but using PME in the
scheme you have above is probably not much slower than the rlist=1.1
below.

> So if they're the same that would mean for ffopls protein simulations 
> 
> rlist                    = 1.1
> coulombtype              = PME
> rcoulomb-switch          = 1.05
> rcoulomb                 = 1.1
> 
> vdwtype                  = Cut-off
> rvdw-switch              = 0
> rvdw                     = 1.1
> 
> ? 
> 
> Do I need to switch on DispCorr? 
For OPLS you want dispcorr=ener (from that particular JACS paper). They
do not have a pressure correction, and hence have the pressure off by
rougly 200 bar.

> While we're at it, I have always used a different temperature coupling
> group for each type of [ molecule ] in the .top, so I have e.g.
> 
> tcoupl                   = nose-hoover
> tc-grps                  = Protein  SOL NA+ CL-
> tau-t                    = 5 5 5 5
> ref-t                    = 300 300 300 300
> 
> This has propagated from an mdp I was given years ago (even though
> I've switched tc algorithm). But I've never really seen why it
> isn't always best just to tcoupl the whole system. I would think
> averages should be the same, but T-fluctuations will be reduced by
> somthing like sqrt(N_grp/N_system) with the groups coupled
> separately. Any comments? 
Yes, if you couple the whole system as one thing you will see that the
equipartition theorem is violated! Water will heat up and Vanderwaals
molecules (lipids, proteins) will cool down. This is mainly a cut-off
effect and hence not as bad when using PME. It has also to do with shake
(I suspect), i.e. poor coupling between internal and external degrees of
freedom. You could combine the Na+ and Cl- in one group.
It doesn't help to use Nose-Hoover iso Berendsen either, it gets  worse
even (my unpublished results...)

-- 
Groeten, David.
________________________________________________________________________
Dr. David van der Spoel, 	Biomedical center, Dept. of Biochemistry
Husargatan 3, Box 576,  	75123 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://zorn.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++