[gmx-users] again, rlist vs rvdw vs rcoulomb and something about energy groups

[m b]

dear all,

I do MD simulations using a cutoff for
both VdW and Coulomb interactions.
I am rather confused about the
parameters used in the input file.
(rlist, rvdw, rcoulomb)
I searched this mailing list
for clarification, but I found two
statements that appear to contradict
each other (just as the documentation
does by the way). The two mails are: 
http://www.gromacs.org/pipermail/gmx-users/2002-June/028443.html
http://www.gromacs.org/pipermail/gmx-users/2003-January/030507.html

Normally I would assume that if one uses
a neighbour list there are (at least) two cutoffs:
One for a sphere around each atom with ALL
(and only) the atoms in this sphere considered when
calculating the energy and the forces on
the primary atom at EACH time step (rvdw,
rcoulomb ?).
A second (larger) cutoff that is used when
calculating the neighbour-list at intervals
of each 10 (or so) time steps (rlist ?)
were of course rlist > rvdw and rcoulomb.
(it is also possible to have the program
calculate this parameter nstlist automatically
thereby optimizing the performance, this is
not implemented in Gromacs, is it ?)
If one uses a switched potential there
is a third cutoff (the onset of the switching
function) that would be shorter than
rvdw and rcoulomb.

this implies: rlist > rvdw,rcoulomb > r*_switch
... in contrast to (part of) the docu, and the
errormsg given by the code, and to one of the
messages cited above.(the message of Anton
Feenstra reminds me rather of another concept,
namely the multiple timestep algoritm which, if 
I am not mistaken, is not implemented
in Gromacs)

In my calculations I now let
rlist < rvdw, rcoulomb, to evade
the error message that Gromacs gives me
otherwise, but if what I wrote above is true
then the physical implications would
be a bit worrying ...
and if I have rlist = rvdw, rcoulomb
would that not mean that the neighbour list
has to be re-calculated at each time-step
to get consistant results ?

So the question is: is what I wrote above
correct or not ? and if not, what is the
precise meaning of these parameters (rlist,
rvdw, etc)

and a related question: If one uses cutoffs
for both coulomb and vdw, are  the two 
contributions calculated in the same inner
loop or not ? If not it might increase the speed
setting rvdw smaller than rcoulomb, otherwise
there would be no point. 

and a third question: I would prefer to use
PME rather than cutoffs for my calcs but 
I am interested in calculating interaction
energies (here between a molecule and a surface,
i.e., energies of adsorption) and hence
I must use energy groups to get the seperate
energy contributions. The reciprocal space
part of Ewald or PME is apparently not considered
by Gromacs when calculating these contributions.
In another mail in this group somebody suggested
to calculate these contributions a posteriori
by making two reruns of the trajectories with the
charges of one of the groups "switched off"
in each run. (did I understand this correctly?)
I assume this would work, but then it would
be quite cumbersome and the statistics would
be much worse.
I wonder if it is really impossible to explicitely
calculate the reciprocal space part of the direct
interaction energy between two groups of atoms
directly within the Ewald/PME double loops ??
This seems to be a long-standing problem
which might be of some interest to many people
(with a large part of all molecular simulation
nowadays being used to calculate protein-ligand
interaction energies) 
However, neither in the literature
nor in the source or docu of any other program
I ever found this problem addressed explicitely.
any ideas anybody ?

I just realise this mail became a bit long-ish ...
sorry for taking your time ... and thanks in
advance for any answers!

cheers,
Michael


__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software
http://sitebuilder.yahoo.com