On Fri, 2014-09-05 at 10:42 +0100, Hannes Loeffler wrote:
> On Thu, 4 Sep 2014 07:11:01 -0400
> Jason Swails <jason.swails.gmail.com> wrote:
>
> > On Thu, Sep 4, 2014 at 4:57 AM, Hannes Loeffler
> > <Hannes.Loeffler.stfc.ac.uk> wrote:
> >
> > > Hi,
> > >
> > > I have the following entry in a parmtop file of mine
> > >
> > > %FLAG DIHEDRALS_WITHOUT_HYDROGEN
> > > %FORMAT(10I8)
> > > 3 6 -9 12 1
> > >
> > > the third index indicates that the non-bonded interactions between
> > > the 1-4 atoms should not be computed (this particular torsion
> > > appears only once in the file = single term dihedral). The
> > > dihedral is from four C atoms in the phenyl ring of toluene
> > > adjacent to the methyl side group.
> > >
> > > Now I understand that to allow mixed scaling the new flags %FLAG
> > > SCEE_SCALE_FACTOR and %FLAG SCNB_SCALE_FACTOR have been introduced
> > > some time in the past to allow explict recording of scaling factors.
> > > This means, I think, that the fifth number above (ICP) points into
> > > the first entry of these flags (the web documentation of the parmtop
> > > format mentions arrays "ONE_SCEE" and "ONE_SCNB"). The values there
> > > are not zero possibly implying that 1-4 scaling should be carried
> > > out contrary to the negative third index.
> > >
> > > So, obvious question: what is really going to happen? My first
> > > guess would have been that ICP(H) overwrites the negative index and
> > > scaling is indeed carried out in sander/pmemd.
> > >
> >
> > No, scaling is not carried out -- those 1-4 interactions are excluded
> > entirely. The negative index determines whether or not to carry out
> > the 1-4 electrostatic energy (of course you can change that negative
> > to a positive and check that the energy does, indeed, change by the
> > exact amount of the scaled 1-4 electrostatic and van der Waals
> > interactions for that 1 pair).
>
> So the simple rule is: negative third index takes precedence over
> ICP/ICPH.
I'm not sure what you mean by this. ICP/ICPH are simply pointers into
the various dihedral parameter arrays. The SCEE and SCNB values indexed
by ICP/ICPH are ignored if the third index is negative (it's always been
that way). The values of 0 for the torsions were added as an error
check -- if implementations of the force field incorrectly compute those
interactions, the energies and forces will become infinite and the
system will blow up. The SCXX_SCALE_FACTOR sections were only added in
Amber11. See the relevant thread on the developer list here:
http://dev-archive.ambermd.org/all/4621.html and the response here:
http://dev-archive.ambermd.org/all/4624.html.
> > Rings are, IMO, the trickiest part of a topology to handle
> > "correctly" in the force field. For all multi-term dihedrals, there
> > is _never_ a case where the first N-1 terms of an N-term dihedral
> > should trigger computing the 1-4 interactions, so those third indexes
> > are always negative and a SCEE/SCNB value of 0 can be safely applied
> > to those terms. In a ring, as you pointed out, it is possible (and
> > quite common) to have torsions specified by a single term, but whose
> > interactions are not computed either because they are included in a
> > different 1-4 interaction (in the case of 6-member rings) or they are
> > excluded because they are 1-3 partners (in the case of 5-member
> > rings). However, those atom types are not guaranteed to exist in
> > just 6-member or 5-member rings, so you can't guarantee that their
> > interactions will never need to be computed. As a result, you can't
> > broadly assume their SCEE/SCNB values to be 0.
>
> The only complication then seems to me to be the 5-rings because the
> decision has been made to exclude possible 1-4 pairs as they are also
> in 1-3 relationship (from a recent conversation with a colleague I had
> the impression that CHARMM does not do that). Otherwise, the simple
> procedure would be to enumerate all 1-4 pairs and eliminate all
> duplicates afterwards.
Actually, 6-member rings are affected as well (all ring sizes from 4 to
6 atoms, really). Consider the following 6-member ring:
C2--C3
/ \
C1 C4
\ /
C6--C5
Atoms C1 and C4 are 1-4 partners. But they are 1-4 partners through
C1-C2-C3-C4 and C1-C6-C5-C4. You only want to compute that interaction
once, so one of those two needs to have a negative third index. If each
of those torsions are specified with only 1 term, then it will appear as
though a unique torsion will have a negative index (and there are 3 such
sets of duplicate torsions in a 6-member ring). And assume this is
benzene. In that case, all atom types are identical and there are only
3 different types of torsions (H-C-C-C, H-C-C-H, and C-C-C-C). The SCEE
and SCNB scale factors obviously cannot be 0 for the C-C-C-C since half
of them are still computed. An alternative would be to define 4 torsion
types, but I don't think that is the approach used by Amber/tleap.
I am almost certain that CHARMM behaves exactly the same way.
Hope this helps,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 05 2014 - 06:00:05 PDT
