On Wed, Jun 15, 2005, Eric J. Sorin wrote:
>
> >
> > OK ... we'll will wait to hear more about this before moving on. In
all
> > honesty, I'm a bit confused by this. Both of us working on this
porting
> > project were under the impression that a perfect match in the torsion
> > definitions always overrode a wildcard match. Can you please explain
the
> > circumstances (or point me to a written source to save yourself the
typing
> > time) in which both should be used? I don't recall ever seeing such a
case
> > before.
Hi Eric:
I need to update you on some new information about parm99. It turns out
that
I was more confused than you were. Junmei Wang (the author of parm99) has
confirmed that his intent was to do what you said: have the specific
torsion
override the wild-card torsion, even if the wild-card version has a
different
periodicity than the specific one.
>
> Here is the wildcard in parm99:
>
> X -CT-CT-X 9 1.40 0.0 3.
JCC,7,(1986),230
>
> Here is one of the specific ones:
>
> H1-CT-CT-OS 1 0.25 0.0 1. Junmei et al,
1999
So, the "intended" interpretation should be that H1-CT-CT-OS should have
*just*
the 1-fold torsion, and not the 3-fold torsion. I had thought (and told
you)
that both would be used.
The "problem" is the LEaP actually acts the way I originally thought,
(although in certain pretty unusual circumstances, such as after a "copy"
command, it might not). So, if we "fixed" things to what Junmei intended,
that would change the de facto way Amber has worked for many years, and
would
cause lots of confusion. [The "fix" would add a specific H1-CT-CT-OS
3-fold
term with a zero force constant, to ensure that the X-CT-CT-X torsion was
really squashed.]
Fortunately, this problem does not come up in proteins, nor does it come
up
with parm94 or parm98 force fields for nucleic acids (which are the ones
that
have been most commonly used.
I'm cc-ing this to the Amber developers' list, because some decisions have
to
be made. My vote is to accept the de facto way LEaP has always worked,
(just
fixing the copy problem, which has only recently been discovered). If
anyone
wants to to look at nucleic acids with Junmei's original intended
parameters,
we can make a "parm99b" or something like that. That way, any existing
parm99 simulations will continue to be valid. And this should mean that
the
GROMACS libraries would not need to be changed.
Junmei, Tom: what do you think of this proposal? Anyone else on this list
is welcome to weigh in as well.
[Thanks to Eric Sorin for raising some of these points, which were
coincidentally also raised by Angela Liu's bug report on June 28, 2005 --
see the mail archives. I'm hoping to get things fully clarified(!) before
we make any public explanations for bug-fixes. But I hope we can get this
all cleared up soon.]
...thanks!....dave case
Received on Wed Apr 05 2006 - 23:49:54 PDT
