On Fri, Mar 25, 2016 at 11:07 AM, David Cerutti <dscerutti.gmail.com> wrote:
> Yes, I agree--in the mdgx topology data structures, for executing the
> energy function, the propers and impropers are all in a list, but each has
> a flag. For parameter development, there is again one long list of
> "dihedral" data structures read from the parm.dat files, and again each o
> them has a flag 0 or 1 to indicate proper or improper.
This is important for preventing an improper type being assigned to a
proper type and vice-versa, but unless I'm misunderstanding you it's not
enough to ensure that you don't "lose" a parameter if proper and improper
torsions share the same type sequence.
The moral here is to use separate lists to store improper and proper
torsion parameters with separate sets of 4-type keys identifying the atom
types involved in each parameter. That they appear in the same list in the
prmtop file is an artifact of convenience. Remember that parameter
assignment was already done inside tleap from two separate lists (one for
propers and one for impropers) *before* they were combined into the same
list in the prmtop. So if a system has both an improper and a proper
torsion containing the same atom type sequence, it wouldn't matter.
Simply put, if mdgx can't store both
IMPROPER ca-nh-ca-c
PROPER ca-nh-ca-c
then you are risking not being able to faithfully store and/or represent
the Amber force field in the general sense.
It may never affect anything in production or it might. It may be a
simple error that causes a crash (due to an unidentified parameter), or it
might not. It's impossible to tell. But it opens up the possibility for a
subtle issue that can have profound importance, but is incredibly difficult
to even detect. I simply don't know...
As a reference, have a look at how ParmEd handles this:
https://github.com/ParmEd/ParmEd/blob/master/parmed/parameters.py#L87-L89
Separate sets of keys for proper, improper (and periodic improper, which
is the functional type Amber uses).
All the best,
Jason
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Mar 25 2016 - 09:30:03 PDT
