Hi Jason,
Are you sure this list comes from the prmtop?
For GB simulations I think it does indeed come from the prmtop but sander and pmemd to the best of my knowledge for periodic simulations throws out whatever is in the prmtop file and instead recomputes the 1-4 list manually in extra points. I might be confusing things here with the exclusion list but you certainly check. I know when I was doing the QMMM work a decade ago I ran into issues with trying to modify things in the prmtop for testing and wondering why it was blitherly ignored. But my memory of specifically what it was is hazy so see if you can trace through where those arrays are actually filled and used (and if it might be just for GB simulations).
All the best
Ross
> On May 18, 2015, at 6:53 AM, Jason Swails <jason.swails.gmail.com> wrote:
>
> Hi everyone,
>
> This is a rather technical discussion regarding the 1-4 nonbonded
> interactions and how they are computed differently for sander and pmemd.
> In particular, sander uses the following logic to identify 1-4
> interactions for the L-J and electrostatic loops:
>
> if ( (k >= 0) .and. &
> (l >= 0) .and. &
> (.not. qmmm_skip) &
> )then
>
> l<0 indicates an improper torsion and k<0 indicates end-groups are
> ignored (e.g., for multi-term dihedrals and ring systems). qmmm_skip
> indicates this pair is in the QM region, which is all well and good. By
> contrast, pmemd has:
>
> if (gbl_dihed(dihed_idx)%atm_k .gt. 0 .and. &
> gbl_dihed(dihed_idx)%atm_l .gt. 0 .and. &
> gbl_fmn(gbl_dihed(dihed_idx)%parm_idx) .ne. 0.d0) then
>
> where gbl_fmn is basically a boolean mask array that is set to 1 when
> the dihedral periodicity is nonzero and 0 otherwise. In practice, there
> is no harm in ignoring dihedrals whose periodicities are 0, since they
> add a constant shift to the energy (and by extension no contribution to
> the force). But the dihedral list in Amber serves a dual purpose -- it
> is both the list of torsion parameters in the system *and* the source of
> the 1-4 nonbonded pairlist. When porting other force fields to work
> with Amber, some programs make use of 0-periodicity torsions
> specifically to tag nonbonded pairs -- a scenario that sander gets
> correct and pmemd currently gets wrong.
>
> My inclination is to simply get rid of gbl_fmn altogether (since it
> *only* serves to ignore 1-4 nonbonded pairs that should, it seems, never
> be ignored). Since this seemed deliberate, though, I wanted to know if
> anybody had insight into why pmemd behaves the way it does, and if there
> is anything I'm missing in my analysis that makes this proposal
> dangerous. But at the very least, we should make sure sander and pmemd
> do the same thing.
>
> Thanks,
> Jason
>
> --
> Jason M. Swails
> BioMaPS,
> Rutgers University
> Postdoctoral Researcher
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon May 18 2015 - 09:00:05 PDT
