Re: [AMBER-Developers] Dihedrals

From: case <>
Date: Wed, 15 Sep 2010 08:12:42 -0400

On Tue, Sep 14, 2010, Jason Swails wrote:
> I have a question about the dihedrals and how they're computed in sander and
> NAB. I looked through ene.f (specifically subroutine ephi) and traced some
> variables, and I found what appeared to be the where the dihedral energy
> term is computed (excluding all 1-4 nonbonds here). What I found is
> epw(jn) = (pk(ic)+cosnp*gamc(ic)+sinnp*gams(ic))*fzi(jn)
> which appears right at first glance. However, when I move over to NAB,
> there's a switch block that requires the periodicity to be 1, 2, 3, 4, or 6
> or it quits with a "bad periodicity" error. I don't see the equivalent of
> that anywhere in sander. Am I missing something? Why the difference.

The NAB code treats the dihedral angle as a function of cosphi, not of phi.
This is more stable, and never involves computing or dividing by sinphi.

> Taking this 1 step further, this limitation in NAB makes it unable to
> correctly simulate some molecules built with ff99SBildn, since that force
> field appears to have periodicities of 5 in 3 dihedrals:
> N -CT-CT-C6 1 0.2317 0.0 -5.
> C -CT-CT-C5 1 0.1042 0.0 -5.
> N -C5-CT-CT 1 0.1298 0.0 -5.
> I believe these will cause grief in NAB, but will they work in sander? I'm
> also assuming sander and pmemd treat them identically? Is it worth
> extending NAB to handle PN=5?

Sounds like we will need to do this. Time for someone to brush up on
multiple-angle formulas in trig, or to fire up Mathematica.


AMBER-Developers mailing list
Received on Wed Sep 15 2010 - 05:30:06 PDT
Custom Search