Re: [AMBER] Dihedral/improper coefficient in AMBER + rmin vs sigma for lennard jones

From: Jason Swails <jason.swails.gmail.com>
Date: Thu, 8 Oct 2015 08:14:15 -0400

On Thu, Oct 8, 2015 at 12:33 AM, conor parks <coparks2012.gmail.com> wrote:

> Hello,
>
> I have two separate questions. If I should submit two seperate requests in
> the future, please let me know. I am just trying to not spam peoples email
> boxes!
>
> 1. I am implementing the amber force field (want to use GAFF parameters) in
> a MD code my research group is developing. Its my understanding that the
> bond and angle spring constants are not multiplied by 0.5. However, its my
> understanding that the functional form of the dihedral/improper energy is
>
> E = 0.5*k*(1+cos(n*phi - d) )
>
> I have a prmtop file that looks like
>
> %FLAG DIHEDRAL_FORCE_CONSTANT
>
> %FORMAT(5E16.8)
>
> 1.55555556E-01 0.00000000E+00 1.10000000E+00
>
> Could someone tell me if these spring constants have already been
> multiplied by 0.5 or not?
>

Yes, this value includes the 0.5 factor. I was pretty sure it did, but
whenever I'm not completely sure about something, I check the code (the
code never lies, after all). The energy expression for dihedrals looks
like this:

epw(jn) = (pk(ic)+cosnp*gamc(ic)+sinnp*gams(ic))*fzi(jn)

Where gamc and gams are cos(phase)*pk and sin(phase)*pk, respectively (fzi
is either 1 or 0). pk is the value read from the prmtop.

2. I am wondering about whether there are issues I should be aware of
> associated with truncation error when converting from the epsilon sigma
> representation of the lennard jones interaction to the amber epsilon rmin
> representation. For instance, in the case of the SPC/E water model, I
> believe the sigma value is given as 3.166. Converting that to Rmin for
> AMBER
>
> rmin = 1.77685742 --> 1.777 if we carry the same precision through
>
> Is it correct to make this kind of truncation if I am going to hand type
> the parameters prior to a simulation, as I need to do in my groups code or
> in lammps since i don't have access to moltemplate, or could even that
> subtle of a change in the parameters lead to disastrous consequences?
>

​Yes and no, depending on your definitions of "correct" and "disastrous".
If part of your validation is going to compare against the answers that
Amber gives (which I would highly encourage), then such truncation will
make that much harder. Your results will undoubtedly differ from the Amber
results due to this truncation, and this will also mask any *other* source
of error that is around the same magnitude if it exists.

As for results, it is unlikely to make a noticeable difference (after all,
variations in the charge on the order of 0.0001 to 0.001 electrons is not
noticeable, and that's a much stronger interaction than L-J).

In my experience, you should always strive to completely eliminate any
precision-related differences in your *input* when doing validation between
implementations. Not doing so introduces an (unnecessary) uncertainty into
your validation. Once validated, such adherence to ultra-high-precision
input can be relaxed.

HTH,
Jason

-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Oct 08 2015 - 05:30:04 PDT
Custom Search