On Tue, Sep 13, 2016 at 12:48 PM, Niel Henriksen <shireham.gmail.com> wrote:
> Thanks Hai,
>
> Earlier, I tested this by changing ntc/ntf to 2 in pysander and didn't see
> any change, so I assumed it was not the culprit. I guess I should have
> also tried it with sander.
>
> But it raises another issue: why don't the pysander results change when I
> set it's ntc/ntf to 2?
>
​Setting ntf=2 may change the bond energy (since it omits calculating some
of the bond energy terms if that bond includes a hydrogen atom). But it
will have no effect on nonbonded interactions.
The only other thing I can possibly think of is that pysander uses ParmEd
behind-the-scenes, and ParmEd scales charges down by 18.2223 when reading
in a prmtop and back up by 18.2223 when writing. It's possible that this
results in some small fluctuations of the prmtop charges that get processed
by pysander due to mild round-off. If this happens to just one atom in the
last decimal place, the atoms are repeated enough that it may add up to be
a ~0.0001 deviation :). That's my best guess as to why this is happening,
but would require more time to confirm than I can afford to spend.
There's also a cut-off point where if the charges are close enough to 0,
sander will set the charge exactly to 0 by smearing the difference over all
of the atoms. This can have some surprising results and was the source of
a non-bug that was posted to the developer list a few years ago (can't find
the thread now, but it's on Bugzilla).
In general, electrostatic reproducibility is *hard*, due to the huge number
of parameters that each program sets differently and long-range nature of
the interaction that can cause fairly significant fluctuations.
All the best,
Jason
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 13 2016 - 11:00:02 PDT
