On Tue, May 21, 2013 at 11:39 AM, Chris Whittleston <csw34.cam.ac.uk> wrote:
> Dear AMBER users and devs,
>
> I've come across something which seems slightly troubling. We interface
> with both AMBER and NAB to obtain the energy, gradient and in the case of
> NAB, the analytic second derivatives of the potential for the purposes of
> locating transition states. As part of this process, we sometimes call NAB
> followed by AMBER for the same coordinates.
>
> In vacuum (igb=0) we get consistent energy and gradient from AMBER and NAB,
> but when using igb=5, we get a completely different energy and gradient for
> identical coordinates. Even more confusingly, although they are consistent
> for igb=2 with the default external dielectric constant, if we set
> extdiel=1, igb=2 also gives inconsistent energy and gradient. This is
> summarised in the attached pdf.
> I was wondering if one of the devs could comment on this. It appears that
> igb=5 must be somehow implemented differently in AMBER and NAB, and
> separately that the effect of extdiel is also different between AMBER and
> NAB.
>
I have 2 comments. Regarding the igb=5 disagreements: I noticed this exact
same thing several months ago and tracked down its origin (the gamma
parameter was set to 4.85 in NAB and 4.851 in sander and pmemd). The 9th
bugfix of Amber 12 adjusts the sander and pmemd value to 4.85 (to be
consistent with NAB and with the original paper). If you make sure you
have an up-to-date sander12 (or make that change by hand where necessary in
the sander source code), then you should get answers that agree exactly---I
have checked this. My version of Amber12 (which does not have your
reported disagreement): with AmberTools 13, this is the output I get from
running the command $AMBERHOME/update_amber --version
$ ./update_amber --version
Version is reported as <version>.<patches applied>
AmberTools version 13.04
Amber version 12.16
Regarding the extdiel factor. Why would you expect setting extdiel=1 in
sander to reproduce the NAB energies? The external dielectric is set to
78.5 by default, and there is no way to change this in NAB short of
changing the epsext variable inside sff.c and recompiling. The internal
and external dielectrics used in the GB equation are fixed in NAB (as they
really should be in sander -- I would be skeptical of other values in
general). If you want to check their consistency, you will need to make
that change and recompile.
HTH,
Jason
P.S. One final comment -- try using 'igb=6' to get a vacuum simulation in
sander. Setting igb=0 gets vacuum energies through the Ewald electrostatic
code, whereas using 'igb=6' gets vacuum energies through the GB code.
While they are equivalent in general, the GB code respects the prmtop
exclusion list and may be more efficient for a brute-force loop-over-pairs.
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue May 21 2013 - 10:30:02 PDT
