Re: [AMBER-Developers] sqm nondeterminism

From: Scott Brozell <sbrozell.rci.rutgers.edu>
Date: Mon, 15 May 2017 14:06:51 -0400

Hi,

On Mon, Apr 24, 2017 at 04:41:09PM -0400, Hai Nguyen wrote:
> On Mon, Apr 24, 2017 at 4:37 PM, Scott Brozell <sbrozell.rci.rutgers.edu> wrote:
> >
> > Some tests using sqm can fail sporadically. This is due to the
> > auto diagonalization routine selection feature.
> >
> > Here is a typical culprit, AmberTools/test/antechamber/tp:
> > ===
> > commit a9f7ccb1d8cef0eca60e34ec4a019bda8cb0e6f9
> > Author: Scott Brozell
> > Date: Tue Jan 31 00:57:55 2017 -0500
> >
> > Crank up the ndiff tolerance: sometimes sqm picks a different diagonalization
> > routine which does not change the energies significantly but does change
> > the coordinates by more than 1.e-3:
> > < | QMMM: Using dsyevr routine (diag_routine=7).
> > ---
> >> | QMMM: Using internal diagonalization routine (diag_routine=1).
> > 93c93
> > < Heat of formation = 25.64120694 kcal/mol ( 1.11188617 eV)
> > ---
> >> Heat of formation = 25.64120695 kcal/mol ( 1.11188617 eV)
> > 123,135c123,135
> > ...
> > < QMMM: 13 13 H -2.2901 -0.9503 0.7145
> > ---
> > ...
> >> QMMM: 13 13 H -2.2734 -0.9589 0.7288
> > ===
> >
> > Apparently, my quick fix for tp was not sufficient since Hai just
> > reported to me a similar tp failure.
>
> > Probably the best practice would be to turn off the auto selection
> > feature for tests that are not testing that feature.
>
> I agree. Getting email reporting AT test failed every night is kind of
> irritating.
> https://travis-ci.org/Amber-MD/ambertools-ci/builds/225061021

On Tue, Apr 25, 2017 at 07:46:28AM -0400, David Case wrote:
> On Mon, Apr 24, 2017, Scott Brozell wrote:
>
> Can you go ahead and do that in master? Then we can consider making a bugfix.

I have updated the sqm manual to clarify that auto selection
(diag_routine = 0)
is the default and have added a comment on these potential effects on
geometries. Has anyone studied diag_routine effects on sander QMMM ?

I have clarified antechamber's docs to note that the -ek option overwrites
previous keywords.

My anecdotal evidence is that sqm diag_routine 3 and 7 are
most competitive with 1.

For reference, here is the syntax for antechamber:

antechamber -i tp.pdb -fi pdb -o tp.mol2 -fo mol2 -c bcc \
   -ek "qm_theory='AM1', grms_tol=0.0005, ndiis_attempts=700, scfconv=1.d-10, diag_routine=3"

scott


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon May 15 2017 - 11:30:02 PDT
Custom Search