I tried with Gaussian and had convergence problems too. I got error messages like 'Rare condition: small coef for last iteration:'. A Google search suggested that this can happen with semi-empirical methods. Checking for the usual suggestions to deal with SCF convergence issues I came across vshift which also happens to be implemented in sqm.
So, the following name list does converge the structure.
&qmmm
qm_theory='AM1',grms_tol=0.0002,tight_p_conv=0,
scfconv=1.0d-9,itrmax=500,pseudo_diag=1,vshift=0.1,
ndiis_attempts=200,ndiis_matrices=6,maxcyc=2000,
qmcharge=-1,verbosity=1
/
Except that the structure gets "folded" (see attachment) with the carboxy groups contacting the amino groups. Not too surprising in vacuum and, I would guess, shows the limits of charge derivation from vacuum structures.
Many thanks again,
Hannes.
________________________________________
From: Jason Swails [jason.swails.gmail.com]
Sent: 08 May 2014 14:25
To: amber.ambermd.org
Subject: Re: [AMBER] sqm convergence failure
On Thu, 2014-05-08 at 13:49 +0100, Hannes Loeffler wrote:
> On Thu, 8 May 2014 08:22:41 -0400
> Jason Swails <jason.swails.gmail.com> wrote:
>
> > On Thu, 2014-05-08 at 11:16 +0100, Hannes Loeffler wrote:
> > > Hi,
> > >
> > > I can't get the attached structure (sqm.in) converged. What
> > > happens is that the wave function won't converge somewhere along
> > > geometry optimisation, at some arbitrary step (not the first step)
> > > depending on name list parameters.
> > >
> > > I have no particular interest in this one. The problem just came up
> > > in a larger test set. The structure is ZINC03814828. Same result
> > > with ZINC03814832 which is almost the same.
> > >
> > > The current name list is an outcome of playing around with various
> > > parameters but without luck. What I wonder is if this could still
> > > be made to converge or if there is a set of known structures which
> > > just cannot be converged? If I replace the carboxy-ethyl group
> > > with propyl the structure converges at scfconv=1.0D-9.
> >
> > I think I may be missing something obvious or perhaps I postprocessed
> > your sqm input file wrong. If I delete the sqm input variables and
> > put '55' at the top of the file (to try and make the file look like
> > an XYZ), I get an image which appears to have 3 hydrogen atoms
> > floating around somewhere in space (image attached).
> >
> > If I didn't make a mistake in my hacky sqm-to-xyz file conversion (it
> > is _very_ possible that I did), this could certainly explain why the
> > structure fails to converge (most of the structure looks fine, but not
> > the 3 labeled H atoms), and more reflects a problem with the starting
> > structure than with sqm.
>
> Many thanks Jason but I believe the structure is sane. Not sure how
> you got this as looks good in my vmd. The C-C distance is 1.54 A and
> the C-H distances are all 1.09.
>
> The structure has been generated as part of a convergence protocol
> which includes MM pre-minimisation steps with the MMFF94 force field.
> The original SDF file contains connectivities and bond order
> information so that should be fine, see attached mol2 file (ignore
> Openbabel's feeble attempt at recognizing residues :-).
Can you try minimizing with mopac or Gaussian (or some other program
that implements the same semiempirical model)? That might help
determine whether the semiempirical model or sqm is to blame...
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
--
Scanned by iCritical.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu May 08 2014 - 11:00:07 PDT
