SUMMARY: G98: opt=stable and freq

From: DE BRUIN theodorus <theodorus.DE-BRUIN #*at*# ifp.fr>
Subject: SUMMARY: G98: opt=stable and freq
Date: Tue, 27 Aug 2002 14:00:01 +0200
Dear All,
 I would like to thank everybody who replied to my question concerning the
 instability of the wavefunction and the frequency analysis that I wanted to
 perform subsequently.
 I wish to point out that the instability did not have a RHF/UHF origin,
 since I used (of course) uB3LYP.
 Initially I used a --link1-- job with guess=read geom=check freq.  However
 in that case I fell back on the old(=instable wavefunction). In my case the
 combination: guess=(read,mix) geom=check freq
 resulted in a freq analysis on the stable wavefunction.
 Thanks again.
 Theo
 REPLIES:
 Frank Jensen [frj #*at*# dou.dk]:
 Try generating the stable wave function (= the one from
 stable=opt) by e.g. guess=alter or guess=mix. This works if the
 wave function is RHF --> UHF unstable. If the instability is
 a symmetry breaking you may need to create your own route card
 where the wave function is first done with stable=opt, and then
 immediately continuing onto the frequency. If that is necessary,
 however, it strongly suggests some (other) problem with your
 wave function, perhaps it is strongly MCSCF, or the DFT grid is
 insufficient.
 Monica Kosa [chmonica #*at*# techunix.technion.ac.il]:
 You can try read optimized wavefunction from the checkpoint file.
 First you run a job with stable=opt keyword. Than you include the keyword
 freq (and delete the stable=opt) and guess=read (in the same input file you
 ran your stable =opt job).
 Reinaldo Pis Diez [pis_diez #*at*# quimica.unlp.edu.ar]:
 I posted a very similar question even before, in 1999!!
 Below I send you the "official" answer, that is, that of Gaussian (my
 original mail is included, too). Note that the suggested recipe is the same
 you tried!!
 Note also that an alternative three-step job is proposed: use first
 stable(opt); then use guess(read) opt to get the true minimum; and finally
 do the freq calculation.
 Dr. Diez,
 There are a couple of points. First, you want to use GUESS=READ and the
 checkpoint file from the STABLE=OPT calculation to have a chance of
 recovering
 the final stable solution. In some nasty cases you may need to use SCF=QC as
 well. Second, note that if the SCF was unstable for the optimization then
 the
 structure no longer corresponds to a minimum and you need to re-optimize to
 get
 the true minimum energy structure. You might want to consider changing your
 protocol and do a STABLE=OPT at the first point, then use GUESS=READ OPT
 with
 SCF=QC if needed. This should get you the stable SCF solution to start with
 and the go from there.
  > Dear CCL'ers,
  >
  > I'm doing some calculations (geometry optimizations) on small systems
  > using DFT as implemented in Gaussian 94 and 98.
  > Once the geometry is optimized, I do a stable(opt) job on the wave
  > function corresponding to that geometry. I find very often that the wf is
  > unstable, but the opt option in the stable keyword works well and
 overcome
  > the unstability problem.
  > Then, I try a frequency calculation using guess(read) and geom(check) but
  > the calculation is performed on the unstable wf instead of the stable
 one!!
  > How can I perform the frequency job using the stable wf? Which
  > combination of keywords am I missing?
  > Thanks in advance.
  > Regards,
  >
  > Reinaldo
 Laurence Cuffe [Laurence.Cuffe #*at*# ucd.ie]:
 I'm only stepping in with advice here as you say a similar question
 in the archives remains un-answered.  I would use the final
 geometry from the stable calculation to start an opt=(calcall)
 calculation with the multiplicity in the input file corrected to
 correspond to that derived from the stable calculation.
 Hope this helps
 Christopher Cramer [cramer #*at*# pollux.chem.umn.edu]:
    We work with unstable wave functions all the time. We find it much more
 efficient to simply run a "stable" calculation (i.e., no
 "=opt"). If there
 is indeed an instability (which the output will tell you), then we proceed
 to
 # u[dft method]/[basis] guess=(read,mix) scf=([qc,nosym,direct,cycles=900])
  [fopt] [freq]
 where the brackets indicate wildcards (e.g., the particular DFT method and
 basis set) or options that may or may not be required to coax the scf to
 convergence (qc is helpful for difficult cases, nosym is essentially
 required for molecules that are not C1, but is better used within scf, so
 that geometry optimization can still take advantage of symmetry, direct may
 not work depending on the size of the system, number of cycles is a
 personal choice...) In this case, you can also add optimization and
 frequency keywords. Note that it is CRITICAL to include the "u" prefix
 to
 the DFT method.
    I seem to recall that the trouble with your approach is that stable=opt
 usually begins with a restricted wave function, and then if an instability
 is found, an unrestricted wave function is optimized (although often it
 seems unable to accomplish this, while guess=mix as listed above is quite
 robust). HOWEVER, when link 1 is entered again because of your use of the
 keyword freq, it defaults back to a restricted formalism, so you simply
 guess the alpha orbitals as an RDFT wave function. My memory may certainly
 be wrong on this particular point -- I haven't tried it in a long time --
 but we usually have excellent success with the protocol I've outlined
 above.
 gaussian.com!help #*at*# gaussian.com
      A two-step calculation is necessary. In the second job, the
 frequency part, use "ub3lyp geom=allcheckpoint guess=read nosymm". If
 you do not explicitly specify the "u" in ub3lyp, the calculation will
 converge to the spin-restricted wavefunction. If this doesn't help,
 please send me your input file and I will attempt to find a set of
 options for you.
 Regards,
 Jim Hess
 ++++++++++++++++++++++++++
 James Hess, Ph.D.
 Customer Support Scientist
 Gaussian, Inc.
 e-mail: help #*at*# gaussian.com
 ++++++++++++++++++++++++++
 Stock [mstock #*at*# chem.ubc.ca] :
 You should be able to get the optimized wavefunction with appropriate
 use of Guess.  Try Guess=Mix, which mixes the HOMO and LUMO.  If that
 doesn't work, check out Guess=Alter, G98 user's reference pp. 98 and
 101.  This will allow you to specify by hand which MOs you want
 occupied.  I'm unfamiliar with the notation used in the manual, so if
 you end up taking this route and you are familiar with it, I'd be much
 obliged for an explanation.  Perhaps a #P Stable calculation would give
 you the info you need to correctly choose the alterations to the
 wavefunction.
 While you're trying to create the stable wavefunction, you can of course
 test it with another stable calculation.
 Juan Pablo Senosiain [senos #*at*# Stanford.EDU]:
 it sounds like you might have a UHF/RHF instability.  This is well
 documented in standard quantum chemistry textbooks.  Try the homolytic
 bond fission using UB3LYP.
 Patrick Musch [musch #*at*# chemie.uni-wuerzburg.de]:
 as far as I know, the SCF routine in Gaussian uses quatrically
 convergent (QC) SCF, so try to read in the stable wavefunction from the
 checkpoint file and do a Link1 for the frequency job GUESS=READ and
 SCF=QC. This might take a little time, since QC SCF is more
 time-consuming than the ordinary algortihm. I hope this might work.
 Olga Dmitrenko [odmitr #*at*# UDel.Edu]:
 Did you try to do first job with opt=stable and link it with the
 control card # guess=read geom=check freq ?
 We are doing similar studies on O-O bond elongation in peroxides.
 Our way is to do UB3LYP with guess=mix.
 **************************************************
 Theodorus de Bruin
 Institut Français du Pétrole
 Division Chimie et Physico-Chimie Appliquées
 Dept. Thermodynamique & Modélisation Moléculaire
 1 - 4 Avenue de bois Préau
 92852 Rueil-Malmaison Cedex
 Tel.: +33 (0)1.47.52.54.38
 Fax.: +33 (0)1.47.52.70.58
 **************************************************