From:	"DAVID W. EWING (216) 397-4742" <EWING ( ( at ) ) JCVAXA.JCU.
Date:	Tue, 15 Jun 1993 14:20 EST
Subject:	Responses to restricted vs. unrestricted semi-empirical question

Here's a summary of the responses to my recent question concerning
restricted vs. unrestricted methods for semi-empirical Hamiltonians.
Thanks to all who responded!  The question:

We're looking at some systems with doublet spin states using MOPAC5.  Our
previous work on some of these systems using ab initio methods told us that
some of the wavefunctions are significantly spin contaminated ( about
1.0).  Question:  should we stick to restricted MNDO, MINDO/3, etc.  or is
it "safe" to use the unrestricted methods?  The question arises because we
are seeing significantly different results restricted vs. unrestricted.

Dave Ewing
John Carroll University
ewing { *at * } jcvaxa.jcu.edu


The answers:
 
From:	IN%"fredvc (+ at +) esvax.dnet.dupont.com" 12-JUN-1993 14:17:38.66

	UHF calculations are attractive because of their (relative) simplicity.
They are *PROBABLY* alright for a qualitative assessment, e.g., for ESR
interpretations, as long as the spin contamination is relatively "low"
(whatever that means!!).  Things begin to get sticky when comparisons of
differing structures/isomers/conformations are involved.

    (1)	Is the level of spin contamination the same?

    (2)	Is the source of the contamination (quartet, sextet, etc.) the same
	for all forms?

	Projection techniques can clean things up a lot. This raises questions
of approach, however.  Do we do a UHF and project at the end, or do a PHF
projecting as we go?  I prefer the latter, thouhg it is easier to get codes for
the former.

	I would tend to come down on the side of RHF methods, just to avoid the
potential problems & ambiguities mentioned above.


  
From:	IN%"SDUDIS%BMSUEM11.BITNET (- at -) FRMOP11.CNUSC.FR"  "Doug Dudis, WPAFB"
14-JUN-1993 04:09:08.66

This sounds like a more general problem.  Specifically, for the polyacetylene
case (and it's oligomers, the all trans polyenes, at least for the larger ones)
the UHF wavefunction gives an INCORRECT geometry (and hence other problems).
The real geometry is dimerized (that is, has double and single bonds) whereas
the UHF wavefunction has someimbalance in it which results in a non-alternate
structure (all bonds equal).  This has been analyzed in some detail in the
lit., and sounds similar to the problem you are encountering.  I apologize for
not giving the references, but I am on sabbatical in Belgium at the moment
and so don't have access to all my references.  As I recall, several of the
papers were in J. Chem. Phys. a few years back.


From:	IN%"rwoods ^at^ biop.ox.ac.uk" 14-JUN-1993 14:09:52.26
 
In response to Dave Ewings question regarding spin contamination in
semiempirical wavefunction I would just like to mention that I ran into
the same problem when I incorporated a finite perturbation routine into
MOPAC6.0.
 
In some spin systems, notably those involving unsaturated carbon-carbon bonds
and those involving nitro groups I could not get SCF convergence with a
UHF-wavefunction regardless of the Hamiltonian (AM1, PM3 etc.).  An examination
of the values of S^^2 indicated that the wavefunctions were converging
slowly, but actually diverged when MOPAC automatically applied any of its
built-in "convergers".  These convergers are applied when certain criteria such
as the number of SCF-cycles have been passed.  When I altered the code so that
the automatic use of convergers such as "PULAY" and "CAMP-KING" was avoided the
SCF slowly converged.
 
I have been told that spin contamination in NDDO methods is an accepted
draw-back.  I would be very interested to hear if anyone else has explored
methods of correcting this problem.

 
From:	IN%"asnegb01 "at@at" asncube.asc.edu"  "Eric G. Bradford" 11-JUN-1993
11:54:47.91

My experience in the past has been that MOPAC UHF will also give
contamination in .  I suggest that you use RHF and its half-electron
approach to handling orbitals which are not fully occupied.


From:	IN%"jstewart' at \`fai.com" 11-JUN-1993 16:01:25.52

For electronic structure, stick with RHF.  For energetics, the RHF method
will give heats of formation a little higher than UHF.  The difference
is likely to be almost constant, so if you are after trends, UHF will
do.


From:	IN%"AHOLDER - at - VAX1.UMKC.EDU"  "Andy Holder" 11-JUN-1993 18:21:55.97

  I would stick with the RHF/HE methods.  Remember that when one
has an open shell in RHF, Dewar's half-electron protocol is used,
which is usually pretty reliable.  The spin-contaminated wave-
functions are not valid.

  An alternative is to do CI and elimiate the contamination prob-
lem altogether.  AMPAC has the best CI stuff in it.


From:	IN%"jan (+ at +) si.fi.ameslab.gov" 11-JUN-1993 18:32:23.96

	About a year ago we implemented many features of MOPAC into our
ab initio code GAMESS.  In addition we added some new features, among them
high spin ROHF.  Since ROHF is an eigenfunction of the spin operator you
avoid the problem of spin contamination.  GAMESS can be obtained by
writing Dr. Mike Schmidt at mike ( ( at ) ) si.fi.ameslab.gov, and is free of
charge.
Hope this helps.


From:	IN%"zheng.,at,.violet.berkeley.edu" 11-JUN-1993 18:39:24.19

 As you said that there are two ways to treat open-shell system, one is using
 Dewar's Half-Electron method which works fine for doublet radical but it's
 relatively slow. The second method is pople's UHF method. Although the
 wavefunction is contaminated, it does give realistic results as far as ESP
 is concerned. HE method will not give you negative electron density. If you
 are interested in energetics, HE is the method of chioce. In the UHF treatment
 electron correlation has been overcalculated, therefore the energetics may
 not be good.

 HE will fail for biradicals or triplet carbene. I usually use UHF when the 
 value is not too far from the that of pure state since it's faster. So again
 it depends on what want to get from your calculations.    Yajun Zheng


From:	IN%"zheng { *at * } violet.berkeley.edu" 12-JUN-1993 12:28:11.35

 In my previous mail I did not say clearly
that uhf method could give negative spin density. HE method will not.
As far as triplet carbene is concerned, we found that the error in calcualated
heat of formation is related to the exchange integral Kij. (Dewar and Zheng
unpublished results)  Yajun



Similar Messages
08/01/1995:  Spin contamination, effect on energy and structure.
12/16/1994:  Spin contamination & AM1 "ROHF" versus UHF
04/16/1997:  Re: CCL:Chem Topic: Spin Contamination
04/08/1997:  Chem Topic: Spin Contamination
08/01/1996:  Re: CCL:M:Heat of formation calculation using MOPAC.
06/14/1993:  Re: Restricted vs Unrestricted Wavefunctions
11/08/1994:  Summary: UHF vs ROHF in semiempirical calc's of radical cations
11/01/1996:  G:SUMMARY: Wavefunction Instability
06/16/1997:  Summary: wfn stability
08/29/1995:  Summary for "Calculation on biradical"


Raw Message Text