From chemistry-request -8 at 8- ccl.net Wed Sep 25 22:36:29 1991
Date: Wed, 25 Sep 91 21:10:04 EDT
From: m10!frisch -8 at 8- uunet.UU.NET (Michael Frisch)
Subject: Re: help for HOMO and LUMO precision
To: chemistry <-at-> ccl.net
Status: RO

    
    Hello, netters!
    
    I am trying to extract HOMO and LUMO values from GAUSSIAN
    output. The problem is when different basis sets are used,
    there will be huge difference for the results. The molecules
    I tried were water and methanol.
    
    Can anybody help me figure out why? Is there any better softwere
    to use for this purpose?
    
    P.S.: Since GAUSSIAN cannot do computations on big molecules, I
          move MOPAC will be my next choice. Then, are the HOMO and
          LUMO values from MOPAC acceptable, comparing with those
          from GAUSSIAN?
    
    Thanks
    
    
    W. Philip Leigh
    
    (1)  97748509 -x- at -x- WSUVM1.CSC.WSU.EDU
    (2)  wlei $#at#$ yoda.eecs.wsu.edu
    

The values of the orbitals evaluated at specific points in space
change only moderately with basis set; the values of the coefficients
depend on the basis functions -- clearly, when you go from minimal
basis to double-zeta, each coefficient of an MBS orbital is replaced
by coefficients of two orbitals.  These coefficients will vary
depending how the relative sizes of the two basis functions and how
tight or diffuse the MO would like to be (given the flexability which
is not present in a minimal basis).

If the minimal basis is particularly poor for a molecule (for example,
an anion, for which the orbitals may be much different in size from
the fixed size imposed by a minimal basis), then the orbitals will
change qualitatively.

As for "big molecules", that depends on your definition of big.  Of
course, semi-empirical methods are cheap and applicable to bigger
systems than ab initio.  On the other hand, it is routine to do a
couple of dozen atoms and several hundred basis functions on modern
workstations with Gaussian 90, especially if you just want to look at
orbitals.  So there's a fair amount you can do ab initio.

Semi-empirical methods constrain you to a minimal basis.  That avoids
the complications of interpreting results from larger basis sets, but
it also means you don't have the flexability of better calculations if
that's what you need.  I would suggest making the choice between ab
initio and semi-empirical methods on the basis of whether the
semi-empirical methods are reliable for the problems you're interested
in.  For stable structures of neutral molecules involving the atoms for
which lots of parametrization information is available (H, C, N, and O)
you will find that the STO-3G MO's and the AM1 MOs are indeed very
similar (provided that you interpret the raw AM1 MO coefficients as
coefficients of symmetrically orthogonalized AOs, as Gaussian does
before printing AM1 orbitals).  If you're primarily interested in
orbitals for this type of system, there isn't much point in doing
STO-3G -- you can either do AM1 (which is as good and cheaper) or
Hatree-Fock with a larger basis set (which is more accurate).  The
farther you stray from the regime for which the semi-empirical method
was parametrized, the more you need to concern youself with its
reliability.

Michael Frisch
Gaussian, Inc.
-------