deficiencies in semi-empirical methods

["charles martin" <chm6 "-at-" midway.uchicago.edu>]

 Hello netters,
 I am hoping that some of you can help me out.  I am trying to find
 specific references describing in detail deficiencies in semi-empirical
 methods as well as attempts and/or successful improvements.
 Here is a list of topics which I am trying to find out about:
 1) The use of different parameterization schemes for specific molecules
  or classes of molecules to treat different properties
  (e.g. excitation energies, ionization potentials,
  heats of atomization/formation, etc.)
 2) Specific examples of calculations demonstrating where semi-empirical
  methods fail but ab initio methods succeed (and vice versa).
 3) Ab Initio calculations of semi-empirical parameters.
   (and/or)
  Models which explicitly describe how correlation contributions enter
  into the semi-empirical parameters.
 4) Inclusion of Rydberg orbitals, d orbitals, etc. into
  semi-empirical valence spaces.
 I realize that there are a lot of works which describe new and improved
 methods (I would be very happy to read any of these sent my way).
 What I am particular looking for, again, however, are documented
 examples detailing the deficiencies of semi-empirical methods.
 I will summarize the information I receive and provide it via email
 to any interested parties.
 Thank you very much in advance:)
 Charles H. Martin
 The University of Chicago
 Department of Chemistry and the James Franck Institute
 chm6 "-at-" quads.uchicago.edu