summary spurious intergrated densities

 any thnaks to everyone who replied. Your comments helped al lot.
 Here is a summary of the answers I received:
         Hi there, I have encountered the same problem a few times, and what you
 have to do is to alter the grid over which the integrals are calculated. Refer
 to  pages 6 and 9 of the "New Methods and Features in Gaussian 92/DFT"
 Notes for a complete description.
         Basically one way to correct this is to compute the integrals using a
 finer grid. Thiscan most easily be accomplished by using the Int=FineGrid
 keyword. (gives about 7000-9000 points per atom compared with the default of
 about 3000 points per atom with Int=Normalgrid).
         If this fails you can define a numnber of even finer grids by following
 the Grid Control Appendix in the Releas Notes on pg 9.
                                 Stephen Decker
                                 decker $#at#$
                                 Dept. of Chemistry,
                                 Univ. of Alberta, Canada
 You need to upgrade to G94...these problems have been fixed...of get G92 Rev
 Level F.
 John M. McKelvey                        email: mckelvey $#at#$ Kodak.COM
 Computational Science Laboratory        phone: (716) 477-3335
 2nd Floor, Bldg 83, RL
 Eastman Kodak Company
 Rochester, NY 14650-2216
      bernhard $#at#$
 Subject: Re:DFT/spurious integrated densities
 few weeks ago i struggle over the same problem.
 I got the tip to add the keywords
 SCF=NoVarACC INT=FineGrid
 which work in ca. 90% of all cases (using B3LYP).
 The first is not fully known to me (didn't find the keyword in the
 g92 manual), but i think it enhance the accuracy of the integrals
 or is related to this point, the second point results in better
 numerics while integration. Don't take it for good, i'm just a
 If this doesn't work, then a slight alteration of the geometry helps
 sometime (nearly in all remaining cases).
 Bernhard Eck
   Dr. Salzner,
    We have seen this problem with G92 and have made a couple of changes
 which you can use with G92 that improve the situation dramatically.
   Add the following Int=FINEGRID to your calculations.  This may be
 sufficient as it improves the integration accuracy.  If not also add
 SCF=NOVARACC which turns off using loose integration cutoffs on
 early SCF iterations.
  Both of these will increase computation time but have proven successful
 enough that INT=FINEGRID is standard in G94.  We have modified the cutoffs
 for early iterations and seldom see this problem with G94.
   Douglas J. Fox
   Director of Technical Support
   help $#at#$