Volume keyword in Gaussian: problem

["Keith E. Gutowski" <gutow001^at^bama.ua.edu>]

 I am trying to compute ionic volumes using the 'volume=tight' keyword in
 Gaussian03.  In the manual, it states for a volume calc:*
 'Tight*
Requests an increased density of points for more accurate integration.
 By default, the volume is computed to an accuracy of about 10%. Use of
 this option is recommended if the computed molecular volume is needed
 more quantitatively.'
Now, when I caculate the volume of PF6- using B3LYP, DZVP2 basis set
 (opt + volume=tight), I get a strange result.  Here is the volume
 output
 for 6 different runs of the same input file (in cubic cm per mol):
 69.815
 57.036
 65.763
 60.465
 68.880
 54.854
Now, I would hope that Gaussian03 could generate better
 reproducibility
 that this, but am I wrong?  I have been trying to track down the source
 of this, but I am a bit stumped.  The only this I can come up with has
 to do with the error on the integrated density.  For each run (PF6- or
 whatever molecule I do), the integrated density is about 4.23D+01,
 while
 the error on the integrated density is about -2.75D+01.   I thought the
 error on the integrated density should be around XXXD-06.  This is
 printed out in the output along with the volume output results.
 Anyone have any idea what may be going on?
 Keith