SUMMARY: A couple of questions about AIM analysis

Hello all,
 I thank everyone for answering my question on AIM analysis.  I was asked by
 some to post a summary, and so I am doing so.  It would appear that the
 overall concensus is that the Gaussian implementation of AIM is not the most
 robust and that more specialized software is more appropriate (i.e., MORPHY,
 AIM2000, etc.).
 First, my original question:
 I have implemented AIM analysis from with Gaussian98W successfully many
 times.  The keywords I use are: SCF(conver=8) density=current
 AIM=bondorders.  The level of theory in all cases has been
 RB3LYP/6-311+G(2d,p) which is the same level of theory as the other
 calculations involving these molecules (e.g., minimizations, frequency
 calculations, NBO analysis).
 However,when attempting the same calculation on one particular structure,
 the calculation fails.  I am given the information below:
  Error termination via Lnk1e in C:\G98W\l609.exe.
  Job cpu time:  0 days  4 hours 16 minutes  8.0 seconds.
  File lengths (MBytes):  RWF=   48 Int=    0 D2E=    0 Chk=    2 Scr=    1
 The Gaussian User's Reference suggests that the AIM code will fail at times
 and cites the particular instance when the message NEWTON-RAPHSON STEP
 FAILED is due to a molecule with rings that has an unusual topology (which
 may well be the case).  Is there any cure for this problem?  Am I able to
 tweak something to get the calculation to go to completion?
 Don't know how to force gaussian to work round difficult topologies, but our
 house code (MORPHY) is available for a modest fee and is it is easy adapt
 internal settings to get round most problems [it will also look at the
 of the laplacian, and elf, and soon other fields like lol]. If you want any
 info let me know.
 Richard Baders code (AIMPAC) is freely available and should be flexible
 to do what you want but I don't have any direct experience of it.
 noj Malcom
 We have done quite a bit of AIM recently but not using Gaussian.  There have
 been several comments about the G98 AIM package that suggest that it is not
 robust, and will fail in cases like yours where there is extreme curvature
 in the atomic surfaces.  I have heard of no "fix" for these cases.  I
 believe that the G98 AIM package is essentially a re-packaging of the
 original AIMPAC bundle of Bader et al.
 We have found that the AIM2000 package has given us very good results,
 although the MORPHY program is also claimed to be very good and may in fact
 be superior if its long-awaited next version ever comes out.  In several
 cases, atomic integrations attempted with the PROAIMV program from the
 AIMPAC bundle failed no matter what parameters we tried, but integration
 with AIM2000 succeeded.  We have also found it much easier to enhance the
 accuracy of integrals with AIM2000.
 Good Luck.
 Phil Hultin
 The Gaussian implementation of AIM is not reknowned for its robustness.
 Their love affair ended a long time ago and the aim in gaussian is an old
 and not maintained version.
 I suggest using aim2000 or aimpac or morphy from a gaussian output wfn file.
 Alexandre HOCQUET
    There is no known cure for this error.  The program is attempting
 to fit a surface so it can do integration within the various regions.
 There are a number of topological features which cause it problems.
 The code is doing a first order search and this is obviously not high
 enough order but no one has taken on the task of implementing a
 higher order search, linear search etc, similar to what is done for
 difficult geometry optimization.
    There are a couple of bond order estimates in the NBO package.  If
 you use
 #  ...  POP=NBORead
 and then after the structure add lines,
 you will see these printed out.
    If you need an estimate of the bond order better than something like
 Mulliken that is my recommendation.
   Douglas J. Fox
   Technical Support
   Gaussian, Inc.
 Dear Professor Gary Breton,
 Gaussian has a multiple of optimization and minimization methods.
 I think you are using the most default method, but I think there are other
 methods available, like QC, GDIIS, NODIIS, etc.
 I am not exactly sure where the error occurred because you cut off the
 output a little too short, but there are other methods.
 I am curious. Why do you use SCF(COVER=8)? The default I think is 7. Is
 there are a need for the extra accuracy in the SCF cycle for AIM
 Ha Yeon Cheong