This mimics almost exactly what we have experienced. However I would
like to add one additional point which I don't think has been mentioned
already. I have found in some calculations the gradients go to exactly
zero which then causes ADF to exit with an error at some later point in
the calculation. It is also worth pointing out these same calculations
worked perfectly with ADF1999.

Dr R J Deeth wrote:
>
> Dear All,
>
> Further to Luigi's message, we too have noticed that the internal
> coordinate optimisation in ADF2000 is nowhere near as reliable as in
> 2.3. We find in particular that torsion terms involving say H atoms
> attached to aromatic rings become unstable and oscillate through +-10
> degrees so that the energies and gradients are low but the angle
> convergence is not met.
>
> Then, on other occasions, we get gradients going through the roof for
> no good reason. As Luigi finds, ADF2.3 performs much better as
> does ADF2000 optimisations in cartesians. However, the latter is not
> so convenient for linear transits.
>
> If this problem with ADF2000 optimisations is a general observation,
> perhaps we should encourage Stan to take a look.
>
> Rob Deeth
>
> =============================================
> = Dr Robert J Deeth
> = Inorganic Computational Chemistry Group
> = Department of Chemistry
> = University of Warwick
> = Coventry CV4 7AL, UK
> = ph: +44 (0)24 76 523187
> = Fax:+44 (0)24 76 524112
> = Web: http:/www.warwick.ac.uk/~mssbq
> =
> = DALTON DISCUSSION 4: Inorganic Reaction Mechanisms
> = 10-13 January, 20002
> = Kloster Banz, near Bamberg, Germany
> = http://www.rsc.org/lap/confs/dd4.htm
> =
> =============================================

-- 
Chris Delfs                             email
Christopher.Delfs_at_anu.edu.au 
Dept of Chemistry                       Tel   +(61)(2) 6125 3071
Australian National University          Fax   +(61)(2) 6125 0760
Canberra  ACT 0200 
Australia