[molpro-user] CBS extrapolations

[Kirk Peterson]

John,

1) it is often more convenient to extrapolate total energies, but I think you would find few people
that would say it is more accurate to do so in general. The one caveat is when you can only use small
basis sets, see #2 below.  With DZ-QZ you might also consider extrapolating the HF energy with
the exponential function.

2) Certainly with CBS extrapolations using the inverse power law formula that molpro uses by
default (L3), using more than a 2-point extrapolation will only make your results worst since this
formula really needs only the largest basis sets to work well.  So I would certainly not do a
DZ-QZ extrapolation with n^{-3}.  Of course if you can only go up to QZ you might compare your
TZ/QZ L3 extrapolation to extrapolating the total energy with the so-called mixed Gaussian-exponential 
formula (EX2 in Molpro) since this has been shown to give statistically slightly better results for relative energies at
this level of basis set.  With basis sets of this size you have to be a bit more pragmatic I'm afraid.

my 2 cents,

-Kirk


On Sep 22, 2010, at 6:49 AM, John Keith wrote:

> Might someone be able to share expertise doing CBS extrapolations on molecules?
> 
> I'd like to benchmark DFT energies of some inorganic complexes against
> CCSD(T) methods extrapolated to the CBS limit.  Basically, I'd like to
> do a similar study as was done recently in Inorg. Chem. (2010) 5546.
> Two questions for someone new at this:
> 
> 1. Molpro's default procedure appears to extrapolate only the
> correlation energy and then add it to the HF energy of the largest
> basis set used in the extrapolation.  When in doubt I normally do the
> default procedure, but I've seen other papers that give reasons why
> extrapolating the total energy, particularly when using Dunning-type
> cc-basis sets is preferred.  Does anyone reading this board have
> strong feelings one way or the other?
> 
> 2. One example in the molpro manual shows an extrapolation using only
> ATZ and AQZ basis sets.  Naturally, a CBS extrapolation using only two
> data points is less ideal than one using 5 or 6, but the size of my
> molecules rules out basis sets larger than AQZ.  Would a CBS(2-4)
> calculation be more meaningful than a CBS(3-4) calculation?  The
> CBS(3-4) calculation is ~0.002 Eh (1.3 kcal/mol) lower than that
> CBS(2-4) calculation. Should I trust either of these approximations,
> or am I better off just benchmarking to a more definite CCSD(T)/AQZ
> value?
> 
> Any advice on the matter would be greatly appreciated.
> 
> Thanks in advance,
> 
> --john
> 
> #####################################################################
> Ph D. John A. Keith
> Institut für Elektrochemie     Email: john.keith at uni-ulm.de
> Universität Ulm                    Telephone: +49-731-50-26061
> Albert-Einstein-Allee 47       URL: http://www.wag.caltech.edu/home/johnk
> Ulm D-89069                       URL:
> http://www.uni-ulm.de/theo_echem/index.shtml
> #####################################################################
> In order to be an immaculate member of a flock of sheep, one must above all,
> be a sheep.
> 
> --Albert Einstein (1879-1955)
> #####################################################################
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user