[molpro-user] cp-corr. Potential DF-LCCSD(T0), numerical accuracy

[Gerald Knizia]

On Tuesday 20 January 2009 16:53, Berger Raphael wrote:
> when trying to calculate a cp-corrected potential curve
> in a close region around a uncorrected minimum
> it seems to give a "noisy" curve for the dE_cp using
> only the default settings. The cp-correction at the
> equlibrium geometry is about 8*10^-4 au.
> The level is DF-LCCSD(T0)/SARC-TZVPP with DKH3
> integrals.

Local CCSD works by neglecting parts of the wave function with small impact. 
Which parts fall below the thresholds depends on the molecular geometry. This 
leads to non-continuous potential energy surfaces unless special care is 
taken about the domains and pairs (lookup mergedom in the manual).
This does not happen if you use conventional CCSD(T), which also usually is 
faster for small calculations.
-- 
Gerald Knizia