[molpro-user] LCCSD(T) is more time and memory consuming than CCSD(T)

[Tatiana Korona]

Dear Leonid,

It is obvious, why it is so: for a small molecule most orbital domains are just 
full domains (i.e. they contain PAOs from all the atoms), so their dimensions 
are necessarily larger than the number of virtual orbitals.

For CCSD(T) the time scaling is v^4o^3, and for LCCSD(T) for small molecules, 
where full domains have been chosen, - N^4o^3, where N is a number of atomic 
orbitals or close to it, i.e. N>v. Additionally, the LCCSD(T) code surely has a 
more complicated structure, what is not so important for really large molecules, 
for which it is devised, but it can has an influence on the performance of this 
code for small molecules, like water.

Best wishes,

Tatiana

  On Sun, 18 Feb 2018, Leonid Shirkov wrote:

> Dear Molpro Community,
>
> I was testing the basis set convergence of LCCSD(T)
> for some small molecules and found out that it is more time
> consuming that conventional CCSD(T) for the same orbital basis set.
>
> My tests for water showed that CCSD(T)/av5z requires 50,m and
> runs much faster than LCCSD(T)/av5z which requires at least 400,m and gives
> "insufficient memory available" error otherwise (two inputs are attached).
>
> I understand that there are many options for LCCSD(T) which have
> effect on its performance, but shouldn't be LCCSD(T), especially the
> DF version, be faster and less demanding than conventional CCSD(T)?
>
>
> Best regards,
> Leonid
>

Dr. Tatiana Korona http://tiger.chem.uw.edu.pl/staff/tania/index.html
Quantum Chemistry Laboratory, University of Warsaw, Pasteura 1, PL-02-093 Warsaw, POLAND