[molpro-user] Closed-shell CCSD code

Werner Győrffy gyorffy at theochem.uni-stuttgart.de
Mon Jul 20 11:19:03 BST 2015


Dear Ugur,

In terms of single reference methods using density fitting, Molpro 
includes local methods such as DF-LMP2, DF-LCCSD(T), DF-LCCSD, DF-LCC2, 
and DF-LADC(2). In terms of properties, we have DF-LMP2 analytical first 
and second derivatives; DF-LCC2 linear response theory and analytical 
first derivatives also for excited states. The canonical results can be 
alway computed by the local code. Furthermore, there are separate 
implementations of the canonical methods for energy calculations and 
analytical first derivatives for DF-MP2, DF-CCSD(T), DF-CCSD; as well as 
for their F12 versions.

Now, regarding your question:

The default implementation of CCSD in Molpro is without density fitting.

In the case of canonical methods, density fitting cannot speed up 
calculations where the method scales higher than O(N^5), such as, for 
example, CCSD which scales as O(N^6) or CCSD(T) which scales as O(N^7). 
Here, I do not consider methods where tensor decomposition techniques 
are used for doubles amplitudes or any 2RDMs. On the other hand, methods 
with O(N^5) or lower scaling are greatly accelerated: for example, HF, 
DFT, MP2, CASPT2 energies and analytical derivatives (all impleneted in 
Molpro). For local methods, linear scaling can be achieved also for 
coupled cluster methods so density fitting helps in reducing the 
calculation cost.

Density fitting in general eliminates the bottleneck of storing 
two-electron integrals.

Regards,

Werner.

On 07/19/2015 12:49 PM, Ugur Bozkaya wrote:
> Hi,
>
> Does the closed-shell CCSD code use the density fitting approximation
> for integrals (fully or partially) to speed up computations?
>
> Thanks,
>
> --
> Assoc. Prof. Ugur Bozkaya
> Department of Chemistry,
> Ataturk University,
> 25240 Erzurum, Turkey,
> Tel: +90 442 231 41 14,
> E-mail: ugrbzky at gmail.com <mailto:ugrbzky at gmail.com>,
> ugur.bozkaya at atauni.edu.tr <mailto:ugur.bozkaya at atauni.edu.tr>.
>
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user
>



More information about the Molpro-user mailing list