DLPNO-CCSD(T) is another technique of Couple-Cluster generalization method which produces the reasonable CCSD(T) results but requires very cheap computation cost compared to the traditional CCSD method. DLPNO-CCSD(T) stands for the domain-based local pair natural orbital methods. It has been implemented in ORCA quantum chemistry software, which developed by Frank Neese.
With DLPNO-CCSD(T) in ORCA, we are able to perform the linear scaling CCSD(T) calculations at the cost of a simple DFT calculation and reaching the experimental results. ORCA's DLPNO capabilities are also available for multi-reference systems as well as for the calculation of properties.
The following is a DLPNO-CCSD(T)/cc-pVTZ calculation of water.
! DLPNO-CCSD(T) cc-pVTZ cc-pVTZ/C
* int 0 1
O 0 0 0 0.0 0.000 0.000
H 1 0 0 1.0 0.000 0.000
H 1 2 0 1.0 104.060 0.000
*
DLPNO-CCSD(T) will give the following output. They are at almost the end of output file.
-------------------------------------------
DLPNO BASED TRIPLES CORRECTION
-------------------------------------------
⁝
Triples Correction (T) ... -0.007718230
Final correlation energy ... -0.278154390
E(CCSD) ... -76.321436435
E(CCSD(T)) ... -76.329154666
⁝
or go to the end of output file and see following
⁝
------------------------- --------------------
FINAL SINGLE POINT ENERGY -76.326535164724
------------------------- --------------------
⁝
The current implementation of DLPNO-CCSD(T) in ORCA is a black-box, where there is many parameters that needed fine-tune for unique complex. Therefore, the developer leaves the setting of this method for us. Three parameters we have to concern are following:
The PNO occupation number, called TCutPNO
Thestrong pair approximation cut-off, called TCutPairs
The domain size parameter, called TCutMK
The default values for each parameter cut-off.
LoosePNO NormalPNO TightPNO Loose2PNO Normal2PNO
TCutPairs e-3 e-4 e-5 e-5 e-5
TCutMKN e-3 e-3 e-4 e-3 e-3
TCutPNO e-6 3.3 x e-7 e-7 e-6 3.3 x e-7
Example input file for manual setting for TightPNO mode with Tight SCF cutoff
! DLPNO-CCSD(T) cc-pVTZ TightPNO TightSCF
or manually set their values in mdci block
! DLPNO-CCSD(T) cc-pVTZ TightSCF
% mdci
# Not construct density matrix
Density None
TCutPairs 1e-5
TCutPNO 1e-7
TCutMKN 1e-4
end
...
To speed up the calculation, you can also use RI approximation (so-called density fitting) and auxiliary basis sets together
! DLPNO-CCSD(T) RIJCOSX def2-TZVP def2-TZVP/C TightSCF
where RIJCOSX is RI-approximation, the RI-J for Coulomb integrals and COSX numerical integration for HF exchange, and def2-TZVP/C is auxiliary basis set for def2-TZVP.
Increasing memory can speed up calculation too!
For large jobs, the calculation may require a large amount of memory. We can set the total amount of (physical) memory for the job using maxcore keyword.
% maxcore MEMORY
Note that the memory is in MB unit and it would be requested per CPU core. Therefore, if you run in parallel mode with N CPU cores, the total amount of memory will be N x Memory! Make sure that your machine has memory enough.
! DLPNO-CCSD(T) TightPNO TightSCF
# Number of CPU processors
% pal nprocs 4 end
# Memory per core (MB)
% maxcore 3000
...
With the above setting, the total memory is 4 x 3000 = 12,000 MB, which roughly is 12 GB
To achieve more accurate results as a standard reference for your own studied system, we can extrapolate the DLPNO-CCSD(T)/BasisSet to complete basis set (CBS) using the following equation.
Energy(DLPNO-CCSD(T)/CBS) = Energy(MP2/CBS) + Energy^(DLPNO-CCSD(T))_(core)
https://sites.google.com/site/orcainputlibrary/coupled-cluster
https://cec.mpg.de/fileadmin/media/Forschung/ORCA/orca_manual_4_0_1.pdf
Rangsiman Ketkaew