32.10 Density-fitted LMP2 (DF-LMP2), LDCSD (DF-LDCSD), and coupled cluster (DF-LCCSD(T0))

Density-fitting LMP2, LDCSD and LCCSD calculations can be performed by adding the prefix DF- to the command name. The input is as follows:


Options for density fitting can be mixed with any options for LOCAL. Options for density fitting can also be given on a DFIT directive (see section 15).

The most important options for density fitting in local methods are

Fitting basis set used in LMP2 and in LCCSD for integrals with up to 2 external orbitals. If a correlation consistent basis set is used (e.g. cc-pVTZ) the corresponding fitting basis for MP2 us used by default (cc-pVTZ/MP2FIT). Otherwise the fitting basis set must be defined in a preceding basis block (see section 11).
Fitting basis set used in LCCSD for integrals over 3- and 4-external orbitals. The default is BASIS_MP2 and this is usually sufficient. However, the accurate approximation of 4-external integrals in LCCSD requires larger fitting basis sets than LMP2. Therefore, in order to minimize fitting errors, it is recommended to use the next larger fitting basis, e.g., BASIS_CCSD=VQZ for orbital basis VTZ.
If LOCFIT=1 local fitting is enabled. This is necessary to achieve linear scaling in DF-LMP2 (see Refs. [11-14]). The errors introduced by local fitting are usually very small, but there are some exceptions. For instance, LOCFIT=1 must not be used in counterpoise calculations, see section 32.9.8) Note that for small molecules LOCFIT=1 can be more expensive than LOCFIT=0.
For further details and options for density fitting see section 15.

molpro@molpro.net 2018-10-21