17.6.4 Orbital domains

In most cases, the domain selection (cf. section 17.4 is uncritical for saturated molecules. Nevertheless, in particular for delocalized systems, it is recommended always to check the orbital domains, which are printed in the beginning of each local calculation. The orbital domains consist of all basis functions for a subset of atoms. These atoms are selected so that the domain spans the corresponding localized orbital with a preset accuracy (alterable with option THRBP). A typical domain output, here for water, looks like this:

 Orbital domains

   Orb.   Atom    Charge       Crit.
   2.1    1 O1     1.17        0.00
          3 H2     0.84        1.00
   3.1    1 O1     2.02        1.00
   4.1    1 O1     1.96        1.00
   5.1    1 O1     1.17        0.00
          2 H1     0.84        1.00
This tells you that the domains for orbitals 2.1 and 5.1 comprise the basis functions of the oxygen atom and and one hydrogen atom, while the domains for orbitals 3.1 and 4.1 consist of the basis function on oxygen only. The latter ones correspond to the oxygen lone pairs, the former to the two OH bonds, and so this is exactly what one would expect. For each domain of AOs, corresponding projected atomic orbitals (PAOs) are generated, which span subspaces of the virtual space and into which excitations are made. Options which affect the domain selection are described in section 17.4. Improper domains can result from poorly localized orbitals (see section 17.6.3 or a forgotten NOSYM directive. This does not only negatively affect performance and memory requirements, but can also lead to unexpected results.

The choice of domains usually has only a weak effect on near-equilibrium properties like equilibrium geometries and harmonic vibrational frequencies. More critical are energy differences like reaction energies or barrier heights. In cases where the electronic structure strongly changes, e.g., when the number of double bonds changes, it is recommended to compare DF-LMP2 and DF-MP2 results before performing expensive LCCSD(T) calculations.

The effect of domain approximations is strongly reduced in explicitly correlated calculations [e.g., DF-LCCSD(T)-F12] and the use of these methods (see below) is therefore strongy recommended (but the F12 option is not available for OSV methods).

molpro@molpro.net 2019-03-22