Diffuse functions needed?

For accurate results on properties like polarizabilities, hyperpolarizabilities, and high-lying excitation energies, diffuse functions are needed. This is especially true for calculations on small molecules. In larger molecules the nature of the relevant virtuals is much more 'molecular', much less Rydberg-like, so that the normal basis sets may be sufficient. Basically all properties calculated through the RESPONSE keyword may require diffuse functions. If you use the EXCITATIONS keyword, the necessity of diffuse functions depends on the type of excitation you are interested in. The lowest excitations do not require diffuse functions, but Rydberg excitations do.

In case of diffuse basis functions the risk of linear dependency in the basis increases. This can be checked, and corrected for with the DEPENDENCY keyword. It is recommended to use this keyword for all calculations involving diffuse functions. A good default setting is

DEPENDENCY bas=1d-4

However, it may be advisable to experiment a bit with the parameter, especially if many linear dependent combinations of AOs are removed. Using too many diffuse functions on a large molecule will lead to insurmountable numerical problems. In such a case it is not only useless, but even harmful, to add many diffuse functions.

In the previous release only some basis sets were provided which contained diffuse functions. These were gathered in the directory Vdiff. Now several ET basis sets have been developed for the elements up to Ar containing some or many diffuse functions. We recommend to use these instead of the Vdiff directory. Most of these basis sets are quite large and not very suitable for large molecules.

In ADF2005.01 augmented basis sets were added in the AUG directory, especially devised for use in in TDDFT calculations, such that one can do a reasonable accurate calculation of excitation energies, with a relatively small basis set, see Del Chong in Mol. Phys. 103 (2005) 749.