Friday, January 3, 2014


Contributed by Frank Jensen

A couple of recent papers illustrate the problems related to using the CP correction for reducing BSSE:

Ł. M. Mentel and E. J. Baerends, JCTC ASAP, DOI: 10.1021/ct400990u
Report that for He-He and Be-Be interaction potentials, the CP correction can be in the wrong direction. The cause is apparently that the basis sets are optimized for the atoms, and thus biased against the complex. The uncorrected interaction energy is thus underestimated and adding the CP correction further underbinds the complex.

Lori A. Burns, Michael S. Marshall, and C. David Sherrill, JCTC ASAP, DOI: 10.1021/ct400149j
Perform a benchmark study using MP2 and CCSD(T) with and without CP corrections, or the average, combined with basis set extrapolation for the A24 benchmark systems + a few extras. Their conclusion is that whether to use the CP, half the CP or no CP depend on the system, method and basis set. Their (weak) recommendation is to use half the CP correction for basis sets of aDZP or aTZP quality to 'avoid the worst errors incurred by either method', and the full CP for larger basis sets and extrapolations.

The latter study most likely has components of the first: A given (fixed) basis set will be (slightly) non-optimum for each fragment and the complex, but which fragment/complex that it is least optimum for will depend on the system and geometry. The inherent basis set over/under-binding of the complex will be modulated by the overbinding by the BSSE. Adding the CP estimate of the BSSE can then lead to either improvement or deterioration of the final binding energy. Noting that all three effects are small, the inherent basis set error can have either sign, the BSSE is always negative and the CP correction is always positive, the combined effect will have significant 'random' errors compared to the exact result, which in magnitude also is small.