Chemistry Reference
In-Depth Information
cancellation of errors when compared to our most reliable results, as discussed
in more detail below.
Pauling Points
As we have been discussing, it takes two components to specify a level of
theory: the basis set and the electron correlation model. In computations of
molecular structure, it has long been known that larger basis sets tend to yield
shorter bond lengths, while more complete treatments of electron correlation
tend to yield longer bond lengths. The best results tend to be those where a
good balance exists between the basis set and the theoretical method; such a bal-
ance comes from favorable cancellation of errors between the incompleteness of
the basis set and the incompleteness of the electron correlation model. The situa-
tion where a good result can arise even though an incomplete model is being used
has been referred to as a
Pauling point
, after the famous chemist Linus Pauling.
Given the extreme expense of large-basis CCSD(T), we have sufficient
motivation to seek theoretical Pauling points for
interactions. Returning to
Tables 2 and 3, we observe that the MP2/aug-cc-pVDZ
0
results are in remark-
ably good agreement with the best theoretical benchmark values. Although
MP2/aug-cc-pVDZ
0
slightly overestimates intermolecular distances compared
to MP2 with the full aug-cc-pVDZ basis, this overestimation is just what is
needed for good agreement (within 0.1
˚
) with the benchmark values. The per-
formance for interaction energies is even more impressive. Although MP2/aug-
cc-pVDZ
0
greatly underbinds compared to MP2/aug-cc-pVDZ, this underbind-
ing is just what is needed for better agreement with the benchmark results. For
the three prototype configurations of the benzene dimer, all of the interaction
energies agree within 0.4 kcal mol
1
of the large-basis CCSD(T) results. Note
that MP2 with the slightly larger aug-cc-pVDZ*, or the full aug-cc-pVDZ basis,
yields much poorer agreement with the benchmark CCSD(T) results. Going to
larger basis sets with MP2 gives even worse results due to dramatic overbinding.
This suggests that, in the context of conventional quantum chemical computa-
tions, the MP2 method with a basis set such as aug-cc-pVDZ
0
may be the most
cost-effective approximation to large-basis CCSD(T). However, it should be
noted that the good agreement between MP2/aug-cc-pVDZ
0
and large-basis
CCSD(T) may not hold for other weakly bound complexes. This is a topic
our research group hopes to investigate in more detail. In the meantime, as dis-
cussed below, there may be some alternative approximations that may be even
more effective at providing high-quality results at reduced computational cost.
p
Resolution of the Identity and Local Correlation
Approximations
Even within the context of a given target level of theory, e.g., MP2/aug-cc-
pVTZ, there exist various approximations that allow one to estimate the desired
