Chemistry Reference
In-Depth Information
the aug-cc-pVDZ basis is large enough to make CCSD(T) computations very
expensive for molecules the size of the benzene dimer or larger—particularly if
they lack any symmetry. For example, in 2008, a CCSD(T)/aug-cc-pVDZ
computation on benzene dimer with no symmetry takes approximately
2 weeks on a 3.2-GHz Intel EM64T Nocona workstation. For routine appli-
cations, then, computational studies require a less expensive procedure. In the
following several sections, we will examine several possible approximations,
all with the aim of maintaining a fairly high reliability in the computational
results.
Truncated Basis Sets
One possible computational shortcut is to reduce the size of the basis
sets used by throwing out certain functions. With respect to the correlation-
consistent family of basis sets, this runs counter to the philosophy used in con-
structing the basis sets, where functions of similar energetic importance are
added together in shells as the size of the basis is increased.
44,63
This may
also degrade the smooth convergence toward the complete basis set limit
and make it harder to perform basis set extrapolation. However, these con-
cerns are more important for benchmark-quality studies than for routine che-
mical applications. We have examined the possible elimination of some groups
of diffuse functions in studies of
interactions,
61,64
and a study by Wilson's
group
65
indicates that some higher angular momentum functions can be
removed from the (nonaugmented) cc-pVXZ basis sets without serious degra-
dation of the total energies of several isolated small molecules.
In earlier studies of the potential energy curves of prototype configura-
tions of the benzene dimer,
61
because we desired many points along the curve,
we found it necessary to use basis set truncations to make the computations
feasible (these computations were performed again later
66
without the basis
set truncations). We originally used a truncated aug-cc-pVDZ basis, which
we denoted aug-cc-pVDZ*, for the CCSD(T) computations, speeding them
up significantly. This is the usual aug-cc-pVDZ basis, except that diffuse func-
tions from hydrogen atoms were removed. For our symmetry-adapted pertur-
bation theory (SAPT) computations,
67
an even smaller basis set was used that
also deleted diffuse
d
functions from carbon; we denoted this basis set aug-cc-
pVDZ
0
. The quadruple-
p
MP2 computations were sped up significantly by
another basis set truncation. Instead of using the full aug-cc-pVQZ basis set
(with an enormous 1512 basis functions for the benzene dimer), we removed
g
functions for carbon and
f
functions for hydrogen, yielding a basis we
denoted aug-cc-pVQZ*.
Any time such approximations are used, it is important to evaluate the
consequences of the basis set truncation on the quality of the results. Table 2
presents optimized intermolecular distances for the configurations of the ben-
zene dimer depicted in Figure 1 using the monomer geometries of Gauss and
