Spin-state energetics
of transition metal complexes remain one
of the most challenging targets for electronic structure methods.
Among single-reference wave function approaches, local correlation
approximations to coupled cluster theory, most notably the domain-based
local pair natural orbital (DLPNO) approach, hold the promise of bringing
the accuracy of coupled cluster theory with single, double, and perturbative
triple excitations, CCSD(T), to molecular systems of realistic size
with acceptable computational cost. However, recent studies on spin-state
energetics of iron-containing systems raised doubts about the ability
of the DLPNO approach to adequately and systematically approximate
energetics obtained by the reference-quality complete active space
second-order perturbation theory with coupled-cluster semicore correlation,
CASPT2/CC. Here, we revisit this problem using a diverse set of iron
complexes and examine several aspects of the application of the DLPNO
approach. We show that DLPNO-CCSD(T) can accurately reproduce both
CASPT2/CC and canonical CCSD(T) results if two basic principles are
followed. These include the consistent use of the improved iterative
(T
1
) versus the semicanonical perturbative triple corrections
and, most importantly, a simple two-point extrapolation to the PNO
space limit. The latter practically eliminates errors arising from
the default truncation of electron-pair correlation spaces and should
be viewed as standard practice in applications of the method to transition
metal spin-state energetics. Our results show that reference-quality
results can be readily achieved with DLPNO-CCSD(T) if these principles
are followed. This is important also in view of the applicability
of the method to larger single-reference systems and multinuclear
clusters, whose treatment of dynamic correlation would be challenging
for multireference-based approaches.