The Paradox of Global Multireference Diagnostics

Modern computational chemistry methods are a useful tool
for modeling
many chemical systems, but they are challenged by multireference species
(e.g., transition metals). A variety of diagnostics have been formulated
to identify such cases. They are typically developed by analyzing
multireference characters of small molecules, and many provide an
average picture of the entire system. We caution the use of such diagnostics
for large systems because large systems may include parts with varying
degrees of multireference characters. Specifically, a small but highly
multireference component may yield a large error in absolute terms,
which may be masked in an average value over the entire molecule.
As the calculation of molecular relative energies often concerns errors
in absolute terms, such a false sense of safety may be detrimental.
A prospective means to tackle this challenge is to use fractional
occupation density to identify potentially problematic components
in a system and then examine this moiety with higher-level computations
on appropriately constructed smaller models.