Descriptor-based approaches are among the most promising candidates
for future materials design. Today databases for various descriptors
exist and identifying a proper one often depends on chemical
intuition. In catalysis such approaches have been successfully applied
for homogeneous systems and surface alloys. However, due to their
complex nature similar general approaches for heterogeneous
single-site catalysts have not been proposed so far.
In this work we present a descriptor-based approach to describe the
adsorption of CO and NO to Fe, Co, Ni and Cu sites in the zeolites
SSZ-13 and mordenite (MOR). Based upon the distribution of Al sites
inside the material we propose a set of 11 different active sites in
SSZ-13 and 15 active sites in MOR. We identify a set of descriptors
based upon electronic structure, stability of active sites and charge
to create a database for such sites. In a second step we use a
combination of a machine learning approach to identify the most
important descriptors for our problem. This allows us to arrive at an
R2 value of over 0.95 in describing the adsorption of CO and NO.
To the best of our knowledge this is the first general,
descriptor-based approach to predict molecular adsorption on
heterogeneous single-site catalysts. Based upon Brønsted-Evans-Polanyi
relationships this fully machine-learning based approach should also
be able to predict reaction barriers for selected reactions, a key
feature for the design of such catalysts.