Abstract
Experimental and calculated values of the activation energies of
elementary steps provide insight into the mechanisms of many important
catalytic reactions. The collections of these values also present an
opportunity to apply knowledge discovery techniques in order to find
more general qualitative patterns that are implicit in the data. Here
we apply classical hierarchical clustering and recently developed
classification profiling methods to 168 steps and activation energies
relevant to carbon dioxide reforming of methane, compiled for eight
different transition metal catalysts. The core fragments of the 168
transition states are defined and used as features. These methods and
features address the basic questions of what are the similarities and
differences among catalysts, and why are they similar or different.
The consistency of the results with experimental observations suggests
that the methods are reliably predictive.
full paper