"Single atom catalysts on amorphous supports: A wild frontier for ab initio calculations"
Several industrially important catalysts for olefin polymerization, metathesis, and epoxidation are atomically dispersed metal centers on amorphous supports. These heterogeneous catalysts elude the usual experimental and theoretical analyses because every site on an amorphous catalyst is different. Moreover, the disorder is “quenched” unlike the predictable dynamical disorder of a liquid environment. We discuss the mysterious initiation process in the Phillips ethylene polymerization catalyst (Cr/SiO2) catalyst at molecular and catalyst pellet scales. At the molecular scale, prevalent bis(ethylene)Cr(II) complexes can generate active Cr(III)-alkyl sites via a new tethered homolysis mechanism. The computationally predicted initiation times and fractions of active sites compare well to experimental estimates. At the pellet scale, the porous support pulverizes itself upon initiation and the Cr/SiO2 catalyst becomes diluted in an expanding sphere of polyethylene. Contrary to intuition, catalyst dilution leads to an effectiveness factor that increases with size of the expanding polyethylene/catalyst spheres. Thus, the observed activity increases with polymerization time may, in part, be due to early transport limitations. Finally, we introduce a new “Importance Learning” algorithm that combines machine learning and importance sampling methods for site-averaged rate calculations in ab initio studies of these catalysts. The new algorithm requires rate calculations at approximately 1000-fold fewer sites than existing strategies, based on results with the same confidence interval.