Kinetic Consequences of Quasi-Harmonic Entropies Calculated with Machine Learning Interatomic Potentials for Microkinetic Modeling

Microporous
catalysts are ubiquitous in chemical processes including
sustainable transformations of biobased feedstocks into fuels and
fine chemicals. The mechanistic insights needed to design next-generation
microporous catalysts can be obtained with ab initio simulations coupled with microkinetic modeling, yet active site
confinement complicates an accurate determination of adsorbate entropies,
which, in turn, affect predictions of rate and equilibrium constants.
In this study, we developed a machine learning force field (MLFF)
strategy to rapidly predict temperature-dependent quasi-harmonic adsorbate
entropies in zeolite Beta, reducing the number of compute-intensive ab initio molecular dynamics calculations needed to construct
a microkinetic model. These entropies directly impacted the kinetics
of a model parallel reaction mechanism. We chose lactic acid dehydration
to acrylic acid on aluminosilicate zeolite Beta to explore the pathway
dependence of unselective product formation and initial deactivation
mechanisms using microkinetic modeling with our MLFF entropy strategy.
The resulting quasi-harmonic entropy approximations led to shifts
in steady-state coverages that impacted reaction orders and product
selectivity. At low lactic acid partial pressures, concerted monomolecular
decarbonylation is favored over Brønsted acid sites, which then
shifts at high lactic acid partial pressures to concerted bimolecular
condensations into lactic acid oligomers. Sequential pathways mediated
by adsorbed alkoxide or carbonyl intermediates have no kinetic relevance
at these conditions. These findings provide a strategy to integrate
quasi-harmonic entropies into microkinetic modeling that is scalable
with reaction temperature and applicable to a wide range of catalysts
and catalytic cycles.