Data-driven discovery of polyoxometalate catalysts for selective ethanol conversion to ethyl acetate.

Abstract

The one-pot oxidation-esterification of ethanol to ethyl acetate (EA) provides a high-atom-economy route for converting bioethanol into a valuable product. Nonetheless, developing effective catalysts is challenging, as balancing EA selectivity with ethanol conversion is complex. Polyoxometalates (POMs) are promising multifaceted catalysts, but their vast chemical space makes identifying optimal catalysts through trial-and-error prohibitively difficult. Herein, we present a strategy that integrates machine learning (ML) and theoretical calculations to accelerate the discovery and understanding of POM catalysts. A dataset of 110 POMs is compiled, and an SVR-XGBoost stacking model is trained to predict POM activity. The model predicts ethanol conversions of up to 96.5% for KPW11Co and EA selectivity of 67.2% for KPW11Ni, which identifies them as superior candidates. Subsequent experiments validated these predictions. In-situ investigations and theoretical calculations reveal distinct reaction pathways for KPW11Co and KPW11Ni. The Co2+ center (3d7) facilitates the activation of O2via an oxidative pathway mediated by ·O2-. The Ni2+ center (3d8) primarily acts as a Lewis acid, coordinating with ethanol to promote direct dehydrogenation. KPW11Ni's distinct reactive pathway makes it exceptional for bioethanol conversion, achieving EA selectivity of up to 80.1%. This research develops a straightforward and reliable prediction framework that integrates descriptor-based ML screening with mechanistic verification, supporting the effective development of POM catalysts and the high-value conversion of alcohols.