Abstract
Identifying molecular catalysts that simultaneously achieve high selectivity, fast turnover, and robust stability remains a central challenge in homogeneous catalysis. Traditional discovery pipelines rely heavily on labor-intensive experimental screening, while existing automated computational workflows typically focus on a single performance descriptor such as enantioselectivity or rate-determining step. Here we present a next-generation, fully automated workflow for direct catalyst design that overcomes these limitations by enabling simultaneous multi-metric catalyst scoring, parallel exploration of multiple reaction pathways, and seamless integration of computational methods spanning quantum mechanical, semiempirical tight-binding, and reactive machine-learning interatomic potentials. This unified physics–AI framework allows rapid, mechanism-guided evaluation of catalytic performance at scale, substantially expanding the accessible design space and accelerating the discovery of high-performing molecular catalysts. We demonstrate the breadth and generality of this approach across three distinct homogeneous catalytic reactions: organocatalyzed asymmetric hydrogenation, Pd-catalyzed C–O cross-coupling, and ansa-metallocene-catalyzed isotactic propylene polymerization. Beyond homogeneous catalysis, the same framework extends naturally to non-catalytic molecular transformations for general reactivity optimization, underscoring its versatility as a powerful, universal tool for molecular design.