From Atoms to Plants

H. Stein

Predicting molecular properties and adsorption energies is no longer the bottleneck in heterogeneous catalysis as modern machine learning potentials and neural networks trained on high-throughput DFT data can deliver millions of adsorption energy predictions. The real challenge lies in what comes next: translating this atomistic data into reaction mechanisms and kinetic models that are actionable at the process level. This talk presents an integrated workflow connecting high-throughput DFT and molecular dynamics simulations to automated reaction mechanism generation and microkinetic model parametrization. Elementary reaction steps are enumerated systematically from surface intermediate inventories, barriers are assigned via Brønsted–Evans–Polanyi scaling relations calibrated across large DFT datasets, and the resulting kinetic parameters are propagated into process simulators. I show how volcano-plot logic breaks down when considering CAPEX and OPEX as different materials may activate (dis)advantageous kinetic pathways. I argue that the community must now focus on closing the loop between atomistic prediction and plant-scale simulation. Siloed tools and intra-lab acceleration are insufficient; what is needed are validated, uncertainty-aware kinetic models that speak the language of chemical engineers and process simulators. Examples from our group illustrate how this pipeline has been applied to real catalytic systems and translated toward industrial scale-up. I will also briefly demonstrate what virtual catalysts are and how they may help in discovering new processes.