A short opinion piece on end-to-end molecular models, graph networks, transformers and matrix completion, for pure and mixture properties, with practical guidance on when they beat UNIFAC and COSMO-RS and a route for carrying them into process simulators. [read]
reading Rittig et al. (2026), Molecular machine learning in chemical process design
2026-10-11 / reading, machine-learning, thermodynamics, process-design