Exploring multiscale modelling combined with machine learning: from electronic structure to pore-scale dynamics in complex materials

Multiscale modelling has become a central strategy in the design and optimisation of advanced materials. In parallel, machine learning (ML) and data-driven approaches are being embedded in this hierarchy to accelerate structure–property prediction, support surrogate modelling, and enable new modes of human–computer interaction. This review examines how ML can mediate a multiscale workflow that connects electronic-structure calculations, atomistic simulations, mesoscopic models, and immersive visualisation. Beyond a general literature survey of the field, the article focuses on four exploratory case studies that exemplify this integration and open new possibilities within the multiscale approach. First, we use ML models trained on Density Functional Theory (DFT) reference data to construct ML-based interatomic potentials for classical molecular dynamics (MD) simulations. Building on these ML capabilities, we explore interactive simulations that combine molecular dynamics with virtual reality (VR), enabling users to navigate and manipulate molecular systems in real time. Next, we perform structural analysis of brine–oil interfaces to extract interfacial molecular features and link them to the interfacial properties using machine learning. Finally, we discuss how MD and ML-derived molecular properties inform Lattice Boltzmann simulations in digital rocks, enabling these mesoscopic models to support pore-scale ML surrogates for fluid dynamics in porous media.