Molecular modelling has become an indispensable tool for understanding and interpreting the behaviour of nanostructured materials. These materials play a central role in the fields of catalysis, sorption, sensing and optical energy conversion. However modelling realistic functional materials and making the connection with experiment is extremely challenging, due to the huge length-time scale gap between theory and experiment. A fundamental question is whether molecular modelling has reached the stage to quantitively predict experimental observables and ultimately provide design rules for discovery of new materials. Addressing this challenge requires a comprehensive understanding of the entire molecular modelling workflow, its limitations, and the opportunities it offers to bridge the gap between atomistic simulations and experiment.
The molecular modelling workflow consists of various key steps : First a reliable structural model is needed to represent the real material. Functional materials are rarely perfect but contain defects, disorder, and spatial heterogeneities that strongly influence their performance. Constructing realistic structural models is a challenge on its own and requires a close synergy with experiment. Second accurate energies are needed for every structural model. Some experimental properties, e.g. selectivities, adsorption isotherms, phase transition temperatures are highly sensitive to the accuracy of the underlying energy evaluation method. Recently Machine Learning potentials (MLPs) have emerged as a promising route towards extending accessible length and time scales by providing surrogate representations of the potential-energy surface. By using advanced learning strategies, it has now become possible to train chemically accurate MLPs opening new perspectives to bridge towards experimental observables.
A third challenge concerns the dynamic nature of materials under operating conditions. In catalytic systems, for example, active sites evolve with time on stream and depend strongly on temperature, pressure, and chemical environment. Making realistic prediction of experimental observables requires extensive dynamical sampling across a broad range of time scales.
Bridging atomic-scale precision and experimental reality requires a direct connection between molecular simulations and experimental observables. Validation against measurable quantities is essential to assess the predictive power of the underlying models. Once this bridge is established, molecular modelling can ultimately guide the design of improved materials. In this talk, I will discuss recent progress towards the predictive design of nanostructured materials. Examples will be drawn from technologically important applications, including carbon capture, solar energy conversion, and the conversion of emerging feedstocks such as CO₂ into value-added chemicals. These examples illustrate how advances at the interface of quantum mechanics, statistical physics and machine learning are bringing modeling closer to experimental reality.
Abstract