Modelling and understanding the structure of graphene oxide materials with machine-learning-driven simulations

Graphene, a two-dimensional sheet of carbon atoms with a honeycomb structure, has evolved from academic interest to large-scale industrial production. Graphene itself is elementally pure, but there exists a wide range of chemically modified materials that are summarily called "graphene oxide" (GO). In GO, the graphene honeycomb is broken up by various types of hydrogen- and oxygen-based functional groups ("hydroxyl", "epoxy", and others, similar to what is found in organic molecules). The functional groups in GO are not just of fundamental interest, but they are thought to be primarily responsible for practical applications - for example, in catalysis, where they define the active centres at which one molecule is transformed into another.

Graphene itself has an ordered atomic structure, but GO much less so, because of the large structural and chemical diversity. This structure is incompletely understood until today, and this has strongly hindered the progress of fundamental research and the commercial exploitation of GO alike.

The aim of the present project is to develop and deploy a new computer simulation methodology, based on machine learning (ML) from accurate quantum-mechanical data, to understand the atomic structure of GO in a way that was not previously possible. A digital "library" of simulated structures will be systematically assembled, representing the range of possible chemical modifications in much more realistic detail than established simulation methods could afford. The project will lead to a new interatomic potential ("force field") that researchers can use to simulate GO on the atomistic scale. The project will also predict a catalogue of accurately computed spectroscopic fingerprints for the microscopic structure of GO materials: this will help to decipher the outcome of various experimental measurements that are used every day in academic and industrial laboratories. The resulting datasets will be openly distributed to maximise the project's academic impact.

The present research is theoretical and computational in nature, but it is expected to have a direct practical implication: by allowing chemists, physicists, and materials scientists to establish new connections between the atomistic structure of GO on the one hand, and technologically relevant properties on the other hand. It is therefore an example for how computational chemistry, accelerated by new machine-learning approaches, can reach a previously unavailable degree of realism in the modelling of complex functional materials.