Discovery of Structure-Property Relationships of Intercalated Graphite Compounds Using Machine Learning

Here, we extract structure-property relationships in graphite intercalated compounds (GICs) using machine learning (ML). We demonstrate the effectiveness of ML to uncover these relationships in GICs, laying the groundwork for future efforts to intentionally design GICs for applications spanning electrochemical energy storage beyond Li-ion batteries, optoelectronics including optical switching, and biomedical sensors and devices (<i>Inorg. Chem. Front.</i>, 2016, 3, 452; <i>Adv. Mater.</i>, 2019 1808213).<br/><u>Example: Optical Switching</u><br/>GICs have varying colors. For example, stage I KC₈ is golden while stage III KC₂₄ is blue. If we want to engineer GICs for optical switching applications, in order to guide our experimental efforts, we must first understand what parameters enable the prediction of the color of a GIC. Graphite is a grey, layered material with an interlayer void space of 3.35 Å and, as a single layer, graphene is the strongest, most flexible, most conductive material at room temperature with a carrier concentration of 10¹⁸ to 10¹⁹ cm^-3 in-plane and 97% broadband transparency.(<i>Adv. Phys.</i>, 2002, 51, 1-186) As guest species insert into the interlayer space between graphite sheets, the space can expand or contract (ranging from 1 Å to over 30 Å), thereby altering the physical properties of the resulting graphite intercalation compound (GIC). Moreover, the graphite carrier concentration can be altered as carriers (i.e. electrons or holes) transfer to the guest species (i.e. acceptor or donor, respectively) with concentrations spanning 10²¹ to 10²² cm^-3 (<i>Adv. Phys.</i>, 2002, 51, 1-186; <i>Science</i>, 1976,192, 1126–1127) thereby altering the electrochemical properties. Using a fine-tree ML model, we elucidated a relationship between the guest species, the interlayer spacing of the GIC, the donor/acceptor nature of the guest, and the stage of the compound, enabling the prediction of the color of a GIC with 80.6 % accuracy.<br/>This research demonstrates the ability to employ ML in order to elucidate structure-property relationships in materials, including 3D GICs. ML also has the potential to guide 2D and few-layer intercalation as we continue exploring differences in intercalation in the 2D realm. For example, the ability to alter the 97% broadband transparency of graphene to become a colored material is intriguing applications for optoelectronics. More importantly, this research opens avenues to intentionally design materials for specific applications, thereby expediting experimental testing.