Although most materials expand when heated, a small minority actually contract or retain the same volume over wide temperature ranges. These unusual materials can help to overcome many of the engineering challenges caused by thermal expansion in, for example, fuel cells or optical components. We have identified a new approach for controlling thermal expansion through fluoride doping of niobium oxyfluorides. A small fluoride content increase leads to zero and then negative thermal expansion. To our knowledge this is the first instance of negative thermal expansion in an oxyfluoride.

If two materials share the same atomic species in the same positions, they will exhibit the same phyiscal properties. Extending this idea further, quantifying how similar different materials are would allow us to predict their properties, given enough data to compare. Machine learning (ML) models can be used to make these predictions, but unfortunately crystal structures are not well-suited to existing ML methods. 

Here, we have developed a "grouped representation of interatomic distances" (GRID) as a way to feed crystallographic structure to machine learning algorithms. Although simple, when combined with Earth mover's distance this representation has the ability to accurately predict physical properties of materials, and reflects chemical intuition. Importantly, GRID can be applied to both crystalline and disordered materials without modification.

Half-metallic materials have the unique electronic property of being conducting in one electron spin direction whilst insulating in the other. They have the potential to vastly improve data processing and storage technology however current half-metallic materials do not meet the standards required. Through computational studies we have identified a new approach for controlling half-metallicity through pressure and chemical composition. The niobium oxyfluoride system is predicted to switch from metallic to half-metallic under applied pressure. The electronic properties can also be controlled by changing the ratio of oxygen and fluorine.

We are hiring! If you are interested in joining the group to study local disorder in crystalline inorganic materials, please get in touch. More details can be found on FindAPhD.

James gave a talk about recent machine learning work within the group at the recent AI4SD conference, and even got captured in cartoon form by ErrantScience!

If you're an early career researcher (ECR) looking for a postdoctoral fellowship, and are interested in machine-driven materials discovery and/or synthesis please get in touch as we would love to work with you! More details available on the RAEng website; deadline 19th April.

We're delighted to welcome two new members to the group this year! Ruizhi joins us as a PDRA from Queen Mary, University of London and will be working on developing new ways of describing crystal structures for use in machine learning. Ellie joins us for her final year Chemical Physics project looking at the effect of oxide doping on the electronic properties of niobium oxyfuorides. Welcome to them both!

We're hiring a postdoctoral researcher! If you are interested in combining machine learning with crystallography in order to discover new materials, please apply for our PDRA position funded by the AI3SD network. Closing date 3rd August 2020.

https://www.vacancies.ed.ac.uk/pls/corehrrecruit/erq_jobspec_version_4.jobspec?p_id=052601

We have studied the negative thermal expansion (NTE) behaviour of oxyphosphate V₂OPO₄ over a wide temperature range using X-ray and neutron diffraction, and have discovered that the NTE arises due to an electronic charge ordering transition from a larger-volume charge-ordered phase to a smaller-volume charge-disordered phase as temperature increases. The highly anisotropic thermal expansion is dominated by the difference in size of V²⁺ and V³⁺, with NTE most apparent along the directions of face-sharing octahedral chains. This material is unusual in remaining semi-conducting throughout the charge order transition, in contrast to many other electronically-driven NTE materials.

We report the effect of iron off-stoichiometry in magnetite (Fe_3-xO₄) on the charge-ordering (Verwey) transition, and the underlying changes in crystal structure. Our detailed analysis reveals that oxidation from Fe²⁺ to Fe³⁺ preferentially occurs at one crystallographic site, while the rest of the charge-ordered structure remains intact. This new 'charge order within a charge order' phenomenom is driven by the unusual electronic trimeron distribution around the special site, causing a locally distinct ionisation potential.

To use machine learning techniques to understand (and predict) structure-property relationships within chemistry, it is first necessary to represent the chemical structure in a computer-readable way. In this perspective review, we outline some of the state-of-the-art methods used to represent structure with applications both for molecular and extended solid systems. We also suggest future areas of focus in the field in order to bridge the divide between machine learning in small molecules and crystalline solids.

We report high-temperature analysis of the local structure in magnetite, Fe₃O₄, which reveals that the low-temperature Verwey state (defined by charge ordering and trimeron bonding) persists (on a local scale) to an astonishing 850 K, the magnetic Curie temperature. This confirms that local Fe-Fe trimeron bonding arises due to magnetic ordering, far above their long-range observation (the Verwey transition).

A fully-funded (fees and stipend) PhD studentship in the discovery of functional mixed anion materials is available with the functional materials group. The project will combine experimental solid state chemistry with data-driven computational approaches to generate new transition metal oxyfluoride materials.

Further details are available, and interested candidates should contact Dr James Cumby (james.cumby@ed.ac.uk) directly.

James travelled to Aalto University, Helsinki to the third Machine Learning for Materials Science (ML4MS) workshop. This interdisciplinary workshop covered a huge range of applications of machine learning to materials science, and James was delighted to be awarded the ‘Best Talk by a Young Investigator’ prize.

• Webcast talks are available
• Thanks must go to the AI3SD network for funding.