High temperature superconductivity has great promise to transform society through the transmission of electricity with zero resistance, though the underlying physics is complex and difficult to predict from first principles, and the space of possible materials is large and equally complex. Machine learning methods have been successfully applied to many complex problems, and recent work has demonstrated such methods may also be viable to predict new functional materials with desirable properties, such as high-temperature superconductivity. In particular, deep learning methods have attracted attention for their ability to consider complex combinations of multiple attributes/features in a nonlinear fashion to predict structured outputs. This PhD project will explore the possibility of using deep convolutional neural networks to extract feature combinations and predict various properties related to superconductivity of materials. These tools will enable other key materials problems for sustainability and net zero to be tackled.
Specifically, the student will work closely with computer scientists, inorganic chemists, physicists, and material scientists to develop tools to predict new materials that may exhibit high-temperature superconductivity. This may involve developing models to identify new chemistries or regions of the periodic table where superconducting states may occur, and/or identifying new ways to improve superconducting properties (such as the transition temperature) in existing materials. As a part of this goal, the student will build models and descriptors to identify shared features in known materials that correlate strongly with the presence of high temperature superconductivity. These approaches have the potential to be expanded to the prediction of other key physical properties of importance for efficient energy use, such as ion transport in electrolytes for solid state batteries, thermoelectric materials for waste heat harvesting and magnetic and electronic materials for information storage.
The deep learning approaches applied will go far beyond the rather obsolete approaches deployed by physical computational science researchers thus far in the literature. This will be combined with the development of appropriate descriptors that use the teams understanding of materials chemistry and physics.
The award will pay full tuition fees and a maintenance grant for 3.5 years. The maintenance grant will be at the UKRI rate, currently £15,609.00 per annum for 2021-22, subject to possible increase . The award will pay full home tuition fees and a maintenance grant for 3.5 years. Non-UK applicants may have to contribute to the higher non-UK overseas fee.
The stipend will be paid in line with the standard UKRI rate and tuition fees covered at the UK rate.
Applications are welcomed from students with a 2:1 or higher master’s degree or equivalent in Computer Science, Chemistry, Physics, or Materials Science, particularly those with some of the skills directly relevant to the project outlined above. Successful candidates will have strong math and programming skills. An interest and/or coursework condensed matter physics is a benefit, though not required.
Applications from candidates meeting the eligibility requirements of the EPSRC are welcome – please refer to EPSRC website.
Please apply by completing the online postgraduate research application form, by clicking on the ‘Ready to apply, Apply online’ button. Please ensure you quote the following reference on your application: Reference CCPR0036 Discovery of high-temperature superconductors using Deep Learning.
Students are encouraged to undertake some teaching duties for the Department, up to a maximum of 6 hours per week in term time, for which they will receive training and be paid at the regular hourly rate (currently £15.99 per hour).
For any enquiries please contact Dr Michael Gaultois on: [email protected]
Machine learning modelling of superconducting critical temperature. arXiv:1709.02727 [cond-mat.supr-con] https://arxiv.org/abs/1709.02727
Element selection for crystalline inorganic solid discovery guided by unsupervised machine learning of experimentally explored chemistry.
Nature Communications 12, 5561 (2021); https://www.nature.com/articles/s41467-021-25343-7