PGI-1 Talk: Hongbin Zhang,TU Darmstadt

Quantum Theory of Materials Seminar

Inverse Design of Materials

Join us at: https://iffvc.fz-juelich.de/b/nat-ham-1kc-oip

Machine learning has been widely applied to obtain statistical understanding and rational design of advanced materials by mapping out the composition-processing-(micro)-structure-property relationships. In this talk, I am going to demonstrate the concept of inverse design and to showcase how it can be carried out in three different flavours, i.e., high-throughput combinatorial computation, Bayesian optimization, and generative deep learning. Taking magnetic materials as an example, a fully-fledged high-throughput workflow has been designed and applied to screen for promising candidate materials as permanent magnets and magnetocaloric, as well as spintronic materials [1]. Such a workflow has been applied successfully on several classes of materials, such as antiperovskite [2], all-d-metal Heuslers, and carbides/nitrides with nano-laminated structures [3]. Furthermore, in order to explore the vast chemical space more efficiently, forward modelling of the Curie temperatures for ferroma gnetic materials has been carried out [4]. This provides the basis for multi-objective optimization, which will be illustrated by figuring out the two-dimensional Pareto front of magnetization and critical temperature. Interestingly, such a generic approach based on Bayesian statistics can be directly integrated with experiments, leading to adaptive design of high-entropy alloys [5] and efficient sampling of x-ray absorption spectroscopy [6].
Last but not least, how generative deep learning can be applied to predict novel stable compounds will be discussed based on our recent implementation based on the generative adversarial network [7].

References
[1] H. Zhang, Electronic Structure 3, 033001 (2021)
[2] H. Singh, et al., npj Computational Materials 7, 98 (2021)
[3] C. Shen, et al., J. Mater. Chem. A 9, 8805 (2021)
[4] T. Long, et al., Mat. Res. Lett. 9, 169 (2021)
[5] Z. Rao, et al., Science 378, 78 (2022)
[6] Y. Zhang, et al., arXiv:2203.07892
[7] T. Long, et al., Acta Mat. 231, 117898 (2022)

Contact

Prof. Dr. Yuriy Mokrousov

Phone: +49 2461 61-4423
Email: y.mokrousov@fz-juelich.de