Tuning spin splittings in 2D compounds using Bayesian optimization

Bayesian analysis offers a data-centered way to search for compounds with targeted properties. We have used Bayesian inference to search for 2D compounds with large spin splittings (SS) together with large band gaps. Our starting point is a collection [Scientific Data 9, 195 (2022)] of calculated SS for all 2D compounds available in the C2DB database. This collection classifies these SS in four different prototypes: Rashba, Dresselhaus and Zeeman and High Order SS. Each prototype has been indicated using different design principles that include the symmetry of the crystal, the types of atoms, the symmetry of the k-point in reciprocal space and the shape of the dispersion relation in the calculated band structures. As we have a large number of SS available, we use the calculated data to infer what are the atoms and crystal structures that maximize the targeted property. We will show the power of this method by focusing on the Rashba SS, but it also works for other properties. The structural clusters that are more likely to have a large RSS are transition metal (TM) dichalcogenides of the Janus type and where two TM share the M sites in MX2. Compounds containing Mo, W, Sb or Bi cations are also more likely to have large RSS. With the obtained results we can elaborate on the prediction of new compounds with optimized properties, followed by ab initio calculations to confirm the results.