Big Data of Materials Science: Critical Role of the Descriptor

Statistical learning of materials properties or functions so far starts with a largely silent, nonchallenged step: the choice of the set of descriptive parameters (termed descriptor). However, when the scientific connection between the descriptor and the actuating mechanisms is unclear, the causality of the learned descriptor-property relation is uncertain. Thus, a trustful prediction of new promising materials, identification of anomalies, and scientific advancement are doubtful. We analyze this issue and define requirements for a suitable descriptor. For a classic example, the energy difference of zinc blende or wurtzite and rocksalt semiconductors, we demonstrate how a meaningful descriptor can be found systematically.

Figure 1

Experimental ground-state structures of 68 octet binary compounds, arranged according to the two-dimensional descriptor introduced by Phillips [2] and van Vechten [1]. Because of visibility reasons, only 10 materials are labeled for each structure. See the Supplemental Material for more details [6].

Figure 2

Calculated energy differences between RS and ZB structures of the 82 octet binary $AB$ materials, arranged by using the nuclear numbers $(Z_A,Z_B)$ as descriptor (top) and according to our optimal two-dimensional descriptor (bottom). In the bottom panel, seven ZB materials with predicted $\mathrm{\Delta }{E}_{AB}>0.5\mathrm{eV}$ are outside the shown window (see Supplemental Material [6]).