Combined Machine Learning and Experimental Study for the Zintl Phase Thermoelectric Ba _{1− <i>x</i>} Sr _<i>x</i> Zn _{2− <i>y</i>} Cd _<i>y</i> Sb ₂ System

Two new Sr and Cd cosubstituted zintl phases of Ba 0.92(2) Sr 0.08 Zn 1.30(3) Cd 0.70 Sb 2 and Ba 0.60(2) Sr 0.40 Zn 0.32(4) Cd 1.68 Sb 2 are successfully synthesized, and these compounds are crystallized in the orthorhombic BaCu 2 S 2 ‐type and the trigonal CaAl 2 Si 2 ‐type phases, respectively. Overall, crystal structures are constructed by the assembly of 1) the structural backbone of the infinite chain and 2) the Ba/Sr mixed site, filling the space within or in‐between anionic frameworks. The structural selectivity between these two title phases can be explained by the radius ratio criterion of cationic‐to‐anionic elements ( r + / r – ). Furthermore, the r + / r – criterion determining the preferred structure is closely related to the Ba/SrSb bond distance and eventually affects the Sr substitution level in both phases. A series of DFT calculations suggest that the BaCu 2 S 2 ‐type phase exhibits a heavily doped semiconductor behavior, while the CaAl 2 Si 2 ‐type phase shows semiconducting behavior. The temperature‐dependent thermoelectric (TE) property measurements prove that Ba 0.60(2) Sr 0.40 Zn 0.32(4) Cd 1.68 Sb 2 exhibits a relatively higher ZT value than Ba 0.92(2) Sr 0.08 Zn 1.30(3) Cd 0.70 Sb 2 due to its higher electrical conductivities and Seebeck coefficients. Machine learning (ML)‐based predictions successfully capture general TE trends of the title phases and show higher ZT values for the BaCu 2 S 2 ‐type phase compared to the CaAl 2 Si 2 ‐type phase.