Artificial intelligence-based screening of dielectric properties in binary ionic-doped yttrium iron garnets

Yttrium iron garnet is a critical ferrimagnetic material, with dielectric properties essential for microwave, magneto-optical, and high-frequency electronic applications. However, the complex composition-structure-property relationships in multi-ion doped systems render large-scale screening via first-principles calculations computationally prohibitive. Here, we developed a dielectric property prediction framework integrating density functional theory calculations with machine learning. We established a database of dielectric constants for single-ion and binary co-doped YIG systems and constructed six physically meaningful descriptors: ionization energy, electronegativity, polarizability, ionic radius, coordination number, and magnetic moment. A SVR model was employed to capture the nonlinear relationship between these descriptors and the dielectric constant. Model parameters were optimized using ten-fold cross-validation and grid search. The results indicate that Pb-Ti co-doping significantly enhances the dielectric constant of YIG, increasing it from approximately 18 in pristine YIG to over 45. The developed SVR model facilitates rapid and reliable screening of high-dielectric compositions while reducing computational costs, and can be extended to ternary and higher-dimensional compositional spaces for high-throughput materials design.