Deciphering the Key Factors Governing Mn4+ Zero-Phonon Line Characteristics via Machine Learning Decoding of Host–Mn4+ Interactions

Achieving wide color gamut in liquid crystal displays relies critically on narrow-band red-emitting phosphors. Mn⁴⁺-activated phosphors are promising candidates due to their sharp emission, yet modulating their zero-phonon line wavelengths remains challenging. This study employs machine learning to decode host–Mn⁴⁺ interactions across 65 distinct hosts (42 fluorides, 6 fluoroxides, 17 oxides). By extracting 29 structural descriptors and leveraging a random forest regression model, we identify nine key features governing ZPL wavelengths. Electronegativity-related parameters dominate (77.83% cumulative importance), while geometric factors (bond angles, distances) also contribute significantly. The model achieves high accuracy (test MAE = 4.133 nm, R² = 0.928), revealing that high electronegativity in secondary-coordination ions enhances Mn–ligand covalency, reducing the E_g → ⁴A_2g transition energy and redshifting the emission peak wavelengths. This work identifies key design principles for Mn⁴⁺-activated fluoride, oxide, and oxyfluoride phosphors, enabling a targeted strategy for discovering next-generation narrow-band red emitters.