Advancing understanding of composition–structure–luminescent properties in laser glass through machine learnings

Rare-earth (RE)-doped laser glasses meet urgent needs in national security and scientific fields, and their optimization has garnered extensive attention. However, the design of these laser glasses often relies excessively on trial-and-error experimentation, leading to significant costs and a lack of scientific guidance. Herein, we propose an integrated method that combines structural descriptors determined from molecular dynamics simulations, a self-constructed luminescent database, and a machine learning algorithm to establish the composition–structure–luminescent property (CSLP) relationship. Using an Nd³⁺-doped commercial silicate laser glass system as an example, the effectiveness of this approach has been demonstrated. The developed CSLP model enables highly accurate predictions of spectral properties, achieving a determination coefficient (R²) greater than 0.94, based on eight structural descriptors. The importance of different structural descriptors on spectral characteristics is ranked and thoroughly discussed, revealing an intrinsic relationship between the first and second coordination shells around RE ions and luminescent behaviors. Furthermore, the generic structural descriptors identified in the CSLP model can be extrapolated to other systems involving different network formers (e.g., silicate and phosphate) and modifier cations (e.g., Li, Na, K, Ba, and Ca). This capability facilitates the design of laser glasses tailored to specific targets, such as large emission cross-sections, extended lifetimes, or reduced quenching effects.