The laboratory of high power laser optical components of Shanghai Institute of Optics and Fine Mechanics (SIOM) of the Chinese Academy of Sciences (CAS) proposed that glass structure gene modeling (GSgM) can accurately simulate the properties or design the compositions of glass with limited data. The study was published in the Journal of Non-Crystalline Solids.  

Neodymium-doped phosphate laser glass (Nd: glass) has long been studied as a gain medium in high-power laser systems because of its wide adjustable range of spectroscopic properties. The high-power and high-energy laser system requires higher performance in terms of the thermo-mechanical, spectroscopic, and anticrystallization properties of Nd: glass.  

Usually, a design methodology for a commercial laser glass is to fine tune the candidate composition within a narrow range to achieve gradual property improvements. The process is inefficient, i.e., higher costs and a prolonged development cycle. Accurate and rapid material design method is the key to shorten the research cycle and accelerate the practical process of new materials.  

The researchers used a baseline Nd:phosphate glass and investigated the effects of K₂O, MgO, and Al₂O₃ on optical properties, coefficient of thermal expansion (CTE), and T_L, which was carried out in conjunction with glass structure analysis using Fourier transform infrared (FTIR) technique.  

Through mathematical statistical analysis, the most closely related structural units were selected to establish the structure-property (S-P) and composition-structure (C-S) model. The prediction accuracy of the model for all properties was ideal, and the correlation coefficient was R² > 0.97, R_adj² > 0.95. The P value (< 0.005) indicated that the model has good convergence.  

Then, the C-S-P simulation platform (glass structure gene simulation method GSgM) was constructed by combining S-P and C-S models, which can realize the bi-directional prediction of compositions and properties. GSgM validation tests were performed by using “bi-directional” modeling steps, P→S→C and C→S→P. The former was to design specific glass genes to meet the target properties, the models satisfactorily match with the experimental results with the less than 0.4% relative error. The latter was predicting the properties of the designed glass, the relative errors in most cases were within 0.5%, except for τf close to 1.73%. 

The use of GSgM methodology moves away from the conventional C-P approach to new glass design and offers more flexibility in handling highly nonlinear glass property response to composition variations, which enable researchers to design new glasses starting from the selection of specific glass structure groups to meet a set of property requirements (or glass composition) all at once and subsequently defining the final glass composition or predicting glass properties.