Machine Learning-Assisted Parameter Dataset Establishment and Query System Design of the “POSTParamQuery” Web Application within the Preferential Occupancy Site Theory (POST) Framework: Practical Applications in Doped Luminescent Materials

Identifying the crystallographic positions of dopant ions in doped inorganic materials is a longstanding challenge that has impeded precise control over material properties. To address this, we introduce a robust theoretical framework along with detailed methodologies to evaluate dopant ion occupancy within a host lattice, taking into account the associated energetic factors. The preferential occupancy site theory (POST) provides a robust framework for predicting lattice sites of the dopant in doped inorganic materials. To facilitate the application of POST in inorganic material design, a comprehensive approach combining parameter dataset and the creation of a user-friendly web application query system named “POSTParamQuery” is presented. First, a standardized dataset of critical parameters, including atomization energy coefficients Jij for various bond types along with empirical bond valence constants (dij0 and bij0) necessary for accurate POST evaluations, was established. Comparing five distinct machine learning (ML) models, identifying XGBoost and Linear Regression model as the one with the best predictive ability, achieving an accuracy (R2) of 0.9229 and 0.9615 in predicting Jij and dij0, respectively. XGBoost and Linear Regression were utilized to predict and establish 4,698 groups of bond parameters for different types of bonds, generating a total of 14,094 standardized parameter data used by POST. Upon this foundation, an accessible web application parameter query system was established, enabling efficient retrieval and analysis of parameter data, thereby streamlining the optimization of material properties. To validate the framework, POST was applied to a series of technologically significant luminescent materials, including Y2O3:Eu3+, Zn3(PO4)2:Mn2+, Sr4Al14O25:Eu2+, Sr2MgSi2O7:Eu2+,Dy3+, Sr2Al6O11:Eu2+, ZnAl2O4:Fe3+, CaGa4O7:Mn2+, and Ca5(PO4)3Cl:Sb3+, Mn2+. Through computational analyses, specific lattice sites occupied by dopant ions such as Eu3+, Mn2+, Eu2+, Dy3+, Fe3+, Mn2+, and Sb3+ were accurately predicted. Notably, The calculated results demonstrate a high degree of consistency with previously reported experimental findings, suggesting that further experiments will likely reinforce this agreement. This outcome highlights the reliability and predictive power of the POST method. This work not only establishes a practical toolset for optimizing existing luminescent materials but also provides a theoretical foundation for the rational design of new functional materials. By integrating standardized parameters dataset, advanced computational methods, and an accessible web application query system, our study advances the field of materials science by bridging fundamental theory with practical applications.