Prediction of crystal lattice parameters of melilites

An integrated database system (DB) on properties of inorganic substances is developed, comprising a database on materials for electronics and a database on the properties of inorganic compounds. This integrated system is accessible online through a single-entry point. The system is used to show the potential of machine learning in identifying regularities within the data and leveraging them for the design of yet-to-be-synthesized compounds. This is demonstrated on the example of materials with a melilite structure, which are promising candidates for new nonlinear optical materials. The prediction accuracy according to cross validation for new melilites exceeds 85%, while the estimation error for their crystal lattice parameters remains below 0.21 Å. For the experimentally validated compound Sr₂Ga₂SiO₇, the predicted and measured lattice parameters show minor discrepancies: 0.05 Å.