Prediction of glass structure by using multiple regression analysis

Multiple regression analyses were applied to the prediction of glass structures, such as O 1s binding energy and fraction of fourfold coordinated boron atoms, N₄. In the case of linear combination of the content of glass constituents, an acceptable prediction accuracy was obtained for O 1s binding energy, and as for N₄, however, a poor agreement was observed between the prediction and measurement. After introducing quadratic and cubic interaction terms into the regression formula, a drastic improvement was achieved in the prediction of N₄. Some regression coefficients were dependent on basicity of each glass constituent, suggesting the feasibility of prediction for the glasses containing novel constituents whose regression parameters have never been determined.