Statistical Learning for the estimation of Judd-Ofelt parameters: A case study of Er3+: doped tellurite glasses

Abstract The active development of optical materials and their corresponding applications requires an in-depth understanding of their optical properties for which Judd-Ofelt (JO) theory is often used. However, calculating the JO parameters ( Ω 2 , Ω 4 , Ω 6 ) require extensive experimental and theoretical work, making conventional calculation often inaccessible. In an endeavour to mitigate these difficulties, we have investigated the relationship between the JO parameters and the bulk matrix composition of a suite of Er3+ doped tellurite glasses. The matrix compositions were used to predict the three JO parameters using Support Vector Machine regression, coupled with sparse Principal Component Analysis. Preliminary results demonstrate a strong relationship between the bulk composition and the JO parameters (root-mean-squared error from 0.2053x10-20 to 0.5915x10-20) and (mean-absolute-error ranging from 0.4038x10-20 to 1.2291x10-20), suggesting that that with further consideration to the data, and additional covariates, we may, in the near future be able to characterize such materials precisely with no, or minimal experimental work.