Self-subtraction improves consistency in spectral curve fitting

Common practice in curve fitting of overlapped spectral features often imparts overextending the number of optimized parameters, resulting in increased model complexity and exacerbated sensitivity to initial conditions, that ultimately leads to inflated uncertainty in values of optimized parameters. We introduce a lightweight operator that unifies two important steps of model initialization: (1) resolution of overlapped bands that exceeds the benefits of the widely used second derivative transform, and (2) bandwidth esti-mation for overlapped features, to achieve a reliable data-driven contraction of optimization complexity and outperform similar methods in terms of speed, flexibility and ease of interpretation. Since only the spectrum at hand is used, the curve fitting process is steamlined by avoiding multivariate models and/or assumptions about the profile line shape included in the choice of a digital filter or a basis function. All statements are reinforced with illustrative theoretical models and x-ray fluorescence spectra obtained from a publicly available database. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper introduces a lightweight operator that reduces the number of parameters in curve fitting, thereby decreasing model complexity and improving stability to initial conditions. By resolving overlapping bands and estimating bandwidth for overlapped features, this method outperforms similar approaches in terms of speed, flexibility, and interpretability.