A spectrum deconvolution method based on grey relational analysis

The extensive usage of X-ray spectroscopies in studying complex material systems is not only intended to reveal underlying mechanisms that govern physical phenomena, but also used in applied studies focused on an insightdriven performance improvement of a wide range of devices. However, the traditional analysis methods for X-ray spectroscopic data are rather time-consuming and sensitive to errors in data pre-processing (e.g., normalization or background subtraction). In this study, a method based on grey relational analysis, a multi-variable statistical method, is proposed to analyze and extract information from X-ray spectroscopic data. As a showcase, the valence bands of microcrystalline silicon suboxides probed by hard X-ray photoelectron spectroscopy (HAXPES) were investigated. The results obtained by the proposed method agree well with conventionally derived composition information (e.g., curve fit of Si 2p core level of the silicon suboxides). Furthermore, the uncertainty of chemical compositions derived by the proposed method is smaller than that of traditional analysis methods (e. g., the least square fit), when artificial linear functions are introduced to simulate the errors in data preprocessing. This suggests that the proposed method is capable of providing more reliable and accurate results, especially for data containing significant noise contributions or that is subject to inconsistent data pre-processing. Since the proposed method is less experience-driven and error-prone, it offers a novel approach for automate data analysis, which is of great interest for various applications, such as studying combinatorial material ?libraries?.

Automated summary

The study proposes a method based on grey relational analysis for analyzing and extracting information from X-ray spectroscopic data, which shows smaller uncertainty in chemical compositions compared to traditional methods. It provides more reliable and accurate results, particularly for data with significant noise contributions or inconsistent data pre-processing. The method offers a novel approach for automated data analysis, which is particularly useful for studying combinatorial material "libraries".