Curvature-Guided Derivative Peak Filtering for Robust Initialization of Sequential XPS Spectral Deconvolution

Abstract

Accurate initialization of peak positions is essential for reliable X-ray Photoelectron Spectroscopy (XPS) spectral deconvolution because nonlinear peak-fitting algorithms are highly sensitive to their starting parameters. Conventional derivative-based peak detection often generates excessive false-positive candidates in noisy or overlapping spectra, reducing fitting stability and reproducibility. This work introduces Curvature-Guided Derivative Peak Filtering (CG-DPF), an interpretable human-in-the-loop framework that combines first-derivative zero-crossing detection with tunable second-derivative curvature thresholding. Unlike classical derivative spectroscopy, which primarily uses the second derivative to identify extrema, CG-DPF applies curvature as a continuously adjustable post-detection acceptance criterion, allowing analysts to balance peak recall against false-positive suppression while preserving transparent parameter selection. The contribution is a reproducible filtering strategy and initialization workflow rather than a new differentiation or smoothing algorithm. CG-DPF was evaluated using a primary benchmark comprising one in-house survey spectrum and seven published XPS survey spectra, with additional external transfer assessments on published oxide thin-film and HAXPES-related datasets. Performance was assessed using reconstruction quality, FWHM compliance, peak count, computational efficiency, and statistical comparison with adaptive multi-scale detection under identical preprocessing and fitting conditions. On the primary benchmark, CG-DPF reduced false-positive initialization from 47 to 5 candidate peaks while maintaining complete FWHM compliance. Across seven independent survey spectra, it achieved a mean Phys. Ratio of 0.66 compared with 0.30 for the Adaptive Medium baseline (Wilcoxon signed-rank test, p = 0.031). External datasets further showed that survey-default parameters do not universally transfer across different materials and acquisition conditions, highlighting the need for analyst-guided parameter tuning.CG-DPF provides an optimizer-independent preprocessing strategy that integrates with conventional Gaussian, Lorentzian, Voigt, and pseudo-Voigt fitting workflows, improving the robustness, reproducibility, and transparency of sequential XPS spectral deconvolution.