Research on a Peak-Deconvolution Method for Raman Spectroscopy for Molecular Structure Characterization of Coal Samples with Different Degrees of Metamorphism

BRIEF REPORTSignificance: Efficient coal utilization requires a precise understanding of its molecular structure. While Raman spectroscopy is a powerful characterization tool, the accuracy of its analysis heavily depends on the peak-fitting method, for which a systematic evaluation is currently lacking. This work innovatively compares five fitting functions across various coals and chars, establishing a more universal deconvolution strategy with goodness-of-fit R > 0.996. This approach significantly improves the accuracy and comparability of structural parameters, offering key methodological support for precise coal utilization.Introduction: The high sensitivity of Raman spectroscopy to carbon structural ordering makes it a core technique for characterizing the molecular structure of coal[6-7]. Precise characterization of coal molecular structure is crucial for revealing coal evolution mechanisms, optimizing resource assessment, and enabling clean utilization. However, significant variations in the number of fitted peaks (2–10) and functions (e.g, Gaussian, Lorentzian) adopted across different studies hinder the cross-comparison of structural parameters[33], compromising the comparability and reliability of Raman spectroscopic data in coal structural characterization. To address this critical scientific issue, this study focuses on optimizing Raman spectral peak deconvolution strategies. Using a controlled-variable approach, the influence mechanisms of the fitting function type (Gaussian, Lorentzian, Gaussian-Lorentzian mixture, Voigt, Pearson Ⅶ) and the number of fitted peaks on the interpretation of characteristic coal Raman bands are systematically decoupled. This investigation establishes optimized peak deconvolution strategies, including the most suitable fitting functions and peak numbers, tailored for different coal types. The optimal number of peaks and fitting function are determined based on the goodness-of-fit (R) and established knowledge of coal structural features. The fitting performance was systematically compared for five functions—Gaussian, Lorentzian, Gaussian-Lorentzian mixture, Voigt, and Pearson Ⅶ—for the first-order spectral region (1000–1800 cm−1) of the coal Raman spectra.　　The precise analysis of coal molecular structure serves as the scientific foundation for elucidating its geological evolution mechanisms, assessing resource potential, and achieving clean and efficient transformation[1-4]. Due to its high sensitivity to the degree of structural ordering and defect states in carbon materials, Raman spectroscopy has become a core analytical technique for characterizing the molecular structure of coal and its thermal conversion products[6-7]. This technique is not only widely applied to trace the structural ordering pathways during coalification[8] and to serve as an effective indicator of organic matter maturity[9], but also to deeply reveal the complex structural evolution of coal during thermochemical conversion processes such as pyrolysis, gasification, and combustion[10-11].　　However, the accuracy and reliability of Raman spectral interpretation are heavily constrained by the strategy employed for deconvoluting complex overlapping spectral bands. Significant discrepancies and challenges exist in current research in this regard. First, the selection of mathematical functions for fitting varies considerably, often involving Gaussian, Lorentzian, mixed Gaussian-Lorentzian, Voigt, or PearsonⅦ functions, lacking systematic comparison and selection criteria[12-14]. Second, the number of sub-peaks set for deconvoluting the first-order Raman region of coal (1000–1800 cm−1) varies widely, ranging from 2 to 10, which introduces substantial subjectivity[33]. This inconsistency in core fitting parameters—namely, the type of function and the number of peaks—directly leads to poor comparability of structural parameters (e.g., ID/IG ratio, defect density) reported across different studies, reduces the overall credibility of the data, and constitutes a major bottleneck in the broader application of Raman spectroscopy for the quantitative characterization of coal structure.　　To address this critical methodological issue, the aim of this study is to systematically optimize and standardize the peak deconvolution strategy for coal Raman spectra. Using coals of different metamorphic grades, typical high-ash coal, and a series of their pyrolysis chars as research subjects, a controlled-variable approach was adopted to quantitatively evaluate the impact of five commonly used fitting functions (Gaussian, Lorentzian, mixed Gaussian-Lorentzian, Voigt, and Pearson Ⅶ) and different numbers of fitting peaks on the accuracy of characteristic band interpretation. By comprehensively considering the goodness-of-fit (R) and the reasonableness of the derived structural parameters, this study ultimately establishes optimized peak deconvolution schemes tailored for different coal types (metamorphic coal, high-ash coal, pyrolysis char). The aim is to develop a more accurate, reliable, and consistent analytical framework for the quantitative characterization of coal structure using Raman spectroscopy.Methods: Coal samples with nine different ranks (vitrinite reflectance, Rmax = 0.59%–7.89%) were collected, including bituminous coal from Yushen, Shaanxi (XBD), anthracite from Ningxia (DWL), and semi-graphite from Hunan (BC), among others. The ash content ranged from 3.42% to 39.66% (details in Table 1). XBD coal was pyrolyzed under N2 atmosphere at 300–900℃ to produce char. Raman spectra were acquired using a Labram Aramis fully automated Raman spectrometer (Horiba Jobin Yvon). The scanning range was 400–4000 cm−1 with a 532 nm Ar+ laser source. Given the micro-scale analysis capability of Raman spectroscopy, bulk sample homogenization was not required for point measurements. Laser power and acquisition time were optimized to prevent sample carbonization or blackening during analysis. To eliminate fluorescence background interference, enhance the signal-to-noise ratio, and ensure data accuracy, baseline correction was applied to all Raman spectra[15]. No smoothing or normalization was performed to preserve signal authenticity. Quantitative structural parameters were obtained through peak deconvolution, where the number of peaks and fitting method selection depend on coal rank[16] and the interpretation of structural information embedded in the Raman spectra[18]. Considering the structural heterogeneity of coal, 6–10 random measurements were taken at different locations on each sample surface, and the average spectrum was used for analysis.Data and Results: The results are presented in the following five parts.　　(1) Symmetry-dependent adaptation mechanism of fitting functions for Raman peaks　　The intensity of Raman peaks is proportional to concentration, forming the basis for quantification[24-25]. Phase mixing leads to peak overlap and asymmetry. Drawing on pharmacopeial methods for chromatographic peak symmetry assessment, a comparative fitting analysis was conducted for the symmetric D1 peak and the tailed G peak. The L-G Sum and Voigtian functions yielded the highest goodness-of-fit (R) for the D1, G, and D2 peaks, demonstrating the best suitability across these peak types. The Gaussian function is more suitable for fitting the full width at half maximum (FWHM) of symmetric peaks, while the Lorentzian function, due to its long-tail characteristics, better adapts to asymmetric peaks (Fig. 1).　　(2) Rank-controlled quantitative deconvolution of Raman peaks and its structural driving mechanism　　The first-order region of coal Raman spectra encompasses the D1 peak (1325–1348 cm−1, A1g vibration, reflecting aromatic layer defects), D2 peak (1609–1615 cm−1, edge defects), D3 peak (1483–1554 cm−1, aliphatic structures/C–C bonds), D4 peak (1137–1221 cm−1, C–C/C–H bonds), and G peak (1570–1596 cm−1, E2g vibration, ideal sp2 stretching)[27-28]. With increasing coal rank, the D4 peak disappears, and the D2 and D3 peaks diminish, reflecting enhanced aromatization and reduced disordered structures[34]. The optimal number of fitted peaks varies significantly with rank: high-rank coals typically require 2–5 peaks[17], whereas low-rank coals necessitate fitting more peaks[35-36]. Experimental analysis reveals that for coal-measure graphite (Fig. 3a), Gaussian fitting leads to significant deviations in peak area ratios and FWHM, rendering it unsuitable; for high-rank coals (Fig. 4b), Lorentzian fitting yields deviant results. Therefore, the combined Lorentzian–Gaussian (L-G) Sum and Voigtian functions provide the best fitting quality.　　(3) Function adaptation mechanism and quantitative correction for Raman deconvolution of high-ash coals　　Previous studies show divergence in deconvolution approaches for high-ash coals, with varying numbers of fitted peaks (2–5) and function choices[37]. This study experimentally demonstrates that high-temperature phase transitions of minerals in coal can generate fluorescent backgrounds or extraneous peaks, interfering with D2 peak detection[40]. The four-peak approach yields the best goodness-of-fit for high-ash coals (R = 0.9983–0.9997), while the five-peak approach reduces R (Table 4). Significant deviations occur with two- and three-peak methods (Fig. 6), consistent with prior conclusions. Under the four-peak scheme, the Pearson Ⅶ function provides the optimal fit (Fig. 7) due to its effective description of band asymmetry. In summary, Raman quantification of high-ash coals is optimized using the four-peak approach combined with the Pearson Ⅶ function, providing an enhanced solution for carbon structural analysis under mineral interference.　　(4) Temperature-controlled multi-peak decoupling and function optimization for pyrolyzed coal char Raman spectra　　Applying a ten-peak method to deconvolute the first-order Raman region of pyrolyzed coal char significantly outperforms the four- or five-peak methods, achieving the highest goodness-of-fit (R) at all temperatures from 300 to 900℃ (Table 5). Among the various fitting functions, Voigtian and L-G Sum functions deliver the best performance (Fig. 9). Notably, R values slightly decrease with increasing temperature, attributed to fluorescent background interference from mineral phase transitions or melting affecting the Raman signal.　　(5) Controlled peak fitting and reproducibility in Raman analysis　　The number of peaks for fitting coal Raman spectra should be selected based on the degree of carbon structural ordering, determined via first- and second-derivative analysis (Fig. 10). To ensure reproducibility, peak positions and widths of the fitting functions were not fixed. Triplicate fittings were performed on each coal sample’s Raman spectrum. Function selection is strongly correlated with coal metamorphic degree: high-rank coals approach ideal graphite vibrational characteristics, while high-ash coals and pyrolyzed chars require adaptation to complex band broadening and mineral interference.