Identification of precursor origins of synthetic graphite anodes in lithium-ion batteries using LDI-MS and multivariate analysis

Abstract

Synthetic graphites have been widely used in industrial applications, including as anodes in lithium-ion batteries. Because they are produced at temperatures above 3000 °C, which generate highly ordered graphitic domains, there is typically no discernible evidence of their precursor materials. In this study, three types of graphite, coal tar based anisotropic graphite, petroleum fluid oil based anisotropic graphite, and coal tar based isotropic graphite, were prepared. Conventional characterization techniques such as X-ray diffraction, Raman spectroscopy, transmission electron microscopy, and even electrochemical performance were unable to distinguish their precursors. Therefore, we introduced laser desorption/ionization time-of-flight mass spectrometry (LDI-MS) combined with multivariate statistical analysis to characterize three graphites prepared from different synthetic precursors as well as two commercial graphites. The resulting LDI-MS spectra were analyzed using principal component, hierarchical cluster, and heatmap analyses, which are widely used in clinical mass spectrometric diagnostics. Notably, LDI-MS coupled with multivariate statistics successfully classified the graphites depending on their precursor materials and processing parameters, such as heat-treatment temperature, whereas conventional analytical tools failed to reveal these differences. These results clearly demonstrate the strong potential of LDI-MS and statistical analysis for the precise characterization of carbon materials and for distinguishing their origins and processing routes.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s40580-026-00551-3.

Introduction

Various precursors including biomass, coal tar (CT), petroleum fluid oil (PFO), and synthetic polymers have been utilized to synthesize carbonaceous materials [1–3]. Among them, CT contains a higher aromatic content and a greater degree of polycondensation [3, 4] while PFO has a lower aromaticity, resulting in a lower coking value and less favorable graphitization properties [5]. The differences in molecular structure between CT and PFO are crucial in determining the quality and performance of the carbonized products [6]. Upon heat treatment at temperatures exceeding 1200 °C, the resulting products typically consist predominantly of carbon, with less than 10 wt% of residual elements. Following extensive heat treatment above 2000 °C, nearly all non-carbon elements are volatilized and removed, leaving behind a structure composed of over 95 wt% carbon. This high-temperature purification promotes the healing of structural defects such as pentagonal and heptagonal rings within the carbon lattice, resulting in well-aligned graphitic layers that exhibit ABAB hexagonal stacking [7–9]. According to the classic carbon development model proposed by Oberlin, fundamental aromatic units act as building blocks that grow into larger graphitic domains [10]. Higher thermal energy promotes the cross-linking of carbon crystallites and their assembly into extended graphitic structures. Although the final structure of these carbon materials is significantly influenced by the precursor type, it remains challenging to determine the original source material after carbonization. This transformation can occur regardless of the precursor with the intermediate structures, which are stabilized into polyaromatic hydrocarbons (PAHCs), for withstanding high-temperature treatment [11, 12]. Consequently, if a well-designed processing protocol is applied to suitable precursors, it is possible to achieve highly ordered carbon materials with excellent crystallinity regardless of their precursors.

Therefore, synthetic graphite produced at temperatures above 3000 °C, which typically exhibits highly ordered graphitic domains, often reveals no discernible evidence of its precursor material [13]. In detail, at temperatures up to 3000 °C, both CT and PFO undergo significant structural transformations. Initially, CT contains a higher aromatic content and a greater degree of polycondensation compared to PFO, leading to differences in their mesophase formation and graphitization behavior. However, as the heat treatment temperature increases, both materials experience a reduction in interlayer spacing (d₀₀₂) and an increase in crystallite thickness (Lc) and lateral size (La), indicating enhanced graphitic ordering. At temperatures around 3000 °C, the structural differences between CT- and PFO-derived graphites diminish, resulting in similar graphitic structures characterized by well-ordered graphene layers and reduced defects [14, 15]. As a result, when handling a graphite sample, it is often impossible to determine whether it is natural or synthetic, what processing route was used in its production, or what further treatments may be necessary to tailor its properties.

In this regard, it is of utmost importance to develop an analytical method to precisely distinguish the graphitized materials based on their precursors and/or processing routes. Conventional analytical methods such as X-ray diffraction (XRD), Raman spectroscopy, and transmission electron microscopy (TEM) have successfully provided the structural information of graphites mainly about their crystallinity including Lc, La, and interlayer spacing (d₀₀₂) [15–17]. In addition to crystallinity, the chemical structure of precursors and their graphitized materials has been investigated using various analytical tools such as Fourier transform infrared (FT-IR), nuclear magnetic resonance (NMR), X-ray photoelectron spectroscopy (XPS), thermogravimetric analysis (TGA), and elemental analyzer (EA) [18, 19]. Even though those analytical methods have succeeded to provide structural information of graphites, they are not sufficient to fully assess the origin of graphitic materials in terms of their precursors and processing parameters owing to their subtle difference, overlapped signals and the lack of sensitivity.

Previously, our group have demonstrated that laser desorption/ionization time-of-flight mass spectrometry (LDI-MS) is a powerful analytical technique for carbon materials including PFO-based graphene-like films [20], carbon fibers [21], and polyethylene-derived carbon and graphitic materials [22]. LDI-MS provides critical structural information based on their mass spectra showing diverse fragmentation patterns depending on their degree of carbonization induced by thermal treatment conditions [20–22]. These characteristics make LDI-MS apart from other conventional analytical tools, and it is particularly useful to analyze the early stage of carbonization accompanied with mesophase formation and conversion into small graphitic domains [20–22]. However, the previous studies are based on the analysis of individual LDI-MS spectrum from carbon materials [23], and this kind of approaches cannot fully harness the potential of LDI-MS techniques such as rapid analytical process, high-throughput analysis, and abundant data from individual mass spectrum [24]. Recently, LDI-MS has been actively combined with multivariate statistics such as hierarchical cluster analysis (HCA), principal component analysis (PCA), and machine learning for the comprehensive analysis of LDI-MS spectra in terms of their pattern recognition and dimensionality reduction, which are useful in LDI-MS analysis of clinical samples for disease diagnostics [25].

Inspired by this feature, we utilized LDI-MS and multivariate statistics to characterize diverse graphites prepared from different synthetic precursors with different processing routes. The graphites were prepared from isotropic and anisotropic CT and PFO by standard graphitization process and fully characterized with conventional analytical tools such as XRD, Raman spectroscopy, and TEM etc. The characterization results were comprehensively analyzed to reveal difference of the prepared graphites, and the graphites were utilized as anode in lithium-ion batteries (LIB) to evaluate their practical applicability depending on the precursor types. Finally, the graphites were characterized by using LDI-MS, and their obtained LDI-MS spectra were statistically analyzed with PCA, HCA, and heatmap analysis, which are extensively utilized in mass spectrometric analysis for clinical disease diagnosis [25]. Notably, the LDI-MS and multivariate statistics have successfully classified the graphites based on their precursor and processing parameter such as heat treatment temperature whereas the conventional analytical tools failed to find their difference. These results clearly indicated the strong potential of LDI-MS and statistical analysis for precise characterization of carbon materials to distinguish their origin and processing routes.

Experimental section

Sample preparation

The CT used in this study was obtained from a steelmaking company (POSCO, Republic of Korea), and the PFO precursor was supplied by a petrochemical company (LG Chem, Republic of Korea). The CT feedstock was refined into soft pitch through a distillation process to remove volatile matter (VM), followed by extraction to separate primary quinoline-insoluble (QI) components. During this refining step, the amount of secondary QI formed was carefully controlled to produce two types of pitch: an isotropic pitch (RCT_iso) and an anisotropic pitch (RCT_aniso). QI plays a crucial role in governing mesophase evolution during thermal treatment, as it inhibits the growth and coalescence of mesophase domains. Consequently, a relatively higher QI level promotes isotropic structural characteristics, whereas a lower QI level facilitates the development of anisotropic mesophase structures. For the PFO precursor, it was refined solely by distillation to remove VM, because PFO is characterized by the absence of primary QI. The refined PFO precursor is referred to as RPFO_aniso. The three refined soft pitch samples were subsequently coked at 480 ℃ for 7 h under 3.5 bar of nitrogen to obtain solid semicoke samples. The semicoke products were then carbonized at 1350 °C under atmospheric-pressure nitrogen to yield calcined coke (CM Furnaces, USA). The resulting calcined coke samples (CCT_iso, CCT_aniso, and CPFO_aniso) were collected in powder form under the size of 74 μm. Powdered calcined coke samples were further graphitized at 3000 ℃ using an induction heating furnace (Linn High Therm, Germany). Through this process, one isotropic graphite (GCT_iso) and two anisotropic graphites (GCT_aniso and GPFO_aniso) were finally obtained.

Characterization

To analyze the chemical composition of the starting materials, CT and PFO, proximate analysis, QI content analysis, gas chromatography–mass spectrometry (GC–MS), ¹H-NMR, and CHNS elemental analysis were conducted. Proximate analysis was performed to determine the VM, fixed carbon (FC), and ash contents of each precursor using a TGA701 analyzer (LECO, USA), following the ASTM D7582 standard method. QI content was measured in accordance with the procedure specified in JIS K2425. The same analytical procedures including proximate analysis, QI content measurement, and CHNS elemental analysis were applied to the three refined precursors (RCT_iso, RCT_aniso, and RPFO_aniso) to compare their compositional changes with the original CT and PFO. For GC–MS analysis, each precursor was diluted tenfold in tetrahydrofuran (THF), and 10 µL of the diluted solution was injected into 11 g of toluene for measurement. This analysis was conducted to identify the differences in chemical structures between CT and PFO. In addition, ¹H-NMR spectra were recorded using a ProPulse 600 MHz NMR system (Agilent Technologies) to determine the aromaticity of each precursor. CHNS elemental analysis was performed using a FlashSmart analyzer (Thermo Fisher Scientific). BBOT, L-cystine, and sulfanilamide were used as standard reference materials, and the amounts of C, H, N, and S were quantified based on thermal conductivity detection after combustion and chromatographic separation.

For the calcined coke samples, yield determination, polarized optical microscopy (POM), proximate analysis, true density measurement, XRD, and Raman spectroscopy were performed. The carbonization yield was calculated from the weight ratio of calcined coke to the initial precursor after coking and carbonization. For microstructural observation, calcined coke samples were cut into pieces larger than 1 × 1 cm², impregnated with epoxy, and polished prior to examination using a polarized optical microscope (Leica, Germany). Proximate analysis was performed following ASTM D7582 as previously described. True density was measured using an AccuPyc 1330 gas pycnometer (Micromeritics, USA) according to ASTM D3179.

XRD measurements were conducted using a D-max 2500 VPC diffractometer (Rigaku, Japan) equipped with Cu Kα radiation (λ = 1.5406 Å). Scans were recorded from 10° to 90° at a rate of 0.5° min^–1. Pure silicon (Kojundo Chemical, Japan) was used as an internal standard to refine peak positions. From the (002) reflection near 26°, the interlayer spacing (d₀₀₂) and degree of graphitization (G.D.) were calculated, and the Lc value was determined using the Bragg’s law, Franklin equation, and Scherrer equation [26]. Raman spectra were obtained using a RAMANtouch system (Nanophoton, Japan) over a spectral range of 500–4000 cm^–1, and the La value was determined from the Tuinstra–Koenig relation [27]. For the graphitized products, XRD and Raman analyses were conducted in the same manner as for the calcined coke samples to evaluate differences in crystallinity. In addition, TEM was employed to directly observe the graphitic layer structures. TEM measurements were performed using an ARM200 instrument (JEOL, Japan).

Electrical performance analysis

The three graphite samples (GCT_iso, GCT_aniso, and GPFO_aniso) were pulverized using a jet mill to prepare them as anode active materials for LIB. The jet-milling process was conducted to achieve a target particle size of 10 μm. The milled graphite powders were then used to prepare electrode slurries by mixing with Super-C as a conductive additive and a binder system composed of SBR and CMC, with the weight ratio of active material, conductive additive, and binder fixed at 96:1:3. The slurry was coated onto Cu current collectors with a thickness of approximately 18 μm. The coated electrodes were subsequently pressed using a roll-press and punched into 14-mm-diameter circular electrodes to fabricate CR2032-type coin half-cells. For cell assembly, a 1 M LiPF₆ electrolyte was used, with a solvent mixture of ethylene carbonate (EC) and ethyl methyl carbonate (EMC) at a 3:7 volume ratio. The fabricated cells were electrochemically characterized using a battery cycler (MIHW-200, NEWARE, China) in a half-cell configuration. Galvanostatic charge–discharge cycling was conducted at a C-rate of 0.1 C, within the voltage window set between 0.005 V and 3.0 V. The initial Coulombic efficiency (ICE) was calculated based on the capacity results from the first discharge and the first charge processes. The specific capacity was determined from the third discharge cycle.

LDI-MS and statistical analysis

The LDI-MS analysis was conducted using an IDSysRT (ASTA, Suwon-si, Republic of Korea) equipped with a 343 nm Nb:YAG (1 kHz) laser in positive ionization mode. The acceleration voltage was 15 kV with laser fluences of 337.83, 440.25, 555.92, 684.84, and 827.00 J/cm² which were calculated from the measured pulse energy and beam spot size. A total of 1000 laser shots were collected for each sample at 100 shots. Every LDI-MS spectrum was obtained by averaging 100 LDI-MS spectra with the custom software provided by ASTA. Samples were prepared by mounting graphite powders on a conductive tape. The MALDIquant R package was utilized for peak binning, smoothing, alignment, and normalization to establish matrices of peak positions and intensities from the LDI-MS spectra. The heatmap analysis, PCA, and HCA were performed on R version 4.2.0 using ComplexHeatmap, factoextra, and apcluster.

Results and discussion

Characteristics of raw and refined precursors

Molecular structure and aromaticity of raw precursors

In this study, CT and PFO were selected as raw precursors for artificial graphite. It is well-known that CT, a by-product of the steel-making coke-oven process, exhibits high aromaticity and high carbon content [28]. In contrast, PFO, a residue from the naphtha cracking process, contains PAHCs but shows lower aromaticity than CT. It is also rich in alkyl substituents, resulting in more aliphatic chain structures. We employed GC–MS and ¹H-NMR analyses to directly compare CT and PFO used in this study to verify whether their characteristics are consistent with those reported in the literature [29, 30]. The results of molecular structure clarified by GC–MS and ¹H-NMR analyses are presented in Fig. 1a and b. As shown in Fig. 1a, CT exhibits peaks with long retention times corresponding to high-molecular-weight aromatic compounds; nitrogen-containing aromatics such as quinoline, indole, and ⁹H-carbazole were also detected. In contrast, PFO presents peaks at short retention times, dominated by low-molecular-weight species and essentially free of heteroatoms (Fig. S1). In Fig. 1b, it is shown that CT presents a strong and broad signal in the aromatic region (δ ≈ 6.5–9 ppm) with about 65.6% aromatic hydrogen, whereas PFO contains only about 42.3%. These results indicate that CT consists of condensed aromatic ring systems, while PFO is dominated by aliphatic rings, reflecting a lower degree of aromatic condensation. The obtained results are well-consistent with previous reports that CT contains heteroatoms such as nitrogen, unlike PFO, and exhibits higher aromaticity [31, 32]. EA further confirmed that CT includes heteroatoms such as nitrogen and sulfur, while PFO was free of these elements (Table 1). In addition, the nitrogen and sulfur contained in CT remained after distillation-based refining, indicating that these heteroatoms cannot be eliminated through such a process. Moreover, it has been reported that the high aromaticity of CT contributes to a higher carbonization yield with less volatile release during heat treatment [33, 34]. To directly validate these aspects, proximate analysis was conducted. In addition, ash content and QI fraction, which can additionally influence the crystallite growth of carbon, were also measured.

Fig. 1.

Characteristics of CT, PFO, RCT_iso, RCT_aniso, and RPFO_aniso; a GC–MS total ion chromatogram b and ¹H-NMR of CT and PFO. c results of proximate analysis and quinoline insoluble amount of CT, PFO and refined precursors

Table 1.

Results of elemental analysis of CT, PFO, RCT_iso, RCT_aniso, and RPFO_aniso

Sample	C (at%)	H (at%)	N (at%)	S (at%)	C/H (wt%/wt%)
CT	89.7	5.08	0.53	0.41	1.47
PFO	90.7	7.36	ND	ND	1.03
RCT_iso	92.4	6.3	1.22	0.75	1.22
RCT_aniso	92.8	4.8	0.99	0.35	1.62
RPFO_aniso	92.4	7.7	ND	ND	1.00

Proximate analysis and QI content of raw and refined precursors

Figure 1c summarizes the proximate analysis and QI contents of both raw and refined precursors. Raw CT exhibits relatively low VM and high FC, with ash content that is nearly absent (∼0.1%). Raw PFO shows higher VM and lower FC, leading to a reduced carbonization yield, and contains ash at undetectable levels. The QI content of CT is about 7.1%, reflecting the presence of high-molecular-weight aromatic compounds, whereas QI is absent in PFO. After refining, RCT_iso and RCT_aniso maintain VM and FC contents similar to those of raw CT, with ash levels of about 0.3% and below detection limits, respectively. Distillation increases QI to 10.7% in RCT_iso, while extraction completely removes QI in RCT_aniso. This indicates that distillation enriches non-volatile residues containing QI, whereas extraction effectively eliminates QI-bearing insoluble species. RPFO_aniso, prepared by distillation of PFO, exhibits reduced VM and increased FC compared to raw PFO, partially alleviating its inherently low carbonization yield. Nevertheless, QI is not formed, underscoring the difficulty of obtaining isotropic precursors from PFO. The above results, consistent with previous reports, confirm the characteristic differences between CT and PFO [35]. High aromaticity and QI content of CT allow flexible tailoring into isotropic or anisotropic precursors and facilitate graphitization, although puffing issues may arise. PFO, by contrast, contains many aliphatic components and lacks QI, resulting in a lower FC content; however, its low ash content and absence of heteroatoms reduce the risk of defect generation. During refining, CT-based precursors experience minimal mass loss and only small compositional changes because of their high-molecular-weight constituents. PFO-based precursors, on the other hand, show a significant increase in FC owing to the removal of light fractions. These differences lead to distinct FC values (e.g., 32.1, 29.8, and 18.5% for RCT_iso, RCT_aniso, and RPFO_aniso, respectively) and subsequent heat-treatment behavior. The molecular and structural characteristics identified in Fig. 1a–c provide the basis for interpreting the graphitization behavior and electrochemical performance discussed in the following sections.

The EA of the raw and refined precursors is summarized in Table 1. The raw CT exhibited carbon and hydrogen contents of 89.7 and 5.08 at%, respectively, with detectable amounts of nitrogen (0.53 at%) and sulfur (0.41 at%). In contrast, the raw PFO showed a slightly higher carbon content (90.7 at%) and significantly higher hydrogen content (7.36 at%), resulting in a lower C/H ratio (1.03) than CT (1.47). These differences clearly reflect the more aliphatic nature of PFO and the more aromatic, heteroatom-containing composition of CT. This interpretation is consistent with the GC–MS results (Fig. 1a and S1), where a higher abundance of peaks corresponding to aromatic compounds is observed for CT compared to PFO. In addition, ¹H-NMR analysis further confirms that CT exhibits a higher degree of aromaticity than PFO, thereby supporting the compositional differences described above. After refining process, the CT-derived precursors (RCT_iso and RCT_aniso) exhibited increased carbon contents of 92.4–92.8% and reduced hydrogen contents, leading to higher C/H ratios compared with the raw CT. Notably, RCT_aniso showed the highest C/H ratio (1.62), which is consistent with its extraction-based refinement that effectively removes low-molecular-weight and hydrogen-rich species. In contrast, RCT_iso, in which QI was retained during the refining process due to the absence of an extraction step, exhibited a slightly lower C/H ratio (1.22) and higher nitrogen and sulfur contents, reflecting the concentration of QI-derived aromatic macromolecules. RPFO_aniso, refined solely by distillation, displayed carbon content comparable to RCT_iso (92.4%) but retained the intrinsically high hydrogen content characteristic of PFO, resulting in the lowest C/H ratio among the refined precursors (1.00). Importantly, nitrogen and sulfur remained below the detection limit, indicating that PFO maintains extremely low heteroatom content even after refinement. Overall, these results demonstrate that the refinement process enhanced the aromaticity (increased C/H ratio) of the CT-derived precursors while reducing heteroatom contents, particularly for RCT_aniso. In contrast, RPFO_aniso retained its aliphatic character but exhibited the cleanest chemical composition with no detectable nitrogen or sulfur. These precursor-specific differences in aromaticity and heteroatom levels are expected to influence their graphitization behavior and the resulting structural ordering of the graphitized products.

Carbonized and graphitized the refined precursors

RCT_iso, RCT_aniso, and RPFO_aniso, first underwent a coking process with 3.5 bar at 480 °C, followed by carbonization at 1350 °C. The carbonized products were also labeled as CCT_iso, CCT_aniso, and CPFO_aniso, and their carbonization yields were 75.6, 72.7, and 25.6%, respectively (Fig. 2a). Even though the FC content of RPFO_aniso was 18.5%, its carbonization yield was 25.6% relatively low compared to RCT-based precursors. It should be noted that the fixed carbon (FC) value is determined by proximate analysis under standardized ASTM conditions at atmospheric pressure, whereas the actual coking (carbonization) process in this study was conducted under pressurized conditions. Such differences in thermal conditions and pressure can influence volatile evolution and carbon retention, leading to a discrepancy between the FC value and the measured carbonization yield. This indicates that the carbonization of VM in RPFO_aniso is limited due to its low content of PAHCs. Figure 2a shows polarized optical micrographs of the carbonized products, confirming that structural differences between isotropic and anisotropic precursors are evident after carbonization. CCT_aniso and CPFO_aniso samples exhibit well-developed anisotropic optical textures with bright and colorful domains, indicating the presence of oriented mesophase structures with varying sizes and orientations. These domains appear as flow-like or mosaic textures. In contrast, CCT_iso shows a dark and featureless appearance without any discernible domain structures, which is characteristic of an isotropic morphology. However, chemical compositions of the samples, including contents of VM, fixed carbon and ash, do not differ significantly, except for a small amount of ash observed in CCT_iso (Fig. 2b–d). Based on their XRD and Raman spectroscopy analyses, it is evident that CCT_aniso and CPFO_aniso have more developed carbon structures than CCT_iso, which results in higher densities for the anisotropic precursor-based carbons (2.13 and 2.12 g/cm³) than isotropic precursor-based carbon (2.04 g/cm³). No significant difference is observed between CCT_aniso and CPFO_aniso, suggesting that, despite distinct precursor types, the given heat treatment produced carbonaceous materials with analogous structural properties. Note that silicon peaks were observed in the XRD patterns due to the addition of standard powder mixed with carbonized samples for calibration purposes. In addition, based on the results summarized in Table 2, the crystallinity of the three calcined coke samples was evaluated. For CCT_iso, the interlayer spacing of (002) plane was larger than the reference value of 0.3440 nm suggested by the Franklin equation [36]; thus, G.D. could not be calculated. In contrast, CCT_aniso and CPFO_aniso exhibited G.D. values of 22.9 and 21.7%, respectively. These results indicate that the anisotropic precursors undergo more efficient crystallite growth during heat treatment than the isotropic precursor. This interpretation was further supported by additional experimental results, as presented in Fig. S2. Moreover, the Raman analysis revealed that CCT_iso exhibited a higher I_D/I_G ratio than the other two samples, corresponding to a smaller La and reflecting its relatively low structural ordering.

Fig. 2.

a Optical micrographs with carbonization yield and the corresponding polarized optical micrographs of the coke samples (CCT_iso, CCT_aniso, and CPFO_aniso). b results of proximate analysis and density. c XRD diffractions and d Raman spectra of the coke samples

Table 2.

The G.D. and structural characteristics of the carbonized samples (CCT_iso, CCT_aniso, and CPFO_aniso) based on XRD diffractograms and Raman spectra

Sample	XRD		Raman
	G.D. (%)	Lc (nm)	ID/IG (-)	La (nm)
CCT_iso	ND	ND	0.9597	5.17
CCT_aniso	22.9	29.3	0.7966	6.23
CPFO_aniso	21.7	25.7	0.8296	5.98

Three samples, GCT_iso, GCT_aniso, and GPFO_aniso, were prepared by graphitization at 3000 °C, and their graphitic structures were also analyzed using XRD and Raman spectroscopy. The XRD patterns in Fig. 3a indicate that all samples clearly exhibit characteristic carbon crystalline planes including (002), (100), (101), (004), and (110), observed at approximately 26.5°, 42.4°, 44.5°, 54.6°, and 76.0°, respectively. Based on the (002) plane, the G.D. and Lc were calculated for each sample. GCT_iso showed a G.D. 90% and a Lc of 32.1 nm, whereas both anisotropic precursor-based graphitized samples (GCT_aniso and GPFO_aniso) showed higher values. An interesting observation is that GCT_aniso and GPFO_aniso samples have analogous structural characteristics. Raman spectra in Fig. 3b show strong G and weak D bands at around 1580 and 1350 cm⁻¹, a typical feature of graphitic materials. Using the intensity ratio of G and D bands, their La values were calculated to be 21.7, 30.3, and 29.0 nm for GCT_iso, GCT_aniso, and GPFO_aniso, respectively. This tendency indicated that the anisotropic precursor-based graphitized samples presented a comparable crystalline quality which was higher than GCT_iso. These characteristics are summarized in Table 3. The development of graphitic layers was also visualized in the TEM micrographs shown in Fig. 3c. As expected, the GCT_aniso and GPFO_aniso samples display more well-developed graphitic layers compared to the GCT_iso sample. Again, no significant difference is observed between GCT_aniso and GPFO_aniso when common analytical techniques such as XRD, Raman spectroscopy, and TEM are applied, indicating that even high-temperature graphitization does not differentiate samples although the anisotropic pitches originated from different precursors.

Fig. 3.

a XRD diffractograms, b Raman spectra, and c TEM micrographs of the graphitized samples (GCT_iso, GCT_aniso, and GPFO_aniso)

Table 3.

The G.D. and structural characteristics of the graphitized samples (GCT_iso, GCT_aniso, and GPFO_aniso) based on XRD diffractograms and Raman spectra

Sample	XRD		Raman
	G.D. (%)	Lc (nm)	ID/IG (- )	La (nm)
GCT_iso	90.0	32.1	0.2282	21.74
GCT_aniso	95.7	45.4	0.1635	30.34
GPFO_aniso	95.7	46.2	0.1711	29.00

Electrochemical performance of the synthesized graphite anodes

The electrochemical performance of the three synthesized graphite samples was evaluated as anodes for LIB. Half-cell tests were conducted using a coin-cell configuration, and the detailed fabrication procedure is illustrated in Fig. 4a. The electrode loading mass and density are summarized in Table S1, confirming that all electrodes were fabricated with comparable loading and density. Galvanostatic charge–discharge cycling was performed at a constant current of 0.1 C for three cycles. The ICE was defined as the ratio of the discharge capacity to the charge capacity obtained during the first cycle. The specific capacity values were taken from the discharge capacities recorded in the third cycle. The first-cycle charge–discharge profiles of the GCT_iso, GCT_aniso, and GPFO_aniso are shown in Fig. 4b–d. All samples exhibited open-circuit voltages (OCVs) close to 3.0 V, which confirms proper assembly of the half-cells. In case of the initial ICE, GCT_iso, GCT_aniso, and GPFO_aniso showed 90.4, 92.6, and 93.2%, respectively. Furthermore, the third-cycle charge–discharge results shown in Fig. 4e–g revealed discharge capacities of 308.5, 346.3, and 347.2 mAh/g for GCT_iso, GCT_aniso, and GPFO_aniso, respectively. Despite being derived from the equal CT precursor, GCT_iso and GCT_aniso exhibited markedly different initial efficiencies and discharge capacities, which can be attributed to differences in texture and corresponding physical properties. In stark contrast, GCT_aniso and GPFO_aniso, which share comparable structural features, displayed similar electrochemical behaviors. These results suggest that anisotropic graphite is more suitable than isotropic graphite for LIB anodes. Considering that commercial LIB anodes are typically based on needle coke–derived anisotropic graphite, the present findings are consistent with practical trends in industry. However, from a reverse-engineering perspective, the results also imply that it is challenging for end-users of artificial graphite anodes to identify the precursor origin based solely on physical and electrochemical properties. As confirmed in the previous section, isotropic and anisotropic graphite powders can be differentiated by their crystallographic features (e.g., XRD and Raman spectroscopy) and electrochemical performance, even without direct observation of bulk textures. Nevertheless, such results provide no information regarding the origin of precursors. Therefore, the development of new analytical techniques capable of classifying graphite powders based on their origin is the utmost issue in the graphite industry.

Fig. 5.

LDI-MS spectra of a GCT_aniso, b GCT_iso, and c GPFO_aniso obtained with different laser fluences

Fig 4. Electrochemical performance of GCT_iso, GCT_aniso, and GPFO_aniso in coin-typed LIB half cell as anode materials; a Assembly process of coin-typed LIB half-cell, b−d 1^st charge−discharge curves, and e−g 3^rd charge−discharge curves. ICE and capacity were determined from 1^st charge−discharge curves and 3^rd charge−discharge curves, respectively.

LDI-MS spectral characteristics of graphitized materials

According to the above sections, it is clear that the distinction of graphite materials based on their precursor is a formidable challenge. The conventional characterization tools for crystalline structure provided analogous analytical results for GCT_aniso and GPFO_aniso such as G. D., Lc, La, and I_D/I_G values. In addition, their performance as a LIB anode was also comparable. Therefore, distinguishing graphite materials solely by analyzing their raw material properties or LIB performance is highly complicated. In this regard, LDI-MS offers a promising analytical approach for differentiating graphite materials due to its high resolution and the abundance of data for multivariate statistical analysis. To explore the potential of LDI-MS in discriminating graphite materials, GCT_aniso, GCT_iso, and GPFO_aniso were analyzed under different laser fluences. The LDI-MS spectra consistently exhibited a series of carbon cluster peaks spaced by 12 Da in the m/z range of 60–336 (C₅–C₂₈) (Fig. 5). This periodic spacing is a well-known fingerprint of graphitic materials and has been used as an indicator of the transformation of petroleum pitches and synthetic polymers into graphitic structures [21, 22]. The presence of these ion clusters in all samples confirms that high-temperature treatment produced graphitic structures regardless of precursor type.

At a laser fluence of 337.83 J/cm², all samples displayed well-defined carbon cluster peaks, and their mass signal intensities increased with rising laser fluence (Fig. 5). These results indicate that laser desorption/ionization (LDI) of the sp² carbon domains was successfully induced under the applied analytical conditions. GCT_aniso exhibited abundant carbon cluster peaks in the low-mass region below m/z 132 (corresponding to the C₁₃ cluster) when the laser fluence exceeded 555.92 J/cm² (Fig. 5a). This progressive shift toward lower m/z values is attributed to laser-fluence-driven fragmentation of graphite materials [37]. We previously reported that LDI of sp² carbon domains in highly crystalline carbon materials requires a higher laser fluence than in low-crystallinity materials; thus, the mass intensities of carbon clusters decrease with increasing crystallinity under identical analytical conditions [20–22]. GCT_iso also showed increasing carbon cluster signal intensities with higher laser fluence; however, its overall signal intensities were noticeably lower than those of GCT_aniso (Fig. 5b). Notably, despite the rise in signal intensity, no carbon clusters appeared in the low-mass region below m/z 132 (Fig. 5b). This result suggests that GCT_iso does not contain large sp² carbon domains that require high laser fluence to undergo LDI. This observation is consistent with its relatively lower crystallinity compared with GCT_aniso, as indicated by its reduced G.D., Lc, La, and I_D/I_G values (Table 3). Unexpectedly, GPFO_aniso exhibited LDI behavior analogous to GCT_iso rather than GCT_aniso (Fig. 5c). Although the carbon cluster signal intensities increased with laser fluence, signals in the low-mass region below m/z 132 remained negligible (Fig. 5c). This is particularly interesting because the conventional characterization tools suggested that GPFO_aniso possesses a crystalline structure similar to that of GCT_aniso, despite their different precursor types. These results imply that LDI-MS is a more sensitive analytical tool for identifying precursor types of graphite materials than conventional XRD or Raman spectroscopy. Considering the intrinsic differences between the precursors such as CT and PFO, the LDI-MS analysis results are reasonable. CT exhibits higher FC, QI, and aromatic content than PFO (Fig. 1), all of which promote the formation of highly crystalline graphite during high-temperature treatment [38].

To further elucidate the differences in the LDI-MS spectra of graphitic materials, it should be noted that the total ion intensity (TII) is closely related to the size and population of sp² carbon domains, which collectively govern laser absorption capacity and thermal dissipation. TII is obtained by summing the intensities of all detected carbon cluster ions, and this value is determined by the balance between laser absorption and thermal dissipation. A larger lateral crystal size (La) enhances the dissipation of locally generated thermal energy during laser irradiation, thereby lowering the TII. Conversely, a thicker crystal (Lc) increases the laser absorption capacity per irradiated area, resulting in an increased TII. Finally, a higher population of sp² carbon domains directly correlates with greater laser absorption capacity. Regarding the LDI-MS spectra, GPFO_aniso exhibited a significantly higher TII (2,532,916) than GCT_iso (2,259,470) despite having a larger lateral crystal size (La: 29.00 nm) than GCT_iso (La: 21.74 nm), a factor that typically lowers TII. However, this is counterbalanced by its thicker crystal (Lc: 46.20 nm) than GCT_iso (Lc: 31.10 nm). This structural contradiction suggests that the TII-enhancing effect of increased crystal thickness dominates over the TII-decreasing effect of a lager lateral crystal size. According to conventional analyses such as XRD and Raman spectroscopy, the population of sp² carbon domains is also higher in GPFO_aniso than in GCT_iso.

Notably, GCT_aniso exhibited the highest TII (2,755,055) despite possessing structural characteristics disadvantageous for laser absorption, a thinner crystal (45.4 nm) and a larger lateral crystal size (30.34 nm) than those of GPFO_aniso. This strongly implies that the substantially higher population of sp² carbon domains in GCT_aniso overcomes these structural disadvantages. Consequently, the high TII in GCT_aniso primarily originates from its abundant sp² carbon domains. This dense population of sp² carbon domains leads to the localized generation of excessive thermal energy during laser irradiation, resulting in more severe fragmentation of carbon cluster ions in the low mass region than GPFO_aniso and GCT_iso.

Multivariate statistical analysis for LDI-MS data interpretation

Although LDI-MS has been employed for the structural analysis of graphite materials, the previous studies have typically relied on the interpretating features in individual LDI-MS spectra such as mass signal intensities and m/z values of the carbon clusters [20–22]. However, LDI-MS spectra are intrinsically high-dimensional as each mass peak constitutes an independent variable. This complexity requires rigorous preprocessing to minimize experimental variability and ensure comparability across multiple samples. Accordingly, all LDI-MS spectra were processed with the MALDIquant R package, which performs peak binning, smoothing, alignment, and normalization [39]. These processes mitigate baseline fluctuations and correct minor experimental shifts, making data matrices suitable for direct and reliable comparison of spectral fingerprints. The resulting matrices were subsequently subjected to PCA, HCA, and heatmap visualization to explore systematic spectral variations of GCT_aniso, GCT_iso, and GPFO_aniso under the different laser fluences. These unsupervised multivariate approaches, which operate without prior class labels, are widely employed for MS-based datasets because they reduce dimensionality, highlight variance-driven structures, and reveal molecular fingerprints that are not readily discernible by inspection of raw LDI-MS spectra [40].

PCA reduces dimensionality of mass peaks in LDI-MS spectra by constructing orthogonal components that capture maximum variance, and the score plot visualizes sample grouping in a low-dimensional space, thereby reflecting differences in the detected ion patterns [41]. At a laser fluence of 337.83 J/cm², the first two principal components such as PC1 and PC2 accounted for 70.3% (PC1 = 61.1% and PC2 = 9.2%) of the total variance (Fig. S3a), whereas at 440.25 J/cm², they explained 75.3% (PC1 = 50.4% and PC2 = 24.9%), enabling clear separation of the GCT_aniso, GCT_iso, and GPFO_aniso (Fig. 6a). At a fluence of 555.92 J/cm², PC1 interpreted 99.6% of the total variance (Fig. S3b), indicating that the dataset became an effectively one-dimensional structure dominated by a single component. This collapse onto PC1 suggests the emergence of strong common ion signals or saturation effect that obscures subtler spectral differences in their LDI-MS spectra (Fig. 5). As a result, GCT_aniso was well separated from GCT_iso and GPFO_aniso in the PCA plot, whereas the GCT_iso and GPFO_aniso overlapped substantially (Fig. 6a). This trend is consistent with their corresponding LDI-MS spectra (Fig. 5), presenting that GCT_aniso showed distinct intense low-mass carbon cluster peak while GCT_iso and GPFO_aniso exhibited an analogous carbon cluster pattern and lacked prominent carbon cluster peaks in the low-mass region.

Fig. 6.

a PCA plot, b HCA dendrogram, and c heatmap analysis of GCT_aniso, GCT_iso, and GPFO_aniso obtained at 440.25 J/cm² laser fluence

While PCA effectively summarizes overall variance and provides intuitive visualization of group separation, it does not inherently assess statistical significance [41]. To complement PCA results, HCA was carried out, which iteratively merges samples based on spectral similarity to construct a hierarchical dendrogram [42]. At 337.83 J/cm² fluence, even though PCA resulted in clear separation, HCA showed partial overlap between GCT_iso and GPFO_aniso, attributed to the 29.7% residual variance unaccounted for PC1 and PC2 (Fig. S4a). At 440.25 J/cm² fluence, HCA achieved stable and well-defined clusters of GCT_aniso, GCT_iso, and GPFO_aniso with the improved intergroup separation compared to 337.83 J/cm² fluence (Fig. 6b), and it is noteworthy that PCA result also provided their clear separation at this fluence. This consistency implied that the 440.25 J/cm² fluence was optimal for both PCA and HCA to induce sufficient LDI of carbon cluster ions from GCT_aniso, GCT_iso, and GPFO_aniso without severe fragmentation which obscured the discrimination of their spectral differences. At over 555.92 J/cm², HCA still constructed well-defined clusters (Fig. S4b), indicating that the three graphite samples were distinguishable, whereas PCA plots showed partial overlap because PC1 accounted for nearly all variance at this condition (Fig. S3b). The divergence underscores the complementarity of PCA and HCA: PCA relies on a reduced dimensional representation, but HCA leverages full-spectrum similarity to perceive subtle intergroup differences even under the fragmentation-dominated high laser fluence conditions.

To further complement PCA and HCA, heatmap analyses were carried out, which simultaneously visualize samples and m/z features with relative intensities encoded by color [43]. This approach revealed distinct spectral fingerprints for GCT_aniso, GCT_iso, and GPFO_aniso. At 337.83 J/cm² fluence, overall mass signal intensity was too weak to clearly resolve precursor type-dependent features (Fig. S5a), whereas at 440.25 J/cm² fluence, sharp and group-specific patterns emerged (Fig. 6c), indicating both mass signal intensity and discriminatory power were also optimized at this condition. In contrast, LDI-MS spectra obtained at 555.92 J/cm² fluence exhibited excessive mass signals that obscured subtle distinction between GCT_iso and GPFO_aniso (Fig. S5b). Consequently, PCA, HCA, and heatmap analyses indicated that the optimal laser fluence was 440.25 J/cm² for LDI-MS analysis of the graphite samples (Fig. 6). Under this condition, the three graphite groups were clearly separated, enabling their reliable classification by precursor types and microstructural anisotropy. Additionally, CPFO_aniso and GPFO_aniso were also analyzed with the identical procedure. The PCA, HCA, and heatmap analyses also achieved the successful separation of CPFO_aniso and GPFO_aniso (Fig. S6), and it implied that the present analytical strategy enabled distinction of carbon materials depending on their processing parameter such as heat treatment temperature. Those results demonstrated that LDI-MS, when integrated with multivariate approaches under the optimized conditions, is a robust and efficient platform for the precise identification and discrimination of graphite materials.

To verify the applicability of LDI-MS combined with multivariate analysis to commercial graphite materials, two commercial anisotropic graphites such as Petroleum-derived commercial graphite anode (PCGA) and Coal-tar-derived commercial graphite anode (CCGA) derived from PFO- and CT-based precursors were analyzed with their laboratory-prepared counterparts such as GPFO_aniso and GCT_aniso, respectively. Their LDI-MS spectra were obtained at the optimized 440.25 J/cm² fluence condition, and the resulting datasets were subjected to PCA, HCA, and heatmap analysis under the identical preprocessing conditions. The PCA score plot showed their clear separation into two distinct clusters: CT-derived (CCGA and GCT_aniso) and PFO-derived graphites (PCGA and GPFO_aniso) which occupied opposite regions of the score space, respectively (Fig. 7a). Along PC1, PFO-derived graphites consistently appeared on the negative score, whereas CT-derived graphites occupied on the positive side, indicating that PC1 acts as the principal discriminatory axis between different precursor types (Fig. 7a). The LDI-MS spectral variables with high loadings on PC1 can be regarded as key chemical markers distinguishing CT- and PFO-derived anisotropic graphites.

Fig. 7.

a PCA plot, b HCA dendrogram, and c heatmap analysis of CCGA, GCT_aniso, GPFO_aniso, and PCGA obtained at 440.25 J/cm² laser fluence

Then, HCA and heatmap analyses were also conducted as complementary tools to distinguish graphites based on their precursor types. Although complete separation of all graphites was not achieved with HCA, the dendrogram reproduced the PCA based precursor classification, forming two main branches corresponding to CT- and PFO precursors (Fig. 7b). These results further supported that the overall clustering hierarchy was governed by the precursor types. Heatmap analysis visually corroborated PCA and HCA results by showing their LDI-MS spectral patterns. PCGA and GPFO_aniso exhibited analogous m/z intensity distribution, confirming their common LDI-MS spectral fingerprints of PFO-derived graphites despite their different origins: GPFO_aniso is home-made; PCGA is a commercial product (Fig. 7c). Likewise, GCT_aniso and CCGA also presented analogous patterns, reflecting their common features of CT-derived graphites (Fig. 7c). As a result, the color contrast between the PFO- and CT-derived graphite groups was clearly discernable, allowing simple visual differentiation (Fig. 7c). Overall, the combined PCA, HCA, and heatmap analyses demonstrated that both commercial and laboratory-prepared anisotropic graphites can be reliably classified depending on their precursor type (CT vs. PFO). Despite differences in their synthesis route and processing history, commercial graphites consistently clustered with their corresponding laboratory-prepared analogues and exhibited nearly identical LDI-MS spectral patterns within each precursor family. Those results confirmed that, unlike conventional XRD, Raman, and TEM analyses which primarily focus on crystallinity, LDI-MS combined with multivariate statistics detected subtle chemical fingerprints that differentiate graphites by precursor origin and microstructural anisotropy based on their LDI-MS spectral pattern. These findings showed strong potential of LDI-MS as a reliable and rapid analytical platform for specific identification of unknown graphite materials.

Conclusion

In this study, we develope a methodology for identifying the precursor origins of synthetic graphite anodes using LDI-MS integrated with multivariate analyses. Although CT and PFO showed fundamentally different chemical characteristics, CT was aromatic, QI-containing, and heteroatom-rich, while PFO was aliphatic and QI-free. These differences disappeared after graphitization at 3000 °C. XRD, Raman spectroscopy, and TEM confirmed that GCT_aniso and GPFO_aniso exhibited nearly identical crystallinity and graphitic layer development. Electrochemical tests also showed that both anisotropic graphites outperformed the isotropic sample but remained indistinguishable from each other, demonstrating that conventional structural and electrochemical analyses cannot reveal precursor origin, posing challenges for quality control and material authentication. In stark contrast, LDI-MS enabled clear differentiation of graphite materials based on their precursor. While all samples displayed characteristic 12 Da-spaced carbon cluster ions, the evolution of these peaks with laser fluence revealed precursor-dependent fragmentation signatures. GCT_aniso produced strong fragment ions at high fluence in low m/z region (< m/z 132), whereas GCT_iso and GPFO_aniso did not despite their structural similarity in conventional analyses. Multivariate analyses (PCA, HCA, and heatmaps) translated these spectral fingerprints into robust classification, clearly separating all graphite types under the optimized fluence (440.25 J/cm²) and distinguishing not only precursor origin but also processing history (e.g., CPFO_aniso vs. GPFO_aniso). In addition, when applied to commercial anisotropic graphites, the combined LDI-MS–multivariate approach accurately classified materials into CT-derived and PFO-derived families, with commercial samples clustering tightly with their laboratory-prepared analogues. This implies that the proposed method can identify precursor origin in synthetic graphite samples prepared under different processing conditions. Therefore, this work offers a practical route for authentication, quality assurance, and origin tracing of industrial graphite powders and sets the foundation for the broader application of mass spectrometric fingerprinting in carbon materials research.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Inchan Yang: Investigation, Validation, Formal analysis, Data curation, Writing-original draft. Yungyeong Bae: Investigation, Validation, Formal analysis, Data curation, Writing-original draft. Jung-Chul An: Investigation, Formal analysis. Taejung Jung: Investigation, Formal analysis. Se Youn Cho: Resources, Supervision. Young-Kwan Kim: Conceptualization, Validation, Resources, Supervision, Data curation, Writing-review & editing. Sungho Lee: Conceptualization, Validation, Resources, Supervision, Data curation, Writing-review & editing.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant (2022R1C1C1008388) and the National Research Council of Science & Technology (NST) grant (CRC23013-000) funded by the Korea government, Ministry of Science and ICT (MSIT, Korea). This research was supported by the Commercialization Promotion Agency for R&D Outcomes (COMPA) grant (2710084822) funded by the Ministry of Science and ICT (MSIT, Korea). This work was also supported by the Technology Innovation Program (RS-2024-00422895) funded by the Ministry of Trade Industry & Energy (MOTIE, Korea).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Competing intrests

The authors declare that they have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.