Raman Spectrum Analyzer 🔬

A Python application for quantitative analysis of Raman spectra of graphene and graphene-like (sp² carbon) materials, with robust support for doped and disordered systems: N-doped, B-doped, SiC-grown graphene, GO, rGO, and dedicated CN-mode detection for g-C₃N₄ under UV/NIR-friendly Raman conditions.

Physical Background

Raman spectroscopy is the primary non-destructive probe of structural quality, defect density, doping level, layer number, and strain in sp² carbon materials. The key spectral features are:

Band	Typical Position (532 nm)	Origin	Dispersive?
D*	~1100–1200 cm⁻¹	Disordered sp³/sp² boundary; OH/epoxy functional groups	Yes
D	~1345 cm⁻¹	Intervalley double resonance at K-point; requires a defect	Yes — 53 cm⁻¹/eV
G	~1580 cm⁻¹	In-plane C–C stretching (E₂g); present in all sp² carbon	No (but doping-sensitive)
D′	~1620 cm⁻¹	Intravalley double resonance; defect-activated	Yes — ~37 cm⁻¹/eV
2D	~2690 cm⁻¹	Two-phonon second order of D; no defect needed	Yes — 106 cm⁻¹/eV
D+G	~2940 cm⁻¹	Combination of D and G phonons; defect-assisted	Yes

Dispersion: The 2D band dispersion is approximately twice the D band dispersion, but this relation is not exact. Reported 2D dispersions range from ~85–106 cm⁻¹/eV (e.g., 99 cm⁻¹/eV for turbostratic graphite and 87 cm⁻¹/eV for HOPG [Barros 2005]; ~100 cm⁻¹/eV for graphene [Ferrari & Basko 2013]), while D band dispersions span ~37–53 cm⁻¹/eV (e.g., 37.4 cm⁻¹/eV for B-doped graphene [Kim 2012]; ~53 cm⁻¹/eV [Dresselhaus 2005]; ~50 cm⁻¹/eV [Wu 2018]). Both values vary with material, doping, and stacking order. This tool uses the approximate 2:1 ratio only to predict laser-dependent peak fitting windows — sufficient for that purpose, not for precise dispersion measurements. Peak fitting windows shift automatically with --laser input.

G-band and doping: The G band is non-dispersive in position but shifts with carrier density. Both electron and hole doping stiffen the C–C bond via removal of the Kohn anomaly, upshifting the G band. The direction and magnitude of the shift distinguish n-type from p-type doping [Pisana 2007].

Material-Specific Raman Signatures

Boron-Doped Graphene

Substitutional boron (0.22 at% in single-layer graphene) introduces chemical disorder: the B–C bond is ~0.5 Å longer than C–C, breaking translational symmetry and activating the D band. Key signatures [Kim 2012]:

I_D/I_G ≈ 7: the extremely high D-band intensity originates from electron–boron elastic scattering before phonon emission — distinct from point defects where the phonon is emitted first.
G-band position: constant as I_D/I_G increases. This is the primary diagnostic separating B-doping from argon-plasma point defects (where G blueshifts monotonically). The p-type stiffening effect (upshift) is exactly counterbalanced by tensile-strain softening from the longer B–C bond.
D′ at ~1620 cm⁻¹: in heavily B-doped samples the D′ intensity equals the G intensity. D′ dispersion in B-doped graphene is ~37 cm⁻¹/eV (vs. ~77 cm⁻¹/eV for the 2D band) — consistent with hole-doped graphene theoretical predictions.
2D downshift: ~4 cm⁻¹ per unit I_D/I_G up to I_D/I_G ≈ 3; beyond that the double-resonance q-vector is modified by strain.
Defect spacing: L_D ≈ 4.76 nm corresponds to 0.22 at% B (Cançado 2011 formula applied to B atoms as vacancy-like scatterers).
I_D/I_D′ ≈ 7: consistent with vacancy-type scatterers [Eckmann 2012], confirming substitutional incorporation rather than edge or sp³ defects.

Nitrogen-Doped Graphene

Nitrogen substitution creates n-type doping. The N–C bond length is closer to C–C than B–C, so structural disorder is lower. Key signatures:

G-band upshift due to Fermi-level elevation (electron doping → Kohn anomaly removal).
2D downshift and broadening: n-type doping shifts 2D to lower wavenumbers and broadens the peak — opposite trend to hole doping [Pisana 2007].
Lower I_D/I_G than B-doping at comparable dopant concentrations.
I_D/I_D′ ≈ 7–13 depending on whether N sits substitutionally (vacancy-like) or as pyridinic/pyrrolic N (sp³-like → higher ratio).

Graphitic C₃N₄ Phase (g-C₃N₄)

g-C₃N₄ has a fundamentally different Raman signature from graphene [Zinin 2009]:

Visible excitation (514/532 nm): strong fluorescence background obscures the spectrum, yielding only two broad overlapping bands near 1357 cm⁻¹ and 1560 cm⁻¹. These are fluorescence artefacts, not the graphene D and G bands — they carry no structural information.
UV (244 nm) or NIR (785 nm) excitation reveals the true triazine fingerprint:
- ~691 cm⁻¹: s-triazine ring breathing mode (ring–C–N in-plane; matches melamine at 677 cm⁻¹).
- ~988 cm⁻¹: second triazine breathing mode (matches melamine at 983 cm⁻¹).
- ~1596 cm⁻¹: C–N ring stretching.
- ~1728 cm⁻¹: high-frequency unassigned mode.
- The D band is absent in UV Raman of g-C₃N₄ — analogous to diamond-like carbon.

⚠️ Analyser note: When a spectrum shows the 691/988 cm⁻¹ doublet, the analyser can now flag g-C₃N₄-specific CN modes through dedicated CN_triazine and CN_bending peak windows. If these modes are detected under visible excitation, the report warns that fluorescence is likely and recommends UV or NIR Raman for quantitative CN-mode interpretation.

Reduced Graphene Oxide (rGO) and GO

Structural evolution from graphite → GO → rGO [Mohan 2017]:

Graphite: sharp G at ~1582 cm⁻¹, very weak D (edge only). D-band intensity is edge-dependent: strong at armchair, weak at zigzag.
GO: broad, overlapping D and G bands; high oxygen content disrupts π-conjugation; low electrical conductivity.
rGO: I_D/I_G increases after reduction as new graphitic nanocrystallites nucleate (π-conjugation is partially restored). The highest conductivity (103.3 S cm⁻¹) was achieved with HI reduction, correlating with the lowest defect density and narrowest D/G FWHM.
D band (~1100–1200 cm⁻¹)*: present in GO/rGO; assigned to the disordered sp³/sp² boundary and OH/epoxy functional groups. I_D*/I_G correlates directly with the C/O ratio and tracks the degree of oxidation [Lee 2021].
FWHM(G): consistent FWHM(G) across samples signals uniform sp² cluster size; increasing FWHM(D) signals more disordered graphitic rings.
I_2D/I_G: CVD graphene shows I_2D/I_G ≈ 0.179 with a single sharp 2D at ~2690 cm⁻¹. rGO shows a distorted or absent 2D region — not a reliable layer indicator.
Functionalised rGO: G downshifts by ~5 cm⁻¹ due to π–π charge transfer from adsorbed molecules (e.g., pyrene) — a Raman indicator of non-covalent functionalisation.

Functionalized Graphene — Substrate Effects

Covalent aryl functionalization of graphene is substrate-dependent [Dierke 2022]:

Graphene on SiO₂: I_D/I_G after functionalization ≈ 0.9; L_D ≈ 12.4 nm.
Graphene on hBN: I_D/I_G after functionalization ≈ 0.3; L_D ≈ 20.9 nm — hBN's flat, dangling-bond-free surface suppresses radical attack.
The substrate effect allows spatial patterning of functionalization density by patterning the underlying substrate.

How This Tool Compares

Capability	This tool	RamanSPy	rampy	Fityk	OriginPro
Global D/G/D′ fit with automatic deconvolution	✅	❌	❌	manual	manual
Laser-energy-dependent fit windows	✅	❌	❌	❌	❌
L_D and n_D via Cançado with E_L⁴ correction + uncertainty	✅	❌	❌	❌	❌
Range-based Eckmann defect-type classification with citations	✅	❌	❌	❌	❌
Stage 1/2 disorder regime detection	✅	❌	❌	❌	❌
Adaptive lineshape (pseudo-Voigt for disordered materials)	✅	partial	❌	manual	manual
Literature knowledge base (113 cited reference values)	✅	❌	❌	❌	❌
Fit-quality validation flags	✅	❌	❌	❌	❌
B/N doping fingerprint	✅	❌	❌	❌	❌
g-C₃N₄ dedicated support	✅	❌	❌	❌	❌
General preprocessing / ML / hyperspectral	❌ (delegates to RamanSPy)	✅	partial	❌	✅
Raman mapping/imaging	❌	✅	❌	❌	✅
Open source	✅	✅	✅	✅	❌

This tool sits on top of the open-source spectroscopy ecosystem: it uses pybaselines for baseline subtraction and optionally leverages RamanSPy's despiking and data loaders when available. Its unique value is a literature-anchored interpretation layer purpose-built for sp² carbons (graphene, GO, rGO, g-C₃N₄), translating raw peak fits into physically meaningful parameters with uncertainty. Users who need general preprocessing, machine learning, or hyperspectral imaging should use RamanSPy as the primary workflow and then feed pre-processed spectra into this tool for domain-specific analysis (see docs/INTEROPERABILITY.md).

Features

Baseline correction — Asymmetric Least Squares (ALS); negative residuals preserved (no clipping)
Peak fitting — Lorentzian (D, G, D′) and Pseudo-Voigt (D+G) using pure scipy
Global D / G / D′ fit — three bands fitted simultaneously to prevent D′ from capturing the G tail
Adaptive G-band fitting — locates true G peak position before fitting; handles doped samples where G sits near or above 1600 cm⁻¹
G+D′ deconvolution — dual-Lorentzian separation of overlapping G and D′ in heavily doped or disordered graphene
SNR-gated peak detection — peak accepted only when R² > 0.75 and SNR > 3 (SNR = amplitude / MAD-based noise estimate)
Pseudo-Voigt area correction — FWHM and integrated area follow the Thompson–Cox–Hastings definition
Quantitative ratios — I_D/I_G, I_2D/I_G, I_D′/I_G, I_D/I_D′
Defect density — L_D (Cançado 2011, full E_L⁴ correction)
g-C₃N₄ CN-mode detection — dedicated CN_triazine (~691 cm⁻¹) and CN_bending (~988 cm⁻¹) windows for UV/NIR-friendly Raman analysis
g-C₃N₄ report logic — gcn4_detected flag plus gcn4_mode_note warning for visible excitation
Disorder stage — Stage 1 / Stage 2 classification with FWHM(G) cross-check
Defect type classification — sp³ / vacancy / edge via I_D/I_D′ [Eckmann 2012]
Layer number estimation — from I_2D/I_G with FWHM(2D) reliability guard
eV-based laser dispersion — all peak windows shift with laser energy per Cançado 2011
Batch processing — analyse entire folders at once
Publication-quality plots — 4 plots per spectrum (300 dpi)
CSV + text reports — structured output ready for further analysis

Peak Windows (532 nm laser)

Peak	Search Range	Fit Window	Line Shape	Physical Origin
CN_triazine	670–715 cm⁻¹	fixed	Lorentzian	g-C₃N₄ triazine ring mode (~691 cm⁻¹)
CN_bending	960–1010 cm⁻¹	fixed	Lorentzian	g-C₃N₄ C–N bending / ring-related mode (~988 cm⁻¹)
D*	1080–1230 cm⁻¹	fixed	Lorentzian	sp³/sp² boundary; OH/epoxy groups in GO/rGO [Lee 2021]
D	1270–1450 cm⁻¹	fixed	Lorentzian	K-point phonon, intervalley double resonance [Ferrari 2001]
G	1540–1680 cm⁻¹ (search)	adaptive ±50 cm⁻¹	Lorentzian	E₂g in-plane C–C stretch [Tuinstra 1970]
D′	global fit with D+G	—	Lorentzian	Intravalley defect-induced [Eckmann 2012]
2D	2580–2780 cm⁻¹	fixed	Lorentzian (or dual)	Second-order overtone; layer-sensitive [Ferrari 2006]
D+G	2850–2960 cm⁻¹	fixed	Pseudo-Voigt	Combination band

G-band strategy: find_peaks in 1540–1680 cm⁻¹ → adaptive ±50 cm⁻¹ window → single Lorentzian. If R² < 0.60, dual-Lorentzian G+D′ deconvolution is triggered automatically. The D′ component is stored in PeakResult.deconv_partner and promoted to the D_prime result slot if it outperforms the standalone fit.

Peak positions shift with laser wavelength for dispersive graphene bands. Dispersive peaks: D ~53 cm⁻¹/eV, D′ ~37 cm⁻¹/eV, 2D ~106 cm⁻¹/eV (Cançado 2011, eV-based). The g-C₃N₄ CN_triazine and CN_bending windows are treated as non-dispersive auxiliary modes.

Quantitative Metrics

Defect Density — L_D

Inter-defect distance L_D (nm) from the laser-energy-corrected Cançado 2011 formula:

L_D² (nm²) = (4.3 × 10³ / E_L⁴) × (A_G / A_D)

where E_L is the laser energy in eV. Valid in Stage 1 only (L_D ≳ 10 nm). Applied to B-doped graphene, where substitutional B atoms behave as vacancy-like scatterers, this yields L_D ≈ 4.76 nm for 0.22 at% B [Kim 2012]. In Stage 2, L_D is set to NaN with a warning.

Disorder Stage Classification

Stage	I_D/I_G trend	FWHM(G)	L_D	Regime
Stage 1	Increases with disorder	Broadens moderately	≳ 10 nm	Crystalline graphene with point defects (Tuinstra–Koenig)
Stage 2	Decreases with disorder	Strongly broadened	< 10 nm	Amorphous / nanocrystalline carbon

Important: High I_D/I_G alone does not mean Stage 2. In B-doped graphene, I_D/I_G ≈ 7 while FWHM(G) remains narrow and the G position is constant — chemical disorder, not structural amorphisation [Kim 2012]. The analyser cross-checks FWHM(G) before assigning Stage 2.

Defect Type — I_D / I_D′

The ratio I_D/I_D′ discriminates between defect types [Eckmann 2012]:

I_D/I_D′	Defect type	Example
~13	sp³ (homoatomic)	H, O functional groups (GO, rGO)
~7	Vacancy-like	Ion bombardment, substitutional B or N
~3.5	Grain boundary / edge	Polycrystalline graphene, nanoflakes

Layer Number — I_2D/I_G

I_2D/I_G	2D FWHM	Layer count
> 2	~25 cm⁻¹ (single Lorentzian)	Monolayer [Ferrari 2006]
≈ 1	Broader, 4-component	Bilayer
< 1	Broad, graphite-like	Few-layer (≥ 3)
Distorted / absent	—	rGO / functionalised (unreliable)

Installation

git clone https://github.com/Hj1308/Raman-analysis.git
cd Raman-analysis
pip install -r requirements.txt

Dependencies: numpy, scipy, matplotlib, pandas, openpyxl

lmfit is optional — install only if you need Voigt / AsymLorentzian fits or uncertainty reporting via the band_config API.

Usage

# Single file (532 nm laser)
python main.py --file spectrum.txt --laser 532

# Single file with custom output directory
python main.py --file spectrum.txt --laser 633 --output ./my_results/

# Batch mode — all .txt/.csv/.xlsx files in a folder
python main.py --folder ./spectra/ --laser 785 --batch

Advanced: `band_config`

from src.peak_fitter import PeakFitter

config = {
    "G":   {"method": "adaptive"},
    "D":   {"lineshape": "Lorentzian"},
    "2D":  {"method": "deconvolve"},      # force dual-Lorentzian (bilayer)
    "D+G": {"lineshape": "Voigt"},        # requires lmfit
}
results = fitter.fit_all_peaks(band_config=config)

Input File Format

Plain text or CSV — two columns (wavenumber, intensity):

# Optional comment lines
1000.0   120.5
1001.0   122.3

Excel (.xlsx): wavenumber in column B, intensity in column C, data from row 4.

Output

For each spectrum:

File	Content
`*_baseline.png`	Raw spectrum + ALS baseline overlay
`*_peaks.png`	Fitted peaks on corrected spectrum
`*_individual.png`	Individual peak fits with residuals
`*_ratios.png`	Intensity ratio bar chart
`*_report.txt`	Full text report (all metrics)
`raman_results.csv`	All parameters in one CSV (batch-compatible)

PeakResult Fields

Field	Type	Description
`center`	float	Peak position (cm⁻¹)
`amplitude`	float	Peak height (a.u.)
`fwhm`	float	Full width at half maximum (cm⁻¹)
`area`	float	Integrated area (Thompson–Cox–Hastings for Pseudo-Voigt)
`r_squared`	float	Goodness of fit
`snr`	float	Signal-to-noise ratio (amplitude / MAD noise)
`found`	bool	True if R² > 0.75 and SNR > 3
`is_deconvolved`	bool	True when G was separated from D′ by dual-Lorentzian
`deconv_partner`	PeakResult \| None	D′ component from G deconvolution
`is_split_2D`	bool	True when 2D fitted with dual-Lorentzian (bilayer)
`twoD_fwhm_warning`	bool	True when FWHM(2D) > 35 cm⁻¹ (layer count unreliable)
`center_stderr`	float \| None	Standard error on center (lmfit backend only)
`fwhm_stderr`	float \| None	Standard error on FWHM (lmfit backend only)

Known Limitations

2D Band — Bilayer / Trilayer

Single Lorentzian for 2D is appropriate for monolayer only. In bilayer graphene, 2D splits into 4 sub-components; single-peak fitting underestimates FWHM and may misassign layer count. Multi-component 2D fitting is on the roadmap.

Ferrari et al. (2006) Phys. Rev. Lett. 97, 187401

L_D in Stage 2

Cançado 2011 formula is valid in Stage 1 only. When Stage 2 is assigned, L_D is set to NaN with a warning in L_D_note. Use FWHM(G) as the primary disorder metric for Stage 2 samples.

g-C₃N₄ and Visible Excitation

Visible-excitation spectra of g-C₃N₄ are dominated by fluorescence. The apparent 1357/1560 cm⁻¹ bands carry no reliable structural information [Zinin 2009]. The analyser now includes dedicated UV/NIR CN-mode support through the 691/988 cm⁻¹ windows, but visible-range spectra remain fundamentally fluorescence-limited and should be interpreted with caution.

Layer Number in rGO / Functionalised Graphene

I_2D/I_G is not a reliable layer indicator for rGO or functionalised graphene; the 2D band depends on preparation method, not layer count [Mohan 2017].

D* Band

The D* peak window (1080–1230 cm⁻¹) is defined, but quantitative I_D*/I_G analysis and C/O ratio correlation are not yet implemented. See roadmap.

G+D′ Overlap Extremes

Dual-Lorentzian deconvolution may not fully separate G and D′ in extreme cases. More complex multi-band models are available via the band_config API.

Roadmap

v2.4 — In Development

Feature	Scientific Basis	Priority
D band quantification*	I_D*/I_G ↔ C/O ratio in GO/rGO; oxidation degree tracking [Lee 2021]	🔴 High
B-doping fingerprint flag	Auto-detect: constant G position + I_D/I_D′ ≈ 7 + I_D/I_G ≫ 1 [Kim 2012]	✅ Implemented
Fitting uncertainty (scipy)	Error bars from covariance matrix for all users (no lmfit required)	🔴 High
Multi-Lorentzian 2D fitting	4-component decomposition for bilayer/trilayer [Ferrari 2006]	🟡 Medium
Carrier-density estimator refinement	G-band shift → carrier density → n/p doping type; scaling bug corrected in current main branch	✅ Implemented (further refinement possible)
Stage boundary refinement	FWHM(G) + A_D/A_G combined metric; 0D vs. 1D defect discrimination [Wu 2018]	🟡 Medium
Dispersion slope validator	Multi-wavelength: D slope ≈ 53 cm⁻¹/eV, 2D ≈ 106 cm⁻¹/eV; deviation flags contamination [Cançado 2011]	✅ Implemented
arPLS baseline	Asymmetrically reweighted PLS for fluorescence-heavy spectra (GO, g-C₃N₄)	🟡 Medium
NIR/UV mode for g-C₃N₄	Dedicated windows at 691/988 cm⁻¹; analyzer warning/note for visible excitation [Zinin 2009]	✅ Implemented
Batch statistics	Mean ± std across all samples; ratio heatmap for batch runs	🟢 Planned
Substrate effect report	Flag I_D/I_G contrast on hBN vs. SiO₂; L_D comparison [Dierke 2022]	🟢 Planned
Electrochemical doping tracker	G-band shift vs. gate voltage; cyclic voltammetry-compatible input	🟢 Planned

Project Structure

Raman-analysis/
├── main.py                  ← CLI entry point
├── requirements.txt
├── README.md
├── examples/
│   ├── graphene_test.txt    ← synthetic test spectrum
│   └── generate_test_spectrum.py
├── results/                 ← output directory
└── src/
    ├── __init__.py
    ├── loader.py            ← file reading & preprocessing (txt, csv, xlsx)
    ├── baseline.py          ← ALS baseline correction (no clipping)
    ├── peak_fitter.py       ← global D/G/D′ fit, adaptive G, G+D′ deconvolution, SNR gate
    ├── analyzer.py          ← ratio calculations & defect analysis (L_D E_L⁴-corrected)
    ├── plotter.py           ← matplotlib visualisation (300 dpi)
    └── exporter.py          ← CSV + text report export

Citation

@software{raman_analyzer_2026,
  author    = {H.J},
  title     = {Raman Spectrum Analyzer: Quantitative Analysis of Graphene Raman Spectra},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21109057},
  url       = {https://doi.org/10.5281/zenodo.21109057}
}

Plain text: H.J (2026). Raman Spectrum Analyzer. Zenodo. https://doi.org/10.5281/zenodo.21109057

Scientific References

Core Raman Methods

Tuinstra, F. & Koenig, J.L. (1970) J. Chem. Phys. 53, 1126
Ferrari, A.C. & Robertson, J. (2001) Phys. Rev. B 64, 075414 — disorder stage classification
Ferrari, A.C. et al. (2006) Phys. Rev. Lett. 97, 187401 — 2D band and layer number
- Lucchese, M.M. et al. (2010) Carbon 48, 1592 — defect-density / inter-defect-distance framework for disordered graphene
Cançado, L.G. et al. (2011) Nano Lett. 11, 3190–3196 — E_L⁴-corrected L_D; eV-based dispersion
Ferrari, A.C. & Basko, D.M. (2013) Nature Nanotechnology 8, 235–246 — peak conventions & line shapes

Defect Characterisation

Eckmann, A. et al. (2012) Nano Lett. 12, 3925–3930 — I_D/I_D′ defect type discrimination
Pimenta, M.A. et al. (2007) Phys. Chem. Chem. Phys. 9, 1276–1290 — disorder in graphite-based systems
Claramunt, S. et al. (2015) Sci. Rep. 5, 19491 — G+D′ overlap in GO/rGO

Doped Graphene

Kim, Y.A. et al. (2012) ACS Nano 6(7), 6293–6300 — B-doped graphene; G invariance, D×7, L_D = 4.76 nm
Pisana, S. et al. (2007) Nature Materials 6, 198–201 — G-band shift and Kohn anomaly

Graphitic C₃N₄

Zinin, P.V. et al. (2009) Chem. Phys. Lett. 472, 69–73 — triazine modes at 691/988 cm⁻¹

Reduced Graphene Oxide

Mohan, V.B. et al. (2017) Graphene Technology 2(4–7), 6–21 — rGO structural evolution; conductivity vs. I_D/I_G
Lee, J. et al. (2021) Carbon 183, 814–822 — D* band in GO/rGO; I_D*/I_G vs. C/O ratio
Dierke, T. et al. (2022) ACS Appl. Nano Mater. 5, 4966–4971 — substrate-dependent functionalization
Hummers, W.S. & Offeman, R.E. (1958) J. Am. Chem. Soc. 80, 1339 — GO synthesis

Line Shape

Thompson, P., Cox, D.E. & Hastings, J.B. (1987) J. Appl. Crystallogr. 20, 79–83 — Pseudo-Voigt profile

Changelog

v2.4 — 2026-07-03

Feature #9: dedicated g-C₃N₄ CN-mode support via CN_triazine (670–715 cm⁻¹) and CN_bending (960–1010 cm⁻¹)
Added gcn4_detected and gcn4_mode_note to analyzer output
Added g-C₃N₄ CN-mode section to formatted report
Visible-excitation warning for g-C₃N₄ CN-mode detection; UV/NIR-friendly note for non-visible excitation
Fixed carrier_density_cm2 scaling in _estimate_doping by removing an erroneous extra factor of 10^12
Full test suite passes after Feature #9 and carrier-density fix

v2.3 — 2026-07-02

Global D / G / D′ simultaneous fit (_fit_D_G_Dp_global)
SNR gate: R² > 0.75 and SNR > 3; MAD-based noise estimate
Pseudo-Voigt corrected: Thompson–Cox–Hastings area definition
L_D: full E_L⁴ Cançado 2011 formula; Stage 2 → NaN + warning
FWHM(2D) guard: twoD_fwhm_warning when FWHM(2D) > 35 cm⁻¹
Removed graphitization_pct (no literature basis)
Baseline clipping removed (negative residuals preserved)
eV-based peak dispersion in get_peak_windows

v2.2 — 2026-07-01

Adaptive G-band window + dual-Lorentzian G+D′ deconvolution
PeakResult extended: is_deconvolved, deconv_partner, is_split_2D
Removed lmfit hard dependency

Author

H.J — Researcher, Chemistry

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
docs		docs
examples		examples
results		results
src		src
tests		tests
tools		tools
validation		validation
.gitattributes		.gitattributes
.gitignore		.gitignore
README.md		README.md
ROADMAP.md		ROADMAP.md
app.py		app.py
build_knowledge.py		build_knowledge.py
main.py		main.py
preprocessing.py		preprocessing.py
requirements.txt		requirements.txt
streamlit_app.py		streamlit_app.py
uncertainty.py		uncertainty.py

Latest commit

History

Folders and files

Repository files navigation