Microchemical analysis of Leonardo da Vinci's lead white paints reveals knowledge and control over pigment scattering properties

Leonardo da Vinci’s technical practice is renowned for innovative methods such as sfumato, yet explicit recipes for his materials are largely undocumented. A microsample from the background of The Virgin and Child with St. Anne, taken near the transition between brown rocky landscape and bluish mountains, presents two adjacent lead white layers offering a unique opportunity to study material choices. Lead white comprises cerussite (PbCO3) and hydrocerussite (Pb3(CO3)2(OH)2), whose formation depends on pH and historical manufacturing (stack process). While synchrotron XRD has quantified HC/C ratios in microsamples, radiation dose and facility access are challenging. Recent work showed HC and C have distinct deep-UV PL, suggesting a non-invasive imaging route. This study investigates whether hyperspectral PL can semi-quantitatively discriminate HC and C at the microscale across the stratigraphy, validates PL with synchrotron HR-XRD, and interprets findings in light of historical practices and Leonardo’s aesthetic aims, especially regarding scattering properties and blue haze effects.

Historical sources indicate that lead white made by the stack process could be post-processed (e.g., washing in vinegar, levigation) to alter quality, composition, and particle size, sold at varied prices. Prior synchrotron XRD studies of Renaissance paintings typically found lead white with HC/(HC+C) between 60–80 wt% and large particles, consistent with reconstructed stack-process synthesis. Extending to R<50 wt% required prolonged corrosion and increased CO2; R>80 wt% could result from early collection. Photoluminescence studies established distinct deep-UV excited emissions for cerussite (~2.8–2.9 eV) and hydrocerussite (~2.0–2.1 eV). Painters’ use of different grades and possible blue-hued scattering effects have been hypothesized in Northern European art, with recent MA-XRD suggesting discriminating lead white use in Vermeer.

Sample and stratigraphy: A cross-section from the painting background shows an imprimitura (α) of white lead in oil, an overlying nearly white blue-toned layer (β) containing sparse ultramarine, minor silicates, and red lake, and a more complex top layer (γ) with earth and copper pigments, quartz, lead-tin yellow, and aluminosilicates. Lead white was present in α and β.

Synchrotron high-angle resolution X-ray diffraction (SR-HR-XRD): Two fragments (50–80 µm) were isolated: a β fragment (pure β layer) and an α fragment (primarily α but with minor β contamination indicated by lazurite). Measurements were performed at ESRF ID22 with a 1×1 mm² beam at 35 keV. Samples in capillaries were rotated; 20 min scans were repeated totaling 1–4 h. For α: 2≤2θ(°)≤22 (0.93≤d(Å)≤10.14), enabling 152 reflections for C and 49 for HC; for β: 4≤2θ(°)≤14 due to lower SNR. Rietveld refinement (FullProf Suite; Thompson–Cox–Hastings profile with spherical harmonics for Lorentzian and axial broadening for Gaussian) provided compositions (±1 wt% from C/HC calibration mixtures) and microstructural modeling. Uncertainties were scaled by Bérar’s factor. Peak broadening was analyzed via Williamson–Hall plots (βcosθ = 0.9λ/L + 4εsinθ) to extract crystallite size L and strain ε; anisotropy treated using families of reflections (nh nk nl series) to reconstruct platelet geometry.

Synchrotron deep UV-excited photoluminescence (SR-µ-PL): Hyperspectral PL micro-imaging was conducted at POLYPHEME (DISCO, Synchrotron SOLEIL) with 250 nm excitation (5 eV), a 40×/0.6 NA objective, emission range 400–680 nm (1.82–3.2 eV), 8 s integration per spectrum. A 30×20 pixel map (3 µm step) covered 90×60 µm. Spectral unfolding: positions of constituent bands were derived from the second derivative of a smoothed average spectrum; an n-Gaussian model (band positions fixed; amplitudes and widths σ optimized) was fitted using Newton’s method with soft parameter bounds and repeated convergences. For each spectrum, amplitudes were normalized to mitigate local absorption/reabsorption effects. Emission amplitude maps for key bands (1.85, 2.03, 2.94 eV) were combined into false-color images. A semi-quantitative metric Γ = ([HC]/[C])α/([HC]/[C])β = AHC,α/AHC,β was defined under the assumption that Φ and ε at 5 eV are constant within each phase across layers, enabling comparison to XRD-derived R via Γ = (R/(1−R))α/(R/(1−R))β.

Optical demonstration of scattering effects: Model films (60 µm) of HC in linseed oil were prepared with two granulometries: large crystallites (>20 µm) and submicrometric (<400 nm). Reflectance spectra (400–800 nm) were recorded to assess wavelength-dependent scattering and hue.

• XRD compositions: β fragment is cerussite-rich with R = HC/(HC+C) = 46 ± 2 wt%; α fragment is hydrocerussite-rich with R = 65 ± 2 wt% (actual HC likely slightly higher due to minor β contamination in the α fragment).
• Crystallite size and strain (Williamson–Hall): In α, HC forms platelets 110 ± 20 nm thick with ~10:1 aspect ratio and low, quasi-isotropic strain ε = (4.7 ± 0.2) × 10^−4. For β, limited SNR precluded full WH analysis for HC, but peak widths suggest similar magnitudes. Cerussite shows near-zero Scherrer broadening, indicating particle sizes >500 nm.
• PL spectral assignments (5 eV excitation): 2.94 eV band attributed to C; 2.03 eV and 2.32 eV to HC; 1.85 eV to Mn2+ impurity emission; 2.51 and 2.97 eV to oxidized drying oil binder. Hyperspectral PL maps show α dominated by HC emission (green), β dominated by C emission (blue).
• PL–XRD agreement: Semi-quantitative ratio Γ from PL amplitudes equals 2.1; Γ from XRD-derived R equals 2.2, validating PL-based phase discrimination within ~5%.
• Provenance of pigment grades: Both α and β fall in a region of HC thickness vs composition indicative of post-synthesis treatment to reduce crystallite size. β’s composition and submicrometric grains are consistent with acidic post-treatment (e.g., washing/grinding in vinegar), which dissolves HC and reprecipitates fine C, yielding R<50 wt% and polydisperse submicrometer particles. Levigation in water would alter size distribution without significantly changing composition.
• Optical scattering demonstration: Submicrometric HC films exhibit reduced reflectance toward longer wavelengths and a perceivable blue hue, consistent with increased Rayleigh scattering by <100 nm fractions; large-crystallite HC shows high, nearly flat reflectance across the visible.

The presence of two adjacent lead white layers with distinct HC/C ratios and microstructures suggests Leonardo deliberately selected different pigment grades for specific functions. The α imprimitura contains a typical stack-process HC-rich lead white (R ~65 wt%), whereas the β layer contains a cerussite-rich, post-treated, fine-particle grade (R ~46 wt%). Historical records, including Leonardo’s own notes in the Arundel 263 manuscript listing purchases of two lead whites at different prices, support access to multiple qualities and conscious selection. Two rationales are proposed: (1) rheological control—fine granulometry facilitates extremely thin, subtle glaze layers characteristic of Leonardo’s technique; (2) optical control—polydisperse, submicrometric lead white enhances Rayleigh scattering of blue light, potentially imparting a subtle blue hue suited to atmospheric perspective effects (e.g., distant mountains and sky). The deep-UV PL imaging, validated against synchrotron HR-XRD, demonstrates a non-invasive, spatially resolved approach to mapping lead white phases in complex stratigraphies, linking microstructure, composition, and visual outcomes. These findings align with broader observations in historical paintings where lead white grades may have been discriminated to achieve specific optical effects.

Leonardo da Vinci used distinct grades of lead white in The Virgin and Child with St. Anne, selecting an HC-rich, coarser pigment for the imprimitura and a cerussite-rich, post-treated, fine-particle pigment for an overlying layer. Synchrotron HR-XRD quantified compositions and revealed crystallite morphologies, while deep-UV hyperspectral PL imaging provided semi-quantitative, spatially resolved phase mapping consistent with XRD. The results, supported by Leonardo’s notes on purchasing different lead white qualities, indicate deliberate control over rheological and scattering properties to achieve aesthetic goals, including subtle blue hues via Rayleigh scattering. Future work should examine additional samples from Leonardo’s oeuvre for similar usage, expand fundamental PL studies of pigments and binders, develop lab-based deep-UV PL micro-imaging (and time-resolved PL) coupled with Raman, and bridge micro-to-macro scales with non-invasive macro-PL/XRF scanning.

• Micro-sample representativity is inherently limited; sampling of Leonardo’s works is particularly constrained due to their exceptional status.
• PL-based quantification is semi-quantitative, relying on assumptions of constant quantum yields and absorption coefficients across layers for each phase; deviations may introduce bias.
• Low pigment content and signal-to-noise in certain layers (β) limit full microstructural analysis (e.g., Williamson–Hall for HC) and precise particle size estimation for the finest fractions (high polydispersity).
• The inferred post-synthesis treatments are deduced from composition–size trends and historical sources; direct documentary evidence of specific treatments for this sample is unavailable.