The four panels are made from colourless and pot-metal yellow, green, blue, and purple glass, as well as a few flashed red pieces. Additionally, the details are depicted in black with grisaille and in yellow, orange, or red by silver staining. The real-thickness colourimetric coordinates of the investigated glass pieces are shown in Fig. 2 on the CIE 1931 xy colour diagram. The objective description of the colours proposed in the subsequent text is based on the approximate colour areas of the CIE.
The analytical results are presented in four sections, following the four quality parameters: (1) the glass composition, (2) the glass forming technique, (3) the transparency and hue of the colourless glass, and (4) the characteristics of the colouring technology.
Compositional characteristics
The analytical study of the four panels by combining UV–Vis-NIR absorption spectroscopy with p-XRF reveals that glass of different compositional groups was employed to build each panel. In the first step, we distinguished seven optical groups for colourless glass, four for blue glass, three for purple glass, two for red glass, three for green glass, and six for yellow glass (see Additional file 7: Figure S5). These groupings were made based on optical parameters, namely the [Fe2+] calculated from the Fe2+ absorption at 1100 nm [34], the UVAE, the colour coordinates, the Co2+ absorption bands in the visible region [39, 44], and the presence of any other chromophore absorption band (see Additional file 6: Tables S2, S4, S6, and S8 and Additional file 7: Figure S5) [33, 45]. This approach showed its potential to distinguish optical groups that are consistent with the different glass compositional groups [39, 46]. Therefore, we consider that glass pieces with the same optical characteristics (i.e., belonging to the same optical group) have the same composition.
In the second step, we attempt to link the optical groups with the glass composition. For that purpose, at least one piece of glass from each optical group was measured via p-XRF. The results were first interpreted following the approach of Dungworth [47] and then confirmed with the method of Adlington and Freestone [40]. The former is based on the measured K2O:CaO ratio, whereas the latter consists of considering Rb, Sr, and Zr as substitutes for K2O, CaO and Ti respectively, to overcome the p-XRF unreliability of the lighter elements because of the effect of the glass corrosion. The glass composition groups were identified based on the flowcharts proposed by Dungworth [47] and Schalm and colleagues [48]. The results indicate that the four panels were assembled from two different glass compositional groups, K-rich glass and Ca-rich glass. K-rich glass shows three compositionally distinct groups, whereas Ca-rich glass shows two groups that are consistent between the four panels and coherent with the dating (see Additional file 6: Tables S3, S5, S7, and S9 and Additional file 8: Figure S7). The two infills in the border of panel AV.1165 (glass pieces 21 and 22) show a nineteenth- or twentieth-century composition, probably industrial soda, because their K2O and Rb contents measured by p-XRF are close or equal to 0.
The repartition of the glass composition groups in the four panels is reported in Fig. 3. An interesting aspect highlighted by the study of the glass composition groups is the fact that for panels AV.1164, AV.1165, and AV.1166 the backgrounds (landscape and borders) are almost exclusively rendered with Ca-rich glass. In addition, Ca-rich glass is the only glass composition group used for the depiction of the two nuns, i.e., the characters with the lower position in the religious hierarchy (in panels AV.1164 and AV.1166). Most of panel AV.1163 is made from K-rich glass, including the background. Only the green pieces of panel AV.1163 have a Ca-rich glass composition.
Figure 4 shows the repartition of the subgroups for the K-rich and Ca-rich glass groups, emphasising the use of a specific K-rich glass subgroup (K2) for the most important characters (Virgin and bishop in panel AV.1163, Virgin and Christ in panel AV.1164) or for parts of the images highlighted by the artistic composition (heads and attributes in AV.1165 and AV.1166). The K-rich glass subgroup K3 is used for two other characters of high status: the Christ (AV.1163) and Saint John (AV.1164). The same glass chemical subgroup is used for the bodies of Saint Dorothy (AV.1165) and Saint Clare (AV.1166). This raises the question of whether the stained-glass makers were able to recognise, to some extent, differences in the glass sheets prior to composing the window, giving them the possibility of a material selection. Can we link the glass composition to the glass-forming technique? Particularly concerning the colourless glass parts, which are mostly used for the characters, were the stained-glass makers able to distinguish K-rich glass from Ca-rich glass based on their colour and/or transparency, as reported in historical sources? These two aspects, the glass-forming technique, and the transparency and hue are investigated in the next two sections.
Forming technique and visual characteristics
Concerning the glass-forming technique, two main types of production traces could be observed on the glass pieces from the four panels. First, we observed numerous elongated bubbles arranged in straight and parallel lines, suggesting the glass was produced with the cylinder forming technique (Fig. 5). These bubbles are visible on the exterior side of most of the yellow glass pieces used for the borders (panels AV.1164, AV.1165, and AV.1166) and for the landscape of panel AV.1163. All these glass pieces have a Ca-rich glass composition.
Second, blue glass pieces used for the sky in panels AV.1164 and AV.1165 show many inclusions on the interior side of the glass surface (Fig. 6). A high number of inclusions is a typical accidental feature of the cylinder glass-forming technique that happens when pouring the glass on an uncleaned stretching table. These glass pieces also have a Ca-rich glass composition and belong to different optical groups (B2, B3, and B4).
Two pieces show a single isolated inclusion on their surface (Fig. 7). The presence of a single inclusion is not sufficient to identify the glass forming method, as it could come either from the crucible, glass-making tools, or stretching table in the cylinder technique.
None of the other glass pieces show bubbles, inclusions, or production traces. This is especially true for the K-rich glass pieces. Visually, the K-rich glass pieces are also very clear, smooth, and even. In addition, K-rich glass is, on average, slightly thicker and discrepant in thickness than Ca-rich glass (2.37 ± 0.51 vs 2.20 ± 0.34). These observations lead us to think that the K-rich glass was probably produced with the crown glass-forming technique.
Most of the glass pieces are very well conserved and show no corrosion. This is rather surprising because K-rich glass is more sensitive to alteration than Ca-rich glass. Only glass pieces with a K-rich glass K1 composition show substantial pitting.
In synthesis, we see coherence between the glass-forming technique and its composition. The tendency observed is that Ca-rich glass was produced via the cylinder forming technique as it shows more accidents (bubbles or inclusions). On the opposite, the K-rich glass pieces—except of the presence of a single isolated inclusion at the surface of a purple piece of panel AV.1166—visually appear flawless, and thicker, and were most probably produced with the crown glass-forming technique. This means that the glaziers could visually distinguish different glass qualities based on the glass surface characteristics.
Transparency and hue
In addition to the better visual properties of K-rich glass, literature [13] indicates that K-rich glass is also clearer (i.e., better decoloured) and more transparent. Therefore, in this paragraph we aim to verify if this statement is correct, specifically for colourless glass, based on optical parameters.
First, the calculation of the optical transparency reveals that K-rich glass (mostly related to crown glass production) is not always the most transparent glass; on average, Ca-rich glass has a transparency value of 51.9 ± 2.6% (44.3 ± 8.8% for K-rich glass).Footnote 3
Second, we investigate hue. Although all of the colourless glass pieces have a “white” colour (see Fig. 2), we can observe differences in hue. In colourless glass, the hue is given by iron and manganese. Iron is the main impurity of glass; it enters the glass through the sand and imparts a yellowish, greenish, or blue hue to the glass, depending on the redox conditions of the batch and furnace [33]. Manganese was added to the glass batch either to oxidise iron, leading to a more yellowish colour, or to compensate for the effect of iron colouring, resulting in a more greyish hue [49, 50].
For all the investigated colourless glass pieces, the decolouration with manganese is confirmed by the normalised a colour coordinate ranging between − 1 and − 5 and the presence of the quite well-defined Mn2+ absorption band at 430 nm (see Additional file 6: Tables S2, S4, S6, and S8 and Additional file 7: Figure S5a) [51]. Besides, although the glass pieces from both compositional groups contain similar iron contents (0.42 ± 0.08 wt% for K-rich glass and 0.48 ± 0.09 wt% for Ca-rich glass), Ca-rich glass appears slightly yellow or green (see Fig. 2). This last observation is in accordance with what historical sources report, i.e., that Ca-rich or Rhenish glass was greener [13].
The calculations of the colour differences (ΔE) in the CIE Lab colour system for the real thickness of the material highlights that Ca-rich glass could easily be distinguished from K-rich glass with the naked eye (ΔE = 5.8Footnote 4). Thus, stained-glass makers were able to recognise, to a certain extent, colour differences in the colourless glass sheets prior to composing the window, giving them the possibility of material selection for the colourless glass.
The four studied panels were assembled from glass pieces having different compositions. Not only there seems to be a link between the glass compositions and the parts of the image represented, but also between the glass composition, the forming technique, and the transparency and hue of colourless glass. We highlighted that the stained-glass makers were able to recognise differences in the glass sheets based on their surface appearance, and the transparency and hue of colourless glass. In the next section we investigate the characteristics of the coloured glass pieces.
Glass colouring
In ancient glassmaking, the chromophore palette available to pot-metal glass was limited. Nevertheless, a wide range of colours could be produced by combining chromophores in different concentrations and by applying different production processes [52]. With this in mind, one might expect that the colouring process selection was related to the targeted colour. We aim to verify if the glaziers indeed selected a colour, or a glass coloured in a specific way.
Yellow
Two distinct colouring methods were applied to obtain the yellow colour in the four panels: pot-metal glass and silver staining.
All of the yellow glass pieces of the four panels show a broad absorption tail before 600 nm, which originates from the strong absorption in the UV region and at 380, 420, and 440 nm due to Fe3+ [33]. However, six separate spectral groups (i.e., Y1, Y2, Y3, Y4, Y5, and Y6) can be recognised (see Additional file 7: Figure S5b), highlighting distinct production conditions but leading to only two colour groups. The first group of glass, which has a white, greenish-yellow hue, concerns the optical groups Y1 and Y2, which both belong to the Ca-rich glass family. Glass pieces from these optical groups present a strong Fe2+ absorption peak around 1100 nm. These glass optical groups were probably produced in oxidising conditions. Glass pieces from the optical group Y3 are also characterised by a Ca-rich composition and by a strong Fe2+ absorption peak around 1100 nm. However, the spectra show a stronger ferri-sulphide complex absorption band, typical for reduced samples [35], leading to a darker, yellowish-orange hue. Three other optical groups (i.e., Y4, Y5, and Y6) also show a strong ferri-sulphide complex absorption band and a yellowish-orange hue. Glass pieces from these groups all have a K-rich composition. The spectra are characterised by the presence of weak Co2+ absorption bands and lower Fe2+ absorption in respect to the white, greenish-yellow glass pieces. Groups Y4, Y5, and Y6 are differentiated by their ferri-sulphide and Fe2+ absorption strengths.
Pot-metal yellow glass was only used for the landscape and all the borders. Two exceptions concern the halo of Saint Clare in panel AV.1166 (glass piece 18, belonging to the optical group Y5) and the golden pastoral staff of the bishop in panel AV.1163 (glass piece 12 measured, belonging to the optical group Y5). In all the other cases, the halos and golden objects were rendered by silver stain. Silver stain was also applied to represent the hairs of the characters, the cross of the crucifixion, and only in AV.1165, for clothes, basket, and flowers related to Saint Dorothy. Silver staining is a complex colouring technology that involves metallic silver and empirical knowledge to produce the exact desired shade. Indeed, the final silver stain colour depends on numerous parameters, including the silver compound type, the base glass composition, the firing temperature, and the baking process’s duration. A variation of one parameter is considered a change of recipe. The study reveals that the silver stain was mainly applied on K-rich glass. On Ca-rich glass, the stain was applied in two specific occurrences on panel AV.1165: for the clothes of St John and on the blue glass to represent the leaves and the stem of the flower held by St Dorothy. The silver staining has a typical optical signature linked to the silver absorption band at around 470 nm, but previous analytical researches have also stressed that the silver stain technology is responsible for variations in the distribution and size of silver nanoparticles inside the glass matrix, resulting in different optical spectra and, consequently, different glass colours [37, 53]. The optical analysis of the investigated stained parts reveals three groups of colours, two darker yellow hues (orange and orange-pink), two lighter ones (greenish-yellow and yellow-green), and two blue-green hues (greenish-blue and blueish-purple) because the stain was applied on blue glass (see Fig. 2). The stains applied on blue glass will be discussed in the section on green glass.
Based on the silver absorption peak characteristics (i.e., peak position and shape) combined with the Ag:Cu ratio, we further subdivided each colour group into recipes (see Additional file 6: Tables S2, S4, S6, and S8 and Additional file 9: Figure S6). It clearly appears that the painters selected the stain recipes based on the targeted colour and on the specific iconographic part they wanted to depict. The first recipe (silver stain 1) stands for gold because it was used to render the monstrance in AV.1166. It has a pure orange hue. This stain recipe has the highest Ag:Cu ratio (7.11), indicating that the content of silver is much more important than the content of copper (with concentrations of 1372 ppm and 193 ppm, respectively). Darker, and especially orange or red, staining layers are attributed to the presence of copper [45, 54] but can be obtained from silver depending on the rate of dilution in ochre, the temperature, and number of firings [12, 54, 55]. The second recipe (silver stain 2) produces an orange-pink hue. It was applied to parts that could be red in reality, i.e., the bishop’s mitre and his cope’s embroideries (AV.1163). The optical spectra from stains 2 and 3 look alike, but silver stain recipe 3 leads to a greenish-yellow hue. This stain was also applied for the bishop’s clothing in panel AV.1163. However, the glass is slightly different because the silver stain 2 was applied on a K-rich glass K3, whereas silver stain 3 was applied on K-rich K2. The Ag:Cu ratio calculated for both stains is also different (2.62 for silver stain 2 and 0.01 for silver stain 3). However, we have to remark that the area of this stain was very small, and despite our effort to align the p-XRF spot on the stain layer, it is possible, but not certain, that the measurement area was shifted. Six other recipes lead to a greenish-yellow colour. First, the silver stain 4 was applied to render the saints’ halos and hair in both panels AV.1164 and AV.1165, and for the bottom part of the cross in panel AV.1164. This stain is characterised by an Ag:Cu ratio of 4.38 ± 0.8 and a silver peak position at 414.1 ± 0.2 nm. Second, silver stain 5 shows a close silver peak position (at 416 nm) but shows a secondary silver peak [53]. The Lorentzian-shaped silver peak is complete, allowing us to estimate the very small silver nanoparticle size to around 0.81 nm [37]. This stain was identified for the rendering of the flowers in the basket of the child accompanying St Dorothy (panel AV.1165). Another, and peculiar, greenish-yellow stain was used for the Christ and the Virgin in panel AV.1164. It is a copper-rich stain. Indeed, the Ag:Cu ratio of the third recipe for greenish-yellow stain (silver stain 6) is very low (0.20 ± 0.2), indicating that the stain contains more copper than silver in proportions, which is unexpected for a light-yellow stain. In this case, there is less probability that the p-XRF spot dimensions caused an error because the measured stained area was relatively large. Silver stain 6 has a Lorentzian-line shape peak located at 418 nm. The silver nanoparticle size of 1.9 nm was calculated from the FWHM of the peak (66 nm). The fourth and fifth greenish-yellow stains (silver stains 7 and 8) have silver peaks with a silver shape. However, the Ag:Cu ratio calculated from the p-XRF measurements are completely different. Silver stain 7 shows a ratio of 4.3, whereas silver stain 8’s Ag:Cu ratio is 0.4. In addition, the silver stain 8 also leads to a slightly different hue (yellow-green). Although silver stain 7 was used for the Virgin in panel AV.1163 and the child’s head in AV.1165, stain 3b was applied for the Christ in panel AV.1163. A last light-yellow recipe (silver stain 9) corresponds to the clothes of Saint John in panel AV.1165. This stain shows a silver peak position at 424 nm, i.e., shifted 10 nm in respect to the other stains that leads to a lighter colour. The switch in peak position could be due to the different glass matrix (here, Ca-rich) or to a different production parameter. Moreover, this stain tends more towards the yellow-green hue than the greenish-yellow hue.
Red
The red glass’s colouration is a direct result of copper. Indeed, the strong absorbance before 600 nm in the optical spectra of the four analysed red glass pieces (see Additional file 7: Figure S5c) indicates that the glass colouration is due to Cu0 [33]. In addition, all of the red glass pieces of the four panels are likely to be flashed glass. This type of red glass is more often reported as being used in stained-glass windows.Footnote 5 In addition, we observed the colourless layer on the edges of a glass piece from panel AV.1165.
The p-XRF measurements performed on the red glass pieces indicate that both sides of the flashed glass pieces were made from the same subgroup of K-rich glass (K3). However, the red glass pieces could be separated into two optical groups. Optical group R2 has a purplish-red hue (see Fig. 2), and optical group R1 has a purer red colour. The difference in colour may be linked to different iron concentrations, which is indicated by the differences in absorbance strength for Fe2+ (see Additional file 7: Figure S5c).
Blue
We observe two shades of blue. Light blue glass was used for the sky in panels AV.1164 (nine pieces) and AV.1166 (four pieces). Dark blue pieces were used for the sky (five pieces in panel AV.1163 and eight pieces in panel AV.1165) and for the clothes and the coat of arms in panel AV.1164 (6 pieces in total).
Cobalt is the blue chromophore used in the four panels (see Additional file 7: Figure S5d). When linking the blue hue with the glass composition, it appears that light blue glass always has a Ca-rich signature (optical groups B3 and B4), whereas Ca-rich (optical group B1) and K-rich (optical group B2) glass groups were used for dark coloured blue pieces. The cobalt contents measured by p-XRF highlight that to produce a deep blue colour, more cobalt is needed in Ca-rich glass than in K-rich glass. Indeed, to obtain the same dark blue colour, 1700 ppm of cobalt is needed in Ca-rich glass, versus 800 ppm in K-rich glass. In Ca-rich, 800 ppm of cobalt leads to a light blue colour. This phenomenon, as mentioned in the literature, is due to, according to their sizes, different alkali ions exerting different influences on oxygen, which reflect the Co–O bond strength [45]. Because Ca-rich glass typically contains less potassium than K-rich glass, it is de facto less efficient than other glass-compositional groups in producing a blue colour.
Purple
Visually, it was possible to distinguish two shades of purple, substantiated by the colour coordinate calculations (see Fig. 2). Four glass pieces were cut in a lighter reddish-purple or purplish-pink glass and used to represent the Virgin’s clothes (AV.1163) and those of Saint Dorothy (AV.1165). Conversely, the clothing of the child accompanying Saint Dorothy in panel AV.1165 was made from a dark blueish-purple piece, the colour of which can be described as a true purple. The same purple colour was used to render Saint Clare’s habit (AV.1166).
All purple glass pieces have a K-rich composition (K3). The difference in the observed colours is linked to the addition of a second colouring agent for the blueish-purple parts (optical group P3). Indeed, optical spectroscopy highlights the presence of Mn3+ absorption bands close to 490 nm [51] in all of the purple glass pieces; instead, the Co2+ bands around 525, 595, and 645 nm [33] are present for the blueish-coloured pieces. The Co2+ absorption bands are also present, but almost imperceptible, in reddish purple glass pieces belonging to optical group P2 (see Additional file 7: Fig. S5e). Due to its low absorption, the effect of Co2+ on the glass colour is mostly negligible. Therefore, we can conclude that the lighter reddish was obtained by colouring with manganese, whereas the darker purple glass appears more blueish because it was coloured with manganese and cobalt. The lighter and the darker blueish-purple pieces have a similar thickness, with an average thickness of 1.95 ± 0.3 mm. This means that adjusting the thickness of the material was not enough to create a variation in the observed colours. Moreover, the hues are different: the darker shade is more blueish, as indicated by its lower normalised b colour coordinate (− 11.16 ± 3.1 versus − 4.06 ± 1.8), whereas the normalised a colour coordinate, corresponding to the red component, is much higher in lighter purple glass (18.15 ± 1.4 versus 8.28 ± 1.2). Deep purple glass obtained from cobalt has only been found once, namely in a stained glass rose window from the Siena Cathedral in Italy, dating from the end of the thirteenth century [56]. No other existence of pot-metal cobalt-manganese glass from the fifteenth century was found in the literature. Technologically, the addition of cobalt to a purple glass batch typically relates to enamel productions, which only occurred from the end of the sixteenth century onwards [57].
Green
Except for one piece, i.e., the habit of Saint Dorothy in panel AV.1165, all of the 16 green glass pieces are connected to the rendering of the landscape.
One single piece (glass piece number 10 of AV.1166) was coloured with Cu2+ (see Additional file 7: Figure S5f, optical group G3). This colouring agent shows a large absorption band at 780–800 nm [33, 58, 59], giving the glass a blue-green hue (see Fig. 2). For all the other green pieces, the recorded optical spectra show high absorption values, especially around the iron and cobalt bands, indicating that the green glass pieces were coloured by iron and cobalt (see Additional file 7: Figure S5f). They all belong to the K-rich composition group (optical group G1), except one piece that has a Ca-rich composition (optical group G2).
Finally, two blue pieces were stained in panel AV.1165 with the objective of producing a green colour [55] to represent the leaves and the stem held by St Dorothy. We recognise two recipes because the shape of the silver peak appears to be different. Silver stain 10 shows only one silver peak, whereas silver stain 11 also presents a secondary silver peak. No additional information could be retrieved about these recipes because the stained layers could not be measured by p-XRF.