One of the most important questions of art history is the chronology of the three doors that are preserved today. The Ravello and Monreale doors can be dated by the donor inscription and the construction of the Monreale Cathedral, but it is still unclear when the Trani door was built. According to the outdated, developmental narrative argument, the Ravello door would have been made before the Trani door, because the Trani door has more ornamentation [1,2,4,10]. New approaches also take workshop processes into consideration. It is striking that the Ravello door reuses a panel from Trani and complements it. The kneeling angels that form the round-arched finial of the Trani door also appear in the upper register of Ravello (Figure S1 and S2). However, as the Ravello door is rectangular, the angel panels were supplemented by a triangular segment in which Sol and Luna now appear. Based on this reasoning, the order of the doors would then be Trani (prior to 1179), Ravello (1179), and Monreale (around 1185/86, taking into account also the construction history of the cathedral). However, looking at the chemical composition of the metal parts and the way they were made, things get more complicated. Looking at the panels on the three doors, it is immediately obvious that most of the panels are exactly the same as those on the other doors (Table 1). This begs the question of how they could be so similar:
- Were the same molds used to produce the exact same motif on the panels of the different doors ?
- Were the same models used for the production of similar panels [4]?
- Were the motifs copied from one door to create the model for the next door?
- Was there another way of reproducing the panels from one door for the other doors [26]
- In the following, we aim to clarify the different production steps and the chronological order of the production of Barisanus’ doors from Trani, Ravello, and Monreale.
3.1 Chemical Analyses and Principal Component Analyses (PCA)
3.1.1 Chemical analyzes
Initially, a univariate approach was used to determine the average composition of the doors. All three surviving doors were made of leaded bronze. The different states of preservation, particularly in terms of corrosion, made comparison difficult. Consequently, we will focus on the alloying elements only: Cu, Sn, Pb, Zn, and Sb (for Monreale only); concerning the PCA, these elements are considered as quantitative. Other elements, such as As, Ni, or Ag, were considered as qualitative and only as present or absent.
There are significant differences in chemical composition between the doors: while the doors from Trani contain between 2-20 wt.% Sn with the majority of the points analyzed showing a content of 11-13 wt. %, the doors from Ravello contain much more Sn, between 2-22 wt.% with a unimodal frequency distribution peaking at 14-19 wt.%, while the door from Monreale contains about 2-25 wt.% Sn (Figure 3a), showing an heterogeneity of composition, probably related to a thicker and more mixed corrosion patina. The amount of Pb (Figure 3b) in the alloy also differs from door to door: Monreale shows with around 1-5 wt.% the lowest amounts of Pb for most of the analyses, partially overlapping with Trani, which shows a composition between 1-8 wt.%. Some exceptions are noticed in the frequency distribution plot for both doors. Ravello is clearly distinguished by the average Pb content, with a unimodal distribution centered around 9-12 wt.%. A particular case are the doors from Monreale: most of the original panels associated with Barisanus contain significant Sb amount of up to 7 wt.%, shown by the bimodal distribution in Figure 3c with a peak around 4-5 wt.%. Some of the panels, frames, or rosetta also contain some Zn, which does not necessarily point to a more recent reproduction of these elements, except for some points of analysis mostly from the Monreale door, clearly visible from Figure 3d.
3.2.2 PCA
A first computational analysis was performed on the entire set of data arranged in a matrix of 761 points of analyses in the rows and 8 columns for the alloying elements. Figure S5a, supplementary material shows the biplot of the analyses of the three doors: PC1 vs PC2 with a total explained variance of 52,3%. The doors can be grouped into two macroclusters, which differ for composition, coherently with the first analytical results.
The lower cluster includes both the Trani and Ravello doors, colored in blue and red, respectively, and partially some points of the Monreale door, colored in dark yellow. Its composition is clearly different from the others. As shown in Figure S5a, the partial overlap occurs along the Cu-Sn direction. Sn shows considerable variation for all doors and is not a discriminant in the PCA analysis. Differences are due to the amount of Pb and Sb, which is inhomogeneous as visible from some points distributed far from the main cloud. Particularly, the amount of Sb in the alloy appears to be much higher for the cluster that makes up the Monreale door, located in the upper part of the graph. The graph shows a few points outside the main cloud. One of these points, belonging to the Ravello door, is particularly noteworthy and is located in the upper left quadrant (iron_D9-RA) and corresponds to a decorative iron element. It has been identified as an anomaly and therefore ruled out.
A new PCA is then performed (Figure S5b). In the calculation of the principal components, the two variables Ag and As were also excluded, as they slightly affect the variance and the information carried. The biplot PC1 vs. PC2, with a total explained variance of 53.6 % shows a less clear chemical differentiation, related only to the concentration of Sb and Pb. Other anomalies were noted: the Monreale points Ro_blB2, Ro_BlC2, Ro_BlD1, which all correspond to a decorative element (rosetta) located on the lower left (Bl) of panel B2, C2 and D1, which have a high amount of Fe from 5 to 10 wt.%. This is the result of contamination by iron door nails, which led to the deposition of rust on the surface by leaching. These points were therefore excluded from the analysis.
A more systematic analysis was then carried out, focusing only on the panels that make up all the doors, to further detail any compositional anomalies or similarities. Figure 4 shows the result of the computational PCA analysis on the panels.
Excluding the anomalies and the points with a superficial heterogeneity due to a variable thickness of patina, three clusters with a defined range of coordinates can be distinguished, each cluster mostly coinciding with a door. However, some points do not have the general composition of the door and therefore need to be discussed. Panels A7, B7 and D7 of the Monreale door belong to the upper right cluster and not to the lower cluster, which describes most of the door's composition. These panels are clearly of modern manufacture; they were added during the 17th century (see above), with a Zn content above the average composition of the door between 2 and 3 wt. % vs < 1 wt. % and a null Sb content < 1 wt. %), with the highest Zn composition for the points A7_5 and A7_6 (8-12 wt.%). Looking at a variable at a time, corresponding to a single element, each door has a specific average panel composition (Figure 5), which remains consistent with what has been previously observed.
However, certain points are situated at considerable distances from the typical compositional range of the doors:
- analyses of panel II4 (points II4_2, II4_3, II4_4 and II4_6) from the Trani door demonstrate a composition consistent with the upper left cluster, which encompasses the majority of the Ravello door. Although the alloy differs from that of the Ravello door, the high Pb content (ranging from 14 to 42 wt. %) leads to an erroneous attribution to the upper left cluster.
- analysis 3 on the panel II3 (point II3_3) of the Trani door reveals a Zn content that does not align with the door's general composition, exhibiting a 8.5 wt. %, even if closer to the upper right cluster. Upon closer examination of the points of analysis, it became evident that a visual discontinuity existed, which was likely the result of a repair that was conducted during the early 20th-century restorations (see above, section 2).
- analyses 5 and 6 of the panel A7 demonstrate a high Fe content (3 wt.%), which categorizes them in the lower right quadrant. It is noteworthy that a high Ag content is observed (between 2 and 12 wt. %, not visible in the PCA), which may be related to a silvering procedure.
Once it was established that the single doors exhibited a slight but statistically significant compositional difference, further analyses were conducted on the single doors, focusing on the decorative elements, namely the frames and panels for the Monreale and Ravello doors, in order to assess a potential compositional consistency. For the Trani door we did not differentiate between frames and panels as they were cast as one unit. The PCA elaboration of each single door is visible from Figure 6 to 11.
Upon examination of the elaboration performed on the Trani door (Figure 6), it becomes evident that a main cloud is present at the center of the graph, indicating that the elemental composition of the door is highly consistent. Some points, whose coordinates are distant from the main cloud, require further examination, as they clearly distinguish the content of Sb (upper left quadrant) or Pb, Zn and Fe (lower quadrants). Panel A6 (analyses in red in Figure 6) exhibits a higher Sb content, approximately 1.5-2 wt.%, in comparison to the average alloy composition (<< 1 wt. %) utilized for the door. Furthermore, panels A7, A8, D5, D7 and II6 exhibit the same elemental content indicated in red in both the biplot and the score plot in, Figure 6a and 6b, which may indicate the utilization of a distinct alloy, compositionally compatible to the Monreale door (visible also from Figure 4), for the fabrication of these scenes. Nevertheless, it is not possible to ascertain any chronological implications related to a change of alloy supply.
Some points located in the lower quadrants (II and III) were previously discussed (II4_2 and II4_3, high Pb content; II3_3, repair). Additionally, several other points were identified: the II3_4 point is also a repair (with the same Zn content as II3_3); the C7_9 point is a modern decorative element, a rosette, with a Zn content of around 23 wt.%, which is incompatible with the ancient alloy composition; the C4_7 point is an outlier, as an anomalous Fe content of 6 wt.% is related to contamination from iron nails. Figure 7 displays the PCA performed on panels and frames of the Monreale door.
The points of analysis are arranged in two main clusters and along an imaginary line that connects Pb to Sn (component 2). Two distinct compositions can be identified according to the percentage of elements present. The cluster on the left is characterized by a higher concentration of Sn and a lower concentration of Pb, and vice versa. The majority of the points belonging to the left cluster are derived from panels, and a similar assumption can be made for the right cluster (frames). The frequency distribution plot (Figure 8) indicates that there is a statistically significant difference in the average composition of panels and frames.
This finding is consistent with the PCA elaboration, which indicates that the Sn content in the panels is higher on average with a unimodal distribution peaking at 18-20 wt.%, compared to the frames with a bimodal distribution peaking at 11-12 and 15 wt.%, suggesting a slight distinction in the composition of the alloys used. Furthermore, a partial overlap of compositions is observed in the frequency distribution, as evidenced by the partial overlap of the red points representing the frames, which are situated within the left cluster and thus exhibit a composition similar to that of the panels. Additionally, modern panels produced in the 17th century are plotted (green points of analyses in Figure 7). It is evident that, apart from a stylistic difference in appearance, they exhibit a composition that is more closely aligned with the right cluster (frames).
Furthermore, we sought to ascertain whether there existed a compositional correlation between frames bearing the same decorative motifs. A PCA on all frames of the Monreale door is presented in Figure 9. The plots presented show three different groups of frames according to the stylistic interpretation of [1].
The majority of the frames in groups 1 (black), group 2 (red) and 3 (green) exhibit a stochastic arrangement in the graph, derived from a varying chemical composition. It appears that the samples can be divided into two subgroups based on their Sb content. However, the actual differences in Sb are relatively minor, and the distinction is an artefact of the elaboration process. Exceptions are the frames of group 2 C1_4 and C6_4, which exhibit a markedly lower Sn content of2 to 5 wt.% versus 14 wt.% on average for group 2, and are produced from an alloy more akin to those employed in the other groups. It is not possible to ascertain whether these frames were produced from a different alloy or are the result of remelting. However, given the information currently available, both hypotheses appear plausible.
The PCA elaboration of the panels and frames of the Ravello door is displayed in Figure 10 The points of analysis appear to be distributed around a main cloud in the center of the graph, with some anomalies. With values of more than 25 wt. % points B2_1 and B2_3 clearly show a very high Pb content which is far from the average of the door and is likely to be related to heterogeneities of the surface, probably due to a non-uniform corrosion patina. Upon distinguishing the panels from the frames, it became evident that there was a distribution pattern towards the Sn variable. This was evidenced by the fact that the panels exhibited a slightly higher amount of the element in average (see frequency distribution plot, Figure 11), although this could not be statistically significant.
The frame on the left of panel A2 (point A2_4) and on the left of panel D1 (D1_5 and D1_8, two replicates of the same point) have a chemical composition that is significantly different from the main cloud. These frame elements, like all the other undecorated flat frame elements, were added in the 19th century. They are composed of a modern quaternary alloy (7 wt. % Sn; 2 wt.% Pb; 3 wt. % Zn). During the last restoration, these frame elements were rearranged and are now mainly found in the upper part of the left wing. The round, undecorated buttons are all from the 19th century.
3.2.3 The question of antimony in the doors of Monreale
The high levels of Sb, up to 7 wt.% by weight, in the panels of the Barisanus’ Door at Monreale raise several questions about the origin of this element. In the 12th century, antimony was not yet known as a pure element, so it must have entered the alloy in the form of a mineral: either by using an antimony rich copper ore, or by adding an antimony mineral such as Stibnite, antimony(III) sulphide ( Sb₂S₃) to the melt, or cosmelting both types of ore.
However, single objects made entirely of antimony, or high-antimony copper alloys, are already known from prehistory. The earliest Sb-rich artifacts are known from Egypt: a potential vase made of antimony from about 3000 BCE and a copper sheet plated with antimony from 2500-2000 BCE; and from Israel; some of the Nahal Mishmar hoard objects were made of Sb-rich arsenical copper [27]. Pliny the Elder, in his Natural History, describes various ways of producing antimony sulphide as early as 77 CE [28] it was used mainly for cosmetics, medicine, and coloring of glass. However, the earliest description after Pliny was written by Vannoccio Briniguccio in 1540 [29] who noted also that stibnite was imported by Venice from Germany, in order to add it to bells, allegedly enriching their sound.
Taking into account contemporary sources about Sicily’s mines - for instance Zakariya al-Qazwini (or ‘Al Qazwîni; 1203-1283) noted that “Sicily has different mines for gold, silver, copper, iron, lead; and they also have alum, antimony, vitriol, ammonia salt and mercury” [30] - we can assume that Barisanus (or his workshop), indeed, might have used local metal sources for the production of the doors, even though there is yet no documentation about mining activities in the 12th century.
Antimony is a commonly found metal in the mineral deposits of the Peloritani Mountains. It exists in the form of antimonite (Sb3S2) and is also found as a component of complex sulphosalts such as tetrahedrite ((Cu,Fe,Zn,Ag)12Sb4S13) and bournonite (PbCuSbS3). These minerals are exploited for various metals, particularly copper when found alongside chalcopyrite (CuFeS2) and bornite (Cu₅FeS₄) [31]. A noteworthy discovery was made at the Selinunte site, where a 3 kg piece of antimony (likely Sb3S2?) was found [32], indicating early mining activities in the area. Although information about local mining operations prior to the 12th century is unknown, it is documented that mining for different metals happened during the 14th century [31]. At Roccalumera, an antimonite mineralisation associated with antimoniferous and slightly argentiferous galena was also explored during the last centuries: in this case, the antimony produced contained discrete traces of arsenic, thus likely was not used for the production of Barisanus’ doors.
As we do not have any indications or contemporary descriptions that antimony was added either as mineral to copper or co-smelted with copper minerals, or added in the form of stibnite to copper, it is most likely that Barisanus (or his workshop) used local copper made from Sb-rich copper ores such as bournonite for the casting of the panels of the door of Monreale.
3.3 Manufacture of the doors
The digital twins of the doors and the resulting orthophotos provide an excellent basis for accurate, precise and repeatable measurements of the dimension of the various elements of each of the doors. The orthophotos are free of distortion, allowing us to compare the dimensions of each motif panel with each other. The Barisanus doors offer a unique opportunity to study the same motif in three different doors. Before the preparation of orthophotos of these doors the following approach would have been not feasible, and now a multitude of questions as to the foundry practices can be addressed quantitatively. Most importantly, the question of model making and mechanical means of reproducing models with a sort of intermediate mold. The method is straightforward. Several measurements of each of the motif panels were taken and subsequently compared. There is no doubt that these motif panels were made by employing some means of mechanical reproduction process, because the motif panels are too similar in themselves.
If Barisanus used a mother mold and therefore a mother model to produce all the wax models for all the doors, it is expected that the panels are all of the same dimensions. If the dimensions are differing, a more complex model would be needed to explain their manufacture.
Solid shrinkage, or volume reduction, can be observed in almost all metals that cool down to room temperature after solidification. It is not to be confused with liquid shrinkage, which happens in the process of the transition from the liquid to the solid state, and can be compensated for by an adequate gating and feeding system. The solid shrinkage we are concerned with here is the amount of volume reduction taking place after solidification. It is directly linked to the coefficient of expansion. Depending on the geometry of the model and the mold, there may be shrinkage obstacles such as ribs or plungers which will result in different shrinkage values. To compensate solid shrinkage modern foundry practice anticipates this by increasing the dimensions of the casting pattern or casting model. This is called the shrinkage allowance and it is specific for each material. Shrinkage is therefore the key to ordering the production sequence of the panels: The smaller the dimensions, the later it was cast.
Next to the solid shrinkage of the metal, there are two more factors to be considered, the dry shrinkage of the mold material, as well as the liquid and solid shrinkage of the bees wax [26]. It suffices here to concentrate on the solid shrinkage of the cast objects, as this presents the minimum of the expected difference in the dimensions of model and cast. Here the use of beeswax is assumed, although we know from Theophilus [33] that “adeps” was used for large objects such as bells (for a discussion see [34]). Adeps may be translated with tallow and its properties are sufficient if the models are simple, but it does not lend itself to either a warm climate, or intricate modeling operations. Both restrictions apply for Barisanus and his workshop.
Table 2 shows the dimensions of two of the ten motifs examined which are present on all three doors. In some cases the panels are present twice on the same doors. These were selected to show that the method may yield promising results as to the question regarding the production sequence of the individual motif panels. For ease of comparison the percentage of the dimensional difference is always calculated in relation to the smaller of the two panels in question. This is in line with modern foundry practice where the shrinkage is expressed in relation to the cast object and not in relation to the dimensions of the model.
We can see that the Trani motif panels are the largest in both cases. This is also true for the other eight Trani motif panels measured. For further details and a detailed discussion of the expected variability in shrinkage values, see [26]. From a foundry perspective, these are the panels that were cast first. For St George, the respective dimensional changes suggest the sequence: Trani – Ravello – Monreale. Dimensional differences range from 1.5 to 2.5 ±0.2 %. This agrees well with the solid shrinkage values reported for copper-based alloys. The dimensional change from the Ravello to the Monreale door is somewhat smaller but still plausibly explained by solid shrinkage. As St George appears twice in the Ravello doors, it was possible to compare the dimensions of panels (Figure 12). Table 2 shows that these fall quite well into the range of the measurement uncertainty. It can therefore be concluded that they were made from the same model.
Looking at the St John the Baptist panel (Figure 13), it is immediately clear that the production sequence is different in terms of dimensions: Trani – Monreale – Ravello. The dimensional changes are also greater, and it seems plausible that the Monreale panels might not have been cast off the Trani panels directly, but were cast off from a door that no longer exists, presumably from the Bari doors. Two explanations seem plausible at this stage:
- The Monreale motif panels were cast off panels which were made by using the Trani panels, and are now lost;
- The Monreale copies were replicated by another method resulting in greater shrinkage.
The Ravello panels of St John do appear to have been replicated from the Monreale panels.
Looking at just these two panels, it is clear that a simple hypothesis for the production sequence of Barisanus’ doors can no longer be entertained. The other eight panels follow the same pattern. Trani was produced first, followed by Ravello and Monreale, and in half the cases the Ravello panels were produced before the Monreale ones. In the other half of the panels, the order of Ravello and Monreale is reversed [26].
The workshop used at least two different methods of joining panels and frames. In Trani's case, the frame elements were part of the panels (i.e. cast together). These 'frame-panels' were nailed to the wooden support, overlapping in part the neighboring panels. On the other hand, the panels and frames of the doors at Ravello and Monreale were made independently. The panels were nailed to the wooden support, the frame elements placed over the gaps between the panels were also nailed to the wooden support, and finally, at the corners of the panels where the frame elements meet, large, decorative bosses were nailed to the wooden support to cover the joints of the frame elements [1,13].