Fig 1.
Selection of Byzantine Glass weights representing the different typologies.
(a) Weights with cruciform monogram, belonging to batch 11. (b) Weight with box monogram. (c) Weight with box monogram enclosed by an inscription. (d) Imperial weight with imperial busts and box monogram; (e) Bust of an eparch enclosed by an inscription. Two weights produced with identical die, but different base glasses. (f) So-called Arab-Byzantine weight with pseudo-Cufic inscription. Objects and images under a CC BY-NC-SA 4.0 license with permission from the British Museum, original copyright The Trustees of the British Museum (a-d, f); and from the Bibliothèque nationale de France (e), original copyright IRAMAT-CEB Orléans.
Table 1.
LA-ICP-MS data of glass standards in comparison with published values.
Corning A corresponds to given data from [45], Nist 612 corresponds to data in [46], a after [47] and b after [48].
Fig 2.
Comparison of the results obtained with the big cell and the standard S155 cell.
The graph shows the correspondence of the experimental results for Corning A, NIST SRM612 and the two Byzantine glass weights BM S322 and BM 1983,1108.1 obtained in the big cell and the standard Resonetic S155 analytical cell. The deviation between the two analytical conditions generally remains below 10%, but exceeds 20% at low concentrations of light elements, where the sensitivity of the big cell is notably compromised.
Fig 3.
Histogram of the mass of 271 Byzantine glass weights.
The mass distribution of Byzantine glass weights from the British Museum (BM) and the Biblothèque nationale de France (BnF) coincide roughly with the weight values of Byzantine gold coinage: 4.5 g (nomisma), 2.25 g (semissis), 1.5 g (tremissis). Data were binned at 0.1 g intervals.
Fig 4.
Principal component analysis of LA-ICP-MS data.
Six compositional groups can be separated based on silica-derived elements (Mg, K, Ca, Al, Fe, Ti, Zr, Sr) as well as varying concentrations of Li and B that are associated with the fluxing agent. Principal components 1 and 2 amount to approximately 60% of the overall variability. The contribution of each element to PC1 and PC2 is indicated by the vectors (represented on an enlarged scale).
Fig 5.
Frequency of the different production groups represented among the BM and BnF collections.
The distribution of artefacts according to compositional categories is similar in the two national collections, demonstrating that this is a representative cross-section of known Byzantine glass weights. Error bars indicate the standard deviation derived from the variance of a Poisson distribution.
Fig 6.
Base glass characteristics of the identified compositional groups.
(a) Lime and alumina concentrations of the Byzantine glass weights (excluding outliers and large weights, S1 Table) indicate differences in the silica source. Asterisks indicate reduced and normalised data. (b) Correlation between zirconium and titanium. (c) Different lithium and boron levels suggest differences in the natron source. Note that the Magby group is not included in this graph due to the significant plant ash component. (d) Strontium to calcium ratios point to different strontium concentrations in the carbonate fraction of the silica source. The correlation between Mn and Sr/Ca provides evidence that manganese minerals present an additional but highly variable source of strontium in the Foy-2 and Foy-2 high Fe glasses.
Fig 7.
Soda, magnesia and iron oxide variations.
(a) The higher soda content in Foy-2 and Foy-2 high Fe possibly reflects a closer proximity to the mineral soda deposits in Egypt, whereas the elevated levels of magnesia in the Magby samples indicate the addition of a plant-ash component. Asterisks indicate reduced and normalised data. (b) Correlation between vanadium and iron of the six glass groups with the cobalt samples singled out.
Fig 8.
Trace element patterns of the primary production groups.
Average trace element concentrations for the six production groups, normalised to the mean values in the continental crust [65].
Fig 9.
Chemical characterisation of the cobalt colorant in Levantine I and Foy-2 glass weights.
Cobalt concentrations are correlated with nickel to a varying degree in Levantine I and Foy-2 glasses. The high correlation coefficient of the linear regression of the Levantine I group reflects the use of a single, homogenous cobalt source.
Fig 10.
Hierarchical clustering reveals production batches.
Batches were identified by hierarchical clustering with the programme Cluster 3.0 [71], and visualised using Treeview [72]. Pearson correlation (uncentered, with average linkage) was carried out on log2-transformed, mean-centered data that had been normalised for each element, and are shown on a false colour scale (right). Samples belonging to the same batch display highly correlated compositional profiles.
Fig 11.
Trace element patterns of two cobalt blue samples of batch 4.
The two cobalt blue samples (BM 1980,0611.38/red; BM 1980,0611.39/black) have identical base glass compositions, while minor variations in Co, Ni, Mo and As relate to the cobalt colorant. Concentrations are given as wt% (left) and ppm (right panel) on a logarithmic scale. Error bars reflect twice the relative standard deviation of the Corning A standard measurements applied to the average value of the two samples.