In the present study, 53 fragments of core-formed glass vessels and 3 fragments of glass beads are investigated. The samples come from various vessels types. Specifically, the vessel assemblage consists of 25 alabastra, 8 oinochoai, 7 amphoriskoi, 4 hydriskoi, 1 aryballos and 8 unidentified glass vessels. The beads assemblage consists of 2 eye beads and 1 bead with white decorative trails (Table 1). The majority of glass vessels bear opaque white or/and yellow decorative zigzag or straight trails, while there are few examples of blue and turquoise decorative trails (Fig. 2).
Small pieces (2–3 mm) were cut from the vessel fragments using a diamond cutting disc. The samples were mounted in a resin block. The resin block was ground with silicon carbide paper of various grits (600, 800, 1200, 2500) and then polished using diamond paste of 6--3 and 1 μm.
Scanning electron microscopy/energy dispersive X-ray spectrometry (SEM/EDX)
Scanning electron microscopy was the analytical technique used for the detection of major and minor elements of 45 glass samples (42 vessels and 3 beads). A JEOL (JSM-6510LV) scanning electron microscope, coupled with an energy-dispersive X-ray spectrometer Oxford instruments, was used. All samples were analysed under high vacuum, with an operating voltage at 20 kV and working distance for each sample of 15 mm. The calibration of the system was performed with geological standards and the accuracy/precision it was established by analysing standard reference materials (NIST SRM620, SRM1831 and SRM612). The analyses are in close agreement with the expected values and are presented in Table 2. The relative error between the expected and measured values is approximately 5 % for most of the oxides. Due to submicrometer beam size, 5 analyses of 300 s were performed on each sample, and the mean value was calculated for each element.
Electron probe micro analysis (EPMA)
Electron probe microanalyses were carried out by Dr. Andy Tindle in order to detect major and minor elements in 11 fragments of core-formed vessels (see Table 5). A Cameca SX50 is located at the Department of environment, earth and ecosystems, The Open University was run at 20 kV accelerating voltage and 20 nA beam current. Quantitative analyses were produced for all major, minor and trace elements detected. Analyses were performed using a defocused electron beam of 20 μm in diameter so as to prevent the volatilisation of light (low atomic number) elements such as soda and magnesium. Multi-element glass standards (Corning glass standards and NIST) were analysed on a regular basis so as to establish and monitor the accuracy and precision of the machine. Table 3 provides a comparison of quoted oxide results compared with our measured results with associated standard deviations. The levels of detection varied from 170 ppm for CaO to 1200 ppm for CuO. A ZAF programme was used to correct and quantify the results.
Laser-ablated inductively plasma mass spectrometry (LA-ICP-MS)
Trace element analysis was performed by laser ablation—inductively coupled plasma—mass spectrometry (LA-ICP-MS). The LA-ICP-MS instrument consisted of a NewWave UP193FX excimer (193 nm) laser system with built in microscope imaging coupled to an Agilent 7500 series ICP-MS. Laser ablation craters were set at 70 μm, the laser being fired for 45 s at 10 Hz and a typical fluence of 2.8 Jcm−2. Data was collected in a time resolved analysis mode, with a gas blank being measured before a series of ablations on glass samples, calibration standards and quality control standards, were carried out. Calibration standards bracketed the samples and QC over a period of 1 h or less.
Calibration of the system was performed using NIST SRM610 trace element glass standard. The measured and expected values are presented in Table 4. NIST SRM612 was used for quality control purposes.
Results and discussion
Type of glass
All Satricum samples are of a soda-lime-silica type (Sayre and Smith 1961). Silicon dioxide (SiO2), the main glass former, varies between 68.61 % wt. and 74.01 % wt. with a mean value of 70.58 % wt. The main source of SiO2 can be either sand or crushed quartzite pebbles. Sand, a less pure source than quartz, exhibits elevated amounts of impurities such as aluminium oxide (Al2O3) and iron oxide (Fe2O3) (Nicholson and Henderson 2000). In the present study, the mean values of Al2O3 and Fe2O3 are 2.08 % wt. and 0.24 % wt., respectively, close to the typical values of such impurities found in sands; although we are not suggesting this as a source, sand from the Belus river on the Levantine coast is thought to have been a suitable sand source used during antiquity, containing 2.98 % wt. Al2O3 and 0.325 % wt. Fe2O3, values that do not change significantly the base glass composition (Brill 1988). When evaluating potential sand sources, it should be considered that they exhibit variability in their chemical composition and not all sands are suitable for glassmaking (Brems et al. 2012; Degryse 2014).
Potassium oxide (K2O) and magnesium oxide (MgO) are found in concentrations below 1 % wt., indicating that natron was the flux used to introduce a mean sodium oxide (Na2O) level of 17.55 % wt. The evaporite deposits consist of sodium-rich minerals such as natron (Na2CO3.10H2O) and/or trona (Na2CO3.NaHCO3.2H2O) (Shortland 2004; Henderson 2013). Calcium oxide (CaO, with a mean value 6.90 % wt.) can be introduced in the glass either but unlikely as a deliberate additive or usually as in our case as an impurity in the form of marine shells in coastal sands (Henderson 2013) or freshwater shells in river sands.
The majority of samples have a deep blue colour. Their colouration is attributed to the simultaneous presence of both cobalt (Co) and copper (Cu) which have mean values of 1132 and 1713 mg/kg, respectively. Two samples have a turquoise colour because of the elevated amount of Cu (5542 and 5307 mg/kg), while four samples are brown but without significantly high values of colourant elements (e.g., Fe, Mn, Cu). Their colouration is probably due to varied furnace atmospheres (Henderson 2000). These samples have elevated boron concentrations (above 350 mg/kg), which is an additional chemical discriminant, while the mean value in boron of the remaining samples is 177 mg/kg. Opacity in yellow and white glasses is due to the presence of antimony Sb and lead Pb which have mean values of 717 and 2049 mg/kg, respectively. Calcium and lead antimonate crystals are responsible for the opacity of ancient glass (Lahlil et al. 2008).
Al2O3 is present in varying amounts in most glass artefacts in antiquity. It is introduced in the glass batch most of the time as an impurity in the sand used in glassmaking (Jackson et al. 2005; Nicholson and Henderson 2000); we should not exclude its deliberate addition (Beltsios et al. 2012) which is not the case in the samples in this study. Therefore, Al2O3 is a possible way to differentiate between primary sand sources. A bi-plot of SiO2 and Al2O3 in Satricum samples and coeval samples from Spina, Italy (Arletti et al. 2011) and Rhodes Island, Greece (Triantafyllidis et al. 2012) is shown in Fig. 3. Furthermore, in the same figure, the concentrations of two hypothetical glasses derived from suitable Italian sands for glassmaking, according to Brems et al. (2012), are plotted.
All samples can be divided in 3 basic groups according to their SiO2 and Al2O3 levels. The first group (group A) has values of SiO2 and Al2O3 above 68 % wt. and below 1.5 % wt., respectively, consisting of 7 Satricum samples and 2 samples from Spina (low Al2O3 group). The second group (group B) has values above 68 % wt. and 1.5 % wt. respectively, consisting of the majority of Satricum samples (40 samples out of 56) and 6 samples from Spina, showing elevated values in both elements (Table 5). Finally, the third group (Group C) has values below 68 % wt. and above 1.5 % wt., respectively, consisting of all the Rhodian samples with the addition of 8 Satricum samples and two samples from Spina. The sample close to group A can be considered either as a group A sample with marginal SiO2 content or as an outlier (SAT.11).
Evidence for the use of different sand sources can be derived from trace element characterization. Forty-five out of the total of 56 samples were analysed by means of LA-ICP-MS for the determination of their trace element compositions. The 45 samples analysed are part of group A and B samples (Fig. 3, Tables 1 and 6). Unfortunately, due to sampling criteria, none of group C samples were analysed by LA-ICP-MS to identify their trace element composition. By plotting Nd against Sr, an interesting correlation can be seen (Fig. 4). There are two groups of samples with positive correlations between the two elements but with different slopes. Samples with a higher slope are part of the samples that form group A (see above) with only one addition (sample Sa.23), which has a basic glass composition similar to group B samples (high Al2O3, high SiO2). A similar positive correlation between these two elements was found in plant ash Late Bronze Age glass from Egypt and Greece (Henderson 2013). It has been suggested that the latter correlation was derived from the fact that plant ash was contributing to the excess of Nd and therefore the combination of plant ash and sand created this correlation (Henderson et al. 2010; Henderson 2013). This study is different because natron-based glasses have no contribution from plant ash so a different explanation needs to be sought. Sr is an element associated mainly with Ca and by implication primarily derived from shells in the sands used; however, a contribution from feldspars or/and heavy minerals in the silica source cannot be excluded (Henderson et al. 2005; Degryse et al. 2006, 2010). Nd is presumed to be related to accessory minerals such as zircon present in sand (Brems et al. 2014). Thus, differences in the Nd/Sr ratio between the two groups could reflect use of sands with different proportions of “contaminant” minerals.
Strontium (Sr), barium (Ba) and zirconium (Zr) are three elements, associated with various minerals in rocks or sediments and hence in derived sands. Their concentration varies and reflects the local geology of the sand precursors. Zirconium is expected to be primarily present as zircons in sands while Sr is connected with the presence of Ca which derives from shells (aragonite) or/and limestone. Barite (BaSO4), which is found as concretions in sands and sandstones, is the likely main mineral in sands. These three elements can therefore be possible independent markers for differentiating sand sources used to make glasses.
In Fig. 5, a bi-plot of Sr versus Zr concentrations, samples divide into two groups. The majority of the samples form a group with low Zr values (29–42 mg/kg) and varying Sr compositions (270–562 mg/kg). Nine samples have elevated values of Zr (above 50 mg/kg). Five of them (yellow diamonds) are part of group A samples and show a positive correlation. This clear distinction, between part of group A and the majority of group B samples, strengthens the idea that they were manufactured with different raw material sources. Four samples have values “between” the two main groups. This could reflect a third group or could be a result of recycling. This distinction is also obvious in our Ba against Zr plot (Fig. 6) where the majority of samples have elevated Ba and lower Zr values while there are some samples with elevated Zr values (∼50 < Zr < 130 mg/kg) suggesting use of different sands (Shortland et al. 2007; Henderson 2013). The five samples (yellow diamonds) are again positively correlated while the four “recycled” have a distinct position in the plot. A comparison with relative levels of barium and zirconium in late Hellenistic and early Roman vessel glasses shows a clear distinction from many Satricum samples, especially for Satricum samples containing below c. 200 mg/kg (Thirion-Merle 2005; Henderson 2013, Fig. 8.3).
Among the 56 glass samples discussed in this paper, there are 53 fragments of glass vessels and 3 fragments of glass beads, the balance being sub-samples from the same objects of different colours. The 53 fragments are dark blue (47), brown (4) and turquoise (2); the 3 bead samples are dark blue. The deep blue colour, in all the vessel and bead samples, is associated with the presence of cobalt (Co) and copper (Cu) in varying concentrations, while the turquoise colour in 2 fragments is associated with high Cu levels. The 4 brown samples do not show anomalous values in any of the colourant elements so it can be assumed that their colouration is due to controlling the furnace atmosphere and therefore will be excluded from the following analysis.
As it can be seen in Fig. 7, the majority of samples have elevated values of both Co and Cu suggesting that they belong at the CoCu cobalt colouring glass category identified by various scholars for earlier glass samples (Shortland and Eremin 2006; Shortland et al. 2007; Smirniou and Rehren 2013). They are positively correlated with a Co:Cu ratio close to 1. There is also a cluster of samples (yellow ellipse) (Sa.14, Sa.36, Sa.37, Sa.38 and Sa.39) which have lower Co values. Also, samples Sa.5, Sa.21 show low Co (91 and 86 mg/kg) and high Cu contents (5542 and 5307 mg/kg) which is to be expected since they are turquoise. Five samples (Sa.4, Sa.16, Sa.41, Sa.42 and Sa.45) can be considered as Co-blue glasses since their Cu values are below 850 mg/kg, as suggested by Smirniou and Rehren (2013). The two glass beads (samples Sa.43–44) show high Cu content (above 3000 mg/kg) and especially Sa.44 show high Co content (2683 mg/kg). The two beads, as it is seen on Fig. 7, do not fall in the same CoCu line like the majority of samples. Triantafyllidis (2001) has suggested that glass used to make beads was often recycled. Furthermore, Pliny the Elder in Natural History (XXXVI. 199) writes about recycling activities associated with the manufacture of glass beads. Therefore, the glass used for these two beads could be a result of recycling.
There are various minerals rich in cobalt, that have been proposed as colourant sources in ancient glasses, such as cobaltite (CoAsS), absolane (a mixture of MnO and CoOOH), trianite (2Co2O.CuO.6H2O) and skutterudite ((Co,Ni,Fe)As3) (Henderson 2013). Another common source of cobalt is cobaltiferous alums which are found mainly in Egypt but there are also alum ores in Iran, Turkey and Germany as well (Kazmarczyck 1986; Henderson 2013). Therefore, elements that can be associated with either minerals or alums include Al, Mn, Ni, Zn, As, Fe. In Fig. 8, the correlation between Mn and Co for vessels and beads of deep blue and turquoise colours is presented.
According to the plot, three major groups of samples can be distinguished, while there are five samples which are “outliers” (falling in the upper, lower and left part of the plot). The first group (n = 9) has low Mn values (between 100 and 200 mg/kg) and variable Co values (700–1550 mg/kg). The majority of the samples (n = 19) fall in the centre of the plot having elevated values of both Co and Mn while the third group (n = 5) has low Co and high Mn contents. We can show that three different cobalt sources were used for the coloration of these samples.
Concerning the “outliers”, all of them are group A samples except from the blue bead (Sa.44), which has high Mn content (2341 mg/kg) and elevated amount of Co (2683 mg/kg Co). Combined with the fact that it has a notable level of Cu (4495 mg/kg), its composition can be considered to be a result of recycling. Among group A samples, there are two turquoise samples (Sa.21 and Sa.5) having low Co values (their coloration is due to high levels of Cu: 5307 and 5542 mg/kg, respectively) and lastly there are two samples with the highest amount of Mn (Sa.4 and Sa.16), which is another indication that group A samples were manufactured not only with different primary raw materials (sand, see Fig. 3) but also with different secondary additives, indicating a totally different glassmaking tradition.
Excluding the outliers, if we plot the rest on a ternary diagram, the correlation between Mn, Co and Ni becomes clear. In this diagram, we can note that the samples fall into three different groups with low, middle and high values of Mn. The high and low Mn samples seem to coincide and fall almost precisely into the two groups that Abe et al. (2012) have identified (the yellow ellipse in Fig. 9). The yellow ellipse in the middle of the graph is for samples coming from Egypt of the eighteenth Dynasty and therefore we can assume they have a totally different source of Co. According to Abe et al. (2012), Co in the lower group of samples is deriverd from Mn-rich cobaltiferrous ores such as asbolane which was mined at Iran. It is rather unclear what the Co source of the other two groups (with middle and low Mn content) is but cobaltiferrous alums from the Dakhla and Kharga oases in Egypt should be excluded since they show different chemical compositions in these trace elements (Shortland et al. 2006). Furthermore, differences in Mn values of cobalt blue glasses were also noted by Henderson (2000), where the chemical analyses of European Iron Age blue glass revealed a change in Co source in the second c. BC, from a Sb-rich source to a new Mn-rich source. In our assemblage ,there is no correlation between Sb and Co, and we can assume that Sb was introduced in the glass as an opacifying agent rather than as an impurity of the Co source.