1 Introduction

1The present study is part of a wider review of chemical data on glass from various sites in northern and central Italy, and dated from the 4^th to the 13^th century A.D. Here focus is only devoted to the medieval period (7^th-13^th centuries A.D.), because in this time-span, a “revolution” in raw materials use becomes apparent in glass compositions, in both the Islamic world and the West (Newton, Davison 1999 ; Henderson 2013). For some reason, whether political or not, natron, the source of the alkali used in Roman and late-antique period, started to become exhausted (Shortland et al. 2006). In both areas, a search for new alkali sources appeared throw the glass compositions, and the result was that plant ash came to be used. In the West, following a period of transition of about 200 years between ca. 800 and 1000 A.D., tree ash has become the main source of alkali for the manufacture of the massive quantities of glass needed for the windows of cathedrals in Northern Europe. The ash of inland plants contains potash, which began to replace soda as the regular source of alkali (Henderson 2013). In the Middle East and Southern Europe, the alkali source is generally believed to be ash from marsh plants such as Salicornia spp., which grow on the shores of the Mediterranean and Atlantic coasts (Tite et al. 2006 ; Henderson 2013). It introduces sodium in the batch, like natron-based glass, together with quantities of potassium and magnesium in the range of up to 2-3 wt% of K₂O and MgO. This new pattern of alkali use was set for hundreds of years to come, plant ash first being imported from the Middle East to Italy (Henderson 2013).

2It must not be forgotten that part of the soda-lime-silica glass produced in Europe may have been obtained by recycling previous soda-lime cullet and/or glass tesserae, rather than by direct use of sand and natron or plant ash (Jackson 1996). In this context, it is the response to the impact of an important change in raw material supply, i.e., from glass based on natron to one based on plant ash - as discerned among the chemical compositions of Italian glasses dated to both before and after the event - that is considered here. In particular, the chemical compositions of Italian samples were compared with major natron and soda ash reference groups of the period of interest (Gratuze 1988 ; Gratuze, Barrandon 1990 ; Freestone et al. 2000 ; Foy et al. 2003  ; Phelps et al. 2016), by means of statistical multivariate methods. Various compositional groups, related to different raw materials and production technologies adopted in the time frame and geographical area considered, are identified and discussed, allowing us to advance hypotheses on the evolution of glass production technology and on main commercial routes for the period of interest.

2 Datasets

3The total dataset considered in our wider review (the original set on which the cluster-analysis was made) is composed of 1043 samples with dates ranging from the 4^th to the 13^th century A.D., furtherly subdivided into 717 from Italy and 326 from literature, which were used for comparisons with Italian samples. In the present study, two subsets of 397 Italian samples and 223 comparison samples, all dated from the 7^th to the 13^th century, are only considered and discussed (fig. 1).

4For the Italian dataset, the geographical area considered in this study is all the northern Italy and part of the central Italy, north of Farfa in the Lazio region (fig. 2). The selected sites are characterised by a chronology as precise as possible, as deduced by the relative references (fig. 2), and the majority of samples were previously characterised from the archaeological and typological viewpoints.

5In the Italian subset, chemical analyses mainly carried out on vessels and window glasses are included. Glassworking wastes and chunks are also considered, due to their particular interest to the present study, as they can provide a unique insight into glass trading and working in the area. Glass samples intentionally coloured and opacified, such as mosaic tesserae, were excluded, except for gold tesserae, due to the hypothesis that the base-glass used to produce such kind of tesserae is comparable with that used for vessels.

6The considered features for each sample are a selection of eleven chemical elements (i.e., SiO₂, Na₂O, CaO, Al₂O₃, K₂O, MgO, Fe₂O₃, TiO₂, MnO, P₂O₅, Sb₂O₃), plus the age of each sample and the site of provenance.

7The literature groups subset, considered in the present study, is composed of 326 samples, used for comparisons with the Italian glass samples and coeval in age. This subset is composed of Groups 2.1, 2.2, and 3.3 of Foy et al. 2003, Groups N-1, N-2, N-3 and N-4 of Phelps et al. 2016, group Levantine 3 of Freestone et al. 2000 and Group 3 of Gratuze 1988 (fig. 3-4).

8The considered features for each sample of the literature groups are the same adopted for the Italian dataset.

3 Cluster analysis

9From the statistical point of view, all and only the 11 chemical elements, listed in the previous section, were considered for each sample of both Italian and literature subsets. The first step was to impute missing values using two models : Amelia II and Random Forest. However, due to lack of data for Sb₂O₃, this leads not to reject the hypothesis of inconsistency of the imputed data for this feature, although Sb₂O₃ was taken into account in clustering models, only when this feature was measured. The second step was to apply some cluster analysis models to the whole dataset (Italian data plus literature data) : “top down” approach was used, following the “divide et impera” rule. This technique determined a phylogenetic tree of our cluster model. In the cluster analysis each step followed the same schema :

• Variable selection : Pearson Correlation, Spearman Correlation and Normalized Mutual Information (under hypothesis of empirical distribution and normal distribution) were taken into account to select the variables on which to perform the analysis.
• Distance selection : a distance was chosen between Euclidean, Maximum, Manhattan, Canberra, Divergence, extended Jaccard.
• Outliers : outliers were isolated according of Local Outlier Factor, Adjusted Boxplot models.
• Cluster analysis : HCA was used as main model, meanwhile Optics, Fuzzy C-Means, Affinity Propagation and Finite Normal Mixture as control models.

10Note that this cluster analysis led to a partition that includes even the age of each sample and its site of provenance, features that are not taken into account in the clustering algorithm.

11Various compositional groups, detailed in the section “Results”, were obtained. The chemical features of each Italian group are summarised into tables which report, for each element, the values to draw a Tukey box-and-whisker plot, i.e., the minimum and maximum whiskers, the lower and upper quartile and the median. To compare the chemical data of each Italian group with those of reference groups, tables with the values to draw a Tukey box-and-whisker plot are also reported for the relative reference groups, when present. For an easier visualisation of the results, Tukey box-and-whisker plots are also added for each chemical element considered in the present cluster analysis. Whenever any imputed data in a feature of the cluster are present, the correspondence element is signalized in grey in tables and box-and-whisker plots (fig. 5).

12In the tables on geographical locations, the presence of window glass, tesserae, glass working wastes and chunks is also reported for each Italian group. The remaining samples are to be intended as vessels.

13The chronological ranges of each compositional group were obtained by dividing the whole reference period (4^th-13^th centuries A.D.) into 20 bins of 50 years each (step 1), and counting how many times each bin was mentioned in the data inside the considered cluster (step 2). The frequency in each bin answers the following question : how many samples could be present in the considered bin ? Observe that the frequency of each bin is the once at the beginning of the interval (step 3). A sample was defined to be “part of the queue”, whenever its dating range overlaps, only in part, the most frequent dating range for the considered cluster. A sample was defined to be an “outlier”, whenever its dating range does not overlap the most frequent dating range for the considered cluster (step 4). It is defined as a high frequency interval the neighbourhood of the modal value of the dating range (fig. 6).

14The number of outliers for each Italian site is reported in fig. 2.

15All the references used for carrying out the cluster analysis will be reported elsewhere.

4 Results

16The results of cluster analysis allow us to subdivide the Italian dataset, dated from 7^th-13^th centuries A.D., into four macro-compositional groups on the basis of the flux used : 220 samples are produced with natron ; 44 with soda ash, 23 show an intermediate composition between natron and soda ash glass and 13 samples are obtained by using potash ash as flux. The above Italian samples may be further subdivided into various compositional groups, listed in the following subsections. Within each macro-period, the number and location of samples identified in Italian compositional groups, the chemical features and the chronological range are reported for each group. The chemistry and chronology of Italian groups are also compared, when possible, with those of the corresponding reference group. Box-and-whisker plots of Italian groups, compared with the reference groups, when applicable, are also shown for each considered element for an easier visualisation of their chemical features.

17It should be pointed out that in the tables and figures on geographical locations and chronology of each compositional group, samples belonging to the considered group, but not included into the chronological range here considered (i.e., 7^th-13^th centuries A.D.), are reported for the sake of completeness, but not discussed in the present paper.

4.1 Natron groups

18Five groups, all composed of glass samples produced with natron as a flux, are identified in Italy. The groups are here named : Ita-N1, Ita-N3, Ita-N4, Ita-N6, Ita-N7.

4.1.1 Group Ita-N1

Chemistry

19The group Ita-N1 shows high silicon (SiO₂=69.23-70.37 wt%), calcium (CaO=8.00-10.77 wt%), and aluminium (Al₂O₃=2.86-3.32 wt%), low sodium (Na₂O=13.79-14.72 wt%) and absence of manganese (fig. 7).

20This group matches the literature groups N1 (Phelps et al. 2016), previously named Levantine 1 (Freestone et al. 2000), and Fra 3.3 (Foy et al., 2003), although there are differences in manganese and antimony contents between the Italian and the literature groups. Group Ita-N1 shows antimony contents up to 0.12 wt % as Sb₂O₃ and no manganese, Group Fra 3.3 has manganese contents up to 1.10 wt% as MnO and no antimony, Group N1 shows neither manganese nor antimony (fig. 7-8). Antimony could be here overestimated by the imputation model but, giving a sense to this value, the model may suggest a possible presence of antimony in the composition of this group.

Geographical locations

21The group Ita-N1 is not very common in northern Italy, being composed of only 10 samples, most found in sites located into north-eastern area (fig. 9). The most numerous samples (5, among which also 2 wastes are identified) are located in Grado, letting us to suppose that this town was the main importer of this kind of glass, which is probably produced along the Syro-Palestinian littoral (Freestone et al. 2000).

Chronology

22In the time range from the 7^th to the 13^th century A.D., group Ita-N1 spans from the 7^th to the first half of the 8^th century with high frequency in the 7^th century A.D., in good accordance with the high frequency of group N1 of Phelps et al. 2016, dated until the 11^th century A.D. Group Fra 3.3 shows its maximum frequency from the second half of the 7^th to the first half of the 8^th century A.D. (fig. 10).

4.1.2 Group Ita-N3

Chemistry

23The group Ita-N3 is characterised by lower silicon (SiO₂=67.00-69.05 wt%), aluminium (Al₂O₃=2.45-2.51 wt%) and higher iron (Fe₂O₃=0.80-1.30 wt%) and titanium (TiO₂=0.25-0.36 wt%) than group Ita-N1, suggesting sand relatively rich in heavy accessory minerals. It is chemically quite similar to group N3 (Phelps et al. 2016), although there are differences in silicon, potassium, magnesium, manganese and antimony contents between the Italian and the literature groups. Group Ita-N3 shows silicon from 67.00 to 69.05% as SiO₂, potassium from 0.36 to 0.57% as K₂O, magnesium from 0.60 to 0.99% as MgO, antimony from 0.35 to 0.47% as Sb₂O₃ and manganese from 0.12 to 0.24 wt% as MnO, while group N3 has silicon from 68.04 to 71.64% as SiO₂, potassium from 0.17 to 0.42% as K₂O, magnesium from 0.35 to 0.65% as MgO and does show neither manganese nor antimony (fig. 11-12). Note that, as pointed out in Ita-N1 results, antimony could be overestimated by the imputation model.

Geographical locations

24The group Ita-N3 is uncommon in Italy, being composed of only 5 samples, all from Venice (fig. 13), suggesting that Venice was to be considered probably the main importer of this kind of glass.

25Chronology

26In Italy, this group shows high frequency from the 9^th to the 11^th century A.D., about one century later than reference group N3 of Phelps et al. 2016. Late sample from Venice (Ven_A-18_3, dated from the 11^th to the 13^th century A.D., Verità, Toninato 1990) is on the queue of the distribution (fig. 14).

4.1.3 Group Ita-N4

Chemistry

27The group Ita-N4 shows very low calcium (CaO < 3 wt%) and high aluminium (Al₂O₃=4.25-4.93 wt%), iron (Fe₂O₃=1.78-1.84 wt%) and titanium (TiO₂=0.48-0.55 wt%), suggesting sand low in carbonates and rich in feldspars and heavy accessory minerals. It matches the group N4 of Phelps et al. 2016, although the Italian group has higher magnesium, manganese and antimony than the literature group (MgO=0.98-1.10 wt% vs 0.81-0.87 wt%, MnO=0.04-0.15 wt% vs 0.04 wt%, Sb₂O₃=0.13-0.41 wt% vs 0.00 wt%) (fig. 15-16). Note that, as pointed out in Ita-N1 results, antimony could be overestimated by the imputation model.

Geographical locations

28The group Ita-N4 is very rare in Italy (fig. 17), but 2 samples were found in Venice-Torcello (Tor1618, Brill 1999 and Tor T1694, Verità et al. 2002), suggesting that Venice probably was to be considered the main importer of this kind of glass, as for group Ita-N3.

Chronology

29In Italy, this group dates from the 7^th to the 10^th century A.D., with high frequency in the 8^th century A.D., in good accordance with the chronology of the literature group N4 (fig. 18).

4.1.4 Group Ita-N6

Chemistry

30The group Ita-N6 shows high contents of iron (Fe₂O₃=0.58-1.33 wt%), titanium (TiO₂=0.12-0.18 wt%), potassium (K₂O=0.50-1.15 wt%) and magnesium (MgO=1.00-1.43 wt%). Manganese and antimony varies from negligible to 1.98 and 0.25 wt% as MnO and Sb₂O₃, respectively. This group is chemically similar to sub-group Fra2.1 of Foy et al. 2003, here named as Fra2.strong, although the Italian group shows higher silicon and antimony and lower iron than the literature group (SiO₂=63.00-68.32 wt% vs 62.93-65.95 wt% ; Sb₂O₃=0.01-0.25 wt% vs negligible ; Fe₂O₃=058-1.33 wt% vs 0.91-3.30 wt%) (fig. 19-20).

31The presence of antimony content let us to hypothesize that in the production of group Ita-N6, a little quantity of Roman antimony-colourless cullet or, more generally, previous glass cullet containing antimony was added to the batch, without excluding a primary Egyptian provenance for this kind of glass, as recently testified (Schibille et al. 2017).

Geographical locations

32The group Ita-N6 is very common in Italy, being composed of 104 samples coming from northern and central Italy, although the most numerous samples (>10 samples) are located in Nogara and Ovaro (fig. 21).

Chronology

33In the time range from the 7^th to the 13^th century A.D., group Ita-N6 spans from the 7^th to the 11^th century A.D. and shows high frequency in 7^th century A.D., in good accordance with the high frequency of the literature group, which dates until the 9^th century A.D.

34Late samples from Ferrara (Frr_F_1031, Frr_F_1380, Frr_F_98613, Frr_F_986A, dated from the 9^th to the 13^th century A.D., Verità, Toninato 1990) are on queue of the distribution (fig. 22).

4.1.5 Group Ita-N7

Chemistry

35The group Ita-N7 has lower contents of iron (Fe₂O₃=0.48-1.11 wt%), magnesium (MgO=0.61-1.00 wt%) and manganese (MnO=0.01-1.29 wt%) and higher antimony (Sb₂O₃=0.01-0.81 wt%) than Ita-N6. This group is chemically similar to sub-group Fra2.2 of Foy et al. 2003, here named as Fra2.weak (fig. 23-24).

36The high antimony lets us to hypothesis that in the production of group Ita-N7, as for group Ita-N6, Roman antimony-colourless cullet was consistently recycled.

Geographical locations

37The group Ita-N7 is very common in Italy, being composed of 99 samples from Italy, almost all located in northern Italy, except for 9 samples from Lazio region (Farfa) and 21 from Tuscany (San Genesio, Pieve di Pava and Pieve di Coneo) (fig. 25).

Chronology

38In the time range from the 7^th to the 13^th century A.D., the group Ita-N6 spans from the 7^th to the 11^th century A.D., with high frequency into the 7^th century A.D. The literature group (Fra2.weak) spans from the 7^th to the 9^th century A.D., with high frequency into the first half of the 8^th century A.D.

39Late samples from Venice (Ven_A-15_2, Ven_A-15_5, Verità, Toninato 1990) and Ferrara (Frr_F_1452, Frr_F_4939, Verità, Toninato 1990), dated from the 9^th to the 13^th century A.D., are on queue of the distribution (fig. 26).

4.2 Soda-Ash groups

40Three groups, all composed of glass samples produced with soda ash as flux, are identified in Italy. The groups are here named : Ita-NA1, Ita-NA2 and Ita-NA3.

4.2.1 Group Ita-NA1

Chemistry

41The group Ita-NA1, produced with soda ash as flux (K₂O=1.95-3.15 wt% and MgO=2.02-3.58 wt%), is characterised by high silicon (SiO₂=64.71-69.57 wt%) and iron (Fe₂O₃=0.37-1.13 wt%), and low aluminium (Al₂O₃=1.01-2.15 wt%), suggesting the use of high quality sand, but less pure with respect to that for the production of group Ita-NA2 (see later). This group seems to be comparable to the group Egy3 of Gratuze 1998 and Gratuze and Barrandon 1990. The main differences between Italian samples and Egyptian samples seem to be slightly higher values in silicon and slightly lower values in sodium, aluminium and titanium in Italian with respect to the Egyptian samples (SiO₂=64.71-69.57 wt% vs 64.50-67.44 wt% ; Na₂O=10.75-13.45 wt% vs 12.00-15.90 wt% ; Al₂O₃ = 1.01-2.15 wt% vs 1.40-2.55 wt% ; and TiO₂ =0.06-0.12 wt% vs 0.09-0.22 wt%, respectively) (fig. 27‑28). The imputation models suggest that Italian samples could be mixed with recycled Roman glass, even though the MnO values are underestimated and Sb₂O₃ values are overestimated. If this hypothesis is true the little differences highlighted above could be justified.

Geographical locations

42The group Ita-NA1 is composed of 29 samples from Italy, with a prevalence of samples identified in north-eastern Italy (fig. 29). The most numerous samples are identified in Grado, suggesting that this town could be the main importer of this kind of glass.

Chronology

43This group in Italy dates from the second half of the 8^th to the 12^th century A.D., with high frequency in the 10^th century A.D., while Egypt 3 dates from the second half of the 10^th century to the first half of the 12^th century A.D. with high frequency from the second half of the 10^th to the first half of the 11^th century A.D.

44Early samples from Farfa (Frf_3890, Frf_5657, Frf_5661, 8^th-9^th century A.D., Brill 1999), Ferrara (Frr_F_1262, Frr_F_1569, Frr_F_4758-1298, Frr_F_4914, 8^th-12^th century A.D. Verità, Toninato 1990), and Venice (Ven_A-18_2, 8^th-12^th century A.D., Verità, Toninato 1990), and late sample from Torcello (Tor_1622, 11^th-13^th century A.D., Brill 1999) are on the queue of the distribution.

45Late samples from Vicenza (Vic_VI226/1, Vic_VI226/2, 13^th century A.D., Silvestri et al. 2005) could be considered as outliers (fig. 30).

4.2.2 Group Ita-NA2

Chemistry

46The group Ita-NA2, produced with soda ash as flux (K₂O=2.22-2.97 wt% and MgO=2.40-3.77 wt%), is characterised by relatively high silicon (SiO₂=69.00-70.05 wt%) and low aluminium (Al₂O₃=1.12-1.42 wt%) and iron (Fe₂O₃=0.30-0.58 wt%), suggesting the use of very pure sand, rich in quartz and low in feldspars and heavy minerals.

47It is comparable to the group Levantine 3 (Freestone et al. 2000), although the Italian samples show slightly higher values in sodium, aluminium, potassium, and magnesium and slightly lower values in silicon and manganese than the literature group (Na₂O=12.90-14.23 wt% vs 11.45-13.10 wt% ; Al₂O₃=1.12-1.42 wt% vs 0.84-1.21 ; K₂O=2.22-2.97 vs 1.49-2.09 wt% ; MgO=2.40-3.77 wt% vs 2.55-2.79 wt% ; SiO₂=69.00-70.05 wt% vs 67.59-71.96 wt% ; MnO=0.33-0.93 wt% vs. 0.78-1.08 wt%) (fig. 31-32).

Geographical locations

48The group Ita-NA2 is very uncommon in Italy, being composed of only 3 samples from Italy, mainly located in Venice (fig. 33), which was probably to be considered the main importer of this kind of glass.

Chronology

49In Italy, this group dates from the 11^th to the 12^th century A.D., in good accordance with the chronological diffusion of the literature group.

50Early sample from Venice (Ven_A-15_1, 8^th-12^th century A.D., Verità, Toninato 1990) is on the queue of the distribution.

51Early sample from Farfa (Frf_3870, 8^th-9^th century A.D., Brill 1999), could be considered as an outlier (fig. 34).

4.2.3 Group Ita-NA3

Chemistry

52The group Ita-NA3 is characterised by K₂O and MgO contents generally higher than 2.5 wt% and by high phosphorous contents (P₂O₅=0.26-0.56 wt%), suggesting that they are obtained by soda ash as flux. They also show low silicon contents (SiO₂=656.75-65.45 wt%) and high aluminium contents (Al₂O₃=43.89-6.93 wt%), suggesting an impure sand source (fig. 35-36).

53This group shows close similarities with late glass produced into Elsa valley (Tuscany) (Fenzi et al. 2013).

Geographical locations

54The group Ita-NA3 is composed of 12 samples from only two Italian sites (Savona in north-western Italy, and Farfa in central Italy) (fig. 37).

Chronology

55In Italy, this group dates from the 10^th to the first half of the 13^th century A.D., with high frequency in the the first half of the 13^th century A.D.

56Early samples from Farfa (Frf_3886, Frf_3888, Frf_3889, Frf_5654, Frf_5658, 8^th to 9^th century A.D., Brill 1999) could be considered as outliers (fig. 38).

4.3 Mixed groups

57Two groups show intermediate compositions between natron and soda-ash glass, and are here named Ita-N.NA1 and Ita-N.NA2.

4.3.1 Group Ita-N.NA1

Chemistry

58The group Ita-N.NA1 has an intermediate chemical composition between natron and soda ash glass, having K₂O from 1.61 to 2.38 wt%, MgO from 1.45 to 2.63 wt% and P₂O₅ from 0.10 to 0.41 wt% (fig. 39-40). This may indicate that an intermediate glass batch was used to produce this type, one part probably composed of the “new” glass batch (with plant ash as flux) and the remaining part obtained by recycling earlier natron glass.

Geographical locations

59The group Ita-N.NA1 is composed of 18 samples from Italy, equally distributed among the selected sites in northern and central Italy (fig. 41). There is no clue to understand where this glass is produced, but it seems to be quite common in Italy.

Chronology

60In Italy, this group dates from the 8^th to the 12^th century A.D., with high frequency from the 9^th to the 11^th century A.D.

61Early samples from Grado (Gra_GRA185/2B, 5^th-8^th century A.D., Silvestri et al. 2005), and Monte St. Martino (MSM_SM3, 6^th-8^th century A.D., Uboldi, Verità 2003), are on the queue of the distribution.

62Late samples from San Genesio (SGe_11, Cagno et al. 2012b) and Savona (Sav_1429, Cagno et al. 2012a), dated to the beginning of the 13^th century A.D., could be considered as outliers (fig. 42).

4.3.2 Group Ita-N.NA2

Chemistry

63The group Ita-N.NA2 is characterised by K₂O ranging from 1.90 to 6.20 wt% and MgO from 1.83 to 3.10 wt% and by P₂O₅ from 0.10 to 1.15 wt%, suggesting that they are obtained by mixing soda and natron as flux. It also shows high contents of Fe₂O₃ (from 2.00 to 3.48 wt%) and TiO₂ (from 0.20 to 0.47 wt%), as well as of Al₂O₃ (from 2.50 to 9.10 wt%), suggesting an impure sand source (fig. 43-44). This group seems not comparable to any other literature group, even though is not too far away from group Egypt 3 of Gratuze 1988. It could be interpreted as a “garbage” group, made with low quality raw materials.

Geographical locations

64The group Ita-N.NA2 is quite uncommon in Italy, being composed of 5 samples, found in north-western Italy (Carvico and Savona) and Tuscany (San Genesio) (fig. 45).

Chronology

65In Italy, this group dates from the 8^th to the 12^th century A.D., with high frequency in the 10^th century A.D.

66Late samples from San Genesio (Sge_53, SGe_17, beginning of the 13^th century A.D., Cagno et al. 2012b) could be considered as outliers (fig. 46).

4.4 Potassium-ash group, Ita-KA1

67One group, here named as Ita-KA1, is obtained with potash-ash as a flux.

Chemistry

68The group Ita-KA1 is obtained with potash ash as a flux and consequently, it shows chemical composition different from those of both natron and soda ash groups, having low sodium (Na₂O=0.00-1.56 wt%), and high potassium (K₂O=12.00-23.76 wt%) and calcium (CaO=3.05-26.76 wt%) (fig. 47-48). This is mostly a macro-group having us no samples to compare in the dataset here considered.

Geographical locations

69The group Ita-KA1 is composed of 13 samples from Italy (fig. 49), typologically all identified as window glass or glass chunks (perhaps the so called “smoothers”). As per literature this kind of glass seems to be produced in northern-Europe and, from the data we have analysed, it seems to be uncommon in Italy, being identified in only three sites, all located in Northern Italy.

Chronology

70This group is typical of only mid-Medieval times, having dates ranging from the 10^th to the 12^th century A.D., with high frequency in the 11^th century A.D.

71Early samples from Loppio (Lop_26276, Lop_26275, 6^th-7^th century A.D., Silvestri et al. 2016) could be considered as outliers (fig. 50).

5 Discussion and conclusions

72The present paper allows us to give a perspective look to the chronological evolution of glass production technology in Italy and main commercial routes from the 7^th to the 13^th century A.D., by means of a review of analytical data on Italian samples (chemistry, geographical locations and chronology), published until now, which are compared with major natron and soda ash literature groups for the period of interest (fig. 51). It should be stressed here that all the hypothesis advanced in the present study are to be considered as valid, only within the dataset used.

73Results of cluster analysis have identified various compositional groups, which demonstrate how the Italian picture is quite complex : natron and soda ash groups are identified, together with intermediate compositions and potash ash glass. This picture should not come as a surprise, given that the time-span considered in this study covers about seven centuries and that during the mid-Medieval period the transition from the natron to the plant ash used as flux occurred.

74Starting from the natron groups, the groups Ita-N6 and Ita-N7 show the widest geographical and numerical diffusion, having been identified about 200 samples in northern and central Italy. Chronologically, these groups date from the 7^th to the 11^th century A.D., but show high frequency in the 7^th century A.D. Groups Ita-N6 and Ita-N7, for which an Egyptian provenance is attested (Schibille et al. 2017), are here interpreted as obtained also by recycling, due to antimony contents identified which let us to hypothesize that in the production of the above groups, Roman antimony-colourless cullet was consistently recycled.

75The high number of samples of groups Ita-N6 (104 samples) and Ita-N7 (99 samples) suggests both that practice of recycling was common in northern and central Italy from the Early Middle Ages onward and that commercial links were active between Italy and Egypt.

76The other compositional groups, obtained with natron as flux, are attested in Italy with few samples : 10, 5 and 2 samples are documented for groups Ita-N1, Ita-N3 and Ita-N4, respectively. Groups Ita-N1, Ita-N3 and Ita-N4 are chemically comparable to groups N1, N3 and N4 of Phelps et al. 2016, which are similar to Levantine 1 of Freestone et al. 2000 and Fra 3.3 of Foy et al. 2003, to groups Egypt 2 and Egypt 1 of Gratuze 1988, Gratuze and Barrandon 1990 (fig. 3).

77Chronologically, the group Ita-N1 shows high frequency in the 7^th century A.D., and the most numerous samples are located in Grado, suggesting that commercial links of north-eastern Italy with the Levant, where group N1 is probably produced, are active.

78It is interesting to note that the groups Ita-N3 and Ita-N4 are all found in Venice, suggesting that these kinds of glass probably enter in Italy by this city, although in different periods, from the 7^th to the 10^th century A.D. for Ita-N4 and from the 9^th to the 11^th century A.D. for Ita-N3.

79We can note that no sample, chemically comparable to group N2 of Phelps et al. 2016, similar to Levantine 2 of Freestone et al. 2000 (fig. 3), is identified in Italy until now.

80All the samples of compositional groups, obtained with natron as flux, date from the 7^th to the 11^th century A.D., with periods of high frequency quite comparable of those of literature groups, except for group Ita-N3 (fig. 51). It starts diffusing in Italy about one century later the period of high frequency of the corresponding literature group N3 (9^th century A.D. vs 8^th century A.D.).

81In Italy, the soda-ash glass starts to diffuse from the 9^th century A.D., as testified by the compositional groups identified and here named Ita-NA1 and Ita-NA2 (fig. 51).

82The most numerous soda ash group (29 samples) is Ita-NA1, chemically comparable to group Egypt 3 of Gratuze 1988, and the samples are generally identified in north-eastern Italy, with maximum chronological frequency in the 10^th century A.D. (fig. 51).

83Another soda ash group, here defined as Ita-NA3, is composed of 12 samples, all located in two sites : Savona in north-western Italy and Farfa in central Italy. The chemical composition of this group, obtained by soda ash and sand (with very high alumina content), shows similarity with the glass produced in Elsa Valley in Tuscany during the 14^th century A.D. (Fenzi et al. 2013), and this may advance the hypothesis of a possible local production for this group.

84The group Ita-NA2, chemically comparable with the group Levantine 3 of Freestone et al. 2000 (fig. 3) is few documented in Italy, with only three samples identified, and high chronological frequency from the 11^th to the 12^th century A.D. (fig. 51).

85The different relative abundance between groups Ita-NA1 and Ita-NA3, probably produced in Egypt and Israel respectively (Schibille et al. 2017), seems to suggest a preference in commercial routes between Italy and Egypt.

86In Italy, the intermediate compositions between natron and soda ash glass, here named Ita-N.NA1 and Ita-N.NA2, start to be diffused from the 8^th cent. A.D.

87The most numerous group (18 samples), generally well distributed among the selected sites, is Ita-N.NA1. It was probably obtained by mixing “new” soda ash batch “Egypt 3” and earlier natron glass. However, as ash from plants is naturally variable in composition, due to the geological and climatic conditions in which the plants grew, or to the different ways in which they were burnt, it cannot be completely excluded that species of plant, not used for samples belonging to group Egypt 3, were used to make this kind of glass. The group Ita-N.NA1 shows high chronological frequency from the 8^th to the 11^th century A.D. (fig. 51).

88Finally, it is interesting to underline that, in Italy during mid-Medieval times (11^th-12^th centuries A.D.) 13 samples, obtained with potash ash as flux, typical of Northern Europe glass production (Henderson 2013), are also identified and here grouped as Ita-KA1 (fig. 51). These samples are found in Sant’Agata Bolognese (Emilia Romagna) and Loppio (Trentino Alto Adige) and are composed of smoothers. This group is also identified in Pavia (Lombardia), but, in this case, samples are composed of window glass.

89The occurrence of smoothers obtained with potash ash as flux may advance hypothesis on correlation between this archaeological type and composition. The same does not happen for window panes, because in the present dataset this type shows also alternative compositions.

90In conclusion, the present study allows us to hypothesis that from the 7^th to the 13^th century A.D., in northern and central Italy, the practice of recycling was common and that commercial links were mainly active between Italy and Egypt for both natron and soda ash glass, although contacts with Levantine area and northern Europe are also testified.

Figures

Fig. 1 - The total dataset ; grey areas represent subsets of Italian and reference samples, considered and discussed in the present study.

Fig. 2 - Subset of Italian samples, considered and discussed in the present study. For each site, region, chronology, references, and number of samples (and outliers), also shown.

Fig. 3 - Subset of literature groups, considered and discussed in the present study.

Fig. 4 - Subset of literature samples, considered and discussed in the present study, subdivided by site. For each site, region, chronology, references, number of samples and literature groups also shown.

Fig. 5 - Tukey box-and-whisker plot (Tukey 1977).

Fig. 6 - Steps to obtain chronological ranges of each compositional group described in the present study.

Fig. 7 - Chemical composition of group Ita-N1, compared with that of literature groups Fra3.3 (Foy et al. 2003) and N1 (Phelps et al. 2016).

Fig. 8 - Tukey box-and-whisker plots of group Ita-N1, compared with those of literature groups Fra3.3 (Foy et al., 2003) and N1 (Phelps et al. 2016).

Fig. 9 - Geographical locations of Italian samples, grouped into Ita-N1.

Fig. 10 - Chronology of group Ita-N1, compared with that of literature groups Fra3.3 (Foy et al. 2003) and N1 (Phelps et al. 2016).

Fig. 11 - Chemical composition of group Ita-N3, compared with that of literature group N3 (Phelps et al. 2016).

Fig. 12 - Tukey box and whisker plots of group Ita-N3, compared with those of literature group N3 (Phelps et al. 2016).

Fig. 13 - Geographical locations of Italian samples, grouped into Ita-N3.

Fig. 14 - Chronology of group Ita-N1, compared with that of literature group N3 (Phelps et al. 2016).

Fig. 15 - Chemical composition of group Ita-N4, compared with that of literature group N4 (Phelps et al. 2016).

Fig. 16 - Tukey box and whisker plots of group Ita-N4, compared with those of literature group N4 (Phelps et al. 2016).

Fig. 17 - Geographical locations of Italian samples, grouped into Ita-N4.

Fig. 18 - Chronology of group Ita-N4, compared with that of literature group N4 (Phelps et al. 2016).

Fig. 19 - Chemical composition of group Ita-N6, compared with that of literature group Fra2.strong (Foy et al. 2003).

Fig. 20 - Tukey box-and-whisker plots of group Ita-N6, compared with those of literature group Fra2.strong (Foy et al. 2003).

Fig. 21 - Geographical locations of Italian samples, grouped into Ita-N6.

Fig. 22 - Chronology of group Ita-N6, compared with that of literature group Fra2.strong (Foy et al. 2003).

Fig. 23 - Chemical composition of group Ita-N7, compared with that of literature group Fra2.weak (Foy et al. 2003

Fig. 24 - Tukey box and whisker plots of group Ita-N7, compared with those of literature group Fra2.weak (Foy et al. 2003).

Fig. 25 - Geographical locations of Italian samples, grouped into Ita-N7.

Fig. 26 - Chronology of group Ita-N7, compared with that of literature group Fra2.weak (Foy et al. 2003).

Fig. 27 - Chemical composition of group Ita-NA1, compared with that of literature group Egy3 (Gratuze 1988).

Fig. 28 - Tukey box and whisker plots of group Ita-NA1, compared with those of literature group Egy3 (Gratuze 1988).

Fig. 29 - Geographical locations of Italian samples, grouped into Ita-NA1.

Fig. 30 - Chronology of group Ita-NA1, compared with that of literature group Egy3 (Gratuze 1988).

Fig. 31 - Chemical composition of group Ita-NA2, compared with that of literature group Lev3 (Freestone et al. 2003).

Fig. 32 - Tukey box-and-whisker plots of group Ita-NA2, compared with those of literature group Lev3 (Freestone et al. 2003).

Fig. 33 - Geographical locations of Italian samples, grouped into Ita-NA2.

Fig. 34 - Chronology of group Ita-NA2, compared with that of literature group Lev3 (Freestone et al. 2003).

Fig. 35 - Chemical composition of group Ita-NA3.

Fig. 36 - Tukey box-and-whisker plots of group Ita-NA3.

Fig. 37 - Geographical locations of Italian samples, grouped into Ita-NA3.

Fig. 38 - Chronology of group Ita-NA3.

Fig. 39 - Chemical composition of group ItaN.NA1.

Fig. 40 - Tukey box-and-whisker plots of group Ita-N.NA1.

Fig. 41 - Geographical locations of Italian samples, grouped into Ita-N.NA1.

Fig. 42 - Chronology of group Ita-N.NA1.

Fig. 43 - Chemical composition of group Ita-N.NA2.

Fig. 44 - Tukey box-and-whisker plots of group Ita-N.NA2.

Fig. 45 - Geographical locations of Italian samples, grouped into Ita-N.NA2.

Fig. 46 - Chronology of group Ita-N.NA2.

Fig. 47 - Chemical composition of group Ita-KA1.

Fig. 48 - Tukey box-and-whisker plots of group Ita-KA1.

Fig. 49 - Geographical locations of Italian samples, grouped into Ita-KA1.

Fig. 50 - Chronology of group Ita-KA1.

Fig. 51 - Chronological evolution from the 7^th to the 13^th century A.D. of Italian compositional groups identified and discussed in the present study, compared, when possible, with the corresponding literature groups.