Tracing glass production in urban centers along the Silk Roads in the early Islamic period

Summary

For most of the 1^st millennium CE, glass was produced in a few industrial centers in the eastern Mediterranean and traded throughout the ancient world in the form of raw glass and finished objects. The chemical variability of glass is therefore limited, allowing economic and cultural exchange networks to be traced from production to consumer sites. This system of production and trade changed toward the end of the millennium. Using elemental analyses of glass from Merv (Turkmenistan), we reconstruct the transformation of the glass industry in the 9^th century. Our data show that raw glass no longer traveled in large quantities over long distances, but that primary productions multiplied in urban centers along the medieval Silk Roads. We propose that the old model of a globalized glass trade disintegrates by the 9^th century in favor of a more flexible and diverse production model, reflected in a variety of localized compositional groups.

Graphical abstract

Highlights

• •Centralized production and globalized trade of glass breaks down in the 9^th century CE
• •Analytical data reveal the multiplication of primary glassmaking in cities along the Silk Roads
• •Urban centers as possible stepping stones for the transfer of glassmaking technologies
• •Raw glass no longer traveled in large quantities over long distances in the Islamic world

Analytical chemistry; Spectroscopy; Archeology

Introduction

Roman and early medieval glass has a strikingly consistent compositional fingerprint due to large-scale primary production in Egypt and the Levant and the use of mineral soda (Na₂CO₃ · NaHCO₃ · 2H₂O) as the fluxing agent.^1,2 In the ninth and tenth centuries, the manufacture of raw glass in the Islamic world underwent a remarkable diversification and change in recipe; soda was now obtained from the ashes of halophytic plants, a tradition that had long been established in the western provinces of the Sasanian Empire.^{3,4,5,6,7,8,9,10,11} Following the model advocated by Sayre and Smith,¹² according to which glass can be categorized into compositional and by extension into geographical groups, scholars have attempted to define supra-regional production zones of Islamic soda-rich plant ash glass (e.g.,^7,13 and references therein), despite their much greater variability in composition. In the attempt to identify regional patterns, previous studies have often ignored or underestimated the possibility that we are dealing with a multiplication of highly localized glass productions, resulting in a patchwork of unique compositional groups. This corresponds with the spread and development of glazed and slip-painted pottery across Iran and western Central Asia in the early medieval period.¹⁴ In view of this, the question arises as to the significance of regional production zones and whether the concept of provenance is still archaeologically and historically meaningful. Crucially, whereas ceramic production sites are easily recognized through their kilns and kiln debris, the equivalent places of glass making and working have not been traced archaeologically as yet in these regions.

A global perspective and large-scale approach are needed to recognize regional specificities, and compositional mapping has the potential to detect general patterns in the types of glass from different regions.^7,13 In the present study, we have analyzed 146 glass fragments found at Merv (Turkmenistan) using laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). This represents a 100% sample of an assemblage excavated in one part of Merv and dating to the ninth century. Based on the compositional data in direct comparison with data from other urban centers along the northern Silk Roads, notably Nishapur, Bukhara, and Samarqand, we argue that the prevalence of specific compositional groups identified among the glass finds from Merv reveals the existence of local production and a limited regional distribution of these local glass groups (for ubiquity analysis see¹⁵). These groups can be clearly distinguished from glass produced in the western regions of the Islamic realm (e.g., Spain, the Levant, Egypt, Mesopotamia), so that area-specific characteristics can be defined. The picture that emerges shows that most of the major urban hubs along the Silk Road developed their own primary and secondary glass production during their respective periods of prosperity. Glass was and remained a material of trade and exchange, although probably more in the form of finished objects than in the form of large quantities of raw glass, either in their own right or as containers for sought-after commodities.

The scale of production and the way in which glass traveled during the early Islamic period has important implications for the interpretation of glass finds from consumer sites and the type of research question that can be addressed by the analytical study of archaeological glass assemblages. The system of production appears to have changed in the long term following the introduction of new plant-ash based recipes. In certain geographical areas, plant-ash glass was still produced on a large scale for the export in the form of raw glass, as evidence from Tyre (Lebanon) shows.⁸ However, this centralized production system inherited from antiquity was over time and at least in some areas replaced by a small-scale approach to the manufacture of raw glass geared to a more local or regional market as the analytical data from ninth-century Merv demonstrate. It is likely that the glass raw materials were extracted from geographically nearby areas, giving the glass a sufficiently diagnostic profile of trace elements and rare earth elements. Variations of the raw materials as well as mixing and recycling or the addition of additives can blur the chemical fingerprints of glass. To better understand the system of the production and consumption of glass in the early Islamic period, the relationship between compositional characteristics of vitreous materials from different sites therefore needs to be re-assessed in light of the variations in raw materials and the extent to which these characteristics are transferred to the finished objects.

Our analysis of the glass assemblage from the ancient city of Merv (Arabic Marw or Marw al-Shāhijān) in present-day southern Turkmenistan explores its relationship to other glass finds from the ninth century. The Merv oasis is connected via long-established trade routes with Bukhara and Samarqand to the northeast, and the Iranian plateau to the southwest (Figure 1). This highly developed and agricultural oasis was occupied since the Bronze Age and became a frontier zone of successive Iranian states (Achaemenid, Parthian, Sasanian) from the sixth century BCE onwards. After the assassination of the last Sasanian ruler, Yazdgard III (r. 633–651 CE) on the edge of Merv, the city surrendered to the Arabs and became a springboard for Muslim armies penetrating deeper into Central Asia and northern Afghanistan. Written sources from this period survive that celebrate the city as a center of many crafts such as metalworking and textile production. Its urban fortunes declined after a devastating invasion by the Mongols in 1221 CE.

Figure 1.

Map of Khurasan and Mawarannahr, indicating the location of Merv in the context of other urban centers along the Silk Road network in red

Created using world map from ArcGIS hosted by Esri, USGS (https://worldmap.maps.arcgis.com/apps/mapviewer/index.html), including Silk Road layer (26ef53db4a80441cb9e3fe93d9e71caa; V-07/03/2024).^16,17,18

The walled portions of the massive site cover an estimated 12 km² and have been the subject of almost continuous archaeological research since the 1890s. Amongst these was the International Merv Project, a collaboration between the former Academy of Sciences of Turkmenistan, University College London, and the British Museum, with excavations and systematic surveys carried out across the site between 1992 and 2000 (St J. Simpson, monograph forthcoming). Important results for the early Islamic period included the location and excavation of part of an industrial quarter within Gyaur-Kala, once the Seleucid-Sasanian city but by this period a suburb of the 400-hectare Islamic site of Sultan-Kala immediately to the west. Early Islamic pottery kilns and a workshop specializing in the production of crucible steel were explored in Gyaur-Kala, and the latter was dated by the stratified coins and pottery to the ninth century CE.^19,20,21 The assemblage also included fragmented glassware of different forms, including window glass, beakers, and bottles (Figure 2). These differ in form and decoration from the earlier Sasanian glassware, analyses of which indicate it to have been brought from Mesopotamia.²² Despite the modest status of the excavated workshop, the frequency of glass far exceeds that of the earlier periods, corresponding to a broad trend seen at other early Islamic sites across Iran and western Central Asia. With the permission of the Turkmen authorities, the assemblage was exported in its entirety to allow fully quantified scientific analysis of the compositions with the primary intention being to seek where this glass was produced.

Figure 2.

A selection of glass samples from Merv analyzed in this study

© The Trustees of the British Museum. Shared under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license. Further samples can be viewed on the British Museum online collection database (https://www.britishmuseum.org/collection).

Results

The elemental composition of the glass samples from Merv (Table S1) was determined by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) at IRAMAT-CEB in Orléans (France) using a well-established protocol.^23,24,25 Full details of the experimental procedures, calibration and quantification of the data are given in the STAR Methods section.

Base glass types

All of the glass samples analyzed in this study from early medieval Merv are without exception soda-rich plant ash glasses (Table 1) that have relatively high soda (>11 wt %) as well as elevated magnesium and potassium oxide contents (>1.5 wt %) with at times very high magnesia concentrations (>7 wt %). Phosphorus, in contrast, is surprisingly low in some cases (0.07–0.61 wt %). Early Islamic soda-ash glasses can be subdivided on the basis of elements characteristic of the plant ash component such as Na, Mg, P, Cl, K, Li, Rb, and Cs, while useful discriminators related to the silica source are Al, Ti, Cr, Zr, Th and rare earth elements (REEs).^7,13,25 The glass finds from Merv are highly variable in terms of the minor and trace element concentrations that cover an entire order of magnitude (Table S1). Alumina, for example, ranges from about 1 wt % to over 10 wt %, titanium lies between 270 ppm and over 2000 ppm, chromium between 6 ppm and 170 ppm, and zirconium between 21 ppm and 225 ppm (Table 1). Based on the alumina contents, Cr/La and Th/U ratios that indicate variations in mineral impurities in the glassmaking sands, as well as the manganese oxide concentrations, we can clearly divide the data into several compositional groups and thus into different primary production events (Figure 3). 29 samples have remarkably high Cr/La ratios, especially in relation to the alumina contents, resembling glass of Mesopotamian origin.¹³ However, most of the samples from Merv have low Cr/La ratios (<4), for which a Central Asian provenance can be assumed. Given their prevalence in this assemblage, a local production is proposed. An exception is a small group of samples with particularly high Al contents that show very different trace element patterns (Figure 4A), the origin of which is not known. In the following, we have subdivided the results into local glass types (Merv G1-G3) and imports (Merv Meso1-2, Merv Sam, Merv high Al) for the sake of clarity, even if this anticipates part of the discussion.

Table 1.

Mean concentrations of major, minor, and selected trace elements obtained by LA-ICP-MS for the different glass groups from Merv

	Na2O	MgO	Al2O3	SiO2	P2O5	Cl	K2O	CaO	TiO2	MnO	Fe2O3	Li	Cr	Cu	Rb	Sr	Zr	Cs	Ba	La	Ce	Hf	Ta	Pb	Th	U
Merv G1 (n = 42)	14.9	4.09	5.93	59.8	0.44	0.49	4.68	8.01	0.22	0.04	1.24	24.3	27.4	90	43.7	409	97.6	1.09	293	10.7	19.7	2.37	0.25	34.0	2.94	0.96
stdev	0.7	0.60	0.61	1.9	0.04	0.11	0.37	1.03	0.03	0.01	0.22	2.0	3.9	268	5.3	64	12.9	0.15	84	1.1	2.0	0.28	0.03	124	0.28	0.15
Merv G2 (n = 23)	14.3	3.78	3.29	64.5	0.38	0.57	3.99	8.08	0.13	0.08	0.75	16.3	16.5	189	30.7	486	68.3	0.65	185	7.57	14.3	1.67	0.15	30.1	2.01	0.63
stdev	1.8	0.62	0.70	2.0	0.05	0.18	0.61	1.20	0.04	0.08	0.23	3.8	5.0	211	6.4	162	18.2	0.17	47	1.81	3.2	0.41	0.04	41.3	0.43	0.13
Merv G3 (n = 27)	14.0	3.94	2.60	65.1	0.40	0.54	3.82	7.65	0.09	1.00	0.65	14.6	13.4	320	25.9	472	50.3	0.54	446	6.66	11.6	1.28	0.12	182	1.65	0.75
stdev	1.2	0.56	0.35	1.3	0.03	0.06	0.21	1.07	0.02	0.35	0.09	2.0	4.0	360	4.0	93	16.5	0.12	250	0.88	1.4	0.35	0.02	146	0.21	0.15
Merv Meso 1 (n = 14)	15.4	3.77	3.08	63.0	0.34	0.64	2.70	8.79	0.16	0.81	1.04	16.4	81.2	243	14.1	446	81.6	0.23	265	7.13	13.2	1.95	0.16	136	1.74	0.77
stdev	1.2	0.66	0.69	2.5	0.06	0.15	0.57	0.94	0.04	0.85	0.29	5.3	27.3	544	5.2	68	39.7	0.12	150	1.68	3.1	0.87	0.04	316	0.33	0.13
Merv Meso 2 (n = 7)	15.6	3.63	1.78	66.3	0.24	0.47	2.76	6.10	0.09	1.13	1.62	27.7	159	46.3	9.7	239	22.9	0.20	116	9.02	11.4	0.58	0.10	10.8	1.18	0.70
stdev	0.1	0.05	0.04	0.3	0.00	0.01	0.03	0.07	0.00	0.02	0.02	0.3	6	1.5	0.1	3	0.2	0.03	2	0.11	0.1	0.02	0.00	1.2	0.02	0.02
Merv Sam (n = 6)	15.0	5.68	1.28	68.2	0.10	0.65	2.45	5.38	0.07	0.60	0.40	24.6	25.9	10.6	12.2	396	90.6	0.14	140	5.05	9.2	2.15	0.09	3.01	1.30	0.54
stdev	2.3	1.04	0.35	4.0	0.02	0.10	0.54	0.71	0.02	0.27	0.13	3.7	4.3	4.8	1.2	91	53.1	0.03	33	1.48	2.8	1.15	0.03	1.77	0.38	0.13
Merv high Al (b) (n = 3)	16.6	4.11	9.11	55.6	0.35	0.52	4.73	7.28	0.21	0.05	1.15	34.6	66.0	307	49.4	655	60.1	1.16	398	19.4	35.1	1.39	0.65	126	5.65	0.90
stdev	0.0	0.06	0.55	1.0	0.01	0.01	0.13	0.12	0.01	0.01	0.15	2.0	1.1	28	4.6	19	0.7	0.03	7	3.5	6.5	0.02	0.01	16	0.39	0.02
Merv high Al (a) (n = 5)	15.8	4.60	7.91	53.9	0.38	0.50	4.13	10.0	0.31	0.05	2.21	21.4	32.9	19.0	54.5	414	96.9	1.15	413	18.0	33.7	2.51	0.39	9.99	5.34	1.42
stdev	1.5	0.83	0.35	0.8	0.04	0.06	0.42	0.9	0.02	0.00	0.16	1.1	3.9	2.8	4.6	36	20.0	0.18	12	1.2	2.1	0.53	0.03	4.03	0.44	0.13

Please note that the groups were divided on the basis of their Al, Cr, La, Th, U, and Mn contents and only data from glass that was not intentionally colored was taken into account. Oxides and chlorine are given as wt % and trace elements in [ppm]. The full set of data is given in Table S1.

Figure 3.

Base glass characteristics of the samples from Merv analyzed in this study were divided into compositional groups

(A) There is a clear distinction between glass with relatively low Al₂O₃ [wt %] and high Cr/La and samples with low Cr/La but highly variable Al₂O₃ contents; (B) despite a wide range of Al₂O₃ concentrations, most samples show a fairly constant Th/U ratio, while Merv G3, Merv high Al (B) and the Mesopotamian glasses deviate from this; (C) Mn [ppm] versus Al₂O₃ [wt %] levels point to differences in the glassmaking recipes.

Figure 4.

Average trace element patterns of the different glass groups from Merv

(A) Similarities between Merv G1-G3 suggest a common geological origin of the silica source, whereas both high Al groups show different patterns; (B) glass imported from the Mesopotamian region has very different trace element patterns compared to the local groups as exemplified by Merv G2 (same data as in a). Individual data points were normalized to the upper continental crust²⁶ and subsequently averaged.

Local glass types (Merv G1, G2, and G3)

Among the glass with low Cr/La ratios (<4) a high aluminum group (Al₂O₃ > 4.5 wt % used as cut off) was identified previously on the basis of the high levels of mineral impurities in the glassmaking sand.¹³ This group, which is referred to as Merv G1 (n = 47), has consistently higher metals and REEs compared to Merv G2 and G3 and shows a slight positive europium anomaly (Figure 4A). It is a relatively homogeneous group with on average 15 wt % soda, 60 wt % silica, 6 wt % alumina, 8 wt % lime, 4 wt % magnesia, and 4.5 wt % potash (Table 1). As far as the signature of the plant ash component is concerned, Merv G1 has slightly higher potash contents along with higher concentrations of lithium, phosphorus, rubidium, and cesium (Table 1). With the exception of two samples (Merv 584, 614), the manganese content remains at the background level (<300 ppm; Figure 3C). If we assume that manganese was added as (de-) colorant during primary production (e.g.,²⁷), this means that Merv G1 is the product of a distinct primary glassmaking event.

The presence of Mn with a cut off at 4000 ppm is used as a criterion to separate the remaining glass with low Cr/La ratios (<4) and alumina below the 4.5 wt % threshold into two further groups, Merv G2 and Merv G3 (Figure 3C). The two groups are close in terms of the base glass composition (Table 1), the main difference being that Merv G3 tends to have lower aluminum (mean Al₂O₃ ∼2.6 wt % compared to ∼3.3 wt % in Merv G2), lower metals and REE contents, but higher uranium relative to thorium (Figure 4A). This adds a further justification to divide these samples into two subgroups. In addition, V, Sr, Mo, Ba and W are somewhat enriched in Merv G3 compared to Merv G2. These slightly higher values of some traces are probably due to accessory elements present in manganese-bearing minerals, as recently found in late antique glass from Jalame.²⁸ MnO contents in Merv G3 vary from about 0.5 wt % to 2 wt %, but there is no systematic difference in color. All glass fragments typically have a light green color (Table S1).

The minor and trace elements associated with the silica source vary more-or-less continuously across Merv G1, Merv G2, and Merv G3. The close compositional relationship of the Merv groups is also seen in a principal component analysis that partially separates Merv G2 and G3 (Figure S1). The correlations between the silica-related elements such as Al₂O₃, TiO₂, Fe₂O₃, V, Cr, Y, Zr, Nb, Hf, Ta, Th, and the entire suite of lanthanide rare earth elements, are very high with R > 0.8 overall and >0.9 for the REEs (Table S2). The correlation matrix moreover shows that even the alkali metals (especially Li, Rb, and Cs, less so K₂O) are strongly correlated with the aforementioned elements. This strong positive correlation suggests that the three Merv groups (G1-G3) are closely related and originate from a similar, if not the same silica source. It also implies that part of the minor alkali metals are derived from the silica source and do not exclusively represent the plant ash composition. The only exception is Na₂O, which shows no clear correlation with any of the other elements analyzed, and is obviously the central component of plant ash.

In terms of the plant ash characteristics, magnesium, phosphorus, chlorine, potassium, and calcium concentrations are relatively homogeneous across both low aluminum groups Merv G2 and G3. Although potash contents are marginally lower in Merv G2 and G3 than in Merv G1, the MgO/CaO, K₂O/P₂O_5, and Li/K₂O ratios are unchanged (Figure S2), pointing to the use of cognate plant ashes and/or ashing processes. The chlorine concentrations of all three groups lie around the 0.5 wt % mark (Table 1). In short, the elements commonly associated with the plant ash form a common cluster of Merv G1, G2, and G3 and do not allow the fluxing agent to be distinguished further.

Imported glass types (Merv Meso1 & Meso2, Merv Sam)

The glass finds with high Cr/La ratios (>4) differ significantly from the other glass samples from Merv. Chromite-bearing silica sources appear to be characteristic of Mesopotamian glass production from the Late Bronze Age²⁹ through the Sasanian and into the Islamic periods, evidenced by analytical data from Samarra,¹⁰ Veh Ardašīr,^30,31 Ctesiphon^13,32 and Siraf.²⁷ The three glass groups from Merv with elevated Cr/La ratios in combination with low to moderate aluminum oxide levels can therefore all be attributed to a Mesopotamian origin (i.e., the Tigris-Euphrates basin), even though they show very different compositional characteristics. Based on the impurities in the silica sources (Figure 3), the phosphorus concentrations, and the MgO/CaO ratios (Figure S2), three broad groups can be distinguished.

A group of six samples has characteristics similar to glass from ninth-century Samarra in Iraq and is accordingly referred to as Merv Sam.¹⁰ These samples have the lowest mineral impurities, the highest magnesium, and MgO/CaO ratios, and exceptionally low phosphorus contents (P₂O₅ < 0.15 wt %; Table 1; Figure S2). More specifically, two samples (Merv 501, 505) match the composition of the very clean Samarra 1 group, while the other four are closely related to Samarra 2, although one sample (Merv 500) exceeds the Zr and Hf levels of the Samarra 2 group (Figure S3). Both Samarra groups are thought to have been manufactured in the vicinity of Samarra itself although no direct glassmaking evidence from this period has been found.¹⁰

Group Merv Meso1 has considerably higher phosphorus and calcium concentrations, and all silica-related impurities are higher than in Merv Sam, including all REEs, but zirconium and hafnium are a little lower (Table S1; Figure 4B). The main difference to the local group Merv G2 is a much higher chromium and lower Rb and Cs content. Judging from these characteristics, a Mesopotamian provenance is proposed. The exact origin of Merv Meso1 is unclear although glass production sites are known in southern Iraq,^33,34 and there is a certain similarity in composition with the Sasanian glass from Veh Ardašīr.^30,31 The trace and rare earth elements lie more or less between the averages of Sasanian 1a and Sasanian 2, as far as can be judged from the published data, which unfortunately lack some of the trace elements such as Th and U (Figure S4). Merv Meso1 shows consistently higher Ba contents, which is probably linked to their higher Mn levels. The manganese contents, which range from background levels (Mn < 300 ppm) to MnO >2 wt %, emphasize the heterogeneity of Merv Meso1 as a group. The plant ash signature of Merv Meso1 also differs from the Sasanian glass from Veh Ardašīr by on average lower K₂O/P₂O₅ and MgO/CaO ratios (Table S1).

The group, provisionally labeled Merv Meso2, is exceptional in many respects. It has relatively little aluminum (Al₂O₃ < 2 wt %), and less strontium, zirconium, and hafnium compared to all other groups (Table 1; Figure 4B). Merv Meso2 also shows a pronounced negative cerium anomaly and there is a general decline in the heavier lanthanide rare earths. The lithium content is similar to that of Merv Sam and part of the Merv Meso1 samples; the boron levels are higher than in any of the other glass groups (∼350 ppm); while MgO, K₂O and CaO are comparatively low (Table 1). Manganese at a concentration of about 1 wt % indicates a deliberate addition as a decoloring agent. Merv Meso2 agrees with Sasanian 1 in terms of the alkali and alkaline earths but differs drastically from all Mesopotamian as well as Central Asian glasses. Its high chromium concentration nonetheless suggests a Mesopotamian origin. The compositional data of the 7 fragments belonging to Merv Meso2 are within the margin of error of the LA-ICP-MS analysis, meaning that they may well belong to the same primary and secondary production batch.

Merv high Al and outliers

Nine glass samples from Merv have outstandingly high alumina concentrations (Al₂O₃ > 7 wt %). The trace element patterns of a subset of 5 of these high Al glasses (subgroup A) behave similarly to those of the Merv G1-G3 groups, but with lower proportions of some high field strength elements such as Zr and Hf compared to the REEs (Figure 4A). In contrast, the remaining 4 high Al samples (subgroup B) have very unusual trace element patterns pointing to the use of a silica source that is largely depleted in the heavy REEs, but that is exceptionally high in Cr, Nb, Ta, and Th relative to the other elements (Figure 4A). These features are due to a significant contribution of feldspar to the glass raw materials. No particularities were recognized with regard to the elements of the plant ash used for these high Al glasses, but for some high lime concentrations (Merv high Al (A)), and elevated lithium and rubidium levels (Merv high Al (B); Table S1).

Two different classes of high alumina glass have been reported in the literature, one of which is a mineral soda glass and associated with south Asian production, and the other is a soda-rich plant ash glass believed to be representative of Central Asian glassmaking.^{35,36,37,38,39,40,41,42} The Merv high Al groups have some resemblance with the glass found in Xinjiang in western China⁴³ and the Ferghana Valley in eastern Uzbekistan⁴⁰ in terms of the major and minor elements. However, the high Al glass from Merv tends to have higher MgO and CaO (Figure S5A) and lower K₂O and Cl contents than the glass from Xinjiang and Ferghana with equivalent high alumina levels (>4.5 wt %), indicating differences in the plant ash component. As far as the ratio of Al₂O₃ to TiO₂ is concerned, Merv high Al (A) lies in the extension of the regression line of the glass from the Ferghana Valley, while Merv high Al (B) deviates from the glass found in both Xinjiang and Ferghana (Figure S5B). No trace element data have been published for Xinjiang and Ferghana, which makes it impossible to say more about the origin of these high Al glasses at this stage. Nevertheless, there seems to be an increased occurrence of high Al glass in Central Asia east of Merv and Bukhara, suggesting an eastern Central Asian origin of these samples.

Apart from these sub-groups, there are 2 individual samples that cannot be clearly assigned at present, due to their unusual relative concentrations of the major components. For example, they have among the lowest MgO, CaO, and highest Cl and K₂O concentrations (Table S1). These two outliers are not included in the figures and are not described further.

Colors and recycling markers

There is no systematic difference in color among the glass groups from Merv. Instead, the vast majority of the glass finds are light green (greenish aqua), regardless of the manganese concentration, which was added to some glass from Merv as a decolorizing agent to oxide the bluish Fe²⁺ to a pale yellow Fe³⁺ (Table S1). Manganese is a clear additive in Merv G3 and Merv Meso2. The samples of Merv G1 and both Merv high Al groups are for the most part free of Mn. The other compositional groups are not uniform. While individual samples of Merv G2, Merv Meso1, and Merv Sam indicate the intentional addition of Mn, its presence at low concentrations (MnO <0.5 wt %) may be the result of an inadvertent admixture of cullet or the uneven incorporation and mixing of the batch (Figure 3C).²⁸

Cobalt is responsible for the coloration of the decoration on 9 blue samples belonging to the Merv G2 group (Co > 150 ppm). Cobalt is strongly correlated with zinc consistent with one of the three different types of cobalt (Ni, Zn, Cu) that have been identified among early Islamic glass assemblages.⁴⁴ The closest match to the Merv cobalt signature (Figure S6) are a handful of glass beads from Viking Age Ribe in Denmark,⁴⁵ a scratch-decorated plate from Samarra¹⁰ and a scrap glass from Raqqa,³² both from the ninth century, and a late eighth-century glass weight from Egypt.⁴⁶ Copper is also elevated in the cobalt-blue glass, and one sample (Merv 625) additionally has higher As, Sn, Sb, Pb and Bi levels, which are all strongly correlated and are probably the result of a lead-bearing ingredient.

Copper is present in 16 samples at concentrations that have a visible effect on the color (CuO > 0.5 wt %), resulting in either a greenish blue/turquoise or dark green glass (Table S1; Figure 5). Tin, antimony, arsenic, and lead are more or less enriched in these samples, which is likely related to the coloring process and the raw materials used for coloring. Copper colored fragments were identified among the Merv G1 and G2 groups, Merv Meso1, as well as one high Al (B) specimen.

Figure 5.

Elements associated with the coloring of the glass

(A) Cu and Co single out the intentionally colored samples and illustrate that the Co-blue samples also have elevated Cu concentrations; (B) Cu is correlated with Sn, suggesting the use of a Cu-Sn alloy as the coloring raw material; (C) contents of Cu and Pb > 50 ppm point to the incorporation of colored cullet and the recycling of glass.

Copper and lead concentrations above the impurity threshold of the silica source (MUQ for Cu ∼32 ppm, for Pb ∼20 ppm)²⁶ can be used to estimate the incidence of recycling among the naturally colored glass in each compositional group (Figure 5C). If we apply a less stringent threshold of 50 ppm for both Cu and Pb, all but five of the unintentionally colored samples from Merv G3 exceed this limit, suggesting that at least 80% of the Merv G3 samples contain recycled material (21 out of 26). In Merv G2, only 5 out of 21 samples have Cu and Pb as well as Mn at a level of natural impurities (i.e., 76% are recycled). Merv G1 seems to be the group with the least recycling, here 29 out of 41 samples (70%) show no clear traces of recycling. It should be kept in mind that the absence of elevated transition metals may result from selective recycling.

There are no obvious recycling markers in the Merv Meso2 and Merv Sam groups, and only 5 of 14 samples from Merv Meso1 show evidence of the incorporation of some copper-bearing cullet. Merv high Al (A), the group with the more conventional trace element pattern (n = 5) shows low levels for both Cu and Pb. In contrast, all Merv high Al (B) samples represent recycled material according to the same criteria (Cu and Pb > 50 ppm).

Discussion

Central Asian glass groups compared to local glass from Merv

To assess the extent of local glass production and to explore its relationship with glass finds from other sites, the results from Merv (groups G1-G3) need to be compared with legacy data of early Islamic glass along the medieval trade routes. A shift in the chemical make-up of glass assemblages from Iran toward Central Asia was observed in connection with the ninth-century glass from Nishapur (Iran).^9,13 The glass from Nishapur has since been revisited, and new data from urban centers in Transoxiana have been published,^3,9,25 allowing for a more nuanced interpretation of the data from Merv. For example, Merv G1 (Al₂O₃ > 4.5 wt %) had been considered as representative of Central Asian glass. Even though the glass assemblage from Merv exhibits typical regional characteristics, such as high and variable alumina and lithium concentrations, the assemblage as a whole reflects above all the multiplication of primary glass production in the Islamic period.

None of the glass compositions reported from multiple sites in Iran,⁹ Bukhara,²⁵ the Samarqand region³ and Ghazni⁴⁷ align with Merv G1 or either sub-type of the Merv high Al groups (Figure 6). Only very few samples from other archaeological contexts have alumina concentrations in excess of 5 wt % and those that come close tend to have different trace element patterns. In the case of Merv G2 and Merv G3, there is also no exact match with published data. The overlap of Merv G3 with the groups Iran G1a & G1b in terms of the alumina contents⁹ is resolved when considering the trace elements that are consistently higher in the Iranian samples (Figure S7). The accessory minerals in the silica source used for the production of Iranian glass G1a and G1b appear to be closer to Merv G1, despite the fact that the Iranian samples have significantly lower Al contents. There is also a pronounced difference in the K₂O/Cs ratios and absolute K₂O levels, the latter tending to be lower in the Iranian groups (Figure 6C).

Figure 6.

Merv G1-G3 and Merv high Al groups analyzed in the present study compared with early Islamic plant ash glass from other Central Asian sites

(A) TiO₂/Al₂O₃ versus Al₂O₃/SiO₂ clearly separate the Merv groups from all other Central Asian assemblages; (B) U and Th ratios of the glass from Merv deviate clearly from those of contemporary glass from Bukhara and the Samarqand region; (C) K₂O/Cs ratios in comparison with K₂O separates the Iranian groups G1 and G2 from glass found further east (Merv, Bukhara, Samarqand). Data sources: Iranian groups G1 & G2;⁹ data from Bukhara²⁵; glass from the Samarqand region³; some finds from Ghazni in Afghanistan.⁴⁷

The recently published data from the area around Samarqand³ do not offer a compositional match for the Merv groups either. The glass from both Kafir Kala and Cholakpeta tends to have lower high field strength elements such as Zr and Hf relative to aluminum and REEs (Figure S8). In terms of the plant ash component, the samples from Samarqand have higher P₂O₅ relative to potassium. Moreover, the data from Samarqand show unusually high antimony values,³ which are rarely found in early Islamic glass. The data of the Islamic glass from Ghazni (Afghanistan) are again not directly comparable, because the assemblage is extremely heterogeneous and tends to be of a later date.⁴⁷ The bulk of the Ghazni fragments have higher REEs relative to Zr and Hf, and higher Th concentrations (Figure S8).

The contrast in composition between Merv G1-G3 and other available datasets confirms that both Merv G1 and G2 and possibly also G3 are independent groups and locally produced. The mineral impurities associated with the silica source vary continuously across the data of all three groups as evidenced by the strong positive correlation of the metals and REEs. This means that geologically related silica sources were used for all three Merv groups with a different proportion in feldspars, pyroxene, and other accessory minerals that result in a proportional increase in heavy elements and REEs. The variations in absolute composition may be due to hydraulic effects during sediment deposition⁴⁸ and/or a chronological shift since the finds cannot be dated more precisely than the ninth century. The variability is thus an essential feature of the glass found at Merv, and reinforces its unique regional character. The silica-related elements behave linearly to each other and the ratios remain largely constant. Exceptions are the samples of Merv G3, where the elemental correlations between Ti and Nd or Th and U deviate from that of Merv G1 and Merv G2 (Figures 4A and 6B). Merv G3 shows instead some similarities with the Bukhara 1 group.²⁵ While the trace element patterns are largely congruent, the Bukhara samples have lower TiO₂/Al₂O₃ ratios and much higher U contents (Figure 6). Since all but 4 samples of the Merv G3 group show clear signs of recycling, it is reasonable to assume that Merv G3 may be the result of mixing of different base glasses, thus blurring group structures and affiliation. Another clue that Merv G3 may contain recycled material from other sites such as Bukhara is the presence of MnO. While Merv G3 (similar to Bukhara 1) has more-or-less continuous concentrations of MnO from about 0.5 wt % to 2 wt %, the local groups Merv G1 and G2 are notable for the absence of MnO.

The cobalt blue samples from Merv are additional evidence of independent local glass production. In a recent review of early Islamic cobalt blue glass, it was found that most of the cobalt blue samples from the eighth to eleventh centuries have Mesopotamian base glass characteristics with Cr/La > 4.⁴⁴ The cobalt blue samples from Merv are exceptional because they all belong to the local Merv G2 glass group, and the cobalt colorant is of the high zinc variant, which was primarily detected in glass beads from Ribe in Denmark but not systematically among contemporary glass from, for example, Samarra, Nishapur or Bukhara (Figure S6). Nothing is known about the cobalt mines that were exploited during the early Islamic period, or how the cobalt ores might have been processed, but it is clear that the cobalt used in Merv has a different origin to that used in association with most Mesopotamian and Iranian glass groups. This further supports the geographical differentiation of the glass of Merv.

Discrete glass production versus regional signatures

Judging from our data, glass in the major urban centers along the medieval trade routes forms part of the tradition of soda-rich plant ash glassmaking typical of the entire early Islamic world, but with local peculiarities. It can be assumed that most of the glass finds from Merv, as well as from other regional centers, originate from the local production of both raw glass and finished objects. The site-specific compositional characteristics can sometimes be masked by recycling and/or trade, as suggested by the Merv G3 group. Of course, the existing database on early Islamic glass from Central Asia is still very limited and it is likely that there are numerous compositional endmembers that have not yet been identified. However, the fact that in many cases we can recognize relatively uniform groups indicates that the overall movement of large quantities of glass in general, and raw glass in particular, was very limited. Clear examples in support of this conclusion are the glass assemblages recovered from Gorgan (Iranian G2),⁹ Bukhara²⁵ and above all Merv, from where statistically significant numbers of samples have been analyzed and which can be conclusively separated into discrete groups (Figure 6). Merv G1, Merv G2, Iranian G2, and Bukhara form clear compositional clusters, yet they also share some region specific features. Compared to Mesopotamian glass, vitreous material from the eastern Islamic provinces of Khurasan and Mawarannahr derive from silica sources consisting of younger felsic rocks (such as granite) rich in feldspar and pyroxene.^6,7

In view of the intrinsic variability, however, it is difficult to define clear regional characteristics. By and large, the glass groups Merv G1-G3 conform to the criteria previously proposed for Central Asian glass having low Cr/La and moderate to high Al and REEs.^6,7 If Merv G2 and G3 are accepted as local, Central Asian groups, then the Merv glass covers an even wider range of compositions than previously thought, including glass with alumina <2 wt % and relatively low levels of other accessory elements. The considerable variability of elements associated with the silica source appears to be a recurring phenomenon of glass production in the more peripheral regions of the Islamic world. Similar observations have been made in relation to soda-rich plant ash glass produced in the Iberian Peninsula in the early Islamic period.⁹ Hence, differences in the absolute concentrations do not necessarily imply different origins. As long as the ratios between elements are stable, similar silica sources may well have been used, pointing to a common origin (Table S2). Rather, the diversity of composition can be seen as a specific feature of early Islamic glass productions.

The heterogeneity and the seemingly random combination of different silica raw materials and plant ashes,^7,30,31,49 make the interpretation of analytical data of early Islamic plant ash glass challenging and raise the question of the utility of regional traits. The decentralization and diversification of primary glassmaking, such as glazed pottery production, during the early Islamic period is now a well-established fact thanks to the increasing number of analytical studies.⁶ The evidence from Merv allows us to take these considerations even further. The old model whereby large quantities of raw glass (or glass in general) were traded over long distances to supply local secondary glass workshops, allowing the tracing of distribution networks, cannot be sustained and may have only ever applied to regions such as the Mediterranean and Mesopotamia where waterborne transport via sea, river or canal was key. Instead, in times of prosperity, powerful urban hubs set up their own glass industries to meet most of their local demand as long as they had access to the necessary resources. The diversification of primary glass production through the Islamic realm therefore makes the perennial question of provenance increasingly irrelevant. The more pertinent research questions might therefore be why certain places did not develop their own local production of raw glass, as seems to have been the case in Abbasid Nishapur,^6,9 and why some materials and objects were still traded over long distances, as demonstrated by finds of Islamic luster and scratch-decorated glass vessels in the Famen temple in China.⁵⁰ It is significant that, just as in the preceding Sasanian period, Mesopotamian glass continued to arrive at Merv in the form of finished products and not as raw glass, because the composition of these samples is clearly different and retains its distinct nature. Therefore, a new conceptual model is proposed that links the changes in the archaeo-vitreous record to the changing local or regional socio-economic and/or cultural conditions in the Islamic world.

Limitations of the study

Early Islamic plant ash glass is compositionally highly complex and no primary production site has yet been archaeologically documented on the Iranian plateau or Central Asia. It is thus challenging to identify commonalities between assemblages, not least because it is strictly speaking not possible to demonstrate the identity of different glass samples. The internal variability of raw materials and glass recycling adds to the complexity of data interpretation. To bypass these difficulties, the present study uses patterns of circulation derived from new data in combination with an extensive database of glass compositions to trace relationships between archaeological sites and changes over time and space. Nevertheless, the coverage of early Islamic glass assemblages is still sketchy, and further studies with high resolution LA-ICP-MS and isotopic data are needed to fully grasp the diversity of glass production and the processes of technical transfer strategies in the medieval world. As more data become available, advanced statistical tools including machine learning could be leveraged to substantiate the diversification of glass production and the proposed changes in the organization of the early Islamic glass industry.

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Nadine Schibille (nadine.schibille@cnrs.fr).

Materials availability

This study did not generate new unique materials beyond the analytical data given in the tables.

Data and code availability

Data reported in this study are all included within this publication, including in the supplemental information.

This paper does not report original code.

Acknowledgments

We thank the British Museum for access to the vitreous materials and support for this project. This project received funding from the European Research Council under the European Union’s Horizon 2020 Research and Innovation Programme (Grant Agreement 647315, to N.S.). The funding organization had no influence in the study design, data collection and analysis, decision to publish, or preparation of the article.

Author contributions

Conceptualization, A.M. and N.S.; formal analysis, A.M. and N.S.; investigation, A.M., N.S., and St.J.S; resources, A.M., N.S., and St.J.S; writing—original draft, N.S.; writing—review and editing, A.M., St.J.S, and N.S.; visualization, N.S.; funding acquisition, N.S.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Other
Iranian groups G1 & G2	Schibille et al.9	N/A
Bukhara data	Schibille et al.25	N/A
Samarkand data (Kafir Kala and Cholaktepa)	Chinni et al.3	N/A
Ghazni data	Fiorentino et al.47	N/A
Samarra groups 1 & 2	Schibille et al.10	N/A
Veh Ardašīr data	Mirti et al.30,31	N/A
Ctesiphon data	Henderson et al.32	N/A
Siraf data	Swan et al.27	N/A
Xinjian data	Brill43	N/A
Ferghana data	Rehren et al.40	N/A
SRM NIST610 & NIST612 data	Jochum et al.51	N/A
Corning glass standards A, B, C & D data	Vicenzi et al.52 / Wagner et al.53	N/A
ELEMENT XR mass spectrometer	Thermo Fisher	https://www.thermofisher.com
M50E excimer 193 nm laser	Resonetics	https://resonetics.com

Experimental model and study participant details

Archaeological samples

In this study, 146 glass fragments from Merv were analysed, corresponding to 150 data points due to 4 objects (Merv 511, 513, 514, 621) with cobalt- or copper-blue decorations, representing the entire ninth-century assemblage from Merv at the British Museum. All samples are translucent and most are weakly coloured as a result of iron naturally present in the silica source. About two thirds of the fragments (n = 99) have a slightly greenish colouration, while eleven have a more pronounced green colour, another fourteen have a bluish green tinge. Four samples have a light bluish and three specimens a yellowish tinge (Merv 519, 529, 556). Ten fragments are blue, four can be considered colourless, three are black appearing and lastly there is one turquoise semi-transparent sherd (Table S1).

The analysed assemblage included 143 vessels and three window glass fragments. Most of the vessels were monochrome plain and free-blown (n = 123), but the collection also contained mould-blown vessels (n = 27) and some decorated with fine cobalt or copper-blue trails on their rims. Some vessel forms and decorative techniques can be discerned such as bottles (n = 13), flasks (n = 6), bowls (n = 24), a jar (Merv 616) and a stopper (Merv 609). The majority of sherds, however, are too fragmentary to create a detailed typology. Identifiable features include folded and fire-polished rims and low push-ups, mould-blown with spiralled (Merv 601) or diagonal ribs (e.g. Merv 537, 583, 604, 614), repeating horizontal rows of sunken squares (e.g. Merv 591), and applied trail decorations (Merv 499, 547, 592b). All available information on the fragments including photographs for most of them, can be found on the British Museum's collection website (https://www.britishmuseum.org/collection) under the inventory number given in Table S1.

Method details

LA-ICP-MS

Small fragments of the glass finds were mounted in eight epoxy resin blocks (16 to 21 samples per block) and ground to remove excess resin and any alterations to facilitate the analyses of healthy glass samples. Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) was carried out in the laboratory of the Institut de Recherche sur les ArchéoMATériaux (IRAMAT) Centre Ernest-Babelon (CNRS Orléans, France).^23,24,25 All eight resin blocks were placed together with the standard reference materials (SRM) in the ablation cell of a Resonetics M50E excimer 193 nm laser sampling system. The laser operated at a maximum energy of 5 mJ, a pulse frequency of 10 Hz and a stationary laser beam diameter of typically 100 μm, reduced to 80 μm when saturation of the spectrometer caused by elevated manganese occurred. The analytical protocol includes a pre-ablation of 15 s before the acquisition of 58 different isotopes from lithium to uranium for 27 s. This corresponds to 9 mass scans for each isotope. The isotopes were selected to minimize possible interference: Li7, B11, Na23, Mg25, Al27, Si28, P31, Cl35, K39, Ca44, Ti47, V51, Cr52, Mn55, Fe57, Co59, Ni60, Cu65, Zn64, Ga71, As75, Se78, Rb85, Sr88, Y89, Zr90, Nb93, Mo95, Ag107, Cd111, In115, Sn118, Sb121, Cs133, Ba137, La139, Ce140, Pr141, Nd146, Sm147, Eu153, Gd157, Tb159, Dy163, Ho165, Er166, Tm169, Yb172, Lu175, Hf178, Ta181, W182, Pt195, Au197, Pb204, Bi209, Th232, U238.²⁴ An argon/helium flow (1 l/min Ar + 0.65 l/min He) transports the sample material to the plasma torch where it is dissociated, atomized and ionized in the plasma at high temperature (8000°C).

A Thermo Fisher Scientific ELEMENT XR mass spectrometer equipped with a discrete dynode detector system was used to record the signal intensities. Thanks to a combination of a secondary electron multiplier (SEM) with two operating modes (counting and analog) and a Faraday collector, the linear dynamic range is more than 12 orders of magnitude, which means that major, minor and trace elements can be measured in a single run regardless of their absolute concentration and isotopic abundance. Peak-jump acquisition mode is implemented with four points per peak for the counting and analog acquisition modes and ten points per peak for Faraday detections. Most ions are quantified with a secondary electron multiplier (SEM) and the Faraday detector is used only for ²³Na, ²⁷Al and ³⁹K. The analytical sequence provides for regular analysis of glass reference materials (NIST610, NIST612, Corning A, B, C and D, APL1) at the beginning, at intervals of 12-15 samples and at the end of the sequence, and blank measurements before and after the analysis of the SRMs. NIST610, Corning B, C, D and APL1 are used to calculate the average response coefficient for each element and thus convert the signals (counts/second) into fully quantitative data. NIST612 and Corning A are used as unknowns to validate the results and verify accuracy and precision (Table S1). Accuracy and precision are estimated by comparing the average results for the SRMs with certified values^52,53,51 and the standard deviations of the repeated measurements, respectively.

Quantification and statistical analysis

The previously described calibration procedure with ²⁸Si as internal standard and five external reference glasses (NIST610, Corning B, C and D, APL1) is used to calculate the response coefficient factor K_y.⁵⁴ Concentrations are calculated using the net average intensity count rates for each isotope. Assuming that the sum of the various glass components is about 100% (or close to), the concentrations of the individual elements can be calculated as:

$$\%Y_m{O}_n=\frac{I_Y\times {\alpha }_Y}{I_{Si}\times K_Y}/\sum \frac{I_X\times {\alpha }_X}{I_{Si}\times K_X}$$

where I_Y, I_X and I_Si are the net intensity count rates corrected for isotopic abundance of isotopes for elements Y, X (all 58 elements) and Si (internal standard). α_Y and α_X are the conversion factors from elements into oxides, while K_Y and K_X are the response coefficient factors according to

$$K_Y=\frac{I_{Y_{std}}\times {\left[Conc\right]}_{{Si}_{std}}}{I_{{Si}_{std}}\times {\left[Conc\right]}_{Y_{std}}}$$

where [Conc]_Ystd and [Conc]_Sistd are the concentrations of element Y and Si in the standard material.

Published: January 20, 2025