The technology of tin smelting in the Rooiberg Valley, Limpopo Province, South Africa, ca. 1650–1850 CE

Journal of Archaeological Science 37 (2010) 1656–1669

Contents lists available at ScienceDirect

Journal of Archaeological Science journal homepage: http://www.elsevier.com/locate/jas

The technology of tin smelting in the Rooiberg Valley, Limpopo Province, South Africa, ca. 1650–1850 CE Shadreck Chirikure a, Robert B. Heimann b, David Killick c, * a

Department of Archaeology, University of Cape Town, Rondebosch 7700, South Africa ¨ rlitz, Germany Am Stadtpark 2A, Go c Department of Anthropology, University of Arizona, 1009 E South Campus Drive, Tucson, AZ 85721-0030, USA b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 24 June 2009 Received in revised form 18 January 2010 Accepted 19 January 2010

A substantial indigenous tin-smelting industry arose in the Rooiberg valley of northern South Africa in the second millennium CE. This study concentrates upon tin-smelting slags and refractory ceramics from two archaeological sites that date between ca. 1650 CE and ca. 1850 CE. These were studied by optical and electron microscopy, wavelength-dispersive x-ray fluorescence (WD-XRF), inductively-coupled plasma mass spectrometry (ICP-MS), and electron microprobe (EMPA). The slags are predominantly glassy; high SnO and relatively low SiO2 contents indicate that tin is a major glass-forming element. Comparison of slag chemistries with the mineralogy of ore deposits and host rocks shows that alluvial cassiterite was used at one of the sites, while cassiterite from hard-rock mining was smelted at the other site. Since few preindustrial tin slags have been studied, we compare our findings to other published examples, mostly from southwest England. Ó 2010 Elsevier Ltd. All rights reserved.

Keywords: Metallurgy Smelting Slags Tin South Africa Rooiberg

1. Introduction Although iron and copper were used in southern Africa from 200 to 300 cal CE, there was apparently no tin or bronze until the thirteenth century CE, some four centuries after the connection of this region to the Muslim maritime trade of the Indian Ocean (Miller, 2002; Killick, 2009a). Bronzes have been excavated in contexts dated between the thirteenth and nineteenth centuries in northern South Africa, Botswana and Zimbabwe (Miller, 2002; Chirikure et al., 2007; Denbow and Miller, 2007). The locations of these finds are shown in Fig. 1. Only one tin ingot has been found in good archaeological context (at Great Zimbabwe, and dated between 1450 and 1550 cal CE; T.N. Huffman, pers. com.). Twelve more tin ingots of undoubtedly indigenous manufacture have been found at surface or in uncontrolled excavations over the last century (Friede and Steel, 1976; Killick, 1991; Grant, 1999; Chirikure et al., 2007). We do not yet know when the production of tin and bronze began in this region. The only known precolonial tin mines in southern Africa (and indeed in all of Africa south of Nigeria) are in the Rooiberg valley, Limpopo Province of South Africa (Fig. 1). These were reopened from 1905 by miners of European descent, * Corresponding author. E-mail address: [email protected] (D. Killick). 0305-4403/$ – see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.jas.2010.01.026

and early mining geologists left excellent descriptions, plans and sections of some of the ancient shafts and surface workings that they encountered (Recknagel, 1908; Trevor, 1912; Baumann, 1919; White and Oxley Oxland, 1974). Baumann (1919) estimated the amount of ore mined in the Rooiberg Valley before 1905 at about 20,000 tons, and thought that some 2000 tons of tin ore might have been produced from it. A more recent estimate by Rooiberg mine geologists gave the total mass of rock mined before 1905 as 180,000 tons, while the estimate of tin ore produced was lowered to 1000 tons (Rozendaal and Kellaway, 1988, cited in Grant, 1994). Until recently this industry has received surprisingly little attention from archaeologists. Surface finds of bronze, arsenical copper, tin, ores and slags from Rooiberg have been periodically analysed and reanalyzed (Wagner and Gordon, 1929; Friede and Steel, 1976; Grant, 1994; Grant et al., 1994; Chirikure et al., 2007; Miller and Hall, 2008), and exploratory archaeological survey and excavation was undertaken by Hall (1981) and by Mason and Steel (Mason, 1986). Slender evidence of the antiquity of tin mining in the valley comes from radiocarbon dating of a wooden pit prop from an ancient gallery (Hall, 1981), and of a piece of charcoal trapped within a tin ingot (Grant, 1994). The calibrated ages (2 sigma) for both dates fall between 1400 and 1640 cal CE. There are also copper deposits in the Rooiberg valley, and two AMS radiocarbon dates on charcoal trapped in an arsenical copper ingot

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

Fig. 1. Location of archaeological sites in Southern Africa where bronze and tin artefacts have been excavated. Map courtesy of Simon Hall.

found at surface produced calibrated dates (2 sigma) between 1025 and 1285 cal CE (Grant et al., 1994). The results reported here were obtained on slags, ores and refractory materials recovered during four months of archaeological survey and excavation in the Rooiberg Valley in 2006 and 2007 by a team led by Simon Hall and David Killick. Reports on the archaeological sites, and on lead isotope provenance studies of southern African tin and bronze artefacts, are in preparation. Here we concentrate on the evidence for the smelting of tin at two sites. 2. The archaeological sites The Smelterskop site is an extensive stone walled site that covers the entire top of a low hill, about 2 km north of the largest cluster of ancient mines (Fig. 2). Stratigraphic evidence suggests that tin smelting at Smelterskop mostly post-dated the construction of the walls, but ten radiocarbon dates from the various stratigraphic levels all fall within the radiocarbon ‘‘black hole’’ from 1650 to 1950 cal CE. Single-grain optically stimulated luminescence (OSL) dating is in progress. At least four tons of metallurgical debris are present on Smelterskop in multiple dumps up to 25 cm thick. These dumps are dominated by heavily slagged tuye`res, but also contain black glassy slag, charcoal, and rare pieces of flat-faced gritty refractory ceramic, which are interpreted as pieces of furnace wall lining. The Elandsberg Ledge smelting site is a small scatter of slag and tuye`re pieces discovered in 2007 among low stone walls about 1.5 km north of Smelterskop. Radiocarbon dates of 243  32 BP (AA-77933) and 373  32 BP (AA-77934) were obtained on wood charcoal. The younger date (1523–1955 cal CE at 2 sigma; calibrated with OxCal 4.0) was on loose charcoal, while the older (1445– 1634 cal CE at 2 sigma) was on charcoal physically entrapped in slag. Slags were not tapped at either Smelterskop or Elandsberg Ledge, but solidified around pieces of charcoal within furnaces.

1657

Fig. 2. Location of archaeological sites (starred) discussed in this article. Only the northern (Rooiberg township) group of precolonial tin mines is plotted. Map courtesy of Simon Hall.

Almost all of the slags had been broken into pieces with a maximum dimension not larger than 4 cm, and one area on Smelterskop seems to have been devoted to slag processing, as it is covered with slag crushed into pieces of 5–10 mm. We assume that this was done to recover tin prills. The slags are black and glassy with conchoidal fracture. Masses of black slag containing white crystals adhere to the distal ends of tuye`res. We recovered no crucible fragments at either site, and found no furnaces. Nor did we find any furnaces during our archaeological survey of the Rooiberg valley. None have been reported from the valley since Wagner and Gordon (1929) illustrated several furnaces of stone slabs in the Blaauwbank donga (an area of intense sheet and gulley erosion about 5 km east of the Rooiberg A3 mines – see Fig. 2). No trace of these furnaces now remains.

3. The tin deposits in the Rooiberg Valley The major tin deposits of South Africa are of Precambrian Age and are genetically associated with the intrusion of the Bushveld Magmatic Province (BMP) between 2.05 and 2.06 billion years ago. The tin mineralization occurs within metasomatically altered sediments (probably originally quartzites) in the Rooiberg Fragment, a roughly triangular remnant of older Transvaal Supergroup and Rooiberg Group rocks that was floated up by the emplacement of the BMP. The Rooiberg Fragment is surrounded and underlain by highly evolved BMP granites, one of which is thought to have been the source of the tin-bearing fluids (Rozendaal et al., 1995; Schweitzer et al., 1995; Crocker et al., 2001; Labuschagne, 2004). Shear and tension fractures in the host rock served as channels for epithermal tin-bearing solutions emanating from the granite. The tin ores that were exploited by the ancient miners consist of fracture fillings, together with cylindrical pipes and strata-bound

1658

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

pockets that have penetrated into and replaced the host rock. Beyond the ore bodies the host rock was transformed into an arkose by infusion with sodium- and potassium-bearing fluids (Rozendaal et al., 1995; Labuschagne, 2004). The ore mineral assemblage is tourmaline, cassiterite, carbonates (ankerite and magnesian siderite), and sulphides (predominantly pyrite, with rare pyrrhotite, chalcopyrite, bismuthinite, sphalerite and galena) (Stumpfl, 1960; Leube and Stumpfl, 1963a, 1963b; Crocker et al., 2001). The structure of the pipes and pockets is generally concentric, with an outermost zone of reddish orthoclase, an inner zone of strongly sericitized arkose (‘greisen’) impregnated with tourmaline and cassiterite, and a center consisting of ironrich tourmaline (schorl) intergrown with cassiterite. Sulphides are mostly commonly emplaced along fissures but may occur also as fist-sized agglomerates in the center of the pockets. Iron-rich carbonates are found ubiquitously as a late stage, replacing all earlier minerals. The sequence of crystallization varied at each of the major ore deposits in the valley, but the overall genetic sequence of ore formation from older to younger is as follows: orthoclase – tourmaline (sometimes with apatite) – cassiterite – pyrrhotite – quartz and pyrite – chalcopyrite – galena, sphalerite, stannite – carbonate (with or without a second generation of cassiterite) (Stumpfl, 1960; Leube and Stumpfl, 1963b; Labuschagne, 2004). Fig. 3 is a micrograph of a piece of tourmaline–cassiterite ore. Small fragments (<200 microns) of remnant cassiterite are often seen in thin sections of slags from both sites. These often show cyclic twinning and compositional zoning, visible as various shades of brown. Tin has been mined since 1905 at six separate locations in the Rooiberg valley. At least three of these were exploited before the arrival of Europeans. These were the Leeuwpoort deposits at the southern end, and two deposits at the northern end – those on the farm Blaauwbank, and those within and around the modern town of Rooiberg, where about 28 ha of pre-European dumps, trenches and filled-in shafts up to 40 m deep were noted (Recknagel, 1908; Baumann, 1919). These early accounts noted the presence of a gossan (locally known as ‘‘ouklip’’) of iron oxides and hydroxides, containing remnant cassiterite, over much of this outcrop. This ancient mine complex is about 2 km south of Smelterskop (Fig. 2) and is assumed to have been the source for the ores smelted there.

4. Methods Polished thin sections and polished blocks of samples were prepared at the University of Cape Town. Major element analyses of slags, tuye`res, pottery and furnace floor fragments were performed on borate glass fusion disks by wavelength-dispersive XRF (Philips 1480) in the Department of Geology, University of Cape Town, under the supervision of Dr David Reid, who developed the protocol for analysis of tin- and zirconium-rich samples. Trace elements were measured on pressed powder pellets by XRF, and on samples in solution by ICP-MS (Perkin–Elmer/Sciex Elan 6000). Polarized light microscopy, scanning electron microscopy and microprobe analyses were undertaken by all three authors at the University of Arizona. Thin sections were studied in transmitted and reflected polarized light, and polished blocks in reflected light, in plane polarized (PPL) and cross-polarized (XPL) illumination. Selected specimens were also examined in UV, blue and green epifluorescence. Scanning electron microscopy in conjunction with EDX analyses was performed to determine the chemical composition of the glassy parts of the slag samples. Electron microprobe analyses of glasses and spinel crystals in a few slag samples were also made, using natural mineral standards and a CAMECA SX-50 wavelength-dispersive microprobe. The reporting of the concentration of elements present in different oxidation states requires some comments. Analyses performed by XRF, ICP-MS and/or ICP-AES yield elemental concentrations that were converted to mass percent oxides by calculation. This conversion is entirely founded on conventions based on stable valence states of the elements under scrutiny, in the present case tin and iron. Since the partitioning of these elements into Sn2þ and Sn4þ and Fe2þ and Fe3þ was not experimentally determined, the most probable theoretical valence states were assumed, i.e. Sn2þ in the case of slags (Tables 1 and 2) and Sn4þ in the case of tuye`res, pottery and furnace walls (Tables 2 and 3). More contentious is the representation of the total iron content. In this paper it is always reported as Fe2O3 even though the presence of spinels and (rarely) fayalite clearly indicates that a large amount of iron is in its divalent oxidation state (as expected from the relevant Ellingham diagram, see Yener and Vandiver, 1993). While for slags this is clearly inconsistent relative to our reporting of tin as SnO, this allows us to directly compare ore, intermediate smelting products, ceramics and slags on the same ternary diagram (Fig. 4b) and also to employ Dietzel’s approach to glass composition (see below). There is also some inconsistency in our reporting of trace elements. The Smelterskop slags were analysed for trace elements by ICP-MS in 2006. The Elandsberg ledge slags were excavated and analysed in 2007, at which time the ICP-MS laboratory was not operational. A smaller suite of trace elements were therefore measured in these slags by XRF. 5. Results 5.1. Major and minor element chemistry of slags and ceramics

Fig. 3. Rectangular laths of weathered tourmaline (light grey), subhexagonal grains of cassiterite (medium grey) and iron hydroxide weathering products (dark grey and black) in a piece of tin ore from mine dumps near Rooiberg township. (Transmitted plane-polarized light).

Nineteen slag samples from Smelterskop (Table 1) and six slag samples from Elandsberg Ledge (Table 2a – analyses EBL-4, 5, 7, 8, 9 and 10) were analysed for their major and minor oxide contents by WD-XRF. When examining these tables, note that the powders used to prepare the glass fusion discs for XRF were weighed before and after heating overnight at 100  C. The change in mass, which represents loss of adsorbed water, is labelled H2O. The dried samples were then placed overnight in an oxidizing muffle furnace at 850  C and cooled in a desiccator to room temperature before being weighed again. The change of mass after this step is recorded as LOI (Loss On Ignition). Positive LOI is attributable to loss of

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

1659

Table 1 Chemical compositon in mass% of Smelterskop slags measured by XRF. H2O is loss of mass (water) at 100  C. LOI (loss on ignition) is change of mass after heating in an oxidizing atmosphere at 850  C. Sample

SiO2

TiO2

Al2O3

Fe2O3

Cr2O3

MnO

MgO

CaO

Na2O

K2O

P2O5

SO3

SnO

H2O

LOI

Total

SK-2 SK-4 SK-8 SK-9 SK-10 SK-11 SK-12 SK-13 SK-14 SK-15 SK-17 SK-19 SK-20 SK-21 SK-22 SK-23 SK-24 SK-25 SK-26

45.78 36.85 44.16 32.18 47.37 27.98 41.53 36.24 38.71 43.64 32.36 34.47 44.67 39.12 41.00 35.95 53.00 28.70 20.88

1.32 1.16 1.43 2.34 1.37 1.07 1.38 0.89 2.07 1.81 1.07 1.49 1.42 1.50 1.44 1.17 2.34 1.06 0.89

10.71 12.12 8.74 6.22 13.85 9.42 11.98 11.84 8.32 12.53 10.40 9.82 13.07 14.51 12.37 12.53 22.07 9.77 7.42

12.83 22.27 11.80 30.02 12.72 11.42 11.67 18.90 34.47 12.24 18.37 27.18 17.84 13.91 14.22 21.02 5.23 10.91 9.77

0.06 0.04 0.05 0.06 0.05 0.04 0.05 0.04 0.05 0.05 0.04 0.05 0.04 0.05 0.05 0.04 0.06 0.04 0.03

0.07 0.09 0.06 0.12 0.09 0.08 0.08 0.18 0.13 0.17 0.11 0.14 0.11 0.09 0.08 0.08 0.10 0.09 0.10

2.47 3.71 1.84 1.17 3.33 2.90 2.66 2.91 1.74 3.42 2.70 2.32 3.99 3.69 3.12 3.12 5.82 3.01 1.85

10.22 8.68 16.77 1.66 6.20 5.77 4.09 3.35 3.12 8.21 3.67 4.14 5.62 2.62 4.88 4.34 4.03 6.76 2.08

0.63 0.89 0.47 0.33 0.76 0.65 0.67 0.88 0.53 0.85 0.80 0.58 0.77 0.99 0.74 0.85 1.47 0.58 0.56

1.97 2.39 4.89 1.55 3.25 1.58 2.30 1.76 1.92 2.58 1.57 1.81 2.31 2.07 2.49 1.59 3.28 1.65 1.01

0.53 0.94 0.56 0.58 0.61 0.59 0.42 0.44 0.61 0.56 0.81 0.80 0.50 0.46 0.45 0.53 0.34 0.81 0.31

0.03 0.02 0.01 0.02 0.02 0.01 0.01 0.01 0.02 0.02 0.02 0.01 0.01 0.03 0.01 0.02 0.01 0.01 0.01

14.93 12.95 10.41 28.25 12.15 42.56 26.38 26.13 11.40 15.51 31.06 19.50 11.04 24.03 20.78 21.36 1.71 40.65 61.74

0.06 0.12 0.12 0.06 0.05 0.07 0.06 0.05 0.05 0.04 0.04 0.05 0.04 0.03 0.02 0.03 0.06 0.06 0.03

2.34 2.86 1.98 4.96 2.14 4.83 3.61 4.32 3.48 2.33 3.75 2.87 2.10 3.01 1.76 3.07 0.42 4.15 6.62

99.28 99.37 99.34 99.60 99.68 99.31 99.67 99.30 99.66 99.30 99.27 99.49 99.33 100.09 99.89 99.56 99.10 99.95 100.06

structural water and carbon dioxide, but most of the slag samples in Tables 1 and 2 have negative LOI, meaning that they gained weight on ignition. This is attributed to oxidation in the muffle furnace of prills of metallic tin in the slag, and perhaps also to oxidation of Fe2þ to Fe3þ. In current archaeometallurgical practice the determination of H2O and LOI in slags is often neglected. This practice is deplorable, since it often leaves analytical totals far short, or far in excess, of 100%. We can make this point more concretely by considering the case of boron, which cannot be accurately measured by XRF, ICP-MS or electron microprobe. In our case it is essential to estimate the boron content of slags, as the ore at Rooiberg consists of cassiterite intergrown with tourmaline (Fig. 3). Tourmaline contains about 12 mass% B2O3, so the amount of boron in the slags provides essential information about the degree of beneficiation of ores before smelting. Because H2O and LOI were both measured in our slags, we are able to determine analytical totals (less boron) for the slags in Table 1. These totals are in all cases greater than 99.10 mass%, and in most cases greater than 99.40 mass%. There must therefore have been less than 10 mass% tourmaline in the ores as charged to the furnace, and in most cases much less. This implies that the ores were effectively beneficiated (presumably by crushing, then panning in water) before being smelted. The low inferred boron levels in these slags also provide an explanation for the fact that in two seasons of excavation at Smelterskop we found only one piece of cassiterite ore. Presumably the ore was brought to the site as a beneficiated powder.

The Smelterskop slag compositions vary widely within the following ranges: SiO2 (20.9–53.0 mass%), Al2O3 (6.2–22.1 mass%), Fe2O3 (5.2–34.5 mass%), CaO (1.7–16.8 mass%), MgO (1.2–5.8 mass%) and SnO (1.7–61.7 mass%) (Table 1). The Elandsberg Ledge (EBL) slags (Table 2a) show comparable scatter within the following ranges: SiO2 (35.6–70.3 mass%), Al2O3 (3.4–9.2 mass%), Fe2O3 (5.0– 11.1 mass%), CaO (6.3–15.5 mass %), MgO (1.3–3.4 mass%) and SnO (3.3–20.9 mass%). In contrast to the slags from Smelterskop, the EBL slags contain more SiO2, less Fe2O3 and SnO, and some contain substantial amounts of ZrO2 (3.6–9.0 mass%) and TiO2 (4.7–10.9 mass%). The one sample of ore recovered from EBL (Table 2a, sample EBL-6) is clearly not representative of the ore actually smelted there, as it contains little ZrO2 or TiO2. Samples of tuye`res, pottery and furnace wall/floor were also analysed (Table 2a, analyses EBL-1, 2 and 3; and Table 3). When compared to the slags, these ceramics show lower scatter for SiO2 (64.7–76.7 mass%), Al2O3 (8.2–19.0 mass%), Fe2O3 (3.8–10.2 mass%), CaO (0.2–1.9 mass%), MgO (0.5–1.6mass%), K2O (0.9–2.5 mass%) and TiO2 (0.4–0.6 mass%). The SnO2 content is generally below 1% (see also Hall and Grant, 1995) with the exception of two samples of furnace floor or wall (EBL-3 in Table 2a and SK-3 in Table 3), which have SnO2 contents of 5.1% and 6.9% respectively. The much higher SnO2 contents of the furnace wall samples are presumably a consequence of the infiltration of SnO vapour into the ceramic during smelting and its subsequent oxidation. The trace element compositions of the tuye`res from Smelterskop (Table 4) are fairly consistent, and also correspond to that of

Table 2a Chemical composition (major and minor oxides) in mass% of Elandsberg Ledge tuye´res (EBL-1,2), furnace wall (EBL-3), slags (EBL-4,5,7,8,9,10) and cassiterite–tourmaline– quartz ore (EBL-6), measured by WD-XRF. Sample

SiO2

Al2O3

Fe2O3

Cr2O3

MnO

MgO

Na2O

K2O

P2O5

SO3

NiO

SnO2

EBL-1 EBL-2 EBL-3

67.96 67.34 75.95

TiO2 0.48 0.49 0.52

14.99 15.07 8.24

9.58 10.17 3.87

0.09 0.1 0.04

0.03 0.12 0.08

0.98 1.07 0.73

CaO 0.19 0.19 1.92

0.43 0.36 0.38

1.69 1.75 2.08

0.15 0.12 0.12

0.02 0.04 0.02

0.04 0.03 0.04

0.06 0.08 5.05

EBL-4 EBL-5 EBL-7 EBL-8 EBL-9 EBL-10

40.98 41.94 70.35 44.94 42.02 35.65

10.21 7.32 4.74 10.23 10.88 8.29

7.65 8.95 3.37 8.82 9.23 6.03

9.32 11.13 5.02 9.45 7.36 9.99

0.13 0.13 0.06 0.17 0.15 0.11

0.25 0.26 0.06 0.29 0.12 0.16

1.87 2.53 1.32 3.35 2.64 2.51

6.32 8.09 3.86 6.96 15.47 9.03

0.63 0.63 0.38 0.67 0.57 0.66

1.82 1.55 0.67 1.89 2.45 2.33

0.98 0.64 0.48 0.93 1.13 1.07

0.03 0.04 0.01 0.03 0.03 0.02

0.01 0.04 0.02 0.01 0.01 0.01

EBL-6

50.73

0.46

1.03

4.15

0.03

0.03

0.44

0.04

0.41

0.02

0.05

0.01

0.01

SnO

11.81 14.71 9.31 7.08 3.32 20.93 41.01

ZrO2

H2O-

LOI

0.15 0.15 0.19

0.91 0.68 0.09

2.07 1.54 0.28

Total 99.82 99.31 99.04

9.01 4.12 3.62 6.41 5.38 5.91

0.01 0.03 0.02 0.02 0.01 0.01

2.21 2.64 2.91 1.96 1.19 3.63

98.82 99.47 100.38 99.29 99.58 99.08

0.16

0.07

1.29

99.94

1660

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

Table 2b Trace element content in ppm of Elandsberg Ledge tuye´res (EBL-1,2), furnace wall (EBL-3), slags (EBL-4,5,7,8,9,20) and cassiterite-tourmaline ore (EBL-6), measured by WDXRF. Sample

Mo

Nb

Y

Sr

U

Th

Rb

Pb

Ba

EBL-1 EBL-2 EBL-3 EBL-4

2 3 3 *

11 13 57 525

16 15 25 4052

21 21 68 274

<3.6 <3.6 <3.6 238

14 18 18 *

210 210 142 *

16 22 *

602 119 1698 120

Sc

EBL-5 EBL-7 EBL-8 EBL-9 EBL-10

* * * * *

823 626 640 677 550

1598 3107 2563 2251 2092

303 419 191 807 312

274 217 205 207 149

* * * * *

* * * * *

* * * * *

2299 99 246 985 207

4 107 142 190 93

<1.9 <1.9 <1.9 <1.9 <1.9

5 7 6 8 <1.3

EBL-6

<1.1

413

18

33

49

40

11

*

125

27

<1.9

96

26 26 3 <1.3

Zn

Cu

Ni

Co

Mn

Cr

Hf

W

50 55 9 <1.9

111 93 50 8

187 192 33 6

9 13 3.00 <2.4

262 927 602 1799

297 295 112 200

V 49 50 23 506

**

**

6 5 15 10 5

<2.4 <2.4 <2.4 <2.4 <2.4

1914 448 2030 888 1119

130 159 247 264 122

1198 484 739 1163 386

** ** ** ** **

** ** ** ** **

4

6

262

70

48

*Analyses unobtainable due to severe spectral overlap by Zr or Sn, **other traces detected but not subject to calibration.

the pottery sherd SK-6, suggesting that the tuye`res and the pottery were both made from the same clays, presumably local. The composition of potsherd SK-18 is however strikingly different, with relatively high levels of Fe2O3, Cr2O3, NiO and MgO, low levels of the trace elements V, Zr, Hf and Sc, and high levels of HREE. This pot was probably imported from an area where there are clays derived from mafic rocks. The compositions of the Smelterskop slags, tuye`res and pottery are displayed as the triangular sections CaO–Al2O3–SiO2 (Fig. 4a) and Fe2O3–Al2O3–SiO2 (Fig. 4b) of the quaternary system CaO– Fe2O3–Al2O3–SiO2, renormalized after removal of their MgO, alkali, TiO2 and SnO contents and LOI. In the CaO–Al2O3–SiO2 plot (Rankin diagram, Rankin and Wright, 1915) the slags occupy tight scatter ellipses with fairly constant SiO2 content (between 60 and 70 mass%) but varying CaO and Al2O3 contents (Fig. 4a). The ceramics are very poor in CaO and hence straddle the binary join Al2O3–SiO2. Shown is also the theoretical compositions of anorthite (CaAl2Si2O8) and the statistical center of gravity of the scatter ellipse of the slags at 9.9 mass% CaO, 20.8 mass% Al2O3, and 69.3 mass% SiO2. (This position in the diagram corresponds also to the ternary eutectic anorthite-mullite-tridymite with a melting point of 1345  C). In the ternary Fe2O3–Al2O3–SiO2 plot (Fig. 4b) the slags spread out widely owing to the scatter of the normalized Fe2O3 content from 6.5 to 43 mass%. Their normalized alumina and silica contents vary only between 10 and 28 mass% and 47 and 66 mass%, respectively. The average composition is 25.1 mass% Fe2O3, 17.4 mass% Al2O3 and 57.5 mass% SiO2 (open circle in Fig. 4b). As in the Rankin diagram, the tuye`res and the pottery occupy a tight scatter ellipse with a statistical center of gravity at 7.9 mass% Fe2O3, 14.4 mass% Al2O3 and 77.7 mass% SiO2. The normalized composition of the potsherds SK-6 and SKOP-3 (0.72 mass% CaO, 17.5 mass% Al2O3, 81.8 mass% SiO2) and the normalized average composition of the tuye`res (0.58 mass% CaO, 15.0 mass% Al2O3, 84.4 mass% SiO2) are so close that both appear to have been produced from the same type of clay. Note that in both diagrams the ratios of these three elements in the slags are significantly different from the same ratios in the tuye`res and furnace lining. Clearly the slags cannot have formed by simple reaction between cassiterite and the furnace refractories – there is excess CaO and Fe2O3 in the slags. As noted above, we can eliminate the possibility that there was more than a small amount of schorl (iron-rich tourmaline) in the ore charged to the furnace. Grant (1994) suggested that bone may have been added as a flux, but this is clearly ruled out by the low phosphorus contents of the slags (0.1–1.1 mass% P2O5; Table 1). The same applies to the apatite that occurs in some parts of the ore body (Stumpfl, 1960). Fuel ash from charcoal may have contributed some calcium, but is not a likely source of iron. We did find a large piece of pure haematite

and a grindstone with red ochre staining on Smelterskop, but we have not seen relict haematite grains in any of the thin sections of the slags. The nearest outcrop of haematite is a banded iron formation some 20 km NNW of the site. Another possible source of

Fig. 4. (a) Triangular diagram CaO–Al2O3–SiO2 of compositions of slags, tuye`res and pottery recalculated from data in Tables 1 and 2. (b) Triangular diagram Fe2O3–Al2O3– SiO2 of compositions of slags, tuyeres and pottery recalculated from data in Tables 1 and 2.

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

1661

Table 3 Chemical composition (in mass%) of tuye`res (SK-1,7 and 16; SKOP-1,2,5,6,7,8,9,10), furnaces wall (SK-3) and pottery (SK-6,18; SKOP-3), measured by XRF. Sample

SiO2

TiO2

Al2O3

Fe2O3

Cr2O3

MnO

MgO

CaO

Na2O

K2O

P2O5

SO3

NiO

SnO2

H2O

LOI

Total

SK-1 SK-3 SK-6 SK-7 SK-16 SK-18 SKOP-1 SKOP-2 SKOP-3 SKOP-4 SKOP-5 SKOP-6 SKOP-7 SKOP-8 SKOP-9 SKOP-10

73.48 75.49 70.58 64.74 74.46 65.94 72.66 76.61 68.08 60.50 68.76 72.32 66.42 70.97 66.85 73.14

0.58 0.52 0.57 0.51 0.58 0.43 0.48 0.62 0.43 0.34 0.58 0.49 0.49 0.51 0.46 0.63

11.36 8.46 12.76 13.59 12.04 11.63 10.64 13.04 13.16 18.97 15.33 13.27 14.38 13.76 13.21 12.39

4.91 4.23 6.77 7.86 5.04 9.07 6.59 5.14 7.57 6.57 8.21 8.86 9.44 7.51 7.66 8.77

0.04 0.04 0.06 0.07 0.04 0.16 0.07 0.04 0.07 0.03 0.07 0.09 0.09 0.08 0.08 0.08

0.06 0.05 0.08 0.08 0.05 0.14 0.03 0.08 0.06 0.03 0.10 0.08 0.04 0.05 0.03 0.05

0.71 0.54 1.16 1.22 0.76 4.93 0.68 0.97 0.94 0.54 1.55 0.91 0.84 0.78 0.74 1.09

0.39 0.75 0.51 0.85 0.36 0.54 0.37 0.31 0.37 0.89 0.73 0.29 0.24 0.26 0.42 0.94

0.51 0.19 0.31 0.22 0.46 0.27 0.26 0.55 0.29 1.17 0.32 0.20 0.24 0.13 0.15 0.25

1.61 1.89 2.19 2.26 1.63 1.61 1.55 1.62 1.98 0.87 2.48 1.84 1.88 1.74 1.50 1.97

1.41 0.14 1.12 1.49 0.46 0.37 0.25 0.05 0.24 0.27 0.08 0.11 0.42 0.24 0.14 0.15

0.01 0.02 0.11 0.01 0.03 0.03 0.01 0.01 0.01 0.03 0.01 0.02 0.02 0.02 0.02 0.01

0.01 0.01 0.02 0.02 0.01 0.04 0.02 0.01 0.02 0.01 0.02 0.02 0.02 0.02 0.02 0.02

0.01 6.88 0.01 0.01 0.06 0.11 0.04 1.27 0.09 0.01 0.23 1.02 0.30 0.36 0.04 2.38

1.73 0.16 0.93 2.33 1.26 0.63 2.39 0.11 0.89 2.55 0.31 0.25 1.31 0.80 3.19 0.02

2.59 0.31 2.12 4.06 2.68 3.43 4.07 0.67 5.44 7.29 1.58 1.06 3.34 2.49 5.71 0.07

99.41 99.68 99.30 99.32 99.92 99.33 100.11 101.10 99.46 100.07 100.36 100.83 99.47 99.72 100.22 101.82

iron oxide is the gossan through which many of the old mines were driven. Blocks of gossan around Rooiberg township today are a light or dark brown and appear to be composed of iron oxyhydroxides with or without quartz. Both CaO and FeO could also derive from the dark red carbonates – siderite (FeCO3) and ankerite (Ca (Fe, Mg, Mn)CO3) – that are associated with the tin mineralization (McDonald, 1912; Stumpfl, 1960; Leube and Stumpfl, 1963b). The specific gravity of siderite and ankerite (3.7–3.9) is higher than that of tourmaline (2.9–3.2), but it is still so much lower than cassiterite (6.8–7.1) that it seems unlikely that sufficient amounts would be retained in panning to account for the iron in the slags (15–30 mass% Fe2O3). Deliberate additions of carbonates, haematite or gossan would seem to be necessary to attain these levels. 5.2. Mineralogy and microstructures At Smelterskop free slags (those not attached to the distal ends of tuye`res) are generally glassy. In transmitted plane-polarized white light (PPL) they are typically a light honey brown in thin section, with obvious schlieren structures visible because of variation in the refractive index of the glass as a result of variation in chemical composition (Fig. 5). The glass almost invariably contains thermally fractured, partly dissolved, quartz grains. Thermal transformation of quartz to cristobalite has taken place along the cracks in the quartz. The occurrence of cristobalite establishes a minimum processing temperature, as the equilibrium temperature of the transformation of high-quartz to high-cristobalite is around 1050  C (see, for example Heimann, 1977) but does not give any clue to the actual operating temperature. In practice a superheat of at least 100  C above the liquidus temperature of the system would have been required to allow these slags to flow freely. Incompletely-dissolved grains of cassiterite are also present in some samples (Fig. 6). The more tin-rich slags from Smelterskop (Table 1) have higher ratios of crystals to glass, with abundant secondary cassiterite crystallized from the melt as skeletal laths or blocky crystals (Fig. 7). The Smelterskop slag specimens typically contain many tiny opaque crystals (Figs. 5 and 6) that often display strong red-orange colour under green epifluorescence illumination. Their compositions, and those of the adjacent glasses, were investigated in several samples by SEM/EDX and by microprobe analysis. Microprobe analyses (in atomic %) are given in Table 5, and show that these are complex spinels containing tin, vanadium and titanium in addition to the more usual iron, magnesium and aluminium oxides. Some of the spinels show marked zoning; Fig. 7 shows two spinels with

cores approximating hercynite-spinel solid solutions (Fe,Mg)Al2O4 and outer layers that become progressively richer in SnO and poorer in Al2O3. Tin spinels have not been previously investigated, so a separate paper on them has been submitted to a mineralogical journal. The spinel crystals in the glassy slags have apparently swept the glass clear of Fe3þ, V3þ and Ti4þ ions, so that these glasses around them show a considerably lower iron content in microprobe and SEM analyses than portions of the glass that lack spinels. This explains their lighter colour. This finding is in accord with the redox equilibrium 2½Fe3D  D ½Sn2D  4 ½Sn4D  D 2½Fe2D 

(1)

3þ

established for Fe -tinted float glasses (Johnson and Johnson, 2005). We suggest that on freezing of the slag the species on the right side of eq. (1) combine to form a ‘‘stannic’’ ulvo¨spinel (tin spinel), Fe2þ(Fe2þSn4þ)O4. All of the slags contain spherical tin prills. These rarely contain inclusions of the intermetallic compounds SnFe and SnFe2, known to tin smelters as hardhead. Slags with higher tin contents also contain skeletal laths of cassiterite precipitated from the melt (Figs. 6 and 7). Conversely, slags with the highest iron contents may contain laths or equiaxed crystals of fayalite (Fe2SiO4), though these

Table 4 Trace element content (in ppm) of a typical tuye`re (SK-16) and an exotic potsherd (SK-18) from Smelterskop, measured by ICP-MS.

Sc V Cr Co Ni Cu Rb Sr Y Zr Nb Cs Ba La Ce Pr Nd Sm Eu

SK-16

SK-18

9.23 59.5 150 11.2 53.8 47.5 118 34.0 13.3 101 8.94 2.99 319 29.2 53.7 6.5 24.6 4.3 0.89

19.9 112 830 46.3 303 59.0 165 34.9 11.4 68.4 4.39 6.70 358 22.8 49.4 4.89 18.6 3.24 0.84

Gd Tb Dy Ho Er Tm Yb Lu Hf Ta Pb Th U Zn Ga As Se Mo W

SK-16

SK-18

3.55 0.47 2.84 0.55 1.58 0.23 1.53 0.22 2.50 0.14 15.7 12.7 1.92 46.3 25.9 21.8 0.32 0.95 26.9

2.84 0.40 2.41 0.48 1.34 0.19 1.32 0.20 1.82 n.d. 54.2 6.60 1.37 51.4 26.8 14.9 0.63 1.05 13.0

1662

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

Fig. 5. Microstructure of a typical glassy tin slag from Smelterskop in transmitted PPL. Tiny opaque spinels and a relict quartz crystal are suspended in a glass with schlieren of varying colour and composition. The round bright areas are entrapped gas bubbles.

were seldom noted. Crystals of anorthite feldspar (Ca2Al2Si2O8) have also crystallized from the melt in those slags with the highest calcium contents. The black slags adhering to the ends of the tuye`res contain many white crystals. Several polished thin sections were cut to sample a gradient from unaltered tuye`re into the slag. The tuye`re fabrics consist of 20–40 area% of subangular to subrounded quartz, with subordinate alkali and plagioclase feldspar (sometimes sericitized) in a dark red clay matrix. The slag has attacked the clay fabric of the tuye`res very aggressively, freeing quartz crystals into the slag. The quartz crystals are heavily cracked by thermal expansion and display undulose extinction. They were actively reacting with the slag, as the size of the remnant quartz crystals diminishes with distance from the tuye`re/slag interface. Two samples of the flat-faced gritty ceramic slabs recovered from the slag dumps were prepared as polished thin sections. These have very different microstructures from the tuye`res, being predominantly composed of subangular quartz sand (50–60 area%) with rare feldspar and interstitial clay. Some cracks and voids near

Fig. 6. A zoned remnant grain of cassiterite surrounded by slag consisting of opaque cubes (spinels) and slender transparent laths (cassiterite) in a clear glass. (Transmitted PPL).

Fig. 7. Secondary electron image of a tin-rich slag sample (SKR1B) from Smelterskop. The white round areas are tin prills; the light skeletal laths and blocky crystals are cassiterite. The zoned spinels have cores of (Fe, Mg)Al 2O4 encased in complex exteriors containing Fe, Sn, Zr, Al, Mg, Ti and O. The black area is a gas bubble and the grey matrix is a Sn–Fe–Ca–Al–Si–O glass.

the outer surfaces have diffuse grey linings when viewed with a low-power objective; at high power these areas resolve into felted masses of tiny perfect parallel-sided colourless crystals with very high birefringence. These are cassiterite, and can only have penetrated the ceramic as a vapour phase. Loss of tin as SnO vapour is a well-known problem in modern tin smelting, so we assume that SnO vapour formed under strongly reducing conditions impregnated the pores of the kiln wall and was reoxidized there to SnO2. These observations correlate well with the bulk chemistry – note the higher SiO2/Al2O3 ratios and much higher SnO2 contents for analyses EBL-3 (Table 2a) and SK-3 (Table 3) when compared to analysed tuye`res from each site. The high-Zr, high-Ti slags from the Elandsberg Ledge site (Table 2a) have very different mineral assemblages. Most contain a much higher ratio of crystals to glass than the glassy Smelterskop slags, and all

Table 5 Composition for three Smelterskop slags of spinel crystals and the glasses in their immediate vicinity in atomic%, as determined by electron microprobe. (Oxygen calculated by stoichiometry). Sample

Si

Mg

SKR1b-3 SKR1b-4 SKR1b-5 SKR1b-8 SKR1b-10 SKR1b-12 SKR1b-glass

0.52 0.41 0.44 0.54 0.34 0.81 14.7

2.68 10.36 0.15 3.25 9.20 0.18 3.25 8.97 0.16 5.23 14.54 0.45 5.65 21.54 0.41 3.03 9.07 0.13 6.01

Al

Mn

Fe

Ti

V

O

3.81 4.78 4.78 2.83 1.15 4.52 8.3

0.11 0.11 0.12 0.07 0.07 0.11

22.09 21.92 22.12 16.54 11.73 21.98 4.8

2.48 1.99 2.01 1.32 0.66 2.14 0.12

1.41 1.78 1.80 1.91 1.52 1.53 0.02

56.38 56.38 56.35 56.57 56.94 56.49 62.7

1.05 1.22 0.36 0.19 12.4 0.31 0.32 11.9

0.02 0.05 0.02 0.02

15.46 19.44 10.48 10.96 3.6 0.04 11.86 2.7

2.81 4.35 0.91 1.02 0.16 1.66 0.12

0.73 2.43 1.07 1.84 0.1 1.12 0.1

56.79 56.48 57.06 57.04 63.5 56.97 64.5

0.01 0.01

0.01 0.01 0.32 0.09 0.33 0.03 0.36

0.17 0.09 0.06 0.75 0.11 0.3 0.02

57.08 57.06 62.8 57.05 62.6 57.12 62.8

SKR2a-18 0.50 SKR2a-19 0.35 SKR2a-20 0.47 SKR2a-21 0.48 SKR2a-glass1 12.7 SKR2a-24 0.40 SKR2a-glass2 13.5

4.94 4.01 5.91 5.51

SKJEb-29 SKJEb-30 SKJEb-glass1 SKJEb-31 SKJEb-glass2 SKJEb-32 SKJEb-glass3

9.83 27.95 0.05 9.76 27.83 0.14 6.6 8.31 25.76 1.11 7.1 9.05 27.32 0.49 6.7

0.03 0.03 13.3 0.11 13.3 0.07 13.7

17.5 11.13 23.4 22.22 5.1 5.65 21.66 7.0

Cr

Sn

0.20 0.53 0.31 0.73

0.03 0.04 9.6 0.09 8.8 0.09 9.0

0.02 0.02

4.82 5.03 3.2 6.73 3.3 5.5 3.3

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

contain relict quartz and zircon crystals (Fig. 8). These have been aggressively attacked by the molten slags, and the zircons (ZrSiO4) are often enclosed within rims of baddeleyite (ZrO2) (Fig. 9), which have precipitated from the Zr-rich melt immediately around the dissolving zircon. Abundant tiny (2–10 micron) cubes, octahedra and irregular crystals, honey brown in transmitted light, are dispersed through these slags. Although these are too small for accurate analyses by microbeam techniques, the EDAX spectra have strong peaks for Zr, Ti, Ca, Fe and O. Some of these crystals are isotropic and are thus identified as zirkelite – (Zr, Ti, Ca, Fe)O2. Others appear identical in plane-polarized light but are anisotropic; these are probably zirconolite – (Ca,Ce)Zr(Ti,Nb,Fe3þ)O7 – which has monoclinic, orthorhombic and trigonal forms (Strunz and Nickel, 2001). Miller and Hall (2008) have previously identified zirkelite in an anomalous high-Zr, high-Ti slag from Smelterskop, while zirconolite has been previously noted in a modern tin slag from Pennsylvania by Farthing (2002). Zirkelite and zirconolite crystals formed by reaction of zircons with a melt rich in Ti4þ ions. There are no relict titanium minerals in these thin sections, but in all cases the slags with zirkelite and/or zirconolite also contain crystals of rutile (TiO2) that have crystallized from the melt. These are rectangular laths, X-shaped interpenetrant twinned crystals or dendrites (e.g. Fig. 9) with high relief and are weakly pleochroic from red-brown to purple. Some of the slags have also precipitated acicular crystals of ilmenite (FeTiO3) (Fig. 8). The slag with the highest tin content (EBL-10) has also precipitated dendrites and spaghetti-like curved strands of colourless cassiterite. In all sections the matrix is a glass, varying from clear through yellow-brown to deep brown. The iron contents of the EBL slags are low (maximum 11.1 mass%) so there is no fayalite, spinel or wustite in any of these sections. From their morphology, however, it is clear that the slags were free flowing when molten. 5.3. The glassy state in tin slags The composition of ceramics including glasses may be expressed in simplified form by grouping components with chemically similar character. The components SiO2, Al2O3 and CaO behave like glass network formers (NF), network intermediates (NI) and network modifiers (NM), respectively. Glasses with SiO2 contents above 60 mass% and CaO contents up to 50 mass% are highly viscous below 1200  C and tend to solidify as vitreous slags

Fig. 8. Relict zircons (large subrounded crystals) in a slag consisting of ilmenite (opaque) and rutile (medium grey) in clear glass. (Transmitted PPL).

1663

Fig. 9. Backscattered electron image of a relict zircon crystal (grey) that was being dissolved by the slag; bright crystallites of baddeleyite (ZrO2) have precipitated from the adjacent melt. Isolated crystals of baddeleyite in the melt have overgrowths of zirconolite (CaZr(Ti, Fe3þ)O7). The dark grey dendrites are rutile (TiO2) and the light grey matrix is a glass.

(McCaffery et al., 1932; Kingery et al., 1976). However, to account for the influence of other NFs such as SnO2, NMs such as MgO, Na2O, and K2O as well as NIs such as Fe2O3, TiO2 and ZrO2, the sums of these oxides were plotted on a ‘glass triangle’ as shown in Fig. 10. These additional oxides shift the center of gravity of the slag composition only slightly towards the NI corner and hence away from the NF corner but nevertheless have some influence on the flow point, lowering it to below 1200  C (Krohn et al., 2005). Whether an oxide acts as glass network former (NF), intermediate (NI) or modifier (NM) depends on the attractive force between the cation R and the oxygen ions O in the R–O coordination polyhedron forming the short range ordered (SRO) glass structure. This ‘bond strength’ can loosely be quantified by the socalled Dietzel parameter (Dietzel, 1941) F ¼ 2z/a2 (z ¼ charge of cation, a ¼ distance R–O in Å) with F > 2.5 for network former oxides, 1.0 < F < 2.0 for network intermediates, and F < 1.0 for network modifiers. Network intermediates can further be

Fig. 10. Ternary diagram of the compositions of slags from Smelterskop (Table 1; filled circles) and Elandsberg Ledge (Table 2; open circles) recalculated so that the sum of network-forming, network-modifying and network-immediate elements equals 100%. The square is the calculated average composition.

1664

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

subdivided into oxides with 1.7 < F < 2.0 (NI(1); oxides able to replace SiO2 as NFs, for example SnO2) and those with 1.0 < F < 1.6 (NI(2); oxides acting as either NFs or NMs depending on their concentration). The following approximate F-values are assumed here:    

network formers NF: B2O3 3.22, SiO2 3.14 network intermediates NI(1): TiO2 2.08, SnO2 1.89, Al2O3 1.69 network intermediates NI(2): Fe2O3 1.57, ZrO2 1.55, SnO 0.96 network modifiers NM: MgO 0.95, FeO 0.87, CaO 0.69, Na2O 0.35, K2O 0.27

plots the concentration in glass of SnO in mass% against Al2O3 and SiO2, respectively. Note that the negative linear correlation of SnO vs. SiO2 concentration is even better than with the XRF analyses: [SnO] ¼ 97.5–1.77$[SiO2]; r2 ¼ 0.94. The SnO vs. Al2O3 correlation is markedly improved: [SnO] ¼ 95.6–5.18$[Al2O3]; r2 ¼ 0.89. In contrast, the poor negative correlation between SnO and FeO shown in Fig. 13 can be expressed by a second-degree polynomial – [SnO] ¼ 89.41–6.69$[FeO] þ 0.12$[FeO]2 with s ¼ 2.001. These findings strongly suggest that under reducing conditions tin oxide replaces silica as the glass-forming oxide compound, whereas FeO acts as a network modifier in accord with the view by Johnson and Johnson (2005). According to the value of the Dietzel parameter (Dietzel, 1941) for Sn2þ (F ¼ 0.96 Å2) and Sn4þ (F ¼ 1.89 Å2) these species behave like conditional network formers and true network formers, respectively (Carbo´ No´ver and Williamson, 1967; Ishikawa and Akagi, 1978) but Fe2þ (F ¼ 0.87 Å2) acts as a strong network modifier. High concentrations of SiO2, Al2O3 and in particular FeO in the glass go along with low concentrations of SnO and vice versa. This is also true for the concentrations of K2O, MnO and TiO2. These findings stand in stark contrast to modern tin-smelting slags, in which tin is said to be positively correlated with silica (Tylecote et al., 1989). The smaller the FeO/SnO ratio the higher is the viscosity. There appears to be also a relation between the FeO/SnO ratio and the ‘glassiness’ of the slags: in the range 0.8 < [FeO/SnO] < 4.0 slags contain comparatively many spinel and olivine crystals whereas slags with [FeO/SnO]< 0.8 appear to be more or less fully glassy. This is consistent with the notion that replacement of glass-forming oxides by SnO increases the connectivity in the glass network by elimination of non-bridging oxygen. The practical implication of this is that in less glassy samples – i.e. samples with high FeO/SnO ratios and thus lower viscosity – lower tin losses during smelting may be expected.

Fig. 10 shows that the glass composition of the slags from Smelterskop and Elandsberg Ledge consists of network-former (NF) oxides (SiO2 þ SnO2 þ SnO) between 50 and 80 mass%, network intermediate (NI) oxides (Al2O3 þ Fe2O3 þ TiO2 þ ZrO2) between 15 and 40 mass%, and network modifier (NM) oxides (CaO þ MgO þ Na2O þ K2O) below 20 mass%. (Since ZrO2 is known to be a conditional network intermediate oxide it has been included into this glass-forming oxide group.) The statistical center of gravity of the scatter ellipse is at NF ¼ 60.4 mass%, NI ¼ 29.4 mass% and NM ¼ 10.2 mass% (mean Smelterskop: NF ¼ 61.2 mass%, NI ¼ 29.3 mass%, NM ¼ 9.5 mass%; mean Elandsberg Ledge: NF ¼ 57.7 mass%, NI ¼ 30.0 mass%, NM ¼ 12.3 mass%). Owing to the high proportion of NFs, these slags are expected to solidify during quenching in the glassy state. Although the silica content of the glassy slags is rather low (average: 38.1 mass%; range 20.9–53.0 mass%), tin oxide (average: 22.8 mass%; range 1.7–61.7 mass%) is acting as an effective conditional network former in line with assumptions by Karim and Holland (1995) and Holland et al. (2003). As expected the amounts of SiO2 and SnO are inversely correlated (Fig. 11). The equation of the fitted line is [SnO] ¼ 85.2– 1.64[SiO2] with r2 ¼ 0.85. The correlations of other oxides (Al2O3, Fe2O3, MgO, CaO) with tin oxide are poor or nil. For example, alumina correlates only weakly with SnO as shown by the low r2-value of 0.32 of the equation [SnO] ¼ 48.4–2.24$[Al2O3]. Since our XRF analyses are bulk analyses, calculations based on them include remnant (undissolved) crystals in the slags, especially quartz and cassiterite. These correlations improve markedly if we use data from energy-dispersive spot analyses of glasses (as measured on the SEM) instead of the bulk XRF analyses. Fig. 12

ICP-MS trace element analyses of the Smelterskop slags (Table 6) show tungsten concentrations in the range of 1300–5000 ppm, but their contents of tantalum (0.2–3.0 ppm) and niobium (20– 80 ppm) are low. This is the expected pattern for a low-temperature cassiterite-sulphide tin ore assemblage (Taylor, 1979). This pattern distinguishes Rooiberg ore from the high-temperature pegmatite

Fig. 11. Plot of SnO vs. SiO2 in slags from Smelterskop and Elandsberg Ledge (bulk compositions, measured by WD-XRF).

Fig. 12. Plots of SnO vs. Al2O3 (top) and SnO vs. SiO2 (bottom), as determined by EDS spot analyses of the glass phase within three tin slags from Smelterskop.

5.4. Trace elements in slags and ceramics

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

1665

Table 6 Trace element content (in ppm) of some Smelterskop slags, measured by ICP-MS.

Fig. 13. Plots of SnO vs. FeO, as determined by EDS spot analyses of the glass phase within three tin slags from Smelterskop.

tin deposits, such as those of Zimbabwe, which are rich in Ta and Nb but poor in W (von Knorring and Condliffe, 1987). The rare earth element (REE) ratios shown in Fig. 14 also conform to the expected pattern for cassiterite-sulphide deposits. Zirconium levels in the Smelterskop slags analysed here vary from 300 to more than 6000 ppm. Zircon is not part of the ore assemblage, but is an accessory mineral in the host rock (arkose). Since zircon (ZrSiO4) has relatively high specific gravity (4.6–4.7) it would tend to be retained with cassiterite (6.8–7.1) during beneficiation of the ore by panning. Some of the slags from Elandsberg Ledge (Table 2a) contain much higher concentrations of zirconium (up to 8.70 mass% ZrO2). As noted above, this suggests that these ores derive from alluvial placers in which zircon from the arkose would be concentrated with cassiterite, titanium minerals – e.g. rutile (SG 4.2–5.5) or ilmenite (SG 4.7–4.8) – and iron, copper or lead sulphides. This deduction is also supported by the fact that the EBL slags which have high zirconia contents also have higher contents of yttrium, as xenotime (YPO4; SG 4.3–5.1) is often concentrated in alluvial placer deposits. One piece of slag from Smelterskop, published by Miller and Hall (2008), also has high ZrO2 and TiO2 and was interpreted as produced by smelting ore from an alluvial placer deposit. Strontium levels in the Smelterskop slags are between 125 and 327 ppm, but are much lower in the tuye`res and pottery. Although wood ash is often considered a potential source of Ca and Sr in early metallurgical slags there is no detectable pattern of covariation of these elements with K and Na, which are ubiquituous in fuel ash. Thus at least some of the Ca and Sr likely derives from the carbonates that cross-cut and post-date the emplacement of cassiterite in the northern Rooiberg valley deposits. The strontium levels of the Elandsberg Ledge slags are significantly higher than those of the Smelterskop slags (range: 200–800 ppm; measured by XRF) and appear to be correlated with high zirconia (Table 2b). Levels of uranium and thorium are much higher in the slags (64–137 ppm U; 28–74 ppm Th) than in the tuye`re and potsherd (Table 4: 1–2 ppm U; 7–13 ppm Th). Uranium and thorium readily substitute for tin within the crystal lattice of cassiterite, and tin slags have attracted some investigation as potential radiation hazards (Farthing, 2002). The high U and Th levels in Rooiberg cassiterite have two useful consequence for archaeology. Firstly, tuye`res within the slag heaps at Smelterskop have experienced very high dose rates of gamma radiation (as measured by buried dosimeters), enhancing the precision of measurement in optically stimulated luminescence dating of the tuye`re fabric (James Feathers and Dana Drake Rosenstein, pers. com. to DJK). Secondly, as the radioactive isotopes of U and Th decay to 208Pb, 207Pb and 206 Pb after crystallization, while the content of 204Pb remains fixed, the lead isotope signature of cassiterite crystals evolves through time. In a separate study, we show that a lead isotope isochron obtained on tin prills from Smelterskop gives the correct geological

SK-11

SK-12

SK-13

SK-14

SK-15

SK-17

SK-19

Sc V Cr Co Ni Cu Rb Sr Y Zr Nb Cs Ba La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Tm Yb Lu Hf Ta Pb Th U

67.5 1238 124 10.1 26.1 14.2 66.4 219 30.0 507 48.8 1.54 316 139 248 27.8 103 18.5 7.45 13.6 1.56 7.41 1.25 3.39 0.51 3.90 0.71 10.6 1.09 78.0 33.1 83.5

70.3 1162 161 6.57 16.8 11.1 99.5 187 33.2 718 47.7 1.46 438 145 259 28.6 106 18.4 7.91 13.6 1.66 8.22 1.41 3.91 0.60 4.52 0.80 15.9 1.30 26.4 33.0 69.7

44.4 813 138 13.2 17.9 8.91 71.7 167 15.7 473 25.6 1.49 278 21.9 39.0 4.52 17.0 3.28 1.30 3.42 0.51 3.07 0.62 1.98 0.33 3.01 0.60 10.7 0.61 17.8 19.3 74.0

51.1 853 202 3.37 8.15 7.45 80.8 125 77.7 6236 68.5 1.54 315 324 593 62.1 224 36.2 11.4 25.2 3.12 16.4 3.07 8.95 1.42 11.0 1.76 109 3.09 21.3 74.2 64.1

80.9 1469 174 3.73 11.8 6.48 93.7 321 36.7 791 70.8 1.07 737 158 281 31.1 118 21.5 9.32 17.1 2.11 10.5 1.71 4.38 0.64 4.84 0.85 17.2 2.28 19.0 41.0 85.5

50.1 977 121 11.0 19.7 8.15 68.6 168 20.6 610 32.5 1.80 298 45.6 83.3 9.32 34.8 6.29 2.41 5.58 0.74 4.18 0.81 2.40 0.41 3.50 0.71 13.7 0.85 111 25.7 93.1

58.4 1026 161 6.45 13.2 13.5 75.9 171 48.6 2613 52.2 1.39 335 208 383 40.8 149 24.7 8.73 17.7 2.14 10.9 1.97 5.60 0.85 6.61 1.12 43.8 1.77 35.8 53.7 72.9

Zn Ga As Se Mo W

57.5 44.8 11.3 0.80 6.08 3899

106 56.5 12.1 0.45 8.61 3286

51.2 59.5 19.3 n.d. 9.11 3473

111 37.3 16.3 2.77 9.11 3316

92.2 66.7 15.8 n.d. 10.7 4451

61.2 49.0 18.1 n.d. 8.61 4096

85.4 48.7 12.0 1.24 18.2 5022

Sc V Cr Co Ni Cu Rb Sr Y Zr Nb Cs Ba La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Tm Yb Lu Hf Ta Pb Th U Zn Ga

SK20

SK21

SK22

SK23

SK24

SK25

SK26

70.3 1409 180 4.40 9.95 6.52 86.0 327 41.8 1355 49.0 1.71 603 146 264 29.2 108 18.6 7.14 14.0 1.76 9.08 1.68 4.90 0.78 6.11 1.11 28.1 1.45 11.7 59.9 110 72.5 61.9

78.4 1328 184 11.5 17.6 9.72 68.8 219 55.3 699 76.6 1.47 612 46.9 64.3 9.16 34.0 7.26 2.95 10.3 1.78 11.2 2.23 6.43 0.96 7.05 1.20 15.9 1.90 60.8 34.5 80.2 81.0 70.4

71.3 1241 184 5.52 15.6 7.19 106 239 41.4 628 49.4 1.80 593 89.8 161 18.6 68.9 12.4 5.13 10.4 1.48 8.23 1.59 4.72 0.74 5.64 0.97 14.0 1.26 16.2 35.9 85.8 73.3 62.8

47.3 840 149 11.5 14.4 10.6 59.9 200 18.7 716 29.8 1.34 284 25.5 46.8 5.44 20.6 4.07 1.63 4.06 0.62 3.69 0.74 2.33 0.39 3.42 0.68 16.0 0.83 26.6 28.3 85.1 66.0 63.3

96.4 1707 203 0.97 5.39 8.83 92.5 255 58.3 1687 83.5 2.17 396 312 581 62.7 231 39.1 14.9 26.6 3.07 14.6 2.50 6.89 1.08 8.39 1.52 34.1 2.75 5.04 67.5 137 101 22.2

66.7 1108 116 10.4 23.3 29.0 67.6 235 35.1 537 39.3 1.41 262 144 258 29.0 109 19.6 7.85 14.4 1.72 8.32 1.43 3.89 0.59 4.43 0.79 11.0 0.72 86.7 35.1 82.1 55.1 45.8

39.5 583 60.7 14.8 24.7 96.3 43.3 103 26.8 294 20.1 1.17 175 138 254 27.7 104 18.5 7.46 12.9 1.47 6.79 1.15 3.05 0.45 3.42 0.58 6.23 0.20 624 27.8 49.2 39.2 33.8

1666

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

Table 6 (continued)

As Se Mo W

SK20

SK21

SK22

SK23

SK24

SK25

SK26

13.7 n.d. 8.25 4715

15.0 1.59 5.06 2737

13.2 0.092 7.04 3771

13.7 0.85 10.7 3592

6.53 0.98 2.24 1286

11.8 1.41 5.17 2837

26.4 2.69 5.58 1466

age (ca. 2.05 billion years), and thus serves as a ‘‘fingerprint’’ for this ore deposit (Molofsky et al., in press). The distribution of rare earth elements (REE) deserves special attention. The REE data in Table 6 are reorganized in Table 7 as sums of light rare earth elements (LREE) between La and Sm, heavy rare earth elements (HREE) between Eu and Lu, and the sum of Sc and Y. Fig. 14 shows a triangular plot of these three groups of REE in slags, as well as a representative potsherd and tuye`re (Table 8; data from Table 4). The slags show a very low and constant ratio of HREE of 8– 10 mass% but a trend towards LREE (50–85 mass%) associated with a decrease of the sum of Sc þ Y from 40 to 10%. Grant (1999) has shown that tin ingots recovered from the Rooiberg area form a tight grouping on LREE, HREE, and Ta/Sc ratios, and our results accord well with his. Grant also compared these ratios to REE ratios in some other African cassiterites, and claimed that the Rooiberg ratios are distinct. The main weakness of this argument is that most of the other deposits were represented by only one or two analyses. More study of trace elements in these other tin deposits would be needed to determine whether REE provide a reliable means of inferring the provenance of tin ingots in southern Africa. In contrast to the slags typical ceramics are much richer in HREE (60%) with only 30% LREE and about 10% Sc þ Y (Table 8). Petrographic analysis shows that the rock and mineral fragments in the ceramics derive largely from the arkose, and presumably the clays derive from alteration of the feldspars in the arkose to clay minerals. 6. The technology of tin smelting at Rooiberg Tin smelting in modern industrial practice is a two-stage process. In the first stage cassiterite is reduced with coke in reverberatory furnaces or toploader kilns (Wright, 1982): SnO2 D 2CO / Sn D 2CO2

(2)

The slags originating from the reduction of cassiterite in the presence of some silicate gangue contain much tin as a viscous tin

Table 7 Concentrations (in ppm) of light rare earth elements (LREE; La – Sm; A), heavy rare earth elements (HREE; Eu – Lu; B) and Sc þ Y (C) in Rooiberg slags.

Sc Y S(Sc þ Y) SL-REE SH-REE

SS %C %A %B S(A þ B þ C)

Sc Y S (Sc þ Y) SL-REE SH-REE

SS %C %A %B S(A þ B þ C)

SK-11

SK-12

SK-13

SK-14

SK-15

SK-17

SK-19

67.5 30 97.5 536.3 39.78 673.58 14.47 79.62 5.91 100

70.3 33.2 103.5 557 42.63 703.13 14.72 79.22 6.06 100

44.4 15.7 60.1 85.7 14.84 160.64 37.41 53.35 9.24 100

51.1 77.7 128.8 1239.3 82.32 1450.42 8.88 85.44 5.68 100

80.9 36.7 117.6 609.6 51.45 778.65 15.1 78.29 6.61 100

50.1 20.6 70.7 179.31 20.74 270.75 26.11 66.23 7.66 100

58.4 48.6 107 805.5 55.62 968.12 11.05 83.2 5.75 100

SK20

SK21

SK22

SK23

SK24

SK25

SK26

70.3 41.8 112.1 565.8 46.56 724.46 15.47 78.1 6.43 100

78.4 55.3 133.7 161.62 44.1 339.42 39.39 47.62 12.99 100

71.3 41.4 112.7 350.7 38.9 502.3 22.44 69.82 7.74 100

47.3 18.7 66 102.41 17.56 185.97 35.49 55.07 9.44 100

96.4 58.3 154.7 1225.8 79.55 1460.05 10.59 83.92 5.45 99.96

66.7 35.1 101.8 559.6 43.42 704.82 14.44 79.4 6.16 100

39.5 26.8 66.3 542.2 37.27 645.77 10.27 83.96 5.77 100

metasilicate, SnSiO3 (Carbo´ No´ver and Williamson, 1967) that cannot be reduced with coke. This combined tin is recovered by resmelting with lime and coke (eq. (3)), or by heating with iron scrap or ferrosilicon (eq. (4)) in a reverberatory furnace according to SnSiO3 D CaO D C / Sn D CaSiO3 DCO

(3)

SnSiO3 D Feðor Fe5 Si2 Þ / FeSiO3 DSn

(4)

Hence there are two generations of tin – raw ‘primary’ tin (eq. (2)) and ‘secondary’ tin from resmelting slag (eqs. 3 and 4). The raw tin is contaminated with reduced iron that reacts with tin to form Sn2Fe or SnFe (Mao and Dahn, 1999), the so-called ‘hardhead’. Removal of ‘hardhead’ during refinery operations involves reheating under oxidizing atmospheres at temperatures slightly above the melting point of tin (232  C), at which temperature the solubility of iron in tin is at a minimum (0.001%). The smelting technology employed at Rooiberg was evidently very different. There appears to have been only a single stage of smelting. We have argued, from differences in the composition of the slags, that two types of ore were used. Slag samples from the Elandsberg Ledge site (EBL) have high ZrO2 and TiO2 contents. This suggests the use at EBL of alluvial tin ore placer deposits, in which detrital zircon and titanium minerals were concentrated together with cassiterite. Zircon is an accessory mineral in the arkose that hosts the tin ore deposits in the northern Rooiberg Valley (Stumpfl,

Table 8 Concentrations (in ppm) of light rare earth elements (LREE; La – Sm; A), heavy rare earth elements (HREE; Eu – Lu; B) and Sc þ Y (C) in Rooiberg ceramics (tuye`res and pottery).

Sc Y S(Sc þ Y) ¼ C SL-REE ¼ A SH-REE ¼ B

SS

Fig. 14. Ternary plot of light rare earth elements (LREE) and heavy rare earth elements (HREE) against the sum of scandium and yttrium for Smelterskop slags and ceramics.

%C %A %B SA þ B þ C (%)

SK-16

SK-18

9.23 13.3 22.53 45.06 95.31 162.9 13.83 27.66 58.51 100

19.9 11.4 31.3 62.6 129.28 223.18 14.02 28.05 57.93 100

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

1960; Leube and Stumpfl, 1963b). Erosion around the outcrop would therefore liberate both zircon and cassiterite, which would be concentrated in the same placer deposits in the stream below the outcrop. The source of the titanium minerals in the placers is not known, but the rhyolites of the Elandsberg and Rooiberg ranges are a possible source. (There are no placers in the stream today, as pumping of water from the modern mine has scoured the stream bed to bare rock). With one exception (Miller and Hall, 2008) the analysed slags from the Smelterskop site have low contents of zirconia and titania and thus appear to derive from zircon-poor tin ore won by underground mining. We have argued that these ores were highly concentrated – certainly much more concentrated than in modern tin smelting at Rooiberg – and relatively little slag was produced per unit of tin metal. The beneficiation was presumably carried out by panning the crushed ores in the stream between the A3 mine and Smelterskop. Although the abundance of slagged tuye`re ends on Smelterskop indicates that smelting took place in furnaces rather than crucibles, we have found no furnaces, and believe that they must have been deliberately destroyed. Several small pieces of sandy furnace lining were however discovered in the slag heaps. The two analysed examples (Table 2a, analysis EBL-3; Table 3, analysis SK-3) show >5 mass% SnO2, which microscopic examination shows has been deposited from a vapour phase in cracks and voids. This evidence, together with the lack of slag attack, indicates that these are pieces of wall rather than floor. Since these pieces are flat rather than curved, they seem to indicate that the furnaces were rectangular in plan. The only furnaces previously discovered in the Rooiberg valley (Wagner and Gordon, 1929) were unlined round or elliptical structures made of piled rocks; it is not clear whether these were smelting or crucible furnaces. The scarcity of the furnace linings relative to the numbers of tuye`res on Smelterskop suggests that linings were reprocessed in some fashion, while the tuye`res were simply discarded. Since the ratios of rare earth elements (REE) in the analysed slags show dominance of LREE, whereas the claybased materials show dominance of HREE (Fig. 14) it does not appear that the linings were simply resmelted, but we are unable to suggest what else may have been done with them. We did not find any masses of primary smelted tin, but three ‘‘bun ingots’’ have been found in the Rooiberg valley over the last century (Friede and Steel, 1976; Grant, 1994). All are elliptical, with dimensions (major axis/minor axis/depth) of 95/75/35 mm, 150/95/ 58 mm and 156/100/51 mm. The masses were 800, 2100 and 2000 g respectively (Grant, 1994, Table 1). The shapes of these ingots indicate either (1) that the bases of the smelting furnaces were elliptical bowls, or (2) that tin was tapped from the furnace into elliptical basins. All three of these are of very pure tin, with a maximum contents of 0.46% Fe and 0.36% Cu in several samples of each, with most analyses much lower than these maxima (Grant, 1994, Table 2). The prills within our slags are generally very pure, as measured by EDAX. In only a couple of instances did we find hardhead inclusions within the metallic tin. Lead isotopic analysis (Molofsky et al., in press) of eight tin ingots discovered over the last century from northern South Africa and southern Zimbabwe shows that all derive from tin ores of Bushveld Magmatic Province age (2.05 billion years). Since Rooiberg is the only known prehistoric tin mine in the Bushveld Magmatic Province, these were probably all produced there. Miller and Hall (2008, Table 8) provide point-count data on the abundance of hardhead in seven of these ingots, including two of the three from Rooiberg. In six of the seven instances, the ingots contained between 0.2% and 2.8 volume% hardhead, corresponding to at most 1.0 mass% iron. The one exception was an ingot previously reported

1667

by Killick (1991), which contained 16.0 volume% hardhead, corresponding to 3.76 mass% iron. The evidence does therefore suggest that tin produced at Rooiberg was generally very pure and thus needed little or no refining. Grant (1994) has argued that the amount of iron oxide in the furnace feed at Rooiberg was kept low to avoid loss of tin to the formation of hardhead, and that in order to avoid this large losses of tin to the slag must be accepted. We agree with his arguments. The slags that we have analysed are rich in tin, most of which is chemically bound as a glass-forming element, and the contents of iron and calcium are variable but always low when compared to prehistoric iron- and copper-smelting slags. It is however apparent from the major chemical data that the Smelterskop slags cannot have been produced simply by reaction between cassiterite and the tuye`res. As noted above, there is excess calcium and iron in the slags. We have been unable to determine how these elements entered the slags. Fuel ash could have been a source of calcium, but not iron; additions of gossan or haematite could have supplied iron but not calcium; and ankerite could have added both elements, plus some magnesium. At present we cannot decide which of these possibilities apply, or whether the additions were deliberate. The chemical compositions of these slags are certainly very different from precolonial copper-smelting slags in southern Africa, all analysed examples of which are heavily fluxed with iron oxides (Miller and Killick, 2004). How the tin-workers developed the lowiron slags left at Smelterskop, which must represent a late stage of the technology, is a question that cannot be answered until earlier tin-smelting sites are discovered and studied. At Smelterskop an extensive scatter of at least 1000 kg of broken slag suggests attempts to increase tin yields by recovering prills. Since the size of the smallest broken fragments is 8–10 mm, and almost all prills seem in thin section are smaller than 1 mm, this would not appear to be worth the effort involved. Possibly the broken slag represents only the first part of the process. It may have been waiting for transport down to the stream between the hill and the mines for crushing and panning, but the site was abandoned before all the broken slag could be processed. We did not find any crucible sherds at Smelterskop or the EBL site. On present evidence, both are solely smelting sites, and the tin produced there was taken elsewhere to be made into ingots, or to be alloyed with copper to make bronze. The only site in the Rooiberg valley that has produced metalworking crucibles so far is Rooikrans (Fig. 2) where two sherds had attached crucible slags containing drops of copper, tin and bronze (Hall, 1981; Miller and Hall, 2008). There are at least six small deposits of copper ore in the valley (Crocker et al., 2001), and thin surface scatters of arsenical copper ore and slag, smelted arsenical copper and small copper bars of rectangular section were recovered in the Blaaubank drainage (Fig. 2) during field survey in 2006. 7. Comparisons with other preindustrial tin slags In spite of the importance of tin and bronze in world history, there has been remarkably little study of preindustrial tin production technology. Almost all of the published comparative data comes from Cornwall and Devon in southwestern Britain (Tylecote et al., 1989; Malham et al., 2002; Farthing, 2002; Malham and McDonnell, 2003). These can be divided into three groups by the smelting technology employed. The first group is from furnaces that are presumed to have been blown by hand (17th century BCE to 14th century CE); the second comprises slags from water-powered ‘‘blowing houses’’ (14th–18th centuries CE); and the third group of slags is from reverberatory furnaces (18th–20th centuries). Slags of similar composition to those reported here from Rooiberg

1668

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669

occur in the first and second groups from Cornwall and Devon, but not in the third. The earliest analysed tin slag, from a burial at Caerloggas (ca. 1600 cal BCE) has around 40 mass% SnO, but very low FeO and CaO. Those from Crift Farm, a hand-blown smelting site dated around 1200 cal CE, have compositions quite similar to the lower range of Rooiberg slags (10–20 mass% SnO), though the iron and calcium oxide contents are generally lower, and the sodium and potassium contents higher in the Crift Farm slags (Malham et al., 2002: Table 1). These differences appear to relate to differences in the host rocks – granite in the case of Crift Farm, arkose at Rooiberg – and perhaps to deliberate additions of iron and calcium carbonates at Rooiberg in excess of remainders of the sideritic-ankeritic envelope of the mined ore. Slags from the medieval ‘‘blowing houses’’ generally have lower tin contents than Rooiberg slags, but there are exceptions that include Retallack, Cornwall (16–17th centuries CE; Tylecote et al., 1989), and Longstone, Devon (early 16th century CE.; Malham and McDonnell, 2003). Both of these have strikingly similar major element compositions to some of our Smelterskop slags. As a group, early and medieval tin slags from southwestern England tend to have higher TiO2 contents (2–15 mass%) than Smelterskop slags, reflecting the concentration of accessory rutile and ilmenite from the granites with cassiterite in English alluvial placers. These are more like our Elandsberg Ledge slags, which also appear to derive from alluvial placers. The only mineralogical studies of prehistoric and historic tin slags from Cornwall and Devon are those of Farthing (2002), whose samples derived mostly from reverberatory furnaces. She did however identify tin prills, spinels, relict quartz and mullite in samples from Crift Farm, and tin prills, zircon dendrites, and relict xenotime, potassium feldspar and quartz in ‘‘blowing house’’ slags from Merrivale and Week Ford. Ilmenorutile had crystallized from the melt in the Merrivale slag, and an unidentified Al–Si–Ti–Zr phase from the Week Ford samples (Farthing, 2002, pp. 72–83, 188– 190). The reverbatory furnace slags, which have much higher iron and much lower tin contents than Rooiberg slags, contained a wider range of minerals, including olivines, spinels, aenigmatite, rho¨nite, hedenbergite, anorthite, and melilite, along with metallic tin, iron and tungsten (Farthing, 2002, pp. 84–108, 188–190). 8. Conclusions The Rooiberg tin mines and Smelterskop were first described in print over a century ago. The preindustrial mines have been largely obliterated by subsequent mining, and much archaeological material has been removed from Smelterskop by casual visitors. The site was also damaged by the extensive excavations of Mason and Steel in the 1980’s, which were never published. This study has documented some aspects of the tin-smelting technology in use at Smelterskop in the late stages of tin production between ca. 1650 and 1850 CE. (Attempts to date this more precisely by optically stimulated luminescence (OSL) are in progress). The tin-workers at Smelterskop used cassiterite ore obtained by hard-rock mining. The Elandsberg Ledge smelting site may be a little earlier (15th–17th centuries cal CE) and used ores obtained from alluvial placer deposits. We know nothing as yet about earlier stages of tin mining and smelting in southern Africa, which may extend back to the thirteenth century (Killick, 2009b). Nor are the origins of the technology understood. Since there is no evidence at all for either tin or bronze in southern Africa before the thirteenth century, about 300 years after the first evidence of connection of the Limpopo valley to the Indian Ocean maritime trade, were these technologies introduced from abroad? Were there changes in the technology of tin production over time? Was tin at first combined with local copper

to make bronze, and only later traded within southern Africa as tin ingots? Why has evidence for prehistoric production of tin only been found at Rooiberg, when there are at least a dozen other tin deposits in northern South Africa, Swaziland and Zimbabwe? And was all the tin produced consumed within southern Africa, or was some exported through the Indian Ocean maritime trade? Answers to these questions must await future research. Acknowledgements Field archaeology and laboratory studies were funded by U.S. National Science Foundation Grant BCS-0542135. Field archaeology at Rooiberg was also supported by grants from the South African National Research Foundation project to Dr Simon Hall (University of Cape Town). Dr Hall kindly supplied Figs. 1 and 2 in this paper. The authors are particularly grateful to Dr David Reid (Department of Geology, University of Cape Town) for developing the XRF protocol for analysis of tin- and zirconium-rich materials, and to Dr Andreas Spath for the ICP-MS trace element analyses. At the University of Arizona, we thank Gary Chandler and Steven Hernandez (Materials Science and Engineering) for assistance with SEM, and Dr Ken Domanik (Lunar and Planetary Sciences) for help with the microprobe. Dr Duncan Miller and Dr Simon Hall discussed these results with us, and practical assistance rendered by Thomas Fenn and Dana Drake Rosenstein is much appreciated. This paper has been greatly improved by helpful criticisms from Dr Thilo Rehren (University College London) and two anonymous reviewers, to who we express our gratitude. References Baumann, M., 1919. Ancient tin mines of the Transvaal. J. Chem. Metall. Min. Soc. S. Af. 19, 120–132. Carbo´ No´ver, J., Williamson, J., 1967. The crystallisation and decomposition of SnOSiO2 glass. Phys. Chem. Glasses 8 (4), 164–168. Chirikure, S., Hall, S., Miller, D., 2007. One hundred years on: what do we know about tin and bronze production in southern Africa? In: La Niece, S., Hook, D., Craddock, P. (Eds.), Metals and Mines Studies in Archaeometallurgy. British Museum Press, London, pp. 112–119. Crocker, I.T., Eales, H.V., Ehlers, D.L., 2001. The fluorite, cassiterite and sulphide deposits associated with the acid rocks of the Bushveld Complex. Counc. Geosci. Mem. 90, Pretoria. Denbow, J., Miller, D.E., 2007. Metal working at Bosutswe, Botswana. J. Afr. Archaeology 5, 280–313. Dietzel, A., 1941. Strukturchemie des Glases. Die Naturwissenschaften 29, 537–547. Farthing, D.J., 2002. The Mineralogy of Tin Slags. Ph.D. dissertation, Johns Hopkins University, Baltimore. Friede, H.M., Steel, R.H., 1976. Tin mining and smelting in the Transvaal during the iron age. J. S. Afr. Inst. Min. Metall. 76, 461–470. Grant, M.R., 1994. Iron in ancient tin from Rooiberg, South Africa. J. Archaeol. Sci. 21, 455–460. Grant, M.R., 1999. The sourcing of southern African tin artifacts. J. Archaeol. Sci. 26, 1111–1117. Grant, M.R., Huffman, T.N., Watterson, J.I.W., 1994. The role of copper smelting in the pre-colonial exploitation of the Rooiberg tin field. S. Afr. J. Sci. 90, 85–90. Hall, S.L., 1981. Iron Age Sequence and Settlement in the Rooiberg-Thabazimbi Area. MA thesis, University of the Witwatersrand. Hall, S.L., Grant, M.R., 1995. Indigenous ceramic production in the context of the colonial frontier in the Transvaal, South Africa. In: Vincenzini, P. (Ed.), Proc. 8th CIMTEC: The Ceramics Cultural Heritage. Techna Srl, Faenza, pp. 465–473. Heimann, R.B., 1977. High-temperature and high-pressure modifications of silica. Miner. Sci. Eng. 9, 57–63. Holland, D., Howes, A.P., Dupree, R., Johnson, J.A., Johnson, C.E., 2003. Site symmetry in binary and ternary tin silicate glasses – 29Si and 119Sn nuclear magnetic resonance. J. Phys. Condens. Matter 15, S2457–S2472. Ishikawa, T., Akagi, S., 1978. The structure of glasses in the system SnO-SiO2. Phys. Chem. Glasses 19 (5), 108–114. Johnson, J.A., Johnson, C.E., 2005. Mo¨ssbauer spectroscopy as a probe of silicate glasses. J. Phys. Condens. Matter 17 (8), R381–R412. Karim, M.M., Holland, D., 1995. Physical properties of glasses in the system SnOSiO2. Phys. Chem. Glasses 36 (5), 206–210. Killick, D.J., 1991. A tin lerale from the Soutpansberg, northern Transvaal, South Africa. S. Afr. Archaeol. Bull. 46, 137–141. Killick, D.J., 2009a. Agency, dependency and long-distance trade: east Africa and the Islamic World, ca. 700–1500 C.E. In: Falconer, S., Redman, C.A. (Eds.), Polities

S. Chirikure et al. / Journal of Archaeological Science 37 (2010) 1656–1669 and Power: Archaeological Perspectives on the Landscapes of Early States. University of Arizona Press, Tucson, pp. 179–207. Killick, D.J., 2009b. Cairo to Cape: the spread of metallurgy through eastern and southern Africa. J. World Prehistory 22, 399–414. Kingery, W.D., Bowen, H.K., Uhlmann, D.R., 1976. Introduction to Ceramics, second ed. John Wiley & Sons, New York. Krohn, M.H., Hellmann, J.R., Mahieu, B., Pantano, C.G., 2005. Effect of tin oxide on the physical properties of soda-lime-silica glass. J. Non-Cryst. Solids 351 (6–7), 455–465. Labuschagne, L.S., 2004. Evolution of the ore –forming fluids in the Rooiberg tin field, South Africa. Counc. Geosci. Mem. 96, Pretoria. Leube, A., Stumpfl, E.F., 1963a. The Rooiberg and Leeuwpoort tin mines, Transvaal, South Africa, part I: general and structural geology. Econ. Geol. 58, 391–418. Leube, A., Stumpfl, E.F.,1963b. The Rooiberg and Leeuwpoort tin mines, Transvaal, South Africa. Part II. Petrology, mineralogy and geochemistry. Econ. Geol. 58, 527–557. Malham, A., Aylett, J., Higgs, E., McDonnell, J.G., 2002. Tin smelting slags from Crift Farm, Cornwall, and the effect of changing technology on slag composition. Hist. Metall. 36, 84–94. Malham, A., McDonnell, J.G., 2003. The characterization of medieval and postmedieval tin smelting slags. In: Proc. Intern. Conf. ‘Archaeometallurgy in Europe’, Milan, Italy, Sept 24–26, 2003, vol. 1, pp. 565–573. Mao, O., Dahn, J.R., 1999. Mechanically alloyed Sn–Fe(-C) powders as anode materials for Li-ion batteries. II. The Sn–Fe system. J. Electrochem. Soc. 146 (2), 414–422. Mason, R.B., 1986. Origins of Black People of Johannesburg and the Southwestern Central Transvaal, AD 350–1880. University of the Witwatersrand Archaeological Research Unit, Johannesburg. McCaffery, R.S., Lorig, C.H., Goff, I.N., Oesterle, J.F., Fritsche, O.O., 1932. Determination of viscosity of iron blast-furnace slags. Trans. Am. Inst. Mining Metallurg. Engng 100, 88–121. McDonald, D.P., 1912. The occurrence of sideroplessite and ankerite in the cassiterite lodes at Rooiberg. Trans. Geol. Soc. S. Africa 15, 107–112. Miller, D.E., 2002. Smelter and smith: metal fabrication technology in the Southern African early and late iron age. J. Archaeol. Sci. 29, 1083–1131. Miller, D.E., Hall, S.L., 2008. Rooiberg revisited – the analysis of tin and copper smelting debris. Hist. Metall. 42, 23–38.

1669

Miller, D.E., Killick, D.J., 2004. Slag identification at southern African archaeological sites. J. Afr. Archaeology 2, 23–48. Molofsky, L.J., Chesley, J.T., Killick, D.J., Ruiz, J. A novel approach to lead isotope provenance studies of tin and bronze. Archaeometry, in press. Rankin, G.A., Wright, F.E., 1915. The ternary system CaO–Al2O3–SiO2. Am. J. Sci. 39, 1–79. Recknagel, R., 1908. On some mineral deposits in the Rooiberg District. Trans. Geol. Soc. S. Africa 11, 83–106. Rozendaal, A., Kellaway, W.F., 1988. The Geology of Rooiberg Tin. Pamphlet issued by Geology Department. Rooiberg Tin Ltd. Rozendaal, A., Misiewicz, J.E., Scheepers, R., 1995. The tin zone: sediment-hosted hydrothermal tin mineralization at Rooiberg, South Africa. Mineralium Deposita 30, 178–187. Schweitzer, J.K., Hatton, C.J., de Waal, S.A., 1995. Regional lithochemical stratigraphy of the Rooiberg group, upper Transvaal Super Group: proposed new subdivision. S. Afr. J. Geol. 98, 245–255. Strunz, H., Nickel, E.H., 2001. Strunz Mineralogical Tables, ninth ed. Schweitzerbart, Stuttgart. ¨ ber einige Beobachtungen an den Zinnlagersta¨tten von RooiStumpfl, E.F., 1960. U berg, Transvaal. Neues Jahrbuch Mineral Abhand 94, 162–180. Festband Ramdohr. Taylor, R.G., 1979. Geology of Tin Deposits. Elsevier, Amsterdam. Trevor, T.G., 1912. Some observations on ancient mine working in the Transvaal. J. Chem. Metall. Min. Soc. S. Af. 12, 267–275. Tylecote, R.F., Photos, E., Earl, B., 1989. The composition of tin slags from the southwest of England. World Archaeol. 20, 434–445. von Knorring, O., Condliffe, E., 1987. Mineralized pegmatites in Africa. Geol. J. 22, 253–270. Wagner, P.A., Gordon, H.S., 1929. Further notes of ancient bronze smelters in the Waterberg district, Transvaal. S. Afr. J. Sci. 26, 563–574. White, H., Oxley Oxland, G.St.J, 1974. Ancient metallurgy practices in the Rooiberg area. S. Afr. Inst. Min. Metall. 74, 269–270. Wright, P.A., 1982. Extractive Metallurgy of Tin, second ed. Elsevier, Amsterdam. Yener, K.A., Vandiver, P.B., 1993. Tin processing at Go¨ltepe, an early bronze age site in Anatolia. Am. J. Arch. 97, 207–238.