At the 1928 meeting of the British Association for the Advancement of Science in Glasgow, the ‘Sumerian Committee’ of the BAAS made its first report, including the analysis of 34 bronze objects (Desch, Reference Desch and Copper1929). The Committee, constituted of leading British archaeologists and metallurgists, was established to ‘report on the probable source of the supply of copper used by the Sumerians’. Thus was born one of the first interdisciplinary projects carrying out the chemical analysis of archaeological copper alloy objects, with the express aim of provenance. Nearly 100 years later, it is perhaps time to reflect on such activity.

This Element seeks to chart the development, degrees of success, and suggests a possible re-focussing for one of the major activities in scientific archaeology – the use of chemical and isotopic measurements on archaeological artefacts to determine the origin of the raw materials used to make these objects, commonly referred to as provenance studies. The focus here is on inorganic materials, particularly copper alloys, ceramics, and lithic materials, since several hundred thousand analyses have been published on these categories; numerically, studies on other materials generally pale into insignificance. Organic materials, particularly amber, have been historically important, and others, such as textiles, the production and trade of which have been key economic activities, are under-represented in the provenance literature because of poor survival and also the need for more specialized analytical techniques such as proteomics and light stable isotope ratios.

Inorganic provenance studies were widely adopted from the 1960s onwards because in principle they can elucidate trade and exchange patterns in the ancient world, and, therefore, contribute to studies of contact between societies, either in terms of trade in materials or other forms of social transfer of goods. Perhaps even more significantly, it can provide proxy evidence for the exchange of ideas. The rise of provenance was facilitated by two parallel developments – the increasing availability of instrumental methods of chemical analysis, and changing theoretical concepts of the role of material culture within archaeology. The growing scepticism in some quarters towards provenance studies from the 1980s onwards was partly the consequence of a gradually increasing recognition of the complexity of the production processes for all but the simplest of artefacts, including a growing appreciation of the potential for recycling in some materials, particularly glass and metals. However, rather than signalling the end of materials analysis as a tool for provenance studies, these potentially confounding features can present interesting new challenges and unexpected opportunities for the modern archaeologist. In fact, they transform the concept of provenance from the apparently simple question of ‘where does this object come from?’ to the much more interesting one of ‘how did humans manage and use the raw materials at their disposal to produce these artefacts?’ Given that the ultimate aim of archaeology is to understand past human societies, this seems to provide a very fruitful and important avenue for future research.

1 The Provenance Hypothesis

Provenance in this context means identification of the source of the raw materials used to make archaeological objects. For ceramics, this corresponds to the source of the clay used, and perhaps the temper added. For copper alloys, it can be interpreted as the mine from which the copper ores are extracted, but it might involve multiple mines if alloying metals (tin, zinc) are added. Archaeologically, the term can be extrapolated from the direct identification of source to include the matching of a set of artefacts (the unknown group) with another set (the control group), the implication being that they come from the same place, without necessarily identifying the specific geological source(s). The former exercise can be considered as provenance-to-source, and the second provenance-to-match. This definition is very different to that used in art history, where provenance means the lifehistory (biography) of the artefact, ideally documenting the sequence of all owners (and hence locations) of a particular work of art since its creation. Some authors, particularly in the USA (e.g., Price and Burton, Reference Price, Burton, Price and Burton2011: 213), have promoted the use of the term provenience to define the ‘birthplace’ of the object, and provenance to signify the ‘resume’ (biography) of the object. Although this is an important distinction, and embraces the art historical definition, the majority of archaeologists simply use the term provenance to cover both of these definitions, perhaps taking the view that ‘birthplace’ is part of ‘biography’.

It is important to emphasize that ‘provenance-to-match’ has a much longer history in archaeology than scientific provenance studies. Similarities in material culture rapidly became one of the key markers for defining cultural groupings, and particular forms of ceramics, such as Roman transport amphorae, or red-gloss Samian ware, arranged into intricate typologies, have been key indicators of trade and exchange across the empire from Spain to India. These parallels are deduced from visual study of form – careful classification of shape, manufacturing details, and decorative features – often supplemented by visual examination of fabric – the colour and texture of the ceramic paste. Thus, the framework for studying provenance was already in place when Weigand et al. (Reference Weigand, Harbottle, Sayre, Earle and Ericson1977) observed that ‘in many instances there will exist differences in chemical composition between pottery from different sources that will exceed, in some recognizable way, the differences observed within pottery from a given source’. They termed this the ‘provenience postulate’, and suggested that it was the basis of all studies involving provenance attribution using chemical analysis. In 2001, Wilson and Pollard (Reference Wilson, Pollard, Brothwell and Pollard2001: 507–508) attempted to clarify and systematize the assumptions behind the scientific provenance of archaeological materials by setting out six criteria for the ‘provenance hypothesis’:

1. i) The prime requirement is that some chemical (or isotopic) characteristic of the geological raw material(s) is carried through (unchanged, or predictably relatable) into the finished object.

2. ii) That this ‘fingerprint’ varies between potential geological sources available in the past, and that this variation can be related to the geographical (as opposed to perhaps a broad depositional environment) occurrences of the raw material. Inter-source variation must be greater than intra-source variation for successful source discrimination.

3. iii) That such characteristic ‘fingerprints’ can be measured with sufficient precision in the finished artefacts to enable discrimination between competing potential sources.

4. iv) That no ‘mixing’ of raw materials occurs (either before or during processing, or as a result of re-cycling of material), or that any such mixing can be adequately accounted for.

5. v) That post-depositional processes either have no effect on the characteristic fingerprint or that such alteration can either be detected (and the altered elements or sample be discounted) or that some satisfactory allowance can be made.

6. vi) That any observed patterns of trade or exchange of finished materials are interpretable in terms of human behaviour. This presupposes that the outcome of a scientific provenance study can be interfaced with an existing appropriate socio-economic model, so that such results do not exist in vacuo.

The first requirement reflects the idea of the ‘fingerprint’ – a characteristic element, set of elements, or isotopic composition which passes through from the source material to the object, ideally with no change. The possibility of a quantifiable change to the fingerprint through the various steps of the processing (depending on material) has to be acknowledged, but presents significant challenges in reality – it would require considerable supporting evidence to conclude that the difference observed in the fingerprint between source and product is due to processing (e.g., volatilization of certain elements at high temperatures) rather than signifying something else. The crux of the hypothesis is captured in points (ii) and (iii), particularly if the aim is provenance to source. Different geological sources (of ore, clay or rock, and so on) can only be distinguished if the fingerprint varies between alternative sources, if the internal variation in the fingerprint is less than that between sources, and the measurement technology is capable of measuring these differences. It also relies on the assumption that sources are geographically discrete. It might be of limited use archaeologically if a geological source of clay consists of a large chemically homogeneous river valley or flood plain (e.g., the Nile, or the Indus), rather than specifically located clay deposits (although it might be the case that what appears to be a chemically homogeneous deposit using one set of indicators (e.g., major elements) might show significant trends when a different set (e.g., trace elements) is used). Point (v) requires that what is measured as a fingerprint is (ideally) unaffected by post-depositional factors such as selective corrosion or contamination from groundwater.

Point (iv) – no mixing or recycling – is perhaps the issue that has dominated theoretical discussions of provenance. It is self-evident that, at least for provenance-to-source, any mixing of material from sources with different fingerprints will make it more difficult to assign an object to a specific source. Depending on the number of potential sources involved, and the magnitude of diversion in the measured fingerprint, it could simply result in less confidence in the assignment of object to source, or it could give rise to the creation of an entirely ‘fictitious source’ – the mixture resulting in data which, when plotted on an appropriate graph, appear to form a coherent source group, but which do not actually correspond to any real source. This is a case where provenance-to-match has a distinct advantage – if the characteristics of the unknown objects match those in the control material, then a common source can be proposed, even if it is itself unknown. The issue of mixing and recycling is discussed further next.

The final point was effectively a plea to interpret the results of any provenance study in terms of real human behaviour, rather than relying on abstract arrows on maps showing the movement of objects, apparently without the aid of human intervention. Trivially, this might involve thinking about how objects can move – as trade, gifts, tribute, war booty, and so on – and also about the mode of transport – maritime, riverine, or land. However, we must avoid the trap of assuming that human activities in the past carried the same meaning as they do today. Uniformitarianism is not a reliable guide in archaeology. Trade and exchange do not necessarily reflect purely commercial activities or ‘market forces’. Marcel Mauss (1872–1950) showed how in many societies gift exchange was at the heart of creating and maintaining relationships both within and between social groupings (Mauss, Reference Mauss1923–1924). Karl Polyani (1886–1964) proposed that there were three modes of exchange: reciprocity, redistribution, and market exchange (Polyani, Reference Polyani1944), although not necessarily mutually exclusive. Gift exchange is a form of reciprocity; redistribution implies some centralized control of distribution, often via a central depot, and market exchange involves a specific central location, but not necessarily a financial transaction. The particular mode of exchange is clearly linked with considerations of scale and organization – trading is very different in the context of a centralized imperial economy such as China or Rome, compared with differently regulated trading between independent tribal groups, or between traders working across the borders of settled sedentary populations and nomadic pastoralists. Renfrew and Bahn (Reference Renfrew and Bahn2020: 371) have combined these social and economic considerations with various forms of settlement organization to produce a series of models for the exchange of physical commodities, some of which are redrawn in Figure 1. These range from direct contact between A and B to intermediate markets (A to market, market to B) to down-the-line trade (A to B to C to D, and so on). The form of exchange has an important influence on the distribution of particular objects, and also on how we should interpret provenance – as discussed in Section 7.1, in down-the-line trade, time taken to travel could be a factor in understanding the significance of finding objects from A at site D.

Figure 1 Some simple models of trade and exchange,

redrawn from Renfrew and Bahn (Reference Renfrew and Bahn2020: 371)

2 The Origins of Chemical Analysis in Archaeology

All of the examples discussed in this Element are predicated on the chemical analysis of inorganic objects. This application of the art of chemical analysis to archaeological artefacts has a long prehistory. The traditional methods of assay for gold – by separating the gold from silver by fire, or using the touchstone – have been known since at least the 2nd millennium BCE (Pollard, Reference Pollard2016). It is undoubtedly the case that miners and metalworkers were able to assay ores and precious metals long before the advent of analytical chemistry, and, indeed, the need for such assay provided an impetus for the development of chemistry (Greenaway, Reference Greenaway1962, Reference Greenaway1964). Cuneiform tablets from Mesopotamia (from the early 2nd millennium BC) describe the quantitative assay of gold by fire (Levey, Reference Levey1959). For example, one 1st millennium BCE text says ‘2 minas 2 shekels of gold were put into the furnace, 10 1/2 shekels of them were lost in the furnace, 1 mina 51 1/2 shekels of dark gold came out of the furnace’ (Levey, Reference Levey1959: 33). Since 60 shekels = 1 mina, we can calculate that the original purity of this gold was around 91.4%. The Medieval Arab scholar Geber (Abu Mūsā Jābir ibn Hayyān, c. 721–c. 815 CE) shows a knowledge of the purification of gold, referring to gold ‘sustaining the Tryal of the cupel, and cement’ (Holmyard, Reference Holmyard1928: 63), and his works were translated into English by Richard Russel in 1678. From at least the late 13th century CE, the Royal Mint (established within the Tower of London around 1279 CE) has routinely assayed the fineness of English gold and silver coinage (Watson, Reference Watson1962). The gold coinage of Henry III was certified as ‘fine’ (i.e., 24 carat, or 100% gold) in 1257 CE, and Edward III established a gold currency at 99.48% purity in 1343. This declined over the next two centuries, until the ‘Great Debasement’ of Henry VIII when it declined to 83.33%, to be reinstated to 99.45% by Edward VI in 1550. Such evidence comes from the results of an ancient ceremony known as the ‘Trial of the Pyx’ carried out at the Mint since the late 13th century, in which the Miles Argentarius (Assay Master) certifies the fineness of the coinage, the method of which is ‘differing but little from the modern fire-assay of silver’ (Watson, Reference Watson1962: 6). Hence, the assay of precious metal precedes modern chemical analysis by many centuries.

Many surviving medieval European texts give increasingly clear descriptions of the process of precious metal purification, and also for assaying base metal ores. Theophilus’ On Divers Arts, written c. 1110–1140 CE, probably by the Benedictine monk Roger of Helmarshausen in Hesse, central Germany (Hawthorne and Smith, Reference Hawthorne and Smith1963), describes in Book III (Chapter 23) how to refine silver in a porous ash-lined cupel, using added lead to promote the oxidation of impurities. He also describes (Chapter 33) how to cement gold in order to purify it using a process which involves creating a ‘sandwich’ of thin sheets of gold alternating with layers of a mixture of recycled ground ceramics or burnt clay (two-thirds) with common salt (one-third) moistened with urine, heated in a fire-tested ‘casserole dish’. Heating the casserole for 24 hours causes the salt to remove impurities, and, after a number of repetitions of the process, pure gold is obtained. The fineness of the original gold can be established by comparing the weight of the refined gold with that of the original. Further chapters (69 and 70) describe how to separate gold from copper in gilded scrap metal by cupellation in a bone-ash crucible, and how to separate gold from scrap gilded silver by heating with sulfur, which allows both gold and silver to be recovered.

The first European book to give a very clear description of assaying is the Probierbüchlein, possibly first published in Germany in 1518, although Annaliese Sisco and Cyril Stanley Smith (Reference Sisco and Smith1949), the translators of this text into English, believe that the first edition is that of 1534 printed in Maydeburg. It was produced in numerous editions through the 16th and 17th centuries, and contained clear practical instructions for the purification of gold and silver by cementation, but also procedures for dissolving metals and ores in mineral acids for parting or assay, much as is still done today. It was the main source of such information in Europe until Lazarus Ercker’s Beschreibung allerfürnemisten mineralischen Ertzt und Berckwercksarten (1574) (Sisco and Smith, Reference Sisco and Smith1951), which provided the first widely available European textbook for miners on assaying ores. This was extensively translated across Europe, including an English version published by Sir John Pettus (Reference Pettus1683) as Fleta Minor, or, the Laws of Art and Nature in Knowing, Judging, Assaying, Fining, Refining and Inlarging the Bodies of Confined Metals. According to Pettus’ translation, Ercker’s introduction says:

Clearly, the arts of assay were well-known long before analytical chemistry, and also that these methods all involved trial by fire – essentially a smaller-scale version of the processes required to reduce metals from their ores, or separate gold from silver, and therefore requiring access to metallurgical furnaces and facilities. It was not until the late 18th century CE that a different method emerged in Europe, ultimately giving rise to quantitative gravimetric analytical chemistry. This consisted of precipitating a known compound of a particular element out of a solution created by dissolving the sample in a suitable solvent. By employing a sequence of specific precipitations, a series of different elements can be quantified from the same solution. By weighing the amount of sample dissolved, and weighing the dried precipitate(s), the proportion of the precipitated element in the sample can be calculated, providing allowance can be made for the form in which the element is precipitated – for example, if tin (Sn) is precipitated as tin oxide (SnO₂), it would require correction by a factor of (119/(119 + 2 × 16)), or 0.79, to reflect the proportion of oxygen in the compound. This could not have been calculated in this way until the atomic weights of the elements had been established, which began with John Dalton’s ‘New System of Chemical Philosophy’ (Dalton, Reference Dalton1808; 1827). Before that, an empirical observation would have been made (by fire) to calculate the proportion of metallic tin in the oxide precipitate.

Thus, at least in the chemical laboratory, trial by fire gradually gave way to gravimetric analysis, originally known as the ‘humid method’. The first systematic exposition was that of Torbern Bergman (1735–1784) at the University of Uppsala, Sweden, who published a protocol for the aqueous gravimetric analysis of gemstones entitled ‘Disquisitio chemica de terra gemmarum’ (Bergman, Reference Bergman1777). The big advance here was the use of an alkali fusion to bring the gemstone into solution, but Bergman’s precipitation protocol was not very rigorous, and was subsequently improved by Martin Heinrich Klaproth (1743–1817) in Berlin (Klaproth, Reference Klaproth1792/3) and Nicolas-Louis Vauquelin (1763–1829) in Paris (Vauquelin, Reference Vauquelin1799). These three analytical protocols have been re-published and compared by Oldroyd (Reference Oldroyd1973).

The earliest report of the quantitative gravimetric chemical analysis of a metal appears to be that of Gustav von Engeström (1738–1813), who published a paper in 1776 on the composition of the imported white copper alloy paktong from China, which he found to contain approximately 29% nickel (with some cobalt). A year previously, von Engeström (Reference von Engeström1775) had published an analysis of imported zinc oxide from China, thus showing that he was engaged in understanding the nature of these imports – ‘industrial espionage’ to allow Europe to compete with Chinese technology. Bergmann himself published ‘Dissertatio Chemica de Analysi Ferri’ in 1781. Given these dates for the early chemical analyses of metals in the late 1770s, it is remarkable to note that the earliest published chemical analyses of archaeological metal artefacts can be traced to 1777 by Johann Christian Wiegleb (1732–1800), read to the Kurmainzische Akademie Nutzlicher Wissenschaften, Mainz, on 2 April 1777 (Wiegleb, Reference Wiegleb1777; Pollard, Reference Pollard2018). He used nitric acid to dissolve the metal, but only measured tin, and assumed that the rest was copper. A few years later, Michel Jean Jérome Dizé (1764–1852) published the tin content of five Roman, one Greek, and two Gallic copper alloy coins (Dizé, Reference Dizé1790). Klaproth, in his better-known publication dated to 1792/3 (but actually read to the Royal Academy of Sciences and Belles-Lettres of Berlin on 9 July 1795, and published in 1798), reported the chemical analyses of six Greek and nine Roman coins, measuring directly copper, lead, and tin, and, in some samples, iron and silver – thus providing the earliest realistic analyses of ancient coins, justifiably earning him the title of the first archaeological chemist (Caley, Reference Caley1949; Pollard, Reference Pollard2016), in addition to his renown as a mineral chemist.

3 The First Expressions of Provenance

Within a few decades of these first chemical analyses of archaeological objects in the late 18th century, the idea soon began to emerge that the chemical characteristics of exotic objects – those deemed not to be local to the point of recovery – might give some indication of where they had come from. The first such explicit observation was that of Karl Christian Traugott Friedeman Göbel (1794–1851), Professor of Chemistry and Pharmacy at Dorpat, Estonia, from 1828. His volume (Göbel, Reference Göbel1842) entitled ‘Ueber den Einfluss der Chemie auf die Ermittelung der Völker der Vorzeit oder Resultate der chemischen Untersuchung metallischer Alterthümer insbesondere der in den Ostseegouvernements vorkommenden, Behufs der Ermittelung der Völker, van welchen sie abstammen’ was a landmark publication in many ways. The title (‘About the influence of chemistry on the determination of the peoples of the past, or results of the chemical analysis of metallic antiquities’) for the first time explicitly linked the chemistry of metallic antiquities to the ‘determination of the peoples of the past’. It was also the largest compilation of archaeological bronze analyses up to that date (119), of which 55 were his own (recording copper, tin, zinc and lead) and the remainder were from other analysts, including 32 re-published from Klaproth. In his introduction, he articulated a set of three culturally orientated questions (pp. 2–3):

1. 1) To what extent can it be demonstrated historically at what time certain peoples either inhabited the areas where such antiquities are found or entered them on war or trading expeditions?

2. 2) What historical traditions do we have that this or that people preferably made or used certain metals and metal compositions of a certain form and chemical composition and for what purposes?

3. 3) If it can be proven that different peoples made metal compositions of similar or the same chemical composition, which people made them earlier? at what time? and what is the chemical constitution of these ancient metal masses?

Critically, he noticed that Greek metal contained tin but no zinc, but Roman contained either tin or zinc. He used this criterion to identify the source and period of the metal finds in Northern European graves. Although we can now see that this is an oversimplification, it marks a major step in our social interpretation of the composition of ancient metals.

His first question is effectively the first published articulation of the concept of archaeological provenance, albeit articulated within the hyper-diffusionist model of cultural transfer prevalent at the time (see Section 4). This idea crystallized further in the mid 19th century into the theory of what was to become provenance studies. The clearest statement of this comes not from metal studies but from stone, in a pair of papers published by Augustin-Alexis Damour (1808–1902). Damour (Reference Damour1865: 313) stated that:

This is essentially the definition of the provenance hypothesis – if an object can be confirmed to be of distant origin, then the only conclusion is that either the object or the raw material must have been transported to the region. Until the early 20th century, however, chemical analyses were generally carried out by gravimetry (wet chemistry) as already described, which required large amounts of sample and considerable skill to produce adequate analyses. Consequently, studies involving a large number of samples were limited – the largest compilation until the early 20th century of copper alloy analyses was that of von Bibra (Reference von Bibra1869), which reported 1,249 metal analyses from 91 different analysts, of which 602 were his own. By this time gravimetric analysis was capable of measuring at least 11 elements in copper alloys (Cu, Sn, Zn, Pb, Ag, Fe, Sb, As, Ni, Co, S), at levels down to between 0.01% and 0.1%, depending on the element, although subsequent work has suggested that some of von Bibra’s trace element data (especially Ni) are unreliable (Caley, Reference Caley1939: 82).

With the advent of instrumental means of chemical analysis from the early 20th century onwards, and even more intensely following the Second World War, large numbers of archaeological objects could be systematically analysed, and several large-scale projects began, initially on copper alloys, but subsequently on ceramics and lithic materials. The earliest form of instrumental analysis was termed optical emission analysis, which was based on experiments involving the emission of light resulting from heating or sparking gases and solids carried out by Kirchoff and Bunsen (Reference Kirchoff and Bunsen1860). The first spectrometers were used in the iron industry for the detection of trace elements by 1880, but the earliest paper on the analysis of archaeological metal was not published until 1921 (Baudoin, Reference Baudouin1921). The replacement of gravimetry by spectroscopy was by no means instantaneous. Earle Radcliffe Caley (1900–1988), one of the leading archaeometallurgists of the 20th century, and also an outstanding historian of metallurgy (translating many of the Medieval and later texts already discussed), clearly preferred gravimetry to the instrumental methods available at the time, even into the 1960s. In his compilation of the analyses of ancient metals (Caley, Reference Caley1964), he devotes 13 pages to a detailed description of gravimetric methods for the analysis of copper alloys and corrosion products (pp. 81–93), compared to 5 on emission spectrography (pp. 93–97). The reasons are not hard to understand, even though instrumental methods of chemical analysis provided a step-change in the scale of the application of analytical chemistry, in terms of the speed of analysis, sample throughput, and also in the range of trace elements measurable. A key advantage of gravimetry is that each precipitation, when carried out correctly, gives an independent estimate of the quantity of a particular element in the sample. If all elements present in the sample (above, say, 0.01%) are measured, then the sum (the analytical total) of all the estimates should be very close to 100%. The quality of the analysis can therefore be verified by the proximity of the total to 100%. Caley, in his analysis of some Chinese bronzes, labelled the averaged totals of duplicate analyses combining gravimetric and spectrographic analyses for such alloys as ‘satisfactory’ if the total was greater than 99.85%, ‘less satisfactory’ for totals around 99.65%, and ‘much less so’ at 99.3% (Caley et al., Reference Caley, Chang and Woods1979: 187). In principle, instrumental methods are capable of similar levels of accuracy, and some modern analyses do approach these values, but often it is not possible to recover a truly independent analytical total because of internal corrections and iterations within the calibration software. The combined protocol of gravimetry for major elements and spectrometry for trace elements employed by Caley in his later work was in fact very common in the major programmes of chemical analysis for archaeological materials carried out in Germany, the Soviet Union, and China up until the 1980s or even later. This involved measuring the major elements gravimetrically (in copper alloys, this included Cu, Sn, Pb, Zn, and often Fe), but quantifying the trace elements (usually including As, Sb, Ni, Ag, and sometimes Bi, Co, and Au) by optical emission spectrometry. It is, of course, the case that trace elements below 0.01% in the sample cannot be measured by gravimetric methods without taking very large samples, and are more easily quantifiable by modern sensitive instrumental methods of analysis.

4 The Archaeological Framework

The period from around the 1950s saw a blossoming of many scientific techniques in archaeology ranging from botany and geophysics to zoology, partly stimulated by the advent and impact of radiocarbon dating from the middle of the century. This gave rise to a ‘Golden Age’ of archaeological materials analysis, in which the developing field of materials science was applied across a range of archaeological materials in order to answer a number of basic questions around the physical nature of objects, and particularly to understand the technological processes involved in their production. A large part of this ‘Golden Age’ was also devoted to the provenance of materials, particularly after the development of mass spectrometry for isotopes of lead in the 1970s (sometimes termed ‘isotope archaeology’, to parallel isotope geology). Although the intellectual roots of provenance can be traced back to the mid 19th century, and a number of large-scale provenance studies using optical emission started in the 1930s, it is probably true to say that these initiatives would not have been as influential had there not been a change in theoretical approaches to archaeological interpretation over the same period.

There are multiple origins across the world for what is now archaeology. When Leonard Woolley (1880–1960) excavated a large number of Mesopotamian artefacts laid out side-by-side, accompanied by cylindrical stone scrolls describing each object in three languages, in his excavations at Ur (modern Iraq) in 1925, he had discovered what has become known as the ‘World’s First Museum’ (Hopkins, Reference Hopkins2021: 43). These objects, covering 1,500 years of Mesopotamian history, had been accumulated and displayed by the Neo-Babylonian Princess Ennigaldi-Nanna, High Priestess of Ur, 547–c. 530 BCE. Across the ancient world, objects from earlier occupants of the dynastic throne were highly valued in order to lend credibility to the later occupants. Perhaps the best-documented example of this is to be found in China, where there is a long tradition of connoisseurship focussed on the objects of the past, especially those with inscriptions. The most obvious manifestation of attempts to inherit credibility is the rise of the Song Antiquarian Movement during the second half of the 11th century CE of the Song dynasty (960–1279 CE), stimulated both by scholarly curiosity but also by the desire of the Imperial Court to recreate the rituals of former dynasties, and hence reinforce perceptions of cultural continuity and heavenly authority (Sena, Reference Sena2019). The famous Song scholar-official Ouyang Xiu (欧阳修; 1007–1072 CE) collected inscriptions from bronzes, stone stelae, and rock carvings, including purchasing bronzes from excavations or antique shops. His collection, numbering a thousand ink rubbings of inscriptions, covered the time period from the 10th century BCE King Mu of the Western Zhou dynasty through to the Five Dynasties immediately preceding the Song, and was published in his Records of Collecting Antiquity (歐陽文忠公文集) in 1062 CE. The Song dynasty also saw the publication of the oldest surviving illustrated catalogues of antiquities, including the Illustrated Investigations of Antiquity (考古圖; 1092 CE) by Lü Dalin (呂大臨; c. 1042-c. 1090).

In Europe, Classical Archaeology (the archaeology of the literate Mediterranean world) can be traced back to the revival of interest in the art of the classical world in late Renaissance Italy (15th century CE) and the subsequent rise of the ‘Grand Tour’ amongst the elites of western Europe. By the 18th century, no noble house was complete without a cabinet of curiosities, consisting of specimens of geological, archaeological, and mythological origin, in part collected during tours of southern Europe, and, following Napoleon’s campaign of 1798–1801, Egypt. A separate branch of European archaeology, leading to prehistoric archaeology, began as an antiquarian pastime for the wealthy, with many burial mounds and other structures, often on their own estates, being opened to recover the ‘treasures’ therein. As in China, a more structured approach to the buried past was given in England by the need of the Tudor State to emphasize its connections with the mythological past, signified by the appointment of John Leland (1503–1552) as King’s Antiquary in 1533 (Trigger, Reference Trigger1989). This paved the way for the publication in 1586 of Britannia by William Camden (1551–1623), providing a survey of what was then known about Roman England, John Aubrey’s (1626–1697) Monumenta Britannica, written between 1665 and 1693 and originally existing as two manuscripts (Bodleian MS Top. Gen. c. 24–5) but not published until 1980 (Legg and Fowles, Reference Legg and Fowles1980), and the well-known work of William Stukeley (1687–1765) at Stonehenge and elsewhere, including Stonehenge. A Temple Restor’d to the British Druids, published in 1740.

This period of archaeology is generally known as the antiquarian stage. With increasing professionalization (or, before that, the emergence of a number of privately sponsored scholars, such as Giovanni Belzoni (1778–1823), Sir Austin Henry Layard (1817–1894), Heinrich Schliemann (1822–1890), Sir Marc Aurel Stein (1862–1943), and Howard Carter (1874–1939)), a newer philosophical framework emerged, termed culture-historical archaeology. This used material assemblages to define distinct cultures, often with ethnic and nationalistic overtones. Drawing heavily on emerging theories of Darwinian evolution and its anthropological interpretations, this began to see the many prehistoric and modern ‘primitive societies’ as stages on the inevitable progression from ‘barbarism’ to ‘civilization’. Crucially, the dominant view was that such societies were incapable of advancing themselves from internal forces, and that the primary mechanism of technological change was the diffusion of ideas from outside. This culminated in the so-called hyperdiffusionist philosophies, which postulated that the major advances in human society (e.g., farming, tool use, metallurgy, but also cultural practices such as embalming) emerged once only (often attributed to ancient Egypt), and diffused throughout the world. The most obvious proponent of such ideas was the Australian anatomist and Egyptologist Grafton Elliot Smith (1871–1937), who published a number of influential volumes, including The Ancient Egyptians and the Origin of Civilization (Reference Smith and Sir1911), The Migrations of Early Culture (Reference Smith and Sir1915), and The Diffusion of Culture (Reference Smith and Sir1933). He argued, for example, that the practices of artificial mummification began in Egypt and spread across the world, ultimately influencing the Inca in south America (Elliot-Smith, Reference Smith and Sir1916).

This idea of cultural stasis, only changed by the arrival of people and ideas from elsewhere, was heavily rooted in the self-perception of the colonial powers during the 18th and 19th centuries, who believed that colonization was a mechanism for the improvement of the colonized. These views have since been refuted by the multiple freedom movements of the 20th century, reinvigorating the study of many indigenous cultures, from the Amazon to the subcontinent of India. Associated with this was the rise of processual (or ‘New’) archaeology in the 1960s, which recognized that isolated cultures were capable of internal evolutionary processes (Trigger, Reference Trigger1989). Much greater emphasis was placed on understanding the ecological, technological, economic, and social forces experienced by ancient societies. This was in part enabled by the increasing adoption within archaeology of scientific methodologies developed elsewhere, including palaeoenvironmental techniques (pollen, plant macrofossils, and zooarchaeological analysis) and dating techniques, starting with the ‘radiocarbon revolution’ from the 1950s onwards. Material analysis also became part of this New Archaeology, enabling the detailed study of technological processes from a physical and chemical examination of the artefacts themselves, but also from the characterization of production debris, residual manufacturing structures, and also the raw materials (see, for example, several chapters in Pollard et al., Reference Pollard, Armitage and Makarewicz2023a).

Thus, the mid 19th century idea of provenance found a fertile environment within the New Archaeology. Although originating within a strong cultural diffusionist framework, where the identification of intrusive material objects was taken as indications of the presence of external influences, the theory rapidly morphed into a new model, although with little overt self-reflection (exceptions include Weigand et al. (Reference Weigand, Harbottle, Sayre, Earle and Ericson1977) in the context of turquoise, Hunt (Reference Hunt2012) for ceramics and Pernicka (Reference Pernicka, Roberts and Thornton2014) for metals). The emphasis on economic factors as being part of the evolutionary pressure on society meant that the identification of trade (interpreted in its widest sense) became an important part of archaeological research. Fortuitously, this combined with the increased availability of instrumental analytical methods to allow the initiation of several massive programmes on the chemical analysis of archaeological artefacts.

5 Provenance in Practice

There is no standard methodology for carrying out a successful provenance study, but there are some broad general principles. The first requirement is the generation of a meaningful starting hypothesis. For example, the question ‘where do these objects come from’ is strictly speaking unanswerable, since it implies a comprehensive knowledge of all the possible sources available at the time of manufacture, which is unlikely for a number of reasons, including the sparseness of such data, and the possibility that viable ancient sources remain unknown. In contrast, ‘do these objects (X) come from A, B or C’ is a more straightforward question, provided sufficient and representative (control) samples are available from sources A, B, or C. The answer, however, should take the form ‘given the data available for sources A, B or C, source A (or whichever) is the most likely source for X, providing that there is no similar but unknown source’. It might be possible using multivariate statistics or kernel density modelling to assign a probability to the proposal that X comes from A, B, or C, but it should be recognized that these probabilities are conditional on the variables measured and the samples analysed. Similar constraints apply to provenance-to-match questions – ‘do these two groups of objects have the same characteristics’ is in principle answerable, but it is important to recognize that, statistically speaking, differences can be established to a required degree of probability, whereas similarity has to be inferred. This means that if a match between X and A is determined to have a low probability (i.e., below a specified critical value), then it can safely be assumed that X and A are different. If, however, the similarity between X and A is accepted at a certain degree of probability (usually 95% confidence), it cannot be assumed that X and A are the same – only that, according to the variables measured and the number of samples analysed, X is consistent with A. It is then up to the person interpreting the data to decide whether, taking all possible sources of evidence into consideration, it is reasonable to accept the hypothesis that they are the same. Such additional evidence might include technological (e.g., are they made in the same way?), archaeological (e.g., is the distribution of objects of type X the same as that of type A?), or art historical (e.g., do A and X use identical decorative motifs and techniques?).

Once the hypothesis to be tested has been established, it is then necessary to decide on the mode of analysis, and the interpretational technique(s) to be used. Clearly, the choice of analytical technique dictates the range of elements or isotopes that can be determined, and hence defines the fingerprint available. The earlier analytical techniques (gravimetry, optical emission (OES), atomic absorption (AAS)) have been completely replaced by a range of newer methods, including inductively coupled plasma methods (ICP-AES or ICP-MS), which in general offer a wider range of measurable elements, with higher sensitivities (i.e., lower minimum detectable levels) allowing much lower levels of certain elements to be measured, combined with faster sample throughput and lower levels of sampling damage. Neutron activation analysis (NAA), which was the benchmark technique for many years for the measurement of trace elements, has now virtually disappeared because of the lack of access to nuclear reactors, but has also largely been replaced by ICP techniques. Other methods available include X-ray fluorescence (XRF) and electron microscopy (SEM, electron microprobe, and so on). Each method has its own strengths and weaknesses, which must be evaluated in the context of the research question being addressed, not least of which is to ask whether a particular technique is capable of measuring the elements (or isotopes) most likely to provide a fingerprint with sufficient sensitivity and precision. For a more detailed description of these techniques as they are used in archaeology, see Pollard et al. (Reference Pollard, Batt, Stern and Young2007).

Likewise, there is no simple rule to define the ‘best’ approach for any particular material. It is generally accepted that, for manufactured materials (e.g., metal, glass, and ceramic), it is better to focus on minor (0.1–1%), trace (0.001–0.1%), or ultra-trace elements (<0.001%), since the major elements (>1%)Footnote ¹ tend to be deliberately controlled for physical or aesthetic reasons. Within the trace and ultra-trace elements, the rare earth elements (REE) provide a special set of elements consisting of the lanthanides (La, Ce, Pr, Nd, Pm, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, and Lu), often supplemented by Y and Sc, which have virtually indistinguishable chemical properties, but provide a graded series of atomic numbers (from 57 to 71) and hence gradually increasing ionic radius (size). Because of this chemical similarity but size differential, the series is often systematically fractionated by particular geochemical processes – sometimes to enhance the lighter elements (LREE), and sometimes the heavier (HREE) – and thus are often used in geochemistry to elucidate processes (Henderson, Reference Henderson1984). They therefore also provide a promising starting point for archaeological provenance studies, especially for ceramics and glass. They are, however, rarely used as measured, since there is a strong odd-even (atomic number) effect, meaning that a plot of the REE’s in, say, a clay would show a sawtooth profile. To counter this, it is usual to normalize the REE profile using the abundance of the REE’s in a specified chondrite (stony meteorite, representing primeval composition). Providing all data have been normalized to the same chondrite, REE profiles can be directly compared. Another powerful method in geochemistry using selected trace elements is the practice of plotting Nb/Ta against Zr/Hf, particularly in granitic rocks, but also in ore deposits (Dostal and Chatterjee, Reference Dostal and Chatterjee2000). Zirconium (Zr) and hafnium (Hf) are d-block transition metals with the same outer electron configuration but in successive periods – hence, they have very similar chemical properties but different atomic weights (40 and 72, respectively), and therefore the ratio of their abundances, just as with the REE series, provide a sensitive process indicator. The same is true of niobium (Nb) and tantalum (Ta). Plotting the two ratios together can be a very effective way of studying the evolution of the continental crust. Little use so far has been made of these particular ratios in archaeology, but similar elemental ratios have been used effectively for the discrimination between sands of different sources, used to manufacture glass (Degryse, Reference Degryse2014: 71). Perhaps the most influential pair of ratios to date has been 1000Zr/Ti plotted against Cr/La (Shortland et al., Reference Shortland, Rogers and Eremin2007: fig. 6; Walton et al., Reference Walton, Shortland, Kirk and Degryse2009), which has been shown to discriminate between early Mesopotamian and Egyptian glass – an important tool in the discussion about the origin of glass. Titanium (Ti) and Zr form a pair similar to Zr and Hf; chromium (Cr) and lanthanum (La) do not form such a pair (both are d-block transition elements, but not from adjacent periods, and having different valence), but Cr is generally associated with iron minerals, and La with clay minerals. As such, this pair of elemental ratios has also proved useful in archaeological studies of glass.

Most, but not all, elements exist in more than one isotopic form. An isotope is an atom which has the same atomic number (number of protons in the nucleus) but a different number of neutrons, giving different atomic weights. Since atomic number dictates the chemistry of the element, isotopes have identical chemical properties, but different weights, which gives rise to variation in those properties which depend on kinetics. For example, carbon (C) is defined as that element which has six protons in the nucleus, but it has three isotopes: carbon-12 (¹²C), with six protons and six neutrons; carbon-13 (¹³C), with six protons and seven neutrons; and the radioactive carbon-14 (¹⁴C), with six protons and eight neutrons. The most common of these isotopes is ¹²C, which constitutes 99% of all C, followed by ¹³C, which is around 1%, and the radioactively unstable isotope is very rare, consisting of about one atom in a million million (1 in 10¹²) of ¹²C.

Following the first experimental confirmation of the existence of isotopes (in the gas neon) by Francis William Aston (1877–1945) in Cambridge, UK (Aston, Reference Aston1920a), between 1920 and 1925 Aston produced a spectacular series of seven papers on the mass spectrometry of an increasing number of elements. In 1920, the number of elements whose precise atomic weight (and hence number of stable isotopes) was known was 11 (Aston, Reference Aston1920b). By 1925, 56 elements were listed (over half of the 80 non-radioactive elements known), including many metals (Fe, Co, Ni, Cu, Zn, Ag, Sn, Sb, Ba, Hg), but not lead (Aston, Reference Aston1925). The four stable isotopes of lead (²⁰⁴Pb, ²⁰⁶Pb, ²⁰⁷Pb, and ²⁰⁸Pb) were first measured by Nier in 1938. This was particularly important for isotope geochemistry, since three of these lead isotopes (²⁰⁶Pb, ²⁰⁷Pb, and ²⁰⁸Pb) are end members of the uranium (²³⁸U and ²³⁵U) and thorium (²³²Th) radioactive decay chains. Hence the natural variation in lead isotopes is much greater than for other isotopes, and also measuring isotope ratios within these decay chains provides some of the fundamental isotope geochronometers in the earth sciences (e.g., Faure, Reference Faure1986).

The idea of using variation in lead isotope ratios for provenancing archaeological objects was presented by Robert H. Brill (1929–2021) and Jesse Marion Wampler at a meeting held in Boston, USA, entitled Application of Science in the Examination of Works of Art, in September 1965, published in 1967 (Brill and Wampler, Reference Brill, Wampler and Young1967a). Before this appeared in print, Grögler et al. (Reference Grögler, Geiss, Grünefelder and Houtermans1966) published the lead isotopic analysis of ten Roman lead pipes and ingots from across Europe, comparing these data with ores from the UK, Germany, Austria, Italy, Jugoslavia, Portugal, Spain, and Greece, thus staking the claim to be the first published lead isotope study in archaeology. Two archaeological samples from Portugal, for example, found matches – one with ores from the southern Algarve and the other with Rio Tinto. These and similar matches were sufficient for the authors to observe that archaeological lead objects could be matched with potential ore sources. The larger and better-known publication of lead isotopes on archaeological materials is that of Brill and Wampler (Reference Brill and Wampler1967b), who reported carrying out 230 measurements, of which 70 were on lead ores (from Greece (Laurion), England, and Spain) and 160 on lead from archaeological samples, with the explicit purpose of establishing the provenance of the artefacts. The catalogue of the samples reported in this particular paper only lists 36 archaeological metals, all of which are lead objects, apart from one (no. 188), which was lead extracted from a Chinese early Zhou dynasty bronze vessel, the results of which were found to be ‘entirely different from the leads found in any of the other archaeological samples studied’ (Brill and Wampler, Reference Brill and Wampler1967b: 76). The actual measurements are not given in the paper. A further eight results are discussed, consisting of seven from samples of glass, and one from the glaze of a Roman vessel from Caerleon, Wales (UK).

Although these papers established beyond doubt the potential for lead isotope analysis on archaeological artefacts, the principal focus was on lead objects, which are relatively rare from archaeological contexts. Sample 188 (the Zhou dynasty bronze) in Brill and Wampler (Reference Brill and Wampler1967b) showed that sufficient lead could be extracted from bronze for isotope measurement, but that sample contained 11.6% lead, which is much higher than in most European Bronze Age bronzes. It was not until the late 1970s that extraction of traces of lead from copper (and silver) alloys was routinely established, allowing the method to become a major contributor to provenance studies of non-ferrous metals, especially in the Mediterranean (Gale and Stos-Gale, Reference Gale and Stos-Gale1982). Isotopic data (most commonly lead, but more recently a wider range of elements such as tin, copper, antimony, strontium, boron, and so on) now provide an extra dimension (in that the isotopic values are generally independent of the abundance of the element), which can be used either in combination with trace elements or alone.

With the increased capacity for generating chemical (and subsequently isotopic) data from archaeological samples, interpretation of the subsequent data became an increasingly pressing problem. In the early days of instrumental chemical analysis, ingenious ways were developed to compare large sets of data. In ceramics, for example, the datasets provided by OES or AAS typically contained abundances of 10 or 12 elements (expressed conventionally as oxides), typically including SiO₂, TiO₂, Al₂O₃, CaO, MgO, Na₂O, K₂O, FeO, and MnO₂, plus a few trace elements (e.g., Cr, Ni, and so on). One obvious approach is to display the data as a set of bi-plots – CaO versus MgO, Na₂O versus K₂O, and so on. This can be extremely effective if the analyst hits upon the ‘right’ set of plots; but with nine elements measured, there are 36 possible combinations of bi-plot, which would have been laborious to plot at the time. In an attempt to consider all elements at the same time, a system was devised in which they were presented as a set of logarithm-scale columns, one for each element (oxide), sometimes multiplied by an arbitrary number to bring the data into range. Thus, Figure 2 (from Catling and Jones, Reference Catling and Jones1977) shows a comparison of ceramic data from Chania (Crete) with the so-called Theban Stirrup Jars (TSJ). The shaded area for each column represents the 80% confidence level for each element, with the horizontal bar representing the mean (average) concentration.

Figure 2 Comparison of two groups of pottery analyses, one from Chania (Crete: left-hand set of columns) and another representing Theban Stirrup Jars (TSJ’s: right-hand set). The coloured area for each oxide represents the concentration range associated with an 80% level of confidence, and the white bar within each area is the mean concentration. Note that the vertical axis is logarithmic, and that some oxides are multiplied by 10 or 100 for ease of presentation. Discrimination is made by visual pairwise comparison – in this case, most pairs overlap, but Mn is higher in TSJ and both Cr and Na are higher in Chania III.

Redrawn from fig. 1f in Catling and Jones (Reference Catling and Jones1977).

Such plots (and variations on the same theme) seem cumbersome now, but did have a number of advantages in a pre-computer era. A visual match can be made by comparing equivalent columns for each dataset, and a match would be accepted if the majority of columns appeared to cover the same range – this was elegantly expressed in this case as ‘TSJ compositions can be collectively accommodated within the Chania characteristics.’ Alternative presentations, in which the set of elements are horizontally linked by a band joining the averages and upper and lower limits, which could then be superimposed from different assemblages, allowed such comparisons to be made more easily. The method could also be extended to other sets of data, such as the composition of copper alloys, or glasses (e.g., Figure 3). There is naturally a degree of subjectivity in deciding when the columns or ribbons ‘match’, but it does allow for the range of variation in each set of data, rather than simply relying on comparing averages. On the other hand, because of correlations between elements, it probably overemphasizes the matching (or non-matching) between the sets of data, since it treats each element as fully independent.

Figure 3 Comparison of five oxides in five important types of Old World glass, redrawn from Sayre and Smith (Reference Sayre and Smith1961; fig. 1). For each glass type (2nd millennium BC, Antimony rich), the five horizontal lines link the mean composition for each glass type, enabling similarities and differences to be easily seen. For Islamic lead only, the lozenge indicates the mean value for that element, and the vertical extent of the lozenge shows the standard deviation for that element (the standard deviations are omitted from the other types for clarity).

Such visual attempts to match these datasets were gradually replaced during the 1970s by an avowedly multivariate approach based on treating each measured element (or oxide) as a variable in a multivariate space defined by these variables – essentially treating each analysed sample as a point in multivariate space, with coordinates equal to the value of the concentration of each element. This approach was applied particularly following the rise of NAA, which was capable of measuring many more elements simultaneously than OES or AAS, thus requiring a more sophisticated approach to data analysis. It was also, of course, enabled by the availability of computers capable of dealing with the calculations required. Multivariate techniques, such as hierarchical cluster analysis (CA) and principal components analysis (PCA) (Hodson, Reference Hodson1969) and Mahalanobis distance-based metrics (Bieber et al., Reference Bieber, Brooks, Harbottle and Sayre1976) began to be applied to archaeometric data (Glascock and MacDonald, Reference Glascock, Macdonald, Pollard, Armitage and Makarewicz2023). Such methods have been incorrectly described as multivariate statistics, which in general they are not. Mainly they provide methods for dimensional reduction, which allow high-dimensional data to be reduced to fewer dimensions, so that the data can be presented in two dimensions (i.e., can be printed) with minimal distortion of the relational information. In some versions (e.g., discriminant function analysis, DFA, and more recently Bayesian and kernel density methods: Pollard et al., Reference Pollard, Ma, Bidegaray, Liu, Pollard, Armitage and Makarewicz2023b), probabilities of group membership can be generated, but these are only valid in the context of the data presented. Although more rigorous statistical tests can be applied to evaluate the discrimination between multivariate groups, such as Hotellings-T² test – the multivariate extension of Student’s t-test – they are rarely, if ever, used in archaeology.

Figure 4, from Hodson (Reference Hodson1969), shows a redrawn dendrogram resulting from the clustering of 100 analyses of European bronzes published as part of the SAM project (see below: 90 published by Junghans et al. (Reference Junghans, Sangmeister and Schröder1960) and 10 by Schubert and Schubert (Reference Schubert, Schubert and Mozsolics1967)). The data consisted of 11 elements (Sn, Pb, As, Sb, Ag, Ni, Bi, Au, Zn, Co, and Fe), and the distance between the samples was measured by squared-mean Euclidean distance (SMED). Clustering was by average-link cluster analysis (ALCA). The dendrogram is plotted with the 100 samples numbered from left to right across the top, with increasing dissimilarity (decreasing similarity) plotted downwards. Hodson identified 16 clusters (as numbered across the top), and subsequently plotted the samples belonging to each group on a map of Europe. He concluded that the method had ‘performed surprisingly well’. It is important to note, however, that the choice of distance metric and clustering algorithm can drastically affect the appearance of the dendrogram (Pollard, Reference Pollard, Aspinall and Warren1983), which is itself a 2D representation of (in this case) 11-dimensional data, and will therefore be distorted to some degree.

Figure 4 Simplified dendrogram showing the agglomerative clustering of 100 analysed European bronzes, mostly SAM data (see text). Modified from Hodson (Reference Hodson1969). Sample numbers are listed horizontally, below the 16 groups identified by Hodson. Similarity between samples decreases down the diagram – horizontal ties mark the dissimilarity between adjacent samples or groups. For clarity, the linkages within the larger groups have been deleted. The dark horizontal line shows the level of dissimilarity identified as significant by Hodson. This is often arbitrary, and a small number of samples are ungrouped at this level (samples 39, 21, and 22).

Isotopic data, initially the three ratios recorded in lead isotope analysis, present a different set of problems, but came with an already-standardized presentation format, derived from isotope geochronology. This consisted of producing a pair of plots – in the early days of archaeological applications (Brill and Wampler, Reference Brill and Wampler1967b), these were ²⁰⁶Pb/²⁰⁴Pb versus ²⁰⁶Pb/²⁰⁷Pb, and ²⁰⁸Pb/²⁰⁷Pb versus ²⁰⁶Pb/²⁰⁷Pb. Because the ²⁰⁶Pb/²⁰⁷Pb axis is common to both figures, Brill and Wampler plotted these two one above the other (redrawn in Figure 5, to better separate the two plots). In this particular figure, archaeological objects of lead from Sardinia, England, and Greece are plotted against specific ore samples, and estimated ore fields defined by modern mineral samples from ancient mining regions. The ore fields are marked as ellipses labelled L (Laurion, Greece), S (Spain), and E (England), although it was observed at the time that, because of the low resolution of the measurements and the low number of samples, these attributions are uncertain since group L, for example, also contains ores from Iran. The interpretation of lead isotope data stimulated a great deal of debate during the 1980s and 1990s, and subsequently alternative modes of presentation have been proposed to bring out the archaeological rather that the geological context of the samples, such as plotting the inverse concentration of lead against one of the lead isotopic ratios (1/Pb versus ²⁰⁶Pb/²⁰⁴Pb) (Pollard and Bray, Reference Pollard and Bray2015).

Figure 5 Early plot of lead isotope data on archaeological lead samples, compared to various ores fields. Redrawn from Brill and Wampler (Reference Brill and Wampler1967b; ill. 2). Lower diagram plots ²⁰⁸Pb/²⁰⁷Pb versus ²⁰⁶Pb/²⁰⁷Pb for 13 archaeological lead samples (from Sardinia, England, and Greece) plus 13 lead ore samples, from Sardinia, England, and Laurion, Greece. Upper diagram plots ²⁰⁶Pb/²⁰⁴Pb versus ²⁰⁶Pb/²⁰⁷Pb for same samples. Unlabelled ellipses indicate estimated measurement errors at the time.

6 The ‘Golden Age’ of Provenance Studies

This section is intended to briefly outline some of the major provenance projects undertaken over the last 100 years, without pretending to be either comprehensive or conclusive. It takes a deliberately historical approach, rather than attempting to summarize the latest work, since this allows the development of the intellectual underpinnings to be examined. The approach is by material, rather than by archaeological culture, geographical region or time period. This is because, although the archaeological context of the question is all-important, the practical constraints are largely dictated by the nature of the material being studied, and it is therefore preferable to think about the theory of provenance studies from the perspective of each material.

It has long been realized that different materials pose different challenges in terms of the complexity of any proposed provenance study. In principle, natural lithic materials present the least problems in this context – most lithic materials require little or no extractive processing other than digging or cutting out, individual objects cannot be made from mixed (different) sources without the fact being obvious, and lithics offer limited opportunities for recycling into new objects, apart from the process of reduction, by which a tool is reshaped or re-sharpened. None of these processes are likely to alter the chemical or isotopic composition of the object, thereby reducing considerably the potential constraints on provenance. This is clearly not the case with manufactured materials – principally ceramics, metals, faience, and glass. The simple fact that the raw materials have to be extracted, processed, and manufactured at high temperatures suggests that there may be some issues with the relative stability of the ‘fingerprints’ from source to artefact. Beyond this, there is the obvious potential for raw materials from different sources to be mixed, and for some finished materials (particularly metals and glass) to be recycled. A fuller discussion of the impact of these factors on the theory of provenance is deferred to the succeeding sections.

6.1 Lithics

Interest in the provenance of the stones of megalithic monuments has a very long history, and probably represents one of the earliest scientific analyses in archaeology. Stukeley (Reference Stukeley1740: 5) reports the microscopic examination of fragments of stone from Stonehenge carried out by himself at the Royal Society, London, on samples collected by Halley in 1720. He concludes that the stones came from ‘the gray weathers, upon Marlborough downs’, but the altar and ‘pyramidals’ are much harder, and from elsewhere. Thus began 250 years of research into the geological origin of the ‘bluestones’ at Stonehenge – now agreed to be the Preseli Hills in west Wales, some 180 miles (290 km) from Stonehenge (Parker Pearson et al., Reference Parker Pearson, Pollard and Richards2019). This, of course, then gave rise to the question of how such large stones were transported over these distances in the Neolithic.

Obviously, worked stone tools have been a major indicator of human activity since the beginning of the human species. Since not all types of stone are equally suitable for chipping and shaping, it is inevitable that particular sources began to be favoured for tool production. Thus, identifying the geological source of the stone gives invaluable information about human behaviour, although in such cases it is quite likely that visual or petrological identification might be sufficient to achieve this aim.

6.1.1 Flint

This is less true in more homogeneous fine-grained rock such as flint. The first study of prehistoric flint mines in England and northern France was carried out by Sieveking et al. (Reference 75Sieveking, Craddock, Hughes, Bush and Ferguson1970) using AAS to measure Al, Mg, K, and Fe in flints from six mine sites, followed by principal components analysis, suggesting that discrimination was possible. The same samples were subsequently re-analysed (along with other samples) by Aspinall and Feather (Reference Aspinall and Feather1972) using NAA to measure 15 trace elements (Na, Cs, Sc, Ta, Cr, Fe, Co, La, Ce, Sm, Eu, Tb, Yb, Th, and U). There was considerable scatter in the data, but it was confirmed that Continental and British sources could be clearly distinguished, based on elements such as Cr and Th. Much of the early work on archaeological flint was summarized at a conference in 1983, published by Sieveking and Hart (Reference Sieveking and Hart1986).

6.1.2 Obsidian

Although strictly a natural volcanic glass rather than a rock, obsidian has been an important medium for the manufacture of bladed tools, traded over considerable distances, and has consequently received considerable attention in terms of provenance. Obsidian tools were also of interest because of the phenomenon known as hydration, in which the alkali elements in the surface are exchanged for hydrogen ions from the water in the surrounding burial environment. The resulting hydration layer (an alkali-depleted hydrated silica gel) can be seen under the optical microscope in cross section, and it has been argued that the thickness of the layer is indicative of the time elapsed since initial exposure. Although it is not necessarily a linear relationship (since ion exchange is a chemical process, and is hence controlled by environmental parameters), obsidian hydration dating is still used in some parts of the world, particularly where organic preservation is poor (Liritzis and Laskaris, Reference Liritzis and Laskaris2009).

In terms of provenance, the initial assumption was that each obsidian flow (corresponding to the later stages of a rhyolitic volcanic eruption) should exhibit a unique and homogeneous pattern of trace elements because of the evolution of magma composition during eruption, and hence obsidian flaked from a particular flow should have a unique chemical fingerprint (e.g., Wright, Reference Wright1968). This combination of geological specificity and long-distance trading makes obsidian a highly attractive target for chemical and isotopic study. Because obsidian is a natural glass, its major element composition (SiO₂, CaO, K₂O, Na₂O, and so on) is constrained by the chemistry of the magma and the physical requirements of the glass transition conditions (largely viscosity), so any variation between flows is expected to be most easily seen in the trace elements. Although early work in the Mediterranean utilized OES methods to measure 16 trace elements, the most significant of which were deemed to be Ba and Zr (Renfrew et al., Reference Renfrew, Cann and Dixon1965: see Figure 6), the analytical focus elsewhere rapidly fell on NAA (e.g., Frison et al. (Reference Frison, Wright, Griffin and Gordus1968) in the northwestern plains of the USA, and Gordus et al. (Reference Gordus, Wright and Griffin1968) from a wide range of geological sources across North America), measuring a range of elements, including Mn, Sc, La, Rb, Sm, Ba, Zr, Na, and Fe. The archaeological significance of these studies is considerable. Gordus et al. (Reference Gordus, Wright and Griffin1968) point out that obsidian was common on Hopewell sites in the Illinois valley (c. 200 BCE–300 CE), but the nearest sources are in Mexico, New Mexico, Yellowstone National Park, and on the Pacific Coast, the closest of which is 1,500 miles (2,400 km) away. Location to source therefore illustrates the existence of some form of long-distance trade networks. Perhaps even more strikingly, in the Mediterranean obsidian was exploited from the early Neolithic onwards (c. 8000 BCE). Moreover, many of the major sources are found on islands (e.g., Sardinia, Lipari, Palmarola, Pantelleria, and Melos), indicating not only long-distance trade in the Neolithic, but also some competence in boat-building and sailing (Dixon et al., Reference Dixon, Cann and Renfrew1968).

Figure 6 Plot of log concentration of barium versus zirconium as measured by OES from various obsidian sources in the Near East, showing good separation of the major sources. Plotted from data presented in Table 2 of Renfrew et al. 1966.

Groupings are those proposed by Renfrew et al.

Like many areas of archaeological science, as more work is done and more data published, the original assumptions can be seen as oversimplifications. This is less the case in obsidian studies than in some other areas, but there is still some ambiguity between potential sources. Gale (Reference Gale1981), for example, pointed out that trace element analysis was not capable of completely resolving the known obsidian sources around the Mediterranean (the islands, plus Anatolian, Armenian, Hungarian, and Slovakian sources), and proposed using the strontium isotope ratio (⁸⁷Sr/⁸⁶Sr) to discriminate between them, which was one of the first uses of this particular isotope system within archaeology. This ratio varies geologically because ⁸⁶Sr is stable, but ⁸⁷Sr is a daughter of ⁸⁷Rb, so the ⁸⁷Sr/⁸⁶Sr ratio varies in geological materials depending on the initial ⁸⁷Rb/⁸⁶Sr ratio, and the geological age of the deposit. Given that obsidian flows tend to be short-term events, but occurring over a wide range of geological periods, it is likely that different flows, even in the same location, will have different strontium isotope ratios because of magmatic evolution. For the Mediterranean sources of obsidian, a plot of ⁸⁷Sr/⁸⁶Sr versus Rb gave very good separation, and allowed archaeological samples to be assigned to specific sources. Gale also noted, however, that a plot of Sr versus Rb gave ‘completely sufficient’ resolution of these sources, and compared this with the rather complicated ‘simple discrimination diagrams’ produced by Renfrew et al. (Reference Renfrew, Cann and Dixon1965), in which [Li+(Ca/100)+(Mg/10)] was plotted against [(Zr/2)+Nb+Pb+(Fe/100)]. Although these linear combinations seem somewhat arbitrary now, it must be remembered that, prior to principal components analysis and discriminant function analysis, it was quite common to empirically produce such combinations in an attempt to maximize the visual discrimination between source groups.

Despite these complexities, particularly around the Mediterranean, the method of trace element analysis has become so widely accepted that suitably calibrated portable XRF spectrometers can now be used in the field to provenance huge numbers of obsidian artefacts in a very short time (Tykot, Reference Tykot2021).

6.1.3 Marble

Another important lithic material for provenance studies has been marble, used throughout the classical world and elsewhere for public and private architecture, and for statuary. Marble is a metamorphic form of limestone, and has restricted occurrences. Some famous ancient marble quarries, such as Carrara in Tuscany (Italy) or Paros in the Aegean, have continued in use into modern times. The history of provenance studies of classical marble has followed a familiar trajectory, which could be generalized to the history of such studies for most materials, at least in Europe:

• Inferences on provenance made from classical and biblical written sources up to the late 18th century

• Visual examination, leading to microscopy and petrography, starting in the 19th century

• Chemical analysis, starting with gravimetry in the 19th century but leading to spectrography in the mid 20th century

• Trace element analysis by NAA from the mid 20th century, replaced by ICP-MS at the end of the 20th century

• Application of isotope measurements in the later 20th century.

The application of isotope methods came relatively early for archaeological marble. Because marbles are carbonates, and because there had been considerable geological interest in the light stable isotope ratios of carbon (¹³C/¹²C) and oxygen (¹⁸O/¹⁶O) in marine carbonates, significant success was achieved from a relatively early date using these isotopes in archaeological material (Craig and Craig, Reference Craig and Craig1972). In this study, modern samples of marbles from the ancient quarries of Naxos, Paros, Mount Hymettus, and Mount Pentelikon were separable using these measurements, and archaeological samples from a number of extant monuments were attributed to certain of these sources (see Figure 7).

Figure 7 Stable isotope analyses ( $\delta$ ¹³C versus $\delta$ ¹⁸O) of modern samples from four ancient sources of marble (circles: Naxos, Penteli, Paros, and Hymettus), with eight archaeological marble samples (squares). The two archaeological samples from Athens are associated with the group for Penteli; one of the two from Delphi is assigned to Paros, and both samples from Naxos are close to one of the two groups identified as Naxos. The other three are unattributed.

Based on data presented in fig. 1 and table 1 of Craig and Craig (Reference Craig and Craig1972).

For the light stable isotopes (e.g., H, C, O, N, S), the data are conventionally reported as δ values, in which the specified ratio in a sample is converted to a value relative to the same ratio in an agreed international standard material, in units of parts per thousand (referred to as ‘per mil’, ‰). For carbon, the ratio is ¹³C/¹²C, and for oxygen it is ¹⁸O/¹⁶O, and the standard material for these isotopes in inorganic carbonates is PDB (Pee-Dee Belemnite), a fossil bivalve from the Cretaceous Peedee formation in South Carolina, USA. The definition is:

$\delta (‰)=\left[\frac{R-R^{*}}{R^{*}}\right]\times 1,000,$

where R is the ratio in the sample and R* the ratio in the standard. Figure 7 is a plot of δ¹³C versus δ¹⁸O for five of the major marble sources around the Mediterranean, showing very good discrimination.

7 Cracks in the Façade

Concerns about the veracity of scientific approaches to the provenance of archaeological materials have come essentially from two directions: one archaeological and one theoretical. Even from the inception of large-scale approaches to provenance in the mid 20th century, some archaeologists have expressed the view that the results of such studies were either ambiguous and unhelpful, or, in some cases, simply wrong. Such views, examples of which have already been given, were usually based on the lack of consistency between ‘scientific’ approaches and other sources of archaeological evidence, and prompted fierce debates about the role of science in archaeology (e.g., Dunnell, Reference Dunnell1993). These criticisms were often well-founded, based on aspects of the methodologies employed, such as the adequacy of the sampling of archaeological objects and prospective sources of raw materials, but also on the interpretative frameworks employed. As the size of the datasets increased (both in terms of numbers of samples, but also in the range of characteristics measured), ever more complex mathematical approaches have been applied. Although theoretically justifiable (except it must be remembered that most multivariate approaches are empirically rather than statistically based), the increasing abstraction of the results simply contributed to the growing unease about the validity of these approaches.

The second strand of concern arose from theoretical issues about the nature of the processing of the materials being studied, and the relative lack of consideration of the articulation between material culture and human behaviour (Pollard and Gosden, Reference Pollard and Gosden2023). The fundamental concept of scientific provenance – that the object carries within it some ‘fingerprint’ of the raw materials from whence it came – has been elaborated upon but not fundamentally reconsidered since the mid 19th century. It has, however, become increasingly obvious to many researchers that the production and circulation systems of many classes of object are potentially far more complicated than this simple postulate would suggest. Such factors include the following:

1. i) Theoretical consideration of time in the transport of objects

2. ii) The consequences of raw materials processing

3. iii) The effects of high temperature processing

4. iv) The effects of mixing and recycling

8 Towards a New Provenance Hypothesis

For copper alloys, the potential for, if not the inevitably of, recycling has prompted a theoretical rethink about the nature of archaeological copper objects. In a model referred to as ‘Form and Flow’ (Bray et al., Reference Bray, Cuénod and Gosden2015), an attempt was made to conceptually separate the material of the object from its form. The form of the object is an instantiation at a particular point in time – it might be an axe or a cauldron at one particular time, but then, along with other scrap and fresh metal, could be recycled into a bell. The form has changed, but there is some continuity in the life of the material. Under this model, the concept of provenance changes from one in which there is an assumed one-to-one relationship between object and source, to one in which the characteristics of the metal flow (chemical and isotopic) are affected by a number of factors, one of which is the source of the metal, but other influences can include the admixture of metal from other sources, or the reintroduction of metal into the flow by recycling. This can be seen as a flowing river, having a source consisting of one or more ore sources, but with different materials being introduced along flow. By monitoring the changes in composition over time, the biography of the metal flow can be defined. In the simplest case, where the metal has a single source and there are no further additions, this reduces to the conventional provenance postulate, but in general it is expected to be more complicated than that.

Although this conceptualization might have relevance for copper circulation, and probably by extension to gold and silver, and possibly even for glass, it is difficult to articulate a parallel model for other materials. Here we fall back on a long-term observation – despite all of the obstacles to successful provenance studies articulated in Section 6, there are many situations in which the results are thought to be ‘correct’. Does this mean that the objections are irrelevant, perhaps peripheral at best, or is there some other factor at work? The key difference is that archaeology is not geology. Geological factors are, of course, significant – clay has particular properties in particular locations, and ditto metal ores, glassmaking sands, and so on. But these properties are then manipulated by human intervention – clays are processed and mixed, ores are selected and blended, and sands are selected and purified. For example, the large-scale Roman terra sigillata factories in Gaul (modern France) have been shown to have produced consistent body compositions over many decades (e.g., at Lezoux: Argyropoulos, Reference Argyropoulos1995). This suggests careful selection and blending of clays and other resources to provide a body with consistent composition having the required physical properties – stiff enough to be moulded, sufficiently calcareous to fire to a glossy red surface, and so on. On an even larger scale, the many billions of pieces of porcelain produced at Jingdezhen, Jiangxi Province, China, for over 1,000 years have been shown to have consistent chemical compositions – there are variations according to quality, vessel size and shape, and some evolution over time (Pollard and Wood, Reference Pollard, Wood, Olin and Blackman1986), but the over-riding message is that the people responsible for making the potting clay from the raw materials at Jingdezhen (for this was a separate profession) were exerting considerable control over the selection of the raw materials before the clay was passed to the potters. A similar story emerged in Yunnan in the 19th century with respect to the selection of copper ores for smelting, as revealed by Wu Qujin.

From a provenance perspective, therefore, the obvious differences between archaeology and geology are the interventions made by humans in the steps between the extraction of the raw materials and the production of the finished products, as well as those affecting the life and discard of the objects. In large-scale operations, we are dealing not just with the interventions of a few individuals, but with the sequential activities of a large and complex social organization. In modern terminology, what we are witnessing is the extensive application of ‘quality control’ over the raw materials of antiquity, exercising huge selective pressures on the exploitation of the available ‘resource-scape’.

Where does this leave scientific provenance studies in archaeology – the endeavour embraced so enthusiastically nearly 100 years ago, and which has provided one of the major applications of chemical analysis to archaeology? It is safe to say that for many studies of lithics and obsidian, the existing methodology is and continues to be perfectly satisfactory. The general lack of high temperature processing, and the inability to recycle apart from obvious physical reduction or repurposing, makes it an ideal material. Moving to metals and glass, the situation is less clear. There will undoubtedly be some cases where the traditional model of provenance works well. An example might be the high-status bronze vessels produced for the Royal family during the Shang dynasty in Bronze Age China (c. 1200 BCE). As far as we can tell, these objects did not contain recycled metal, and the raw material supply chain would most likely have been highly controlled, selecting only the finest quality materials (Liu et al., Reference Liu, Pollard and Cao2020). In such cases we might optimistically expect that a definitive statement can be given about the sources of the raw materials (although the exact sources still remain elusive!). More generally, however, we must expect varying degrees of uncertainty. Despite the many suggestions made to date, it is still difficult to determine the degree of recycling in any particular object, or even in assemblages of objects.

In 1983, Ursula Martius Franklin (1921–2016) published a schematic diagram showing the selection and manufacturing processes for three classes of materials (Figure 8). Class I is stone (‘in which human intervention consists of careful choice of naturally available materials’). Her Class II is ‘the naturally occurring constituents – clay, water, and temper – must be selected, prepared, and mixed in proper proportions. From this new composite raw material the object is shaped, … heated to more or less elevated temperatures, producing a new material’. Class III (metals) require a two-part operation, one in which the raw material is extracted from its ore, followed by further processes in which the raw metal is refined, mixed (alloyed, possibly with recycled material), and further heated and then converted (cast or worked) into a finished product. In Figure 8, we have added glass to Class III, since glass is often produced at a primary centre, and then coloured and shaped at a secondary centre (see Section 6.3). The significance here for these three classes of material is that it also conveniently, if somewhat simplistically, reflects their suitability for provenance studies – Class I is very suitable, with a good prospect of provenancing to geological source; Class II is suitable for provenance to match, but less likely to be provenanced to source; Class III is potentially too complex to give provenance to source in many cases, but can give provenance to match, although even this is limited by alloying and recycling.

Figure 8 Schematic classification of materials selection and processing.

Adapted from Franklin (Reference Franklin and Kuwayama1983), with the addition of glass to Class III.

For the more complex (Class II and III) materials, an important line of enquiry lies in the combination of a detailed materials science-based study of the individual steps in the manufacturing process (via raw materials, manufacturing debris, semi-finished products, and so on, if available) combined with targeted provenance studies at each stage. It is clearly easier to study the provenance of actual raw materials recovered from an archaeological excavation of a production site than it is to infer the provenance from the finished or partially processed material; equally, it is probably more fruitful to provenance copper before it is fully refined, or alloyed with other metals. In other words, if the material is available, it is preferable not to attempt to decipher the complex origins of the raw materials from the finished object, but to dissect the process into its component parts. This offers the possibility not only of better understanding the production process but also of unravelling the sources of the different components added at different stages of production.

As always, however, another approach is to change the question. Rather than asking about the provenance of a set of objects, in line with the arguments developed, it might be more fruitful to consider the same data as evidence for the degree of quality assurance exerted on the raw materials. This is perhaps the same as saying that it is easier to resolve the question of ‘provenance-to-match’ than it is to ask the question of ‘provenance-to-source’. Implicitly, it is accepting the notion that quality assurance of raw materials dominates the manufacturing process, and makes it easier to match, for example, unknown terra sigillata with known kiln assemblages than it is to identify the actual geological source of the clay. But it is worth thinking beyond these practicalities. The fact that such ‘provenance-to-match’ studies can be successful indicates that the practices of human control over the selection and processing of raw materials can be studied directly by looking at the variability within the chemical and isotopic data, and also by documenting changes in these factors. Essentially, we are looking at deliberate, repetitive, and consistent human actions manipulating the materials selected from the available ‘resource-scape’, both locally and occasionally over longer distances. As long ago as 2003, Buxeda and Kilikoglou pointed out that measurements of total variation in ceramic datasets estimated from log-ratio transformations provided an important indicator of ‘technologically induced variability, and to the occurrence of alteration and contamination processes’, thus providing a tool for such studies (Buxeda and Kilikoglou, Reference Buxeda i Garrigós, Kilikoglou and van Zelst2003: 197). The theme that the (largely ignored) dispersion of chemical data, as opposed to the simple average values, contained important information was further developed by Michelaki and Hancock (Reference Michelaki and Hancock2011).

The idea that making pots was a direct reflection of human selection within the landscape – itself not a new idea – was elegantly demonstrated by Michelaki et al. (Reference Michelaki, Braun and Hancock2015). They were considering ceramics from two small Neolithic communities in southern Calabria, Italy, and proposed that these ceramics could be regarded as ‘congealed taskscapes’ – physical manifestations of the multiple interactions required to construct a pot between people, materials, and landscapes. From a consideration of the different clay resources in the local landscape (4 km radius), and using a combination of X-ray diffraction, optical microscopy, and instrumental NAA, they reconstructed the sources that the potters used, and therefore where they went in the landscape to get their raw materials. Almost incidentally, this also highlighted where they did not go. This revealed a preference for inland resources rather than coastal regions – possibly predicated on the physical properties of the clays, but also potentially related to other perceptions of the suitability of the landscape.

This, and several other examples, shows that chemical analysis within archaeology can reveal many other aspects of past human behaviour, beyond the simple question of ‘where does the raw material for this object come from’. The principal requirement is a well-formulated question. Given that archaeology is ultimately about reconstructing human behaviour in the past, this suggests that a broader interpretation of provenance studies, incorporating the study of selective human interventions within the resource-scape, and the complexity of material production processes can actually be far more informative than the question first envisaged in the original concept of provenance.