The reconstruction of the first copper-smelting processes in Europe during the 4th and the 3rd millennium BC: where does the oxygen come from?

Appl Phys A (2010) 100: 713–724DOI 10.1007/s00339-010-5651-y

The reconstruction of the first copper-smelting processesin Europe during the 4th and the 3rd millennium BC: where doesthe oxygen come from?

E. Burger · D. Bourgarit · A. Wattiaux · M. Fialin

Received: 30 September 2009 / Accepted: 9 March 2010 / Published online: 10 April 2010© Springer-Verlag 2010

Abstract From the end of Chalcolithic times (end of the4th millennium BC) up to the end of the Bronze Age (1stmillenium BC), copper production increases dramatically inWestern Europe. However, due to the scarcity of technology-related archaeological data, the technological backgroundsustaining the transition to mass production modes remainspoorly understood. The main archaeological clues concern-ing metal production stem from the metallurgical waste,namely copper slags. Those complex materials may be agenuine chemical footprint of the process. In particular, itmay bring new insights on one main issue of the process re-construction: the origin of the oxygen in the system. A newanalytical methodology based on both mass-balance calcu-lation and quantification of Fe3+ contents in copper slags(Mössbauer spectroscopy, electronic microprobe and Syn-chrotron µ-XANES at the Fe-K-edge) has been set up. Thismethodology enables us to distinguish between the solid andgaseous sources of oxygen in a broad range of working con-ditions, thus yielding new features for the understanding ofthe first smelting processes dealing with copper sulphides inWestern Europe 4000 years ago.

E. Burger (�) · D. BourgaritCentre de Recherche et de Restauration des Musées de France(C2RMF), CNRS UMR171, 14 quai François Mitterrand,75001 Paris, Francee-mail: [email protected]: +33-1-69086923

A. WattiauxInstitut de Chimie de la Matière Condensée de Bordeaux,ICMCB, CNRS, Université de Bordeaux 1, 33608 Pessac, France

M. FialinService CAMPARIS-IPGP-CNRS, campus Jussieu, 4 placeJussieu, case courrier 110, 75252 Paris Cedex 05, France

1 Introduction

From the end of Chalcolithic times (end of the 4th mil-lennium BC) up to the end of the Bronze Age (1st mil-lenium BC), copper production increases dramatically inWestern Europe [1, 2]. Whereas previously mainly confinedin the domestic sphere and devoted to ornaments or pres-tige items, metallurgy progressively becomes an industrialactivity where metal tools and weapons replace lithic ones.Yet in Western Europe little archaeological data are availableon the associated copper-production activity, namely coppersmelting.

Copper smelting of sulphide minerals, the most abundantcopper ores, can be modeled through reactions (1) and (2),where “O” represents the total amount of oxygen. Duringreaction at 1150–1250°C, two or three immiscible liquidphases are synthetized: (i) the metallic copper, (ii) an en-riched copper sulphide called matte and (iii) a molten slag(iron oxides and silicates) [3].

CuFeS2 + “O” = (CuxFeySz)matte + (1 − x)Cu

+ (1 − y)FeO + (1 − z)SO2 (1)

FeO + SiO2 = 〈Fe–Si–O〉slag (2)

The copper slags are the main archaeological remainsthat can be found on protohistoric copper-smelting sites.The microstructure of archaeological slags consists of ran-domly distributed submicrometric to millimetric crystal-lites. These crystals are mostly made of iron and coppersulphides (Cu–Fe–S) or oxides (Fe3O4, Cu2O, CuFeO2),silica compounds (quartz, tridymite, cristobalite) and iron-containing silicates (olivines (Fe, Mg)2SiO4, clinopyroxens(Fe, Mg)CaSi2O8) [4]. They are often embedded in a glassy

714 E. Burger et al.

matrix, as in the samples studied here. The microstructureand composition of archaeological slags may provide a reli-able footprint of the conditions during their formation. Thephysico-chemical analysis of metallurgical wastes is thusone of the most powerful tools used by the archaeometallur-gists to collect information about the process [5, 6].

Most archaeological records and associated archaeomet-allurgical investigations so far agree on the fact that the tran-sition to copper mass production during the beginning ofthe Bronze Age is linked with a lowering in the slag viscos-ity, enabling a much better separation of the metallic copperand thus a better production rate [7]. Recent investigationsof slags [8–11] shed light on two main features responsi-ble for this evolution: (i) a better mastering of the over-all chemical composition of the slag and (ii) an improve-ment of reducing conditions within the reactor, which de-creases the amount of magnetite responsible for high viscos-ity [7, 10–12]. Concerning this second feature however, therelationships between redox conditions and working con-ditions are not straightforward. The main reason for thisis that the redox conditions within the ancient charcoal-powered metallurgical reactors pertain to at least two dif-ferent sources of oxygen.

On the one hand, gaseous oxygen is brought by naturalor artificial draught and reacts with the charcoal accordingto the Boudouard equilibrium (3):

C(s) + CO2(g) = 2CO(g) (3)

One of the main process parameter involved is the height ofthe charcoal layer, and consequently the height of the fur-nace. In the following we will refer to this gaseous oxygensource as “pO2”.

On the other hand, the solid source of oxygen may stemfrom the initial content of copper oxide in the charge, thusproviding information on either the initial ore or a possi-ble previous partial roasting step (solid-state oxidation) ofthe copper sulphide-based ore. We will refer to this secondsource of oxygen through the initial atomic ratio of oxygento sulfur in the charge “O:S”.

The aim of this paper is to present the specific method-ology which has been developed at the C2RMF in orderto distinguish between the solid source of oxygen and thegaseous one. Therefore, copper-smelting experimental mod-eling has been carried out, as well as in-depth investigationof both the synthetic products obtained and the archaeolog-ical slags. Most ancient copper slags bear high amounts ofiron, mainly present in the two Fe2+ and Fe3+ oxidationstates since equilibrium temperature and oxygen fugacitylay within the process conditions range, respectively around1000–1200°C and 10−5–10−10 atm [13]. We have focusedon the determination of the oxidation state of iron and on itsmicrostructural distribution, by using several techniques in-cluding Mössbauer spectroscopy, XANES spectroscopy and

electronic microprobe. The new methodology has then beentested on three archaeological case studies.

2 Experimentals

2.1 Experimental modelling of copper smelting

Some 50 experiments modeling slagging processes werecarried out in a laboratory tubular electric furnace, at 1200°Cisotherms. Main isotherms duration were 30 min and cool-ing rate of 500°C/min, but some tests at 8 h and slow coolingof 10°C/min were performed. Five different oxygen partialpressures (pO2) were tested (pO2 = 0.21, 10−3, 10−4, 10−7

and 10−10 atm). These were imposed by using buffering gasmixtures of CO and CO2 at a flow rate of 1 L/min, with a0.1 atm uncertainty. The charge consisted of a 3 g mixture of63 µm to 500 µm powders composed of natural chalcopyrite(CuFeS2), malachite (CuCO3.Cu(OH)2 and quartz (SiO2),placed within a pure kaolinite self-made crucible. More de-tails on the experimentals have been exposed in [7]. The O:Sratio varied according to the relative amounts of chalcopy-rite and malachite, assuming that the copper oxide reactingwith chalcopyrite is tenorite (CuO) according to the mala-chite decomposition at 400°C (4):

Cu(CO3)Cu(OH)2 → CuO + CO2 + H2O (4)

Quartz was added in steochiometric proportion to form fay-alite (Fe2SiO4), meaning atomic Fe:Si ratio equals to 2.Seven different O:S ratio were tested (0; 0.8; 2; 3; 4; ∞). Inthe following, the experiments will be labelled “X/10−y”,X being the O:S ratio and 10−y being the pO2 (atm).

2.2 Analytical procedure

2.2.1 Mineralogical and chemical analysis

After weight and density determination as well as macro-scopic observations, the synthetic products and the archae-ological slags were cut in two parts. One part was finelycrushed, the other part was embedded in epoxy resin andmechanically polished.

Bulk analysis of the slag powders has been carried outby Proton Induced X-Ray Emission (PIXE) at the AGLAEfacility (C2RMF). A 3 MeV proton beam was brought to airthrough a 0.1-µm thick Si3N4 foil, and focused to a diam-eter of 50 µm on the target (5 mm-diameter pellets of slagpowders). Samples were mechanically rastered so the beamcovered a square area of 2 × 2 mm2.

The main crystallized phases were identified by X-raydiffraction on ∼100 mg slag powder in detector scan mode(cobalt anticathode, 0.04◦/s step between 15° and 80° in

The reconstruction of the first copper-smelting processes in Europe during the 4th and the 3rd millennium 715

an additive mode during ∼8 h). Slag mineralogy and mi-crostructure were locally investigated by optical microscopyand SEM-EDS (JEOL JSM-840; 20 kV and FEG-PhilipsXL30; 2 kV) on the polished sections.

2.2.2 Mössbauer spectroscopy

57Fe Mössbauer spectra were obtained at room tempera-ture (293 K) using a spectrometer operating with a conven-tional constant acceleration method. γ -rays were emitted bya 57Co/Rh source. Isomer shifts are related to room tempera-ture Fe metal. Samples were slags finely ground powders, ofwhich the iron concentration (∼10 mg/cm2) provides negli-gible thickness effects. The fitting of the Mössbauer spectrawas carried out using the Hesse and Rubatsch method [14].

2.2.3 µ-XANES and electronic microprobe

Distribution of Fe2+ and Fe3+ between the different consti-tutive phases of the slag has been investigated using spatiallyselective techniques: µ-XANES for the fayalite cristallitesand electronic microprobe for the glassy matrix.

Fe-K-edge X-ray absorption near edge structure(XANES) spectra were collected on polished section slagsin fluorescence mode. Fayalites crystallites were first iden-tified thanks to elemental cartography on major elements Feand Ca. As the beam was focused down to 1 × 1 µm2, fay-alites bigger than 10 µm were selected. Main focus has beenon the pre-edge region, therefore the spectra were collectedin two steps. First a rapid scanning from ∼20 eV below to∼100 eV above the Fe-K-edge (7080–7240 eV) with 0.2 eVsteps: 2 scans at 0.5 s/step were summed up. Secondly, forthe pre-edge region (7100–7119 eV), 13 scans at 3 s/stepwere summed up. The oxidation state of iron was quanti-fied using the method developed by Wilke on the pre-edgefeatures of the spectra [15].

Local Fe3+ measurements in the glassy matrix were ob-tained using electronic microprobe, by measuring the shiftof iron Lα peak, according to the methodology developedby Fialin [16]. The peak positions were determined by usingthe peak search routine of the Cameca PC-controlled-SX100type. In order to limit the beam damage, the experimentalprotocol followed a low beam power (15 keV; 90 nA) anda short time for data collection (30 s). As a consequence,in order to improve the precision of Fe3+/ΣFe measures,we used the average of 60 accumulated peak search runs(for a total time of 30 min). As the beam was focused downto 10 µm2, glass zones bigger than 15 µm were selected.This current protocol gives statistical uncertainties of about+/−7 wt% measured on reference hematite (Fe2O3). Notethat the accuracy and precision on Fe3+/ΣFe dramaticallydepend on the total Fe wt% concentration.

3 Results

3.1 Comparing the roles of O:S and pO2 in the syntheticsmelting products

Depending on the conditions, the synthetic products are con-stituted by the soldification of two or three immiscible liquidphases, namely the metallic copper, the slag, and a resid-ual copper-rich sulphide called matte (Fig. 1). Note that inmost cases, all the iron has gathered within the slag phase.The copper mass balance within the synthetic products isreported in Table 1. It puts into light a total reaction be-tween CuFeS2 and initial CuO, whatever the initial input.As a consequence, the main source of oxygen in the systemappears to be the solid one, as expected by the difference ofreaction kinetics between gaseous and solid oxygen in suchsystems [17], as well as the disproportion of the amounts ofoxygen brought in by sources (Table 2).

Yet, global Fe3+ contents within the slag phase, as mea-sured by Mössbauer spectroscopy (Table 3), point out a clearinfluence of both O:S and pO2 on the slag redox state: asexpected, a monotonous increase of the Fe3+ amount withthe quantity of oxygen is observed in the system whateverits origin. For example, at a fixed pO2 of 10−4 atm, theamount of Fe3+ increases from 0 to 100% when O:S variesfrom 0 to 4. Similarly, at a fixed O:S of 2, amounts of Fe3+range between 24% and 100% as pO2 increases from 10−7

to 10−1 atm. Note that a given amount of Fe3+ may beachieved either by the “gaseous source of oxygen” path orthe “solid source of oxygen” one, as seen notably for the2/10−3 and 2.5/10−4 conditions yielding both around 60%Fe3+, or for the 2/10−4 and 2.5/10−10 conditions leadingto some 35% Fe3+.

Fig. 1 Cross section of a typical synthetic product of in-lab cop-per-smelting simulation. The product is constituted by (1) a centraldroplet of copper, (2) a residual matte and (3) a slag overflowing onthe top

716 E. Burger et al.

Table 1 Comparison between theorical (considering CuO as theunique source of oxygen) and experimental copper mass balancewithin final smelting products. These results have been verifiedfor O:S ratio in the range 0.8 > O:S > ∞, and pO2 in the range

10−3 > pO2 > 10−10 atm. The good adequacy between theorical andexperimental copper mass balance puts into light a total reaction be-tween CuFeS2 and CuO, whatever the initial input

Experi-mentalO:S

Copper massbalance

Corresponding reaction considering CuO as the uniquesource of oxygen

AssociatedO:S(±0.1)

0.8 90% Cu1.66Fe0.16S + 10% Cu CuFeS2 + 1.78CuO → 1.49Cu1.66Fe0.16S + 0.3Cu + 0.76FeO + 0.51SO2 0.9

2 10% Cu2S + 90% Cu CuFeS2 + 4CuO → 1/2Cu2S + 4Cu + FeO + 3/2SO2 2

2.5 100% Cu CuFeS2 + 5CuO → 6Cu + FeO + 2SO2 2.5

3 10% CuFeO2 + 90% Cu CuFeS2 + 6CuO → 20/3Cu + 1/3Fe3O4 + 1/3CuFeO2 + 2SO2 3

4 10% CuFeO2 + 30% Cu2O + 60% Cu CuFeS2 + 8CuO → 2Cu2O + CuFeO4 + Cu + 2SO2 4

∞ 100% Cu2O CuO → 1/2Cu2O + 1/2O2 ∞

Table 2 Calculated amounts of oxygen provided by the solid and gaseous source after 30 min (with a gas-flow of 1 L/min). For those conditions,and pO2 in the range 10−3 > pO2 > 10−10 atm, the solid oxygen is several order of magnitude higher than the gaseous oxygen

O/S 0.8 2 2.5 3 4

Chalcopyrite (mmol) 6.4 3.4 2.9 2.5 2

Fe within the slag (mmol) 4.8 3.4 2.9 2.5 2

Solid Oxygen (mmol) 10.2 13.6 14.5 15 15.7

− log(pO2) 1 3 4 7 10

Gaseous Oxygen after 30 min (mmol) 100 1 0.1 10−6 10−8

O(s):O(g) (assuming O(s) = 10 mmol for every experiments ∼0.1 ∼10 ∼100 ∼1.107 ∼1.109

Table 3 Influence of O:S and pO2 on the global Fe3+/ΣFe within thesynthetic slag. Global Fe3+/ΣFe were measured by Mössbauer spec-troscopy, except for the values marked by an asterisk, which have been

roughly estimated from the nature of the crystallized phases observedwithin the microstructure

O/S 0 0.8 2 2.5 3 4

Chalcopyrite (mmol) 10 6.4 3.4 2.9 2.5 2

Fe within the slag (mmol) 0 4.8 3.4 2.9 2.5 2

Solid Oxygen (mmol) 0 10.2 13.6 14.5 15 15.7

−Log(pO2) Gaseous Oxygenafter 30 min (mmol)

Global %Fe3+

1 100 80% 100% 100% > Fe3+ > 66%∗ 100% Fe3+∗

3 1 63% 100% > Fe3+ > 66%∗ 100% Fe3+∗

4 0.1 0%∗ 7% 35% 60% 100% > Fe3+ > 66%∗ 100% Fe3+∗

7 10−5 24% 100% > Fe3+ > 66%∗ 100% Fe3+∗

10 10−8 36% 100% > Fe3+ > 66%∗ 100% Fe3+∗

The reconstruction of the first copper-smelting processes in Europe during the 4th and the 3rd millennium 717

Thus, in a first approximation, the smelting process maybe described as the succession of reaction (5) and equilib-rium (6):

CuFeS2 + CuO → matte + Cu + FeO (5)

FeO + O2 = Fe3O4 (6)

3.2 Fe3+ distribution within the slags

3.2.1 Mineralogy of the slags

3.2.1.1 The three types of synthetic slags The experimen-tal slags consist of iron-rich silicate glasses containing ran-domly distributed crystallites, which natures depend onthe oxydo-reducing conditions (Fig. 2 and Table 4). Threetypes of slag microstructure could be distinguished. Type A(Fig. 2a) corresponds to the highest reducing conditionstested. It consists of micrometric droplets of quartz and crys-tallized zones of fayalite. During the cooling, the crystal-lization of fayalite, an iron-rich silicate (Fe2SiO4) provokesthe silica saturation of the remaining melt. This leads to thecrystallizations of droplets of quartz. Type B (Fig. 2b) is as-sociated to the intermediate reducing conditions, it shows inaddition to type A microstructure (quartz and fayalite) poly-hedral crystals of magnetite (from 1 µm to 20 µm). Type C(Fig. 2c) slags were obtained under the most oxidizing con-ditions. They exhibit crystals of copper oxides: delafossite(CuFeO2) at O:S = 3, with addition of cuprite (Cu2O) atO:S = 4. Note that the proportion of glassy phase increaseswith reaction time as seen by XRD for experiment 2/10−7.Elemental analysis has shown that the input of Al stem-ming from the corrosion of the kaolinite crucible (7 to 12%Al2O3) may be responsible for this.

3.2.1.2 Partial versus full reaction within the archaeolog-ical slags In-depth characterization of the archaeologicalslags is to be found elsewhere [8–11]. The slags of LaCapitelle du Broum (Hérault, France) and Riparo Gaban(Trentino, Italy) refer to the typical Chalcolithic slag miner-alogy encountered in low copper-production mode contexts[11]. Both the overall high viscosity of the smelted prod-uct and the viscosity of the melted phase prevent full sep-aration of the liquid metal by gravity and thus an efficientprocess. They are highly heterogeneous, porous with lowdensity (SG ∼ 2.7) with abundant unreacted or partially re-acted minerals aggregated by a liquid phase (Fig. 3a). On thecontrary, the Early Bronze Age slags of Saint-Véran (HautesAlpes, France) are homogeneous, dense (SG ∼ 3.5), mas-sive with very little porosity, testifying to complete melting(Fig. 3b). These slags can be clearly assigned to mass pro-duction modes, where the low slag viscosity promotes easyseparation of metal and subsequent high production rates.

Fig. 2 SEM micrograph (back scattered electron) of synthetic slagscross sections. (a) Type A (here Experiment 0.8/10−4): droplets ofquartz (black); crystals of fayalite (gray). (b) Type B (here Experiment2.5/10−4): residual quartz (black); droplets of residual matte (white),cristals of fayalite (dark gray) and magnetite (light gray). (c) Type C(here Experiment 4/10−4): crystals of cuprite (light gray), delafossite(dark grey) and cristobalite (black)

718 E. Burger et al.

Table 4 Influence of O:S and pO2 on the macrostructure of final product (M: Matte, S: Slags, MC: Metallic Copper) and the microstructure ofsynthetic slag (see Fig. 2 for a description of A-type, B-type and C-type slags)

− log(pO2) O/S = 0 0.8 2 2.5 3 4 ∞

1 – – – – – – –

3 M M + S(TypeA) M + S(TypeB) – S(TypeC) S(TypeC) Copper Oxide

+ MC + MC + MC + MC

4 – – M + S(TypeB) – S(TypeC) S(TypeC)

+ MC + MC + MC

7 – – M + S(TypeB) S(TypeB) S(TypeC) S(TypeC)

+ M + MC + MC + MC

10 – – S(TypeB) – – –

+ MC

Fig. 3 Macroscopic pictures of archaeological slags cross sections.(a) Riparo Di Gaban (Chalcolithic); (b) Saint Véran (Early BronzeAge)

Approximately 40 of the 50 slags of La Capitelle investi-gated correspond, for the melted phase at least, to the B-typedefined previously for the synthetic slags, except that augite-like silicates were observed instead of fayalite (Fig. 4a).The 10 remaining slags belong to the C-type. For Riparodi Gaban, all investigated slags pertain exactly to the A-type(Fig. 4b), whereas the slags of Saint Véran are similar to theB-type with a mix of fayalite and augite-like clinopyroxensas the main crystalline silicates (Fig. 4c). The mass-balanceestimations in the archaeological slags are reported in Ta-ble 5.

3.2.2 Overall Distribution of Fe2+ and Fe3+

3.2.2.1 Synthetic slags Mössbauer spectra enabled us todifferentiate between five types of Fe according to theirstructural environment (Table 6 and Fig. 5). The relativeamounts of the different types of Fe in the synthetic slagsvary according to the working conditions, and particularlythe pO2 and the O:S (Fig. 6). The increase of global Fe3+content is closely related to the increase of magnetite, asfar as slags of type A and B are concerned. In the type Cslags however, delafossite formation modifies the Fe2+ andFe3+ distribution drastically. Best example may be seen inexperiments 2/10−3 (type B) and 2.5/10−4 (type C), whichboth exhibit the same total amount of Fe3+ (around 60%Fe): because of the consumption of Fe3+ by delafossite,less magnetite can form, thus forcing the Fe2+ to enter anon-magnetic octahedral environment. Note that this showsa non-equivalence of O:S and pO2 towards Cu oxidation:a larger O:S enhances copper oxide contents. Actually, it ismost probable that the CuO stemming from the decomposi-tion of malachite (reaction (4)) has never been reduced andhas reacted with the iron oxide according to equation (7).

CuO + FeO → CuFeO2 (7)

The reconstruction of the first copper-smelting processes in Europe during the 4th and the 3rd millennium 719

Fig. 4 SEM micrograph (back scattered electron) of archaeologicalslags cross sections. (a) Riparo Di Gaban (Chalcolithic): Residualquartz (black); droplets of residual matte (white), and cristals of fay-alite (light gray). (b) La Capitelle (Chalcolithic): pyroxens (black),magnetite (light gray). (c) Saint Véran (Early Bronze Age): fayalite(light gray) and magnetite (white)

3.2.2.2 Archaeological slags The relative amounts ofFe3+ within the archaeological slags are reported in Table 7.The slags from la Capitelle were not analysed by Möss-bauer spectroscopy. A mean volumic proportion of 30% ofmagnetite and 70% of pyroxens (Ca, Fe)Si2O6 has been esti-mated by observation of the polished sections (Fig. 6). If oneassumes that pyroxens bear only Fe2+ atoms, this yields toa minimum of 44% of Fe3+. Note that at Gaban, Fe3+ con-tents lay just over its saturation limit (∼20 wt% at 1200°C[18]). This explains the very few magnetite amounts ob-served (type A slag).

The distributions of Fe2+ and Fe3+ in the differentphases are represented in Fig. 7. While global Fe3+ con-tents are similar to those observed in the Type A andB synthetic slags, the proportion of non-magnetic Fe3+-silicate compound is much larger in the archaeological slags(about 20%).

3.2.3 The Fe3+-silicate compounds

In the synthetic slags, such increase of the amounts of Fe3+silicate has been observed when reaction time is increasedfrom 30 min to 8 h (Fig. 8). This Fe3+ could be located invarious phases. In order to check if this apparent increasecould be linked to an oxidation of the glassy matrix, theevolution of FeLα peaks in the glass were investigated byelectronic microprobe (Fig. 9). For the synthetic slags, sam-ples showing the largest glass zones were choosen, namelythose obtained after low cooling rate (10°C/min), in orderto match the 10 µm spatial resolution of the technic. In afirst approximation indeed, the cooling rate seems not toplay a major role in the Fe distribution, as seen for exper-iment 2/10−7 when comparing the two cooling rates tested,namely 10 and 500°C/min (Fig. 8 and Table 8). In both typeC (experiment 2.5/10−4) and type A (experiment 0.8/10−4)

slags investigated, the amount of Fe3+ does not evolve sig-nificantly with increasing reaction time. For the type C slag,delafossite and magnetite consume all Fe3+. For the archae-ological slags, one slag from Saint Véran was investigated,showing a mean content of 39 ± 4% of Fe3+.

Yet, both XRD and SEM investigations show that thesynthetic as well as the archaeological slags are highlycrystallized. Consequently, three hypotheses can be for-mulated to attribute the large Fe3+ amounts within non-magnetically ordered crystallites. First, oxidized forms offayalite—laihunite Fe2+Fe3+

2(SiO4)2—have been alreadybeen observed within archaeollogical slags, as reported inliterature [19, 20]. However, this hypothesis might be dis-carded according to the observations made by XANES on5 slags from both Chalcolithic (Riparo di Gaban) and EarlyBronze Age (Saint Véran) slags (Fig. 10). Second, oxidizedforms of pyroxens have also been already observed in nat-ural cristals [21]. Third, slags may contain nanometric sized

720 E. Burger et al.

Table 5 Estimations of copper mass balance in the archaeological slags. The estimation is made by combining (i) the elemental analysis ofresidual matte (SEM-EDS) and (ii) the global Cu:S ratio measured by PIXE on slag powders

Archaeological Initial sulphidic Final residual matte (% mol) (SEM-EDS) Final global molar Final copper mass

site ore Cu Fe S Cu:S (PIXE) balance in the slags

La Capitelle Cu11FeSb4S13 56±2% 0±1% 43±2% 5.0 ± 3 20% matte (CuS) + 80% Cu

Riparo Gaban CuFeS2 40±7% 20±7% 40±2% 1.4 ± 3 50% matte (Cu2FeS2) + 50% Cu

Saint Véran Cu5FeS4 100±3% 0±3% 0±3% → ∞ Cu (no matte)

Table 6 Hyperfine parameters, physical and chemical properties of the five different types of iron observed within experimental and archaeologicalslags by Mössbauer spectroscopy

Magneticordered iron

δ (mm/s) Δ (mm/s) Oxidation state Environment Hypothesis

1.01–1.18 2.2–2.7 Fe2+ Octaedric Fayalite? Glass?

0.36–0.39 1.1–1.7 Fe3+ Octaedric Oxidised Fayalite? Glass?

0.39 0.81–0.87 Fe3+ Octaedric Delafossite

Non magneticordered iron

δ (mm/s) Δ (mm/s) Oxidation state Environment Hypothesis

0.66–0.67 39–44 Fe2+ Octaedric (?) Magnetite

0.27–0.29 Fe3+ Octaedric (?) Magnetite

Fig. 5 Typical Mössbauer spectrum of a B-type slag

magnetite cristallites, which would behave as superpara-magnetic particles [22].

4 Discussion

4.1 Separation of the O:S and pO2 effects

According to the laboratory experiments, the copper sulphide-based ore may be smelted along the following two-step

mechanism. First, depending on the amount of solid oxy-gen in the charge (e.g. O:S), part or all sulphur is removedby a reaction with the copper oxide (Table 1). Simultane-ously, the newly-formed iron oxide reacts with the quartzto form the slag (2). Secondly, an equilibrium is establishedbetween the slag and the gas (6), thus controlling the oxida-tion state of the iron. Indeed, because of its lower density,the slag floats on top of the melt and stays in close contactwith the gas phase. It is well known that kinetics of iron oxi-dation in a silica-based melt is controlled by the diffusion ofthe oxygen [23]. Our experiments have shown that the rel-ative proportion of Fe3+/ΣFe ratio do not evolve betweena 30 min reacted slag time and a 8 h one, thus proving theequilibrium, probably thanks to convection movements inthe low viscosity slags (32 Pa.s). Moreover, the two differ-ent cooling rates tested did not lead to measurable variationsof the global Fe3+/ΣFe ratio in the slag.

The distinction of the specific effects of the O:S ratioand the pO2 makes possible to identify both key parameterswithin the ancient processes by the sole investigation of partof the final product, namely the slag. Accordingly, a two-step analytical methodology was elaborated, using the cor-respondence between Fe3+ and both source of oxygen thathave been calibrated in this study (Table 3).

First, if the initial ore is approximately known, and ifone assumes that the archaeological metallurgical wastesput into light are representative of the final product of the

The reconstruction of the first copper-smelting processes in Europe during the 4th and the 3rd millennium 721

Fig. 6 Influence of O:S andpO2 on the iron distribution ofiron within synthetic slags,measured by Mössbauerspectroscopy (1200◦C, 30 minisotherm, 500◦C/min coolingrate)

Fig. 7 Iron distribution withinarchaeological slag, measuredby Mössbauer spectroscopy.Two slags from Saint Véran(Early Bronze Age) and twoslags from Riparo Di Gaban(Chalcolithic)

Table 7 Global Fe3+/ΣFe within the archaeological slags (La Capitelle and Riparo Di Gaban: Chalcol-ithic, Saint Véran: Early Bronze Age). Theses measures were performed by Mössbauer spectroscopy, exceptfor slags from La Capitelle, where Fe3+/ΣFe were estimated from the surfacic proportions of crystallizedphases in SEM pictures of slag polished sections

Archaeological slags Global %Fe3+ (±5%)

Saint Véran 1 32%

Saint Véran 2 37%

Gaban 1 22%

Gaban 2 23%

La Capitelle 44%∗

722 E. Burger et al.

Table 8 Local Fe3+/ΣFe measured within the glassy matrix on three synthetic slags obtained after low cooling rate (10°C/min) and one ar-chaeological slag from Saint Véran. Fe3+/ΣFe were estimated from the shift of Lα peak of iron as seen by electronic microprobe. These resultsrepresent the average of 60 accumulated peak search runs

Sample Type of slag Isotherm duration %Fe total %Fe3+/ΣFe

0.8/10−4 A 30 min 18% 3.9 ± 1.6%

0.8/10−4 A 8 h 19% 7.7 ± 5.9%

2.5/10−4 C 8 h 22% 1 ± 1%

Saint Véran B 8 h 22% 1 ± 1%

Fig. 8 Influence of the cooling rate (500°/min vs. 10°/min) and thereaction time (30 min vs. 8 h) on the iron distribution within syn-thetic slag, measured by Mössbauer spectroscopy (1200°C, O:S = 2and pO2 = 10−7 atm)

process, the initial O:S ratio in the charge is estimated bymeasuring the quantity of sulphur being removed from theinitial ore. As experimental investigations have shown thatmore than 80% of the sulphur is removed as SO2 [7, 13],such estimation is carried out by the quantification of the re-maining sulphur in the archaeological waste product. Con-sidering the fore-mentioned approximations, in most archae-ological cases the O:S ratio may be estimated with an accu-racy of about 20% at the most.

Secondly, once the O:S is determined, the pO2 can beestimated by measuring the total ratio of Fe3+/ΣFe withinthe slags. A reasonable approximation of this ratio is pro-vided by the amounts of magnetite and delafossite as shownby our experimentations on synthetic slags. Indeed, it hasbeen shown above that main Fe3+ is to be found in mag-netite and/or delafossite, whereas main Fe2+ is to be foundin fayalite, as far as the relative amounts of glassy phase andof clinopyroxen are negligible. In particular, no ferrifayalite

Fig. 9 Position of the FeLα and FeLβ peak maximum for a Fe2+ bear-ing (Fe0.94O) and a Fe3+ bearing (Fe2O3) oxides standard, for a beamenergy of 7 keV. The Fe2+ Lα peak position is found to be shiftedby 2 eV toward lower energies compared to Fe3+ Lα

or laihunite (Fe2+Fe3+2(SiO4)2) has been found, contrary

to previous investigations of protohistoric copper slags byMössbauer spectroscopy [19, 20]. In such a case, quantifi-cation of crystalline phases can be performed by Rietveldrefinement of X-ray diffraction diagrams, for instance. How-ever, large amounts of glass due notably to high amounts ofdivalent cations such as Ca2+ and Mg2+ are encounteredquite often in protohistoric slags. Moreover, these cationsand particularly Ca2+ are also responsible for pyroxen for-mation, since both phases can host appreciable quantities ofFe3+. In such a case, Mössbauer spectroscopy proves nec-essary for an accurate determination of the Fe3+/ΣFe ratioand consequently the pO2.

4.2 Application to ancient smelting processes

The application of the proposed methodology to the threearchaeological cases under study yield three different O:Sranging from 0.6 to 2.3 (Table 9). The input of both theO:S thus calculated and the global Fe3+ contents previ-ously measured (Table 3) yield an approximate value of the

The reconstruction of the first copper-smelting processes in Europe during the 4th and the 3rd millennium 723

Fig. 10 (a) Illustration of a typical Fe-K-edge XANES spectrum andthe fitting of model extracted pre-edge intensities and centroïd posi-tions. The spectrum was modelled using three Gaussian peaks for thepre-edge and two further Gaussian peaks to model the backgroundfrom the main edge in this spectral region. (b) Pre-edge parameters (Fe

co-ordination and oxidation state) of samples plotted in the variogramafter [15]. Dashed lines between fields indicate the variation of pre-edge parameters assuming binary mixtures of respective end-members.Positions of centroïd energy show that, for each measurement, the fay-alites are totally ferrous

Table 9 Estimations of the archaeological O:S ratio from the esti-mated copper mass balance (see Table 5) and the corresponding chem-ical reaction (considering CuO as the unique source of oxygen). This

application yields three archaeological O:S ratio ranging around 1 forChalcolithic sites (La Capitelle and Riparo Di Gaban) and around 2.3for the Early Bronze Age site (Saint Véran)

Archaeological site Chemical Reaction O:S

La Capitelle 17Cu11FeSb4S13 + 231CuO → 114CuS + 304Cu + 17FeO + 68Sb + 107SO2 1.0 ± 0.2

Riparo Gaban 4CuFeS2 + 5CuO → 3Cu2FeS2 + 3Cu + FeO + 2SO2 0.6 + 0.1

Saint Véran Cu5FeS4 + 9CuO → 14Cu + FeO + 4SO2 2.3 + 0.4

pO2 which might have prevailed during the ancient smelt-ing processes. Hence, at Saint Véran, the couple (O:S ∼2.25; Fe3+/Fetot = 33 ± 6%) leads to a pO2 close to10−10 atm. At La Capitelle, the low O:S (∼1) and the rela-tively high Fe3+ (>44%) reveal a very oxidizing atmospherewith a pO2 larger than 10−4 atm. Much more surprising,despite the low Fe3+ content (around 22%), the very lowO:S (∼0.6) at Riparo Gaban witnesses a very oxidizing at-mosphere with a pO2 similar to the one inferred for LaCapitelle.

As far as the proposed new methodology is concerned,one sees clearly how the sole consideration of the Fe3+ con-tent in the slag can be misleading in an attempt to find out thepO2 which used to prevail in the associated ancient metal-lurgical processes. Solid oxygen indeed represents the mainsource of oxygen in these charcoal-powered systems, thusthe initial input of oxygen drives most of the process. Yet,the pO2 still plays a detectable role. Moreover, it representsa major witness of the ancient process working conditionsand deserves, in this respect, to be properly quantified.

Discussing the process parameters responsible for thevariations of O:S and pO2 is beyond the scope of this pa-per, for further details see [13]. Nevertheless, a few indica-tions may be given here in order to illustrate the fact that,at this stage of our knowledge, the apparently low accuracyachieved by our method is much sufficient to bring a hand-ful of new data to the field. Hence, as far as the O:S ratiois concerned, even if one takes into account the maximumvalue estimated for Riparo Gaban, the ratio remains the low-est of the three sites and may reveal a very partial roastingof the ore due to a bad mastering of the solid-state oxidationprocess. On the contrary, the craftsmen of Saint Veran wereable to synthesize the more or less exact quantity of cop-per oxide required for further total co-smelting. Note thatthe Cu:S ratio in the intial ore is much higher at Saint Veran(5/4) than at Riparo Gaban (1/2), which might have been anappreciable help during the roasting step [10]. The interme-diate situation noticed at La Capitelle may be mainly due toan input of solid oxygen as natural ore as suggested by thearchaeological record [9], rather than to a better masteringof the roasting process when compared with Riparo Gaban.

724 E. Burger et al.

Under such a hypothesis, well mastering of the sorting of themixed ore can be inferred at this site despite the uncertaintyof our method.

As far as the pO2 is concerned, the calculations by Re-hder [24] would yield a height of charcoal of approximately60 cm in the furnaces of Saint Véran, whereas the heightof the reactors of La Capitelle and Riparo Gaban would nothave exceeded 30 cm. These results are compatible with thefew data available so far on the evolution of the furnace de-sign during this period [25]. Moreover, these results yieldfor the first time quantitative data on the reactor morphology,while the very existence of a superstructure in protohistoriccopper-smelting furnaces remains unclear.

5 Conclusion

Solid oxygen is shown to be the main source of oxygen inancient copper-smelting processes. Therefore, the quantifi-cation of its initial input in the reacting system proves to benecessary to reconstruct the processes. The determination ofthe mass balance of the reacting system, which is based onthe waste product characterization recovered on archaeolog-ical sites, yields an approximative estimation of this oxygeninput. This said, the pO2 prevailing in the furnace has alsoa marked influence on the final product formation, which isstrongly correlated to the solid oxygen effect. Thus, once thesolid oxygen source is characterized, global quantificationof Fe3+ relative content by Mössbauer spectroscopy enablesthe quantification of this second source of oxygen.

Despite the relatively low accuracy of this new method-ology, mainly due to the often very fragmentary archaeolog-ical data available, it appears very efficient for the compar-ison of ancient smelting processes. Hence, the applicationof this methodology to three protohistoric smelting sites in-volved in various production modes has shown that bothsources of oxygen have been the purpose of technical im-provement.

Although spatially-resolved Fe3+ quantification was car-ried out for the first time in the archaeometallurgical field, itsinformative potential may still be under exploited. Indeed,the local oxidation state of the amorphous and crystallinephases in the waste product may provide new markers ofthe thermal history of the slag formation and thus clarify theprocess parameters. Moreover, this oxidation state may help

addressing one main issue of the ancient process reconstruc-tion, namely the very origin of the solid oxygen, be it stem-ming from natural ore or from anthropic transformation.

Acknowledgement Marine Cotte should be warmly acknowledgedfor her assistance during the XANES measurements at the ESRF.

References

1. F. Bertemes, V. Heyd, in Cultures et sociétés du Bronze ancien enEurope (CTHS, Paris, 1996), pp. 3–36

2. L. Carozza, C. Marcigny, L’âge du Bronze en France (La décou-verte, Paris, 2007)

3. A.K. Biswas, W.G. Davenport, Extractive Metallurgy of Copper,2nd edn. (Pergamon, Elmsford, 1980)

4. P.J. Mackey, Can. Metall. Q 21(3), 221–260 (1982)5. H.G. Bachmann, The Identification of Slags from Archaeologi-

cal Sites. Occasional Publication, vol. 6 (Institute of Archaeology,London, 1982)

6. T. Rehren, E. Pernicka, Archaeometry 50(2), 232–248 (2008)7. E. Burger, Unpublished Ph.D. thesis, Univ. Pierre et Marie Curie,

France (2008)8. G. Artioli, I. Angelini, E. Burger, D. Bourgarit, in Proceed-

ings of International Conference “Archaeometallurgy in Europe”,Aquileia, Italy (CD Rom) (2007)

9. D. Bourgarit et al., in Proceedings of International Conference“Archaeometallurgy in Europe”, Milano, Italy (2003), pp. 431–440

10. D. Bourgarit, P. Rostan, E. Burger, L. Carozza, B. Mille, G. Arti-oli, Hist. Metall. 42(1), 1–11 (2008)

11. D. Bourgarit, in Metals and Mines (Archetype Publications Ltd,London, 2007), pp. 3–14

12. A. Hauptmann, Zür frühen Metallurgie des Kupfers in Fe-nan/Jordanien (Deutsches Bergbau Museum, Bochum, 2000)

13. E. Burger, D. Bourgarit (in preparation)14. J. Hesse, A. Rubartsch, J. Phys. E: Sci. Instrum. 7(7), 526–530

(1974)15. M. Wilke, F. Farges, P.-E. Petit, G.E. Brown, Jr., F. Martin, Am.

Mineral. 86, 714–730 (2001)16. M. Fialin, A. Bezos, C. Wagner, V. Magnien, E. Humler, Am. Min-

eral. 89, 654–662 (2004)17. E. Burger, D. Bourgarit, V. Frotté (in preparation)18. A. Muan, Trans. AIME 193, 965–976 (1955)19. H. Moesta, Interdiscip. Sci. Rev. 11(1), 73–87 (1986)20. B. Mette, Metalla 10(1/2), 1–122 (2005)21. G. Amthauer, G.R. Rossman, Phys. Chem. Miner. 11(1), 37–51

(1983)22. D.J. Dunlop, Phys. Earth Planet. Inter. 26, 1–26 (1981)23. V. Magnien, D.R. Neuville, L. Cormier, J. Roux, J.-L. Haze-

mann, D. de Ligny, S. Pascarelli, I. Vickridge, O. Pinet, P. Richet,Geochim. Cosmochim. Acta 72(8), 2157–2168 (2008)

24. J.E. Rehder, in The Beginnings of Metallurgy (Deutsches BergbauMuseum, Bochum, 1999), pp. 305–315

25. P.T. Craddock, Paléorient 26(2), 151–165 (2000)