Available online at www.sciencedirect.

com

Geochimica et Cosmochimica Acta 75 (2011) 17181732 www.elsevier.com/locate/gca

Gold solubility in oxidized and reduced, water-saturated mac melt

Aaron S. Bell a,, Adam Simon a, Marcel Guillong b
a

Department of Geoscience, High Pressure Science and Engineering Center, University of Nevada Las Vegas, 4505 Maryland Pkwy, Las Vegas, NV 89154-4010, USA b ETH Institute for Isotopengeologie/Mineral.Rohstoe, Clausiusstrasse 25, Zurich, CH 9082, Switzerland Received 31 August 2010; accepted in revised form 2 December 2010; available online 4 January 2011

Abstract We have performed experiments to evaluate Au solubility in natural, water-saturated basaltic melts as a function of oxygen fugacity. Experiments were carried out at 1000 C and 200 MPa, and oxygen fugacity was controlled at the fayalite-magnetitequartz (FMQ) oxygen fugacity buer and FMQ + 4. All experiments were saturated with a metal-chloride aqueous solution loaded initially as a 10 wt% NaCl eq. uid. The stable phase assemblage at FMQ consists of basalt melt, olivine, clinopyroxene, a single-phase aqueous uid, and metallic Au. The stable phase assemblage at FMQ + 4 consists of basalt melt, clinopyroxene, magnetite-spinel solid solution, a single-phase aqueous uid, and metallic Au. Silicate glasses (i.e., quenched melt) and their contained crystalline material were analyzed by using both electron probe microanalysis (EPMA) and laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). Measured Au concentrations in the quenched melt range from 4.8 lg g1 to 0.64 lg g1 at FMQ + 4, and 0.54 lg g1 to 0.1 lg g1 at FMQ. The measured solubility of Au in olivine and clinopyroxene was consistently below the LA-ICP-MS limit of detection (i.e., 0.1 lg g1). These melt solubility data place important limitations on the dissolved Au content of water-saturated, Cl- and S-bearing basaltic liquids at geologically relevant fO2 values. The new data are compared to published, experimentally-determined values for Au solubility in dry and hydrous silicate liquids spanning the compositional range from basalt to rhyolite, and the eects of melt composition, oxygen fugacity, pressure and temperature are discussed. 2011 Elsevier Ltd. All rights reserved.

1. INTRODUCTION The geochemical behavior of Au as a trace component dissolved in silicate liquids, common magmatic silicate minerals, and oxide minerals has implications for processes ranging from the genesis of magmatichydrothermal ore deposits (i.e., porphyry-, high-suldation epithermal-,and layered-mac intrusion-hosted deposits) (Naldrett and Duke, 1980; Hedenquist et al., 1994; Audetat et al., 1998; Frank et al., 2002; Simon et al., 2005; Bell et al., 2009) to

Corresponding author. Address: Dept. of Earth and Planetary

Science, American Museum of Natural History, USA. Tel.: +1 212-769-5459. E-mail address: abell@amnh.org (A.S. Bell). 0016-7037/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.gca.2010.12.022

core formation and the chemical dierentiation of the early bulk earth (ONeill et al., 1995; Ringwood, 1966; Cottrell and Walker, 2006; Righter et al., 2008). Three proposed origins for metallically-fertile, siderophile-element-enriched silicate liquids are: (1) the low degree partial melting of oxidized, sulde-free asthenospheric mantle (Mungall, 2002); (2) the non-mutually exclusive process of siderophile element enrichment in residual liquids, driven by the crystallization and dierentiation of sulde-free basaltic magmas (Mustard et al., 2006); and (3) the large degree of partial melting of asthenospheric mantle that fully incorporates the stable mantle sulde phase(s) into a sulde undersaturated silicate liquid. Mantle derived silicate liquids are widely accepted as the primary agent of chemical mass transfer for Au and other siderophile elements from mantle to crustal reservoirs. Therefore, it is essential to understand

Au solubility oxidized and reduced melts

1719

the thermodynamic and compositional variables that exert inuence on the capacity of mantle derived basaltic liquids to dissolve and potentially fractionate Au and other siderophile elements during open system degassing and crystallization processes. Solubility data dening the maximum Au capacity of silicate liquids in equilibrium with metallic alloys or sulde phases are essential for the purpose of modeling Au delivery into crustal magmatic systems that source ore forming magmatichydrothermal systems. Extant experimental studies have demonstrated the importance of oxygen fugacity (fO2) and to a more limited extent, the eects of pressure and bulk composition, on the solubility of many siderophile elements in silicate liquid. Several experimental studies have concluded that the variation of the fO2 imposed on select silicate liquids alters, by orders of magnitude, the solubility of many siderophile elements in the silicate liquid (Borisov and Palme, 1996, 2000; Ertel et al., 1999; Blaine et al., 2005; Brenan et al., 2005). Experiments with anorthitediopside eutectic liquids demonstrate that the solubility of most siderophile elements decreases linearly with the imposed decreasing oxidation state of the liquid, following a slope dened by the oxidation state of the stable metallic oxide species in the melt. Additionally, the small, yet variable, eect of pressure on the measured solubility of Ni, Pt, and Pd in silicate melt is also an important factor (Ertel et al., 2006; Righter et al., 2008). Botcharnikov et al. (2010) experimentally quantied the effects of dissolved Cl and S on Au solubility in andesitic and rhyodacitic liquids at fO2 $ NNO, 1050 C, and 200 MPa. Their data indicate that the addition of Cl or S to the melt appear to be correlated with an increase in the solubility of Au in the silicate liquid. Jego et al. (2010) quantied the relationship between Au solubility in intermediate silicate melt (i.e., dacitic; $65 wt.% SiO2) and report that increasing fO2, at 1000 C and 400 MPa in a S free assemblage, is correlated with an increase in the solubility of Au in the silicate melt. Jego et al. (2010) also report data from a limited number of S-bearing experiments that are consistent with the results of Botcharnikov et al. (2010) suggesting that the addition of S to intermediate to felsic silicate liquids is correlated with an increase in the measured Au solubility values in these melt compositions. The aforementioned studies have advanced signicantly our collective understanding of the solubility behavior of select siderophile metals in silicate melt. However, there are relatively few experimental data that constrain the solubility behavior of siderophile metals in natural H2O-, Cl-, and S-bearing mac liquids saturated with crystalline silicate and oxide phases. Further, there are few data that constrain the Au-sequestering potential of modally-abundant silicate phases which may exert a signicant eect on the budget of siderophile elements during the evolution of mac magmatic systems. In this study, we performed experiments to elucidate the inuence of fO2 on Au solubility in H2O-saturated, Cl- and S-bearing basaltic silicate liquids, in equilibrium with clinopyroxene and olivine at FMQ and with clinopyroxene and magnetite-spinel solid solution at FMQ + 4, to quantify both the solubility of Au in the silicate liquid and the solubility of Au in clinopyroxene and olivine in equilibrium with the experimental melt to better

understand the eects of crystal fractionation on the Au content of basaltic melts during crystallization along liquid lines of decent. 2. EXPERIMENTAL DESIGN AND RATIONALE 2.1. Experimental techniques and PT conditions Experiments were performed in near-horizontal (i.e., the vessel and furnace are both at a positive angle of 10 relative to the horizontal lab bench), rapid quench titanium zirconium molybdenum (TZM) pressure vessels and a Shaw-membrane-equipped, internally heated gas pressure vessel (IHPV). All experiments were carried out in either Ar H2 or ArCH4H2 mixed gas pressure media at 1000 C and pressures ranging from 195 to 200 MPa. Measured temperature gradients imply that the temperature uncertainty in the TZM runs at any position within the charge is 12 C from the reported run temperature. Temperatures in the IHPV experiments have been measured to be 5 C from the reported run temperatures. Pressure was monitored by using a bourdon tube strain gauge with a precision of 5 MPa in both the IHPV and TZM experiments. 2.2. Starting materials Starting materials consisted of a dolerite powder (tungsten carbide milled; <5 lm nal grain size) from the lower chilled margin of the Ferrar Dolerite, McMurdo Dry Valleys, Antarctica, and a NaClKClHCl aqueous solution. The mineralogy of the starting dolerite consisted of orthopyroxene, plagioclase, augite, and minor oxide phases (Boudreau and Simon, 2007; Hersum et al., 2007). The chemical composition of the starting dolerite is presented in Table 1. The starting aqueous solution was prepared to a total salinity of 10 wt.% NaCl eq. The molar ratios of NaCl:HCl and NaCl:KCl in the aqueous solution were set to unity with a RCl concentration of 1.70 molal. All experiments were contained in Au capsules, xing the Au activity of the experimental system at unity. Starting assemblages, run durations, P, T and fO2 values are listed in Table 2. Some short-duration experiments contained a pre-fractured cylindrical (2 mm OD 5 mm length), inclusion-free chip of quartz in the outer Au capsule in an unsuccessful attempt

Table 1 Major element composition of starting dolerite powder. SiO2 Al2O3 FeOtotal MgO CaO Na2O K2O TiO2 P2O5 MnO Total 53.61 14.55 9.85 6.55 10.59 1.77 0.79 0.63 0.1 0.18 98.72

1720

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732

Table 2 Run conditions for all experiments and quantied Au solubility of the quenched melts. Au concentrations are presented as the mean of 713 unique LA-ICP-MS analyses per quenched experimental glass; the 1r standard deviation follows the mean Au solubility to the right. The 2r error reported is the LA-ICP-MS analytical error derived from counting statistics Run TZM 11 TZM 20 TZM 12 TZM 21 IHPV 1-2 IHPV 1-1 TZM 23 TZM 30 TZM 31 IHPV 1 Duration (H) 7 18.5 24 25 144 172 6 12 24 192 fO2 FMQ + 4.0 0.5 FMQ + 4.0 0.5 FMQ + 4.0 0.5 FMQ + 4.0 0.5 FMQ + 4.0 0.5 FMQ + 4.0 0.5 FMQ 0.5 FMQ 0.5 FMQ 0.5 FMQ 0.1 Pressure 195 195 195 195 200 200 195 195 195 200 MPa 10 MPa 10 MPa 10 MPa 10 MPa 5 MPa 5 MPa 10 MPa 10 MPa 10 MPa 5 Temperature C 990 12 990 12 990 12 990 12 1000 5 1000 5 990 12 990 12 990 12 1000 5 Au 1r (lg g1) 1.46 0.11 1.10 0.35 4.90 0.21 2.90 0.69 1.05 0.03 0.64 0.03 0.35 0.12 0.09 0.01 0.26 0.10 0.49 0.05 2r Analytical error 0.22 0.12 0.36 0.24 0.16 0.1 0.1 0.04 0.08 0.12 SiO2 added Yes Yes Yes Yes No No Yes Yes Yes No

to trap the aqueous phase as synthetic uid inclusions. As discussed below, the long-duration experiments, i.e., those that best represent steady-state conditions, did not contain quartz. 2.3. Capsule conguration The experimental capsule conguration (Fig. 1) consisted of a small length of Au tubing (1 mm ID, 1.2 mm OD 10 mm length) welded at the base and loaded with approximately 1013 mg of powdered dolerite which was compressed into the bottom of the tube. The top of this capsule was mechanically crimped, but not welded. This smaller capsule containing the dolerite was then loaded into a larger Au capsule (4.8 mm ID with a 5 mm OD 20 mm length) along with 30 lL of starting aqueous solution which was pipetted into the base of the outer capsule. This capsule conguration was designed to allow us to recover the silicate melt as a cylindrical aliquot of glass (i.e., quenched melt) + crystalline material that preserves the geometry of the uidmelt interface. The outer capsule was triple crimped and welded shut. Masses of the capsules were monitored both before and after welding to ensure that the solution was not lost during welding. The volume of solution loaded was carefully selected to ensure that the

calculated volume of the aqueous phase at the PT conditions of the experiment (approximated using the density predicted by the algorithm of Driesner et al., 2007) would not exceed the total available volume of the experimental capsules. Based on the analog phase relations for the NaClH2O system (Driesner et al., 2007), the aqueous uid in the experiments is a single phase, supercritical uid at run conditions. 2.4. Control and estimation of experimental fO2 values Oxygen fugacity of the experiments was xed at either $FMQ + 4 or $FMQ. Low fO2 values were achieved through the addition of CH4 and H2 to the Ar pressure medium in the TZM and IHPV experiments, respectively. Oxidizing experiments were run at the intrinsic fO2 of the TZM and IHPV pressure vessels. Several experimental studies have utilized pure-Ar as a pressure medium in TZM and IHPV apparatuses along with NiPd or hydrogen redox sensors to measure fO2 (Popp et al., 1984; Matthews et al., 2003; Berndt et al., 2005; Schuessler et al., 2008). All published experimental studies that have measured the intrinsic fO2 of an Ar-pressure medium TZM and IHPV vessel consistently document that the fO2 of these vessels, accounting for subtle variation in metal content of the vessel, ranges from FMQ + 3.5 to FMQ + 4.5. The fO2 of the IHPV utilized in the current study has been measured as FMQ + 3.5. Based on the available published fO2 measurements and that measured for the IHPV used in the current study, we assumed an oxidation state of FMQ + 4 0.5 for the oxidizing runs. Reducing experiments in the TZM vessels were performed with a xed partial pressure of 6.5 bars of CH4 that was pre-charged into the pressure vessel at ambient conditions. Using the equilibrium constant for the CH4 dissociation reaction and a tabulated hydrogen fugacity (fH2) coecient, it is possible to estimate the fH2 of the pressure medium allowing us to calculate the approximate fO2 of the experiment at run conditions. Calculated fO2 values of the reducing experiments are approximately FMQ 0.5 log units. Hydrogen fugacity in the IHPV experiments was xed by using a Shaw membrane and continuously monitored with a pressure transducer. All experimental charges were saturated with a supercritical aqueous phase in the MeClnH2O (where n is proportional to the charge of the metal) system. In order

Fig. 1. Schematic conguration.

diagram

of

the

experimental

capsule

Au solubility oxidized and reduced melts

1721

to explore the potential for negative deviation the aH2O due to non-ideal mixing in the uid phase, the aH2O of our experimental system was calculated for the analog H2ONaCl system by using the equation of Aranovich and Newton (1996). Calculated aH2O values are 0.96 for the aqueous phase at experimental PT conditions. This value is in good agreement with observations that aH2O tends toward unity due to non-ideal mixing in low salinity uids near the critical point (Webster and Mandeville, 2007) The eect of aH2O on the calculated fO2 values of an experiment is shown in Fig. 2. 3. ANALYTICAL METHODS 3.1. Electron microprobe analysis The glass cylinders and glass-hosted crystals (i.e., cpx and ol at FMQ, and cpx and mt-sp at FMQ + 4) recovered from each run were mounted in epoxy and polished for electron probe microanalysis (EPMA). Phase identication and textural relationships were documented by using energy dispersive spectrometry (EDS) and back-scattered electron (BSE) imaging. Silicate glass, clinopyroxene, and olivine were analyzed for major elements by using a four spectrometer JEOL 8900 SuperProbe at the University of Nevada, Las Vegas to perform wavelength dispersive spectrometry (WDS). Analysis of the silicate glasses employed a 15 kV accelerating potential and a 3 nA beam current used in conjunction with a de-focused 10 lm beam in order to minimize alkali diusion (Morgan and London, 2005). Analysis of pyroxene and olivine utilized a 10 nA beam current with a 2 lm beam diameter. A ZAF matrix correction algorithm was applied to calculate element concentrations. Natural

glass and crystal standards were used for all analyzed elements. Standards for glass analyses included: VG-568 rhyolite glass for Si, Na, and K; sillimanite for Al; hornblende for Ca; VG-A99 basaltic glass for Fe and Ti; kaersutite for Mg; scapolite for Cl; and rhodonite for Mn. The basaltic glass standards VG-A99 and VG-2 were analyzed periodically as blind standards to monitor the accuracy of the glass analytical routine. The concentration of H2O in the glass was determined by EPMA dierence and should be considered a semi-quantitative estimate. 3.2. LA-ICP-MS analysis Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) analyses were performed on the glass and crystalline silicate phases at the Institute for Isotopengeologie and Mineral Rohstoe at the ETH in Zurich, Switzerland. All analyses utilized a homogenized Eximer ArF laser (k = 193 nm) coupled with a Perkin-Elmer 6100 DRC quadrupole ICP mass spectrometer. A carrier gas mixture consisting of He (1.15 L min1) and H2 (5 ml min1) was employed to increase sensitivity and to lower the limits of detection for all elements analyzed (Guillong and Heinrich, 2007). Ablation crater size was varied from 40 lm for glass to 20 lm for clinopyroxene crystals. The repetition rate for the ablation was xed at 10 Hz with energy densities ranging from 10 to 14 J cm1. The SILLS software package (Guillong et al., 2008) and NBS-610 glass standard were utilized for all data processing and reduction. Internal standards of Al and Si (quantied by using EPMA as described above) were utilized in data reduction for homogenous glasses (i.e., long duration experiments). Compositionally, inhomogeneous short duration experiments (discussed below) were reduced

Fig. 2. Plot denes the magnitude of the displacement of position of the FMQ redox buer as a function of water activity in the experimental system. Actual experimental water activities and oxygen fugacity values are marked with the heavy black rectangle. Potential positive deviations of aH2O from unity (not contoured on the diagram) will lead to higher calculated experimental fO2 values (e.g. aH2O = 1.3 yields fO2 = DFMQ + 0.3).

1722

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732

188 via normalization of the major oxide components of the melt to the anhydrous major element analytical totals determined by EPMA. Data reduction by both methods yields statistically identical concentrations of Au and other trace components. The basaltic glass standard VG-A99 was analyzed intermittently as a monitor of analytical accuracy. Additionally, the Au/Ca count-rate ratios for NIST-610 were carefully evaluated for each standard analysis in order to ensure that a potentially inhomogeneous distribution of Au in the NIST-610 glass standard or sensitivity drift with respect to Au was manifested as variability in the calculated Au concentrations of the experimental glasses. No systematic error caused by either inhomogeneous Au distribution in the NIST 610 or sensitivity drift were observed in any of the analytical blocks; thus, the apparent variability in the analyzed Au concentrations, discussed in detail below, represents accurately the actual, subtle variability in the Au content of the experimental glasses. 4. RESULTS AND DATA 4.1. Spatial relationships: analytical transects and phase distributions Recovered cylinders of glass including the glass-hosted crystalline material, were mounted horizontally in epoxy wafers such that the full length of the glass cylinder was exposed and available for analysis. This orientation preserved the original uidmelt interface geometry and allowed us to perform EPMA and LA-ICP-MS analytical transects along the full length of the original melt column. Crystalline material was homogenously distributed throughout the volume of quenched melt in the short duration experiments. The majority of crystalline material in some longer duration runs was conned to the portion of the liquid column opposite the uidmelt interface. Fig. 3 is a back-scattered electron (BSE) image of a recovered glass (i.e., quenched melt) cylinder and illustrates dierences in the spatial distribution of crystalline material. We interpret the observed texture to represent density-driven crystal settling in long duration runs. Clinopyroxene and olivine crystals were analyzed throughout the entire volume of melt when possible. 4.2. Stable phase assemblages The stable phase assemblage varied with the fO2 imposed on the run. Phase assemblages from experiments performed at FMQ + 4 contained silicate liquid, clinopyroxene, magnetite-spinel solid solution, and a single-phase aqueous uid. Phase assemblages recovered from experiments performed at FMQ contained silicate liquid, clinopyroxene, olivine, and a single-phase aqueous uid. The modal abundance of crystalline material was estimated visually to be 3035%, with clinopyroxene being the most abundant crystalline phase at both oxidizing and reducing conditions. The observed clinopyroxene crystals are generally less than 25 lm maximum dimension. Olivine abundance in reducing experiments is low relative to the total abundance of clinopyroxene, with ratios of clinopyroxene/olivine that are
Fig. 3. Backscattered electron image of experimental glass cylinder recovered from IHPV1-1. The arrow represents the typical EPMA transect from the base of the melt column to the uidmelt interface, the latter is not visible in this photomicrograph. The small bright spots represent the class of large metallic nuggets discussed in the text (excluding the large piece of experimental capsule visible in the upper left portion of the image).

approximately 20:1. The grain size of the crystalline material was observed to coarsen with increasing run duration. 4.3. Major element compositions of the quenched melt Table 3 presents the melt composition data for all EPMA analytical transects from a given experiment. The glasses recovered from the long-duration runs are chemically homogeneous (Table 4). EPMA data from an analytical transect of the compositionally homogenous melt from IHPV1-2 are presented in Table 4. We observe some variability in the chemical composition of the silicate liquid in short duration experiments in both oxidizing and reducing experiments (Table 5). Compositional data from an EPMA analytical transect, set orthogonal to the uidmelt interface in short-duration run TZM 21, are presented in Table 5. These data illustrate the tenor and magnitude of the observed compositional variability in the short duration experiments. Concentrations of Na, K and Cl in the melt recovered from quartz-bearing short-duration runs generally increase monotonically approaching the exchange interface from the base of the melt column. The SiO2 content of the melt also increases approaching the uid melt interface, however this increase is generally more restricted and conned to the 100200 lm of melt immediately adjacent to the uidmelt interface. The increasing concentrations of Na, K, Cl, and Si in the melt are interpreted to reect the diusion of Na, K, Cl, and Si into the portions of the liquid that are under-saturated (i.e., reective of disequilibrium with the aqueous phase) with respect to these components. The volumetric extent to which these uid components have interacted with the melt is controlled by the relative dierences in the diusivities of each

Au solubility oxidized and reduced melts Table 3 Average composition of the quenched melt. The 1r standard deviation is contained within parentheses to the right of the average value. The # analyses indicates individual point analyses of quenched glass per experimental charge # Analyses

1723

Na2O

(0.14) (0.10) (0.21) (0.13) (0.26) (0.08) (0.33) (0.12) (0.09) (0.10)

component; thus, un-surprisingly we observed that the network modifying cations Na1+ and K1+ have diused into the melt to a much greater volumetric extent than has the network forming cation IVSi4+. The measured concentrations of Al, Mg, Ca, and Fe decrease from the base of the melt column to the uid melt interface only in the short duration runs. The measured elemental concentration gradients in these runs may reect simple dilution of the melt from the addition of 58 wt.% SiO2 via introduction as a soluble component from the aqueous phase. The depletion of Al, Mg, Ca, and Fe near the uid melt interface may alternatively reect the leaching of these components into the chloride bearing aqueous solution. The exact nature and origin of the compositional gradient in the melt are not specically relevant to Au solubility and will be discussed in detail in a separate publication. We highlight here that the measured Au concentrations along the entire height of the melt column remain relatively consistent except for two short-durations. 4.4. Solubility of Au in quenched melt The measured Au concentrations in silicate glass are presented in Table 2 and plotted as a function of fO2 and experimental run duration in Figs. 4 and 5, respectively. The measured concentrations of Au in the silicate liquid are consistent between runs of the same oxidation state. The concentrations of Au in the melt from experiments at $FMQ + 4 range from a high of 4.9 lg g1 to a low of 0.64 lg g1, with a mean (1r) of 1.98 lg g1 1.45 lg g1. The concentrations of Au in the melt from experiments at $FMQ have a much more restricted range from 0.54 lg g1 to 0.1 lg g1, with a mean (1r) of 0.35 lg g1 0.18 lg g1. The reported solubility values are the average of 713 separate LA-ICP-MS analyses along the entire length and width of the recovered glass cylinder. 4.5. Solubility of Au in clinopyroxene and olivine Quantifying the concentration of Au, by using LA-ICPMS, in the equilibrium crystalline phases proved to be extremely dicult owing to the relatively small size (i.e., 5 35 lm) of clinopyroxene and olivine crystals recovered from most experiments. Furthermore, analyses of small diameter crystalline material would inadvertently include some volume of quenched liquid, yielding a mixed LAICP-MS signal that required de-convolution of the contribution from the crystal and the quenched melt. An additional analytical complication was that clinopyroxene and olivine crystals were often intergrown with Au particles, as displayed in Fig. 6, rendering an analysis contaminated with a large mass of metallic Au. Making use of the fact that Sc concentration in the clinopyroxene was a factor of 45 greater than that of the melt, the count rate 45Sc was monitored in order to delimit the crystal only portion of the signal from the mixed crystal-melt portions of the signals. The study of Brenan et al. (2005) indicates that crystalline olivine may be incredibly sluggish, even at high temperatures, to equilibrate with the siderophile element content of silicate melts. In light of this observation, it is

H2O Total Cl K2O CaO MgO FeO TiO2 Al2O3 SiO2 Run ID

TZM 11 TZM 12 TZM 20 TZM 21 TZM 23 TZM 30 TZM 31 IHPV 1 IHPV 1-1 IHPV 1-2

55.4 (2.07) 53.88 (0.80) 53.85 (2.24) 54.65 (0.68) 54.95 (3.70) 54.13 (0.61) 55.30 (4.33) 51.64 (0.33) 51.10 (0.27) 51.52 (0.22)

15.82 15.30 15.57 15.80 14.77 16.54 14.65 17.36 17.30 17.04

(0.72) (1.10) (0.67) (0.56) (0.77) (0.17) (0.99) (0.13) (0.15) (0.10)

0.65 0.49 0.64 0.64 0.62 0.77 0.65 0.68 0.67 0.66

(0.08) (0.08) (0.10) (0.08) (0.09) (0.11) (0.13) (0.05) (0.08) (0.11)

4.28 3.61 5.25 4.12 5.70 5.32 5.16 3.70 4.41 4.27

(0.59) (0.22) (1.04) (0.09) (2.39) (0.50) (2.36) (0.12) (0.11) (0.11)

3.74 3.70 4.03 3.94 3.93 3.16 3.70 3.91 3.99 4.10

(0.49) (0.24) (0.43) (0.16) (0.57) (0.11) (0.71) (0.10) (0.12) (0.09)

7.53 7.63 7.38 7.90 7.59 7.10 7.35 8.32 8.44 8.50

(1.0) (0.71) (0.86) (0.56) (1.59) (0.26) (1.73) (0.11) (0.08) (0.13)

2.16 2.50 2.32 2.01 2.19 2.50 2.34 2.66 2.64 2.55

(0.70) (0.21) (0.48) (0.35) (0.98) (0.18) (0.89) (0.08) (0.05) (0.05)

2.45 2.20 2.56 2.47 2.42 2.34 2.34 2.86 2.65 2.51

1.50 1.78 1.20 1.33 1.27 1.29 1.24 1.93 2.12 1.98

(0.17) (0.08) (0.09) (0.18) (0.15) (0.09) (0.24) (0.07) (0.10) (0.09)

92.88 91.90 92.65 92.76 93.32 92.98 92.71 92.81 93.58 92.92

(0.29) (0.64) (0.30) (0.42) (0.60) (0.48) (0.53) (0.51) (0.39) (0.30)

7.12 8.1 7.35 7.24 6.62 7.02 7.29 7.19 6.42 7.08

14 10 13 19 13 14 13 16 14 15

1724

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732

Table 4 EPMA analytical transect for IHPV1-2. The run time was 142 h. SiO2 51.64 51.48 51.86 51.27 51.34 51.61 51.14 51.48 51.73 51.43 51.40 51.49 0.21 Al2O3 17.08 17.10 17.25 17.03 17.03 16.99 16.88 17.12 17.10 17.11 16.93 17.06 0.10 TiO2 0.77 0.85 0.59 0.64 0.66 0.71 0.77 0.65 0.65 0.61 0.66 0.69 0.08 FeO 4.40 4.40 4.42 4.44 4.25 4.25 4.25 4.06 4.23 4.09 4.27 4.28 0.13 MgO 3.96 4.13 4.13 4.23 4.29 4.14 4.15 4.05 3.94 4.10 4.16 4.12 0.10 CaO 8.52 8.41 8.23 8.32 8.58 8.55 8.78 8.41 8.61 8.60 8.51 8.50 0.15 K2O 2.68 2.60 2.68 2.67 2.66 2.57 2.60 2.56 2.53 2.52 2.64 2.61 0.06 Na2O 2.73 2.80 2.59 2.85 2.63 2.59 2.56 2.64 2.57 2.70 2.49 2.65 0.11 Cl 1.93 1.93 1.81 2.02 2.02 1.96 1.88 2.12 2.03 2.04 1.97 1.97 0.09 Total 93.33 93.35 93.38 93.13 93.16 93.14 92.63 92.76 93.03 92.90 92.74 93.05 0.26

Table 5 EPMA analytical transect for TZM 21. The run time was 25 h. SiO2 53.62 54.02 53.40 53.86 53.36 53.32 56.31 57.22 57.59 58.51 59.32 55.50 2.32 Al2O3 16.01 16.23 16.56 16.29 16.45 16.64 15.03 15.42 15.17 14.57 14.19 15.69 0.85 TiO2 0.75 0.55 0.49 0.61 0.57 0.75 0.67 0.61 0.69 0.62 0.59 0.63 0.08 FeO 4.65 4.41 4.73 4.58 4.67 4.53 4.21 3.66 3.68 2.76 2.45 4.03 0.80 MgO 4.03 4.08 4.13 4.11 4.02 3.93 3.39 3.24 3.20 2.78 2.91 3.62 0.52 CaO 8.16 8.07 8.43 7.72 8.19 8.31 6.76 6.47 6.19 5.65 5.39 7.21 1.14 K2O 1.58 1.73 1.60 2.14 1.67 1.70 2.63 3.00 3.04 3.53 3.58 2.38 0.80 Na2O 2.44 2.39 2.26 2.40 2.47 2.30 2.56 2.52 2.43 2.62 2.74 2.47 0.14 Cl 1.49 1.48 1.41 1.59 1.58 1.57 1.15 1.32 1.49 1.45 1.47 1.45 0.13 Total 92.54 92.85 92.89 93.11 92.76 92.87 92.60 93.26 93.15 92.20 92.37 92.78 0.33

Fig. 4. Au concentration plotted as a function of oxygen fugacity. The dashed line is the linear regression including all of data in the current study; the dotted line is the linear regression of only the long duration experiments. The lled circles are all of the Au solubility data from Botcharnikov et al. (2010), who report that the increasing Au concentration in their study corresponds to increasing chlorine and sulfur activities in the melt. Error bars on the plot represent the 1r standard deviation based on the mean of each datum.

Au solubility oxidized and reduced melts

1725

Fig. 5. Au concentration as a function of run duration. High Au concentrations in the short duration experiments may reect the higher Au solubility in relatively oxidizing conditions in the rst few hours of the experiments run at the intrinsic oxidation state (i.e., FMQ + 4.5). The xs represent reducing runs and the circles represent oxidizing runs. Error bars represent the 1r standard deviations; where no error bars are visible the standard deviations are smaller than the symbols.

analyses of both olivine and clinopyroxene ranged from 0.1 to 0.05 lg g1. We suggest that these data constrain the Au solubility of both clinopyroxene and olivine to <0.1 lg g1. 4.6. Attainment of equilibrium in the experimental silicate melts The EPMA data for the recovered glasses from longduration experiments (i.e., >100 h run times) indicate that the recovered glasses are compositionally homogeneous with respect to all elements, including Au. While this does not explicitly demonstrate equilibrium, it is a necessary condition for equilibrium. Demonstrating absolutely the attainment of equilibrium with respect to Au in the silicate liquids is challenged by the fact that our experiments approached equilibrium by saturating the silicate liquid with Au from the low Au concentration side. To our knowledge, the demonstration of equilibrium with respect to siderophile element solubility has to date not been demonstrated through experimental reversals (i.e., approaching the equilibrium concentration from both sides of the reaction). Similarly, no thermodynamic data for dissolved AuO species in silicate liquids are available such that equilibrium Au solubilities can be calculated from a purely theoretical approach. Thus, in order to evaluate the potential time lengths required to reach equilibrium with respect to Au solubility in our experiments, we calculated the length of time required for Au to fully diuse throughout the entire volume of silicate liquid in a given experiment; in this application of this calculation to our experimental system, we have assumed that Au diusion of from Au-saturated to Au-undesaturated portions of the liquid represents the largest kinetic barrier to attainment of equilibrium Au concentration in the melt. There are no data

Fig. 6. Backscattered electron image of a LA-ICP-MS crater from the analysis of a clinopyroxene crystal. The crater contains a large metallic Au particle; a second glass-hosted gold nugget at the interface between melt and another crystal is denoted by the arrow.

possible that the Au concentrations measured in crystalline material from the short duration runs may be out of equilibrium with the Au content of the early silicate liquids. Despite our best attempts to reduce the LA-ICP-MS limits of detection for Au by employing H2 in the carrier gas (cf., Guillong and Heinrich, 2007) and utilizing the largest possible beam diameter for analyzing the crystalline material, the measured Au concentrations were consistently below the limit of detection. The limits of detection for the

1726

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732

in the literature which explicitly constrain the diusion of Au in hydrous silicate melts; however, following the method of Mungall (2002) we have estimated a diusion coecient using an eective ionic radius of 1.54 A for Au1+ coecient 9 2 1 for Au of 1.7 10 m s for the our experimental melt at 1000 C. Using a very conservative diusion length scale for our experimental conguration of 5 mm, we calculate that Au should be homogenous distributed in the melt in approximately 4 h.. However, as discussed below, the measured variability in Au solubility in short-duration runs (i.e., runs of less than 24 h in duration) indicates that a homogenous Au distribution was not achieved. We also note that the long-duration experiments in the current study have run times exceeding 100 h. These run times are consistent with a plethora of published experimental studies that document that run times >100 h at the experimental conditions used in the current study are sucient for silicate liquids to fully equilibrate with the experimental metal capsule. (Candela and Holland, 1984; Webster, 1992; Frank et al., 2002; Simon et al., 2004, 2005, 2006, 2007; Botcharnikov et al., 2010; Jego et al., 2010). 5. DISCUSSION 5.1. Insights from LA-ICP-MS signal processing and equilibrium metallic micronuggets LA-ICP-MS analysis of quenched silicate melt aords unique insights into the spatial distribution of Au in the melt, beyond facilitating the simple exclusion of Au particles and micronuggets from signals. The presence, and treatment during signal processing, of metal micronuggets in silicate glass has been discussed in a number of studies (e.g., Cottrell and Walker, 2006, Simon et al., 2008; Jego et al., 2010) with no consensus on whether the observed micronuggets should be included or ltered (i.e., removed) from LA-ICP-MS transient signals during signal processing. In the current study, we observed the presence of both large gold nuggets, i.e., visible in BSE images (Fig. 6), and small scale inhomogeneties in the LA-ICP-MS signal which are not correlated with Au particles visible in BSE images. We interpret the BSE-observable Au nuggets as equilibrium metallic phases that coexisted with the melt and not as metal nuggets which nucleated and grew during quench. These observed large Au particles, often associated with vesicles or crystalline material were generally visible and easily avoided during LA-ICP-MS analysis. A second type of cryptic nugget was identied only during LA-ICP-MS signal processing of experimental silicate glasses with low Au concentrations that generally displayed inhomogeneous Au count rates (1r ranges from 30% to 50% relative) even in the areas of melt where we observed no visible metallic particles. The results from previous studies of the solubility of metals in silicate melts suggest that similar heterogeneities observed in nominally nugget-free portions of LA-ICP-MS signals may represent nano-nuggets that cannot be spatially resolved and, thus, cannot be removed during signal processing (Ertel et al., 1999). Such cryptic nano-particles contaminate the signal with excess metallic Au, yielding concentrations that are not

representative of the total dissolved metal content of the quenched liquid. To explore whether or not the small-scale Au inhomogeneties (<50% relative to the mean count-rate) in the transient signal are in fact Au particles that require ltering during signal processing, we compared the transient signal of Au to those of other elements which should exist only as fully dissolved components of the melt. Comparisons of the raw count-rate of 197Au with the raw count-rates of 173Yb, 238U, and 232Th for an individual LA-ICP-MS signal reveal a striking similarity with respect to the lack of countrate homogeneity. The measured isotopes 173Yb, 238U, and 232Th were chosen for the comparison because the measured concentrations of these elements overlap with that of the measured Au concentrations. For example, the mean count rate for 197Au from one analysis of glass in experiment IHPV1-1 is 339 counts s1, with a 1r standard deviation of 111 counts s1 (32%). Mean count rates and standard deviations at the 1r level for the same analysis for 173Yb, 238U, and 232Th are 392 240 (59%), 971 359 (22%), and 3417 1012 (29%), respectively. Based on the comparison of the standard deviations from the mean count rates, the homogeneity of Au concentrations within the integration interval are comparable to the concentrations of other trace elements endemic to the composition of the starting material, and as such, we suggest that these variations represent the stochastic variability of the signal itself, rather than the contribution of an additional metallic phase. Fig. 7 contains the raw LA-ICP-MS signal used for the comparison. The origin of the heterogeneous nature of the signals is unclear; however, invoking a cause and eect relationship between the signal and Au particles is inconsistent with the observation of similar variability in the signals of elements with no documented nuggetting behavior. The exact interpretation of the link between LAICP-MS signals and the physical state of the analytes remains speculative at best. We suggest that the small-scale inhomogeneous nature of the Au signals may be related in some capacity to the application of the LA-ICP-MS analytical technique and that the observed small-scale perturbations (less than 2r from the mean count rate) in the Au transient signal should be included during signal processing. To lter (i.e., remove) these perturbations results in an articially lower calculated metal solubility in the silicate liquid. When concentrations are plotted as a function of ablation time, none of the elements used in the count-rate comparison display the large concentration spikes that are the hallmark of the nuggeting behavior in the siderophile elements. At the 2-sigma level the intrinsic variability concentrations calculated at each data acquisition point in the spectra (Appendix 1) dier by no more than a factor of 1.52.0. We feel that this observation, in particular, is the most convincing piece of evidence that the variability in Au count rates cannot be attributed to cryptic Au particles. We recommend that large Au particles comprising a distinct class of micro-nuggets should be routinely ltered out of the LA-ICP-MS signals. However, cryptic contamination of LA-ICP-MS signals by sub-micron metallic particles does not appear to manifest itself as time variable intensities with high standard deviations (50% relative); hence, these cryptic Au particles should not be ltered.

Au solubility oxidized and reduced melts

1727

Fig. 7. LA-ICP-MS signal from IHPV1-1. Note the similarity in the one sigma standard deviations for the count rates of Au, Th, Pb, and U. No individual point in this signal corresponds to 1r < 50% relative, indicating that no large nuggets are present.

5.2. Au solubility in silicate melts Previous studies of Au solubility in hydrous and dry basaltic liquids over a range of pressures and temperatures have reported Au solubility values ranging from <1 lg g1 to 60 lg g1 (Borisov and Palme, 1996; Jugo et al., 1999; Frank et al., 2002; Simon et al., 2003; Brenan et al., 2005; Simon et al., 2005; Bell et al., 2009; Jego et al., 2010; Botcharnikov et al., 2010) The general conclusion from these studies is that Au solubility displays a rst order dependence on oxygen fugacity values greater than the FMQ redox buer and is generally accepted to be present in the melt as an oxide species. Dissolution of Au into silicate liquids as a monovalent oxide species was proposed by Borisov and Palme (1996) in following the form: 1 Au O2 AuOmelt 0:5 4 1

Rearranging the equilibrium constant for the above reaction in logarithmic form yields: log10 AuO0:5 1 log 10f O2 K 4 2

The Au concentrations measured in the experimental melts in this study do not strictly adhere to the anticipated slope of 0.25 predicted by the linear relationship between Au solubility as a monovalent oxide species and the imposed fO2 value. The slope of the linear regression in Fig. 4 is 0.15 with a standard error of 0.049 at the 2r condence interval; thus the deviation of the calculated slope from 0.25 is relatively minor considering the error of the regression slope. Despite the scatter in the data and the dierence in the slope, the measured Au concentrations

from experiments at $FMQ are consistently lower than the measured Au concentrations from the experiments at FMQ + 4. The deviation of the Au concentrations in the reducing experiments from the values predicted by the relationship in Eq. (2) may reect the changing speciation of Au in silicate liquids with decreasing fO2. For example, we calculate a predicted Au solubility value of 0.18 lg g1 at FMQ for Au as a monovalent oxide species. Calculation of this value was accomplished by using the Au solubilities obtained from IHPV 1-1 and IHPV 1-2 in conjunction with the theoretical slope for Au dissolution as the species AuO0.5. This value is a factor of two lower than the measured value of 0.35 lg g1. We speculate that the elevated Au concentrations at low fO2 values are caused by the formation and increasing abundance of either AuCl or AuS, or both, complexes in the melt. Furthermore, the possibility of a stable, zero valence Au-silicide species, as suggested by Borisov and Palme (1996) cannot be excluded from consideration. The exclusion of the high Au solubility values in the 24 h TZM runs, perhaps caused by initial fO2 values higher than FMQ + 4 during attainment of the imposed fO2 (see below for explanation), further decreases the slope of the observed AufO2 relationship, strengthening the evidence for alternative Au species. The potential existence of a chloride species, however, seems most plausible given the high chlorine activity in the experimental melt. The probability of the existence of an Ausulde species is assumed to be zero in basalt melt at FMQ + 4 because S is present dominantly as sulfate in silicate liquid at this redox condition (Mavrogenes and ONeill, 1999; Jugo et al., 2005). Botcharnikov et al. (2010) report an apparent linear correlation between Au concentration and Cl activity in andesitic melts at fO2 values of approximately DNNO + 0 and propose the existence of a AuCl species in the melt. We

1728

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732

suggest that AuO0.5 is the most prominent Au species in chloride-bearing silicate melts at the high oxidation states prevalent over the rst several hours at run temperature. Thus, fO2 is the primary variable controlling Au solubility in the early experimental liquids. As fO2 values inside the charge adjust to the imposed fO2, the dominant Au speciation may shift to either AuCl or AuS complexes, muting the eect of fO2 while emphasizing the eects of the chlorine and sulfur activities of the melt. This nding is consistent with the results from Botcharnikov et al. (2010). Interestingly, no consistent relationship between the composition of the melt and Au content is observed. In experiments TZM 20 and TZM 21 the concentration of Au increases subtly by a factor of 1.52.0 (i.e., an absolute increase of 1 lg g1 to 1.8 lg g1) in the volume of melt adjacent the uidmelt interface. The origin of this Au concentration gradient appears to be correlated with the total alkali, Cl, and Si content of the liquid such that portions of the melt richer in alkalis, Cl, and Si contain higher concentrations of Au. However, this relationship is not observed in all experiments and awaits future study. The observed temporal evolution of the Au contents of the melts is signicant for the experiments run at FMQ + 4. The temporal variation of Au solubility in the melt is plotted in Fig. 5. The highest measured Au concentrations in silicate glass are for the experiments of 24 h duration. The measured concentrations of Au in silicate glass subsequently decrease with increasing run time. We suggest that the temporally increasing Au concentrations are caused by the initially highly oxidizing conditions of the charge present before internal redox equilibrium is attained. In contrast, experiments at reducing conditions have high fH2 values such that the osmotic equilibrium of H2 between the pressure vessel and the charge attains redox equilibrium more quickly relative to the experiments at intrinsic oxygen fugacity values (cf., Gaillard et al., 2002). Temporal variability in the measured Au content of the silicate liquids is not observed in the experiments at $FMQ. Inadvertent, subtle variations in the ratio of atmospheric gas to aqueous solution loaded into the charges may also aect the initial oxidation state of the charge, causing Au to dissolve readily into the melt as an oxide species during the initial hours of the experiment. 5.3. Au concentration in clinopyroxene and olivine Data from the long duration experiments for Au solubility in olivine are consistent with data from Brenan et al. (2005). They report maximum Au concentrations in olivine ranging from 0.1 lg g1 to 0.005 lg g1. The highly incompatible nature of Au in the lattice of both clinopyroxene and olivine phases is not unexpected considering the relatively large ionic radius and the probable 1+ oxidation state of Au. As a monovalent cation, the diadochy of Au in the olivine lattice appears to be extremely limited. Data from the current study suggest that Au compatibility on the clinopyroxene lattice appears similar in magnitude to that of olivine, but the relatively high LA-ICP-MS limits of detection prohibit making more than semi-quantitative comparisons. Potential coupled substitution mechanisms involving Au1+ in the M2 site accompanied by Fe3+ or Al in the M1

site, or alternatively the presence of IVAl may enhance Au compatibility in clinopyroxene. It seems unlikely that coupled substitutions involving Au1+ would greatly enhance the capacity of clinopyroxene to sequester signicant quantities of Au. The extremely low Au activities endemic to natural silicate liquids are likely the dominant factor dictating the Au content of the crystalline material of, even considering potential coupled substitution mechanisms. However, compositionally controlled enhancement of partition coecient values for Au between pyroxenes and melt cannot be excluded from further consideration and may have important implications for the evolution of the Au budgets of highly fractionated silicate liquids. 5.4. A global comparison of Au solubility data in melts with relevance to ore-forming magmatic systems In Fig. 8 we present the new data from the current study and published Au solubility data that allow us to explore the eects of fO2, pressure, temperature, melt composition, and melt volatile content on the solubility of Au in a silicate liquid saturated with metallic Au. The fO2 values for all data plotted in Fig. 8 have been recast relative to the FMQ buer in order to normalize for the eects of temperature. We include only data from Au-saturated experimental assemblages. Several experimental studies of Au solubility in silicate liquids have been performed in hydrous, Cl-, S-, bearing haplogranite assemblages. Bell et al. (2009) quantied Au solubility in hydrous S- and Cl-bearing haplogranite; the silicate melts were saturated with intermediate solid solution and pyrrhotite, or a Cu Fe-sulde melt. They used LA-ICP-MS to quantify Au solubility, and report values that range from 0.35 lg g1 to 0.019 lg g1 at 800 C and 150 MPa. Simon et al. (2007) report a Au solubility value of 0.60 lg g1, quantied by using LA-ICP-MS, in pyrrhotite-saturated haplogranite at 800 C and 120 MPa. Studies by Simon et al. (2003) and Simon et al. (2005) report Au solubility values ranging from 0.15 lg g1 to 0.79 lg g1, quantied by using LA-ICPMS, in hydrous Cl-bearing haplogranite at 800 C over pressures ranging from 100 MPa to 145 MPa. Frank et al. (2002) quantied Au solubility in a hydrous Cl-bearing haplogranite at 150 MPa and 800 C by using secondary ion mass spectrometry; they report a mean Au solubility of 1.0 lg g1. Jugo et al. (1999) report a mean Au solubility of 4.0 lg g1 in hydrous, Cl-bearing haplogranite, saturated with intermediate solid solution and pyrrhotite at 850 C and 100 MPa. It is tempting to associate the increased measured Au solubility in the silicate liquid in the Jugo et al. (1999) study with the presence of sulfur, but we highlight that the study by Bell et al. (2009) included the identical sulde assemblage and the measured Au solubility values in the latter study are approximately one order of magnitude less than those reported by Jugo et al. (1999). We suggest that perhaps the measured Au concentrations in the melt in the Jugo et al. (1999) study may reect the use of the bulk technique of instrumental neutron activation analysis (INAA) which may include Au micronuggets in the analysis; hence, articially increasing the measured Au concentration.

Au solubility oxidized and reduced melts

1729

Fig. 8. Plot contains all of the published Au solubility values that are compared and discussed in Section 5.4 of the text.

The solubility of Au in intermediate composition melts has been the focus of several recent studies by Botcharnikov et al. (2010) Jego et al. (2010) and Zajacz et al. (2010). Botcharnikov et al. (2010) determined Au solubility in water-saturated andesitic and rhyodacitic melts as a function of Cl and S concentrations at fO2 $ NNO, 1000 C, and 200 MPa. LA-ICP-MS determined Au concentrations in their study range from 0.32 lg g1 to 0.77 lg g1 in andesitic melt and from 0.33 lg g1 to 0.83 lg g1 in dacitic melt. The lowest values in the quoted concentration range represent the S- and Cl-free liquids, whereas the larger values represent a measured increase in Au concentration of the silicate liquid that is presumed to be associated with increasing S and Cl activity in the silicate liquid. Jego et al. (2010) report concentrations of Au ranging from 0.03 lg g1 to 0.25 lg g1 in S-free, water-saturated melt of dacitic and andesitic composition at 1000 C, 400 MPa, and fO2 xed at $NNO 1, $NNO +1.5 and $NNO +3. Jego et al. (2010) reported that the solubility of Au in the silicate liquid is correlated with increasing fO2, in a manner consistent with the discussion above. Jego et al. (2010) also report data from one series of S bearing, hydrous dacite melts that are consistent with the results of Botcharnikov et al. (2010) wherein a positive correlation between Au solubility and S abundance is observed. Jego et al. (2010) report that Au solubility increases from $0.1 lg g1 to between $1 and $4 lg g1 with the addition of S to dacite melt held at an fO2 of $NNO-1. Zajacz et al. (2010) report Au solubility values, determined by using LA-ICP-MS, in water-saturated andesite at 1000 C and 200 MPa, fO2 of NNO-0.4, of 80 ng g1 in a S-free assemblage and a maximum of 0.65 lg g1 in the S bearing assemblage. Gold solubility has also been quantied in anhydrous S- and Cl-free magmatic systems over a wide range of fO2 values. Borisov and Palme (1996) report values ranging from 0.30 lg g1 to 1.0 lg g1 in anorthitediopside melt at 1300 1400 C and atmospheric pressure. Brenan et al (2005) have

quantied Au solubility in natural, anhydrous, S-free basaltic liquids by using LA-ICP-MS. They reported values range from 12.0 lg g1 to 0.86 lg g1. Oxygen fugacity values in both of these studies were varied over multiple orders of magnitude (Fig. 8); thus, the large range in reported values is attributed to the large range of imposed fO2 values. Both of these studies convincingly demonstrate that Au solubility in silicate liquid, in the absence of volatiles (i.e., H2O, Cl, S) is a function of fO2. Jugo et al. (2005) quantied Au solubility in a sulde- and sulfate-saturated anhydrous basalt (lherzolitic coexisting assemblage of olivine, orthopyroxene, clinopyroxene and spinel) at 1300 C, 1 GPa and fO2 of FMQ 1 (sulde saturated) and FMQ + 1.9 (sulfate saturated). They used LA-ICP-MS to measure Au concentration in the quenched silicate glass and report that Au solubility increased from 0.52 0.08 lg g1 in the reduced, sulde saturated assemblage, to 4.7 0.9 lg g1 in the oxidized, sulfate-saturated assemblage. The authors do not speculate as to whether the measured increase in Au solubility is strictly a function of fO2, the presence and speciation of S in the silicate liquid (i.e., sulde vs. sulfate), or both. We note that the results of Jugo et al. (2005) are consistent with observations in the current study as well as those reported in Botcharnikov et al. (2010) and Jego et al. (2010) suggesting a symbiotic relationship between Au solubility and dissolved sulfur in silicate liquids that are basaltic to intermediate in composition (i.e., contrast these results with experimental data from rhyolite assemblages). The new and published Au solubility data (Fig. 8) indicate that Au solubility in silicate liquids, across the range of geologically plausible pressure, temperature, bulk composition, including potential Au complexing volatiles S and Cl, and oxidation state documented for the evolution of magmatichydrothermal systems (i.e., from $FMQ 1 to $FMQ + 3), varies roughly one order of magnitude from $0.05 ug g1 to $1 ug g1. The Au solubility data for volatile-free basaltic liquids (i.e., Borisov and Palme, 1996; Brenan et al., 2005) indicate a positive

1730

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732

correlation between Au solubility and fO2 (see Section 5.2). The reported Au solubility values obtained for volatile-free basaltic liquid are elevated slightly, albeit less than one order of magnitude, with respect to the Au solubility values obtained from water-saturated, Cl- and S-bearing basalt (i.e., compare data from the current study with those from Borisov and Palme (1996) and Brenan et al. (2005) in Fig. 8). We suggest that this may reect the higher temperatures of the dry mac melts following the general relationship observed by Borisov and Palme (2000) for Au concentration and temperature in dry haplobasaltic melts. There are no data with which to evaluate the relationship between the solubility of Au in rhyolite melts and the imposed fO2 of the system; i.e., all experiments to quantify Au solubility in rhyolite melts have been performed at approximately NNO 0.5 (Fig. 8). There is no reported observed systematic relationship between Au solubility and the activity of S in the experimental data from rhyolite melt compositions (i.e., Bell et al. (2009) experiments were sulde saturated; Frank et al. (2002) experiments were sulde free; Simon et al. (2003) experiments were sulde free; Simon et al. (2005) experiments were sulde free; Simon et al. (2007) experiments were sulde saturated; Jugo et al. (1999) experiments were sulde saturated). The range of sulfur concentrations in the sulde-saturated rhyolite melts spans the range observed for natural sulde-saturated rhyolite melts; i.e., $50 lg g1 to $300 lg g1. We also note that the Cl concentration in these experimental melts varies across the range observed for natural felsic liquids and there is seemingly no systematic relationship between Au solubility and the Cl content of these experimental melts. The available data for intermediate melt compositions (i.e., dacites from Jego et al. (2010), rhyodacites and andesites from Botcharnikov et al. (2010), and andesites from Zajacz et al. (2010), respectively) suggest that increasing the activity of S in an intermediate to andesitic melt is correlated with an increase in the measured Au solubility in the silicate liquid. The absolute magnitude of the increase in Au solubility, over the range of experimentally imposed sulfur activities, is approximately one-half of one order of magnitude in the Botcharnikov et al. (2010) study, approximately one order of magnitude in the Jego et al. (2010) study, and approximately one order of magnitude in the Zajacz et al. (2010) study. These data clearly illustrate that the presence of S has a measurable eect on the solubility of Au in intermediate to mac silicate liquids. However, the available data do not conclusively identify the speciation of Au in these intermediate to mac liquids, and, thus, we feel that it is premature to advocate for a dominant Ausulfur species in the silicate liquid. Further, the lack of any observed correlation between Au solubility and the S content of rhyolite melts, over a wide range of S concentrations, suggests that perhaps melt structure plays a role in controlling Au solubility in evolved silicate liquids. Clearly, additional experiments are required to fully address the speciative role of S, and Cl, on the behavior of gold, and other siderophile elements, in mac to intermediate silicate liquids. The Au solubility data presented and discussed above have important implications for the ore-potential of magmatic systems. In order to apply these experimental data

to natural systems it is necessary to consider the dierence between the concentration of Au in a silicate liquid that is in equilibrium with metallic Au and the concentrations of Au in natural silicate melts. Frank et al. (2002) report that the concentration of gold in natural silicate melts is approximately three orders of magnitude below that required for metallic Au saturation (aAu = unity). Thus, it is reasonable to expect that most silicate liquids, across a wide range of compositions at geologically reasonable pressures, temperatures, fO2, and fS2, will remain Au-undersaturated during their evolution. The available experimental data indicate that increasing concentrations of volatiles, such as S and Cl, will increase the absolute concentration of Au in mac to intermediate melts; however, the measured magnitude of this increase (i.e., from $0.1 to $1 lg g1), assuming that the positive eect of S and Cl on the dissolved concentration of Au in these liquids scales linearly from Au-saturation to the lower Au concentrations in natural melts, cannot be solely responsible for the ore-generative ability of a magmatic system. Gold-undersaturated melts inherit a nite Au budget from melting the ultimate source material (perhaps a few ng g1), thus the absolute Au concentration of the melt is dependent on the initial Au input from the source as well as the partitioning behavior of Au during ascent and fractionation. The data presented and discussed above suggest that the absolute abundance of Au-complexing volatiles does not modify signicantly the source-inherited metal budget in Au undersaturated magmas but, rather will alter Au partitioning behavior among the melt, aqueous, and crystalline phases. Certainly, more work is required to constrain comprehensively the eect of the role of dissolved volatiles on Au solubility in melts across the compositional range from basalt to rhyolite; however, in light of the available data (Fig. 8), we suggest that while magmatic variables such as fO2 and dissolved sulfur certainly play a deterministic role in moderating the absolute abundance of Au in mac to intermediate silicate liquids, the critical controls on the ore potential of any magmatic system are the timing and the nature of the volatile phase(s) that is exsolved from the silicate melt (cf. Burnham, 1979; Candela and Piccoli, 2005). The timing of volatile phase saturation is critical owing to the requirement, in terms of ore potential, that the volatile phase must percolate through the magma and vent to the overlying country rock (Candela, 1991). A silicate melt that saturates with a volatile phase after signicant sulde crystallization clearly has a low ore potential, irrespective of the original Au content of the melt. The composition of the volatile phase, specically the abundance of metal-scavenging ligands (i.e., S and Cl), controls the mass transfer of Au, and other metals, from the melt to the potentially ore-formative volatile phase. Thus, it is not necessarily as important that higher concentrations of S and Cl in the silicate melt increase the Au-carrying capacity of the melt, but rather that higher concentrations of S and Cl, under the optimal redox conditions where the majority of the S is present as S2, lead to signicantly greater mass transfer of the Au from the melt to the volatile phase and this is potentially the most important factor in the development of a magmatichydrothermal ore deposit.

Au solubility oxidized and reduced melts

1731

6. CONCLUSIONS The new experimentally determined Au solubility data place important constraints on the maximum Au capacity of water-saturated mac silicate liquids, over the range of fO2 documented for ore-generative magmatichydrothermal systems, containing geologically realistic (with respect to ore forming systems) concentrations of chlorine and sulfur. A comparison of the new data with published Au solubility data indicates that the precise nature of siderophile element speciation in silicate liquids at the fO2 values expected for most magmatichydrothermal systems remains enigmatic. There is clearly a positive relationship between Au solubility and the fO2 of the system; however, data from the current study and available literature indicate that the addition of magmatic volatiles such as Cl and S may act as complexing anions that enhance the overall Au solubility in mac to intermediate silicate melts, as well as altering Au speciation at low, geologically relevant fO2 values. The presence of water in silicate liquids seems to have little eect on the Au content of silicate melts, based on comparisons of Au solubility data in both hydrous and anhydrous studies. The existence of prominent Au species in silicate melts other than simple oxides has not yet been conrmed by spectroscopic techniques and must be inferred through experimentally produced solubility data wherein all variables are xed except for the studied volatile. Perhaps suitable analog melts containing appropriately high quantities of Au may be used in the future to quantitatively measure Au speciation. Furthermore, the eect of pressure on the Au capacity of Cl- and S-bearing melts has not yet been expanded to the conditions through which natural mantle derived arc magmas ascend through the upper mantle, crust, and eventually into the near surface. High pressures may further promote the formation of Au-chloride and/or Au-sulde species in the melt. The compatibility of Au as lattice bound components in clinopyroxene and olivine, common silicate phases present in basaltic magmatic systems and their upper mantle sources, was constrained to be <0.01 lg g1. If coupled substitution mechanisms indeed aect Au compatibility on the pyroxene lattice, simple static partition coecients may not accurately reect the partitioning behavior of Au between silicate melt and pyroxene and olivine. Following from this study, Au concentration should be strongly enriched in sulde free, H2O undersaturated liquids undergoing open system fractionation of clinopyroxene and olivine. Au partitioning behavior in silicate liquidsuldeoxide aqueous uid systems is a complex function of oxygen fugacity, the fugacities of other volatile components of the melt, and perhaps to a much lesser extent the bulk composition of the liquid. These complexities challenge our ability to model siderophile element behavior in naturally fractionating, volatile bearing basaltic liquids exhibiting open system behavior. More experimental work involving consistent, high precision Au solubility measurements is required before a quantitative model of the Au behavior in natural magmas can be fully developed. Finally, we emphasize that simple fractionation models utilizing static partition coecient values for Au between silicate melt, crystalline silicates, suldes,

and aqueous uids will be complicated by the temporal evolution of the magmatic fS2, fO2, and fHCl.
ACKNOWLEDGEMENTS This work was supported primarily by National Science Foundation Grant EAR 0609550, and additional funding from the UNLV High Pressure Science and Engineering Center (HiPSEC) via the US Department of Energy Cooperative Agreement Nos. DE-FC08-01NV14049 and DE-FC8806NA27684. AB thanks the Society of Economic Geology, Geological Association of Canada/Mineralogical Association of Canada, the UNLV Department of Geoscience and the UNLV Graduate and Professional Student Association for funding assistance during the course of this study. We thank Bruce Marsh for providing the starting dolerite, and John Beckett and Edward Stolper for allowing AB to perform exploratory experiments in their laboratory. We thank James Webster for performing experiments in his IHPV apparatus for us and Chris Heinrich for providing access to the LA-ICP-MS lab at ETH and Jung Hun Seo for his assistance with LA-ICP-MS analyses. The manuscript beneted signicantly from the careful, thorough reviews by Jim Mungall, Roman Botcharnikov, Dick Henley and an anonymous reviewer that improved the paper both scientically and stylistically.

APPENDIX A. SUPPLEMENTARY DATA Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.gca.2010. 12.022.
REFERENCES Aranovich L. Y. and Newton R. C. (1996) H2O activity in concentrated NaCl solutions at high pressures and temperatures measured by the brucite-periclase equilibrium. Contrib. Mineral. Petr. 125, 200212. Audetat A., Gunther D. and Heinrich C. (1998) Formation of a magmatichydrothermal ore deposit: insights with LA-ICP-MS analysis of uid inclusions. Science 279, 2091. Bell A., Simon A. and Guillong M. (2009) Experimental constraints on Pt, Pd and Au partitioning and fractionation in silicate meltsuldeoxideaqueous uid systems at 800 C, 150 MPa and variable sulfur fugacity. Geochim. Cosmochim. Acta 73, 57785792. Berndt J., Koepke J. and Holtz F. (2005) An experimental investigation of the inuence of water and oxygen fugacity on dierentiation of MORB at 200 MPa. J. Petrol. 46, 135. Blaine F. A., Linnen R. L., Holtz F. and Brugmann G. E. (2005) Platinum solubility in a haplobasaltic melt at 1250 C and 0.2 GPa: the eect of water content and oxygen fugacity. Geochim. Cosmochim. Acta 69, 12651273. Borisov A. and Palme H. (1996) Experimental determination of the solubility of Au in silicate melts. Mineral. Petrol. 56, 297 312. Borisov A. and Palme H. (2000) Solubilities of noble metals in Fecontaining silicate melts as derived from experiments in Fe-free systems. Am. Mineral. 85, 1665. Botcharnikov R. E., Linnen R. L. and Holtz R. (available online) Solubility of Au in Cl and S-bearing hydrous silicate melts. Geochim. Cosmochim. Acta 74, 23962411. Boudreau A. and Simon A. (2007) Crystallization and degassing in the Basement Sill, McMurdo Dry Valleys, Antarctica. J. Petrol. 48, 13691386.

1732

A.S. Bell et al. / Geochimica et Cosmochimica Acta 75 (2011) 17181732 Morgan G. and London D. (2005) Eect of current density on the electron microprobe analysis of alkali aluminosilicate glasses. Am. Mineral. 90, 1131. Mungall J. E. (2002a) Roasting the mantle: Slab melting and the genesis of major Au and Au-rich Cu deposits. Geology 30, 915 918. Mungall J. E. (2002b) Empirical models relating viscosity and tracer diusion in magmatic silicate melts. Geochim. Cosmochim. Acta 66, 125143. Mustard R., Ulrich T., Kamenetsky V. and Mernagh T. (2006) Gold and metal enrichment in natural granitic melts during fractional crystallization. Geology 34, 85. Naldrett A. and Duke J. (1980) Platinum metals magmatic sulde ores. Science 208, 1417. ONeill H., Dingwell D., Borisov A., Spettel B. and Palme H. (1995) Experimental petrochemistry of some highly siderophile elements at high temperatures, and some implications for core formation and the mantles early history. Chem. Geol. 120, 255 273. Popp R., Nagy K. and Hajash A. (1984) Semiquantitative control of hydrogen fugacity in rapid-quench hydrothermal vessels. Am. Mineral. 69, 557562. Righter K., Humayun M. and Danielson L. (2008) Partitioning of palladium at high pressures and temperatures during core formation. Nat. Geosci. 1, 321. Ringwood A. (1966) Chemical evolution of the terrestrial planets. Geochim. Cosmochim. Acta 30, 41104. Schuessler J. A., Botcharnikov R. E., Behrens H., Misiti V. and Freda C. (2008) Oxidation state of iron in hydrous phonotephritic melts. Am. Mineral. 93, 14931504. Simon A. C., Frank M. R., Pettke T., Candela P. A., Piccoli P. M. and Heinrich C. A. (2005) Gold partitioning in melt-vaporbrine systems. Geochim. Cosmochim. Acta 69, 33213335. Simon A., Pettke T., Candela P., Piccoli P. and Heinrich C. (2003) Experimental determination of Au solubility in rhyolite melt and magnetite: constraints on magmatic Au budgets. Am. Mineral. 88, 1644. Simon A. C., Pettke T., Candela P. A., Piccoli P. M. and Heinrich C. A. (2004) Magnetite solubility and iron transport in magmatic-hydrothermal environments. Geochim. Cosmochim. Acta 68, 49054914. Simon A. C., Pettke T., Candela P. A., Piccoli P. M. and Heinrich C. (2006) Copper partitioning in sulfur bearing magmatic systems. Geochim. Cosmochim. Acta 70, 55835600. Simon A. C., Pettke T., Candela P. A., Piccoli P. M. and Heinrich C. (2007) The partitioning behavior of As and Au in a haplogranite - vapor at magmatic conditions in sulfur-free and sulfur-bearing systems. Geochim. Cosmochim. Acta 71, 1764 1782. Simon A., Pettke T., Candela P., Piccoli P. and Heinrich C. (2008) The partitioning behavior of Ag in a vapor-brine-rhyolite melt assemblage. Geochim. Cosmochim. Acta 72, 16381659. Webster J. D. (1992) Fluid-melt interactions involving Cl-rich granites: Experimental study from 2 to 8 kbar. Geochim. Cosmochim. Acta 56, 679687. Webster J. D. and Mandeville C. W. (2007) Fluid immiscibility in volcanic systems. In Fluid-Fluid Equilibria in the Crust (eds. A. Leibscher and C. Heinrich). Reviews in Mineralogy and Geochemistry, vol. 65, pp. 313362. Zajacz Z., Candela P. and Piccoli P. (2010) Hyperlink andesite melt-volatile equilibria and its implications for the formation of magmatic-hydrothermal ore deposits. Geological Society of America Abstracts with Programs, vol. 42, p. 46. Associate editor: Edward M. Ripley

Brenan J. M., McDonough W. F. and Ash R. (2005) An experimental study of the solubility and partitioning of iridium, osmium and gold between olivine and silicate melt. Earth Planet. Sci. Lett. 237, 855872. Burnham C. W. (1979) Magmas and hydrothermal uids. In Geochemistry of Hydrothermal Ore Deposits (ed. H. L. Barnes). Wiley, New York, pp. 71136. Candela P. A. (1991) Physics of aqueous phase evolution in plutonic environments. Am. Mineral. 76, 10811092. Candela P. A. and Piccoli P.M. (2005) Magmatic processes in the development of porphyry-type ore systems. In Economic Geology and the Bulletin of the Society of Economic Geologists 100th Anniversary Volume (eds. J. W. Hedenquist, J. F. H. Thompson, R. J. Goldfarb and J. P. Richards), pp. 2538. Cottrell E. and Walker D. (2006) Constraints on core formation from Pt partitioning in mac silicate liquids at high temperatures. Geochim. Cosmochim. Acta 70, 15651580. Driesner T. and Heinrich C. (2007) The system H2ONaCl. Part I: Correlation formulae for phase relations in temperature pressurecomposition space from 0 to 1000 C, 0 to 5000bar, and 0 to 1 XNaCl. Geochim. Cosmochim. Acta 71, 48804901. Ertel W., ONeill H. S. C., Sylvester P. J. and Dingwell D. B. (1999) Solubilities of Pt and Rh in a haplobasaltic silicate melt at 1300 C. Geochim. Cosmochim. Acta 63, 24392449. Ertel W., Walter M. J., Drake M. J. and Sylvester P. J. (2006) Experimental study of platinum solubility in silicate melt to 14 GPa and 2273 K: implications for accretion and core formation in Earth. Geochim. Cosmochim. Acta 70, 25912602. Frank M., Candela P., Piccoli P. and Glascock M. (2002) Gold solubility, speciation, and partitioning as a function of HCl in the brine-silicate melt-metallic gold system at 800 C and 100 MPa. Geochim. Cosmochim. Acta 66, 37193732. Gaillard F., Scaillet B. and Pichavant M. (2002) Kinetics of iron oxidation-reduction in hydrous silicic melts. Am. Mineral. 87, 829837. Guillong M. and Heinrich C. (2007) Sensitivity enhancement in laser ablation ICP-MS using small amounts of hydrogen in the carrier gas. J. Anal. At. Spectrom. 22, 14881494. Guillong M., Meier D. M., Allan M. M., Heinrich C. A. and Yardley B. (2008) SILLS: A MATLAB-Based Program for the Reduction of Laser Ablation ICPMS Data of Homogeneous Materials and Inclusions: Mineralogical Association of Canada Short Course 40, Vancouver, B.C., pp. 328333. . Hedenquist J. and Lowenstern J. (1994) The role of magmas in the formation of hydrothermal ore deposits. Hersum T., Marsh B. and Simon A. (2007) Dolerite partial melting of granitic country rock, melt segregation, and re-injection as dikes, McMurdo Dry Valleys, Antarctica. J. Petrol. 48, 21252148. Jego S., Pichavant M. and Mavrogenes J. A. (2010) Controls on gold solubility in arc magmas: An experimental study at 1000 C and 4 kbar. Geochim. Cosmochim. Acta 74, 21652189. Jugo P., Candela P. and Piccoli P. (1999) Magmatic suldes and Au: Cu ratios in porphyry deposits: an experimental study of copper and gold partitioning at 8508 C, 100 MPa in a haplogranitic meltpyrrhotiteintermediate solid solutiongold metal assemblage, at gas saturation. Lithos 46, 573589. Jugo P., Luth R. and Richards J. (2005) Experimental data on the speciation of sulfur as a function of oxygen fugacity in basaltic melts. Geochim. Cosmochim. Acta 69, 497503. Matthews W., Linnen R. L. and Guo Q. (2003) A ller-rod technique for controlling redox conditions in cold-seal pressure vessels. Am. Mineral. 88, 701707. Mavrogenes J. A. and ONeill H. S. C. (1999) The relative eects of pressure, temperature and oxygen fugacity on the solubility of sulde in mac magmas. Geochim. Cosmochim. Acta 63, 1173 1180.