Introduction

Efforts over the past 20 years have produced much data concerning upper mantle oxygen fugacity, and although agreement is not universal, a relatively consistent picture is emerging. The relative oxygen fugacity (Δlog fO2 [FMQ]) measured for lithospheric upper mantle is heterogeneous, where trends in oxygen fugacity have been noted to occur with metasomatism (e.g., McCammon et al. 2001), partial melting (e.g., Bryndzia and Wood 1990) and tectonic environment (e.g., Wood et al. 1990; Ballhaus 1993; Parkinson and Arculus 1999).

Calibration of an olivine-orthopyroxene-garnet oxybarometer (Gudmundsson and Wood 1995) has enabled oxygen fugacity determinations to be extended to much greater depths, in particular to 170 km beneath the Fennoscandian Shield (Woodland and Peltonen 1999) and to 220 km within the Kaapvaal craton (Woodland and Koch 2003). While a global picture is still lacking, the general pattern of relative oxygen fugacity (Δlog fO2 [FMQ]) decreasing with depth demonstrated by these studies supports earlier predictions of a reduced upper mantle within the garnet peridotite facies (Wood et al. 1990; Ballhaus and Frost 1994; Ballhaus 1995; Wood et al. 1996).

Studies of mantle xenoliths from the Slave craton have revealed a number of similarities with the Kaapvaal craton, but also a number of differences (Kopylova et al. 1999), including the identification of two segments in the upper part of the Slave mantle with different bulk compositions, representing different degrees of depletion from partial melting (Kopylova and Russell 2000; Kopylova and Caro 2004). A redox profile through the Slave cratonic mantle would therefore provide complementary data for constructing a more global picture of oxygen fugacity variation in the deeper part of the lithospheric mantle, and the opportunity for a more detailed examination of the relation between oxygen fugacity and the degree of depletion.

While the number of studies reporting oxygen fugacities for mantle rocks continues to grow, the nature of what controls oxygen fugacity remains an open question. Canil et al. (1994) examined the relative buffering capacities of different equilibria (involving Fe, C, H or S) based on their estimated mantle abundance, and concluded that no single equilibrium could be considered to dominate overall. A more recent comparison of Fe2O3 and carbon abundance showed that concentrations vary depending on rock type, suggesting that the nature of buffering varies for different regions of the mantle (Luth 1999).

To address all of these issues, we have determined a redox profile for a 130-km-thick mantle segment in the Slave craton that transects three depth facies of peridotite and segments of different bulk compositions. We employed Mössbauer spectroscopy to measure Fe3+ concentrations in mantle garnet and spinel and calculated Δlog fO2 (FMQ) from equilibria involving these accessory minerals. The data demonstrate a strong correlation between depletion and oxygen fugacity in the spinel peridotite facies, but little correlation in the garnet peridotite facies. In addition, our data suggest that iron equilibria have the prevailing capacity to buffer oxygen fugacity in the cratonic peridotitic mantle.

Sample description

Peridotitic mantle xenoliths were collected for this study from the drill core of the Jericho and the 5034 (Gahcho Kue cluster) kimberlite pipes. The kimberlites are located in the northern and southeastern parts of the Slave Craton (northern Canada, Fig. 1 in Kopylova and Caro 2004). The Jericho pipe was dated as Middle Jurassic (172±2 Ma by Rb–Sr and U–Pb geochronology) and the 5034 kimberlite as Middle Cambrian (539±2 Ma by the Rb–Sr method on phlogopite) (Heaman et al., in press). Both kimberlites are diamondiferous and host spectacular suites of fresh mantle xenoliths.

Fig. 1a, b
figure 1

Range of Fe3+/Σ Fe in (a) spinel and (b) garnet in upper mantle xenoliths of the Slave craton. Equilibration temperatures for spinel are based on OW-geotherm estimates, while those for garnet are based on BK–AlBK estimates. Spinels show distinct values depending on rock type due to the depleted nature of spinel peridotite (different rock types are grouped by dashed boxes). Garnet values plot roughly along the literature trend (Woodland and Koch 2003 and references therein), which is indicated by the shaded region. Symbols are as follows: open triangle spinel in N Slave spinel peridotite, open square spinel in N Slave spinel–garnet peridotite, solid square garnet in N Slave spinel–garnet peridotite, solid circle N Slave low-T garnet peridotite, solid diamond N Slave high-T garnet peridotite, solid triangle N Slave pyroxenite and megacrysts, square with plus SE Slave coarse low-T garnet peridotite, square with cross SE Slave deformed low-T garnet peridotite

Full size image

Five types of the Jericho mantle xenoliths (N Slave) were selected for the study: spinel peridotite, low-T spinel–garnet peridotite, low-T garnet peridotite, high-T garnet peridotite, and megacrystalline pyroxenites grading into crystalline intergrowths of megacrysts. The petrology of garnet-bearing rock types and their mineral analyses were reported in Kopylova et al. (1999), and their bulk compositions can be found in Kopylova and Russell (2000). Mineral analyses for the Jericho spinel peridotites that have not been previously published are listed in Table 1. The Gahcho Kue peridotites (SE Slave) are described in Kopylova and Caro (2004). For this study, we selected the peridotite that has coarse and deformed textures. Note, however, that deformed peridotite of the N Slave almost always shows temperatures higher than the conductive geotherm and is therefore classified as high-T peridotite, whereas deformed peridotite of the SE Slave shows temperatures that plot along the conductive geotherm and is therefore classified as low-T peridotite.

Table 1 Mineral compositions of spinel-bearing peridotites from N Slave (Jericho pipe)
Full size table

Experimental procedure

Garnet and spinel concentrates were prepared by hand picking under a binocular microscope. The garnet was treated with an acid wash of diluted HCl and HF, and in the case of strongly altered garnets, an additional wash of strong HF was used. Single grains of garnet and spinel were chosen from each monomineralic concentrate using a binocular microscope with selection criteria based on appearance (transparent and free of alteration), diameter and thickness. Grains were mounted between pieces of cellophane tape and masked with 25-μm-thick Ta foil drilled with a 500-μm hole. Dimensionless absorber thicknesses for garnet were generally between two and three (5–7 mg Fe/cm2) and for spinel between four and six (10–15 mg Fe/cm2).

Mössbauer spectra were recorded at room temperature (293 K) in transmission mode on a constant acceleration Mössbauer spectrometer using either a nominal 1.85-GBq 57Co source in a 6-μm Rh matrix (conventional source) or a nominal 370-MBq 57Co high specific activity source in a 12-μm Rh matrix (point source). The velocity scale was calibrated relative to 25 μm α-Fe foil using the positions certified for National Bureau of Standards (now called the National Institute of Standards and Technology) standard reference material no. 1541; line widths of 0.28 mm/s (conventional source) and 0.36 mm/s (point source) for the outer lines of α-Fe were obtained at room temperature. Further description of the method using a point source (Mössbauer milliprobe) is given in McCammon et al. (1991) and McCammon (1994).

Mössbauer spectra for garnet and spinel were fitted to Lorentzian and Voigt doublets depending on spectral resolution with either one (garnet) or two (spinel) doublets assigned to Fe2+ and one doublet assigned to Fe3+ according to current models in the literature (e.g., Amthauer et al. 1976; Li et al. 2002) using the commercially available fitting program NORMOS written by R.A. Brand (distributed by Wissenschaftliche Elektronik GmbH, Germany) and RECOIL written by K. Lagarec and D. Rancourt (distributed by Intelligent Scientific Applications Inc, Canada). Absorption due to Fe3+ was well resolved in all of the spectra, and values for Fe3+/ΣFe were calculated based on relative area ratios corrected for thickness effects and differences in recoil-free fraction. The latter correction was based on the Debye model with assumed Mössbauer temperatures of 340 and 400 K for garnet Fe2+ and Fe3+, and 330 and 400 K for spinel Fe2+ and Fe3+, respectively (Amthauer et al. 1976; De Grave and Van Alboom 1991). The uncertainties given for the Mössbauer determination of Fe3+/Σ Fe are based on errors from the fit to the given model, and uncertainties in the model itself.

Results and calculation of oxygen fugacity

Results for Fe3+/ΣFe in garnet and spinel are presented in Tables 2 and 3. Calculations of Fe3+/ΣFe from the electron microprobe data based on stoichiometry deviated significantly from these data; hence only Mossbauer determinations of Fe3+/ΣFe were used in the present work.

Table 2 Garnet Fe3+/ΣFe, pressure, temperature and relative oxygen fugacities for garnet-bearing equilibria in Slave peridotites and pyroxenites
Full size table
Table 3 Spinel Fe3+/ΣFe, pressure, temperature and relative oxygen fugacities for spinel-bearing equilibria in N Slave peridotites
Full size table

Variations of Fe3+/ΣFe in spinel and garnet reveal several patterns. Fe3+ concentrations in spinel show a striking contrast between spinel and spinel–garnet peridotite (Fig. 1a), which is likely due to the depleted nature of N Slave spinel peridotite (Kopylova and Russell 2000). Fe3+ concentrations in garnet are correlated with temperature (Fig. 1b), similar to observations already in the literature (indicated by shaded area in Fig. 1b) (Luth et al. 1990; Canil and O’Neill 1996; Woodland and Peltonen 1999; Woodland and Koch 2003), which may be related to both a temperature-dependent partitioning of Fe3+ between minerals (Canil and O’Neill 1996) and an increase in the bulk Fe2O3 concentration with depth (Woodland and Peltonen 1999).

Garnet thermobarometry and oxybarometry

For the garnet peridotite assemblage we used two sets of equilibrium P–Ts that were found to be consistent with the mineralogy and petrology of the Slave peridotite. Both sets of P–T estimates satisfy independent petrological constraints, e.g., potentially diamondiferous samples plot in the diamond stability field, and spinel–garnet peridotites plot around the spinel–garnet transformation curve (Kopylova et al. 1999; Kopylova and Caro 2004). These are temperature and pressure estimates made by (1) the two-pyroxene geothermometer of Brey and Köhler (1990) (BK) combined with the Al-in-opx geobarometer of Brey and Köhler (1990) (AlBK); and (2) the two-pyroxene geothermometer of Finnerty and Boyd (1987) (FB) combined with the Al-in-opx geobarometer of MacGregor (1974) (MG); both are listed in Table 2. The BK temperatures are approximately 50–100°C higher than the FB temperatures, while the two pressure estimates show no systematic variation and their difference is generally < 5 kbar.

We calculated oxygen fugacity using the garnet-olivine-orthopyroxene oxybarometer of Gudmundsson and Wood (1995) (GW), incorporating the typographical correction noted by Woodland and Peltonen (1999). Oxygen fugacities were calculated relative to the FMQ buffer (Δlog fO2 [FMQ]) based on Fe3+/ΣFe Mössbauer determinations in garnet. We calculated relative oxygen fugacities using both sets of P–T estimates (FB–MG and BK–AlBK) and found no systematic variation between the two results (Table 2). For a comparison of relative uncertainty contributions, the typical error of 0.02 in garnet Fe3+/ΣFe for these samples produces a variation in Δlog fO2 of 0.4–0.5 log-bar units, while a temperature uncertainty of 100°C produces a variation in Δlog fO2 of only 0.1 log-bar units. Pressure uncertainty has a slightly larger effect than temperature, where a 3-kbar variation in pressure produces a Δlog fO2 variation of 0.3–0.4 log-bar units.

Spinel thermobarometry and oxybarometry

It is more difficult to assess equilibrium conditions for the spinel–orthopyroxene–olivine paragenesis. As a first approach (1), we calculated temperatures using the olivine-spinel thermometer of O’Neill and Wall (1987) (OW) at pressures consistent with the N Slave geotherm (Kopylova et al. 1999). Using this method all spinel peridotites were determined to have equilibrated at 30 kb, and the spinel–garnet peridotites at 35–40 kb (Table 3), which is consistent with the estimated location of the spinel–garnet transformation curve for the N Slave peridotites (Kopylova et al. 1999). To check for consistency in temperature, we compared the OW temperatures with those determined from two-pyroxene thermometry (FB or BK), again assuming equilibration on the N Slave geotherm, and found that the BK temperatures were generally higher by 20–100°C, while the FB temperatures were lower by 0–80°C. As a second approach (2) we calculated temperature using a two-pyroxene thermometer for lherzolites (FB or BK) at pressures estimated based on the Al-in-opx geobarometer (MG or BK) for garnet-bearing samples, and on the intersection of temperatures with the N Slave geotherm for garnet-free rocks (Table 3). The extrapolation of the N Slave geotherm to garnet-free peridotite assumes a common steady-state thermal regime for the garnet-bearing and the garnet-free mantle, but this is supported by analysis of the surface heat flow and the surface heat generation on the N Slave craton (Russell and Kopylova 1999). As seen in Table 3, however, the uncertainty in pressure is larger than for the garnet-bearing assemblage.

We calculated relative oxygen fugacity using three different formulations of the spinel–olivine–orthopyroxene oxybarometer: (1) Ballhaus et al. (1991) (B); (2) O’Neill and Wall (1987) updated by H.S.C. O’Neill (personal communication) Footnote 1 (xOW) and (3) Wood (1990) (W). Oxygen fugacities were calculated relative to the FMQ buffer (Δlog fO2 [FMQ]) based on Fe3+/ΣFe Mössbauer determinations in spinel. To simplify comparison between different oxybarometers, we based all calculations on PT estimates from the OW geothermometer intersected with the N Slave geotherm. The effect of uncertainties in pressure, temperature and composition on Δlog fO2 is similar for all three oxybarometers, where a temperature uncertainty of 200°C produces a variation in Δlog fO2 of 0.2 log-bar units, and a pressure uncertainty of 10 kbar produces a variation in Δlog f 2 of 0.3–0.4 log-bar units. For comparison, the typical error of 0.02 in spinel Fe3+/ΣFe for these samples produces a variation in Δlog fO2 of 0.2 log-bar units.

The three formulations of the spinel–olivine–orthopyroxene oxybarometer give significantly different estimates (up to two log-bar units different) for relative oxygen fugacity in spinel-bearing peridotites (Table 3). This contrasts with experiments by Wood (1990, 1991) who found agreement using the same three formulations to within one log-bar unit. Also, Gudmundsson and Wood (1995) reported an experiment with co-existing olivine, orthopyroxene, spinel and garnet, where Fe3+/ΣFe could be determined for the latter two phases. We applied the three spinel oxybarometer formulations to their reported data, and found agreement of all Δlog fO2 values to within 0.5 log-bar units of the value determined from the garnet-olivine-orthopyroxene oxybarometer of Gudmundsson and Wood (1995). So why should samples from the Slave mantle yield such widely differing Δlog fO2 values compared to these experiments? The answer probably lies in the vastly different conditions and compositions between the experiments and the Slave mantle samples, including in temperature (1,200 vs. 600–1,000°C), redox conditions (near FMQ vs. at least two log-bar units below) and spinel Cr/(Cr+Al) composition (0–0.6 vs. 0.7–0.9). Indeed, Ballhaus et al. (1991) noted that application of the spinel oxybarometer could lead to problems in the garnet lherzolite and diamond stability fields arising from nonstoichiometry or Cr2+ substitution in spinel. In addition, they advised caution at lower temperatures due to the effects of cation ordering. These uncertainties question the reliability of all spinel oxybarometer formulations for samples from the cratonic mantle (low-temperature, highly reducing, high Cr concentrations in spinel).

In the absence of experimental calibrations under the appropriate conditions, we have therefore evaluated spinel oxybarometer results based on independent information regarding the assemblage. There is strong petrographical and petrological evidence that spinel in the N Slave spinel–garnet peridotite is equilibrated with garnet and other peridotite minerals:

  1. 1.

    Spinel is in textural equilibrium with the peridotitic assemblage. The crystals are large, anhedral and interstitial, typical of primary peridotitic spinel and uncharacteristic of secondary euhedral fine-grained spinel after garnet or serpentinized olivine.

  2. 2.

    The garnet composition and chemical correlation trends between elements in garnet depend on the presence or absence of spinel in the paragenesis as summarised in Kopylova et al. (2000).

  3. 3.

    Compositions of olivine, orthopyroxene and clinopyroxene in spinel-bearing garnet peridotite are systematically different from those in spinel-free garnet peridotite (Kopylova et al. 1999).

  4. 4.

    Temperatures calculated using the Ni concentration in garnet thermometer are lower in spinel-bearing peridotite than in spinel-free peridotite (Kopylova et al. 1999).

  5. 5.

    Two-pyroxene and olivine-garnet temperatures computed for spinel-bearing garnet peridotite are systematically lower then those for spinel-free garnet peridotite (Kopylova et al. 1999).

  6. 6.

    Equilibrium pressures and temperatures for spinel–garnet peridotite (Kopylova et al. 1999) are consistent with the experimentally modelled transition from the spinel to garnet peridotite facies for a rock of a given olivine and spinel composition (O’Neill 1981).

The equilibrium between spinel and garnet in the peridotite assemblage implies that any estimate of relative oxygen fugacity involving spinel should be identical to that involving garnet, which suggests that the lowest set of Δlog fO2 estimates in Table 3 is the most appropriate. We have therefore adopted the O’Neill and Wall (1987) estimates from Table 3 for the discussion below, but we emphasise that it does not imply a general endorsement of the formulation over the others, but simply indicates that the formulation reproduces the oxygen fugacity most closely for these particular conditions.

Whole rock concentrations of Fe3+ and V

Spinel Fe3+ concentrations are significantly different for spinel peridotite compared to spinel–garnet peridotite. We examined how these concentrations relate to the bulk chemistry of the rocks, particularly bulk Fe2O3 and the abundance of V which also records oxygen fugacity (Canil 2002 and references therein).

The bulk Fe2O3 concentration is lower in spinel peridotite than in spinel–garnet peridotite or low-T and high-T garnet peridotite (Table 4). However only a minor portion of Fe3+ resides in primary peridotitic minerals (Canil et al. 1994). We supplemented whole rock Fe2O3 abundance determined by wet chemistry by calculation of total Fe2O3 abundance in primary minerals based on the Fe3+ content of individual minerals, their modal abundances and total Fe concentrations (O’Neill et al. 1993; Canil et al. 1994).

Table 4 Bulk and mineral compositions, mineral modes and estimated Fe2O3 residing in primary minerals in N Slave peridotite
Full size table

Modal abundances and bulk compositions were available from the N Slave dataset of Kopylova and Russell (2000). We used the bulk chemical data averaged for each rock type, Fe3+/ΣFe values for spinel and garnet from Mössbauer spectroscopy, Fe3+/ΣFe values for clinopyroxene based on the olivine-orthopyroxene-clinopyroxene oxybarometer (Luth and Canil 1993) (spinel facies) or garnet-clinopyroxene Fe3+ partitioning (Canil and O’Neill 1996) (garnet facies), and Fe3+/ΣFe values for orthopyroxene based on orthopyroxene-clinopyroxene Fe3+ partitioning in both the spinel and garnet facies (Canil and O’Neill 1996). Following the observations of O’Neill et al. (1993), Canil et al. (1994) and our own unpublished data for mantle olivines, we assumed Fe3+/ΣFe to be zero in olivine for all samples.

The calculated total Fe2O3 abundances averaged for each rock type are listed in Table 4, along with bulk chemical data. Spinel peridotite shows a greater chemical depletion than spinel–garnet peridotite as inferred from higher MgO and Mg-numbers of bulk compositions and minerals as well as lower bulk FeOtotal (Kopylova and Russell 2000). The lower total Fe2O3 in spinel peridotite translates to an even greater reduction of Fe2 O3 concentration in spinel, because Fe3+ is preferentially concentrated in that phase. Hence depletion results in more reducing conditions, where spinel peridotite shows Δlog fO2 (FMQ) values more than one log-bar unit lower than for spinel–garnet peridotite (Table 3).

Spinel peridotite has lower concentrations of V (Table 4 and Fig. 2), but they cannot be taken as evidence for more oxidising conditions during melting in the Archean as proposed by Canil (2002). The low concentrations of bulk V in N Slave peridotite are often found in orthopyroxene-rich peridotite (Fig. 2), but the modes of orthopyroxene in these rocks are too high for simple melt residues (Kopylova and Russell 2000). The peridotite, therefore, cannot be generated by melt depletion, and the whole rock V concentrations, if modelled on such a premise (Canil 2002) leads to erroneous results (Lee et al. 2003).

Fig. 2
figure 2

Variation of V (ppm) with modal orthopyroxene (vol%) for N Slave peridotite. Whole rock V contents were analysed using XRF (Table 1 in EPSL Online Background Dataset, Kopylova and Russell 2000), while orthopyroxene modes were determined by image analysis and recalculation of bulk chemical analyses as described in Kopylova and Russell (2000)

Full size image

Oxygen fugacity of cratonic peridotite

The final Δlog fO2 estimates for spinel-bearing rocks from the Slave mantle calculated using the modified O’Neill and Wall (1987) spinel-orthopyroxene-olivine oxybarometer are plotted as a function of depth in Fig. 3a, and as expected from the significantly different Fe3+/Σ Fespinel ratios in spinel–garnet and spinel peridotites, spinel-garnet peridotite shows equilibration at higher Δlog fO2 than garnet-free spinel peridotite (Fig. 3a). This conclusion is independent of both oxybarometer and thermobarometer choice. To illustrate the variation in Δlog fO2 for different thermobarometer solutions, we plotted two different estimates for each sample in Fig. 3a: (1) based on BK temperatures and AlBK pressures; and (2) based on OW temperatures projected onto the N Slave geotherm. Different thermobarometric estimates for each sample are joined by a line, and illustrate that despite a large uncertainty in pressure (up to 9 kbar), Δlog fO2 estimates for the same sample vary by <0.5 log-bar units. Therefore spinel–garnet peridotite equilibrated at more oxidising conditions than garnet-free spinel peridotite, irrespective of their estimated depth of origin, and independent of which spinel oxybarometer is used for the calculation.

Fig. 3a–c
figure 3

Oxygen fugacity (relative to FMQ) of spinel and garnet peridotites and pyroxenites as a function of depth. a Relative oxygen fugacities estimated using the spinel-orthopyroxene-olivine oxybarometer (modified formulation of O’Neill and Wall 1987) for PT estimates based on (1) the OW thermometer and the projection onto the N Slave geotherm (large symbols) and (2) the BK–AlBK thermobarometer (small symbols). Estimates of Δlog fO2 (FMQ) for the same samples using different PT estimates are connected by tie-lines. b Relative oxygen fugacities estimated using the garnet-orthopyroxene-olivine oxygen oxybarometer (Gudmundsson and Wood 1995) (GW) for PT estimates based on the AK–AlBK thermobarometer. Symbols are the same as for Fig. 1. Outlined P-Δlog fO2 fields for spinel and spinel–garnet peridotite from spinel-based oxybarometry are transferred from Fig. 3a, and the shaded fields highlight the trends for N Slave spinel–garnet peridotite and N Slave garnet peridotite. The graphite-diamond equilibrium (Berman and Simon 1955) is based on the N Slave geotherm (Kopylova et al. 1999). c Relative oxygen fugacities estimated using the GW oxybarometer for all garnet-bearing rocks from the present work (solid circles) compared to values from the literature: large bold outlined field (Lesotho xenoliths, Woodland and Koch 2003); hatched field (South Africa xenoliths, ibid.); small white field (Fennoscandian xenoliths, Woodland and Peltonen 1999 as corrected by A.B. Woodland, personal communication). Buffer curves were calculated based on the N Slave geotherm (Kopylova et al. 1999) from Eggler and Baker (1982) (EMOG/D) and Ballhaus et al. (1991) (IW)

Full size image

The final Δlog fO2 estimates for garnet-bearing rocks from the Slave mantle calculated using the Gudmundsson and Wood (1995) garnet-orthopyroxene-olivine oxybarometer are plotted as a function of depth in Fig. 3b. The Δlog fO2 estimates for spinel–garnet peridotites are close to those calculated for the same rocks using the spinel-based oxybarometer (reproduced as a dashed box in Fig. 3b to facilitate comparison), although as noted in the previous section, the Δlog fO2 values calculated using spinel-based oxybarometry for diamondiferous cratonic spinel–garnet peridotites probably overestimate relative oxygen fugacity.

Garnet peridotites show a general reduction in Δlog fO2 with depth, where Δlog fO2 values of garnet peridotites are lower than those for spinel–garnet peridotites (Fig. 3b). Superimposed on this general trend is a degree of scatter arising from anomalously high- or low-Fe3+ concentrations in garnet for some of the N Slave coarse garnet peridotites. Relative oxygen fugacities of N Slave high-T peridotite also vary, even for samples from the same depth, and show no obvious trend with depth. These variations reflect a larger chemical variance in mineral compositions of pyroxenes and garnet that is also recorded in the large scatter of the BK temperature estimates. The relative oxygen fugacities of N Slave pyroxenites and megacrysts are similar to values for the N Slave high-T peridotites. The trend in relative oxygen fugacities of spinel-free coarse and deformed peridotites from the SE Slave are essentially identical to those from the N Slave. Even though SE Slave peridotites were equilibrated on a lower geotherm (Kopylova and Caro 2004), there is no observable effect on relative oxygen fugacity.

The relative oxygen fugacity estimates of spinel and garnet-bearing peridotites from the Slave mantle are consistent with all redox constraints implied by the sample petrology. Firstly, they are more oxidising than the IW buffer, where Ni is expected to reside in olivine instead of precipitating as metal (O’Neill and Wall 1987). This Δlog fO2 constrains the minimum fugacity and is applicable to peridotite with Ni-bearing olivine. Secondly, all of our relative oxygen fugacity estimates are more reducing than EMOG/EMOD buffer curves at the corresponding pressures (Fig. 3b). This fO2 constrains the maximum fugacity of peridotites equilibrated in the diamond stability field (above 940°C; Kopylova et al. 2000) based on the known presence of harzburgitic diamond in the Jericho and 5034 pipes.

Woodland and Koch (2003) reported relative oxygen fugacities for garnet peridotites from the Kaapvaal craton extending to depths of nearly 220 km, supplementing previous Kaapvaal data of Luth et al. (1990) and Canil and O’Neill (1996) (recalculated by McCammon et al. 2001). Also, Woodland and Peltonen (1999) presented relative oxygen fugacity data for the mantle beneath the Fennoscandian Shield from spinel–garnet and garnet peridotites to a depth of approximately 170 km. This literature data, together with our estimates, define a broad array of Δlog fO2 (FMQ) decreasing from −2 to −4 log-bar units over depths of 110 to nearly 220 km (Fig. 3c).

The reduction in relative oxygen fugacity with depth appears to be generally continuous within the data scatter for all of the garnet-only peridotite datasets. We note that the “kink” in the redox-depth profile reported by Woodland and Koch (2003) arises through their addition of Δlog fO2 data from spinel oxybarometry to the depth profile (their samples Let 19, Let 23 and Liq 9). If these data are omitted and only data from garnet-based oxybarometry are plotted, the trend is relatively linear for all data. This highlights the difference in oxygen fugacity variation between spinel and garnet peridotite, and is discussed further below in the context of redox control within the mantle.

All datasets converge in the deepest region which is dominated by deformed, high-T peridotite samples pervasively altered by asthenospheric melts that show Δlog fO2 values only one log-bar unit greater than the iron-wüstite buffer. The observation of such reducing conditions at these depths supports the growing consensus that the asthenosphere is relatively reduced (Green et al. 1987; Wood et al. 1990; Ballhaus and Frost 1994; Wood et al. 1996), instead of oxidised with a reduced lithosphere floating on top (Haggerty and Tompkins 1983; Haggerty 1986; Haggerty 1999).

The reduced nature of cratonic peridotitic mantle is fully consistent with all mineralogical constraints on oxygen fugacity available for these rocks, including the absence of carbonate in mantle xenoliths. Experiments have demonstrated that carbonate is stable in the presence of MgSiO3 (as orthopyroxene or silicate perovskite) to at least 26 GPa (Katsura and Ito 1990); yet it is diamond and graphite, not carbonate, that are found in mantle peridotites under cratons. The lack of carbonate in mantle xenoliths hence rules out conditions more oxidising than the EMOD buffer (~FMQ-1). The overall reduced affinity of upper mantle minerals hosted by diamond (reviewed in Navon 1999) supports this estimate. Existing data suggests that off-cratonic subcontinental mantle may be reduced to a similar degree as the cratonic mantle (Pearson et al. 1994). Graphite and moissanite have been found in the off-cratonic subcontinental mantle (Lutkov et al. 1988), where relative oxygen fugacities of FMQ-3 to FMQ-1 have been reported (O’Neill and Wall 1987).

Oxygen fugacity and chemical depletion of the mantle

A number of studies have reported a correlation between depletion and oxygen fugacity in spinel-bearing peridotite (e.g., Wood et al. 1990; Bryndzia and Wood 1990; Daniels and Gurney 1991; Kadik 1997), which has been attributed to the slightly incompatible behaviour of Fe3+ during partial melting (e.g., Bryndzia and Wood 1990). Recently, Lee et al. (2003) modelled the redox evolution of residues produced by progressive melt depletion for fractional melting in a system closed to O2 exchange. If the combined O2 content of the melts and solid residue remains constant, Δlog fO2 (FMQ) of the residues steadily decreases by 1.5 log-bar units as melting proceeds from ~5 to ~25%. Even though Δlog fO2 estimates are not available for high degrees of melting typical of cratonic peridotites (20–40%, Walter 1998), it is clear that a strong depletion (>5%) leads to reduced residues.

A correlation between depletion and oxygen fugacity is also observed in the spinel-bearing peridotite data from the N Slave craton (Table 4). The higher bulk MgO content of depleted samples is negatively correlated with bulk Fe2O3 according to the trend reported by Canil et al. (1994) (Fig. 4a). The lower bulk Fe2O3 concentration and amount of Fe2O3 present in primary phases in spinel peridotite leads to a significantly lower relative oxygen fugacity compared to spinel–garnet peridotite (Fig. 4b). This trend is most evident for averaged whole rock values (grey regions in Fig. 4) and less evident for individual samples (data points in Fig. 3). The individual samples are the few for which both whole rock chemistry and Mössbauer analysis were performed, and were not originally chosen to be representative of all samples within the given facies.

Fig. 4
figure 4

Effect of depletion (indicated by bulk MgO) on whole rock Fe2O3 content and relative oxygen fugacity. Grey regions are constructed from average values for N Slave peridotites plus one sigma standard deviations for each of the four peridotite facies (spinel, spinel–garnet, low-T garnet and high-T garnet) (Table 4). Solid circles indicate values for individual samples, while the star refers to fertile garnet peridotite (sample 23-5). Some of the individual analyses lie outside the grey regions because values deviate more than one standard deviation from the average. a Effect of depletion on whole rock Fe2O3 content. The N Slave samples follow the general trend reported by Canil et al. (1994) (dotted region). b Effect of depletion on relative fO2 in the spinel-peridotite facies. The lower Fe2O3 contents in garnet-free spinel peridotite lead to lower relative fO2. c Effect of depletion on relative f O2 in the garnet-peridotite facies. The effect of whole rock Fe2O3 content variations on relative fO2 is small in garnet peridotite as discussed in the text

Full size image

For comparison, we can evaluate the change in relative oxygen fugacity expected in an isochemical mantle with constant Fe2O3 abundance across the spinel–garnet peridotite transition. The change in relative oxygen fugacity depends critically on the partitioning of Fe3+ between garnet and spinel (Ballhaus and Frost 1994). N Slave spinel–garnet peridotites show K gt–spD values for Fe3+ of roughly 0.2, implying that most Fe3+ remains concentrated in spinel, even in the presence of garnet. The reaction between spinel peridotite and spinel-garnet peridotite should therefore be associated with essentially zero change in relative oxygen fugacity (Fig. 3b in Ballhaus and Frost 1994), making the contrast in relative oxygen fugacity between spinel and spinel–garnet peridotite in the N Slave mantle even more striking. There the transition between spinel and spinel–garnet peridotite is not isochemical, but shows a distinct chemical boundary due to a less depleted composition, and the increased abundance of Fe2O3 in spinel results in an increased relative oxygen fugacity in spinel–garnet peridotite.

A correlation between depletion and oxygen fugacity in the spinel peridotite facies is expected to be strong because compositional changes in the residue during partial melting do not work against the effect of depleted Fe2O3 concentrations in reducing oxygen fugacity: (1) the modal abundance of spinel, orthopyroxene and clinopyroxene decreases with increased degrees of partial melting, and since these phases contribute similar amounts to whole rock Fe2 O3 abundance (O’Neill et al. 1993), the tendency to concentrate Fe3+ in spinel with increasing depletion is reduced, and (2) olivine and orthopyroxene are on opposite sides of the oxygen fugacity equilibrium:

$$ 2{\text{Fe}}^{{2 + }} {\mathop {{\text{Fe}}^{{3 + }} _{2} {\text{O}}_{4} }\limits_{{\text{sp}}} } + 6{\mathop {{\text{FeSiO}}_{3} }\limits_{{\text{opx}}} } = 6{\mathop {{\text{Fe}}_{2} {\text{SiO}}_{4} }\limits_{{\text{ol}}} } + {\text{O}}_{2} $$
(1)

(e.g., Wood 1991); hence coupled compositional variations have less effect on oxygen fugacity. The correlation between depletion and oxygen fugacity, however, may be obscured in many spinel peridotite suites (e.g., Parkinson and Arculus 1999; Ionov and Wood 1992) which may represent mixtures of variously depleted residues from different protoliths.

The relation between depletion and oxygen fugacity may be different in the spinel-free garnet peridotite stability field. Fertile garnet peridotite samples have been reported from the N Slave mantle that show bulk chemical patterns distinct from both low-T and high-T garnet peridotite (Fig. 6 in Kopylova and Russell 2000), and are represented in our dataset by sample 23-5. The garnet Fe3+/ΣFe value is similar to those for other high-temperature peridotites (Fig. 1b and Table 2), but the significantly higher modal abundance of garnet in fertile peridotite leads to a higher calculated total Fe2O3 abundance (Table 4 and Fig. 4a). While this higher abundance is consistent with a more fertile composition, it does not necessarily result in a higher oxygen fugacity due to other compositional changes in the system (Fig. 4c). For example, the higher iron concentrations of olivine and orthopyroxene in the more fertile assemblage increase the activities of fayalite and ferrosilite, which then decrease oxygen fugacity through the equilibrium for spinel-free garnet peridotite:

$$ 2{\text{Fe}}\mathop {_{\text{3}}^{{\text{2 + }}} {\text{Fe}}_{\text{2}}^{{\text{3 + }}} {\text{Si}}_3 {\text{O}}_{12} }\limits_{{\text{gt}}} = 4\mathop {{\text{Fe}}_2 {\text{SiO}}_4 }\limits_{{\text{ol}}} + 2\mathop {{\text{FeSiO}}_3 }\limits_{{\text{opx}}} + {\text{O}}_2 $$
(2)

(Gudmundsson and Wood 1995). An Mg-number shift from 0.92 to 0.90 in olivine, for example, decreases Δlog fO2 (FMQ) for sample 23-5 by nearly one log-bar unit.

The pattern in the garnet peridotite facies that depletion does not necessarily translate to a more reduced residue may be relevant on a more global scale. Compositional changes in the residue during partial melting tend to cancel out effects due to depleted Fe2O3 concentrations (which would normally lead to decreased oxygen fugacity), because (1) the modal abundance of garnet (the dominant contribution to whole rock Fe2O3) decreases (Table 4, see also Walter 1998), hence concentrating Fe3+ in the remaining garnet (which increases oxygen fugacity), and (2) olivine and orthopyroxene iron concentrations decrease, hence increasing oxygen fugacity. Pressure also plays a role, since the activity of the skiagite component in garnet is reduced with increasing pressure through the negative ΔV term for reaction 1 (e.g., Wood et al. 1990; Wood et al. 1996). This would reduce oxygen fugacity in deeper, more fertile assemblages compared to shallower, more depleted compositions. To compare the present results with other worldwide occurrences, we combined literature data of garnet peridotite whole rock analyses (Canil et al. 1994) with estimates of their relative oxygen fugacities (McCammon et al. 2001), and found a lack of correlation similar to that illustrated in Fig. 4c.

Control of oxygen fugacity in the cratonic peridotite mantle

Discussion regarding the control of oxygen fugacity in the mantle has been ongoing for more than two decades (see reviews by Arculus 1985; Wood et al. 1990; Luth 1999). If the mantle is assumed to be a closed system with no oxygen exchange with external sources, then oxygen fugacity is an intrinsic quantity that is completely determined by the bulk composition at a given pressure and temperature. The degree to which the system preserves this oxygen fugacity in the event of fluid infiltration or other influx of oxygen reflects the buffering capacity of the different equilibria, which depends in part on the concentration of oxidised species (e.g., Canil et al. 1994; Ballhaus 1995; Luth 1999). An appraisal of the relative buffering capacities of different equilibria (involving Fe, C, H or S) showed that no single equilibrium is considered to dominate overall (Canil et al. 1994), contributing to the generally accepted consensus that factors controlling oxygen fugacity likely vary in different parts of the mantle.

Within the cratonic mantle the control of oxygen fugacity is more unequivocal. Wood et al. (1990) argued that oxygen fugacity should decrease with depth within the garnet peridotite facies based on a highly negative ΔV term for the reaction, assuming that iron equilibria involving Fe3+ in garnet dominate over other equilibria in controlling oxygen fugacity. This was further quantified by Ballhaus (1995) and Wood et al. (1996), establishing the dominance of the pressure effect in reducing oxygen fugacity. These estimates were made before the first fO2-depth profiles in the cratonic mantle were reported (Woodland and Peltonen 1999; Woodland and Koch 2003; this work), which emphasises the important evidence provided by these depth profiles of oxygen fugacity controlled primarily by garnet-based iron equilibria in the cratonic mantle.

Phase transitions involving iron-bearing phases should influence oxygen fugacity in a mantle where redox relations are dominated by iron. The depth variation of oxygen fugacity in the spinel–garnet perodotite facies should therefore be different compared to the garnet peridotite facies, partly due to the different ΔV terms for the redox reactions. The zero-pressure volume change for solids in the spinel–garnet peridotite reaction,

$$ {\text{Fe}}\mathop {_{\text{3}}^{{\text{2 + }}} {\text{Fe}}_{\text{2}}^{{\text{3 + }}} {\text{Si}}_3 {\text{O}}_{12} + }\limits_{{\text{gt}}} \mathop {{\text{Fe}}_3 {\text{O}}_4 }\limits_{{\text{sp}}} + 2\mathop {{\text{FeSiO}}_3 }\limits_{{\text{opx}}} = 5\mathop {{\text{Fe}}_2 {\text{SiO}}_4 }\limits_{{\text{ol}}} + {\text{O}}_2 , $$
(3)

is −17.8 cm3 (relative to FMQ), which is significantly smaller than the corresponding value of −26.4 cm3 for solids in the garnet peridotite reaction (Eq. 2) (based on data from Holland and Powell 1990 and Woodland and O’Neill 1993); hence we would expect a significantly greater reduction in oxygen fugacity with depth within the garnet peridotite facies. This is suggested by N Slave data that show a slope in Δlog fO2 (FMQ) versus pressure of 0.09–0.11 log-bar units/kbar for garnet peridotite, compared to only 0.04 log-bar units/kbar for spinel–garnet peridotite (Fig. 3b). This is further supported by the global dataset which shows a slope of roughly 0.09 log-bar units/kbar for garnet peridotite (Fig. 3b). In a similar way we can rule out a dominant control of oxygen fugacity by pyroxene-based iron equilibria (Luth and Canil 1993) because of the positive volume change relative to the FMQ buffer, implying that relative oxygen fugacity would increase with depth in a mantle dominated by pyroxene-based redox equilibria (Luth 1999).

The oxygen fugacity of the cratonic mantle is therefore related to its chemical and mineralogical boundaries, and is distinct in mantle domains of different composition. This rejects models where oxygen fugacity is controlled externally. However the oxygen fugacity measured from peridotites may not be representative of the redox state in all parts of the cratonic mantle, for example where eclogite is also present. The lower oxygen fugacities of depleted rocks could explain an observed strong preference of elemental carbon for cratonic harzburgite and eclogite. While average Archean mantle contains 85% lherzolite and 15% harzburgite, 85% of peridotitic diamonds are associated with depleted harzburgite (Grutter et al. 2003). Melting and depletion of harzburgites were likely to have occurred in the spinel-depth facies, even though harzburgites are now found at greater depths consistent with the presence of diamond. The shallow melting is suggested by extremely low concentrations of Al in cratonic peridotite, which cannot be achieved by melt extraction in the presence of garnet (Canil and Wei 1992; Kelemen et al. 1998; Canil 2002). Cratonic eclogite also represents the refractory residue produced by extraction of silicic trondjemite-tonalite-granodioritic magmas (Ireland et al. 1994; Barth et al. 2001). The lower oxygen fugacities of depleted rocks could also explain a geographical coincidence and excellent depth correlation between an ultra-depleted harzburgitic layer (Griffin et al. 1999) and a segment with low electromagnetic conductivity (Jones et al. 2001) of the Central Slave craton. The layers are mapped both geochemically and with a magnetotelluric survey down to ~100–150 km depth coincident with the graphite–diamond transformation. The higher abundance of graphite which has been invoked to explain the conductive shallow mantle of the central Slave (Jones et al. 2001) is therefore consistent with the reduced character of ultra-depleted harzburgites.

If the redox state of the mantle is controlled by Fe equilibria, then partial melts should be more oxidised than their spinel-bearing mantle sources. Indeed, MOR basalts have higher oxygen fugacity (up to three log units Δlog fO2) than their parent abyssal peridotites (Wood et al. 1990). A compilation of the most recent data on the redox state of arc mantle and arc magmas show a consistent difference of up to two to three log-units Δlog fO2 (Figs. 1 and 8 in Lee et al. 2003). Thus, the observed redox state of some magmas is consistently lower than that of the mantle protoliths and can constrain only the maximum fO2 of the mantle.

Concluding remarks

The following picture of the cratonic upper mantle has emerged as a result of this study. The subcontinental mantle is a mixture of domains with distinct compositions, and in the case of closed system behaviour, each domain can control its own oxygen fugacity that is pre-determined by bulk composition, pressure and temperature. Predominant peridotitic segments interleave with eclogitic lenses that may have a different oxygen fugacity. Diamonds are stable in both lithologies, and carbonates are unstable in the cratonic peridotitic mantle because of the reduced conditions (Δlog fO2 [FMQ] = −1.5 to −4.5). The cratonic mantle is primarily buffered by the exchange of Fe2+ and Fe3+ between accessory spinel and garnet, and Fe2+ between olivine and pyroxenes; hence there is a change in oxygen fugacity with the disappearance of spinel with increasing depth in the peridotitic mantle. Relative oxygen fugacities of garnet-bearing cratonic peridotite decrease with depth in the range 80–210 km, reaching conditions one log-bar unit above the IW buffer. The reducing conditions within the cratonic mantle may be similar to those for off-cratonic mantle below continents.

Partial melting of the mantle redistributes Fe3+, producing reduced residues and oxidised magmas. The availability of oxygen in the process of partial melting is not necessarily externally imposed, and thus melting cannot always be characterised by its “redox state”. Segments of the mantle that experienced more extensive melting and hence extraction of Fe2O3 are more reduced, where the reduction is more pronounced in spinel peridotite than in garnet peridotite. Elemental carbon favours concentration in these depleted rocks, both harzburgite and eclogite. Small amounts of material representative of this depleted reduced cratonic upper mantle are captured by diamonds as mineral inclusions. Subsequent partial melting of the mantle produces carbonate-rich magmas such as kimberlites, and while fluids associated with this magmatism are more oxidised than complementary reduced cratonic mantle, they can also be trapped in diamond.