The origin of the negative niobium tantalum anomaly in subduction zone magmas

Abstract

In order to determine the distribution of niobium and tantalum between clinopyroxenes and aqueous fluids, we measured in two separate sets of experiments niobium and tantalum solubility in clinopyroxenes and in aqueous fluids, respectively. The solubility of niobium in clinopyroxenes was experimentally investigated in the system CaMgSi₂O₆–NaAlSi₂O₆–Nb₂O₅–H₂O at 1.5 GPa and 700–1100 °C using piston-cylinder experiments. In the presence of excess Nb₂O₅, CaNb₂O₆ coexists with a clinopyroxene. The solubility of niobium in the pyroxene increases drastically with the aluminum content. While the solubility of niobium is in the order of 100–300 ppm by weight for Al-poor clinopyroxenes, it reaches 4 wt.% for clinopyroxenes containing 10 wt.% Al₂O₃. Microprobe analyses suggest that Nb is incorporated in the clinopyroxene as NaNbAl₂O₆ component. The solubility of CaNb₂O₆ in aqueous fluid was determined by the direct visual observation of the dissolution of CaNb₂O₆ crystals in an aqueous fluid using an externally-heated diamond anvil cell. At 1.5 to 1.7 GPa and 800–1000 °C, an aqueous fluid saturated with diopside dissolves only 20–100 ppm by weight of CaNb₂O₆, i.e. niobium solubility in the fluid is orders of magnitude below the solubility in aluminous clinopyroxenes. Experiments on the solubility of CaTa₂O₆ in clinopyroxenes and in aqueous fluid suggest that Ta behaves generally similar to Nb, but with the notable exception that the solubility of CaTa₂O₆ in aqueous fluids and aluminous pyroxenes is approximately a factor of five lower than the solubility of the corresponding niobium compound. Our results imply that (1) the fluid/clinopyroxene partition coefficient for Nb and Ta is between 0.1 and 0.001 for aqueous fluids containing little dissolved silicates and for clinopyroxenes of a composition realistic for the subducted slab or the subarc mantle. Only for very Al-poor clinopyroxenes may the partition coefficient approach unity. (2) The depletion of Nb and Ta in subduction zone magmas and fluids is related to the intrinsically low solubility of Nb and Ta in water-rich and solute-poor fluids in the shallow parts of a subduction zone. (3) The presence of rutile in the subducted slab is not a necessary requirement for the development of the negative Nb and Ta anomaly. (4) Fluid transport of niobium and tantalum will affect the Nb/Ta ratio in the subarc mantle only under exceptional circumstances. (5) Previous reports of high fluid/clinopyroxene partition coefficients for Nb and Ta are probably related to the Al-free or Al-poor composition of the systems studied.

Introduction

The negative niobium tantalum anomaly, i.e. the relative depletion of Nb and Ta as compared to other highly incompatible elements, is probably the most prominent geochemical feature of magmas produced in subduction zones (e.g. Pearce, 1982). While many other incompatible trace elements are enriched by orders of magnitude in these magmas compared to N-MORB, the concentrations of Nb and Ta are comparable to or sometimes even lower than in N-MORB (Stolz et al., 1996). Numerous models have been proposed to explain this relative depletion of Nb and Ta (e.g. (Kelemen et al., 2004) and references therein). The most popular models invoke either (i) melt formation in the mantle wedge above the subducting slab in the presence of rutile, which retains Nb and Ta or (ii) melting in the mantle wedge following a selective enrichment of incompatible elements due to hydrous fluids, which causes a relative depletion of Nb and Ta, as these elements may not be easily transported by fluids. The absence of rutile in normal mantle peridotite and the rather high solubility of rutile in basalts (Ryerson and Watson, 1987) make hypothesis (i) unlikely. Therefore, it is widely believed that the trace element abundance pattern in arc magmas largely reflects the relative mobility of incompatible trace elements in aqueous fluids, with large ionic lithophile elements (LILE) being very mobile, while high-field strength elements (HFSE) are much less mobile and therefore relatively depleted in the source region of most arc magmas in the mantle wedge (e.g. Tatsumi et al., 1986: McCulloch and Gamble, 1991: Hawkesworth et al., 1994, Arculus, 1994). Indeed, Münker et al. (2004) have recently shown that Nb and Ta may be enriched by fluid metasomatism in the mantle wedge only to a very small extent.

Experimental studies have shown that rutile strongly retains Nb and Ta in the presence of an aqueous fluid, with fluid/rutile partition coefficients being < 0.01 (Brenan et al., 1994). Since rutile is a common accessory mineral in eclogites, it is plausible to assume that fluids released from a subducted slab will be depleted in Nb, Ta and other HFSE elements. Therefore, the negative Nb Ta anomaly is often simply attributed to the presence of rutile in the dehydrating slab, with the understanding that the geochemical signature of the fluids released from the slab remains unchanged when the fluid percolates through the mantle wedge until it reaches the zone of melting. Models of the thermal structure of subduction zones (Davies and Stevenson, 1992) as well as direct evidence from seismic tomography (Wiens and Smith, 2003), however, suggest that the zone of melting is usually located several tens of kilometers above the dehydrating slab. It is rather difficult to see how an aqueous fluid should travel such a long distance through the mantle wedge without reequilibrating with the surrounding mantle peridotite. This problem becomes even more severe in the light of recent experimental studies, which demonstrate that these fluids have very low viscosity (Audétat and Keppler, 2004), potentially combined with low wetting angles at high pressures and high solute contents (Mibe et al., 1999), so that they are likely to infiltrate grain boundaries. Chemical exchange between subduction zone fluids and the mantle wedge has also been suggested based on geochemical arguments. In particular, such an exchange may explain why enrichment of incompatible trace elements does not always directly correlate with radiogenic isotope ratios (Hawkesworth et al., 1994), as one would expect if only material from the subducted slab were transported into the zone of melting.

We suggest that retention of Nb and Ta by rutile in the dehydrating slab is not a sufficient explanation for the negative Nb Ta anomaly in arc magmas. Even if the fluids released from the dehydrating slab were low in Nb and Ta (Becker et al., 2000, Scambelluri and Philippot, 2001), the high-field strength element depletion in the fluid would probably disappear again by scavenging of Nb and Ta from the minerals of the mantle wedge, unless the mantle wedge itself contains minerals that retain Nb and Ta. The ultimate cause of the Nb Ta anomaly must, therefore, be related to the partitioning of trace elements between aqueous fluids and the mantle peridotite.

The partitioning of trace elements between mantle minerals and aqueous fluids has been investigated in several studies (Brenan et al., 1995, Keppler, 1996, Stalder et al., 1998, Kessel et al., 2005). While there is general agreement that large ionic lithophile elements such as Rb or Ba strongly partition into the fluid, some major discrepancies exist with respect to the partitioning of high-field strength elements. While Keppler (1996) suggests that Nb and Ta are depleted in an aqueous fluid in equilibrium with clinopyroxene, both Brenan et al. (1995) and Stalder et al. (1998) report an enrichment of Nb and Ta in fluid in equilibrium with major mantle minerals. Brenan et al. (1995), however, noted that the fluid/clinopyroxene partition coefficient of Nb strongly depends on the aluminum content of the pyroxene. A change in Al content by about 1 wt.% reduces the partition coefficient by two orders of magnitude, and for pyroxenes with about 1.5 wt.% Al₂O₃ (the maximum alumina content in the clinopyroxenes studied by Brenan et al.) the partition coefficient is close to unity. Extrapolating these data to higher alumina contents representative of upper mantle pyroxenes would suggest that Nb becomes compatible in the clinopyroxene and strongly depleted in the aqueous fluid.

High solubilities of high-field strength elements in aqueous fluids at high pressure were suggested by early studies of rutile solubility, (Ayers and Watson, 1993) which appeared to imply solubilities in the order of 2 wt.% at 1–2 GPa and temperatures around 1000 °C. However, there are two recent independent studies (Audétat and Keppler, 2005, Tropper and Manning, 2005) which demonstrate that rutile solubility is two to three orders of magnitude lower than previously reported. This suggests that the solubility of other high-field strength elements in fluids may also be intrinsically low.

Among the major constituents of the upper mantle, clinopyroxene is the most likely host of high-field strength elements, while the contents of incompatible trace elements in both orthopyroxene and olivine are usually orders of magnitude lower (Green, 1994, Eggins et al., 1998). Garnet may contain a noticeable fraction of high-field strength elements, but its importance is probably subordinate compared to clinopyroxene. Amphibole could be another important host of Nb and Ta (Ionov and Hofmann, 1995) but it is not a common phase in mantle peridotites.

In this study, we reexamine the partitioning of Nb and Ta between aqueous fluids and clinopyroxenes with particular attention to the effect of alumina on the fluid/mineral partition coefficients of these elements. Moreover, we use an experimental approach that is radically different from all previous studies in order to overcome some of the inherent difficulties in measuring trace element partitioning between fluids and minerals.

Equilibrium in the trace element distribution between minerals and fluids is not easy to achieve. While in mineral-melt partition experiments the mineral of interest can be conveniently crystallized from the melt by slow cooling, this is generally not possible with fluids. Accordingly, attainment of equilibrium would require diffusion of high-field strength elements through the mineral of interest. The very low diffusion coefficients of these elements in silicates make it rather unlikely that equilibrium could be reached in such a way. One conceivable solution to this problem is the use of glasses or other amorphous substances as starting materials (Stalder et al., 1998, Kessel et al., 2005). The mineral of interest grows during the experiment at the expense of the glassy starting material in the presence of fluid and an equilibrium distribution of trace elements between fluid and mineral may ultimately be achieved. However, the glassy starting materials are thermodynamically highly unstable under run conditions, i.e. they have much higher free energies and much higher solubilities in the fluid than the stable minerals. Accordingly, the composition of the fluid initially generated in these experiments may be far away from equilibrium.

Analyzing the equilibrium fluid composition is another problem in these studies. In general, some material will precipitate from the fluid during quenching from high-pressure, high-temperature conditions so that the fluid cannot be simply withdrawn from the capsule and directly analyzed after the experiment. One way to solve this problem is to calculate the fluid composition by mass balance from the analyses of the solid phases alone. This is in principle possible, although for high-field strength elements this can lead to serious errors, because these elements may be strongly concentrated in traces of accessory oxide phases (e.g. Brenan et al., 1995) that are unlikely to be detected in the run products. Mass balance calculations that overlook these traces of accessory phases may, therefore, erroneously suggest that most of the high-field strength elements partitioned into the fluid. Accordingly, alternative ways to analyze the fluid have been developed, most notably the diamond-trap technique (Ryabchikov et al., 1989). In this method, the fluid infiltrates during the run into the cavities in a layer of diamond powder, the so-called diamond trap. After the experiment, the entire content of the trap can be analyzed; since this includes material precipitated from the fluid during quenching, a quite realistic analysis of the high-pressure, high-temperature fluid composition is in principle possible. However, this method has its own difficulties. If glassy starting materials were used, the first fluids generated in the experiments will be out of equilibrium. They will nevertheless immediately infiltrate the diamond trap, where they may then precipitate various minerals that are stable under run conditions, possibly including accessory phases rich in Nb and Ta. These phases will then be included in the analyses of the “fluid” despite the fact that they were not stably dissolved in the fluid under run conditions. This problem with the diamond-trap technique could be avoided by using stable minerals as starting materials. However, in this case, as noted above, the slow diffusion of most trace elements through these solid minerals makes the attainment of equilibrium during realistic run durations unlikely. We therefore suspect that several previous studies of the fluid/mineral partitioning of Nb and Ta were affected by one or the other problem mentioned above. In order to overcome these problems, we designed a new method for estimating fluid/mineral partition.

Fig. 1 shows the result of an exploratory piston-cylinder experiment in the system CaMgSi₂O₆–NaAlSi₂O₆–Nb₂O₅–H₂O. The run product consists of clinopyroxene crystals, quenched melt and a few crystals of CaNb₂O₆. This phase was found to be stable over a wide range of temperatures and bulk compositions at 1.5 GPa. In similar experiments with Ta instead of Nb, CaTa₂O₆ formed as the phase containing excess Ta that was not dissolved in melt or clinopyroxene. In order to determine the partitioning of Nb and Ta between clinopyroxene and fluid, we therefore first measured the solubility of Nb and Ta in clinopyroxene in equilibrium with CaNb₂O₆ or CaTa₂O₆. These experiments were carried out by crystallizing the clinopyroxene from a silicate melt in equilibrium with the excess Nb or Ta phase. In a separate set of experiments, we then determined the solubility of CaNb₂O₆ and CaTa₂O₆ in aqueous fluid by direct visual observation of the dissolution of small grains of known size in a diamond anvil cell. In these experiments equilibrium is quickly achieved by complete dissolution of the solid in the fluid; diffusion in the solid phase is not a rate-limiting step. The fluid/clinopyroxene partition coefficient of Nb and Ta is then simply obtained as the ratio of the solubility in the fluid and the solubility in the clinopyroxene. This approach eliminates the main problems in measuring fluid/mineral partition coefficients: equilibrium is quickly achieved and fluid compositions are unambiguously determined by in-situ observation.