Main

We studied minerals in 22 spinel peridotite samples, dredged from seven locations along the central Indian ridge (CIR) and its fracture zones. These samples are petrographically residual, lacking plagioclase and crosscutting magmatic veins. All samples are harzburgites except for one orthopyroxene-bearing dunite and two lherzolites. Clinopyroxenes (cpxs) have rare-earth-element patterns that are depleted in light rare-earth elements (LREEs), and all cover the most depleted range of the global abyssal peridotite spectrum3,4. Whereas the heavy rare-earth elements (HREEs) have a limited range in concentration (∼2–10 times chondritic), the LREE abundances vary by more than an order of magnitude (Table 1).

Table 1 REE concentrations in cpx and spinel Cr numbers of CIR peridotites
Full size table

To evaluate the relationships between major and trace elements, we selected all published global abyssal peridotites for which both mineral major-element and cpx trace-element data are available. This includes samples from three locations on the southwest Indian ridge (SWIR) and two on the American–Antarctic ridge (AAR)3,4. In addition, we used data from drill cores at Hess deep (East Pacific Rise)5 and the MARK area (Mid-Atlantic Ridge near Kane fracture zone)6, giving a reasonable sampling of global abyssal peridotite occurrences.

First we tested which trace elements correlate with major-element indicators of partial melting. The most common of these is the chromium number (Cr#) in spinel7,8 (Fig. 1). (Cr# is the molar Cr/(Cr + Al) ratio.) In the CIR peridotites spinel Cr# values range from 0.17 to 0.57, which is 80% of the global abyssal peridotite spectrum. There is a well defined correlation between moderately incompatible elements, such as HREEs dysprosium, erbium and ytterbium, with spinel Cr# (Fig. 1). All elements were fitted with an exponential line, as the fractional melting equation is an exponential function. Fits were poor (see r2 values in Fig. 1) for highly incompatible elements (for example, Ce, Sm, Ce/Yb) and good for less incompatible elements (for example, Dy, Yb). Only highly incompatible elements, it seems, are highly decoupled from major-element mineral chemistry. Cerium, the most incompatible element in this selection, shows the largest variation at a given Cr#. Both strong Ce depletion at low Cr# and large variations towards higher concentrations at higher Cr# suggest that analytical uncertainty at low concentrations is not the only responsible factor. A variety of processes can explain LREE enrichments in mantle rocks9,10,11,12,13.

Figure 1: Mineral major- and trace-element systematics in residual abyssal peridotites.
figure 1

a, Cr number (Cr#) in spinel versus (Ce/Yb)N in cpx. (Subscript N indicates normalization to chondritic values.) Theoretical melting models predict that these parameters are negatively correlated, but they are in fact highly decoupled. b-f, Plots of Cr# in spinel versus incompatible element concentration in coexisting cpx CeN (b), SmN (c), Ti (µg g-1) (d), DyN (e) and YbN (f) show progressively tighter correlations expressed by the correlation coefficient (r2). Numbers in parentheses are number of locations for that portion of mid-ocean ridge. All points are sample averages. Data from refs 3–6, 23 and Table 1; chondrite normalizing values from ref. 24.

Full size image

Correlation coefficients calculated between trace elements and three major-element melting indicators (Cr# in spinel, Al2O3 and Mg number (Mg#) in cpx) confirm the visual impression of Fig. 1. (Mg# is the molar [Mg/(Mg + Fe)] ratio.) This is summarized in Fig. 2, where the experimentally determined cpx/melt distribution coefficient (Kcpx/ld) of each measured trace element14 is plotted versus r2 for the correlation with spinel Cr#. The HREEs (Dy, Er, Yb) in residual cpx are well correlated with spinel Cr# (r2 ≈ 0.9) and the aluminium content of cpx (r2 ≈ 0.75, for n = 57). Correlations with middle rare-earth elements (MREEs) (Sm, Eu) and Ti are markedly lower but still significant. LREEs (La, Ce and Nd), Sr and Zr are only weakly or not at all correlated. The Mg# of residual silicate phases is often regarded as a powerful monitor of melting, which increases with increasing degree of melting. However, Fig. 2 shows that Mg# in cpx does not correlate significantly with other melting indicators. This is probably due to its small absolute range in residual oceanic mantle silicates (between 0.90 and 0.93 for cpx), and is thus largely analytical.

Figure 2: Plot of r 2 versus the cpx/liquid distribution coefficient.
figure 2

Values for r 2 were obtained by plotting melting indicators (spinel Cr#, cpx Al2O3 content, and cpx Mg number, Mg#) versus the logarithmic concentration of each trace element. All values of Kcpx/ld are from ref. 14, except Sr (ref. 25). HREEs are well correlated with spinel Cr#, and trace elements become progressively decoupled from spinel Cr# with increasing incompatibility. There is no significant correlation between cpx Mg# and any of the trace-element concentrations.

Full size image

The good correlation between the major elements and the moderately incompatible trace elements (MREE and HREE) indicates that both are highly useful tracers of partial melting in the mantle. Furthermore, the presence and absence of correlations provide us with a guide on how to interpret abyssal peridotite mineral data. Considering that REE abundances in cpx can be used for quantitative melting modelling3, the good correlation of spinel Cr# with HREE concentration in cpx enables us to develop an empirical equation that describes the extent of melting (F) as a function of spinel Cr#. As a limiting case, we use pure fractional melting in the spinel stability field to calculate the degree of melting after starting conditions given in ref. 3, except for a lower initial cpx mode of 0.14. Mineral/melt partition coefficients were taken from ref. 14. The measured concentrations of Dy, Er and Yb each yield a degree of melting F, which is plotted versus the Cr# of the associated spinel (Fig. 3). A logarithmic fit to all these calculated points yields an equation for the degree of melting (in per cent) as a function of the spinel Cr#: F = 10 ln(Cr#) + 24, calibrated for spinel Cr# values between 0.10 and 0.60 (thick line in Fig. 3). A similar equation has been derived for mantle rocks in Bay of Islands ophiolite15.

Figure 3: Calculated degree of melting versus measured Cr# of spinel.
figure 3

Degrees of melting are calculated from concentrations of Dy, Er and Yb in cpx, yielding three independent degrees of melting for each spinel Cr#. All individual model points are only shown for pure fractional melting (open symbols). The exponential fit (thick curve) is based on average degree of melting for each sample. The same procedure yields melting trajectories for batch melting (dashed curve) and critical melting with 1% residual melt porosity (thin curve) using equations after ref. 26. The degree of fractional melting F (in per cent) as a function of Cr# can be expressed as F = 10 ln(Cr#) + 24 for Cr# values between 0.1 and 0.6. Most and least depleted Cr# compositions from the MARK area are shown. Experimental data shown agree with the melting trend for cpx-bearing residues shown as filled diamonds16.

Full size image

In order to assess how robust this equation is, we applied different melting models using otherwise identical starting parameters. The average model melting curve assuming batch melting (dashed line) and critical melting with 1% residual melt porosity (thin line) are also shown in Fig. 3 (the model data points themselves are not shown). The small difference between pure fractional and critical melting with a relatively high melt porosity confirms that the moderately incompatible HREEs are not sensitive to changes in melt porosity within the melting column, as long as cpx remains in the residue. We note that Fig. 3 by itself does not indicate whether melt extraction is by fractional or batch mechanisms.

As an independent test, spinel Cr#–liquid fraction pairs from 10-kbar batch melting experiments16 are plotted as filled symbols in Fig. 3. These are consistent with our calculated melting curve as long as cpx is present in the residue. By treating major elements chromium and aluminium as trace elements, spinel compositions can be modelled directly to a first approximation (partition coefficients are calculated from the melting experiments of ref. 16). Batch and fractional melting models yield similar spinel compositions, both of which agree with our results. By contrast, in our models F is calculated from incompatible trace elements Dy, Er and Yb in cpx, which show considerable differences between fractional and batch mechanisms (Fig. 3). The experimental data plot above our fractional melting curve at low F. This is because the starting composition chosen for this particular experiment is relatively rich in chromium, resulting in a significantly higher initial spinel Cr# than observed in the most fertile abyssal peridotites, and in a melting trend that is shifted towards higher Cr# values at a given melt fraction.

Using both major- and trace-element melting indicators, some inferences can begin to be drawn regarding the geodynamic controls on melting globally. First, if only non-hotspot peridotites are considered, the degree of melting appears to correlate with spreading rate17. Although moderately incompatible trace elements confirm this inference, this relationship must be treated with great caution. Hess deep and the MARK area are the only locations on their respective ridges yet investigated for both major and trace elements, and the comparatively well investigated SWIR, AAR and CIR show large compositional variation. Therefore, regional-scale ‘forcing functions’ deserve more attention.

A geodynamic control on the degree of melting, one which is superimposed on the effect of the spreading rate, is the vicinity to a transform offset. Most abyssal peridotites are collected at fracture zones. Mantle rocks from Hess deep, on the other hand, were generated in the centre of a spreading segment far from a transform fault5, which may explain their extreme depletion. In our database, only three other locations are not situated directly at a major fracture zone (MARK area, Green Rock hill on CIR 25° S, and CIRCE 93 samples on CIR axis 12° S). The last two are among the most depleted samples from the CIR, indicating that all other CIR fracture zone samples may have been affected by a transform fault effect18,19. Our quantitative tool of Fig. 3 can now be used to estimate the degree of melting for oceanic mantle rocks where often only spinel data are available. The MARK area is of particular importance, because a large set of mantle melting residues contains evidence for this transform fault effect20. Peridotites directly at Kane fracture zone have low Cr# values of around 0.17, a value that systematically increases to 0.41 in highly depleted residues near the ridge centre (Fig. 3). Applying our melting function under fractional melting conditions, we estimate the minimal magnitude of the transform fault effect. The most depleted peridotite corresponds to 15% melting and the most fertile sample to 5%. Thus, the minimal difference in the degree of melting, which may be caused by deepening of the final depth of melting towards the transform fault, is 10% (that is, a factor of three).

Further, the vicinity of a hotspot is believed to increase the degree of melting. The strong trace-element depletion in the Bouvet fracture zone peridotites has been attributed3 to this phenomenon. Comparing these depleted Bouvet fracture zone samples to the fertile Islas Orcadas fracture zone peridotites away from the Bouvet hotspot provides an estimate of the additional hotspot-imposed melting, because of the identical spreading rate at the SWIR and the location at a fracture zone. For the SWIR, the additional degree of melting related to the vicinity of the Bouvet hotspot is 8%. However, the Bouvet fracture zone mantle rocks are not uniquely depleted. Harzburgites from Green Rock hill have nearly identical major- and trace-element compositions to the Bouvet samples. Yet, no nearby hotspot has been detected in the area of Rodrigues triple junction. It therefore seems that the geodynamic combination of near-hotspot, transform fault and slow-spreading ridge can produce similar residual mantle rocks as a hotspot-free, non-transform, intermediate-spreading setting. However, if the global abyssal peridotite data set is filtered for samples from hotspots and transform faults, peridotite compositions from the few locations left do show a correlation with spreading rate21. Mantle temperature variations should in principle also produce strong effects on the degree of melting in the peridotite residues22.

Another question is whether the observed correlations shown in Fig. 1 may also be inherited from the mantle source before melting. The calculations that are based on these observations assume that the premelting Cr# values and HREE concentrations are uniform. It is also possible that the two parameters were already similarly correlated before the most recent (mid-ocean-ridge basalt) melting event. If this were true, we would expect that the calculated degree of melting should not be correlated with tectonic parameters such as spreading rate. Further, this would imply that the systematic depletion trend of the MARK area peridotites20 is fortuitous and not directly related to a transform fault effect. Such inherited depletion might be correlated with abundances of radiogenic isotopes. Future studies that focus on the relationship between melting and regional tectonics will help resolve this issue.