Introduction

Olivine is commonly the earliest abundant crystallizing phase in mafic igneous rocks and therefore has the potential to record information about melts closer to the original, unfractionated compositions than other, better studied, phenocryst phases such as clinopyroxene. Mg-rich, mantle-derived melts usually crystallize olivine as the liquidus phase (Melluso et al. 1995; Larsen and Pedersen 2000; Krishnamurthy et al. 2000; Upton et al. 2002), and experiments indicate that the liquidus phase field of olivine expands at low-pressure, near-surface conditions (Ulmer and Sweeney 2002; Canil and Bellis 2008). Olivine may be preceded by Cr-rich spinels in picrites, basalts and other mantle-derived melts (Bell and Williamson 1994; Kamenetsky et al. 1995; Chung et al. 2001), but these have limited stability and abundance due to the rapid decrease of Cr concentrations that they cause in the melts. Furthermore, they may enter a reaction relationship with later melts and so may only be preserved where encased in olivine (Thy 1983; Foley 1985).

Distinguishing early crystallizing olivines from crystals with other origins is particularly important for MgO-rich volcanic rocks such as picrites, kimberlites and olivine lamproites, for which the proportion of olivines crystallized in equilibrium with the erupted melt is debated (Kerr et al. 1996; De Bruiyn et al. 2000; Zhang et al. 2008). The proportion of exotic olivines is important for deciphering the possible range of primary lamproitic melts, as olivine-rich lamproites with 42–45 wt% SiO2 may be either primitive (Jaques et al. 1984; Foley 1993) or represent silica-richer primary melts carrying extra olivine (Mitchell 1995). Kimberlitic olivines are thought to be a mixture of phenocrysts, mantle xenocrysts and “macrocrysts” of uncertain origin (Skinner and Clement 1979; Hayman et al. 2008; Brett et al. 2009), so that the nature and composition of primary kimberlite melts continues to be hotly debated, with opinions ranging widely from carbonatitic to H2O-rich but relatively carbonate-poor silicate melts (Canil and Bellis 2008; Kjarsgaard et al. 2009). The distribution of trace elements in the various olivine populations may prove to be the key to resolving these questions.

Despite their potential importance for elucidating the early crystallization history and origin and olivine populations, trace elements in olivine have rarely been studied systematically due to the generally lower concentrations of most trace elements than in other common minerals, many of them close to detection limits for routine analysis even with Laser-ICP-MS. Exceptions are a few compatible transition elements, amongst which Ni, Cr and Mn are sufficiently abundant that useful analyses may be obtained by electron microprobe using high sample currents and long counting times (Sobolev et al. 2005, 2008; Kamenetsky et al. 2008). However, these elements give petrogenetic information that differs from that usually sought from incompatible trace elements.

The ugandites of the western branch of the East African Rift are olivine-rich kalsilite-bearing leucitites containing various sizes of olivines (Tappe et al. 2003; Rosenthal et al. 2009). The high MgO contents of up to 22 wt% in the whole rocks indicate that many of these olivines may be accumulated phenocrysts, and some possibly xenocrysts. Olivines with these possible diverse origins and ages were therefore analysed in this study for a wide range of trace element concentrations in order to characterize trace elements in olivines in this type of alkaline mafic rock, and also to investigate the range of trace element patterns present. The Laser-ICP-MS at the Research School of Earth Sciences in Canberra is ideal for this work because of the combination of a 193-nm wavelength Excimer laser with a rapid wash-out sample cell and a capability for line scans with a slit of high aspect ratio (Spandler and O’Neill 2010).

A short summary of relevant olivine/melt trace element partitioning studies

Numerous studies of phenocryst/matrix pairs in mafic volcanic rocks as well as partition coefficients determined from experimental crystallization of primitive melt compositions have led to the identification of expected characteristics for trace element concentrations and partitioning between olivine and melt in magmatic systems. Most studies have been carried out on basaltic systems, for which it has been shown that very few trace elements show compatible or even near-compatible behaviour in olivine and that these are restricted to the first row transition elements (Kennedy et al. 1993; Dunn and Sen 1994). Of these, nickel shows the highest partition coefficients (up to 48; Duke 1976; Nabelek 1980; Norman et al. 2005) and D-values above 1 have been found for Cr (up to 5.2), Co (up to 6.3; Duke 1976) and Mn (up to 1.79; Watson 1977). Most partition coefficients for vanadium are below 0.3, as are all known values for Ti, Sc and Ga (Dunn 1987; Nielsen et al. 1992; Ewart and Griffin 1994), whereas for Cu and Zn, compatible values have been reported only in basanites (up to 4 for Cu and 1.4 for Zn; Adam and Green 2006). The rare earth element (REE) patterns of olivines show that all are strongly incompatible, with lowest values for the LREE (4–5 orders of magnitude below 1) and highest values for HREE (generally 0.01–0.05; Kennedy et al. 1993; Dunn and Sen 1994; Adam and Green 2006; Evans et al. 2008).

The effects of variation in melt composition on the partitioning of trace elements are debated: it is often thought that mineral/melt partition coefficients increase towards melts with higher SiO2 or MgO contents (Green 1994; Bédard 2005), whereas variations as a function of pressure or composition within the mafic spectrum (i.e. <54 wt% SiO2) are less clear. Partition coefficients in strongly alkaline, but mafic, lavas have recently been shown to deviate strongly from the behaviour of the same elements in common basaltic melts. Several transition elements show higher partition coefficients in lamproites from Antarctica and Spain: D Ni ranges from 42–95, D Cr from 3.2–5.75, D Co from 6.0–9.3 and D Mn from 1.5–2.2 (Foley and Jenner 2004; Prelevic and Foley 2007). Particularly for Ni and Co, these values are much higher than typical for experiments on basaltic systems. D-values for titanium were slightly lower than in basaltic systems (D Ti 0.004–0.006 compared to 0.008–0.02). O’Neill and Eggins (2002) concluded that melt compositional effects may be particularly strong for the high field strength elements (HFSE) due to complexing with calcium. In this respect, the current investigation of the strongly alkaline but also CaO-rich kamafugite series may help to clarify dependencies on melt composition: the lamproites mentioned above are peralkaline, but poor in CaO.

A further recent development in olivine-melt partitioning concerns the timescale of survival of trace element signals in igneous olivines. Divalent cations have been shown to be sufficiently mobile that they may be used to gauge the rate of ascent of xenoliths (Köhler and Brey 1990), and experiments by Spandler et al. (2007) showed that the trivalent REE may be transported considerable distances through olivine on the timescale of magmatic crystal fractionation. De Hoog et al. (2010) identify a set of elements (Cr, Al, V, Sc, Ca and Na) that may be useful as thermometers. Given that diffusion rates are often thought to be negatively correlated with cation charge, it may be expected that more information on early magma compositions may be gained from higher charged cations such as P5+, Ti4+, Zr4+, Nb5+ and Ta5+. However, new data on diffusion rates indicate that Ti, Zr and Hf may diffuse more rapidly than other 4+cations (Spandler and O’Neill 2010) and that 5+, 3+and 1+cations may be more complex than previously thought (Mallmann et al. 2009; Grant and Wood 2010). Complex zoning in phosphorus was described by Milman-Barris et al. (2008), who noted a lack of consistency of correlations of zoning of P with trivalent elements, possibly indicating systematic differences in diffusion behaviour. The question of differential diffusion rates is important given that vanadium concentrations in olivine have been used to read palaeooxidation states in the lithospheric mantle (Canil and Fedortchouk 2001).

Samples: olivine-bearing volcanics of western Uganda

Four samples (three ugandites C-4035, C-6095 and C-6099, and a leucite basanite C-2896) were selected for trace element analyses in olivine from a series of volcanic rocks from the Toro Ankole and Virunga volcanic fields of western Uganda studied by Tappe et al. (2003) and Rosenthal et al. (2009). The ugandites (olivine kalsilite leucitites in modern classifications) selected contain abundant olivines, some euhedral, others rounded, that form about 90% of the phenocryst population (Fig. 1). Consequently, these samples have amongst the highest MgO contents of the rock suite (16.8–22.6 wt%; Rosenthal et al. 2009). Other minerals in the ugandites are smaller and minor phenocrysts of clinopyroxene and leucite set in a groundmass of leucite, clinopyroxene, oxides (principally titanomagnetite) and glass. The leucite basanites are characterized by phenocrysts of clinopyroxene and lesser amounts of olivine and leucite in a groundmass of leucite, clinopyroxene, nepheline, kalsilite, oxides and glass. These were included to maximize the potential for characterizing olivines with different origins. Ugandite C-4035 and leucite basanite C-2896 had the glassiest groundmasses, enabling average trace element concentrations to be compiled from integration of ten 140-μm spot analyses of the groundmass and thus the calculation of partition coefficients between olivine and coexisting melt. The whole-rock compositions of these two rocks are reproduced in Table 1.

Fig. 1
figure 1

Photographs of olivine showing spot analysis craters, troughs carved by line scans and alteration zones in olivines of Ugandite C-4035. a Traverse of a line scan through two central zones (outlined by red and green dotted lines) and rim zone; b ablation holes in central zone (blue arrow), avoiding rim zone (green arrow). Sloping crack (red arrow) results in an alteration zone of the type listed in Table 6; c round ablation hole (140 μm) in small phenocryst chosen to avoid rim zones. All photographs taken with a Keyence digital microscope with lighting and contrast optimized to view zonation in olivines

Full size image
Table 1 Whole-rock compositions of ugandite C-4035 and leucite basanite C-2896
Full size table

Analytical methods

Trace element analyses in olivines were performed using an ArF EXCIMER laser (193 nm wavelength) coupled to an Agilent 7500 ICP-MS at the Research School of Earth Sciences, Australia National University, Canberra. Analyses were made by ablation of polished thin sections (50–100 μm thick) using a pulse energy of 100 mJ and a repetition rate of 5 Hz, with spot sizes of 40 or 140 μm or line scans. Line scans were achieved by placing a metal slit in the laser path (Spandler and O’Neill 2010), resulting in a rectangular laser pulse with an aspect ratio of about 1:20 impinging on the sample, and by controlled motor-driven movement of the sample orthogonal to the long axis of the laser rectangle. The polished samples were placed in a sample chamber designed in-house at the Research School of Earth Sciences, consisting of a large sample chamber driven under a small, stationary, hopper-shaped ablation chamber. Using a He–Ar gas mixture, this set-up achieves exceptionally rapid wash-out times, minimizing the time lag between ablation and analysis in the ICP-MS, which is advantageous particularly for line scan analysis of materials with inhomogeneous concentrations. This sample chamber set-up has been used in several studies of silicate minerals with low trace element concentrations (Eggins et al. 1998; Witt-Eickschen and O’Neill 2005; Evans et al. 2008; Spandler et al. 2007).

29Si was used as an internal standard to correct for differences in ablation yield on the basis of microprobe analyses of olivines or glass. The external standard was NIST SRM 612. Details of the data reduction procedure, including correction procedures and detection limits, are given in Eggins et al. (1998). Backgrounds were collected for about 30 s before beginning ablation, and signals were collected for 20–40 s for spot analyses and 120–350 s for line scans. For multi-isotope elements, the isotopes 49Ti, 53Cr, 60Ni, 63Cu, 67Zn, 71Ga and 88Sr were used for analysis.

The major element composition of olivine is relatively simple, so that the number of interferences stemming from argides of the major elements Si, Mg and Fe are limited in number, mostly falling between the masses 64–70, and thus addressing Cu, Zn for the elements analysed in our palette. Furthermore, Scandium is a monoisotopic element with an interference from 29Si16O which needs correction. We analysed 63Cu, for which the 23Na40Ar interference is no problem given the low Na content of olivines. For the calculation of partition coefficients, however, this interference is a problem if NIST SRM 612 is used for standardization because it contains 14 wt% Na2O. Therefore, we standardized 63Cu in an extra data reduction run to BCR-2G, which contains only 3.16 wt% Na2O. The negligible residual error from this interference is then only one-third of the 2-sigma error.

Three zinc isotopes are troubled by potential overlaps from Mg or Si; 24Mg40Ar on 64Zn, 26Mg40Ar on 66Zn and 28Si40Ar on 68Zn, whereas the 67Zn isotope has only 4% abundance and an interference from 27Al40Ar which is a concern of similar nature to the 23Na40Ar on 63Cu, namely a potential error source for the glass standard, but not for the olivines. We ran tests to quantify the magnitude of the overlaps by analysing NIST SRM 616 as an unknown and standardizing it to both NIST SRM 612 (2.11 wt% Al2O3; 77 ppm Mg) and ugandite olivine (130 ppm Al; 48.58 wt% MgO). The most significant interference is 28Si40Ar on 68Zn, and so the use of this zinc isotope is not recommended for olivines or other silicate minerals as it will result in up to 8–10 ppm “phantom” zinc. We present data for 67Zn, for which a minimal correction of about 2 ppm has been performed in olivines where a NIST glass with ca. 2 wt% Al2O3 is used. De Hoog et al. (2010) recommended the use of 66Zn because their tests showed only limited formation of 26Mg40Ar in the plasma. Our tests produced a 66Zn/67Zn isotope ratio of 6.7679, which is within 0.5% of the natural isotope ratio of 6.8049, confirming that analysis of zinc on mass 66 is an equally good choice to 67Zn.

The mass overlap between 45Sc and 29Si16O cannot be resolved by the quadrupole ICP-MS and so must be corrected. We obtained the correction factor by measuring 43Ca, 44Ca, 28Si and 29Si in extra calibration runs: 44Ca is falsified by an interference with 28Si16O, whereas 43Ca is unaffected. Taking the difference between the observed 43Ca/44Ca and the natural isotope ratio and assuming the production of 28Si16O and 29Si16O in equal proportions, we quantified the effect of 29Si16O on 45Sc. This amounted to about 7.8% or 0.25 ppm Sc in the ugandite olivine analyses. This agrees well with the 0.2 ppm correction noted by de Hoog et al. (2010).

Results

Results are given in a series of tables and plots. Table 2 presents results for 24 trace elements in spot analyses of large (>0.1 mm) and small olivine phenocrysts in ugandite sample C-4035. Results for the individual spot analyses are given in Online Resource 1, in which the major element compositions by electron microprobe and Mg-values (100 Mg/(Mg + Fe)) are also given. The concentrations are remarkably uniform, and no consistent correlations of trace element concentrations with Mg-value could be observed, in contrast to the behaviour found in ocean island basalts (Sobolev et al. 2007). This is in keeping with the conclusions from earlier studies that the ugandites are near-primary melts and their variation represents mixtures at source, whereas the Virunga rocks (including the leucite basanite) have experienced crystal fractionation (Rosenthal et al. 2009; Platz et al. 2004; Chakrabarti et al. 2009). Standard deviations for individual trace elements average 19 and 23% for large and small phenocrysts, respectively, with lowest deviations for cobalt (2.7 and 5%) and highest for strontium (52 and 106%). The results in Table 2 can be understood to represent the main ugandite phenocryst population against which deviations in zoned crystals can be measured and assessed. The results for the first row transition elements (Sc-Zn, with the exception of vanadium) exhibit the lowest standard deviations, averaging 7 and 12% for large and small phenocrysts, respectively. The groundmass in this sample is the finest grained and most homogeneous amongst the samples analysed, with standard deviations for individual elements for ten 140-μm spots varying between 10 and 71% and averaging only 32%. The average analysis (Table 2) therefore produces the most reliable partition coefficients, which are presented in Table 3 for comparison with olivines from other areas. The ranges of partition coefficients for large and small phenocrysts are very small for all elements (average difference of 1.3%) when compared to the diversity of partitioning values even for basaltic rocks. For example, D Ni averages 31.2 and 34.4 in small and large ugandite phenocrysts (Online Resource 2), compared to a range from 2.2 to 25 from experiments in basalts (Dunn and Sen 1994; Kennedy et al. 1993).

Table 2 Trace element compositions of phenocrystic olivines and groundmass in ugandite C-4035 and in leucite basanite C-2896
Full size table
Table 3 Phenocryst/groundmass partition coefficients for trace elements in Ugandan olivines
Full size table

Intracrystal zonation within the main phenocryst population is not strongly accentuated (Fig. 2), but some general tendencies can be outlined. Trace element concentrations in cores and rims in three olivines shown in Fig. 2 are listed in Table 4. Figure 2a illustrates a line traverse from rim to rim passing through the core of an olivine in ugandite C-4035. The core in Fig. 2a has higher concentrations of Cr and Ni and lower concentrations of Mn, Sc, Ti, Cu and V relative to the rim. The trends for Cr, Ni, Mn and Sc are the same as those between large and small phenocrysts (Table 2), indicating that zonation in this crystal reflects the melt evolution path. Some crystals show increasing Cr from core to rim and no change in Ni, possibly indicating the presence of an older core. Zonation in other crystals is, however, more complex, and does not follow the trends defined by large to small phenocrysts. The crystal in Fig. 2b has a phosphorus-rich core, which correlates with lower Ti and Al, whereas no change is seen in elements that otherwise show zonation (Sc, Mn, Cr, Ni).

Fig. 2
figure 2

Intensity (counts per second) versus time plots of line scan analyses of zoned olivines in ugandite C-4035. Low signals during the first 30 s are gas backgrounds taken before the laser is turned on. a Core with higher Cr and Ni and lower Ca concentrations, but with near-constant Co, Li, Al and P. The two subgrain boundaries show strong enrichment in Al and inconsistent enrichments in Ca, V and Ti. b Example of stronger zonation with cores richer in P and poorer in Al and Ti, but with essentially no zonation in Ni, Cr, Co, Sc and V

Full size image
Table 4 Core and rim analyses of olivines in the main ugandite phenocryst population
Full size table

Table 5 lists trace element analyses for eight crystal cores with unusual concentrations that deviate from those of the main phenocryst population. These are summarized from all four samples studied; partition coefficients are not given because of the lack of reliable groundmass concentrations for two of the samples. Of the five analyses from ugandites, analyses 1, 4 and 5 are relatively similar to the main phenocryst population, with deviations for some elements, and so may represent phenocrysts from related melts. Analysis 1 is notably high in lithium (8.6 ppm), whereas REE and Y resemble the usual phenocrysts. Analysis 4 has high Ni (3,286 ppm) and low Mn (899 ppm), whereas REE are lower than usual for analyses 4 and 5. Analysis 3 has considerably higher concentrations of many trace elements (Cr 1,260 ppm; MREE and HREE, Al, Ga and Zr twofold enrichment; Sr tenfold enrichment) but probably originates from a related melt type. In contrast, the very low concentrations of many elements in analysis 2 (Cr 1.2 ppm, Ti 12 ppm, V 0.21 ppm, Co 107 ppm) indicate that it is probably a xenocryst and may be of non-magmatic origin.

Table 5 Zone analyses from larger olivines
Full size table

Unusually high trace element concentrations in olivines are often attributed to falsification by infiltration of host melts along cracks, particularly in rock types strongly enriched in silicate-incompatible elements such as kimberlites and lamproites (Bedini and Bodinier 1999; Hiraga et al. 2004). Given that kamafugites and leucite basanites belong to the potassic rocks series, the effects of alteration by infiltration of host melts have been investigated by line scan traverses in several crystals, and examples are illustrated in Fig. 3 and tabulated in Table 6. These show that cracks are characterized by high concentrations of Al, Ca, Cu, Ga, Sr, Zr, Nb, La and Ce that are generally between two and ten times the unaltered concentrations, but may be enriched in Sr by up to 500 times. The apparent width of the alteration zones (cracks) is exaggerated in the line scans (Fig. 3) due to their often falling at acute angles relative to the polished surface.

Fig. 3
figure 3

Intensity (counts per second) versus time plots of a line scan analysis across an alteration zone in olivine in ugandite C-4035. The alteration along the crack (140–170 s) is strongly enriched in Sr, Al, Cu, Ga and V. Nickel shows a slight decrease due to the dilution effect. Line scan analyses in general show enrichment in these elements with respect to spot analyses

Full size image
Table 6 Trace element compositions of sections of scan analyses
Full size table

Discussion

Comparison of results collected by spot analyses and line scans

The accurate analysis of trace element concentrations in olivines is important for the correct assessment of trace element budgets in the mantle sources of igneous melts, and also of the possible contamination by infiltration of host melts, particularly alkali-rich melts which are also enriched in many incompatible trace elements. Laser ablation ICP-MS has become the most widely used method for analysing trace elements in situ in minerals, but challenges are still posed by colourless minerals, by submicroscopic inclusions of minerals and fluids below the surface, and by elements with very low concentrations; all of these problems apply to olivine. The first problem is eliminated by the use of the 193-nm wavelength laser, which results in clean ablation and allows constriction of the beam by a high aspect ratio slit whilst preserving sufficient power for ablation. The second and third problems can be approached by optimizing analysis in either spot or line scan mode. The advantages and disadvantages of spot and line scan analysis are briefly assessed here.

The need to recognize and eliminate inclusions from the analysis is especially important for minerals such as olivine with generally low trace element concentrations because even very small inclusions (mineral or fluid) rich in trace elements will greatly falsify olivine analyses. Inclusions are usually recognized by time-resolved analysis in spot mode and removed by appropriate selection of the depth interval for data reduction; most published data for trace element concentrations in minerals were collected in this way. In the line scans used here, the analysed volume is a trough of ca. 2 μm depth instead of a deep hole (compared in Fig. 1), meaning that inclusions can be recognized optically and a line scan path can be chosen to avoid them. The option to eliminate small inclusions (generally fluid) after analysis as for spot analyses remains. The length of line scans is limited by crystal size and the degree of their zonation, whereby the blurring of zone boundaries is minimized in the RSES-ANU set-up by the rapid wash-out in the double chamber system.

Compared to spot analyses, line scans eliminate the uncertainty associated with possible loss of sample material to crater walls in deep holes, but introduce the possibility of systematic surface contamination owing to analysis being limited to the uppermost 2 μm. Polishing materials used in preparation of the polished surface may be concentrated in the uppermost few micrometres. For this reason, we do not present analyses for Pb here. A comparison of the line scan results in Table 6 with the 140 μm spot analyses in Table 2 shows that consistently higher concentrations are found in the line scans for Sr (by a factor of >8), La and Ce (>3), Cu (>2) and Zn (>1.3). However, results for 16 other trace elements with widely different chemical behaviour (e.g. Na, P, Sc, Ti, Co, Ga, Y) show no evidence of surface contamination.

The line scans provide a better opportunity to quantify the chemical effects of alteration along cracks or sub-grain boundaries than do spot analyses because results can be assessed together with knowledge from optical microscopy. The concentration of trace elements, particularly incompatible ones, along grain boundaries has been claimed to be widespread in olivines and mantle xenoliths in incompatible-element-enriched melts such as kimberlites and is important for the interpretation of trace element and isotopic signals and age information (Wittig et al. 2009). The potassic alkaline rocks studied here are also enriched in many incompatible elements (e.g. Sr, Ba, Zr; Tappe et al. 2003; Rosenthal et al. 2009) and so allow assessment of these effects. The results for line scan sections that run across cracks in Fig. 3 show enrichment in Al, Ca, Cu, Ga, Sr, Zr, Nb, La and Ce. This palette includes the same four elements (Sr, La, Ce and Cu) that are enriched generally in line scans with respect to spot analyses, and strontium has the highest levels in both cracks and line scan analyses. This prompts the interpretation that surface contamination may be largely due to the smearing out of alteration material in cracks during polishing and not due to the polishing agents themselves. Despite some extreme enrichments in these few elements, many trace elements remain almost unchanged (e.g. Li, Na, P, Sc and the first row transition elements), so that assumption of a global increase of all “incompatible elements” in cracks and along grain boundaries are not justified.

Zonation of minor and trace elements in olivine phenocrysts

Zonations of trace elements in the main phenocryst population are not strong, but some consistencies can be recognized. Decreasing concentrations of Ni and Cr and increasing concentrations of Sc and Mn can be recognized both in zonation of large olivines (Fig. 2a), and between large and small olivines (Table 2), and so these trends are thought to track the early chemical evolution of the melt. The stronger decrease in Cr (605–460 ppm) than in Ni (2,800–2,460 ppm) for the crystal in Fig. 2a amounts to a 31% decrease in the Cr content of the melt compared to only 14% for Ni, assuming the same partition coefficients apply. This documents co-crystallization of Cr-spinel at this early stage of melt evolution. Several core compositions could be identified that show markedly different concentrations of some elements, and/or trends that contradict those seen in other crystals. These cores usually have abrupt boundaries to zones more typical of the main phenocryst population; for example, the core zone with low Al and Ti and high P in Fig. 2b. The correlation of Ti and Al may indicate a coupled substitution (Spandler and O’Neill 2010). Other core zones presented in Table 5 show higher (analysis 3) or lower (analysis 5) levels of most trace elements, high Ni and lower Mn (analysis 4), or high Rb and Li (analysis 1). These differences are partly due to the analyses coming from different samples of ugandite, but are also partly remnants of earlier phases of melt evolution (analysis 4 with high Ni). Analysis 2 with exceptionally low levels of transition elements (30.7 ppm Al, 1.2 ppm Cr, 12 ppm Ti, 350 ppm Ni) is considered to be a xenocryst, although its Mg-value (92.8) is not markedly different from the main population. It is unlikely that any of the cores described are derived from mantle xenoliths, as the xenolith suite in this volcanic field consists of mica pyroxenites devoid of olivine (Lloyd et al. 1987).

Systematic variation of elements such as Mn, Ni and Co with the Mg-value as found in olivines in Hawaiian basalts (Sobolev et al. 2007) is not seen. The Mg-values of ugandite olivines varies from 91.8 to 87.5 (electronic supplement 1), but plots against trace elements fail to show strong correlations. The correlations seen in Hawaii have been interpreted to be caused by the involvement of recycled pyroxenite (Sobolev et al. 2005, 2007). Although pyroxenites are thought to be involved in the source of Ugandan volcanic melts on the basis of trace element concentrations and radiogenic isotopes (Rosenthal et al. 2009), they probably are not derived from recycled ocean crust, but rather from low-degree melts deeper in the mantle that resulted in phlogopite pyroxenite veins (Tappe et al. 2007; Rosenthal et al. 2009). This indicates that care must be taken to distinguish between the type of mantle pyroxenite and its origin, because the compositions of mantle pyroxenites are likely to vary considerably. The zonation patterns seen in the ugandite olivine population is complex and frequent, but is not strong in most cases. The unusual olivine cores that deviate from the trends of the main phenocryst population (Table 5) make up less than 10% of the crystals analysed, which indicates that the ugandites carry relatively few xenocryst olivines despite their high modal olivine contents. Furthermore, these deviant olivines appear to be mostly from related melt types within the plumbing system of the volcanoes, indicating mixing between related kamafugite melts, which is supported by the variation in radiogenic isotope characteristics (Rosenthal et al. 2009).

Zoning in trace elements with highly charged cations has recently received attention owing to discussion whether they record crystal growth rates and preserve early magmatic information better than cations with lower charge (Milman-Barris et al. 2008; Spandler et al. 2007). Recent results from natural olivine crystals and from experiments have led to the expectation that zonation in Al and P may be preserved when zonation of other cations is obliterated by rapid diffusion. Milman-Barris et al. (2008) describe complex zonation of phosphorus in olivines that tracks skeletal crystal growth shapes and correlates with the trivalent cations Al and Cr, whereas the rapidly diffusing divalent cations show no zonation. However, Spandler et al. (2007) and Spandler and O’Neill (2010) demonstrated experimentally that diffusion of trivalent cations (REE) through olivine at magmatic temperatures is rapid, implying that the preservation of zonation should be prevented except where crystal growth takes place very rapidly. Similarly rapid diffusion of Y3+ in natural olivine xenocrysts has also recently been documented (Qing et al. 2010).

In olivines from both ugandites and leucite basanites zonation in elements ranging in cation charge from Mn2+ to P5+ are preserved, and even Na1+ shows zonation in the ugandites (Table 5). Phosphorus concentrations in the Ugandan olivines are much lower than in the zones described by Milman-Barris et al. (2008) and differ in correlating negatively with Al (Fig. 2b). This confirms that partitioning patterns and diffusion rates of cations with different charges are complex (Grant and Wood 2010; Qing et al. 2010; Spandler and O’Neill 2010) and require more investigation. There is a tendency for positive correlation between the concentrations of high-charge cations P, Ti, Al, Cr, but there are exceptions such as a very low Cr core that is rich in most trace elements (Table 5, analysis 7). This crystal is also enriched in divalent cations such as Mn and Co, showing that trace element behaviour in the natural Ugandan olivines does not follow the expectations from experiments in many cases. Possible reasons for the divergence of results are that the olivines crystallized very rapidly, particularly in the ugandites that preserve Na zonation, or that the difference in temperature or other variables has a greater effect on diffusion rates than recognized to date.

Titanium preserves zonation, correlating with Al, but not with P (Fig. 2b). D Ti is around 0.007, similar to values seen in the lamproites and other kamafugites, although erratically high values up to 2.7 are seen in Spanish lamproite phenocrysts. Similar behaviour is seen for Zr, including the erratic values for Spanish phenocrysts, but higher D Zr in basalts argues against a simple application of the negative correlation found between D Zr and CaO content of the rock found by Evans et al. (2008).

Comparison of olivines between Ugandan volcanic fields and with other volcanic rocks

Trace elements in olivines of the western rift volcanics: ugandite and leucite basanite

The ugandite and leucite basanite samples come from different volcanic fields in the western rift of East Africa: the ugandites are from the southern Bunyaruguru field of the Toro Ankole region, 150–200 km northeast of the Virunga field where the leucite basanite occurs. Whilst the volcanics of this rift are generally summarized as having high K2O, CaO and low SiO2 (Foley et al. 1987), the melt compositions of ugandites and leucite basanites differ quite considerably in detail: the ugandites studied here have lower SiO2 (40.8–41.8 vs. 45.6%), Al2O3 (5.8–7.2 vs. 12.8%) and Na2O (0.77–0.9 vs. 2.15%), but higher MgO (16.8–22.6 vs. 9.5%) and K2O (2.9–5.15 vs. 2.8%) than the leucite basanites (Rosenthal et al. 2009; Table 1). These differences represent the northern part of a stronger, more regional trend of increasing K2O and CO2 contents and decreasing SiO2 towards the north within the western rift, which are thought to mirror regional changes in the composition of primary, mantle-derived melts (Pouclet et al. 1981; Tappe et al. 2003; Rosenthal et al. 2009). The isotopic characteristics and consistently high Mg# (74.9–80.2) of the ugandites indicate near-primary, mantle-derived, melt compositions (Rosenthal et al. 2009), whereas the leucite basanites have well-defined trends at lower Mg# (the sample studied here has Mg# of 59.5), indicating fractionation controlled by the main phenocryst phases olivine and clinopyroxene. The differences in initial melt compositions and in the degree of fractionation therefore cause differences in the trace element composition of olivines.

Given these different magmatic histories, the trace element compositions (Table 2) and partition coefficients (Table 3, Fig. 4) are similar in most cases. The most marked differences are that the leucite basanite olivines have higher DCa and DAl and less fractionated LREE/HREE. The La/Yb ratio of the groundmass (Table 2) also shows an opposite effect with 35 in the leucite basanites compared to 127 in the ugandites. Partition coefficients are also appreciably lower in the leucite basanites for Li and Gd, and marginally lower for P, Ti, Mn, Co and Yb, but higher for La and Ce, and marginally higher for Sc, Cr, Cu, Sr, Y, Nb and Zr. Trace element concentrations in olivine, however, generally correlate with those in the groundmass, indicating that the variation in partition coefficients is barely significant. Most of these differences are probably due to the greater degree of fractionation in the leucite basanites, including longer co-crystallization with other phenocryst phases, particularly clinopyroxene. It can be noted here that the REE partitioning pattern of the ugandites is similar to that of the Gaussberg lamproite (Mg# = 70; Foley and Jenner 2004) which also has a very similar La/Yb (average 122); it is the fractionated leucite basanite that does not conform. Co concentrations do not relate to Ni in any simple way. Also shown in Fig. 4 are D olivine/melt values determined from experiments on basanite by Adam and Green (2006). These generally agree well for elements where the D-values for ugandite and leucite basanite are similar (e.g. Mn, Co, Ni, Ga, Y), but in most other cases, the experimental basanite D-values are higher, but intermediate for the LREE (Fig. 4).

Fig. 4
figure 4

A comparsion of olivine/matrix partition coefficients for trace elements in ugandite C-4035 from the Bunyaruguru volcanic field (filled circles) and leucite basanite C-2896 from the Virunga volcanic field (open squares). Partition coefficients are very similar for most elements, apart from Li, Sc, Ca and the REE. The steeper REE partitioning pattern of the ugandite is more typical of magmatic olivines. Stars indicate experimental partition coefficients measured on a less potassic basanite, and vertical bars indicate the range of values from the experimental basanite study (Adam and Green, 2006)

Full size image

Comparisons with other alkaline and basaltic volcanics

The most promising trace elements for deciphering the early history of mafic melts are the transition elements, owing to their relatively high concentrations in olivines, and highly charged elements such as phosphorus, niobium, titanium and zirconium, as these may better preserve zoning due to their slow but varying diffusion rates (Milman-Barris et al. 2008; Spandler and O’Neill 2010). The concentrations of trace elements in olivines from the ugandites and leucite basanites of Uganda are compared with those in other volcanic rocks in a series of plots in Figs. 5-7. Figure 5 compares nickel, for which olivine is the major host, with titanium which shows variable concentrations thought to reflect its concentration in the host rock (de Hoog et al. 2010). In this plot, olivines from Mg-rich mantle-derived melts such as kimberlites and picrites lie in a restricted band between 2,500 and 3,100 ppm Ni (Brett et al. 2009; de Hoog et al. 2010). The mantle is represented by olivines from diamond-bearing peridotites and inclusions in diamond (Sobolev et al. 2009). Olivines in fractionated melts such as rims on kimberlitic olivines or in the melilitites of Oldoinyo Lengai (Brett et al. 2009; Keller et al. 2006) show lower Ni contents, whereas much higher Ni contents—up to 7,000 ppm—are found in olivine phenocrysts in ocean island basalts and in Spanish lamproites (Neumann et al. 1999; Prelevic and Foley 2007). The ugandite olivines show Ni contents typical of primitive mantle-derived high-MgO melts, but have the highest Ti contents (180–330 ppm) yet identified in such rocks, in keeping with the high TiO2 in the whole-rock (3.94 wt%; Table 1). The Ni contents of olivines in the leucite basanites (average 2,280 ppm) indicate a less advanced stage of fractionation than implied by the Mg-value of 59.5, arguing that the olivines are relatively early phenocrysts carried in a more fractionated melt.

Fig. 5
figure 5

Concentrations of nickel and titanium in olivines from ugandites (red circles) and leucite basanites (yellow circles) compared with those from other volcanic rocks and diamond-bearing mantle rocks. Primitive mantle–derived melts lie in a band between 2,500 and 3,100 ppm Ni; those deviating from this band either crystallized from fractionated melts (lower Ni) or involve melting of pyroxenite in their sources (higher Ni). All olivines with >100 ppm Ti are volcanic. The relatively high Ni in leucite basanite olivines argues against crystallization from a strongly fractionated melt. Symbols and data sources: Kimberlites (green triangles; black borders are rim analyses—Brett et al. 2009), Spanish lamproites (yellow diamonds—Prelevic and Foley 2007), ocean island basalts (blue squares—Neumann et al. 1999), Oldoinyo Lengai olivine melilitites (grey diamonds—Keller et al. 2006), Baffin picrites (blue diamonds—de Hoog et al. 2010), inclusions in diamonds (black squares—Sobolev et al. 2009), diamond-bearing peridotites (white squares—Sobolev et al. 2009)

Full size image

In the plot Cr vs. Mn (Fig. 6), there is also a central band for MgO-rich mantle-derived rocks between 700 and 1,300 ppm Mn, whereby the ugandite and leucite basanite olivines extend to higher Cr contents (640 and 810 ppm, respectively) than mantle rocks (<500 ppm). High Mn concentrations above this band are typical of ocean island basalts and melilitites (Neumann et al. 1999; Hagerty et al. 2006; Keller et al. 2006). These high Mn contents correlate with Zn, Al, Co and Sc and are probably derived from recycled pyroxenite in the source (Sobolev et al. 2007). Here again, the high Cr contents in the leucite basanite olivines appear to indicate early crystallization before fractionation lowered the Mg value of the melt.

Fig. 6
figure 6

Variation of Cr and Mn concentrations in volcanic olivines. Olivines crystallized from primitive mantle-derived melts generally have 700–1,300 ppm Mn—the Ugandan olivines fall in this range with high Cr contents. Olivines with >1,500 ppm Mn are interpreted as involving melting of recycled pyroxenites in their sources. The mica pyroxenites in sources of the Ugandan rocks have contrasting trace element features to those of recycled ocean crust. Symbols and data sources as for Fig. 5, plus Hawaiian basalts (orange squares—Hagerty et al. 2006), Gaussberg lamproites (black crosses—Foley and Jenner 2004), mid-ocean ridge basalts (grey triangles—Jeffries et al. 1995)

Full size image

In terms of partition coefficients, the highest partition coefficients are found for nickel, and compatible behaviour (D > 1) is found also for Co (3.79–4.69), Cr (1.66–3.2) and in some cases Mn (0.84–1.395). Values for all large phenocrysts are within the range for small phenocrysts for these elements. The D-values for all three elements lie within the range found for basalts (Fig. 7; Duke 1976; Villemant et al. 1981; Beattie 1994; Norman et al. 2005), but D Cr is notably higher than for kamafugites from Brazil (Melluso et al. 2008).

Fig. 7
figure 7

Co-variation of Ni partition coefficients with those of Cr and Co shows a general correlation for magmatic olivines, with olivines from the Ugandan volcanics occupying the central part of the range. D Ni > 50 is known only from lamproites, and olivines in these rocks splay out from the near-linear relationship (note that the xenocrysts in the Spanish lamproites [stars] may be mantle-derived). Most basaltic olivines show much lower D-values for the transition metals, with D Ni correlating well with K2O of the rock. Data sources: Foley and Jenner (2004), Prelevic and Foley (2007), Norman et al. (2005), Kennedy et al. (1993), Adam and Green (2006), Duke (1976), Seifert et al. (1988), Gaetani and Grove (1997)

Full size image

The olivine-matrix partition coefficients for nickel are 23–38 in the ugandites and 34–38 in the leucite basanites. The highest D Ni for natural phenocryst/matrix pairs measured by in situ techniques are for K-rich rock compositions, and strong positive correlations are seen between D Ni and K2O/Na2O (Fig. 8a), K2O/Al2O3 (Fig. 8b) and K2O, suggesting that enrichment of melts in potassium is the major factor that leads to high DNi rather than peralkalinity. This is probably principally due to the depolymerizing effect of alkalis on melt structure, which is greater than the effect of water (Mysen and Cody 2005). The relative effect of potassium is greater than that for sodium, as gauged by the magnitude of the offset of phase boundaries in Na- and K-rich systems, thus promoting the stability of the less polymerized phase (Kushiro 1975; Gupta et al. 1987; Foley et al. 1986). There is no correlation between D Ni and Mg# (Fig. 8c) or SiO2, indeed the entire range of D Ni known to date is found in melts rocks with 50 ± 2 wt% SiO2 (Fig. 8d). Despite the larger number of olivines analysed, the Ugandan results show less variation in D Ni than in lamproites, which show a range of values straying to higher D Ni relative to the simple linear correlation (Fig. 8a, b; Foley and Jenner 2004; Prelevic and Foley 2007). Higher D Ni for a given K2O/Na2O or K2O/Al2O3 are also found for basanitic melts in peridotite melting experiments (Adam and Green 2006) and for Kilauean tholeiites (Norman et al. 2005); the reasons for this are unclear.

Fig. 8
figure 8

Values from the Ugandan volcanics for the olivine/matrix partition coefficient for nickel fill out the central part of the range for magmatic olivines, defining a good correlation with K2O, K2O/Al2O3 and K2O/Na2O (a, b), indicating K2O to be the main factor influencing D Ni. No good correlation is seen with SiO2 (the entire D Ni range is taken up by rocks with 50 ± 2 wt% SiO2) or Mg#, contrary to the general dependence on MgO seen amongst basaltic rocks (Bédard 2005). Data sources: [1] Foley and Jenner (2004); [2] Prelevic and Foley (2007); [3] Kennedy et al. (1993); [4] Norman et al. (2005); [5] Adam and Green (2006); [6] Gaetani and Grove (1997); [7] Nabelek (1980)

Full size image

Vanadium has been used as a gauge for the oxidation state at the time of crystallization, controlled by the decreasing V3+/V5+ speciation with increasing oxygen fugacity because olivine takes up only V3+ (Canil and Fedortchouk 2001). Results for ugandites and leucite basanites from the western rift are consistent at D V around 0.015. The slightly lower values for Gaussberg lamproite (0.01; Foley and Jenner, 2004) and higher values for the Spanish lamproites (0.06–0.21; Prelevic and Foley, 2007) give a relative order of oxidation states in agreement with oxygen fugacities estimated from Fe3+/(Fe3++Fe2+) of chromite inclusions in olivine by Foley (1985). Applying the V/Sc method of Mallmann et al. (2008) and Mallmann and O’Neill (2009), the calculated fO2 at crystallization of the ugandite olivines varies between ∆FMQ-0.54 and ∆FMQ + 1.38. The average value is slightly higher for the small phenocrysts than the large phenocrysts (FMQ + 0.84 vs. FMQ + 0.64), possibly indicating minimal oxidation during emplacement which is much more marked for lamproites (Foley 1985). D V for olivines in the Brazilian kamafugites (0.0168; Melluso et al. 2008) is similar to the western rift kamafugites analysed here, implying crystallization at a similar oxygen fugacity slightly below the Ni-NiO buffer. D V for Kilauean tholeiites are similar, whereas most basaltic to andesitic olivines show higher D V (0.02–0.09: Duke 1976; Luhr and Carmichael 1980; Ulmer 1989), implying lower fO2 conditions during the early stages of crystallization. The agreement between results from D V in olivines and Fe3+/(Fe3++Fe2+) in spinels indicates that fO2 information is often preserved in olivines.

Scandium shows slightly higher partition coefficients in leucite basanites (0.015–0.018) than in ugandites (0.011–0.012), whereas D V for the Brazilian kamafugites are even lower at 0.047 (Melluso et al. 2008). Values in basalts and lamproites are generally >0.1 (Nielsen et al. 1992; Ulmer 1989; Foley and Jenner 2004; Prelevic and Foley 2007), and higher still in chondrites (up to 0.84; Kennedy et al. 1993).

DLi in ugandites (0.36) is similar to basalts (generally 0.3–0.4), although some higher values exist for basaltic andesites (up to 1.3; Dunn and Sen 1994). DLi for the leucite basanites is notably lower (0.13). Corroded “xenocrysts” in Spanish lamproites show similar values to the magmatic olivines studied here, whereas later phenocrysts show erratic values much higher than those seen in most volcanics (0.65–1.8; Prelevic and Foley 2007).

Data are sparser for Cu, Zn and Ga in olivine. Also, care must be taken in assessing values for Cu and Zn because of the varying choice of isotope measured, bearing in mind the problems noted in the “Analytical methods” section. This may apply to experimental determinations on basanitic melts of peridotite that imply compatibility of both Cu and Zn (D Cu 1.2–4.1; Adam and Green 2006). Zinc is apparently more incompatible in the Ugandan volcanics (D Zn 0.44–0.49) than in any basaltic rocks measured to date, but more measurements are needed to tell whether this may also due to analytical problems.

DGa is low in the ugandites (0.006–0.007), but lower still in Spanish lamproites (<0.0024; Prelevic and Foley 2007). The few published values range from 0.005–0.3; Ewart and Griffin 1994; Canil and Fedortchouk 2001; Adam and Green 2006).

Comparison of trace elements in olivines from ugandites and metasomatically enriched peridotites

The volcanic rocks of the western rift have potassium-rich, low-silica compositions similar to those expected for melts that cause metasomatic enrichment of the mantle lithosphere and are known from isotopic and trace element studies to be essential precursors to many kinds of magmatism (Foley 1992; Litasov et al. 2005; Jiang et al. 2006; Maria and Luhr 2008; Conticelli et al. 2009). Equilibration of infiltrating melts with peridotitic olivines may result in similar trace element characteristics to those seen in the ugandites.

A brief survey of the available data indicates that magmatic olivines differ from peridotitic olivines clearly in their higher Ca and Ti concentrations. Magmatic olivines from ocean island basalts and melilitites as well as the ugandites all have >600 ppm Ca and >100 ppm Ti (Fig. 9a; Neumann et al. 1999; Keller et al. 2006), whereas peridotitic olivines have less than 600 ppm Ca and <1–100 ppm Ti (O’Reilly et al. 1997; van Achterbergh 2004; Kaeser et al. 2006; Zheng et al. 2007; Rehfeldt et al. 2008; de Hoog et al. 2010). Exceptions are found particularly in olivines from Kaapvaal craton peridotites that show petrographic evidence for enrichment by melts (up to 340 ppm Ti; van Achterbergh 2004; Rehfeldt et al. 2008) and rift-related overprints from eastern Antarctica (Foley et al. 2006) with up to 1,200 ppm Ca. Trace element patterns in metasomatized peridotites appear to define two trends; one towards high Ti that follows the ugandite characteristics, and a second to high Ca and Ti from the rift-related areas and ocean islands (Fig. 9a; Gregoire et al. 2000; Neumann et al. 2002; Foley et al. 2006). The Al vs. Ti pattern is very similar, with ugandites following kimberlites to high Ti without a notable increase in Al concentrations (Fig. 9b). The high Al and Ca trend is also characterized by higher Mn and Zn, following the enrichment of ocean island basalts in Ni, Mn, Al, Co and Zn, probably linking these trace element signals with pyroxenite recycling as hypothesized for high Ni in ocean island basalts (Sobolev et al. 2005, 2007). Neither the ugandites nor the Kaapvaal peridotitic olivines show enrichments in all these trace elements, probably indicating that the metasomatic agent in the Kaapvaal is a Ti-rich low-silica melt that is not derived from recycled ocean crust.

Fig. 9
figure 9

Variation of Ca and Al with Ti concentrations in volcanic olivines compared to mantle olivines represented by inclusions in diamond and diamond-bearing peridotites. The trends to high Ti contents without strong increase in Ca or Al of the ugandites (red circles) follow the trend of kimberlites (green triangles). The high Ca–Ti signature of plumes or recycled ocean crust (blue squares) may be present in diluted form in the leucite basanites (and Oldoinyo Lengai olivine melilitites) but is not seen in the ugandites. Symbols and data sources as for Fig. 5

Full size image

Conclusions

Although many of the traditionally analysed “incompatible” suites of trace elements are at concentrations below the detection limit in olivine phenocrysts, optimized Laser-ICP-MS analyses enabled the determination of concentrations and phenocryst/matrix partition coefficients for up to 34 trace elements to assess their usefulness as monitors of early magmatic history. A comparison of line scans representing shallow near-surface trough analyses with spot analyses representing analysed volumes equivalent to vertical cylinders shows large spot analyses to be preferable with respect to line scans for the determination of elements with concentrations near the detection limits. Line scan analyses are contaminated by trace elements concentrated on alteration zones (Al, Ca, Cu, Ga, Sr, Zr, Nb, La and Ce), which are smeared out over the surface during polishing. This must be counteracted by pre-ablation of the surface by laser shots prior to quantitative analysis if line scan analyses are to be made.

The main phenocryst population of the ugandites shows common, but relatively subtle zonation, particularly decreasing Ni and Cr, and increasing Mn and Sc towards the rims. Despite the high modal content of olivines in the ugandites with 16–22 wt% MgO, less than 10% of the olivines appear to be xenocrystal, and most of these are derived from related kamafugitic melt types from the volcanic plumbing system. Olivines in the leucite basanites have higher Ni (average >2,200 ppm) and Cr contents than would be expected from the fractionated rock composition (Mg-value 59), indicating crystallization at an early stage.

Leucite basanites have partition coefficients generally similar to ugandites, but markedly lower for Na and Li and higher LREE/HREE. Partition coefficients for Ni are high relative to basalts, but not as extreme as in lamproites (Foley and Jenner 2004), whereas D-values for Zn, Ga and Sc are low. Intermediate values for Cr, Co and Mn overlap with those known from olivines in basalts. Zonation is preserved for elements with cation charges from 5+ to 2+(P, Ti, Zr, Cr, Al, Mn, Ni) but does not follow the correlations seen in basalts (Milman-Barris et al. 2008).