Volume 22, Issue 3 e2020GC009212
Research Article
Free Access

Trachytic Melt Inclusions Hosted in Clinopyroxene Offer a Glimpse Into Samoan EM2-Endmember Melts

J. V. Adams

Corresponding Author

J. V. Adams

Department of Earth Science, University of California Santa Barbara, Santa Barbara, CA, USA

Correspondence to:

J. V. Adams,

[email protected]

Search for more papers by this author
F. J. Spera

F. J. Spera

Department of Earth Science, University of California Santa Barbara, Santa Barbara, CA, USA

Search for more papers by this author
M. G. Jackson

M. G. Jackson

Department of Earth Science, University of California Santa Barbara, Santa Barbara, CA, USA

Search for more papers by this author
First published: 11 February 2021
Citations: 3


We present major and trace element data on clinopyroxene-hosted, glassy melt inclusions (MIs) from a highly enriched (high 87Sr/86Sr = 0.71856) lava (ALIA-D115-18) from the Samoan hot spot. Following correction for post-entrapment crystallization, the MIs are trachytic (62.6–65.4 wt% SiO2, 1.5–3.0 wt% MgO, and 7.2–10.6 wt% total alkalis (Na2O + K2O)) ALIA-D115 lavas dredged off the submarine flanks of Savai'i Island, western Samoa have previously been suggested to exhibit binary endmember magma mixing between a mafic, low 87Sr/86Sr magma and an evolved, high 87Sr/86Sr magma (the Enriched Mantle 2 endmember; EM2). In major and trace element space, these ALIA-D115 lavas form a mixing trend between more primitive Samoan shield lavas and the MIs of this study, suggesting the MIs could represent the EM2-derived mixing endmember. Recent efforts to constrain the EM2 mixing endmember composition shows remarkable overlap with the MI compositions, supporting the argument they could represent the 87Sr/86Sr mixing endmember to the ALIA-D115 lavas and are derived from the EM2 endmember. Additionally, one melt inclusion contains Fo73 olivines not in equilibrium with the trachytic melt, suggesting the melt interacted with a mafic magma prior to entrapment within host clinopyroxene. Projecting the ALIA-D115 mixing trend in 87Sr/86Sr versus SiO2 and K2O space shows that the trend intersects an average MI composition at 87Sr/86Sr values of ∼0.725, thereby offering a window into more extreme compositions associated with the EM2 endmember.

Key Points

  • Reconstructed clinopyroxene-hosted melt inclusions from an enriched (high 87Sr/86Sr) Samoa hot spot lava are trachytic

  • Melt inclusion chemical variability is not the result of post-entrapment crystallization or diffusion, but represents magma heterogeneity upon capture

  • The melt inclusions are a mixing endmember for a suite of enriched Samoan lavas and may be evidence of Enriched Mantle two derived liquids

Plain Language Summary

Droplets of silicate melt (i.e. melt inclusions) captured in the mineral clinopyroxene were measured for their chemistry. The melt inclusions and their clinopyroxene hosts are found in one lava recovered from the submarine flanks of Savai'i Island—an oceanic island formed along the Samoan hot spot track. The melt compositions represent silicate liquids present during magma evolution beneath Samoa. Lavas erupted at Samoa uniquely show extreme chemical signatures associated with ancient continentally derived sediment that has been recycled into Earth's interior via subduction. These diagnostic recycled materials have subsequently been incorporated into the Samoan plume, although the composition and origin of this material remains enigmatic. The melt inclusions provide a window into the composition of this recycled material and this study shows the melt inclusions may have been derived from a sediment-infiltrated source. Remarkably, mixing between the melt inclusions and other compositionally distinct lavas from Samoa can explain the range in chemistry of lavas dredged from Savai'i Island. The Samoan melt inclusions help us understand the origin and evolution of the source contributing to the volcanic rocks of the Samoan hotspot, and ultimately, the chemical evolution of the Earth's mantle.

1 Introduction

Melt inclusions (MIs) have been used for decades to provide a glimpse into the petrogenetic evolution of lavas (e.g., Anderson & Wright, 1972; Danyushevsky et al., 2002; Kent, 2008; Kent & Elliott, 2002; Lowenstern, 1995; Melson, 1983; Plank et al., 2013; Sobolev, 1996). Melt inclusions represent liquids present in evolving magmas captured by growing phenocrysts and can be used to better understand the evolution of a magma. Conceptually, these captured melts are physically isolated from the effects of subsequent (post-entrapment) magma mixing, assimilation, or fractionation that ultimately modify magma compositions. They present compositional snapshots in time, although following entrapment MI compositions may be modified by post-entrapment crystallization (PEC) or by post-entrapment diffusive exchange (PED) with their host crystal. MIs often reveal highly heterogenous compositions, or distinct compositions not realized by the associated erupted products (e.g., Kamenetsky, 1996; Kamenetsky et al., 1997; Kent et al., 1999; Maclennan, 2008; Reinhard et al., 2018; Rowe et al., 2011; Saal et al., 1998; Sobolev et al., 2000; Sobolev & Shimizu, 1993; Sours-Page et al., 1999; Wallace, 2005). Compared with the whole rock, these differences in composition are likely a direct result of sampling different time frames during melting and mixing processes, or even different mantle sources than are present in whole rock lavas (e.g., Jackson & Hart, 2006; Saal et al., 1998; Sobolev, 2007). Other work suggests anomalous melt inclusion compositions could also be the result of dissolution-reaction-mixing processes at the grain-scale within the magma chamber due to injections of new magma that results in partial dissolution of mush zones (Danyushevsky et al., 2004). In either case, melt inclusions can be used to help better understand lava petrogenesis.

Here, we focus on reconstructed major and trace element compositions of uniformly trachytic clinopyroxene-hosted melt inclusions found within the second-most-enriched (high 87Sr/86Sr = 0.71856) whole rock lava from the Samoa hot spot (whole-rock chemistry can be found in Jackson et al., 2007). This lava, ALIA-D115-18, was dredged off the submarine flanks of Savai'i Island, western Samoa, and similar to other lavas from the same dredge haul—(ALIA115 dredge haul) showing the most geochemically enriched 87Sr/86Sr values ever measured in relatively fresh whole rock ocean island basalts (OIB)—this lava exhibits the “enriched mantle 2” (EM2) geochemical signature (Jackson et al., 2007). The EM2 signature is believed to be the result of addition of ancient subducted terrigenous sediment to the mantle source region (Adams et al., 2021; Edwards et al., 2019; Jackson et al., 2007; White & Hofmann, 1982). However, important questions remain regarding the composition of EM2 or its source, including the precise composition of the EM2 endmember, the petrogenetic history of EM2-derived OIB melts, and how these melts mix with non-EM2 melts in magma chambers. A recent study by Adams et al. (2021) uses binary mixing theory, high precision 87Sr/86Sr and 143Nd/144Nd analyses of single-crystal magmatic clinopyroxene, and major and trace element data of clinopyroxenes within the ALIA-D115-18 sample to calculate the EM2 magma mixing endmember as being trachytic in composition. Further, this trachytic composition is shown to represent a magma mixing endmember, that when mixed with a much more mafic endmember, can reproduce the chemical signatures observed in the entire ALIA-D115 lava suite from which ALIA-D115-18 is derived. The work presented here uses the clinopyroxene-hosted melt inclusions in the ALIA-D115-18 sample to place further constraints on the composition and petrogenetic history of EM2. We emphasize that, remarkably, the range of calculated trachytic EM2 endmember model melts from Adams et al. (2021) bear a striking resemblance to the trachytic melt inclusions of this study.

We suggest below that the clinopyroxene-hosted trachytic MIs in sample ALIA-D115-18, once corrected for post-entrapment crystallization, are EM2-endmember melts captured during the magma mixing process. One clinopyroxene-hosted MI has two partially resorbed Fo73 olivine crystals, suggesting interaction of the trachytic melt with a mafic magma as the olivines and the surrounding trachyte melt are not in equilibrium. This observation is consistent with two-component mixing between a mafic (olivine-bearing) magma and a much more evolved magma that has been inferred from relationships between major and trace element compositions and radiogenic isotopic ratios in the Samoan magmatic suite alluded to above (Adams et al., 2021; Edwards et al., 2019; Jackson et al., 2007). This raises several important questions: what is the origin of these high-silica MIs? What is their relationship to the ALIA-D115-18 whole rock and the EM2 source? Once corrected for possible post-entrapment modification, do they represent relatively primary high-silica partial melts that were trapped in growing clinopyroxenes at depth during magma mixing? These questions are addressed in the following sections.

2 Samples and Methods

Melt inclusions of this study are hosted in clinopyroxenes from a single Samoan hotspot lava (ALIA-D115-18), dredged off the submarine flanks of Savai'i Island, western Samoa. Twenty-two MIs were studied with diameters ranging from ∼15 to 100 µm. The MIs are generally spherical or ellipsoidal in cross-section and are mostly glassy; some contain minor amounts of crystal inclusions consisting of clinopyroxene ∼ ilmenite >> olivine (Figure 1). Of the 22 inclusions, vapor bubbles are observed in two MIs occupying between 11 and 26 vol% of the MI volume, one shows significant amounts of crystal inclusions representing ∼10 vol% of the MI (Figure 1a), one has ∼2 vol% of ilmenite (Figure 1b), and, most significantly, 20 MIs (91%) show <2 vol% of crystal inclusions. The crystal volume estimate was made assuming an ellipsoidal geometry for both the MI and the crystals.

Details are in the caption following the image

Back-scatter electron (BSE) images of select melt inclusions. (a) Melt inclusion with the largest amount and variety of crystal inclusions. (b) and (c) are relatively pristine melt inclusions with some ilmenite crystal inclusions nucleating on the melt inclusion walls. (d) Relatively pristine melt inclusion showing clinopyroxene needles that have nucleated in the inclusion (e) one of two melt inclusions that show direct reaction of the melt inclusion walls with the surrounding clinopyroxene. The reaction texture is denoted by the jagged appearance of the melt inclusion wall. In addition, there is a slightly brighter region around the melt inclusion edge compared with the rest of the host clinopyroxene, indicative of compositionally distinct clinopyroxene. (f) Melt inclusions showing minimal daughter crystals. All of the above melt inclusions also show laser and ion probe pits, and remnant gold coat in regions of low topography is visible as bright regions in the images. See Supporting Figure S10 for BSE images of the other melt inclusions analyzed in this study.

For volatile and trace element analysis of the melt inclusions by ion probe, the sample surface was pre-sputtered prior to analysis. Volatile (H2O, Cl, F, and S) elements in the melt inclusions were analyzed on the Cameca IMS 6F ion microprobe housed at the Department of Terrestrial Magnetism (DTM), Carnegie Institution of Washington following methods from Hauri (2002), where uncertainties are ± 10% (2urn:x-wiley:15252027:media:ggge22451:ggge22451-math-0001) for volatile element concentrations. Trace element (Sc, Sr, Nd, Y, Zr, Ba, Ce, Dy, Nb, La, Yb, Hf, and Gd) analyses of the glassy portions of the melt inclusions were also carried out on the DTM Cameca IMS 6F and following techniques in Peterson et al. (2017), where precision on trace element analyses is better than 10% (2urn:x-wiley:15252027:media:ggge22451:ggge22451-math-0002). Secondary standard ALV 519-4-1 was measured along with the melt inclusion trace element measurements. These measurements can be found in Data Set S1. Major element concentrations in the melt inclusions were measured on the JEOL Superprobe housed at the Geophysical Laboratory, Carnegie Institution of Washington, and methods and uncertainties follow Hauri (2002): a 15 kV accelerating voltage was used, and a 10 nA, defocused (10 µm) beam was employed during analysis. Lastly, it is important to note these melt inclusions were never heated (e.g., by heating stage or furnace) for rehomogenization. New major, trace, and volatile element data on melt inclusions are shown in Data Set S2.

Major element concentrations in the hosting clinopyroxenes and the olivine inclusions in the mt1gr11 MI (Data Set S2) were also measured on the JEOL Superprobe housed at the Geophysical Laboratory, Carnegie Institution of Washington, and methods and uncertainties follow Hauri (2002). Trace elements were measured in the host clinopyroxenes (Data Set S2) by LA-ICP-MS at the University of California Santa Barbara (UCSB) using a Photon Machines Excite 193 Excimer laser coupled to an Agilent 7700 quadrupole ICPMS. A 15 µm spot size was used and the reference material NIST612 was analyzed every 8–10 unknown analyses; analyses and the 2RSD on each element can be found in Data Set S3. Each trace element measurement reported in Data Set S2 is an average of three to four spot analyses spread across each clinopyroxene grain. Additionally, we measured 17 compositional profiles in four of the clinopyroxenes hosting melt inclusions and the major and trace element data can be found in upporting Data Sets S4 and S5, respectively. Major and trace element analyses were carried out in several transects across four clinopyroxenes, extending from near the clinopyroxene-melt inclusion boundary to distal regions of the clinopyroxene, and analyses were carried out at UCSB. Major elements were analyzed on the EPMA at UCSB; an accelerating voltage of 15 kV was used with a 20 nA beam with a spot size of 2 µm.

Correction for PEC was performed on each individual MI. For each MI, 2% increments of the host clinopyroxene composition (for major, trace, and volatile elements—assuming zero concentration in the clinopyroxene for the volatiles, as volatiles were not measured in the clinopyroxenes) was added back into each MI bulk composition until the MI was saturated in clinopyroxene of Mg# matching the respective host composition Mg#. This iterative calculation was performed using the MELTS modeling software (Ghiorso & Sack, 1995; Gualda et al., 2012; the most applicable modeling software available for igneous phase equilibria calculations) in which, after each incremental addition of host clinopyroxene to the MI, the new bulk MI composition was input into MELTS to calculate the stable mineral assemblage from the liquidus to ∼100°C below the liquidus. Once the calculated liquidus phase was clinopyroxene with a Mg# matching that of the real host clinopyroxene value, the process of clinopyroxene addition into the MI was complete. In the MELTS calculations, we used the Quartz-Magnetite-Fayalite oxygen buffer to dictate the ferrous to ferric Fe ratio—all Fe was converted back to ferrous Fe at the end of the calculation and thus, only total FeO is reported in Data Set S6. We tested a range of pressures for each calculation and results showed that the temperature of clinopyroxene saturation and the Mg# of that clinopyroxene for a given bulk composition were insensitive to pressure (i.e., the clinopyroxene saturation temperature and Mg# were within typical analytical uncertainties at a range of pressures). Although there are other daughter crystals within some of the MIs, the amount of clinopyroxene added back into the MIs far outweighs the volume percentage of other crystals, such as ilmenite (which may not all be daughter crystals, see below), so other daughter crystals were neglected from the reconstruction calculation. The original melt inclusion compositions and host clinopyroxene data can be found in Data Set S2. The reconstructed MI compositions and the amount of clinopyroxene addition to the MI to account for PEC can be found in Data Set S6.

3 Results

3.1 Major and Trace Elements

The original melt inclusions of this study were corrected for minor amounts of post-entrapment crystallization of the host clinopyroxene. The amount of clinopyroxene added back into the original melt inclusions ranges from 16% to 24% by mass, with most requiring less than 20%. The original measured and reconstructed melt inclusions are trachytic and show limited variability in the major elements with the exception of silica, Na2O, and K2O. The reconstructed melt inclusions lie between ∼62.6–65.4 wt% SiO2, 1.5–3.0 wt% MgO, and 7.2–10.6 wt% total alkalis (Na2O + K2O) (see Data Set S6 for full compositions).

Trace elements have been measured in 16 of the 22 inclusions. The reconstructed incompatible trace element compositions show much more variability compared with major elements (see Data Set S6 for major and trace element compositions). There is significant variability in the REEs, as can be seen in Figure 2, that is not explained by PEC (as discussed in the following section). A comparison of the melt inclusions with the ALIA-D115-18 whole rock and average Ta'u and Ofu lavas, which represent the least geochemically enriched lavas from Samoa, shows that most of the ALIA-D115-18 MIs are more enriched in Ba, Nb, K, and some light rare-earth elements (LREEs), and depleted in Ti and some heavy rare-earth elements (HREEs), compared with other Samoan whole rock lavas (Figure 2). Nb and Ti show distinct depletions for all of the MIs, likely related to ilmenite removal from the melt pre- or post-entrapment. Additionally, many of the MIs show positive or slight negative Sr anomalies (where Sr anomaly refers to the deviation of Sr above or below bracketing elements with similar incompatibility in a primitive mantle normalized spider diagram).

Details are in the caption following the image

Primitive mantle normalize trace element patterns of clinopyroxene-hosted melt inclusions (MIs) (blue circles) from the ALIA-D115-18 whole rock lava (red squares; Jackson et al., 2007). The data are shown in comparison with other Samoa whole rocks that have geochemically depleted 87Sr/86Sr compositions (average compositions for Ta'u and Ofu whole rocks; samples with <7wt% MgO have been excluded). The orange and gray diamonds represent average Samoan lava compositions from Ta'u and Ofu Islands, respectively, which are the most geochemically depleted shield volcanoes in the eastern Samoan region. The dark blue triangles are a high silica, high 87Sr/86Sr (0.71234) Savai'i xenolith glass, 90SAV1-1, from Hauri et al. (1993). The inset spiderdiagram shows rare earth elements only. All of the MIs plotted are the reconstructed MIs from Data Set S6.

3.2 Volatiles

Water, sulfur, and the halogens have been measured in 18 of the 22 MIs. The original uncorrected MI compositions show H2O contents ranging from 0.4 to 1.1 wt%, S ranging from 67 to 504 ppm, Cl ranging from 781 to 2916 ppm, and F ranging from 19 to 2674 ppm. Although the MI volatile elements were not analyzed in the host clinopyroxene, important for making the correction for PEC, we simply assume the host clinopyroxenes have concentrations of 0 ppm for all volatiles (a valid assumption considering volatiles are highly incompatible in clinopyroxene) to make the PEC correction. The reconstructed volatile ranges are 0.3–0.9 wt% H2O, 56–418 ppm S, 648–2449 ppm Cl, and 16–2219 ppm F (see Data Set S6). A comparison of any reconstructed volatile element with melt inclusion size shows no correlation, suggesting the variation in volatile concentrations, and in particular H2O, has not been affected by diffusive loss (Cabral et al., 2014; Kent, 2008). F concentrations are highly variable across the MI suite. In the H2O, Cl, and S dataset, one MI has anomalously low volatile concentrations compared with the rest of the MIs, and this inclusion is also the MI that shows the most faceted shape (Mt2gr16A; Figure 1f). We do not have an explanation for the low volatile concentrations in this inclusion, except to suggest that the inclusion may have ruptured during ascent, allowing some volatile loss, prior to reannealing.

4 Discussion

4.1 Post-entrapment Modification

Modification processes affecting MIs post-entrapment have been studied extensively by many authors (Baker, 2008; Bucholz et al., 2013; Cottrell et al., 2002; Danyushevsky et al., 2000, 2002; Gaetani et al., 2012; Gaetani & Watson, 2000, 2002; Kent, 2008; Kress & Ghiorso, 2004; Nielsen et al., 1998; Portnyagin et al., 2008; Qin et al., 1992; Schiano & Bourdon, 1999; Sobolev, 1996; Spandler et al., 2007). The two main ways in which MIs can be chemically modified post-entrapment are by: (1) nucleation and growth of the host mineral on the MI walls by reaction between the edge of the melt inclusion and host clinopyroxene, or nucleation and growth of phases within the MI that are different than the host mineral, and (2) diffusive re-equilibration of the MI with the host crystal. These modification processes and their effects on the MIs of this study are evaluated below.

4.1.1 Diffusive Modification

If diffusive re-equilibration has played an important role in modifying the melt inclusion compositions, we might expect a correlation between MI size and composition, with the smallest inclusions closest to equilibrium due to greater diffusive equilibration compared with large melt inclusions (e.g., Baker, 2008; Cottrell et al., 2002; Nielsen et al., 1998; Qin et al., 1992; Sinton et al., 1993). Of course, this is dependent on the fact that the MIs have not reached equilibrium among all inclusions, which in turn is dependent on magmatic residence times. Magmatic residence time calculations, based on Fe/Mg inter-diffusion within clinopyroxenes from the ALIA-D115-18 lava, show residence times about on the order of <100 years from clinopyroxene crystallization to eruption (Adams et al., 2021). This rather short residence time is insufficient to allow incompatible and most major elements in the MIs to fully re-equilibrate with their host based on diffusion calculations. To elucidate this fact, Figure 3 shows a series of plots of melt inclusion size versus composition for selected major and trace elements, for several incompatible (e.g., K2O, Cl, La, and Sr) and two compatible (e.g., CaO and Sc) elements (or element oxides) in clinopyroxene. These plots show no clear correlations. Additional comparisons (not shown) of other chemical species, including both major and trace elements, also show a lack of correlation with melt inclusion size. This is consistent with the argument that magmatic residence times were very short and thus, PED of the melt inclusions has been negligible.

Details are in the caption following the image

Melt inclusion size versus composition. If diffusive modification has played a significant role in modifying melt inclusion chemistry, a linear trend will be observed when comparing melt inclusion incompatible elements and size, as smaller inclusions will be affected more rapidly by diffusion compared with larger inclusions. Melt inclusion sizes and reconstructed compositions can be found in Data Sets S2 and S6, respectively.

Additionally, boundary-layer driven diffusion can also be assessed as a means of modifying MI composition using a simple test of plotting element diffusivity in melt versus trace element standard deviation measured using the composition of all 22 MIs (Kent, 2008; Michael et al., 2002). Accordingly, fast diffusing elements should show little variation (low standard deviation) because diffusion will rapidly reduce (“smooth out”) trace element variability in the clinopyroxene, while slow diffusing elements should show the most variability (high standard deviation). We test this in Figure S11 with a comparison of compositional standard deviation versus known diffusivities in trachytic melts at 1150°C for MI trace elements, and we observe no distinct correlations. This also supports the hypothesis that diffusion has played a negligible role in modifying the MIs and supports previous work from Adams et al. (2021) that magmatic residence times were too short for diffusion to have modified the bulk composition of the melt inclusions significantly.

4.1.2 Post-entrapment Crystallization

Of the melt inclusions presented in this study, 20 out of 22 show 2 vol% or less crystallinity. The crystal volume estimate was made assuming an ellipsoidal geometry for both the MI and the crystals. Of the 20 MIs with 2 vol% crystallinity or less, two show <<1 vol% crystal inclusions, but contain vapor bubbles. These two MIs (mt2gr9 and mt2gr8b; Figures S10i and S10h, respectively) are also the only two MIs that are surrounded by clinopyroxene “halos," which are rings of clinopyroxene—ranging from ∼10 to 30 µm and encircle the MIs—with distinct mean atomic number observed only in back-scatter electron (BSE) imaging. These halos could be related to PEC. Compositional gradients in the clinopyroxenes going away from the MIs, including clinopyroxenes hosting halos (discussed in Text S1 and denoted by a gray dashed line in Figures S3 and S4), do show some evidence of compositional variability related to PEC. However, those inclusions that do show gradients consistent with PEC have very similar major element chemistry to those that do not show evidence of PEC, suggesting PEC is not the primary process for generation of trachytic compositions in the MIs, but we do acknowledge that PEC has occurred. Two MIs obviously exhibit evidence of PEC at the inclusion-clinopyroxene boundary: in one inclusion, this resulted in formation of orthopyroxene (melt inclusion Mt1gr12; see Figure S10e), and in another inclusion, the result is a jagged MI rim surrounded by compositionally distinct clinopyroxene (indicated by the brighter regions surrounding the MI in the BSE image—not to be confused with the previously descrived “halos” which are much larger and completely surround the MI—compared with the slightly darker host clinopyroxene) that returns to the host clinopyroxene composition with distance from the MI edge into the clinopyroxene (Mt2gr23; see Figure 1e).

Of the two MIs that show greater than 2 vol% crystallinity, one shows about 2.5 vol% of predominantly ilmenite (Mt2gr44: Figure 1b). The second of the two MIs with >2 vol% crystallinity (Mt1gr11; Figure 1a) contains crystals of Fo73 olivine + ilmenite + clinopyroxene representing a crystallinity of ∼10 vol% of the MI. However, we argue these were not crystallized directly from the melt inclusion post-entrapment, but rather were trapped along with the melt simultaneously. In this single MI that exhibits the olivine, ilmenite, and clinopyroxene, the crystals appear isolated within the bulk volume of the MI and apparently have not grown from the MI walls. Observing the MI in 3D would help determine if the apparent isolation is definitive (e.g. Kent, 2008; Kress & Ghiorso, 2004; Roedder, 1984; Sobolev, 1996). However, perhaps more significantly, if these crystals did precipitate directly from the melt inclusion following entrapment, one expects that the MI would significantly differ in major element composition compared to other melt inclusions with little or no daughter crystals; this is not the case. Furthermore, the presence of Fo73 olivine in the trachytic melt inclusion provides further evidence that the olivine did not grow post-entrapment, as the olivine is out of equilibrium with the trachytic melt (and thus, the hosting clinopyroxene) according to the MELTS MI reconstruction calculations, which found no stable olivine in the equilibrium assemblage at a range of pressures, temperatures, and water concentrations tested. Additionally, no olivine phenocrysts exist in the ALIA-D115-18 whole rock, suggesting olivine was not crystallizing from the magma that gave rise to this sample, and further, the olivines are similar in composition to the range of olivine compositions found in forsteritic olivines in other more primitive Samoan shield lavas (see Jackson & Shirey, 2011). Lastly, the olivine and ilmenite within this inclusion appear visually to be intergrown suggesting both the ilmenite and olivine were trapped together and did not grow from this MI. To summarize, all but two MIs show much less than 2 vol% crystallinity with two showing 2–10 vol%. We argue that some of the crystals in the MIs are not related to PEC but were instead present at the time of magma mixing and MI entrapment.

Nevertheless, based on the reasoning above, it is clear that some PEC has occurred. This is perhaps most apparent in the fact that the original melt inclusions are not in equilibrium with their host-clinopyroxenes. Approximately 14%–24 % addition of clinopyroxene back into the “raw” MIs was needed to bring them into equilibrium with their host clinopyroxene (see Data Set S6 for exact values). However, again, given the reasoning above, we argue that the primary process controlling the compositional variability observed in the MIs is not PEC. The reconstructed compositions can be seen (in addition to the measured MI compositions) in Figures 4 and 5 and Data Set S6.

Details are in the caption following the image

Major elements versus SiO2 showing the relationship between the ALIA-D115 lavas (red squares; data from Jackson et al., 2007; the ALIA-D115-18 sample is outlined in bold), other Samoan whole rock lavas (orange diamonds), Samoan pillow glasses (purple circles), and the reconstructed ALIA-D115-18 melt inclusions (blue circles). Also plotted for reference are the original MI compositions (grey circles) in addition to the inferred mafic (black star) and inferred silicic (low MgO silicic EM2 melt endmember 1 = green star, high MgO silicic EM2 melt endmember 2 = pink star) mixing endmembers calculated by Adams et al. (2021). The high silica (trachytic), high-87Sr/86Sr Savai'i xenolith glass from Hauri et al. (1993) is also shown for reference as a dark blue triangle. The ALIA-D115 lavas show a mixing trend extending from more primitive Samoan lavas and pillow glasses to the melt inclusions suggesting the melt inclusions could be a mixing endmember. The rest of the Samoan lavas (orange diamonds) that are more evolved show fractional crystallization trends. In addition, a mixing line (gray hatched line) has been calculated using a melt inclusion (mt2gr21; SiO2 = 65.4 wt%, TiO2 = 0.98 wt%, Al2O3 = 16.2 wt%, FeO = 1.2 wt%, MgO = 1.5 wt%, CaO = 4.4 wt%, and Na2O + K2O = 9.7 wt%) and a primitive Samoan whole rock lava composition (OFU-04-05; SiO2 = 45.7 wt%, TiO2 = 5.0 wt%, Al2O3 = 12.5 wt%, FeO = 12.4 wt%, MgO = 8.7 wt%, CaO = 11.4 wt%, and Na2O + K2O = 3.7 wt%) (Hart & Jackson, 2014) as endmembers. The mixing line captures the ALIA-D115 lavas quite well. The contribution of a trachytic MI composition to the mixture that is required to generate the most radiogenic (highest 87Sr/86Sr) ALIA-D115 lava is about 70%–80%, consistent with that proposed by Adams et al. (2021).

Details are in the caption following the image

Select trace element plots comparing the ALIA-D115 lavas (red squares; data from Jackson et al., 2007; the ALIA-D115-18 sample is outlined in bold), other Samoan whole rock lavas (orange diamonds), Samoan pillow glasses (purple circles) and the ALIA-D115-18 melt inclusions (blue circles). Trace element data show the ALIA-D115 mixing trend is very clearly extending between more primitive Samoan whole rock lavas (or glasses) and the trend of the MIs of this study. The Samoa whole rock lavas show a clear fractional crystallization trend (positive slope) while the ALIA-D115 lavas form an orthogonal trend (negative slope) extending toward the MIs. Using the same two mixing endmembers as in Figure 4: the mt2gr21 melt inclusion (Y = 17.4, La = 87.0, Nd = 54.0, Ce = 163.5, K = 55,417) and a primitive Samoan whole rock lava (OFU-04-05; Y = 32.0 ppm, La = 38.5, K = 11,539 ppm, Nd = 44.1, Ce = 81.9). A mixing line has been calculated for trace elements (gray hatched line). The contribution of a trachytic MI composition to the mixture that is required to generate the most evolved and highest 87Sr/86Sr lava from Samoa is about 70%–80%, consistent with that proposed by Adams et al. (2021).

4.2 Origin of Trachytic Melt Inclusions

Several recent studies have shown the ALIA-D115 lavas—a suite of lavas from the same submarine dredge of a Samoan volcano (Savai'i) as the ALIA-D115-18 lava studied here—to be a result of magma mixing between a low 87Sr/86Sr, mafic endmember and a high 87Sr/86Sr, evolved composition representing the EM2-derived endmember (Adams et al., 2021; Edwards et al., 2019; Jackson et al., 2007). While EM2 has high 87Sr/86Sr and EM2 lavas at Samoa are relatively evolved, previous work has shown that EM2 cannot be derived from shallow contamination by assimilation of sediments (Jackson et al., 2007) or oceanic crustal material, given that oceanic crust has insufficient radiogenic 87Sr/86Sr to produce the values of the ALIA-D115 lavas. Furthermore, the Pb isotopic compositions of modern marine sediments in the Samoan region and the ALIA-D115 lavas form a diverging, non-overlapping trend such that the highest 87Sr/86Sr lavas are furthest from the marine sediment field (see the following work for more detail: Jackson et al., 2007; Edwards et al., 2019). Adams et al. (2021) present high precision Sr and Nd isotope measurements on individual clinopyroxene porphyrocrysts from the ALIA-D115-18 whole rock lava, and reveals the highest 87Sr/86Sr (0.723888) ever recorded for the oceanic mantle (for comparison, the highest 87Sr/86Sr for an ALIA-D115 whole rock is 0.720469; Jackson et al., 2007). Thus, the EM2-derived endmember melt must have 87Sr/86Sr at least as high as this value. Adams et al. (2021) advances the knowledge of the EM2-derived mixing endmember by using the compositions of ALIA-D115 whole rock lavas, together with magma mixing theory, to derive estimates of the compositions of the two magma mixing endmembers contributing to the ALIA-D115 magmatic suite. The generation of the ALIA-D115 lava compositions requires a mafic, low 87Sr/86Sr endmember and an endmember with high 87Sr/86Sr and high SiO2 (i.e., the EM2-derived endmember). Potential EM2-derived mixing endmembers calculated by Adams et al. (2021) are trachytic in composition (SiO2 = 62.2–65.5 wt%, MgO = 0.5–2 wt%, Na2O + K2O = 10–11 wt%) with estimated 87Sr/86Sr values between 0.7249 and 0.7316. Remarkably, the range of EM2-derived endmember melt compositions from Adams et al. (2021) overlap the reconstructed trachytic MIs presented in this study (Figure 4). Thus, the question remains: what is the origin of these trachytic MIs? Are they melts formed by fractional crystallization of a mafic parental melt that was characterized by radiogenic 87Sr/86Sr? Or do they represent partial or total melts of a more silica-rich and higher 87Sr/86Sr source material? In the case of the latter, the reactivity of such relatively high-silica melts in an ultramafic (peridotitic) mantle is problematic for their mobility. In either case, the MIs were captured during the magma mixing process that formed the ALIA-D115 lavas.

Results of this study suggest the reconstructed MIs are remarkably similar to an independently derived EM2-mixing endmember proposed by Adams et al. (2021). In order to further assess this hypothesis, we can compare the MI chemistry with that of the ALIA-D115 whole rocks, other Samoan whole rock lavas, and Samoan pillow glasses to better understand the mixing process. Figure 4 shows major element variation of the ALIA-D115-18 MIs with whole rock lava compositions from the ALIA-D115 and other Samoan whole rock lavas and pillow glasses. The ALIA-D115 lavas form an array extending from the more primitive Samoan whole rocks toward the ALIA-D115-18 melt inclusions. Also shown are two EM2 mixing endmembers proposed by Adams et al. (2021), one a lower MgO trachytic endmember (green star), a higher MgO tracytic endmember (pink star), and a calculated mafic mixing endmember (black star). The major element composition of the EM2-derived mixing endmembers overlaps with the major element composition of the melt inclusions of this study, with the exception of FeO, which is about 1.5–3.0 wt% lower in the ALIA-D115 MIs than the inferred EM2-derived endmember. This suggests that the MIs could in fact be representative of the EM2-derived mixing endmember predicted by Adams et al. (2021). However, in major element space (Figure 4), the trend formed by ALIA-D115 whole rock lavas could be mistaken for a fractional crystallization trend (with the mineral assemblage olivine + clinopyroxene + plagioclase + ilmenite). Nevertheless, unequivocal evidence of the MIs being a mixing endmember of the ALIA-D115 lavas is seen in trace element and isotope space (Figures 5 and 6).

Details are in the caption following the image

87Sr/86Sr versus SiO2 and K2O comparing Samoan whole rock lavas (orange diamonds), ALIA-D115 whole rock lavas (red squares; data from Jackson et al., 2007; the ALIA-D115-18 sample is outlined in bold), Samoa pillow glass (purple circles), a high silica (trachytic), high 87Sr/86Sr Savai'i xenolith glass (Hauri et al., 1993; dark blue triangles), and the inferred mafic (black star) and two silicic mixing endmembers (low MgO EM2 silicic endmember 1 = green star, high MgO EM2 silicic endmember 2 = pink star) calculated by Adams et al. (2021). The figure allows an estimate of the 87Sr/86Sr composition of the melt inclusions (not yet measured) using whole rock trends and melt inclusion major element compositions. The gray shaded region is the range of SiO2 and K2O values in the reconstructed MIs of this study. The thick gray line within the gray shaded region is the average MI value. The dashed line is a mixing line through the ALIA-D115 lavas, which is shown to illustrate an extension of this mixing array into the MI composition space (gray shaded region). The mafic mixing endmember is the same Ofu lava used in the previous mixing figures and the silicic mixing endmember uses the average MI SiO2, K2O, and Sr concentration to estimate the 87Sr/86Sr value. The dashed mixing line (showing little curvature due to the similarity of Sr concentration between the two mixing endmembers) meets the average MI value at a 87Sr/86Sr value of ∼0.725 for both the SiO2 and K2O plots. This value is within the range of values calculated by Adams et al. (2021), but is most similar to the value calculated for the higher MgO EM2 endmember of 0.7249. We chose to use an MI composition as an EM2 mixing endmember instead of those calculated by Adams et al. (2021) because we wanted to independently evaluate whether the MIs correspond to these previously calculated endmembers.

In comparison of the LREE and K with Y (Figure 5), the Samoan whole rock lavas and pillow glasses form a linear, positively sloping fractional crystallization trend, whereas the ALIA-D115 lavas form an orthogonal (and negatively sloping) trend connecting the array formed by Samoan lavas with the array formed by the Samoan MIs. As an example, Figure 5 presents a comparison of K and Y (where both elements are incompatible in the suite of phases observed in the host lava and melt inclusions), where the ALIA-D115 lavas show a negatively sloping trend, nearly orthogonal to that of the other Samoan lavas and pillow glasses, that cannot be the result of fractional crystallization. This is because fractional crystallization will generate positively-sloping trends for incompatible elements like K and Y (as can be seen for the other Samoan lavas and pillow glasses). Therefore, we argue that the trend formed by the ALIA-D115 lavas is the result of magma mixing in which the trachytic MIs represent one mixing endmember and the Samoan basalts reflect the other mixing endmember; this is consistent with the results of Adams et al. (2021), who identified the ALIA-D115 lavas to be the result of mixing between a mafic (low 87Sr/86Sr) and a silicic (high 87Sr/86Sr) endmember. Although we cannot yet measure the 87Sr/86Sr in the MIs due to their small size, their isotopic compositions can be inferred by extrapolating existing trends in the Samoan magmatic suite. Figure 6 shows 87Sr/86Sr versus SiO2 and K2O for the ALIA-D115 whole rocks, other Samoan whole rocks, and two mixing endmembers of Adams et al. (2021). The data support a mixing trend for the ALIA-D115 whole rock lavas, where lavas associated with higher SiO2 and K2O have higher 87Sr/86Sr. Using this trend, we can extrapolate 87Sr/86Sr compositions by employing the SiO2 and K2O concentrations of the melt inclusions. An extrapolation of the ALIA-D115 mixing trend into the range of SiO2 and K2O observed in the MIs show that on average, the MIs could have a 87Sr/86Sr of ∼0.725, which is within range of the values predicted by Adams et al. (2021; Figure 6). Lastly, further evidence that the MIs could represent a mixing endmember is that one melt inclusion hosts two Fo73 olvines, not in equilibrium with a trachytic liquid, and indicative of interaction with a more mafic magma. Thus, there are multiple lines of evidence that interaction of mafic and silicic magmas played a central role in the formation of the ALIA-D115 lavas, and the melt inclusions provide us with the first glimpse of actual melts that match the trachytic composition of the EM2 endmember melt predicted by Adams et al. (2021).

After reconstruction for PEC, the Si-rich nature observed in the MIs represents the Si-rich nature of liquids captured during the formation of the ALIA-D115-18 whole rock and hosting clinopyroxenes. If the MIs are in fact melts that ultimately derived from an EM2 source, what process has controlled their composition? There are two obvious potential mechanisms that could explain the observed compositions: 1) the EM2-derived liquids began as mafic primary liquids that experienced significant fractional crystallization prior to entrapment and the MIs were captured by growing clinopyroxene, or 2) the MIs formed as a result of partial melting, to different degrees (highly affecting the trace element variability), of the EM2-derived source directly, with minor subsequent pre- or post-entrapment fractional crystallization (i.e., the primary mantle melts were silicic).

In reality, it is likely a combination of both mechanisms, but a few points seem clear: 1) the enrichment in LREEs compared with HREEs is indicative of garnet in the petrogenesis of the melt inclusions (Figure 2), and 2) if the melt inclusions were a result of originally mafic melts that underwent significant fractional crystallization, we expect that their trace element signatures would reflect this process. Trace element signatures in Figure 2 show that the MIs exhibit depletions in Nb and Ti, which could be related to removal of Ti-rich phases, like ilmenite (i.e., during crystallization after partial melting from the source). The Nb and Ti depletion could also be due to ilmenite crystallization after entrapment, however, PEC of ilmenite appears to be minimal in these inclusions with the exception of one MI (mt1gr11) in which the ilmenite appears to have intergrown with olivine and been trapped along with the MI. Additionally, the host clinopyroxenes themselves are depleted in Nb and Ti, suggesting ilmenite removal prior to their growth. The MIs are depleted in Sc (Data Set S2) relative to the ALIA-D115 lavas (Jackson et al., 2007), suggesting that they could have had significant clinopyroxene removal as well, during pre- and post-entrapment. This seems reasonable as ilmenite and clinopyroxene are common phases present in the ALIA-D115 lavas. However, another very common phase in the ALIA-D115 lavas is plagioclase, which is never seen in any of the MIs of this study. If the MIs had undergone significant differentiation prior to entrapment, plagioclase would likely have been removed, but this is certainly not the case as the MIs are quite enriched in Sr: while some inclusions exhibit slight negative Sr anomalies—the hallmark of plagioclase crystallization—many do not (some show positive anomalies, some show no anomaly, and others show negative anomalies; Figure 2). Further, we note this is likely not a result of pressure or water effects suppressing plagioclase crystallization according to fractional crystallization simulations using MELTS (performed for the PEC calculations as described in the Methods section) and the reconstructed MI compositions as the starting bulk compositions. MELTS simulations were performed at a range of pressures and water concentrations (where water concentrations were those of the reconstructed MI bulk compositions and show plagioclase saturation shortly after clinopyroxene crystallization begins in all simulations. More specifically, plagioclase saturates ∼5 oC–50 oC after initiation of clinopyroxene crystallization, corresponding to ∼2–15% fractionation of the original liquid bulk composition, with the variability in saturation temperature depending predominately on the water concentration of the MI (i.e., as MI water concentrations increase, the change in temperature between clinopyroxene and plagioclase saturation increases, with plagioclase saturating at increasingly lower temperatures after clinopyroxene saturation at the liquidus). This suggests that plagioclase removal is not responsible for the high silica nature of the inclusions. Thus, it seems that the argument that the MIs owe their ultimate origin to mantle-derived mafic melts that underwent significant differentiation is ambiguous, and does not fit the MI geochemistry. Alternatively, if the trachytic liquids in the MIs are a direct result of partial melting of a mantle source (i.e., the primary mantle-derived melts were trachytic), how have these liquids survived without reacting with the surrounding peridotite mantle during ascent through the asthenosphere and oceanic mantle lithosphere?

There are a few possibilities that could explain the origin of the trachytic MIs. One of the simplest explanations is that they are derived from partial melting of a sediment or sediment-enriched source. These melts then mix with more primitive Samoan basaltic melt (derived by partial melting of peridotite) to create the ALIA-D115 mixing trend. Problematically, this requires silicic melts to remain isolated from the ambient ultramafic mantle—including during ascent through >60 km of Pacific mantle lithosphere—until time of mixing with the basaltic magmas in shallow magma chambers. It has been previously documented that siliceous melts are highly reactive with ultramafic bulk compositions (Arai et al., 2003; Bianchini et al., 2011, 2015; Shimizu et al., 2004). Nevertheless, there is evidence for isolation of silica-rich veins in peridotite mantle xenoliths found in alkali basalts from Tallante, Spain. The high-silica narrow dikes (quartz-diorite bulk composition) are thought to be the result of slab melts that are able to remain isolated form the surrounding mantle through an initial reaction between the high silica activity silicate melt and low silica activity olivine creating an orthopyroxene reaction barrier at the contact, via the classic reaction relationship of SiO2 + Mg2SiO3 = 2MgSiO3 (i.e., a reaction between the silicic melts and the peridotite wallrock;Arai et al., 2003; Bianchini et al., 2011, 2015; Shimizu et al., 2004). However, the Tallante xenoliths record a process that occurred at the cm-scale. In order for the process to transport trachytic liquids out of the mantle source, it would have to operate on a larger scale, allowing the trachyte liquid + peridotite reaction to create opx-rich wall rock to buffer the silicate melts, allowing transport >60 km through the peridotite oceanic mantle lithosphere without completely reacting with the surrounding peridotite. However, if this process does occur at Samoa, where conduits transporting trachytic melts are mantled by opx to facilitate melt transport to the surface, then a key question is why do we not see these silicic primary liquids erupted at other hotspot locations, like Hawaii, where partial melting of Si-saturated eclogite yields Si-rich liquids (Hauri, 1996; Sobolev, 2007; Sobolev et al, 2000, 2005; Yaxley & Sobolev, 2007).

Another explanation for the formation of the trachytic MIs is a scenario in which ancient recycled sediment, captured by the Samoan plume source, partially melts and fertilizes the surrounding mantle to form a hybridized peridotite + sediment mixture. Thermodynamic phase equilibria modeling by Adams et al. (2021) of various hybrid peridotite + sediment bulk compositions suggests that the partial melts are basanitic to phonolititc (depending on the extent of melting) and that subsequent differentiation of these melts does not lead to trachytic liquids. That being said, the exploratory model space is quite large and these models capture a small portion of conceivable starting compositions (including the fraction of recycled sediment, fraction of peridotite, and/or fraction of recycled mafic material), pressures, percentage of source melting prior to differentiation, redox conditions, total volatiles, etc. There are some experimental studies, however, that show partial melting of hybridized peridotite + sediment-like bulk compositions can in fact produce trachytic liquids similar to the MIs of this study (Gao et al., 2019; Wang et al., 2017). However, the same problem noted earlier remains: how are silicic liquids transported through the peridotitic mantle lithosphere to the surface without reacting with the wall rock?

Lastly, there is evidence that the EM2 source has been enriched not only by addition of sediments, but also carbonatitic material. Carbonatitic metasomatism beneath Samoa has been proposed by many authors (Ashley, 2019; Ashley et al., 2020; Hauri et al., 1993; Hauri & Hart, 1994). Evidence for carbonatite metasomatism is clearest in Samoan mantle peridotite xenoliths that show evidence of interaction with carbonatitic liquids based on their LREE enrichment and Zr and Ti depletions in clinopyroxene (Hauri & Hart, 1994; Hauri et al., 1993). More recently, Ashley et al. (2020) performed an analysis of water systematics and major and trace element characteristics in Samoa peridotite xenoliths (focusing on orthopyroxenes) revealing variable depletions in water contents of nominally anhydrous minerals, and in addition to their trace element characteristics, indicate metasomatism of carbonatitic and silicate liquids. Further, previous studies have suggested that partial melting of mantle enriched with carbonatitic liquids can result in phonolitic to trachytic liquids (Ashwal et al., 2016; Hauri et al., 1993; Loges et al., 2019; Tsuno and Dasgupta, 2011). To this end, Hauri et al. (1993) found trachytic glass (69% SiO2, 8% total alkalis; see main text Figures 3-6 for plotted compositions of the glass) hosted in a carbonatite metasomatized peridotite xenolith from the Samoan island of Savai'i—the same volcano that erupted the ALIA-D115-18 melt inclusions examined here—which they interpret to be the product of a reaction between carbonatitic melt with peridotite. Critically, this liquid is nearly trachytic and has elevated 87Sr/86Sr (0.712344)—further suggesting a potential role for carbonatitic metasomatism in the origin of EM2 and supporting the presence of high silica melts in the mantle. An important question to address is whether sufficient trachytic melt with high 87Sr/86Sr values can be generated in metasomatic peridotites, and sequestered into crustal magma chambers, to give rise to the Si-rich Samoan EM2 endmember lavas and the trachytic melt inclusions they host.

5 Conclusions

Three key questions were raised in this work: 1) Have the trachytic MIs been significantly modified by post-entrapment processes, 2) how are the trachytic MIs related to their host whole rock and the other ALIA-D115 lavas, and 3) what is the origin of the trachytic MIs? We believe the MIs, after compositional reconstruction that involves clinopyroxene addition, represent the most extreme EM2 compositions in global OIB. In both major and trace element space, the MIs lay at one end of the ALIA-D115 mixing array, suggesting the silicic MIs are the high 87Sr/86Sr mixing endmember with an affinity to EM2. This idea is consistent with previous studies that the melts derived from the EM2 mixing endmember beneath Samoa are highly evolved and geochemically enriched (high in 87Sr/86Sr). Projecting the ALIA-D115 mixing trend in 87Sr/86Sr versus SiO2 space and in 87Sr/86Sr versus K2O space shows that the trend intersects the average MI composition at 87Sr/86Sr values of ∼0.725 (consistent for both SiO2 and K2O), which is within the range of the independently calculated values predicted by Adams et al. (2021) of 0.7249–0.7316. Trace element signatures suggest the MIs could have undergone some fractionation prior to entrapment in the clinopyroxenes. Alternatively, the MIs could be representative of melts derived directly from the EM2 source with little fractionation prior to entrapment, but this requires that the trachytic liquids were somehow preserved against reaction with peridotitic mantle during vertical transport through 60 km of oceanic mantle lithosphere. Further partial melting thermodynamic calculations of potential EM2 source compositions as well as laboratory studies of partial melting of various mafic and ultramafic sources subject to sediment addition or modification by metasomatism should be further explored to better understand the formation of these trachytic melt inclusions.


M. G. Jackson acknowledges the mentorship and generosity of Erik H. Hauri, who provided unfettered access to microbeam facilities during his 16-months tenure as a post-doc at DTM. M. G. Jackson also acknowledges support from NSF grants OCE-1912931, EAR-1900652, and OCE-1736984. The authors thank Stan Hart and Hubert Staudigel for generously providing us with a Samoan sample for this work. J. V. Adams acknowledges Gareth Seward and John Cottle for help with major and trace element analyses on the clinopyroxenes. All authors would like to thank Michael Rowe, Leonid Danyushevsky, and one anonymous reviewer for their extremely constructive and helpful reviews that greatly improved this manuscript. The authors declare no conflicts of interest.

    Data Availability Statement

    All geochemical data from this study are available in the Supporting data sets of this article and can also be found on the IEDA EarthChem online database at the following link: https://doi.org/10.26022/IEDA/111729.