Signature of deep mantle melting in South Iceland olivine

We present new high-precision major and trace element data on olivine macrocrysts from various volcano-tectonic settings in Iceland and use these data as a proxy for mantle mode and melting conditions. Within individual sampling sites examined (seven lavas and one tephra) olivine-dominated fractional crystallization, magma mixing and diffusive re-equilibration control observed variability in olivine composition. High-pressure fractional crystallization of clinopyroxene may have lowered Ca and increased Fe/Mn in one olivine population and subsolidus diffusion of Ni and Fe–Mg affected the mantle-derived Ni/Fo ratio in some compositionally zoned olivine macrocrysts. Interestingly, magmas erupted at the southern tip of the Eastern Volcanic Zone (SEVZ), South Iceland, have olivines with elevated Ni and low Mn and Ca contents compared to olivines from elsewhere in Iceland, and some of the SEVZ olivines have relatively low Sc and V and high Cr, Ti, Zn, Cu and Li in comparison to depleted Iceland rift tholeiite. In these olivines, the high Ni and low Mn indicate relatively deep melting (Pfinal > 1.4 GPa, ~ 45 km), Sc, Ti and V are compatible with low-degree melts of lherzolite mantle, and elevated Zn may suggest modal (low-olivine) or geochemical (high Zn) enrichment in the source. The SEVZ olivine macrocrysts probably crystallized from magmas derived from olivine-bearing but relatively deep, enriched and fertile parts of the sub-Icelandic mantle, and indicate swift ascent of magma through the SEVZ lithosphere.


Icelandic mantle
Although many of the details remain obscure, Earth's mantle is evidently heterogeneous in mineralogy and chemical composition (Allègre and Turcotte 1986). A major source of this heterogeneity is subducted oceanic crust that contributes to the mantle source of oceanic island basalts (Hofmann and White 1982;Hauri 1996;Chauvel et al. 2008). In Iceland, there is abundant isotopic evidence for recycled oceanic crustal material in the upper mantle Chauvel and Hémond 2000;Kokfelt et al. 2006;Peate et al. 2010). Anyhow, isotope systematics do not yield straightforward information about mantle lithology, and it is yet unclear whether the recycled oceanic crust resides in the mantle as a discrete rock type, i.e. pyroxenite or eclogite, or if it has been completely mixed with the dominant mantle peridotite (Green and Ringwood 1963).
Insights into the modal composition of the mantle can be acquired by studying the major and trace element Communicated by Mark S Ghiorso.

Electronic supplementary material
The online version of this article (https ://doi.org/10.1007/s0041 0-019-1580-8) contains supplementary material, which is available to authorized users. composition of magmas that represent partial melts of the mantle. This is, however, prone to uncertainty, as the nature of mantle melts is dependent on not only the modal composition of the mantle but also melting temperature (T), pressure (P) and the degree of melting (F). In addition, mantle-derived magmas are usually modified in the crust by low-P fractional crystallization, magma mixing and crustal contamination, which may obscure the traits they inherited from the source. Nevertheless, with careful treatment of the data, the quantity of possible endmember mantle source rock-types can be estimated, and the inferred mantle components can be further assessed by comparing their assumed melt productivity to crustal thickness (Shorttle and Maclennan 2011;Brown and Lesher 2014;Shorttle et al. 2014;Lambart 2017). Using this approach, it has been established that melting of depleted lherzolite alone is unlikely to produce the compositional variability (especially the high Fe and low Ca) in Icelandic lavas (Shorttle and Maclennan 2011). A minor (4-15%) olivine-free pyroxenite or mixed pyroxenite-peridotite hybrid seem to be required in the mantle source (Shorttle et al. 2014).
Besides utilizing the major and trace element composition of magmas, the nature of the mantle source can be probed by analyzing forsteritic olivine macrocrysts. The composition of forsteritic olivine reflects the composition of its near-primary parental magma, and hence the mode (Sobolev et al. 2005) and/or melting conditions (Li and Ripley 2010;Matzen et al. 2013Matzen et al. , 2017a in the mantle source. In Iceland, olivine compositions have been found to be rather consistent with shallow melting of lherzolite (Herzberg et al. 2016), although some lavas have olivines with higher Ni than expected by models of fractional crystallization of peridotite-derived melts (Shorttle and Maclennan 2011;Herzberg et al. 2016;Neave et al. 2018). These Ni-rich olivines have been suggested to indicate melting of enriched, olivine-free or olivine-poor, mantle below Iceland (Sobolev et al. 2007(Sobolev et al. , 2008Shorttle and Maclennan 2011), or explained by mixing of variably differentiated lherzolitederived melts in the crust (Herzberg et al. 2016).
We present new olivine trace element data from various volcano-tectonic settings in Iceland, gathered using high probe current microprobe and laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) methods. The primary finding are the relatively Ni-enriched and Mn-and Ca-poor olivine macrocrysts from Eyjafjallajökull and Vestmannaeyjar volcanic systems, exceptional in the context of pre-existing data (Sobolev et al. 2007;Shorttle and Maclennan 2011;Neave et al. 2018). Some of these olivine crystals are also enriched in Ti, Zn, Cu, Li, Cr and depleted in Sc and V compared to olivines from rift zone tholeiite (Háleyjabunga). We propose that the parental magma of these olivine macrocrysts equilibrated at high temperatures and pressures with a fertile olivine-bearing mantle source.

Olivine as indicator of mantle mode and melting conditions
In mantle lherzolite, abundant residual olivine limits the Ni content of partial melts by effectively sequestering Ni. In contrast, low-degree partial melts of olivine-free pyroxenite are, in most cases, rich in Ni, and poor in Mn and Ca as garnet and clinopyroxene in the residuum preferentially sequester Mn and Ca, but not Ni. Ni, Mn and Ca are stored in early-formed Fo-rich olivine in proportion to their content in primitive melts, even if the melt is later modified by fractional crystallization and magma mixing. Because of this, trace elements in olivine should record the lithology of the mantle source. Sobolev et al. (2005) were the first to use high-Ni, low-Mn and low-Ca in olivine as an indicator of olivine-free pyroxenite in the mantle. Since this work, olivine composition has been used to estimate the mantle lithology for a wide range of volcanic provinces (e.g. Sobolev et al. 2007Sobolev et al. , 2008Herzberg 2011;Herzberg et al. 2014;Trela et al. 2015).
Using olivine as a mantle proxy is complicated by variation in T, P, and oxygen fugacity (fO 2 ) that affects element partitioning in mantle melting and subsequent olivine crystallization. Recent experimental work suggests that Ni and Mn distribution coefficients between olivine and melt ( D ol∕liq Ni , D ol∕liq Mn ) are, directly or indirectly, dependent on temperature of mantle melting (Li and Ripley 2010;Putirka et al. 2011;Matzen et al. 2013Matzen et al. , 2017a. D ol∕liq Ni is relatively low for high-temperature peridotite melting at high pressures, which results in partial melts relatively enriched in Ni. Consequently, when a deep mantle-derived magma rises to a lower temperature and pressure environment, D ol∕liq Ni is higher, resulting in crystallization of Ni-rich olivine (Matzen et al. 2013(Matzen et al. , 2017a. D ol∕liq Mn correlates positively with MgO in the melt, and MgO-rich melts of peridotite are generated at higher temperature; therefore, high-temperature mantle melts have lower MnO and they crystallize low-Mn olivine macrocrysts (Matzen et al. 2017b). This, and the global correlation between lithospheric thickness and olivine compositions, has sparked a debate on whether the variation in Ni and Mn in olivines reflect the depth of mantle melting or variation in mantle mode (Niu et al. 2011;Heinonen and Fusswinkel 2017;Putirka et al. 2018). Mn partitioning between partial melts and mantle residue may also be susceptible to water content in the mantle (Balta et al. 2011), and the extent of parental melt polymerization (NBO/T) has been proposed to affect Ni partitioning during olivine crystallization (Wang and Gaetani 2008). Also, the slower diffusion of Ni compared to Mg and Fe in olivine lattice during crustal storage can affect element ratios in zoned olivine (Lynn et al. 2017).
Because complexly influenced by parameters of melting and crystallization, olivine compositions need not to be directly related to the volume of a certain rock type in the mantle, e.g. olivine-free pyroxenite. Nevertheless, olivine macrocrysts are useful, as their composition echoes the evolutionary history of an individual melt, i.e. they record fractional crystallization and magma mixing, and the most primitive olivines in a system mirror the composition of the original mantle-derived melt in a way glass or wholerock composition of magmas nearly never do. Olivine composition is a high-detail mantle proxy, valuable when treated in view of other mantle source indicators, such as major, trace element and isotopic composition of magmas.

Sample locations
We analyzed olivines from eight locations, which represent varying volcano-tectonic environments in Iceland. These include Háleyjabunga and Mosfellsheiði lavas in southwestern Iceland; Kistufell hyaloclastite in the central highlands; Berserkjahraun lava of the Snaefellsnes Volcanic Zone (SVZ); and Stórhöfði, Hvammsmúli, Brattaskjól lavas and Eyjafjallajökull 2010 summit tephra from the southern tip of the Eastern Volcanic Zone (SEVZ, Table 1, Fig. 1). The majority of olivine trace element data gathered earlier from Iceland are from the neovolcanic lavas of SVZ and the Northern Rift Zone (Sobolev et al. 2007(Sobolev et al. , 2008. To gain new insights, we focused our sampling efforts on SEVZ (Fig. 1), where high-precision olivine analyzes had not been done earlier. Olivine major and trace element compositions from Háleyjabunga and Kistufell have been presented by Sobolev et al. (2007), and Háleyjabunga has also been revisited by Shorttle and Maclennan (2011) and Neave et al. (2018). Háleyjabunga and Kistufell are the end-members in terms of Ni and Mn content in the earlier Sobolev et al. (2007) highprecision data set from Iceland, and we re-analyzed them for assessment of data quality (cf. Supplementary Material).

Olivines from rift-zone tholeiites
The sampling locations Háleyjabunga, Mosfellsheiði and Kistufell represent the central parts of the mid-oceanic rift that lies across Iceland (Fig. 1); here the amount of decompression melting is generally the largest (Saemundsson 1979). Háleyjabunga is a 13 ka (Saemundsson 1995) lava shield situated near the tip of the Reykjanes Peninsula, known for its depleted isotopic and trace element character (Thirlwall et al. 2004). Our Háleyjabunga sample has the most magnesian olivines in our sample suite with forsterite (Fo) content up to 91 (Fo = 100 × atomic Mg/(Mg + Fe 2+ ). Mosfellsheiði is a younger than 700 ka (Saemundsson et al. 2010) olivine-tholeiite lava flow field from the outskirts of Table 1 The sample locations in this study with a collection of available geochemical data (Jakobsson 1979;Wiese 1992;Jóhannesson 1994;Breddam et al. 2000;Thirlwall et al. 2004Thirlwall et al. , 2006Peate et al. 2009Peate et al. , 2010Saemundsson et al. 2010;Sigmarsson et al. 2011) See Supplementary Material for data sources affiliated to respective cells a Rock types according to Jakobsson et al. (2008). All samples are in fact subalkaline in composition b Eyjafjallajökull 2010 summit eruption tephra is a mixture of benmoreite and basalt  (Breddam 2002), close to the inferred center of the Iceland plume at the rift axis (Wolfe et al. 1997).
The Kistufell magma has been interpreted as plume-derived near-primary melt, and it has up to Fo 88.7 olivines (Breddam 2002). The exact age of Kistufell is unknown, but is probably close to the end of the last glacial period (Breddam 2002).

Olivines from off-rift volcanic zones
The sampled lavas and tephra from SVZ and SEVZ (Berserkjahraun, Hvammsmúli, Brattaskjól and Stórhöfði lavas, and Eyjafjallajökull 2010 tephra) were erupted from off-rift volcanic systems situated in areas of minimal crustal extension ( Fig. 1). Off-rift volcanoes are peripheral to the main rift zone and they produce alkaline or mildly alkaline lavas enriched in incompatible elements compared to rift tholeiites (Jakobsson 1972;Hemond et al. 1993). This indicates relatively small degree of mantle melting (Saemundsson 1979), while isotopic composition (especially high 206 Pb/ 204 Pb) suggests a more enriched mantle source (Kokfelt et al. 2006;Peate et al. 2010). This covariation of small degree mantle melting and enriched isotopic composition is commonly explained by a model where smaller degree melting allows preferential tapping of enriched and fusible veins set in a more refractory mantle (Chauvel and Hémond 2000;Fitton et al. 2003;Kokfelt et al. 2006).
The tectonic setting of the two major off-rift volcanic zones, SVZ and SEVZ (Fig. 1), is different. SVZ has been interpreted as a zone of weakness that is the remnant of the Snaefellsnes-Skagi paleo-rift (Hardarson et al. 1997), active from 15-20 to 10-12 Ma (Martin et al. 2011). In contrast, magmatism in South Iceland is related to the southward propagation of the Eastern Volcanic Zone (EVZ), which is the youngest (activated > 3 Ma, Martin et al. 2011) volcanically active zone in Iceland, and the most profuse magma producer in historical times (Thordarson and Larsen 2007). In this area, including the Eyjafjallajökull and Vestmannaeyjar volcanic systems, active extension is minimal. Of SEVZ and SVZ lavas, the latter have higher La/Yb N (2-12 vs 2-7), indicating smaller degrees of melting, as well as lower ε Nd , and lower 3 He/ 4 He (Peate et al. 2010).
Three locations from SVZ were analyzed for olivine trace elements compositions by Sobolev et al. (2007): Enni, Sydri-Raudamelur and Ydri-Raudamelur. Due to these SVZ olivine data already existing, we sampled only one SVZ lava, Berserkjahraun ( Fig. 1, Table 1). Berserkjahraun is one of the most primitive of SVZ lavas (Peate et al. 2010), erupted within the Ljósufjöll volcanic system, and inferred to be younger than 4000 a in age (Jóhannesson 1994). Our sample of Berserkjahraun lava has Fo 83.2-90.1 olivines.
Our SEVZ sample set includes olivine macrocrysts from three samples from Eyjafjallajökull volcanic system and one sample from Vestmannaeyjar volcanic system (Fig. 1, Table 1). Brattaskjól and Hvammsmúli samples are from macrocryst-rich transitional basalts (often referred to as ankaramites) from the southern slope of the volcano. Hvammsmúli is 600 ka in age (Wiese 1992), a 45 m thick outcrop that has been interpreted as a remnant of a lava lake (Loughlin 1995), whereas Brattaskjól is a less studied macrocryst-rich basalt 4 km east of Hvammsmúli. Our Hvammsmúli and Brattaskjól samples have up to Fo 89.8 and Fo 89.2 olivines, respectively.
Our "Eyjafjallajökull 2010" sample comprises forsteritic olivines from tephra of the Eyjafjallajökull 2010 eruption. This eruption was a mixed eruption of benmoreite and basalt, which carried olivine with diverse compositions (Fo 74-87 ). The 2010 summit eruption followed an effusive flank eruption on the Fimmvörðuháls pass, and it was the intrusion of Fimmvörðuháls parental magma into a silicic magma chamber that ignited the summit eruption (Sigmarsson et al. 2011). The olivine crystals of Fimmvörðuháls have been extensively studied (Sigmarsson et al. 2011;Viccaro et al. 2016;Pankhurst et al. 2018), and due to the co-genetic nature of Fimmvörðuháls basalt and Eyjafjallajökull 2010 tephra, they likely carry similar olivine macrocrysts.
Our olivine samples from Vestmannaeyjar were collected from the Stórhöfði lava flow field at the southern tip of Heimaey Island. According to Jakobsson (1968Jakobsson ( , 1979, Stórhöfði lava erupted during the Holocene and is slightly older than 5470 ± 160 years in age (Mattsson and Höskuldsson 2003). It is the most primitive lava flow field on Heimaey, compositionally similar to the Surtsey 1963-1967 eruption (Mattsson and Oskarsson 2005), and it has up to Fo 89 olivines. The Stórhöfði sample examined in this study is from a tube-fed lava flow related to a late phase of the eruption (Mattsson and Oskarsson 2005).

Electron probe microanalysis
The microprobe analyses of major and trace element compositions of olivine crystals were conducted at the Institute of Earth Sciences, University of Iceland using the JEOL JXA-8230 electron microprobe. Olivine macrocrysts were hand-picked from crushed and sieved (Ø = 0.1-4.0 mm) rock samples and mounted in 1-inch epoxy molds or glass slides, except for the Mosfellsheiði sample, for which olivines were analyzed directly from 1-inch glass slides made of polished rock. Polished sample surfaces were cleaned with ethanol and carbon coated before the electron probe and laser ablation analyses. For the high-precision trace element analyses of olivine cores, we utilized a modified version of the highprobe current protocol by Batanova et al. (2015) with an acceleration voltage of 20 kV and 500 nA beam current. These analyses were focused on forsteritic olivines using low brightness of BSE images as an indicator of high Mg content, and the analyses targeted the geometric, compositionally homogenous, center of the olivine crystals. The resulting instrumental precisions for Ni and Mn are 0.52% and 0.48% relative standard deviations, respectively. When measuring traverses across olivine zonation, we used shorter analysis time, 15 kV accelerating voltage and 20 nA probe current. This setup delivers precise Fo measurements (~ 0.5% standard error) but relatively low precision on trace elements, e.g. an average 6.5% standard error for Ni. Detailed information of analytical conditions and standards are given in Supplementary Material.

LA-ICP-MS analysis
LA-ICP-MS analyses were carried out at the Department of Geosciences and Geography, University of Helsinki using a Coherent GeoLas Pro MV 193 nm laser ablation system connected to an Agilent 7900 s quadrupole ICP-MS equipped with high-sensitivity ion lens configuration and Pt interface cones. The following isotopes were included in the measurement program: 7 88 Sr, 137 Ba. The olivine macrocrysts were ablated up to three times at the center of the crystals with a spot size of 44 μm and laser energy densities of 7 J/ cm 2 . Sample measurements were bracketed by analyses of reference materials NIST SRM 610 (Spandler et al. 2011) and GSE 1G (Jochum et al. 2005) at fluences of 10 J/cm 2 . External standardization using GSE 1G synthetic basalt glass yielded better reproduction of EPMA compositional data than NIST SRM 610, and therefore, all data reported here were externally standardized against GSE 1G. Data reduction was carried out using the SILLS software package (Guillong et al. 2008). To avoid problems arising from Fe-Mg zoning in some olivine grains and the different sampling spot sizes of LA-ICP-MS and EPMA, internal standardization was done using Si concentrations as determined by EPMA. Further information on the analytical procedure and complete summary of instrumental parameters is given in the Supplementary Material.

Compositional zonation in olivine
Olivine crystals in the analyzed samples vary in size (0.3-3.4 mm in diameter) and compositional zoning (Fig. 2). The largest crystals analyzed are from Hvammsmúli We analyzed the zoning patterns along rim-to-core traverses in olivines from Hvammsmúli, Brattaskjól and Mosfellsheiði. Nearly all of the analyses of Hvammsmúli (19/19), Brattaskjól (10/12) and Mosfellsheiði (3/3) revealed decoupling of Fo and Ni in the Mg-Fe zoned parts of the crystals (Fig. 3), a phenomenon earlier reported by Lynn et al. (2017) from Hawaiian olivines. In Hvammsmúli olivine 'Pos-1a_Ol1', the Ni content becomes constant 190 μm from the crystal edge, whereas Mg-Fe zonation extends 520 μm from the crystal edge to core (Fig. 3).
The broadest Mg-Fe zoning analyzed from Hvammsmúli olivines is 700 μm thick from rim to core (olivine Pos-1b_Ol15, Fig. 2c), while Ni zoning only extends ~ 500 μm from the rim (Fig. 2c). Zoning patterns in normally zoned Mosfellsheiði and Brattaskjól olivines are variable and thinner. Ni zonation extends 30-180 μm and 20-100 μm from the crystal edge towards the crystal cores for Mosfellsheiði and Brattaskjól olivines, respectively, while the width of the Mg-Fe zoned rims varies in the range 120-220 μm in Mosfellsheiði and 50-100 μm in Brattaskjól.

Major and trace elements in olivine
Major and trace element composition in the olivine macrocrysts examined is illustrated in Figs. 4 and 5, and the full dataset is included in the Supplementary Material. Olivine data from Kistufell and Háleyjabunga are found consistent with earlier analyses acquired by Sobolev et al. (2007) and Neave et al. (2018) (see Supplementary Material). The lowest Ni and highest Mn relative to Fo content were measured in olivines from Háleyjabunga and Berserkjahraun. Olivines from these two sampling locations define continuous linear trends on Mn vs. Fo and Ni vs. Fo plots (Fig. 4) and, at Fo 90 , both have ~ 2500 ppm Ni and ~ 1200 ppm Mn. Kistufell and Mosfellsheiði olivine macrocrysts have higher Ni and lower Mn, and define a single trend in the Mn-Fo space (Fig. 4a). Kistufell has primitive Fo 88.15 olivines, whereas Mosfellsheiði olivines are relatively evolved (Fo <86 ).
Compared to olivines from the rift-zones and SVZ, the SEVZ olivine macrocrysts have higher Ni and lower Mn contents at a given Fo content (Fig. 4) Selected olivine macrocrysts from Hvammsmúli, Brattaskjól, Kistufell and Háleyjabunga were also analyzed with LA-ICP-MS for concentrations of Li, Na, Al, P, Ca, Sc, Ti, V, Cr, Mn, Co, Ni, Cu and Zn. At a given Fo, SEVZ olivines (Hvammsmúli and Brattaskjól) have lower Sc, and higher Li, Ti, Cr, Cu, and Zn, than Háleyjabunga (Fig. 5). The separate trends of SEVZ and Háleyjabunga olivines are outside of the analytical uncertainty; however, the compositions of Kistufell olivine macrocrysts overlaps the complete compositional variability detected in many trace elements (Sc, Ti, V and Al; Fig. 5). There is no significant difference in Co content, at a given Fo, between the locations. Brattaskjól olivine macrocrysts stand out with elevated Cr and Na in relation to the other samples. Our analyses of Háleyjabunga olivines are in agreement with earlier measurements by Neave et al. (2018), although our analyses have marginally higher Co contents. In turn, our SEVZ olivines resemble Neave et al. (2018) Stapafell olivines, with the exception that SEVZ olivines have higher Ti, Zn and Cu contents (Fig. 5).

Ni, Mn and Ca in Iceland olivine macrocryst record
The major and trace element composition of Háleyjabunga and Berserkjahraun olivine macrocrysts resembles MORB olivines with low Ni and high Mn contents, and are compatible with the Herzberg (2011) model of olivine crystallizing from peridotite-derived melts and their derivate melts along Liquid Line of Descent (LLD, Fig. 6) at low pressure. Kistufell and Mosfellsheiði olivines, although richer in Ni and lower in Mn, are also consistent with peridotite melting, as long as melt mixing is allowed (red mixing lines in Fig. 6b). Mixing of primary and fractionated magmas can slightly elevate Ni in olivine relative to its Fo content (Herzberg et al. 2016). Ni contents in Mosfellsheiði olivines do not change over a range of Fo content (Fig. 6a), which may indicate cotectic crystallization of Ol + Plg + Cpx. Ca content in Háleyjabunga, Berserkjahraun, Kistufell and Mosfellsheiði olivine is compatible to Herzberg (2011) model of olivine crystallization from a peridotite source, except for a few Berserkjahraun olivines (Fig. 6c). In essence, the SVZ and rift-zone olivines in our dataset conform with lherzolite-sourced partial melts and the previous olivine data Some olivines from Eyjafjallajökull and Vestmannaeyjar volcanic systems at SEVZ, however, have higher Ni contents and Fe/Mn ratios than previously published from Iceland (Fig. 6a, b). High Fe/Mn and Ni correlate in the high-Fo SEVZ olivine macrocrysts, and some of the high-Ni and high-Fe/Mn olivines, especially from the Eyjafjallajökull 2010 tephra, have low Ca (Fig. 6c). Many of the SEVZ olivines are outside the compositional range expected to crystallize from peridotite-derived melts as defined by Herzberg (2011) and may imply melting of an olivine-free mantle source (Sobolev et al. 2005) or relatively deep mantle melting combined with shallow olivine crystallization (Matzen et al. 2013(Matzen et al. , 2017a. Trace elements concentrations in olivine couple strongly with Fo content. Thus, meaningful comparison of olivine compositions from different localities is achieved by normalizing the trace element contents by FeO or MgO, and by limiting the comparison to high-Fo olivines. In Fig. 7, we compare high-precision Fo >86 olivine compositional data from SEVZ and other localities in Iceland (Sobolev et al. 2007, and this study) by utilizing Kernel density estimation (Green et al. 1988) of the probability density distribution of olivine compositions. Although SEVZ olivine compositions overlap with olivine from elsewhere in Iceland (Rift-zones and SVZ), Fig. 7 exhibits clearly how SEVZ has higher Ni and lower Mn in relation to olivine Fo. The Iceland rift-zone and SVZ olivine Ni data is somewhat bimodal, with two Ni(Mg/Fe)/100 probability density peaks at ~ 0.6 and ~ 0.7. This is not an artefact of combining our and Sobolev et al. (2007) datasets, but the result of elevated Ni in some Icelandic olivine. Herzberg et al. (2016) explained the elevated Ni in some Iceland olivine by mixing of near-primary and evolved melts derived from the same initially low-Ni, high-Mn peridotite sourced partial melt (see fig. 7 in Herzberg et al. 2016). The bimodality in SEVZ data (Fig. 7b) is solely produced by the Eyjafjallajökull 2010 olivines that have high Fe/Mn.

Indications of melt mixing
The SEVZ high-Fo olivines exhibit large variability in Ni, Mn and Ca contents, which decreases with decreasing Fo (Fig. 6). This is most evident in Brattaskjól and Hvammsmúli olivines, with Fo <88.2 grains conforming to a crystal line of descent at low pressure and Fo 88.3-89.5 olivines exhibiting high variability in trace element contents (Fig. 8). In this high-Fo olivine population, there is an inverse correlation in Ni and Mn (R 2 = 0.498), and Ni and Ca (R 2 = 0.486). We interpret this compositional characteristic by mixing of near-primary mantle melts with initially varying Ni, Mn, and Ca.
The Hvammsmúli and Brattaskjól Fo >88.2 olivines may initially have had variable Fo contents, which may have been later suppressed by diffusive re-equilibration with a carrier melt during crustal storage (Costa and Morgan 2010). During this process, variation in some trace elements (e.g. Ca) could have been retained due to their slower diffusion compared to Mg-Fe (e.g. Dohmen et al. 2007). Therefore, in theory, the high variability in trace element compositions in similar Fo olivines in Hvammsmúli and Brattaskjól could be a consequence of this incomplete re-equilibration process. This model is, however, hampered by the similar diffusion rates of Mn and Mg-Fe (Dohmen et al. 2007). Mn concentrations should have been similarly suppressed during diffusive re-equilibration as Fo contents, and we do not see this in the data (Fig. 8). In addition, the detected variation in Ca in olivines is so large (~ 1100 ppm, Figs. 6 and 8b), that producing it by olivine fractionation is improbable. Due to this reasoning, we favor mixing of two melts with significantly different Ca, Mn and Ni contents as the explanation for the high variability in the SEVZ olivine trace element contents. For Brattaskjól and Hvammsmúli, the higher Fo mixing end member has high Ni and relatively low Mn and Ca, whereas the more evolved member has lower Ni and higher Mn and Ca (Fig. 8).
High compositional variability is typical for high-Fo olivine-hosted melt inclusions of Icelandic magmas (Maclennan 2008a, b;Neave et al. 2013). This variability in the melt inclusion record (e.g. highly varying La/ Yb) decreases as olivines become more evolved (ferrous), and the average composition of the olivine-hosted melt inclusions approaches the composition of the host magma. Maclennan (2008a) concluded that this phenomenon is consequential to mixing and homogenization of diverse mantle melts in the mid-lower crust. The major and trace element trends in Brattaskjól and Hvammsmúli olivine populations may indicate the same phenomenon: mixing of mantle melts derived from different depths and different Compositions of model olivines crystallizing from accumulated nearfractional melts of KLB-1 mantle lherzolite (Hirose and Kushiro 1993) are indicated by the black line (cf. Sect. 5.5 for model details). Stippled field outlines the model olivine compositions in equilibrium with F5-40% melts of modally enriched KG2 peridotite (Kogiso et al. 1998). Labels in black and blue correspond to the degree of partial melting in percent (F%) and melt derivation pressures in GPa, respectively. Grey arrow marks the compositional vector produced by fractional crystallization of olivine; however, assumption of olivine-only crystallization has been shown true only for Háleyjabunga (Neave et al. 2018). Average instrumental 2σ errors are indicated by grey bars for the trace elements, whereas errors in Fo are smaller than the symbols used. Olivine Fo content is taken from EPMA analyses, whereas all trace element data is from LA-ICP-MS analyses. Háleyjabunga and Stapafell olivine compositions from Neave et al. (2018) are included for comparison ◂ rock-types from the mantle melting column. Furthermore, the variability in Brattaskjól and Hvammsmúli olivine trace element compositions indicates that the Fo 88.2-89.6 olivine population has had sufficiently short residence time in the magmatic environment not to be completely homogenized by diffusive re-equilibration.

High-P fractionation of clinopyroxene
Early fractionation of clinopyroxene has been proposed to explain elevated Ni and Fe/Mn and low Ca in olivines in some of the lavas of the North Atlantic Igneous Province Higher values can be seen to represent higher contribution of pyroxenite mantle (e.g. Sobolev et al. 2007) or higher temperature difference between mantle melting and olivine crystallization (Matzen et al. 2013(Matzen et al. , 2017a (Hole 2018). Also, particular geochemical trends in Vestmannaeyjar lavas in SEVZ, specifically decreasing CaO/ Al 2 O 3 and Sc/Y with lowering MgO, have been explained by high-P (0.6-1.0 GPa) clinopyroxene crystallization (Furman et al. 1991;Mattsson and Oskarsson 2005). Against this backdrop, and considering that we assume olivine-dominated fractional crystallization for the Iceland magmas (Fig. 6), it is important to evaluate whether high-P clinopyroxene crystallization may have influenced the SEVZ olivine record. The stability field of clinopyroxene expands at the expense of olivine at high P and clinopyroxene becomes a liquidus phase in progressively more primitive melts (O'Hara 1968); clinopyroxene crystallization can thus commence in near-primary mantle melts at high pressures. Crystallization of Ca-rich clinopyroxene lowers Ca and elevates Fe/Mn in the melt, and olivine crystallizing along with clinopyroxene may, therefore, be imprinted by higher Fe/Mn (Herzberg et al. 2013) and lower Ca (Hole 2018) than olivine crystallizing in the absence of clinopyroxene. Furthermore, Ni is not as compatible in clinopyroxene as in olivine, and therefore, the Ni content of mafic magmas can stay relatively elevated if clinopyroxene crystallization is initiated early in the crystallization sequence (Hole 2018).
It is unlikely that high-P fractionation of clinopyroxene could have produced the high Ni and low Mn contents in Fo >88 olivines in Brattaskjól, Hvammsmúli and Stórhöfði, as clinopyroxene fractionation does not enrich melts in Ni, and thus cannot elevate Ni in olivines above the initial mantle-derived values. In addition, the trends of Ni and Ca against Fo content in olivines are not suggestive of clinopyroxene fractionation (i.e. steep depletion in Ni and enrichment in Ca in lower-Fo olivine, Fig. 6) for these localities. Olivine macrocrysts from the Eyjafjallajökull 2010 eruption tephra, however, may have been affected, as they are relatively evolved (Fo <86.9 ), very poor in Ca and Mn, and show an increase in Fe/Mn (Fig. 6a) at Fo 82-85 . Moreover, the relatively flat Ni versus Fo trend is suggestive of cotectic crystallization of olivine + clinopyroxene + plagioclase (Fig. 6b). These trace element characteristics comply with fractionation of clinopyroxene and olivine at high pressures, and thus the highest Fe/Mn values acquired from Eyjafjallajökull 2010 olivines (Figs. 6a and 7b) may have been affected by the early initiation of clinopyroxene crystallization.

Ni/Fo-decoupling in zoned olivine macrocrysts
Decoupled Fe-Mg and Ni in olivine rims likely reflects faster diffusion of Fe-Mg than Ni during re-equilibration of olivine and carrier melt (Lynn et al. 2017;Neave et al. 2018). We observed this decoupling in nearly all Hvammsmúli, Mosfellsheiði and Brattaskjól olivines. Considering the maximum thickness of olivine zonation in these localities, this phenomenon may have affected Ni/Fo ratio (by lowering Fo in relation to constant Ni) in crystal cores of Hvammsmúli olivines smaller than 1.4 mm, and Mosfellsheiði and Brattaskjól olivines smaller than 0.5 mm and 0.25 mm in diameter, respectively. Luckily, these locations host large olivine macrocrysts; therefore, only a few (potentially 13 out of 79) olivine core analyses of Hvammsmúli and none of the olivine core analyses of Mosfellsheiði or Brattaskjól may have been affected. In general, the Iceland olivine macrocryst record is dominated by large homogenous cores with thin zones of diffusive re-equilibration at crystal rims (e.g.  Thomson and Maclennan 2013). Hence, accidental analyses of Fe-Mg zoned olivine rims (with non-equilibrium Ni/Fo ratio) are improbable and unlikely to affect the current Iceland olivine trace element record (Shorttle and Maclennan 2011;Sobolev et al. 2007;Neave et al. 2018; and this study) to any significant degree.

Sc, Ti, V and Zn and their partitioning in mantle melting
In addition to the apparent enrichment in Ni coupled with lower Mn and Ca contents, SEVZ olivines have different Li, Sc, Ti, V, Zn and Cu contents in comparison to Háleyjabunga olivines (Fig. 5). This could be solely consequential to the greater depth and lower degree of mantle melting (F) at SEVZ, and it is uncertain, whether variation in the mantle source can be discerned based on these trace elements. To inspect this, we modelled partitioning of Sc, Ti, V and Zn during adiabatic near-fractional melting of KLB-1 lherzolite (Hirose and Kushiro 1993) at 1480 °C potential temperature (Matthews et al. 2016) and 0.5-4.0 GPa. In addition, we ran a similar melting model for the KG2 olivine-bearing pyroxenite-peridotite hybrid (Kogiso et al. 1998). Sc, Ti, V and Zn were chosen, because their partitioning has been determined at high-P conditions and sufficient partitioning coefficients were available (Davis et al. 2013;Laubier et al. 2014). Cu and Li were not modelled as their partitioning is yet poorly constrained, and since they are moderately incompatible to incompatible elements in mantle melting, their enrichment in SEVZ olivines (Fig. 5) likely reflects lower F, not variation in mantle mode (Ryan and Langmuir 1987;Lee et al. 2012). Al-depletion in SEVZ olivines in comparison to Háleyjabunga olivines, in turn, is likely consequential to higher Fo of Háleyjabunga olivines and the Mg-dependent partitioning of Al (Herzberg and O'Hara 2002). The modelling of mantle melting was done with the pMELTS software (Ghiorso et al. 2002) with the ALPHAMELTS 1.8. front end (Smith and Asimow 2005). We used the correct version of pMELTS garnet model (Berman and Koziol 1991), and did not include Na 2 O, K 2 O, Cr 2 O 3 and MnO into the initial mantle compositions. pMELTS has not been calibrated for MnO (Ghiorso et al. 2002), Cr 2 O 3 was not included to avoid over-stabilization of spinel (Asimow et al. 1995), and Na 2 O and K 2 O were omitted to facilitate reliable estimation of the solidus (Lambart et al. 2012(Lambart et al. , 2016Jennings et al. 2017). All models were done without imposed constrains on fO 2 , although ƒO 2 = FQM -1 was used to set Fe 2+ /Fe 3+ for the initial state. A 0.5 wt% threshold of melt extraction was used in all models, and a simple 1D melt aggregation column was assumed when assessing the composition of accumulated near-fractional partial melts at various pressures. After modelling mantle melting, we calculated the Fo and trace element content of olivine in equilibrium with the model mantle melts at 1 atm utilizing the parameterization of Herzberg and O'Hara (2002) and appropriate partitioning coefficients. The Supplementary Material provides the complete model conditions, initial mantle source compositions and partitioning coefficients.
Modally enriched mantle components, such as KG2, are thought to reside in the sub-Icelandic mantle as slivers in a more refractory matrix (Shorttle et al. 2014). Hence, our model melts of a one-component near-fractional mantle melting of KG2 do not fully suffice as an analogy for potential enriched Iceland melts. We opted for this simple setup, as developing a complex multi-component melting model for the sub-Icelandic mantle was not the aim of this study (cf. Shorttle and Maclennan 2011;Rudge et al. 2013;Shorttle et al. 2014;Lambart 2017 for existing parameterizations). The model olivine compositions of the KLB-1, and especially the KG2, partial melts are mainly illustrative and restricted by the limitations of pMELTS (cf. Rudge et al. 2013 and the Supplementary Material). Fo content in the KG2 model olivines should be treated with caution, as pMELTS performs poorly when calculating major element compositions in partial melts of modally enriched mantle lithologies (e.g. Lambart et al. 2016). The model olivine compositions are, however, useful for a qualitative estimation of whether typical KLB-1 lherzolite can explain the detected trace element variability in olivines, or whether lithological variation in the source is suggested.
The model olivine compositions crystallizing from KLB-1 and KG2 partial melts are shown in Fig. 5 and given in the Supplementary Material. The KLB-1 solidus is crossed at 3.2 GPa in the garnet stability field and garnet is in the residue until the spinel-garnet transition at 1.77 GPa. Spinel disappears from the residue at 1.63 GPa. The large stability range of garnet in comparison to spinel reflects limitations in the pMELTS garnet model and is partly due to omitting Cr 2 O 3 from the calculations. Melt aggregation occurs over a large P-interval with a melt-fraction of 2% at 2.61 GPa, 5% at 2.22 GPa, 10% at 1.86 GPa, 15% at 1.60 GPa, 20% at 1.27 GPa and 25% at 0.68 GPa (Fig. 5a). According to the pMELTS model, the KG2 mantle is already 5% molten at 4 GPa, and this melt portion was handled as an isobaric batch melting at 4.00 GPa. Its melt fraction increases to 10% at 2.11 GPa, 15% at 1.78 GPa, 20% at 1.47 GPa, 25% at 1.30 GPa, 30% at 1.15 GPa, 40% at 0.99 GPa and 45% at 0.85 GPa. The accumulated partial melts are ferric, which manifests as low Fo content of the model olivines (Fig. 5).
The Sc, Ti, and V concentrations in the analyzed olivine crystals conform with model olivines crystallizing from partial melts of lherzolite mantle (KLB-1). Háleyjabunga olivines show compatibility with a high-degree melt of a spinel lherzolite, whereas Brattaskjól and Hvammsmúli Page 13 of 19 43 olivines seem to have crystallized from accumulated lowdegree melts of mixed spinel and garnet lherzolite source (Fig. 5). Lower Sc in SEVZ olivines can be explained by greater depth of melting (Fig. 5a, higher D Sc in garnet lherzolite compared to spinel lherzolite), whereas high Ti in SEVZ olivines is compatible with the lower F expected in off-rift setting (Fig. 5b). Differences in V can be also attributed to greater melting depth beneath SEVZ (Fig. 5c), but due to the strong dependency of V partitioning on ƒO 2 (Canil 1997;Laubier et al. 2014), the lower V in olivines may also reflect more oxidized nature of the SEVZ magmas. KG2sourced model olivines have comparable Sc, Ti and V contents to those of KLB-1, although similar Sc and Ti values are reached at higher melt fractions (Fig. 5a-c).
Our KLB-1 lherzolite melting model does not explain the elevated Zn in SEVZ olivines, as Zn content is rather insusceptible to melting degree and melt derivation depth. In turn, KG2 model olivines have high Zn in comparison to KLB-1 melts (Fig. 5d), and thus the high Zn in SEVZ olivines may indicate the contribution of a modally (more pyroxene, less olivine) or compositionally (elevated Zn) enriched source in the mantle melting. Elevated Zn/Fe is a known property of many OIB magmas and has been proposed to indicate olivine-free pyroxenite in the mantle source (Le Roux et al. 2010Roux et al. , 2011. Alternatively, the lower Fo content in SEVZ olivines compared to Háleyjabunga, potentially implying a more differentiated host melt, may also contribute to the apparent Zn-enrichment in the SEVZ olivines. The olivineliquid partitioning of Zn during fractional crystallization of basalts is yet inadequately constrained. In essence, although Sc, Ti and V contents of the SEVZ olivines conform to olivine crystallizing from partial melts of KLB-1 lherzolite, the Zn enrichment may indicate that some enriched lithologies partake in melting in the mantle below. KG2-like peridotite-pyroxenite hybrid could be a good candidate for a potential enriched mantle lithology below Iceland, as it produces magmas with high-Fe/low-Ca (Shorttle et al. 2014) and, according to our analysis, does not significantly modify Sc, V, and Ti in olivines compared to 'normal lherzolite' values.

Origin of the SEVZ olivine signature
High Fe/Mn and Ni, and low Ca in olivine are attributes of olivine crystallized from a partial melt of olivine-free pyroxenite mantle (Sobolev et al. 2007). Following this rationale, the strongest affinity to 'pyroxenite source' is shown by the Eyjafjallajökull 2010 olivine macrocrysts, but also the other SEVZ samples show partial contribution of olivine-free source (Fig. 7). The highest-Ni Hvammsmúli, Brattaskjól and Stórhöfði olivines in SEVZ are comparable to, e.g., the ~ 70 Ma Quepos Terrane olivine from magmas of the Galapagos plume, for which olivine-free pyroxenite mantle source has been inferred (Trela et al. 2015). The parametrization of Gurenko et al. (2010): implies 44-52 wt% and 64-77 wt% contribution of olivinefree pyroxenite-derived melt (X px ) to the parental melt of Hvammsmúli high-Ni (> 3000 ppm) and Eyjafjallajökull 2010 olivine macrocrysts, respectively.
Recent thermodynamic modelling suggests that the silicic melts of olivine-free pyroxenite are prone to react with the surrounding peridotite, and thus less likely to preserve their coherence and compositional signature and eventually crystallize olivine in the crust (Lambart 2017). In addition, as the concentrations of the transitional elements in the SEVZ olivines are compatible with a garnet lherzolite mantle source, large input from olivine-free source seems unlikely. Although some of the SEVZ olivines were found to be Znenriched, this does not necessarily argue for olivine-free source, as Zn-enrichment is also expected in partial melts of enriched olivine-bearing mantle types (e.g. KG2). Furthermore, the lower Ca in olivines does not indicate the absence of olivine in the mantle source either but may imply lower degree of mantle melting and source enrichment in clinopyroxene or garnet.
As large-scale contribution of magmas directly derived from olivine-free mantle (Px-1 or similar, Sobolev et al. 2007) seems unlikely for SEVZ magmatism, we proceed on deciphering the SEVZ olivines by means of relatively deep melting of olivine-bearing mantle. Also, even if the mantle signature in the SEVZ olivines was due to contribution of melts from olivine-free pyroxenite source, the mantle melting would occur deep anyhow, due to the lower solidus temperature and greater fusibility of olivine-free pyroxenite in comparison to common mantle peridotite (Lambart et al. 2016). Mn are, directly (the former) and indirectly (the latter), temperature dependent (Matzen et al. 2013(Matzen et al. , 2017a, and elevated Ni and Mn in olivine can thus be related to the temperature difference between mantle melting and olivine crystallization (−ΔT m-c ). The ratio between NiO in crystallizing equilibrium olivine (NiO ol @T c ) and NiO content in olivine in the mantle (NiO ol @T m ) relates to the temperature of olivine crystallization (T c ) and mantle melting (T m ) as follows (Matzen et al. 2017a): (1)  Spice et al. 2016). It is to be noted that the selection of T c has only minor effect on the calculated ΔT m -c within reasonable values (e.g. 1266-1447 °C), but the formalization is susceptible to uncertainty regarding Ni in mantle olivine (Matzen et al. 2017a).
With these parameters, Eq. 3 gives 75 ± 3 °C for the −ΔT m-c , i.e. it indicates mantle-melt equilibration at 75 ± 3 °C higher T than the T of olivine crystallization. Using the slope of the olivine-saturated liquidus (55 °C/GPa, Sugawara 2000), ΔT m-c corresponds to 1.36 ± 0.06 GPa ΔP. If we assume olivine crystallization at surface (at 1 atm) and 3100 kg/m 3 average density for crust and refractory uppermost mantle, this pressure corresponds to ~ 45 km final mantle melting depth. This must be considered as the absolute minimum value for the inferred depth of melt equilibration in the mantle (final melting depth), as primitive olivine macrocrysts of Hvammsmúli and Stórhöfði likely crystallized at considerable pressure and depth, not at the surface. Crystallization pressures up to 0.6-1.0 GPa for Vestmannaeyjar and Eyjafjallajökull magmas have been proposed (Jakobsson et al. 1973;Furman et al. 1991;Sigmarsson 1996;Keiding and Sigmarsson 2012). If these pressures of olivine crystallization are considered and correspondingly elevated T c 's are used (1349-1371 °C), final melting pressures of 2.0-2.5 GPa (i.e. depths of 66-81 km) are indicated. These pressures are close to the inferred pressure of spinel-garnet phase transition for KLB-1 lherzolite mantle (2.14-2.17 GPa, Jennings and Holland 2015).
Our estimation of final mantle-melt equilibration at > 45 km depth is considerably deeper than the crustal thickness at SEVZ (Darbyshire et al. 2000;Fedorova et al. 2005;Jenkins et al. 2018), but in line with the depth of the lithosphere-asthenosphere boundary (LAB) according to glacial isostatic adjustment studies (Árnadóttir et al. 2009; Barnhoorn et al. 2011). However, there is considerable uncertainty regarding the LAB in Iceland, with resent estimates of ridge centered lithospheric thicknesses varying in 10-60 km range (Pagli et al. 2007;Bjarnason and Schmeling 2009;Barnhoorn et al. 2011;Rychert et al. 2018). We do not view the olivine compositions as a direct proxy for LAB depth (as done in, e.g. Matzen et al. 2017a), but consider the SEVZ high-Ni olivines as the crystallization products of deep accumulated melts that avoided mixing and equilibration at the top of the melting column in the mantle (which may or may not correspond to LAB). This view is supported by (1) the previously discussed mantle melt mixing trend in Brattaskjól and Hvammsmúli olivine populations, where the high-Ni olivines represent a deep endmember in the presence of apparent shallower mantle melt; and (2) the likelihood of mantle-melt equilibration pressures greater than 2 GPa supported by the evidence for deep olivine crystallization and "garnet signature" (low Sc) in olivine crystals. Essentially, the SEVZ high-Ni/low-Mn olivines are an indication of the survival of mantle melts from the deep parts of the mantle-melting column to the T-P-conditions of olivine crystallization.

What makes SEVZ special?
High-Ni, low-Mn olivines seem to be characteristic for SEVZ but absent in SVZ and the rift-zone magmas. This is noteworthy, especially regarding the trace element enriched SVZ lavas (e.g. Berserkjahraun), which, alike SEVZ lavas, have compositional and isotopic characteristics arguing in favor of melt extraction from relatively deep in the mantle (e.g. Kokfelt et al. 2006). Why the apparent signature of deep mantle melting in olivines at SEVZ in particular? It is likely that deep mantle melts are in fact ubiquitous in all rift-zones and off-rift zones in Iceland, but they are usually diluted by melt mixing in the mantle or mid-lower-crust (Maclennan 2008a;Neave et al. 2013). Although magmas originating from deep portions of the mantle can erupt at the rift zones (as indicated by trace element enriched lavas like Stapafell and Gaesafjöll, Shorttle and Maclennan 2011), survival of deep melts is most likely to occur at off-rift settings, like at SEVZ, where the degree of mantle melting is low (Kokfelt et al. 2006). Furthermore, as magmatism at SEVZ is a relatively recent phenomenon (younger than 3 Ma, Martin et al. 2011), the underlying mantle may have not yet been depleted of the most fusible lithologies (e.g. KG2), which further biases the melt production to increased depth. Sufficiently refractory shallow mantle (e.g. harzburgite, Shorttle et al. 2014) and effective vertical melt transport could also inhibit melt mixing in the mantle and promote magma derivation from depth. Channelized melt transport (Spiegelman and Kelemen 2003;Weatherley and Katz 2012;Rudge et al. 2013), in contrast to melt extraction by diffuse (equilibrium) porous flow (Lundstrom et al. 1995;Niu 1997;Dijkstra et al. 2003), may be the required melt transfer mechanism for survival of deep mantle melts and potentially important in the mantle below SEVZ.
The nature of the crustal magma plumbing systems beneath Eyjafjallajökull and Vestmannaeyjar is also likely to contribute to the crystallization, and survival, of high-Ni/low-Mn olivine macrocrysts. Even if a parcel of deep mantle melt were to keep its coherency during transit through the mantle and manage to crystallize equilibrium olivine in the crust (with high-Ni, low-Mn), this olivine would need to be erupted relatively fast to avoid re-equilibration with accumulated magmas with diverse composition during storage in the crust or uppermost mantle. Olivine macrocrysts in Iceland are rarely in equilibrium with their carrier melts, and are prone to be modified by diffusive re-equilibration (Thomson and Maclennan 2013). For example, if the Hvammsmúli and Brattaskjól olivine population (Fig. 8) had resided in a magma reservoir for extended time, the compositional signature of deep origin would have been erased. The originally heterogeneous olivine cargo (high variability in Ni, Mn and Ca in high-Fo olivines; Fig. 8) would have re-equilibrated diffusively with itself and the carrier magma, suppressing extremes in Ni, Mn and Ca. In the end, homogenized olivine population would have only modest enrichment in Fe/Mn and "Kistufell-like" Ni and Ca (Fig. 6) and the original subpopulation of high-Ni olivines might only be identifiable from olivine-hosted melt inclusions or slowly diffusing trace elements.
There are indications of short-lived and poorly developed deep magma storage systems below Eyjafjallajökull and Vestmannaeyjar. U-series disequilibria (Sigmarsson 1996) and absence of equilibrium phenocryst assemblages (Mattsson and Oskarsson 2005) in Vestmannaeyjar lavas suggest rapid differentiation within a deep and cold crust and short magmatic residence times. Furthermore, deep seismicity prior to and during the Eyjafjallajökull 2010 eruption suggests near-direct magma derivation from the mantle (Tarasewicz et al. 2012). Relatively fast transport of mantle melts limits crystal-melt equilibration and magma mixing, promoting survival of compositional extremes in the mineral cargo. Rather little is known about the lower-crustal magma storage zones in SVZ, but the maturity of SVZ volcanism could hint at well-established magma reservoirs. SVZ lowdegree primitive magmas may reside in the deep crust (or shallow mantle) long enough that magmas aggregated from different depths in the mantle melting column are better mixed and equilibrated. This limits compositional variability in populations of primitive olivine, and suppresses a potential "deep melting signature" (high-Ni/low-Mn in olivines).

Conclusion
In each of the sampling locations, major and trace element trends in olivine populations reflect events of fractional crystallization, magma mixing and diffusive re-equilibration. Two Eyjafjallajökull transitional basalts also show large variability in the Ni, Mn and Ca contents of primitive Fo 88.2-89.6 olivines, which we consider as a mixing trend of melts equilibrated at different levels in the mantle. Eyjafjallajökull and Vestmannaeyjar volcanic systems in SEVZ have anomalous (high-Ni and Fe/Mn-enriched) olivines compared to previously published olivine data from Iceland. These olivine macrocrysts likely crystallized from magmas that equilibrated with the mantle at relatively high temperatures and pressures, at depths of > 45 km and possibly as deep as 66-81 km near the spinelgarnet peridotite transition. The evidence for deep mantle melts that avoided mixing and kept their coherency while migrating up through the upper mantle and the crust is an indication of effective vertical magma transport below SEVZ.