A mantle-source solution to the enigma of bimodal arc volcanism

Silicate melts in arc environments are dominated by mac (low-silica) and silicic (high-silica) compositions, often generating a characteristic bimodal pattern. We investigate the whole arc crust and show that the plutonic lower crust shares the bimodal pattern of melts from volcanoes. This key observation reveals that, contrary to some explanations of bimodal volcanism, variation in mantle source and mantle processes must fundamentally control bimodalism. We also recognise bimodalism in Th/La composition of the whole arc crust and suggest a new working hypothesis: bimodalism originates by melting of distinct sub-arc mantle sources, one dominated by relatively dry peridotite and the other by hydrous pyroxenite. The two groups of primary melts fractionate along distinct liquid lines of descent that lead to relatively dry mac melts (Th/La~0.1) versus hydrous silicic melts (Th/La>0.2) by 65–80% fractional crystallisation. Common crustal processes such as crystal fractionation, assimilation, reactive ow and/or magma mixing may also lead to differentiation of both groups.


Full Text
Subduction zones are the most tectonically active regions on Earth. They produce more variable volcanic rock compositions than any other tectonic environment and are crucial to understanding many fundamental geological processes, such as continental growth, geochemical cycles, tectonic-climate coupling, ore genesis, earthquakes, volcanism, and other natural hazards 1

. A basic characteristic of melts
in arc environments is that they are bimodal 2 , meaning that ma c and silicic compositions are common, whereas intermediate melts are relatively rare. Although this bimodality has been recognised and debated widely for nearly 100 years, it is yet to be convincingly explained 3 . Here, we use whole-rock and isotope compositions of plutonic rocks in deep crustal arc sections to develop a new hypothesis for bimodality in arcs. We show that the bimodal pattern extends throughout the entire arc crust from volcanoes to the lower crust, indicating that the ultimate cause of bimodality cannot be a crustal phenomenon. Instead, we propose that bimodalism originates by melting of distinct sub-arc mantle sources dominated by relatively dry peridotite or hydrous pyroxenite. Crustal processes that were previously invoked to explain bimodality in igneous systems also occur, but their in uence is limited to secondary modi cation of bimodal melts that originate in the mantle.
A long-recognised pattern of bimodal distribution in the composition of igneous rocks, known as the Bunsen-Daly or Daly gap, has puzzled geologists for over 100 years, producing many extant and contradictory models for its origin that fuel ongoing debate [2][3][4] . Suggested causal processes include liquid immiscibility, stalling of intermediate compositions, crystallization and/or separation of particular minerals, and partial melting of the crust 5-7 , but consensus is lacking.
In subduction zone settings, ma c and silicic melt inclusions dominate, as opposed to whole rock compositions where intermediate rocks are very common 2 (Fig.1); the whole rocks comprise mostly crystal-rich units with up to 90% of minerals in disequilibrium with their bulk-composition 8 . Recently, Meade et al. 3 returned to crustal assimilation as a major factor in the explanation of bimodality in continental volcanism. However, Keller et al. 9 argued for progressive fractional crystallization during magma ascent to produce intermediate and silicic magmas, emphasizing the fractionation of ma c cumulates. Similarly, numerical considerations of the thermal effects of ma c intrusions support melting of the crust or, at smaller scales, reactive ow through crystal mushes to create bimodal suites containing low degrees of magma mixing [10][11][12][13] . The breadth of these studies shows that the bimodal geochemical pattern is well documented, but its origin may be multifaceted.
The mechanisms that have been proposed to create the bimodal geochemical pattern are complicated and invariably assigned to the crustal component of the arc system (e.g., deep hot zone, MASH [melting, assimilation, storage, homogenization] zone, shallow hot zone, cold storage and remobilisation 11,13,[14][15] ). Most of these models are based on inferred magma chamber or crystal mush processes occurring dominantly in the upper crust 4 (<15 km) but some studies suggest that the fundamental bimodal character had already been generated below this level 2 . An additional limitation to these models is the common assumption of a relatively homogeneous peridotite mantle source for the initial arc magmas. However, the rst stages of mantle melting almost universally occur in a mixed source region consisting of a minimum of two distinct rock types 16 . In the case of the mantle wedge above subducting slabs, these comprise a mixture of peridotite and metasomatised components involving serpentinite through to hydrous pyroxenite 17,1 with increasing temperature deeper into the mantle wedge above the subducting slab (Fig.1).
The fundamental bimodal character of volcanism in arcs 2 is investigated here by using a new compilation of whole-rock geochemical data from eroded Mesozoic arcs: Fiordland (New Zealand), Kohistan (Pakistan) and Talkeetna (Alaska, USA), which are the Earth's premier deep arc sections (Fig.1).
This compilation presents consistent peaks in composition at 50-54 wt.% and 70-75 wt.% SiO 2 ( Fig.1; complete references and compiled data in Supplementary Table 1). The pattern is the same as that identi ed in melt inclusions from volcanic rocks (red line in Fig.1; ref.2), demonstrating that the bimodal character applies to the entire arc crust from volcanoes down to the lower crust. Thus, a deeper origin within the mantle is implicated as the ultimate source of the bimodal geochemical pattern, discounting all explanations that rely on crustal processes alone.
The formation of hydrous pyroxenite by interaction between slab-sediment-derived partial melt and peridotite in the mantle wedge above a subducting plate is predicted by experiments from temperatures as low at 675°C 17,18 . Partial melting experiments on dry pyroxenites show that they may produce primary melts that overlap or are up to 5-15 wt.% higher in SiO 2 than melts of peridotite 19 (B and A, respectively on Fig.1). However, experiments show both source types are capable of forming a range of primary melt compositions. While current models focus on ux melting of peridotite hydrated by water released from the subducting slab (A, Fig.1), our hypothesis also invokes ux melting of hydrous pyroxenite (B, Fig.1), as return ow in the mantle wedge may draw pyroxenite to depths and temperatures where it will partially melt 20 . Large areas of hydrous pyroxenite occur in arc-related ultrama c massifs and are often thought to originate as cumulates 21 ; these may become involved in melting above the subducting slab. Phlogopite pyroxenite rocks are expected to be common products of melt-rock reaction in the mantle wedge 17,18 . However, most experiments on pyroxenite have been aimed at the deep recycling of dry rocks in the oceanic environment 19,22 , and so may be of limited relevance to the hydrous conditions above subducting slabs. Melting points of hydrous pyroxenites 23 compared to peridotite 29 . The felsic plutons and melt inclusions form steep arrays on Figure 3, increasing in Th/La (Fig.3, black arrows). We require new experiments to better understand partitioning of Th and La during ux melting of peridotite and hydrous pyroxenite, and how these elements behave during magmatic fractionation. These details will help explain the bimodalism of Th/La in arc crust (Fig.3).
The observation of bimodal compositions in both surface volcanoes and the plutonic base of arcs ( Fig.1) indicates that signi cant magmatic evolution occurs by melt-rock reaction during transport through the sub-arc mantle (arrows A to C or B to D, Fig.1). Additionally, some evolution may occur in the lower to middle crust via combinations of crystal fractionation, assimilation, reactive ow and/or magma mixing (C to E and D to F, Fig.1; 13,[30][31] ).
In our case study of the lower and middle crust of Fiordland, New Zealand, the ma c and silicic plutonic components are spatially distributed into paired inboard (Western Fiordland Orthogneiss, WFO, dry, ma c, rear arc) and outboard (Separation Point Suite, SPS, wet, silicic, front arc) belts (Fig.2). Our new and compiled published isotopic data (Supplementary Table 2) shows that the paired plutonic belts in Fiordland have similar mantle-like Sr-Nd isotopic compositions to each other (Fig.4), precluding signi cant ancient crustal contributions in the petrogenesis of the silicic suite 3,15 and relating the two mantle sources. The distinction in major element compositions (Fig.2) and the simple spatial pattern in Fiordland likely re ects well-separated relatively dry peridotite-and hydrous pyroxenite-dominated sub-arc mantle sources (Fig.1). These relationships are di cult to explain with the other proposed causes of bimodality 2-8,10-15 , though additional isotopic systems (e.g., Pb) could be analysed to investigate this further. An equally heterogeneous sub-arc involving two mantle sources is inferred for paired volcanic belts, such as NE Japan, Kamchatka, Kurile and Chile [32][33][34] . We envisage that the distribution and degree of mixing of the sub-arc mantle endmembers controls the presence or absence of such spatial geochemical patterns. In contrast, relatively homogeneous volcanic and plutonic crust in many arcs is produced by melting of a uniform sub-arc mantle source.
In summary, a heterogeneous distribution of variably hydrous source rocks in the sub-arc mantle imparts a fundamental bimodal geochemical signature to magmatic arcs. Relatively dry peridotite and hydrous pyroxenite endmembers form primary melts that follow very different fractionation paths. This is essential to explain the existence not only of bimodal volcanism at Earth's surface, but also of bimodal plutonism in the lower and middle crust (Fig.1). It places the solution to the enigma of arc bimodalism at sub-crustal depths within the sub-arc mantle (Fig.1) and attributes it to reactions between peridotite of the mantle wedge and uids or melts from the subducting slab to form bimodal source rocks comprising less-reacted peridotite and metasomatised components such as hydrous pyroxenite. This working hypothesis provides a critical foundation for future research of melting processes, the causes of the bimodal composition of arc volcanism, and how and where primary melts are modi ed on their way to the surface.

Methods
Geochemical databases: The geochemical compilation of whole-rock data from (1) deep-arc sections for Kohistan and Talkeetna, (2) peridotite and pyroxenite xenoliths, and (3) melt inclusions compositions are from the Georoc database (http://georoc.mpch-mainz.gwdg.de/georoc/). The data was re ned by removal of altered samples (Supplementary Table 1    Epsilon Nd versus initial 87Sr/86Sr ratio at crystallisation ages for Earth's premier exposed deep arc sections ¬-Fiordland (New Zealand), Kohistan (Pakistan) and Talkeetna (Alaska, USA). The Fiordland data includes the older long-lived Darran Suite and two Cretaceous suites comprising inboard (Western Fiordland Orthogneiss, ma c, rear arc) and outboard (Separation Point Suite, silicic, front arc) belts.
Smaller circles and diamonds, this study; larger circles and other symbols are from literature (Supplementary Table 2).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.