Consequences of a crystal mush-dominated magma plumbing system: a mid-ocean ridge perspective

Abstract

Crystal mush is rapidly emerging as a new paradigm for the evolution of igneous systems. Mid-ocean ridges provide a unique opportunity to study mush processes: geophysical data indicate that, even at the most magmatically robust fast-spreading ridges, the magma plumbing system typically comprises crystal mush. In this paper, we describe some of the consequences of crystal mush for the evolution of the mid-ocean ridge magmatic system. One of these is that melt migration by porous flow plays an important role, in addition to rapid, channelized flow. Facilitated by both buoyancy and (deformation-enhanced) compaction, porous flow leads to reactions between the mush and migrating melts. Reactions between melt and the surrounding crystal framework are also likely to occur upon emplacement of primitive melts into the mush. Furthermore, replenishment facilitates mixing between the replenishing melt and interstitial melts of the mush. Hence, crystal mushes facilitate reaction and mixing, which leads to significant homogenization, and which may account for the geochemical systematics of mid-ocean ridge basalt (MORB). A second consequence is cryptic fractionation. At mid-ocean ridges, a plagioclase framework may already have formed when clinopyroxene saturates. As a result, clinopyroxene phenocrysts are rare, despite the fact that the vast majority of MORB records clinopyroxene fractionation. Hence, melts extracted from crystal mush may show a cryptic fractionation signature. Another consequence of a mush-dominated plumbing system is that channelized flow of melts through the crystal mush leads to the occurrence of vertical magmatic fabrics in oceanic gabbros, as well as the entrainment of diverse populations of phenocrysts. Overall, we conclude that the occurrence of crystal mush has a number of fundamental implications for the behaviour and evolution of magmatic systems, and that mid-ocean ridges can serve as a useful template for trans-crustal mush columns elsewhere.

This article is part of the Theo Murphy meeting issue ‘Magma reservoir architecture and dynamics'.

1. Introduction

Magma chambers have traditionally been considered to be magma-filled reservoirs, where melt evolution is controlled by the removal of crystals through crystal settling and/or crystallization along the walls of the chamber. However, this view is now being challenged, and a new paradigm is emerging where magma reservoirs are considered to be filled predominantly by crystal mush [1–3]. This paradigm shift was predominantly initiated by the realization that melt-filled chambers are difficult to maintain over geologically significant periods of time, particularly in the upper continental crust, and by petrological observations (see [2] for a review). At mid-ocean ridges, the realization that the plumbing system is dominated by crystal mush came relatively early, and was driven primarily by geophysical observations: the large melt-filled chambers once considered to underlie the ridge axis [4,5] were not found by geophysical surveys in the 1980s (see [6] for a review). Instead, the data indicated the presence of crystal mush.

As discussed by Cashman et al. [2], the emerging mush paradigm requires a reassessment of the processes controlling plutonic and volcanic evolution. The mid-ocean ridge magmatic system provides an ideal opportunity to do so. This is because it features a combination of observables that is difficult to find in other geodynamic settings: it is an active system, which enables geophysical data to constrain the nature of the magma plumbing system (e.g. [6]), yet both plutonic and volcanic sections are available for sampling. Hence, the plutonic–volcanic connection can be assessed directly, and the significance of petrological and geochemical observables can be placed within the framework of known magma chamber architecture. Furthermore, mid-ocean ridges do not suffer from complications introduced by the interaction of the magmatic system with older and compositionally different country rocks: all observables relate directly to the magmatic system itself.

In this paper, we use petrological observations on the mid-ocean ridge system to reflect on some of the consequences of a magma plumbing system dominated by crystal mush. We first cover the geophysical constraints on the nature of mid-ocean ridge magma chambers, and then discuss the role the crystal mush plays in melt transport, the compositional evolution of melts and the generation of cryptic fractionation. We conclude that mush processes exert a fundamental control on the physical and chemical evolution of the mid-ocean ridge magmatic system.

2. The geophysical framework: a mush-dominated magma plumbing system

The nature of the mid-ocean ridge magma plumbing system is best constrained at the fast-spreading East Pacific Rise, where a wealth of seismic data have been obtained over the last three decades. Seismic reflection data have established that the top of the magmatic system is formed by a melt-bearing body 1–2 km below the seafloor, situated at the transition from the sheeted dykes to lower crustal gabbros [7–10]. This so-called axial melt lens is hundreds of metres to a few kilometres wide, and tens of metres thick [9]. It is present along a large proportion of the ridge axis (approx. 85% at 9–10°N on the East Pacific Rise), but is not continuous, with numerous disruptions present, commonly associated with ridge axial discontinuities [11]. Although the axial melt lens contains enough melt to possess a negative seismic velocity anomaly relative to the overlying sheeted dykes, more recent work has shown that the axial melt lens is not uniformly melt-rich [12]: it is segmented on both a kilometre and hundreds of metres scale into crystal-rich and crystal-poor portions [13,14]. Furthermore, melt lens depth varies over short spatial scales [11,15]. Hence, the axial melt lens is a dynamic horizon, with changes in depth, melt and crystal proportions in both space and time.

The axial melt lens overlies a large region with attenuated seismic velocities. Seismic tomography and seafloor compliance data indicate that this region, referred to as the low-velocity zone, is typically approximately 7 km wide, and extends to the base of the crust at approximately 6 km depth [16–18]. It is interpreted to be a crystal mush. Proportions of melt are difficult to determine and model-dependent, but estimates range from a few to potentially a few tens of per cent [16,17]. Within the low-velocity zone, a number of melt lenses have been found in recent years [19]. These occur both on axis and up to 8 km off-axis [20]. Furthermore, a number of sills have been detected at the Moho [21]. Overall, the seismic character of the East Pacific Rise changes little along the ridge axis, indicating that it approximates steady state over geological time.

Intermediate-spreading ridges, such as the Juan de Fuca Ridge, appear to have a similar seismic structure to fast-spreading ridges: they, too, feature a segmented axial melt lens with a variable crystal content [22,23]. However, the axial melt lens here is deeper than at the East Pacific Rise, indicating an overall cooler thermal structure. At slow-spreading ridges, mantle upwelling is sufficiently slow and cooling sufficiently efficient that a steady-state melt lens cannot be maintained, and a regular Pacific-type, 6 km thick layer-cake igneous crust may not form. Some segments (e.g. Vema) expose crust that appears to conform to the classic Penrose stratigraphy of gabbros–sheeted dykes–lavas, but the seismically defined crustal thickness is variable along axis, and anomalously thin at the ends of ridge segments [24]: such along-axis variation appears to be the norm at slow spreading rates [25]. Other, less magmatically robust segments locally expose mantle peridotite, with or without gabbroic intrusions, indicating that slow-spreading lithosphere along such segments may be composed of lithospheric mantle intruded by gabbroic plutons, such that a pure igneous crust does not form [26]. Consistent with these observations, axial melt lenses are not normally present at slow-spreading ridges. So far they have only been detected at two locations on the Mid-Atlantic Ridge: at 57°N on the Reykjanes Ridge [27], and Lucky Strike (37°N), which is near the magmatically robust Azores platform [28]. As at fast-spreading systems, the melt lens at Lucky Strike overlies a region of crystal mush. At other slow-spreading segments for which seismic data are available, only crystal mush has been detected [29,30]. Overall, these observations imply that the generally lower amount of melt supply and the cooler thermal structure at slow spreading rates lead to significant variations in the nature of the plumbing system: in contrast to fast-spreading ridges, the system is not in steady state. Instead, because of the generally cooler thermal structure, emplaced melts quickly turn to the observed mush bodies, and they cannot sustain an axial melt lens.

Overall, the seismic data of mid-ocean ridges indicate that the magma plumbing system is dominated by crystal mush [6].

3. Consequences of a mush-dominated magma plumbing system

The prominence of crystal mush in the plumbing system of mid-ocean ridges has a number of consequences for the magmatic processes occurring in the system, for which we may look for evidence in the rock record. In the sections below, we discuss three of these: modes of melt transport, buffering of melt compositions and cryptic fractionation.

(a). Melt transport

Melt transport mechanisms in crystal mush are predicted to differ from those in melt-filled magma chambers surrounded by solid wall rock. In the latter situation, melt transport can readily occur by dyking, particularly in a mid-ocean ridge setting, where extension creates optimum conditions for dyke formation to occur. By contrast, in a crystal mush, melts are predicted to migrate by porous flow [31], provided that the interstitial melt is positively buoyant with respect to the surrounding crystal framework. To determine whether melts are buoyant in a mid-ocean ridge crystal mush, we modelled the differentiation of a primitive mid-ocean ridge basalt (MORB) using MELTS v. 1.2 [32]. We chose the proposed Indian Ocean primary melt of [33] (also modelled in [34]; composition listed in the electronic supplementary material, table S2), assuming 0.15% H₂O, and simulated fractional crystallization at 2 kbar. Although the exact crystallization sequence will depend on the assumed melt composition, water content and pressure, most MORB will follow a similar crystallization path, and we assume that the results in terms of density evolution are generally applicable. The results are shown in figure 1, which plots the density of the melt and the density of the bulk cumulate assemblage crystallized at each step as a function of temperature. As can be seen, the density of the melt at its liquidus is 2.68 g cm⁻³; in the temperature interval 1270°C–1120°C, where the majority of crystallization occurs, melt density changes very little. Although the density of the fractionated crystals changes significantly as a function of the saturation temperatures of the different phases, the density of the melt is always below that of the crystals. Hence, melts in a mid-ocean ridge crystal mush are predicted to be positively buoyant throughout their crystallization histories.

Figure 1.

Density evolution of a primary MORB, and the cumulates crystallized from it, during progressive fractional crystallization. Model curves were calculated using MELTS v. 1.2 [32], and the primary MORB composition is that of [33] (see electronic supplementary material, table S2), assuming 0.15% H₂O. The density of the cumulate plotted is the modal proportion multiplied by the density of the phases at each increment of crystallization: it thus represents the bulk density of the cumulate produced at each step. (Online version in colour.)

The upward migration of buoyant interstitial melts may be further aided by compaction. Although it has been suggested that compaction may not play as significant a role in generating cumulate rocks as once thought in all plutonic systems [35], we posit that, compared to continental-hosted intrusions, the lower oceanic crust provides an environment where compaction is most likely to occur. This is because: (i) the mush cools slowly, allowing sufficient time for compaction to occur; (ii) interconnected mush thickness is as much as 4–5 km at fast-spreading ridges, providing a significant gravitational driving force; and (iii) the regime of continual intrusion coupled with wholesale bulk deformation of the system will act to enhance the efficiency of melt expulsion. Evidence that compaction may indeed play a role in mid-ocean ridge mush systems is provided by three observations. The first is that minor deformation twins in plagioclase and kink bands in olivine are very common: an example is shown in figure 2. These structures are present even if they are not associated with any solid-state deformation, such as the fast-spreading East Pacific Rise, where solid-state deformation appears to be completely absent [36,37]. Hence, we argue that these textures result from the compaction of the crystal mush: compaction will lead to the mechanical interaction between different mineral grains as melt is expelled and the modal proportion of crystals increases [38,39], leading to minor deformation of the crystal lattice through dislocation creep. Considering that the melt proportion in most of the mid-ocean ridge crystal mush is considered to be less than 20% (see §2), such particle interaction appears inevitable. Particle interaction is also responsible for the second line of evidence for compaction: the occurrence of pressure solution driven by impingement of different grains [40]. A final consideration is magmatic flow, which is expected to be common at mid-ocean ridges because of the continuous extension of the crust, and, at fast-spreading ridges, because the lower crust is mechanically coupled to the flow of the underlying mantle as it rolls over beneath the ridge axis [41]. Magmatic flow results in simple shear-dominated deformation of the crystal mush. As a result, it is common for oceanic plutonic rocks to be foliated and/or lineated (figure 2), and the same is true for plutonic rocks from the Oman ophiolite [36,41–44]. The fabrics in these rocks are defined by aligned crystals that contain no features characteristic of solid-state deformation, such as dislocation structures and subgrains [36]. This indicates that the deformation occurred in the presence of melt [45,46], i.e. deformation operated when the system was a crystal mush. If deformation occurs in a crystal mush, this will lead to expulsion and focusing of interstitial melts into melt-rich zones, thus facilitating melt segregation from the compacting crystal framework [47–49]. Hence, although the degree to which compaction occurs in relatively small continental magma chambers remains an open question [35], the observations of the mid-ocean ridge crystal mush suggest that compaction is an important process in this setting.

Figure 2.

Photomicrographs illustrating microtextural evidence for compaction in oceanic cumulates. (a) Full thin section scan of troctolite from IODP Hole U1309D at Atlantis Massif (30°N, Mid-Atlantic Ridge). A magmatic foliation, outlined by orange dashed lines (S1), is defined by subparallel plagioclase laths. Plagioclase laths are characterized by tapered deformation twins, with some prominent examples outlined by the white ellipses. (b,c) Close-ups of red rectangular boxed areas in (a), showing a series of kink bands in olivine (outlined by the red arrows) and deformation twins in plagioclase. Ol, olivine; Plag, plagioclase; S1, foliation. (Online version in colour.)

Combined, the buoyancy forces and (deformation-assisted) compaction will lead to the upward migration of the interstitial melts by porous flow. The crystal framework through which the melts flow is highly variable in composition: although it will be dominated by a fairly simple mineral assemblage of olivine, plagioclase and clinopyroxene, observations from lower crustal sections across the spreading rate spectrum indicate that modal proportions and mineral compositions vary on a centimetre to metre scale (e.g. [33,37,50]). As a result, it is highly unlikely that percolating melts retain equilibrium with the mush framework during their ascent. This will lead to reactions between the melts and framework: hence, porous flow is expected to be reactive. A significant number of observations in the last decade have firmly established the footprint of this process in the mid-ocean ridge rock record [51–57]. These have recently been covered in depth in [34], and we only recap the main findings here. The first is that evidence for reactions between crystal and migrating melt are nearly ubiquitously preserved in oceanic plutonic rocks [34] as well as lower crustal xenoliths in MORB [56,57], and are manifest on a grain scale by both reaction textures (e.g. dissolution fronts, symplectites) and mineral compositions. From a compositional viewpoint, zoning in major elements of cumulus (i.e. former framework) crystals towards domains of melt flow is associated with unusual enrichment and fractionation of trace elements [54]. The second main finding is that the same trace element enrichment and fractionation that is found on a grain scale also occurs on the scale of the entire lower crust [54]. This suggests that the lower crustal crystal mush acts as one system, and that it is permeable throughout. This is consistent with the fact that the lower crust, at least at Hess Deep in the Pacific, forms an overall upward differentiation sequence. Hence, the lower crustal crystal mush is an interconnected system, evolving upwards, with a significant role for reactive porous flow.

The consequences of reactive porous flow are profound: melt evolution no longer obeys the laws of fractional crystallization. From a major element perspective, reactions may change the MgO–FeO–CaO–Al₂O₃ relationships, as these are the dominant elements in the main phases involved [53]. MgO, in particular, is sensitive to reactive flow, as Fe–Mg exchange between migrating melts and crystals is rapid. In the case where evolved melts interact with a primitive matrix, this can lead to unusual mineral and melt compositions, with equilibration with the matrix leading to Mg# typical of a primitive melt composition, but enriched incompatible elements (e.g. Ti) [53]. From a trace element perspective, it is striking that the reactive geochemical signature of trace element enrichment and fractionation that is present in the lower crust is very similar to the trace element distributions in MORB [54]. This led [34] to argue that MORB trace elements bear an imprint of reactive porous flow in the lower oceanic crust.

However, porous flow is unlikely to be the only melt transport mechanism operating in the lower oceanic crust. Observations in the Oman ophiolite, an analogue for fast-spreading oceanic crust, indicate that the sheeted dykes and lavas are in Fe–Mg equilibrium with cumulates from the deeper part of the lower crust, but not with those from the upper part of the lower crust [58]. Similarly, the upper gabbros in Hess Deep are not in equilibrium with the overlying sheeted dykes and lavas [59]. These data require that melts were extracted from the deeper parts of the lower crustal crystal mush without only limited interaction with the overlying, more evolved parts of the mush, and that extraction was rapid and efficient enough to transport melts upwards by several kilometres to erupt. Melt transport by porous flow would not satisfy either of these criteria.

Such rapid, channelized melt flow requires that melts can be efficiently extracted from the mush. One mechanism for this to occur is by concentrating (interstitial) melts into melt-rich zones. Such self-organization may occur as a result of crystal settling and compaction, aided by deformation, as discussed above. This is consistent with numerical models of crystal mushes, which indicate that compaction of an internally homogeneous mush will lead to melt segregation and concentration (e.g. [60]). Once melts have segregated into a sufficiently large melt body, they may travel upwards through the mush as porosity waves [61,62]. Alternatively, their buoyancy may destabilize the overlying crystal mush, and melt rapidly travels to the top of the system [63]. In addition to melt segregation, replenishment of the mush is likely to play a significant role in the occurrence of channelized melt transport. It is a very effective means of unlocking, or fluidizing, the mush, leading to rapid upward transport of melts in crystal-poor ‘chimneys’ [64–66]. Hence, processes that are inherent to replenished crystal mushes such as those of the lower oceanic crust may facilitate melts to be extracted and erupted.

From a mid-ocean ridge perspective, these dynamic processes operating in the crystal mush not only explain the occurrence of both porous and focused melt transport, but they may also facilitate magma mixing, provide an explanation for the presence of diverse populations of phenocrysts/antecrysts in MORB, and can account for observed magmatic fabrics. The role of replenishment and magma mixing in MORB is well documented [67–72], and is usually assumed to occur in melt-rich reservoirs. Within a crystal mush, mixing may occur between the replenishing melt and the interstitial melt present in the mush. In effect, this would produce effects akin to in situ fractionation, where interstitial melts are returned to a melt reservoir in a static magma chamber crystallizing inwards from its walls [73]. In the mush variant discussed here, the interstitial melt is mobilized from a crystal mush by the destabilizing effect of replenishment [64,66,74–76]. Either way, interstitial melts are mixed with a larger body of melt, so the geochemical effects are likely to be similar. It has been noted that the incompatible trace element distributions of MORB, which are over-enriched relative to fractional crystallization predictions [54,67], can be explained by in situ fractionation [77], or, alternatively, by a contribution of trace element-enriched interstitial melts to MORB [34,54]. Mush destabilization by replenishment can account for either of these. It can also account for the observation that MORB commonly contains diverse populations of phenocrysts and antecrysts, each recording a different petrogenetic history [74–76,78–80], as well as xenoliths of crystal mush [56,57]: when mush destabilizes and melts are transported upwards through it, they may pick up crystals and crystal clots from different parts of the system. Furthermore, even when crystals originate from a single location, the dynamics of mixing of melt and mush can be such that each crystal may take a different composition–time path, leading to the generation of multiple populations [64]. Finally, focused melt transport through a crystal mush may explain the occurrence of vertical magmatic fabrics in gabbros from the top third of fast-spreading lower oceanic crust. Such fabrics have been documented both in Pacific lower crust [36,44] and in the Oman ophiolite [42,81,82], and are defined predominantly by aligned plagioclase defining both a sheeted dyke-parallel foliation and approximately vertical lineation. Although some workers argue that the fabrics record downward flow of cumulates produced in the overlying axial melt lens (e.g. [83–86]), the fact that the deeper, layered gabbros are more primitive than the foliated gabbros, and that, in contrast to the layered gabbros below, the foliated gabbros are not layered, makes this scenario unlikely. Instead, we argue that the fabrics record the focused upward transport of melt through the mush, with the shear stress along the melt–mush interface leading to the alignment of the crystal framework surrounding the melt channel [36,42,82].

In summary, melt transport in the mid-ocean ridge crystal mush occurs both by porous and focused flow. Porous flow is an inevitable consequence of crystal mush systems, and, because porous flow is reactive, has major implications for melt evolution. It is supplemented by rapid, channelized flow that results from a combination of replenishment and the concentration of interstitial melts in the mush into melt-rich bodies. These punctuated events are associated with the destabilization of the crystal mush, and may account for the incompatible trace element distributions in MORB, the diversity of its crystal cargo and vertical fabrics in plutonic rocks.

(b). Buffering of melt compositions

Fractional crystallization has long been the paradigm for the differentiation of magmas [87], and this has also traditionally been invoked to explain the differentiation of MORB (e.g. [88]). However, if melts evolved in a crystal mush rather than a melt-filled magma chamber, fractional crystallization may be less effective. As outlined in §3a, reactions between migrating melts and the cumulus framework may play an important role in the compositional evolution of interstitial melt, and these melts may be remobilized and mixed with replenished melts during melt extraction. The record preserved in oceanic gabbros indicates that these reactions are generally (though not exclusively) between melts that are evolved compared to the melts that formed the framework crystals: much of the associated mineral zonation is normal [34,54]. A second type of reaction that may occur is associated with replenishment: hot, primitive melts are not in equilibrium with lower crustal cumulates, and will induce melting reactions [89–95]. Combined, these reactions serve to reduce the compositional variance of the melts in the system; both evolved and primitive melts react with the mush in an attempt to gain equilibrium with it.

In the end member scenario where these two types of reactive process go to completion, crystallization will approach equilibrium crystallization, forcing all melts onto the three-phase olivine–plagioclase–clinopyroxene cotectic (assuming that those are the phases in the mush). In this case, melt–mush reaction would buffer all melt compositions, leading to MORB with little major element variability. In nature, full equilibrium is unlikely to be obtained: preserved reaction textures and zoning patterns (e.g. [34,56,57,89]) indicate that reactions do not tend to go to completion. Furthermore, there is significant small-scale mineralogical and compositional variability in the lower oceanic crust (e.g. [33,37,50]). Hence, even if melts achieve local equilibrium with their host, continued migration will lead to disequilibrium once more as they reach the next unit. However, different elements will react at different rates: those for which diffusive exchange is rapid are more likely to attain equilibrium, at least on a local scale, whereas slow diffusing elements require extensive melting and/or dissolution–reprecipitation reactions and are less likely to attain equilibrium. In this context, it is instructive to look at FeO–MgO relationships: Fe–Mg exchange between olivine and melt is rapid [96], so Fe–Mg is the best candidate for examining a potential role of the crystal mush in buffering melt compositions. That Fe–Mg buffering occurs during melt porous flow in the mush is clear from the mineral compositions of clinopyroxene in oceanic gabbroic rocks. Some of these, especially those occurring as interstitial grains or oikocrysts in primitive troctolitic cumulates, contain high TiO₂ contents, indicating that they crystallized from evolved melts. However, they have high Mg# and are in Fe–Mg equilibrium with the surrounding forsterite-rich olivine [53]. Combined, the textures and compositions indicate that evolved, clinopyroxene-saturated melt percolated the troctolite, acquiring a higher Mg# through equilibration of the host olivine [34,53]. Slow-diffusing elements, in this case Ti, reflect the high degree of differentiation of the melt, whereas the fast-diffusing Fe–Mg relationships show an apparent primitive origin as a result of diffusive re-equilibration.

In figure 3, we plot histograms of the Mg# of the East Pacific Rise and Mid-Atlantic Ridge glass datasets. These were downloaded from the PetDb database [97,98], and comprise data for 3825 Pacific and 3607 Atlantic glasses. We calculated Mg# assuming Fe³⁺/Fe^T = 0.12. The most striking feature is the narrow range present in the distributions: the standard deviations are 4.2 mol% for the Atlantic and 6.2% for the Pacific. The small variation in Mg# in samples representing thousands of kilometres of ridge volcanism indicates that there is a magmatic process that modulates the Fe–Mg systematics of melts extracted from the plumbing system. This could be replenishment mixing: melt-rich reservoirs undergoing a cyclic evolution of replenishment, tapping and fractionation can reproduce the global MORB MgO distribution [67]. Alternatively, the narrow distribution may result from the scenario outlined above, where the reaction of both evolved interstitial and primitive replenishing melts with the mush framework provides a buffering capacity. At the least, the observation of Fe–Mg buffering during reactive flow in the lower oceanic crust indicates that this scenario is a viable one.

Figure 3.

Histograms of Mg#, calculated using Fe³⁺/Fe^T = 0.12, for Atlantic (a) and Pacific (b) glass data from the PetDB database [97,98]. (Online version in colour.)

Another consideration is the role of temperature. The Pacific glass data have a lower mean Mg# (57.7) than their Atlantic counterparts (61.1), and have a non-Gaussian distribution, characterized by a skew towards the low-Mg# end (figure 3). The lower mean Pacific Mg# has been interpreted to reflect the temperature of the underlying magma plumbing system, with slow-spreading magma chambers thought to be more thermally insulated and hotter than fast-spreading magma chambers, leading to a lower mean degree of differentiation [99]. The low-Mg# tail has been interpreted to mark either the dying stage of replenished–tapped–fractionated magma chambers [67], or the interaction of melts with cool, hydrothermally altered crust at ridge discontinuities such as propagating ridge tips and overlapping spreading centres [100]. Crystal mush processes may provide an alternative explanation for this observation: numerical models of the thermochemical evolution of crystal mush show that one of the consequences of melt migration through a mush results in the collection of relatively cool, evolved melts at the top of the section; hence, the mush becomes stratified with respect to temperature and composition [62]. These model predictions match remarkably well with the mineral compositional profile of the Pacific lower crust at Hess Deep, which shows an upward differentiation trend, with the uppermost gabbroic rocks representing crystallization from highly evolved melts [54,59]. Furthermore, seismic data indicate that the crystal mush is most (vertically) extensive at fast-spreading ridges compared to slow-spreading ridges (see §2); hence, it would be predicted to develop more advanced compositional stratification. The result, then, is that Pacific lavas have, on average, a lower Mg#, and that they extend to lower Mg#, than slower-spreading lavas.

(c). Cryptic fractionation: the clinopyroxene paradox

It has long been recognized that a large majority of MORBs have major element trends indicative of fractionation of clinopyroxene, which is particularly manifest in a decrease in CaO with decreasing MgO [101]. However, clinopyroxene phenocrysts are rare, and experimental data indicate that many MORBs are not saturated in clinopyroxene at low pressure [102,103]. This issue has been referred to as the pyroxene paradox [69,104].

Although the pyroxene paradox has been recognized for decades, it has largely remained a qualitative concept. There are very few large MORB suites for which both mineral modes of phenocryst phases, required to quantify the abundance of clinopyroxene, and major element data, to test for clinopyroxene fractionation, are available. To address this, we have collected modal proportions of phenocrysts from MORBs from the Gakkel Ridge (Arctic Ocean). The Gakkel Ridge is the slowest-spreading mid-ocean ridge on Earth, with spreading rates decreasing westwards from 14.6 to 6.3 mm yr⁻¹ [105]. Because the rate of mantle upwelling scales with spreading rate, the magnitude of mantle melting and heat advection to the oceanic crust at ultraslow spreading rates is lowest. As a result, crustal thickness is significantly reduced, whereas lithospheric thickness is increased [106]. At Gakkel, lithospheric thickness has been estimated to be as much as 35 km [107]. Hence, for the purpose of examining the pyroxene paradox, Gakkel Ridge is a good candidate: with the lithosphere being thick, melt evolution might occur at elevated pressures compared to slow- and fast-spreading ridges, leading to increased stability of clinopyroxene (e.g. [88,108,109]). If clinopyroxene is rare even under these conditions, which favour its crystallization, this would suggest that there is an intrinsic process in the mid-ocean ridge plumbing system which prevents much of the clinopyroxene that has crystallized from being erupted as phenocrysts.

The Gakkel Ridge basalts studied herein were collected during the international Arctic MOR Expedition (AMORE 2001 [110]) and sample approximately 850 km of the ridge axis. Samples were collected from different tectonomagmatic segments of the ridge to ensure a good representation of Gakkel MORB as a whole (see [110] for details on tectonomagmatic segmentation). To obtain mineral modes, element maps of 95 samples were obtained, which were subsequently transformed into phase maps (see Methods in the electronic supplementary material for details).

The result of our modal analysis is presented in the electronic supplementary material, table S3, and illustrated in figure 4. The average modal proportion of Gakkel basalt phenocrysts is 8.5% plagioclase, 2.1% olivine and 0.1% clinopyroxene (figure 4b). Of the 95 samples studied, 94 contain olivine phenocrysts, 62 contain plagioclase and 13 contain clinopyroxene (figure 4a): this amounts to 99%, 65% and 14% of the sample set, respectively. Whereas olivine and most of the plagioclase occur as individual crystals, there are only three samples in which clinopyroxene occurs simply as phenocrysts: in the majority of clinopyroxene-bearing samples, clinopyroxene phenocrysts occur alongside glomerocrysts of intergrown clinopyroxene and plagioclase (figure 5). One sample contains clinopyroxene only as plagioclase–clinopyroxene glomerocrysts. A detailed textural study of these glomerocrysts indicates that they represent fragments of crystal mush that were entrained in the basalt prior to eruption [111].

Figure 4.

Modal analysis of 95 basalts from the Gakkel Ridge (Arctic Ocean). (a) Olivine and plagioclase are present as phenocrysts, but clinopyroxene is not. Within individual samples, clinopyroxene may occur as phenocrysts, glomerocrysts or both. (b) Mean modal proportions of phenocrysts in the Gakkel basalts. (Online version in colour.)

Figure 5.

Line drawing of glomerocryst in a Gakkel basalt. The glomerocryst is composed of a network of lath-shaped plagioclase, with granular to interstitial olivine. Clinopyroxene forms anhedral crystals within the plagioclase framework. Scale bar is 1 mm. (Online version in colour.)

In order to test whether the Gakkel basalts contain a chemical imprint of clinopyroxene fractionation, we have examined their MgO–CaO relationships. For this purpose, we have compiled all Gakkel glass analyses (n = 516) from the literature [110,112–117]. In order to relate the compositions of the literature data to the modal data on our samples, we also present new glass data on a subset (n = 34) of our Gakkel samples (electronic supplementary material, table S4; see the electronic supplementary material for methods). Our new glass data overlap the literature data, indicating that our samples, for which modal data are available, are representative of the broader Gakkel melt compositions.

The MgO–CaO relationships are shown in figure 6. As is typical for MORB (e.g. [118]), the data form a broad array with a positive slope (i.e. decreasing CaO with decreasing MgO), with significant variation at the high-CaO end. We superimpose the low-pressure liquid lines of descent of two of the primitive Gakkel glasses, which we modelled using rhyolite MELTS v. 1.2 [32], assuming a pressure of 0.1 kbar and oxygen fugacity of QFM-1. It is immediately clear that the data do not represent a simple cogenetic suite, as those primitive samples with relatively high CaO (approx. greater than 11.2 wt%, including prominent groups at 11.4 wt%, 11.7 wt% and 12.0 wt%) cannot be produced by crystallization of the most primitive of the glasses, which are characterized by relatively low CaO (approx. 11 wt%). Hence, there is clear variation in the glass data related to the composition and/or degree of melting of the mantle source. It is beyond the scope of this paper to reconstruct the detailed petrogenetic history of the Gakkel glasses; nonetheless, the liquid lines of descent are informative when it comes to assessing clinopyroxene fractionation, because the olivine and olivine + plagioclase sections of the liquid lines of descent form essentially horizontal arrays. The only way for melts to decrease in CaO is through clinopyroxene fractionation. At low pressure, this will occur relatively late, as is illustrated in figure 6. With increasing pressure, clinopyroxene will saturate earlier, at higher MgO contents. As such, the low-pressure olivine and olivine + plagioclase parts of the liquid lines of descent form the maximum upper boundary between clinopyroxene-saturated and clinopyroxene-undersaturated melts (e.g. [118]). The observation that nearly all of the glasses fall below this boundary for the high-CaO starting composition, and a large majority do so for the low-CaO starting composition, indicates that, like global MORB [67], most of the Gakkel glasses have a geochemical signature of clinopyroxene crystallization. Hence, the combined modal and compositional data indicate that the pyroxene paradox applies to the Gakkel Ridge: many of the samples have crystallized clinopyroxene, but only 14% contain clinopyroxene, and clinopyroxene on average only forms 0.1% of the phenocryst mode.

Figure 6.

MgO–CaO relationships of glasses from the Gakkel Ridge. The literature data (black open circles; n = 516) are supplemented by 34 new analyses presented herein (green full circles; data in electronic supplementary material, table S4). The liquid lines of descent of two primitive glasses, outlined by the blue and red circles, are shown by the solid lines. Water contents were set at 0.25 wt% and 0.4 wt% for the high-CaO and low-CaO primitive melts, respectively. These were chosen on the basis of their location along the Gakkel Ridge axis, along with the known along-axis gradient in Gakkel glass water content [115]. (Online version in colour.)

A number of different mechanisms have been proposed to account for the pyroxene paradox. One relates to MORB phase equilibria, which are such that, at increasing pressure, the clinopyroxene stability field increases [88,108,109]. Hence, MORB may be saturated at the depth of its source magma chamber, but be undersaturated in clinopyroxene upon eruption onto the seafloor [88,102,119]. This scenario explains why MORBs are commonly undersaturated in clinopyroxene; however, it does not readily explain why clinopyroxene should not occur as phenocrysts. Although some resorption of clinopyroxene phenocrysts could occur during melt transport from the magma reservoir to the seafloor, the rapid transport during dyking events makes it unlikely that nearly all clinopyroxene should be assimilated back into the melt. An alternative, that density separation of clinopyroxene (ρ ∼ 3.25 g cm⁻³) from plagioclase (ρ ∼ 2.65 g cm⁻³) occurs during melt transport, is ruled out by the observation that olivine phenocrysts, which have a similar density to clinopyroxene, are common (figure 4). A second explanation for the pyroxene paradox is provided by magma mixing [68]. If evolved, clinopyroxene-saturated melts mix with primitive, clinopyroxene-undersaturated melts during replenishment events, the resulting mixture will be clinopyroxene-undersaturated. It will, however, have lower CaO at a given MgO than the olivine–plagioclase control line, because the evolved melt will have crystallized clinopyroxene prior to mixing, lowering its CaO content. A similar scenario could be envisaged during in situ fractionation [73], where evolved, clinopyroxene-saturated melts are extracted from crystal-rich portions of the magma chamber and are added back to the main melt reservoir.

Crystal mush offers an alternative explanation for these scenarios [53]. When primitive MORB is emplaced in the lithosphere (be it the mantle lithosphere at (ultra)slow-spreading ridges or the lower crust at fast-spreading ridges), it will start crystallizing. The first increments of crystallization are olivine (±Cr-spinel), quickly followed by olivine + plagioclase. At the point where plagioclase joins the liquidus, the crystallization rate increases dramatically [120]. As a result, the magma turns into mush as a crystal framework starts forming. At what proportion of crystals a framework starts forming has been the subject of considerable recent interest. This so-called critical crystal volume fraction (ϕ_c) is not simply a function of the crystallinity, but also depends on grain shape, the degree of particle alignment, surface roughness and melt viscosity [121,122]. Hence, there is no unique value of ϕ_c: it differs from one system to the next, and within the system as any of these parameters change [121]. Another key aspect is that, even at relatively high melt proportion, crystals may form force chains that can sustain yield strength [121,123]. A consequence of force chains is that the mechanical transition from melt to mush occurs at relatively low crystal proportions [123]. At mid-ocean ridges, the fact that crystallization is dominated by plagioclase, which has a large degree of shape anisotropy, promotes low ϕ_c [122] and hence early mush formation: natural and experimental observations suggest that plagioclase-dominated systems may behave mechanically like a mush at relatively low crystal content (approx. 20–25% [122,123]). Such modal proportion is strikingly similar to the fraction of crystals that have formed from MORBs at lower crustal pressure by the time it reaches clinopyroxene saturation (22% in the Indian MORB modelled in §3a; see fig. 13 in [34]). Hence, by the time clinopyroxene crystallizes, it does so within an olivine + plagioclase framework. Average MORB possesses a Mg# of approximately 60 (though slightly higher at slow-spreading compared to fast-spreading ridges [99]), which requires approximately 40–50% fractionation of a primary melt (Mg# ∼ 72). Hence, given that a framework already forms at 20–25% crystallization, a significant proportion of MORB crystallization is likely to occur in this framework-dominated mode, rather than in a melt-dominated mode. The implication is that, being an interstitial phase in a crystal framework, clinopyroxene does not crystallize as a ‘free’ phase within a pure magma, as models based on melt-filled magma chambers would predict. Hence, clinopyroxene does not readily form phenocrysts suspended in a magma-rich reservoir, and is thus unlikely to erupt: it is essentially trapped in the mush. Nonetheless, melts extracted from the mush will show a chemical signature of clinopyroxene crystallization.

It is difficult to identify parameters that uniquely discriminate between the different explanations for the pyroxene paradox, and in fact the paradox might result from a combination of the different processes discussed above. Nonetheless, three arguments suggest that the mush explanation is a viable one. First, the Gakkel lavas contain few clinopyroxene phenocrysts despite the conditions favouring relatively high-pressure fractionation: hence, there must be something in the magma plumbing system that prevents them from being erupted. Second, ultraslow-spreading ridges are relatively magma poor, and contain no documented melt-rich magma chambers, limiting the opportunity for magma mixing. Instead, slow- and ultraslow-spreading magma plumbing systems are dominated by crystal mush, providing ample opportunity for clinopyroxene to get trapped in a mush. Third, clinopyroxene in oceanic plutonic rocks commonly grows around or in between idiomorphic plagioclase (e.g. [124]). Three examples from the longest recovered section of lower oceanic crust (Hole 735B, Indian Ocean) are shown in figure 7. Although each example has different modal proportions and grain size, they each contain lath-shaped plagioclase that forms a network. Clinopyroxene either grows as a minor interstitial phase defining an intergranular texture (figure 7a), or fills in a larger proportion of the pore space, growing in between and around plagioclase laths in a subophitic manner (figure 7b,c). We take these textures as indications that clinopyroxene crystallized within the framework of previously crystallized olivine + plagioclase. Such textures bear a striking resemblance to the glomerocrysts in the Gakkel basalts, where clinopyroxene also forms within a plagioclase framework (figure 5). These samples provide a direct snapshot of the crystal mush, illustrating that the formation of plagioclase frameworks, and subsequent interstitial growth of clinopyroxene, are an important step in the evolution of melts emplaced in the lower oceanic crust.

Figure 7.

Core photographs of gabbroic rocks from ODP Hole 735B (Atlantis Bank, Indian Ocean). (a) Fine-grained leucogabbro characterized by interstitial clinopyroxene in a plagioclase network. (b) Medium-grained gabbro with higher proportions of clinopyroxene, which grew subophitically to interstitial to a plagioclase network. (c) Medium-grained, subophitic gabbro. Despite the higher proportion of clinopyroxene, clinopyroxene still predominantly crystallized within an interconnected framework of plagioclase. Scale bar is 5 cm in each case. (Online version in colour.)

4. Synthesis

In figure 8, we present our observations and interpretations in the form of a schematic cartoon illustrating the processes operating in lower oceanic crustal crystal mush systems. This cartoon is deliberately not placed in the context of a fast-spreading or slow-spreading ridge: we view the processes we describe as generic to any crystal mush, irrespective of context.

Figure 8.

Schematic of a lower oceanic crustal crystal mush. Melt lens 1 is formed by replenishment of the crystal mush by primary melt, leading to reaction with the surrounding mush and mixing of interstitial melts into the melt lens. Melt lens 2 forms by the segregation of interstitial melts in the mush. Melt transport occurs both by reactive porous flow, aided by compaction, and by channelized focused flow. The latter, which drains melt lens 3, feeds eruptions of MORB, picking up crystal cargo and producing fabrics in the mush en route. (Online version in colour.)

We highlight a number of features. The first is the role of replenishment of melts derived from the upper mantle. Replenishment will lead to the formation of melt-rich lenses in the crystal mush (melt lens 1 in figure 8). Along the margins of these lenses two processes will operate: partial melting of the surrounding crystals, and mixing of the replenishing melt with the interstitial melt of the surrounding mush. Both processes lead to homogenization of the melts ultimately extracted from the crystal mush. The mixing process of interstitial melts into the replenishing melt will achieve a geochemical signature similar to in situ crystallization, which has been called upon to explain the incompatible trace element distributions of MORB.

As emplaced melts crystallize, plagioclase forms a crystal framework after only limited amounts of crystallization (20–25%). As a result, clinopyroxene tends to crystallize within a plagioclase-dominated framework, rather than in a melt-filled reservoir, which is consistent with the textures of oceanic plutonic rocks (figure 7). Clinopyroxene is thus rare as phenocrysts in MORB, despite the near-ubiquitous signature of its fractionation in MORB evolution.

Throughout the crystal mush, interstitial melts migrate through porous flow, driven by a combination of buoyancy and compaction (figure 8). That compaction, aided by deformation in places, plays a role is suggested by the microtextures of oceanic plutonic rocks (figure 2). Because equilibrium between migrating melts and surrounding crystals is unlikely to be the norm, porous flow is reactive. This leads to a wealth of reaction textures and compositions in oceanic plutonic rocks, and melt evolution trends that are characterized by strong incompatible trace element enrichment relative to fractional crystallization models. Interstitial melt may segregate to form a second type of melt lens (labelled 2 in figure 8) and thus self-organize into potentially eruptible bodies. This provides an additional mechanism for interstitial melts to contribute to MORB, in addition to mixing along the margins of melt-rich bodies. In the case of interstitial melts migrating by reactive porous flow, the melts may carry a reactive geochemical signature. This provides an alternative explanation for the incompatible trace element distributions of MORB.

Melt-rich lenses may destabilize, facilitating the rapid, channelized upward transport of melt (melt lens 3 in figure 8). This will enable more mixing and reaction to occur between melts in the channel and interstitial melt of the mush around the channel wall. Furthermore, melts in the channel may pick up a range of crystals and glomerocrysts (figures 4 and 5) from the channel walls, accounting for the diverse nature of crystal cargo in MORB. Plagioclase crystals along the channel margin may align, forming foliations and lineations in the surrounding mush.

The integrated end result of these various processes is the extraction of melts that are relatively homogeneous in terms of major elements (particularly of rapidly diffusing elements, e.g. Fe–Mg, figure 3), have a cryptic fractionation signature and carry a clear signature of magma mixing. However, magma mixing is accomplished in a very different way from that classically assumed to occur in melt-rich magma chambers: it occurs between melt-rich bodies and the interstitial melts of the crystal mush as melts are emplaced, segregated and transported by channelized flow. Ultimately, this leads to trace element distributions that are incompatible with fractional crystallization: the more incompatible elements are over-enriched.

We conclude that the fact that the mid-ocean ridge plumbing system is dominated by crystal mush has a number of fundamental implications for the behaviour and evolution of the system as a whole. Because all of the underlying processes are governed by physical processes, and by the chemical response of the system to these physical processes, we anticipate that the results described herein are inherent to the evolution of crystal mushes in general, irrespective of geodynamic setting. As such, the mid-ocean ridge system can serve as a useful template for trans-crustal mush columns elsewhere.

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material.

Authors' contributions

C.J.L. conceived of the paper, and drafted the manuscript. C.J.M. contributed to critical discussions. E.N.B. analysed the Gakkel samples. All authors read and approved the manuscript.

Competing interests

The authors declare that they have no competing interests.

Funding

This work was supported by a number of NERC grant nos. (particularly NE/I001670/1 to C.J.L. and NE/C509023/1 to C.J.M.). E.N.B. is supported by NERC grant no. NE/L002434/1.