Reaction Between Mid‐Ocean Ridge Basalt and Lower Oceanic Crust: An Experimental Study

Reaction between mid‐ocean ridge basalt (MORB) and crystal mush in the lower oceanic crust has been invoked to explain chemical variations of both MORB and minerals in the lower oceanic crust. Nonetheless, such reactions have been little studied experimentally. We conducted experiments to investigate the mechanisms and chemical consequences of melt‐mush interaction by reacting molten MORB with troctolite at 0.5 GPa. Isothermal experiments demonstrate that melt infiltrates into troctolite with dissolution of plagioclase and olivine. The reacted melts have higher MgO and Al2O3 and lower TiO2 and Na2O contents and crystallize more primitive olivine and plagioclase compared to those crystallized from the unreacted melts, suggesting melt‐mush reaction could result in the formation of high‐Al basalt. The melt compositional variations induced by reaction also significantly affect the calculated pressures for MORB fractionation, indicating that major element‐based barometers for MORB fractionation can only be used reliably if reaction can be ruled out. After reaction, the troctolite contains olivine with plagioclase inclusions and poikilitic clinopyroxene with partially resorbed olivine and plagioclase chadacrysts, indicating that melt‐mush interaction occurs through dissolution‐reprecipitation mechanisms. Clinopyroxene has high Mg# (>83) and elevated Na2O and TiO2 contents, and olivine has different Fo versus Ni correlations from fractional crystallization models, which provide testable parameters for the effect of melt‐mush reaction in the rock record. By comparison with samples from lower oceanic crust and layered intrusions, we propose that melt‐mush reaction plays an important role during magma transport in the crystal mush in both oceanic and continental magma systems.


Introduction
Mid-ocean ridge basalt (MORB) has long been recognized as a probe to the composition of the upper mantle. Crustal level modifications to MORB compositions are largely attributed to fractional crystallization, which can be corrected through parameterizations of experimental data or thermodynamic calculations (e.g., Grove et al., 1992;Langmuir et al., 1992). However, studies on both MORB and plutonic rocks from the lower oceanic crust have provided evidence suggesting that the crustal evolution of MORB is complex, involving more than fractional crystallization alone (e.g., Lissenberg & MacLeod, 2016, and reference therein). Seismic studies indicate that mid-ocean ridges are mostly underlain by crystal mush (Canales et al., 2000;Carbotte et al., 2013;Crawford et al., 1999;Crawford & Webb, 2002;Dunn et al., 2000;Seher et al., 2010;Vera et al., 1990). Hence, melts and crystals coexist throughout the mid-ocean ridge magma plumbing system, providing two possible mechanisms for reactions to occur between melts and crystals. The first is during the emplacement of primitive melts into the mush (Kvassnes & Grove, 2008;Leuthold et al., 2018). Primitive melts are unlikely to be saturated in (all of) the phases present in the mush in which they are emplaced, leading to partial melting reactions. The second is during upward flow of interstitial melts in the mush Gao et al., 2007;Lissenberg et al., 2013;Lissenberg & Dick, 2008). Numerical modeling of this process indicates that melt migrating through such a compacting mush will react with the constituent crystals (Solano et al., 2014), which is consistent with the rock record (Lissenberg & MacLeod, 2016). Therefore, melt-mush reaction is a natural consequence of the emplacement of primitive melt into the lower oceanic crust and the percolation of (evolved) interstitial melt through the crystal mush.
Phenocrysts and xenoliths in MORB samples provide further evidence of magma replenishment and subsequent reaction. Glassy MORB samples are widely known to host olivine and plagioclase crystals, which are commonly not in equilibrium with their surrounding glass (Bennett et al., 2019;Moore et al., 2014;Pan & Batiza, 2002. Some of these samples contain phenocrysts proportions as high as 50% (Bennett et al., 2019;Hellevang & Pedersen, 2008;Lange et al., 2013), suggesting physical mixing and/or chemical reaction between a phenocryst-poor magma with a crystal-bearing mush. Melt-bearing cumulate xenoliths in MORB from the East Pacific Rise and Iceland provide strong evidence for complex melt-mush reaction history in the lower oceanic crust, including the passage of multiple MORB magmas through the network of intergranular pores, and disequilibrium-induced dissolution and subsequent growth of crystals in the mush (Gurenko & Sobolev, 2006;Ridley et al., 2006). High-Al melt inclusion in olivine in MORB (e.g., Danyushevsky et al., 2003;Koleszar et al., 2009;Laubier et al., 2012) and high-Al MORB (e.g., Coumans et al., 2016;Yang et al., 2017), mostly with high MgO (>8 wt.%), were all proposed to form by reaction with a plagioclasebearing cumulate in the lower oceanic crust.
Hence, the available data indicate that disequilibrium induced by both replenishment and porous flow of melt through a crystal mush is common in the magma pluming system and that the resulting reactions may play an important role in the evolution of the oceanic crust. Experimental works have been devoted to the reaction of melt with mantle rocks, including peridotite (e.g., Daines & Kohlstedt, 1994;Lambart et al., 2009;Morgan & Liang, 2003Saper & Liang, 2014;Tursack & Liang, 2012;Van Den Bleeken et al., 2010Wang et al., 2013) and dunite (e.g., Borghini et al., 2018), but very little attention has been paid to reaction of melt with crustal cumulates. Melt-cumulate interaction experiments have been conducted by melting plagioclase-olivine and plagioclase-augite mineral pairs at temperatures much higher than the melt-mineral saturation boundary (Kvassnes & Grove, 2008), which could be applicable during emplacement of primitive melt in a mush. However, in natural conditions, such interaction might not simply occur by melting cumulates and subsequent mixing with preexisting melt, especially during flow of interstitial melts in a primitive cumulate framework. Thus, it remains poorly understood by which mechanisms melt-mush reactions proceed, how and to what extent such processes would change MORB compositions, and hence, how such interaction will affect reconstructions of mantle source compositions and melting processes. We conducted a series of melt-mush reaction experiments to investigate the mechanism and chemical consequences of such process by reacting both primitive and evolved MORB melts with a primitive cumulate (troctolite). The results have enabled a reconstruction of the melt-mush reaction mechanisms and place constraints on the chemical consequences of melt-mush reaction on both the plutonic rocks and the melt.

Experimental Methods
Reactive crystallization experiments characterizing melt-mush interaction were conducted at 1,180 and 1,200°C (corresponding to the estimated liquidus temperatures for the two starting melts) and 0.5 GPa for 6-27 hr using a 19.1-mm piston cylinder apparatus at Brown University. The reaction couple was formed by juxtaposing a powdered MORB against a troctolite in a graphite-lined molybdenum capsule (Figure 1).

Starting Materials
The starting materials for the melts are a powered primitive MORB glass with Mg# (defined as mole fractions of 100 * MgO (MgO + FeO T )) of 64 (KN182-13 D44A, provided by Alberto Saal) and moderately evolved MORB from the East Pacific Rise (Mg# 54; MOA8712-095, provided by John Sinton), with the latter representing average East Pacific Rise basalts. A lower oceanic crust troctolite analog was produced by combining 70 wt.% unzoned plagioclase (anorthite [An] 87, from Oman ophiolite gabbro sample 97OB1, provided by Benoit Ildefonse) and 30 wt.% unzoned olivine (forsterite [Fo] 90, from Kilbourne Hole, New Mexico, Sample KBH-1). In addition, a troctolitic mush analog was produced by combining 45 wt.% plagioclase, 45 wt.% olivine, and 10 wt.% interstitial primitive MORB glass of the same origin. Optically clean mineral grains of approximately 100-150 μm and glass chips of around 1-3 mm were first hand separated under a binocular microscope, mixed based on the appropriate weight percentages, and then ground under ethanol to obtain the desired starting materials. Afterward, the basalts and troctolite were stored in closed containers in a 110°C oven. Table 1 lists the starting mineral and melt compositions.

Experimental Procedure
The experimental procedure was similar to that described in Tursack and Liang (2012) and Saper and Liang (2014). The furnace assembly consists of a molybdenum capsule in an MgO sleeve sandwiched between two crushable MgO spacers in a graphite, Pyrex, and KCl salt sleeve. The molybdenum capsule (6.5-mm outer diameter and 10 mm long) was lined with a graphite inner sleeve (4-mm outer diameter and 2-mm inner diameter), which served as an inert buffer between molybdenum and the silicates. To form a reaction couple, we packed the bottom 2-3 mm of the graphite inner capsule with troctolite and the top 3-4 mm with the powdered basalt. The top of the capsule was then sealed with a 1-mm layer of graphite powder and capped with a molybdenum lid (0.5-1 mm thick; Figure 1). The furnace assembly was held in a 110°C oven for at least 12 hr.
The furnace assembly was wrapped with lead foil and loaded into the piston cylinder apparatus. To run an experiment, the assembly was first cold pressurized to slightly above 0.5 GPa. The temperature was then ramped to 1,180 (evolved melt) or 1,200°C (primitive melt) at a rate of 75°C/min. After 6 hr (27 hr for Run TE_Ls) of reaction at the target temperatures, Runs TE1, TE_mush, and TP1 were quenched, and the others were step-cooled to 1,000 (evolved melt) or 1,050°C (primitive melt) and subsequently quenched following the temperature-time history displayed in Figure 1. Pressure was kept at 0.5 GPa throughout the experiment. Temperature was measured using a W 97 Re 3 -W 75 Re 25 thermocouple and monitored using a Eurotherm controller with an uncertainty of 10°C (Morgan & Liang, 2005). After quenching the experiment, the charge was unloaded, and the molybdenum capsule was extracted and sectioned longitudinally. The sectioned charges were mounted in epoxy, polished, and carbon coated for electron microprobe and scanning electron microscope analyses (detailed methods can be found in Supporting Information S1).
Because we used the liquidus temperatures of the melts during the experiments, the basalt was liquid, and troctolite was mainly solid with varied proportions of interstitial melt. In addition, two step-cooling experiments with the evolved and primitive MORB melts only (i.e., without troctolite; Runs Es and Ps) were conducted as a reference to compare the crystallization sequence with and without reaction at the same pressure. Except for TE_Ls, all runs were reacted at the respective liquidus temperatures for 6 hr. TE1, TP1, and TE_mush were quenched directly after 6-hr reaction. TE2/Es and TP2/Ps were step-cooled to 1,000 and 1,050°C, respectively. TE_FC and TE_SC were cooled with constant cooling rates at 1 and 0.1°C/min, respectively, to compare whether varied cooling rates would affect the results of the reaction experiment. TE_Ls is a long duration run, which was reacted at 1,180°C for 27 hr before step-cooling to 1,000°C. Our reaction experiments produced a variety of reaction textures and mineral compositional variations, depending on the run conditions and the compositions of the reacting melt.  (Figures 2 and S1). Euhedral clinopyroxenes crystallized in the upper part of the melt in TE1 ( Figure S1a). Relatively large olivines (~100 μm in length) crystallized near the melt-troctolite interface, and small olivine and plagioclase crystals are present in the upper part of the melt in Run TP1 ( Figure S1b).
Interstitial melt phase in both TE1 and TP1 runs are volumetrically less than those in TE_mush run due to the absence of melt phase in the starting troctolite in the former two runs ( Figure S1). Nonetheless, melt infiltrated the troctolite in TE1 and TP1, with the amount of the interstitial melt gradually decreasing away from the melt-troctolite interface, forming a melt infiltration zone in the upper part of the troctolite (Figure 2). Plagioclase grains, much smaller than the starting plagioclase and mostly round in shape, were found to be included in olivine grains in the melt infiltration zone (Figure 2b). These plagioclase-bearing olivines are euhedral, in contrast to olivine grains below the zone. Because no plagioclase inclusions were present in the olivine starting material, this observation indicates that the euhedral olivine was precipitated either during or after plagioclase was partially dissolved.

Step-Cooling Experiments: Clinopyroxene Precipitated as Reaction Product
Step-cooling experiments produced olivine-bearing gabbro and clinopyroxene-bearing troctolite from top to bottom ( Figure 3). In the long duration Run TE_Ls, the gabbros contain relatively large clinopyroxene grains, feldspar laths, olivine, sulfide, and melt. Interstitial melt and clinopyroxene were present throughout the troctolite ( Figure 3). Clinopyroxene shows poikilitic textures with olivine and plagioclase chadacrysts ( Figure 3d), forming darker patches in the troctolite relative to interstitial melt bearing troctolite with a lighter green color in the element map ( Figure 3a).

Chemical Consequences of Melt-Mush Reaction 3.2.1. Chemical Variations of Melts in Isothermal Experiments
Overall, melt compositions gradually shift to higher MgO and Al 2 O 3 , lower FeO, and slightly lower TiO 2 and Na 2 O contents toward the melt-troctolite interface ( Figure 4). Mg# in the melt increased progressively from the original values for evolved and primitive melts at 54   Step-cooling is at an average rate of~0.15°C/min to the final temperatures.

Shifts in Melt Fractionation Sequences Following Reaction: Step-Cooling Experiments
To compare the melt fractionation sequence before and after reactions, step-cooling experiments of the two starting melts (i.e., without troctolite) were run. The highest Mg# of clinopyroxene and Fo of olivine in the melt phase in the step-cooling experiments can be used to discriminate whether clinopyroxene or olivine is the first phase to crystallize in the melt ( Figure 6). If the two are in Mg-Fe equilibrium, the melt should be close to olivine and clinopyroxene cotectic at the onset of crystallization. If the former plots above the equilibrium field ( Figure 6), clinopyroxene is likely to have crystallized earlier than olivine and vice versa. The mineral compositions in the step-cooling experiment for evolved melt only (Run Es) are clinopyroxene with Mg# up to 82 and plagioclase with An up to 59 (not in Figure 6 due to the absence of olivine in Run Es). After 6-hr reaction with troctolite in TE1, melt compositions at the melt-troctolite interface have higher MgO and Al 2 O 3 contents. In the corresponding step-cooling experiments, TE2, TE_FC, TE_SC, and TE_Ls, clinopyroxene (Mg# up to 83-86), and olivine (Fo up to 80-85) in the gabbro sections are roughly in Mg-Fe equilibrium ( Figure 6) and are thus close to cotectic. Plagioclase in these runs also have much higher An than those in Run Es (up to 66-76, +17 mol%). Therefore, reaction with a primitive troctolite caused the melt, which was originally saturated with clinopyroxene as ferromagnesian phase, to be cosaturated with olivine. Furthermore, reaction led the melt to crystallize more primitive olivine, clinopyroxene, and plagioclase during magma evolution than it would have done without reaction. Troctolite section is composed of plagioclase (pl, dark gray color representing low-Na plagioclase in the troctolite) and olivine (ol, bright red color) with interstitial clinopyroxene (cpx, darker red color surrounding olivine and plagioclase) in b), and interstitial melt (green color) in (c). Plagioclase in the gabbro has a lighter blue color due to higher Na contents in the plagioclase crystallizing from the melt compared to those in the troctolite. (d) Backscattered electron image for poikilitic clinopyroxene (outlined with yellow dashed curve) enclosing plagioclase and olivine chadacrysts (outlined with white curves and labeled with white fonts) in the troctolite section. Plagioclase relict hosted in olivine suggests reprecipitated olivine grew over resorbed plagioclase. The olivine chadacrysts in the poikilitic clinopyroxene are rounded as a result of dissolution, in contrast with the larger olivine grains (labeled with yellow fonts) not enclosed in clinopyroxene.
The mineral compositions in the step-cooling experiment for primitive melt only (Run Ps) are olivine with Fo up to 84, clinopyroxene with Mg# up to 87, and plagioclase with An up to 68, with clinopyroxene and olivine in Mg-Fe equilibrium, indicating that magma evolution of the primitive melt at the run pressure should start with cotectic olivine and clinopyroxene followed by plagioclase fractionation. After 6-hr reaction with troctolite in TP1, melt compositions at the melt-troctolite interface have higher MgO and Al 2 O 3 contents. In the corresponding step-cooling Run TP2, spinel crystals are found at the interface, and the reacted melt crystallized more primitive olivine (Fo up to 87, +3 mol%) and plagioclase (An up to 77, +9 mol%) and less primitive clinopyroxene (Mg# up to 82, −5 mol%) compared to Run Ps ( Figure 6). The presence of olivine crystals in melt phase near the melt-troctolite interface of Run TP1 also indicates that olivine was the first phase to crystallize after reaction.
Although plagioclase crystallization is likely suppressed by the slightly higher pressure of 0.5 GPa of the experiment compared to typical lower oceanic crustal conditions (0.1-0.2 GPa), the findings of more primitive olivine and plagioclase compositions after reaction with a primitive troctolite should apply at lower pressures as well. Clinopyroxene crystallized from the reacted melt, on the other hand, increased in Mg# from 82 to up to 86 in TE runs but decreased from 87 in Run Ps to 82 in Run TP2. Extensive olivine crystallization in the melt in Run TP2 may have lowered the MgO content of the melt, which thereafter crystallized clinopyroxene with lower Mg# than that in Run Ps.

Troctolite Compositional Variations
The starting olivine in the troctolite has Fo content of~90 and NiO content of~0.37 wt.%. After reaction, the olivine compositions show a general decrease in both Fo and NiO contents toward the melt-troctolite interface in all runs (Figures 7a and 7b). Variations of olivine compositions in the troctolite are controlled by both the starting melt compositions and run durations. In isothermal experiments of TP1 and TE1 with same duration (6 hr), olivine cores show significant shifts in the first 400 μm of the troctolite (Figures 7a and  7b), corresponding to~400-μm melt infiltration (Figures 2 and S1). The far-field olivine in the troctolite retains the same compositions to the starting olivine within error and was hence not affected significantly by reactive melt infiltration. In contrast, the olivine core compositions in the troctolite of long-duration Run TE_Ls are characterized by more evolved compositions through the entire troctolite section and are spatially more homogeneous, indicating that melt infiltrated and reacted to the end of the troctolite.
Olivines show sharper decreases in Fo for TE1 than TP1, as would be predicted by the differences in melt composition: Melt Mg#-olivine Fo relationships ( Figure 5) suggest that chemical equilibrium has been achieved at the interface of both TE and TP runs after 6-hr reaction, indicating Fo in olivine are controlled by the Mg-Fe equilibrium between melt and troctolite. Hence, lower Mg# of the evolved melt leads to lower Fo in the olivines after reaction. In contrast to Fo, Ni in olivine compositions are similar between TP1 and TE1 (Figure 7a). Plagioclase, on the other hand, show clear core-to-rim variations with cores in all runs retaining their original compositions and rims having lower An numbers toward the melttroctolite interface (Figure 7c). Interstitial clinopyroxene in the reacted troctolite in the step-cooling runs has high Na 2 O and TiO 2 contents (up to 0.81 and 2.51 wt.%, respectively) with high Mg# (>83).

Discussion
Melt and mineral compositions vary systematically across experimental charges. Through reaction with troctolite, melts become more enriched in MgO and Al 2 O 3 and depleted in TiO 2 and Na 2 O contents and crystallize more primitive olivine and plagioclase compared to those crystallized from the starting melts. On the other hand, the troctolite shifted to more evolved compositions to be equilibrated with the melt near the melttroctolite interface. These chemical variations record the progression of melt-mush reaction and provide important clues to the mechanisms of melt-mush interaction.

Mechanism of Melt-Mush Reaction: Diffusion-Assisted Dissolution and Reprecipitation
Collectively, the observation of interstitial melt in the troctolite in Runs TE1 and TP1, the gradual decrease in melt fraction further into the troctolite, and plagioclase relicts in euhedral olivines (Figure 2) suggest the incipient melt infiltration into the troctolite and the dissolution of plagioclase + olivine by melt. Plagioclase relicts in olivines can be explained by different dissolution and reprecipitation rates of olivine and plagioclase. Experimental study of the dissolution of olivine and plagioclase in basaltic magma suggests dissolution of plagioclase is~6 times faster than olivine, whereas the growth rate of plagioclase is at least 10 times slower than olivine (Donaldson, 1985). Hence, it is likely that melt dissolved plagioclase rapidly with minor plagioclase reprecipitation, whereas olivine was more likely to be dissolved and reprecipitated simultaneously. Therefore, during dissolution and precipitation of olivine, plagioclase is partially dissolved, and the relic grains were incorporated into growing olivine crystals (euhedral in Figure 2b).
The chemical variations of olivine and plagioclase in the troctolite after reaction show different characteristics, also suggesting different reaction mechanism for olivine and plagioclase. Even the core compositions of olivine crystals have been modified in all the experiments. In contrast to olivine, the cores of plagioclase in all runs retained their original  Roeder & Emslie, 1970). Gray circle and triangle represent the starting melt and olivine compositions of Runs TE1, TE_mush, and TP1, respectively. Blue circle, purple diamond, and orange triangles represent melt and olivine compositions near the melt-troctolite interface for the three runs after reaction, and they are plotted within the field defined by K D of 0.30 ± 0.03, suggesting chemical equilibration was achieved at the interface after reaction. Olivine in both the troctolite (tr) and melt (m) are plotted for Run TP1.  (Grove & Bryan, 1983;Sisson & Grove, 1993) and K d ol/melt from 0.27-0.33 (e.g., Roeder & Emslie, 1970). Experimental charges plotted below the equilibrium field indicate olivine crystalized early than clinopyroxene and vice versa, demonstrating the first phase to crystallize in the gabbro sections changed from cotectic clinopyroxene and olivine to olivine in TP runs (shown by the orange line with an arrow) and from clinopyroxene to cotectic olivine and clinopyroxene in TE runs

10.1029/2019GC008368
Geochemistry, Geophysics, Geosystems compositions, with rims having lower An contents toward the melttroctolite interface (Figure 7c). The difference in core-to-rim compositions of olivine and plagioclase appears consistent with the observed fast diffusion rates of Mg-Fe-Ni in olivine (e.g., Chakraborty, 1997) and slow diffusion of CaAl-NaSi in plagioclase (e.g., Grove et al., 1984;Morse, 1984). To evaluate the control of diffusion on the Mg-Fe equilibrium between melt and olivine at the interface, we modeled the effect of diffusion on the core-to-rim variations of olivine crystals in the troctolite near the reaction interface in Run TE1, using an exact solution method from Crank (1975) and diffusion coefficiences calculated based on Dohmen and Chakraborty (2007), Chakraborty (2010) and Ulmer and Luth (1991). During the 6-hr experimental duration, diffusion should only affect the Fo and Ni contents within~2 μm of the rim ( Figure S2). However, even the core compositions of olivine near the interface are in chemical equilibrium with the interface melt for the isothermal runs, suggesting accelerated rate of reequilibration by diffusion-assisted dissolution and reprecipitation. This is consistent with the presence of plagioclase inclusions in olivine crystals ( Figure 2b). Therefore, the reaction occurring in the troctolite during the isothermal experiments is as follows: The relative portions of plagioclase and olivine dissolved in the melt are discussed in section 4.2.
The isothermal experiments TE1 and TP1 demonstrate that after 6 hr of reaction, melt dissolved troctolite around the margins of olivine and plagioclase, forming an interstitial melt phase. In most cooling experiments, clinopyroxene formed around some of the olivine and plagioclase grains of the troctolite during cooling to form the clinopyroxene-bearing troctolite ( Figure 3d). In the long duration Run TE_Ls, melt-troctolite reaction for 27 hr has produced relatively large porosity in the troctolite, and interstitial melt phase is present in the entire troctolite section (green in the element map in Figure 3a). Upon cooling, clinopyroxene started to form from this interstitial melt around some of the olivine and plagioclase grains ( Figure 3d) and reduced the porosity of the troctolitic mush. The olivine chadacrysts are rounded as a result of earlier dissolution, which is in contrast with the larger olivine grains not enclosed in clinopyroxene (Figure 3d). These observations indicate that the reaction that occurred in the troctolite during the step-cooling experiments included dissolution and reprecipitation of olivine and plagioclase as well as the precipitation of clinopyroxene as the new mineral, following the reaction: (2)

Quantifying Dissolution and Reprecipitation
Systematic variations in the reacted melt compositions in Runs TE1, TP1, and TE_mush can be used to quantify the dissolution (and reprecipitation) of plagioclase and olivine in the troctolite during melt-mush reaction. We used the reaction in equation (1), which includes the roles of both dissolution and reprecipitation of olivine in modifying melt compositions in the isothermal experiments. We neglected plagioclase reprecipitation as a simplification. This is a reasonable assumption since we observed no olivine relics in plagioclase and mostly quench growth of plagioclase rims. The oxides in the melt that change considerably, such as Al 2 O 3 , MgO, TiO 2 , and FeO, are used to calculate the net amount of dissolution. Although experimental study suggested dissolution of plagioclase is~6 times faster than olivine in basaltic magmas (Donaldson, 1985), the actual dissolution rate of plagioclase to olivine may be lower considering Figure 7. Plots of (a) Fo and (b) NiO contents in the olivine cores for Runs TP1, TE1, and TE_Ls and (c) Anorthite (An) in the cores and rims of plagioclase for Run TE1 in the troctolite as a function of distance away from the melt-troctolite interface. Rim analyses were not plotted as most of the olivine are too small to yield good-quality data on the rims due to the possible effect of secondary fluorescence contamination (e.g., Llovet & Galan, 2003). Gray shaded fields represent the starting olivine and plagioclase compositions.
The starting evolved and primitive melt compositions were used to model the assimilation of the starting olivine and plagioclase with varied proportions and the reprecipitation of the olivine at the melt-troctolite interface for each run. The equilibrium between the core compositions of olivine near the interface with melt, which could not be explained by diffusion in 6 hr ( Figure S2), suggests that the ratio of reprecipitated to dissolved olivine is relatively high. Our model suggests that high ratios (>0.6) would lead to decreasing MgO in melt, inconsistent with melt variations. With the reprecipitated to dissolved olivine ratios in the range 0.4-0.6, the interface melt compositions can be accounted for with dissolved plagioclase/olivine ratios from 2:1 to 1:1. An example is shown in Figure 8 with a reprecipitated to dissolved olivine ratio of 0.5.
In this section, we compare the compositional variations in melts and minerals in our experimental charges with those observed in samples from lower oceanic crustal cumulates. We emphasize general mineral compositional trends rather than absolute values and ranges, as the latter depend specifically on melt composition, temperature, pressure, duration, and melt-rock ratio during melt-mush reaction.

Fo-Ni Variations in Olivine in Natural Lower Oceanic Crust Cumulates
Fo and Ni variations in olivine of cumulates have been used as indicators for magma chamber processes (e.g., Coogan, 2014;Rampone et al., 2016;Sanfilippo et al., 2013;Suhr et al., 2008). Our experiments suggest that primitive mush reacting with variously evolved magmas would lead to a decrease in both the Fo and Ni contents of the olivine. Ni contents are predominantly controlled by the change in Ni partition coefficient between evolved and primitive melts. The higher the Mg# of the melt, the lower partition coefficients of Ni in olivine (Hart & Davis, 1978). Thus, the primitive melt in TP1 with higher Mg# would have a lower partition coefficient for Ni in olivine, indicating that reacting with a more primitive melt would decrease the Ni in olivine more rapidly than an evolved melt (Figure 9). Prolonged interaction with evolved melts (Run TE_Ls) leads to very low Ni contents (NiO < 0.1%; Figure 9).
To compare the relative effects of magma differentiation and melt-mush reaction on the Fo versus Ni correlation of olivine in the cumulates, we modeled olivine compositional variations produced by the fractional crystallization of a primary MORB (ALV0527-001-001, Gale, Dalton, et al., 2013) using MELTS (Ghiorso & Sack, 1995). As MORBs are known to undergo sulfide-saturated differentiation before eruption (e.g., Bézos et al., 2005;Yang et al., 2014), and Ni, as a chalcophile element, is partly controlled by sulfide saturation during magma evolution, we modeled both sulfide-saturated and sulfide-undersaturated fractionation trends. Ni contents in the olivine were calculated with composition-and temperature-dependent K D for Ni partitioning between olivine and melt (Matzen et al., 2013) and D Ni of 500 (Peach et al., 1990) for Ni partitioning between sulfide and melt. Sulfide-saturated differentiation used a silicate to sulfide fractionation ratio of 1,000, equivalent to a sulfide segregation rate of~10 ppm/% fractional crystallization as estimated for MORB evolution in Bézos et al. (2005) and Yang et al. (2014). The gentler decreases in Ni for Fo lower than 84 in the modeled fractionation trends are due to the addition of clinopyroxene on the liquidus. In Figure 9, we also plot olivine phenocrysts compositions from MORB, as well as lower oceanic crustal cumulates from fast-(Hess Deep & Pito Deep, East Pacific Rise), slow-(MARK area, Hole 1309D, Mid-Atlantic Ridge) to ultraslow-spreading (Hole 735B, Southwest Indian Ridge) ridges. As demonstrated in Figure 9, a large number of olivines in lower oceanic crustal cumulates and olivine phenocrysts in MORB have both higher and lower NiO at a given Fo content than olivine from both sulfide-saturated and sulfide-undersaturated fractionation models, suggesting that olivine compositional variations in both MORB and cumulates cannot be explained by fractionation alone. Post-crystallization reactions of olivine in cumulates were suggested to have different effects on Fo and Ni contents, which were proposed to account for those olivines with both slightly higher (Coogan, 2014;Suhr et al., 2008) and slightly lower Ni (Sanfilippo et al., 2013) at a given Fo. Our experimental work revealed that Ni variations in olivine after melt-mush reaction would largely depend on the Mg# of the reacted melt. Compared to the olivine from normal fractionation trends before clinopyroxene crystallization, higher Mg# in the primitive melt (Run TP1) would result in olivine with lower Ni, while lower Mg# in the evolved melt (Run TE1) produced olivine with higher Ni contents at high Fo (>84) in the reacted troctolite. Yet prolonged reaction with troctolite by the evolved melt in Run TE_Ls produced olivine with much lower Ni than the fractionation trends after clinopyroxene crystallization at low Fo. Therefore, variably evolved melts reacting with primitive cumulates could be a possible explanation for the wide range of Ni variations in olivine in both lower oceanic crust cumulates and MORB globally (Figure 9).

Origin of High-Mg# Clinopyroxene in the Lower Oceanic Crust
During fractional crystallization at crustal depth (≤2 kbar), MORB is not predicted to crystallize clinopyroxene until after significant olivine and plagioclase fractionation (Grove et al., 1992). Therefore, the Mg# of the magma has already dropped considerably when clinopyroxene saturates, with experiments providing an average Mg# of~83 for the first clinopyroxene crystallized (Grove, 1990;Grove et al., 1992;Grove & Bryan, 1983;Tormey et al., 1987;Yang et al., 1996). Nonetheless, high-Mg# clinopyroxene (Mg# ≥ 85) is not uncommon in the lower oceanic crust and has been found in the Atlantic (Kane Megamullion, Lissenberg & Dick, 2008; IODP Hole 1309D in the Atlantis Massif, Drouin et al., 2009;Suhr et al., 2008), Pacific (Pito Deep, Perk et al., 2007, and Indian (ODP Hole 735B, Dick et al., 2002) oceans. Both high-pressure crystallization (Elthon, 1987) and melt-mush reaction (Lissenberg & Dick, 2008) have been invoked to explain such high-Mg# clinopyroxene. Olivine gabbros collected from within 1 km of the base of the sheeted dike complex at Pito Deep contain clinopyroxene with Mg# up to 89 (Perk et al., 2007), demonstrating that Mg-rich clinopyroxene can also crystallize during MORB differentiation at low pressures.
The Mg# variations of clinopyroxene in the melt portions of TE and TP runs mainly depend on the extent to which olivine crystallization preceded clinopyroxene crystallization. By reaction with troctolite in TP1, the olivine-saturated primitive melt increased its Mg# and crystallized olivine with high Fo (up to 87) as a result ( Figure 6). Such early crystallization of magnesian olivine would deplete the melt in MgO, which explains the relatively low Mg# of clinopyroxene (~82; Figure 6) formed upon the step cooling of this reacted melt (Experiment TP2). In contrast, the Fe-Mg systematics indicate that olivine crystallization did not precede clinopyroxene during the crystallization of the evolved melt, even after reaction (Experiments TE2, TE_SC and TE_Ls; Figure 6). The increased Mg# in the melt after reaction, therefore, leads to the formation of high-Mg# clinopyroxene (Mg# up to 86). Although clinopyroxene crystallization is likely enhanced by the slightly higher pressure of 0.5 GPa of the experiments compared to typical lower oceanic crustal conditions (0.1-0.2 GPa), the presence of high-Mg# clinopyroxene in the TE runs but not in TP runs indicates that the factor controlling high-Mg# clinopyroxene precipitation is not higher pressure as both runs were at the same pressure. Hence, high-Mg# clinopyroxene is most likely to form when a relatively evolved, clinopyroxene-saturated melt reacts with primitive cumulates. This conclusion, which is not sensitive to crystallization pressure, is in keeping with the suggested formation of high-Mg# clinopyroxene in natural oceanic (Lissenberg & Dick, 2008;Lissenberg & MacLeod, 2016) and ophiolitic (Basch et al., 2018(Basch et al., , 2019Rampone et al., 2016;Sanfilippo, Tribuzio, et al., 2015) gabbros.
The clinopyroxenes crystallized in the (reacted) melt portions of the experiments contain moderate TiO 2 and Na 2 O contents, as is expected from fractional crystallization models ( Figure 10). Interstitial clinopyroxene in the troctolite in the step-cooling runs, on the other hand, extends to very high Na 2 O and TiO 2 contents (up to 0.81 and 2.51 wt.%, respectively) with high Mg# (>83), distinct from the clinopyroxene compositions predicted to form by fractional crystallization (Figure 10). We infer that these titanium and sodium-rich clinopyroxenes form from evolved interstitial melts (i.e., high TiO 2 ) that equilibrated with primitive olivine and thus acquired high Mg# (cf. Borghini et al., 2018;Borghini & Rampone, 2007;Lissenberg & Dick, 2008). There are a significant amount of clinopyroxenes in lower oceanic crustal cumulates with compositional shifts toward higher Na 2 O and TiO 2 at high Mg# relative to the fractionation trend (Figures 10a and 10b), indicating their compositions might have been affected by melt-mush reaction (Lissenberg & Dick, 2008;Lissenberg & MacLeod, 2016;Sanfilippo et al., 2019). Furthermore, clinopyroxene from some lower crustal sections of ophiolites and layered intrusions have even higher Na 2 O and TiO 2 contents (Figures 10c and  Dick et al., 2002;Hébert et al., 1991;Lissenberg et al., 2013;Miller et al., 2009;Natland & Dick, 1996;Perk et al., 2007;Ross & Elthon, 1997;Suhr et al., 2008;Yamazaki et al., 2009) and olivine phenocrysts in mid-ocean ridge basalt (MORB; from PetDB database), respectively. Orange dots represent sulfide-undersaturated fractionation trend of a primary MORB (ALV0527-001-001; Gale, Dalton, et al., 2013). From this melt olivine, followed by plagioclase and clinopyroxene is removed by fractional crystallization, and each dot represents a 1°C temperature step, modeled using MELTS (Ghiorso & Sack, 1995). Blue dots represent the sulfide-saturated fractionation trend from the same melt with a silicate/sulfide fractionation ratio of 1,000 based on the sulfide segregation rates during MORB differentiation from Bézos et al. (2005) and Yang et al. (2014). 10d), suggesting an important role for such melt-mush reaction in the formation of both continental and oceanic cumulates.

Implications for MORB Major Element Barometry and the Origin of High-Al MORB
Major element variations of MORB are commonly used to calculate the pressures at which they underwent (partial) crystallization in the crust prior to eruption (e.g., Danyushevsky et al., 1996;Grove et al., 1992;Herzberg, 2004;Villiger et al., 2007;Wanless & Behn, 2017;Yang et al., 1996). Such calculations depend heavily on the MgO-CaO-Al 2 O 3 systematics of the melt with higher pressures resulting from higher Al 2 O 3 and/or lower CaO, reflecting the more prominent role of clinopyroxene crystallization at higher pressure (Lissenberg & Dick, 2008). Our experimental results demonstrate that interaction of melts with primitive cumulate minerals modifies melt compositions, with higher Al 2 O 3 and MgO after reaction in the isothermal experiments (Figure 4). Such compositional changes of the melt may significantly affect the calculated crystallization pressure. To determine the effect of melt-mush reaction on the calculated crystallization pressure, we applied the parameterization of Herzberg (2004) to the evolved melt compositions in TE1 runs, but note that similar results would be obtained with other MORB major element barometers (Danyushevsky et al., 1996;Grove et al., 1992;Villiger et al., 2007;Yang et al., 1996). We chose the melt compositions that match the criteria proposed in Herzberg (2004) to be saturated in clinopyroxene, olivine, and plagioclase (CaO < −0.3MgO + 14.5). As shown in Figure 11, meltmush reaction would result in an increase in the calculated crystallization pressures from~0.1 GPa to above 0.4 GPa, corresponding to an apparent shift in crystallization depth of >12 km. Such a result is  Coogan et al., 2002;Dick et al., 2002;Hébert et al., 1991;Lissenberg et al., 2013;Miller et al., 2009;Natland & Dick, 1996;Perk et al., 2007;Ross & Elthon, 1997;Sanfilippo et al., 2019;Suhr et al., 2008;Yamazaki et al., 2009) and (c and d) Erro-Tobbio ophiolite (Basch et al., 2019) and Panzhihua layered intrusion (Pang et al., 2009). Black dots represent fractional crystallization trend of the same MORB magma as described in Figure 9. Arrows show the effect of melt-mush reaction on the clinopyroxene compositions. consistent with previous predictions by Lissenberg and Dick (2008), who stated that the MgO-CaO-Al 2 O 3 relationships of MORB could result from melt-mush reaction instead of polybaric fractional crystallization. Hence, melt-mush reaction significantly affects MORB majorelement barometers, resulting in erroneous pressures. It follows that such barometers can only be used reliably if reaction can be ruled out.

Conclusions
To investigate the mechanisms and chemical consequences of reaction between basaltic magma and primitive cumulates, we conducted melt-mush reaction experiments by juxtaposing troctolite with either a primitive or an evolved MORB. From our results and comparison to natural data, we draw the following conclusions: 1. Melt-mush reactions proceed through diffusion-assisted dissolution and reprecipitation. The troctolite in the isothermal experiments shows interstitial melt phase after reaction, indicating melt infiltration and dissolution of plagioclase and olivine with plagioclase/olivine ratios of 2 to 1. Even the cores of olivine crystals in the troctolite at the interface have achieved Mg-Fe equilibrium with the melt, and they contain plagioclase inclusions, neither of which can be accounted for by diffusion on the time scale of the experiments, indicating reprecipitation of olivine during reaction. Furthermore, the troctolite in the stepcooling experiments shows both interstitial melt and poikilitic clinopyroxene with rounded olivine and plagioclase chadacrysts, indicating clinopyroxene precipitated upon cooling as reaction product. 2. Through reaction with troctolite, melts became more enriched in MgO and Al 2 O 3 and depleted in TiO 2 and Na 2 O contents and crystallized more primitive olivine and plagioclase compared to those crystallized from the starting melts. On the other hand, the troctolite shifted to more evolved compositions to be equilibrated with the melt near the melt-troctolite interface. Olivines in the troctolite have different Fo versus Ni correlations for different starting melt compositions, resulting from varied Ni partition coefficients with different MgO contents of the melt. Clinopyroxene precipitated in the troctolite has high Mg# (>83) with enriched Na 2 O and TiO 2 contents, distinct from those crystallized in the melts. Therefore, high-Mg# clinopyroxene can result from melt-mush reaction and cannot simply be used as evidence for high-pressure crystallization. 3. The higher MgO and Al 2 O 3 contents of reacted melts significantly affect the pressures inferred from MORB major element barometers. The reacted melt yielded higher calculated pressures (~0.4-0.5 GPa) than that calculated from the starting compositions (~0.1 GPa), indicating that major element-based barometers for MORB fractionation can only be used reliably if reaction can be ruled out. 4. The Fo-Ni correlations of olivine, as well as Mg# versus TiO 2 and Na 2 O correlations of clinopyroxene in cumulate rocks, can be used to identify the role of melt-mush reaction in the chemical variations of Figure 11. Plot of fractionation pressures calculated using the method in Herzberg (2004) as a function of distance away from the melt-troctolite interface in Run TE1. Melt compositions that match the criteria of CaO < −0.3MgO + 14.5 as proposed in Herzberg (2004) were chosen for calculation. Orange arrow shows the trend for the effect of melt-mush reaction on the calculated pressures. Gray line indicates the calculated pressure for the starting evolved melt composition as listed in Table 1.

10.1029/2019GC008368
Geochemistry, Geophysics, Geosystems cumulates. By comparison with cumulates from lower oceanic crust, ophiolite, and layered intrusion, we propose that melt-mush reaction plays an important role during magma emplacement and transport in crystal mush in both oceanic and continental magma systems.