The grain-scale processes of peridotite partial melting were examined at 1340°C and 1.5 GPa using reaction couples formed by juxtaposing pre-synthesized clinopyroxenite against pre-synthesized harzburgite. Reaction between clinopyroxenite and harzburgite produces a melt-enriched, orthopyroxene-free, olivine + clinopyroxene reactive boundary layer. Clinopyroxene compositions vary systematically across the reactive boundary layer with compositional trends similar to the published clinopyroxene core-to-rim compositional variations in the bulk lherzolite partial melting studies conducted at similar P-T conditions. The growth of the reactive boundary layer is at the expense of the harzburgite and is consistent with grain-scale processes that involve dissolution, reprecipitation, and diffusive exchange between the interstitial melt and surrounding crystals.
 At moderate pressures, partial melting of spinel lherzolite produces olivine and melt at the expense of pyroxenes via the reaction cpx + opx + sp → ol + melt. This melting relation has been extensively studied in the laboratory using natural and synthetic starting compositions [e.g., Kushiro, 2001, and references therein]. Most of the studies have focused on the composition of the melt produced at prescribed P-T conditions, and equilibrium was assumed when the melt composition did not change with time [e.g., Stolper, 1980; Takahashi and Kushiro, 1983; Baker and Stolper, 1994]. Although significant progress has been made through laboratory studies, in detail, the grain-scale processes through which peridotite partially melts are still not fully understood. For example, a feature common to most laboratory melting studies using natural starting materials is that residual pyroxenes are chemically zoned [e.g., Stolper, 1980; Takahashi and Kushiro, 1983; Baker and Stolper, 1994; Robinson et al., 1998; Pickering-Witter and Johnston, 2000; Wasylenski et al., 2003]. This zonation has typically been attributed to solid-state diffusion, though the validity of this interpretation has not been carefully examined. Recent theoretical studies have shown that during isothermal-isobaric melting of solid solution forming minerals more complicated processes such as dissolution and reprecipitation are likely to take place [Liang, 2003].
 In order to better understand the grain-scale processes of lherzolite melting, we conducted a set of partial melting experiments at 1340°C and 1.5 GPa using reaction couples formed by juxtaposing pre-synthesized harzburgite and clinopyroxenite. The advantage of the reaction couple method is the production of a thick reaction zone that provides a unique opportunity to study the grain-scale processes of lherzolite melting. As will be shown below, dissolution and reprecipitation play an important role in controlling the distributions of various elements in the melt and residual minerals during lherzolite melting.
2. Experimental Methods
 Melting experiments were carried out in two steps: (1) syntheses of harzburgite and clinopyroxenite, and (2) reaction between the pre-synthesized clinopyroxenite and harzburgite (ol:opx = 54:46, by weight), all at 1340°C and 1.5 GPa. Starting materials were prepared by mixing handpicked optically clean mineral separates from a fertile spinel lherzolite xenolith from central Mexico (sample XDJ-16 from Liang and Elthon ). For simplicity, spinel was not included in this study.
 All experiments were conducted using a 19.1 mm piston cylinder apparatus with a NaCl-Pyrex-graphite-MgO furnace assembly, and a molybdenum capsule consisting of a graphite inner sleeve lined with platinum, similar to those described by Morgan and Liang . Synthesis experiments were run at 1340°C and 1.5 GPa for 48 hrs. After quenching, the cylindrical capsule was cross sectioned, the exposed surfaces were polished to 5 μm finish and dried for at least 24 hrs at 200°C under vacuum. Harzburgite and clinopyroxenite cylindrical half-capsules were then juxtaposed, jacketed with platinum tube to form a reaction couple. To start a melting experiment, the reaction couple was first annealed at 1200°C and 1.5 GPa for 24 hrs, the temperature was then increased to 1340°C at 150°C/min and held at the run condition for 47–79 hrs. Each run product was mounted in epoxy, sectioned longitudinally and polished to 0.3 μm finish.
 Chemical analyses were conducted using a Cameca SX-100 electron microprobe at Brown University. Analytical conditions common to all the analysis were an accelerating voltage of 15 kV, a 20 nA focused beam for minerals, a 10 nA slightly defocused beam for glass, and a 50 nA focused beam for X-ray intensity maps.
3. Results and Discussions
3.1. Experimental Charge
 At the run condition (1340°C and 1.5 GPa), the pre-synthesized clinopyroxenite and harzburgite contain small amounts of melt (<5% and 1%, respectively). Thus our reaction couple consists of melt-bearing harzburgite and clinopyroxenite. Results from two experiments are presented here (duration of 47 and 79 hrs, respectively). Each experimental charge consists of three distinct zones (Figure 1a): (1) clinopyroxene + melt, (2) an orthopyroxene-free reactive boundary layer consisting of olivine + clinopyroxene + melt, and (3) orthopyroxene + olivine + melt. The opx-free reactive boundary layer is located on the harzburgite side of the original clinopyroxenite-harzburgite interface (marked by capsule offset in Figure 1a), suggesting that the reactive boundary layer was formed at the expense of the harzburgite. The thickness of the reactive boundary layer increases with time (up to ∼250 μm for the 79 hr run). Modal abundance of clinopyroxene appears to decrease in the reactive boundary layer with run time (from ∼48% to 29%), whereas those of olivine and the melt increase (50% to 57%, and 3% to 14%, respectively). As highlighted by the X-ray intensity map in Figure 1d, the melt (∼15%) is concentrated in the reactive boundary layer.
3.2. Mineral and Melt Compositions
 Both cpx and opx grains are chemically zoned in the clinopyroxenite and the harzburgite (Figures 1b–1d). Detailed microprobe traverses show that the core-to-rim chemical variations in cpx grains are much smaller than the cpx compositional differences across the reactive boundary layer, especially for the longer duration run. As shown in Figure 2, SiO2, MgO, and FeO in cpx increase towards the reaction front, whereas CaO, Al2O3, and Na2O in cpx decrease from the clinopyroxenite to the harzburgite. In general, opx cores have lower CaO and slightly higher Al2O3 content (asterisks in Figure 3) than opx rims (open triangles in Figure 3). The extent of core-to-rim variation in opx is not correlated with distance from the reaction boundary layer. The olivine composition, in contrast, does not vary significantly throughout the charge.
 Although compositional variations in residual pyroxenes have been noted in most lherzolite melting experiments, core-to-rim compositional variations were reported in only a few studies. Figure 3 compares our measured pyroxene compositions across the clinopyroxenite-harzburgite reaction couple with the pyroxene core-to-rim compositions reported by Robinson et al.  and Pickering-Witter and Johnston . The most striking feature of this plot is that the compositional variations in pyroxenes across our reactive boundary layer are very similar to the pyroxene core-to-rim compositional variations observed in the lherzolite melting experiments. The pyroxene rim compositions from the bulk melting studies (stars in Figure 3) correspond to the pyroxene compositions around the harzburgite-clinopyroxenite interface, whereas the pyroxene core compositions from the previous studies correspond to the pyroxene compositions far away from the reactive boundary layer (i.e., in the interiors of the clinopyroxenite and harzburgite).
 Although small dendritic crystals were observed in the reactive boundary layer, the presence of several large melt pockets in the reactive boundary layer (Figure 1d) allows us to measure the melt composition in that region. In contrast to the systematic variations in cpx across the reactive boundary layer, the melt compositions in this region are more scatter (Figure 2), reflecting the role of local mineralogy and the disequilibrium nature of our experiment. In compositional space, SiO2 (50.8∼52.5%), Al2O3 (15.5∼17.8%), Na2O (4.8∼5.7%), and TiO2 (0.9∼1.05%) define negative but weak correlations with the MgO content of the melt (6.1∼8.4%), where ranges of melt composition are given in the parentheses. No clear correlations were observed for CaO (10.7–12.4%) and FeO (4.9∼6.6%). The FeO-MgO mineral-melt distribution coefficients are 0.27 to 0.39 for olivines and 0.27 to 0.42 for clinopyroxenes in the reactive boundary layer, indicating some quench effect. The SiO2, TiO2, Al2O3, FeO, and CaO abundance in melts from our reaction couple experiments are similar to those reported in bulk lherzolite melting experiments conducted at similar P-T-t conditions [e.g., Baker and Stolper, 1994; Robinson et al., 1998; Pickering-Witter and Johnston, 2000], whereas the high-Na2O and low-MgO characteristics of our melts are similar to those observed in the peridotite-basalt (pyroxenite) mixture or reaction couple experiments [e.g., Kogiso et al., 1998; Takahashi and Nakajima, 2002]. This is not surprising since our experimental charge is in fact a pyroxenite-peridotite reaction couple. A more detailed study of the interstitial melt compositions will be reported elsewhere.
3.3. Grain-Scale Processes
 The presence of concentration gradients in our reaction couple experiments suggests that diffusion plays an important role during clinopyroxenite-harzburgite reaction. Nevertheless, given the relatively slow cation diffusion rates in pyroxenes, it is unlikely that the observed concentration gradients in cpx within the reactive boundary layer are produced by solid-state diffusion alone. For instance, a bulk diffusivity of 10−14 m2/s would be needed to account for the observed concentration variations of Ca, Mg, and Fe in cpx in the reactive boundary layer. This apparent bulk diffusivity is 3–5 orders of magnitude larger than the Mg2+-Fe2+ chemical diffusivity in cpx [Brady and McCallister, 1983] and 3–4 orders of magnitude smaller than the effective binary diffusivities of MgO and FeO in basalt at 1300°C [Lesher, 1994]. Hence diffusion in the melt and in the solid must both contribute to the development of the reactive boundary layer in our experiments.
 The development of the reactive boundary layer is consistent with the following reaction:
where subscripts 1 and 2 designate material in the clinopyroxenite and reactive boundary layer, respectively. The original melt in the clinopyroxenite (melt1) is undersaturated in orthopyroxene and enriched in jadeite component. Hence melt1 reacts with (or dissolves) opx at the clinopyroxenite-harzburgite reaction front. The rate of this reaction is limited by the rate of diffusion in the interstitial melt [Morgan and Liang, 2003]. Incongruent dissolution of opx produces a silica-rich liquid and olivine, as is evident by the increase in olivine grain size in the reactive boundary layer (Figures 1a and 1b). Part of the dissolved opx components mix with melt1, resulting in cpx reprecipitation in the reactive boundary layer (cpx2), as is evident by the increase in cpx grain size in this region (Figures 1b–1d). Hence composition of melt2 is the net result of opx + melt1 reaction, olivine precipitation, and cpx2 reprecipitation in the reactive boundary layer. Similarly, diffusive infiltration of small amount of melt2 into the harzburgite results in opx reprecipitation in the harzburgite. The extent of opx reprecipitation is likely small due to the small fraction of melt in the harzburgite. This interpretation is consistent with the theoretical and numerical analyses of the grain-scale processes of crystal dissolution-reprecipitation in partially molten systems [Liang, 2003]. It also confirms speculation of Kogiso and Hirschmann  regarding the processes of cpx reequilibration in their study of clinopyroxenite melting. The mechanisms outlined here can be used to better understand the grain-scale processes of lherzolite melting in general.
3.4. Applications to Lherzolite Partial Melting in the Laboratory
 It is well known that pyroxenes are chemically zoned in isothermal-isobaric partial melting experiments using natural lherzolite starting compositions [e.g., Stolper, 1980; Takahashi and Kushiro, 1983; Baker and Stolper, 1994; Robinson et al., 1998; Pickering-Witter and Johnston, 2000]. The zoning persists even in long duration runs, while the composition of the melt appears to be homogeneous. The fact that the pyroxene core-to-rim compositional variations in bulk lherzolite melting experiments are similar to the pyroxene compositional variations in our reaction couple experiments (Figure 3) suggests that these systematic variations are produced by similar grain-scale processes, involving dissolution and reprecipitation, in addition to solid-state diffusion.
 The development of pyroxene core-to-rim compositional variations can be understood as follows. During initial stage of partial melting of a spinel lherzolite (e.g., within the first few minutes of an experiment), a transient melt was first produced by disequilibrium melting of pyroxenes at their grain boundaries and grain junctions. The disequilibrium melt dissolves pyroxenes and precipitates olivine. The rate of pyroxene dissolution depends on the rate of chemical diffusion in the melt and the extent of under-saturation between the crystals and the melt. With continuing dissolution and diffusive reequilibration in the melt, compositions of the interstitial melt become uniform throughout the charge. This usually takes a few hours to a day, a time scale defined by cation diffusion in the interstitial melt with length scale given by the mean distance between the pyroxenes. Since cation diffusion rates in minerals are much slower than those in the melt, only the surface of the crystals are in chemical equilibrium with the melt, as noted in numerous experimental studies. It can be shown through a simple mass balance calculation that the weight fraction of pyroxene dissolved is larger than expected by the lever rule. In other words, pyroxenes are over-dissolved (for a detailed discussion of this topic, see Liang ). As the partially molten system continues to equilibrate, over-dissolution is followed by reprecipitation of pyroxenes, which is rate limited by solid-state diffusion. On laboratory time-scales, reprecipitation of pyroxene is too slow to compensate for the initial over-dissolution, giving rise to the observed zoning in pyroxenes. Hence the chemical processes governing the bulk lherzolite partial melting in the laboratory are identical to those observed in our clinopyroxenite-harzburgite reaction experiments. The differences between the two are the initial and boundary conditions. For example, the effective volume of cpx is much larger than that of opx and hence orthopyroxene was completely consumed in our reaction experiments.
3.5. Implications For Pyroxenite-Peridotite Reaction Studies
 Although the primary objective of this study is to understand the grain-scale processes of lherzolite melting, the experimental method developed here can also be used to study the kinetics of pyroxenite-peridotite interaction. The roles of garnet pyroxenite and pyroxenite-peridotite in mantle melting have received considerable attention in experimental petrology lately [e.g., Yaxley and Green, 1998; Kogiso et al., 1998; Rapp et al., 1999; Takahashi and Nakajima, 2002; Hirschmann et al., 2003]. The geometry of our reaction couple is similar to those of Yaxley and Green , Rapp et al. , and Takahashi and Nakajima  in their experimental studies of pyroxenite-peridotite or eclogite-peridotite melting/reactions. The main differences are the starting mineral compositions, the use of the molybdenum outer capsule, and the pre-synthesis step in our experiments. The molybdenum capsule helps to preserve the sample geometry and the offset between the two half-capsules marks the original interface between the clinopyroxenite and the harzburgite at the onset of reaction (Figure 1a). The location of the original interface is essential in quantitative studies of the kinetics of pyroxenite-peridotite interactions.
 Results from our clinopyroxenite-harzburgite reaction experiments show that disequilibrium reaction between a pyroxenite and a peridotite can create a melt and a new peridotite having geochemical features broadly similar to those observed in the peridotite-eclogite reaction experiments [e.g., Yaxley and Green, 1998; Rapp et al., 1999; Takahashi and Nakajima, 2002]. Although more work is needed, the grain-scale processes discussed in this study may help to understand the development of the orthopyroxenite reactive boundary layer observed in the pyroxenite-peridotite reaction experiments.
 We thank Joseph Devine for his help during microprobe analysis, Zachary T. Morgan, Caleb W. Holyoke, III, and Jan Tullis for their help in the laboratory. Helpful reviews were provided by Tetsu Kogiso and Marc Hirschmann. This work was supported in part by NSF grants EAR-0208141 and EAR-9902684.