4.1. General Observations and Interpretation of the Experiments
 Quench crystallization, evidenced by low MgO contents of melt pools, is an intrinsic process operating on the melt compositions in the IR and MIX experiments. However, the relatively large enrichment in alkalis and silica in glasses near the interface does not solely reflect quench modification since these melt pools are the largest, and therefore least affected by quenching. Comparison of quench-modified glass and quench-free glass (in vitreous carbon layers) from the same experiment [Lundstrom, 2000, Figure 2] indicates the following concentration increases due to quenching: SiO2, +1%, Al2O3, +8%, FeO, +7%, CaO, +4%, Na2O, +12% and TiO2, +16%. These changes are small relative to the changes in the melt pool compositions observed spatially in traverses through the IR experiments. Because quench crystallization does affect the results, melt compositions for all IR experiments are presented both in raw form and after correction back to MgO of 10 wt.% through incremental addition of olivine in equilibrium with the melt. This process will not correct all melt pools back to their original composition given that some pyroxene crystallization may also occur; however, it should adequately compensate for the enrichment in incompatible elements from quenching and reduce some of the variability in the profiles.
 The rapid infiltration of Na into partially molten peridotite causes incongruent dissolution of orthopyroxene. Na2O diffuses >600 μm into the peridotite in 10–30 min experiments with the peak in Na2O generally coincident with the peak in SiO2. High SiO2 contents are unlikely to reflect diffusion because Si is typically the slowest diffusing species in silicate melts [Watson and Baker, 1991]. Rather, the correspondence between the Na2O and SiO2 peaks indicates a causal relationship between the addition of Na2O and reaction to form SiO2 rich melt. Each infiltration-reaction experiment shows a decrease in the orthopyroxene mode in the area of alkali infiltration but little reduction far from the interface. Complimentary decreases in orthopyroxene mode and increases in olivine and melt modes in the harzburgite-Cl experiment (IR-37) indicate incongruent dissolution, reflecting the shift in the olivine-orthopyroxene phase boundary with addition of Na to the melt [Kushiro, 1976; Ryerson, 1985].
 Shaw et al.  and Shaw  have shown that silica rich melt forms around orthopyroxene as it dissolves in alkali basalts. However, Shaw  concludes that direct melt-mineral reaction is unlikely to explain the creation of silica rich glasses in mantle xenoliths because continued reaction and diffusion quickly erase any boundary layer around orthopyroxene. The DIA mechanism is distinct from the dissolution experiments of Shaw and co-workers as it reflects addition of alkalis into the peridotite without addition of slow diffusing elements like TiO2 which have opposite effects on melt SiO2 content [Hirschmann et al., 1998]. Indeed, the direct mixture experiment results show that addition of both alkalis and TiO2 and P2O5 to KLB-1 result in melts with small to negligible increases in SiO2 content. In contrast, DIA leads to high silica melts that can be in equilibrium with peridotite over the lengthscale of a xenolith.
 Little is known about the actual melt species diffusing during multicomponent diffusion in silicate melts. Charge neutrality within an experiment must be maintained during diffusion but whether individual ions diffuse or ionic complexes diffuse is not known. During basalt-granite diffusion-reaction experiments, Watson and Jurewicz  observed high rates of Na diffusion with no obvious mechanism for maintaining charge neutrality. These authors suggested that O−2 diffusion compensated for Na+ diffusion. In the basalt-granite case, Na will diffuse in the same direction as Ca (from basalt into granite) whereas activity gradients drive Na and Ca to diffuse in opposite directions during a basanite-tholeiite couple as in the IR experiments. Although a definitive interpretation is not possible, several observations suggest that Ca counter diffusion occurs in response to Na diffusion. First, Ca diffuses faster than all other cations except Na and Li in melt-melt diffusion couples, with poor fits to binary profiles. Second, the CaO glass concentration in IR experiments is low in areas with elevated Na2O concentrations. Although decreasing CaO content could reflect local increase in the clinopyroxene mode, Na2O infiltration should destabilize clinopyroxene resulting in increased CaO in the melt, not decreased. Increased clinopyroxene modes that could explain CaO depletions are not observed in any experiment. Finally, in experiment IR-39, there is also evidence for MgO diffusing out of the peridotite.
 The repeated formation of the boundary zone in the enhanced-Cl basanite experiments may indicate an important mechanism within the DIA process. All 3 experiments produce peaks in CaO and Cl within the reacted peridotite devoid of pyroxene adjacent to the interface. Similar elemental behavior (CaO, TiO2 and Al2O3 enrichment) is observed within other diffusive infiltration-reaction experiments (Z. Morgan, and Y. Liang, An experimental and numerical study of the kinetics of harzburgite reactive dissolution with applications to dunite dike formation, manuscript submitted to Earth and Planetary Science Letters, 2003). Although the peak in CaO in the two lherzolite experiments could reflect clinopyroxene dissolution, this cannot explain the CaO enrichment in the harzburgite experiment.
 Chlorine enrichments may fingerprint boundary zone development during DIA in the mantle. Although Cl diffusion in silicic melts has been studied previously [Bai and Koster van Groos, 1994; Watson, 1991, 1994], little work on Cl diffusion in mafic melts exists. The relatively large Cl diffusion coefficient indicates that Cl can play an active role in DIA. The repeated correlation of CaO and Cl peaks and the observation that Cl diffuses toward the boundary zone from both ends of the couple in IR-36 suggests that Cl activity gradients are primarily influenced by variations in melt CaO content. This is consistent with work by Webster et al.  and Webster and McBirney  that have shown that Cl solubility is highly dependent on melt Ca and Mg contents, possibly due to the formation of Ca-Cl complexes [Webster and Mathez, 2001].
4.2. Is the DIA Process Important to Mantle Melting?
 Both trace element depletions in abyssal peridotites [Johnson et al., 1990] and the observation that MORB is undersaturated with orthopyroxene [O'Hara, 1965; Stolper, 1980] require that some of the melt beneath a ridge ascend by channelized flow [Kelemen et al., 1997]. On the basis of trace element relationships, Kelemen et al.  have shown that dunites likely represent pathways for channelized flow. If silica-poor melt from depth ascends in channels, then gradients in melt silica content will likely exist between the melt in a conduit and the melt in surrounding peridotite at shallow mantle depths, causing DIA to occur. How important this process is in affecting the major element content of melts at the surface remains debatable. Addressing this requires better answers to three questions: (1) Over what depth range and for how long will a parcel of peridotite be exposed to the DIA process? (2) How is sodium originally distributed in the mantle source region and how do Na2O contents of ascending melts vary with depth? (3) How is a melt channel system spatially organized in a melting region?
 The IR experiments show that DIA could occur at 0.5–1 GPa pressure or 15 to 30 km deep in the mantle. The effects of DIA should become even more pronounced at shallower depths (<15 km) because the addition of Na has a greater effect on the melt silica activity coefficient at low pressure [Hirschmann et al., 1998]. Thus estimating that DIA occurs as peridotite ascends from 30 to 15 km depth (or even to the MOHO at fast spreading ridges) is not unreasonable. If the mantle passively upwells beneath ridges [Bottinga and Allegre, 1976; Sleep, 1975; Toomey et al., 1998; Lundstrom et al., 1998] at mm to cm/yr rates, peridotite adjacent to a melt conduit could be exposed to DIA for several million years. This timescale controls the flux of Na2O that the peridotite is exposed to.
 The amount of melting added by DIA will depend on the gradients in Na2O between channels and surrounding peridotite as well as the overall flux of Na2O through the channels, and the advective velocity of intergranular melt in the surrounding peridotite. Initial melts of peridotite should be rich in incompatible elements including alkalis. Alkali basalts, often interpreted as early formed small degree melts, typically have garnet trace element signatures indicating pressures of origin >2.5 GPa. However, melting experiments at >3 GPa relevant to a MORB source with 0.3 wt% Na2O produce low degree melts having low Na2O contents (<2 wt% [Longhi, 2002]). Thus some uncertainty exists about the Na2O contents of ascending basalts and gradients across a melt channel-peridotite interface. However, alkali rich basalts are observed in both MORB and OIB settings, attesting to the presence of melts similar to the basanite used here and possibly indicating sources significantly enriched in Na2O relative to depleted mantle.
 Inferring the gradients in Na that existed in the mantle during melting from observed melts at the surface is difficult because if the DIA process occurs, it acts to reduce any gradient in melt Na2O. Indeed, Na2O concentrations of MORB are much less variable than elements of similar incompatibility [Langmuir and Hanson, 1980] supporting the idea that Na variability may have been reduced by DIA. Melt inclusion Na2O contents, independent of being hosted by olivine or plagioclase, are also much less variable than TiO2, despite the fact that the titanium partition coefficient should be 2–4 times larger than sodium's in the shallow mantle (Figure 12). If DIA impacts melting significantly, mantle sources and initial melts might be considerably more heterogeneous in Na2O than can be inferred from observed melts.
Figure 12. Histograms of melt inclusion Na2O and TiO2 concentrations for both olivine and plagioclase hosted samples from the study of Sours-Page et al. . Despite the fact that Na2O should be more incompatible in mantle minerals, variations in TiO2 are far greater than those of Na2O, regardless of host crystal. A possible explanation for this observation reflects the efficiency of the DIA process in homogenizing gradients in Na2O during the melting process.
Download figure to PowerPoint
 Assuming magma generation beneath a ridge is a steady state process, the flux of Na2O through a melt channel determines the extent of DIA influence on a parcel of peridotite adjacent to the channel. In one scenario, Na2O could be stripped from a wide region of the mantle by early formed deep melts beneath a ridge, ascend up through the melt conduit and disperse back into harzburgite during DIA at shallow depths. In this way, the Na2O flux resulting from melting over a wide area is focused onto a smaller area causing the peridotite at shallow depths to melt more extensively (a geometrical magnification of the amount of Na2O available for DIA). However, even if there is no geometric effect accentuating the Na2O flux, the DIA process could still dramatically alter melting at shallow depths. Effectively, the DIA process re-fertilizes a peridotite in Na2O at shallow depths where the effect of Na2O on melting relations (relative orthopyroxene-olivine stability) is greatest [Hirschmann et al., 1998]. Although Na2O is stripped from peridotite by the time the peridotite arrives at shallow depths in a non-DIA moderated melting process, Na2O will continue to affect the amount and composition of shallow melts in a DIA moderated melting column.
 Possibly the most important aspect to determining the impact of DIA is the spacing between melt channels and the extent of horizontal advective flow between the channels. Although advection is much more important than diffusion during chemical transport in the vertical direction, diffusion could play a role in horizontal transport of Na2O. Pressure gradients could drive advection of melt toward conduits [Stevenson, 1989; Hall and Parmentier, 2000; Spiegelman et al., 2001], in the opposite direction of Na diffusion. If so, advective transport of Na out toward the conduit will likely overwhelm Na diffusion away from conduits. However, the rheology of partially molten peridotite remains poorly understood and the actual flow regime remains highly ambiguous. If buoyancy forces are much greater than the suction forces driving horizontal advection, then the component of horizontal velocity will be small. If so, the rate of Na diffusion can be significant relative to that of advection. If the overabundance of Na in abyssal peridotites is attributable to DIA, then this observation may provide constraint on the problem of melt suction near channels.
 Dunite makes up 5–15% of most ophiolites and its observed distribution can provide constraint on the spacing between melt conduits [Kelemen et al., 1997]. Size and frequency of dunites within a dunite-rich area of the Ingalls ophiolite follow a power law distribution with dunites 1–10 cm in width occurring at a frequency of ∼10/m [Kelemen et al., 1999]. Although further work is needed to quantify channel spacing at different ophiolites, centimeter width dunites occur regularly within ophiolites. Even cm-scale channels would increase the impact of DIA by shortening the lengthscale of diffusion needed to affect the peridotite located between melt channels.