Petrology and geochemistry of the lower ocean crust formed at the East Pacific Rise and exposed at Hess Deep: A synthesis and new results



[1] The geochemistry and petrology of the lower oceanic crust record information about the compositions of melts extracted from the mantle, how these melts mix and crystallize, and the role of hydrothermal circulation in this portion of the crust. Unfortunately, lower oceanic crust formed at fast spreading ridges is rarely exposed at the seafloor making it difficult to study these processes. At Hess Deep, crust formed at the East Pacific Rise (EPR) is exposed due to the propagation of the Cocos-Nazca spreading center westward. Here we review our state of knowledge of the petrology of lower crustal material from Hess Deep and document new mineral major and trace element compositions, amphibole-plagioclase thermometry, and plagioclase crystal size distributions. Samples from the deeper parts of the gabbroic sequence contain clinopyroxene that is close to being in trace element equilibrium with erupted basalts but which can contain primitive (moderate Cr, high Mg#) orthopyroxene and very calcic plagioclase. Because primitive mid-ocean ridge basalts (MORBs) are not saturated with orthopyroxene or very calcic plagioclase this suggests that melts added to the crust have variable compositions and that some may be in major but not trace element equilibrium with shallow depleted mantle. These apparently conflicting data are most readily explained if some of the melt extracted from the mantle is fully aggregated within the mantle but reacts with the shallow mantle during melt extraction. The occurrence of cumulates with these characteristics suggests that melts added to the crust do not all get mixed with normal MORB in the axial magma chamber (AMC), but rather that some melts partially crystallize in isolation within the lower crust. However, evidence that primitive melts fed the AMC, along with steep fabrics in shallow gabbros (from near the base of the dyke complex), provides support for models in which crystallization within the AMC followed by crystal subsidence is also an important process in lower crustal accretion. More evolved bulk compositions of gabbros from the upper than lower parts of the plutonic section are due to greater amounts of reaction with interstitial melt and not because their parental melt had become highly fractionated through the formation of large volumes of cumulates deeper in the crust. Amphibole-plagioclase thermometry confirms previous reports that the initial ingress of fluid occurs at high-temperatures in the shallow gabbros (Tave 713°C) and show that the temperature of amphibole formation was similar in deeper gabbros (Tave 722°C). This thermometry also suggests that fracture and grain boundary permeability for seawater-derived fluids was open over the same temperature interval.

1. Introduction

[2] Remote sensing experiments at fast spreading ridges, especially the East Pacific Rise (EPR), have led to significant advances in our understanding of the subsurface structure of ridge axes. In particular, these studies have resolved the size, shape, and along-axis continuity of the axial magma chamber (AMC) [e.g., Detrick et al., 1987]. This body lies near the base of the sheeted dyke complex, is ∼10–100 m thick, ∼300–1500 m wide, and is nearly continuous along-axis. It is underlain by a region of low seismic velocities suggesting that the sub-AMC region is partially molten and that the melt fraction decreases with depth from the AMC downward [e.g. Dunn et al., 2000]. Despite these important studies, there is little consensus about the high-temperature magmatic and hydrothermal processes operating at fast spreading ridges. For example, different models of melting and melt extraction make different predictions for the diversity of melt compositions added to the crust [e.g., O'Hara, 1985] and models of melt fractionation within the crust predict different cumulate compositions [e.g., O'Hara, 1977; Langmuir, 1989; Korenaga and Kelemen, 1998]. Also, models of hydrothermal circulation and heat extraction within the lower crust, which is intimately linked with melt crystallization and differentiation and lower crustal accretion, make different predictions about the initial temperature of fluid penetration into the lower crust, it's importance in heat extraction and the resulting degree of alteration [e.g., Manning et al., 1996; McCollom and Shock, 1998]. Testing these kinds of models cannot be achieved using remote sensing studies alone.

[3] Because of the general inaccessibility of plutonics formed at fast spreading ridges the Oman ophiolite has commonly been used as an analogue for field, petrological, and geochemical studies of lower crustal processes [e.g., Pallister and Hopson, 1981]. However, neither the palaeo-spreading rate of the Oman ophiolite nor the effect of its probable suprasubduction zone setting on mantle and crustal processes are accurately known. Petrological and geochemical studies of samples from the lower crust formed at the EPR are few and far-between and are restricted to regions near transform faults, where lower crustal processes may be atypical [e.g., Constantin, 1999], and to regions where unusual tectonic activity has exposed “normal” lower crust [e.g., Hékinian et al., 1993; Gillis et al., 1993]. Hess Deep is an example of the latter where the westward propagating Cocos-Nazca ridge has dissected crust formed at the EPR exposing lower crustal and upper mantle lithologies. Here we present a study of the igneous and metamorphic petrology and geochemistry of lower crustal samples formed at the EPR and recovered from Hess Deep. We present new mineral major and trace element data, plagioclase crystal size distributions (CSD) and amphibole-plagioclase thermometry. These data are used to investigate magmatic and hydrothermal processes in the lower crust at fast spreading mid-ocean ridges.

1.1. Hess Deep

[4] Hess Deep is the deepest part of a rift valley caused by the propagation of the western end of the Cocos-Nazca spreading center into the eastern side of the Galapagos microplate (Figure 1). Approximately 1 Ma ocean crust which formed at the fast spreading EPR at ∼2°N, where the half spreading rate is ∼65 mm yr−1, is dissected exposing lower crustal and upper mantle lithologies at the seafloor [Lonsdale, 1988]. This area provides probably the best opportunity to study the lower crust formed at a present-day fast spreading ridge and has thus been the focus of several research cruises.

Figure 1.

Map showing the Hess Deep area in the equatorial Pacific and the sampling locations discussed in this study. Key morphological features are the Hess Deep at 5400 mbsl, the slow-spreading Cocos-Nazca ridge at 4000–4500 mbsf, the intrarift ridge (IrR) that rises to 3000 mbsl north of Hess Deep that ODP Site 894 lies on top of, and the steep bounding scarps north of the intrarift ridge and south of Hess Deep. The samples studied here were recovered from the southern slope of the intrarift ridge along the tracks of Nautile dives 1, 9, and 18 (labeled within the lower left yellow box), at the crest of the intrarift ridge at ODP Site 894, and along Alvin dive tracks in an area along the northern rift valley scarp (yellow box in upper right).

[5] The lower crustal samples discussed here were recovered at Hess Deep during four cruises (Figure 1). The NAZCOPAC cruise explored the intrarift ridge and rift valley scarps with the Nautile submersible [Francheteau et al., 1990; Hékinian et al., 1993]. Two cruises with the Alvin submersible have investigated the upper crust (lavas and dykes) and uppermost gabbros exposed along the northern rift valley scarp; several dredges were also recovered from intrarift ridge [P. Lonsdale, unpublished data, 1990; Karson et al., 1992, 2002]. Ocean Drilling Program (ODP) Leg 147 drilled gabbroic rocks at the crest of the intrarift ridge (Site 894) and peridotites intruded by numerous gabbroic veins, ∼10 km southeast of Site 894 (Site 895) [Gillis et al., 1993]. New data are presented here for samples recovered from >4950 m below sea level (mbsl) on Nautile dives 1, 9, and 18 which explored the southern slope of the intrarift ridge [Francheteau et al., 1990; Hékinian et al., 1993], ODP Hole 894G and the northern rift valley scarp.

[6] The geological context and pseudostratigraphical position of samples from the northern rift valley scarp are relatively well constrained with all of the gabbro samples studied here coming from within <500 m of the base of the sheeted dyke complex [P. Lonsdale, unpublished data, 1990; Karson et al., 1992, 2002]. It has been suggested that ODP Site 894 recovered rocks from very near the top of the gabbroic layer [Natland and Dick, 1996; Pedersen et al., 1996], a conclusion we support on the basis of cooling rate estimates (L. Coogan, unpublished data, 2001) using the method of Coogan et al. [2002a]. In an attempt to study samples from the deeper parts of the lower crust we have restricted our study of samples from the southern slope of the intrarift ridge to those recovered at >4950 mbsl. Although some of these may have been transported downslope, they are likely to include samples representative of deeper levels in the crust than those recovered from the northern rift valley scarp and ODP Hole 894G.

1.2. Analytical Methods

[7] Major and minor element compositions were determined on a JEOL 8600 Superprobe electron microprobe at the University of Leicester using a 15 kV accelerating voltage and 30 nA beam current. A 10 μm beam was used for all analyses ,and the raw data were corrected using a ZAF correction procedure. Trace element compositions of clinopyroxene were determined by ion microprobe at Edinburgh University and laser ablation ICP-MS (LA-ICP-MS) at the University of Hong Kong, using NIST 610 [Pearce et al., 1997] and NIST 612 [Norman et al., 1996], to standardize, respectively. These analytical techniques are described by Coogan et al. [2000a, 2000b] and Coogan et al. [2002b], respectively. Only the cores of clinopyroxene crystal were analyzed by LA-ICP-MS (Table 1) and most ion microprobe analyses are also of crystal cores (Table 2).

Table 1. Representative Compositions of Clinopyroxene Crystal Cores From ODP Hole 894G (LAM-ICP-MS)
ppm (LAM-ICP-MS)
ppm (LAM-ICP-MS)
Table 2. Representative Clinopyroxene Compositions in Samples From the Southern Slope of the Intrarift Ridge
SampleNZ 1-4NZ 1-4NZ 1-6NZ 1-6NZ 1-6NZ 9-3NZ 9-3NZ 9-5NZ 9-6NZ 9-6NZ 9-6NZ 9-12NZ 9-13NZ 9-13NZ 9-13NZ 9-17NZ 9-17NZ 18-3NZ 18-3NZ 18-4
ppm (ion microprobe)

2. Igneous Petrology

[8] Plutonic rocks recovered along the southern slope of the intrarift ridge (Figure 1) by submersible span a wide range of lithologies (gabbronorites, olivine gabbronorites, gabbros, olivine-gabbros, rare troctolites and Cr-spinel-bearing troctolites, anorthosites, and plagioclase-bearing dunites) [Hékinian et al., 1993]. The more primitive lithologies (troctolites, anorthosites and Cr-spinel bearing samples) are restricted to sampling depths >5300 mbsl [Hékinian et al., 1993], whereas olivine-gabbros, gabbros, and gabbronorites occur throughout the entire depth range sampled. It is not clear whether any of these samples are mineralogically layered in the sense used in continental intrusions. Drilling at the crest of the intrarift ridge at ODP Site 894 (∼3000 mbsl; Figure 1) recovered olivine gabbro, gabbro, olivine gabbronorite, gabbronorite, and patches of pegmatitic amphibole-gabbro [Gillis et al., 1993; Natland and Dick, 1996; Pedersen et al., 1996]. Similar lithologies have been recovered from the northern rift valley scarp [Gillis, 1995; Hanna et al., 1999]. These upper gabbros apparently lack modal layering but contain many contacts between lithologies of contrasting grain size [Gillis et al., 1993; Natland and Dick, 1996] as is common in the uppermost part of many ophiolite plutonic sequences [e.g. Pedersen, 1986]. At ODP Site 895, a Moho transition zone assemblage of harzburgite, dunite, troctolite, and gabbro was recovered [Gillis et al., 1993; Arai and Matsukage, 1996; Boudier et al., 1996a; Dick and Natland, 1996; Edwards and Malpas, 1996]. In general the dunite forms a carapace between the harzburgite and the gabbroic lithologies. A striking feature of the gabbroic rocks recovered from Hess Deep is the abundance of orthopyroxene; 25% of all plutonic samples described by Hékinian et al. [1993] are gabbronorite or olivine gabbronorite, 85% of Hole 894G is gabbronorite or olivine gabbronorite [Gillis et al., 1993], and gabbronorite is the dominant lithology recovered from the northern rift valley scarp [P. Lonsdale, unpublished data, 1990; Gillis, 1995].

2.1. Whole Rock Geochemistry

[9] The whole rock compositions of plutonic rocks provide information about the process and efficiency of crystal-melt separation during cumulate formation and melt fractionation. Figure 2a compares the whole rock compositions of gabbros from ODP Sites 894 [Pedersen et al., 1996] and 895 [Gillis et al., 1993], the southern slope of the intrarift ridge [Blum, 1991; Hékinian et al., 1993] and the northern rift valley scarp [Natland and Dick, 1996] with basalts from the East Pacific Rise [Lonsdale et al., 1992; Regelous et al., 1999] and gabbros from the Oman ophiolite [MacLeod and Yaouancq, 2000]. Compositional difference between the basalts and gabbros indicates that these gabbros are, in general, cumulates not frozen liquids. Comparison of the compositional range in the gabbros and basalts with model trends shows that fractional crystallisation better approximates the crystal-melt separation process than equilibrium crystallisation (Figure 2a). In detail, however, many of the Hess Deep plutonics have higher Ti for their Mg# relative to the model cumulates. The simplest explanation of this is that their bulk compositions have been modified by reaction of cumulate crystals with interstitial melt during their solidification (a “trapped liquid shift”) [see, also, Pedersen et al., 1996].

Figure 2.

Comparison of the whole rock geochemistry of gabbros from Hess Deep with MORB from the EPR and gabbros from the Oman ophiolite; (a) Mg# (calculated as 100Mg/[Mg + Fetot]) versus Ti. The modeling trends show the compositions of melts and cumulates derived by 80% equilibrium and fractional crystallization of a suitable parental melt modeled using the MELTS algorithm [Ghiorso and Sack, 1995]. Fractional crystallization better approximates the melt and cumulate compositions than equilibrium crystallization (cumulates from equilibrium crystallization show a similar trend to those from fractional crystallization but with much less change in Mg# or Ti during crystallization). Melt fractionation trends are hatched at 10% crystallization intervals. The scatter of cumulate compositions toward higher Ti abundances for a given Mg# than is predicted by the fractional crystallization trend can be explained by the plutonics being a mixture of cumulus crystals plus some interstitial melt (a trapped liquid shift); (b) N-MORB normalized trace element compositions of average gabbro compositions from Hess Deep and the Oman ophiolite (n = number of analyses). Note the depletion in incompatible elements in the gabbros with respect to N-MORB, which is greater for the light than heavy REE. N-MORB values from Hofmann [1988]. Data sources are given in the text. All data from the Oman ophiolite are from the Wadi Abyad section in the south-central part of the ophiolite following the lithological division of MacLeod and Yaouancq [2000].

[10] The average trace element compositions of samples from ODP Hole 894G [Pedersen et al., 1996] and from >4950 mbsl on the southern slope of the intrarift ridge [Blum, 1991] are shown in Figure 2b. Large positive Sr anomalies and smaller positive Eu anomalies indicate that on average both sample suites are “cumulates” (i.e., have accumulated at least plagioclase) and are not frozen liquids, consistent with the interpretation based on major elements (Figure 2a). The lack of a positive Ti anomaly in either the Hole 894G or southern slope gabbros indicates that their parental melts were not oxide saturated and thus were not highly fractionated. This supports a model for the high incompatible element abundances in the Hole 894G coming from “trapped melt” [cf. Pedersen et al., 1996] rather than being due to crystallization from a highly fractionated parental melt.

[11] Gabbroic rocks from the southern slope of the intrarift ridge have whole rock major and trace element compositions similar to those of lower gabbros from the Oman ophiolite (Figure 2b). Gabbros from ODP Hole 894G and the northern rift valley scarp are much more similar to the upper gabbros in the Oman ophiolite. The main difference between the Hess Deep and Oman gabbros is the positive Ti anomalies in the upper gabbros from the Oman ophiolite which is absent in gabbros from ODP Hole 894G. This reflects oxide accumulation in the Oman upper gabbros but not the Hess Deep upper gabbros.

2.2. Mineral Chemistry

[12] Plagioclase from ODP Hole 894G shows extensive normal, reverse, oscillatory, and patchy zoning (Figure 3a) with some crystals having very anorthitic cores (up to An90; An = 100Ca/[Ca + Na + K]) [Kelley and Malpas, 1996; Natland and Dick, 1996; Pedersen et al., 1996]. The upper gabbros from the northern rift valley scarp also display a similarly large range of anorthite contents and extensive zoning [Gillis, 1995; Hanna et al., 1999]. Samples from the southern slope of the intrarift ridge show a broad range of An87–48; however, higher An contents are more common, and zoning is much more limited, than in samples from ODP Hole 894G and the northern rift valley scarp [Hékinian et al., 1993]. Mafic phases are more evolved in gabbros from ODP Hole 894G (olivine Fo61–71, orthopyroxene Mg#64–74, and clinopyroxene Mg#62–81) [Natland and Dick, 1996; Kelley and Malpas, 1996] and from the northern rift valley scarp (orthopyroxene Mg#56–68, and clinopyroxene Mg#30–80) [Gillis, 1995; Hanna et al., 1999] than in samples from the southern slope of the intrarift ridge (olivine Fo72–88, orthopyroxene Mg#67–88, and clinopyroxene Mg#65–89; [Hékinian et al., 1993]). A striking characteristic of one sample from the southern slope of the intrarift ridge is the occurrence of primitive (Mg#88; Cr2O3 ∼ 0.2 wt %) cumulus orthopyroxene (Table 3; Figure 3b). In comparison with orthopyroxene in troctolite veins within peridotite drilled at ODP Site 895 [Arai and Matsukage, 1996] this orthopyroxene has lower Ti, Al, Cr, and Ca. The occurrence of primitive cumulus orthopyroxene in these plutonics suggests that some primitive melts added to the crust are orthopyroxene saturated even though primitive MORB is not [O'Hara, 1968; Stolper, 1980].

Figure 3.

Photomicrographs showing petrographic characteristics of the Hess Deep plutonics. (a) False color SEM image showing irregularly shaped calcic plagioclase cores (light green) overgrown by more sodic rims (dark green) in sample 147 894G 9R4 11cm; (b) primitive orthopyroxene crystal core (Mg # ∼ 88), altered around its edges, as described in the text (sample NZ 9-6); (c) aligned plagioclase crystals making a crystal shape-preferred fabric parallel to the yellow arrow (sample NZ 9-3); (d) amphibole replacing clinopyroxene in sample 147 894G 9R4 11 cm; (e) macroscopic amphibole vein (running from bottom left to top right) crosscutting igneous minerals in sample NZ 9-3; (f) amphibole filling a microvein along a plagioclase-plagioclase grain boundary (at tip of arrow) in Sample 2212–1516. Note the amphibole replacing the clinopyroxene and small crosscutting vein also. Key is as follows: p, plagioclase; c, clinopyroxene; o, orthopyroxene; a, amphibole.

Table 3. Representative Compositions of Primitive Orthopyroxene in Sample NZ 9-6

[13] The covariation of clinopyroxene and olivine major and compatible minor element abundances is shown in Figure 4. Partial crystallization processes (e.g., fractional crystallization) lead to a decrease in both the Mg# and compatible element abundances of the melt and crystals with rapid decreases in highly compatible minor elements as shown by the field for cumulates from the Mid-Atlantic Ridge and Oman ophiolite in Figure 4. Samples from the southern slope of the intrarift ridge are similar to most oceanic gabbros (Figure 4). In contrast to this, olivine and clinopyroxene from ODP Hole 894G and in some samples from the northern scarp have much higher Cr and Ni abundances for their Mg# than is expected during partial crystallization. Both this and the zoned plagioclase can be explained by the crystallization of primitive crystals from a primitive melt followed by overgrowth from an evolving melt in a crystal mush (most likely a “trapped” interstitial melt). In this scenario, slow-coupled diffusion of solid solution components in plagioclase led to the development of extensive zoning. In contrast, much more rapid Mg/Fe diffusion in the mafic phases allowed extensive re-equilibration resulting in a large decrease in the crystal Mg# and a relatively smaller decrease in their compatible minor element abundances. Thus it appears that the first crystals to form probably had similar compositions throughout the gabbroic sequence (quite primitive) but that reaction with interstitial melt lowered the Mg# of mafic phases in the upper plutonics and led to extensive zoning in plagioclase. This is consistent with the whole rock compositions that suggest that many upper crustal samples are mixtures of cumulus crystals and interstitial melt (Figure 2a).

Figure 4.

(a) Clinopyroxene Mg#-Cr2O3 covariation in Hess Deep gabbros compared with oceanic gabbros from the MAR (Mid-Atlantic Ridge) [Coogan, 1998] and the Oman ophiolite [Browning, 1982]. Samples from the southern slope of the intrarift ridge [Hékinian et al., 1993] fall along the same trend as those from the Mid-Atlantic Ridge and Oman ophiolite, whereas, clinopyroxene from ODP Hole 894G, and some from the northern rift valley scarp have much lower Mg#s for their Cr contents than gabbros from other spreading ridges; (b) olivine Fo-NiO (Fo = forsterite = 100 Mg/[Mg + Fetot]) correlations show a similar pattern to clinopyroxene Mg#-Cr2O3. The relatively high compatible element (Cr and Ni) abundances for a given Mg# in the ODP Hole 894G gabbros can most simply be explained by reaction of interstitial melt and primitive crystals as shown schematically by the red arrows. All data points are sample averages.

[14] Clinopyroxene rare earth element (REE) abundances in gabbroic rocks from ODP Site 894 and the southern slope of the intrarift ridge are shown in Figure 5a5c along with a comparison of the compositions of melts calculated to be in equilibrium with them and MORB from the equatorial EPR (Figures 5d5f). Clinopyroxene in the majority of the samples recovered from the southern slope of the intrarift ridge are in equilibrium with melts similar to, or with slightly lower REE abundances than, erupted equatorial EPR basalts. Some gabbroic samples from ODP Site 895 (C-type of Dick and Natland [1996]), and a few samples from ODP Hole 894G [Gillis, 1996], also contain clinopyroxene in equilibrium with equatorial EPR MORB. However, the REE abundances and LREE/HREE (light REE to heavy REE ratio) of most clinopyroxene from ODP Hole 894G are too high to be in equilibrium with most MORB from the equatorial EPR.

Figure 5.

Chondrite-normalized REE spidergrams for clinopyroxene from (a–c) the different sample suites studied and (d–f) calculated melts in equilibrium with them calculated using the distribution coefficients (D values) of Hart and Dunn [1993]. Clinopyroxene in samples from the southern slope of the intrarift ridge generally have significantly lower REE abundances, and lower LREE/HREE, than clinopyroxene from ODP Site 894. Most of the samples recovered from the southern slope of the intrarift ridge are close to being in equilibrium with MORB from the EPR (grey field in Figures 5d–5f from Lonsdale et al. [1992]) but most of those from Hole 894G have higher REE abundances and higher LREE/HREE than normal EPR MORB. Data from Gillis [1996] and the ion probe data for samples from the southern slope of the intrarift ridge from this study (Table 2) include both crystal cores (solid lines) and rims (dashed lines) whereas the LA-ICP-MS data for samples from ODP Hole 894G (Table 1) are for crystal cores only.

[15] A striking feature of the clinopyroxene compositions is an increase in LREE/HREE ratios with increased REE abundance. This is observed in data from each sampling area (ODP Sites 894 and 895 and the southern slope of the intrarift ridge) and in the combined data (Figure 6a) showing that it is not an analytical artifact. A similar correlation is also observed in whole rock compositions indicating that grain scale partitioning between clinopyroxene and plagioclase does not control the clinopyroxene REE ratios. This fractionation is shown by the co-variation of Ce(n) and Yb(n) in all clinopyroxene from Hess Deep (Figure 6a). There is a ∼40-fold variation in Ce and only ∼10-fold variation in Yb with a general increase in Ce(n)/Yb(n) with increasing REE abundances. The Ce(n)/Yb(n) ratios of clinopyroxene (0.12–0.57; greater than four-fold variation) are more variable than those of MORB from the equatorial EPR (0.67–1.17; less than two-fold variation) [Lonsdale et al., 1992]. Neodymium isotope data for the Site 894 gabbros show that these precipitated from depleted MORB (the average 143Nd/144Nd is 0.51319) suggesting that an enriched parental melt is an unlikely origin of the high Ce(n)/Yb(n) in the Site 894 gabbros [Pedersen et al., 1996].

Figure 6.

Rare earth element systematics in clinopyroxene. (a) Ce(n) versus Yb(n) for all samples from Hess Deep showing a general increase in clinopyroxene Ce(n)/Yb(n) with increasing REE abundances which can be seen in data from each of the sampling areas individually and in the data as a whole. Lower crustal gabbros from Oman [Kelemen et al., 1997a, 1997b] have similar clinopyroxene REE abundances to the most depleted samples from the southern slope of the intrarift ridge; (b) modeling of the composition of clinopyroxene in equilibrium with melts evolving through different melting and crystallization processes. Fractional crystallization cannot explain the change in Ce(n)/Yb(n) in clinopyroxene and equilibrium crystallization can only explain the variation at very low melt fractions remaining (and cannot have operated in “end-member” form as the cumulates would all be identical in composition in this scenario). Models of boundary layer crystallization [Langmuir, 1989] or reaction of cumulate crystals and interstitial melt both fit the data well. Crystallization of melts formed by different extents of melting could also explain the variations observed. The melting model assumes an initial mantle source with Ce(n) = 0.9 and Yb(n) = 0.4 so as to produce a 20% ponded fractional melt with a starting composition in equilibrium with the most depleted clinopyroxenes. The bulk distribution coefficients used in the melting models are DCe = 0.02 and DYb = 0.07 (calculated assuming 8% clinopyroxene; 30% orthopyroxene, 61% olivine and 1% porosity; D values from Hart and Dunn, [1993] and Schwandt and McKay, [1998] and assuming DREEolivine = 0). The starting melt composition for crystallization modeling is taken as the aggregated 20% fractional melt (Ce(n) = 7.5 and Yb(n) = 11.8). The bulk distribution coefficients used in the crystallization models were DCe = 0.05 and DYb = 0.15 (calculated assuming 35% clinopyroxene and 65% plagioclase crystallization; clinopyroxene D values from Hart and Dunn, [1993] and plagioclase (DCe = 0.035 and DYb = 0) calculated from the clinopyroxene D values assuming the relative partitioning between plagioclase and clinopyroxene observed in gabbros from the Mid-Atlantic Ridge [Coogan, 1998]). The difference between the bulk Ce and Yb distribution coefficients used in the modeling are at the high end of the range possible to maximize the fractionation of these elements during modeling of the various processes. The boundary layer crystallization model assumes 15% porosity in the solidification zone at the time of melt extraction and that all of this melt is removed and is just one of a multitude of possible fractionation trends. The trapped melt reequilibration shows the effect of full equilibration of trapped (parental) melt on the most primitive cumulates. All clinopyroxene in equilibrium with melts are calculated using the distribution coefficients of Hart and Dunn, [1993].

[16] The observed increase in Ce(n)/Yb(n) with increasing REE abundances could potentially be explained either by partial crystallization processes or crystallization of different parental melts. Unfortunately, Figure 6b shows that the REEs provide a rather poor data source to use to test between different models for the origin of the variations in Ce(n)/Yb(n). Distinguishing the relative importance of crystallization of incompletely aggregated melts, crystallization of melts that have evolved via boundary layer fractionation, and reaction with interstitial melt will require more data (e.g., trace element zoning profiles in clinopyroxene and full mass balance of mineral and whole rock trace element abundances). The limited trace element zoning profiles in clinopyroxene presented by Natland and Dick [1996] do, however, suggest that reaction with trapped melt mainly changed the composition of crystal rims producing highly zoned crystals. Thus the compositional diversity of crystal cores is unlikely to be dominated by the effects of reaction with interstitial melt.

2.3. Textural Data

[17] The textures of plutonic rocks record important information about their petrogenesis. Herein, we emphasize two characteristics of plutonics from Hess Deep, their fabrics and their plagioclase crystal size distributions (CSDs). More general descriptions of magmatic textures are available elsewhere [Gillis et al., 1993; Hékinian et al., 1993].

2.3.1. Fabrics

[18] Fabrics in plutonic rocks record information regarding both syn- and post-crystallization deformation. Unlike gabbros from slow spreading ridges, in which evidence for crystal plastic deformation at near-solidus temperatures is common, there is very little evidence for crystal plastic deformation in any samples from Hess Deep. Crystal plastic deformation is limited to slight undulose extinction and very scarce deformation twins in plagioclase that is observed in a few samples from all sampling localities. This suggests very minor subsolidus strain. In contrast to crystal plastic fabrics, magmatic fabrics are common in ODP Hole 894G [Gillis et al., 1993; Yaouancq, 1994; MacLeod et al., 1996] and also occur in some samples from the southern slope of the intrarift ridge [Hékinian et al., 1993] and the northern rift valley scarp. These fabrics are defined by the preferred orientation of plagioclase crystals (Figure 3c).

[19] Detailed studies of the gabbro fabrics have only been carried out in samples from ODP Hole 894G and have shown that these are both foliated and lineated. The fabrics are generally better defined in finer grained samples although they occur throughout much of the drill core [Gillis et al., 1993; Yaouancq, 1994; MacLeod et al., 1996; Richter et al., 1996]. Reorientation of some sections of core to geographical coordinates shows that the foliations are steeply dipping (mean dip of 69°) with a nearly north-south orientation, parallel to the EPR, and that there is a steeply plunging lineation [MacLeod et al., 1996]. Measurements of the anisotropy of magnetic susceptibility in samples from ODP Hole 894G show that there is also a magnetic fabric parallel to the plagioclase fabric [Richter et al., 1996]. The fabrics are typical of those formed during magmatic flow by the physical alignment of crystals [Benn and Allard, 1989] suggesting that magmatic flow within the shallow gabbros beneath the EPR is close to axis parallel and near-vertical.

2.3.2. Plagioclase Crystal Size Distributions

[20] The sizes of crystals and their size distribution are a fundamental property of a rock texture providing information regarding the initial accumulation of crystals and the solidification of the crystal mush. Crystal size data can be represented in a number of different ways (e.g., cumulative frequency plots, histograms); in this study we follow the common method of displaying plagioclase crystal sizes on plots of ln(n) versus L, where n is the population density (number of crystals in a given size range per unit volume divided by the size range) and L is the midpoint of the size range (i.e., as CSDs) [Marsh, 1988]. Plagioclase CSDs have been determined for 13 samples from ODP Hole 894G and 5 samples from the southern slope of the intrarift ridge. Plagioclase sizes were measured on photomicrographs at an average magnification of 40 times, and an average of 200 crystals were measured per sample. Two-dimensional measurements of crystal lengths and widths were converted into three-dimensional shapes following the approach of Higgins [1994]. Calculated median crystal aspect ratios [Higgins, 1994] for the Hole 894G plagioclase are 1:1.8:3 and for the deeper samples are 1:1.4:2.5 possibly suggesting slightly more equant crystals occur at greater depths in the crust. Plagioclase CSDs were calculated using the software CSDCorrections version 1.1 of Higgins [2000] assuming an angularity of 0.8, a massive fabric and using the measured crystal widths; the data were binned so as to have five divisions per decade.

[21] Plagioclase CSDs show a ln-linear negative correlation between size and population density at larger crystal sizes, followed at smaller sizes by a depletion in the finest crystal sizes (less than ∼0.2–1.2 mm) as is commonly observed in CSDs (Figure 7). The shape of the CSD can be used to interpret the conditions of plagioclase crystal accumulation (Figure 7a). Physical crystal accumulation, in which coarse crystals are preferentially accumulated with respect to fine ones, will lead to an overenrichment in the coarser crystals with respect to that predicted by extrapolating a straight line through the smaller sizes on a CSD plot (Figure 7a) [Marsh, 1988]. Only one sample shows clear evidence for this (ODP 147 894G 20R3 91 cm; Figure 7n). Calculating the CSD for this sample using crystal lengths instead of widths eradicates this kink suggesting that this may be a measuring artifact (the two largest bins contain a total of only four crystals). This suggests that coarse crystal accumulation, such as might be expected during crystal settling, was not an important process in controlling the resulting CSD. Natland and Dick [1996] suggest that some plagioclase crystals in ODP Hole 894G are the remains of previously larger crystals which broke during cumulate formation. This process should lead to a depletion in coarse crystals in the CSD (Figure 7a); there is no evidence for this in our data.

Figure 7.

Plagioclase CSDs; (a) expected changes in the form of an initially ln-linear CSDs for different physical processes during crystal accumulation [from Marsh, 1988]; (b-s) data from Hess Deep gabbros. The data used to define the ln-linear portion of the plot are shown as filled symbols and other data as open symbols. Sample numbers: from ODP Hole 894G and preceded by ODP 147 894G with the final number being the piece depth in cm (Figures 7b–7n); samples from the southern slope of the intrarift ridge from the NAZCOPAC cruise (Figures 7o–7s) [Hékinian et al., 1993]. See text for details.

[22] To further analyze the CSD data, the slopes and intercepts of the ln-linear portion of the CSD have been calculated along with the average size of the four largest crystals measured (Lmax) [Marsh, 1998] and the length of crystals at the transition from the ln-linear to fine crystal depleted portions of the CSD for each sample (Figure 8). These four parameters correlate such that the coarsest grained samples (highest Lmax) tend to have CSDs with the low intercepts, shallow slopes and have their transition between the ln-linear and fine crystal depleted portions of the CSD at large crystal sizes. There are no systematic differences in any of these parameters between samples from ODP Hole 894G and those from the southern slope of the intrarift ridge. This contrasts with the findings of Garrido et al. [2001] who show that Lmax, slope, and intercept are generally greater in the samples from deeper in the plutonics in the Oman ophiolite than those from shallower levels.

Figure 8.

Plot of CSD intercept versus (a) slope, (b) the average size of the four largest crystals measured (Lmax), and (c) the length of crystals at the transition from the ln-linear to fine crystal depleted portions of the CSD. This is calculated as the midpoint between the last data point of increasing population density with decreasing crystal size and the next smallest data point. No correlation between either mineral or whole rock chemistry and CSD parameters has been found. See text for discussion.

[23] Interpretation of CSDs is complex, especially in plutonic rocks in which grain impingement during growth will have affected the sizes of the resulting crystals and crystal dissolution as well as growth may have occurred during textural maturation (e.g., Ostwald ripening). An important feature of Figure 8 is the correlation between the crystal size at the transition from the ln-linear to fine crystal depleted and the slope and intercept. This depletion in fine crystals occurs at crystal sizes far too large to be an artifact of measurement imprecision at small sizes. It may either reflect inhibited nucleation during the final stages of solidification or destruction of the finer crystals during cumulate maturation by either dissolution or annexation of smaller crystals by larger ones [Marsh, 1998]. Irrespective of which of these processes is responsible for these correlations, it seems likely that samples with low intercepts, gentle slopes, and a large size range of fine depleted crystals have undergone more extensive “cumulate maturation.” Two petrographic features suggest that grain boundary migration and/or surface energy minimization were important in the development of these rocks. First plagioclase-plagioclase grain boundaries are commonly smoothly curved rather than interpenetrating and angular, which suggests that grain boundary migration was an important process [Hunter, 1996]. Second it is common to see twinning that appears to be continuous across a plagioclase-plagioclase grain boundary and in some cases this makes grain boundary identification difficult. This is consistent with crystal annexation having been an important process during cumulate solidification. In combination with the CSD data, these observations suggest that some form of grain coalescence during cumulate formation probably occurred. A more complete discussion of CSDs in oceanic gabbros is given by Garrido et al. [2001].

3. Metamorphic Petrology

[24] Metamorphic assemblages formed by reaction of oceanic plutonics with hydrothermal fluids record the integrated history of fluid-rock interactions that lead to extraction of heat from the lower crust and mass transfer between the crust and hydrosphere. The Hess Deep plutonics have experienced two stages of hydrothermal metamorphism. The first relates to near-axis, high temperature, fluid flow and is probably representative of normal fast spreading mid-ocean ridges [Gillis, 1995; Manning and MacLeod, 1996], and it is this that we concentrate on in the following sections. The second was due to unroofing and is specific to the tectonic setting of Hess Deep. These episodes of alteration have resulted in heterogeneous alteration associated with brittle deformation at amphibolite to zeolite facies conditions.

[25] High temperature alteration can be divided into two types irrespective of the sampling location; groundmass alteration associated with relatively pervasive fluid flow, and alteration associated with more focused fluid flow in fractures. High temperature alteration is more abundant in samples from ODP Site 894 than in samples from either the northern rift valley scarp [Gillis, 1995] or from the southern slope of the intrarift ridge. Samples from the intrarift ridge, where many exposures are fault surfaces related to Cocos-Nazca rifting, locally display overprinting by greenschist to zeolite facies assemblages in net-veined and cataclastically deformed samples. This suggests that this lower temperature fluid flow was related to unroofing [Gillis et al., 1993; Hékinian et al., 1993]. Low temperature alteration is most pervasive (50–100%) within haloes adjacent to chlorite- and zeolite-bearing veins (see section 3.2) and within porphyroclasts and breccia fragments [Hékinian et al., 1993; Früh-Green et al., 1996; Manning and MacLeod, 1996]. In these regions, plagioclase is replaced by assemblages of albitic plagioclase, prehnite, epidote, clays, and zeolites. The mafic phases are pseudomorphed by assemblages of actinolite, chlorite, clay minerals, and iddingsite.

3.1. Groundmass Alteration Assemblages

[26] Groundmass alteration is lowest in nonpegmatitic samples that display only minimal fracturing [Hékinian et al., 1993; Gillis, 1995; Früh-Green et al., 1996; Lécuyer and Gruau, 1996; Manning and MacLeod, 1996]. In these samples, plagioclase generally retains its igneous cation compositions (An content); it is commonly depleted in δ18O with respect to magmatic values in ODP Hole 894G [Früh-Green et al., 1996; Lécuyer and Gruau, 1996; Lécuyer and Reynard, 1996] but not in samples from the southern slope of the intrarift ridge [Agrinier et al., 1995]. The mafic phases are variably (<5–100%) altered. Clinopyroxene and orthopyroxene are replaced by calcic amphibole, minor secondary clinopyroxene and magnetite (±chlorite), and fibrous actinolite with minor talc, respectively. Olivine alteration is characterized by coronas composed of amphibole, talc, chlorite, magnetite, and pyrite. These samples can also show small amounts of low temperature, oxidative, alteration with calcite, and iddingsite replacing olivine.

[27] Centimeter-to-decimeter size pegmatitic patches are more pervasively altered than adjacent, nonpegmatitic lithologies probably due to their coarser grain size leading to more intracrystal cracking during cooling and possibly due to them containing primary hydrous phases [Gillis, 1996; Kelley and Malpas, 1996]. Pegmatites occur along the northern rift valley scarp and in ODP Hole 894G where they comprise <5% of the core [Gillis et al., 1993] but have not been observed in samples from the southern slope of the intrarift ridge. Pegmatitic patches are composed of granular amphibole (multiple generations), plagioclase, Fe-Ti oxides, quartz, zircon, and apatite [Gillis, 1996; Kelley and Malpas, 1996]. Plagioclase is more sodic than in the finer grained lithologies and is commonly altered to assemblages of albitic plagioclase, calcic amphibole, and epidote. Brown to green calcic amphibole rims clinopyroxene and forms discrete interstitial grains. Major and trace element data for pegmatitic amphibole from ODP Site 894 suggest a magmatic origin with subsequent interaction with hydrothermal fluids [Gillis, 1996; Gillis and Meyer, 2001]. Hydrogen isotope ratios for amphibole separates from samples recovered near the crest of the intrarift ridge [Agrinier et al., 1995; Früh-Green et al., 1996], and fluid inclusion data for the Site 894 gabbros [Kelley and Malpas, 1996], suggest the presence of magmatic fluids in these patches.

3.2. Vein Alteration Assemblages

[28] A systematic sequence of fracture infilling has been documented for the core from ODP Site 894. The earliest veins are microscopic (<40 μm), occurring along grain boundaries and within igneous phases and are filled with calcic amphibole [Manning and MacLeod, 1996]. In these the amphibole major element compositions vary dependent on the phase the vein is cutting [Manning and MacLeod, 1996]. Adjacent to amphibole veins, plagioclase compositions are commonly shifted toward lower anorthite contents (generally ≤10% lower) [Manning et al., 1996]. These are cut by progressively later macroscopic veins composed of amphibole, chlorite-calc silicate, chlorite-smectite, and zeolite-calcite [Früh-Green et al., 1996; Manning and MacLeod, 1996]. Integration of vein orientations and mineralogies, with borehole (formation microscanner) and magnetic data, reveals that the chlorite- and zeolite-bearing veins have a strong east-west preferred orientation indicating that they record the unroofing of the gabbroic rocks in response to Cocos-Nazca rifting [Manning and MacLeod, 1996]. Thermal arguments suggest that the earlier amphibole veins probably developed at, or near, the EPR axis [Manning and MacLeod, 1996]. Similar vein assemblages are prevalent within samples from the southern slope of the intrarift ridge [Hékinian et al., 1993; Agrinier et al., 1995]. Gabbros from the northern rift valley scarp display only minimal effects of Cocos-Nazca rifting in that they are cut by early high temperature amphibole vein networks with few chlorite-smectite veins [Gillis, 1995].

3.3. Amphibole-Plagioclase Thermometry of High Temperature Fluid-Rock Reaction

[29] High temperature (>700°C) penetration of seawater-derived hydrothermal fluids into the gabbros recovered at ODP Site 894 and from the northern rift valley scarp has previously been documented for microscopic and macroscopic veins using plagioclase -amphibole thermometry [Manning et al., 1996, 2000; Weston, 1998]. Here we report the results of new thermometric calculations for hydrothermal amphibole in samples from the southern slope of the intrarift ridge and additional data for the northern rift valley scarp (Figure 9; Table 4 and Table 5). Hydrothermal amphibole was identified on the basis of texture (microscopic and macroscopic vein fill, rims on clinopyroxene and olivine; Figures 3d3f) and major element composition such that grains with TiO2 contents ≥1 wt % are likely magmatic and hence were excluded [Gillis and Meyer, 2001].

Figure 9.

Histograms showing the variations in equilibration temperatures recorded by amphibole-plagioclase thermometry. Temperatures are binned in 25°C intervals. (a) Equilibration temperature divided by sampling location. Note the similar temperatures in all areas; (b) equilibration temperature divided by amphibole texture.

Table 4. Summary of Amphibole-Plagioclase Thermometry
LocationTextureTemp Range, °CMeanStandard DeviationNumber of analysesNumber of samplesData source
N. scarpMafic replacement651–8147454683This study
Microscopic vein652–77571438447Manning et al. [2000], Weston [1998], this study
Macroscopic vein725–7877482561This study
Site 894Mafic replacement624–7426975055Gillis and Meyer [2001]
Microscopic vein618–80171533979Manning et al. [1996], 2000; Weston [1998]
Macroscopic vein614–74970025553Weston [1998]
Pegmatite546–74363569283Weston [1998]
S. slopeMafic replacement610–78971949275this study
Microscopic vein656–8007354694this study
Macroscopic vein648–78371740113this study
Table 5. Representative Amphibole Compositions Used for Thermometry
SampleNZ 9-12NZ 9-12NZ 9-12NZ 9-12NZ 9-12NZ 9-3NZ 9-3NZ 9-3NZ 9-3NZ 9-17NZ 9-17NZ 9-17NZ 9-17NZ 9-17NZ 1-6NZ 18-6NZ 18-6NZ 18-4NZ 18-4NZ 18-4
texturecpx repmicromacromacrocpx repcpx repmicromacromacrooliv repoliv repmacromacromacromicrocpx repcpx repcpx repcpx repcpx rep
Cations on Basis of 23 Anhydrous Oxygens
Plag Ab0.50.540.520.410.300.390.390.380.340.340.350.370.360.170.410.410.460.730.610.61
Temp (°C)707656648689704783729720729740784783762754718701782611697714

[30] Similar to previous work, we used the edenite + albite = richterite + anorthite exchange geothermometer developed by Holland and Blundy [1994] and assumed a pressure of 1 kbar for all calculations. This geothermometer is calibrated in the temperature range 500°–900°C with an uncertainty of ±39°C. Shifts in plagioclase compositions adjacent to amphiboles and rapid hydrothermal reaction rates at amphibolite facies conditions [Wood and Walther, 1983] suggest that equilibrium was achieved on a local scale [see also Manning et al., 1996, 2000; Weston, 1998].

[31] Calculated temperatures for different amphibole textures from each sample suite are summarized in Figure 9 and Table 4. The principle finding of these calculations is that there are no systematic differences in mean temperature between sample suites or amphibole textures (Figure 9). This suggests that seawater ingress occurred over approximately the same temperature interval throughout the gabbroic sequence. Microscopic and macroscopic amphibole veins show similar mean temperatures in all of the sample suites (Table 4). Amphibole replacing the igneous mafic phases (clinopyroxene and olivine ± amphibole) has a similar mean temperature to that of the fracture fills but displays a much broader temperature range. In particular, amphibole associated with pegmatitic patches, which is included in the mafic replacement population here, has the broadest temperature range and the lowest average temperatures suggesting a prolonged interaction with hydrothermal fluids [Weston, 1998]. This is consistent with the greater degree of plagioclase alteration in these samples. Fibrous amphibole replacing olivine in a single sample from the southern slope of the intrarift ridge records very high temperatures (740°–789°C) which may reflect the reactivity of olivine with hydrous fluids; however, our data set for this texture is too limited to draw general conclusions. These results confirm that initial seawater penetration into the gabbroic sequence occurred between ∼700° and ∼750°C and continued down to ∼600°C. This is consistent with temperature estimates based on major element compositions and O-isotope data for plagioclase [Agrinier et al., 1995; Früh-Green et al., 1996; Lécuyer and Reynard, 1996] and compositions of secondary clinopyroxene [Gillis, 1995; Lécuyer and Reynard, 1996; Manning and MacLeod, 1996]. Subsequent fluid-rock interaction associated with Cocos-Nazca rifting was initiated at ∼450°C and extended down to ∼150°C [Agrinier et al., 1995; Früh-Green et al., 1996].

4. Discussion

[32] In the following sections we address some basic questions regarding the near-axis evolution of oceanic crust formed at fast spreading ridges. On the basis of the similarities in their petrology and their recovery within <500 m of the base of the sheeted dyke complex, we group samples from ODP Site 894 and the northern rift valley scarp together in this section and refer to these as “shallow gabbros.” To distinguish these and samples from the southern slope of the intrarift ridge we refer to the latter as “deep gabbros.”

4.1. Are Melts Added to the Crust Fully Aggregated? Not All of Them

[33] The compositional variability in melts crossing the Moho is controlled by the composition of melts generated in the mantle and the mechanisms of their extraction. A wide range of melt compositions are generated within the mantle and then mixed, or aggregated, prior to eruption of relatively homogeneous MORB. For example, the compositions of melt inclusions in olivine [e.g., Sobolov and Shimizu, 1993] and plagioclase [e.g., Sours-Page et al., 1999] phenocrysts record some of the preaggregation heterogeneity. Clearly, even erupted lavas are not “completely aggregated”' because lavas with different parental compositions erupt in close spatial and temporal proximity; however, the degree of variability is much less than that generated within the mantle. Oceanic gabbros can record similar information to that of melt inclusions in both the compositions of their constituent minerals [e.g., Coogan et al., 2000a] and in the phase equilibria suggested by cumulate assemblages.

[34] A general crystallization sequence for MORB is: olivine ± spinel to olivine + plagioclase to olivine + plagioclase + clinopyroxene to plagioclase + clinopyroxene + orthopyroxene ± olivine. It has been suggested that the first clinopyroxene to crystallize should have an Mg# < 88 due to an interval of olivine-plagioclase co-precipitation prior to clinopyroxene saturation on the basis of experimental studies [e.g., Elthon et al., 1992]. More primitive clinopyroxene could form either if the parental melts are dissimilar to those on which the experiments were performed, such as having higher Ca/Al [Coogan et al., 2000a] or if clinopyroxene nucleation and/or composition were kinetically controlled in the experiments. Thus it is not entirely clear if the occurrence of high Mg# clinopyroxene requires an incompletely aggregated parental melt or is a normal feature of MORB differentiation.

[35] Phase equilibria studies have shown that primitive MORBs are not in equilibrium with the shallow mantle (they are not orthopyroxene saturated) [O'Hara, 1968; Stolper, 1980]. If primitive MORBs are assumed to be similar in composition to the average Moho crossing melt, then this requires that melts generated at high pressure are extracted from the mantle without equilibrating with the shallow mantle. The occurrence of high Mg# orthopyroxene in the deeper gabbros thus presents strong evidence for crystallization from an incompletely aggregated parent. MORB does not reach orthopyroxene saturation until a significantly lower temperature than clinopyroxene saturation by which stage the melt, and orthopyroxene precipitated from it, will be relatively evolved (Mg#opx ∼ 80). The simplest explanation for the occurrence of primitive orthopyroxene is that there is significant variability in the major element composition of melts crossing the Moho compared to those used in phase equilibria experiments (i.e., that aggregation of the diverse melt compositions produced within the mantle occurs, in part, within the crust). Further evidence that incompletely aggregated MORB crosses the Moho comes from the occurrence of some plagioclase which are more calcic (up to An90) than any expected to be in equilibrium with fully aggregated MORB [e.g., Grove et al., 1992].

[36] Despite the evidence from mineral major element compositions that incompletely aggregated melts were added to the crust, clinopyroxene incompatible trace element compositions suggest that the parental melts were MORB-like (Figure 5). This contrasts with the evidence from melt inclusions that magmas with “depleted” major element compositions also have ”depleted“ incompatible trace element signatures [e.g., Sobolov and Shimizu, 1993]. It also contrasts with the findings of Benoit et al. [1999] who suggest that some orthopyroxene bearing cumulates in the Oman ophiolite formed from “second-stage” melting of the mantle in response to reheating of hydrothermally altered peridotites because the cumulates they describe have very depleted incompatible trace element signatures.

[37] The paradox that primitive melts that are in equilibrium with orthopyroxene, but that have MORB-like trace element compositions, are added to the crust can be explained if either (1) a fully aggregated MORB reacted with the upper mantle during extraction from the mantle or (2) the phase equilibria of a fully aggregated MORB was changed due to the addition of water [Boudier et al., 2000]. These are discussed in turn below.

[38] It is generally thought that melt extraction from the mantle occurs principally in dunite conduits [Kelemen et al., 1997a; Suhr, 1999; Lundstrom et al., 2000] because the dunite can “shield” the melt from reequilibrating with the surrounding orthopyroxene-bearing mantle. However, there is also evidence that at least some portion of melt migrates more diffusely for instance along grain boundaries or in less well-defined conduits in which the wall rock and melt are not completely chemically isolated [Kelemen et al., 1997a; Niu, 1997; Asimow, 1999]. Dissolution-reprecipitation reactions would force melts distributed along grain boundaries into major element equilibria with a harzburgitic residue, while they could retain their MORB-like incompatible trace elements abundances. Direct evidence for melts being driven toward orthopyroxene saturation during transport through the upper mantle comes from the occurrence of dunitic carapaces around gabbroic bodies from ODP Site 895. The dunite carapaces indicate melts migrating through these channels dissolved orthopyroxene from the surrounding mantle and thus must have moved into, or closer to, equilibrium with orthopyroxene. Clinopyroxene within these gabbroic rocks have trace element compositions in equilibrium with fully aggregated MORB [Dick and Natland, 1996] and some of these veins even contain primitive orthopyroxene [Arai and Matsukage, 1996]. A significant problem for models of melt reequilibration with the shallow mantle is that clinopyroxene in abyssal peridotites, including those from Hess Deep [Dick and Natland, 1996], is much too depleted in incompatible trace elements to be in equilibrium with fully aggregated MORB. This requires either that porous flow velocities are too rapid to allow the melt and matrix to fully equilibrate or that flow is partially channelized with only partial isolation of the melt from the surrounding harzburgitic mantle. Thus some melts probably partially reequilibrate with the upper mantle during melt extraction while others are extracted through dunitic conduits and remain out of major element equilibrium with the upper mantle.

[39] The second model is that addition of water to the magma chamber could stabilize orthopyroxene [Boudier et al., 2000]. This seems an unlikely origin for these cumulates because (1) if water entered the lower crust via roof assimilation, the most obvious mechanism, eruptive equivalents would be expected to be found at MORs, which they are not. Alternatively, if the water entered the mantle [e.g., Benoit et al., 1999] and allowed remelting of a depleted source, a depleted REE signature would be expected; (2) no oxides are present in the primitive orthopyroxene-bearing sample as might be expected if they had crystallized under conditions of very high oxygen fugacity as suggested by Boudier et al. [2000]; (3) the primitive compositions would require that this increase in oxygen fugacity occurred without much crystallization which is unlikely if the water came from assimilation; and (4) the orthopyroxene co-exists with plagioclase which would be destabilized by the addition of H2O. Despite these arguments, isotopic studies [e.g., Benoit et al., 1999] are required to conclusively test this hypothesis. Whatever the origin of this primitive orthopyroxene, its occurrence indicates that primitive cumulus orthopyroxene cannot be used as an unambiguous indicator of a suprasubduction zone setting in ophiolite plutonic sequences.

4.2. How Do Melts Fractionate Within the Lower Crust?

[40] Modeling of the whole rock compositions of gabbros and basalts from the EPR (Figure 2) suggests that, in general, the compositional variability in both lithologies is similar to that expected during fractional crystallization (Figure 2). In detail, however, imperfect crystal-liquid separation is required to explain the high incompatible element abundances in some gabbros (Figure 2). The whole rock geochemistry also shows that the deeper gabbros are generally more primitive and have lower incompatible element abundances than the shallower gabbros (Figure 2) [Blum, 1991; Natland and Dick, 1996; Pedersen et al., 1996]. Clinopyroxene trace element compositions support this and show that most of the deeper gabbros crystallized from melts with REE abundances similar to those of erupted lavas. The occurrence of crystal cores that preserve evidence of growth from a primitive melt (high An plagioclase, high Ni olivine, and high Cr clinopyroxene) in gabbros from ODP Hole 894G (Figure 3) [Natland and Dick, 1996; Pedersen et al., 1996] suggests that their parental melt was as primitive as that of cumulates from deeper in the crust and that their more evolved bulk compositions are largely due to reaction with interstitial melt consistent with the interpretation based on whole rock compositions. An alternative hypothesis, that the primitive portions of the crystals in ODP Hole 894G are xenocrysts [Natland and Dick, 1996] and the Site 894 gabbros crystallized from a highly fractionated liquid, is not supported by the plagioclase CSDs, which show little or no evidence of mixing of two populations of plagioclase crystals. Neither is this supported by the lack of a positive Ti anomaly in the average composition of the core (Figure 2b). Positive Sr anomalies show that all of the plutonics have accumulated plagioclase (Figure 2b) and thus none are simply frozen melts. In summary, the data can be explained relatively well by imperfect fractional crystallization of relatively primitive parental melts with the shallow-level plutonics recording evidence of less efficient crystal-melt separation than the deeper plutonics. However, important roles for other processes (e.g., boundary layer fractionation [Langmuir, 1989] and assimilation [O'Hara, 1977]) cannot be ruled out.

4.3. How is the Lower Crust Accreted? From The Top and The Bottom?

[41] Models for lower crustal accretion can be split into two end-member classes based on where crystallization is assumed to occur. In the first class, most crystallization occurs within the AMC, with crystals subsiding downward and moving outward to form the lower crust [Henstock et al., 1993; Nicolas et al., 1993; Phipps Morgan and Chen, 1993; Quick and Denlinger, 1993]. The majority of the latent heat of crystallization is thus released within the AMC and can be removed by the axial hydrothermal system within the upper crust. In the second class of models crystallization occurs in situ throughout the lower crust in a series of sills leading to distributed latent heat release, in turn, requiring the hydrothermal system in the lower crust to efficiently remove this heat [Kelemen et al., 1997b; Korenaga and Kelemen, 1998; MacLeod and Yaouancq, 2000]. Boudier et al. [1996b] propose a hybrid model in which the crust accretes from the top and bottom. The petrology of the Hess Deep gabbros provides evidence for accretion from both the top and the bottom but provides little insight into the relative proportions of each.

[42] Support for models of accretion from the top comes from four observations. First the occurrence of clinopyroxene and olivine crystals with high Cr and Ni abundances, and highly zoned plagioclase with calcic cores, within the shallow gabbros suggest that parental melts at all levels in the plutonics are similarly primitive (Figure 4). Crystal size distributions indicate that these primitive crystals are not xenocrysts. Thus not all melts can have fractionated in multiple sills in the lower crust prior to feeding the AMC. If it is assumed that the melt parental to the gabbros from ODP Hole 894G was typical of those added to the AMC, mass balance indicates that melts must fractionate within the AMC since the upper crust (dykes and lavas) is on average more evolved than this. The crystals formed during fractionation must be removed since the AMC is a steady state feature, and the simplest way to remove them is by downward subsidence. The same observation of primitive melts feeding the upper gabbros has been made in the Oman ophiolite [Coogan et al., 2002b]. Second, steeply dipping magmatic fabrics in samples from Hole 894G suggest subvertical flow of a crystal mush at this level. Given the thermal and mass balance constraints, these are most readily explained by crystal alignment during subsidence of a crystal mush from the AMC along with the upward flow of melt to the AMC [e.g., Quick and Denlinger, 1993]. Third, relatively limited evidence for hydrothermal circulation within the lower crust suggests that heat extraction from the lower crust is unlikely to be efficient enough to drive large-scale crystallization in situ within the deeper parts of the lower crust on-axis. Finally, the larger amounts of reaction with interstitial melt at shallow levels in the plutonics than at deep levels may reflect less efficient compaction of interstitial melt out of this region during crystal subsidence [Henstock, 2002]. This scenario is consistent with the observation from the EPR that the sub-AMC region is characterized by lower seismic velocities than deeper levels [Dunn et al., 2000]. These observations are consistent with thermal arguments that suggest that accretion of the lower crust from the top must be an important process [Henstock et al., 1993; Phipps-Morgan and Chen, 1993; Chen, 2001].

[43] Evidence for crystallization in situ within the gabbroic section at deeper levels than the AMC comes from the occurrence of primitive orthopyroxene crystals within the deeper gabbros from the southern slope of the intrarift ridge. This suggests that some melts that are added to the crust crystallize and/or react with preexisting cumulates, within the deeper parts of the plutonic sequence, prior to full aggregation. It seems improbable that this could occur within the AMC where an incompletely aggregated major element signature would be expected to be rapidly annihilated through mixing (primitive orthopyroxene phenocrysts are, to our knowledge, unknown along the EPR). Thus a composite model of crustal accretion from the top and from the bottom is most likely [Boudier et al., 1996b] although the relative importance of each is unclear.

4.4. How Efficiently Does Hydrothermal Circulation Cool the Lower Crust? Not Very?

[44] The role of hydrothermal circulation in extracting heat from the lower crust is poorly constrained despite its importance in driving crystallization and in altering the bulk composition of the crust and seawater. The initial penetration of seawater-derived hydrothermal fluids into the lower crust occurs at ∼750°C along a microfracture and macrofracture network leading to the formation of amphibole veins and the replacement of some primary mafic minerals [Manning and MacLeod, 1996; Manning et al., 1996]. Rapid formation of amphibole seals the porosity and permeability and inhibits heat extraction. Evidence for this comes from (1) the correlation between amphibole composition and wall rock phase which suggests low mass transport length scales and thus low fluid fluxes [Manning and MacLeod, 1996] and (2) the narrow temperature interval of amphibole-plagioclase equilibration [Manning et al., 1996] and the similar average temperatures for microveins, macroveins and mafic phase replacement (Figure 9; Table 4). This suggests that fluid flow occurred over a small temperature and time interval. In particular, the observation that microfractures form earlier than macrofractures in all samples but have similar mean temperatures [Weston, 1998] suggests that little heat is extracted by fluid flow during vein formation and (3) low calculated fluids fluxes from O- and Sr-isotope systematics [Lécuyer and Reynard, 1996]. More extensive alteration, which continued over a wider temperature interval, in pegmatitic gabbros from ODP Hole 894G [Weston, 1998] (Table 4) may reflect either greater intracrystalline cracking during cooling due to coarser crystal sizes or the higher H2O contents of these lithologies leading to a lower temperature brittle-plastic transition [Hirth et al., 1998]. There is no evidence for pervasive fluid flow at temperatures higher than those recorded by the amphibole-plagioclase thermometry as has been suggested by McCollom and Shock [1998]. However, if this flow occurred in a closed system (i.e., not in contact with the ocean), it is possible that the fluid could transport heat without extensive mass exchange.

[45] The similarity in amphibole-plagioclase equilibration temperatures in shallow and deep gabbros suggests that the temperature of fluid ingress is constant with depth. For any reasonable thermal structure [e.g., Phipps Morgan and Chen, 1993] this suggests lower crustal cracking occurs at different levels in the crust at different crustal ages (i.e., different distances off-axis) [Manning et al., 2000]. There is no evidence for either a decrease in the temperature of initial fluid penetration with depth as predicted by models of the brittle-plastic transition (BPT) [Hirth et al., 1998] or an increase as observed by Manning et al. [2000] in the Oman ophiolite. These discrepancies can be explained in several ways. The water content of the plutonics is likely to decrease with depth in the lower crust similar to the other incompatible elements and also because the effects of assimilation of hydrated roof material will be greatest in the shallow plutonics. This will lead to the shallow gabbros being weaker than predicted by Hirth et al.'s. [1998] dry diabase model and would go some way toward mitigating the predicted decrease in the temperature of the BPT with depth. Alternatively, or additionally, the assumption of Hirth et al. [1998] that all material at temperatures below the BPT has hydrostatic fluid pressure is unlikely to be true due to the rapid sealing of permeability by amphibole precipitation. Variations in strain rate with depth may also effect the temperature of the BPT [e.g., Manning et al., 2000]. Finally, the thermal model assumed by Hirth et al. [1998] is unlikely to be accurate for the EPR as it is based on a model for the slow-spreading Mid-Atlantic Ridge.

5. Summary and Conclusions

[46] A review of the igneous and metamorphic petrology of the lower crust formed at the EPR and exposed at Hess Deep allows us to make some first-order interpretations about magmatic and hydrothermal processes operating at fast spreading mid-ocean ridges.

[47] 1. Moho crossing melts are more compositionally diverse than erupted MORB. Some are in equilibrium with orthopyroxene and calcic plagioclase but have MORB-like incompatible trace element abundances. These unusual compositions are most easily explained if fully aggregated melts react with the upper mantle during melt extraction. Once in the crust, some melts must crystallize in isolation from fully aggregated MORB in order for their signature to be preserved in cumulates.

[48] 2. Comparison of the whole rock compositions of basalts and gabbros suggest that the fractionation of MORB occurs via imperfect fractional crystallization. Mineral compositions and zoning support this hypothesis and suggest the parental melts at all levels in the crust are quite primitive. The final crystal-melt separation is more efficient at deeper than shallower levels in the lower crust as is suggested by seismic studies of the EPR [e.g., Dunn et al., 2000].

3. Lower crustal accretion most likely occurs from the top and bottom [Boudier et al., 1996b], but the relative proportions of each cannot be constrained from present petrological constraints.

[50] 4. High temperature fluid-rock interactions are initiated at ∼750°C and are, in the main, limited to a ≤150°C temperature interval prior to the system being sealed up by amphibole precipitation. This limited temperature interval of hydrothermal circulation, evidence for limited mass transport distances, and O-isotope modeling all suggest limited fluid fluxes.


[51] We would like to thank Jim Natland and an anonymous reviewer for journal reviews and Susumu Umino and Bill White for editorial advice and scientific suggestions. Rolf Pedersen, John Malpas, Jim Natland, and Craig Manning are thanked for the loan of samples and/or for sharing unpublished data. Discussions with Richard Thomas have enhanced our (still limited) understanding of the petrology of the lower crust in the Oman ophiolite. Discussions at the conference “The Geology of Oman” in 2001 inspired us to think harder about the petrology of present-day lower ocean crust formed at a fast spreading ridge. Rob Wilson assisted with electron probe analyses and Richard Hinton and John Craven with ion microprobe analyses. We are grateful to the chief scientists of the NAZCOPAC (J. Francheteau) and ALVIN (P. Lonsdale and J. Karson) cruises for allowing access to some of the samples used in this study. The NAZCOPAC cruise was sponsored and supported by IFREMER and INSU (Institut National des Sciences de l'Univer). LAC was partially supported by Cardiff University and his attendance at the Oman conference was funded by the Royal Society. KMG was supported by an NSERC research grant.