Ce que nous apprennent les roches du manteau

Ce que nous apprennent les roches* du manteau sur la migration des magmas dans le manteau Peter Kelemen * Roches experimentales, volcaniques et du manteau

Minerals in the mantle and lower crust Olivine Orthopyroxene Clinopyroxene Spinel Garnet Plagioclase Mg 2 Si. O 4 - Fe 2 Si. O 4 Mg 2 Si 2 O 6 - Fe 2 Si 2 O 6, etc Ca. Mg. Si 2 O 6 - Ca. Fe. Si 2 O 6, etc (Mg, Fe)(Cr, Al)2 O 4 , etc (Mg, Fe, Ca)3 Al 2 Si 3 O 10, etc Ca. Al 2 Si 2 O 8 - Na. Al. Si 3 O 8 Melting reactions P > 20 kilobars (2 Gpa) Ol + Opx + Cpx + Gnt = melt 8 kb < P < 20 kb Opx + Cpx + Sp = Ol + melt P < 8 kb Opx + Cpx + Plag = Ol + melt if fertile Opx + Cpx + Sp = Ol + melt if depleted

Partial melts of peridotite have increasingly high Si. O 2 at lower pressure. This reflects incongruent melting, producing olivine + relatively Si. O 2 rich melt from pyroxenes + spinel and/or garnet and/or plagioclase. The spread in the data reflects increasing Si. O 2 at a given pressure as the degree of melting rises from moderate to high. As a consequence of the effect of pressure, as they decompress, mantle melts will be saturated only in olivine, and have the capability of dissolving pyroxene via general reactions in which melt 1 + pyroxene = melt 2 + olivine, where melt 2 has higher Si. O 2 than melt 1. Compositions of partial melts of “dry” natural peridotite, compiled in the mid-1990’s. Hirose = Hirose & Kushiro, EPSL 1993, Kinzler = Kinzler & Grove, JGR 1992 a, b & 1993, Baker = Baker & Stolper, GCA 1994, Falloon = Falloon & Green Min & Pet 1987, J Pet 1988, Falloon et al. J Pet 1988 An-p-to-date compilation would add experimental data from Walter, J Pet 1998, Wasylenki et al J Pet 2003, Pickering et al CMP 2000, and Johnston & Schwab, GCA 2004, among others. really low F really high F ~3 to 20% melting

pyroxenes dissolve olivine precipitates, Si. O 2 up peridotite dissolves (even olivine), Mg. O up { { tion m on Depth satura i rat atu es en rox olivine py Partial melts of peridotite have increasingly high Si. O 2 at lower pressure. This reflects incongruent melting, producing olivine + relatively Si. O 2 rich melt from pyroxenes + spinel and/or garnet and/or plagioclase. The spread in the data reflects increasing Si. O 2 at a given pressure as the degree of melting rises from moderate to high. As a consequence of the effect of pressure, as they decompress, mantle melts will be saturated only in olivine, and have the capability of dissolving pyroxene via general reactions in which melt 1 + pyroxene = melt 2 + olivine, where melt 2 has higher Si. O 2 than melt 1 an tle so lid us Temperature liquid adiabat

periodic table in approximate order of crystal/liquid partitioning Rare Earth Elements in order of increasing Z More introductory material: Petrologists & geochemists normalize trace element contents of rocks to solar system, chondritic meteorite, or “primitive mantle” AKA “silicate earth” abundances in order to remove the effects of fusion, neutron capture and fission on elemental abundances, and emphasize fractionation due to terrestrial processes.

Normalized concentrations, showing relatively smooth REE and trace element variation “Normal MORB”, AKA N-MORB, C 1 chondrite and primitive mantle compositions from Anders & Grevesse, GCA 1989 and Mc. Donough & Sun, Chem Geol 1995

Bottom up: Top down: Diffuse porous flow Melting & diapirs Magma fracture Focused porous flow Sills & lenses at “top” MORB composition MORB focusing MORB ascent rate Arc composition Arc focusing φ= WFρ s/( wρ f) STEADY STATE! Hotspot flux, comp, ( = 1) focusing w = kΔρg/(φμ f) “DARCY’S LAW” Steady state melt flux relationship, indicating that production must equal transport flux, modified from Spiegelman & Kenyon, EPSL 1992 and later (with ) from Kelemen et al. , Phil Trans Roy Soc London 1997; Darcy’s Law holds generally for porous flow (grain boundary, tubes, cracks, etc) and relates melt velocity w to permeability, k, density contrast between melt and solid , the gravitational constant g, melt fraction , and melt viscosity f. W = solid upwelling velocity, F = melt fraction at some height, = density of solid s and “fluid” f, w = melt or fluid velocity, = volume fraction of “channels” carrying melt; = 1 means that the entire source region transports melt via diffuse porous flow; alternatively, = 0. 01 would indicate that 1% of the volume of the source region is composed of channels transporting melt.

Most scientists believe that partial melt in an olivine rich matrix occupies olivine triple grain boundaries and four grain intersections in a network resembling this network formed by melt (now glass) and gas bubbles. See Wark et al. , JGR 2003, for photo and extensive discussion.

Von Bargen & Waff Wark, Watson, et al. k = d 2φ3/270 Faul et al. w = kΔρg/(φμ f) k = d 2φ3/c Figures from Wark et al. , JGR 2003 and Wark & Watson, EPSL 1998, illustrating various estimates for permeability of grain boundary networks of basaltic melt in an olivine rich matrix. Sources for estimates: Faul, JGR 1997; Von. Bargen & Waff (also VBW), JGR 1986; RK = Riley & Kohlstedt EPSL 1991; C = Cheadle, 1989 thesis at Cambridge Univ (also see Cheadle et al. , Geology, 2004); MK = Mc. Kenzie EPSL 1989; Wark, Watson et al. = Wark & Watson, EPSL 1998; Liang et al. , JGR 2002; Wark et al. , JGR 2003

Von Bargen & Waff Figures from Wark & Watson, EPSL 1998, illustrating theoretical grain boundary networks determined by equilibrium “wetting angle”, which is a function of solid-solid and solid-liquid surface energy, and experimental results on permeability of quartzite+water±Na. Cl and marble+fluid networks.

Grain size variation: some grains smaller, more melt on triple grain boundaries (= grain edges) At low melt fraction, little or no melt on large grain edges If rock is banded in grain size, low permeability to banding Explanation from Wark et al. , JGR 2003, for slightly lower permeability in experimental quartzite compared to Von. Bargen & Waff theory, in terms of anistropic melt distribution as a function of grain size.

HARZBURGITE (+) OL + SP ( ) hz ol+sp ol OL only ( )

quartzite Fa ul et a l. marble hz ol+sp ol Von Bargen & Waff Wark, Watson, et al. As for previous slide, but also showing similarity of experimental results for melt+olivine±spinel±opx and results for quartzite+H 2 O±Na. Cl.

compositional variation across a large dunite in the Josephine peridotite A percolation threshold, AKA permeability threshold, such as that predicted by Faul, JGR 1997, should be evident in mantle residues as a finite proportion of “trapped melt” that remains in the rock after cooling. One can use mass balance to estimate an “upper bound” for trapped melt in DUNITE by analyzing whole rock and mineral compositions. In this example, Sundberg et al. (Marshall Sundberg, Peter Kelemen & Greg Hirth, in prep. ) analyzed Ca. O in olivine and whole rock. It was assumed that the Ca. O content of spinel is negligable. The Ca. O that is NOT in olivine could represent trapped melt, but it could also represent “cumulate” clinopyroxene or plagioclase that crystallized in pore space from cooling, migrating melt after which the remainder of the melt continued to flow out of the rock. Ca. O could also be added to the rock during alteration, which is evident as minor amounts of serpentine along olivine grain boundaries in this rock. Minor amounts of Ca. O could be removed from the rock during serpentinization of olivine, but for these samples serpentine represents less than 10% of the rock, and the concentration of Ca. O that could be removed from olivine is in any case very low.

upper bound estimate of “permeability threshold” based on upper bound estimate of “trapped melt”, based on Ca. O in whole rock - olivine Inferred upper bound proportion of trapped melt (X melt = [Ca. O in whole rock - Ca. O in olivine*X olivine - Ca. O in spinel (0) * X spinel]/Ca. O in melt) in dunite samples from the Josephine peridotite.

Von Bargen & Waff Wark, Watson, et al. X Faul et al. w = kΔρg/(φμ f) k = d 2φ3/c X Wark, Watson, et al. k = d 2φ3/270 In conclusion, for melt±spinel± 20% opx, the Wark et al JGR 2003 formulation for permeability seems to be (approximately) applicable, and the Faul, JGR 1997 formulation is almost certainly incorrect.

k = d 2φ3/c c is a “geometric factor” Wetting angles may vary depending on crystallographic orientation and mineral At low melt fractions, “unfavorable” grain edges have no melt at all Positive or negative feedback on permeability? In fact, a range of wetting angles can be anticipated to be present, even in olivine + melt systems, because surface energies are not the same on all olivine grain faces. In harzburgites and lherzolites, surface energies of pyroxene/melt are higher than those for olivine/melt, and wetting angles will be larger for triple grain junctions involving pyroxene. As a result, at “low” melt fractions, it can be anticipated that unfavorable triple grain junctions will not be occupied by melt. Two possible consequences can be anticipated. On the left, Wark et al. JGR 2003 (or is this from Wark & Watson, EPSL 1998, I forget) schematically illustrate the possible increase in the “effective grain size” resulting from the lack of melt on some triple grain junctions. In this case, since the permeability depends on the square of the grain size, one might expect that permeability would be higher; this results from the fact that the effective diameter of the remaining pores is larger than if all triple grain junctions contained melt. On the right, Zhu & Hirth, EPSL 2003 illustrate the effect of pyroxene - in this case opx - on the wetting angle in melt bearing triple grain junctions. Because the wetting angle exceeds 60° for ol-opx and opxopx-opx triple grain junctions, it can be anticipated that at low melt fractions melt on these junctions (if any) would be in isolated “pockets” rather than an interconnected “network”, lowering the overall permeability. The relative importance of these negative and positive feedbacks is the subject of ongoing research.

Results from Zhu & Hirth, EPSL 2003, for model mixtures of olivine and opx. Mixtures including ~ 55% or less opx are predicted to have permeability very similar to olivine only rocks (VBW model in upper left figure), while those with 60% or more opx have an effective average wetting angle greater than 60° and are predicted to have a percolation threshold below which most of the melt is in isolated pockets rather than an interconnected grain boundary network.

ol + melt 6 h ol + opx + melt ol + melt 6 h ol ± opx NO initial melt ol + opx + melt ol ± opx NO initial melt Results from Daines & Kohlstedt, Phil Trans Roy Soc London 1993; top two panels show results of experiments in which the left side of an experimental capsule initially contained olivine + ~ 10% melt, and the right side contained olivine + ~ 3% melt. Because of forces arising from surface energy differences between the two sides, melt flows from the melt-rich source on the left to the melt-poor sink on the right, and permeability can be estimated from such experiments. The lower two panels show results of experiments in which the melt-rich source contained olivine + opx + melt, while the sink contained olivine or olivine + 20% opx (no initial melt). This is not an ideal geometry, because in theory the driving force for melt to enter the melt-free sink is infinite at zero melt fraction. However, the experiments showed that, if anything, olivine + 20% melt was MORE permeable than olivine alone at the same melt fraction.

f 3 = 270 m. WFrs/(d 2 Drgrf) from φ= WFρ s/( wρ f) STEADY STATE! ( = 1) w = kΔρg/(φμ f) “DARCY’S LAW” k = d 2φ3/270 Wark et al. Concluding that the Wark et al. , JGR 2003 expression for permeability is approximately correct, and combining this with Darcy’s Law plus the 1 D steady state melt flux from Spiegelman & Kenyon, EPSL 1992, I derived an expression for melt porosity as a function of melt viscosity, solid upwelling rate, fraction of melting at a given depth, densities, grain size, and density contrast between melt and solid. This one is new, but for similar calculations (using the Von. Bargen & Waff permeability expression) see Kelemen et al. , Phil Trans Roy Soc London 1997, Figures 9 & 10 and associated text. Given the porosity, one can calculate permeability, melt velocity, and so on. The calculation shown here uses values appropriate for high Mg. O, relatively low Si. O 2 melts near the base of a melting column beneath a ridge, with density difference between solid and melt of only 250 kg/m^3, and melt viscosity of only 2 Pa s. Given these values, and the steady state assumption, porosity is unlikely to exceed 2% over the entire melting column beneath a fast spreading ridge, and will be less than 1% beneath a slow spreading ridge. Such low porosities, and the steadily increasing values of porosity and permeability at decreasing depth, may tend to stabilize a steady state melt transport system and damp the formation of porosity waves, diapirs and “magmafractures”.

Bottom up: Top down: Diffuse porous flow OK, prefer Wark et al. (for now) field evidence? Melting & diapirs Magma fracture Focused porous flow Sills & lenses at “top” MORB composition MORB focusing MORB ascent rate Arc composition Arc focusing Hotspot flux, comp, focusing Summary to this point. Diffuse porous flow is likely, and there is no evidence for an appreciable permeability threshold, at least in an olivine-rich peridotite at melt fractions greater than 0. 3%. We now move on to field evidence for diffuse porous flow of melt in the mantle

Models of regional pervasive porous flow conflict with structural and seismic evidence that fractures control fluid transportation in the upper mantle. Effects of porous -medium flow have been inferred in studies of mantle peridotite … but are well documented only on scales of centimeters or decimeters. In all these [cases], porous flow is fundamentally controlled by proximity to magmafilled fractures. Nielsen & Wilshire, 1993 The idea that diffuse porous flow of melt is an important mechanism for melt transport has been challenged, as one can see here.

Johnson et al. , JGR 1990 and Johnson & Dick, JGR 1992, showed that peridotite residues of melting beneath mid-ocean ridges closely approximate residues of fractional melting, and thus require melt transport at very low melt fractions.

On the left, another figure from Johnson et al. , JGR 1990, illustrating the large observed variations in Ti and Zr in cpx, and contrasting these with the small variations predicted for “batch melting” - where melt does not leave the rock until melting is complete, so that high melt porosities are attained - and the much larger variations predicted for fractional melting. OK, so it is evident that very small melt fractions must be mobile somewhere in the melting column, and this requires porous flow. As shown on the right, in order to approximate fractional melting, melt must flow laterally - or diagonally, … - into some kind of non-reactive conduit for focused flow that then carries it away from the residual peridotites. If, instead, all the melt were extracted via diffuse porous flow upward through the entire melting region, the composition of melt and residue would approach that of a batch melting system, as shown by, e. g. , Spiegelman & Kelemen, G-cubed 2003. However, the distances over which diffuse porous flow into focused, non-reactive conduits takes place are not evident from the data on mid-ocean ridge peridotite residues. Some scaling is provided via the reasoning of Spiegelman & Kenyon, EPSL 1992. Thus, these data indicate that very small fractions of melt must be mobile somewhere in the melting region beneath ridges (for a critical appraisal of this constraint, see Kelemen et al. , Phil Trans Roy Soc London, 1997, Figure 2 and associated text.

“U-shaped” REE patterns for peridotites from the New Caledonia massifs, from Prinzhofer & Allegre EPSL 1985

Explanation, from Navon & Stolper, J Geol 1987, of U-shaped REE patterns in terms of “chromatographic” exchange between migrating melt and residual mantle peridotite. Similar processes and formulations have been proposed by Kelemen, J Geol 1986; Kelemen et al. , Nature 1990; Ozawa & Shimizu JGR 1995; Ozawa JGR 2001; Takazawa et al Nature 1992; Godard et al EPSL 1995; Vernieres et al JGR 1997

Data on 117 residual mantle peridotites dredged from the mid-ocean ridges, analyzed by Niu, J Petrol 2004. All of these rocks show U-shaped extended trace element variation diagrams. In addition, as shown schematically on the right hand panel, whole rock concentrations of highly incompatible elements are commonly higher than concentrations in cpx, suggesting that melts migrating along grain boundaries did not equilibrate with cpx cores, but did enrich cpx rims and other mineral phases in highly incompatible trace elements.

On the left, diagram from Niu, J Petrol 2004, showing how whole rock compositions commonly have higher light REE concentrations than cpx in the sample. Only the panel outlined in red is “normal” in the sense that in equilibrated spinel lherzolites, cpx is the main host for REE thus has higher REE concentrations than the whole rock. Niu interpreted the late, grain boundary enrichments in light REE and other highly incompatible elements to crystallization of cooling melt, migrating along grain boundaries into thermal boundary layer at the base of the “lithosphere”. On the right, enrichment of residual peridotites in the Wadi Tayin massif of the Oman ophiolite (Hanghoj et al. , in prep) in incompatible elements increases upward, toward the crust-mantle transition zone (MTZ). This is the opposite of the pattern expected for residues of decompression melting, which should be most depleted at the shallowest depths. These data are consistent with the idea that refertilization of residual peridotites occurs as cooling melt migrates upward into thermal boundary layer at the base of a growing oceanic plate.

melt out residual porosity nothing coming in nothing out local melt coming in melt out melt coming in residual porosity nothing out MORB coming in Godard et al, EPSL 2000, considered various alternatives to understand REE patterns in residual mantle peridotites from the Maqsad area in the Samail massif of the Oman ophiolite, and concluded that REE systematics were most consistent with addition of “trapped melt” with the composition of primitive mid-ocean ridge basalt (MORB) like the primitive melts that formed the crust in the Oman ophiolite.

Figures from Suhr et al. , G-cubed, submitted, showing textures indicative of late, refertilization of residual peridotites via crystallization of small “cumulate” pyroxene and spinel grains from melt migrating along residual mineral grain boundaries. Samples from ODP Site 1274 at about 15° along the Mid-Atlantic ridge. Very similar data from the same region were presented by Seyler et al. , CMP 2007.

More textures indicative of crystallization of late, cooling melt along olivine and opx grain boundaries in residual mantle peridotite from the 14 to 16°N area along the Mid-Atlantic Ridge, from Kelemen et al. , ODP Leg 209 Initial Reports, 2004, and Kelemen et al. , Summary, ODP Leg 209 Scientific Results, 2007. Below, a figure from Suhr et al. , G-cubed, submitted, emphasizing that opx-poor harzburgites from Site 1274 have HIGHER incompatible element concentrations than more normal harzburgites and dunites, consistent with the presence of abundant, late, grain boundary refertilization of the opxpoor harzburgites.

hic” reaction al peridotites on a regional le exception ridotite data, asets r really ngth scale transport. nstraints, we the massiff ariation in the da dertaken by o-workers. rom Bodinier, s 1988, l scale ajor and omposition “Decoupling” ace elements low Al high Al low Al light REE d light REE “enriched”

Porphyroclastic (low T) Light REE Enriched (addition of low degree melts) coarse, granular (high T) light REE depleted (“MORB source”) Regional variation in the Ronda peridotite, documented most recently by Lenoir et al. , J Petrol 2001, shows regional scale textural and geochemical variation resulting from “chromatographic” reaction between residual peridotite and melt migrating via diffuse porous flow, which also faciliated high temperature recrystallization of low temperature deformation textures in the pre-existing residual peridotite.