ARTICLESPUBLISHED ONLINE: 8 DECEMBER 2014 | DOI: 10.1038/NGEO2306

Lower-mantle water reservoir implied by theextreme stability of a hydrous aluminosilicateMartha G. Pamato†, Robert Myhill*, Tiziana Bo�a Ballaran, Daniel J. Frost, Florian Heidelbachand Nobuyoshi Miyajima

The source rocks for basaltic lavas that form ocean islands are often inferred to have risen as part of a thermal plume from thelower mantle. These rocks are water-rich compared with average upper-mantle rocks. However, experiments indicate that thesolubility of water in the dominant lower-mantle phases is very low, prompting suggestions that plumes may be sourcedfrom as-yet unidentified reservoirs of water-rich primordial material in the deep mantle. Here we perform high-pressureexperiments to show that Al2SiO4(OH)2—the aluminium-rich endmember of dense, hydrous magnesium silicate phase D—is stable at temperatures extending to over 2,000 ◦C at 26GPa. We find that under these conditions, Al-rich phase D isstable within mafic rocks, which implies that subducted oceanic crust could be a significant long-term water reservoir inthe convecting lower mantle. We suggest that melts formed in the lower mantle by the dehydration of hydrous minerals indense ultramafic rocks will migrate into mafic lithologies and crystallize to form Al-rich phase D. When mantle rocks upwell,water will be locally redistributed into nominally anhydrous minerals. This upwelling material provides a potential source forocean-island basalts without requiring reservoirs of water-rich primordial material in the deep mantle.

H igh-pressure experiments indicate that nominallyanhydrous minerals at >200 km depth in the uppermantle can host the H2O contents inferred for basalt source

rocks (typically 70–700wt ppm) as hydroxyl defects in their crystalstructures1,2. The H2O capacity of ultramafic rocks increases in thetransition zone, as recently confirmed by the discovery of hydrousringwoodite in diamond3. In contrast, the lower-mantle phasesbridgmanite (magnesium silicate perovskite) and ferropericlaseseem to have a much lower H2O solubility1,4,5. It is thereforeproblematic that ocean-island basalt (OIB) sources apparentlyoriginating in the lower mantle have the highest H2O contents6,7.

Several dense hydrous magnesium silicates (DHMS phases) arethermodynamically stable in peridotites within subducting litho-sphere8,9. In the lower mantle, the most important of these phasesare superhydrous phase B (also known as phase C, nominallyMg10Si3O14(OH)4; ref. 10), phase D (hereafter Mg-phase D; nom-inally MgSi2O4(OH)2; ref. 11) and the newly discovered phase H(nominally MgSiO2(OH)2; ref. 9). However, these Mg–Si endmem-bers break down at temperatures lower than those of typical mantlegeotherms, and as a result they cannot form long-term water reser-voirs in the lower mantle. An important issue, therefore, is whetherthese phases have solid solutions that can increase their thermalstability in lower-mantle assemblages.

One potential stabilizing component in hydrous phases atlower-mantle pressures is Al. Al-bearing Mg-rich phase D breaksdown at ∼1,600 ◦C, about 200 ◦C higher than the Mg-phase Dendmember12. Phase H can also accept Al, forming a solid solutionwith the similarly structured phase δ-AlOOH (ref. 13). In certaincompositions, this solid solution is stable even along typical mantlegeotherms at >40GPa (ref. 14). However, Fe counteracts thestabilizing effect of Al addition, such that Fe, Al-bearing phase Din ultramafic compositions may not be any more stable thanthe Mg-phase D endmember12. As a result, it is unclear whetherultramafic rocks in the convecting lower mantle can contain any

hydrous phases. An intriguing possibility is that the relatively highAl contents of recycled oceanic crust could yield greater hydrousphase stability than observed in ultramafic systems. The presenceof phase Egg (ideal formula AlSiO3(OH); ref. 15) in superdeepdiamond inclusions16 implies the presence of at least some hydrousrecycled oceanic crust in the mantle transition zone. Even ifsubduction effectively dehydrates crustal rocks before phase Eggbecomes stable17, mafic lithologies could be rehydrated at greaterdepths by hydrous melts released from surrounding ultramafics.

Recycled sections of mafic oceanic crust are unlikely to bechemically homogenized on the timescales of mantle convectionand will therefore persist as distinct lithologies18,19. There is thusa possibility that mafic components within a mechanically mixedlower mantle could become a focus for H2O, if a suitable host exists.One untested candidate is superaluminous phase D, which has beensynthesized at 25GPa and 1,500 ◦C (ref. 20) in a bulk compositionsimilar to portions of subducted oceanic crust. This phase D hashigh Al and low Fe contents, suggesting that it could be significantlymore stable than the magnesium silicate endmember. In this studywe investigate the stability field of Mg, Fe-free aluminous phaseD (hereafter Al-phase D; ideal formula Al2SiO4(OH)2) to examinewhether it could be a host for H2O in the uppermost lower mantle.

The stability and composition of Al-phase DAl-phase D was synthesized in multianvil experiments between1,460 ◦C and 2,100 ◦C in the simplified Al2O3–SiO2–H2O system.Two starting mixtures were employed with Al/Si ratios of 2:1(mixture 1) and 1:1 (mixture 2) and H2O contents of 13 and19wt% respectively (Supplementary Table 1). The resultingphase assemblages were identified using X-ray diffraction (XRD),electron probe microanalysis (EPMA), electron backscatterdetection (EBSD) and transmission electron microscopy (TEM;Supplementary Table 2). Compositions determined by EPMA arepresented in Supplementary Table 3.

Bayerisches Geoinstitut, University of Bayreuth, 95440 Bayreuth, Germany. †Present address: Department of Geology, University of IllinoisUrbana-Champaign, Illinois 61820, USA. *e-mail: bob.myhill@uni-bayreuth.de

NATURE GEOSCIENCE | VOL 8 | JANUARY 2015 | www.nature.com/naturegeoscience 75

© 2014 Macmillan Publishers Limited. All rights reserved

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2306

26

P (G

Pa)

1,000 1,200 1,400 1,600 1,800 2,000 2,200

Temperature (°C)

30

22

S5021_2

S4517_2

S4523S5081S5113S4517_1

S5050_2

[D]

[L]

[stv]

[cor]

[egg]

[egg]

[D]

[egg]

[δ]

[D]

[egg]

[D]

D

cor

L

stv

eggδ

H2O/2

SiO2 Al2O3/2

eggstv D

δ

δ

δ

δ

δ

coreggDδ

δ cor(D)

L

egg corD

stv

Dcor

L

stv

cor(egg)

L

DeggL

stv Degg L

egg corL

stv

Dstv L

eggL

stv

egg

stvδ

cor

corstv

δD

Figure 1 | Schreinemakers analysis of phase relations in the systemAl2O3–SiO2–H2O. Based on experiments performed between 22 and26 GPa. Larger ternary diagrams show phase assemblages observed in thisstudy at 26 GPa. Smaller ternary diagrams show results from previousstudies or are deduced from known phase relations. Phase abbreviationsare as follows: δ, δ-AlOOH; stv, stishovite; D, Al-phase D; egg, phase Egg;cor, corundum; L, hydrous liquid/melt.

The chemical composition of Al-phase D was found to varywith bulk composition and with temperature. The Al2O3-richcomposition (mixture 1) produced Al-phase D with an atomic Al/Siratio of approximately 2.5:1. The SiO2-rich composition (mixture 2)produced Al-phase D with an Al/Si ratio of approximately 1.5:1buffered by the presence of stishovite. All EPMA analyses ofAl-phase D yielded totals significantly below 100%. These largedeficiencies and their systematic variation with temperature implythat they can be used to provide approximate estimates of H2Ocontents. In the more SiO2-rich composition the resulting chemicalformula is Al1.54Si0.98O6H3.5 at 1,460 ◦C and Al1.73Si1.2O6H2 at2,100 ◦C, corresponding to a decrease in H2O content from 18to 10wt%. Raman analyses of Al-phase D synthesized at 2,100 ◦C(Supplementary Fig. 1) qualitatively confirm a significant hydroxylconcentration in the phase. The wavenumber of the hydroxyl peakindicates strong hydrogen bonding20.

A petrogenetic grid of the Al2O3–SiO2–H2O system at highpressure is presented in Fig. 1. Constraints on pseudo-univariantreactions are provided by the EBSD- and TEM-determined exper-imental phase assemblages, and by the results of previous stud-ies13,15,21. Experimental run product assemblages are indicated byshaded regions in the sketched ternary diagrams. The temperature-dependent mineral compositions are based on EPMA analyses. Thepressure–temperature (P–T ) stabilities and relative positions of eachreaction in Fig. 1 are derived by applying Schreinemakers rulesto the available experimental constraints. A degree of uncertaintyarises from an incomplete knowledge of phase composition. Thisis especially true of stishovite, phase Egg and δ-AlOOH, which arenearly collinear in composition space. In addition, a degree of solidsolution in each phase means that, in reality, each reaction will bemultivariant and consequently of finite width. Despite these caveats,Schreinemakers analysis provides good constraints on phase stabil-ity, and allows some extrapolation to regions where experimentaldata are lacking.

The Si- andAl-rich startingmixtures both produced assemblagesof δ-AlOOH and stishovite at temperatures at or below 1,200 ◦C.δ-AlOOH is inferred to be stable to 1,460 ◦C, and found to containup to 17wt% SiO2 in the more Al-rich composition. As Si-freeδ-AlOOH decomposes at 1,200 ◦C at 26GPa (ref. 21), we infer thatthe addition of a SiO2 component increases its thermal stability.

A three-phase assemblage of stishovite, δ-AlOOH and phase Eggcrystallizes from the Si-rich mixture up to 1,460 ◦C. The absenceof Al-phase D can be explained by its higher H2O content thaneither the bulk composition or δ-AlOOH at these conditions. Twofurther experiments employing a stoichiometric phase Egg bulkcomposition confirmed phase Egg stability at 26GPa and at 1,200 ◦Cand 1,460 ◦C, which is approximately 5GPa higher than previousdeterminations15. It is probable that the discrepancy in reportedstability fields reflects the near collinear compositions of stishovite,δ-AlOOH and phase Egg; the solid solutions of all three phasesmay cause the temperature of the reaction phase Egg→ stishovite+δ-AlOOH to be strongly dependent on bulk H2O concentration.

Al-phase D appears in assemblages produced from both startingmixtures after the breakdown of δ-AlOOH. Up to 1,600 ◦C,Al-phase D coexists with stishovite and phase Egg in the Si-richstarting mixture, after which phase Egg breaks down. The moreAl-rich mixture seems to fall within the compositional field ofAl-phase D, although in some cases minor stishovite was alsoproduced. It is assumed that Al-phase D is stable within the P–Tstability field of δ-AlOOH at 26GPa. The lack of Al-phase D in runsconducted within this field may reflect minor water loss from thecapsules during the experiments. The alternative, that decreasingtemperature results in Al-phase D breaking down to an assemblageof solid phases and H2O-rich liquid, is extremely unlikely.

Topological analysis indicates the presence of an invariant pointin the system located approximately at 1,800 ◦C and 24GPa ([δ] inFig. 1). This invariant marks the pressure where Al-phase D takesover from phase Egg as the most thermally stable hydrous phasein the system. The highest temperature experiment was performedat 2,100 ◦C where Al-phase D coexisted with stishovite and analumina-rich melt (Fig. 2). Although thermal gradients inside thistype of multianvil assembly are expected to be about 200 ◦Cmm−1,the small distance between the sample and the thermocouple(<0.5mm) ensures that Al-phase D must be stable to temperaturesexceeding 2,000 ◦C. This is one of the highest known thermalstabilities of any hydroxide, being rivalled only by Mg-bearingδ-AlOOH, as revealed in recent experiments14. The thermal stabilityof Al-phase D probably reaches a maximum at a pressure above26GPa, and at even higher pressures Si-bearing δ-AlOOH mayreplace phase D as the most thermally stable hydrous phase in theAl2O3–SiO2–H2O system.

The relationship between phase D structure and stabilityThe decomposition temperature of Al-phase D at 26GPa is atleast 800 ◦C above that for Mg-phase D. The results of single-crystal structural refinements (details reported as SupplementaryInformation) can illuminate the reasons why the replacement ofMg and Si with Al has such a remarkable effect on the stability ofthis hydrous structure. The crystal structure of Mg-phase D (spacegroup no. 162; P 3̄1m; ref. 22) is based on a hexagonal close-packedarray of oxygen atoms. The SiO6 and MgO6 octahedra occur intwo separated layers stacked along the c direction, with Mg andSi in the 1a and 2d Wyckoff positions respectively (M1 and M2;Supplementary Fig. 4). Three further octahedral sites correspondingto the 2c and 1b Wyckoff positions (M3 and M4) remain vacant.However, when Al replaces Mg in Al-phase D (space group no. 193;P63/mcm), all six octahedral sites become partially occupied bya random and disordered distribution of Si and Al. M1 and M4become equivalent, as do M2 and M3, resulting in an increase insymmetry. One of the main differences between the structures is

76 NATURE GEOSCIENCE | VOL 8 | JANUARY 2015 | www.nature.com/naturegeoscience

© 2014 Macmillan Publishers Limited. All rights reserved

NATURE GEOSCIENCE DOI: 10.1038/NGEO2306 ARTICLESa

b

20 µm

JEOL COMP 15.0 kV ×110 100 µm WD11 mm

Figure 2 | Scanning electron micrographs of recovered phase D-bearingsamples. a, Backscattered electron image of experiment S5113 recoveredfrom 26 GPa and 2,100 ◦C. Dark quenched melt can be seen in the lowerpart of the Pt capsule, overlain by a layer of stishovite. The bulk of theassemblage comprises grains of Al-phase (dark) and smaller grains ofstishovite (light). b, A backscattered image of a sample recovered from25 GPa and 1,500 ◦C (S4253) in the system H–Fe–Mg–Al–Si–O2. Fe- andAl-bearing bridgmanite (light grey) coexists with grains of Al-phase D(darker grey). Pockets of hydrous melt can be seen at grain boundaries.

that Al–Si disorder in Al-phase D results in essentially undistortedoctahedra of similar size, whereas in Mg-phase D large Mg–Odistances cause the octahedra to be strongly distorted. The lesserextent of octahedral distortion is likely to stabilize Al-phase Drelative to its Mg-bearing counterpart.

Another factor influencing the dehydration temperatures ofphaseD is the strength of theOHbond.Oneway to compare theOHbond strength in each structure is through Pauling bond-strengthsums23. These can be calculated by assuming that the hydrogenposition and site occupancy in Al-phase D are identical to thosedetermined for Mg-phase D through neutron diffraction measure-ments24. Owing to the difference in cation distribution and disorderbetween the two structures the protonatedO site inMg-phase D hasan effective Pauling bond strength of +1.67 compared with +1.42for Al-phase D. This implies that the O site in Al-phase D is moreunderbonded than inMg-phaseD, resulting in a strongerO–Hbondand potentially strengthening hydrogen bonds with adjacent O sitesof the same type.

An important factor in the charge distribution in Al-phase Ddescribed above is the large degree of Al–Si disorder. Althoughthe contribution to the configurational entropy due to Al–Sidisorder has a stabilizing influence on low-pressureminerals such as

feldspar, the preference of Al for octahedral rather than tetrahedralcoordination at pressures of a few gigapascals means that this effectplays only a minor role in much of the upper mantle and transitionzone. Such disorder seems to become important again in the lowermantle, however, where Si andAl both exist exclusively in octahedralcoordination. Al-phase D may be the first member of a new class ofcompletely disordered hydrous aluminosilicates that consequentlyhave high thermal stabilities. The high thermal stabilities recentlyreported for the solid solution between phase H and δ-AlOOHabove 40GPa (ref. 14) may also indicate Al–Si disorder.

The composition of phase D in the lower mantlePhase D contains roughly equal atomic proportions of Fe and Alwhen it coexists with Fe, Al-bearing bridgmanite with compositionssimilar to those existing in lower-mantle ultramafic rocks12,25,26.Fe cancels out the stabilizing effect of Al (ref. 12), such thatphase D in ultramafic compositions is likely to break down at sim-ilar temperatures to the Mg-phase D endmember. In other words,phase D in ultramafic rocks is only likely to be stable within sub-ducting slabs. To investigate the partitioning behaviour in maficcompositions, we performed an additional experiment20 (S4253)at 25GPa and ∼1,500 ◦C using a hydrated bulk composition fab-ricated from 13.6 wt% Al2O3, 21.6% Fe2O3, 33% SiO2 and 31.8%Mg(OH)2. Recovered bridgmanite (Fig. 2b) has the approximatechemical formula Mg0.63Fe0.37Al0.37Si0.63O3, similar to compositionsexpected within subducted basalts27. Coexisting Al-phase D hasthe approximate composition Mg0.24Fe0.16Al1.83H1.5SiO6. The Al/Feratio in this phase D is much higher than that observed in Fe,Al-poor compositions (Fig. 3). We suggest that in hydrous lower-mantle ultramafic rocks, phase D and bridgmanite accommodateFe3+ and Al3+ in roughly equal proportions. As Al contents increase,the strong preference for the coupled substitution (Fe3+Al3+) ↔(Mg2+Si4+) in bridgmanite results in highAl/Fe ratios in phaseD. Asincreasing Al/Fe in phase D expands its thermal stability, subductedcrustal rocks with high Al contents could even host Al-rich phase Dat temperatures well above typical lower-mantle isentropes.

Deep geochemical processing in the mantleThe remarkable stability of Al-phase D has major implications forthe hydrogen budget of the lower mantle. Within subducting slabsin the deep upper mantle and transition zone, it is believed thatAl-poor ultramafic lithologies host most subducted water in theform of hydrous phases including phases A, superhydrous B, D andbrucite8,28. Owing to the low thermal stability of Al-poor hydrousphases and the low H2O solubility of nominally anhydrous mineralsin the lower mantle1,8,9, ultramafic rocks descending through theupper–lower-mantle boundary will become supersaturated withH2O, releasing hydrous melts as they heat up. The resulting meltswill migrate through ultramafic rocks but will form Al-phase Dwithin Al-rich mafic crustal rocks, as shown in Fig. 4. Subductedmafic units have been proposed to be present throughoutthe mantle18,19, owing to the extreme timescales required forhomogenization through chemical diffusion29. Rocks containingAl-rich phase D in the lower mantle would still be able to host∼1,000wt ppm H2O in majoritic garnet and clinopyroxene afterupwelling into the upper mantle1,30.

The process described above implies that H2O in the lowermantle will become preferentially concentrated in mafic rocks. Oneprediction of this premise is that magmas produced from lower-mantle sources that exhibit geochemical evidence for the presenceof recycled material should also be more H2O-rich. Although thereseems to be a general agreement that OIB mantle sources containmore H2O than the depleted mantle, there is some question asto whether this arises from the recycling of hydrated lithosphereor reflects H2O in a primordial source. Several studies havedemonstrated a negative correlation between H2O concentrations

NATURE GEOSCIENCE | VOL 8 | JANUARY 2015 | www.nature.com/naturegeoscience 77

© 2014 Macmillan Publishers Limited. All rights reserved

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2306

MO

RB

Perid

otite

0.1

1.0

10.0

Al/

Feph

ase

D

0.0 0.1 0.2 0.3 0.4

Al3+ + Fe3+pv

Mg-rich phase D

Al-rich phase D

Ref. 12Ref. 26

This study

Figure 3 | Al/Fe ratio in phase D as a function of bridgmanite composition.Cation concentrations based on EPMA reveal that Al/Fe ratios in phase Dincrease sharply as a function of increasing trivalent cation content inbridgmanite. Fe and Al in bridgmanite are reported on a one-cation basis,and all Fe is assumed to be ferric in both phases. Representativebridgmanite compositions in peridotitic25 and mid-ocean-ridge basalt27

(MORB) bulk compositions are shown as grey bars.

atg

A

D + br/shB

law

Transition zone

Upper mantle

∼410 km

∼670 kmLower mantle

Melt

∼1,200 °C∼1,6

00 °C

Al-poor D

Al-rich D

Figure 4 | Potential mechanisms of deep hydrogen transport betweenhydrous phases and melts in a subduction zone. The convecting mantle isshown as a heterogeneous mixture of mantle and recycled crust. Thecrustal layer on top of the subducting slab (dark grey) becomes nominallyanhydrous at 300 km depth, after the breakdown of lawsonite (law).Hydrated lithospheric mantle can carry water to greater depths, buthydrous phases become unstable at high temperatures, releasing melts. Inthe lower mantle, recycled oceanic crust can become rehydrated by thesemelts, forming Al-rich phase D that remains stable even at hightemperatures. Mineral abbreviations are as follows: atg, antigorite; A, phaseA; D, phase D; br, brucite; shB, superhydrous phase B; NAMs, nominallyanhydrous minerals.

of OIB mantle and enriched mantle components (for example,EM1) defined, for example, by extremely radiogenic Sr isotopicratios and considered to result from subducted sediments6,31.This has been interpreted to indicate that the lithosphere isefficiently dehydrated during the subduction process6. However,subducted sediments are not representative of subducted oceanic

crust. Instead, recycled oceanic crust has been linked to a ‘focuszone’ or FOZO component with moderately radiogenic Pb and Srratios (206Pb/204Pb ∼ 20, 87Sr/86Sr ∼ 0.703; ref. 32). H2O/Ce valuesincrease from EM1-influenced basalts towards those with FOZO-like isotopic characteristics6, implying high H2O concentrationsin the FOZO source. The progressive hydration of recycled maficcrust in the lower mantle due to crystallization of Al-phase D,and potentially other hydrous aluminosilicate phases stable at evenhigher pressures9,14, would explain the relationship between higherH2O contents and the FOZOmantle component without needing toinvoke the presence of a wet primordial lower-mantle source.

MethodsStarting compositions were based on analyses of crystals of Al-phase D previouslyreported20, which demonstrate a compositional range in the Al/Si ratio. Twostarting mixtures were fabricated from SiO2, Al2O3 and Al(OH)3 with Al/Si ratiosof 2:1 and 1:1 and H2O contents of 13 and 19wt% respectively (SupplementaryTable 1). In two further experiments a stoichiometric phase Egg composition(AlSiO3OH) was employed. These powders were sealed inside 1-mm-diameterwelded Pt capsules. In most experiments two capsules, each with a differentcomposition, were placed symmetrically either side of a horizontally insertedthermocouple within the multianvil assembly. Multianvil experiments wereperformed at pressures between 22 and 26GPa and at temperatures between1,050 and 2,100 ◦C. Further experimental details are reported as SupplementaryMethods. Recovered samples were analysed using powder XRD, EPMA, TEM andEBSD. A single crystal of Al-phase D with dimensions of 30 × 15 × 15 µm wasrecovered from a sample synthesized at 1,460 ◦C for data collection usingsingle-crystal XRD.

The Supplementary Information includes a full description of the structuralrefinement, further experimental details and analytical information. Runconditions and averaged EPMA analyses are also provided as a separate Excel file.

Received 13 June 2014; accepted 4 November 2014;published online 8 December 2014

References1. Bolfan-Casanova, N. Water in the Earth’s mantle.Mineral. Mag. 69,

229–257 (2005).2. Ardia, P., Hirschmann, M. M., Withers, A. C. & Tenner, T. J. H2O storage

capacity of olivine at 5–8GPa and consequences for dehydrationpartial melting of the upper mantle. Earth Planet. Sci. Lett. 345,104–116 (2012).

3. Pearson, D. G. et al.Hydrous mantle transition zone indicated by ring-wooditeincluded within diamond. Nature 507, 221–224 (2014).

4. Bolfan-Casanova, N., Keppler, H. & Rubie, D. C. Water partitioning betweennominally anhydrous minerals in the MgO–SiO2–H2O system up to 24GPa:Implications for the distribution of water in the Earth’s mantle. Earth Planet.Sci. Lett. 182, 209–221 (2000).

5. Bolfan-Casanova, N., Keppler, H. & Rubie, D. C. Water partitioning at 660 kmdepth and evidence for very low water solubility in magnesium silicateperovskite. Geophys. Res. Lett. 30, 1905 (2003).

6. Dixon, J. E., Leist, L., Langmuir, C. & Schilling, J-G. Recycled dehydratedlithosphere observed in plume-influenced mid-ocean-ridge basalt. Nature 420,385–389 (2002).

7. Saal, A. E., Hauri, E. H., Langmuir, C. H. & Perfit, M. R. Vapourundersaturation in primitive mid-ocean-ridge basalt and the volatile content ofEarth’s upper mantle. Nature 419, 451–455 (2002).

8. Ohtani, E. Water in the mantle. Elements 1, 25–30 (2005).9. Nishi, M. et al. Stability of hydrous silicate at high pressures and water

transport to the deep lower mantle. Nature Geosci. 7, 224–227 (2014).10. Ringwood, A. E. & Major, A. High-pressure reconnaissance investigations in

the system Mg2SiO4–MgO–H2O. Earth Planet. Sci. Lett. 2, 130–133 (1967).11. Liu, L-g. Effects of H2O on the phase behaviour of the forsterite–enstatite

system at high pressures and temperatures and implications for the Earth. Phys.Earth Planet. Inter. 49, 142–167 (1987).

12. Ghosh, S. & Schmidt, M. W. Melting of phase D in the lower mantle andimplications for recycling and storage of H2O in the deep mantle. Geochim.Cosmochim. Acta 145, 72–88 (2014).

13. Ohtani, E., Litasov, K., Suzuki, A. & Kondo, T. Stability field of new hydrousphase, δ-AlOOH, with implications for water transport into the deep mantle.Geophys. Res. Lett. 28, 3991–3993 (2001).

14. Ohira, I. et al. Stability of a hydrous δ-phase, AlOOH–MgSiO2(OH)2, and amechanism for water transport into the base of lower mantle. Earth Planet. Sci.Lett. 401, 12–17 (2014).

78 NATURE GEOSCIENCE | VOL 8 | JANUARY 2015 | www.nature.com/naturegeoscience

© 2014 Macmillan Publishers Limited. All rights reserved

NATURE GEOSCIENCE DOI: 10.1038/NGEO2306 ARTICLES15. Ono, S. High temperature stability limit of phase egg, AlSiO3(OH). Contrib.

Mineral. Petrol. 137, 83–89 (1999).16. Wirth, R., Vollmer, C., Brenker, F., Matsyuk, S. & Kaminsky, F. Inclusions of

nanocrystalline hydrous aluminium silicate ‘‘Phase Egg’’ in superdeepdiamonds from Juina (Mato Grosso State, Brazil). Earth Planet. Sci. Lett. 259,384–399 (2007).

17. Poli, S. & Schmidt, M. W. Petrology of subducted slabs. Annu. Rev. EarthPlanet. Sci. 30, 207–235 (2002).

18. Helffrich, G. R. & Wood, B. J. The Earth’s mantle. Nature 412, 501–507 (2001).19. Hirschmann, M. M. & Stolper, E. M. A possible role for garnet pyroxenite in

the origin of the ‘‘garnet signature’’ in MORB. Contrib. Mineral. Petrol. 124,185–208 (1996).

20. Boffa Ballaran, T., Frost, D. J., Miyajima, N. & Heidelbach, F. The structure of asuper-aluminous version of the dense hydrous-magnesium silicate phase D.Am. Mineral. 95, 1113–1116 (2010).

21. Sano, A., Ohtani, E., Kubo, T. & Funakoshi, K-i. In situ X-ray observation ofdecomposition of hydrous aluminum silicate AlSiO3OH and aluminum oxidehydroxide δ-AlOOH at high pressure and temperature. J. Phys. Chem. Solids65, 1547–1554 (2004).

22. Yang, H., Prewitt, C. T. & Frost, D. J. Crystal structure of the dense hydrousmagnesium silicate, phase D. Am. Mineral. 82, 651–654 (1997).

23. Smyth, J. R. β-Mg2SiO4: A potential host for water in the mantle? Am. Mineral.72, 1051–1055 (1987).

24. Suzuki, A. et al. Neutron diffraction study of hydrous phase G: Hydrogen inthe lower mantle hydrous silicate, phase G. Geophys. Res. Lett. 28,3987–3990 (2001).

25. Ishii, T., Kojitani, H. & Akaogi, M. Post-spinel transitions in pyrolite andMg2SiO4 and akimotoite-perovskite transition in MgSiO3: Precise comparisonby high-pressure high-temperature experiments with multi-sample celltechnique. Earth Planet. Sci. Lett. 309, 185–197 (2011).

26. Litasov, K. D., Ohtani, E., Suzuki, A. & Funakoshi, K. The compressibility ofFe- and Al-bearing phase D to 30GPa. Phys. Chem. Mineral. 34,159–167 (2007).

27. Hirose, K. & Fei, Y. Subsolidus and melting phase relations of basalticcomposition in the uppermost lower mantle. Geochim. Cosmochim. Acta 66,2099–2108 (2002).

28. Hacker, B. R. H2O subduction beyond arcs. Geochem. Geophys. Geosyst. 9,Q03001 (2008).

29. Holzapfel, C., Rubie, D. C., Frost, D. J. & Langenhorst, F. Fe–Mg interdiffusionin (Mg,Fe)SiO3 perovskite and lower mantle reequilibration. Science 309,1707–1710 (2005).

30. Hauri, E. H., Gaetani, G. A. & Green, T. H. Partitioning of water during meltingof the Earth’s upper mantle at H2O-undersaturated conditions. Earth Planet.Sci. Lett. 248, 715–734 (2006).

31. Workman, R. K., Hauri, E., Hart, S. R., Wang, J. & Blusztajn, J. Volatile andtrace elements in basaltic glasses from Samoa: Implications for waterdistribution in the mantle. Earth Planet. Sci. Lett. 241, 932–951 (2006).

32. Stracke, A., Hofmann, A. W. & Hart, S. R. FOZO, HIMU, and the rest of themantle zoo. Geochem. Geophys. Geosyst. 6,Q05007 (2005).

AcknowledgementsThe authors would like to thank G. Gollner, H. Fischer, S. Übelhack, G. Manthilake,H. Schulze, U. Dittman and D. Krauße. This work was funded through the support ofEuropean Research Council (ERC) Advanced Grant ‘DEEP’ (#227893). R.M. is supportedby an Alexander von Humboldt Postdoctoral Fellowship.

Author contributionsT.B.B. and D.J.F. designed the study; M.G.P. performed the experiments and processedthe analytical data with assistance from F.H. (EBSD) and N.M. (TEM); T.B.B. performedthe structural refinement; R.M., T.B.B. and D.J.F. interpreted the analytical data; R.M.,M.G.P. and D.J.F. wrote the paper. All the authors discussed the results and implicationsand commented on the manuscript at all stages.

Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints.Correspondence and requests for materials should be addressed to R.M.

Competing financial interestsThe authors declare no competing financial interests.

NATURE GEOSCIENCE | VOL 8 | JANUARY 2015 | www.nature.com/naturegeoscience 79

© 2014 Macmillan Publishers Limited. All rights reserved