Introduction

The buoyancy of subducted lithosphere is a principle determinant in its descent from, and/or return to Earth’s surface, where less or more buoyant than the enclosing mantle, respectively1,2. Rock density varies in relation to progressive mineralogical change during metamorphism, and is a function of the ambient geotherm, depth and rate of subduction. A crucial density threshold is generated by metamorphic changes at eclogite facies conditions (depths ≈ 60–70 km and P ≈ 1.8–2.2 GPa), which presents as a point of no-return for mafic crust (Fig. 1)3,4. The final destination for most subducted material is shown by seismic tomography to be either the mantle transition zone or the core–mantle boundary5. However, the presence of mafic eclogite at Earth’s surface indicates that dense material can return from high-pressure conditions (>1.8 GPa). Most instances of rocks returning from conditions above the quartz–coesite transition (>2.4 GPa), referred to as ultra-high-pressure (UHP), involve predominantly silicic continental lithosphere, which remains positively buoyant to greater depths2,6.

Figure 1
figure 1

Pressure–temperature (P–T) pseudosection for a mid-ocean ridge basalt (MORB), displaying the metamorphic facies for this protolith and the PT boundaries of key minerals (inset). Different shading intensity of the fields represents changes in variance of the phase assemblage, the red shading represents fields denser than the ambient mantle. Warm and cool slab-parallel geotherms are displayed for reference (dashed red lines). Isopycnals for the metamorphosed MORB are shown in dashed black lines, with assemblage fields above the density threshold delineated in red (3.33 gcm−3). Labelled boxes represent the P–T conditions inferred for natural eclogite terranes, which correlate well with the predicted mineral assemblages (Table S2).

Full size image

The preservation of coesite, diamond and majoritic garnet defining UHP terranes are consistent with their return from depths of 80–160 km (P = 2.4–5 GPa)7,8,9,10. However, interpretations of pseudomorphs after stishovite, exsolved components in garnet, clinopyroxene and olivine or α-PbO2-type TiO2 with and without majoritic garnet, have led to inferences of burial depths approaching the mantle transition zone (400 km: P up to 12–14 GPa) for mixed mafic and felsic terranes11,12,13,14,15,16,17,18. The interpreted exhumation of rocks from these depths is problematic in the context of return-limits established through experimentation (~250 km)19,20,21. Such circumstances require the action of an attached and denser “anchor” to firstly reach such conditions, and then another process to restore positive buoyancy and enable their return to Earth’s surface. The proportion of felsic to mafic material in subducted lithosphere is thus crucial to buoyancy relations at various depths, and presents a test for the validity of the UHP record, yet it is seldom quantified22. Forward predictions of phase equilibria23 for lithosphere comprising varying proportions of lherzolitic mantle and crust of either oceanic or continental composition can establish the pressure–temperature (PT) conditions under which the subducted components will become neutrally buoyant in the upper mantle24,25,26. Oceanic crust is represented here by mid-ocean ridge basalt (MORB), and the more diverse continental crust is explored using averaged andesitic and granitic compositions27. Crucial preservation thresholds – points of no-return – and their determinative buoyancy agents are established by considering the gross changes involved in differing lithospheric components and validated using detail from the natural record. The predicted preservation thresholds reconcile the range of PT conditions commonly recorded by eclogite facies terranes, and enable the veracity of UHP estimates to be queried. The modelled outcomes highlight the importance of identifying robust petrological evidence in buoyant rock types that can be used to corroborate PT conditions inferred from elemental exchange equilibria, such as the former stability of wadeite, K-cymrite or K-hollandite, where rocks have been returned from below the predicted preservation thresholds.

Phase Equilibria Modelling

Pressure–temperature (P–T) pseudosections28,29 for MORB (Fig. 1), andesite (Fig. 2), granite (Fig. 3) and lherzolite (Fig. S1) represent the metamorphism of oceanic and continental lithosphere during subduction. The simple lithospheric models used here assume 7 km of oceanic and 40 km of continental crust with 93 and 110 km of lithospheric mantle, respectively (Fig. 4). The oceanic scenario represents an example of 100 Myr old subducted lithosphere30. Continental crust has more variability in its composition and water content, which can be expected to induce limited departures in its overall density (~0.05 gcm−3) compared to that of the average andesite model (see Supplement Fig. S4). Scenarios involving differing proportions of crust (mafic, andesitic and granitic) and lithospheric mantle (lherzolite and/or serpentinite) are assessed to establish buoyancy constraints on the lithosphere reaching and returning from extreme depths (Fig. 4).

Figure 2
figure 2

P–T pseudosection for a metamorphosed andesite, displaying the metamorphic facies for this protolith and the PT boundaries of key minerals (inset). Different shading intensity of the fields represents changes in variance of the phase assemblage. Warm and cool slab-parallel geotherms are displayed for reference (red dashed lines). Isopycnals for the metamorphosed andesite are displayed in dashed black lines with the labelled boxes representing the P–T conditions inferred for natural eclogite terranes (Table S2). No red shading is present as all phase assemblages density remain less than the ambient mantle.

Full size image
Figure 3
figure 3

P–T pseudosections for metamorphosed dry (a) and wet granite (b), and MORB (c) at upper mantle conditions (P = 4–15 GPa). Different shading intensity of the fields represents changes in variance of the phase assemblage. Warm and cool slab-parallel geotherms are displayed for reference (red dashed lines). Isopycnals are displayed in dashed black lines and labelled boxes represent P–T conditions inferred for natural UHP terranes (Table S2). Assemblage fields that are denser than the ambient mantle are delineated in red and those with similar density to the ambient mantle in orange.

Full size image
Figure 4
figure 4

Predicted density of phase equilibria in oceanic and continental lithosphere (a) and crust (b,c) during cool and warm subduction. The effects of differing proportions of crustal components is shown in (b,c) at various depths. The PT boundaries of key minerals are shown for reference, including the lower limit of UHP conditions (quartz–coesite transition). A conjugate diagram illustrating the dependence of density on the lithospheric components for the warm and cool geotherms is shown in Fig. S4.

Full size image

Ambient upper mantle densities established along an isentropic potential temperature of 1350 °C present a threshold critical to evaluating the buoyancy of subducted lithosphere and its potential return to the surface (Fig. S1). The uniformity of the upper mantle under typical conditions limits variation in its density from 3.33 gcm−3 at low-P (<2 GPa) equivalent to depths of less than 65 km3. However, the density of lherzolite is strongly influenced by the proportion of hydrous minerals (e.g. in serpentinised components); their breakdown is coupled with large changes in density (Fig. S1).

High-pressure conditions induce dense assemblages in crustal rock types (Figs. 1 and 2). The capacity for oceanic crust to exceed the upper mantle density is not dependent solely on it reaching eclogite facies conditions. It is, instead, controlled mostly by high-variance equilibria associated with the breakdown of chlorite and amphibole, and the growth of omphacite and garnet (Fig. 1). These reactions induce densification above 3.33 gcm−3 at depths dependent on the ambient geotherm: oceanic crust experiencing warm subduction will be neutrally buoyant at approximately 65 km (2 GPa) versus 80 km (2.5 GPa) in scenarios of cool subduction (Fig. 1). Continental crust of andesitic composition can be expected to increase density at a lower rate than MORB crust, as garnet and omphacite modes will be lower (Fig. 2). Steps in the density of such silicic and aluminous material are influenced strongly by kyanite mode and the quartz to coesite transition, with values being less than 3.33 gcm−3 across the modelled range (1–4 GPa; Fig. 2).

The calculation of NCKFMASHTO mineral assemblage densities at P > 3.5 GPa is limited by our current understanding of activity–composition models at UHP conditions28,29. However, the higher-pressure density of felsic and mafic crustal end-members (i.e. granitic and basaltic compositions) can be assessed in the reduced NC(K)FMAS(H) system against that of a lherzolitic upper mantle (Figs. 3 and S1)29. Mafic crust can be expected to metamorphose to high-variance, dense equilibria. The stabilisation of stishovite and the dissolution of omphacite into majoritic garnet are predicted to induce densities in basaltic crust that are greater than that of the surrounding lherzolitic mantle (Fig. 3c). Dry granitic crust is predicted to undergo polymorphic transformations of K-feldspar to wadeite at P = 5.5–6.5 GPa or depths of 200 to 220 km, and wadeite to K-hollandite near the stishovite transition (280 km: Fig. 3a). For fluid-saturated granitic crust, wadeite is preferentially destabilised by phengite and K-cymrite with or without kyanite depending on temperature (Fig. 3b). The growth of K-hollandite at the expense of K-cymrite occurs at higher-P (9–12 GPa) equivalent to burial depths of 280–360 km. The stabilisation of wadeite in dry granitic crust induces assemblage densities close to that of the ambient mantle (3.43 gcm−3), that are overall denser than in hydrous assemblages stable at otherwise equivalent conditions (3.20–3.30 gcm−3), consistent with previous experimental predictions (Fig. 4c)3,19,20,21.

Points of No-Return

The slab-pull force exerted by subducting lithosphere is mostly controlled by the density of its mantle component1,4,31. This is, in turn, principally dependent on the ambient geotherm, as the isopycnals (line of equal density) in lherzolite are controlled by its thermal expansivity (Figs. 3 and S1). Oceanic lithosphere is predicted to be denser than 3.33 gcm−3 at much shallower conditions (60 km and P = 1.2–1.4 GPa) compared to continental lithosphere (85 km and P > 2.6 GPa), on account of higher modes of garnet, omphacite and amphibole in its crustal component (Figs. 1 and 2). At depths below these buoyancy thresholds, much of the lithosphere could be expected to be exhumed or involved in a circumstance of stalled subduction, as the pull-force will be terminated or greatly diminished (Fig. 5)32,33,34. These depth estimates for the points of no return will increase if a reduced proportion of the dense lithospheric mantle is subducted due to tectonic scenarios involving lithospheric attenuation or dismemberment (Fig. 4a)2,6.

Figure 5
figure 5

Schematic representation of the predicted depth-limits for the buoyant exhumation of subducted oceanic (a) and continental (b) crust.

Full size image

The majority of exhumed eclogite terranes comprise oceanic and continental crust or hydrated ultramafic material of density less than the upper mantle. Mafic eclogite mostly records peak conditions that correspond with the predicted density thresholds, which are in turn controlled by the subduction geotherm (60 and 80 km or P > 1.7 and 3.0 GPa; Fig. 1). Variability in the P–T conditions experienced by exposed mafic eclogite terranes has been attributed to the extent of mechanical coupling between the crust and the mantle wedge35,36. These are also influenced by the occurrence of serpentinite, which is buoyant (2.95–3.15 gcm−3) and prone to being intensely deformed during metamorphism4,37,38,39. The maximum depths from which mafic terranes may be exhumed are predicted to cluster at 30 and 80 km depending on interface-decoupling of the subduction channel from the overriding lithosphere, and at 55–60 km due to the effects of serpentinite dehydration (Fig. 1)35,36.

The bulk of the continental crust is predicted to remain positively buoyant at UHP conditions shallower than 120 km, beyond which its density progressively approaches 3.33 gcm−3 (P > 3.5 GPa: Fig. 2)26. Granitic crust is predicted to have a maximum burial depth of between 160 and 260 km (Figs. 3 and 4c)19,20,21, dependent on its water content. The presence of hydrous minerals (mafic or felsic) will affect rock density and prograde mineral progression, with dehydration reactions inducing large volume reductions at high-PT conditions following the consumption of water as a separate phase. Dry granitic crust will develop wadeite-bearing mineral assemblages and densities similar to that of the enclosing mantle at depths of 160–170 km (P = 5.5 GPa). The stabilisation of K-cymrite and phengite instead of wadeite in hydrous granitic crust is predicted to result in densities for such compositions remaining less than that of the ambient mantle to depths of 260 km, before marked densification associated with the growth of stishovite and K-hollandite (P = 8–10 GPa: Figs. 3b and 4c)25.

Mineral assemblages associated with UHP conditions general contain coesite, stishovite or majoritic garnet, and mostly occur in mafic or ultramafic protoliths hosted by comparatively buoyant felsic rock26,40,41. The co-occurrence of mafic or ultramafic rock with intermediate or felsic rock types is necessary to generate slab-pull forces that can drag the rocks to upper mantle conditions, before or after the detachment of any crust from its lithospheric mantle (Fig. 4). Such pull forces become increasingly important beyond burial to 80–90 km (2.5–2.8 GPa) due to a decrease in the rate of densification associated with thermal expansion (Fig. 4). For crust dominated by andesite to remain buoyant at UHP conditions (say 120 km or ~3.5 GPa), it need comprise no more than 25% mafic material (Fig. 4). These predictions can account for the occurrence of coesite, rare microdiamond and/or majoritic garnet from depths of up to 160 km in fragmented lithosphere (Figs. 2 and 5)26,40 that had limited basaltic material (<25%)7,8,9, as in several of the Himalayan UHP terranes31. Exhuming terranes with proportional more mafic or ultramafic material requires they be dominated by felsic material6,26,42. The subduction and exhumation of crust dominated by granite requires that there be less than 40% accompanying mafic component, for the crust to be positive buoyancy at depths between 160 and 260 km. Higher proportions of dense rock types will render mixed terranes susceptible to comparatively slow and/or staged exhumation42: they could, for example, initially lose a dense anchor at UHP conditions and be partial uplifted, to then have a second stage of mafic/ultramafic loss whilst stalled at the base (30–60 km or 0.5–1.8 GPa) of the overriding lithosphere (e.g. Dabie–Sulu, the Western Gneiss Terrane & Bohemian Massif42,43,44).

Comparatively few UHP occurrences are inferred to have returned from depths beyond 160 km (P > 5 GPa), but burial to depths of 360–400 km (P ≈ 12 GPa) have been attributed to rocks with pseudomorphous textures posited to be after stishovite, α-PbO2-type TiO2 and/or the presence of exsolution textures in majoritic garnet and omphacite11,12,13,14,15,16,17,18. The exhumation of eclogite from depths of 200–260 km is possible for granitic protoliths, particularly for those with hydrous high-grade assemblages (Fig. 4). However, the interpreted return of eclogite from depths of 300–400 km draws attention to the need of a plausible tectonic process (e.g. Altyn Tagh, Ezgebirge massif or Kokchetav massif)12,13,14,17,45. Mechanisms controlling the re-exposure of terranes with high proportions of mafic/ultramafic material from depths of 300–400 km is even more cryptic (e.g. Alpe Armi or Dabieshan)11,16.

Returning From Ultra-Deep Burial

Caution seems warranted in attributing UHP assemblages to depths greater than 160 km (P > 5 GPa), in the absence of mechanisms that facilitated their positive buoyancy and uplift2,6,9,26. Alternative mechanisms that have been proposed as influential to the buoyant return of UHP terranes to Earth’s surface include their partial melting45,46 and/or the metastable persistence of lower-P phases47. Most UHP eclogite is hosted by comparatively buoyant felsic rock that is commonly retrogressed, and/or apparently does not record passage through UHP conditions, making it difficult to interpret their high-P history26,40,41. Relatively small disparities in burial P estimates (<0.5 GPa) can be attributed to tectonic overpressure, typically in localised high strain domains41,48.

At depths greater than 160 km, high ambient temperature conditions will reduce, but not eliminate the effects of kinetic impediments to metamorphic equilibration49,50,51. Generally, incomplete metamorphism is attributed to a lack of a free-fluid or hydroxyl-bearing phases that aid diffusion-controlled reaction kinetics49,50,51. Paradoxically, exhuming felsic crust from depths within the stishovite or wadeite eclogite facies (160–260 km) seems more feasible for compositions with hydrous mineral assemblages (Fig. 3). The inefficient metamorphism of dry granitic crust could result in either the complete metastable persistence of coesite and K-feldspar, or their partial transformation to wadeite or stishovite (<50% and <20%, respectively). Such inefficiency could result in crustal densities (3.33–3.43 gcm−3) close to the buoyancy limit at depths of 260 km. However, the catalysing influence of H2O makes reaction overstepping on depth-scales greater than 100 km seem less plausible for most subduction systems, particularly if hydrous and anhydrous rocks are intermingled51. The mineral assemblages observed in natural eclogite terranes reconcile well with predictions from fluid-saturated phase equilibria modelling, despite considerable variability in rock compositions and inferred equilibrium volumes (Table S2)52,53. It seems common that progressive metamorphic equilibration is achieved across a large range of P–T conditions50,54. It is also likely that fluid-saturated conditions were experienced by many exhumed UHP terranes, especially mafic compositions. Therefore, the subduction of wadeite or kyanite eclogite to depths of 350–400 km seems unlikely to escape localised metamorphic reaction and densification, and/or involve extensive metastable persistence47. Robust evidence for the subduction and exhumation of granitic crust from depths of 350–400 km might come by establishing pseudomorphs after K-cymrite, wadeite or K-hollandite as well as stishovite, and/or armoured inclusions and exsolution features that are linked to mineral exchange trends13,14,21,55.

Methods

Phase equilibria modelling was performed using THERMOCALC14. Equilibria models for a MORB, serpentinite (Fig. S1) and an andesite were calculated in the NCKFMASHTO chemical system (Na2O–CaO–K2O–FeO–MgO–Al2O3–SiO2–H2O–TiO2–O) utilising version 3.45i23 and the internally consistent thermodynamic dataset 6.2 (updated 6th February 2012)28. Lherzolite (Fig. S1) was modelled in the NCKFMASTOCr chemical system (Na2O–CaO–K2O–FeO–MgO–Al2O3–SiO2–TiO2–O–Cr2O3) for P = 1.0–3.5 GPa utilising version 3.47 of THERMOCALC and the internally consistent thermodynamic dataset 6.33 (updated 23rd June 2017)56. Ultra-high-P conditions (4–15 GPa) in granite, MORB and lherzolite were modelled in the reduced NCKFMAS(H) (Na2O–CaO–K2O–FeO–MgO–Al2O3–SiO2–H2O) and NCFMAS(H) systems with the internally consistent dataset 6.2 (updated 6th February 2012)28,29.

Mineral activity–composition models and abbreviations used for the NCKFMASHTO models include: glaucophane (gl), actinolite (act), hornblende (hb), omphacite/diopside (o/dio)57, feldspars (pl & kfs)58, garnet (g), paragonite (pa), biotite (bi), muscovite (mu), chlorite (chl)59, epidote (cz & zo), chloritoid (ctd), staurolite (st), talc (ta), olivine (ol)28, brucite (br) and antigorite (atg)60, with pure phases of lawsonite (law), albite (ab), rutile (ru), sphene (sph) quartz (q), coesite (coe), sillimanite/kyanite (sill/ky) and H2O. Mineral activity–composition models and abbreviations used for the NCKFMASTOCr system are: clinopyroxene (cpx), garnet (g), orthopyroxene (opx), spinel (sp), olivine (ol) and dry silicate liquid (liq)56. Mineral activity–composition models and abbreviations used in the NC(K)FMAS(H)28,29: orthopyroxene (opx), high-P clinoenstatite (hpx), clinopyroxene (o and jd), olivine (ol), wadsleyite (wad), majoritic garnet (g), corundum (cor), muscovite (mu), K-feldspar (ksp), together with pure phases coesite (coe), stishovite (stv), lawsonite (law), kyanite (ky), wadeite (wa) and K-hollandite (hol) and K-cymrite (kcm).

Pressure uncertainties for the assemblage field boundaries are approximately ±0.1 GPa at the 2σ level61, although the additional uncertainty on activity composition models will subtly increase this error. The densities of the modelled mineral assemblages were calculated within the THERMOCALC software, using the calcsv command23. Density estimates were corrected for free water using its defined Tait equation of state within the activity–composition relations of the thermodynamic dataset28,62. Non-linearity in the equation of state for H2O necessitated the calculation of its density variability at specific imposed conditions in relation to its modal proportions. The relative proportion of free water was determined by setting H2O at molar proportions that just saturated the low-temperature and high-pressure equilibria38 (Table S1). The equated density of all free water was subtracted from its thermodynamically equilibrated solid phase assemblage established in THERMOCALC. Uncertainty on the density calculations are likely to be <1% at the 2σ level25,61. Depth to pressure conversion was based on the PREM relationship63.

The modelled bulk rock compositions are based on a MORB-type eclogite from New Caledonia38, which shows limited effects of alteration across a large PT range64, an averaged andesitic composition of the continental crust27, an averaged of the upper continental crust of granitic composition19 and the KLB-1 lherzolite65. The modelled redox conditions for the MORB and the andesite were fixed at Fe3+/[Fe3+ + Fe2+] = 0.07–0.15, and the dry and hydrated lherzolite at 0.03, being considered appropriate for the crust and upper mantle, respectively56,66. Fluid was considered to be in excess in the MORB, serpentinite, andesite and granite models.

Two geothermal gradients for common subduction zones were used to interrogate the densities of oceanic and continental lithologies (Fig. S4). The cool geotherm is based on the Nankia subduction zone67, whereas the warm geotherm is based on the field array preserved on the Pam Peninsula, New Caledonia38, which approximates a geotherm that falls in between those interpreted for the modern Chilean and Cascadia subduction zones68. Both are consistent with exposed eclogite P–T estimates35,69.

Estimates of the pressure–temperature (P–T) conditions experienced by exposed high-pressure and UHP eclogite terranes worldwide were compiled and contrasted with conditions that could be predicted for their assemblages. The full compilation of eclogite terranes including mineral assemblages and published geothermobarometry results are shown in Table S2. As an additional comparison on the validity of PT and density estimates, the observed mineral assemblages were correlated with assemblages predicted by phase equilibria modelling (Table S2). Collectively, the compilation suggests that the phase equilibria calculations are generally robust for the gross considerations of the constituents of lithospheric buoyancy.