Evaluating the importance of metamorphism in the foundering of continental crust

The metamorphic conditions and mechanisms required to induce foundering in deep arc crust are assessed using an example of representative lower crust in SW New Zealand. Composite plutons of Cretaceous monzodiorite and gabbro were emplaced at ~1.2 and 1.8 GPa are parts of the Western Fiordland Orthogneiss (WFO); examples of the plutons are tectonically juxtaposed along a structure that excised ~25 km of crust. The 1.8 GPa Breaksea Orthogneiss includes suitably dense minor components (e.g. eclogite) capable of foundering at peak conditions. As the eclogite facies boundary has a positive dP/dT, cooling from supra-solidus conditions (T > 950 ºC) at high-P should be accompanied by omphacite and garnet growth. However, a high monzodioritic proportion and inefficient metamorphism in the Breaksea Orthogneiss resulted in its positive buoyancy and preservation. Metamorphic inefficiency and compositional relationships in the 1.2 GPa Malaspina Pluton meant it was never likely to have developed densities sufficiently high to founder. These relationships suggest that the deep arc crust must have primarily involved significant igneous accumulation of garnet–clinopyroxene (in proportions >75%). Crustal dismemberment with or without the development of extensional shear zones is proposed to have induced foundering of excised cumulate material at P > 1.2 GPa.

The foundering and removal of dense continental crust 1 is part of the gross-scale geochemical and geodynamic cycling in magmatic arcs 2,3 . Foundering is predicted to occur in most arc systems 4 , though the short-lived nature of root detachment greatly restricts its direct observation via geophysical methods 2,5 . The re-incorporation of mafic crust into the mantle by foundering is interpreted to progressively contribute to refining the average composition of the continental crust to andesite 2,4 . Periodic crustal thickening, root removal and rebound during long-lived subduction can influence plate-scale cycles, including lithospheric contraction and extension 6 . Systematic overturn of the arc system through foundering can also stimulate magmatic flare-up 7 , linked to the melting of descending crust and the mantle 3,4 .
Foundering is a consequence of a gravitational instability when the lower crust develops a density greater than that of the underlying mantle (~3.33 gcm −3 ) [8][9][10][11] . The rheology of the crust and mantle are also important 8 . Typically, foundering of the base of the continent is considered to reflect either (1) Rayleigh-Taylor instabilities ("drip") or (2) delamination, whereby modelling predicts that the base of the lithosphere peels away and sinks 2,12 . The exact mechanics of foundering in compositional stratified lower crust are not generally considered or known. The initial densification can occur by two main mechanisms. Igneous fractionation, involving the production of dense cumulate material, is thought to occur in most magmatic arcs 2,4 . Metamorphism can also increase the density of lower arc crust through: (i) its conversion to eclogite 9,11 ; (ii) anatectic melt-extraction and the production of restite 13 ; or (iii) large-scale metasomatic alteration 14 . The scarcity of deep arc exposures limits our understanding of the relative importance of these mechanisms.
We present a case study of well-exposed deep arc crust to evaluate the efficiency of closed-system metamorphism to produce crust that is denser than the upper mantle. Eclogite, with or without high-P granulite, is predicted to be a consequence of crustal thickening, dependent on lithospheric heat flow 15,16 and composition. The conversion of crustal rocks to eclogite has been well studied through direct experimentation 15 and the investigation of natural examples 9,17 . These studies assume the lower crust is primarily of basic (gabbroic) composition, but
The Breaksea Orthogneiss records incomplete metamorphism and patchy deformation (D 1 ) that occured during cooling at high-pressure conditions (1.8-2.0 GPa) subsequent to its plutonic emplacement 21 . Post-D 1 decompression to P ≈ 1.0-1.4 GPa and T ≈ 650-750 °C is recorded by diopside-albite symplectite 21 that partially to completely pseudomorphs omphacite and its inferred to correspond to a stage of gneiss dome formation 35,36 . Upper amphibolite facies assemblages developed locally in D 2 shear zones that are thought to be related to orogenic collapse 29,35 . This work explores outcrops at Breaksea Tops away from the effects of the major D 2 shear zones (Fig. 1a). In accordance with the persistence of protolith relicts in the orthogneiss, terminology specific to both igneous and metamorphic stages is used to clarify inferred processes. Previous work. Basic and monzodioritic, now eclogite and granulite, gneiss of the Breaksea Orthogneiss preserve gradational contacts in outcrop and linear first-order trends in Harker plots of whole-rock composition (Fig. 1c) 21 . Variations in the proportions of garnet, clinopyroxene, orthopyroxene and plagioclase in the orthogneiss are mostly attributed to cumulate processes and magma redox conditions that preceded high-grade D 1 deformation 23,[34][35][36] . Interlayered near-monomineralic garnetite and clinopyroxenite retain delicate microstructure and mineral chemistry consistent with an igneous origin 23 . Garnet in basite components can also interpreted to be of igneous origin because its major and rare earth element (REE) characteristics are similar throughout the spectrum of ultrabasic to intermediate compositions and chondrite-normalised REE patterns lack positive Eu anomalies 23 . As garnet and omphacite collectively form 90% of basite layers, this is also then consistent with most omphacite in the eclogite layers being of cumulus origin 23 .
Similar igneous mineralogy is observed in monzodioritic gneiss that is interlayered with the basite, garnetite and clinopyroxenite; mineral REE concentrations overlap among all rock types. Igneous and metamorphic garnet are distinguished in the monzodioritic gneiss by microstructural and chemical features: (1) large euhedral garnet grains in clusters with omphacite retain heavy-REE-enriched patterns that overlap with those of igneous garnet in eclogite and garnetite; and (2) idioblastic garnet grains forming coronae that separate omphacite and plagioclase are heavy-REE depleted and have positive Eu anomalies consistent with metamorphic growth during consumption of plagioclase. In places, foliated (S 1 ) assemblages of omphacite, garnet, plagioclase, kyanite and rutile are consistent with metamorphic equilibration at conditions of the omphacite granulite sub-facies 34 . The nature of the S 1 fabrics varies between exposures in the mouth of Breaksea Sound and those at Breaksea Tops 33 . Outcrop relationships at Breaksea Sound preserve S 1 with both vertical and near horizontal orientations that define concentric, decimetre-scale gneiss domes spatially related to large extensional D 2 shear zones 35,36 . These dome structures are not observed at Breaksea Tops. In addition, the fabric trajectories are oblique to the unit boundaries, inconsistent with gross-scale diapiric flow (Fig. 1a).

Results
Field Relationships. Garnet, clinopyroxene and plagioclase, with or without orthopyroxene, K-feldspar, quartz and kyanite, form monzodioritic gneiss that makes up much of the Breaksea Orthogneiss (Fig. 1b). There are both gradational and sharp contacts between layers and cumulate pods of basite, garnetite and clinopyroxenite. An igneous layering at Breaksea Tops is locally cut by a moderately dipping (~65°), north-striking gneissic foliation (S 1 ) with an associated L 1 mineral stretching lineation plunging towards the southeast (Fig. 1a). Foliation trajectories are consistent with only local deviations in strain shadows to basite pods (Fig. 1a). Whole rock major and trace element compositions throughout the gneiss are remarkably consistent, defining a restricted range in silica and total alkali content between monzogabbro and monzodiorite (SiO 2 = 48-55 wt.% and Na 2 O + K 2 O = 4-7 wt.%) (Fig. 1c) 23 . The bulk composition of the gneisses resembles those predicted for the lower crust (Table 1) 18,19 . Petrography. Mineral assemblages are subtly distinct across the compositional spectrum of the Breaksea Orthogneiss, although textural relationships remain largely similar. Monzodioritic gneiss comprises varying proportions of plagioclase, omphacite, orthopyroxene, igneous (Type 1) and metamorphic (Type 2) garnet, rutile, K-feldspar, quartz, kyanite, ulvöspinel, hornblende and apatite. Large (300-1000 μm) Type 1 garnet and omphacite grains form mm-scale grain clusters in monzodioritic gneiss, that are variably elongated (Fig. 2b). Grain cores of Type 1 garnet have rutile exsolution lamellae and euhedral inclusions of feldspar and omphacite. The facetted feldspar inclusions are generally antiperthite or comprise composite inclusions of plagioclase surrounded by narrow K-feldspar rims. Exsolution-free rims to the Type 1 garnet cores have rounded omphacite and plagioclase inclusions. Omphacite may locally preserve euhedral shapes (Fig. 2b) and have facetted plagioclase inclusions and/or fine oxide exsolution lamellae in grain cores. Some portions of the gneiss contain coarse orthopyroxene (800-1000 μm) intergrown with omphacite and ulvöspinel, with only minor Type 1 garnet. In sites of higher strain, Type 2 garnet coronae separate elongated omphacite from plagioclase. Quartz, K-feldspar and rutile are intergrown with Type 2 garnet, commonly as vermicular inclusions (Fig. 2c). Type 1 garnet is absent or occurs in low modes in areas of Type 2 garnet abundance. The coarse-grained feldspar-rich matrix enclosing the garnetomphacite clusters is dominated by plagioclase, with minor K-feldspar, quartz, kyanite, garnet and omphacite (Fig. 2b). K-feldspar occurs interstitial to plagioclase with minor quartz. Kyanite occurs as tabulate and acicular grains, aligned within S 1 in the feldspar-rich matrix of high-strain samples.

P-T path and Mineral Equilibria
Modelling. Pressure-temperature pseudosections (P-T) are presented ( Fig. 3) for monzodioritic and gabbroic compositions observed in the Breaksea Orthogneiss (Fig. 1c). The pseudosections indicate the dependence of mineral assemblage on changes in pressure, temperature and composition 37,38 . The modelled conditions encompass the granulite -eclogite transition for gabbroic and monzodioritic protoliths ( Table 1). The upper pressure stability of plagioclase marks the lower boundary of the eclogite facies ( Fig. 3a,b). The main effect of lowering the bulk-rock silica content from a monzodiorite to a basite protolith is that the eclogite facies boundary is encountered at lower pressure conditions (1.9 cf. 1.8 GPa: Fig. 3a,b) 33 . Gabbroic compositions have less feldspar (plagioclase and K-feldspar) and quartz, and higher modes of garnet and hornblende at the expense of omphacite when contrasted with monzodioritic compositions at the eclogite facies boundary (Fig. 3).
Mineral assemblage and texture define two high-P stages in the monzodioritic gneiss of the Breaksea Orthogneiss. Granoblastic, HREE-enriched Type 1 garnet intergrown with coarse omphacite is interpreted as igneous (Fig. 2b) 23 . Hypersolvus feldspar and high-Zr rutile inclusions in Type 1 garnet reflect T > 950 °C (Table 2 and Fig. S1). The high-Ti content, and rutile exsolution in Type 1 garnet potentially extends these estimates to above 1000 °C 39 . Part of the igneous equilibria are likely accounted for by the quinivariant field (L-o-g-q-pl-ru) for monzodioritic compositions (Fig. 3a). The igneous assemblages are partially recrystallized by metamorphic reaction textures involving Type 2 garnet intergrown with jadeitic omphacite, albitic plagioclase, quartz and rutile (Fig. 2c). Metamorphic mineral textures are consistent with recrystallization both accompanying the development of S 1 and post-dating it. Zirconium-in-rutile thermometry from Type 2 grains within necklace textures suggests growth at T ≈ 850 °C, and match other mineral exchange estimates for the peak metamorphic conditions ( Table 2) 21 . The quadrivariant field (o-g-ksp-pl-ru-q-ky-mu) best accounts for the mineral modes and S 1 textures (Fig. 3a) 33 . Feldspar and garnet exsolution textures and mode changes, including a reduction in anorthitic plagioclase and the growth of jadeitic omphacite, in the monzodioritic gneiss are best explained by a high-P cooling history. Key high-variance changes induced by isobaric cooling at 1.8 GPa mostly involve the breakdown of anorthitic plagioclase and growth of omphacite. Garnet mode initially increases with cooling to ~900 °C, before decreasing with further cooling to accommodate Fe-Mg exchange related to omphacite mode increases at lower T. Equivalent changes could be induced by a PT path involving pressure increasing from ~1.5 to 2.0 GPa, but such a path is predicted to be associated with larger changes in the type and composition of the Ca-Na clinopyroxene than are observed. An equivalent isobaric cooling PT path at lower pressure conditions (<1.5 GPa) in a composite pluton of gabbroic or monzodioritic composition for the Malaspina Pluton would encompass an equivalent series of reactions involving diopside instead of omphacite and higher plagioclase modes (Fig. 3) 24 .
Rock densities calculated from mineral equilibria modelling of the granulite -eclogite transition have trends dependent on both composition and PT conditions (Fig. 3a,b). In general, higher pressure conditions induce higher densities in both of the compositions modelled, but the most pronounced changes are associated with the destabilisation of plagioclase at the lower boundary of the eclogite facies. As key mineral assemblages associated with this change have boundaries with high dP/dT values, density should also increase during recrystallization on a PT path involving cooling (e.g. 950 to 850 °C) at constant pressure (Fig. 3a,b), assuming that the rocks fully    (Fig. 3). Assuming a density of 3.33 gcm −3 for the upper mantle, monzodioritic gneiss will be capable of foundering at a pressure 0.15-0.2 GPa above the granulite -eclogite transition, whereas basic gneiss may founder at ~0.5 GPa below the transition (Fig. 3). Conditions critical to foundering are thus not strictly dependent on the eclogite facies boundary or garnet production. The higher pressure and lower temperature changes in monzodioritic gneiss density are mostly associated with K-feldspar stability and omphacite mode (Fig. 3a). In basic gneiss, the higher garnet and omphacite modes give higher rock densities despite the presence of plagioclase and melt stability (below the eclogite facies, Fig. 3b).

Discussion
The veracity and scale of processes plausibly resulting in, and derivative from a gravitational instability remain controversial, and are complicated by the continental crust being heterogeneous 2,3 . Using the case study of the Zealandia magmatic arc, we resolve the blend of intrinsic and extrinsic parameters that have the potential to facilitate the foundering of the lower continental crust, and contrast this detail with macroscopic features presented by the case study.
How to founder the lower crust. To induce foundering of the lower crust, geological processes need to form rock of density higher than the majority of the continental crust and the upper mantle 8 . The key minerals required to produce such high densities in common crustal rocks are garnet (>3.6 gcm −3 ) and omphacite (~3.16-3.4 gcm −3 ), their co-occurrence characterising the eclogite facies 15 . The high-variance reactions delineating the granulite -eclogite transition are controlled by plagioclase breakdown 15,33 , which will be accompanied by a density step in the profile of basic or felsic lower crust. However, at high-T conditions much of the change is controlled by high variance pressure-dependent reactions in granulite: the progressive consumption of the albite component in plagioclase facilitates the growth of omphacite via solid solution, involving the substitution of the Ca-Tschermak's and jadeite molecule 15 . The substitution is coupled with the consumption of grossular-rich garnet through Fe-Mg exchange with omphacite 15 . However, the positive dP/dT of the relevant high-variance reactions means that near isobaric cooling can also lead to eclogite formation (Fig. 3). Metamorphism of the Breaksea Orthogneiss incompletely records cooling from high-T conditions associated with igneous crystallization, metamorphic minerals being distinguished by sodic plagioclase, and marked changes in the jadeite content of clinopyroxene and grossular content of Type 2 garnet. Igneous garnet, clinopyroxene and hypersolvus feldspar record high igneous crystallization temperatures (>950 °C) in grains that were patchily metamorphosed and deformed at lower-temperature conditions (~850 °C). The patchy metamorphism is marked by the development of comparatively dense metamorphic mineral assemblages (Figs 3 and 4). Similar, though less pronounced densification is predicted for igneous rocks cooling at lower pressure conditions, in the Malaspina Pluton (Fig. 3) 24 .
The eclogite facies boundary is commonly considered a minimum depth to form rock of density adequate to induce crustal foundering 15,18 . However, the upper pressure limit of plagioclase is sensitive to changes in bulk-rock composition and is predicted to occur at lower pressure conditions in more basic compositions (Fig. 3) 15,33 . In addition, rocks that recrystallized in the eclogite facies need not have densities exceeding that of the upper mantle. The densities of high-P intermediate and basic gneiss are mostly dependant on omphacite and garnet mode. The proportions of both these minerals are in turn dependent on pressure and temperature conditions, as can be shown by comparing the different PT conditions required for gabbroic and monzodioritic protoliths to have densities above that of the upper mantle (3.33 gcm −3 ). Metamorphosed gabbroic protoliths are predicted to have densities above 3.33 gcm −3 in plagioclase-bearing equilibria, even at suprasolidus (melt-present) conditions (Fig. 3b). In contrast, metamorphosed monzodioritic protoliths are predicted to have densities less than 3.33 gcm −3 until the progressive breakdown of K-feldspar at T ≈ 750 °C, some 0.1-0.2 GPa beyond the granulite -eclogite transition (Fig. 3a). Mineral mode changes modelled across the transitions are marked mostly by plagioclase consumption and omphacite growth. Garnet may be either a product or a reactant (Fig. 3). The densification of intermediate to basic rocks is therefore not as dependant on garnet growth as commonly considered 15,18 . Basic compositions have greater capacity to develop higher garnet modes that attribute higher densities.
Mineral assemblage changes associated with a given rock moving from the granulite to eclogite facies do not always run to completion 38 . The metastable persistence of porphyroclastic or phenocrystal material at elevated metamorphic temperature conditions (e.g. >750 °C) is not generally explicitly considered but seems quite common 10,17,20 . For this case study, the efficiency of metamorphism in the Breaksea Orthogneiss would seem to have been limited by a comparatively short timescale (c. 10 Myr), heterogeneity in strain and restricted fluid/melt availability 23,30,31,34 . The high proportions of relict minerals would have enhanced buoyancy of the lower arc crust, and reduced its chance of foundering (Fig. 4).
The average silica content of the lower continental crust is generally thought to be above 50 wt.% (Table 1) 18,19 presenting difficulties for closed-system metamorphism to induce a gravitational instability at common crustal thicknesses. Mechanisms that will be more effectual in producing dense lower crust include: (i) igneous accumulation; (ii) partial melt extraction; and (iii) large-scale metasomatism. Varying combinations of basic cumulate and intermediate rocky types are observed in most deep arc systems 2,18 , their formation related to arc melting. The presence of garnet-clinopyroxene basite in both the Breaksea Orthogneiss 23 and Malaspina Pluton highlights the potential for igneous cumulates to densify the deep crust in arc systems, all cumulate compositions having densities higher than mantle peridotite (Fig. 4). However, the potential for the Fiordland basite to founder is dependent on its proportion relative to the interlayered, buoyant monzodioritic gneiss (Fig. 4). The proportion of garnet-clinopyroxene cumulate in the Breaksea Orthogneiss would need to exceed 60% to reach densities appropriate for its foundering (Fig. 4). In thinner arc systems (~40 km), the proportions of cumulate would potentially need to be even greater (above 75%) on account of higher plagioclase mode (Fig. 4) 24 .
Foundering mechanisms: crustal dismemberment. The continental crust in most of Earth's magmatic arcs is approximately 40 km thick (~1.2 GPa at its base) 40 , with Cordillera Zealandia having been at least 65 km thick (~1.8 GPa) and probably thicker 21 . Primitive crustal material similar to, or more basic than, the Breaksea Orthogneiss is inferred to underlie Fiordland based on seismic velocity models 41 . This would extend arc crust to depths greater than 80 km 7,32 . The presence of such material underneath the Breaksea Orthogneiss is supported by its depleted Sr/Y and REE ratios and Hafnium isotopic signatures 7,24 . Few examples of such thick arc crust are evident in the geological record, plausibly because of root detachment 40,42 . Rock densities of both igneous and metamorphic stages in the Breaksea Orthogneiss are too low to potentially induce foundering, mostly due to its composition. It seems reasonable to assume that the overall composition of the seismically-imaged material is similar, as it also did not founder 41 . The foundering of any monzodiorite-dominated crust thus seems negligible, but a significant proportion of the Cretaceous arc crust between 1.2 and 1.8 GPa has not be identified and is missing. In addition, a large proportion of residual material (~25% by volume) complementary to felsic high Sr/Y batholiths that accumulated between 1.2 and 1.0 GPa is not recognized in the Fiordland section 24 . Theas features are consistent with material having been lost from the Fiordland arc, but perhaps not from the level represented by the Breaksea Orthogneiss (1.8 GPa).
Two main mechanisms can be envisaged for density sorting above the arc base: (1) diapirism and counter-flow related to the concentric dome fabrics present in the WFO 36 ; or (2) crustal excision along extension shear zones (Fig. 5). Density sorting via counter-flow diapirism is difficult to reconcile in Fiordland on account of: (i) the 100 m to km-scale of the domes 36 being 1 to 2 orders of magnitude too small to have controlled appreciable foundering; (ii) the patterns are developed in material that remained positively buoyant (even for eclogite); (iii) the domes in Breaksea Orthogneiss at Breaksea Sound likely formed after its decompression to ~1.2 GPa (albite-diopside symplectites) 21 and are spatially linked to the RISZ (Fig. 1a); (iv) gross scale counter-flow pathways are not observed at the deeper levels recorded in the Breaksea Orthogneiss at Breaksea tops (Fig. 5). It is possible that the dome patterns reflect a larger process that completely removed dense material elsewhere 36 , but that then begs the question of how to make the dense material. At the conditions attributed to dome formation (~1.2 GPa), even pure basite would need to cool and recrystallize below T ≈ 600 °C to be negatively buoyant relatively to the upper mantle (Fig. 3b) and there would need to be a lot more of it than is observed.
Igneous processes are the most viable mechanism to produce a crust with a high-density layer of cumulate material. The composition of WFO plutons that accumulated at P < 1.8 GPa indicate that significant crystal fractionation or crustal mixing must have occurred, leaving a dense residue 24,30,32 . The large volume of HREE-depleted WFO magma compositions that accumulated at depths between 1.0 and 1.2 GPa, together with observed garnetclinopyroxene cumulates, supports the concept of a cumulate pile having underlain this part of the arc crust 22,24 . We posit that the missing section of the crust comprised up to 75% garnet-clinopyroxene basite to account for the high Sr/Y ratios in the Malaspina Pluton 2,24 .
The posited cumulate material would have overlain more buoyant material (Breaksea Orthogneiss), creating a crustal sequence that is distinct to most generalized predictions of arc lithosphere 18,19,42 . As diapiric counter-flow would seem to have been incapable of density sorting in the Cretaceous Zealandia arc, it is proposed that crustal excision was a driver. Large-scale shear zones present an ideal mechanism to facilitate the excision of dense material from the crust and the RISZ juxtaposes 1.2 and 1.8 GPa components presented by the Malaspina Pluton and the Breaksea Orthogneiss 35 . Extensive conjugate shear zones record c. 111 Ma displacements of up to 9-25 km 29,36,43 that are localized within, and adjacent to, WFO plutons 22 . Dense cumulate portions in the section between 40 and 65 km (Fig. 4), could have feasibly migrated both laterally and vertically along active shear zones during the period of fault activity 31,36 . The hanging wall material would have been suitably located to subsequently founder into the mantle 12 . The architecture could also possibly induce the return flow of depleted mantle material, to account for harzburgite xenoliths in the Breaksea Orthogneiss 21,44 . In addition to contribution to decimetre-scale dome perturbations in the WFO 36 . If such a crustal dismemberment mechanism eventuated, the Zealandia system involved foundering during extension at the end-stages of a magmatic flare-up cycle 3,6 . Extension could therefore be a viable period for crustal destruction and the initiator of a new Cordillera tectonic cycle 6 .

Methods
Crushed whole-rock powders were analysed by X-ray fluorescence (XRF) spectrometry on a PANalytical PW2400 at the University of New South Wales for major element compositions. Protolith classification (Fig. 1c) followed Middlemost 45 . Mineral chemistry was determined on a CAMEBAX SX100 electron microprobe (Macquarie Geoanalytical) with a 15 kV accelerating voltage and a beam current of 20 nA. Zr-in-rutile analyses were performed on a Agilent 7700cs Inductively Coupled Plasma Mass Spectrometer (ICPMS), with attached 213 nm Nd:Yag laser ablation microprobe (LAM). Operating conditions involved a c. 60 s background period count prior to laser ablation, and a c. 100-120 s analysis using a 50 μm beam diameter and 5 Hz pulse repetition rate. The analysis of NIST 610 glass during each session provided an external standard and standard reference material BCR2, was analysed as an internal standard. Precision was 1% relative standard deviation, based on 1σ distribution on Zr.
Phase equilibria modelling was performed in the NCKFMASHTO chemical system (Na 2 O-CaO-K 2 O-FeO-MgO-Al 2 O 3 -SiO 2 -H 2 O-TiO 2 -O) using THERMOCALC version 3.45i 37 and the internally consistent thermodynamic dataset 6.2 (updated 6 th February 2012) 46 . Mineral activity-composition models and abbreviations used include tonalite melt (L), hornblende (hb), omphacite/diopside (o/dio) 47 , feldspars (pl & kfs) 48 , garnet (g), muscovite (mu) 49 and ilmenite (ilm) 50 . Pure phases include rutile (ru), quartz (q), kyanite (ky) and H 2 O. The modelled bulk rock compositions are based on XRF analysis of a monzodiorite (1203 C) and a non-cumulate gabbro (P73824 22 : Fig. 1c). The modelled redox conditions (Fe 3+ /[Fe 3+ /Fe 2+ ] = 0.18) are generally considered appropriate for arc magmas 51 . Modelled water content (1 wt.%) is based on the observed modal proportions of hydrous phases (hornblende). The mineral chemistry is model dependent and shows partial differences to the observed compositions, although all changes match relative variations. Pressure uncertainties for assemblage field boundaries are approximately ± 0.1 GPa at the 2σ level 52 . The use of sodic-calcic pyroxene at suprasolidus conditions produced  43 ). Possible foundering mechanisms involving diapiric counter-flow 36 or crustal excision (this study). limited difference in field boundaries compared to the calibrated augite model 47 , on account of the low melt modes. It was utilised due to the importance of Na-Ca exchange.
Zirconium-in-rutile thermometry were calculated with the calibration of Tomkins et al. 53 for inclusions in Type 1 garnet and intergrowths with Type 2 garnet (Table 2). Rutile is intergrown with quartz and is assumed to have grown in equilibrium with zircon. In the absence of zircon, the temperatures are considered as minimum estimates 53 . The precision on the estimates are 3-4%, encompassing the analytical uncertainty (~1% RSD) and a ±3% uncertainty in the calibration 53 . Ternary feldspar thermometry was applied to inclusions in Type 1 garnet, using calibrations at 0.8 54 and 2.0 GPa 55 (Fig. S1). The reintegration of mineral compositions was calculated based on proportions established from BSE images (ImageJ) and electron microprobe analysis. The temperature dependence of analysed Ti concentrations in Type 1 garnet provides an additional thermometer 39 . The thermometer was calibrated with and without orthopyroxene and therefore only provides approximate temperatures that match calibration curves. Neglecting exsolved Ti in rutile needles results in an underestimation of the garnet T.
Rock samples were measured for dry density using a Mettler Toledo AG-204 scale with density determination kit. Conservative uncertainties of 0.5% on the rock measurements are based on the variability of observed mineral modal abundance. Densities of modelled mineral assemblages were calculated internally within the THERMOCALC software, using the calcsv command 37 . Uncertainty on the calculations are likely to be <1% at the 2σ level 52 . Measured sample densities correlated with predicted values suggesting that the effects of expansion or contraction at P-T were within error.