Poly-cyclic Metamorphic Evolution of Eclogite: Evidence for Multistage Burial–Exhumation Cycling in a Subduction Channel

Abstract

(Ultra-)high-pressure [(U)HP] eclogites provide direct evidence for metamorphic processes at mantle depths in subduction zones. Textural evidence and pressure–temperature ( P – T ) path reconstruction allow a unique opportunity to constrain the geodynamic processes along the plate interface. To better understand the dynamics of subduction channels, we present a detailed petrographical, petrological and pseudosection modeling study of a metamorphically complex HP eclogite sample from a tectonic block of the Akeyazi (U)HP metamorphic terrane in the Chinese Tianshan. Mineral chemical variations and textural discontinuities show that all the major phases of the rock (garnet, epidote, clinopyroxene and amphibole) display multiple growth zones compatible with a poly-cyclic evolution including two burial–exhumation cycles. The precise P – T path derived from thermodynamic modeling based on effective bulk compositions suggests that the eclogite underwent complex burial–exhumation cycles with two P – T loops indicated by the occurrence of polyphase garnet and mutually replaced and regrown amphibole. It is suggested that the eclogite sample, which represents a relic of tectonically detached oceanic crust, underwent two burial–exhumation cycles during convective flow in a subduction channel. This natural occurrence is consistent with numerical simulation studies showing that pressure cycling is to be expected for deeply subducted oceanic crust in a subduction channel. The modeled exhumation process for the eclogite sample corresponds to a previously suggested subduction channel exhumation model for the Akeyazi (U)HP rocks.

INTRODUCTION

Subduction zones are vital elements of plate tectonics and play a key role in deciphering crustal evolution, volcanism and material recycling. High-pressure/low-temperature (HP/LT) blueschist- to eclogite-facies terranes are widely regarded to include exhumed fragments of subducted continental or oceanic crust. Therefore, metamorphic studies of eclogites may reveal important information about their P – T evolution and associated tectono-metamorphic processes at great depth in subduction zones (e.g. Ernst & Liou, 2008 ).

The results of thermodynamic modeling of subduction zone-derived metamorphic rocks often indicate a rather simple prograde P – T evolution from blueschist- to eclogite-facies conditions, followed by a rapid and thus often isothermal decompression (e.g. Agard et al. , 2009 ). The reconstruction of their P – T paths commonly is straightforward, as the rocks record a successive subduction and exhumation sequence and the development of mineral assemblages and rock textures is a response to a continuous P – T evolution (i.e. a single P – T loop). However, some subduction zone-derived high-pressure and low-temperature rocks reveal polymetamorphic characteristics, which complicate the reconstruction of P – T trajectories, as the observed mineral assemblages and rock textures are the result of several distinct tectono-metamorphic events or thermally or fluid-induced phases of reactivity (e.g. Austrheim, 1987 ; Ganne et al. , 2003 ; Ague & Baxter, 2007 ; Gaidies et al. , 2008 ; van der Straaten et al. , 2008 ; Kabir & Takasu, 2010 ; Herwartz et al. , 2011 ).

Burial–exhumation cycles may involve long-term episodic tectonic events (producing ‘polymetamorphism’) resulting from orogeny-scale shortening–extension switches ( Lister et al. , 2001 ; Brueckner, 2006 ), and also include short-lived low-amplitude pressure cycles in a single event (producing ‘poly-cyclic metamorphism’) owing to alternating shortening and extensional deformation ( Beltrando et al. , 2007 ; Harris et al. , 2007 ; Rubatto et al. , 2011 ) or alternatively owing to complex subduction channel dynamics ( Cloos & Shreve, 1988 ; Gerya et al. , 2002 ; Blanco-Quintero et al. , 2011 ; Malatesta et al. , 2012 ; Klemd et al. , 2015 ). Polymetamorphism and poly-cyclic metamorphism are characterized by multistage mineral assemblages and polyphase mineral growth, and both have been reported from some subduction zone-derived HP/LT metamorphic rocks (e.g. Austrheim, 1987 ; Cesare, 1999 ; Ganne et al. , 2003 ; Ye et al. , 2009 ; Herwartz et al. , 2011 ; Rubatto et al. , 2011 ; Chen et al. , 2013 a , 2013 b ; Regis et al. , 2014 ). This is of particular importance as poly-cyclic metamorphism may be a result of P – T cycles induced by multiple burial and exhumation cycles in a single subduction event, which differ from a single P – T loop commonly described for HP/LT rocks.

A subduction channel along the plate interface is believed to act as an effective exhumation pathway for high-pressure and ultrahigh-pressure (UHP) rocks ( Cloos, 1982 ; Cloos & Shreve, 1988 ; Gerya et al. , 2002 ; Warren et al. , 2008 ; Malatesta et al. , 2012 ). Numerical simulations of oceanic subduction zones suggest that fragments dismembered from the descending slab may undertake complex trajectories during their exhumation within the low-viscosity material within the subduction channel ( Gerya et al. , 2002 ; Krebs et al. , 2008 ; Blanco-Quintero et al. , 2011 ; Malatesta et al. , 2012 ). Accordingly, the P – T paths of high-pressure metamorphic rocks in subduction channels may be highly variable, and P – T trajectories with large-scale convection movements recording a complex, multiple burial–exhumation process are expected (e.g. Gerya et al. , 2002 ; Gerya & Stöckhert, 2006 ; Li & Gerya, 2009 ; Blanco-Quintero et al. , 2011 ; Krebs et al. , 2011 ).

The Akeyazi (U)HP metamorphic terrane is interpreted to be a large, well-exposed tectonic mélange, exhumed through subduction channel dynamics [see Klemd et al. (2015) for a recent review]. However, up to now no textural or petrological features supporting cyclic burial–exhumation processes have been reported for the (U)HP rocks of this terrane. This study is focused on a detailed P – T path reconstruction of a single, poly-cyclic eclogite sample from the Akeyazi (U)HP metamorphic terrane to constrain the record of multiple burial–exhumation processes in subduction zones. The results suggest that in general some eclogites may indeed experience complex burial–exhumation cycles in a channel-like structure along the plate interface during a single, continuous subduction event.

GEOLOGICAL SETTING

The Tianshan HP/LT belt extends east–west for about 1500 km from NW China, via Atabashi in Kyrgyzstan, to Fan–Karategin in Tajikistan ( Gao et al. , 1995 ; Tagiri et al. , 1995 ; Volkova & Budanov, 1999 ), separating the Yili–Kazakhstan–Kyzylkum and the Tarim–Karakum cratons ( Volkova & Budanov, 1999 ). It represents an oceanic subduction/collision zone in the South Tianshan Orogen (e.g. Gao et al. , 2009 , 2011 ; Hegner et al. , 2010 ; Klemd et al. , 2015 ). In the Chinese part of the Western Tianshan, a wedge-shaped (U)HP/LT belt, which is internationally known as the Akeyazi metamorphic terrane (AMT), extends for at least 200 km along the South Central Tianshan Suture and consists of a suite of metasedimentary, mafic and ultramafic rocks (Fig. 1a; Gao & Klemd, 2003 ). It is mainly composed of blueschist-, eclogite- and greenschist-facies metasedimentary rocks and some mafic metavolcanic rocks with normal and enriched mid-ocean ridge basalt (N-MORB and E-MORB), ocean island basalt (OIB) and arc basalt affinities ( Gao & Klemd, 2003 ; John et al. , 2008 ).

Eclogites occur within blueschists or mica schists as pods, boudins, thin layers or massive blocks and have been interpreted to represent a tectonic mélange ( Gao, 1997 ; Gao et al. , 1999 ; Gao & Klemd, 2003 ; Wei et al. , 2009 ). Most of the eclogites have experienced peak metamorphic conditions between 480 and 580 °C at 1·4–2·4 GPa on a regional scale (e.g. Klemd et al. , 2002 ; Wei et al. , 2003 ). Ultrahigh-pressure metamorphism was suggested either by the presence of coesite inclusions in garnet in eclogites and host mica schists ( Lü et al. , 2008 , 2009 ) or—in the absence of UHP minerals—by thermodynamic modeling ( Wei et al. , 2009 ; Tian & Wei, 2013 ). The discovery of coesite indicates that some of the eclogites and metapelites have indeed experienced UHP metamorphism in the AMT. In addition, the intimate interlayering of HP and UHP rocks ( Fig. 1b ) suggests that the rocks were derived from varying depths within the descending slab and then juxtaposed during exhumation by channel flow ( Klemd et al. , 2011 , 2015 ). Metamorphic studies of metapelites show that some mica schists have experienced UHP metamorphic conditions with or without coesite relics ( Lü et al. , 2008 , 2012 ; Yang et al. , 2013 ), whereas others have experienced only HP metamorphic conditions ( Wei & Powell, 2006 ; Wei et al. , 2009 ; Lü et al. , 2012 ; Li et al. , 2015 ).

Fig. 1

Geological map of the Akeyazi (U)HP/LT metamorphic terrane in northwestern China [modified after Li et al. (2013) ]. (a) Regional tectonic map. (b) Detailed geological map showing the area from which the eclogite sample was taken. The localities of HP and UHP rocks from previous studies are indicated (e.g. Klemd et al. , 2002 , 2011 ; Wei et al. , 2003 , 2009 ; Lin & Enami, 2006 ; Lü et al. , 2008 , 2009 , 2012 ; Li et al. , 2012 , 2013 ; Tian & Wei, 2013 ; Yang et al. , 2013 ). The red and white stars represent the localities of UHP rocks deduced by occurrence of coesite and thermodynamic modeling, respectively.

The age of peak metamorphism in the AMT belt is constrained by multi-point Lu–Hf isochron ages from four blueschist- and eclogite-facies rocks, which yield consistent garnet growth ages of c . 315 Ma ( Klemd et al. , 2011 ), confirming previous Carboniferous subduction and collision ages for this region ( Gao & Klemd, 2003 ). Uranium–Pb secondary ionization mass spectrometry (SIMS) ages of c . 320 Ma of metamorphic zircon rims (containing omphacite inclusions) from eclogites ( Su et al. , 2010 ) confirm the Lu–Hf garnet ages and are in agreement, within error, with an eclogite SIMS U–Pb rutile age of 318 Ma ( Li et al. , 2011 ), and a sensitive high-resolution ion microprobe (SHRIMP) zircon U–Pb age of 320 Ma obtained for a UHP metapelite ( Yang et al. , 2013 ). The timing of peak metamorphism appears to be associated with a major dehydration-related fluid release event that caused vein formation associated with reactive fluid flow in the high-pressure rocks at about 317 ± 5 Ma (e.g. John et al. , 2012 ). White mica Ar–Ar and Rb–Sr ages cluster at 311 Ma and have been interpreted to represent greenschist-facies isotopic re-equilibration at shallow crustal levels during exhumation, and thus suggest a maximum time frame of 320–311 Ma for this subduction–exhumation cycle ( Klemd et al. , 2005 ; Wang et al. , 2011 a ).

SAMPLE DESCRIPTION AND PETROGRAPHY

The investigated eclogite sample (L1010-2) was collected from the upper area of the Akesayi river, which is a branch of the main Akeyazi river ( Fig. 1b ). The sample comes from a loose decimeter-sized boulder ( Fig. 2a and b ) and consists of garnet ( c . 20 vol. %), omphacite ( c . 42 vol. %), phengite ( c . 15 vol. %), amphibole ( c . 7 vol. %), epidote ( c . 6 vol. %), quartz ( c . 2 vol. %) and pyrite ( c . 4 vol. %), with titanite, rutile, apatite and zircon as accessory minerals ( Fig. 2c–e ). This eclogite has distinctive textural features including multistage growth of garnet and epidote, as well as glaucophane and barroisite overgrowths, which have not been reported previously from the (U)HP rocks of the Akeyazi terrane.

Fig. 2

(a) Field photograph of the Akeyazi terrane and the sample locality. (b) Hand specimen of the eclogite sample investigated in this study. (c, d) Photomicrographs of the thin section (3.5 × 2.2 cm) used for mineral abundance estimation; (c) is in plane-polarized light and (d) is in cross-polarized light. (e) Photomicrograph of the area indicated in (c) showing the structure and mineral assemblages of the polycyclic eclogite sample with multistage garnet growth. Mineral abbreviations are after Whitney & Evans (2010) .

Garnet

The size and shape of the garnets and the nature of their mineral inclusions indicate the presence of two types of garnet: a large-size garnet (Grt-A, 1–5 mm) that mostly occurs as idioblastic porphyroblasts ( Figs 2e and 3a ), and a small-sized garnet (Grt-B, <0·5 mm) that shows irregular, xenoblastic shapes ( Fig. 3b ).

Fig. 3

Representative back-scattered electron (BSE) images of garnet in the eclogite. (a) Garnet porphyroblast (Grt-A) showing two stages of garnet growth, Grt_1 and Grt_2, separated by a sharp boundary highlighted by the dotted line. Areas indicated by black rectangles are enlarged in (c)–(f) to illustrate the inclusions in garnet. (b) Irregular small Grt-B consisting mainly of Grt_2 with or without Grt_1 inclusions. (c) Core of Grt_1 contains jadeite, omphacite (Omp_1) and quartz inclusions. Intergrowths of jadeite and omphacite are observed. (d) Mantle of Grt_1 contains jadeite, aegirine–augite, omphacite (Omp_1) and quartz inclusions as well as possible lawsonite pseudomorphs (epidote + paragonite assemblages). (e) Grt_2 contains omphacite (Omp_1), epidote (Ep_1) and quartz inclusions. (f) Jadeite, omphacite, paragonite, titanite and ilmenite inclusions in Grt_2.

Grt-A is usually distinctly zoned, displaying two generations (Grt_1 and Grt_2) ( Figs 2e and 3a ). The dominant idioblastic garnet (Grt_1) often has thin rims of xenoblastic Grt_2, and the boundary between them is usually sharp ( Figs 2e and 3a ). The core and mantle domains of Grt_1 contain inclusions of mainly quartz, jadeite, aegirine–augite, omphacite and minor possible lawsonite pseudomorphs (epidote + paragonite assemblages), rutile and pyrite ( Fig. 3a, c and d ), whereas the rim domain of the Grt_1 is usually inclusion-free except for some tiny rutile and apatite inclusions ( Fig. 3a ). Similarly, Grt_2 (surrounding Grt_1) in Grt-A also contains quartz, jadeite, omphacite and epidote, possibly lawsonite pseudomorphs (Ep and Pg), ilmenite, and rutile/titanite inclusions in the inner portions, whereas the outer parts are usually inclusion-free ( Fig. 3a, e and f ).

Grt-B mainly consists of Grt_2 ( Supplementary Material Fig. SM1b ; supplementary material is available for downloading at http://www.petrology.oxfordjournals.org ), which occasionally contains tiny Grt_1 grains in the core area ( Fig. 3b;Supplementary Data ), and contains the same mineral inclusion suite ( Fig. 3b ) as the Grt_2 domain in Grt-A.

Glaucophane and barroisite

Glaucophane frequently contains a barroisite core ( Fig. 4a–c ), and in places it is also rimmed by barroisite ( Figs 3a and 4). Detailed petrographic studies indicate that (1) the barroisite inclusions (Brs_1) occasionally enclose tiny glaucophane relics (Gln_1) ( Fig. 4 ), (2) Brs_1 is extensively overgrown by a second generation of glaucophane (Gln_2), and (3) Gln_2 is rimmed by later barroisite (Brs_2) ( Figs 3a and 4). Thus the amphiboles underwent the following textural transition: Gln_1 → Brs_1 → Gln_2 → Brs_2 ( Fig. 4 ).

Fig. 4

Representative photomicrograph, BSE images and element X-ray maps of amphibole in the eclogite. (a–c) Representative structures showing the cyclic transitions of glaucophane and barroisite (Gln_1 → Brs_1 → Gln_2 → Brs_2); (d–f) X-ray maps (Na, Ca and Fe) of amphibole with successive glaucophane and barroisite replacement. Arrows indicate isolated glaucophane relics (Gln_1) in barroisite (Brs_1).

Rare glaucophane inclusions were also found in garnet (Grt_1), whereas Fe-barroisite (Brs_2) occurs along cracks in garnet (Grt-A) ( Figs 2e and 5d ).

Fig. 5

Representative BSE images of omphacite, epidote and rutile/titanite in the eclogite. (a) Omphacite in the matrix shows two generations (Omp_1 and Omp_2) and Ep_1 is overgrown by Ep_2. (b) Barroisite (Brs_1), which contains early stage glaucophane (Gln_1) relicts, is overgrown by the second stage glaucophane (Gln_2). Ep_1 is enclosed by Ep_2, which contains many mineral inclusions (such as Omp_1). (c) Various glaucophane and barroisite intergrowths and rutile replacement by titanite that contains Omp_2 and Ep_2 inclusions. (d) Possible lawsonite pseudomorph (epidote–paragonite intergrowth) in garnet, barroisite and epidote in garnet cracks. Fe-rich barroisite is Brs_2 and Ep_1 inclusions are partly transformed to Ep_2 during the second exhumation process.

Sodic clinopyroxene

Sodic clinopyroxene in the eclogite occurs as omphacite, jadeite and aegirine–augite. Omphacite is the dominant matrix mineral and back-scattered electron (BSE) images reveal two generations ( Fig. 5a ), Omp_1 (grey) and Omp_2 (dark). In the matrix dark Omp_2 occurs as the dominant and latest-stage clinopyroxene type, whereas light Aeg–Aug and grey Omp_1 occur as patchy sectors (or small relict inclusions) in Omp_2 ( Fig. 5a;Supplementary Data ). Jadeite occurs only as inclusion in garnet ( Fig. 3 ) and is the predominant pyroxene inclusion type in Grt_1. However, a few Omp_1 and Aeg–Aug inclusions, both locally intergrown with jadeite, were also observed in the mantle of Grt_1 ( Fig. 3c and d ). In contrast, Omp_1 is the main pyroxene inclusion type in Grt_2 (both in Grt-A and Grt-B), which contains few jadeite inclusions in its early growth increments ( Fig. 3b, e and f ).

Epidote

Epidote displays two matrix types: an early textural generation (Ep_1) is slightly darker in BSE images than a later generation (Ep_2; Fig. 5a and b ). Ep_1, with an irregular or corroded shape, occasionally contains an allanite core and some quartz and glaucophane (possible Gln_2) inclusions ( Fig. 5a and b ). In contrast, xenoblastic Ep_2, which encloses Ep_1 relicts, contains omphacite inclusions (mainly Omp_1; Fig. 5a and b ). Epidote and Fe-barroisite (Brs_2) also occur along garnet cracks ( Figs 2e and 5d ), along which quartz is the main mineral ( Fig. 2e ). The epidote in garnet is mainly Ep_1, and Ep_2 also occasionally replaces the Ep_1 inclusions ( Fig. 5d ).

Other minerals

Phengite usually occurs as a matrix mineral ( Fig. 3a ) whereas paragonite (intergrown with epidote) exclusively occurs in garnet, probably representing one phase of the mineral assemblage that replaces lawsonite ( Figs 3d and 5d ). Rutile and titanite are the main Ti-phases, with titanite often surrounding partially resorbed rutile grains ( Figs 3a and 5c ). In addition, titanite contains many mineral inclusions such as Omp_2, Ep_2, phengite and quartz ( Fig. 5c ). Pyrite contains omphacite, epidote, quartz, rutile, titanite and apatite inclusions.

ANALYTICAL METHODS

Major element compositions of minerals were obtained by electron microprobe analysis (JEOL JXA 8200) at the GeoZentrum Nordbayern (GZN), University of Erlangen–Nürnberg, Erlangen, Germany. Quantitative analyses were performed using wavelength-dispersive spectrometers (WDS) with an acceleration voltage of 15 kV, a beam current of 15 nA, a 3 μm beam size and 10–30s counting time. The same instrument was also used to acquire the BSE images. Natural minerals and synthetic oxides were used as standards, and the ZAF procedure was used for data correction. X-ray maps for selected elements in garnet, omphacite and epidote were conducted in WDS mode at GZN with an acceleration voltage of 15 kV, a beam current of 100–200 nA, a 2–7 μm pixel size and dwell time of 100 ms. For amphibole the X-ray maps were performed in WDS mode at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS) with an acceleration voltage of 15 kV, a beam current of 100 nA, a 0·5 μm pixel size and dwell time of 40 ms. Representative major element analyses are given in Table 1 .

Table 1: Representative major element composition of minerals in the eclogite

Mineral:	Grt	Grt	Grt	Grt	Grt	Grt	Amp	Amp	Amp	Amp	Aeg–Aug	Omp	Omp	Jd	Jd	Ph	Pg	Ep	Ep	Rt	Ttn	Ilm
Location:	Grt_1c	Grt_1m	Grt_1r	Grt_2c	Grt_2m	Grt_2r	Gln_1	Brs_1	Gln_2	Brs_2	matrix	Omp_1	Omp_2	in Grt_1	in Grt_2	matrix	in Grt_1	Ep_1	Ep_2	in Ttn	matrix	in Grt_2
SiO 2	38·32	38·45	38·70	38·61	38·09	37·87	56·24	50·83	55·39	47·19	54·84	55·55	55·65	57·21	58·05	52·34	50·98	38·48	38·47	0·05	32·14	0·04
TiO 2	0·15	0·00	0·07	0·20	0·14	0·00	0·04	0·32	0·07	0·27	0·15	0·09	0·11	0·05	0·03	0·28	0·66	0·20	0·10	96·15	38·14	50·19
Al 2 O 3	21·25	21·33	21·71	20·21	21·53	21·75	10·60	9·50	10·72	11·44	6·32	10·39	9·10	15·45	14·70	27·03	36·92	27·36	24·88	0·02	1·01	0·00
Cr 2 O 3	0·00	0·00	0·00	0·00	0·00	0·00	0·00	0·03	0·00	0·05	0·03	0·04	0·04	0·02	0·00	0·03	0·01	0·00	0·02	0·04	0·17	0·01
FeO	30·97	31·25	28·99	29·65	29·16	29·42	14·47	17·02	16·32	21·61	18·27	11·33	9·23	8·95	12·42	3·97	1·16	0·00	0·00	2·84	0·57	46·08
Fe 2 O 3	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	n.d.	7·87	10·86	n.d.	n.d.	n.d.
MnO	0·70	0·60	0·34	0·39	0·44	0·13	0·04	0·01	0·00	0·06	0·00	0·03	0·02	0·01	0·01	0·00	0·05	0·01	0·01	0·02	0·01	2·98
MgO	1·50	1·72	2·00	1·34	1·61	1·70	9·02	10·13	7·52	6·84	3·29	4·61	6·83	2·29	0·81	2·85	0·35	0·04	0·02	0·01	0·00	0·04
CaO	8·01	7·40	9·16	9·02	9·95	10·03	1·39	5·92	1·50	6·43	7·23	9·72	12·30	4·37	1·86	0·01	0·27	23·59	23·39	0·39	28·06	0·08
Na 2 O	0·04	0·00	0·00	0·00	0·00	0·00	6·97	5·21	7·12	4·91	10·58	9·18	7·45	12·37	14·01	0·65	6·70	0·01	0·01	0·00	0·08	0·00
K 2 O	0·00	0·00	0·00	0·00	0·00	0·00	0·01	0·17	0·05	0·39	0·00	0·01	0·00	0·00	0·00	9·93	1·39	0·00	0·00	0·00	0·03	0·00
F	0·00	0·00	0·00	0·00	0·00	0·00	0·00	0·02	0·00	0·01	0·00	0·03	0·00	0·00	0·01	0·01	0·00	0·00	0·00	0·00	0·00	0·00
Cl	0·00	0·00	0·00	0·00	0·00	0·00	0·00	0·00	0·01	0·00	0·00	0·00	0·00	0·00	0·00	0·00	0·02	0·00	0·01	0·00	0·00	0·00
Total	100·95	100·75	100·97	99·41	100·91	100·90	98·78	99·15	98·70	99·18	100·70	100·96	100·72	100·74	101·90	97·10	98·51	97·63	97·76	99·52	100·21	99·42
Si	3·040	3·054	3·040	3·101	3·005	2·985	7·750	7·212	7·749	6·876	1·979	1·975	1·987	1·989	1·999	3·438	3·160	2·990	3·009	0·001	1·042	0·000
Al	1·985	1·995	2·010	1·911	2·000	2·019	1·721	1·587	1·766	1·962	0·268	0·435	0·383	0·633	0·596	2·091	2·695	2·489	2·291	0·000	0·039	0·000
Cr	0·000	0·000	0·000	0·000	0·000	0·000	0·000	0·003	0·000	0·006	0·001	0·001	0·001	0·001	0·000	0·002	0·000	0·000	0·001	0·000	0·004	0·000
Ti	0·009	0·000	0·004	0·012	0·008	0·000	0·004	0·034	0·007	0·030	0·004	0·002	0·003	0·001	0·001	0·014	0·031	0·012	0·006	0·980	0·930	0·970
Fe 3+	0·000	0·000	0·000	0·000	0·000	0·000	0·496	0·653	0·332	0·754	0·505	0·240	0·151	0·218	0·338	0·000	0·000	0·511	0·709	0·000	0·000	0·000
Fe 2+	2·055	2·076	1·906	1·991	1·924	1·939	1·171	1·366	1·578	1·879	0·047	0·097	0·125	0·043	0·020	0·218	0·060	0·000	0·000	0·032	0·015	0·990
Mn	0·047	0·040	0·023	0·027	0·029	0·009	0·005	0·001	0·000	0·007	0·000	0·001	0·001	0·000	0·000	0·000	0·003	0·001	0·001	0·000	0·000	0·064
Mg	0·177	0·204	0·234	0·160	0·189	0·200	1·853	2·143	1·568	1·486	0·177	0·244	0·364	0·119	0·042	0·279	0·032	0·005	0·002	0·000	0·000	0·001
Ca	0·681	0·630	0·771	0·776	0·841	0·847	0·205	0·900	0·225	1·004	0·279	0·370	0·471	0·163	0·069	0·001	0·018	1·964	1·960	0·006	0·974	0·002
Na	0·006	0·000	0·000	0·000	0·000	0·000	1·863	1·433	1·931	1·387	0·740	0·633	0·516	0·834	0·936	0·083	0·805	0·002	0·002	0·000	0·005	0·000
K	0·000	0·000	0·000	0·000	0·000	0·000	0·002	0·031	0·009	0·072	0·000	0·000	0·000	0·000	0·000	0·832	0·110	0·000	0·000	0·000	0·001	0·000
Cation	8·00	8·00	7·99	7·99	8·00	8·00	15·07	15·36	15·17	15·46	4·00	4·00	4·00	4·00	4·00	6·96	6·91	7·97	7·98	1·02	3·01	2·03
O	12	12	12	12	12	12	23	23	23	23	6	6	6	6	6	12	12	13	13	2	5	3
XFe	0·92	0·91	0·89	0·93	0·91	0·91	0·39	0·39	0·50	0·56	0·21	0·28	0·26	0·27	0·32	0·44	0·65
Alm	69·43	70·37	64·96	67·40	64·50	64·74				WEF	25·37	36·00	48·19	16·28	6·53		Ps	0·17	0·24
Grs	23·01	21·36	26·28	26·27	28·19	28·28				Jd	24·54	40·36	36·80	62·02	59·65
Prp	5·98	6·92	7·98	5·42	6·34	6·68				Ae	50·09	23·63	15·01	21·70	33·82
Spss	1·59	1·36	0·78	0·91	0·97	0·30

X_Fe = Fe ²⁺ /(Fe ²⁺ + Mg); Ps, pistacite; Prp, pyrope; Alm, almandine; Grs, grossular; Spss, spessartine; WEF, wollastonite + enstatite + ferrosilite; Jd, jadeite; Ae, aegirine; c, core region of mineral; m, mantle region of mineral; r, rim region of mineral;. n.d., not determined; Fe ³⁺ was calculated assuming stoichiometric mineral compositions.

Trace-element analyses of minerals were performed in situ on polished thin sections by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) at the GZN using a single-collector quadrupole Agilent 7500i ICP-MS system equipped with an UP193Fx Argon Fluoride New Wave Research Excimer laser ablation system. The glass reference material NIST SRM 612 was used as standard for external calibration. LA-ICP-MS measurements were conducted using a spot size of 25 μm diameter, a laser frequency of 15 Hz and 0·50 GW cm ⁻² and a fluence of 2·48 J cm ⁻² . The carrier gas consisted of a mixture of 0·65 l min ⁻¹ helium and 1·06 l min ⁻¹ argon. Acquisition time was 20 s for the background and 25 s for the mineral analysis. The internal standards used were Si for the silicates, Ca for the Ca-phases and Ti for the Ti-phases, all of which were determined by electron microprobe analysis (EMPA). Reproducibility and accuracy, which were determined for NIST SRM 610, were usually <6% and <10%. The trace-element concentrations were calculated using GLITTER Version 3 ( van Achterbergh et al. , 2000 ). Representative trace-element analyses of the minerals are given in Supplementary Data .

The bulk-rock major element composition was determined by X-ray fluorescence spectrometry (XRF) on fused glass disks, using a Philips PW 2400 XRF spectrometer at the GZN. The loss on ignition (LOI) was determined prior to major element analysis using a pre-ignition method after heating the samples (1 g) for 12 h at 1300 °C. The major elements SiO ₂ , TiO ₂ , Al ₂ O ₃ , Fe ₂ O ₃ , MnO, MgO, CaO, Na ₂ O, K ₂ O, and P ₂ O ₅ , along with a subset of trace elements (Ba, Cr, Ga, Nb, Ni, Pb, Rb, Sr, Yh, V, Y, Zn, Zr), were analyzed. Generally, precision and accuracy were better than 1% (2σ) for most elements and 0·5% for SiO ₂ .

MINERAL CHEMISTRY

Garnet

Garnet has 63–72 mol % almandine, 20–29 mol % grossular, 4–9 mol % pyrope and 0–2 mol % spessartine components ( Tables 1 and 2 ). The X-ray maps and a compositional profile show that Grt_1 and Grt_2 are chemically zoned ( Figs 6 and 7 a).

Fig. 6

Element X-ray maps (a–e) of garnet (Grt-A) showing two-stage growth and BSE image of the same grain (f).

Fig. 7

Chemical compositions of garnet, omphacite, epidote and amphibole in the eclogite sample. (a) Compositional profile intersecting the center of the garnet porphyroblast (Grt-A) in Fig. 3a showing two-stage growth. (b) Ternary classification diagram for sodic clinopyroxenes after Morimoto et al. (1988) , showing the distinct aegirine–augite and omphacite generations, which display a linear evolution. (c) Chemical discrimination of two groups of epidote with various Al and Fe concentrations. (d, e) Chemical composition of amphiboles in the eclogite [classification after Hawthorne & Oberti (2006) ].

Table 2: Mineral assemblages and compositions during metamorphic evolutions

In Grt_1, Mn decreases slightly from the core to the mantle, and strongly decreases at the rim; however, at the outer rim, X_Spss [= 100Mn/(Fe + Ca + Mg + Mn)] values show a subtle increase ( Figs 6 and 7 a). In contrast, Mg shows a reversed behavior when compared with Mn. The Ca content decreases from the core to the mantle, is rather constant at the mantle domain, and increases rapidly at the rim; X_Grs [= 100Ca/(Fe + Ca + Mg + Mn)] values reach their highest at the outer rim ( Figs 6 and 7 a). Iron shows a reversed zoning pattern compared with that of Ca ( Figs 6 and 7 a). The major compositional ranges of the core–mantle–rim domains of Grt_1 are given in Table 2 . When normalized against chondrite, garnet displays a rare earth element (REE) pattern with a relative light REE (LREE) depletion and enrichments of the middle (MREE) and heavy REE (HREE) ( Supplementary Material Fig. SM4a and b ). Trace-element zoning of Grt_1 is characterized by core–mantle–rim decreases of MREE (e.g. Tb: 3·13 ppm in the core, 1·14 ppm in the mantle and 0·096 ppm in the rim) and Y (e.g. 93·4 ppm in the core, 45·6 ppm in the mantle and 33·0 ppm in the rim), and increases of HREE (e.g. Yb: 4·32 ppm in the core, 5·84 ppm in the mantle and 9·61 ppm in the rim) ( Supplementary Material Fig. SM4e and f , Supplementary Data ). Thus, when compared with the HREE, the MREE vary from slightly enriched in the core and mantle to strongly depleted in the rim of Grt_1 ( Supplementary Material Fig. SM4a and b ).

The Grt_2 of Grt-A and Grt-B displays two separate zones of chemical zoning ( Fig. 7a ). At the initial growth stage, the almandine and spessartine components increase and the pyrope and grossular components decrease ( Figs 6 and 7 a; Supplementary Material Fig. SM1 ). Subsequently, during the second stage, the almandine and spessartine components decrease and the pyrope and grossular components increase towards the rim ( Figs 6 and 7 a; Supplementary Material Fig. SM1 ), similar to the zoning patterns in the mantle and rim domains of Grt_1. At the outermost rim of Grt_2, the Mn content drops significantly ( Figs 6d and 7a;Supplementary Material Fig. SM1 ). Detailed core–mantle–rim major compositional variations of Grt_1 are given in Table 2 . The trace element pattern of Grt_2 is slightly different than that of Grt_1 in showing lower MREE and Y concentrations and increasing core–rim trends ( Supplementary Material Fig. SM4e ).

Sodic clinopyroxene

The four generations of clinopyroxene described above vary in composition ( Table 1 ). In the Jd–Ae–WEF (Wo, En, Fs) ternary diagram ( Morimoto et al. , 1988 ), aegirine–augite and omphacite plot in three separate groups and display a linear evolution ( Fig. 7b ). Aegirine–augite plots near the boundary of the aegirine and aegirine–augite fields, whereas Omp_1 and Omp_2 classify as typical omphacites ( Fig. 7b ). Compositional X-ray maps indicate that the linear evolution from aegirine–augite to omphacite is strongly controlled by Fe, Na, Mg and Ca concentrations ( Supplementary Material Fig. SM2 ). The FeO and Na ₂ O concentrations decrease gradually from Aeg–Aug via Omp_1 to Omp_2 (e.g. FeO: from 18·27 to 9·23 wt %; Na ₂ O: from 10·58 to 7·45 wt %) and are balanced by increasing MgO and CaO concentrations (e.g. MgO: from 3·29 to 6·83 wt %; CaO: from 7·23 to 12·30 wt %). Accordingly, the X_Aeg [= 100Fe ³⁺ /(Fe ³⁺ + Al ^VI + Fe ²⁺ + Mg)] values decrease from 46–52 in Aeg–Aug, via 21–33 in Omp_1, to 14–22 in Omp_2. The X_Mg [= 100Mg/(Fe ³⁺ + Fe ²⁺ + Mg)] values increase from 21–28 in Aeg–Aug, via 32–49 in Omp_1 to 47–58 in Omp_2 ( Table 2 ). However, Aeg–Aug has lower X_Jd [= 100Al/(Fe ³⁺ + Al ^VI + Fe ²⁺ + Mg)] values (27–39) whereas Omp_1 and Omp_2 share a uniform higher X_Jd range (35–44). In addition, Aeg–Aug has higher X (Fe ³⁺ ) [= Fe ³⁺ /(Fe ³⁺ + Fe ²⁺ )] values (0·89–0·94), whereas the values of Omp_1 (0·55–0·80) overlap those of Omp_2 (0·51–0·72) ( Table 2 ).

Jadeite, which occurs only as inclusions in garnet, consists of 55–66 mol % jadeite component ( Fig. 7b;Table 1 ). The X_Aeg values of most jadeites range between 25 and 39; however, occasionally jadeite contains no aegirine component ( Fig. 7b ).

Glaucophane and barroisite

Glaucophane is the matrix sodic amphibole according to the classification of Hawthorne & Oberti (2006) , whereas Fe-glaucophane mainly occurs as inclusions in garnet ( Fig. 7d ). Gln_1 and Gln_2 have similar major element compositions ( Table 1 ), but Gln_1 has slightly higher Fe ³⁺ and lower Al ³⁺ contents in the C-site than Gln_2 ( Fig. 7d;Table 1 ). The sodic–calcic amphibole associated with glaucophane is classified as barroisite ( Fig. 7e ). Generally, Brs_1 has higher SiO ₂ (50·83 wt %) and lower FeO (17·02 wt %) contents than Brs_2 (47·19 wt % and 21·61 wt %, respectively) ( Fig. 7e;Table 1 ). Barroisite along garnet cracks ( Fig. 5d ) has even higher FeO concentration (e.g. 26·5 wt %) when compared with matrix Brs_2 and is classified as ferro-barroisite ( Fig. 7e ).

The concentrations of most trace elements, except Li and Ti, in amphiboles are below the detection limit ( Supplementary Material Table SM1 ). It is noteworthy that Brs_1 has a higher Mn concentration than Gln_2, 285 ppm vs 63·7 ppm, respectively ( Supplementary Material Table SM1 ).

Epidote

The analyzed epidote-group minerals have high Fe ₂ O ₃ contents (>6 wt %) and are classified as epidote ( Table 1 ), although allanite is preserved in some of their core domains ( Fig. 5b ). Compositional X-ray maps of epidote indicate that Ep_1 is relatively enriched in Al and Ep_2 in Fe ( Supplementary Material Fig. SM3 ). The Al and Fe distribution shows a good negative correlation ( Fig. 7c ). CaO variations between Ep_1 and Ep_2 are insignificant ( Supplementary Material Fig. SM3 ). Generally, Ep_1 has higher Al ₂ O ₃ (27–29 wt %) and lower Fe ₂ O ₃ (6–8 wt %) contents than Ep_2 (23–26 wt % and 10–12 wt %, respectively) ( Fig. 7c;Table 1 ).

The concentrations of most trace elements vary insignificantly between Ep_1 and Ep_2 ( Supplementary Material Table SM1 ). Ep_2 has slightly lower ∑REE values than Ep_1 ( Supplementary Material Table SM1 ); however, they display similar chondrite-normalized REE patterns ( Supplementary Material Fig. SM4d )

Other minerals

The Si content of phengite is c . 3·45 a.p.f.u. and the Na content of paragonite clusters between 0·81 and 0·86 a.p.f.u. Phengite has high concentrations of Cs, Rb and Ba whereas the HFSE and REE concentrations are below the detection limit ( Supplementary Material Table SM1 ). Rutile and titanite are the main carriers of Ti, Nb and Ta ( Supplementary Material Table SM1 ). Titanite also contains considerable amounts of MREE. Apatite has F contents of 3·6–3·8 wt %, whereas the Cl contents are usually below the detection limit.

PSEUDOSECTION MODELING

P – T pseudosection modeling is currently the most common technique used to obtain thermobarometric information on HP–UHP rocks, and on the shape of their P – T trajectories (e.g. Konrad-Schmolke et al. , 2006 ; Powell & Holland, 2008 ; Groppo & Castelli, 2010 ). The chemical zoning of minerals (such as garnet and phengite) and the respective bulk-rock composition allows calculation of a set of P – T ranges, which can be used to reconstruct the metamorphic P – T path. The resultant P – T evolution frequently helps to understand the driving forces of metamorphism, paricularly the geodynamic environment of subduction-related terranes (e.g. Groppo & Castelli, 2010 ; Massonne, 2013 ; Meyer et al. , 2014 ).

Calculation method

The P – T pseudosections for the studied eclogite were calculated in the model system MnNCKFMASH(O) [MnO–Na ₂ O–CaO–K ₂ O–FeO–MgO–Al ₂ O ₃ –SiO ₂ –H ₂ O(–O)] considering the actual mineral assemblage. The fluid phase was assumed to be pure H ₂ O and in excess. TiO ₂ and P ₂ O ₅ were neglected as they predominantly occur in rutile/titanite and apatite, and are incorporated only in minor amounts in the main silicate minerals in the P – T range of interest. Accordingly, CaO and SiO ₂ accounting for titanite and apatite were deducted from the bulk-rock compositions; FeO* was also subtracted because of the presence of pyrite in the sample (for details see Supplementary Material ). However, MnO was considered for the pseudosection calculation owing to its critical role in stabilizing garnet towards lower temperatures, especially at lower pressures (e.g. Droop & Harte, 1995 ; Konrad-Schmolke et al. , 2006 ). Furthermore, Fe ₂ O ₃ was included because of the occurrence of significant volumes of Fe ³⁺ -bearing minerals, such as epidote, omphacite and amphibole, in the eclogite sample ( Tables 1 and 2 ) as well as its effect on the stability fields of those minerals. This is of particular importance for amphibole stability at high-grade metamorphic conditions (e.g. Warren & Waters, 2006 ; Diener & Powell, 2012 ). Because naturally occurring ‘oceanic’ eclogites have higher Fe ³⁺ contents ( John & Schenk, 2003 ; John et al. , 2010 ; Rebay et al. , 2010 ; Chen et al. , 2013c ) than unaltered MORB (0·12–0·16) ( Bezos & Humler, 2005 ; Cottrell & Kelley, 2011 ), X (Fe ³⁺ ) was set at 0·20, combining the mineral modes and microprobe data (see Groppo & Castelli, 2010 ). However, for simplicity the X (Fe ³⁺ ) value is treated as a constant at each stage (see below) considering the ferric iron transport within amphibole, omphacite and epidote. This is acceptable given that calculating with different X (Fe ³⁺ ) values has an insignificant influence on the pseudosection and deduced P – T path (for details see Supplementary Material and Supplementary Data ; see Li et al. , 2015 ).

Pseudosection calculations were performed using the Perple_X software package ( Connolly, 1990 , 2005 ; version 6.6.7) and the internally consistent thermodynamic database of Holland & Powell (1998) and update (hp02ver.dat file). The following solid-solution models were used: Gt(HP) for garnet ( Holland & Powell, 1998 ), Omph(GHP) for omphacite ( Green et al. , 2007 ), cAmph(DP) for amphibole ( Diener et al. , 2007 ), Mica(CHA) for phengite ( Coggon & Holland, 2002 ), Chl(HP) for chlorite ( Holland & Powell, 1998 ) and Ep(HP) for epidote ( Holland & Powell, 1998 ). Lawsonite and quartz were treated as pure end-member phases.

Effective bulk composition (EBC) calculations

A prerequisite of pseudosection calculation and thermobarometrical application is that thermodynamic equilibrium was maintained between minerals, or between porphyroblasts and matrix throughout the entire P – T evolution. In this case, effective bulk compositions are critical for phase equilibrium calculations in metamorphic rocks, because the bulk composition considered available for mineral reaction may change during the P – T evolution as a result of element fractionation into compositionally zoned garnet porphyroblasts. For instance, the garnet core is chemically isolated during the growth of the garnet mantle and rim (e.g. Stüwe, 1997 ; Evans, 2004 ). Therefore, the modeling of a set of pseudosections based on the EBC representative of each equilibration volume and the estimation of P – T ranges using the garnet isopleth thermobarometry method for each pseudosection is believed to result in a more accurate reconstruction of the P – T path (e.g. Gaidies et al. , 2006 ).

Although the garnets in the present study do not have a distinctive bell-shaped Mn profile, indicating possible minor element fractionation during their growth (see Tian & Wei, 2013 ; Regis et al. , 2014 ), calculation of an EBC is still necessary to achieve equilibration volumes and therefore obtain a precise P – T path (for details see Supplementary Material and Supplementary Material Fig. SM6 ). The bulk-rock compositions effectively reacting during each stage of garnet growth (including Grt_1 and Grt_2) were calculated following the method described by Evans (2004) and Gaidies et al. (2006) . This method employs a Rayleigh fractionation model ( Hollister, 1966 ) based on the measured Mn content of garnet to generate composition vs modal proportion curves for garnet, followed by vector estimations of crystal fractionation ( Supplementary Material Fig. SM7 ). Examples of detailed calculations have been given by Fazio et al. (2008) and Groppo & Castelli (2010) ; details of the EBC calculation are given in the Supplementary Material . Garnets were divided into several growth increments on the basis of their compositional zoning patterns and relative Mn contents ( Supplementary Material Fig. SM7a and b ). The core-to-rim domains of Grt_1 were divided into three increments (Grt_1a, Grt_1b and Grt_1c), and a similar division was employed for Grt_2 (Grt_2a, Grt_2b and Grt_2c). The representative effective bulk compositions of six growth stages (EBC-1a to EBC-2c) used for the pseudosection calculations are summarized in Table 3 . Despite the observation that sodic clinopyroxene occurs as several types during the P – T evolution, bulk-rock composition fractionation as a consequence of its development was not considered owing to the presence of only very rare relicts of Aeg–Aug and the limited compositional difference between Omp_1 and Omp_2 ( Fig. 7b;Table 1 ), which thus have an insignificant influence on EBC variations compared with garnet (e.g. Groppo & Castelli, 2010 ).

Table 3: Effective bulk compositions (EBCs) used for calculation of pseudosections (wt %)

L1010-2	XRF	XRF*	EBC
			1a	1b	1c	2a	2b	2c
SiO 2	45·19	44·69	50·62	50·90	52·82	54·45	54·68	55·77
TiO 2	2·42
Al 2 O 3	14·36	14·36	16·26	16·17	15·46	14·85	14·77	14·38
FeO			11·95	11·67	9·75	8·22	8·02	7·04
Fe 2 O 3	19·32	14·65	3·32	3·24	2·71	2·28	2·23	1·95
MnO	0·16	0·16	0·18	0·17	0·11	0·08	0·07	0·06
MgO	3·53	3·53	4·00	4·06	4·41	4·68	4·72	4·93
CaO	8·31	7·46	8·45	8·47	8·62	8·65	8·63	8·54
Na 2 O	3·18	3·18	3·61	3·68	4·22	4·68	4·75	5·06
K 2 O	1·43	1·43	1·61	1·65	1·89	2·10	2·12	2·26
P 2 O 5	0·29
LOI	1·58
Total	99·77	89·46	100·00	100·00	100·00	100·00	100·00	100·00

XRF* is bulk composition after subtracting element contents accommodated in rutile, titanite, apatite and pyrite from the whole-rock XRF composition; X (Fe ³⁺ ) was set at 0·20 and considered as constant.

Pseudosection and P – T estimates

Based on various effective bulk compositions representing the equilibrated compositions during the growth from Grt_1a to Grt_2c ( Table 3 ), six P – T pseudosections in the MnNCKFMASHO system were calculated for the eclogite sample ( Fig. 8 and Supplementary Material Fig. SM8 ). Garnet, omphacite and phengite are stable phases under all considered P – T conditions. Amphibole is absent in the HT–HP field. Lawsonite stabilizes in the LT–HP field, whereas the epidote-group minerals appear in the low-pressure field ( Fig. 8 and Supplementary Material Fig. SM8 ). Chlorite stabilizes in the low-temperature field (e.g. <490 °C), whereas quartz appears only in the high-temperature field (e.g. >500 °C) where chlorite is absent owing to the growth of Grt_1 ( Fig. 8a and Supplementary Material Fig. SM8a–c ). The chlorite-out and quartz-in isograds are almost parallel to the P -axis at temperatures of c . 480 °C and 500 °C respectively. As the effective bulk composition changes they shift progressively towards the LT–HP field at pressures >2·0 GPa ( Fig. 8 and Supplementary Material Fig. SM8 ). In addition, two types of omphacite are present in the LT–LP field during Grt_1 nucleation ( Fig. 8a ), and only one omphacite type is left during Grt_2 growth according to the changes of the EBCs ( Supplementary Material Fig. SM8 ). However, the pseudosection and the calculated mineral assemblages at high temperatures (e.g. >520 °C) vary insignificantly with EBC variations ( Fig. 8 and Supplementary Data ).

Fig. 8

P – T pseudosections (calculated using Perple_X) in the system NMnCaKFMASHO for the eclogite sample calculated with the effective bulk composition for stage I (a) and stage IV (b). Areas of the pseudosections outlined by the dashed rectangles are contoured with grossular and pyrope isopleths. P – T segments are derived from the intersection of compositional isopleths corresponding to the actual compositions of garnet and constrain the P – T conditions of its growth at each of the stages. The complete set of pseudosections for stages I–III and IV–VI with all the compositional isopleths for garnet is given in Supplementary Material Fig. SM8 . The rapid Exhumation_1 to a shallow level (long-dashed line) is deduced from the occurrence of Ep_1 in both Grt_2 and the matrix. A composite P – T path (lines with arrows) deduced from segments of single pseudosections shows that the eclogite sample underwent a complicated P – T evolution.

For each pseudosection, the relative P – T conditions were constrained for each garnet growth stage using the method of garnet isopleth thermobarometry (e.g. Gaidies et al. , 2006 ; Groppo & Castelli, 2010 ). The measured mineral composition of the Grt_1 core (Prp 5·9–6·2 and Grs 23–20; Table 2 ) relates to a short P increase of initial garnet nucleation from 2·3 to 2·5 GPa at 505–510 °C in the stability field of Grt + Omp + Amp + Ph + Lws ± Qz ( Fig. 8a ). The mantle composition of Grt_1 (Prp 6·4–7·3 and Grs 21–22; Table 2 ) constrains an initial P decrease from 2·4 GPa to 2·3 GPa at 510–515 °C in equilibrium with the mineral assemblages of Grt + Omp + Amp + Ph + Lws + Qz ( Supplementary Material Fig. SM8b ). The Grt_1 rim reveals an increase of the pyrope component to a maximum value (from 7·8 to 8·5) and a subsequent decrease (from 8·5 to 7·3) at the outermost rim, as well as a continuously increasing grossular component (from 22 to 28) ( Fig. 7a;Table 2 ). These compositional changes reflect a significant P decrease and a limited change in T (from 2·4 to 2·2 GPa and 515 to 525 °C and subsequently from 2·2 to 2·0 GPa and 525 to 520 °C) in equilibrium with the mineral assemblage Grt + Omp + Amp + Ph + Lws + Qz that also occurs as inclusions in the mantle domain of Grt_1 ( Supplementary Material Fig. SM8c ).

The incipient growth of Grt_2 ( Fig. 8b ), the abrupt decrease of X_Prp [= 100Mg/(Fe + Ca + Mg + Mn)] values from 7·3 to 4·4 and the slight decrease of X_Grs values from 29 to 26 ( Fig. 7a;Table 2 ) reveal a short P increase– T decrease period (from 1·95 to 2·15 GPa and 520 to 480 °C), whereas the subsequent increase in pyrope and grossular components (from 4·4 to 5·8 and from 26 to 27, respectively) suggest a second short T increase and minor P decrease (from 480 to 500 °C at c . 2·1 GPa) ( Fig. 8b ). In addition, the chemical composition of the Grt_2 mantle (Prp 6·0–7·0 and Grs 27–28) and the Grt_2 rim (Prp 6·8–7·1 and Grs 28–29; Table 2 ) records a second P decrease– T increase period from 2·1 to 1·95 GPa at 505–520 °C in the stability field of Grt + Omp + Amp + Ph + Lws + Qz ( Supplementary Material Fig. SM8e and f ).

Reliability of the modeling results and error estimates

Uncertainties regarding the P – T estimates (see above) determined by isopleth thermobarometry derive from systematic errors including thermodynamic data uncertainties, random analytical imprecisions as a result of inaccurate microprobe analyses, and the effects of disequilibrium and mineral fractionation (hereafter ‘geological errors’) (e.g. Worley & Powell, 2000 ; Dachs et al. , 2012 ). The internally consistent thermodynamic dataset of Holland & Powell (1998) contains calculated uncertainties for the enthalpies of formation of the mineral end-members, which can be propagated through calculations. Unfortunately, uncertainties on tabulated entropies, which can be as high as 10% and are mainly based on mismatches between results from entropy estimation techniques and calorimetric measurements ( Dachs et al. , 2012 ), are not considered. Perple_X Gibbs energy minimization algorithms do not allow estimation of the P – T uncertainties related to the adopted thermodynamic database (e.g. Fazio et al. , 2008 ), but are thought to be in the same range as those propagated in T HERMOCALC calculations (average P – T , Powell & Holland , 1994 ). Therefore the average P – T thermobarometric calculation was used to estimate the uncertainties regarding the thermodynamic database and T HERMOCALC calculation (e.g. Powell & Holland, 1994 , 2008 ; Worley & Powell, 2000 ). Even though the calculated average T condition (495 ± 25 °C) derived from T HERMOCALC for the studied sample corresponds well to that obtained from the pseudosections, the average P estimate (1·82 ± 0·18 GPa) is somewhat lower, owing to the statistical parameters such as sigfit and e* (see Meyer et al. , 2013 ). However, our estimates of propagated P – T uncertainties of ± 0·18 GPa and ±25 °C are believed to represent the best possible estimate of pseudosection calculations uncertainties.

The analytical error is derived from counting statistics, standardization and correction procedures in the electron microprobe routine. The EPMA error is typically <1% relative, constraining the associated uncertainty on grossular and pyrope contents to about <0·3 and <0·1, respectively ( Table 2 ). Because the isopleths are spaced at about P / X_Grs = 0·1 GPa and T / X_Prp =10 °C in the P – T ranges of interest ( Fig. 8 and Supplementary Material Fig. SM8 ), the corresponding uncertainties of the P – T calculations derived from analytical errors are about 0·06 GPa and 2 °C. Mapping the composition of porphyroblasts prior to point analysis is helpful in reducing the input of the geological error into the compositional dataset. In addition, calculating the changes of the effective bulk composition minimizes the geological error caused by crystal fractionation and disequilibrium.

In summary, the uncertainties associated with the P – T calculations are proposed to lie in the range of 0·18 GPa and 25 °C, thereby suggesting no significant overlap of the calculated pressures, although a limited temperature overlap occurs between the P – T segments (Min ^T = 480 °C, Max ^T = 525 °C; Min ^P = 1·95 GPa, Max ^P = 2·48 GPa). However, because our main aim is to decipher the relative and not the absolute P – T values and we use an internally consistent approach for the whole dataset, we consider the resulting trends of the P – T evolution as robust.

The combination of pseudosection modeling based on the effective bulk composition and garnet isopleth thermobarometry allows the modeling of a well-defined P – T path for the eclogite sample, which is in accordance with our petrological observations.

However, minor discrepancies with regard to the clinopyroxene and lawsonite stability fields between the pseudosection and the actual sample are also present. For example, jadeite is present as inclusions in garnet cores and mantles; however, jadeite is absent in the pseudosection and two immiscible clinopyroxenes occur only in the LT–LP field ( Supplementary Material Fig. SM8 ). This discrepancy may be related to the high miscibility of omphacite solid solutions, whereas the absence of lawsonite in the natural mineral assemblage is probably due to its late consumption during retrograde replacement. This breakdown of large amounts of lawsonite (>6 vol. %), during decompression seems to have affected the omphacite and amphibole compositions of the actual eclogite sample, as both differ slightly from those in the calculated pseudosections. These discrepancies are thought to be acceptable considering the uncertainties in the thermodynamic database as well as the geological assumptions (e.g. EBCs ignoring polyphase sodic clinopyroxene, amphibole and epidote; or Fe ³⁺ estimation) used for the modeling. In addition, practical experience indicates that these discrepancies have little effect on the distribution patterns of garnet component isopleths and the application of garnet isopleth thermobarometry (e.g. Groppo & Castelli, 2010 ). Therefore, the shape and trend of the calculated P – T trajectories are thought to be those of the trajectory actually taken by the studied eclogite in the subduction channel.

DISCUSSION

Episodic metasomatism or multiple-stage metamorphism

Chemical zoning of minerals and their inclusion assemblage, especially in porphyroblastic minerals, are distinctive features in metamorphic rocks, and may provide important information on the minerals’ growth history ( Tracy, 1982 ; Farber et al. , 2014 ). For example, study of garnet zoning is an important tool in determining the P – T evolution of metamorphic rocks. Normal garnet zoning is usually attributed to Rayleigh fractionation during prograde growth ( Hollister, 1966 ), although the mineral phases in chemical equilibrium with garnet, as well as dehydration reactions, are also important parameters influencing zoning patterns ( Konrad-Schmolke et al. , 2006 ; Farber et al. , 2014 ). Monophase chemical zoning in metamorphic minerals is commonly observed, whereas polyphase mineral generation is relatively rare. Polyphase mineral growth in metamorphic rocks is mainly recorded by minerals such as garnet and is generally interpreted as a result of polymetamorphism (e.g. Ganne et al. , 2003 ; Gaidies et al. , 2008 ; Ye et al. , 2009 ; Chen et al. , 2013 a ). Alternatively, episodic metasomatism caused by fluid–rock interaction can also be responsible for such polyphase mineral growth (e.g. Stowell et al. , 1996 ; Compagnoni & Hirajima, 2001 ; Nabelek et al. , 2006 ; Li et al. , 2013 ). Fluid-derived vein networks, which are direct evidence of fluid-induced metasomatism of eclogites and blueschists, are commonly observed in the Akeyazi terrane (e.g. Gao & Klemd, 2001 ; Gao et al. , 2007 ; John et al. , 2008 , 2012 ; van der Straaten et al. , 2008 ; Beinlich et al. , 2010 ; Li et al. , 2013 ). Thus, it seems plausible that episodic metasomatic processes were responsible for polyphase mineral growth in the Akeyazi eclogite sample. However, petrological observations argue against this scenario.

Oscillatory-zoned garnet in HP rocks is the result of either episodic metasomatic processes with continuous growth but episodically changing composition of element supply (e.g. Yardley et al. , 1991 ; Stowell et al. , 1996 ) or cyclic tectonic instabilities (e.g. Garcia-Casco et al. , 2002 ; Kohn, 2004 ). In the present case, the growth features of garnet (two generations) do not support continuous growth during episodes of varying material supply, as the first generation garnet was overgrown by a compositionally different second generation garnet. Additionally, it is similar to oscillatory zoning observed in an eclogitic garnet from the northern serpentinite mélange in Cuba, which was interpreted to be the result of P – T fluctuations ( Garcia-Casco et al. , 2002 ). Furthermore, the REE patterns of the different epidote and garnet generations are almost identical (Ep_1 vs Ep_2; Supplementary Material Fig. SM4 ) or slightly different (Grt_1 vs Grt_2, Supplementary Data ), which is interpreted to be inconsistent with an episodic material supply. However, this is not regarded as robust evidence as trace elements do not always unambiguously act as effective indicator for poly- or poly-cyclic metamorphism owing to the fact that the element budgets may change during fluid infiltration.

The glaucophane–barroisite cyclic replacement in combination with polyphase garnet, epidote and sodic clinopyroxene does, however, provide compelling evidence for multiple P – T cycles. Episodic fluid-induced metasomatic processes cannot easily produce the cyclic transition between glaucophane and barroisite observed in the studied eclogite, as this is generally interpreted to be due to P – T fluctuations. However, the prerequisite is that glaucophane and barroisite display real cyclic replacement textural relationships, rather than being coexisting equilibrium phases. The existence of miscibility gaps between sodic and calcic amphiboles has been recognized for a long time, and the typical coexisting phases are glaucophane and actinolite ( Triboulet, 1978 ; Ernst, 1979 ; Reynard & Ballèvre, 1988 ; Smelik & Veblen, 1992 ). In contrast, textural relationships in samples containing glaucophane and barroisite suggest that the glaucophane typically did not grow at the same time as the barroisite ( Ernst, 1979 ) and the barroisite overgrowing glaucophane has typically been interpreted as a late replacement feature ( Reynard & Ballèvre, 1988 ). Therefore, the textural features (i.e. the mutual replacement or overgrowth) of glaucophane and barroisite rather than equilibrium textures (i.e. discrete adjacent grains with sharp, straight grain boundaries) are interpreted to represent cyclic transformations owing to P – T fluctuations.

In summary, the multistage growth of garnet and epidote, the polyphase transition of sodic clinopyroxene, and mutual overgrown amphibole in the eclogite sample are interpreted as a result of fluctuating P – T conditions instead of episodic metasomatic processes.

Metamorphic evolution and P – T path

By synthesizing the different segments of the P – T path from each of the pseudosections, a P – T path was reconstructed ( Fig. 8a ). The textural observations and mineral compositions described above, as well as the modeled P – T path, indicate that the studied eclogite sample underwent complex multistage eclogite-facies metamorphism. Six stages of metamorphic mineral growth are distinguished. The textures and mineral assemblages of each metamorphic stage are described below.

Stage I: prograde HP metamorphism (Subduction_1)

The earliest stage of metamorphism (stage I) recorded in the eclogite is represented by the cores of the garnet porphyroblasts (Grt_1) and their coarse-grained quartz, moderate- to fine-grained jadeite, omphacite (Omp_1), rutile and occasional pyrite inclusions ( Fig. 3a and c ). At this early stage, the modeled pseudosection indicates that the garnet crystallized at the expense of chlorite and Gln_1 ( Fig. 9a;Table 4 ). Significant amounts of lawsonite were present in the eclogite during this stage. This is in accordance with the presence of box-shaped epidote–paragonite inclusions in Grt_1 that are interpreted as pseudomorphs after lawsonite. It is also supported by the presence of relict lawsonite inclusions (associated with epidote–paragonite intergrowth) in garnet and dolomite in other Tianshan eclogites (e.g. Li et al. , 2013 , 2014 ; Du et al. , 2014 a ). Jadeite and omphacite were the main matrix phases before and at the onset of garnet growth ( Fig. 9a ). Coexisting jadeitic clinopyroxene and omphacite generally occur as a result of the presence of a temperature-dependent miscibility gap at a Jd _60–85 composition (e.g. Green et al. , 2007 ). However, the studied prograde jadeitic clinopyroxene and omphacite inclusions in garnet have compositions that do not overlap the Jd compositional range characteristic for immiscibility (e.g. Lin & Enami, 2006 ; this study), which was also observed in eclogites from other localities (e.g. Tsujimori et al. , 2005 ; Groppo & Castelli, 2010 ). Therefore, the equilibrium mineral assemblage of stage I is garnet + jadeite + omphacite + lawsonite + glaucophane + phengite + chlorite + quartz + rutile ( Fig. 9a ).

Fig. 9

Schematic P – T evolution of the poly-cyclic eclogite sample. Chlorite and lawsonite are inferred from pseudosection modeling and the other minerals are based on actual observations. (a) Stage I: start of garnet nucleation at the expense of chlorite and amphibole, corresponding to the prograde subduction stage. Significant amounts of lawsonite are present in the eclogite sample, whereas jadeite and omphacite are considered to be the main matrix phases. (b) Stage II: initial decompression stage corresponding to the growth of the Grt_1 mantle. Omphacite occurs as the main matrix phase. (c) Stage III: first exhumation stage recorded by the growth of the Grt_1 rim, Brs_1 replacing Gln_1 and the replacement of lawsonite by epidote. (d) Stage IV: cooling and re-subduction stage showing the initial crystallization of Grt_2, glaucophane (Gln_2) regrowth at the expense of barroisite; decomposition of epidote and reappearance of jadeite. (e) Stage V: second exhumation process, which is characterized by the continuous growth of Grt_2, barroisite (Brs_2) and Ep_2 overgrowths of Gln_2 and Ep_1, respectively. Omp_1 is the main matrix phase at this stage. (f) Stage VI: rapid isothermal decompression stage. Lawsonite is nearly completely consumed, rutile is replaced by titanite and the matrix omphacite is Omp_2.

Table 4: Major mineral assemblages of eclogite during multiple subduction–exhumation processes (numbers in parentheses refer to points in Fig. 10)

Stages	Dynamics ( Fig. 10 )	Garnet	Na-clinopyroxene	Amphibole	Epidote	Ti-phases
I	Subduction_1 (0–1)	Grt_1c	Jd + Omp_1	Gln_1		Rt
II–III	Exhumation_1 (1–3)	Grt_1m/r	Omp_1 + Jd + Aeg–Aug	Brs_1	Ep_1	Rt (+ Ttn)
IV	Cooling (3–4)	Grt_2c	Jd + Omp_1	Gln_2		Rt
IV	Subduction_2 (4–5)	Grt_2c	Omp_1	Gln_2		Rt (+ Ttn)
V–VI	Exhumation_2 (5–6)	Grt_2m/r	Omp_2	Brs_2	Ep_2	Ilm + Ttn (+ Rt)

The X_Grs isopleths in garnet are extremely P -dependent (i.e. X_Grs decreases with increasing pressure) whereas increasing temperature produces increasing X_Prp reflecting the T dependence of the X_Prp isopleths ( Fig. 8a ). Thus the Grt_1 core domain with decreasing grossular and increasing pyrope components towards the mantle domain constrains a prograde P – T path with a rapid P increase and a slight T increase, typical for burial during subduction (hereafter Subduction_1).

Stage II: initial decompression (Exhumation_1a)

The second metamorphic stage (stage II) is recorded by the mantle domain of Grt_1 and its inclusions, which are similar to those that occur in the Grt_1 cores ( Fig. 3a and d ). The abundant inclusion distribution in Grt_1 at stage I and II is believed to be a consequence of high garnet growth rates. This stage reveals slight increasing pyrope and grossular components in garnet ( Figs 6 and 7 a) suggesting an initial decompression process with subtle heating under HP conditions ( Supplementary Material Fig. SM8b ; Fig. 8a ). At this stage, omphacite is starting to dominate the matrix ( Fig. 9b ), and the glaucophane (Gln_1) replacement by barroisite (Brs_1) and the lawsonite consumption by garnet and epidote were initiated ( Fig. 9b ). Therefore, the equilibrium mineral assemblage of this stage is garnet + omphacite + jadeite + aegirine–augite + lawsonite (± epidote) + glaucophane (± barroisite) + phengite (± chlorite) + quartz + rutile ( Fig. 9b;Table 4 ).

The growth of Grt_1 is interpreted to be reaction-controlled as indicated by the subtle variations of almandine, grossular and pyrope components ( Fig. 7a ). However, Mn and the MREE are most probably more supply controlled owing to fractional crystallization (e.g. Hollister, 1966 ; Hickmott et al. , 1987 ).

Stage III: first retrograde metamorphism (Exhumation_1b)

This metamorphic stage (stage III) is represented by the almost inclusion-free rim of Grt_1 ( Fig. 3a ). The rare occurrence of inclusions demonstrates that the Grt_1 growth rate is slow and chemical equilibrium was probably fully achieved at this stage. Two garnet domains are distinguished by the behavior of the pyrope component, which increases in the inner rim and decreases in the outer rim ( Figs 6b and 7a ), whereas the grossular component increases continuously from the inner to the outer rim ( Figs 6c and 7a ). Accordingly, decompression and simultaneous heating to the maximum temperature and subsequent cooling during decompression are revealed on the basis of garnet isopleth thermobarometry ( Supplementary Material Fig. SM8c ).

Subsequently, the eclogite sample may have been rapidly exhumed until reaching the epidote stability field (e.g. 1·7 GPa at 520 °C, Fig. 8 ) as is indicated by the occurrence of Ep_1 in both Grt_2 and the matrix ( Figs 3e and 5a, b ). During Exhumation_1, lawsonite was significantly decomposed to epidote (Ep_1) and Grt_1, as revealed by the rapid increase of Ca in the Grt_1 rim domain ( Figs 6c and 7a ). Glaucophane (Gln_1) is probably also involved in the breakdown of lawsonite, and is almost completely replaced by barroisite (Brs_1; Figs 4 and 9 c). Jadeite and Aeg–Aug, which occur only as minor relics, are almost completely replaced by Omp_1 ( Fig. 9c ). Based on these observations, the mineral assemblage in stage III is garnet + omphacite + epidote (± lawsonite) + barroisite (± glaucophane) + phengite + quartz + titanite (± rutile) ( Fig. 9c;Table 4 ). The transition of Jd/Aeg–Aug to Omp_1 and the high Ca content of the Grt_1 rim require the introduction of CaO that may have been liberated during the breakdown of lawsonite, whereas the released FeO is incorporated into Ep_1.

The LREE that were released during lawsonite breakdown are usually incorporated into epidote ( Tribuzio et al. , 1996 ). As a result of elemental fractionation, Mn, Y and MREE concentrations in the garnet decrease rapidly from the inner rim towards the outer rim of Grt_1 ( Fig. 7a and Supplementary Material Fig. SM4a, b and e ), whereas the HREE (e.g. Yb and Lu) concentrations are generally flat ( Supplementary Data ). This is in contrast to the well-accepted ‘bell-shaped’ growth zoning model for garnet, which indicates HREE enrichment towards the core owing to fractionation (e.g. Hickmott et al. , 1987 ; Hickmott & Shimizu, 1990 ). The zoning pattern of Grt_1 requires further study, as it may have been the consequence of changing garnet–matrix HREE partition coefficients during garnet growth (e.g. Hickmott & Spear, 1992 ; Schwandt et al. , 1996 ). However, this is beyond the scope of this work.

Stage IV: recompression (Cooling and Subduction_2)

This metamorphic stage (stage IV) is recorded by the compositions of the thin Grt_2 (in Grt-A) core and its fine-grained jadeite, omphacite (Omp_1) and quartz inclusions ( Fig. 3a, e and f ). Grt_1 and Grt_2 in Grt-A are separated by rather sharp boundaries (textural discontinuities), although their concentric compositional zoning indicates continuous growth ( Fig. 7a ). This contrasts with the pelitic schists in the Western Alps, in which garnet with textural and chemical discontinuities represents distinct pre-Alpine and subsequent Alpine metamorphic events (e.g. Ganne et al. , 2003 ; Gaidies et al. , 2008 ). The pyrope and grossular components of the Grt_2 core decrease first, and then increase towards the rim ( Fig. 7a ). Based on the pseudosection modeling and the isopleth calculation, the variations of X_Prp and X_Grs suggest an increasing pressure at a decreasing temperature ( Fig. 8b ), thereby probably revealing a short burial period (Subduction_2). This is followed by a temperature increase at a constant pressure indicating a short thermal relaxation in the subduction zone. During this stage Grt_2, which encloses Grt_1 (Grt-A) and also forms xenoblastic Grt-B grains ( Fig. 9d;Supplementary Material Fig. SM1 ), experiences rapid growth. As a consequence of the temperature decrease, epidote starts to decompose and lawsonite appears ( Fig. 9d ). In addition, glaucophane (Gln_2) crystallizes at the expense of barroisite ( Fig. 4 ) and some jadeite reappears, probably at the expense of omphacite ( Fig. 3f ). Therefore, the mineral assemblage of this stage is similar to that of stage II with some modifications including two types of garnet, multiple occurrences of glaucophane and a matrix dominated by Omp_1 ( Fig. 9d ). It is noteworthy that at this stage the Mn content in the Grt_2 core increases (from Grt_1 to Grt_2 in Grt-A) slightly ( Fig. 7a ). Generally, an Mn increase is considered as a result of intracrystalline diffusion by garnet resorption (e.g. Florence & Spear, 1991 ; Kohn & Spear, 2000 ), external Mn introduction by fluid infiltration or breakdown of Mn-rich minerals during garnet growth (e.g. Tracy, 1982 ). No indication of garnet resorption was observed in this sample and the temperature is rather low (<530 °C) for sufficient diffusion. In addition, external fluid infiltration seems unlikely (see discussion above). Therefore it is thought that the slight Mn increase at this stage is due to the breakdown of Mn-bearing minerals such as barroisite (285 ppm Mn in Brs_1) and their replacement by glaucophane (64 ppm Mn in Gln_2) ( Supplementary Material Table SM1 ).

Stage V: heating and decompression (Exhumation_2)

This exhumation period (stage V) is defined by the mantle increments of Grt_2 and its inclusions; that is, omphacite (Omp_1), epidote (Ep_1), quartz, ilmenite and titanite ( Fig. 3a, e and f ). The pyrope and grossular components continuously increase towards the rim in Grt_2 and are thus interpreted to display a decompressional heating process ( Supplementary Material Fig. SM8e ), during which lawsonite starts to decompose. Furthermore, barroisite (Brs_2) crystallizes at the expense of glaucophane (Gln_2), and Ep_2 replaces Ep_1 ( Figs 4, 5a, b and 9e). Jadeite disappears at this stage and Omp_1 is the main matrix mineral ( Fig. 9e ). This is supported by the wide distribution of Omp_1 inclusions in Grt_2 and Ep_2 ( Figs 3e and 5a, b ). Therefore the mineral assemblage of this stage is thought to be garnet + omphacite + epidote (± lawsonite) + barroisite (± glaucophane) + phengite + quartz + ilmenite/titanite (± rutile) ( Fig. 9e;Table 4 ). The Cpx transformation releases FeO that may have been incorporated by Ep_2 as indicated by its high Fe content ( Table 1 ).

Stage VI: second retrograde metamorphism (Exhumation_2)

The last stage of exhumation is characterized by the Grt_2 rims and the equilibrium mineral assemblage ( Table 4 ) that occurs as mineral inclusions in the rims and as matrix minerals ( Fig. 9f ). At this stage, lawsonite almost disappears from the equilibrium mineral assemblage, and Omp_2 and some Omp_1 (with or without Aeg–Aug) relicts occur as matrix phases ( Fig. 5a;Supplementary Material Fig. SM2 ). Titanite appears at the expense of rutile, in accordance with Omp_2 and Ep_2 inclusions in the titanite ( Fig. 5c ).

This stage is probably followed by a rapid isothermal decompression, which is commonly recorded in the Tianshan eclogite-facies rocks (e.g. Klemd et al. , 2002 ; Wei et al. , 2009 ; Li et al. , 2012 ; Tian & Wei, 2013 ).

Multiple burial and exhumation cycles along the plate interface

The P – T path constraints from thermodynamic modeling show that the eclogite sample underwent a multistage burial and exhumation evolution during its overall ascent along the slab interface ( Fig. 10 ). Although the modeled P – T estimates used for the reconstruction of the P – T trajectory may slightly overlap within the uncertainties of the thermobarometric calculation (±25 °C and ±0·18 GPa), the overall shape of the P – T trajectory is robust and reflects a multiple burial and exhumation history. This conclusion is supported by the alternating glaucophane and barroisite replacements as recorded by single grains (transitions of Gln_1→Brs_1→Gln_2→Brs_2; Fig. 4 ), and the multistage growth of garnet and epidote.

Fig. 10

Schematic illustration of the multiple subduction and exhumation processes in the subduction channel. The critical points (kink points of the P – T loops) are labelled 1–6. The thickness of the subduction channel is greatly exaggerated in comparison with the thickness of the overriding blocks. The downgoing oceanic crust is subducted to mantle depths and undergoes HP–UHP metamorphism. The metabasaltic slices are detached into the subduction channel and exhumed upwards within the buoyant sediments towards the hot mantle wedge, where the temperature is higher. These blocks or slices are recoupled to the descending oceanic crust and overriding sediments to undergo the second subduction process. Finally, the return channel flow carries the eclogite sample upwards to undergo the second exhumation process and return to a shallow level in the crust.

Subduction-related multiple burial and exhumation cycles have previously been reported for blueschists and eclogites in other metamorphic terranes (e.g. Beltrando et al. , 2007 ; Harris et al. , 2007 ; Umhoefer et al. , 2007 ; Kabir & Takasu, 2010 ; Blanco-Quintero et al. , 2011 ). These processes have also been described as pressure cycles ( Beltrando et al. , 2007 ) or Yo-Yo subduction process ( Rubatto et al. , 2011 ; Regis et al. , 2014 ). In general, burial–exhumation cycles involve three scenarios in subduction-related settings, based on their temporal and spatial characteristics: (1) two separate burial–exhumation cycles produce ‘polymetamorphic HP rocks’ of a slab or terrane owing to two separate orogenic events (e.g. Herwartz et al. , 2011 ; Wakabayashi, 2012 ); (2) the subducted slab underwent two (or more) burial–exhumation cycles during a single orogenic event as a result of switches between tectonic shortening and extension, thereby producing ‘poly-cyclic metamorphism’ ( Lister et al. , 2001 ; Brueckner, 2006 ; Beltrando et al. , 2007 ; Harris et al. , 2007 ); (3) tectonic slices detached from the subducted slab may be coupled by corner flow in the subduction channel and may have undergone partial burial–exhumation cycles as displayed by ‘poly-cyclic metamorphism’ along the subduction interface ( Garcia-Casco et al. , 2002 ; Kabir & Takasu, 2010 ; Blanco-Quintero et al. , 2011 ; Rubatto et al. , 2011 ; Regis et al. , 2014 ). The first scenario is improbable for the eclogite from our study, as the South Tianshan Orogen has been interpreted to have formed as a result of a single orogenic event during late Paleozoic times ( Gao et al. , 1998 , 2009 ; Klemd et al. , 2011 , 2015 ). In addition, the Tianshan eclogite-facies rocks show a well-constrained peak metamorphic garnet age (315 ± 5 Ma), excluding the possibility of two orogenic events (see Klemd et al. , 2011 ). Scenario 2 is characterized by tectonic shortening–extension switches that are generally accompanied by sudden collapses and surges of an orogenic belt (e.g. Beltrando et al. , 2007 ). This is not compatible with the continuous tectonic evolution of the South Tianshan orogen ( Gao et al. , 1998 , 2009 ; Wang et al. , 2011 b ). Therefore, the multiple burial–exhumation events observed in the studied sample are considered to be a result of subduction channel dynamics (scenario 3). Similar burial–exhumation cycles recorded by polyphase textures of garnet and amphibole were reported from poly-cyclic eclogites in the northern serpentinite mélange in Cuba ( Garcia-Casco et al. , 2002 ; Blanco-Quintero et al. , 2011 ) and the Sambagawa metamorphic belt in Japan ( Kabir & Takasu, 2010 ), both of which have been interpreted as a result of large-scale convective movement of tectonic blocks in a serpentinite subduction channel (e.g. Blanco-Quintero et al. , 2011 ). Additionally, multiple burial–exhumation events revealed by tectonic units or detached tectonic slices are also believed to be consistent with subduction channel tectonics as suggested by numerical simulations ( Gerya et al. , 2002 ; Gerya & Stöckhert, 2006 ; Li & Gerya, 2009 ).

In subduction zones, the downgoing slab is subducted to mantle depths. Mechanical coupling between the plates and dehydration reactions may result in weakening and fracturing of the surface of the descending upper crustal section (e.g. John et al. , 2008 , 2010 ; Konrad-Schmolke et al. , 2011 ; Warren, 2013 ). This could cause decoupling and detachment of metabasaltic slices or tectonic units that are transferred into a subduction channel-like structure along the plate interface (e.g. Gerya et al. , 2002 ; Warren et al. , 2008 ; Agard et al. , 2009 ; John et al. , 2010 ). In this study the detachment occurs at a depth of c . 75 km in the subduction zone (point 1; Fig. 10 ). Subsequently, detached metabasic crustal slices became exhumed within weak serpentinites and sediments ( van der Straaten et al. , 2008 ) and/or low-viscosity material between the overriding and the subducting plates as a result of buoyancy and underplating during continuing subduction. The temperature increases during the upward motion (point 2) of the sample towards the hot mantle wedge until a shallower depth at c . 60 km (point 3, or even shallower to c . 45 km) is reached. This exhumation process results in the replacement of Gln_1 by Brs_1 and of lawsonite by Ep_1. However, such metabasaltic blocks or slices—as revealed by the thermodynamic modeling of the studied sample—may become coupled again with the continuously subducting slab and buried for a second time (points 4 and 5; Fig. 10 ). This process may lead to the second HP metamorphic event as recorded by the growth of the core increment of Grt_2 and the second increment of glaucophane (Gln_2). Subsequently, the positive buoyancy of material in the subduction channel may carry the eclogite upwards for a second time (point 6; Fig. 10 ) and finally to a shallow level in the crust. Numerical simulations have confirmed the potential for channel flow convection at great depth during continuing subduction ( Gerya et al. , 2002 ), and modeled P – T trajectories reveal ‘yo-yo’ tectonics in subduction channels ( Gerya et al. , 2002 ; Gerya & Stöckhert, 2006 ; Li & Gerya, 2009 ; Blanco-Quintero et al. , 2011 ).

Further implications for the exhumation mechanism of the Akeyazi (U)HP/LT rocks

The eclogite sample in our study appears to have undergone multiple burial and exhumation processes along the subducting plate interface. This may shed some light on the exhumation mechanism of intimately interlayered HP and UHP rocks elsewhere in the AMT (e.g. Klemd et al. , 2002 , 2015 ; Wei et al. , 2003 ; Lü et al. , 2008 , 2009 ; Li et al. , 2012 , 2015 ; Tian & Wei, 2013 ).

The subduction channel model is interpreted to be consistent with the interlayering of HP and UHP rocks and their partial back-transformation into blueschists in the AMT ( van der Straaten et al. , 2008 ; Klemd et al. , 2011 , 2015 ; Li et al. , 2015 ). According to this model, the different slices should show variable P – T paths and ages, and the overall arrangement of the blocks and the matrix should reveal great structural complexities (e.g. Hacker & Gerya, 2013 ); in addition, pressure cycling of single domains or samples is to be expected (e.g. Gerya et al. , 2002 ; Gerya & Stöckhert, 2006 ; Hacker & Gerya, 2013 ). The following features are all in accordance with the Tianshan tectonic mélange concept ( Gao et al. , 1999 ; Gao & Klemd, 2003 ): (1) the chaotic juxtaposition of HP and UHP rocks ( Fig. 1 b ) reveals the structural complexity in the subduction channel; (2) the P – T paths of different Tianshan eclogite-facies rocks vary significantly (see Li et al. , 2015 , fig. 10 ); (3) the various peak metamorphic ages of the AMT eclogite-facies rocks range mainly between 320 and 310 Ma (e.g. Du et al. , 2014 b ; Klemd et al. , 2015 ); furthermore, (4) the investigated eclogite sample shows a poly-cyclic evolution interpreted to have been caused by multiple burial and exhumation cycles (pressure cycling), which is a likely characteristic of a subduction channel (e.g. Cloos, 1982 ; Gerya et al. , 2002 ; Blanco-Quintero et al. , 2011 ; Hacker & Gerya, 2013 ). Therefore, the subduction channel model is proposed to explain the most likely exhumation mechanism of the HP and UHP rocks from the AMT. This scenario is in agreement with the initial subduction of oceanic crust and accretionary wedge sediment removal by subduction erosion ( Liu et al. , 2014 ) and is therefore similar to that for accretionary-type subduction terranes, whereby peak metamorphic conditions are much higher than previously recorded (see Klemd et al. , 2015 ).

The multiple mineral growth, poly-cyclic metamorphism and the very complex P – T path of the studied eclogite sample appear to be completely different from those described for all previously studied eclogite samples from the Akeyazi (U)HP metamorphic terrane. However, the proposed tectonic model is in good agreement with that of previous studies, which favored the juxtaposition of tectonic blocks with different P – T paths and metamorphic ages during exhumation in the subduction channel. In general, the eclogitic blocks, which were exhumed in the subduction channel, show variable P – T paths and only a small number of samples may experience burial–exhumation cycles and record pressure cycling ( Gerya et al. , 2002 ). That is probably the reason why poly-cyclic metamorphism has not been observed up to now in previous studies of the (U)HP rocks in this metamorphic terrane. However, lawsonite-bearing eclogites containing distinct core–rim structures (with sharp boundaries) and glaucophane overgrowth of barroisite (see Du et al. , 2014 a , fig. 2 ) seem to indicate that more (U)HP samples from the Akeyazi metamorphic terrane underwent poly-cyclic metamorphism.

The preferred matrix material within the channel-like structure along the plate interface is thought to be serpentinite, which has a low viscosity as a result of mantle wedge hydration in the upper plate by fluids released from the subducted oceanic crust and overlying sediments (e.g. Cloos, 1982 ; Guillot et al. , 2001 ; Gerya et al. , 2002 ; Federico et al. , 2007 ; Agard et al. , 2009 ). This is compatible with the presence of serpentinite mélanges in the Alps ( Federico et al. , 2007 ), Cuba ( Blanco-Quintero et al. , 2011 ) and the Dominican Republic ( Krebs et al. , 2008 , 2011 ), where the subduction channel model was successfully identified as an effective exhumation mechanism for HP–UHP rocks. However, serpentinite occurrences in the Akeyazi metamorphic terrane are very rare; almost all mafic blueschists and eclogites occur in a matrix of metasediments (see Klemd et al. , 2015 ). Thus the main concern is whether such metasedimentary material has similar rheological characteristics to serpentinite. Unfortunately, little is known about the rheology of matrix material in subduction channels. In the original papers in which the subduction channel concept was described ( Cloos, 1982 ; Cloos & Shreve, 1988 ), the weak matrix was proposed to consist of sediments. Both sediments and serpentinites were considered to have a low density and a constant low viscosity in the numerical modeling of subduction channel processes by Gerya et al. (2002) . Consequently, in the case of the present study it is suggested that the metasedimentary matrix material shows similar rheological characteristics to serpentinite and is thus thought to have been responsible for the exhumation of eclogitic blocks in the subduction channel (e.g. Malatesta et al. , 2012 ). This is supported by the occurrence of eclogitic boudins with no or little deformation within the metasedimentary matrix. In addition, the post-peak metamorphic P – T paths of both eclogite and blueschist boudins and metasedimentary host-rocks (HP and UHP) are identical, showing near-isothermal decompression. This indicates that the mafic boudins and their metasedimentary host-rocks shared a common rapid exhumation path owing to the buoyancy of the metasedimentary matrix in the subduction channel ( Wei et al. , 2009 ; Klemd et al. , 2011 , 2015 ).

CONCLUSIONS

(1) Petrological evidence suggests that the studied Tianshan eclogite sample underwent poly-cyclic metamorphism and multistage mineral growth. Both garnet and epidote show multistage growth with textural discontinuities and chemical variations. Amphibole displays alternating glaucophane and barroisite transitions (Gln_1 → Brs_1 → Gln_2 → Brs_2), indicating a complex burial–exhumation evolution of the eclogite sample.

(2) Based on the effective bulk compositions, thermodynamic modeling shows that the eclogite sample underwent a complex P – T path composed of six segments corresponding to different increments of the garnet, including initial burial (Grt_1 core) and ascent (Grt_1 mantle), subsequent exhumation (Grt_1 rim), cooling and re-burial (Grt_2 core), and final exhumation (Grt_2 mantle and rim).

(3) The burial–exhumation cycles, which are recorded by the petrographical–petrological characteristics of the eclogite and confirmed by pseudosection modeling, are interpreted to be a result of a complex burial and ascent of metabasaltic blocks or slices within a subduction channel-like plate interface.

(4) The pressure cycling of the eclogite sample is in accordance with the previously suggested channel flow model for the Akeyazi (U)HP belt, which is characterized by the juxtaposition of HP and UHP rocks, various ages and P – T paths, and multiple burial–exhumation cycles.

ACKNOWLEDGEMENTS

T. Jiang is thanked for assistance in sample preparation. The authors are grateful to M. Hertel for help with the XRF analyses, to H. Brätz for assistance with the LA-ICP-MS whole-rock analyses, and to M. Meyer for help with the electron microprobe analyses. We thank M. Meyer and Y. D. Sun for helpful discussions. J. Hermann is thanked for his editorial handling and helpful suggestions. Thorough and constructive reviews by D. Regis, H. P. Schertl and an anonymous reviewer are highly appreciated and substantially improved the paper.

FUNDING

This work was supported by the National Natural Science Foundation of China (41390445, 41502053), China Postdoctoral Science Foundation (2015T80135, 2014M560114) and the Deutsche Forschungsgemeinschaft (KL692/17-3). This publication is a contribution to IGCP Project 592 and the Marie Curie ITN ZIP.

SUPPLEMENTARY DATA

Supplementary data for this paper are available at Journal of Petrology online.