Phase equilibrium modelling and implications for P–T determinations of medium‐temperature UHP eclogites, North Qaidam terrane, China

In contrast to low‐T eclogites (garnet growth zoning preserved) or high‐T eclogites (garnet diffusionally homogenized at peak conditions), medium‐temperature eclogites pose additional challenges to P–T determinations due to the partial preservation of garnet zoning. The Dulan area, in the southeastern part of North Qaidam ultrahigh‐pressure terrane, exposes minor eclogites hosted by ortho‐ and paragneisses. Four fresh, medium‐temperature eclogites contain the paragenesis Grt+Omp+Rt+Qz/Coe+Ph±Ky±Zo. In all samples, garnet XMg shows little zoning, suggesting diffusional modification, and precludes the use of pyrope+almandine+grossular isopleth intersections to determine a P–T path. However, in one sample, sharp zoning in grossular content suggests that grossular growth compositions are preserved. Since garnet pyrope+almandine compositions appear to be modified, we instead use the intersections of grossular and garnet volume isopleths to define a prograde P–T path. This approach yields a path from ~17 kbar and ~410°C to ~35 kbar and ~625°C with a gradient of ~5–9°C/km through the lawsonite stability field. Peak P–T conditions determined from the intersection between Si pfu in phengite and Zr‐in‐rutile isopleths are ~26–33 kbar and ~625–700 °C for the four eclogites. These conditions are close to the limit of the lawsonite stability field, suggesting that fluid released from lawsonite breakdown may have promoted re‐equilibration at these conditions. These peak conditions are also in good agreement (within 3 kbar and 50°C) with garnet–omphacite–phengite (±kyanite) thermobarometry in three of the four samples. We regard the phengite–rutile constraints as more reliable, because they are less sensitive to uncertainties associated with ferric iron compared to conventional thermobarometry. Phase equilibrium modelling predicts that the retrograde assemblage of amphibole+zoisite formed at ~60 km. Infiltration of external fluids was likely required for the growth of these hydrous minerals. Based on the comparison of P–T estimation methods applied in this study, we propose that the garnet grossular+volume isopleths can recover the prograde P–T path of medium‐temperature eclogites, and that the combination of phengite+rutile isopleths represents a more robust approach to constrain peak P–T conditions.


K E Y W O R D S
Dulan area, North Qaidam UHP terrane, phase equilibrium modelling, UHP eclogites, Zr-in-rutile thermometry 1 | INTRODUCTION Ultrahigh-pressure (UHP) eclogites are products of subduction to mantle depths (Chopin, 2003;Gilotti, 2013). Mineral assemblages preserved in these rocks record metamorphism taking place in the subducting slab which can be used to place constraints on the P-T evolution of UHP terranes, leading to the better understanding of largescale geodynamic processes (e.g., Hacker, Abers, & Peacock, 2003;Poli & Schmidt, 2002). However, mediumtemperature eclogites (MT; 650-750°C) pose additional challenges to P-T determinations due to the partial preservation of garnet zoning, in contrast to low-T eclogites (garnet growth zoning preserved) or high-T eclogites (garnet diffusionally homogenized at peak conditions).
Phase equilibrium diagrams have proven to be a powerful approach to obtain robust P-T information and elucidate phase relations in metabasic rocks (e.g., Rebay, Powell, & Diener, 2010;Wei & Clarke, 2011). However, the relatively simple mineral assemblages typical of eclogites (i.e., gar-net+omphacite) represent a challenge for phase equilibrium modelling and conventional thermobarometry because the compositions and proportions of minerals changes slowly within the garnet+omphacite field (>17 kbar; e.g., Groppo, Rolfo, Liu, Deng, & Wang, 2015;Hacker, 2006;Krogh Ravna & Terry, 2004;Štípská & Powell, 2005). Trace element thermometry is independent of the data, assumptions, and uncertainties of phase equilibrium modelling and conventional thermobarometry, and therefore can improve the precision of the peak temperature constraints. In contrast to peak P-T evidence, the prograde P-T record is more susceptible to diffusional modification of the mineral compositions by elevated temperatures at peak stages, but slow diffusing elements should better preserve prograde evidence (Caddick, Konopásek, & Thompson, 2010;Konrad-Schmolke, Handy, Babist, & O'Brien, 2005).
In this paper, we evaluate and discuss the applicability of phase equilibrium modelling, conventional thermobarometry, and Zr-in-rutile thermometry to MT-UHP eclogites. We present a detailed petrological study of four eclogites from the Dulan UHP area in the North Qaidam terrane, explore the prograde and peak P-T evolution of the eclogites, and discuss the tectono-metamorphic evolution of the area.

| GEOLOGICAL SETTING
The North Qaidam UHP terrane is a~350 km NW-SEtrending terrane, represents an early Palaeozoic continental suture, and preserves UHP evidence in eclogites, peridotites, and quartzofeldspathic gneisses at multiple locations (Song et al., 2003(Song et al., , 2014Zhang et al., 2017). This terrane is bounded on the northeast by the Qilian terrane, and on the southwest by the Qaidam basin ( Figure 1a; Mattinson et al., 2007). The North Qaidam eclogites are hosted in the Proterozoic Dakendaban Group gneiss (Mattinson et al., 2007;Song et al., 2003). This unit comprises a sequence of metasedimentary and metaigneous rocks locally intruded by granite and granodiorite (Mattinson et al., 2007).

| Analytical methods
To constrain the metamorphic evolution of the Dulan area, four eclogites were selected that contain well-preserved peak mineral assemblages, textures, and assemblages suitable for thermodynamic modelling (i.e., including kyanite and/or phengite; Figure 1b). Mineral proportions were obtained by image analysis of BSE mosaics of whole thin sections using JMicroVision software (Roduit, 2008), and were refined with visual estimations where the BSE brightness is similar for different minerals (e.g., kyanite, quartz). X-ray maps and major element mineral chemistry analyses were performed using a five-spectrometer JEOL JXA-8500F field emission electron microprobe housed at Washington State University. Well-characterized natural and synthetic materials were used as standards for calibration, and a ZAF correction routine was applied. The operating conditions were: accelerating voltage of 15 keV, beam current of 20 nA, and beam diameter of 10 μm for mica and 3 μm for the rest of the minerals.

| Results
Dulan eclogites commonly present a well-developed eclogitic foliation formed by oriented garnet, omphacite,    Figure 3. The white box in h shows the location of the garnet in Figure 6. p: pseudomorph rutile, and in some samples kyanite, zoisite, phengite, and amphibole as well ( Figure 2). Fractures filled with secondary quartz, calcite, and/or epidote cut the foliation, and are associated with alteration of the primary eclogitic mineral assemblage to fine grained symplectites ( Figure 2).
Garnet occurs as xenomorphic elongated inclusion-poor crystals up to~1 mm long (Figure 2a). Garnet is typically clustered with omphacite, rutile, and other garnet. In some garnet clusters (up 2.5 mm in size), individual grains are indistinguishable in the X-ray maps (Figure 2a). Most garnet is compositionally homogeneous, but a few crystals show moderate zoning, with increasing Prp (~20-24%) and Alm (~40-50%), and decreasing Grs (~38-25%) from core to rim; Sps and X Mg show little zoning ( Figure 3a; Table 1).
in kyanite. In some areas, zoisite appears to be growing along the grain boundaries within the kyanite (Figure 2b). Zoisite presents first order interference colours, and some grains display anomalous blue interference colours. Scarce crystals have oscillatory zoning. Some zoisite grains contain quartz pseudomorphs after coesite ( Figure 4a).
Hypidiomorphic omphacite varies in size from 0.5 to 1.5 mm long (Figure 2e inclusions in omphacite and garnet; matrix crystals are completely replaced by fine-grained symplectites of Bt+Pl ( Figure 2e). Phengite yields Si pfu = 3.52-3.53, X Mg = 0.92-0.94, and X Na = 0.04-0.05 ( Figure 5b; Table 1). Epidote (X Fe 3+ = 0.06-0.15; Figure S1 and Table S1) occurs as oriented hypidiomorphic to xenomorphic crystals up to~250 μm long ( Figure 2f). Epidote appears in the matrix; in some areas, it is unclear if epidote is sited on grain boundaries within garnet clusters or if it forms inclusions in garnet ( Figure 2f). Most epidote is inclusion-free, presents first-order interference colours, and some grains display anomalous blue interference colours. Scarce grains contain omphacite and garnet inclusions. Some crystals display strong compositional zoning with low-Ca and high-X Fe 3+ cores, and high-Ca and low-X Fe 3+ rims ( Figure S1 and Table S1). Low-Ca content in the crystallographic A-site suggests the presence of REE and/or Sr. Amphibole occurs as intergranular crystals 500-750 μm long, typically replacing omphacite (Figure 2e,f). Amphibole in this eclogite is hornblende and pargasite ( Figure S1 and Table S1).
Garnet occurs as idiomorphic porphyroblasts 300 μm to 1.5 mm long (Figure 2g,h). Most grains contain phengite, amphibole, and epidote inclusions (Figure 2g,h). Inclusions are constrained to the garnet core and mantle (Figure 2g,h). Garnet presents strong chemical zoning, with an overall increase of X Mg (~0.29-0.41%), Prp (~18-32%), and Alm (~44-48%), and decrease of Grs (~34-23%) and Sps (~2-0.8%) from core to rim ( Figure 6; Table 1). Within the mantle, garnet shows a zoning reversal characterized by an increase of grossular and a decrease of pyrope ( Figure 6 and Figure S2). Compositional cross-cutting relations (i.e., zoning sharply truncated by resorption surfaces followed by garnet overgrowth of different composition) suggest two phases of garnet growth each bounded by a resorption event ( Figure S2). Garnet can be separated into two groups. Garnet 1 includes the grossular-rich core and the zoning reversal in the mantle; these zones are truncated by resorption surface 1. Garnet 2 forms grossular-poor rims; resorption surface 2 locally truncates both garnet 1 and garnet 2 ( Figure S2). Omphacite grains are in contact with both resorption surfaces suggesting that the resorption events happened during eclogite facies conditions ( Figure S2).

F I G U R E 6
Selected garnet X-ray compositional maps and corresponding core-rim profile from eclogite D161B. Warm and cold colours correspond to high and lower concentrations respectively. Colour scale is not the same between the different maps. Black box shows garnet overgrowth with different composition (see text). The location of the X-ray map is shown in a white box in Figure 2h. The green shaded region "a" on the profile corresponds to the "a" point shown in Figure S6. The black arrows on the profile delimit the two different garnet phases (see text and Figure S2) Omphacite occurs as hypidiomorphic crystals 500-700 μm long (Figure 2g,h). This mineral is partially replaced by symplectites of Cpx+Pl along grain boundaries (Figure 2g,h). Omphacite yields X Mg = 0.84-0.91, Jd = 0.28-0.36, and Acm = 0.01-0.03 ( Figure 5a; Table 1).
Amphibole occurs as small grains and porphyroblasts 500-700 μm long. Two inclusions in garnet are barroisite and pargasite, and amphibole in the matrix is hornblende, typically replacing omphacite ( Figure S1 and Table S1).

| Methods
In order to understand the metamorphic evolution of eclogites from the Dulan area, phase equilibrium modelling was coupled with Zr-in-rutile thermometry and conventional thermobarometry. For samples D197, D205, and D167, fracturing, resorption and recrystallization, and diffusion likely affected garnet (see discussion) obscuring the prograde P-T path, but peak conditions and constraints on the retrograde conditions were obtained. For sample D161B, a prograde P-T path was obtained using the methodology described below. For all samples, peak conditions were determined using the intersection of the maximum Si pfu in phengite isopleths and Zr-in-rutile isopleths. The P-T area showing the observed amphibole and epidote volume is interpreted to be the conditions where the rock stopped recording P-T information.

| Calculation parameters for phase equilibrium diagrams
Phase equilibrium diagrams were constructed using Per-ple_X 6.7.4 (version October 2015; Connolly, 1990Connolly, , 2009 and the internally consistent thermodynamic data set tc-dS55 updated August 2004). Modelling was performed in the 10-component MnO- system, using the following activitycomposition (a-x) relations for solid solution phases: garnet, epidote, and talc ; biotite (Powell & Holland, 1999); chlorite (Holland, Baker, & Powell, 1998); white mica (Coggon & Holland, 2002); amphibole (Dale, Powell, White, & Elmer, 2005); and omphacite (Diener & Powell, 2012). Additional pure phases included were lawsonite, kyanite, quartz/coesite, zoisite, and albite. Water was considered to be in excess due to the abundance of hydrous prograde, peak, and retrograde minerals. We acknowledge that the upper temperature limit used in the models may exceed the wet solidus for metabasic rocks, as demonstrated in recently parameterized a-x relations for solid solution phases (Green et al., 2016). However, the Green et al. (2016) melt model was calibrated for use at crustal P-T conditions (<13 kbar), thus, is unsuitable for examining phase equilibria in the eclogite facies. Furthermore, there is no evidence of partial melting in the studied samples. Uncertainties in the location of the mineral assemblage boundaries of the most P-T sensitive assemblages are ±1 kbar and ±50°C (Palin, Weller, Waters, & Dyck, 2016;Powell & Holland, 2008).
The bulk compositions used to perform the phase equilibrium modelling were obtained from XRF, and from modal abundances estimates combined with microprobe analyses (i.e., reconstructed bulk-rock compositions; Table 2). Whole-rock XRF major element data were obtained with a ThermoARL XRF spectrometer housed at Washington State University. The XRF whole-rock compositions were modified to account for the following scenarios: (a) Minerals that were not included in the modelling but their compositions have an important contribution to the whole-rock composition (i.e., apatite). The phosphorus content was decreased to 0%, and CaO was decreased a corresponding amount assuming that phosphorus is bound exclusively in apatite. (b) The effect of ferric iron (Fe 3+ ) on the stability of the mineral assemblages (also modified in reconstructed bulk-rock compositions). Mineral compositions, and T-X and P-X diagrams (X Fe 3+ = 0-0.5) were used as a guide to constrain the oxidation state of the samples . A value of X Fe 3+ = 0.05 was used to construct all diagrams based on satisfactory reproduction of the observed mineral assemblages and compositions of the samples (Table 1). The prograde path was constructed for sample D161B by modelling the garnet growth, using the intersection between modelled garnet volume isopleths and its corresponding grossular content. Because of the low Ca diffusion rate, its concentration is more likely to represent the original concentration in garnet than Fe, Mg, or Mn (Carlson, 2006;Carlson, Pattison, & Caddick, 2015;Chakraborty & Ganguly, 1992;Chernoff & Carlson, 1997;Florence & Spear, 1991;Giuntoli, Lanari, & Engi, 2018). To construct the P-T path, a garnet was selected with a representative concentric compositional zoning from sample D161B, using different X-ray maps and different garnet core-rim compositional profiles ( Figure S2). The selected garnet is shown in Figures 2h and 6, and Figure S2. Each measured grossular composition was assigned to its corresponding cumulative volume value within the garnet. For example, from the core to rim profile, the first analysis has a volume 1 (V 1 ), calculated using radius 1 (R 1 ), and a Grs content 1 (X 1 ). The next point will have a V 2 , calculated using R 2 , and a corresponding X 2 . Finally, the total volume value of the garnet (i.e., 100%) was scaled to the total abundance of garnet in the sample. To calculate the volume, the garnet was assumed to have a dodecahedral shape for the volume calculations. All the information for the 29 points used to build the prograde P-T path is shown in Table S2. Garnet fractionation was not accounted for in building the P-T path. A test of the effect of garnet fractionation shifted point 29 by~50°C and~3 kbar, to lower pressure and temperature. Furthermore, recent studies suggest that chemical fractionation of zoned minerals in metabasic rocks does not have a significant effect on the P-T results (St-Onge, Rayner, Palin, Searle, & Waters, 2013). Notwithstanding, we acknowledge that the lack of chemical fractionation will provide a qualitative location of the grossular and volume isopleths.

| Conventional thermobarometry
Conventional thermobarometry was applied to the studied eclogites following Krogh Ravna (2000) and Krogh Ravna and Terry (2004). In order to better constrain the peak conditions, eight possible mineral compositions combinations were used for samples D197, D205, and D161B, and four combinations for eclogite D167. For eclogites D205, D167, and D161B, the garnet-omphacite thermometer and garnetomphacite-phengite barometer were used. For eclogite D197, the garnet-omphacite-kyanite-phengite-coesite/ quartz equilibrium was used. The minerals used for the P-T calculations are interpreted to be in textural equilibrium; chosen compositions are from (a) garnet rims (or near rim if the outermost analyses show evidence of back diffusion); (b) omphacite with the highest jadeite content; and (c) phengite with the highest Si pfu. The chemistry of the minerals used for the calculations are shown in Table S3.

| Zr-in-rutile thermometry
Thin sections were examined under reflected-light microscopy and BSE before analytical measurements were made. Rutile was selected avoiding cracks, grain boundaries, bad polishing, and nearby zircon (<100 μm away). Measurements that contain more than 0.3 wt% SiO 2 (suggesting contamination from nearby silicates), overlapped pits, scratches, and/or holes were excluded from the results.
Measurements of rutile compositions were performed using a JEOL JXA-8500F field emission electron microprobe at the GeoAnalytical Lab in Washington State University, utilizing five wavelength-dispersive X-ray spectrometers and ProbeForEPMA data acquisition and reduction software (Probe Software, Inc., Eugene, OR). Accelerating voltage was 15 kV, with a beam diameter of 5 μm, and a beam current of 200 nA. Well-characterized natural and synthetic materials were used as standards for calibration. A ZAF intensity correction routine, and a MAN background intensity correction were applied (see Appendix S1 for details on the technique). The Zr detection limit using this setup is 12 ppm.
Zr-in-rutile temperatures were calculated using the pressure-dependent calibration of Tomkins, Powell, and Ellis (2007), and reported errors are one-sigma. Results are shown in Table 3, and individual analyses for each sample are shown in Table S4.

| Phengite-zoisite-kyanite eclogite D197
The peak mineral assemblage of eclogite D197 is Grt+Omp+Ph+Ky+Rt+Zo+Qz/Coe. Omphacite and zoisite contain quartz pseudomorphs after coesite (Figure 4a), indicating that they were stable at UHP conditions. The phase equilibrium diagram constructed for the XRF whole-rock composition fails to predict zoisite at UHP conditions, inconsistent with our petrographic observation of quartz pseudomorphs after coesite in the zoisite. Therefore, a reconstructed bulk composition was used to build the phase diagram. The reconstructed bulk composition is significantly higher in Al 2 O 3 wt % due to the abundance of kyanite in the thin section compared with the part of the sample  (Table S3). (d) Summary of the P-T constraints. Mineral assemblage constraints are shown in different shaded areas. The grey star and grey polygon correspond to those in (b) and (c) respectively analysed by XRF. Zoning in garnet (Figure 3a) appears to be modified from original growth zoning (see discussion below). Furthermore, phase equilibrium modelling fails to predict the prograde P-T path using the intersections between the Grs and Prp isopleths ( Figure S3). The phase equilibrium diagram for eclogite D197 shows that the observed peak assemblage is stable in a large field from >580°C and~18-30 kbar (Figure 7a). The peak assemblage extends into both the coesite and quartz stability fields. The peak temperature is constrained using the average Zr content of rutile from the sample ( Table 3). The Zr-in-rutile isopleth intersects with the maximum phengite Si pfu (Table 1) at~700°C and 29 kbar (Figure 7b). The calculated peak conditions are in the field of the observed mineral assemblage. Garnetomphacite-kyanite-phengite-coesite/quartz equilibrium yields 725-800°C and~26-32 kbar (Figure 7c). These conditions overlap the observed assemblage field, and the Ph+Grt+Omp+Ky+Coe/Qz field.

| Phengite eclogite D205
The peak mineral assemblage for sample D205 is Grt+Omp+Ph+Rt+Qz/Coe. Amphibole and epidote porphyroblasts are texturally late and are interpreted to represent retrograde phases. The grossular and pyrope contours from the garnet core (point a in Figure 3b) intersect at 560°C and~26 kbar in the Tlc+Ph+Grt+Bt+Omp+Lws field ( Figure S4). There are no possible intersections within the P-T diagram that can reproduce the observed zoning in the garnet′s mantle and rim (Figure 3b and Figure S4). Therefore, a reliable P-T path cannot be calculated using the garnet zoning.
The phase equilibrium diagram shows that the peak assemblage is stable at >620°C and >16 kbar (Figure 8a). The peak assemblage extends into both the coesite and quartz stability fields. The peak temperature is constrained using the average Zr content of rutile from the sample ( Table 3). The Zr-in-rutile isopleth intersects with the maximum Si pfu in phengite (Table 1) at~660°C and~29 kbar (Figure 8b). These conditions are within the peak assemblage stability field. Garnet-omphacite-phengite thermobarometry yields~735-763°C and~35-37 kbar (Figure 8c). These conditions overlap the peak assemblage field.
The retrograde evolution of this eclogite is characterized by the presence of amphibole and epidote poikiloblasts. The assemblage field containing both minerals combined with the isopleth of observed amphibole volume,~10-15%, delimits the retrograde path to~590-650°C and~17-21 kbar (Figure 8a,d). The phase equilibrium diagram predicts that amphibole is the first mineral to form during decompression followed by epidote.

| Phengite eclogite D167
The peak mineral assemblage of eclogite D167 is Grt+Omp+Ph+Rt+Qz/Coe. Omphacite preserves quartz pseudomorphs after coesite, suggesting that it was stable at UHP conditions (Figure 4b). Amphibole and epidote are interpreted as retrograde phases. Phase equilibrium modelling was performed using both XRF and reconstructed whole-rock compositions; however, only the results from the latter are presented here (see discussion). Similar to the samples described above, phase equilibrium modelling fails to predict the prograde P-T path using the intersections between the grossular and pyrope isopleths (Figure 3c and Figure S5).
The phase equilibrium diagram for sample D167 reveals that the peak assemblage is stable at >650°C and >22 kbar (Figure 9a). The peak assemblage extends into both the coesite and quartz stability fields. The peak temperature is constrained using the average Zr content of rutile from the sample ( Table 3). The Zr-in-rutile isopleth intersects with the maximum Si pfu in phengite (Table 1) at~675°C and 32 kbar (Figure 9b). The peak conditions are within the Ph+Grt+Omp+Lws+Coe stability field, close to the boundary of the peak assemblage stability field. Garnetomphacite-phengite thermobarometry yields~700-715°C and~31-32 kbar (Figure 9c). These conditions overlap the peak assemblage stability field. Constraints on the retrograde history are provided by the presence of amphibole and epidote. The assemblage field containing both minerals combined with the isopleth of observed amphibole volume,~5-10%, delimits the retrograde path tõ 570-770°C and~15-22 kbar (Figure 9a,d). The phase equilibrium diagram predicts that epidote is the first mineral to form during decompression followed by amphibole.

| Phengite eclogite D161B
The peak mineral assemblage for eclogite D161B is Grt+Omp+Ph+Rt+Qz/Coe. Zoisite and amphibole porphyroblasts (Figure 2g) are interpreted to represent retrograde phases. Epidote and amphibole elongated inclusions, commonly multiphase (Figure 2h), are interpreted as lawsonite pseudomorphs (see discussion). The hand sample shows strong compositional banding, and the phase equilibrium diagram built with the XRF whole-rock data fails to predict the complete observed assemblage. Therefore, a reconstructed bulk composition was used to build the phase diagram. Similar to the samples described above, phase equilibrium modelling fails to predict the prograde P-T path using the intersections between the Grs and Prp isopleths (Figure 6 and Figure S6).
The phase equilibrium diagram shows that the peak assemblage is stable at >610°C and >16 kbar (Figure 10a).
The peak assemblage extends into both the coesite and quartz stability fields. The peak temperature is constrained using the average Zr content of rutile from the sample ( Table 3). The Zr-in-rutile isopleth intersects with the maximum Si pfu in phengite (Table 1) at~625°C and~26 kbar (Figure 10b). These conditions are within the Ph+Grt+ Omp+Lws+Qz stability field, close to the boundary of the peak assemblage field.
The prograde P-T path for eclogite D161B was determined by modelling garnet growth using the intersections between garnet grossular and volume isopleths, calculated from the 29 points of the garnet compositional profile (Figures 6 and 11; Table S2). The phase equilibrium diagram reveals that the garnet volume increase as P-T increases, garnet volume isopleths are closely spaced in a temperature range of 500-600°C, and as the volume isopleths get to the Ph+Grt+Omp+Qz/Coe field, they become more widely spaced (Figure 11a). Garnet volume isopleths are strongly temperature dependent in the lawsonite stability field, whereas in the zoisite field, volume isopleths have a gentle negative slope (Figure 11a). Grossular content generally decreases as P-T increases, but in the Ph+Grt+Omp+Qz/Coe field, grossular slightly increases with P-T (Figure 11b). The P-T point where the first garnet crystalizes lies outside the P-T range of the diagram, so the prograde P-T path begins at~410°C and~17 kbar, with a garnet volume of~1% and a value of~35% Grs  (Table S3). (d) Summary of the P-T constraints. The retrograde conditions are constrained by the observed amphibole volume (5-10%). Mineral assemblage constraints are shown in different shaded areas. The grey star and grey polygon correspond to those in b and c respectively (point 1 in Figure 11c; Table S2). These conditions are in the Chl+Amp+Ph+Grt+Omp+Lws+Qz stability field. Garnet growth continues to the mantle composition at~540°C and~28 kbar, with a garnet volume of~18% and a value of~25% Grs (point 15 in Figure 11c; Table S2). These conditions are in the Tlc+Ph+Grt+Omp+Lws stability field. Then, the model predicts a pressure reversal tõ 550°C and~26 kbar, with a garnet volume of~18% and a value of~28% Grs (point 16 in Figure 11c; Table S2). These conditions are in the Tlc+Ph+Grt+Omp+Lws+Qz stability field. According to our P-T path, the garnet rim forms at~625°C and~35 kbar. These conditions correspond to a garnet volume of 30% and a Grs of 22% (point 29 in Figure 11c; Table S2). These conditions are in the Ph+Grt+Omp+Lws+Coe stability field.
The retrograde evolution of this eclogite is characterized by the presence of amphibole and zoisite. The assemblage field containing both minerals combined with the isopleth of observed amphibole mode,~10-15%, delimits the retrograde path to~590-640°C and~17-21 kbar (Figures 10a  and 11c). The phase equilibrium diagram predicts that the first mineral to form during decompression is amphibole followed by zoisite (Figures 10a and 11c).
The P-T path shows an apparent pressure drop (~3 kbar) during the prograde burial at~545°C (Figure 11). This reversal is a consequence of the increase in Ca in the garnet mantle, which appears to be a common feature in Dulan UHP eclogites (Figure 6 and Figure S2; Meyer, Mattinson, & Zhang, 2015). The apparent pressure reversal might reflect one or more of the following: (a) a major tectonic event. Published studies suggest that oscillatory zoning in  (Table S2). (b) P-T path and grossular isopleths (Table S2). (c) Summary of the P-T constraints. Calculated gradients obtained for this eclogite are shown in the lower left corner. Numbered stars show P-T conditions discussed in the text. The retrograde conditions are constrained by the observed amphibole volume (10-15%). Mineral assemblage constraints are shown in different shaded areas. The grey star and grey polygon correspond to those in Figure 10b,c respectively minerals can reflect burial and exhumation cycles (e.g., García-Casco, Torres-Roldán, Millán, Monié, & Schneider, 2002;Rubatto et al., 2011). Our phase equilibrium diagrams suggest a decompression event (~9 km); and/or (b) a disequilibrium and/or kinetic effect. The lack of correlation between the garnet Ca zoning with Fe, Mg, and Mn suggests that Ca might have been transport controlled (Chernoff & Carlson, 1997;García-Casco et al., 2002;Konrad-Schmolke et al., 2005;Spear & Daniel, 2001). However, because diffusion might have affected the rock at peak conditions (see discussion below), the lack of zoning correlation (i.e., Ca with respect Fe, Mg, and Mn) might not be indicative of this process. A change in the kinetics of crystal growth could also influence the zoning (Kohn, 2004;Pattison & Tinkham, 2009;Stowell, Zuluaga, Boyle, & Bulman, 2011). As shown in Figure 11a,b, once the sample crosses the 5% garnet volume isopleth, the garnet growth rate is predicted to increase until it reaches the 17% volume isopleth where chlorite leaves the assemblage and the garnet growth rate is predicted to sharply decrease (i.e., in the Tlc+Ph+Grt+Omp+Lws stability field), and then increase again once it crosses the 18% volume isopleth (i.e., when enters to the Tlc+Ph+Grt+Omp+Lws+Qz stability field). The increase in Ca in the garnet mantle occurs when the garnet growth rate is predicted to change from slow to rapid, suggesting that the Ca oscillation could be a consequence of the change of growth rate; and/or (c) episodic Ca release due to break down of Ca-bearing minerals at this stage (e.g., apatite or carbonates; Indares, White, & Powell, 2008;Groppo, Rolfo, & Indares, 2012); and/or (d) fluid-rock interaction. Garnet growth and resorption events might be triggered by fluid pulses (Ague & Axler, 2016;Faryad, Klápová, & Nosál, 2010;Giuntoli et al., 2018). As recently suggested by Viete et al. (2018), these fluid pulses might induce overpressure cycles (drained and undrained conditions), causing pressure fluctuations without changes in depth. Given that the phase equilibrium diagram is a simplified qualitative model of a natural process, it is also possible that the apparent pressure reversal is a result of the uncertainties related to the modelling, but the general trend of the P-T path should be reliable (de Capitani & Petrakakis, 2010;Powell & Holland, 2008).

| Peak conditions
Eclogites from the southern part of the Dulan area yield peak conditions of~700°C and~29 kbar for eclogite D197,~660°C and~29 kbar for eclogite D205, and 625°C and~26 kbar for eclogite D161B (Figure 12). Eclogite D167 from the north part of the Dulan area yields 675°C and~32 kbar ( Figure 12). These conditions are in the Grt+Omp+Ph+Rt+Qz/Coe±Ky±Zo stability field, with the exception of eclogites D161B and D167, where the peak conditions are obtained in the lawsonite stability field. Eclogite D161B yields the lowest temperatures conditions of our sample set, agreeing with the preservation of the strongest compositional zoning of all our samples, similar to the temperature-zoning relations of different types of eclogites in the Western Gneiss Region, Norway (e.g. Carswell & Cuthbert, 2003;Cuthbert, Carswell, Krogh-Ravna, & Wain, 2000).
Our P-T constraints have a significantly narrower range than those previously obtained with conventional thermometry (Ren et al., 2016;Song et al., 2003Song et al., , 2006Song et al., , 2014Zhang, Ellis, et al., 2010;Zhang, Meng, et al., 2009;Zhang, Mattinson, et al., 2010;, and indicate that the southern and northern part of the Dulan area experienced the same peak conditions (Figure 12). Diffusional modification of garnet (see discussion below) might partially account for the scattered P-T conditions from conventional thermobarometry in previous work.
For all eclogite samples, the peak conditions obtained using the phengite-rutile approach yield P-T conditions in  Ren et al. (2016) and Zhang, Ellis, et al. (2010). The red area corresponds to the P-T conditions for the HP granulites (Yu et al., 2011(Yu et al., , 2014. The grey area represents previous P-T estimates using conventional thermobarometry for eclogites of the Dulan area (Ren et al., 2016;Song et al., 2003;Zhang, Ellis, et al., 2010;Zhang, Mattinson, et al., 2010;Zhang, Meng, et al., 2009;. Metamorphic facies and abbreviations follow Liou, Tsujimori, Zhang, Katayama, and Maruyama (2004) the coesite stability field (Figure 12), with the exception for eclogite D161B (Figure 12). The peak P-T conditions obtained from all the eclogites imply depths of~80-100 km, and are consistent with the petrographic observation of quartz pseudomorphs after coesite in zoisite and omphacite reported in this study (Figure 4), and coesite inclusions in garnet and omphacite reported in the area (Figure 1; Zhang, Mattinson, et al., 2010;Zhang, Meng, et al., 2009;). Peak P-T estimates of all our samples fall close to the limit of the lawsonite stability field (Figures 7d, 8d, 9d, and  11c). Fluid release associated with lawsonite breakdown is expected to be a main source of fluids at eclogite facies conditions (Groppo & Castelli, 2010;Kerrick & Connolly, 2001;Peacock, 1993). These fluids can re-equilibrate mineral compositions, and therefore, the P-T conditions recorded by phengite and rutile may record the P-T point where lawsonite breakdown liberated fluids rather than the maximum pressure of the P-T path (Figure 11c; Wei, Qian, & Tian, 2013;Wei et al., 2009), supported by our results.

| Retrograde metamorphism
The lack of connection between the last point of the D161B calculated P-T path (i.e., point 29 in Figure 11c), and the peak conditions constrained using the phengiterutile approach (grey star in Figure 11c) might reflect resorption of garnet during exhumation, supported by the observation of resorption surface 2 ( Figure S2). This process would consume the garnet that would have resulted in points along the missing space between the obtained peak conditions and the modelled P-T path (e.g., Groppo et al., 2015;O'Brien, 1997). The black box in Figure 6 shows garnet overgrowth with a composition similar to the expected composition for the missing P-T path (Figure 11b). Furthermore, the resorption event might have been enhanced by fluid pulses (see below), similar to the prograde path (Giuntoli et al., 2018).
Textural relations suggest that epidote and amphibole in samples D205, D167, and D161B crystallized during retrogression. We interpret the formation of these minerals to be compelling evidence of fluid infiltration at early-retrograde conditions (Figures 8d, 9d, and 11c). Dehydration of hydrous minerals, such as lawsonite and talc, liberate large amounts of fluids that can hydrate surrounding UHP rocks in the subduction zone (Clarke et al., 2006;de Sigoyer, Guillot, Lardeaux, & Mascle, 1997;Massonne, 2012;Palin, St-Onge, Waters, Searle, & Dyck, 2014;Pawley, 1994;Schmidt, 1995;Wei & Clarke, 2011;Wei et al., 2009). Fluids related to lawsonite or talc breakdown in eclogites at greater depths in the subduction zone can be responsible for hydrating eclogite facies assemblages in rocks located at lower depths, triggering the growth of hydrous minerals, such as amphibole and epidote at depths of~60 km.
Amphibole compositional zoning is expected to evolve to glaucophane if the retrograde P-T path experiences significant cooling, however, the observed amphibole zoning in eclogite D205 (calcic cores to sodic-calcic rims; Figure S1 and Table S1) indicate that the eclogite experienced high enough temperatures during exhumation to be in the calcic amphibole stability field (e.g., Forneris & Holloway, 2003;Weller, Wallis, Aoya, & Nagaya, 2015;Wei et al., 2013;Wilke, O'Brien, & Altenberger, 2010). Our results indicate that the Dulan eclogites experienced a retrograde P-T path to temperatures of~580-660°C, enough to obliterate any evidence of lawsonite, but cold enough to preclude partial melting (Figures 8d, 9d, and 11c; Clarke et al., 2006;Wei & Clarke, 2011). These results contrast with the hypothesis that eclogites from the south part of the Dulan area experienced wide-spread granulite facies conditions during exhumation (Ren et al., 2016;Song et al., 2003Song et al., , 2014Zhang et al., 2015). Furthermore, phengite in metabasic rocks is expected to disappear on the solidus (Schmidt & Poli, 1994;Schmidt, Vielzeuf, & Auzanneau, 2004); the presence of this mineral in the eclogites from the southern part of the Dulan area is inconsistent with the idea that these eclogites experienced retrogression to granulite facies conditions and partial melting.

| Whole-rock composition
Our results indicate that, even though MT-UHP eclogites present relatively simple mineral assemblages, compositional heterogeneities can affect the results of phase equilibrium modelling. A good example is eclogite D197, where the phase equilibrium diagram calculated for the modified XRF data fails to predict the observed mineral assemblage. Our results suggest that it is important to compare XRF and reconstructed bulk compositions because, even if the compositional heterogeneities are not a disequilibrium feature, they can have important effects in the topology of the phase equilibrium diagram (Guevara & Caddick, 2016;Lanari & Engi, 2017;Palin et al., 2016). For instance, we calculated phase equilibrium diagrams for eclogite D167 using both XRF data ( Figure S7) and a reconstructed bulk composition (Figure 9). The phase equilibrium modelling using both compositions yields relatively similar peak P-T conditions (±3 kbar), but the lawsonite stability fields shifted to higher temperatures, and the zoisite stability fields shifted to higher pressures in the phase equilibrium diagram built using the reconstructed bulk composition (Figure 9 and Figure S7). As a result, the peak conditions lie very close to the limit of the lawsonite stability field for the diagram calculated using the reconstructed bulk composition, similar to results for the other samples and consistent with the hypothesis that phengite-rutile P-T conditions reflect fluid release and re-equilibration due to lawsonite breakdown. The diagram calculated for the reconstructed bulk composition is our preferred result because this approach is expected to better reproduce the observations made at the thin section scale, although this technique might introduce biases, such as point-counting mis-estimations related to geological uncertainties (cf. Palin et al., 2016).

| Garnet diffusion: complexities of phase equilibrium modelling
Our results demonstrate that modelling of garnet zoning using the compositional isopleth approach (i.e., intersection between multiple compositional isopleths) is not suitable for building complete P-T paths for the studied eclogites ( Figures S3-S6). The observed zoning of X Mg , Alm, and Prp (Figures 3 and 6) is not predicted by the phase equilibrium diagrams (Figures S3-S6). The garnet X Mg in eclogites D167 and D205 shows a subtler increase from core to rim than expected, whereas in eclogite D197, the X Mg is nearly homogenous through the garnet (Figure 3). This suggests that diffusion of Fe and Mg has modified the original garnet composition, changing X Mg and producing the observed patterns (Caddick et al., 2010;Carlson, 2006;Konrad-Schmolke, O'Brien, de Capitani, & Carswell, 2008;Konrad-Schmolke et al., 2005;Štípská & Powell, 2005). Other evidence that diffusion affected garnet is the presence of low-Mg patchy zones (that might be closer to the original garnet core composition) found in garnet from eclogite D161B ( Figure 6). With the exception of sample D161B, garnet zoning in the samples is weak and uncommon. Figures 3 and 6 show examples of the strongest zoning found in the studied eclogites. The lack of wide-spread zoning suggests that most of the eclogites re-equilibrated, obscuring prograde P-T conditions (Caddick et al., 2010;Konrad-Schmolke et al., 2005, 2008. The long HP-UHP event experienced by the Dulan area at moderate peak temperatures (Mattinson et al., 2006) might have facilitated the compositional modification off garnet.
Even though eclogite D161B displays Prp and X Mg zoning (Figure 6), the zoning is less than expected ( Figure S6), suggesting diffusional modification and questioning the reliability of using the pyrope and almandine isopleths for P-T path determination ( Figure S6). For instance, without diffusional modification, garnet is predicted to show a stronger increase of Alm (0.22-0.62), Prp (0.01-0.26), and X Mg (0.05-0.35), and a stronger decrease of Sps (0.42-0.01) from core to rim compared to the observed zoning ( Figure S6e). By contrast, the preservation of fine-scale features in Ca zoning suggests that Ca zoning is more likely to represent the original garnet composition, consistent with the low diffusion rate of Ca compared to Fe and Mg (Carlson, 2006;Carlson et al., 2015;Chakraborty & Ganguly, 1992;Chernoff & Carlson, 1997;Florence & Spear, 1991;Giuntoli et al., 2018). Therefore, the P-T path constructed for eclogite D161B using the garnet volume and the Grs content should represent a valid constraint.

| Conventional thermobarometry versus phase equilibrium modelling
Peak conditions calculated with conventional thermobarometry are within~3 kbar and~50°C of the conditions obtained using the phengite-rutile approach (Figures 7d,  9d, and 11c), with the exception of eclogite D205, where conventional thermobarometry yields substantially higher P-T conditions ( Figure 8d). As has been widely discussed (e.g., Carswell, O′Brien, Wilson, & Zhai, 1997;Hacker, 2006;Krogh Ravna & Terry, 2004;Powell & Holland, 2008), there are large uncertainties associated with the garnet-omphacite thermometer associated with Fe 3+ in omphacite. We interpret the peak conditions obtained using the phengite-rutile approach to be more reliable because it is less sensitive to uncertainties associated with ferric iron compared to conventional thermobarometry. Furthermore, complementing phase equilibrium modelling (e.g., Si pfu in phengite isopleths) with independent thermometry (e.g., Zr-in-rutile thermometry) appears to be a robust way to obtain peak P-T conditions because the thermometry is independent of the data and assumptions made in phase equilibrium modelling (e.g., Powell & Holland, 2008;Štípská & Powell, 2005;Wei & Clarke, 2011). Therefore, our discussion of the peak conditions of the Dulan area is made using the results obtained with phengite-rutile approach.

| Sensitivity of conventional thermobarometry results
As mentioned above, conventional thermobarometry of eclogite D205 yields substantially higher P-T conditions than using the phengite-rutile approach (Figure 8d). According to the calibration of the conventional thermobarometry used in this paper, the highest Grs 2 × Prp garnet value will yield the highest P-T conditions (Krogh Ravna & Terry, 2004). A simple test was made in eclogite D205 to evaluate the sensitivity of the conventional P-T results as a function of the garnet composition. For this purpose, the highest jadeite content in omphacite and the highest Si pfu in phengite were used as fixed values, and the garnet composition was changed for the P-T calculations ( Figure 13; Table S5). As shown in Figure 13a, some garnet grains show homogenous Grs 2 × Prp compositions, and others display irregular Grs 2 × Prp zoning that we interpret as evidence of compositional modification. Our results show that garnet composition has a strong influence in the obtained P-T conditions, yielding a range of values of 550-745°C and~28-36 kbar ( Figure 13b; Table S6). These results show that the garnet rim values yield the highest P-T conditions and the core values yield the lowest P-T result, consistent with the increase of Grs 2 × Prp from core to rim (Table S6). The peak conditions obtained from the phengite-rutile approach are close to the values from the garnet mantle and inner rim (Figure 13b; Table S6). These garnet zones display similar Grs 2 × Prp composition to that of most garnet at the thin section scale (Figure 13a). Results from the other samples also suggest that the highest Grs 2 × Prp does not necessarily correspond to the peak conditions reached by the eclogite. For instance, in eclogite D197 and D161B, the lowest Grs 2 × Prp values belong to the rims (opposite to eclogite D205), and in eclogite D167, the Grs 2 × Prp zoning is irregular and does not show major changes through the sample (Table 1 and Table S3). These results suggest that the traditional methodology to select mineral compositions for conventional thermobarometry that will yield maximum P-T conditions (e.g., Carswell, Wilson, & Zhai, 2000;Krogh Ravna & Terry, 2004) Table S6

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article.
Appendix S1. MAN background intensity correction. Figure S1. Composition profiles of minerals in eclogite D197. Figure S2. Garnet textural setting, and compositional zoning from eclogite D161B. Figure S3. Eclogite D197 compositional isopleths for the calculated phase equilibrium diagram. Figure S4. Eclogite D205 compositional isopleths for the calculated phase equilibrium diagram. Figure S5. Eclogite D167 compositional isopleths for the calculated phase equilibrium diagram. Figure S6. Eclogite D161B compositional isopleths for the calculated phase equilibrium diagram. Figure S7. Eclogite D167 phase equilibrium diagram and peak conditions from Zr in rutile thermometer. Table S1. Representative microprobe analysis of amphibole and zoisite from studied eclogites from the Dulan area, China. Table S2. Prograde P-T path information from eclogite D161B. Table S3. Microprobe analyses of minerals used for conventional thermobarometry.