Polyphase inclusions in an UHP gneiss

Chapter 4 Polyphase inclusions in an UHP gneiss 4.1 Introduction Mineral inclusions in zircon and garnet have proved to be a major sources of information on the evolution of UHP rocks in the Kokchetav and elsewhere (Sobolev et al., 1991; Hermann et al., 2001; Katayama & Maruyama, 2009). Intensive studies have led to the finding of an extensive list of UHP indicators in the Kokchetav: diamonds, high-si phengite, coesite, high-k pyroxene and high-si titanite (Sobolev et al., 1991; Hermann et al., 2001; Katayama & Maruyama, 2009; Ogasawara et al., 2002). Investigation of inclusions has also led to the discovery of new minerals kokchetavite (polymorph of KAlSi 3 O 8 ) and kumdykolite (polymorph of NaAlSi 3 O 8 ) (Hwang et al., 2004, 2009) that were both found as microinclusions in K-rich clinopyroxene. It has also been reported that Kokchetav rocks contain polyphase inclusions, which are interpreted to represent crystallised melts. Korsakov & Hermann (2006) found rich associations of polyphase inclusions in the diamondiferous garnet-diopside rock. The inclusions were interpreted as carbonatitic melts and two types of silicate melts. Microdiamonds from Kokchetav gneisses have inclusions of silicate glass with high P and K contents, which is interpreted to represent former melts (Hwang et al., 2006). Contrastingly, diamonds from marbles and garnet-pyroxene rocks contain inclusions of hydrous high-k fluids (Hwang et al., 2005). The diamonds also contain fluid inclusions rich in carbonates and water. Numerous authors conclude that such fluids and melts played an important role in diamond formation (Hwang et al., 2005; De Corte et al., 2000; Sitnikova & Shatsky, 2009; Korsakov et al., 2004). Polyphase inclusions are also reported in many other UHP complexes. The Erzgebirge complex in Germany has remarkable rocks with ubiquitous polyphase 171

172 inclusions in garnets, often with diamonds (Stockhert et al., 2001). Metamorphic diamonds by themselves often contain glassy inclusions with a composition similar to granite (Hwang et al., 2006). UHP metapelites from Eastern Greenland contain polyphase inclusions interpreted as crystallized melts (Lang & Gilotti, 2007). Polyphase inclusions found in garnet and kyanite in rocks from Dora Maira, Italy, were interpreted as inclusions of supercritical fluids by Ferrando et al. (2009). In the UHP complexes of Dabie-Shan, China, there are pyroxenites with polyphase inclusions interpreted as fluid precipitates (Malaspina et al., 2006b). Chapter 3 presented data on the geochemistry of the Kokchetav UHP gneisses and demonstrated that UHP metasediments experienced melting and melt extraction, which resulted in strong depletion of UHP gneisses in LREE, Th and U. The bulk rock compositions provide information on the overall effect of the UHP melting, but the specific details of this process are unclear. In this chapter, I present the detailed descriptions of the UHP garnet-biotite gneiss from the Barchi Kol that contains abundant polyphase inclusions in garnet. Petrography and mineralogy of this sample is investigated and experiments on homogenization of inclusions demonstrate that these inclusions represent former melts. Compositions of inclusions are used for the reconstruction of the metamorphic and geochemical evolution of the sample during UHP metamorphism. 4.2 Analytical methods Phase relations were analysed in polished thin sections via back-scattered electron (BSE) images on a JEOL 6400 scanning electron microscope (SEM) at the Centre for Advanced Microscopy, ANU. The phase compositions were determined by EDS, using an acceleration voltage of 15 kv, a beam current of 1 na and an acquisition time of 120 s. Hydrous glass was not stable under the focused electron beam and a significant decrease in Na 2 O occurred during acquisition. Analyses were performed by area scan whenever possible but often, only spot analyses were possible due to tiny size of inclusions. This problem was overcome by applying a correction to the measured values. Observations during 2 minutes long acquisition showed the average loss of Na 2 O was20%±7%. Na concentrations measured in glass were corrected for this value. A special sample preparation method was used to ensure the reduction of contamination and the preservation of garnets, which later were used for the experimental re-homogenisation of inclusions. The sample was crushed to a grain

173 size of < 3mm with a hydraulic press. Then, a quarter of the sample was powdered in an alumina mill for bulk rock analysis. The rest of the sample was sieved and garnets 1-0.6 mm in diameter were selected for further experiments. The fine grained (<0.6 mm) fraction of the sample was used for heavy minerals separation. The bulk rock composition was measured by the same procedure as in Chapter 3. Trace elements in minerals and glasses were analysed using the LA-ICP-MS facility at RSES, ANU, using a pulsed 193 nm ArF Excimer laser with 100 mj energy at a repetition rate of 5 Hz, coupled to an Agilent 7500 quadrupole ICP-MS (Eggins et al., 1998). Laser sampling of minerals was performed in an He Ar H 2 atmosphere using a small spot size of 16 37 μm. Data acquisition was performed by peak hopping in pulse counting mode, acquiring individual intensity data for each element during each mass spectrometer sweep. Counting was performed for 60 seconds, including a gas background measurement of 20 25 seconds. The LA data were processed by an Excel spreadsheet created by Charlotte Allen. Mineral compositions were calculated using NIST-612 (Pearce et al., 1997) as the external standard and SiO 2 content measured by SEM was used as an internal standard for garnet and tourmaline, Ti for rutile and Ce for monazite. BCR-2G was employed as secondary standard (Norman et al., 1998). Measurement of K content was a challenge because K 39 has an atomic mass close to that of Ar 40, the most abundant ion in ICP-MS. This results in a K 39 high background, far higher than counts from NIST-612 glass. Hence BCR glass, which has a high content of K 2 O, was used as a standard for calculation of K 2 O content in analysis of inclusions. 4.3 Results 4.3.1 Sample description Sample B94-26 originates from drill hole 111, in the Barchi Kol area, at a depth of 44.5 m. Major minerals include Grt, Bt, Kfs, Pl and Qtz. The rock is composed bands with different proportions dark (Grt and Bt) and light (Qtz and Fsp) minerals. The most apparent parts of the sample are melanosome (dominated by Grt and Bt) and leucosome (mostly composed of Qtz and Fsp, Fig. 4.1 a), however withing melanosome there are smaller bands with elevated fraction of Qtz and Fsp and large garnet porphyroblasts. The banding dips at 70 o relative to the vertical direction of the drill core. The melanosome is very rich in garnet of irregular shape, with bands composed predominantly of garnet with an ag-

174 gregate of biotite and feldspar-quartz between them (Fig. 4.1). Bands enriched in Qtz-Fsp contain large garnet grains (up to 15 mm, Fig. 4.2) while biotite rich zones contain smaller grains (<3 mm). The leucosome is dominated by finegrained quartz and feldspars and also contains small euhedral grains of garnet. Bt-Pl symplectite are coarse grained aggregates of biotite flakes and plagiocalse (Fig. 4.1). Plagioclase in symplectite contains small exolutions of Kfs. Bt-Pl symplectites are abundant in the garnet-rich bands and are also present around the large garnet grains (Fig. 4.1 d). Dolomite aggregates were observed in the matrix of melanosome part of the sample. Accessory minerals include: Rt, Zrn, Mnz, tourmaline and sulfides (FeS 2, gersdorffite NiAsS and ZnS). Rutile grains have thin exsolutions of ilmenite. Small sulfides grains are disseminated though feldspar matrix. Large garnet grains contain large, rounded mineral inclusions of Ky, Rt, Phe and abundant polyphase solid inclusions (Fig. 4.1). Kyanite is observed in the sample as rounded inclusions in garnet or as corroded grains between garnet and Bt-Pl symplectite (Fig. 4.1 d, 4.2). In melanosome matrix dolomite is present. Several features differentiate sample B94-26 from many other Grt-Bt gneisses from Barchi Kol unit; the presence of large garnet grains, Bt-Pl symplectites, and rounded inclusions of rutile and kyanite in garnet. Importantly, garnet contains abundant polyphase inclusions. 4.3.2 Mineral compositions Garnet Garnet composition was studied in thin sections and in garnet separates mounted in epoxy. Melanosome hosted garnet (Grt-M) have compositions considerably different from leucosome hosted garnet (Grt-L). The most apparent difference is higher Ca content in Grt-M (Alm 48 63.5 %, Grs 18 23.5 %, Sps 1 5.7 %, Py 12 26.5 %). Large garnets in melanosome have homogeneous composition and in some small garnets rims show Mn and Fe increase and Ca and Mg decrease. Grt-L have lower Ca content (Alm 53 72.2 %, Grs 7 16 %, Sps 0.9 3.5 %, Py 20 28.5 %; Fig. 4.3, Table 4.1). From core to rim, Ca, Mn and Mg increase, and Fe decreases. The rim compositions in Ggrt-M and Grt-L are similar. Garnet inclusions in zircon have compositions similar to Grt-M and high Mn content (Alm 53 57 %, Grs 19 %, Sps 2.4 4 %, Py 21.4 23.4 %). Na content in garnet is 0.03 0.09 wt.% and is similar in both groups of garnet. Grt-L have high Y (80 200 ppm), HREE, and P (1000 1500 ppm). Grt-M garnets

175 μ μ Figure 4.1: Sample petrography. a) Scanned images of polished section of sample B94-26 in directions perpendicular (left) and parallel (right) to the drill core. Dotted line shows the boundary between leucosome (L) and melanosome (M) parts of the sample. b-c) Abundant polyphase inclusions in garnet. d) Kyanite and phengite mineral inclusions in the large garnet. e) Small garnet grain with core rich in polyphase inclusions. f) Bt-Pl symplectite consuming kyanite grain adjacent to garnet. g) Tourmaline and rutile crystals in the matrix of the sample.

176 Figure 4.2: BSE image of the large garnet grain. The garnet contains mineral inclusions (Ky, Rt, Phe) and numerous polyphase inclusions. Dashed areas depict parts of the garnet with abundant polyphase inclusions. Bt-Fsp symplectite (Sym) are adjacent to the garnet, which is sitting in a Qtz-Kfsp matrix. have lower Y (50 100 ppm), HREE, and P (300 or 600 ppm). Zr content in both Grt-M and Grt-L garnets slightly decreases from 25-30 ppm in cores to 10 15 ppm in rims. In contrast Grt-M garnets have concave REE patterns with higher LREE content and lower HREE than Grt-L (Fig. 4.3). Grt-L garnets have REE patterns with flat M-HREE distribution from Sm to Lu and a decreasing trend for LREE (Fig. 4.3). In cores of large Grt-M garnets, a Eu anomaly is absent but the majority of both Grt-M and Grt-L garnets have small negative Eu anomalies (Eu/Eu*=0.5 0.7) in their REE patterns. Micas and feldspars Biotite is the only mica present in the matrix of sample B94-26. The biotite in the matrix contains 2.7 2.8 Si pfu and 3.9 4.5 wt.% of TiO 2. Phengite was exclusively found as inclusions in large garnet porphyroblasts and in zircons.

177 Figure 4.3: Composition of garnet: Grt-M garnet from melanosome, Grt-L garnet from leucosome, Zrn inc inclusions in zircon, Grt-new garnet crystallized in homogenization experiments. a) Triangular Ca-Mg-Fe plot. Arrows indicate compositional change from core to rim in type-l garnets. b) Diagram of spessartine vs. grossular components in garnet. c) REE patterns of garnets from sample B94-26. d) Compositions of phengite and biotite inclusions in garnet/zircon and matrix grains.

178 Phengite inclusions in garnet contain 3.1 Si pfu and 1.7 2 wt.% TiO 2. Zircons contain phengite inclusions with 3.20 3.30 Si pfu and 2.66-3.4 wt.% TiO 2, and biotite inclusion with 3.0 Si pfu and 3.9 wt.% TiO 2. In the sample matrix the feldspar minerals are K-feldspar (Ab 6 12%, An 0.5 1%, Or 96-86%) and plagioclase (Ab 78-75%, An 8 22%, Or 2 10%). Kfs contains exsolutions of plagiocalse and vice versa. Kfs inclusion in zircon has lower content of anortite component (Ab 9.6%, An 0.3%, Or 90%) than matrix Kfs. Rutile Rutile is present as large (up to 1 mm) matrix grains and inclusions in garnet. Rutile grains in both textural contexts have ilmenite exolution lamellas. Nb content in the rutile is 3000-5000 ppm, the Nb/Ta ratio varies from 7 to 22 and U content is 70-90 ppm (Table 4.1). Zr concentrations in rutile inclusion in garnet and in matrix grains are similar (790-1090 ppm) and and application of Zr-inrutile thermometer (Tomkins et al., 2007) indicates temperatures of 735-770 o C at a pressure of 10 kbar. Zircon The zircons are subdivided into four domains based on their textures and trace element compositions (Fig. 4.4, 4.5). Domain 1 commonly forms the cores of the larger grains and has low CL emission with weak zoning (Fig. 4.4). This domain has low Th/U ratios, high content of HREE and steep REE patterns (Fig. 4.5). Domain 1 contains 400-500 ppm P, 60-300 ppm of U and 34-72 ppm of Ti. The Tiin-Zrn thermometer (Ferry & Watson, 2007) indicates crystallization temperature of 840-970 o C. Domain 2 forms cores or mantles around domain 1 with high CL emission and weak concentric zoning. This domain has a similar MREE content to domain 1 but lower HREE contents. Domain 2 has very high Ti content at 100-300 ppm, for which the Ti-in-Zrn thermometer (Ferry & Watson, 2007) gives temperature of 1050-1130 o C. Domain 3 forms the mantles and rims on most grains. It has a CL with high to medium intensity. This domain is characterised by flat HREE pattern, high but variable Th/U ratios and Ti content of 30-125 ppm (corresponding to a temperature 860-1050 o C). Domain 4 is scarce and forms rims, it can be recognized by its low CL emission. This domain has REE patterns similar to domain 3 but has high content of Th and U. Ti content is as low as 10ppm, corresponding to 750 o C. Zircon domains 2 and 3 contain

179 Figure 4.4: CL images of zircons from sample B94-26. Numbers from 1 to 4 mark different growth domains of zircons. inclusions of Phe, Bt, Grt and calcite. Monazite Some of the polyphase inclusions contain small (< 3 μm) monazite grains. However monazite was not found as grains either in the thin sections or within inclusions in garnet. One large grain (> 200μm) of monazite was found in the heavy minerals concentrate and its composition was determined by LA-ICP-MS. The monazite grain has high Th (19 wt.%, Th/La 1.2), low HREE content (Yb 5-6 ppm, Y 670-750 ppm) and U (1.4 wt.%, Th/U 14). The concentration of Sr in the monazite grain is as high as 4300 ppm (Table. 4.1) Other minerals Schorl tourmaline forms blue and brown grains with zoned coloration and irregular shape in the Qtz-Fsp-Bt matrix (Fig. 4.1). Tourmaline has a REE pattern with enrichment in LREE and positive Eu anomaly (Table. 4.1). Dolomite occurs in the melanosome and has an Fe-rich composition (Ca 0.5 Mg 0.35 Fe 0.15 CO 3 ). Inclusions of calcite were found in zircon (Ca 0.95 Mg 0.01 Fe 0.04 CO 3 ). A variety of sulfide minerals were detected (FeS 2, gersdorffite - NiAsS and ZnS) as tiny grains dispersed predominately in the rock-forming feldspars. Also sulfides are present in the polyphase inclusions.

180 Figure 4.5: Zircon composition. a) REE, Th and U patterns of different domains of zircons. b) Histogram of temperatures determined by the Ti-in-zircon thermometer (Ferry & Watson, 2007). c) Th/U vs. temperature.

181 4.3.3 Polyphase inclusions Polyphase inclusions are abundant in garnet and occasionally found in rutile. Polyphase inclusions often occupy central parts of small garnet grains (Fig. 4.1e). In large garnet crystals (> 2mm), inclusions concentrate in irregular zones, spreading from central to outer parts of garnet (Fig. 4.2). Zones rich in polyphase inclusions also contain rounded inclusions of kyanite and rutile (Fig. 4.1d, 4.2). Most inclusions have polygonal negative crystal shapes and dimensions varying from less than 1 μm to 1 mm (Fig. 4.6). Large inclusions have radiating cracks filled with chlorite (Fig. 4.6 b,d,e). Inclusions smaller than 10 μm rarely have cracks around them (Fig. 4.6 c). The inclusions contain numerous silicate minerals: Qtz, Bt, Phe, Ky, Pl, Kfs, Ba-feldspar (with up to 16.3 wt.% of BaO) and Chl. Accessory minerals in inclusions are Rt, Ilm, Ap, Zrn, baddeleite, Aln, Mnz, REE-carbonates and sulfides. Laser ablation analysis of polyphase inclusions Measurement of major and trace element compositions of the polyphase inclusions in melanosome garnets was performed by LA-ICP-MS in thick section. The incomplete ablation is a major problem in reliably estimating of the bulk composition of the polyphase inclusions. Additionally inclusions exposed on the surface of thin section are most accessible for analysis but a fraction of these inclusions are lost during polishing. If a large laser spot size is used and the complete ablation of the inclusions is ensured, then the fraction of the inclusions in the total mass ablated is small and extrapolation to bulk inclusion composition is unreliable. Exposed and subsurface inclusions were ablated together with/without host garnet. The time resolved ablation patterns showed large compositional variations during course of the analysis due to presence of minerals with different compositions. Overall integration of signal averaged out some of these variations, but compositions of the polyphase inclusion still are highly variable (Fig. 4.7). Variations in concentrations of LREE, Th, U, Zr, Nb and relative abundances are particularly high. Therefore, in order to characterize the composition of polyphase inclusions and determine whether they can be transformed into a melt phase, I performed experiments on the homogenization of inclusions at high P and T.

182 Figure 4.6: Multiphase inclusion in garnet from sample B94-26 (BSE images). a) View of inclusions-rich garnet. Numerous inclusions of various sizes are visible. b) Polyphase inclusion with association Qtz + Rt + Ap + Chl +Mnz. The inclusion has cracks. c) Two small inclusions without visible cracks. d) Inclusion with association Qtz + Bt + Pl + baddeleite (ZrO 2 ). This association is unstable because baddeleite and quartz should react to form zircon. e) Example of a large inclusion with numerous cracks filled by chlorite. It contains a number of grains of Ba-rich feldspar (up to 10 wt.% of BaO). f) Unusual triangular inclusion with a Ba-rich mineral and sulfide grain.

183 Figure 4.7: Primitive Mantle (PM McDonough & Sun (1995)) normalised trace element content of polyphase inclusions in comparison with the composition of the host garnet (host Grt), which is shown by bold red. Arrows show that content of an element is below detection limit of LA-ICP-MS. Experiments on the homogenization of polyphase inclusions Experiments on the homogenization of fluid and melt inclusions are performed routinely with volcanic and hydrothermal samples (Roedder, 1984). Recently homogenization of inclusions also was applied to high grade metamorphic rocks, but the substantial difference is that, in this case, the homogenization must be performed at high pressure in order to stabilize minerals hosting inclusions (Malaspina et al., 2006b; Bartoli et al., 2011). Works on fluid and melt inclusions often aim at estimating conditions at which inclusions formed. In particular the temperature of inclusions homogenisation, with relevant corrections, can be a reliable estimate of the temperature of their formation. Instead, in this work, the main purpose of experiments was the transformation of heterogeneous associations into the homogeneous glass suited for analysis by micro beam methods. Another aim is to demonstrate that inclusions indeed represent trapped melt. Such experiments should be performed at conditions where garnet is stable, the hydrous granitic melt is above the solidus and temperature is high enough for complete dissolution of secondary accessory minerals. These requirements dictate the choice of experimental setup. The experiments were performed at high

184 temperature (900-1000 o C) in order to homogenize inclusions and to dissolve completely the accessory minerals. High pressure is necessary for the stabilization of garnet. The experiments were performed with in a powdered matrix medium in order to cushion garnet grains from the strain occurring in the press assembly. In order to prevent leakage of the fluid from the inclusions most of the experiments were performed with an addition of Al(OH) 3, which releases water on heating. High water activity at the conditions used in the experiment presents a problem because it facilitates the formation of matrix melt and destabilizes garnet. As an attempt to cope with this effect some of experiments were performed with oxalic acid (H 2 C 2 O 4 ), which produces H 2 O and CO 2 upon ignition and results in low activity of H 2 O in fluid (Table 4.2). Experiments on the homogenisation of polyphase inclusions were performed on a piston-cylinder press at RSES, ANU. The experiments were run in 3.5 mm gold capsules at temperatures 900 and 1000 o C and pressure of 10-20 kbar. Several experiments were performed with garnets separated from the crushed sample (Table 4.2). The capsule was filled with layers of garnet grains separated by layers of matrix powder (Fig. 4.8). In the experiment C3388 garnet grains were set in a mix of 90% SiO 2 and 10% Al(OH) 3. A one hour long run at 1000 o C and 20 kbar produced melt, Qtz, Ky and two types of garnet. Approximately 10 % of garnet after the experiment was represented by homogeneous grains with small glass inclusions and had composition similar to type-l garnet. In an optical microscope, inclusions appear transparent and isotropic without bubbles or crystalline phases (Fig. 4.8). An exception was the presence of a tiny grain of gersdorffite (NiAsS) in one glassy inclusion (Fig. 4.8). The other 90% of garnet recrystallized and obtained spongy textures after the experiment. The spongy garnet had lower Ca content and higher Mg# than the original garnet (Fig. 4.3). The matrix was composed of quartz and melt with variable major composition similar to andesite: SiO 2 =60%, Al 2 O 3 =13% (Table 4.3). Kyanite was present in melt near garnets. Experiment D1348 was performed with a large garnet grain (5 2.4 2 mm) cut from the sample melanosome (Fig. 4.1 a) with a diamond saw. The garnet grain was placed into the capsule and immersed into the matrix of SiO 2 +Al(OH) 3. In this run large fraction the garnet survived the experiment and glassy inclusions were present in garnet, though most of the inclusions have thin cracks. In this experiment garnet had type-m composition both before and after experiment. Additional experiments were performed at various conditions (Table 4.2), but were unsuccessful, generally due to decomposition of garnet. In order to reduce

185 a b c 100 μm 100 μm d Grt new melt SiO2 e Grt new Ky inclusions Grt inclusion Grt glass FeNiAsS 100 μm 100 μm Figure 4.8: Experiment for the homogenisation of polyphase inclusions. a) Scheme of experimental charge. Garnet layers in the capsule were interlayered by a mix of SiO 2 Al(OH) 3. b) Optical image of polyphase solid inclusions before experiment. c) Homogeneous inclusions after an experiment at 1000 o C,inthe system Grt+SiO 2 +Al(OH) 3. (d-e) Experimental products contain two types of garnet: garnet (Grt) with homogenized inclusions and spongy garnet (Grt new) recrystallized during experiment. melting during experiment, I also performed experiments at a lower temperature (900 o C) and with reduced water vapor activity/pressure. This was achieved by running experiment D1300 with mix of SiO 2 with oxalic acid (H 2 C 2 O 4 ). Inclusions in these experiments homogenized partially, with some silicate crystals (Bt, Pl, etc.) still present. Experiments with different set up produced inclusions in garnets with different compositions. Homogenized inclusions in garnet from melanosome (Grt-M) produced in experiment D1348 will be denoted as type-m inclusions and inclusions in Grt-L from experiments C3388 and D1300 will be marked as type-l inclusions. In two experiments with homogeneous glass composition (C3388 and D1348) major element composition of glasses was determined by EDS and trace

186 element composition by LA-ICP-MS. Major element composition of the homogenized inclusions In general, homogenized inclusions are felsic melts with oxide totals 90%. These melt totals indicate high water content in melts. Melts have 58-68 wt.% SiO2 and 6-10 wt.% of FeO (Table 4.3). Their compositions on a TAS diagram are close to syenite/andesite. Experimental melts coexisting with garnet and mica at 30-45 kbar conditions have MgO and FeO contents < 1.3 wt.% at temperatures 1000 o C (Hermann & Spandler, 2008). The high content of FeO and MgO in homogenized inclusions is interpreted to result from interaction of melt with the host garnet and for small inclusions host garnet also might make contribution. Fortunately both host garnet and garnet dissolved in inclusion have the same composition and thus can be treated as one component. A value of 0.8 wt. % FeO was chosen as original iron content in the melt (Hermann & Spandler, 2008). Then fraction of garnet component in the inclusions was calculated from equation 4.1 and garnet component subtracted from analyses. Such calculations show that typical inclusions contain 10-25 % of garnet component. Uncorrected and corrected compositions of inclusions normalized to anhydrous 100 % are presented on Table 4.3 and Fig. 4.9. After subtraction of the garnet component, inclusions have broadly granitic composition with 2-4 wt.% of Na 2 O and 2-5 wt.% of K 2 O (Fig. 4.9). Average SiO 2 content is 73.4 wt.% 4 in type-l and 71 wt.% in type-m inclusions, which is in excellent agreement with experimental melts which contain 68.7-74.9 wt.% of SiO 2 (Hermann & Spandler, 2008). There are some differences in major element compositions of type-m and -L inclusions. Type-L inclusions are peraluminous granitic melts and have ASI from 0.8 to 1.7 and average 1.2. Type-M inclusions have lower content of Na + K and higher Ca and Al (ASI is also close to 1.2) than experimental melts. These particularities can be due to the crystallization of garnet of different composition on inclusion walls and thus lead to an incorrect subtraction of a garnet component. Another possible explanation is partial loss of alkalies through fractures which are visible in garnets before and after experiments. Type-L inclusions have a significant Cl content of 0.4-1 wt.%, whereas in type-m inclusions Cl content is close to limit of detection of EDS microprobe (<0.3 wt.%) (Fig. 4.9). 4 Here compositions are recalculated to anhydrous basis

187 Figure 4.9: Major element compositions of glassy inclusions obtained after homogenization experiments. a) Na 2 O vs. K 2 O. b) SiO 2 vs. Na 2 O + K 2 O. c) Na 2 O + K 2 O vs Cl. Trace element composition of homogenized inclusions Homogenized polyphase inclusions often are very small (3-10 μm, rare 15μm) and their LA-ICP-MS analysis represent mixes of garnet with glass thus a calculation of an absolute concentration is necessary. The compositions of the inclusions can be calculated if there is an element with an independently constrained concentration in the inclusions an internal standard C El std inc (Halter et al., 2002). The mass fraction of an inclusion in the ablation analysis X can be calculated from the concentration of the internal standard element by the formula: X = CEl std mix CEl std host C CEl std inc El std host (4.1) Where Chost El is the concentration of element El in the host mineral, CEl mix is the measured concentration and Cinc El is the inferred concentration in the inclusion. Then, for all other elements, the concentration in the inclusion can be calculated using the formula: Cinc El = Chost El CEl host CEl mix (4.2) X According to this equation uncertainty in internal standard will propagate

188 to uncertainty in the concentration of all elements and thus the choice of the internal standard needs careful consideration. Elements compatible in garnet in garnet cannot be used as internal standards because their decrease in garnetinclusion mixes is covered by noise of signal from ablation with small spot size. Na, K and Rb are the elements which are most suitable as internal standard because they have a low content in garnet and their concentrations in melts can be estimated from SEM analyses and/or experimental data. Concentrations of Na and K measured in inclusions vary significantly and the primary reason for this is the difficulty of microprobe measurement of volatile elements in micron sized inclusions of different shape and size. Though corrections have been applied to the Na content in the inclusions it is possible that analytical uncertainty was not accounted for completely. On the other hand experimental studies are quite consistent regarding granitic composition of HP-UHP melts derived from metapelites, both Na 2 O and K 2 O vary significantly depending on melting conditions and bulk system composition (Hermann & Spandler, 2008), but Na 2 O + K 2 O is close to 8 wt.%. This value is very close to Na 2 O + K 2 O of type-l inclusions. In type-m inclusions Na 2 O + K 2 O is systematically lower (average 5.6 wt. %) than in experiments. According to experimental data by (Hermann & Rubatto, 2009) K/Rb ratio in melts buffered by phengite is 10-30% higher than K/Rb ratio in the starting composition. Therefore Rb content in the inclusions can be estimated by applying a coefficient to K content in the inclusions, but Rb has the advantage of being measured by LA-ICP-MS more reliably than K. The Rb content in the inclusions was estimated at 250 ppm from the K/Rb ratio of bulk rocks and the K content of the inclusions. Therefore Na, K and Rb may be used as an internal standard either of Na, K and Rb, either through the measured concentrations in the inclusions, or using the values predicted by experiments. Here the compositions of the inclusions were recalculated to the average concentrations of Na 2 O and K 2 O in type-m and -L inclusions measured by SEM and was extrapolated to Na 2 O + K 2 O content of 8 wt.%, which is an average of the compositions of the melts from the study by Hermann & Spandler (2008). Then X was calculated as average of estimates from Na, K and Rb. The fractions of inclusions in the LA-ICP-MS analyses (X) estimated using these constraints are in good agreement between each other (Fig. 4.10). This is considered as the most robust and consistent approach and inclusions composition calculated this way is reported in Fig. 4.12 and in Table 4.4. Another way is to calculate for Na 2 O and K 2 O content measured by SEM, which predicts a higher fraction of inclusion in the analyses and thus lower trace

189 Figure 4.10: Estimates of fractions of homogenized inclusions in ablated material from concentrations of Rb and K, in comparison with estimates obtained from Na concentrations. Estimates from different elements agree between each other close to 1:1, thus demonstrating internal consistency. See the text for details. Figure 4.11: Trace element composition of homogenized inclusions. Chondrite (McDonough & Sun, 1995) normalised patterns of measured garnet-inclusion mixes and composition of the host garnet (host Grt). Arrows show elements which concentrations in garnet are below detection limit. elements concentrations. Elements compatible in garnets (in particular HREE) cannot be reliably estimated in inclusions because two reasons: firstly uncertainty on garnet component is secondly, even if concentrations of these elements would be estimated they still will be completely useless because high garnet capacity for these elements will result in easy and rapid exchange with host garnet either during exhumation or

190 Figure 4.12: Average compositions of two types of homogenized inclusions and composition of protolith and sample B94-26. during experiments. Therefore only elements highly and moderately incompatible in garnet will be considered in further discussion. LA-ICP-MS analysis of the homogenized inclusions revealed increased concentrations relative to the host garnet for many elements: Na, K, As, Rb, Sr, Zr, Nb, Cs, Ba, La, Ce, Pr, Nd, Th and U (Fig. 4.11). Compositions of the homogenized inclusions are much more consistent than those of polyphase inclusions, but there remains a large variability which can be explained only by variable inclusion compositions. In both type-m and type-l inclusions have similar concentrations of Na, K, Rb and Cs. Ba content in type-m inclusions is around 1000 ppm and in type-l inclusions it is highly variable (200-4000 ppm). Sr content is higher in type-m (500-700 ppm) than in type-l inclusions (30-130 ppm). There are significant differences in concentrations of other elements, in particular LREE, Th, U, Zr, Nb, As. Type-M inclusions have high LREE content (>100 ppm La) whereas type-l inclusions have a lower LREE content (< 50 ppm La). Also type-m inclusions have high concentration of Th and U and Th/U ratio of 6, and type-l inclusions have high U content but low Th, and thus low Th/U ratios of 0.2-0.8. Type-M inclusions have much lower Zr content (30-100 ppm) than type-l inclusions (230-1000 ppm). Type-L inclusions have high As content (100-1000 ppm in type-l compared to 30-130 in type-m), which is also confirmed by observation of gersdorffite - NiAsS in those inclusions (Fig. 4.8).

191 4.4 Discussion 4.4.1 Origin of polyphase inclusions There are several mechanisms for the formation of polyphase inclusions in minerals: 1) capture of melt and its later crystallization; 2) capture of mineral aggregates during growth; 3) capture of melt together with crystal(s); 4) capture of mineral inclusion and its later decomposition with formation of melt (Perchuk et al., 2005) or polyphase aggregate. Several features indicate that the polyphase inclusions represent trapped melts, rather than mineral aggregates captured during garnet growth. (1) Most inclusions can be homogenized to a single melt phase. (2) After the subtraction of dissolved garnet, the inclusions have the composition of peraluminous granitic melt, similar to high pressure melts produced in experiments (Hermann & Spandler, 2008). (3) Inclusions have significant concentrations of incompatible trace elements (Zr, LREE, Th, U), which are mostly insoluble in major minerals, but can have high concentration in the melt. (4) Inclusions formed by the capture of mineral aggregates or melt together with crystal(s) should have wide variations in composition corresponding to the abundances of phases in the original inclusion. However, inclusions have a reproducible composition (Fig. 4.9, 4.11), which cannot be assigned to any simple mineral mixture. In crustal rocks, crystallization of melt inclusions sometimes produces simple mineralogy and textures similar to that of granites - nanogranites (Cesare et al., 2009). Instead, polyphase inclusions in sample B94-26 have a rich association of minerals (over 15 phases ) and some components are present in different minerals: Zr is found in zircon and baddeleite; LREE are present in monazite/allanite/lree-carbonates, Ti is found in rutile or ilmenite. Some of the associations present in the inclusions cannot be in equilibrium, for instance quartz associated with baddeleite in some inclusions (Fig. 4.6) is particularly good evidence for absence of equilibrium, because these minerals should react to form zircon. Thus, variable mineral associations in inclusions are evidence for the complex crystallization history of polyphase inclusions in UHP gneiss. One reason for such variability can be decrepitation and loss of fluid phase at various PT conditions during exhumation. Composition of inclusions can be modified by several processes, in particular by partial loss of melt/fluid through cracks, and loss or intake to/from host mineral by diffusion. In sample B94-26, rows of very small (less then 1 μm) inclusions

192 of chlorite are commonly present near the large polyphase inclusions (Fig. 4.6, 4.8). Some LA-ICP-MS trace element analyses of garnets are enriched in Rb, Ba and K. This could be explained by the contamination of the garnet analyses by chlorite inclusions. At the same time, garnet with chlorite inclusions have the same content of LREE and Th as non-contaminated garnet. LREE and Th can be transported rather via melts, whereas LILE can also be liberated by aqueous fluids (Spandler et al., 2007). Therefore the low LREE content in chlorite inclusions indicates that they were formed from aqueous fluids originated by decrepitation of inclusions, from fluid exsolution during crystallization of melt. However, small inclusions usually do not have cracks and only small inclusions survive homogenization experiments. Therefore I assume that homogenized polyphase inclusions are likely to preserve their original composition. Diffusion can significantly affect composition of polyphase inclusions, especially in the case of inclusions hosted by garnet, which has high content for several trace elements. However, content of trace elements like LILE, LREE and some of HFSE in garnet is very low. Even if diffusion of these elements was fast enough, but low capacity of garnet would prevent from efficient transport from inclusions to outer space. On the other hand change of PT conditions will result decreased solubility of some elements in garnet and their diffusion out of grain or exsolution to lamellas and inclusions. Melt inclusions can become sink for the elements which solubility in garnet decreased. Hence elements with high concentrations in garnet can be easily affected by diffusion and thus should considered with caution. Stoeckhert et al. (2009) proposed that polyphase inclusions with diamonds in rocks from Erzgebirge, Germany, experienced loss of fluid during exhumation. Stoeckhert et al. (2009) also concluded that decrepitation of inclusions is controlled by the rate of decompression and that in order to achieve brittle failure of the host garnet, the decompression had to be extremely fast. If decompression is not fast enough then inclusions should expand by plastic deformation of garnet. On this basis Stoeckhert et al. (2009) concluded that exhumation was very fast. Chlorite filled fractures around some of polyphase inclusions in the Kokchetav gneiss B94-26 are similar with those of the Erzgebirge inclusions. Peak PT parameters and exhumation path were similar in The Kokchetav and in Erzgebirge. Hence, the conclusions of Stoeckhert et al. (2009) that the exhumation was very fast are likely applicable to the Kokchetav gneiss as well. This conclusion is in agreement with short upper limit on exhumation time estimated by U-Pb dating in previous works (Hermann et al., 2001) and in this study in Chapter 2.

193 4.4.2 Interpretation of homogenization experiments In experiment C3388 homogenized inclusions composed of glass with high content of K 2 O, Na 2 O and Cl (Table 4.3, Fig. 4.9). Matrix melt and secondary inclusions in the spongy garnet have low quantity of these components. This difference in the composition of inclusions and matrix melt demonstrates the isolation of inclusions from matrix melt because Na and K are able to diffuse rapidly in melt (Acosta-Vigil et al., 2006). In experiment D1348 some fraction of biotite/feldspars were intergrown with large garnet grain used as starting material. In experiment D1348 matrix melt obtained elevated Na and K concentrations from biotite and feldspars which were together with garnet porphyroblast and the difference with inclusions and matrix melt is less apparent. Experiments C3388 and D1300, performed with garnet separates, produced homogeneous and/or partly homogeneous inclusions in garnets. The garnet grains with inclusions had compositions (low content (2-5 %) of CaO), which are equivalent to leucosome garnet cores (Grt-L) and are significantly different from the garnets occurring in melanosome (Grt-M with 7-8 % CaO). The inclusions encountered in these experiments correspond to type-l, with low LREE content. Only in experiment D1348 (with a large garnet grain cut from the sample) has high-ca (Grt-M) garnet survived the experiment and this experiment produced type-m inclusions with high LREE concentrations. Therefore my homogenization experiments C3388 and D1300 resulted in efficient selection of garnets and inclusions: Grt-M garnets with high CaO content were not stable during the experiment and recrystallized completely; only Grt-L garnet with low CaO content survived. Experiments C3388 and D1300 give a biased result, indicating that only low LREE inclusions are present in the sample. This study of the UHP gneiss sample demonstrates that performing only one type of experiment with either separated grains or with single large garnet grain can give biased result and careful, extensive petrographic study of the sample is absolutely necessary for correct interpretation of homogenization experiments. 4.4.3 Estimation of temperatures of inclusions formation LREE content of melts buffered by monazite and or allanite has strong temperature dependence and can be used as geothermometer (Montel (1993), see also Chapter 1). The application of the monazite solubility thermometer from Chapter 1 requires estimates of several parameters: LREE and H 2 O content in the melt, pressure and α - activity of LREE in monazite. For the protolith of the

194 Kokchetav UHP gneiss, α is 0.94 (Chapter 3 and Table 3.1). Type-M inclusions likely formed at peak conditions and thus pressure can be estimated as 50 kbar. The content of LREE is 780-1160 ppm and the calculated temperature is 980-1020 o C. This is a minimum temperature estimate because it is possible that LREE concentration in the melt was not buffered by monazite. Type-L inclusions occur in retrogressed leucosome and pressure of their formation can be estimated at 20 kbar. LREE content in these inclusions is 40-220 ppm corresponding to low temperatures of 600-750 o C. These temperatures are also minimum estimate because monazite may have not been saturated during formation of these inclusions. The uncertainty in the calculation of the absolute composition of the inclusions from the mixes data propagate an uncertainty in calculated temperatures. If the composition of the type-m inclusions is estimated from measured concentration of Na 2 O and K 2 O then temperatures are only slightly lower at 930-980 o C. 4.4.4 LREE, Th and U evolution during melting Sample B94-26 is severely (80%) depleted in LREE, Th and U relative to sedimentary protolith (see Chapter 3, section 3.8.1, Tables 3.1, 3.2 and 3.3). The Th/U ratio of the bulk rock B94-26 is 3, slightly lower than Th/U ratios 4-8 observed in the protolith (Table 3.2). This weak enrichment of U relative Th can be due to the effect of residual zircon. Type-M inclusions have Th/U ratios of 7±3, which are within range of Th/U ratios of the protolith. These high Th/U ratios of type-m inclusions indicates their formation at a time when monazite was completely dissolved in the melt because in the presence of residual monazite/allanite melts acquire low Th/U ratios (Chapter 1). Type-M inclusions have very high LREE content 8 times higher than estimated for the protolith (KMC, Table 3.1). Therefore, the extraction of type-m melt is likely responsible for the depletion of the gneiss in LREE, Th and U (Fig. 4.13). Type-L inclusions have low LREE contents and low Th/U ratios between 0.3-1, with an average of 0.5. The occurrence of these inclusions in leucosome garnet indicates their formation during the exhumation and cooling stage. A possible mechanism can be the formation of additional melt by decomposition of phengite on decompression, which diluted the residual high-lree melt (recorded in type- M inclusions) that remained in the rock after peak melting and melt extraction. Then type-l inclusions were formed when monazite saturation was reached again and monazite consumed Th and caused the enrichment of the melt in U. The low Th/U ratios of these inclusions can be linked with domain 4 of zircons which also

195 Figure 4.13: PT diagram for sample for sample B94-26 with PT path based on Hermann et al. (2001) with changes according Auzanneau et al. (2006). Ellipses with numbers from 1 to 4 denote stages of zircon growth (zircon domains 1 4) and red polygon is for type M inclusions and blue polygon for type L inclusions. Gray lines show location of reactions of phengite breakdown, with phengite disappearance right/below lines: (a) reaction Phe(Ms)+Qtz(Coe)=melt from (Auzanneau et al., 2006), (b) reaction Cpx+Phe+Qtz=Bt+Pl+Grt+melt (Auzanneau et al., 2006), (c) phengite upper stability limit (Hermann & Spandler, 2008). have low Th/U ratios and were formed at low temperatures. Another possibility is that, together with melt inclusions, garnet captured small zircon grains with high U content, causing high Zr contents in type-l inclusions. Type-M inclusions have Th/La ratios of 0.8. Type-L inclusions tend to have high Th/La ratios with an average of 2. Thus the monazite grain from the heavy mineral concentrate which has very high Th content and Th/La ratio of 1.2, can be interpreted as having formed by crystallization of type-l melt in leucosome. The origin of these local variations is unclear, however, they seem to play a minor role in bulk Th/La fractionation because the Th/La ratio of sample B94-26 is 0.3, within the range of 0.3-0.6 in protolith. Therefore, the bulk rock composition is in agreement with insignificant fractionation of Th from La during UHP melting as proposed in Chapter 1.

196 4.4.5 Hosts for LILE and their behaviour during melting Section 3.7.3 demonstrated from bulk geochemistry data that the behaviour of LILE was controlled by the degree of melting and melt extraction, and overall LILE are weakly depleted in the majority of UHP gneisses. In sample B94-26, LILE are now hosted in micas and feldspars. Phengite is absent as a matrix mineral in sample B94-26 and found only as inclusions in garnet and zircon. LILE concentrations were determined by LA-ICP-MS in matrix plagoclase, K-feldspar and biotite and also in phengite and biotite inclusions in zircon and in phengite inclusions in the large garnet porphyroblast (Table 4.1, Fig. 4.14). In the latter case, inclusions were ablated together with the host zircon. The trace element composition of micas was reliably estimated by using Al as internal standard, because zircon has negligible LILE contents, Al and K. The results are presented on the Fig. 4.14. There is a remarkable similarity between LILE contents between sample B94-26 and in the protolith for the Kokchetav UHP gneisses (Fig. 4.14 a; see Chapter 3 for protolith estimate). K-feldspar, plagioclase and biotite have different concentrations of LILE: Pl hosts Sr, Kfs concentrates Rb, Sr and Ba, Bt has high content of Rb, Cs and Ba (Fig. 4.14 b). Altogether, these minerals are capable of hosting all LILE present in the rock. High pressure mica inclusions in garnet and zircon have a higher content of Sr and Ba than matrix biotite and 3-5 times higher concentrations of Rb, Cs and Ba than the bulk rock composition (Fig. 4.14 c). Thus, the presence of 20-30 % mica of this composition would easily accommodate all the Rb, Cs and Ba present in the rock. The concentration of Sr in mica inclusions in zircon is lower than in the bulk rock and mica inclusions in garnet have Sr comparable to the bulk rock. In inclusions of type-m and -L, concentrations of LILE are quite similar (Fig. 4.14 d) with the exception of Sr: in type-l inclusions Sr is lower than the bulk rock content, whereas in type-m inclusions Sr is 2-3 times higher. Phengite and biotite inclusions were observed in domain 3 of zircons, and close to zircon rims. Thus these micas were formed during decompression, but a high content of Si and Ti indicates their HP-HT origin. Compositions of micas and feldspars show that in the current assemblage, LILE are fractionated between Pl, Kfs and Bt, but at HP conditions they were mostly concentrated in phengite. Type-M inclusions formed at peak conditions have only slightly higher concentration of Rb, Cs and Ba than the bulk rock and protolith, thus extraction of this melt would not produce a large depletion in these elements. Sr content in inclusions is higher than in the rock and protolith, thus it is possible that

197 Figure 4.14: LILE concentrations in the bulk rock and in protolith (a) and comparison with: b) LILE concentrations in rock forming minerals, c) LILE concentration in the high pressure mica inclusions in garnet and zircon, d) LILE concentration in homogenized inclusions. real protolith had high Sr content. Sr was incompatible and possibly similar Sr concentration in restite and protolith occur only by coincidence. 4.4.6 Hosts for HFSE and their behaviour during melting In sample B94-26 concentration of HFS elements (Ti, Nb, Ta, Zr and Hf) are similar to protolith, suggesting moderately incompatible behaviour during UHP melting (section 3.7.4). Sample B94-26 contains 6500 ppm Ti. Ti content in type-l inclusions is 1000-3000 ppm and 3000-6000 ppm in type-m inclusions. Concentrations in type-m inclusions are in agreement with Ti solubility of 2800-5500 in 1050-900 o C UHP melts buffered by rutile (Hermann & Rubatto, 2009). However, these concentrations should be considered with caution because garnet can have significant Ti content and will modify Ti content in inclusions. Nb content is 26-40 ppm in type-m inclusions and it is < 10 ppm (averaging 3 ppm) in type-l inclusions

198 whereas in bulk sample B94-26, Nb content is 15 ppm. It is expected that type- M inclusions have Nb/Ta ratio <10, and the extraction of that melt caused an increase in the Nb/Ta ratio from 10 in protolith to 13 in restitic sample B94-26. However Ta was not measured in the inclusions due to expected low concentrations. Overall, the extraction of type-m melt had very little effect on Ti content, but it reduced Nb content in the rock and presumably depleted Ta more strongly than Nb. Sample B94-26 has elevated Nb/Ta ratio of 13, relative to Nb/Ta ratio of 10 in the protolith. Rutile contains 2800-5200 ppm Nb and 140-600 ppm of Ta. Nb/Ta ratios in rutile are highly variable between 7-22. In garnet Nb and Ta concentrations are very low (<0.1 ppm). Phengite and biotite have high Nb content (10 ppm Nb, Table 4.1) and Nb/Ta ratio in both phengite and biotite is around 30, more than two times higher than bulk rock. The fractionation of Nb/Ta ratio by phengite is huge, in particular when compared with rutile, which has D rt/melt Nb /D rt/melt Ta 0.4-1 and is close to unity in low-t granitic melts (Schmidt et al., 2004; Xiong et al., 2011). Therefore, the increased Nb/Ta ratio in sample B94-26 can be explained only by the Nb/Ta fractionation in the residual phengite (see also discussion in section 3.7.4). Though rutile was present in sample B94-26 it is clear that rutile was not the only major host for Nb and Ta in sample B94-26 and it is phengite which has made large positive effect on Nb/Ta ratio. Large variation in Nb content and Nb/Ta ratios in rutile in sample B94-26 is probably related to Ti exsolution from HP garnet and reactions of rutile with mica. Zr and Hf bulk rock concentrations in Kokchetav gneisses are mostly unaffected by UHP melting (section 3.7.4). Type-M inclusions have low Zr content (30-80 ppm) and type-l inclusions have high Zr content (500-700 ppm). These concentrations are in complete disagreement with inferred high temperatures for type-m inclusions and low temperature for type-l inclusions as zircon saturated melt at 1000 o C and 45 kbar contains 260 ppm Zr (Hermann & Rubatto, 2009). Also low Zr content is in disagreement with the common observation of Zr bearing minerals in polyphase inclusions (zircon and baddeleite). Zr content increases in garnet with increasing temperature and pressure, and experiments by Rubatto & Hermann (2007a) estimated very high solubility (200-800 ppm Zr in garnet), though these data need to be confirmed. In the garnets from sample B94-26 Zr content is only 10-20 ppm. Therefore one explanation for the high/low Zr concentrations in inclusions can be diffusion of Zr between garnet and its inclusions. Another explanation can be the heterogeneous capture of small zircon grains together with melt. The difficulties in interpretation of Zr concentration in inclu-