The role of sulfate-rich fluids in heavy rare earth enrichment at the Dashigou carbonatite deposit, Huanglongpu, China

Abstract The Huanglongpu carbonatites are located in the north-western part of the Qinling orogenic belt in central China. Calcite carbonatite dykes at the Dashigou open pit are unusual due to their enrichment in heavy rare earth elements (HREE) relative to light rare earth elements (LREE), leading to a flat REE pattern, and in that the majority of dykes have a quartz core. They also host economic concentrations of molybdenite. The calcite carbonatite dykes show two styles of mineralogy according to the degree of hydrothermal reworking, and these are reflected in REE distribution and concentration. The REE in the little-altered calcite carbonatite occur mostly in magmatic REE minerals, mainly monazite-(Ce), and typically have ΣLREE/(HREE+Y) ratios from 9.9 to 17. In hydrothermally altered calcite carbonatites, magmatic monazite-(Ce) is partially replaced to fully replaced by HREE-enriched secondary phases and the rocks have ΣLREE/(HREE+Y) ratios from 1.1 to 3.8. The fluid responsible for hydrothermal REE redistribution is preserved in fluid inclusions in the quartz lenses. The bulk of the quartz lacks fluid inclusions but is cut by two later hydrothermal quartz generations, both containing sulfate-rich fluid inclusions with sulfate typically present as multiple trapped solids, as well as in solution. The estimated total sulfate content of fluid inclusions ranges from 6 to >33 wt.% K2SO4 equivalent. We interpret these heterogeneous fluid inclusions to be the result of reaction of sulfate-rich fluids with the calcite carbonatite dykes. The final HREE enrichment is due to a combination of factors: (1) the progressive HREE enrichment of later magmatic calcite created a HREE-enriched source; (2) REE–SO42– complexing allowed the REE to be redistributed without fractionation; and (3) secondary REE mineralisation was dominated by minerals such as HREE-enriched fluorocarbonates, xenotime-(Y) and churchite-(Y) whose crystal structures tends to favour HREE.


Introduction
The light rare earth elements (LREE; La to Sm) and the heavy rare earth elements (HREE; Eu to Lu) are now classified as critical raw materials by the European Commission (European Commission, 2017) due to the growing demand for these strategic elements. This has created a heightened interest in understanding the origin of economic REE deposits (Williams-Jones et al., 2012;Wall, 2014).
This investigation follows from the recent work of Smith et al. (2018) on the Huanglongpu calcite carbonatites. They investigated the HREE enrichment of the carbonatites by looking at the progression from magmatic REE-bearing minerals to later hydrothermal REE-bearing minerals. They demonstrated that each subsequent REE-bearing mineral stage resulted in preferential leaching of the LREE and enrichment in the HREE. This is associated with a shift to more silica-rich hydrothermal conditions leading to quartz growth, and finally with sulfate mineralisation. They documented magmatic HREE enrichment seen in late-stage magmatic calcite. Subsequently, hydrothermal alteration of magmatic REE phases, notably monazite-(Ce), was accompanied by growth of secondary REE minerals, in particular britholite-(Ce) and Ca-REE fluorocarbonates, which show relative HREE enrichment compared to magmatic phases.
The Huanglongpu carbonatite is mined as a molybdenum deposit, but is considered to be part of a key REE-producing area of China, although its reserves and grades for the REE are unpublished (Kynický et al., 2012). The presence of quartz with associated fluid inclusions, together with hydrothermal HREE-rich minerals, make it suitable for investigating the role of secondary hydrothermal fluid in the REE enrichment of carbonatites.

Geological setting
The Huanglongpu deposit lies in the north-western part of the Qinling orogenic belt in central China (Xu et al., 2007;Fig. 1). This orogenic belt hosts numerous ore deposits and is divided by the Shangdan suture into two parts: North and South Qinling. The northern border of the North Qinling belt is defined by a normal fault associated with the Cenozoic rift basin to the north. The southern Qinling belt is further divided by the Mianlue suture which was extensively reworked by Late Mesozoic thrusting (Xu et al., 2010). A comprehensive summary of the North and South Qinling belt lithologies is given in Xu et al. (2010). The Qinling orogenic belt was incorporated into Rodinia during the Grenvillian orogeny, followed by rifting, then subduction-related tectonics in the Palaeozoic and early Mesozoic, leading to Cretaceous to Palaeogene reactivation, and finally Neogene extension. The detailed tectonic history can be found in Ratschbacher et al. (2003).
The Qinling orogenic belt is an important metallogenic belt, hosting the most important Mo ore camp in China and including several world-class Mo deposits (Mao et al., 2008). The mineralisation is mostly hosted by granitic porphyry bodies with a few porphyry-skarn deposits. However, in the Huanglongpu area ( Fig. 1) the Mo mineralisation is associated with unusual carbonatite veins. The Huanglongpu area contains an extensional structure hosting ore bodies that extend over a distance of 6 km, mainly controlled by a northwest-striking fault zone (Xu et al., 2010;Fig. 1a). There are five ore fields currently being mined ( Fig. 1a): Yuantou, Dashigou, Shijiawan I and II and Taoyuan, with a total reserve of 0.18 million tonnes of MoS 2 . Apart from Shijiawan I, which is hosted by porphyry and has intruded into the Neoarchean Taihua gneiss, these are associated with carbonatite dykes. Most of the carbonatite ore fields are also related spatially to porphyry and porphyry skarn Mo deposits which yield Re-Os ages ranging from 148 Ma to 112 Ma (Mao et al., 2008). However, the carbonatite molybdenite yields Re-Os ages from 209 Ma to 221 Ma and monazite-(Ce) U-Pb and Th-Pb ages from 208.9 ± 4.6 Ma and 213.6 ± 4.0 Ma (Huang et al., 1994;Stein et al., 1997;Song et al., 2016) indicating that the mineralisation of the carbonatites is not related to overprinting by the porphyry-related systems. The samples analysed in this investigation are from the Dashigou open pit.

The Dashigou open pit
Individual molybdenite-bearing calcite carbonatite dykes at Dashigou are generally between 0.1 and 1 metre in width, and extend for several tens of metres (Fig. 1b). The main set of dykes predominantly dips N to NNW at steep angles, 50-88°N, (Smith et al., 2018;fig. 1b). Regardless of the range of dyke orientations observed by Smith et al. (2018), including some east-dipping conjugate dykes, they are understood to be closely contemporaneous (Smith et al., 2018;fig. 3e). The country rock gneiss has been fenitised next to the carbonatites (Fig. 1c), and now consists mainly of K-feldspar, phlogopite, pyrite and calcite, with minor amounts of monazite-(Ce) and titanite. However, the fenite was apparently unaffected by the later hydrothermal alteration which corroded and altered monazite-(Ce) in the samples we studied.

Mineralogy of the calcite carbonatites
The carbonatite dykes are dominated by calcite and are referred to here as calcite carbonatite. The majority of the dykes contain central quartz lenses (Fig. 1c). A few dykes show evidence for multiple opening events (Table 1). The calcite carbonatite dykes consist of 50 to 80 vol.% calcite, with typical individual crystals ranging from less than 0.1 cm to 2 cm in size, together with sulfates (celestine-baryte), sulfides (molybdenite, pyrite, galena and sphalerite), K-feldspar, phlogopite, albite, REE minerals and apatite. Quartz only occurs as lenses and is only observed in the calcite carbonatite dykes and not the surrounding fenite. The lenses are composed of quartz grains, ranging from less than 100 μm up to 3 cm, containing μm-sized birefringent inclusions discussed later in this manuscript. Smith et al. (2018) described the paragenesis of the calcite carbonatite minerals. Molybdenite can occur as disseminated grains, as fracture infill within the calcite carbonatite, with other sulfides, or in fracture infill within the fenitised gneiss. Celestine-baryte was inferred to be hydrothermal in origin and Smith et al. (2018) documented a crosscutting relationship with the calcite carbonatite dykes. Pyrite and sphalerite are later than cogenetic molybdenite and pyrite.
The REE mineral assemblages of the calcite carbonatite dykes (Smith et al., 2018) preserve a range of replacement textures making it possible to establish a history of REE mineral growth. The initial, magmatic stage associated with the calcite carbonatites consists of monazite-(Ce), bastnäsite-(Ce), parisite-(Ce) and aeschynite-(Ce) possibly with magmatic burbankite-(Ce); all these minerals are relatively enriched in LREE. Subsequent hydrothermal events first led the monazite-(Ce) to be altered to produce a second generation of apatite, and this was in turn replaced and overgrown by britholite-(Ce) accompanied by the crystallisation of allanite-(Ce). Bastnäsite-(Ce) and parisite-(Ce) were replaced by synchysite-(Ce) and röntgenite-(Ce). Aeschynite-(Ce) was altered to uranopyrochlore and then pyrochlore with uraninite inclusions (Smith et al., 2018). These hydrothermal replacement minerals show HREE enrichment relative to the magmatic precursors.
The magmatic calcite shows some late HREE enrichment indicating that the Dashigou magmatic carbonatite system evolved to HREE-richer conditions (Xu et al., 2007;Smith et al., 2018). The composite dykes observed in the field suggest a sequence of calcite carbonatite pulses.
There is significant variation in the extent of hydrothermal alteration between calcite carbonatite samples, with magmatic REE phases well preserved in some, but mostly overprinted in others. Smith et al. (2018) concluded that the hydrothermal alteration was linked to the deposition of some of the quartz including the fluid inclusions which they host.

Methodology
Samples were collected from the Dashigou open pit in June 2016, within a 100 metre radius around location WGS84 49S, 410530E, 3803439N (Fig. 1a). The open pit exposes calcite carbonatite dykes intruded into country rock gneiss (Fig. 1). Four samples were also collected from the Yuantao mine dump a few kilometres north of the Dashigou open pit (WGS84 49S, 409886E, 3804124N). The two fenite samples used in this study were in direct contact with the associated calcite carbonatite samples.
Mineral and fluid-inclusion parageneses were established using optical petrography followed by scanning electron microscopy (SEM) at the University of Leeds using a FEI Quanta 650 field-emission gun scanning electron microscope operated at 20 kV employing back-scattered electron (BSE) and cathodoluminescence (SEM-CL) imaging.
Oxygen isotope measurements were undertaken on selected polished areas of some carbonatite samples mounted in an epoxy block. Prior to being mounted, the selected samples were mapped using BSE. Stable-oxygen-isotope measurements were carried out at the Edinburgh Ion Microprobe Facility (EIMF) using a Cameca 1270 mass spectrometer operating with a primary Cs + beam, in multi collector mode, operating at conditions of ∼5 nA with a net impact energy of 20 KeV (+10 kV primary and -10 kV secondary). The beam size was ∼12 μm; standards were: calcite (EIMF) and baryte (Stern S0237). The instrument was calibrated before each session and its drift was corrected by starting and finishing a session by analysing 5 to 10 points of the reference standards and then 5 points every ten analyses. Linear least squares was applied as the correction for drift. Reproducibility of the calcite and baryte standards can be found in Appendix 1 (Supplementary materialssee below).
Raman spectroscopy was carried out in the GeoResources Laboratory, at the University of Lorraine, France. The fluid-inclusion phases were analysed separately to identify solids and Raman-active solution and gas components. Analyses were performed using a LabRAM Raman microspectrometer (Horiba Jobin Yvon) equipped with a Linkam THMS600 microthermometry stage. The excitation beam was a monochromatic green laser at 514 nm (ionised argon 74.53 nm, with a working temperature of -128°C and a Pn of 500 mW). For each solid and gas analysis a spectrum was taken directly next to the fluid or solid inclusion to allow subtraction of the quartz and air N 2 peaks from the sample spectrum. The subtractions were made in Horiba LabSpec software, and CrystalSleuth software by RRUFF (Laetsch and Downs, 2006) was used to identify the solids. The gas proportions within the fluid inclusions were estimated using the method of Dubessy et al. (1989).
Microthermometry measurements were carried out at the University of Leeds using a Linkam MDS600 Heating Freezing Stage attached to a TMS93 temperature programmer and controller and an LNP cooling system mounted on an Olympus BX50 microscope. Double polished wafers were between 200-300 μm thick. The stage was calibrated using synthetic fluid-inclusion wafers containing pure water fluid inclusions (0°C and 374.1°C) and H 2 O-CO 2 fluid inclusions (-56.6°C and 31.1°C) provided by Ronald Bakker, Leoben University, Austria.
Bulk rock analyses were carried out by Activation Laboratories Ltd. with the sample preparation undertaken at the University of Leeds. Samples were powdered using a steel pestle and mortar and an agate TEMA barrel and sieved to <200 μm. The methods 4Litho and 4F-Total sulfur were used on the calcite carbonatite and country rock samples. The methods and detection limits are detailed in Appendix 1.

Calcite carbonatites
The calcite carbonatite samples have been divided into two broad alteration categories, little altered and altered, based on the extent to which new REE minerals have replaced the original magmatic ones (Table 1). No correlation of alteration with the different dyke orientations was observed. The little-altered samples host predominantly magmatic REE minerals. Monazite-(Ce) is the main primary magmatic REE mineral in the calcite carbonatite samples, and is usually fairly pristine in the little-altered samples (Fig. 2a), although in some instances it can show overgrowth and fracture-fill of late magmatic molybdenite (Fig. 2b). In the altered samples (Table 1) monazite-(Ce) is at least partially corroded and is sometimes fully replaced by new HREE-enriched phases (Fig. 2c), with later HREE-enriched phases in fractures (Fig. 2d). The HREE-enriched phases are predominantly Ca-REE-fluorocarbonates, uraninite, xenotime-(Y) and churchite-(Y) together with unidentified Y and REE-bearing silicate minerals with a variable amount of Ca and Fe, and unidentified Ti, Nb, U, Y and REE-bearing oxide phases with a variable amount of Fe and Ca ( Fig. 2c) (Table 2). Smith et al. (2018) described the REE mineral paragenesis, demonstrating that the hydrothermal REE mineralisation becomes more HREE enriched with each subsequent hydrothermal event. Xu et al. (2007) first reported that the Huanglongpu carbonatites are anomalously rich in the HREE, exhibiting a distinctive flat REE pattern. Xu et al. (2010) investigated calcite carbonatites from the Yuantou, Shijiawan and the Dashigou deposits and reported that total REE concentrations could exceed 3000 ppm. They found that ΣLREE/(HREE+Y) ratios ranged from 0.6 to 2.8. We have obtained a wider range of Σ(LREE/HREE+Y) ratios in this study, from 1.1 to 17.3, correlated with the alteration categories described above, with the little-altered calcite carbonatite samples showing ΣLREE/HREE ratios ranging from 9.9 up to 17.3 and the altered calcite carbonatite samples ratios ranging from 1.1 to 3.8 (Table 3). ΣHREE + Y totals also overlap with those reported by Xu et al. (2007; show the HREE-enrichment documented by Xu et al. (2010). Instead, they exhibit a LREE-enriched pattern typical of carbonatites worldwide (Woolley and Kempe, 1989) and similar to that of the fenites (Fig. 3, green lines). The fenite REE pattern reflects its REE mineralisation consisting mainly of unaltered monazite-(Ce), which has been observed in other carbonatite associated fenites (e.g. Thair and Olli, 2013;Midende et al., 2014;Trofanenko et al., 2016). In contrast, the altered calcite carbonatites in this study (Table 1, shown in red on Fig. 3) are HREE-enriched and plot close to the samples described by Xu et al. (2010) with the most HREE-enriched samples showing a positive Y anomaly (Fig. 3).

Quartz lenses within calcite carbonatite dykes
Irrespective of the host dyke's orientation, SEM-CL imaging reveals the presence of three quartz generations making up the quartz lenses that occur within them (Fig. 4). The dominant generation is the earliest, Qz-1; it is moderately luminescent and shows some poorly-defined growth zonation. Generation Qz-2 occurs as non-luminescent overgrowths and porosity infills, merging into fracture fills cutting grains of Qz-1. The last generation, Qz-3, is also a non-luminescent quartz and occurs as thread-like fracture fills cutting both earlier generations. While Qz-1 lacks fluid inclusions and appears patchy and strained in cross polars, both Qz-2 and Qz-3 are cloudy and contain fluid inclusions.

Fluid-inclusion paragenesis and hydrothermal HREE mineralisation
Smith et al. (2018) showed that secondary quartz grew at the same time as the later hydrothermal REE minerals that are enriched in the HREE. The relationship between HREE minerals and Qz-2 and Qz-3 partially replacing magmatic calcite is shown in Fig. 5. HREE phases occur at the interface between calcite and Qz-2 ( Fig. 5a) and in calcite fractures associated with celestine-baryte mineralisation related to Qz-3 (Fig. 5b).

Calcite and quartz stable oxygen isotope data
The samples analysed are from two distinct dykes and data is presented in Table 4. Secondary quartz data is from both Qz-2 and Qz-3 (non-luminescent quartz, Fig. 4) as these could not be distinguished during the analyses. Because of the difficulty in finding exact points for analysis in the ion probe, traverses were made across areas with all quartz generations to ensure that primary and secondary generations would be analysed. It is clear from the results that both Qz-1 and secondary quartz (Qz-2 and Qz-3) have similar compositions with tightly constrained populations of data that differ by no more than the analytical uncertainty (c. 0.25‰). Average compositions are close to 10‰ SMOW (e.g. DSG 002 Qz-1 δ 18 O ranges from 9.72‰ to 10.43‰ and secondary quartz δ 18 O ranges from 9.57‰ to 10.18‰;

Fluid-inclusion petrography and microthermometry
Three samples of quartz lenses from altered carbonatites were selected for the fluid-inclusion study ( Fig. 3; Table 5): DSG 002, DSG 437 and DSG 433 (Table 1). Fluid inclusions range in size from <5 μm up to 70 μm but are commonly between 10 μm to 40 μm. Due to the cloudiness of Qz-2 and Qz-3, we were unable to collect freezing measurements such as Tm CO 2 and Tm CLA . Fluid inclusions are abundant in Qz-2 and mainly occur as clusters or in fluid-inclusion-rich zones in cloudy quartz (Fig. 6a), more rarely in planar arrays and linear trails. The fluid inclusions usually have quite irregular shapes (Fig. 6a,b,d) and are predominantly LCO 2 + VCO 2 + Lw and LCO 2 + VCO 2 + Lw + S(s) type fluid inclusions (where LCO 2 = liquid CO 2 , VCO 2 = gas CO 2 , Lw = aqueous liquid and S(s) = from one to five solids). They show variable proportions of CO 2 relative to the aqueous liquid, even within the same fluid-inclusion cluster (Fig. 6a). Some inclusions have a homogenised CO 2 bubble at room temperature and are referred to as LCO 2 + Lw fluid inclusions ( Fig. 6a; Table 5; Appendix 2). About 15% of the fluid inclusions in Qz-2 contain one or more solids (Fig. 6a-f; Tables 2,5). These solids are frequently birefringent and individual fluid inclusions can contain up to five different solids, but there is only ever one opaque solid (Fig. 6d). On rare occasions, fluid inclusions are not CO 2 -bearing and only comprise an aqueous solution and solid(s) (Lw + S) with the solid taking up most of the inclusion volume.
The Qz-2 generation may also contain small (<15 μm) birefringent solid inclusions near the fluid inclusions, within the same cluster or fluid-inclusion trail. These solid inclusions are of comparable size to the solids observed within the fluid inclusions (Fig. 6a). Microthermometric data for Qz-2 fluid inclusions show a scatter of Th CO 2 from +12°C to +30.9°C, yielding a density of the CO 2 phase of 0.85 to 0.47 g cm -3 (Valakovich and Altunin, 1968). CO 2 always homogenised to the liquid phase.
In Qz-3, fluid inclusions occur in planar arrays decorating healed fractures ( Fig. 6g-i) or as linear fluid-inclusion trails in finer healed fractures (<50 μm). Overall, the Qz-3 fluid inclusions are smaller than those in Qz-2 ranging from <5 μm to 25 μm. These fluid inclusions tend to have similar shapes within a single array, either elongate or with a more rounded shape (Fig. 6h-i). A wider variety of fluid-inclusion types are hosted by Qz-3 ( Fig. 7) but they are predominantly of LCO 2 + VCO 2 + Lw type, similar to those in Qz-2, but with more variability in the relative proportion of the CO 2 and aqueous phase (Fig. 6h). Some of these fluid inclusions have a homogenised CO 2 bubble below room temperature. In aqueous inclusions (Lw + V) there is a reasonably constant liquid/vapour ratio (c. 80:20). Lastly, there are monophase Lw inclusions which are small (<15 μm) and tend to occur in wider healed fractures, typically from 100 μm to 400 μm (Fig. 6i). They are clearly secondary. Lw and Lw + V type fluid inclusions can occur in the same fluid-inclusion trail and in this instance the vapour bubble proportion is usually lower at ∼10% (Fig. 6i). Small (<15 μm) birefringent solid inclusions are also hosted by Qz-3, however they tend to be less abundant than in Qz-2.

Raman spectroscopy
About 200 fluid inclusions were analysed from Qz-2 and Qz-3 in the three samples investigated (Table 5; Appendix 2). Compared to Qz-2, Qz-3, fluid inclusions are less suitable for Raman spectroscopy and microthermometric analysis as the majority are smaller than 20 μm. A summary of the fluid inclusion Raman spectroscopy data is provided in Appendix 2. When possible, the same fluid inclusions were analysed for solution, solid and gas. Both solid-bearing and solid-absent inclusions were analysed for solution components and gave comparable results.

Qz-2 fluid inclusions
The aqueous liquid phase of approximately 100 CO 2 -bearing fluid inclusions was analysed by Raman spectroscopy. Sulfate (SO 4 2-) ions were detected in solution in 90% of the fluid inclusions, from the peak at 981 cm -1 . A few inclusions without detectable sulfate in solution have sulfate solids and it is likely that where sulfate was not found this was a result of the small volume of liquid available. HSO 4 was not detected during the analysis (1050 cm -1 ), precluding very acidic conditions (pH < 2).
In approximately 52 CO 2 -bearing fluid inclusions, the vapour phase composition was determined. It is dominated by CO 2 and N 2 with CO 2 between 68% to 100%. Ninety percent of the fluid inclusions have a CO 2 content ≥95%. H 2 S, CH 4 and H 2 were only detected as trace constituents (<1%) of the vapour phase. No difference in vapour phase composition was detected between solid-bearing inclusions and the inclusions without solids.
The majority of the solids hosted in the fluid inclusion in Qz-2 are sulfates (Fig. 8). They include arcanite, anhydrite, celestine, aphthitalite and some other unidentified sulfates (Table 2); a few calcite crystals were also found (Figs 6a-f; 9). Note that celestine and glauberite have very similar Raman spectra and they can be hard to differentiate; when this was the case they are noted as celestine/glauberite in Appendix 2. Whewellite, thénardite, anglesite, dolomite, molybdenite, gypsum, muscovite and an unidentified phosphate mineral were also detected in fluid inclusions ( Table 2). The opaque solids could not be identified as they moved under the laser beam.  Table 1. Solid inclusions also occur hosted directly in the quartz and these are predominantly sulfates (mainly anhydrite with one unidentified sulfate) and calcite.

Qz-3 fluid inclusions
Sulfate was only detected in about half of the 27 LCO 2 + VCO 2 + Lw fluid inclusions analysed from Qz-3 (Table 5) and, as with Qz-2, fluid inclusions with and without detectable sulfate in solution occur within the same inclusion trails. Like Qz-2, CO 2 is the main component of the vapour phase, but in these inclusions is always >99%, with N 2 occurring in trace amounts only. The majority of the solid inclusions hosted directly in Qz-3 are calcite with some unidentified phosphate and sulfate particles (Fig. 8).
The aqueous portions of two Lw fluid inclusions were analysed and SO 4 was detected in both. About 75% of the 16 Lw fluid inclusions also had detectable sulfate in solution (Table 5).
Compared to Qz-2 fluid inclusions, Qz-3 fluid inclusions lack sulfate solids and sulfate is not always detected in solution. Likewise, there are fewer solid sulfate inclusions in Qz-3.

Fluid-inclusions origin and reconstructed composition
The LCO 2 + VCO 2 ± Lw ± S(s) fluid inclusions in the Qz-2 and Qz-3 generations have variable CO 2 /aqueous phase ratios (Fig. 6a,h), from some that are nearly pure CO 2 inclusions to others than contain multiple solids, occurring in the same inclusion trails. This suggests that the fluid inclusions result from heterogeneous trapping of different phases that coexisted at the time of fluid-inclusion formation, including an aqueous liquid, supercritical CO 2 and sulfate solids. Although the solids could be daughter minerals, formed after trapping during cooling of a homogeneous fluid, the compositional variability (Figs 5; 7) and the fact that they show no signs of dissolution during heating before the inclusions decrepitate between 225°C to 325°C, makes this unlikely. Accidental trapping is the preferred interpretation, particularly as some of the same solid phases also occur in the quartz as solid inclusions of similar size (Fig. 6a). We infer that these solids were present in the fluid while Qz-2 was being precipitated. Hence, the composition of these inclusions does not represent any single fluid. Regardless of this limitation, we have attempted to estimate the compositions of a number of the individual inclusions.
Due to the chemically complex sulfate-rich fluid and sulfate solids within the fluid inclusions, interpretation of the microthermometric measurements is problematic due to the lack of knowledge of the precise chemical system. The presence of sulfate crystals in these fluid inclusions precludes conventional estimation of their dissolved load as NaCl equivalent salinity. Instead, we used visual proportion estimates to calculate the bulk composition of individual solid-bearing fluid inclusions (Shepherd et al., 1985;Fig. 6a;Appendix 2). Due to the clear evidence of heterogeneous trapping, these estimates represent the composition of the material trapped in the fluid inclusions, not a true fluid composition.
Fluid inclusion DSG 002 4 (Fig. 6a) is a LCO 2 + VCO 2 + Lw + S fluid inclusion hosting a single crystal of arcanite consisting of 15% of the inclusion by volume, with 15% CO 2 -rich vapour phase and 70% arcanite-saturated aqueous phase (the volumes were estimated using visual proportion estimates as presented in Appendix 2). These proportions give a bulk fluid-inclusion composition of 33 wt.% K 2 SO 4 equivalent or ∼3 molal K 2 SO 4 . Some other solid-bearing fluid inclusions have multiple sulfate crystals taking up >15% of the fluid-inclusion volume and must therefore have an even higher sulfate content (Fig. 6). The sulfate-bearing LCO 2 + VCO 2 + Lw fluid inclusions hosted in Qz-2 have similar petrographic characteristics to the LCO 2 + VCO 2 ± Lw ± S(s) fluid inclusions. Hence, their salinities have also been estimated as equivalent K 2 SO 4 using the visual estimate proportion method assuming a K 2 SO 4 saturated solution. The majority of these fluid inclusions have the CO 2 bubble occupying from 10% to 60% of the fluid inclusion. This corresponds to a salinity between 6-10 wt. % K 2 SO 4 equivalent, or <0.7 molal K 2 SO 4 .  Interpretation and discussion

Origin of the quartz lenses
The systematic localisation of the quartz lenses within the dyke cores ( Fig. 1) indicates they are genetically linked to the calcite carbonatite and formed in the same stress field, which is consistent with the lack of quartz lenses in the surrounding rocks. The quartz lenses are not intergrown with the calcite indicating they are not conventional magmatic phases. Both Qz-2 and Qz-3 are believed to be of hydrothermal origin due to the volume for volume replacement texture observed, their characteristic non-luminescent CL emission (Fig. 4) and the presence of fluid inclusions. The Qz-3 Lw + V fluid inclusions Th TOT temperatures point to low hydrothermal temperatures (120-222°C).
The origin of the original quartz (Qz-1 in this study) from the Huanglongpu calcite carbonatites, which accounts for just over 50% of the quartz lenses in our samples overall, has been discussed by Song et al. (2015). They suggested a magmatic origin of the quartz from Si enriched carbonatitic liquids which might have been produced by intensely fractional crystallisation of nonsilicate minerals. Bai et al. (2019) expanded on the work of Song et al. (2015) and concluded that quartz precipitated mainly before calcite, which is consistent with the calcite veins locally crosscutting the quartz lenses observed in this study, but must also be reconciled with the fact that quartz lenses only occur within calcite carbonatite dykes. Bai et al. (2019) also favoured a magmatic origin, but they ascribed the coarse, nearly monomineralic quartz grains to a supercritical C-H-O-Si magmatic fluid intermediate between a hydrous melt and an aqueous solution. Our stable oxygen isotope data (Table 4) indicates very small fractionation between quartz and calcite, which is indicative of temperatures in excess of 500°C (Zheng, 1999) and is consistent with a magmatic environment (Table 4). We conclude that the quartz veins formed from a fluid of magmatic origin, but were subsequently reworked by hydrothermal activity.

Conditions of hydrothermal activity
There is no direct measurement of the temperature of Qz-2 and Qz-3 precipitation but the quartz luminescence and the fluid inclusions provide some indications. The brighter luminescence of Qz-1 compared to the later quartz generations indicates a higher temperature for growth. The later quartz generations exhibit low CL emissions, typical of hydrothermal quartz (Rusk and Reed, 2002;Rusk et al., 2006). Both Qz-2 and Qz-3 show very similar luminescence (Fig. 4) and on the basis of the literature on non-luminescent hydrothermal quartz precipitation, they have probably precipitated at temperatures below ∼450°C.
The CO 2 -bearing fluid inclusions hosted in Qz-2 and Qz-3 decrepitated upon heating before homogenisation. It is likely that decrepitation only occurs above the temperature of trapping and, if so, the maximum trapping temperature is below the range 225 to 325°C. The homogenisation temperature Th TOT of the Lw + V fluid inclusions hosted in Qz-3 give a temperature range from 120 to 220°C which may be close to the true trapping temperature as the system was experiencing significant fluid flow. This is consistent with the low luminescence from Qz-2 and Qz-3 (Fig. 4) and with the association of hydrothermal bastnäsite-(Ce) with them. Shu and Liu (2019) estimated that, in the Dalucao carbonatites, hydrothermal bastnäsite-(Ce) formed between 147°C and 323°C while the pervasive hydrothermal bastnäsite-(Ce) stage of the Maoniuping carbonatite deposit also crystallised at low temperature, between 160°C to 240°C (Zheng and Liu, 2019). Both these studies invoke a fluid with high (SO 4 ) -2 content, alongside Cland F -. It should be noted that the similarity in oxygen isotope composition between different quartz generations (Table 4) suggests that the hydrothermal fluid from which they grew had an oxygen composition dominated by interaction with magmatic silicates under similar conditions to those at which the hydrothermal quartz was precipitated.

Possible sources of hydrothermal fluids and their dissolved components
This investigation has demonstrated that the fluid inclusions in the hydrothermal quartz (Qz-2 and Qz-3) are very variable. Where solids are present they vary in both abundance and mineralogy between adjacent inclusions. The ratio of CO 2 vapour phase to aqueous fluid is also variable. We also note that the sulfate-rich fluid could not have also transported all the associated cations, as baryte and celestine are nearly insoluble (Blount, 1977;Monnin, 1999). We conclude from these lines of evidence that the population of solid-bearing inclusions results from heterogeneous trapping, but the issue of how such a complex, multi-phase fluid came to be present must still be explained.
We note however that similar CO 2 -bearing fluid inclusions with sulfate solids and opaque crystals hosted in quartz and calcite from a quartz-bearing carbonatite deposit in the north Qinling region were attributed by Song et al. (2016) to trapping of a carbonatite fluid exsolved at subsolidus conditions. Sulfate-rich fluids have also been found in the Maoniuping REE deposit, from the Himalayan orogenic belts, China within a similar carbonatite setting (Xie et al., 2009(Xie et al., , 2015. They described extremely sulfate-rich melt-fluid inclusions in fluorite and suggested a sulfate-rich supercritical orthomagmatic fluid originated from the unmixing of the carbonatite melt. Alkali-sulfate salt melt inclusions have been identified in carbonatites by Panina and Motorina (2008). There are two difficulties in applying these models to the examples described here. Firstly the heterogeneous character of the inclusions makes it very unlikely that their contents correspond to any single fluid. Secondly, at Dashigou, no melt inclusions or other sulfate-rich phases are present in the early, high temperature Qz-1. Instead, we infer that the variety of the fluid inclusions provides evidence for at least three different types of fluids associated with hydrothermal Qz-2 and Qz-3: a sulfate-rich aqueous fluid, a more dilute aqueous fluid and a carbonic fluid. Although the sulfate content of these hydrothermal fluids might have been derived directly from a carbonatite parent, this cannot be demonstrated, and it is possible that the sulfate is derived from oxidation of reduced sulfur under surface conditions. Acid sulfate geothermal waters are common in many modern volcanic geothermal systems formed at similar temperatures to those inferred for the hydrothermal activity here (Lewis et al., 1997;Smith et al., 2010). Boiron et al. (1999) described sulfate-rich fluid inclusions associated with retrograde quartz veins from Ouro Fino (Brazil) and pointed out that the composition of the fluid was similar to some present day geothermal systems where feldspars have been destroyed and so cannot buffer the Na/K ratio. However, S isotope data from sulfides and sulfates are consistent with a magmatic sulfur source with δ 34 S of ∼1‰ (Huang et al., 1984;Song et al., 2016).
The more dilute fluid present in many Qz-3 fluid inclusions is also consistent with the introduction of a fluid of geothermal origin, as it is common in modern geothermal fields for fluids with different ligand concentrations to occur within the same system (Henley and Hedenquist, 1986;Lewis et al., 1997).
The carbonic fluid observed in the majority of the fluid inclusions hosted in Qz-2 and Qz-3 could have resulted from the dissolution of calcite within the carbonatite (Fig. 10, model 1). Nonetheless, CO 2 -rich fluids have also been commonly described as orthomagmatic fluids resulting from carbonatite magma unmixing (Rankin, 2005;Bühn et al., 2002).
It is unlikely that metal and sulfate ions now present in Qz-2 inclusions were introduced in a single fluid. It is clear from the varied mineralogy of the inclusions that the proportions of major cations in sulfate inclusions are very variable, and in any case the bulk fluid composition cannot have existed as a single phase under the conditions inferred for the hydrothermal activity, if at all. The Maoniuping REE carbonatite deposit host sulfate-rich multisolid daughter-crystal fluid inclusions (Xie et al., 2015; e.g. ADC and ADV fluid inclusions), which behaved differently as they did meet the expectations for a single-phase origin. However, unlike these fluid inclusions, none of the CO 2 -bearing fluid inclusions at Dashigou homogenised upon heating, which is evidence for heterogeneous trapping.

Model for the hydrothermal alteration at Huanglongpu
Two possible models for the hydrothermal alteration and HREE enrichment of the calcite carbonatites at Dashigou, based on inclusion mineralogy and fluid-inclusion results are presented in Fig. 10. They are intended to demonstrate how a sulfate-rich fluid is able to interact with a source of cations to produce a multi-phase fluid.
Model 1 is based on the reaction of a single acid sulfate fluid with the calcite carbonatite and its hosts. This generates CO 2 and precipitate sulfates by the neutralisation of the acid fluid by wallrock reactions (e.g. H 2 SO 4 + CaCO 3 = CaSO 4 + CO 2 + H 2 O). The sulfate minerals present in the fluid inclusions and in quartz are mainly salts of K (arcanite), Ca (anhydrite), Sr (celestine) and Na (aphthitalite) (Fig. 9), and these cations are readily available within wallrock calcite carbonatite dykes or the surrounding fenite (e.g. H 2 SO 4 + 2KAlSi 3 O 8 = K 2 SO 4 + Al silicates). In this model, local production of CO 2 explains why it is associated closely with the sulfate fluid.
Model 2 involves the mixing of two externally-derived fluids, one sulfate-rich and one CO 2 -rich, with cations carried by the carbonic fluid, and does not explicitly require chemical interaction with the carbonatite or fenite. This model can also result in the precipitation of sulfate salts, but requires cations to be introduced in a carbonic fluid even though they, and the CO 2 , are already available in the rocks.
Irrespective of the origin of the initial alteration and precipitation of sulfates, the growth of Qz-3 appears to be associated with the introduction of a more dilute fluid to the system and this is also shown in Fig. 10.
While Qz-2 occurs as thread-like infills linked by a cobweb texture, Qz-3 always occurs infilling healed fractures (Fig. 4). We propose a late fracturing event led to the introduction of the more dilute water, leading to Qz-3 precipitation and the less saline fluid inclusions. Loss of CO 2 at this stage is likely as aqueous Lw + V and Lw fluid inclusions become common (Fig. 10).

Hydrothermal REE enrichment
The HREE enrichment of the Dashigou carbonatite leading to the uncommon flat REE pattern (Fig. 3)    HREE-enrichment model in which magmatic, HREE-enriched calcite with a relatively flat REE pattern (their fig. 7a) provided a baseline source for the secondary REE mineralisation. The magmatic calcite has lower REE contents than the magmatic REE phases, but it accounts for over 50% of the carbonatite deposit and is more reactive than the REE minerals, making calcite a viable HREE-enriched source. The primary magmatic LREE-rich mineralisation consisted of monazite-(Ce) (e.g. sample HLP15, ΣLREE/(HREE+Y)= 587470, table 4) and fluorocarbonate (bastnäsite-(Ce) and parisite-(Ce)) mineralisation. These early REE minerals show steep, LREE enriched, chondrite-normalised patterns as demonstrated by their high ΣLREE/(HREE + Y) ratio (Smith et al., 2018; fig. 8A-B). The sulfate-rich fluid leached the REE from the magmatic carbonatite without fractionation and then precipitated secondary HREE-enriched phases with flat REE patterns (e.g. britholite-(Ce), Smith et al., 2018; sample HLP10, ΣLREE/(HREE + Y)= 11, fig. 8C-E; table 6). The hydrothermal REE minerals are still Ce-dominant but have much higher HREE content than found typically in carbonatite related systems.
Hydrothermal Qz-2 and Qz-3 grew alongside secondary REE minerals as the carbonatites became enriched in HREE (Smith et al., 2018), suggesting that the fluid inclusions contain information about the fluid responsible for the REE transport and deposition. The associated fluid inclusions have high sulfate contents and Qz-2 and Qz-3 emplacement correlates overall with widespread growth of sulfate minerals and deposition of secondary quartz in the calcite carbonatite dykes.
Rare earth element transport is also likely to be an important factor for HREE-enrichment. Williams-Jones (2008, 2014) and Migdisov et al. (2016) identified two categories, ligands leading to REE transportation (Cland SO 4 2-) and ligands leading to REE deposition (F -, OH -, CO 3 2-, HCO 3 and PO 4 2-). At Dashigou, sulfate is inferred to be the main complexing agent Fig. 9. Laser Raman spectroscopy spectra of some trapped solids hosted in Qz-2 fluid inclusions from the Dashigou calcite carbonatites note that the background signal from the quartz were subtracted. (a) Arcanite spectrum of a LCO 2 + VCO 2 + Lw + 3S, the solid on the right is calcite and the solid in the middle is an unidentified sulfate solid (DSG 433 2f-1). (b) Anhydrite spectrum of a LCO 2 + VCO 2 + Lw + S (DSG 002 13a-1). (c) Celestine spectrum of a LCO 2 + VCO 2 + Lw + 2S, the celestine crystal is accompanied with an unidentified opaque (DSG 437 8b-1). (d) Aphthitalite spectrum in a LCO 2 + VCO 2 + Lw + 3S, the opaque and the small transparent solid are unidentified (DSG 433 6a).
for REE transport from the fluid-inclusion data. The flat REE profile of the hydrothermally altered calcite carbonatites (Fig. 3) is consistent with experimental studies on the REE-SO 4 2complexes at hydrothermal temperatures from 100°C to 400°C Williams-Jones, 2008, 2014;Migdisov et al., 2016). These studies agree that sulfate is a non-selective ligand as it does not fractionate the LREE from the HREE. The stability of REE-SO 4 2complexes is not strongly dependent on atomic number and they have similar formation constants across the lanthanide series. In contrast, HREE-Cl 2+ complexes show an overall lower stability than REE-sulfate complexes and as this decreases along the lanthanide series, a chloride-dominated fluid will fractionate the LREE from the HREE. In contrast a sulfate fluid is able to transport the LREE and the HREE equally. One case study on a natural carbonatite system has demonstrated previously the importance of REE sulfate transport at low temperature below 240°C (Zheng and Liu, 2019). Sulfate transport has also been observed in sedimentary basin systems at low hydrothermal temperature between 100°C and 120°C (Richter et al., 2019).
The chloride content of the Dashigou fluids were not measured but no trace of Cl-bearing salts were observed suggesting only a minor amount of this ligand is present in the fluids. The precipitation of sulfate minerals within the calcite carbonatite dyke as the sulfate fluid encountered Ca, Na and K is inferred to have destabilised the REE-SO 4 2complexes and triggered the deposition of the hydrothermal REE minerals. While sulfate transport does not account for the HREE enrichment specifically, it plays an important role in linking REE mineralisation to the growth of sulfate minerals. The final HREE enrichment factor is thought to be the stabilisation of secondary minerals whose crystal structures tend to favour HREE, in particular HREE-enriched fluorocarbonates, uraninite, xenotime-(Y), churchite-(Y) and other minerals which were not always possible to identify; Smith et al. (2018) also reported britholite-(Ce) and Ca-REE fluorocarbonate mineralisation. These fractionate the HREE out of the fluid they are crystallising from. An additional mechanism that could also have concentrated the HREE is the co-precipitation of hydrothermal gangue LREE enriched phase such as sulfate minerals.
In summary, a HREE-enriched source is probably important for the final HREE enrichment. The late-stage magmatic HREE enrichment of the calcite, the main gangue mineral of the carbonatite, provides a source baseline allowing a potential secondary HREE enrichment (Xu et al., 2007;Smith et al., 2018;Bai et al., 2019). However, it is also possible that HREE-enrichment is enhanced at the hydrothermal stage. This might have happened indirectly, by selective removal of LREE in a Cl fluid (Smith et al., 2018), although no evidence for such a fluid has been seen, or directly through transport and precipitation of REE. Sulfate-REE complexation facilitates non-selective transport of the REE, and precipitation of REE minerals might then occur in response to precipitation of sulfate minerals. Stabilisation of secondary HREE-enriched minerals in the hydrothermal environment could have selectively concentrated HREE.

Conclusions
The calcite carbonatites at Dashigou can be divided into two groups on the basis of hydrothermal alteration of magmatic minerals, including monazite-(Ce); little-altered calcite carbonatites exhibit a normal carbonatite LREE-enriched REE pattern, but more altered dykes have enhanced levels of HREE and show an unusual flat REE pattern (Fig. 3). These observations suggest that the rocks became enriched in HREE in the course of the hydrothermal alteration which was accompanied by the growth of secondary minerals which favoured incorporation of HREE (Smith et al., 2018). Alteration took place in at least two phases whose conditions are not well-defined but are between 450°C and 120°C. The hydrothermal fluid responsible for the alteration was remarkably sulfate-rich, and the bulk composition of heterogeneously-trapped fluid inclusions can exceed 30 wt.% K 2 SO 4 equivalent. The REE were transported as sulfate complexes and probably precipitated due to the precipitation of sulfate gangue minerals. The simplest model for the process involves interaction of an externally-derived acid sulfate hydrothermal fluid with carbonatite and fenite (Fig. 10). Calcite and silicates neutralised the acidity of the sulfate-rich fluid, generating CO 2 and acquiring cations (K, Ca and Na) that precipitated sulfates. Late fracturing was accompanied by renewed quartz precipitation from a more dilute fluid lacking CO 2 .
Overall, the HREE enrichment of the Dashigou carbonatite is believed to be due to a combination of factors. Late-magmatic HREE enrichment of the calcite, the main mineral of the carbonatite, provided a source baseline. Secondary HREE-enrichment was made possible by a fluid rich in sulfate, a non-selective REE ligand able to transport and concentrate the REE equally, but in complexes that are destabilised by the precipitation of sulfate minerals. Final HREE-enrichment was achieved by the growth of secondary REE minerals which favour HREE, notably HREE-enriched fluorocarbonates, uraninite, xenotime-(Y) and churchite-(Y).