Mineralization of the Bayan Obo Rare Earth Element Deposit by Recrystallization and Decarbonation

The rare earth elements (REEs [La-Lu + Y]) are classified as critical metals because of their importance in renewable energy and high-technology applications (Goodenough et al., 2018). Although many minerals and rock types contain appreciable levels of REEs, minable concentrations of these elements are restricted to relatively few deposits worldwide (Chakhmouradian and Wall, 2012). China continues to dominate the global supply of REEs but has been gradually reducing its production and exports because of environmental concerns (Xie et al., 2016). Consequently, there has been an increasing international effort to locate new sources of these elements. The Bayan Obo deposit is the world’s largest deposit of REE ores and produces the bulk of REE exports from China (Xie et al., 2016). Dolomite rock of debatable origin (sedimentary rock vs. carbonatite; Chao et al., 1997; Xu et al., 2008) hosts most of the Bayan Obo orebody. The genesis of the REE mineralization has been hotly debated since the discovery of the deposit in 1927, including suggestions of a synsedimentary origin (Chao et al., 1997), metasomatic reworking of sedimentary carbonate rocks by carbonatitic (Smith and Henderson, 2000; Yang et al., 2009) or subduction-derived fluids (Ling et al., 2013), and a carbonatite source (Xu et al., 2008). New geochemical evidence supports the interpretation that the dolomite rock is of carbonatitic origin and that the first mineralization stage took place in the Mesoproterozoic (Song et al., 2018; Yang et al., 2019; Chen et al., 2020). Numerous lines of evidence have suggested a long history of the deposit, possibly extending from ~1300 Ma to ~400 Ma (Zhang et al., 2003; Zhu and Sun, 2012; Smith et al., 2015; Song et al., 2018). However, there is no consensus concerning the number of mineralization stages, with some studies suggesting as many as eleven (Chao et al., 1992). The dolomite rock displays a stratiform-like structure and evidence of strong metasomatic reworking (Smith et al., 2015). The rocks of the Bayan Obo deposit are characterized by a wide range of crustally derived and mantle-derived radiogenic and stable isotopic compositions, and therefore the provenance and chemical composition of the postdepositional metasomatic fluids that interacted with the dolomite rock are still poorly understood. Several distinct styles of hydrothermal alteration have been recognized and occur in complexly superposed mineral assemblages containing fluorite, riebeckite, aegirine, and phlogopite (Smith, 2007; Smith et al., 2015; Deng et al., 2017). The ambiguity of textural and temporal relations between the primary and alteration parageneses complicates interpretation of the REE mineralization.

Carbonatites are known to contain the highest concentrations of REEs (n × 10^2–4 ppm) of any igneous rock and are considered to be an exploration target of major importance. Although there are more than 500 known carbonatites in the world, only a few are currently being mined for the REEs (e.g., the Bayan Obo, Maoniuping and Dalucao, and Muluozhai deposits in China, Mountain Pass in the United States, and Mount Weld in Australia; Weng et al., 2015; Smith et al., 2016; Verplanck et al., 2016). The key question that needs to be addressed in the case of Bayan Obo is how a dolomite carbonatite became the host for this giant REE deposit, dwarfing in size the REE deposits hosted by carbonatites elsewhere.

Here, we report in situ analyses for C and O stable isotopes and Sr and Nd radioisotopes of precursor and recrystallized dolomite grains and REE minerals. The samples were obtained from the dolomite host rock, recovered as part of a 1,776-m-long drill core. The C, O, and Sr isotope compositions provide new constraints on the nature of metasomatic fluids and mineralization processes at Bayan Obo. A combination of in situ dating and Nd isotope analyses of REE minerals provides a clear record of two mineralization stages, in which the REE grade of the Mesoproterozoic carbonatites was enhanced by recrystallization and CO₂ degassing at ~400 Ma.

The Bayan Obo REE deposit is located on the northern margin of the North China craton, bordered by the Bainaimiao arc to the north (Fig. 1). The basement rocks in this region consist of Archean-Paleoproterozoic metamorphic rocks that are unconformably overlain by an ~8.5-km-thick sequence of the Mesoproterozoic Bayan Obo Group rocks. These sediments were deposited within the Bayan Obo marginal rift during the Mesoproterozoic continental breakup of the North China craton (Zhang et al., 2017). The Mesoproterozoic rifting was accompanied by the emplacement of basalts, basaltic trachyandesites, and carbonatite dikes (Yang et al., 2011). The Bayan Obo Group comprises 18 lithological units (H1-H18), including metasandstones, limestones, slates, dolomite rocks, and metavolcanics. The H1-H9 units, in ascending order from base to top, are all exposed in the Bayan Obo area (Bai and Yuan, 1985). Chao et al. (1997) classified the H1-H8 units as metasedimentary rocks. The presence of quartz-hosted melt inclusions, however, shows that the H2 unit is of magmatic origin (Xie et al., 2020). The REE-Fe orebodies are hosted in the H8 dolomite, whereas the H9 slate unit is the dominant wall rock for the Bayan Obo orebodies (Zhang et al., 2003). The H9 unit has been divided into four subtypes: biotite type, calcite-biotite type, calcite type, and K-feldspar type, according to mineral assemblages. Xie et al. (2020) suggested that the calcite-biotite type is a carbonatite, as the mineral assemblage and texture are similar to those of a biotite-rich calcite carbonatite. Near the ore deposit, a few dozen Mesoproterozoic carbonatite dikes, composed mainly of dolomite, calcite, and rare Sr, Ba, and REE carbonate minerals, intruded the Bayan Obo Group low-grade metasediments and basement rocks, which were fenitized adjacent to the margins of the dikes (Le Bas et al., 2007). Based on their mineralogical composition, the dikes have been classified as dolomite, dolomite-calcite, and calcite carbonatites (Wang et al., 2002). Late Paleozoic Hercynian dioritic-granitic plutons were emplaced to the southeast of the deposit and include quartz monzonite, monzonitic granite, and biotite granite (Zhang et al., 2003; Ling et al., 2014).

The H8 unit extends sublatitudinally over a distance of 18 km, has a maximum width of 3 km in plan view, and is composed of a spindle-shaped stratified body. Rare xenoliths of mafic metamorphic rock have been found in the H8 carbonatites (App. Fig. A1A). The unit comprises texturally variable (inequigranular, fine- to coarse-grained, massive to foliated) dolomite, with subordinate riebeckite, phlogopite, fluorite, fluorapatite, magnetite, and REE minerals (App. Fig. A1B-D). The proportions of the minerals in the dolomite rocks are quite variable. Depending on the sample, the proportions of riebeckite and phlogopite may vary from 1 to 20 vol %, and the fluorite content varies from 1 to 20 vol %. The H8 rocks underwent extensive deformation and metasomatic reworking. It contains three main types of ore, namely disseminated, banded, and massive. The REE contents of the three ore types are <3, 3 to 6, and 6 to 12 wt % REE₂O₃, respectively (Xu et al., 2008). The studied drill core is from the East orebody and varies from 1 to 6 wt % in its total light (L)REE (La-Sm) content (Song et al., 2018).

The major element compositions of dolomite crystals in the drill core were analyzed using a JXA-8100 electron microprobe (EMP) at Peking University. The operating conditions were a 15-kV accelerating voltage and a beam current of 10 nA, with an electron beam defocused to 10 μm. A set of appropriate matrix-specific standards and optimal instrumental conditions (detector type, beam settings, and counting statistics) were carefully chosen by performing multiple measurements. All raw data were corrected using a ZAF (Z, atomic number; A, absorption; F, fluorescence) procedure.

In situ laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS) at Peking University was used to measure the abundances of selected trace elements in the dolomite, with an Agilent 7500Ce mass spectrometer coupled to a 193-nm ArF excimer laser. The diameter of the ablation spot ranged between 40 and 60 μm depending on the size of the individual mineral grains. The measurements were done at a laser energy density of 5.5 J/cm² and a repetition rate of 5 Hz. Standard NIST610 glass was used for external calibration, and the CaO content of the same minerals determined independently by EMP analysis was used as an internal standard. The analytical uncertainty for most trace element concentrations was within 5% based on repeated analyses of standards NIST 612 and NIST 614.

In situ Th-Pb dating of monazite and bastnäsite in polished thin section was carried out using a quadropole ICP-MS (Agilent 7700×) with a laser ablation system (ASI RESOnetics S-155, 193-nm wavelength) at Nanjing FocuMS Technology Co. Ltd. Analyses were performed with a beam diameter of 24 μm and a repetition rate of 6 Hz. The initial conditions were set at ThO⁺/Th⁺ <0.3% to minimize the production of molecular compounds. Each analysis consisted of an approximately 20-s background acquisition and a 65-s sample acquisition. The monazite standard 44069 (424 ± 1 Ma; Aleinikoff et al., 2006) and the bastnäsite standard K-9 (118 ± 1 Ma; Sal’nikova et al., 2010) were used as the external standards to monitor the instrumental shift and the laser-induced U-Th-Pb fractionation. Offline data selection and integration were performed by using ICPMSDataCal software, and age calculations were processed using the ISOPLOT program (Ludwig, 1994). The uncertainties associated with the age determinations are quoted at 1σ and ages were calculated at the 95% confidence level. Additional details of the analytical procedure are presented in Yang et al. (2014).

After electron microprobe analysis and scanning electron microscope-backscattered electron (SEM-BSE) imaging, small pieces of polished thin sections of dolomite samples were cut and mounted in Buehler EpoFix epoxy resin, together with the following carbonate reference materials: Hammerfall dolomite, UW6220 dolomite, and GTS144 ankerite. In situ C and O isotope analyses of carbonates were carried out using an ion microprobe SHRIMP-SI in the Research School of Earth Sciences, Australian National University. The analytical conditions were similar to those outlined in detail by Hu et al. (2018). Analyses of the standards were typically performed after every three to five unknowns to correct for instrumental mass-dependent fractionation. A 10-kV, 2-nA primary beam of ¹³³Cs⁺ ion was focused to a ~6-μm diameter on the sample surface. Negative ions of ¹²C^–, ¹³C^–, ¹⁶O^–, and ¹⁸O^– were measured with Faraday cups. All the data for the unknowns were calibrated initially against Hammerfall dolomite (δ¹³C_{Vienna-PeeDee Belemnite (V-PDB)} = –0.28 ± 0.07‰, and δ¹⁸O_{Vienna-standard mean ocean water (V-SMOW)} = 21.43 ± 0.02‰, 2σ; Hu et al., 2018) and further corrected for matrix effects, using a method similar to that of Śliwiński et al. (2016a, b). Plots of in situ C isotope ratios versus Fe contents of the dolomite from different paragenetic settings do not show any obvious correlation (App. Fig. A2). The corrected ¹⁸O/¹⁶O ratios are reported in standard δ¹⁸O notation, relative to SMOW, and the ¹³C/¹²C ratios are reported relative to V-PDB.

The in situ Nd isotope compositions of monazite and bastnäsite in thin section were measured by multi-collector (MC) ICP-MS using a Thermo-Finnigan Neptune instrument coupled to a 193-nm ArF excimer laser ablation system at the Institute of Geology and Geophysics, Chinese Academy of Sciences. Prior to analysis, the Neptune MC-ICP-MS was tuned and optimized for maximum sensitivity using JNdi-1 standard solution. The diameter of the laser spot and frequency were adjusted to between 20 and 24 μm and 4 and 10 Hz, respectively, depending on the Nd concentration and the sizes of individual mineral grains in the thin sections. Each spot analysis incorporated an approximately 60-s signal acquisition. The ¹⁴⁷Sm/¹⁴⁹Sm and ¹⁴⁷Sm/¹⁴⁴Sm values were used to calculate Sm mass bias and ¹⁴⁴Sm composition, respectively (Yang et al., 2008). The ¹⁴⁶Nd/¹⁴⁴Nd ratio was used as interference correction for ¹⁴⁴Sm. Normalized ¹⁴³Nd/¹⁴⁴Nd and ¹⁴⁷Sm/¹⁴⁴Nd isotope ratios were calculated using the exponential law (Depaolo and Wasserburg, 1976). The ¹⁴⁷Sm/¹⁴⁴Nd ratio was then further calibrated externally against the ¹⁴⁷Sm/¹⁴⁴Nd ratio of monazite standards, Jeffson and Nama (Liu et al., 2012), during the analytical sessions. The average measured ¹⁴³Nd/¹⁴⁴Nd of Nama is 0.511886 ± 22 (2σ, n = 22), and ¹⁴³Nd/¹⁴⁴Nd of Jeffson is 0513087 ± 24 (2σ, n = 9); these are consistent with the recommended values (¹⁴³Nd/¹⁴⁴Nd= 0.511896 ± 32 [2σ; Nama]; ¹⁴³Nd/¹⁴⁴Nd = 0.513057 ± 93 [2σ; Jeffson]; Liu et al., 2012). More detailed information on the in situ Nd isotope analytical procedure employed in the present work is available in Yang et al. (2008).

In situ Sr isotope compositions of dolomite were measured using a RESOlution laser ablation system coupled to a Nu Plasma II MC-ICP-MS at the State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences (Wuhan). The analyses of dolomite grains were carried out using a spot size of 50 to 75 μm with a repetition rate of 10 Hz and an energy density of 4 to 5 J/cm². The isotopic ratios were quantified in static multicollector mode at low resolution, using seven Faraday collectors and a mass configuration array from ⁸²Kr to ⁸⁸Sr to monitor variations in Kr, Rb, and Sr signals. An in-lab coral standard was analyzed as the external standard. The average ⁸⁷Sr/⁸⁶Sr isotope composition obtained for the coral standard is 0.70920 ± 8 (2σ, n = 20), which corresponds to the recommended value of 0.70923 ± 4 (2σ; Pu et al., 2005). The analytical procedure and data reduction strategy are described in more detail in Tong et al. (2015).

The C-O isotope compositions of dolomite in drill core were measured at the Institute of Geochemistry, Chinese Academy of Science, using an Isoprime continuous-flow isotope-ratio mass-spectrometer (Elementar). The results are reported in conventional delta notation, as per mil (‰) variation relative to the V-PDB and SMOW standards for carbon and oxygen, respectively. The Sr and Nd isotope compositions of the whole rocks were analyzed in solution using a VG AXIOM MC-ICP-MS instrument at Peking University. Mass fractionation corrections for the Sr and Nd isotope ratios were normalized to an ⁸⁶Sr/⁸⁸Sr ratio of 0.1194 and an ¹⁴⁶Nd/¹⁴⁴Nd ratio of 0.7219, respectively. Repeated measurements of standards NBS987 and JNdi yielded an average ⁸⁷Sr/⁸⁶Sr value of 0.710254 ± 14 (2σ, n = 6) and an ¹⁴³Nd/¹⁴⁴Nd value of 0.512094 ± 34 (2σ, n = 6), respectively, consistent with the recommended values of 0.710248 (NBS987) and 0.512113 (JNdi). To calculate the initial Sr-Nd isotope values, the Rb, Sr, Sm, and Nd concentrations of whole-rock samples were measured in solution by ICP-MS (Thermo Fisher Scientific X-Series II). Repeated analyses of well-characterized standards indicate that the accuracy of trace element measurements was better than 10%.

Core from depths of 200 to 1,776 m in drill hole #WK15-05 was collected for petrographic observation and geochemical analyses (App. Fig. A3A, B). The sample interval in the core was ~20 m for thin sections and 50 m for whole-rock analyses, respectively. The examined drill core samples are composed of large amounts of fine-grained and minor coarse-grained dolomite rocks. Some fine-grained dolomite rocks contain relatively large dolomite porphyroclasts (0.4–3.4 × 0.3–1.8 mm in size) with serrated margins immersed in a fine-grained (0.01–0.03 mm across) dolomite matrix (Fig. 2A, B; App. Fig. A3C). The porphyroclasts represent remnants of precursor dolomite after deformation and hydrothermal reworking. The matrix grains are interpreted to be the products of dynamic recrystallization, as shown by the polygonal shapes and triple grain junctions of the crystals (Fig. 2C, D; App. Fig. A3D) and indicated by the locally developed preferred orientation (App. Fig. A3D). Under cathodoluminescence (CL), the dolomite porphyroclasts display weak red-luminescent rims and nonluminescent cores (Fig. 2B), whereas the recrystallized, finegrained dolomite, which is associated with fluorite and REE minerals, is nonluminescent (Fig. 2D). Some red-luminescent dolomite rims are surrounded by bright orange rims, which implies two stages of dolomite growth (Fig. 2E, F).

Monazite (ideally, LREEPO₄) is a common REE mineral in the studied samples. It normally occurs as clusters of minute crystals associated with apatite, which is probably a precursor phosphate phase (Fig. 3A), as well as disseminated grain clusters and in veinlets containing REE fluorcarbonates (particularly, bastnäsite LREECO₃F) and fluorite; monomineralic veinlets of monazite were also observed (Fig. 3B). The Th-Pb ages of monazite and bastnäsite from paragenetically different settings were measured in situ (i.e., in polished thin sections) in H8 unit samples from the 1,482 to 1,769 m depth interval (App. Table A1). When combined with the previously reported Sm-Nd isochron dates of whole-rock and REE mineral samples, the data cover a wide range, with two strong peaks at ~1.3 Ga and ~400 Ma (Fig. 4). Anchimonomineralic monazite veinlets yielded a much more restricted range of dates clustering around 400 Ma (Fig. 3B), in comparison with monazite grains intergrown with bastnäsite and apatite (~340–980 Ma; Fig. 3A). The in situ Th-Pb age analyses were performed on disseminated anhedral (50–200 μm) and large (up to 0.8 cm in length) bastnäsite grains. Their ages cluster at ~420 Ma (Fig. 4).

The precursor and recrystallized dolomite varieties can be clearly distinguished on the basis of their chemical compositions (App. Tables A2, A3). The recrystallized variety has higher Fe and Mn contents relative to the porphyroclasts. Their compositions form two partially overlapping fields, consistent with chemical reequilibration of the porphyroclasts during recrystallization (Fig. 5A, B). Both varieties contain elevated Sr abundances (~2,550 and ~2,070 ppm, respectively), which are similar to Sr levels in carbonatitic dolomite (Chakhmouradian et al., 2016) and higher than the sedimentary carbonate rocks in the Bayan Obo area (<200 ppm; Zhang et al., 2003). The porphyroclast and recrystallized dolomite have similar average chondrite-normalized (CN) REE profiles (Fig. 6). The profiles of dolomite porphyroclasts are smooth and exhibit a slight negative slope (La/Yb_CN = 2–15). The recrystallized variety yields either negatively or positively sloping profiles (La/Yb_CN = 0.2–15) that overlap with those of porphyroclasts. The LREE contents of porphyroclastic and recrystallized dolomite are similar (i.e., La = 2.5–21 vs. 0.6–27 ppm, respectively). Many recrystallized grains exhibit a small positive Eu anomaly (Eu/Eu* = 1.1–2.5), which is not the case for the dolomite porphyroclasts. Lower La/Nd_CN and La/Ho ratios, compared to the precursor dolomite, have been determined in many of the recrystallized grains (Fig. 5C, D), but their Y/Ho ratios are indistinguishable. The absence of any departure of the Y/Ho ratio from typical carbonatitic values (Chakhmouradian et al., 2016) at extremely variable La/Ho values suggests LREE mineral deposition during dolomite recrystallization (Bau and Dulski, 1995).

In situ oxygen isotope analyses of the precursor and recrystallized dolomite yield a similar range of values (δ¹⁸O_V-SMOW = 10.3–16.9‰), whereas the carbon isotope values vary significantly (App. Table A4). The porphyroclasts are generally enriched in heavy C relative to the recrystallized dolomite (– = –1.09 to 2.37 vs. –3.59 to 0.79‰, respectively). The drill core samples yield a similar range of δ¹⁸O_V-SMOW values (12.0–15.3‰) in comparison to the in situ analyses but a more limited range of δ¹³C_V-PDB ratios (–2.3 to 0.4‰), which overlaps with both precursor and recrystallized dolomite compositions (App. Table A5). The C-O isotope signatures of the dolomite samples are distinct from those of both mantle-derived carbonatites and metasedimentary dolomite of the Bayan Obo Group (Fig. 7A).

In situ Sr isotope analyses of the dolomite show a large variation (App. Table A6). The porphyroclasts are characterized by low ⁸⁷Sr/⁸⁶Sr ratios (0.70241–0.70394), which increase slightly towards their rims. The Sr isotope ratios of the recrystallized variety show some overlap with the precursor values but are generally higher and far more variable (0.70288–0.71409). A similar, albeit more limited, range of values was recorded for drill core samples (0.70288–0.70533; App. Table A7). The high Sr content and low Rb/Sr ratio of the Bayan Obo dolomite imply that the present-day ⁸⁷Sr/⁸⁶Sr ratios accurately record the composition of its source and were negligibly affected by ⁸⁷Rb decay. In an ⁸⁷Sr/⁸⁶Sr vs. δ¹³C_V-PDB plot, the two dolomite varieties occupy separate compositional fields with relatively little overlap (Fig. 7B).

The Nd isotope ratios were measured independently for the same monazite and bastnäsite grains and used to calculate the initial ¹⁴³Nd/¹⁴⁴Nd_(t) ratios (App. Table A8). The REE phases gave broadly similar T_CHUR(Nd) model ages ranging from 1.20 to 1.35 Ga, which imply a common LREE source. Combined with the initial Nd isotope ratios of the drill core samples, calculated for the peak ages of 1300 and 400 Ma, the measured REE mineral ages and εNd_(t) values show a robust linear correlation (Fig. 8).

The precursor porphyroclastic dolomite is characterized by low ⁸⁷Sr/⁸⁶Sr ratios, which are consistent with a mantle origin and similar to those of the Mesoproterozoic carbonatite dikes uncontaminated by wall-rock feldspathic material (Le Bas et al., 2007). However, this dolomite is characterized by higher δ¹³C_V-PDB values than those typically observed in mantle rocks (–7 to –5‰; Ray et al., 1999), implying that its source was enriched in ¹³C (or depleted in ¹²C). The possible processes that could cause such enrichment in mantle-derived melts or fluids are fractional crystallization, assimilation of sediments, or addition of subducted crustal materials to the mantle C reservoir (Xu et al., 2014). Fractional crystallization and sediment contamination can be ruled out, as these processes generate concomitant enrichment in ¹³C and ¹⁸O (Ray and Ramesh, 2000), which is not observed (Fig. 7A). Enrichment of ¹³C by the incorporation of recycled oceanic carbonates (δ¹³C_V-PDB ≈ 0‰; Veizer et al., 1992) into the mantle source region remains a possibility. Paleotectonic reconstructions demonstrate that between 2.3 and 1.9 Ga, the northern edge of the North China craton was an active continental margin characterized by southward subduction (in present-day coordinates) and Andean-type magmatism, which culminated with the amalgamation of the North China craton with the Columbia supercontinent by ~1.8 Ga (Kusky et al., 2016). Multiple manifestations of arc magmatism southeast of Bayan Obo (Yang and Santosh, 2015) and the presence of high-pressure eclogite xenoliths derived from a recycled slab in Paleoproterozoic carbonatites ~300 km further inland (Xu et al., 2018) provide unambiguous evidence that subducted material was present in the mantle beneath this part of the North China craton. Subduction processes could have played a key role in REE enrichment of the mantle sources of the Bayan Obo carbonatites (Xu et al., 2014; Hou et al., 2015).

The recrystallized dolomite is indistinguishable from the porphyroclasts in its O isotope signature, but is characterized by lower δ¹³C_V-PDB and higher ⁸⁷Sr/⁸⁶Sr values (Fig. 7B). Their δ¹⁸O_V-SMOW values are significantly higher than the primary mantle-derived carbonatite (Taylor et al., 1967). The negative shift in δ¹³C_V-PDB values in the recrystallized dolomite is probably a reflection of isotope fractionation due to degassing (Suwa et al., 1975; Valley, 1986; Demény et al., 1994). It is well known that the Bayan Obo dolomite rock underwent strong metasomatism (Smith, 2007; Smith et al., 2015; Deng et al., 2017); hence, the observed changes in δ¹³C_V-PDB must be discussed in the context of metasomatic reworking.

Three possible scenarios should be considered for the origin of the metasomatic fluids responsible for the chemical and isotopic reequilibration of the precursor dolomite.

1. Mantle-derived fluids are typically enriched in ¹²C and depleted in ⁸⁷Sr relative to fluids of crustal provenance. Archean-Paleoproterozoic and Mesozoic mantle peridotite xenoliths from the North China craton show that the subcontinental lithospheric mantle is characterized by relatively low Sr isotope ratios (0.7030–0.7060; Zhang et al., 2020; Zou et al., 2020). Thus, a mantle origin for the fluids can probably be ruled out based on the high levels of radiogenic Sr in the recrystallized samples (Fig. 7B).

2. Granite-derived fluids could, in principle, cause the observed shift in ⁸⁷Sr/⁸⁶Sr ratios, but are not likely to reset the C isotope ratio because CO₂ solubility in non-arc silicic-rich melts is very low (Lowenstern, 2001). More importantly, the Bayan Obo granitoids were emplaced during the late Permian (Zhang et al., 2003), i.e., they are younger than the bulk of the REE mineralization (Fig. 4). The older (Silurian-Devonian) granitoids, which intruded the northern margin of the North China craton, on the other hand, have relatively low ⁸⁷Sr/⁸⁶Sr ratios (~0.706; Fig 7B; Zhang et al., 2014); i.e., they cannot explain the extremely radiogenic signature of some of the recrystallized dolomite (Fig. 7B). Considering the abundance of silicate minerals in the H8 unit and the low values of the partition coefficients for Sr between fluids and silicate (or carbonate) melts (Song et al., 2016), the high Sr content and initial ⁸⁷Sr/⁸⁶Sr ratio recorded in the recrystallized dolomite would have required massive fluid infiltration relative to the precursor rocks.

3. Subduction zone processes are commonly proposed as an alternative source of the fluids responsible for the reworking of the H8 unit (Ling et al., 2013). The northernmost margin of the North China craton is delineated by the Solonker suture, along which the Paleo-Asian Ocean closed to form the southern sections of the Central Asian orogenic belt (Fig. 1). The suture initially developed during the accretion of the Bainaimiao arc onto the passive margin of the North China craton at 437 to 453 Ma (Eizenhöfer and Zhao, 2018). Subduction-zone fluids are dominated by aqueous alkali and aluminosilicate components and characterized by elevated Sr and Cl levels (Manning, 2004). With increasing hydration and proximity to metasedimentary rocks at the top of the subducted slab, an aqueous (Sr, Cl, Si)-bearing fluid would have a radiogenic ⁸⁷Sr/⁸⁶Sr signature consistent with our data (Scott et al., 2019; Fig. 7B). Furthermore, fluid-inclusion studies indicate that the ore-forming fluids involved in the development of the Bayan Obo REE mineralization mostly have compositions in the H₂O-CO₂-NaCl system (Smith and Henderson, 2000).

Here, we propose that the H8 dolomite carbonatite underwent textural, trace-element, and isotopic reequilibration with an externally derived fluid, which caused the observed C-Sr isotope decoupling. The negative shift in δ¹³C_V-PDB and precipitation of alkali silicates (phlogopite, riebeckite) and fluorite can be modeled as Rayleigh devolatilization involving a fluid capable of mobilizing fluorine and silicon from the subduction zone:

This formulation provides useful constraints on isotope fractionation in natural systems (Valley, 1986): δf = 1,000 × (F^α-1 – 1) + δi, where F is the C or O fraction remaining in the rock after the above reaction, α is the CO₂-dolomite fractionation factor, and δi and δf are the initial and final isotopic values of the dolomite, respectively. We used fractionation factors for C and O determined experimentally (Chacko et al., 1991), and assumed that the H8 dolomite reequilibrated at T = 400° to 500°C because of the homogenization temperature of fluid inclusions trapped during the main REE depositional stage at Bayan Obo >400°C (Smith and Henderson, 2000; Weng et al., 2015). The initial δ¹³C_V-PDB (0.72‰) and δ¹⁸O_V-SMOW (13.51‰) values were used as the average C-O isotope values of the precursor dolomite porphyroclasts. Using the average δ¹³C_V-PDB ratio of recrystallized dolomite (–1.33‰) as the final carbon isotope composition, the F(carbon) value was calculated to be ~0.58 (Fig. 7A); i.e., approximately 42% of C was released as CO₂ from the H8 unit during its metasomatic reworking. If we consider that the contribution of organic C (δ¹³C_V-PDB ≈ –25‰) to the subducted fluid was 0.24% (Plank and Manning, 2019), ~40% of the Bayan Obo dolomite was degassed (App. Fig. A4). Typical subduction zone fluids contain low CO₂ (Manning, 2004), and organic carbon (CH₄) was found only in trace amounts in fluid inclusions from mineralized parageneses (Smith and Henderson, 2000). Therefore, the negative shift of carbon in Bayan Obo dolomite is not likely to have been caused by organic carbon in subduction-derived fluids. Degassing reactions normally lower the δ¹⁸O_V-SMOW value of reequilibrated carbonate minerals by no more than 3‰ because of the calc-silicate limit (F> 0.6), even at high levels of decarbonation (Valley, 1986). However, ¹⁸O is partitioned preferentially into carbonates relative to their associated silicate phases (Chacko et al., 2001), and fluids derived from subducted metasediment-dominated rocks show enrichment in ¹⁸O (δ¹⁸O_V-SMOW ≥9.6‰; Scott et al., 2019). Thus, dolomite recrystallization involving such fluids is not expected to have a significant effect on the primary O isotope signature. A similar example of the degassing of carbonate minerals with little O isotope depletion has been reported by Wei et al. (2020).

Decarbonation during metasomatic reworking will result in volume loss. Volume loss occurred in response to negative specific volume changes during decarbonation reactions (1, 2, and 3) and was probably accompanied by the closure of reaction-induced porosity by creep, grain-boundary sliding, and pressure solution as fluid overpressures dissipated (Balashov and Yardley, 1998). These processes are manifested by the zones of recrystallized dolomite and the formation of fluorite and monazite-apatite segregations, stringers, and schlieren in the H8 unit. The volume change due to decarbonation can be semiquantitatively estimated given that the precursor rock was effectively composed of monomineralic dolomite. If Rayleigh volatilization follows a normal calc-silicate decarbonation trend via the stoichiometry of reactions (1) and (2), the approximate volume change in the precursor rock would be around –9 and around –6% for phlogopite and riebeckite formation, respectively, when F(carbon) ~0.6. However, reactions (1) and (2) are constructed on the basis of conserved Fe and Mg, which is unlikely in an aqueous-chloride solution, so this volume reduction may not be realistic (Yardley, 2005). Using SUPRCRT92 to calculate solid phase molar volume (V_m) changes on reaction, and assuming a Rayleigh devolatilization parameter of ~0.6 (i.e., 40 mol % Ca(Mg,Fe)(CO₃)₂ loss), fluorite formation via reaction (3) would result in a volume change of about –25% at P = 1 kbar and T = 25° to 450°C (using V_{m dolomite} = 64.35 cm³/mol, V_{m phlogopite} = 149.66 cm³/mol, V_{m riebeckite} = 273.6 cm³/mol, and V_mfluorite = 25.54 cm³/mol; Robie and Bethke, 1962; Johnson et al., 1992). Calcium conservation on fluorite formation is more reasonable given the low solubility of fluorite. It is clear that reactions (1), (2), and (3) operated in combination, and the Mg and Fe released during dolomite dissociation were incorporated in ferromagnesian silicates, both in the H8 unit and in the adjacent fenites (Smith, 2007), resulting in significant localized volume loss. The corresponding release of CO₂ as a fluid was recorded in the chemistry of the fluid inclusions (Smith and Henderson, 2000).

The Nd isotope data suggest a two-stage evolutionary history involving the separation of an LREE-bearing carbonatite magma from the mantle at ~1.3 Ga and subsequent periodic remobilization of the REEs until ~400 Ma (Fig. 8). There is no isotopic evidence to suggest a significant influx of LREEs from external sources, such as the Paleozoic granitoids. The late-stage metasomatic hydrothermal reworking, however, played a significant role in the REE re-enrichment of the Bayan Obo deposits.

The decarbonation process mentioned above would have led to the decomposition of the precursor dolomite and precipitation of the recrystallized dolomite, fluorite, and silicate minerals. The precursor and recrystallized dolomite display similar chondrite normalized REE profiles and relatively consistent near-chondritic Y/Ho ratios (Figs. 5, 6), suggesting that the decomposition of dolomite resulted in the enrichment of REEs that are readily immobilized in low-solubility phases such as monazite (Van Hoozen et al., 2020). Therefore, the decarbonation process caused REE enrichment of the residual dolomite body with these elements.

Two questions pivotal to the genesis of the Bayan Obo deposit are when and why such extensive recrystallization of the H8 dolomite took place. The published Sm-Nd isochron ages of whole-rock and mineral samples and Th-Pb ages of monazite and bastnäsite cover a wide range from 1400 to 300 Ma (Fig. 4). It is now widely accepted that REE minerals were deposited at Bayan Obo over a period of at least 1 billion years (Song et al., 2018). In this study, the in situ dating of monazite associated with apatite in paragenetically similar samples gave a wide range of dates, from ~340 to 980 Ma, whereas monomineralic monazite veinlets are characterized by relatively consistent isotopic characteristics and an age of ~400 Ma. This discrepancy indicates that the 1300 to 400 Ma date range may not represent the REE depositional ages. The older dates could arise from variable degrees of Pb loss from the Mesoproterozoic isotopic system during dolomite reaction with a fluid and recrystallization (Chen et al., 2020). The monazite was affected by a thermal event at ~400 Ma, which resulted in isotopic reequilibration and resetting of the Th-Pb system. The tectonic evolution of the Bayan Obo area is inferred to have involved regional metamorphism during two major lithospheric events: crustal extension and rifting in the Mesoproterozoic and arc-continent collision in the Ordovician-Silurian (Fig. 4; Tang and Yan, 1993; Xiao et al., 2003; Eizenhöfer and Zhao, 2018). Both events are well documented geochronologically, which enabled us to place the development of the REE mineralization in a temporal context. The emplacement of REE-rich carbonatites occurred in the Mesoproterozoic in response to rifting, which plays an essential role in global carbon cycling (Foley and Fischer, 2017). Following its separation from the Columbia supercontinent, the northern edge of the North China craton became a passive continental margin (Zhang and Zhao, 2016). The onset of southward (in present-day coordinates) subduction in the Ordovician was accompanied by the release of fluids carrying a metasedimentary isotopic signature, which triggered hydrothermal reworking and decarbonation of the H8 unit and remobilization and concentration of REEs to minable levels. The plate convergence lasted until ~410 Ma, but postcollisional magmatism was recorded for another 50 Ma (Ma et al., 2019), which explains the commonly reported Devonian ages of the Bayan Obo monazite (Song et al., 2018).

The C-Sr isotope evolution of the mineralized H8 unit at Bayan Obo indicates that the precursor rock underwent dynamic recrystallization, metasomatic reworking, and decarbonation, which resulted in volume loss and REE remobilization by fluids. The driving forces for these secondary processes were plate convergence, subduction, and the release of slab-derived fluids in the Silurian. The silica- and halogen-bearing, ⁸⁷Sr-rich fluids were responsible for the isotopic and trace-element reequilibration of rock-forming dolomite, whereas the convergent tectonics determined the synclinal shape of the REE-rich H8 carbonatite and probably facilitated fluid ascent and circulation within it. These processes resulted in a wide range of complex textures and ore types. Although the size of the orebodies implies voluminous carbonatitic magmatism, our data support syndeformational REE enrichment owing to decarbonation and volume loss without any need for external REE contributions.

We thank A.E. Williams-Jones, Charles Beard, Zengqian Hou, and an anonymous referee for their constructive comments and suggestions to improve the manuscript. This research was financially supported by the National Natural Science Foundation of China (41825008, 92162219) and the Guangxi Natural Science Foundation (2020GXNSFGA297003). ARC acknowledges support from the Natural Sciences and Engineering Research Council (Canada). MS acknowledges support from the UK-RI Natural Environment Research Council grant NE/V008935/1. JK was supported by the Czech Science Foundation GACR EXPRO (grant number 19-29124X).

Cheng Xu is a professor of geochemistry at Peking University (since 2010) and Dean of the College of Earth Sciences at Guilin University of Technology (since 2020). Xu received a B.Sc. degree at the China University of Geosciences (Wuhan) in 1999 and a Ph.D. degree at the Institute of Geochemistry, Chinese Academy of Sciences, in 2004. He worked at the Institute of Geochemistry from 2003 to 2010. His research focuses on carbonatite genesis and REE mineralization. Xu also studies HREE mineral origin in parental granites, whose weathering crusts form HREE deposits in South China.