Rare earth elements (REEs) are essential metals for modern technologies. Recent studies suggest that subcontinental lithospheric mantle (SCLM) remelting, previously fertilized by subducted marine sediments, leads to formation of REE-bearing rocks. However, the transfer mechanism of REE-rich sediments from the subducted slab to the overlying mantle wedge is unclear. We present high-pressure experiments on natural REE-rich marine sediments at 3–4 GPa and 800–1000 °C to constrain the phase relations, sediment melting behavior, and REE migration during subduction. Our results show recrystallization into an eclogite-like assemblage, with melting only occurring at 4 GPa, 1000 °C, experiments. Regardless of melting behavior, REE are refractory and mostly hosted by apatite. Buoyancy calculations suggest that most of the eclogite-like residues would form solid-state diapirs, ascending to the SCLM, resulting in the REE-fertilized source. Such flux may be required for substantial REE transport during subduction, as a foundation for economic-grade mineralization.

Rare earth elements (REEs) are indispensable to zero-emission technologies and other technological applications. Nearly all REE resources are hosted by alkaline-silicate rocks and carbonatites derived from mantle sources fertilized by subduction while formed in intracontinental, rift-related, or syn- to post-collisional settings (Goodenough et al., 2021; Beard et al., 2023). A model by Hou et al. (2015a) explains these REE-bearing rocks by metasomatism of the subcontinental lithospheric mantle (SCLM) by fluids derived from subducted REE-rich marine sediments, while their melting is hindered by the thick lithosphere at >2.5 GPa. This newly REE-enriched metasomatized mantle is then available for subsequent melting events, which lead to crustal formation of REE deposits (Hou et al., 2015a). This model well demonstrates the two-stage REE migration during slab-mantle interaction, but the process of SCLM REE metasomatism, and why its melting is delayed, are poorly constrained.

Experimental petrology methods are highly informative for studying Earth’s deep geochemical processes. Previous experimental studies have examined melting behavior and trace element transfer in subducted sediments by using either synthetic or natural materials, including pelitic sediments with varying CO2 and H2O contents (e.g., Tsuno and Dasgupta, 2012); global subducting sediments (Hermann and Spandler, 2008); mixtures of marine clays and terrigenous sediments (Mann and Schmidt, 2015); volatile-free sediments (Spandler et al., 2010); iron-calcareous clays (Thomsen and Schmidt, 2008); H2O-saturated clay (Nichols et al., 1994); greywacke and pelites (Schmidt et al., 2004); metagreywacke (Auzanneau et al., 2006); and marine limestone (Chen et al., 2021). These experiments exhibited either no sediment melting behavior or low REE contents in the produced fluids or melts. Alternatively, Behn et al. (2011) suggested that density and viscosity contrasts between subducted sediments and their overlying mantle would lead to formation of buoyant solid-state flux diapirs. Based on sedimentary layer thickness and the geotherm and dynamic parameters of subducted sediments, diapirs would form at ±40 km, at the slab depth below the arc (at ~2–4 GPa; Behn et al., 2011). Thus, diapirism provides a new possibility for REE migration during subduction in lieu of mobility in fluids or melts.

REE-rich marine sediments are diverse and contain a variety of mineral assemblages and REE hosts. Their subduction has been considered a major contributor to the mantle underneath REE deposits (Hou et al., 2015a; Ren et al., 2021). We conducted experiments on three representative natural REE-rich marine sediments with ~1000–2000 ppm REE + Y (ΣREY) contents (Tables S1 and S2 in the Supplemental Material1) from the Pigafetta Basin, western Pacific Ocean. An objective of this study is to remove the uncertainties inherent in the use of synthetic and REE-depleted starting materials. Additionally, our reaction experiments can be used to constrain the changes in phase relations of these sediments at subduction conditions. Combined with buoyancy calculations, we provide new experimental constraints on REE migration during slab-mantle interaction.

We carried out 12 experiments on REE-rich marine sediments at 3–4 GPa and 800–1000 °C. Silicate melts occurred only at the 4 GPa, 1000 °C, experiments, which constrains the solidus temperature of the selected sediments to between 900 and 1000 °C. Phase modes were estimated by mass balance calculation (Fig. 1; see supplemental text and Table S3) and selected scanning electron microscopy (SEM) images shown in Figure S2.

All experiments but one produced an eclogite-like assemblage consisting of garnet and clinopyroxene along with minor and accessory phases (apatite, mica, K-feldspar, and quartz/coesite; Fig. 1; Fig. S1). Such results agree with most experimental studies using sedimentary materials (e.g., those of Schmidt [2015]) and observations of eclogitic xenoliths in natural REE-bearing rocks (e.g., Xu et al., 2018). Sample number 64-30 at 1000 °C did not produce garnet. Apatite occurs in all experiments, with textures varying from euhedral to disseminated (Fig. S1). Mica is the only observed hydrous mineral and formed in all melt-absent experiments. Mica rims were dissolved in our 3 GPa, 900 °C, runs, potentially indicating incipient melting (Fig. S1). K-feldspar only exists in the 3 GPa experiments.

Major and trace element compositions of all run products are given in Tables S4 and S5. Almost all garnet, clinopyroxene, mica, and apatite in 3 GPa experiments exhibit higher Mg + Fe values than in 4 GPa experiments, and the halogens are mainly hosted by apatite, mica, and melts (Table S4). Melts are alkali silicate with high concentrations of Si and Na + K (Table S4), consistent with trachyte-dacitic, similar to most previously experimentally produced liquids (e.g., those listed by Schmidt [2015]; Fig. 2). It is worth emphasizing that melts contain only ~600 ppm ΣREY (Table S5; Fig. S5). In contrast, apatite and garnet are the principal REE hosts, with ΣREY as high as 30,000 ppm and 2400 ppm, respectively (Table S5). Other elements of all the phases show consistent variation trends; e.g., negative anomalies for Nb, Ta, Sr, Ti, and Ce, and positive anomalies for Th and U (Table S5). The Ce anomalies are inherited from the starting materials and do not indicate exceptional oxidizing conditions during the experimental runs.

Melting Behavior

Previous experiments on sediments resulted in a wide range of melting temperatures (670–1250 °C) at 2.5–5 GPa (Fig. 2). Our experiments estimate the solidus of natural REE-rich marine sediments with ~10 wt% H2O (Table S1) between 900 °C and 1000 °C at 3–4 GPa, which is comparable to that reported by Johnson and Plank (1999). Such temperatures are ~300 °C higher than the wet solidus of sediments (Fig. 2). As is widely proposed, the melting reaction for sediments is quartz/coesite + mica + clinopyroxene ± fluid = melt + garnet (Carter et al., 2015). The lack of garnet in our run LMD752-30 indicates that melting occurred via the reaction clinopyroxene + garnet ± quartz/coesite = melt, known to take place at similar subduction conditions (Kessel et al., 2005). The different reactions likely result from differences in bulk starting composition, as the starting material for this run (sample 64-30) contained the lowest Mg, Mn, and Fe contents, such that garnet is easily destabilized.

The question persists as to whether sediments can melt at the slab-top condition of the subducting zone. Recent thermal modeling (Syracuse et al., 2010) predicts slab-top temperatures of ~750–850 °C in most subduction zones at 2.5–4.5 GPa, with some offset toward higher temperatures of up to 950 °C in some subduction zones with relatively young, hot oceanic crust (e.g., Cascadia). Generally, most previously determined solidi of sediments are well above the temperatures that are present within the subduction zones, except for the H2O-saturated solidus (Fig. 3). Most H2O-saturated experiments are designed to recreate underlying dehydrating lithologies. However, such effects would be eliminated at <2 GPa because roughly two-thirds of the water in fully hydrated oceanic crusts would be lost at forearc depths (Ague and Nicolescu, 2014). Thus, we experimentally suggest that most sediments are unlikely to substantially melt except in the hottest subduction zones.

Refractory Rare Earth Elements

Even if partial melting of REE-rich sediments would occur in hot subduction zones, it would have minor impact on REE migration. Residual mineralogy directly impacts the behavior of trace and minor elements in partial melts. Some examples include monazite, apatite, and epidote-group minerals for light REEs (LREEs) (Hermann and Spandler, 2008; Skora and Blundy, 2010; Carter et al., 2015) and garnet for heavy REE + Y (HREE + Y) (Hermann and Spandler, 2008). These residual minerals, which carry substantial ΣREY, could remain stable at temperatures higher than the solidi of sediments, such as garnet (up to 1400 °C; Klemme et al., 2002) and apatite (up to 1050 °C; Johnson and Plank, 1999).

In our study, ΣREY abundances of melts are consistently low, with garnet and apatite coexisting with the melts. Noticeably, the apatite role in our experiments is not limited to the LREEs but also HREEs and Y. For example, apatite in run LMD752-46 contains ΣREY of ~24,000 ppm (~17,000 ppm LREEs, ~2600 ppm HREEs, and ~3900 ppm Y). In contrast, garnet contains much less, at ~18 ppm LREEs, ~260 ppm HREEs, and 630 ppm Y. In the garnet-absent experiment LMD752-30, apatite exhibits the highest ΣREY (~31,000 ppm), whereas the melt contains merely 291 ppm. Although fluids were assumed to be the primary fluxes transferring REEs in subduction zones (Hou et al., 2015a), they have been experimentally proved weak REE mobility via comparable experiments (e.g., Johnson and Plank, 1999; Carter et al., 2015; Skora et al., 2015). Hence, we conclude that REEs within marine sediments would be hosted by the residual minerals (in this study, mostly apatite and garnet) rather than released by melts or fluids under subduction zones.

In melting experiments of other slab components such as mid-ocean ridge basalt and altered oceanic crust (e.g., Carter et al., 2015), REEs exhibit similar immobility behavior and tend to be controlled by residual minerals. In this case, traditional fluid/melt-migration models seem insufficient to drive the slab-mantle REE recycling. A more efficient flux transporting REEs from subducted slabs to the mantle wedge should therefore be proposed.

Diapirism of REE-Rich Sediment

Several studies have speculated that diapirism could provide an additional recycling pathway of buoyant subducted materials between slabs and overlying mantle wedges (e.g., Behn et al., 2011; Marschall and Schumacher, 2012). The dynamic processes of this solid-state flux are controlled mostly by the relative density and viscosity contrasts between the sediments and overlying mantle, the thickness of the sedimentary layer, and the geotherm and dynamic parameters and processes of the subducting slab (Gerya and Meilick, 2011). We calculated the density of the average compositions of our sediment samples and compared them to those of the overlying mantle (Fig. 3). These sediments are characterized by low densities (2900–3362 kg/m3 at <6.0 GPa; Fig. 3) compared to the mantle peridotite (3200–3400 kg/m3; Jull and Kelemen, 2001) and are generally higher than those of marble (2091–3028 kg/m3; Chen et al., 2021) and average ultrahigh-pressure metasediment (2800–3300 kg/m3; Behn et al., 2011). In terms of the viscosity, previous numerical calculations of wet quartz (which is ~100 times less viscous than wet olivine; Behn et al., 2011) are applicable for our sediment samples.

For hot (Colombia-Ecuador) and cold (Kermadec) slab-top geotherms (Syracuse et al., 2010), the density contrast at <6 GPa is shown in Figure 3. Considering the invariably lower density and viscosity of our sediments (although the magnitude of the density contrast significantly decreased at ~2.7 GPa due to significantly increasing temperature along the subduction geotherm), we propose that these REE-rich marine sediments could also detach from the subducting slabs, forming buoyant diapirs (Fig. 3). This indicates that diapirs derived from marine sediments could result in massive REE storage within the SCLM, a flux overlooked in previous studies.

Tectonic Implications

Our model explains why REE-bearing rocks are closely associated with subduction zone belts but only form after protracted subduction has ceased (Beard et al., 2023). It is plausible that subduction has enriched the mantle wedge with REEs via refractory diapirs. Once the slab is delaminated, such REE-enriched zones melt due to decompression, heating of the mantle wedge, or a hot influx (likely CO2-rich) from the mantle. Although both the alkaline silicate rock and carbonatite typically host REE mineralization, ~50% of the global rare earth oxides are hosted by carbonatite-related deposits (Weng et al., 2015). We propose that such a difference is largely attributed to their various CO2 contents. In our model, CO2 has no direct effect on REE mobility from slab to wedge. Instead, CO2 could promote REE migration and mineralization in the following processes:

  1. Carbonate within sedimentary layers could assist REE-rich diapir formation by enhancing the density contrast (Fig. 3) and thickening the sediments (Behn et al., 2011).

  2. CO2 greatly elevates the solidus (Fig. 2; Schmidt, 2015), which would ensure a concentrated formation of carbonatite, rather than multiple melting as in alkaline silicate rocks (Beard et al., 2023), thereby diluting REEs.

  3. Carbonatitic melts derived from the CO2-fertilized SCLM may play a key role in secondary REE enrichment, as it extracts REEs from the silicate-carbonatite complex melts (Martin et al., 2013; Anenburg et al., 2021).

Importantly, this model also implicates the formation of zonal mineralization in continental collision zones. The widely distributed continental collision–related porphyry Cu deposits (PCDs) are also considered to form by SCLM remelting, previously modified by fluids derived from subducting slabs (Hou et al., 2015b). We also suggest that different series of element fluxes derived from subducted slabs may result in differential migration of ore elements in subduction zones and furthermore form variable fertilized magma sources (Fig. 4). These sources, after being remobilized by later geological processes, would form different ore belts according to the formation depth along the previous subduction suture in continental collision zones. Excellent examples in previous studies show that this is indeed widespread (Hou et al., 2015a, 2015b). Spatially, the PCDs from southeast Tibet are parallelly distributed with the Mianning-Dechang REE belt (in Sichuan, China), indicating the geographical fractionation and transfer of ore elements from subducted slabs to the mantle wedge. Cu-rich fluids are released first from the slab and infiltrate the mantle wedge, and then REE-rich residues form diapirs detaching from the slabs and mechanically ascending to the mantle wedge (Fig. 4). Accordingly, this model provides direction and reference for further study of REE migration during subduction and future ore prospecting of subduction- and collision-related zones.

1Supplemental Material. Experiment design, analytical and calculating methods, and data sources for this study. Please visit https://doi.org/10.1130/GEOL.S.24578953 to access the supplemental material; contact editing@geosociety.org with any questions.

This research was funded by the National Natural Science Foundation of China (grant 92162216) and the National Key R&D Program of China (grant 2022YFF0800902). We thank all editors and reviewers for the help and insightful comments.

Gold Open Access: This paper is published under the terms of the CC-BY license.