Introduction

Since the Moon was found not to be completely devoid of volatile elements1,2, numerous studies over the past decade have revealed that the volatile (e.g., water and halogen) distribution on the lunar crust and in the lunar interior is heterogeneous. Most of our knowledge on the halogen distribution in the Moon is based on studies of lunar apatite from different crustal lithologies, including mare basalts, magnesian-suite (Mg-suite), alkali-suite rocks, and ferroan anorthosite (FAN)3,4,5,6,7,8,9. These apatite studies identified two volatile reservoirs with distinct F/Cl ratios in the Moon. The parental magmas of primary crustal lithologies such as the Mg-suite rocks generally have much lower calculated F/Cl ratios (~1 to 5) compared to those of the secondary mare basalts (~20–50)6. This observation has been further verified by halogen abundance measurements of lunar pyroclastic glass beads and olivine-hosted melt inclusions (see a detailed review by McCubbin et al.5).

The origins of these two distinct source regions have been explained within the context of the lunar magma ocean paradigm. McCubbin et al.10 and Potts et al.8 proposed that the source region of mare basalts consists of cumulates from the lunar magma ocean with F » Cl, largely influenced by the much lower mineral-melt partition coefficients of cumulate minerals for Cl compared to F8. In contrast, parts of the primary crust, represented by various Apollo highland lithologies, are generally KREEP-rich (i.e., enriched in potassium, rare earth elements, and phosphorus). Parental magma of most Mg-suite rocks are thought to be associated with the urKREEP reservoir referred as the last dreg of the lunar magma ocean (LMO)11,12,13,14. The F/Cl ratios of urKREEP are estimated to be low, between ~0.5 and 4.4, based on halogen abundances in lunar soil10 and apatite15, respectively. This range is close to the inferred range of F/Cl in the source of Mg-suite rocks, and the involvement of urKREEP is also consistent with the KREEP signature in many of these samples.

However, the identification of KREEP-poor Mg-suite rocks suggest that the origin of some primary crustal rocks may not involve the KREEP component16,17,18, implying that the KREEP component may not be the “petrologic driver” of the formation of Mg-suite magmas19. This agrees well with the observation by remote sensing of the widespread distribution of Mg-suite lithologies on the lunar farside, which is thought to be poor in KREEP20,21,22. As the recent Chang’e-6 sample-returned mission to the lunar farside likely returned new Mg-suite samples in an area not affected by KREEP23,24, it is important to reassess the role of KREEP in Mg-suite formation from a halogen perspective.

Here, we aim to reassess the formation and evolution of the halogen reservoirs in the Moon by studying the partitioning of Cl between nominally anhydrous minerals (NAMs) that are key to the formation of the lunar crust and co-existing lunar melt. Unlike apatite, which forms only during the final stages of magma crystallization, fractionating F/Cl ratios of final dregs of melt6, continuously crystallizing NAMs can faithfully record nearly constant F/Cl ratios without modifying F/Cl ratios in melt because F and Cl are generally highly incompatible. Several recent studies have reported the F and Cl contents in lunar NAMs, including alkali feldspar, plagioclase and pyroxenes from highland rocks25,26,27,28,29. These results show large variations in F/Cl ratios ranging from less than one to hundreds, suggesting the existence of heterogeneous halogen reservoirs or lunar crust alteration by late processes. The F/Cl ratios of parental magma of non-KREEP Mg-suite rocks also remain enigmatic, given the paucity of halogen partition coefficients between NAMs and silicate melt under lunar-relevant conditions8,30,31.

In this study, we report new Cl partitioning data between plagioclase, orthopyroxene and melt at lunar conditions derived from high-temperature, high-pressure experiments, and derive a set of partition coefficients that can be used at pressures, temperatures, and water contents that are relevant to the Moon. Combined with our previously reported F partition coefficients31, we are able to reconstruct the F/Cl ratios of melts in equilibrium with lunar crustal minerals, providing new insight into the evolution of the halogen budgets of the lunar crust and the origin of the Mg-suite.

Results

The composition of the starting material for the experiments was based on a residual melt composition formed at around 80% solidification of a nominally dry lunar magma ocean (experiment LBS6 from ref. 32, Supplementary Data 1). High-pressure and high-temperature experiments were conducted at pressures of 0.5-1 GPa and temperatures of 1120–1240 °C. Full details of the starting material synthesis, high-pressure, high-temperature experimental techniques, and analytical methods are provided in the Methods. Representative back-scattered electron microscope images of run products are shown in Fig. 1. Experiments run at 0.5-1 GPa and 1120–1240 °C all produce orthopyroxene and plagioclase in equilibrium with melt. The experiments with low melt fractions (e.g., at 1120 oC), in addition, contain quartz and rutile. Most experiments show sharp contacts between liquid and crystals (Fig. 1) consistent with mineral-melt equilibrium. In case of suspected bulk disequilibrium in terms of phase relations or core-to-rim orthopyroxene zonation in experiments at low temperatures (e.g., Fig. 1), the major and chlorine element abundances of mineral grains were obtained from crystal rims at a distance of ~10 μm from the crystal-melt interface. Such locations are closest in chemical equilibrium with the surrounding melt pool, while avoiding the analytical problem of crystal analyses being contaminated by secondary fluorescence from adjacent melt. For orthopyroxene grains with rims thinner than 10 μm, some analysis points fall within 5 μm of the crystal-melt interface. Quantitative secondary fluorescence (SF) modeling on the extent of the Cl contribution from Cl-rich melt shows that SF has no significant effect on the reported Cl abundances in mineral phases at either 10 or 5 μm distance from the crystal-melt interface (Supplementary Fig. S1, Supplementary Information). The analysis positions in homogeneous orthopyroxene grains (e.g., in JQCl9 and JQCl10) and all plagioclase grains were obtained from positions in the cores of several randomly selected crystals.

Fig. 1: Representative back-scattered electron images of experimental run products.
figure 1

a Experiment JQCl6 (48 h) run at 0.5 GPa and 1120 oC. b Experiment JQCl2 (24 h) run at 0.5 GPa and 1120 oC. c Experiment JQCl8 (22 h) run at 0.5 GPa and 1160 oC. d Experiment JQCl10 (24 h) run at 1 GPa and 1240 oC. Ol, olivine; Opx, orthopyroxene; Qtz, quartz; Rut, rutile; Liq, liquid.

Full size image

Chlorine is detected in all orthopyroxene (37–210 μg/g) and plagioclase (34–824 μg/g), well above the electron microprobe detection limit of about 28 μg/g, while the silicate melt contains 3.06–8.90 wt.% Cl contents. As the orthopyroxene grains with zonation show both Mg-rich rims and Fe-rich rims, orthopyroxene-liquid Fe-Mg exchange coefficients, KD have been calculated for individual analyses for further equilibrium state evaluation. Only individual orthopyroxene rim compositions with KD ranging from ~0.2–0.433,34,35 were used for each experiment and for partition coefficient calculation (Supplementary Data 2). Although the orthopyroxene Mg-rich rims generally have disequilibrium KD values (e.g., 0.16 in JQCl1), their Cl abundances show no systematic difference with Fe-rich rims. The calculated Cl mineral/melt partition coefficients (\({D}_{{{{\rm{Cl}}}}}^{\min /{{{\rm{melt}}}}}\)) are presented in Table 1. The Cl contents of individual phases and partition coefficients for quartz and rutile are not discussed in the main text. Full detailed descriptions on individual mineral phases, and Cl contents of all phases are provided in the Supplementary Information.

Table 1 Experimental conditions and mineral-melt Cl partition coefficients
Full size table

In order to evaluate the quality and reproducibility of the derived Cl partition coefficients, we have conducted duplicate experiments with varying experimental durations. The Cl partition coefficients show no correlation with Cl contents in melt (Supplementary Fig. S2), suggesting it obeys Henry’s law. At 1120 oC and 0.5 GPa, three experiments with durations of 24–48 h have different mineral phases composition between one 24 h experiment and two 48-h experiments (Supplementary Data 2). The 48-h experiments have Cl contents of 8.21 ± 0.19 wt% and 8.90 ± 0.57 wt% in the silicate melt, but the 24-h experiment has lower 5.94 ± 0.14 wt% Cl abundance. This indicates that bulk equilibrium was not reached in this 24-h experiment at a temperature of 1120 oC. However, all three experiments yield consistent \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\) and \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) (Table 1 and Fig. 2), indicating 24 h is sufficient to reach equilibrium Cl partitioning at temperatures of 1120 oC and above. This conclusion is further supported by the consistent \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\) and \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) values in duplicate experiments at 1160 oC and 0.5 GPa (Table 1 and Fig. 2).

Fig. 2: Absence of systematic effects of temperature, pressure and water content on Cl partitioning between silicate melt and minerals.
figure 2

a \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\) values from this study and literature as a function of temperature and pressure, in which part data of ref. 40 show outliers with anomalous high values for orthopyroxene; (b) \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) values from this study and literature as a function of temperature and pressure. Error bars are shown in 1σ when available. Two experimentally determined very high values for orthopyroxene (0.16 ± 0.09 and 0.015 ± 0.008)74 are not plotted, due to their large uncertainties and use of experimental conditions relevant for the Earth’s mantle, not the Moon. c \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\) values as a function of water content in glass; (d) as a function of water content in glass. The blue shaded area represents the mean value with its standard deviation of our experimental data (excluding the data with high uncertainties, dotted points). Error bars represent 1σ standard deviations. Duplicate experiments are circled in the figure. Literature data are from ref. 40 and ref. 8.

Full size image

Although we attempted to do experiments that were nominally anhydrous, in practice, this is impossible to achieve in terrestrial laboratories. Nominally anhydrous experiments of the type performed in this study contain on the order of 500 μg/g water36,37. In the absence of hydrous minerals, this water would all concentrate in the melt phase. To assess if water concentrations affect the behavior of Cl, we measured the water contents in the silicate glasses from our experiments. These range from 0.20–1.00 wt.% H2O (Supplementary Data 3). \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\) and \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) values show no dependence on pressure, temperature, or water content within the experimental temperature and pressure range and in the measured range of water contents. Recent studies have shown a significant increase in \({D}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) for water itself as the amount of water in silicate melt is decreased below 1 wt.%38,39. This is not observed in the case of Cl partitioning. The mean \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\) and \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) values are calculated as 0.0019 ± 0.0006 and 0.0024 ± 0.0006 (Supplementary Data 4), respectively, excluding some high values that are accompanied by much larger errors (e.g., \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Plag}}}}/{{{\rm{melt}}}}}\) = 0.0077 ± 0.0051, in experiment JQCl4). Our data are in good agreement with existing data (where available) obtained at lunar conditions8 and most of the data focused on terrestrial conditions40 (see detailed discussion in the Supplementary Information about the exception related to their \({D}_{{{{\rm{Cl}}}}}^{{{{\rm{Opx}}}}/{{{\rm{melt}}}}}\)).

Discussion

Widespread Cl enrichment on the lunar nearside

A growing number of studies have reported halogen concentrations in lunar crust silicate materials25,26,27,28,29. Mosenfelder et al.25,26,27 reported the F concentrations in plagioclases from FAN and troctolite, whereas Cl concentrations generally fell below detection limits. Two data points from FAN 60015 show detectable high Cl concentrations (0.2 and 14.4 μg/g, respectively) but without coupled enrichments in F and water. Barrett et al.29 analyzed low halogen contents in plagioclases from two paired lunar meteorites, troctolitic breccia NWA 5744 and troctolitic anorthosite NWA 10140. The F and Cl concentrations from these two samples are consistently low at 0.6–1.0 μg/g F, and ~0.03–0.06 μg/g Cl.

Simon et al.28 reported a comprehensive dataset of volatiles (including halogen and water) contents in lunar silicate minerals, including plagioclase, alkali feldspar, pyroxenes and quartz from Apollo breccia samples. They observed a correlation between hydrogen enrichment and increasing microstructural evidence of shock, suggesting that at least some of the hydrogen was added to the lunar crust by impact-induced exchange and/or partial assimilation. Their samples also show a Cl enrichment which is decoupled from hydrogen and F enrichment (Fig. 3), which is consistent with the observations of ref. 27. For example, the alkali feldspar in felsite 14321,1062 has low volatile abundances (generally <2 μg/g H), but variable Cl concentrations ranging from 0.02 to 4.67 μg/g, a variation by over two orders of magnitude. This felsite 14321,1062 is a silicic clast and exhibits no impact-induced hydrogen addition28. Thus, these Cl variations represent either F/Cl ratio variations in their equilibrium magma or a secondary alteration process.

Fig. 3: Variations of hydrogen and halogen abundances in lunar silicate minerals.
figure 3

a Variation of Cl and H concentrations in lunar silicate minerals. b Variation of F and Cl concentrations in lunar silicate minerals. F/Cl ratio fields of minerals (plagioclase and orthopyroxene) crystallized from magma with low F/Cl ratios (0.5-4.4, KREEP-like) and high F/Cl ratios (lunar mantle-like) are plotted. The hydrogen addition trend by asteroid or comet impacts comes from ref. 28. Literature data include Apollo materials28 and meteoritic plagioclases (NWA 5744 and 10140)29.

Full size image

Chlorine enrichment also occurs in different minerals within the same sample. In troctolite 14303,363 (Fig. 3b), the orthopyroxene contains 3.88 μg/g F and 0.15 μg/g Cl, whereas the plagioclases have 1.39 μg/g F but high Cl contents of 1.3 μg/g, an order of magnitude higher than in orthopyroxene. Using the halogen partition coefficients (Table 1 and Supplementary Data 3), the magma in equilibrium with these orthopyroxenes would have a F/Cl ratio of 1.8, which is 23 times higher than the F/Cl ratio (0.08) calculated on the basis of plagioclase data. This discrepancy in F/Cl ratios of their parental magma cannot be explained by fractional crystallization.

Variations in halogen contents in silicate minerals can be influenced by magma crystallization, but it is unlikely to generate the observed variations in F/Cl ratios by over an order of magnitude unless the magma is highly evolved and apatite crystallization occurs. We modeled the F/Cl ratios in minerals crystallized from different magmas with different F/Cl ratios (see modeling details in Methods). Nearly all samples plot within the field of low F/Cl ratios, similar to or lower than those of urKREEP (0.5 to 4.4) (Fig. 3b).

It is important to note that volatile degassing may contribute to an increase in the F/Cl ratio, as Cl is more prone to degassing than F41,42,43. This is evidenced by measurement in orthopyroxene in 14303,363, which shows low Cl content and a high F/Cl ratio. However, neither degassing processes nor KREEP contamination can explain data with F/Cl ratios lower than 0.5. In addition, the extremely low F/Cl ratios observed are unlikely to be caused by fluorapatite crystallization for two reasons. Fluorapatite crystallization could potentially decrease F/Cl ratios in residual melts, as the partition coefficients of F between apatite and melt (4.4–19) are generally higher than those of Cl (1.1–5)44. However, OH, which has a much lower partition coefficient (0.07–0.24) than both F and Cl, would be expected to become enriched in the residual melt - yet such enrichment is not observed. Second, although on the scale of an individual sample, alkali feldspar from felsite 14321,1062 shows a wide range of Cl concentrations, from 0.02 to 4.67 μg/g, this variation is coupled with nearly constant F concentrations. This is also inconsistent with the concomitant crystallization of fluorapatite.

Vapor phase crystallization of a variety of minerals, including oxides, halides, iron, and alkali-bearing phases, has been reported in vugs within Apollo 14 breccias45, and vapor-phase related metasomatism could likely explain the observed high Cl and heavy Cl isotopes (high δ37Cl value) in lunar samples in the past15,46. It has been suggested that lunar ejecta blankets, such as the Fra Mauro unit, could have initiated non-volcanic fumarole activity45. Several studies proposed that the metal chloride vapors released were enriched in 37Cl, which altered the primary composition of primitive lunar crust samples15,46,47 (e.g., ‘rusty-rock’ 66095, granulite 79215,15, and Apollo 14 apatite). Cl-bearing vapor condensation has also been proposed as a mechanism to explain volatile contents in lunar glass beads (74220, 15426) and 6609548, which could result in high δ37Cl values in lunar samples. For example, vapor condensation from LMO degassing has been suggested to account for the elevated Cl contents and high δ37Cl values observed in some FAN samples49.

A conclusion on which metasomatic process caused the Cl enrichment cannot be drawn without Cl isotopic data. Whether this Cl enrichment occurs only on the lunar nearside, or if it also affected the lunar farside, and which metasomatic process may be responsible, remain open questions. An analyses of halogen abundances in the recently returned Chang’e-6 samples23,24 could shed light on the geographical distribution of Cl enrichment, while Cl isotopic analyses could shed light on whether the metasomatic process(es) are comparable between nearside and farside.

Origin of non-KREEP Mg-suite rocks revealed by F/Cl ratio

After excluding the data whose equilibrium melts have low F/Cl ratios (<0.5) caused by Cl metasomatism, nearly all the silicate minerals from Apollo Mg-suite rocks still suggest that their parental magma had low F/Cl ratios, closely resembling those of KREEP (Fig. 3). This consistency is attributed to the KREEP affinity of the Apollo nearside Mg-suite19,50,51. While the non-KREEP Mg-suite rocks NWA 5744 and 10140 lack a KREEP component17,18, their plagioclases equilibrated with magma containing 19.3–32.2 μg/g F and 12.5–25 μg/g Cl. Using the same Monte Carlo simulation, the F/Cl ratio of the Mg-suite parental magma was calculated using the results of 1,000,000 simulations. The F and Cl contents in plagioclase from non-KREEP Mg-suite rock were considered within the specified ranges of 0.6–1 μg/g for F and 0.03–0.06 μg/g for Cl. The partition coefficients for F and Cl were assumed to follow normal distributions without negative values. The derived F/Cl ratio of the parental magma is 1.51 ± 0.72, similar to those in urKREEP (Fig. 3b). To account for the origin of the low F/Cl ratios in the parental magma of these non-KREEP Mg-suite rocks, a lunar mantle source with a similar low F/Cl ratio would be required.

Previous studies have thoroughly reviewed various petrogenetic models for the origin of Mg-suite rocks19,50,51, suggesting that partial melting of a hybrid source composed of mixed cumulates, whether from the deep or shallow mantle, is responsible12,13,14,50,51,52,53,54,55,56. Nearly all models require an early olivine cumulate as the Mg-rich source and a later addition or assimilation of plagioclase as a source for Al and Ca. Prissel and Gross55 proposed that lower mantle cumulates, composed of 93% olivine and 7% orthopyroxene, could generate plagioclase-undersaturated, Mg-rich primary magma to account for the observed Mg-suite lithologies. In contrast, Elardo et al.13 and Elardo and Manosalva14 suggested a hybrid source composed of olivine and plagioclase, emphasizing the critical role of KREEP components in the generation of Mg-suite primary magma. Recently, the Mg-suite clast found in the Chang’e-5 sample has been proposed to originate from an orthopyroxene-rich source region56. In this study, we selected three candidate mantle sources—(1) olivine and orthopyroxene; (2) olivine and plagioclase; (3) orthopyroxene and plagioclase—to test whether the F/Cl ratios of parental magma for non-KREEP Mg-suite rocks that do not appear to be affected by Cl metasomatism can be reproduced.

The modeling details and model parameters are shown in Methods and Supplementary Information, and results are depicted in Fig. 4. The \({D}_{F}/{D}_{{Cl}}\) ratios in orthopyroxene and plagioclase are similar (~13) but much higher than that olivine \({D}_{F}/{D}_{{Cl}}\) ratio of ~1 (Supplementary Data 4). Thus, the addition of orthopyroxene or plagioclase in olivine cumulates increases the F/Cl ratios in the mantle source and partial melts derived from the mantle source significantly. Starting from a LMO with F/Cl ratios from 1–5, we find that only the olivine-rich mantle source (>70 wt.% olivine) can produce the required low F/Cl ratio parental magma for the non-KREEP Mg-suite rocks, and only for models with an initial F/Cl ratio in the LMO of 1 (Fig. 4a). If the mantle source is orthopyroxene-dominated, no matter how much plagioclase is admixed in the source, the generated partial melts would have high F/Cl ratios ranging from ~10–60 (Fig. 4b).

Fig. 4: Monte Carlo simulation of F/Cl ratios and associated standard deviations (1σ) of partial melts from different cumulates.
figure 4

a Partial melts from olivine-dominated mantle cumulates mixed with varying amounts (wt.%) of orthopyroxene or plagioclase. b Partial melts from orthopyroxene-dominated mantle cumulates mixed with varying amounts (wt.%) of plagioclase. The F/Cl ratio of the non-KREEP Mg-suite parental magma derived from plagioclase in this study is shown for reference. Ol, olivine; Opx, orthopyroxene; Plag, plagioclase.

Full size image

We note that olivine-rich cumulates may possess insufficient halogen budgets due to the highly incompatible nature of Cl in olivine. If we assume a halogen-rich initial LMO (F = 5 and Cl = 5 μg/g, close to the maximum value estimates from McCubbin et al.5, a mantle cumulate composed of 90% olivine and 10% orthopyroxene (from a residual LMO after 50% solidification) would contain only low levels of F (0.053 μg/g) and Cl (0.029 μg/g). Even minimal 1% partial melting of this cumulates would only yield a magma with approximately 5 μg/g F and 3 μg/g Cl. This parental magma would need to undergo at least ~80% fractional crystallization to reach the calculated lowest halogen concentrations (~19 μg/g F and 12 μg/g Cl) in the magma that equilibrated with non-KREEP Mg-suite plagioclase. This is inconsistent with the high-Mg# olivine (~80) and high-An plagioclase (~98) observed in the paired NWA 5744 and 10140 meteorites17.

Instead, to reach the required F/Cl ratios of these non-KREEP Mg-suite rocks, the involvement of a trapped instantaneous residual liquid (TIRL) component in the source is required. With >2% TIRL (with a composition equivalent of that of the LMO after 50% solidification), even the orthopyroxene-rich mantle source is able to produce a parental magma with KREEP-like F/Cl (Fig. 5). This TIRL not only archives the necessary halogen budgets in the mantle source but also increases the source’s abundances of other incompatible trace elements. By considering varying weight percentages of TIRL in the mantle source, this approach also offers an alternative explanation for the high REE content in non-KREEP Mg-suite plagioclase14.

Fig. 5: Monte Carlo simulation of F/Cl ratios and associated standard deviations (1σ) in partial melts from orthopyroxene-dominated mantle cumulates.
figure 5

Partial melts are derived from 10% batch melting of an orthopyroxene-dominated mantle cumulates mixed with varying amounts (wt.%) of trapped instantaneous residual liquid (TIRL). The F/Cl ratio of the non-KREEP Mg-suite parental magma is shown for reference as the orange band. Opx, orthopyroxene; Plag, plagioclase.

Full size image

Olivine-hosted melt inclusions generally exhibit high F/Cl ratios in both low-Ti and high-Ti lunar basalts (~15–20)2,57,58,59, indicating their mantle source contains significant orthopyroxene and/or plagioclase. This observation agrees with McCubbin et al.10, who proposed that the TIRL content in these mantle sources cannot exceed 1 wt.%. Unlike these mare basalts, the addition of TIRL in the mantle source of non-KREEP Mg-suite rocks is crucial. Although the presence of TIRL in the mantle source (with a composition equivalent of that of the LMO after 50% solidification) increases incompatible element concentrations, its REE concentrations remain orders of magnitude lower than those of the urKREEP reservoir60. This composition does not meet the enrichment level required to generate abundances observed in KREEP-rich troctolites (e.g., >15% KREEP addition in their source)13. In addition, the low solidus of TIRL contributes to a reduction in the effective melting temperature of early Mg-rich cumulates, thereby promoting the onset of early Mg-rich magmatism on the Moon. In conclusion, incorporating TIRL into the mantle source, whether it be an olivine- or orthopyroxene-dominated hybrid mantle, appears to be required to explain the presence of non-KREEP Mg-suite rocks on the Moon.

Methods

Starting material synthesis and experimental techniques

Synthesis of starting materials followed the methods described in our previous work on fluorine mineral-melt partitioning31. A residual melt composition formed at around 80% solidification of a nominally dry lunar magma ocean (experiment LBS6)32 was chosen as the starting material. This starting material is in the simple system CaO-Al2O3-MgO-SiO2-FeO-TiO2. Three weight percent Cl was added to this composition in the form of CaCl2 without changing the major element ratios in LBS6 (Supplementary Data 1). Starting materials were prepared by mixing appropriate amounts of high-purity (99.5–100.05%, Alfa Aesar) oxide powders (MgO, Fe2O3, Al2O3, TiO2, SiO2) and carbonates (CaCO3). As described in ref. 31, the oxides MgO, Al2O3, TiO2 and SiO2 were fired overnight at 1000 oC and then stored at 110 oC, and the other chemicals were stored at 110 oC prior to use. After mixing the components under ethanol in an agate mortar for 1 h, the mixture was dried in air and decarbonated in an iron-saturated Pt crucible in a box furnace by gradually raising the temperature from 650 to 1000 oC in 7 h31. The resulting mixture was melted in an iron-saturated Pt crucible for 30 mins at 1550 oC to promote homogeneity and quenched to a glass by immersing the bottom of the Pt crucible in water. The glass was subsequently crushed, dried, and reground under ethanol with appropriate amounts of CaCl2 powder in an agate mortar for 1 h and then kept at 110 oC until use.

High-pressure and high-temperature experiments at pressures of 0.5–1 GPa and temperatures of 1120–1240 oC were conducted in a QUICKpress piston cylinder press using a half-inch (12.7 mm) diameter talc-pyrex cell at the Vrije Universiteit Amsterdam, using methods described in detail in our previous work31. Temperatures were measured using a PtRh30-PtRh6 (type B) thermocouple. Graphite-lined Pt capsules are used in all experiments. As in ref. 31, a hand-machined graphite bucket, with an ID of 0.7 mm, OD of 1.6 mm and a length of 3-4 mm, was filled with starting material, closed with a graphite lid and inserted in a Pt capsule, with an ID of 1.7 mm, OD of 2 mm, and a length of ~6 mm. Capsules loaded with starting material were kept at 300 oC in a furnace overnight to dry before welding. The graphite-lined Pt capsules impose an oxygen fugacity close to that of the carbon-carbon monoxide (CCO) buffer36, ensuring that iron in our experiments is present predominantly as Fe2+. This is close to the current best estimates of oxygen fugacity conditions in the LMO61,62. Experiments were kept at 800 °C and target pressure for 30 mins for sintering31. Then they were heated to a near-liquidus temperature of 1300 °C for 30 min and cooled to the target temperature at a rate of 5 oC per 30 mins. Subsequently, the experiments were kept at their final target temperatures for 22 to 48 h. At the end of experiments, samples were quenched by turning off the power, after which temperatures typically dropped to values below 50 oC in approximately 20 s. Experimental conditions are summarized in Supplementary Data 2.

Electron microprobe analysis

Experimental run products were mounted in epoxy and polished, and carbon-coated for electron microprobe analysis. The chemical compositions of the run products (minerals and glasses) were determined using a JEOL JXA-8530F field emission electron probe microanalyzer (EPMA) at the Universität Münster. As previously described in Zhang et al.63 and Flemetakis et al.64, the electron beam size was set to a diameter of 2 and 10 μm for the mineral phases and glasses, respectively. Major element abundances were determined using a beam current of 15 nA and operated at 15 kV accelerating voltage. Counting times were 15 s for peak and 5 s for background. Chlorine abundances in the minerals and glasses were determined in a second step using counting times of 120 sec on peak and 60 sec on background at beam currents were 60 nA. The electron microprobe gave a detection limit of ~28 μg/g for Cl analysis. Both analyses, major elements and Cl, were merged using the Offline-matrix correction EPMA software. The experiment results of mineral major and Cl compositions are provided in the Supplementary Data 2.

Water content measurement

The water contents of the glasses from our experiments were measured by a combination of a Fourier-transform infrared spectrometer (FTIR, JASCO IRT-5200) and a Raman spectrometer (JASCO NRS-5100gr) at the Geodynamics Research Center (GRC), Ehime University. Most samples were analyzed by Raman spectrometer only, because most of the recovered samples with quench cracks were difficult to doubly polish, and the area of quenched glass pockets in the polished run products was too small to enable FTIR measurements (with the sole exception of sample No. JQCl9). In order to assess the validity of the Raman measurements, three water-bearing standard glasses were synthesized using high-pressure apparatuses and measured by FTIR as well. Two water-bearing standard glasses (JB-1b_#1 and JB-1b_2#) were synthesized from a mixture of water-bearing natural alkaline basalt (JB-1b)65, and different amounts of dry oxides after decarbonation (Si-Mg-Ca-Al-Fe) using a 2000-ton multi-anvil apparatus at GRC, Ehime University. One standard glass (PLW1-18N10) was synthesized from dry oxides after decarbonation by a piston-cylinder press at the Vrije Universiteit Amsterdam.

The water content of quenched glass (\({C}_{{{{{\rm{H}}}}}_{2}{{{\rm{O}}}}}\) in weight percent) in these doubly polished standard glasses and sample JQCl9 was determined by FTIR using the following equation:

$${C}_{{H}_{2}O}=\frac{{M}_{{{{{\rm{H}}}}}_{2}{{{\rm{O}}}}}A}{{\rho }_{{{{\rm{glass}}}}}d\epsilon }\times 100$$
(1)

where \({M}_{{{{{\rm{H}}}}}_{2}{{{\rm{O}}}}}\) is the molar weight of H2O (18.02 g/mol), \(A\) is absorbance at the OH peak, \({\rho }_{{{{\rm{glass}}}}}\) is glass density, \(d\) is glass thickness, and \(\varepsilon\) is an excitation coefficient. Here we assumed \({\rho }_{{{{\rm{glass}}}}}\) of 2800 g/L66 and \(\varepsilon\) of 78 L/mol/cm67. Transmitted IR spectra of doubly polished samples below a 100 μm thickness were collected at ~20 Pa with aperture size of 50 × 50 μm2, a spectral resolution of 4 cm−1, and 128 scans (Supplementary Fig. S3). The water content of the glass was determined using Eq. (1), \(d\) of the glass, and \(A\) (the peak height at 3540 cm−1 after the linear background subtraction between 2500 and 3800 cm−1).

The water contents of most quench glasses were measured by Raman spectrometer following method developed in previous studies68,69,70. Raman spectra at 100–4000 cm−1 were collected using a laser having a 532.15-nm excitation wavelength with a × 100 objective lens, 1800 l/mm grating, counting times of 60 sec, and 5–10 times accumulation (Supplementary Fig. S4). The spatial resolution and the spectral resolution are 1–2 μm2 and 0.42 cm−1, respectively. The power input of the laser was reduced to 5.4 mW to minimize possible damage to the samples. A silicon standard having 520 cm−1 peak was used for the daily calibration of the spectrometer.

For the data treatment of Raman spectra, to apply the internal calibration of the previous study70 to our Raman spectra, the effects of elastic scattering of the excitation beam and temperature were corrected by the following equation:

$$I={I}_{{{{\rm{obs}}}}}\left\{{v}_{{{{\rm{o}}}}}^{3}\left[1-\exp \left(-h{{{\rm{c}}}}v/k{{{\rm{T}}}}\right)\right]v/{\left({v}_{{{{\rm{o}}}}}-v\right)}^{4}\right\}$$
(2)

where \(I\) is the corrected intensity, \({I}_{{{{\rm{obs}}}}}\) is the measured intensity of the raw spectrum, \({v}_{{{{\rm{o}}}}}\) is the wavenumber of incident laser light (532.15 nm: 18792 cm−1), \(h\) is the Plank constant (6.62606957 × 10−34 Js), \({{{\rm{c}}}}\) is the speed of light (2.9979 × 1010 ms−1), \(v\) is the measured wavenumber in cm−1, \(k\) is the Boltzmann constant (1.3806488 × 10−23 JK−1), and \(T\) is temperature in K. A spline cubic baseline was fitted using the corrected Raman spectrum in the regions of 100–400 cm−1, an intermediate minimum of 620–680 cm−1, and 1200–1500 cm−1 for Si-O vibrations and 2800–3400 cm−1 and 3700–4000 cm−1 for O-H stretching vibration. Typical spectra after the baseline subtraction are shown in Supplementary Fig. S5. The water content (\({C}_{{{{{\rm{H}}}}}_{2}{{{\rm{O}}}}}\)) in weight percent of the quenched glass was estimated by the following equation:

$${C}_{{{{{\rm{H}}}}}_{2}{{{\rm{O}}}}}\,=a\, {A}_{{{{\rm{w}}}}}/{A}_{{{{\rm{s}}}}}$$
(3)
$$a=0.096\times {C}_{{{{\rm{FeO}}}}}+0.663$$
(4)

where \({A}_{{{{\rm{w}}}}}\) and \({A}_{{{{\rm{s}}}}}\) are area of the O-H stretching vibration region and the Si-O vibration region, respectively. Slope \(a\) is determined using the Eq. (4) with FeO content in weight percent of the glass from electron microprobe analyses. The resulting water contents in quenched glasses are summarized in Supplementary Data 3. In comparison with FTIR measurements, the water contents of synthetic standard glasses and JQCl9 determined by Raman measurements are reproduced with a root-mean-square error (RMSE) of 0.24 wt.% H2O (Supplementary Fig. S6 and Supplementary Data 3).

Modeling F/Cl evolution

The partition coefficients used in our modeling are summarized in Supplementary Data 4. The Cl partition coefficients for orthopyroxene and plagioclase are the mean values with 1 σ standard deviation from this study after exclusion of measurements with abnormally high uncertainty, while F partition coefficients for orthopyroxene and plagioclase are the mean values reported in ref. 31. The number of available olivine-melt partition coefficients is limited and we chose a pair of partition coefficient for F and Cl from Experiment Fe_5 conducted by Potts et al.8. All partition coefficients are from nominally anhydrous experiments close to lunar conditions.

The F/Cl ratio evolution modeling includes two processes. First, we have modeled the F/Cl ratio of minerals (plagioclase and orthopyroxene) crystallized from their equilibrated magmas with different F/Cl ratios. As the plagioclase and orthopyroxene have an identical \({D}_{{{{\rm{F}}}}}/{D}_{{{{\rm{Cl}}}}}\) ratio, the F/Cl ratios of both minerals formed from the same magma are indistinguishable. The modeling results are plotted in Fig. 3b. Second, we have calculated the halogen contents and F/Cl ratios of different mantle sources, including (1) olivine and orthopyroxene; (2) olivine and plagioclase; (3) orthopyroxene and plagioclase.

The halogen concentrations of the mantle sources (composed of LMO cumulates) are controlled by their concentrations in the LMO magmas with which the cumulates equilibrated. McCubbin et al.5 concluded that the bulk silicate Moon (BSM) has 3.1–4.9 μg/g F and 1.0–4.1 μg/g Cl halogen contents with F/Cl ratios ranging from approximately 1 to 5. In our modeling, the F/Cl ratio of the initial LMO is set within this range (Supplementary Table S4). Since halogens are highly incompatible, we can simply model the halogen contents in LMO residual magma after any degree of solidification by assuming all halogens remain in the liquid during crystallization, regardless of whether it is equilibrium or fractional crystallization. This also indicates that the F/Cl ratio of the partial melts formed from a mantle source would be nearly constant even the melt underwent certain degrees of evolution.

For modeling partial melts from an olivine-dominated mantle, we adopted a 1.9% melting model based on ref. 55, with the accumulated fractional melting equation expressed as:

$${C}_{{{{\rm{melt}}}}}={C}_{0}{\left(1/F\right)\left[\right.1-(1-F)}^{\left(\frac{1}{{D}_{{{{\rm{bulk}}}}}}\right)}$$
(5)

To model the F/Cl ratio of partial melts derived from an orthopyroxene-dominated mantle, we applied a 10% batch melting degree following Sheng et al.56, using the batch melting equation:

$${C}_{{{{\rm{melt}}}}}={C}_{0}/\left[{D}_{{{{\rm{bulk}}}}}+F(1-{D}_{{{{\rm{bulk}}}}})\right]$$
(6)

where \({C}_{{{{\rm{melt}}}}}\) is the halogen content of partial melts, \({C}_{{{{\rm{o}}}}}\) is the halogen content of the mantle source, F is the melt fraction, \({D}_{{{{\rm{bulk}}}}}\) is the bulk partition coefficient calculated based on modal abundances of mineral phases. The olivine and orthopyroxene in the mantle source are considered to be in equilibrium with the LMO after 50% solidification and expected to be Mg-rich, and the plagioclase is considered to have formed in equilibrium with the LMO after 80% solidification when plagioclase is expected to occur38,71,72,73. In this modeling, the LMO residual magma after 50% solidification has been considered as TIRL. All the modeling results and associated uncertainties were generated by a Monte Carlo approach on the basis of 1,000,000 simulations using Eqs. (5) and (6). The details and code of the Monte Carlo simulation are provided in the Supplementary Information and Supplementary Software 1.