Carbon distribution in planet Mercury from magma ocean evolution to graphite crust and core composition Authors: Olivier Namur, Bernard Charlier, Camille Cartier, Kaustubh Hakim, Johan Villeneuve, Nicola Tosi, Jasper Berndt, Stephan Klemme, Tahar Hammouda, Maxime Maurice, Maud Boyet, Celia Dalou, Oliver Shorttle Published: 2026 ABSTRACT AbstractMESSENGER observations revealed a primary graphite flotation crust on Mercury, implying substantial carbon retention in its magma ocean rather than sequestration into the core. To investigate the conditions enabling this retention, we conducted high-pressure, high-temperature metal–silicate partitioning experiments over a wide range of oxygen fugacities. Carbon behavior is strongly redox dependent: under relatively oxidizing conditions it is highly siderophile, whereas under the reducing conditions relevant to Mercury it becomes significantly less siderophile, promoting carbon retention in silicate melts and graphite crystallization. Modeling of carbon partitioning between the core, mantle, crust, and atmosphere indicates that oxygen fugacities of IW − 6 to IW − 6.5 best reproduce the graphite crust thickness inferred from MESSENGER data. Under these conditions, Mercury’s core remains relatively carbon-poor ( < 5000 μ g/g), implying that its density deficit is primarily controlled by other light elements, most likely silicon and sulfur. These results link Mercury’s extreme reduction to both its graphite crust and internal chemical structure.Subject terms: Petrology, Inner planets Introduction Mercury, the smallest planet in the Solar System, is characterized by a disproportionately large metallic core (~ 70% of the planet’s mass1,2), which may be due to the removal of a large fraction of the silicate shell of an originally larger proto-Mercury (i.e., ~ 3500 km radius)3. Inversion of MESSENGER geodetic data converges to a core radius of ~ 1950–2020 km surrounded by a silicate shell (mantle and crust) ~ 420–470 km thick2,4,5, corresponding to a pressure at the core-mantle boundary of ~ 5–6 GPa4. Mercury’s surface is notably depleted in iron oxide and enriched in sulfur, indicating highly reducing conditions during mantle melting and crust formation6. These geochemical features suggest that similarly low oxygen fugacity (fO2) conditions prevailed during equilibration between Mercury’s metallic core, silicate magma ocean (MO), and possibly its primary atmosphere. Conservative estimates place Mercury’s fO2 between IW − 3 and IW − 77, where IW refers to the iron-wüstite (Fe-FeO) buffer, although modeling of surface sulfur concentrations further constrains fO2 to IW − 5.4 ± 0.56. Owing to these reducing conditions, it is commonly assumed that Mercury accreted from building blocks compositionally similar to enstatite chondrites8–10. Mercury’s surface includes low-reflectance terrains interpreted as mixtures of silicate material with graphite and, potentially, metallic iron11,12. Spectral reflectance models reproducing observed surface spectra require the presence of approximately 1–3 wt.% graphite11,12. Given its low density, graphite has been proposed to represent remnants of a buoyant primary crust that formed by flotation during crystallization of the global silicate MO13. This primordial crust would have subsequently been disrupted and redistributed, both by impact processes and by mixing with silicate material from the planet’s volcanic secondary crust. In this study, we present a series of high-temperature, high-pressure experiments under oxygen fugacities spanning IW − 3 to IW − 9. We used starting materials representative of Mercury’s mantle and crust to investigate the solubility of carbon in reduced silicate melts and metallic alloys under conditions relevant to Mercury’s differentiation. We find that carbon becomes increasingly lithophile and less siderophile under increasingly reduced conditions. Combining the experimentally determined carbon partitioning behavior with a thermodynamic model of MO degassing14, we evaluate the carbon distribution between Mercury’s core, magma ocean, and primary atmosphere. We then model graphite saturation in the MO to constrain the thickness of a primordial graphite-rich flotation crust that would have formed atop Mercury’s MO as a function of fO2. Modeled graphite thicknesses are also used in combination with MESSENGER data to constrain the most likely fO2 conditions under which Mercury differentiated. Our model has strong implications for the composition of Mercury’s core. Inversion of MESSENGER geodetic data indicates that Mercury’s core must contain several wt.% of light elements1. Yet, under the fO2 conditions that stabilize a graphite crust, we find that C is strongly depleted in the core. Our analysis, therefore, shows that other candidate light elements, Si, and to a lesser extent S, are the only ones that can enter the core in sufficient proportions to explain the geodetic data. These elements also significantly depress the core solidus, which can explain why Mercury’s core has remained molten over 4.5 Gyr15. Although strongly focused on Mercury, our study will be critical to further understand larger, reduced planets such as proto-Mercury3, super-Mercuries or reduced exoplanets16,17, and the early Earth, which dominantly accreted from reduced bodies18. Results and discussion Carbon solubility at saturation in silicate and metal melts We conducted a series of high-pressure, high-temperature metal-silicate melt equilibria experiments using a natural enstatite chondrite (Hvittis, EL6), a natural H5 ordinary chondrite (Tamdakht), a natural carbonaceous chondrite (NWA11345), a synthetic powder formulated to replicate the bulk composition of an enstatite chondrite, and two synthetic powders replicating the bulk composition of Mercury’s Northern Volcanic Plains and High-Mg region19 (Table 1 and Methods). Experiments were performed between 1523 and 2443 K in evacuated silica tubes (100 kPa), internally heated pressure vessels (0.1–0.3 GPa), a piston-cylinder apparatus (1.3–4.0 GPa), and multi-anvil presses (5–26 GPa) (Methods). Details of experimental conditions are given in Table 2 and Supplementary Fig. S1. All experiments were conducted in graphite capsules to ensure carbon saturation. Table 1Chemical compositions (wt.%) of starting materialsSampleSiO2TiO2 /Ti*Al2O3Fe2O3 /Fe*MnO /Mn*MgOCaONa2OK2OP2O5 /P*NiO /Ni*Cr2O3 /Cr*LOITypeReferenceHvittis42.100.02*1.2725.90*0.12*25.500.890.470.040.11*1.67*0.21*–Nat.64EC62.420.102.70–0.1030.811.771.700.20–0.100.10–Syn.6NVP58.790.3013.80–0.1713.255.977.010.17–0.100.10–Syn.6Tamdakht39.290.112.3832.950.3324.772.040.970.110.32––8.12Nat.This studyNWA 1134532.240.122.4934.570.2721.701.770.340.150.28––4.76Nat.This studyHMg53.800.368.38–0.2126.527.432.660.09–0.100.10–Syn.6*Concentrations for these elements are reported in elemental metal form rather than as oxides. Chemical compositions (wt.%) of starting materials *Concentrations for these elements are reported in elemental metal form rather than as oxides. Table 2Experimental conditions and phase assemblagesSampleT (K)ΔIWP (GPa)DurationComp.S sourceSi frac.Phases*s_10831883− 6.605.003.5 hHvittis–0.10sil, met, sul, Ens_11331963− 5.705.003.5 hHvittis–0.10sil, met, sul, Ens_11351923− 6.205.002.5 hHvittis–0.05sil, met, sul, Ens_11361913− 5.505.002.5 hHvittis–0.05sil, met, sul, Ens_11371973− 0.905.003.0 hHvittis–0.00sil, met, sul, Ens_11431873− 7.405.003.0 hHvittis–0.30sil, met, sul, Ens_11781853− 7.805.002.0 hHvittisFeS0.50sil, met, sul, En, Coes_11831993− 0.805.002.25 hHvittis–0.00sil, met, sul, Ens_11841913− 7.905.002.0 hHvittis–0.30sil, met, sul, Ens_11982073− 1.005.003.0 hHvittisFeSFeSisil, met, sul, Ens_12072073− 4.505.003.0 hHvittis–0.02sil, met, sul, Ens_12102073− 5.105.003.0 hHvittis–0.05sil, met, sul, Ens_12112123− 4.305.002.0 hHvittisFeSFeSisil, met, sul, Ens_12122123− 4.905.003.0 hHvittis–0.05sil, met, sul, Ens_12431973− 7.905.002.0 hHvittis–0.42sil, met, sul, Ens_12451973− 7.205.002.0 hHvittis–0.42sil, met, sul, Ens_12472023− 6.605.002.0 hHvittis–0.42sil, met, sul, Ens_12612023− 6.405.002.0 hHvittis–0.42sil, met, sul, Ens_12622023− 7.405.002.0 hHvittis–0.42sil, met, sul, Ens_12632033− 6.405.002.0 hHvittis–0.42sil, met, sul, Ens_176a2444− 2.105.0040 minHvittis–0.00sil, mets_1772292− 2.1026.035 minHvittis–0.10met, Maj, Bdm, Stis_1952345− 5.6026.020 minHvittis–0.10met, Maj, Bdm, Stis_1982360− 4.8026.030 minHvittis–0.05sil, mets_2052349− 5.4026.035 minHvittis–0.10met, Maj, Bdm, Stis_A0241973− 6.503.006.0 hECFeS, S0.20sil, met, suls_A0251793− 6.801.306.0 hECFeS, S0.20sil, met, suls_A0401993− 4.401.306.0 hECFeS, S0.10sil, met, suls_A0411993− 4.502.006.0 hECFeS, S0.10sil, met, suls_A0421873− 6.701.306.0 hECFeS, S0.20sil, met, suls_A0451973− 6.206.006.0 hECFeS, S0.20sil, met, suls_A7021743− 6.502.005.0 hNVP–0.10sil, met, Ens_A7041673− 4.901.505.0 hNVPFeS, S0.10sil, met, sul, Ens_A7101833− 6.002.005.0 hECFeS0.20sil, met, sul, Ens_A7122023− 5.604.003.0 hECFeS0.20sil, met, sul, Ens_B8731973− 1.703.004.0 hEC–0.00sil, mets_B8771973− 6.803.004.0 hEC–0.50sil, mets_B8791973− 6.403.004.0 hECFeS0.50sil, met, suls_B8801973− 5.303.004.0 hEC–0.20sil, mets_A6011753− 6.501.704.0 hEC–0.20sil, mets_FM1501573− 5.8010−924 hTamdakht–0.00sil, met, sul, En, Pls_FM1521623− 5.5010−924 hTamdakht–0.00sil, met, sul, En, Pls_FM1531573− 5.4010−948 hTamdakht–0.00sil, met, sul, En, Pls_FM158b1523− 5.5010−972 hNWA 11345–0.00sil, met, sul, En, Pls_FM1651673− 5.4010−95.0 hTamdakht–0.00sil, met, sul, En, Pls_FM1661673− 5.3010−95.0 hNWA 11345–0.00sil, met, sul, En, Pls_y012_21633− 6.600.106.0 hNVPFeS0.20sil, met, suls_y013_11773− 4.600.104.0 hNVPFeS0.05sil, met, suls_y013_21773− 7.100.104.0 hNVPFeS0.10sil, met, suls_y0141633− 6.300.106.0 hNVPFeS0.20sil, met, suls_y022_11693− 7.500.105.0 hNVPFeS0.50sil, met, suls_y022_21693− 7.600.105.0 hNVPS0.50sil, met, suls_y023_11773− 4.300.103.0 hHMgFeS0.02sil, met, suls_y023_21773− 5.000.103.0 hHMgFeS0.05sil, met, suls_y024_11773− 5.200.103.0 hHMgFeS0.10sil, met, suls_y024_21773− 5.900.103.0 hHMgFeS0.20sil, met, suls_y026_11833− 4.500.103.0 hHMgFeS0.20sil, met, suls_y026_21833− 8.400.103.0 hHMgFeS, S0.50sil, met, suls_y0291633− 6.100.104.0 hNVP-0NaFeS0.05sil, met, suls_y030_11693− 7.100.104.0 hNVPS0.20sil, met, suls_y030_21693− 7.000.104.0 hHMgFeS, S0.50sil, met, sul, Fos_y031_21633− 8.000.103.0 hNVPS0.50sil, met, suls_y032_11773− 7.100.103.0 hNVPFeS, S0.50sil, met, suls_y032_21773− 7.300.103.0 hNVPFeS, S0.50sil, met, suls_y033_11693− 7.100.103.0 hNVPFeS, S0.50sil, met, suls_y035_11583− 2.100.103.0 hHMgFeS0.50sil, met, suls_y035_21583− 6.100.103.0 hNVPS0.20sil, met, suls_y047_21583− 5.100.106.0 hNVPFeS, S0.10sil, met, suls_y048_11728− 5.700.305.0 hHMg–0.10sil, mets_y048_21728− 4.800.305.0 hHMgFeS0.10sil, met, suls_y049_21743− 4.600.104.0 hHMgFeS, S0.10sil, met, suls_y056_11793− 4.000.103.0 hHMg–0.05sil, mets_y056_21793− 6.200.103.0 hHMgFeS0.20sil, met, suls_y083_11793− 3.500.105.0 hHMg–0.10sil, mets_y083_21793− 6.100.105.0 hHMgFeS0.50sil, met, suls_y0971728− 5.800.105.0 hHMg–0.20sil, mets_y0991758− 5.900.504.0 hHMg–0.20sil, met*Sil=silicate melt, Met=metallic alloy, Sul=sulfide, En=enstatite, Fo=forsterite, Coe=coesite, Maj=majorite, Bdm=bridgmanite, Sti=stishovite. Experimental conditions and phase assemblages *Sil=silicate melt, Met=metallic alloy, Sul=sulfide, En=enstatite, Fo=forsterite, Coe=coesite, Maj=majorite, Bdm=bridgmanite, Sti=stishovite. All experimental runs contain an assemblage with a silicate melt and a metal phase. Depending on experimental conditions and starting compositions, additional phases are observed, including sulfides and silicate minerals (enstatite, forsterite, quartz, coesite, stishovite, majorite, and bridgmanite). In all experiments at 100 kPa to 6 GPa, the silicate melt is preserved as perfectly quenched glass. However, in experiments at 26 GPa, dendrites formed upon quenching. Additional details on experimental products are given in the Supplementary Section 1 and representative photomicrographs of the run products are shown in Supplementary Fig. S2. In all run products, electron probe microanalysis (EPMA) was used to analyze the major element composition of the silicate glass and the C content in the metal phase, and secondary ion mass spectrometry (SIMS) was used to analyze the C and H contents in the silicate glass (Methods). Glass and metal compositions are presented in elemental concentrations in Supplementary Data 1. All elements are reported in wt.%, except for C and H in silicate glass, which are reported in μg/g. Silicate melt. Carbon content at graphite (100 kPa to 6GPa) or diamond (26 GPa) saturation in the silicate melt is shown as a function of fO2, expressed as ΔIW in Fig. 1 and provided in Supplementary Data 1. We observe that the carbon content in the silicate melt continuously decreases from 2454 μg/g at IW + 2 to ~ 8 μg/g at IW − 1 to IW − 2, where it reaches a minimum. At more reducing conditions, the carbon content in the silicate melt increases significantly, although the increase is strongly dependent on pressure. In 100 kPa to 2 GPa experiments, the carbon content increases from 4 μg/g at IW − 3 to 388 μg/g at IW − 8. In experiments at 5–6 GPa, it increases from 55 μg/g at IW − 4 to ~ 1990 μg/g at IW − 8, with one outlier at 3489 μg/g (IW − 6.2, 6 GPa). At 26 GPa, the carbon content in the silicate melt seems to be less fO2 dependent, although our experiments cluster at only two fO2 values. It is, however, strongly enhanced by high-pressure conditions reaching 4800–10 400 μg/g at IW − 2 to IW − 5. Data at 26 GPa should, however, be taken with extra caution because the experiments showed dendritic growth in the silicate melt rather than a homogeneous glassy texture (see above and Supplementary Fig. S2). This dendritic growth is known to artificially increase the carbon content by 10–1000 μg/g due to matrix effects during SIMS analyses (“Methods”) compared to calibrating standards (Supplementary Fig. S3). Carbon in the silicate melt. Carbon content at graphite or diamond saturation in the silicate melt as a function of fO2 [ΔIW]. Error bars represent 1σ standard deviations. Carbon species in oxidized and reduced conditions, as determined by Raman spectroscopy, are shown by orange and magenta arrows. The red bar across the vertical color scale corresponds to the estimated pressure at Mercury’s core-mantle boundary. References for literature data are given in the Methods and in Supplementary Dataset 1. At any given fO2 (and pressure), we observe some variability in the silicate melt C contents. This variability can be explained by temperature differences between experiments. Indeed, multiple linear regression analysis of our data shows that C solubility in the silicate melt is independent of silicate melt composition but moderately dependent on temperature, in addition to pressure and fO2 (see Methods; Supplementary Figs. S4–S6). In general, the C concentrations that we observe in our experiments align well with existing literature where fO2 conditions overlap (Supplementary Data 1) and extend their trends towards more reduced conditions. Our results also support and extend thermodynamic models of C solubility previously calibrated on existing experiments20. The speciation of carbon in the silicate melt was determined based on previous literature21,22 and confirmed by Raman spectroscopy (see details on Raman spectra of representative samples in Supplementary Section 2 and Supplementary Fig. S7). Between IW + 2 and IW − 1, carbon is present as C–O species (i.e., CO32−) whereas C dissolves as C–H species at more reducing conditions. Relatively high H contents were measured in our experimental silicate melts (3–3000 μg/g; Supplementary Data 1), confirming C–H speciation. The dominant carbon speciation that we observe seems to depend only on fO2 and not the bulk composition used in the experiment. For synthetic compositions prepared on an H-free basis, we suspect that H was gained from the assembly and surrounding atmosphere during the experiment. Minor Raman peaks attributed to N–H species are observed in the most reducing experiments (< IW − 6). Although high N concentrations in silicate melts can increase carbon solubility at fO2 > IW − 3 due to the formation of C–N species22 (Supplementary Fig. S8), this effect disappears under more reducing conditions because N becomes locked into the silicate as anhydrous N3− linkages22. Therefore, the minor amount of N present in our experiments does not affect the measured carbon concentrations, as the vast majority of our experiments were conducted at fO2 below IW − 3. In our lowest pressure experiments (100 kPa), where we observe the lowest C concentrations, the dissolution of C–H species might be limited by low fCH4 values20, possibly pointing to the presence of other C-bearing species (e.g., carbide) in the melt. However, relatively high (up to 350 μg/g) OH− concentrations were observed at 100 kPa in reduced melts similar to those of this study23, which supports the formation C–H species even at very low pressure20. Metal melt. The carbon contents in the equilibrium metal melt are shown in Fig. 2. They are generally higher than in the conjugate silicate melt and are highly dependent on fO2 (Fig. 2a) and the metal Si content (Fig. 2b), both being strongly coupled6. Indeed, the metal C content is negatively correlated with Si content6,24–26 (Fig. 2b), likely due to the large effect of silicon on the carbon activity coefficient (γC)27. We therefore interpret the decreasing C solubility at graphite saturation under more reducing conditions as being due to the higher metal Si content. We also observe that C is negatively correlated with S (Supplementary Fig. S9), which also likely affects γC. Carbon in metal melt. a Carbon content at graphite or diamond saturation in metallic alloy as a function of fO2 [ΔIW]. b Correlation between Si and C at graphite or diamond saturation in metallic alloy. Experiments performed in silica tubes (100 kPa) are denoted by stars because they behave as outliers (see text). Thermodynamically modeled C saturation (FactSage) in the Fe-Si-C-O system is also shown. Error bars represent 1σ standard deviations. From the literature data, we filtered out any data with > 10 wt.% S because high S contents are known to decrease C solubility80. In detail, when compared to fO2, we note that the metal carbon content does not show an obvious trend between IW and IW − 4, with values on the order of 2.3–8.7 wt.% (mean = 4.25 wt.%), in agreement with previous studies21,24,28. Under more reducing conditions, it drops significantly and steadily, reaching values of 0.12 wt.% at IW − 8, values lower than previously reported25. We do not observe a significant effect of pressure on the carbon content in the metal melt under fO2 conditions more oxidizing than IW − 4, except perhaps for experiments at 26 GPa that have lower C contents (2.3–3.5 wt.%). Under more reducing conditions, however, the pressure effect seems stronger. Low-pressure experiments (< 1 GPa) generally exhibit higher carbon concentrations compared with higher-pressure experiments (5–6 GPa, 26 GPa) experiments: ΔC = 1.93 wt.% at IW − 4 to IW − 6 and ΔC = 0.96 wt.% at IW − 6 to IW − 8, with ΔC being the difference between the average C content at low pressure and the average C content at high pressure. Experiments performed at 100 kPa in silica tubes are an exception to this trend, with low metal carbon contents similar to or lower than those of high-pressure experiments (5–26 GPa). This might be due to the volatilization of carbon in the silica tubes. We note that the lowest metal C concentrations are observed in the most reducing experiments in which some N was detected in the silicate melt. However, compilation of published N-bearing and N-free experiments shows that N has no substantial effect on C solubility in metals (Supplementary Fig. S10). Finally, we note that the evolution of C in metal as a function of fO2 is similar to the trend calculated using thermodynamic simulations in the Fe-Si-C-O system at log10pO2 = − 8 to − 16 (pO2 = oxygen partial pressure) and 1873 K (approximately our mean experimental temperature; Methods; Fig. 2a). Metal-silicate carbon partitioning under reducing conditions By combining the carbon contents of the silicate and metal melts, we estimated the evolution of the metal-silicate partition coefficient (DCm/s) as a function of fO2 and pressure (Fig. 3; see details on error propagation in Methods). Because the carbon content in the silicate melt reaches a minimum around IW − 1 to IW − 2 (Fig. 1), we restricted our analysis to fO2 conditions more reducing than IW − 1. Sigmoid regressions (“Methods”) were performed for three groups of experimental pressures: < 2 GPa (excluding 100 kPa experiments), 5–6 GPa, and 26 GPa. At the lowest pressure conditions, DCm/s is high (~ 5000) from IW − 1 to IW − 5, then decreases to ~ 15 by IW − 8. At intermediate pressures, a similar trend is observed but shifted to lower values, decreasing from maximum values of ~ 900 between IW − 1 and IW − 4 to ~ 1.5 at IW − 8. At 26 GPa, DCm/s shows little dependence on fO2, clustering at values of 2–5, although this trend is poorly constrained. Metal-silicate partitioning of C. Evolution of DCm/s as a function of ΔIW. Dashed lines show sigmoid functions fitted to experiments at < 2 GPa (blue), 5-6 GPa (green), and 26 GPa (yellow). Note that the fitted curve at 26 GPa is poorly constrained due to the low number of experiments, which cluster at two fO2 values. Error bars represent 1σ standard deviations. The observed DCm/s versus fO2 trends differ substantially from those generally assumed for carbon distribution between metal and silicate under reducing conditions. In a recent study, Keppler and Golabek29, using an empirical model calibrated on high-pressure experiments at 1–3 GPa30, suggested that carbon strongly partitions into the metal (Supplementary Fig. S11). According to their model, DCm/s increases monotonically from 8 × 103 at IW − 1 (2000 K) to 2.5 × 105 at IW − 6, a trend that would extend to 8 × 106 at IW − 8 (Supplementary Fig. S11). This discrepancy arises from the limited experimental database used to calibrate their DCm/s behavior, which included only experiments performed at IW − 1 to IW − 2 and is therefore unreliable beyond its calibration range. Our data further highlight the strong effect of pressure on DCm/s, which is primarily driven by the pronounced influence of pressure on C solubility in silicate melts, and secondarily by its moderate effect on the C content of the metal phase; both factors contribute to a decrease in DCm/s with increasing pressure. This effect was not captured in previous models because they were calibrated using experiments conducted over a narrower pressure range (e.g., 1–6 GPa) and an fO2 range (IW − 1 to IW − 2) where the influence of pressure is not as strong28,30,31. Carbon distribution inside planet Mercury The variation in carbon partitioning between metal and silicate as a function of fO2 provides a powerful constraint on the distribution of carbon within Mercury. Here, we calculate the C distribution in the planet’s interior across the range of fO2 values typically proposed for Mercury (IW − 3 to IW − 7;6,7) and extend our calculations to more reducing conditions (IW − 8). We assume a bulk planet composition similar to enstatite chondrites (EH), which are generally considered as the most likely building blocks of Mercury9,10,32 and contain an average carbon concentration of 4097 ± 1436 μg/g (average of 13 samples; Supplementary Fig. S12)33. We take into account that part of the bulk planetary carbon budget lies in the primary atmosphere that was in thermodynamic equilibrium with the MO, which limits the C budget of the planet’s interior (core and MO). To determine the C content of the atmosphere, we use two different methods. (1) Our lowest-pressure experiments contain silicate melts at graphite saturation with 1–70 μg/g C. According to Armstrong et al.34, a MO with such a composition would be in equilibrium with a CO-rich atmosphere (pCO ≈ 9600–88 800 kPa), accounting for 253–2300 μg/g of the planetary carbon budget (Methods). (2) We use the thermodynamically robust magma ocean-atmosphere models MAGWORLD-I14 and Atmodeller35. Model calculations were performed at 1673 K19 using an EH composition for C-H-N-S abundances36 and a magma ocean mass of 1 × 1023 kg (Mercury’s bulk silicate mass; see physical properties in Table 3). From IW − 8 to IW − 3, these model calculations predict respective CO partial pressures between 17–6480 kPa (MAGWORLD-I) and 16–5100 kPa (Atmodeller), equivalent to respective atmospheric contributions to the bulk planet’s carbon budget of 120–1120 μg/g and 137–950 μg/g (Supplementary Fig. S13). These small differences arise from the implementation of solubility laws14,35. Using these atmospheric carbon contents, we calculate that Mercury’s interior contains 2931–4036 μg/g C. Table 3Physical and thermodynamic properties of Mercury, the magma ocean, graphite, and relevant constantsPropertySymbolValueUnitReferencePhysical properties of MercuryMass MercuryMm3.28 × 1023kg1Radius MercuryRm2.44 × 106m1Radius coreRc2.02 × 106m4Radius inner coreRicvariablem4,5Core mass fraction MercuryCMFm0.7–4Gravity Mercuryg3.70m/s2–CMB pressure MercuryPCMB5–6GPa4Physical properties of the magma oceanDensity MOρMO3200kg/m313Volume MOVMO2.62 × 1019m3–Depth MOzMO4.20 × 105m1Surface MOAMO7.48 × 1013m2–Thermodynamic properties of the magma oceanLatent heat MOLH4.00 × 105J/kg82Heat capacity MOCp,MO1405J/kg/K82Thermal expansion MOα3.0 × 10−51/K82Properties of carbon phasesDensity graphiteρgr2260kg/m313Density diamondρdm3250kg/m326Density Fe3CρFe3C7700kg/m315Thickness graphite crustzgVariablem–ConstantsGas constantR8.3144J/mol/K– Physical and thermodynamic properties of Mercury, the magma ocean, graphite, and relevant constants Using the calculated bulk C budget of the planet’s interior, we then determine the partitioning behavior of carbon between the metallic core and the overlying MO. We use our parameterization relating DCm/s to fO2 obtained from our 5–6 GPa experiments (Fig. 3; “Methods”) because they correspond to the most likely conditions of core-magma ocean equilibration4,6. For mass-balance constraints, we consider a core mass fraction of 0.7 (i.e., MO mass fraction = 0.3; Table 3)1 and that differentiation occurs at a temperature of 2173 K at the CMB6. We obtain that Mercury’s MO (blue curve in Fig. 4a) contains 10–400 μg/g carbon if equilibrated between IW − 3 and IW − 6.9 (see full error propagation in Supplementary Fig. S14); the onset of graphite saturation occurs at IW − 6.9. With increasingly reduced conditions (IW − 6.9 to IW − 8), the MO carbon content follows the graphite saturation curve (Supplementary Fig. S15), increasing to ~ 800 μg/g at IW − 8. Similarly, the core is expected to contain 3900–4700 μg/g C between IW − 3 and IW − 6.9 (red curve in Fig. 4a). Once graphite saturation is reached, the core C content decreases sharply, dropping to ~ 1200 μg/g at IW − 8, which matches the C content observed in the metal phase of our most reduced experiments. Carbon distribution in Mercury’s interior. a Evolution of the C content of Mercury’s core and MO as a function of fO2. Note that the core C concentration increases from IW − 3 to IW − 6.9, mainly due to the thinning of the primary atmosphere14,35. The orange curve shows the C contribution that does not enter the core or MO because it is in excess of C solubility in both the metal and silicate (insoluble carbon). In this case, graphite was already present at the onset of magma ocean crystallization and we consider it to be primordial graphite. b Calculated thickness of the graphite flotation crust atop Mercury’s MO. The relative contributions of graphite formed by solidification of the MO and by the accumulation of primordial graphite are shown. Light and dark green areas correspond to fO2 ranges proposed for Mercury’s interior6,7. All calculations were performed at 2173 K and a CMB pressure of 6 GPa6,26. Using our partitioning model, we can also evaluate the speciation of carbon during Mercury’s primordial differentiation. As shown in Fig. 4b, once carbon saturation is reached (IW − 6.9), a substantial fraction of the bulk planetary carbon no longer partitions into the core, MO, or atmosphere. We infer that this represents primordial graphite (insoluble carbon; orange curve) that would have accumulated as a floating crust atop the MO13, corresponding to a total contribution of up to 8000 μg/g when scaled to the MO mass. We further assume that the carbon dissolved in the magma ocean between IW − 3 and IW − 8 ultimately forms graphite upon MO solidification. By summing the contributions of these two reservoirs, we calculate that only a thin graphite crust (0–300 m thick) would have formed between IW − 3 and IW − 6.9. Under more reducing conditions, however, a much thicker graphite crust develops, reaching 4600 m at IW − 8. It is important to note that our model assumes that Mercury’s magma ocean contains sufficient hydrogen to stabilize carbon as C–H species (Fig. 1), which are key to the high C solubility in the MO. High H2 solubility in the MO37 is indeed very likely given the high H contents of EH chondrites38 and the high pH2 of the innermost solar nebula39. Even if this were not the case29, the graphite distribution would not be dramatically different than that shown in Fig. 4. In further detail, the calculated graphite crustal thickness for a H-poor planet is very thin (< 1 m) if formed at fO2 conditions more oxidizing than IW − 6.5, and only increases to 4500 m at IW − 8 due to the accumulation of primordial (insoluble) graphite (Supplementary Fig. S16). The analysis shown in Fig. 4 does not consider the possible formation of diamond or Fe3C layers at the CMB15,26. We nevertheless investigate this possibility in the Supplementary Section 3 and show that such layers cannot exceed 60 m for diamond and 13 km for Fe3C, and that they only form at fO2 conditions more reduced than IW − 7 (Supplementary Fig. S17). These values could potentially increase in the case of crystallization of a C-free or strongly C-poor inner core (Supplementary Fig. S18). Given that Mercury’s present size may reflect large-scale silicate removal from a once more massive proto-planet3,40, we also used DCm/s determined at 26 GPa to evaluate the distribution of carbon in such a more massive body. For a proto-Mercury of EH bulk composition, our calculations predict the formation of a 3–6 km thick graphite layer at the surface and a 2–5 km thick diamond layer at the core-mantle boundary. Oxygen fugacity during Mercury’s differentiation The calculated carbon distribution within Mercury imposes strong constraints on the fO2 conditions of the planet’s interior. Indeed, fO2 is a key parameter for assessing the size of carbon reservoirs. Such a constraint can be obtained from the analysis of low-reflectance material on Mercury’s crust41, which is interpreted as carbon-rich and is thought to originate from a ~ 4 km thick layer containing 1–3 wt.% C11. Considering an average crustal density of 3000 kg/m342, we calculate that the primary graphite crust must have been 40–120 m thick. According to our model (Fig. 4), accumulating this thickness of graphite requires fO2 values around IW − 6 to IW − 6.5. At these conditions, the MO contains 80–150 μg/g C, and the metallic core contains ~ 5000 μg/g C (Fig. 4). More oxidizing conditions produce a much too thin graphite crust whereas fO2 values more reduced than IW − 7 quickly increase the graphite thickness beyond observational constraints. Our model assumes a planet with a bulk C content similar to EH chondrites; a higher bulk carbon content would permit the formation of a thicker graphite crust at any given fO2 (Supplementary Fig. S19), although very reduced conditions (< IW − 5) are sill needed to match MESSENGER constraints on graphite thickness. Our analysis, therefore, substantially narrows the previously proposed fO2 range (IW − 3 to IW − 7;7) and brings it much closer to estimates based on Mercury’s surface sulfur content (IW − 5.4 ± 0.5;6). We also note that our fO2 range is near the low-pressure C-CO equilibrium (IW − 6 to IW − 7) at the temperature of crustal magmatism on Mercury (1573–1773 K)19. The primary graphite crust might therefore have buffered the fO2 of magmas during crustal production. Finally, our fO2 is similar to that of the solar nebula (IW − 6.9)39, suggesting that Mercury might have differentiated in equilibrium with H2-rich nebular gas. Indeed, at IW − 6 to IW − 6.5, a H-poor MO cannot form a graphite crust thicker than a few centimeters (Supplementary Fig. S16). Light elements in Mercury’s core MESSENGER revealed the partly molten nature of Mercury’s core and a significant density deficit compared to pure Fe1. These two findings imply the presence of a substantial fraction (i.e., several wt.%1,2,15,43) of light elements. Given the presence of graphite at Mercury’s surface, carbon has been proposed as a potential light element in the core25, although it has not yet been considered in models of Mercury’s interior. Our analysis above indicates that Mercury differentiated under highly reducing conditions around IW − 6 to IW − 6.5. Under these conditions, carbon cannot be considered an important light element of Mercury’s core because its concentration in the core cannot exceed ~ 5000 μg/g (Fig. 4) or perhaps slightly higher in the outer core if carbon were incompatible during inner core crystallization44. Such concentrations would reduce the solidus temperature of Fe by only 30 K at 6 GPa45 and the bulk core density by less than 2%46. Silicon and sulfur are other candidate light elements that could explain Mercury’s geodetic constraints1,2,4,43. Under Mercury’s highly reducing conditions, Si is abundant in the silicate MO but also enters the core easily3,6,47,48. Conversely, S tends to stay in the MO, though some may still be incorporated into the core6. Consequently, the Fe-Si-S system is critical for understanding Mercury’s core, as its pressure-dependent immiscibility field directly influences the core’s structure and chemical makeup. At 5–6 GPa, immiscibility develops when the light element fraction exceeds ~ 12 wt.%49. The distribution of S within Mercury’s interior can be inferred from its partitioning between the metallic core and the MO (DSm/s50), as well as from the planet’s bulk S content6. Here, we adopt a bulk Mercury composition with 4 wt.% S6. We find that a homogeneous Fe-Si-S(-C) core (i.e., no immiscibility) is only possible at fO2 conditions more oxidizing than IW − 6 (Fig. 5). However, between IW − 6 and IW − 8, we predict an Fe-Si-dominated metallic core with relatively low S content (0–6 wt.% S) overlain by an Fe-S layer 0–120 km thick (Supplementary Fig. S20). Within the fO2 range that we consider the most appropriate (IW − 6 to IW − 6.5; see above), we calculate that the Fe-S layer is thinner than 80 km, which agrees well with previous estimates based on the S and Ti contents of Mercury’s crust6,51. Light element distribution in Mercury’s core. Evolution of the C, S, and Si contents of Mercury’s core. The vertical dashed line shows the fO2 at which immiscibility between a metallic core (Fe-Si-C ± S) and a sulfide (Fe-S) layer develops. The blue shaded area between IW − 6 and IW − 7.8 represents the bulk amount of sulfur that contributes to the formation of the Fe-S layer. Error envelopes represent the propagation of 1σ uncertainties on the Si activity coefficient27,67 and on the metal-silicate sulfur partition coefficient, DSm/s50. A key implication of our model is that under highly reducing conditions (IW − 6 to IW − 6.5), Mercury’s metallic core is Si-rich with a moderate S content (2.5–5 wt.%). Our calculated maximum Si concentrations of 8.0–10.0 wt.% at these fO2 conditions (Fig. 5) align closely with recent geophysical models4,15,43. EH chondrites contain metals with 2.99 ± 0.93 wt.% Si (maximum = 5.42 wt.% Si)52. Achieving a core Si content of 8–10 wt.%, therefore, requires SiO2 reduction from the MO towards the core3. This process increases the Mg/Si ratio of the equilibrium MO to values above 1.2–1.5 (Fig. 6). These values are in good agreement with previous estimates of the bulk Mercury Mg/Si ratio (1.14)3. Core compositions with even higher Si contents are possible at fO2 conditions more reduced than IW − 6 to IW − 6.5 (Fig. 5), but this would produce an unrealistically refractory mantle (Fig. 6). This strongly confirms that ~ IW − 6 is the most realistic fO2 for Mercury’s differentiation. Mercury’s silicate Mg/Si ratio. Evolution of Mercury’s magma ocean Mg/Si ratio as a function of the core’s Si content (wt.%). The black and purple curves consider that Mercury accreted from building blocks containing a metal phase with 2.99 (purple) and 5.42 wt.% Si (black). These values correspond to the average and highest Si contents observed in metallic alloys in EH chondrites. The envelopes propagate the errors on Mg/Si ratios of enstatite chondrites83. The pale orange field shows the Si contents of Mercury’s core that are permissible for fO2 conditions sufficiently reduced (< IW − 6) to produce a graphite primary crust matching MESSENGER observational constraints. A previous estimate of Mercury’s mantle Mg/Si ratio3 and the average Mg/Si ratio of Mercury’s crust84 are shown and compared to the Earth’s mantle and oceanic crust. In summary, our study suggests that Mercury differentiated under fO2 conditions around IW − 6. Under these reduced conditions, Mercury’s interior comprised a metallic core with ~ 5000 μg/g C and a MO with ~ 80–150 μg/g C (Fig. 7a), overlain by a primary CO-rich atmosphere contributing a maximum of 150 μg/g to the bulk planet’s C budget. Progressive solidification of the MO led to the formation of a primary graphite crust that was later disturbed by bombardment and volcanism (Fig. 7c). Progressive crystallization of the core, still active today, likely enriched the outer liquid core in C44. Even more accurate estimates of fO2 as well as of geophysical and crustal properties will likely be possible using the physical, chemical, and spectral measurements expected from BepiColombo. Schematic representation of Mercury’s carbon distribution. a The carbon distribution between the magma ocean and the metallic core during accretion of enstatite chondrite (EC) material and primordial differentiation. b Formation of a graphite crust and a primary atmosphere during differentiation and magma ocean crystallization. The mantle mineralogy is based on recent experiments on EH chondrite compositions65. For convenience, only CO is shown as as gas species in the atmosphere, but other species are expected (mainly CH4, H2, and minor CO2 and H2O)14. Note that C becomes enriched in the outer liquid core during crystallization44. c Late-stage mantle melting, volcanism, and impacts reshaping the primary graphite crust and the secondary volcanic crust. Note that Mercury has lost its primary atmosphere due to meteorite impact erosion and the difficulty to retain an atmosphere under the planet’s low gravity. Our results are not only critical for understanding Mercury but are also highly relevant for metal-rich and reduced exoplanets that may have formed from stellar nebulas with high C/O ratios, therefore favoring low fO2 conditions during differentiation and early saturation in graphite and/or diamond53–56. Future observations from space-borne telescopes such as the James Webb Space Telescope will likely confirm graphite as a major phase in reduced exoplanets17,57. Methods Experiments and methods Starting compositions. We used a range of starting compositions in our experiments (Table 1). In some experiments, the starting material (Supplementary Fig. S1) consisted of a natural enstatite chondrite (Hvittis, EL6) powder, to which varying amounts of Si metal (0–50 wt.%) were added to reduce the intrinsic fO2. In other experiments, we employed natural powders from two distinct chondrite groups: the carbonaceous chondrite NWA 11345 (CM2) and the ordinary chondrite Tamdakht (H5). For a separate set of experiments, we used synthetic powders. These starting materials were designed to replicate the silicate fraction of enstatite chondrites58,59 or the crust of Mercury19 and were composed of high-purity oxides, silicates, and phosphates (TiO2, Al2O3, Cr2O3, MnO, MgO, CaSiO3, Na2SiO3, AlPO4). To these, varying proportions of Fe, Si, and FeS were added (Table 1) to yield (S-bearing) compositions with low intrinsic fO2. Multi-anvil experiments. Experiments at 5 GPa and 26 GPa were performed on multi-anvil presses at the Laboratoire Magmas et Volcans (LMV), Clermont-Ferrand, France. Experiments at 26 GPa were performed using a 1500 t multi-anvil press, and the 5 GPa experiments were performed using a 1000 t press. Pressure calibrations were similar to those described previously for the 1000 t press by Hammouda60 and for the 1500 t press by Boujibar et al.61. The pressure assembly consisted of Cr-doped MgO octahedra of 18 mm edge length for 5 GPa experiments and of 10 mm edge length for 26 GPa experiments, compressed between pyrophyllite gaskets by 32 mm tungsten carbide anvils with truncation edge lengths of 11 and 4 mm, respectively. The samples were contained in graphite capsules that transformed to diamond in the 26 GPa experiments. The assembled octahedra were dried in an oven at 500 K overnight before each experiment. The samples were first compressed to the desired pressure at room temperature, with an associated uncertainty of 0.5 GPa, and then heated to the target temperature and held isothermally for durations of 20 min to 4 h. These durations are sufficient to achieve equilibrium between silicate and metallic liquids under these pressure-temperature conditions. A LaCrO3 heater insulated by ZrO2 sleeves produced the high temperatures. Temperature was monitored using W5Re95/W26Re74 type C thermocouples, with no correction for the pressure effect on the electromotive force, yielding an uncertainty of ± 100 K. Samples were quenched at a rate of approximately 800 K/s down to the glass transition temperature by shutting off the power. One experiment was counducted at 6 GPa at the VU Amsterdam using a protocol similar to that described for the 5 GPa experiments. Piston cylinder experiments. Experiments at 1–4 GPa were performed with a piston cylinder. The 1–2 GPa experiments were conducted at the University of Liège (Belgium) and MIT (USA). The cells were made of barium carbonate, and the samples were contained in graphite capsules. The cells were pressurized at room temperature up to 0.75 GPa, then heated to 1138 K. The pressure was then increased to the target pressure after a 5 min dwell. After the final pressure was reached, heating resumed up to the target temperature. We used a type D thermocouple (W3Re97/W25Re75) to monitor temperature. The difference between the measured temperature and the sample temperature was estimated to be around 20 K. The samples were left at the target temperature for 45 min to 3 h, depending on the temperature. Quenching was acheived by shutting off the electrical current. The 3–4 GPa experiments were performed at the Bayerisches Geoinstitut (BGI, University of Bayreuth) using a talc-Pyrex pressure cell and a graphite capsule within an outer platinum capsule. The protocol was similar to that used in Liège and at MIT, except that a W5Re95/W26Re74 type C thermocouple was used. Internally heated pressure vessel experiments. Experiments at 0.1–0.5 GPa were performed using internally heated pressure vessels at the University of Hannover (Germany). We used graphite capsules surrounded by platinum jackets. Thin (0.1 mm diameter) Pt wire was used to place the sample in the hotspot of the furnace. Pressure was increased to 0.1–0.5 GPa using argon as the pressure medium before heating. Two type S (Pt/Pt90Rh10) thermocouples were used to control the temperature at the top and bottom of the cell, and two others were used to monitor the sample temperature. After a few hours at target temperature, the samples were quenched by applying a current to the Pt wire, causing the sample to fall onto a room-temperature, water-cooled copper plate at the bottom of the furnace. Evacuated silica tube experiments. Experiments at 100 kPa were performed at the Center de Recherches Pétrographiques et Géochimiques (CRPG, Nancy, France). The experimental charges comprised graphite crucibles containing the experimental powder, which were placed within 1.5 to 3 cm3 sealed silica tubes. Before the experiments, the silica tubes were sealed on one side, evacuated under vacuum at 10−2 mbar and 120 ∘C during 24 h, then sealed on the other side. Once sealed, the silica tubes were introduced into a muffle furnace at the target temperature (1523–1673 K). Experimental durations were between 5 and 48 h depending on the temperature. During the experiments, the oxygen in the sample reacted with the carbon in the crucible to produce carbon monoxide gas, theoretically buffering the fO2 at C-CO. The experiments were quenched by removing the tube from the furnace. Electron probe micro analysis (EPMA) The major element contents of silicate glasses were analyzed using a CAMECA SX100 electron microprobe analyzer (EMPA) at the University of Hannover (Germany). Analyses were performed with an accelerating voltage of 15 kV, a beam current of 8 nA, and a defocused beam of 10 μm diameter. We used the following standards for Kα X-ray line calibration: albite for Na, orthoclase for K, wollastonite for Si and Ca, TiO2 for Ti, Fe2O3 for Fe, MgO for Mg, Mn3O4 for Mn, Cr2O3 for Cr, and pyrite for S. Raw data were corrected with the CATZAF software. Metals were measured with a JEOL JXA 8530F EPMA at the University of Munster (Germany). We used a beam current of 15 nA and a defocused beam of 2–20 μm diameter. On-peak counting times were 15–20 s (7.5–10 s for background) for each element. Calibration was performed with pure metal Mg, Si, Fe, Cr, Ni, Mn, pyrite for S, wollastonite for Ca, and TiO2 for Ti. Raw data were corrected with the CATZAF software. For carbon measurements, we first removed the carbon coating on the samples by polishing the sample surface with an Al2O3 powder. The samples were then thoroughly cleaned with an ionic cleaner before being coated with iridium. Carbon contamination in the EPMA is a severe issue for C measurements that is time dependent but also apparently sample dependent. To minimize C contamination we used a liquid N2 ‘carbon trap’ near the EPMA stage, which was refilled on a regular basis during analytical sessions. We also waited more than one hour before starting the measurements after sample exchange. Following Dasgupta and Walker62, C was measured for 10 s on-peak and 5 s on-background. Following Zhang et al.63, C was measured with a beam current of 80 nA. We used a voltage of 15 kV, slightly higher than Zhang et al.63, who used 11 kV. The C content due to unavoidable remaining surface C contamination was assessed by measuring two C-free standards: Fe metal (99.995 wt.% Fe) and FeS (36.18 ± 0.15 wt.% S (1σ; Supplementary Fig. S21). We obtained 197 ± 14 cps (I; intensity) C for the Fe metal and 118 ± 22 cps C for the FeS. As previously noted by Zhang et al.63, C-free S-bearing standards have less cps C than pure Fe, which was interpreted as reflecting the greater thermal conductivity of Fe compared to FeS. The I we obtained on C-free standards corresponds to 0.48 ± 0.03 wt.% C for Fe and 0.74 ± 0.08 wt.% C for FeS. Note that fewer cps on FeS translates into higher C concentrations than Fe due to matrix effects during the ZAP correction. The value we obtained for Fe is identical to previous reports (0.5 wt.%62 and 0.47 ± 0.08 wt.%63). The background C on Fe-rich metals was confirmed by measuring a series of standards with various C contents (Supplementary Fig. S21): SUS546, 3.80 wt.% C; SRM13g, 0.61 wt.% C; SRM12h, 0.41 wt.% C; NSC3, 1.49 wt.% C; NIST633, 0.57 wt.% C; Fe7C3, 8.80 wt.% C; Fe, 0.00 wt.% C; ECRM0901, 1.05 wt.% C; BS190, 0.02 wt.% C; BSCSSS4951, 0.81 wt.% C; SRM1227, 0.97 wt.% C; Fe3C, 6.69 wt.% C; and ECRM0562, 0.82 wt.% C. We obtained a linear relation between C contents and cps C at 80 nA with an intercept at 197 cps, identical to the contamination value determined by repeated measurements of the Fe standard (Supplementary Fig. S22). For all standards listed above, the exact location of the C peak was assessed through wavescans of the JEOL spectrometer L unit (mm; Supplementary Fig. S21). We also measured Si (99.995 wt.%) and obtained negative cps C (− 242 ± 8.34) due to the interaction between the positive background of Si and the position of the C peak (127 mm). Due to the generally low S contents of our samples (0–2.5 wt.% S, except samples at 26 GPa that contained 6–15 wt.% S due to closure of the Fe-Si-S miscibility gap), we did not apply any correction for S mass absorption on the effective C concentration (< 5% at < 10% S and < 20% at < 25% S). Following Zhang et al.63, the C concentration of the experimental samples (Cunknown) was determined as: with I being the intensity (cps) and tanθ being the slope of the I vs cps regression (Supplementary Fig. S22). Finally, due to the effect of the thermal conductivity and influence of background slopes on the positive background of Si, we calculated Iblank as: Due to the high reproducibility of our standard measurements, we estimate the precision of our C measurements to be < 0.1 wt.% and the C detection limit to be ~ 350 μg/g (Supplementary Fig. S23). Secondary ionization mass spectrometry (SIMS). SIMS analyses of silicate glasses were performed with a Cameca IMS 1280 at the CRPG. Samples were repolished with Al2O3 powder to remove the previous C coating. Some samples were kept in epoxy, whereas others (experiments performed at LMV) were mounted in an indium substrate to avoid excessive degassing. Samples were cleaned with Milli-Q water and dried in a vacuum oven at 373 K. Samples were then coated with 5 nm thick gold, and stored under vacuum for 24 h to 7 days prior to introduction into the airlock of the instrument. A focused beam of 133Cs+ ions, with a current between 1 and 1.5 nA and an acceleration voltage of 12 kV, was directed onto a spot approximately 10 μm in diameter and rastered across an area of 30 × 30 μm2. Negatively charged secondary ions were then extracted and accelerated to 10 kV into a double-focusing mass spectrometer. In the focal plane of the secondary ion optics, a mechanical aperture limited the analyzed region to the central 15 × 15 μm2 of the rastered area. To fully resolve 1H16O− from 17O−, a mass-resolving power of approximately 6000 was applied, exceeding the required mass resolution of 4700. During each analytical session, calibration curves for H2O and CO2 were established by repeated measurements of the 12C/30Si and 1H16O/30Si ratios in basaltic and andesitic glass standards. Specifically, we used: SCO (3 μg/g CO2, 0.01 wt.% H2O), KL2-G (5 μg/g CO2, 0.02 wt.% H2O), VG-2 (140 μg/g CO2, 0.29 wt.% H2O), M43 (3172 μg/g CO2, 2.62 wt.% H2O), M48 (178 μg/g CO2, 0.77 wt.% H2O), M35 (1019 μg/g CO2, 4.20 wt.% H2O), M40 (2183 μg/g CO2, 3.07 wt.% H2O), M34 (4 μg/g CO2, 5.70 wt.% H2O), t1 (6 μg/g CO2, 0.03 wt.% H2O) and n72 (0 μg/g CO2, 0 wt.% H2O). The CO2 and H2O contents of these standards were previously determined by Fourier transform infrared (FTIR) spectroscopy and Karl Fischer (KF) titration. The associated uncertainties, calculated from calibration curves obtained across four separate sessions, were below 10%. All carbon in the basalt standards was present as carbonates. In contrast, the reduced experiments performed herein contain C dominantly as CH4. When applying the SIMS calibration derived from oxidized standards to these reduced glasses, we assumed that all carbon species in the glass exhibited similar ionization behavior, so that the bulk carbon quantification was not systematically biased21. Each analytical spot was subjected to 360 s of pre-sputtering, after which at least 10 cycles of ion intensity measurements were collected. During each cycle, signals from 12C, 1H16O, 30Si, and 18O were recorded, and the resulting 12C/30Si and 1H16O/30Si ratios were translated into carbon and hydrogen concentrations based on the calibration curves (Supplementary Fig. S24). The stability of the ion signals was closely monitored throughout the measurement cycles because surface contamination can cause a progressive decrease in intensity. Spots exhibiting an intensity drop exceeding 10% were considered compromised and excluded from further analysis30. Raman spectroscopy. We performed Raman spectroscopy on the quenched silicate melts of representative samples to determine carbon speciation. Spectra on samples at 5 GPa and 26 GPa were acquired using a LabRAM HR800 spectrometer (Horiba) at CRPG, Nancy, equipped with a 600 gr mm−1 grating. To minimize fluorescence and optimize the signal-to-noise ratio, a 514.5 nm laser was used at 10% output power (approximately 1.2 mW on the sample surface). Spectra on samples at 0.1 GPa were acquired using a Horiba Jobin LabRAM HR Evolution at the Department of Earth and Environmental Sciences, KU Leuven with a 100 mW (source power) 532 nm laser and a spectral resolution of 0.26 cm−1. For each measurement, 20–30 accumulations of 1–2 s were collected. Data were acquired from 1400 to 4200 cm−1. Data processing, including baseline correction and plotting of spectra, was performed using Spectragryph v1.2.13. Metal-silicate partition coefficients. The metal-silicate partition coefficient of C (DCm/s) was obtained by dividing the mass of C in the metal phase by that in the silicate glass. Both measurements (by EPMA and SIMS, respectively) are associated with an intrinsic analytical error. The absolute uncertainty of the derived value DCm/s was determined using Gaussian error propagation, assuming that the variables are independent and uncorrelated. The 1σ propagated error, σDCm/s, is calculated by summing the squares of the relative uncertainties as follows: where σm and σs represent the standard deviations of the metal and silica concentrations, respectively, and Cm and Cs represent their corresponding mean values. Oxygen fugacity calculations. The oxygen fugacity in most experiments was calculated from6 as: from which we derive: with R being the gas constant, T temperature (K), and a being activities (x ⋅ γ; x = mole fraction, γ = activity coefficient). We computed the ΔG∘ term of the Si-SiO2 equilibrium using the JANAF thermodynamic database. We consider that aSiO2 = 1 in the silicate glass because many experiments contain SiO2 and to facilitate comparison with previous studies6,47,64. Using lower aSiO2 values would slightly decrease the calculated fO2 values (Supplementary Fig. S25)65. In contrast, γSi has a larger effect on fO2 values. We calculated γSi using the interaction parameter formalism (ε approach66), which takes into account the non-ideal interaction between all components in the Fe-rich metal alloy: and γFe and γi are the activity coefficients of Fe and solute i (e.g., Si), respectively, in the molten alloy consisting of N constituents at the temperature of interest. γi0 is the activity coefficient of solute i when it is infinitely dilute in pure liquid Fe. The interaction parameters εji refer to the measured effects of component i on the activity of component j in the alloy, and are dependent on the mole fractions of i and j (xi and xj, respectively). Interaction parameters and activity coefficients were selected from various studies 67 and extrapolated from 1873 K (the reference temperature, T0) to the temperature of the experiment (T) following the procedure of ref. 27: and Calculated fO2 values are then related to the fO2 of the iron-wüstite (IW) buffer68: with μO2 being the oxygen chemical potential. The pressure effect on fO2(IW) is computed after Huebner69. Some experiments contain a metal phase with low Si content (< 0.1 wt.%). For those, EPMA accuracy at low concentrations is low and fO2 was calculated as: where γFe was calculated similarly to γSi. To calculate aFeO, we used a γFeO value of 1.570. For the silica-tube experiments at 100 kPa, oxygen fugacity is intrinsically buffered by the following equilibrium: The fO2 of this equilibrium was calculated as: for which we assume fCO ≈ pCO = 1 (at 100 kPa). For the gases, we calculated: with enthalpy and entropy values calculated according to the Shomate equation as: and For graphite, we calculated the Gibbs free energy as: with the temperature effect on enthalpy and entropy calculated as: and with all coefficients found in Holland and Powell71. Compilation of published experimental results. We compiled metal-saturated experiments conducted in graphite capsules for which carbon saturation was achieved (Supplementary Dataset 1). Experiments that were not carbon-saturated (e.g., those performed in olivine or MgO capsules with added graphite or carbonate powder) were excluded because DCm/s may be slightly higher under graphite-saturated conditions compared to graphite-undersaturated conditions (Supplementary Fig. S26)28. We further restricted our compilation to studies in which carbon concentrations were measured directly by EPMA or SIMS for metal and by SIMS or FTIR for the silicate; we excluded all datasets in which carbon in the metal was inferred indirectly from low EPMA totals. To ensure that equilibrium was attained, we excluded experiments of short duration (< 1 h) when reported (e.g., ref. 72). We also removed experiments or datasets involving nitrogen22,73–75, as well as those reporting quench textures in the silicate glass31 (Supplementary Fig. S3). For all retained experiments, we recalculated fO2 using the methodology described in Eq. (11) (for Si-free metal) or Eq. (5) (for Si-bearing metal). The final dataset spans pressures of 0.1–0.8 GPa, temperatures of 1613–2473 K, and oxygen fugacities from IW − 5.3 to IW+1.521,24,30,31,34,76–81. We also included a few studies in which carbon was accurately measured in the metal phase but not in the silicate phase61. These studies extend our database to 2673 K and 23 GPa but could not be used to calculate DCm/s. Instead, they were only considered for evaluating carbon solubility in metal. Finally, we excluded the C-saturated experiments of Fichtner et al.28 because, even for relatively long run durations (> 1 h), the reported carbon contents in silicate melts vary widely (500–6000 μg/g) under similar conditions (6 GPa, 1873–2073 K, ~ IW − 1.2). Indeed, many of these experiments plot significantly outside the trends defined by both literature data and our experiments relating fO2 to the carbon content of the silicate melt. This may indicate that experiments performed with graphite in Pt capsules take longer to equilibrate than experiments performed directly in graphite capsules. Modeling the thickness of Mercury’s graphite and diamond layers To calculate the thickness of the graphite crust, a series of parameters are needed. (1) A parameterization of DCm/s as a function of ΔIW. (2) A solubility model for C in silicate melt. In our simulations, the calculated C content of the core is significantly below the C content at graphite (or diamond) saturation in the alloy (0.2 to 8 wt.%), meaning that the conjugate MO is not graphite saturated. As such, a solubility function of C in metal is not strictly necessary. An exception was found at IW − 7 to IW − 8, where C solubility in metal is low, forcing graphite saturation in both the core and the MO. Under these conditions, we limited the metal C content to 1200–7000 μg/g at IW − 8 to IW − 7, in agreement with our experimental data and the FactSage thermodynamic model. (3) Relevant physical properties of Mercury are listed in Table 3. We parameterized the evolution of DCm/s as a function of ΔIW using a set of shifted sigmoid functions. We adopted this strategy to constrain the value of DCm/s at the most reducing condition (IW − 8.5). This approach minimizes the influence of outliers at oxygen fugacities where experimental coverage is sparse and analytical uncertainties in the metal are highest due to low C concentrations. We used three individual functions for experiments at < 1 GPa, 5–6 GPa, and 26 GPa to take into account the large effect of pressure on C solubility in silicate melts. For each case, parameters are stored in a library indexed by integers, with the following definitions: L, the maximum value of the sigmoid function; L, the maximum value of the sigmoid function; k, the slope parameter controlling the steepness of the sigmoid; k, the slope parameter controlling the steepness of the sigmoid; x0, the midpoint of the sigmoid; x0, the midpoint of the sigmoid; xshift, a horizontal shift applied to adjust the curve; xshift, a horizontal shift applied to adjust the curve; ytarget, the desired value that the shifted sigmoid passes through at xshift. ytarget, the desired value that the shifted sigmoid passes through at xshift. The shifted sigmoid function is defined as: where the constant C is chosen to satisfy: Solving for C gives: For experiments < 1 GPa, we regressed the following coefficients: L = 4570 ± 2318, k = 1.55 ± 1.07, x0 = − 4.75 ± 0.12, xshift = − 8.5, and ytarget = 20. For the 5–6 GPa experiments, L = 184 ± 51, k = 4.03 ± 0.92, x0 = − 4.77 ± 0.07, xshift = − 8.5, and ytarget = 1. For the 26 GPa experiments, we had too few experiments to meaningfully calculate the error, but we obtained the regressed parameters L = 3.29, k = 118, x0 = − 5.34, xshift = − 8.5, and ytarget = 0.5. Our calculated fits of DCm/s as a function of ΔIW are shown in Supplementary Fig. S11 and compared with the model of Chi et al.30 as used by Keppler and Golabek29 to estimate the graphite behavior on Mercury. Although the C distribution between Mercury’s core and MO is controlled by DCm/s, the carbon content of each reservoir cannot exceed its solubility limit, after which graphite (or diamond) is exsolved. Here we focused on our experiments performed at 5–6 GPa, corresponding to the pressure of Mercury’s CMB and regressed a relation between C in the silicate at graphite saturation and ΔIW using a multiple linear regression algorithm: Due to the low number of experiments at 26 GPa, we regressed a simpler equation at that pressure: Due to the limited number of 26 GPa experiments, we did not compute the errors on the regressed coefficients. With Eqs. (22) to (23) and Table 3, we calculated the thickness t of the graphite crust. We assumed the planet is spherical and that the graphite crust (gr) is uniform in density and completely surrounds the mantle. The mass of a spherical graphite crust with inner radius Ri and outer radius Rm is Rearranging for Ri, we obtain The crustal thickness t is the difference between the outer and inner radii: In these equations. Mgr is calculated as the product of the mass of the silicate shell by the carbon concentration in the silicate shell, with the mass of the silicate shell calculated as: The carbon concentration in the silicate shell is composed of two contributions. First, the carbon initially dissolved in the MO, whose maximum concentration is defined by Eq. (23). Second, the primordial (insoluble) graphite that does not enter the MO or the core. The carbon contents of the MO and core are calculated using Eqs. (23) and (22), and any remaining carbon contributes to the primary shell as primordial graphite. Note that the C content of the core at all fO2 conditions except the most reduced is far (~ 10 times) below graphite (or diamond) saturation but its maximum C content is controlled by the C content of the MO and DCm/s. We used a similar approach to calculate the thickness of a diamond (and Fe3C) layer at the core-mantle interface using the mass of diamond (Mdm) as: where Rj is the radius of a sphere containing Mercury’s core and the diamond shell. To calculate the mass of carbon in Mercury’s primary atmosphere, we used the algorithm MAGWORLD-I14 and Atmodeller35 for fO2 conditions between IW − 4 and IW − 8. We considered a bulk enstatite chondrite composition. The mass of carbon in Mercury’s primary atmosphere MC,atm is related to the mass of the atmosphere Matm and the mass fraction of carbon-bearing gases (mCO, mCH4, mCO2) as: where μ represents the molar mass. The mass fraction of a gas mi is related to its mole fraction Xi as where ∑k sums over the ratios of mass fraction and molar mass of all considered gases in the C-H-O-N-S chemical system. Finally, Dalton’s law implies that the ratio of the partial pressure of a gas to the total pressure is equal to that gas’s mole fraction. For example, the partial pressure of CO pCO is calculated as the product of XCO and the total pressure Ptotal: Sulfur and silicon distribution within Mercury We used a DSm/s parameterization developed for miscible systems50 to calculate the distribution of S between the core and the MO: where logCS is defined as61: Here we parameterized Eq. (33) to fO2 (ΔIW) through XFeO using Eq. (11), in which we calculate aFeO as XFeO ⋅ γFeO using γFeO = 1.5 70. Using Eq. (33) and Mercury’s bulk sulfur content (4 wt.%6), we computed the S content of the core and the MO. To validate this model, we compared our results with a model of S solubility in silicate melt6 and found that the results are very similar under the most reduced conditions, supporting the development of immiscibility below IW − 6 (Supplementary Fig. S27). The development of FeSi(± S)–FeS immiscibility in the Fe-Si-S system was tracked using a parameterization of the solvus based on six pairs of immiscible melts in experiments in the Fe-Si-S system49. When immiscibility is reached, the FeSi(± S) melts follow the experimental solvus in equilibrium with stoichiometric FeS. The evolution of Si in Mercury’s core as a function of fO2 was calculated using Eq. (5), following the procedure described above to determine γSi as a function of the Fe content in the metal. Thermodynamic simulations The equilibrium thermodynamic calculations of graphite saturation in the Fe-Si-S-O system were performed using the Gibbs free energy minimization Equilib module of the FactSage Thermochemical Software. The outputs of these calculations are shown in Fig. 2. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Code 1 Transparent Peer Review file Author contributions O.N. conceptualized and designed the project. O.N., B.C., C.C., T.H. and M.B. performed the experiments. O.N., B.C., C.C., J.V. and J.B. conducted the analyses. O.N. performed the calculations with help from N.T., M.M. and O.S. O.N. wrote the paper. C.C. prepared the last figure of the manuscript. All authors discussed the results and contributed to the editing of the paper. Peer review Peer review information Nature Communications thanks Fabrice Gaillard and Francesca Miozzi for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by an European Research Council (ERC) grant under the European Union’s Horizon (grant agreement no. 101125126; IronHeart) to O.N. SIMS analyses were funded by the Europlanet program to O.N (grant 17-EPN3-059). T.H. and M.B. acknowledge support from the Program National de Planétologie (CNRS-CNES). C.C. acknowledges financial support from the Program National de Planétologie (PNP) of the Institut National des Sciences de l’Univers (INSU), CNRS. Data availability The experimental data and compositions, including those of the literature, are provided in Supplementary Data 1. Supplementary Figs. and text are displayed in the Supplementary Information. Code availability The Python scripts used for all calculations are submitted as Supplementary Code 1. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-75458-y. References Associated Data Data Availability Statement The experimental data and compositions, including those of the literature, are provided in Supplementary Data 1. Supplementary Figs. and text are displayed in the Supplementary Information. The Python scripts used for all calculations are submitted as Supplementary Code 1.