Sulfur partitioning between aqueous fluids and felsic melts at high pressures: Implications for sulfur migration in subduction zones

Estimation of S loss to inner Au capsules

Gold is commonly used as a capsule material in S partitioning experiments due to its effectiveness in preventing S loss^16,19. To assess the potential loss of S to the inner Au capsules and its impact on the accuracy of the calculated partition coefficients, we analyzed the S content in the inner wall of Au capsules. The compositions of the Au capsules in the Set-1 runs are detailed in Supplementary Table S5. The results indicate that the S concentrations in the Au capsules range from approximately 0 to 100 ppm in most experiments, suggesting minimal S loss to the capsule. However, in the case of runs DN918 and DR928, the inner walls of Au capsules showed elevated S concentrations of around 150 ± 30–310 ± 140 ppm. This increased S content in the capsules of these two runs raises the possibility of an overestimation of the partition coefficients. Given that the inner Au capsule mass is approximately 13 times that of the starting material, we corrected the partition coefficients for these two runs by subtracting the capsule S mass from the initial total S content to determine the true S content in the aqueous fluid. This adjustment yielded corrected partition coefficients of approximately 75 and 2 for DN918 and DR928, respectively (Fig. 3).

Dependency of the fluid-melt partition coefficient of S on (a) fO₂ and melt composition, (b) NaCl content and (c) pressure at 950 ℃. The D_S^fluid/melt decreases with increasing fO₂, NBO/T, addition of NaCl and pressure. NBO/T values are indicated near the symbols. In panels (a) and (c), filled symbols represent D_S^fluid/melt calculated without accounting for S loss to the capsule, while open symbols represent D_S^fluid/melt calculated after correcting for S loss to the capsule. Neglecting this loss can lead to an overestimation of the D_S^fluid/melt. (d) Comparison between the results in this study and previous studies. Literature data were obtained from previous studies^{6,7,13,14,15,16,19,22,28,41,49,50}: S1998 (899 ℃, 0.23 GPa, with white dacite as starting material), B2004 (850 ℃, 0.2 GPa, rhyodacite), S&M2006 (852–858 ℃, 0.15 GPa, peralkaline rhyolite), W2009 (904–1000 ℃, 0.2 GPa, phonolite), Z2013 (1000 ℃, 0.2 GPa, rhyolite), K2010 (850 ℃, 0.05–0.3 GPa, haplogranite), W2011 (900 ℃, 0.2 GPa, haplogranite), Z2012 (1000 ℃, 0.2 GPa, andesite), J&D2013 (850–1000 ℃, 2 GPa, MORB), J&D2014 (950 ℃, 2–3 GPa, MORB), M2016 (900 ℃, 0.2 GPa, dacite), B2018 (850 ℃, 0.2 GPa, haplogranite), D2018 (950–1000 ℃, 3 GPa, pelite melt), X2022 (950 ℃, 1 GPa, dacite). It should be noted that although MORB was used as the starting material in J&D2013 and J&D2014, the produced silicate melts were felsic due to significant crystallization.

Assessment of chemical equilibrium and fO
₂

To ensure the reliability of our experimental results, it is essential to assess the attainment of chemical equilibrium and control of fO₂ during the experiments. Previous studies have established that S partitioning equilibrium between the aqueous fluid phase and silicate melt can be achieved within 48 h at temperatures ranging from 900 to 950 ℃ and 72 h at 800 ℃^4,41. In our experiments, the run duration of 48–72 h at 950 ℃ is considered sufficient to ensure chemical equilibrium. This conclusion is supported by the observed homogeneity of S contents and major elements in the quenched glasses (Tables 1, 2). Additionally, based on previous research on S diffusion in dacitic melt³⁹, we calculated average diffusion coefficients of 28 μm²/s for S²⁻ and 1.6 μm²/s for S⁶⁺ at 950 ℃. These values indicate sufficiently rapid diffusion to facilitate equilibration within the capsule during the experimental run duration. This suggests that the equilibrium for S in both oxidizing and reducing conditions had been attained within the initial few hours. Therefore, we can conclude that chemical equilibrium was achieved in our experiments.

The fO₂ conditions during the experiments were controlled using external fO₂ buffers (CCO, NNO, and RRO), and the fO₂ of the sample (expressed as ΔNNO) was determined using two different methods (Table 1). In runs where both the metal and oxide in the fO₂ buffers coexist after the experiment, the fO₂ of the sample generally matches to the theoretical fO₂ value of the buffer. This is because of the presence of 30–50 wt% H₂O loaded into the sample, which ensures the activity of H₂O in the sample is close to unity or slightly lower. On the other hand, in the two runs (K-32 and XUD-27) that saturated with pyrrhotite and magnetite, the fO₂ can be calculated using the pyrrhotite-magnetite oxybarometer^42,43, resulting in values slightly lower than the fO₂ values of RRO buffer. This discrepancy result from the influence of other volatiles in the fluids, such as H₂S and SO₂. If the fluids are not pure H₂O, the samples will be more reduced than the fO₂ buffer³¹.

Effect of fO
₂
on S partitioning

To investigate the effect of fO₂ on S partitioning, six experiments were conducted using the K-feldspar and dacitic glasses at 950 ℃ and 1 GPa under varying fO₂ conditions with CCO, NNO, and RRO buffers (Table 1). Figure 3 illustrates that as fO₂ increases from NNO-1.4 (oxygen fugacity 1.4 log10 units below the Ni–NiO buffer) to NNO + 2 the D_S^fluid/melt decreases from 391 ± 79 to 48 ± 3 in the K-feldspar system and from 147 ± 40 to 27 ± 1 in the dacite system. This decrease in partition coefficients can be attributed to the transformation of S species in the aqueous fluid from H₂S under reducing conditions to SO₂ and SO₄^2 − under oxidizing conditions⁴⁴, as supported by the Raman spectroscopy of fluid inclusions (Fig. 2). The observed correlation between fO₂ and S partitioning is consistent with previous studies at lower pressures (≤ 0.3 GPa)^{13,16,18,19,20,28,33}. The obtained D_S^fluid/melt values in the K-feldspar system, ranging from 268 ± 37 (NNO) to 48 ± 3 (RRO), are generally in agreement with the values reported for the haplogranite system at 850 ℃ and 0.2 GPa from previous study¹⁹. In their study, the D_S^fluid/melt ranged from 323 ± 14 (NNO) to 74 ± 5 (RRO). It is noteworthy that in the dacite system, the D_S^fluid/melt of 131 ± 32 in the CCO-buffered run (DC918) is slightly lower than 147 ± 40 obtained from the NNO-buffered run (DN918). This may be due to a small amount of S alloying with the inner Au capsule in run DN918, leading to an overestimation of the D_S^fluid/melt (Supplementary Table S5). As mentioned above, the D_S^fluid/melt value in run DN918 is 75 after correction, consistent with the observed correlation of fO₂ and D_S^fluid/melt. Overall, the results highlight the significant influence of fO₂ on S partitioning, with higher fO₂ conditions favoring the incorporation of S into the silicate melt and resulting in lower partition coefficients. The observed trends are consistent with the transformation of S species in the aqueous fluid under varying redox conditions.

Effect of composition of silicate melt and aqueous fluid on S partitioning

Besides fO₂, the composition of the silicate melt is another important factor influencing S partitioning¹⁹. The silicate melts in the K-feldspar system have a similar A/CNK (Al/ (Ca + Na + K)) ratio and significantly lower NBO/T (non-bridging oxygen per tetrahedron, a measure of the degree of polymerization of the silicate melt structure) compared to those in the dacite system. Notably, in run KR916 with the RRO buffer, the quenched glass exhibits significantly lower K₂O content and higher A/CNK ratio compared to the other two K-feldspar experiments. This can be explained by the fact that alkalis more readily enter the aqueous fluid under oxidizing conditions than under reducing conditions⁴⁵. Under similar fO₂ conditions (Table 1 and Fig. 3a) and at identical temperature and pressure, the K-feldspar system (NBO/T = 0, completely polymerized) exhibits higher D_S^fluid/melt values (391 ± 79–48 ± 3) compared to the dacite system (with D_S^fluid/melt = 147 ± 40–27 ± 1 and NBO/T = 0.102–0.129). This could be attributed to the fact that silicate melts with a high degree of polymerization (small NBO/T) tend to have lower S content compared to depolymerized melts characterized by elevated NBO/T¹⁴. This observation is in agreement with previous studies at lower pressures^15,19.

To investigate the effect of Cl on the partitioning of S under oxidizing condition (RRO), three experiments were conducted at 950 ℃ and 1 GPa in the dacite system. The results reveal that the addition of NaCl under oxidizing condition significantly increases the S content in the silicate melt, resulting in a decrease in D_S^fluid/melt (Table 1 and Fig. 3b). Specifically, the S content in the silicate melt increases from 324 ppm without NaCl to 1225–2071 ppm with the addition of NaCl, accompanied by a decrease in D_S^fluid/melt from 27 ± 1 to 3 ± 0.04 (ωNaCl = 10 wt%) and 5 ± 0.3 (ωNaCl = 20 wt%). This observation contradicts previous studies that indicated its independence from NaCl content under oxidizing condition (RRO), but it is consistent with the decrease in D_S^fluid/melt upon the addition of NaCl under reducing condition (NNO)^18,19. The solubility of S in the silicate melt depends on various factors, including FeO_T and CaO in the silicate melt^12,22,46. The nearly constant FeO_T and CaO content among the three experiments implies that the increase in S concentration in the silicate melt is independent of the melt composition (Table 2). Previous research showed that adding 10 wt% NaCl only slightly increases S solubility in fluid-undersaturated hydrous K-feldspar melt under both NNO and RRO buffers at 950 °C and 1 GPa⁴⁷. The higher S content in Cl-bearing runs compared to Cl-free runs may be owing to the non-ideality mixing behavior in the aqueous fluid phase¹⁸. Although the exact mechanisms underlying this effect are not fully understood, it is possible that the presence of NaCl modifies the speciation of S in the aqueous fluid, leading to a higher proportion of less volatile S species.

However, an interesting observation is that the S content in the silicate melt initially increases and then decreases with increasing Cl content, which has also been observed in previous studies^18,29. One possible explanation is that a significant amount of NaCl can alter the aqueous fluid equilibrium under RRO conditions, causing a shift from SO₂ towards SO₃ and related derivatives such as H₂SO₄, HSO₄⁻, and NaSO₄⁻. This can result in a substantial amount of S partitioning into the aqueous fluid phase and potentially forming dilute sulfuric acid, leading to a significant decrease in S content in the silicate melt and an increase in the partition coefficient. A study by Ni and Keppler⁴⁸ confirmed that SO₂, HSO₄^– and H₂SO₄ are the dominant S species in oxidized hydrothermal fluids, supporting this hypothesis. Furthermore, the formation of Na- and S-bearing species in the melt could be a plausible S-favoring mechanism. This mechanism might compete with inhibitory processes, such as potential competition between S and Cl for anion sites in the melt. The interplay of these mechanisms could explain the observed patterns in S behavior. To fully elucidate the role of Cl on S partitioning at high pressure, further experimental and theoretical investigations are required.

Effect of pressure on S partitioning

The effect of pressure on S partitioning has been regarded as less significant compared to factors such as fO₂ and melt composition at crustal pressure^{6,16,19,49,50}. To investigate the effect of pressure, we conducted experiments at 1 GPa and 2 GPa on the dacite system, representing lower crust and upper mantle conditions. Our results revealed a consistent trend of decreasing D_S^fluid/melt with increasing pressure under both reducing (NNO) and oxidizing (RRO) conditions (Fig. 3c). In NNO buffered experiments, D_S^fluid/melt decreased from 147 ± 40 at 1 GPa to 20 ± 2 at 2 GPa. Similarly, in RRO buffered experiments, D_S^fluid/melt ranged from 27 ± 1 at 1 GPa to 20 ± 2 at 2 GPa. The observed decrease in D_S^fluid/melt at higher pressures suggests that pressure does play a role in S partitioning, albeit a secondary one compared to other factors. The smaller decrease in D_S^fluid/melt in the RRO runs at 2 GPa may be due to S loss to the Au capsule or the influence of the NBO/T value. After correcting for S loss to the capsule, the D_S^fluid/melt value in the RRO experiments at 2 GPa is approximately 2. Furthermore, at high pressure, the increased dissolution of silicate components into the aqueous fluid can cause variations in both the NBO/T and A/CNK ratios of the silicate melt. Additionally, alkalis more readily enter the aqueous fluid under oxidizing conditions compared to reducing conditions⁴⁵, resulting in a lower NBO/T and higher A/CNK of the melt.

To eliminate the effect of melt composition or temperature on S partitioning, we compared our experimental results with previous studies that employed similar starting materials or were conducted at similar temperature (Fig. 3d). Generally, the S partition coefficients obtained at 1 GPa and 2 GPa in this study are lower than those from experiments at lower pressures (≤ 0.3 GPa)^16,19. However, Masotta, et al.¹⁵ obtained a D_S^fluid/melt value as low as 1.5 ± 0.3 between aqueous fluid and trachy-andesitic melt with RRO buffer at 950 °C and 0.2 GPa. The discrepancy cannot be explained by differences in melt composition or pressure, as the quenched glasses in both our experiments and theirs have similar NBO/T and A/CNK values. Furthermore, our S partition coefficients do not align with the D_S^fluid/melt values calculated using the regression equation derived from their lower pressure experiments¹⁵, as shown in Supplementary Table S2.

Although run products from previous experiments conducted at 850–1000 ℃ and 1–3 GPa contained S-bearing minerals such as pyrrhotite and anhydrite, as illustrated in Fig. 3, our results fall within the range of D_S^fluid/melt reported in previous studies at high pressures^4,6,7,22. For example, our dacite runs employed experimental conditions similar to a recent study²², where experiments were conducted with dacite + FeS + H₂O at 1 GPa and 950 ℃. Our results are comparable to the partition coefficients obtained with a lower H₂O content in the silicate melt (< 10 wt%) but higher than those obtained with a higher H₂O content in the silicate melt (> 10 wt%) in their study (Fig. 3d). The discrepancy may be due to sulfide or sulfate being undersaturated in our run products. Their run products contained high abundances of pyrrhotite and anhydrite (7–12 wt%), which sequestered most of the S, introducing great uncertainties in the determination of partition coefficients through mass balance calculations. Another possible explanation is that the presence of pyrrhotite and anhydrite might affect the concentration of FeO and CaO in the silicate melt and aqueous fluid phase, resulting in variations in D_S^fluid/melt13.

Solubility of S in silicate melt

Three experiments were conducted to investigate the effect of silicate melt composition on S solubility at 2 GPa and 950 ℃ under RRO buffer conditions (Set-2 runs in Table 1). Additionally, to investigate the impact of pressure, we compared these S solubility results at 2 GPa with those from published experiments at 1 GPa, which employed similar starting materials under comparable conditions except for pressure^20,40. Recent studies have demonstrated that pyrrhotite and anhydrite coexist in the silicate melt at 1 GPa and 950 ℃ under RRO buffer conditions^22,51. However, in our experiments, except for run XUD-27, which contained 1.2 wt% pyrrhotite and trace amounts of anhydrite, the run products were saturated with either sulfide or anhydrite. This is attributed to the absence of CaO in run K-32, preventing the crystallization of anhydrite, or the too-low FeO content to saturate sulfide in run DR926. In the latter case, no FeS was added, and S was introduced into the starting material using dilute sulfuric acid.

In the K-feldspar system, run K-32 produced pyrrhotite and a S concentration of 11818 ± 444 ppm in the silicate melt. The high S solubility can be attributed to the absence of CaO in the silicate melt, as previous studies have demonstrated an inverse relationship between S and CaO concentrations in CaO-poor melts^46,52,53. The S solubility in K-32 exceeds that reported in a previous study conducted under similar conditions⁴⁷, where S concentrations in K-feldspar melts ranged from 1242 to 1593 ppm in runs with 10 wt% H₂O added and up to 4357 ppm in runs with 20 wt% H₂O added at 1 GPa and 950 ℃ under RRO buffer conditions. The higher S solubility in this study can be attributed to the high H₂O content in the silicate melt. It is important to note that the H₂O content in the silicate melt was controlled by the amount of pre-loaded H₂O and its solubility, which is pressure-dependent. Run K-32, with a pre-loaded H₂O content of 30 wt% and an experimental pressure of 2 GPa, exhibited higher H₂O and S solubility compared to the previous study⁴⁷.

In comparison to the K-feldspar system, the experiments with dacitic glass (run XUD-27 and DR926) showed lower S solubility, ranging from 6428 ± 260 to 8699 ± 1000 ppm. Despite the fO₂ in run DR926 being 0.4 log units higher than that in XUD-27, indicating a higher S⁶⁺/S ratio, the S solubility in run DR926 was lower. This discrepancy may be attributed to the lower FeO content in the silicate melt, consistent with the important role that melt composition plays in melt solubility^54,55,56,57. This observation aligns with a previous study²² indicating that S solubility in dacitic melt under intermediate fO₂ conditions is influenced by the H₂O, FeO, and CaO content in the silicate melt. They found that S solubility in dacitic melt increases with the addition of H₂O, ranging from 2013 ± 61 ppm with 10 wt% H₂O to 3372 ± 120 ppm with 15 wt% H₂O, and to 7997 ± 426 ppm with 20 wt% H₂O at 1 GPa and 950 ℃ with similar fO₂²². The S solubility from dacite experiments at 2 GPa in this study is consistent with the experiment with 20 wt% H₂O at 1 GPa in theirs (Table 1). However, the effects of melt composition and H₂O content cannot be clearly separated. Although run XUD-27 was conducted at a higher pressure than run D-1, resulting in a higher H₂O content in the silicate melt, slight differences in CaO content (2.93 ± 0.12 wt% vs. 1.99 ± 0.18 wt%) between these two runs suggest that the variation in S solubility cannot be solely attributed to pressure and the corresponding H₂O content in the silicate melt. Nevertheless, this study demonstrates that highly hydrated magma melts at high pressure up to 2 GPa can dissolve a significant amount of S.

Sulfur mobilization capacity of fluid-saturated felsic magma

Recent investigations have established that achieving maximum S dissolution capacity in fluid-saturated felsic melts requires a high H₂O content under intermediate fO₂ conditions, such as RRO-buffered fO₂²². The transportability of S in fluid-saturated evolved magma can be estimated by considering S solubility in the felsic melt, S partition coefficients, and the mass ratio of aqueous fluid to silicate melt, assuming a constant H₂O content in the parental arc basalt. Combining previous results of S solubility in dacitic melts at 1 GPa and 950 ℃ with the S partition coefficients determined in this study, a preliminary model was developed to estimate the total S content (S_tot) transportable by fluid-saturated dacitic magma under lower crust and upper mantle conditions. The maximum S content in the silicate melt (S_melt) is determined by its solubility, which is controlled by the fO₂ and melt H₂O content (H₂O_melt) at given temperature, pressure, and melt composition. Solubility values were derived by fitting empirical data from previous studies^22,58 (Supplementary Fig. S1). The S content in the fluid under the corresponding redox condition is obtained by multiplying S_melt by D_S^fluid/melt. The S_tot is then determined based on the total H₂O content in the evolved magma and the mass ratio of fluid to melt. Additional information can be found in Supplementary Table S6. These calculations represent a simplified model designed to explore the total possible S load at fixed pressure, temperature, and melt composition, solely as a function of the distribution of 15 wt% H₂O between aqueous fluid and silicate melt. The H₂O distribution is set arbitrarily and may reflect various factors not explicitly considered in this simplified approach. This model is not intended to represent the complexities of S transport within a dynamic magmatic system, but rather to illustrate the potential impacts of fO₂ and H₂O distribution on S partitioning and total S content in a closed system.

In our model, the maximum H₂O content in the silicate melt is limited to 15 wt% because the solubility of pure H₂O in silicate melt is approximately 15 wt% at 1 GPa²². The addition of other volatiles, such as H₂S and SO₂, reduces the H₂O solubility, leading to fluid saturation with lower H₂O content in the silicate melt. Accordingly, we calculated the S_tot assuming an evolving magma containing 15 wt% H₂O, with 10–15 wt% H₂O retained in the silicate melt and the remaining 0–5 wt% H₂O present as fluid. The modeling results indicate that the total S content dissolved by residual melt and exsolved fluids increases with rising fO₂, even though the D_S^fluid/melt decreases with increasing fO₂ (Fig. 4). Notably, under RRO conditions, fluid-saturated magma generally dissolves less S than a hydrous melt with 15 wt% H₂O and without fluid saturation. Although fluids can contain substantially more S than the coexisting silicate melt (D_S^fluid/melt > 1), this does not necessarily result in greater total S transport. For a given total H₂O content, fluid-saturated magma typically dissolves more S than fluid-unsaturated magma (with some exceptions), especially under reduced conditions where S solubility is lower but D_S^fluid/melt is higher. For example, if an evolving magma contains 15 wt% H₂O, the silicate melt can dissolve approximately 8000 ppm S under RRO conditions at 1 GPa and 950 ℃, assuming all H₂O is retained in the silicate melt without fluid saturation (dashed blue line in Fig. 4). However, if fluid saturation occurs with 5 wt% H₂O as fluid and 10 wt% H₂O retained in the silicate melt, the mass ratio of melt to fluid in the magma system would be 94.44 vs. 5.56 wt%. In this case, the silicate melt can dissolve 2557 ppm S (Supplementary Table S6), while the aqueous fluid can dissolve 69,039 ppm S (D_S^fluid/melt = 27). The total S content dissolved in the silicate melt plus fluid would then be 6253 ppm (2557 * 94.44% + 69,039 * 5.56% = 6253 ppm), which is lower than in the case without fluid saturation. This result occurs because S solubility increases significantly with the H₂O content of the silicate melt, and the D_S^fluid/melt value is not high enough to compensate for this increase. In contrast, under NNO conditions, the silicate melt can dissolve 370 ppm S (Supplementary Table S6), and the aqueous fluid can dissolve up to 54,442 ppm S (D_S^fluid/melt = 147). The maximum total S dissolved in the silicate melt plus fluid would then be 3374 ppm S, which is much higher than in the case without fluid saturation (dashed orange line in Fig. 4). Therefore, under reduced conditions, where S solubility is low, fluid saturation is more critical.

Modeled total dissolved sulfur content in fluid-saturated dacitic magma at 950 ℃ and 1 GPa, under varying redox conditions (CCO, NNO, RRO) and a fixed magma H₂O content of 15 wt%. Solid lines depict total S content that can be dissolved in a system with coexisting silicate melt (10–15 wt% H₂O) and aqueous fluid (0–5 wt% H₂O). Dashed lines represent S solubility in purely hydrous melt (15 wt% H₂O). The figure shows that under reduced conditions (CCO and NNO), the total S dissolved in melt plus fluid is significantly higher compared to hydrous melt alone. Conversely, under oxidized conditions (RRO), fluid-saturated magma generally dissolves less S than fluid-undersaturated magma. The addition of NaCl decreases the partitioning coefficient (D_S^fluid/melt), reducing S dissolution and transport capacity compared to Cl-free magma, with or without fluid saturation. See Supplementary Table S6 for detailed data.

In subduction zones, parental arc basalts typically contain 2–6 wt% H₂O⁵⁹, and these hydrous basalts undergo significant crystallization within the lower crust, producing highly hydrated felsic magmas and fluids^22,59,60. An evolving magma that contains 15 wt% H₂O can be readily form through 70% amphibole crystallization of a parental basalt containing approximately 5 wt% H₂O. Primitive arc basalts generally have S concentrations ranging from 900 to 2500 ppm^61,62,63,64. Given the incompatibility of S in silicate minerals and oxides, the S concentration in evolved magma could reach 3000 to 8300 ppm, if not limited by saturation in an S-bearing phase such as pyrrhotite, anhydrite, or fluid. As shown in Fig. 4, for a given H₂O content of 15 wt%, both silicate melt and fluid-saturated magma under RRO conditions can effectively dissolve and potentially transport all the S in the evolved arc magma under lower crustal pressure conditions. In contrast, under reduced condition (CCO and NNO), sulfide saturation is evitable, regardless of whether fluid saturation is present.

While the addition of Cl can indeed influence S speciation in the fluid phase, leading to a decrease in D_S^fluid/melt, it may also enhance metal transport, particularly for Cu and Au, due to potentially high fluid-melt partition coefficients if the D values observed at lower pressures⁶⁵ are applicable to lower crustal conditions. As shown in Fig. 4, the addition of Cl (D_S^fluid/melt = 5 ± 0.3 with ωNaCl = 20 wt%) significantly reduces the ability of fluid-saturated magma to transport S under RRO conditions, resulting in lower transport efficiency compared to Cl-free, fluid-saturated magma and fluid-free, but highly hydrous silicate melt. From this perspective, the addition of Cl does not favor sulfide breakdown or S transport under RRO conditions. However, the formation of Cl complexes can increase the solubility of Cu and Au in the fluid phase, potentially facilitating their transport to shallower depths. In conclusion, the interplay of fO₂, H₂O content, and Cl content in magma and fluid saturation determines the fate of S and metals in lower crustal magma reservoirs, necessitating further investigation.

Our results emphasize that S exhibits a strong affinity for the aqueous fluid phase under upper mantle and lower crust pressures at various redox conditions, even though the S partition coefficient decreases at elevated pressure. The presence of Cl in the aqueous fluid reduces the S partition coefficient, diminishing the capacity of fluid-saturated magma to effectively transport S. Highly hydrous felsic magmas saturated with a significant amount of aqueous fluid provide an efficient mechanism for dissolving sulfide and transporting S from deep to shallow regions. We conclude that highly hydrous magma, particularly under moderately oxidized conditions (~ NNO + 2), could play a crucial role in the formation of giant porphyry ore deposits and act as a deep source for the excess S released during explosive volcanic eruptions.