Magnetite-apatite ores record widespread involvement of molten salts

Kiruna-type magnetite-apatite rocks are important sources of iron, phosphorus, rare earth elements (REEs), and other critical elements. They are also highly contentious in terms of genesis. Some models invoke hydrothermal processes, such as precipitation from hot water or metasomatic replacement (Sullivan et al., 2023). Others argue for magmatic origins such as crystallization from Fe-rich silicate or oxide-phosphate melts (Velasco et al., 2016). Yet others argue for hybrid processes (Reich et al., 2022). At its core, the debate boils down to a basic question: what types of fluids form these enigmatic rocks?

We examined the fluids involved in forming magnetite-apatite rocks through an extensive survey of fluid and melt inclusions from 20 deposits worldwide (Fig. 1; Table 1). The study localities include deposits from different geologic settings and with ages from ca. 2.5 Ga to ca. 2 Ma (Fig. 1; see the Supplemental Material¹ for geologic details and references for each locality). We report detailed evidence that formation of every studied deposit involved molten salts composed of sulfate, chloride, and carbonate components. These fluids are radically different from those that have been previously invoked, but our results show that salt melts are ubiquitous in magnetite-apatite rocks.

Polycrystalline melt inclusions are abundant in the primary minerals of all deposits studied (Figs. 1 and 2; Table 1; see the Supplemental Material). The inclusions are mostly or entirely filled with polymineralic crystals, commonly together with deformed vapor bubbles that occupy up to 40% of the inclusion volume. Aqueous liquid is seldom observed and occupies <1% of the inclusion volume. Within any given assemblage, properties of the inclusions are consistent in terms of mineral phases, volume fractions, and microthermometry. Hence, the inclusions show no evidence of post-entrapment modification such as water loss.

The compositions of the inclusions vary between deposits but are consistent within each deposit, including between different host minerals (Table 1). At room temperature, the inclusions always contain crystalline salts—primarily alkali-calcic sulfate, chloride, and carbonate minerals, in relative proportions that show a modest correlation to the lithologies of the surrounding wall rocks (Figs. 1 and 2; Table 1). Most deposits show significant proportions of two or three salt types (Fig. 1; Table 1), suggesting a continuum of compositional types. Silicate minerals, especially calc-silicates like those found in skarns (Xu et al., 2023), are common in the inclusions (Figs. 1 and 2; Table 1). Inclusions from most deposits also contain significant amounts of iron ± titanium oxides (i.e., hematite, magnetite, and ilmenite; Fig. 2B), even up to 80 vol% in some cases. Phosphate minerals, especially monazite, are present in inclusions from half of the studied deposits (Fig. 2D). Most of the minerals in the inclusions are anhydrous (Table 1), implying that the trapped melts were water-poor (<1 wt% H₂O).

Anhydrite is the most common sulfate in the inclusions, along with variable amounts of baryte, cesanite, gypsum, glauberite, and celestine (Table 1; Fig. 2; Table S2 in the Supplemental Material). Chloride salts include sylvite, halite, hibbingite, and rinneite (Table 1; Fig. 2E; Table S2). Inclusions rich in chlorides differ starkly from aqueous brine inclusions in that they contain little or no liquid water. Carbonates, especially calcite and dolomite (along with lesser magnesite, natrite, and trona; Table 1; Table S2), are less common in the inclusions compared to sulfates and chlorides, but sometimes dominate (Figs. 1 and 2). Notably, where carbonates are hosted in inclusions in silicate minerals, they are always accompanied by calc-silicates, suggesting chemical exchange between the melt and the host mineral.

Silicate phases in the inclusions are primarily calc-silicates such as andradite and diopside. Calc-silicates even dominate in some deposits, but are always accompanied by sulfates, chlorides, and/or carbonates (Fig. 1). Silicate minerals that make up the metasomatic haloes surrounding the orebodies—albite, scapolite, epidote, and actinolite—are common in the inclusions (Table 1). Quartz, uncommon in most magnetite-apatite rocks, sometimes appears in the inclusions but is subordinate to carbonates (Fig. S21A), indicating silica-undersaturated and calc-silicate normative bulk compositions.

Heating experiments (detailed in the Supplemental Material) reveal that the polycrystalline inclusions begin to re-melt at temperature (T) as low as 285 °C and are fully molten at 615 °C to >1100 °C (Table 1; Fig. 3). Chloride-rich inclusions generally show the lowest first-melting T (T_f), as low as 285 °C. Carbonates and sulfates melt mostly in the range of 650–900 °C, consistent with eutectic relationships of multicomponent systems. Oxides, phosphates, and calc-silicates show the highest melting T > 1000 °C (Figs. 3B and 3C). Immiscible separation between calc-silicate–rich and salt-rich melt is common in the inclusions at high T, especially between 900 °C and 1100 °C for inclusions rich in chlorides and >1100 °C for the inclusions rich in calc-silicates. In some cases, the inclusions decrepitated prior to complete melting, suggesting high internal pressures. None of the inclusions could be quenched to glass even at cooling rates of >200 °C/s, suggesting that these are low-viscosity ionic liquids.

The occurrence of aqueous liquid or vapor inclusions in the studied deposits is highly variable (Table 1), and samples from some deposits host no aqueous fluid inclusions at all. When present, aqueous inclusions are sometimes only vapor-rich, in other cases vapor-rich plus brines, and in yet other cases only brines (Table 1; see details in the Supplemental Material).

The core implication of these results is that molten salts are ubiquitous in magnetite-apatite deposits. In contrast to recent suggestions that such melts are unusual, local phenomena (Reich et al., 2022), we find evidence for their involvement worldwide, from Archean through to recent times, and at different tectonic settings and formation depths. Similar melt inclusions also occur in other magnetite- and apatite-rich rocks (Table 1), including phoscorite (magnetite-apatite-olivine) in carbonatite pipes (Palmer, 1998; Solovova et al., 1998) and ultramafic alkaline complexes (Veksler et al., 1998; Nikolenko et al., 2020); magnetite-apatite dikes in alkalic porphyries (Kamenetsky et al., 1999); nelsonite (ilmenite-magnetite-apatite) in anorthosite complexes (Frost and Touret, 1989); and diopside-titanite-magnetite dikes associated with iron oxide veins (Bakker and Elburg, 2006). These rocks represent diverse geologic settings, but each shares a similar mineral assemblage and hosts abundant salt melt inclusions (Table 1). The most straightforward conclusion from these data is that molten salts are integral to the formation of magnetite- and apatite-rich rocks wherever they occur.

In light of the evidence for molten salts, the next obvious questions are: where do they come from, and what role do they play in ore formation? Previous studies have invoked saline aqueous fluids of either evaporitic origin (Barton and Johnson, 1996; Li et al., 2015; Yan and Liu, 2022) or magmatic-hydrothermal origin (Knipping et al., 2015; Hu et al., 2020). We acknowledge the potential involvement of such fluids, but we stress that the essentially dry salt melts reported here are quite distinct from both. Inclusions similar to those described here were previously interpreted as “hydrous saline melt” formed by phase separation of a magmatic-hydrothermal fluid (Broman et al., 1999), but the absence of water in the inclusions, along with the abundance of silicate, oxide, and phosphate minerals within, suggest a non-hydrothermal origin. The great formation depth of some studied deposits (Table S1) further discounts low-pressure condensation from a hydrothermal fluid. In our view, the most likely principal sources of these salt melts are hiding in plain sight, arising from interaction between mantle-derived silicate magmas and surface-derived chemical sedimentary rocks.

The most straightforward process to generate carbonate-, sulfate- and chloride-rich melts is by melting of limestones and evaporites. Many magnetite-apatite deposits are emplaced into limestones (Fig. 1), and magnetite-apatite rocks show both spatial correlations and isotopic similarities with evaporite sequences (Barton and Johnson, 1996; Li et al., 2015; Tornos et al., 2017; Bain et al., 2020, 2021; Peters et al., 2020). We contend that evaporite-derived aqueous brines are not the key factor, as such brines circulate in numerous hydrothermal settings, from orogenic belts (Morrissey and Tomkins, 2020) to the roots of many porphyry copper deposits (Runyon et al., 2019), without generating magnetite-apatite rock. The key difference is whether the evaporite package melts. Similarly, in the case of carbonate rocks, we can make an analogy to calc-silicate skarns, which are thought to be primarily hydrothermal-metasomatic in origin. The key difference, again, is whether limestone melting generates carbonate-rich melts. Such melts are potent metasomatic agents that can generate calc-silicates (Vasyukova and Williams-Jones, 2022) and crystallize to antiskarn (Anenburg and Mavrogenes, 2018) and endoskarns, the latter representing transitional cases between skarn and magnetite-apatite systems (Xu et al., 2023).

Melting of limestones and evaporites should be suspected wherever hot magmas intrude such rocks. Whereas the solidus T of monomineralic salt is generally high (for example, 1460 °C and 1339 °C for anhydrite and calcite, respectively), eutectic relationships of multicomponent systems dramatically lower the melting T. The eutectic between CaSO₄ and CaCO₃ is at 977 °C (Treiman, 1995) and that between CaCl₂ and CaSO₄ is at 635 °C (Freidina and Fray, 2000). By analogy to CaCl₂-CaSO₄-CaF₂ (Arbukhanova et al., 2009), the eutectic between CaCl₂, CaSO₄, and CaCO₃ is likely <600 °C. Addition of alkali chlorides lowers the eutectic even further (Walter et al., 2020), as do phosphate and fluoride (Treiman, 1995), and even small amounts of water (Durand et al., 2015). Hence, halide-bearing carbonate melts persist to ~200 °C (Anenburg et al., 2020), and our results show that melts dominated by sulfates, chlorides, and carbonates are stable down to <300 °C (Table 1).

While melting of chemical sediments supplies molten salts, a related industrial process—namely, smelting—provides insight into how these ingredients influence ore formation. Smelting is the process whereby metals are extracted from a rock by first melting it, then prompting separation of the melt into two immiscible liquids (liquid “matte” enriched in the metal of interest, and liquid “slag” composed of the unwanted impurities) through strategic addition of suitable fluxes (Moore, 1981). The fluxes serve multiple roles: lowering the melting T and melt viscosity, triggering immiscible phase separation, and sequestering alkalis, SiO₂, Al₂O₃, and other impurities in the buoyant calc-silicate slag. Sulfate, chloride, and especially carbonate salts are among the chief fluxes used in industrial smelting; especially limestone in the case of traditional iron smelting (Moore, 1981).

In magnetite-apatite deposits, assimilation of sulfate, chloride, and carbonate salts plays much the same role as they do in smelting: triggering immiscible separation of iron-rich liquids and promoting selective removal of silicate impurities. A process analogous to industrial smelting unfolds when hot, mantle-derived magmas encounter and assimilate fluxes—the same ones used in industrial furnaces—in the form of chemical sedimentary rocks. Just like in an industrial furnace, addition of carbonate, sulfate, and chloride fluxes triggers immiscible separation of two or more liquids while also lowering the solidus T and viscosity. Molten salts then assist in both forming and refining the nascent ores by accumulating iron, phosphate, and REEs (Anenburg et al., 2020; Pietruszka et al., 2023), and by removing calc-silicate impurities (Tornos et al., 2024). Snapshots of the early stages of such immiscible separation are recorded in melt inclusions in the andesites from El Laco, Chile (Pietruszka et al., 2023, 2024). As phase separation proceeds, significant volume fractions of iron oxides and REE phosphate in the inclusions record concentration of iron and phosphorous in the “matte” (Fig. 2B; Fig. S16A). Other components are sequestered in the low-viscosity and chemically reactive residual salt melt, stripping away impurities and generating haloes of calc-silicate rock analogous to industrial slag. The calc-silicate rich inclusions represent aliquots of contaminated melt trapped amid this zone-refining process, whereby most of the principal impurities are removed.

For decades, magnetite-apatite deposits have defied attempts to define a coherent genetic model. We suggest that this is largely because one of the key ingredients—molten salts—has been overlooked. Many previous studies seem to have assumed that the fluids must be similar to those in other well-known ore deposit types, such as porphyry-copper deposits, even though the ores themselves are markedly different. Widespread evidence for molten salts in magnetite-apatite deposits offers a new perspective on what these systems represent and reconciles why they never fit neatly into the traditional “magmatic” or “hydrothermal” categories. Our interpretations of the sources and roles of these fluids remain speculative (they are working hypotheses, and need further investigation), but evidence for their widespread involvement is concrete and direct. Hence, models for how these rocks form must explicitly address the role of molten salts, not as an occasional or anomalous factor, but as a basic commonality across this entire category of deposits. More broadly, this means that our view of crustal geologic fluids must expand to include molten salts as an essential type that gives rise to distinctive mineralization and metasomatism.

We thank Michael Anenburg, Dan Harlov, and an anonymous reviewer for constructive comments that helped us improve the paper. This study was supported by Natural Sciences and Engineering Research Council of Canada Discovery Grants to M. Steele-MacInnis (RG-PIN/2018-04370) and J.M. Hanchar (RG-PIN/004649-2015), and by the Spanish Grant NANOMET PID2022-138768OB-I00 funded by MCIN/AEI/10.13039/50110001133 to F. Tornos. R.S. Bottrill publishes with the permission of the Director of Mines, Tasmania.