Accumulation of magnetite by flotation on bubbles during decompression of silicate magma

Magnetite (Fe3O4) is an iron ore mineral that is globally mined especially for steel production. It is denser (5.15 g/cm3) than Earth’s crust (~2.7 g/cm3) and is expected to accumulate at the bottom of melt-rich magma reservoirs. However, recent studies revealed heterogeneous fluid bubble nucleation on oxide minerals such as magnetite during fluid degassing in volcanic systems. To test if the attachment on fluid bubbles is strong enough to efficiently float magnetite in silicate magma, decompression experiments were conducted at geologically relevant magmatic conditions with subsequent annealing to simulate re-equilibration after decompression. The results demonstrate that magnetite-bubble pairs do ascend in silicate melt, accumulating in an upper layer that grows during re-equilibration. This outcome contradicts the paradigm that magnetite must settle gravitationally in silicate melt.

Flotation can happen from the moment of magnetite crystallization and first bubble nucleation; i.e., from the time of fluid exsolution until the density of the magnetite-bubble solution becomes higher than the surrounding melt or until the suspension reaches the top of the melt-rich magma chamber, likely a more crystalline or mushy layer. The depth range is variable and dependent on many parameters: -water content: The higher the water content, the earlier (deeper) the fluid exsolution begins.
-density of fluid: The densiy of the fluid is dependent on the amount of dissolved NaCl eq and dissolved metals, such as Fe. The lower the content of solutes, the further the suspension can ascend.
-density of the surrounding melt: The density of the surrounding melt is dependent on the melt composition and dissolved water concentration. The more mafic and dryer the melt, the further the suspension can travel. However, less water content means higher viscosity of the melt and may hinder the process.
-amount of magnetite in suspension: The higher the amount of magnetite crystals in the suspension, the more difficult it is for the exsolved fluid bubbles to lift the magnetite. For example, when the abundance of magnetite exceeds 37 vol% of a suspension that contains 35 wt% NaCl eq and 7.2 wt% dissolved Fe in an andesitic melt with a density of 2.27 g/cm 3 (see Knipping et al. 2015a for calculation), the suspension would become negatively buoyant.
-location of the melt-rich magma reservoir in the crust: The more shallow the melt-rich magma reservoir is located (thinned crust), the shallower the suspension can ascend.
In Fig. S1, the pressure and temperature range over which magnetite flotation is possible was calculated by using the MELTS software for thermodynamic modeling for an andesite that contains 5.75 wt% H 2 O at a fO 2 =NNO+3. The model results indicate that magnetite flotation is possible in magma reservoirs in Earth's upper crust from ~2 to ~10 km assuming a lithostatic geobaric gradient of 28 MPa/km.

S2: Velocity of magnetite suspension
The velocity of the magnetite suspension can be calculated by using Stoke's law (Eq. S1) due to its dependency on density contrasts, melt viscosity and bubble size.
Eq. S1 ρ s equals the density of the magnetite-fluid-suspension, which is dependent on the proportion of magnetite (5.2 g/cm 3 ) and fluid (0.5 g/cm 3 ) in the suspension, ρ m equals the melt density (2.27 g/cm 3 ) 10 , η equals melt viscosity (2.1 log kg/m*s) 42 , g is the gravitational force (9.81 m/s 2 ) and R is the bubble radius. Stoke's law usually calculates the sinking velocity of particles. Therefore, positive buoyant particles have a negative velocity. According to the experiments, re-equilibrium is reached after at least 72 h; i.e., by 72 h all bubbles ascended through the melt column and accumulated between the capsule wall and the melt and no bubbles are anymore existent within the melt. Thus, a minimum velocity of 3000 µm/72 h = 42 µm/h can be assumed for the suspension. This velocity translates to at least 365 m/1000 years on a natural scale. In a 1000 m thick magma reservoir it would take approximately (1000 m / 0.365 m/a =) 2700 years to reach re-equilibrium; i.e., to theoretically float all bubblemagnetite-pairs that could ideally accumulate into a (100 µm/3000 µm * 1000 m =) 33 m thick magnetite layer at the roof. Therefore, magnetite flotation is a very fast and efficient process on a geologic scale.    bubble (a,b), by several bubbles (c,d) or magnetite aggregates are attached to one or more bubbles (e,f).  Transmitted light (a,b) and BSE (c,d) images of an experiment conducted at a constant final pressure of 150 MPa without prior decompression and equilibrated for 3 days. The results reveal a heterogeneous distribution of magnetite and exsolved fluid bubbles. Innumerable small magnetite crystals (< 10 µm) are efficiently attached to the exsolved fluid bubbles that accumulated at the top of the capsule, while the bottom of the capsule is depleted in magnetite. This is in contrast to the fluid-absent static experiment at 250 MPa wherein large magnetite crystals (< 100µm) settled gravitationally to the bottom of the melt column (Fig. 3a,e). Fig. S7. BSE image of a natural magnetite sample from the Los Colorados IOA deposit in comparison with reflected light and/or BSE images of magnetite and glass from decompression experiments of the current study. (A) shows an overview BSE image including a typical inclusion-rich (black spots) magnetite core and pristine magnetite rim discovered at Los Colorados. (B) is the enlargement of an inclusion in the magnetite core and exhibits its polycrystalline nature. The inclusion-rich magnetite cores observed at Los Colorados are interpreted as igneous magnetite, since polycrystalline silicate inclusions only homogenized at magmatic temperatures (T>975 °C) 43 . The experiments of this study (C, D and E) reveal that sudden supersaturation of the melt caused by decompression/degassing results in fast magnetite growth, such as hopper growth 28 , where several silicate melt inclusions can be entrapped within euhedral appearing crystals. The size and habitus of the experimental magnetite inclusions are very similar to those in natural samples. This provides further evidence that polycrystalline silicate inclusions in oxides are an igneous growth feature; i.e., magnetite entraps melt as melt inclusions that crystallize during cooling into polycrystalline silicate inclusions. Elemental compositions are normalized to 100 %. P1D represents the composition of the starting glass. *wt% mgt was calculated by difference to the starting composition. For Cl-bearing experiments the addition of Fe induced by the added fluid as FeCl 3 (0.54 wt% Fe addition to the system) was taken into account prior to by-difference calculations.