Magnesium metal (Mg) is well-known for its mechanical properties, which include exceptional specific stiffness, high specific strength, and mechanical damping capability. These favorable characteristics make it extremely advantageous in a wide number of industries, including transportation vehicles (Schumann, 2005; Skszek, 2015; Magnesium Vision, 2020; Powell et al, 2021). Its high oxidation energy and affinity for hydrogen make it also favorable for energy storage (Bergthorson, 2018; Röntzsch and Vogt, 2019). A significant barrier to greater use for these applications has been the lack of a Mg primary production technology with comparable cost, energy use, and environmental impact to those of the steel or aluminum which it would replace. Furthermore, the US and EU have both named Mg metal a critical material with very high supply risk and difficulty of substitution (European Commission, 2020; Federal Register, 2022), creating urgency for new production technologies.

Mg is commercially produced in two distinct ways: electrolysis of magnesium chloride (MgCl₂) or thermal reduction of magnesium oxide (MgO) using the Pidgeon process. Historically, electrolysis has accounted for nearly 75% of the global magnesium supply. This uses anhydrous magnesium chloride (MgCl₂) or carnallite (MgCl₂·KCl) feedstock made from brines, operates at temperatures ranging between 928 and 993 K, and produces liquid Mg metal and chlorine (Cl₂) gas at cathodes and anodes, respectively. This process results in life cycle cradle-to-gate greenhouse gas (GHG) emissions of 5–7 t carbon dioxide equivalent (CO₂e) per t Mg product (Ehrenberger, 2013; Xie et al, 2018; Ehrenberger, 2020), largely because of the expensive chloride drying operation, and handling of chlorine gas. A new variant at Alliance Magnesium uses a hydrogen anode to produce anhydrous HCl, and reduces the GHG impact to as low as 2.5 t CO₂e/t Mg (Fournier, 2021).

In the early 21st century, as China became the world’s largest producer of magnesium, the low cost of labor and thermal energy in China enabled the Pidgeon process to become dominant despite using more energy than electrolysis. Around 80 percent of the world’s primary magnesium metal is now produced in China via the Pidgeon method which is based on thermal reduction between 1373 and 1473 K with ferrosilicon as a reducing agent. The higher energy requirements, and use of coal and coke oven gas for thermal energy, result in GHG emissions usually around 20–27 t CO₂e/t Mg (Cherubini et al, 2008; Ehrenberger, 2013; Ehrenberger, 2020), with sulfur, soot and other emissions as well.

The introduction of the Hall-Héroult cell in the late 1800s dramatically reduced the cost of aluminum metal by using low cost aluminum oxide feedstock. A similar process produces most light rare earth metals such as neodymium. Unfortunately, two properties of magnesium prevent this process from reducing the cost and energy consumption of MgO reduction. First, Mg metal density is lower than all molten halide salts, so it would float and short-circuit the cell. Second, metallic Mg dissolves in many salts, resulting in electronic current and decreased efficiency.

To address both of these issues, Yerkes proposed a magnesium oxide electrolysis cell with a reactive cathode in the 1940s (Yerkes, 1947). The cathode of the cell is a dense liquid metal, such as tin, lead, copper or silver, which absorbs reduced magnesium metal to form an alloy with high density and low magnesium metal activity. Kang et al recently expanded on the concept using a MgF₂-LiF-MgO electrolyte with low melting point around 1023 K, low density, low volatility, and high conductivity, and measured current yield over 90% (Park et al, 2017; Kang et al, 2019; Lee et al, 2020; Lee et al, 2021a; Lee et al, 2021b; Jeoung et al, 2022; Kang et al, 2022). There are three important challenges with this electrolyte and process. First, MgO solubility is low, around 0.5–0.8 wt% at the 1,083–1123 K operating temperature range (Kim et al, 2020), compared with 8–15 wt% for alumina in cryolite (Zhang et al, 2003), such that maximum current density is likely low. Second, Ca and other electropositive raw material impurities accumulate in the electrolyte bath with no proposal for removing them. Third, Mg separation from the reactive cathode metal by vacuum distillation is a slow, costly, capital-intensive and energy-intensive batch process which consumes 5–7 MWh/t Mg product (Ditze and Scharf, 2008).

The variation on this process presented here would mitigate these issues by using MgF₂-CaF₂-MgO electrolyte with high MgO solubility (Krishnan et al, 2005; Krishnan, 2006), followed by gravity-driven multiple effect thermal system (G-METS) distillation (Powell, et.al, 2020; Telgerafchi et al, 2021; Rutherford et al, 2021; Espinosa et al, 2022; McArthur et al, 2023). G-METS can potentially reduce magnesium distillation energy consumption, capital cost, and other operating costs by as much as 90%. CaF₂ replaces LiF to inexpensively mitigate Ca accumulation in the electrolyte. MgF₂ addition balances the CaO impurity via the in situ reaction: MgF₂ + CaO → MgO + CaF₂. This is analogous to AlF₃-NaF-Al₂O₃ electrolyte in the Hall-Héroult process with AlF₃ addition to balance Na₂O, which is the most prominent impurity in Bayer process alumina. This electrolyte also builds on years of research on MgF₂-CaF₂-MgO molten salt for MgO electrolysis using solid oxide membrane (SOM) anodes (Krishnan et al, 2005; Pal and Powell, 2007; DeLucas et al, 2008; Ditze and Scharf, 2008; Gratz et al, 2014; Guan et al, 2014) which place a yttria-stabilized zirconia (YSZ) solid electrolyte between the molten salt and anode. This research showed MgO solubility in MgF₂-CaF₂ eutectic is as high as 22 wt% (Krishnan et al, 2005), but maximum measured current yield reached just 67% (Pati, 2010). One drawback is the higher eutectic point of MgF₂-CaF₂, which is approximately 1,223 K, such that the electrolysis process must operate above 1,250 K, cf. MgF₂-LiF enables a lower process operating temperature of 1,053 K (Lee et al, 2021b).

The work presented here builds on a prior techno-economic analysis (TEA) of MgO electrolysis using this method by the authors (Rutherford et al, 2021). A design concept shows how a large electrolysis cell and potline could operate using YSZ SOM anodes, including using magnetohydrodynamics (MHD) for stirring. New experiments demonstrate molten salt electrolysis of MgO using a liquid tin (Sn) cathode in MgF₂-CaF₂-MgO electrolyte with both carbon and YSZ SOM anodes. These experiments show over 90% current yield using carbon anodes, and 84% current yield using YSZ SOM anodes, the latter of which is the highest Mg current yield to date with SOM anodes. And expansions on the prior TEA study include a detailed cost model of YSZ SOM anodes, and a cradle-to-gate GHG emissions estimate for the process.

The flowsheet and stream flow rates of the proposed Mg production process, using molten salt electrolysis with reactive cathode followed by G-METS distillation, are shown in Figure 1. The cathode begins as Sn-Mg alloy with 5–10 wt% Mg, or a similar alloy with lead. MgO reduction increases its volume to form between 60–40 and 66–33 Sn-Mg alloy, from which most of the Mg is later removed via G-METS distillation. The majority of the Sn is recovered post-distillation and circulated back into the electrolysis cell, with a small amount added to make up for losses.

Because the alloy in the distiller is mostly Sn, it cannot use the untreated steel alloys envisioned for Mg and rare earth magnet recycling, where everything in the distiller would be at least 50 wt% Mg. The distiller would thus need a liner material, such as electrodeposited molybdenum (Takeda et al, 2022) or a ceramic coating. Make-up Sn cost can be reduced by feeding tin oxide (SnO₂) in place of Sn, this would be preferentially reduced in the electrolysis cell over MgO to feed the cathode.

Other minor impurities in the MgO feed such as silicon, iron, and aluminum may build up in the liquid Sn, requiring occasional purging to keep their concentrations low and prevent them from interfering with process operation. One potential method for such purging could be cooling to a temperature close to the Sn melting point to reduce solubility of impurities and precipitate them out as a sludge.

Oxide ions from MgO react at the anode. Carbon anodes emit CO₂ while lowering the electrical energy requirement for the process due to the energetically favorable chemical reaction C + 2 O²⁻ → CO₂ + 4 e⁻. While YSZ SOM anodes provide the unique opportunity to prevent direct CO₂ emissions from the process, the oxygen evolution reaction and resistance of YSZ significantly increase the voltage and energy requirement of the process, as described in the prior TEA study (Rutherford et al, 2021). An anode design featuring a YSZ electrolyte coating on porous electronically-conducting perovskite anode, such as La_0.8Ca_0.2MnO_3-δ, rare earth nickelates, or GDC-Pr₆O₁₁ (Nechache and Hody, 2021) could reduce the SOM thickness and cell resistance considerably. In either scenario, the cell would operate around 1,323 K, with 1 A/cm² anode current density requiring cell voltage of about 5 V and 8 V for carbon and SOM anodes, respectively.

A plant with carbon anodes would look very similar to a Hall-Héroult aluminum smelter, but one with YSZ SOM anodes would look quite different. Figure 2 shows a potential cell design, consisting of 30 clusters of 15 anodes each, for a total of 450 anodes, each carrying 400 A. These clusters can be switched off and removed independently to replace the anodes in the case of breakage or routine maintenance to address wear. The cathode uses steel rails joined by cast iron to graphite blocks similar to a Hall-Héroult cell. The cell and bus arrangement in Figure 2 would produce a mostly vertical magnetic field, which would interact with the radial component of current from each SOM tube to create vortices. This is important because the YSZ SOM cell lacks CO₂ gas lift stirring which drives mass transfer in cells with carbon anodes; MHD can instead provide that stirring.

Experiments characterized reduction of MgO from a 42 wt% MgF₂ 53 wt% CaF₂ 5% MgO (by mass) molten salt electrolyte with a reactive Sn cathode, forming a Mg-Sn alloy at the liquid tin cathode. Two experiments used a carbon anode, and one used a YSZ SOM anode with 3 wt% Y₂O₃ in the electrolyte to prevent yttria leaching from YSZ (Gratz et al, 2013).

Figure 3 shows the experimental apparatus used for electrolysis experiments. The experimental gantry and sealed furnace tube were made of aluminum and stainless steel respectively. To prevent oxidation of the graphite crucible or other components, argon gas entered the sealed furnace pipe at a rate of roughly 1 standard liter per minute (SLPM), displacing air and other gasses out through an exit tube. Exiting gases passed through an oxygen sensor before being vented away from the experiment. When oxygen content fell below 0.1%, heating and electrolysis could begin.

Carbon anodes were 0.75″ diameter graphite rods (graphitestore.com fine extruded rod). The YSZ SOM anode consisted of a 0.75″ outer diameter (OD) 2–3 mm thickness closed-end YSZ solid electrolyte tube, with 5–10 g silver inside acting as the anode, and 0.25″ diameter graphite rod current collector for simplicity and robustness.

Salts were first baked in a box furnace at 300°C for 4 h to prevent any adverse water reactions due to bath salt or oxide hydration in ambient air. In powder form, the various salts have low densities, so they were blended and melted together in a graphite crucible at 1,150°C for 12 h to reduce their solid volume by more than 50%. After cooling, the solidified salt block was extracted from the crucible and broken apart using a rock crusher to <2 cm pieces. This mixture was much more dense than its previously powdered form, making room for the other 1/3 total experimental mass of salts within the crucible. In cases with smaller salt quantities, the entire prebaked mass of salts fit within the crucible to be pre-melted.

To prevent current leakage from the molten salt reaction throughout the apparatus framing, various techniques of electrical insulation were used. As seen in Figure 3, the crucible cage used 3 ceramic bushings to prevent contact between the anode and stainless steel heat separating baffles. An alumina pipe separated the stainless steel cathode current collector from the baffles as well as the molten salt. The graphite crucible was placed atop an alumina plate and wrapped in ceramic fiber insulation to prevent current leakage to the body of the cage assembly. Similar precautions were taken outside the furnace; atop the lid assembly, various PTFE bushings are used to insulate the electrodes from the couplings which hold them in place.

For two of the experiments, a secondary alumina crucible separated the liquid Sn-Mg cathode from the graphite crucible, following Lee et al. (2020). (In the first carbon anode experiment the tin was in the bottom of the main graphite crucible.) Solid tin shot or lump (99.9+% pure, MetalShipper.com) was added to the alumina crucible before setting it within the larger graphite crucible. This nested crucible design meant the Sn-Mg cathode, when molten, was contained within the alumina crucible, only contacting the molten salt at the top interface of the crucible; the stainless steel cathode current collector also did not have electrical connection with the molten salt except via the Sn-Mg cathode. Once set, the alumina pipe was lowered and secured flush with the rim of the alumina crucible, the steel cathode was then lowered through the pipe until it rested atop the solid tin shot. The anode was then also lowered into position, a few centimeters from the bottom of the graphite crucible. After the electrodes were in place, the crushed pre-melted salt was added to the graphite crucible, concealing the alumina crucible entirely. Finally, any remaining prebaked salt mixture was added to the larger crucible before setting and sealing the crucible cage into the furnace pipe apparatus.

Because these experiments resulted in relatively low Mg concentration in Sn, a first “experiment zero” measured the concentration of Mg picked up by Sn during static exposure to the molten salt bath, without electrolysis. This experiment began with the following ingredients in a graphite crucible: 48.8 g tin lump (99.8% Strem) with a blend of powders consisting of 30.5 g MgF₂ (42.1 wt%), 38.3 g CaF₂ (52.8 wt%) and 3.7 g MgO (5.1 wt%). The crucible and its contents were heated in argon gas in stages to 1,050°C as follows: ramp to 500°C at 10°C/min, hold for 2 hours, ramp to 800°C at 10°C/min, hold for 30 min, and then ramp to 1,050°C at 10°C/min. This condition was held for 30 min, then the furnace power was shut off and the system cooled. The ramp-hold procedure was intended to drive off any moisture as water vapor; if ramped rapidly to 1,050°C, some of it might have reacted with fluoride salts to form HF.

All electrolysis experiments used this same heat-up procedure, though the one with a YSZ SOM anode used a slower heating rate of 2°C/min for all ramps. These ramp and set temperatures were controlled at the furnace thermocouple, which was often 20°C–50°C above the measured temperature inside the crucible.

In Sn-bath exposure (experiment zero), Mg:Sn mass ratio in the metal phase following exposure was measured using ICP-OES and reported as 0.0001050, 0.0001229 and 0.0001501. This indicated that the Sn gained approximately 0.0126 wt% Mg from contact with the bath; this is the value of y’ _Mg in Eq. 2.

As discussed above, Mg production operating cost using the above method is potentially low. The biggest energy and cost savings relative to conventional MgCl₂ electrolysis is in the much simpler and less energy-intensive drying of MgO, though the inability to use a multi-polar cell for MgO reduction raises the minimum electrolysis energy consumption.

Operating cost is sensitive to prices of electricity, MgO raw material, carbon or YSZ anodes, and tin for the reactive cathode. These experiments with carbon anodes have demonstrated current yield of 94%–97% which is comparable to 93%–95% in the aluminum industry, though it remains to be seen whether this persists over longer time and at scale.

It is possible that the current yield in the SOM anode experiment is lower than with carbon anodes due to the presence of Y³⁺ in the electrolyte. That is, although Y₂O₃ is slightly more stable than MgO, the preponderance of F⁻ ions in the electrolyte could make Mg more stable, as YF₃ is less stable than MgF₂, and yttrium is reduced. Y³⁺ in the electrolyte dramatically reduces yttria leaching from the SOM, so removing it would require using an alternative such as calcia-stabilized zirconia (CSZ) or magnesia-stabilized zirconia (MSZ) for the SOM instead of YSZ. With lower conductivity of CSZ or MSZ, it would be necessary to make the SOM thinner in order to avoid higher resistance and more energy consumption.

As mentioned, lead (Pb) is a potential alternative to tin for this process, with many advantages. At the time of this writing, Pb at $2,050/t is much less expensive than Sn at $26k/t. The higher density of Pb allows for more Mg absorption while maintaining higher density than the electrolyte. That higher starting Mg fraction, plus the higher activity coefficient for Mg in Pb than Sn, reduce the energy required for distillation. And Fe solubility in Pb is much lower than in Sn, so a nickel-free stainless material of construction for the distiller, such as alloy 440C, would not corrode as much with liquid Pb-Mg alloy. In production plant design, these advantages must be carefully weighed against the much more stringent handling protocols required for industrial use of Pb.

Regarding other potential reactive liquid metal cathodes.

• Cu and Ag melting points are much higher than those of Sn and Pb, making it much harder to remove electronegative raw material impurities such as Fe, Al and Si via temperature swing.

Fundamentally, this process is very similar to Hall-Héroult aluminum (Al) production. Because Mg is divalent rather than trivalent, electrolysis produces 50% more Mg by mole and 33% more Mg by mass per A·h charge than Al, particularly if the 94%–97% current yield measured here holds up at scale. Cell voltage with carbon anodes is very similar to that of Al cells, so electrical energy consumption would be lower than for Al, and MgO raw material cost is also very close to that of alumina. At scale, this process could therefore produce Mg at equal or lower cost vs. Al. With the same carbon anodes, current bus, cell configuration and materials, raw material feed and product material withdrawal systems, and electrical infrastructure, this Mg production process could even retrofit into an existing or curtailed alumina smelter to dramatically reduce capital cost.

For production using carbon anodes, G-METS distillation is likely the biggest technical risk: a continuous distillation process would have low capital and labor costs as well as low energy usage, but has not been proven. If continuous distillation is not feasible, conventional distillation with 6–10 h cycle time and 5-7 kWh/kg energy usage would raise the cost of Mg-cathode separation considerably.

Using YSZ SOM anodes, the biggest cost increase vs. carbon anodes is for energy due to the high electrical resistance of the zirconia membrane, resulting in $230/t Mg product additional cost, vs. $51/t in higher costs for the anodes themselves, at 4,000-h lifetime. If anode lifetime is just 2,000 h, then anode cost is about $100/t, etc. If a low-cost support could replace much of the YSZ thickness without sacrificing lifetime, then reductions in cell voltage and energy cost, along with reduced zirconia powder use, could result in lower overall cost despite the increased anode manufacturing complexity.

Though a full LCA is not yet complete, emissions calculated for this process are relatively low. In particular, using hydroxide raw material from brines or seawater instead of carbonates, and substituting electrical energy for fossil fuel energy in raw material production, can dramatically reduce emissions. As renewable electrical energy cost continues to fall relative to fossil fuel energy, this can potentially make the lowest-emissions scenario also the one with lowest cost.

At 94%–97% with carbon anodes and 84% with YSZ SOM anodes, the current yields measured in this study are the highest for electrolytic MgO reduction in molten MgF₂-CaF₂ salt. This is likely due to the use of a reactive tin cathode which reduces the activity of the Mg product, which both increases the reduction reaction driving force, and also reduces metallic Mg dissolution and electronic conductivity in the electrolyte.

Costs with YSZ SOM oxygen-producing anodes are around 10%–15% higher than with carbon anodes. Based on the model assumption of 4,000-h membrane lifetime, and cost estimates of YSZ powder and SOM tube manufacturing at scale, cost of the YSZ membranes themselves only increases Mg production cost by about 2% relative to carbon anodes; the remainder of the cost increase is due to increased cell resistance resulting in higher electrical energy usage.

A large fraction of Mg production GHG emissions using this process is due to raw material production and carbon anode consumption. Using MgCl₂ from brines or seawater, electrical energy for NaOH precipitating agent, electrical heating energy, and YSZ SOM anodes can dramatically reduce process emissions.