Atmospheric ISRU & Propellant

The air as feedstock

Carbon Dioxide at 6 Millibars: Gathering the Raw Material

The feedstock is free, unlimited and everywhere. Getting enough of it into a pipe is the engineering.

The Martian atmosphere is 95.1% carbon dioxide, with 2.6% nitrogen and 1.9% argon along for the ride. It requires no prospecting, no excavation, no site selection and no planetary-protection argument. Every other body in the series makes its propellant from something that must first be found, dug and hauled; on Mars the plant sets up wherever the lander stops and starts breathing.

The catch is density. At 0.020 kg/m³ and 610 Pa, a plant must process an enormous volume of very thin gas to collect a useful mass flow, so the intake stage is a real machine rather than a duct. Two approaches exist: mechanical compression, which means a scroll or piston compressor running continuously in dust-laden air, or cryogenic adsorption, which freezes CO₂ onto a cold surface at night and releases it by day, trading electrical power for thermal cycling and time. Both have terrestrial ancestry; neither has flown at production scale.

Intake is where the failure population concentrates. Compressors are rotating machinery in an abrasive environment, which puts them in the fail-suddenly category rather than the degrade-gracefully one — a distinction that drives the redundancy architecture below.

MOXIE

MOXIE: The Only In-Situ Propellant Hardware Ever Flown

122 grams of oxygen, produced 16 times over two years, at up to 12 grams an hour.

The Mars Oxygen In-Situ Resource Utilization Experiment rode Perseverance to Jezero Crater and ran between April 2021 and August 2023. It compressed Martian air, heated it to roughly 800 °C, and split CO₂ into oxygen and carbon monoxide across a solid-oxide electrolysis stack — a ceramic membrane that passes oxygen ions and nothing else. Across 16 runs it produced about 122 g of oxygen at up to 12 g/hr, at roughly 98% purity, including runs at night, during dust season, and across the full range of atmospheric density the site offered.

What MOXIE settled is that the chemistry, the ceramics and the thermal cycling work in the real environment, which is not a small thing for a technology whose failure mode is cracked ceramic. What it did not settle is scale. A crew of four needs roughly 3.4 kg of oxygen a day for breathing alone — about 140 g/hr, or 12 times MOXIE’s peak rate for life support. Propellant production runs two orders of magnitude above that again. The gap between a toaster-sized demonstrator and a plant producing hundreds of kilograms a day is the entire unfunded middle of the technology.

Two oxygen routes

Dissimilar Redundancy on 78% of Propellant Mass

Water electrolysis and direct CO₂ electrolysis share no feedstock, no physics and no equipment. No other body offers that.

Methalox is 78% oxygen by mass at the mixture ratios real engines run. Oxygen can be made two entirely separate ways on Mars. The water route — prospect, excavate, transport, purify, then electrolyze — has a long chain with a mining operation at the front of it. The atmospheric route — intake, compress, heat, solid-oxide electrolysis — has no mining, no water, and few moving parts.

Because they overlap in nothing, a failure in one cannot propagate to the other. That is true dissimilar redundancy on the majority of propellant mass, and it is a property no other destination in the series has: lunar oxygen comes from regolith reduction and nothing else, belt oxygen from water and nothing else.

The recommendation that follows is cheap and specific: run life-support oxygen on the atmospheric route. A crew of four needs roughly 7 kW and a few hundred kilograms of hardware for the whole duty. Doing it that way means that losing the ice mine — the most exposed element in the entire chain, the one with prospecting risk, excavation risk and planetary-protection exposure — costs the launch window rather than the crew. It is the cheapest dissimilar redundancy available anywhere in the architecture, and the hardware has already flown.

Sabatier

The Sabatier Loop and Why O/F = 4.00 Is an Identity

The net reaction is the exact reverse of methane combustion, so the ratio cannot come out any other way.

Methane is made by the Sabatier reaction, CO₂ + 4H₂ → CH₄ + 2H₂O, an exothermic reaction over a nickel or ruthenium catalyst at 300–400 °C, discovered in 1897 and run continuously aboard the International Space Station since 2010. On Mars the hydrogen comes from electrolyzing local water, and the water Sabatier produces is recycled back into the electrolyzer.

Combining the two steps gives a net reaction of CO₂ + 2H₂O → CH₄ + 2O₂, which produces oxygen and methane in a 4:1 mass ratio. That 4.00 is a stoichiometric identity rather than a computed result — the net reaction is precisely the reverse of methane combustion, so conservation of atoms forces it. Two independent derivations agreeing on 4.00 are not corroboration; they are the same arithmetic twice.

This matters because the literature frequently quotes a smaller number and is often accused of error. It is not an error. A plant that imports hydrogen from Earth rather than electrolyzing local water runs at O/F = 2.00 exactly, because the imported hydrogen supplies what would otherwise have come from splitting water. Both figures are right for their own architecture, and confusing them has produced a decade of avoidable argument.

Engines want roughly 3.6, not 4.00, so a plant running the water route produces a modest oxygen surplus — conveniently, the same surplus that life support and the storm shelter want. The alternatives to a surplus are methane pyrolysis, which recovers hydrogen at the cost of discarding carbon, or accepting a hydrogen import. Zubrin’s original Mars Direct took the import route: about 6 t of hydrogen carried from Earth, atmospheric CO₂ for everything else, no mining at all, and roughly 100 kWe. It works, and the price is storing liquid hydrogen for years on a surface where boiloff is worse than it is in orbit.

Power

The Propellant Plant Sizes the Reactor, Not the Habitat

Three to six times the power of everything else on the site, in the architecture most programs assume.

Under direct-to-surface the plant must make 683 t of methalox in a 780-day window, about 876 kg a day, which draws 438–912 kW continuous. A 150 kWe outpost is a rounding error beside it. Under the orbit-staged ferry the same plant makes 107 t at about 137 kg a day and draws 68–142 kW — the difference between inventing a megawatt-class space fission reactor and using the 40 kWe class already under contract. The usual sequence of building the outpost and then adding ISRU has the dependency backwards: the plant sizes the power system, and the power system sizes everything else.

A reporting correction goes with those figures. Propellant production is an energy integral over a window, not a power rating — the kilowatt numbers are 780-day averages, and the plant should be designed to a duty cycle rather than run flat out. That is what makes it the site’s load-follower: it is large, continuous, genuinely interruptible, and its product is storable. Grid precedence runs reactors flat out, life support and habitat off the top, peaky industrial loads from the buffer, and the propellant plant absorbing whatever is left, curtailed first when anything else needs power. With several trains the curtailment is granular — shed a train, no fast control loop required.

The chain

ISRU Is a Serial Chain, and Redundancy Is Bought at the Stage

Duplicating one stage buys nothing if another is single-string.

The plant is five stages in series: extraction, electrolysis, Sabatier, liquefaction, storage. Availability is set by the weakest stage, so the useful question is not how many plants there are but which stage fails and how.

Three half-size trains, not two full-size ones. N trains each sized at 1/(N−1) of the requirement gives total installed capacity of N/(N−1): two trains need 2.00× the requirement, three need 1.50×, four 1.33×. Two full-size units is the intuitive answer and the expensive one — three half-size trains give the same single-failure guarantee for a quarter less hardware.

Buffering does the rest. Water is the universal interstage buffer, and at the ferry architecture’s 65 kg/day of net mining, about 2 t of stored water decouples every stage in the chain for 30 days — and it is the same substance as the life-support reserve and the storm-shelter shielding, so the tonnage does triple duty. Buffer as water, not as gas.

Liquefaction and storage

Holding Cryogens for 780 Days on a Warm, Dusty Surface

Making the propellant is the well-studied half. Keeping it is where the mission is actually lost.

Propellant made early in a window must survive until departure, which means holding liquid oxygen at 90 K and liquid methane at 111 K for up to two years in daylight, dust and an 80 K daily temperature swing. Boiloff is worse on the Martian surface than in orbit, because the tank sees conducted heat from the ground and convected heat from the air in addition to radiation.

Burial is the answer, and the numbers are decisive: a buried tank in regolith at a thermal conductivity of 0.09 W/m·K sees about 945 W of heat leak, against roughly 113 kW for the same tank standing exposed. That is a factor of 120, bought with an excavator.

Two corrections apply to the naive version of that plan. First, cold surfaces on Mars cryopump the atmosphere: CO₂ freezes onto the ground around a buried cryogenic tank and forms a conductive halo, so the 136:1 insulation advantage is a beginning-of-life figure that degrades toward 21–72:1 unless the tank carries a sealed vapor barrier. Second, ordinary multilayer insulation does not work here — the mean free path of CO₂ at Martian pressure is about 9.5 µm, so millimeter-scale MLI gaps conduct freely, while fine-pore aerogel with 20 nm pores performs at close to its vacuum value. The insulation choice is set by gas kinetics, not by temperature.

One free asset sits in the same section. A tank farm holding hundreds of tonnes of methalox is, incidentally, a 206 MWh chemical energy store — larger than any battery bank the site will ever have, and already paid for.

Buffer gas

Nitrogen, Argon and the Things the Atmosphere Gives Away

4.5% of the air is inert gas, and that is the only free buffer gas anywhere off Earth.

A habitat atmosphere is mostly not oxygen. The buffer gas that makes up the rest — 63% to 76% of the mix depending on the chosen pressure and oxygen fraction — must be shipped from Earth at every other destination in the series. The Moon has no nitrogen worth the name. The belt has none. Mars has 2.6% nitrogen and 1.9% argon in an atmosphere that a compressor is already processing for carbon dioxide, so buffer gas arrives as a byproduct of the propellant plant.

Argon is the easier catch and is perfectly serviceable as a pressure buffer. Nitrogen is the one that matters longer term, because a settlement that intends to grow food needs nitrogen for the biology, not just for the air. Neither stream is large, and both take real separation work — but the raw material is already in the pipe.

The carbon monoxide that solid-oxide electrolysis produces as its other output is usually vented. It is also a fuel, a reducing agent for metallurgy, and a feedstock for carbonyl-process metal refining. Nothing in the current architecture uses it, which makes it the most obvious unclaimed byproduct on the site.

Record

Provenance and Data Sources