Revisiting LUNOX and an ISRU critique

Posted on April 21, 2026 by Kirk Sorensen

There’s a particular NASA mission concept from 1993 that deserves more attention than it gets, and a 2007 critique of lunar in-situ resource utilization that deserves to be read alongside it. Taken together, they map out why lunar ISRU has spent thirty years stuck in an architectural cul-de-sac — and, more usefully, they point toward the way out.

The Joosten/Guerra LUNOX Architecture

In 1993, Kent Joosten and Lisa Guerra at NASA/Johnson Space Center proposed a lunar outpost mission architecture called LUNOX.[1] It was a serious piece of work, developed in the wake of Mike Griffin’s First Lunar Outpost (FLO) study that had concluded in 1992 with a price tag so eye-watering it had effectively killed the program. LUNOX was a maverick attempt to rescue the idea of a return to the Moon by fundamentally rethinking the propellant mass equation.

The central insight was simple and obvious. A lunar lander that has to bring its return propellant from Earth is carrying enormous amounts of dead mass through two full gravity wells. Oxygen is the dominant component of chemical propellant mass. The Moon has oxygen bound to metals in regolith. If you can break those strong chemical bonds between oxygen and metals in the regolith, and liberate, purify, and liquefy the oxygen, you can shrink the lander’s size dramatically. That further shrinks the launch vehicle that sends it, and collapses the program cost.

Joosten’s numbers made the case. His “Phoenix” manned lander massed just 33,684 kg after translunar injection, compared to the 95,000 kg of FLO’s corresponding vehicle. The ascent propellant was 10,165 kg of lunar-produced liquid oxygen; only 2,492 kg of liquid hydrogen needed to come from Earth. A Shuttle-derived heavy-lift launcher (essentially the Shuttle-C design) could handle the job. Total program cost was estimated at $19.6 billion in FY1993 dollars for a first crewed landing in 2005.

The surface infrastructure was elegant. The first cargo flight delivered a 7,269 kg LUNOX plant and a 5,110 kg nuclear reactor in the 40–60 kWe class. The second flight brought a fleet of small robotic vehicles: two “Loader” bulldozers that collected 500 kg/hour of ilmenite-rich regolith, two “Tanker” rovers that moved the produced liquid oxygen, and two “Hauler” rovers for heavy equipment. The chemical process was hydrogen reduction of ilmenite in a fluidized-bed reactor, followed by solid-state high-temperature electrolysis of the resulting water, with Stirling-cycle liquefaction for storage. Annual production capability was 24,000 kg of LOX at roughly 4% extraction efficiency after beneficiation.

LUNOX was to be international. Russian Energia boosters would launch all the unmanned cargo, the Americans would develop the crew hardware and launch the crewed missions on Shuttle-derived vehicles, and the payloads would be cross-compatible so either partner could carry any element if the other pulled out. It was a plausible, well-engineered architecture.

It went nowhere. The International Space Station absorbed the agency’s human spaceflight bandwidth, Mars Direct absorbed its long-term exploration imagination, and LUNOX was filed away. When the U.S. returned seriously to lunar planning a decade later, under Constellation, ISRU had been re-scoped as a second-order problem to be solved after the basic architecture was in place.

Rapp’s Critique of ISRU

In 2007, Don Rapp — a former senior research scientist and division technologist at JPL who had managed Mars in-situ propellant production research there — published a two-part essay in The Space Review titled “The problems with lunar ISRU.”[2] The second part, in particular, is a clinically devastating cost-benefit analysis of ISRU as it was being planned within the Constellation/ESAS architecture.

Rapp’s argument ran as follows. In the ESAS architecture, ISRU would save approximately 4 metric tons of ascent oxygen propellant per crewed mission, twice per year — 8 MT/year total. With a LEO-to-lunar-surface “gear ratio” of about 4:1, that translates to 32 MT/year of mass savings in low Earth orbit, or roughly one avoided cargo delivery every four years at $1.2 billion per cargo mission. Net savings: about $300 million per year.

The investment required to realize those savings, by Rapp’s accounting, was staggering. Prospecting campaigns to locate polar ice (the assumed feedstock): ~$5 billion across LRO, multiple long-distance rovers, and a ground-truth mission. Technology development for excavation, processing, and autonomous operations: ~$6 billion, including a nuclear reactor required only because the candidate sites were in permanent darkness at the poles. In-situ test and validation of one-tenth-scale and full-scale demonstration systems: ~$8 billion. Total: ~$19 billion.

Rapp’s conclusion was direct: saving $300 million per year against a $19 billion investment required more than sixty years to break even, and that was before discounting for the time value of money. For any reasonable financial framing, ISRU in the ESAS architecture simply did not pay back.

The critique cut deeper than the numbers, though. Rapp’s real point was structural. ISRU in the Constellation architecture was an add-on, not a foundation. The lunar sorties had to be fully capable of landing, ascending, and returning without ISRU, because the architecture couldn’t be allowed to depend on a not-yet-validated surface plant. That independence requirement ate all the leverage. The ascent propellant was only 4 MT — a modest fraction of total mission mass. The descent propellant, at 20–25 MT, was the real prize, and the ESAS architecture had no path to capturing it. ISRU was being evaluated as a marginal optimization on a system designed not to need it, and when evaluated that way, it failed the cost-benefit test.

Rapp ended with five conditions under which lunar ISRU could justify itself. ISRU must be built into the fabric of the campaign from the start. An extended robotic precursor phase must establish the plant before crewed missions arrive. Polar ice must be the feedstock (his view, given the architectures under discussion). Oxygen must be retained as ascent propellant, ideally with hydrogen as well. And ISRU must address descent propellants, not just ascent.

The Architectural Lesson

Read together, LUNOX and Rapp’s critique tell a coherent story. LUNOX was architecturally correct in 1993: ISRU was designed into the mission from the start, every mass budget assumed lunar-produced oxygen, and the lander simply could not close its mass equation without the LUNOX plant. Rapp’s five conditions are, almost point for point, a description of what LUNOX already was. The mission designers of 1993 understood what the mission designers of 2007 had forgotten.

LUNOX imagined using hydrogen as the way to reduce a particular oxide, ilmenite, and it might have done better. Hydrogen reduction of ilmenite at 4% extraction efficiency meant 25 tons of regolith had to be mined, beneficiated, heated, and processed for every ton of oxygen produced. That’s a staggering materials-handling burden that drives plant mass up, reactor power requirements up, and robotic-vehicle complexity up. The LUNOX plant itself massed 7.3 tons precisely because it had to move that much mass to produce a useful oxygen stream. Hydrogen reduction of ilmenite also recovers, at best, only one-third of the oxygen even in the ilmenite itself — and ilmenite is a minor phase in most of the lunar surface. The process is feedstock-limited to mare basalts, specifically the ilmenite-rich ones, which constrains site selection and demands significant upstream beneficiation before processing.

Fluorination followed by electrical reduction is another chemistry that gets all of the oxygen out of the regolith at reasonable temperatures, produces silicon and aluminum as byproducts rather than discarding a slag, and works on any silicate phase the Moon has to offer. Fluorination as a strategy was sitting in the open literature when Joosten was doing his LUNOX study.

Why Fluorine Works on Lunar Rock

The foundational argument for fluorination of lunar materials is not just that fluorine is the most reactive element, though it is. The deeper argument is an acid-base argument, and it was laid out in its cleanest form by Donald Burt of Arizona State University in an abstract presented at the 19th Lunar and Planetary Science Conference in 1988.[3] Burt developed the idea more fully in a 1992 conference paper,[4] and Seboldt and colleagues at the German Aerospace Research Establishment and the University of Bonn provided experimental validation the following year.[5]

The chemistry works as an exchange reaction. For any metal oxide MO, the reaction with fluorine runs

    \begin{displaymath} 2\text{MO} + 2\text{F}_2 \rightarrow 2\text{MF}_2 + \text{O}_2 \end{displaymath}

Two fluorines go in for every oxygen that comes out. Burt, drawing on his earlier work in petrological acid-base theory, treated the combination “$\text{F}_2\text{O}_{-1}$” as an exchange operator with the properties of a Lewis acid — effectively, the anhydride of HF. The question of whether a given mineral will fluorinate readily then reduces to a question of how basic its constituent oxides are. The more basic the oxide, the more favorably it exchanges two fluorines for an oxygen.

The striking result is that the affinity sequence for $\text{F}_2\text{O}_{-1}$ exchange among common lunar elements is periodic: K > Na > Li > Ca > Mg > Be, and so on, tracking the acid-base character of the oxides themselves. Compare this to the affinity sequences for oxygen alone or fluorine alone, which are irregular and don’t follow obvious periodic patterns. The fluorine-oxygen exchange reaction, uniquely, tracks the chemistry most directly.

This matters because lunar minerals are basic. The dominant phases — anorthite with its CaO and Al2O3, olivine and pyroxene with their MgO and FeO — are built from basic oxides. Silica is the most acidic oxide in the lunar assemblage, and even it fluorinates readily because the SiF4 product is volatile and continuously leaves the reaction site, exposing fresh surface. The Moon’s mineralogy is, from a fluorination standpoint, about as favorable a feedstock as you could ask for. The same cannot be said for terrestrial crustal rocks, which are much more silica-rich and therefore more resistant to fluorine attack.

Seboldt’s experimental work at Bonn demonstrated this directly. Four lunar soil simulants — two highland-type, two mare-type — were fluorinated at temperatures between 623K and 923K. At 720K (roughly 450C, a temperature that’s essentially trivial to reach), approximately 80% of the oxygen was extracted from any of the four compositions. Above 800K, olivine-free highland simulants gave near-total oxygen release. Framework silicates like anorthite and chain silicates like pyroxene decomposed easily; olivine was the laggard, because its reaction products form solid passivating coatings rather than volatile SiF4 that clears the surface. For highland sites where olivine is essentially absent, this is a non-issue.

What Fluorination for LUNOX Would Have Looked Like

A LUNOX architecture built around fluorination of highland anorthite rather than hydrogen reduction of mare ilmenite would have had dramatically different infrastructure requirements. The 4% extraction efficiency becomes 80%. The 25 tons of regolith handled per ton of oxygen becomes 2.8 tons. The plant mass shrinks, the reactor shrinks, the robotic excavation fleet shrinks. The feedstock is no longer mare ilmenite but the anorthositic highlands — which cover the majority of the lunar surface and are accessible from most landing sites. Silicon for solar cells and aluminum for structure come out as byproducts of the same process.

The fluorine itself, which must be imported from Earth, is imported as a stable salt rather than as the reactive gas. Burt’s 1988 abstract proposed NaF as the transport form, with sodium reduction of the intermediate fluorides recovering metallic Al and Si, and Na2O/CaF2 exchange producing CaO and closing the fluorine loop back to NaF for electrolysis. In the same abstract, Burt noted an attractive alternative: LiF in place of NaF. Lithium is nearly as strong a reducing agent as calcium and markedly stronger than sodium, LiF melts at a lower temperature than NaF (advantage for electrolysis), and LiF is 38% lighter than NaF on a per-fluorine basis — so 38% less Earth-launched mass delivers the same fluorine inventory to the Moon. The LiF route also reduces magnesium from its fluorides, which opens olivine and pyroxene as feedstocks and makes mare regolith as workable as highland anorthite. It’s a pathway with genuinely distinct virtues, and the choice between Na and Li routes is one of the real engineering trades left open by the existing literature.

The remaining unsolved problem in Burt’s scheme was the electrolysis of the molten fluoride salt to regenerate F2 at an inert anode. Fluorine is aggressive enough to attack nearly any conventional electrode material, including platinum. Burt proposed lanthanide-doped CaF2 as a solid-state conductor with the right electronic properties, citing contemporary work on defects in solid fluorides. That specific materials-science path has not yet been demonstrated at production scale, though the intervening thirty-eight years have produced a large body of rare-earth molten fluoride electrolysis knowledge that has moved the adjacent problems considerably forward. The Seboldt group proposed an alternative: recycle fluorine via HF rather than via NaF, using mature industrial HF electrolysis as the electrochemical step and atomic hydrogen reduction of the mixed fluorides as the recycling chemistry. Either path closes the loop; the second uses entirely proven industrial technology.

Applying Rapp to the Present Day

The deeper reason to revisit both papers now is that the current Artemis architecture is repeating Constellation’s structural mistakes almost verbatim. ISRU is again being treated as a second-order add-on. The Human Landing System program has no architectural dependency on lunar-produced propellant. The mission-critical systems are being designed to close without ISRU, which means ISRU — when it arrives, if it arrives — will again be evaluated as a marginal optimization on a system that doesn’t need it. Rapp’s critique will apply to Artemis ISRU with exactly the same force it applied to Constellation ISRU, and the conclusion will again be that the numbers don’t close.

The way out is the same way out LUNOX found in 1993: design the architecture around ISRU from the start, accept an extended robotic precursor phase, and choose a feedstock that doesn’t require $5 billion of prospecting to locate. Highland anorthite meets that last test trivially. It’s everywhere, it’s well-characterized from Apollo, and we’ve known its composition since 1971. The prospecting phase that Rapp costed at $5 billion is, for an anorthite-feedstock architecture, a few lines in a site-selection trade study.

Fluorination also addresses Rapp’s fifth condition — capturing the descent propellant mass — in a way that no oxygen-only ISRU approach can. Fluorination of anorthite produces metallic aluminum as a coproduct of oxygen. Aluminum burns with liquid oxygen at a specific impulse of 270-280 seconds. It’s not the best propellant combination you could imagine, but it’s one where both components come from the same feedstock at the same plant. Once you can produce both Al and LOX on the Moon, you’re no longer limited to saving ascent mass. You can top off descent propellants from local production too — not on the first mission, but on missions after the plant has run long enough to build up inventory. That’s the architectural unlock Rapp was looking for and that the water-from-polar-ice approach cannot provide, because polar ice gives you only hydrogen and oxygen.

The nuclear power question, which Rapp called out as potentially a show-stopper, also looks different in an anorthite-fluorination architecture. You are no longer forced to operate in the permanent darkness of polar cold-traps to reach your feedstock. A surface reactor at a mid-latitude anorthositic site can run continuously on a predictable day-night cycle with buffer storage, and the reactor itself can be a technology that has an independent terrestrial market rather than an exploration-only development program. A liquid-fluoride-salt reactor has direct chemistry overlap with the fluorination plant it’s powering — the same containment alloys, the same fluoride-salt handling expertise, the same electrochemical toolkit. The reactor and the ISRU plant stop being two separate expensive development programs and start being two applications of one technology base.

The Thirty-Year Lesson

The LUNOX study is thirty-three years old. Rapp’s critique is nineteen years old. The fluorination chemistry they could have used is thirty-eight years old in its foundational form and thirty-three years old in its experimental validation. None of what I’ve described here requires inventing anything new. It requires reading papers that have been in the open literature for a generation, applying a cost-benefit framework that Rapp laid out two decades ago, and noticing that the chemistry trade and the architecture trade point the same direction.

Joosten was right about the architecture and might have picked a better chemistry. Rapp was right about the economics of ISRU-as-afterthought. Burt and Seboldt were right about fluorine as the universal oxidant and electricity as the universal reductant. Putting the three together — LUNOX’s mission architecture, Rapp’s discipline about economic closure, and the fluorination chemistry approach that makes the numbers actually work — gives you a blueprint for lunar ISRU that has a shot at paying back its investment. It’s worth thinking seriously about it.

References

  1. Joosten, B. K., and Guerra, L. A. (1993). “Early Lunar Resource Utilization: A Key to Human Exploration.” AIAA 1993-4784.
  2. Rapp, D. (2007). “The problems with lunar ISRU.” The Space Review, two-part essay.
  3. Burt, D. M. (1988). “Lunar Production of Oxygen and Metals Using Fluorine: Concepts Involving Fluorite, Lithium, and Acid-Base Theory.” Lunar and Planetary Science XIX, pp. 150–151.
  4. Burt, D. M. (1992). “Lunar Mining of Oxygen Using Fluorine.” The Second Conference on Lunar Bases and Space Activities of the 21st Century, NASA CP-3166, Vol. 2, pp. 423–428.
  5. Seboldt, W., Lingner, S., Hoernes, S., Grimmeisen, W., Lekies, R., Herkelmann, R., and Burt, D. M. (1993). “Lunar Oxygen Extraction Using Fluorine.” In Resources of Near-Earth Space, pp. 129–147.
Posted in ESAS, Lunar Commerce, Lunar Exploration and Development, Space Transportation | 5 Comments

Queqiao, first use of the powered lunar swingby

Posted on April 19, 2026 by Kirk Sorensen

In 2006, no one had flown the powered lunar swingby to EML2. In 2018, the Chinese did it, and they cited Farquhar while they were doing it.

For 20 years I’ve been entranced by the astrodynamic advantages of the Earth-Moon L2 libration point, reached via a trajectory technique called the powered lunar swingby. The argument has always rested on solid physics — Farquhar published it in 1971, Edelbaum confirmed it in 1970, and I’ve been repeating it to anyone who would listen since a thread I opened on NASASpaceFlight.com in January 2006. But when I advocated for it at NASA in 2006 it lacked a flight demonstration.

It has one now, but it wasn’t NASA that flew it.

Queqiao

On May 20, 2018, China launched a small satellite called Queqiao — “Magpie Bridge,” from a Chinese folk tale about lovers separated by the Milky Way and reunited once a year by a bridge of magpies — on a Long March 4C rocket from Xichang. Its mission was to serve as a communications relay for the Chang’e 4 spacecraft, which China intended to land on the far side of the Moon. Since the far side faces permanently away from Earth, any lander there is completely cut off from direct radio contact. A relay satellite at EML2, with simultaneous line-of-sight to both the Earth and the lunar far side, is the elegant solution — exactly as Farquhar described in his original 1971 paper.

Queqiao took 24 days to reach its destination. A direct mission to low lunar orbit takes four or five days; the longer journey was not a shortcoming but a deliberate design choice. The Chinese mission planners evaluated three trajectory options for reaching EML2: direct transfer, low-energy transfer, and the powered lunar swingby. They chose the swingby. Their mission documentation explicitly cites Farquhar’s 1971 paper for the direct transfer trajectory as a comparison case, and selected the swingby over it specifically to save propellant.

On May 25, Queqiao flew past the Moon at an altitude of just 100 kilometers. At closest approach — perilune — it fired its thrusters for 912 seconds, producing a velocity change of 203 m/s. That burn, at the most propulsively efficient point in the trajectory, redirected the spacecraft outward toward EML2 rather than continuing on a hyperbolic escape trajectory. Four days later it approached the EML2 region, and on June 14, 2018, after a final 66 m/s insertion burn, it entered its operational halo orbit — a southern halo orbit around EML2 with a vertical amplitude of 13,000 km, placing it between 47,000 and 79,000 km from the Moon at various points in its orbit.

That halo orbit gave Queqiao continuous line-of-sight with both Earth and the lunar far side. On January 3, 2019, Chang’e 4 became the first spacecraft in history to land softly on the far side of the Moon, in the Von Kármán crater within the South Pole-Aitken Basin. Every data packet from that lander and its Yutu-2 rover has since traveled up to Queqiao at EML2 and back down to Earth. As of this writing, Queqiao is still operational, still in its halo orbit, still relaying data — more than six years after launch.

Why This Matters

Let me be specific about what Queqiao demonstrated, because each element maps directly onto the architecture I’ve been advocating.

The powered lunar swingby trajectory works. The 203 m/s perilune burn that redirected Queqiao from a lunar flyby onto an EML2 transfer orbit is precisely the maneuver Farquhar described and that I have cited repeatedly as the key to making EML2 accessible for far less delta-V than any alternative. It is no longer a theoretical trajectory. It has been flown, successfully, by an operational spacecraft, with a mission-critical relay function depending on its success.

The EML2 halo orbit is operationally viable. One of the standard objections to EML2 as a staging location is that the halo orbit is unstable and requires continuous station-keeping. This is true — but the station-keeping cost is modest, and Queqiao’s six-plus years of continuous operation in that orbit demonstrates that the maintenance burden is entirely manageable for an operational mission. The Chinese technical documentation reports small orbit correction maneuvers approximately every nine days; the propellant budget for this was clearly acceptable for a satellite designed to last at least three years and still operating at twice that.

The far-side communications geometry is exactly as advertised. Farquhar’s 1971 paper proposed the EML2 halo orbit specifically as a relay location for far-side operations, noting that it provides continuous line-of-sight to both Earth and the far side simultaneously. Queqiao demonstrated this is not just geometrically correct but operationally reliable through a landing, rover deployment, and years of surface operations in one of the most demanding communications environments in the solar system.

The halo orbit can be selected to maximize polar coverage. The specific orbit chosen for Queqiao — a southern halo with 13,000 km vertical amplitude — was selected after comprehensive trade analysis to ensure visibility of the lunar far side and south polar region. This is exactly the kind of orbit selection flexibility that makes EML2 attractive as a general-purpose staging location: you can tune the halo orbit parameters to optimize for whatever operational requirement matters most, whether that’s polar coverage, Earth link margin, or surface access geometry.

Farquhar’s Idea, Finally Realized

Robert Farquhar proposed the EML2 communications relay concept in 1971. He spent the following decades arguing for it within NASA, getting various versions of it considered and then set aside, watching the idea resurface repeatedly without ever quite making it to flight. When I started the NSF thread in 2006, I was drawing on his work and making the same arguments he’d been making for thirty years. The ESAS study had dismissed libration point rendezvous without ever properly evaluating the powered swingby trajectory. The Gateway, when it finally got serious attention, was placed in the right orbit but wrapped in an architecture that made it difficult to reach and hard to justify.

It was China that finally flew the mission Farquhar had been describing since the Nixon administration. The Wikipedia article on halo orbits notes this explicitly: although spacecraft had used halo orbits in the Earth-Sun system since 1978, China was the first to realize Farquhar’s original idea of a communications relay satellite in a halo orbit around the Earth-Moon L2 point. The first. In 2018. Forty-seven years after the paper.

Farquhar died in 2015, three years before Queqiao launched. He did not live to see his idea validated in hardware. That is a loss, and I note it with genuine sadness. He was one of the most creative astrodynamicists of the twentieth century, and the trajectory you see in every Queqiao mission graphic — the graceful arc past the Moon, the outward coast to EML2, the capture into the halo — is his.

Queqiao-2 and a Different Approach

China launched a follow-on satellite, Queqiao-2, in March 2024. Queqiao-2 did not go to EML2. Instead it entered a highly elliptical frozen orbit around the Moon — 200 km by 16,000 km, inclined at about 62 degrees — designed to provide coverage of the lunar south polar region that China is targeting for its next phase of exploration. The frozen orbit has the attractive property of requiring essentially no active station-keeping for up to ten years, because its orbital parameters were chosen to be naturally stable against the Moon’s lumpy gravity field over that timescale. For a satellite intended to support a specific regional surface program over a decade, this is a sensible choice.

But Queqiao-2 gives up a few things relative to Queqiao-1’s EML2 position. The frozen orbit provides excellent south polar coverage but does not offer simultaneous line-of-sight to the entire far side. It cannot serve as a relay for operations at arbitrary far-side locations the way the EML2 halo orbit can. And it is not a staging node for anything — it is a dedicated relay satellite, nothing more. Queqiao-1 at EML2 is also a relay satellite, but it occupies a location from which, in principle, you could stage landers to the surface, accumulate propellant for onward missions, and expand into a full operational node. The frozen orbit satellite cannot grow into that role.

The two satellites are not really competing — they serve different immediate needs. But the contrast between them illustrates the distinction I’ve been drawing throughout this series between a location that is merely useful for one specific task and a location that is architecturally foundational for an entire program.

What Queqiao Adds to the Blog Series

The previous posts in this series made the case for EML2 staging on the basis of trajectory mechanics, delta-V budgets, station-keeping costs, and the geometric advantages of the halo orbit. All of that analysis was correct — it was correct in 1971 when Farquhar published it, it was correct in 2006 when I was arguing it on an internet forum, and it remains correct today.

What Queqiao adds is proof. Not simulated proof, not analytical proof, but operational flight proof that a spacecraft can be launched from Earth, execute a powered lunar swingby at 100 km perilune altitude with a 203 m/s burn, coast to EML2, insert into a halo orbit with a 66 m/s burn, and then operate reliably in that orbit for years while performing a mission-critical relay function for surface assets 65,000 km away.

Every number in that sequence is consistent with what Farquhar calculated and what I’ve been citing. The perilune burn magnitude, the coast time, the insertion delta-V, the halo orbit geometry — all of it came out as predicted. The trajectory is not exotic. It is not experimental. It is proven, repeatable, and understood.

When I argue that a crewed transfer vehicle should execute a powered lunar swingby to reach an EML2 station for about 330 m/s total, I am now arguing for a trajectory whose essential physics has been demonstrated by an operational spacecraft. The scale is different — Queqiao is a small satellite and a crewed vehicle is a larger, more complex system — but the astrodynamics are identical. The powered swingby works. EML2 is reachable. The halo orbit is livable.

China flew Farquhar’s trajectory. Now someone needs to build Farquhar’s station.

A Note on the Radio Astronomy Payload

Another aspect of Queqiao’s mission connects directly to an argument I made in the lunar space elevator post about the scientific value of the EML2 location.

Queqiao carries a science instrument: the Netherlands-China Low Frequency Explorer, or NCLE, a radio astronomy experiment developed jointly by Dutch and Chinese scientists to detect faint radio signals from the early universe. Operating at EML2, with the Moon providing a partial shield from Earth’s radio emissions and the halo orbit giving broad sky coverage, NCLE is attempting observations in the low-frequency radio regime — below 30 MHz — that are completely inaccessible from Earth due to ionospheric absorption and man-made interference.

This is exactly what I described in the space elevator post: EML2 as a platform for low-frequency radio astronomy, with the Moon serving as a noise shield. Queqiao is not doing this from the lunar surface, and the NCLE doesn’t have the sensitivity of the large arrays that have been proposed for far-side surface deployment — but the principle is being demonstrated right now, operationally, by a satellite in EML2 halo orbit. The location is as good for radio astronomy as the theoretical arguments suggested.

Queqiao is, in miniature, the EML2 node I’ve been describing across this entire series: a relay station, a science platform, reached via powered lunar swingby, operating in a halo orbit around EML2, providing communications support for far-side surface operations. It is small and it is a satellite rather than a crewed station, but every element of the concept is there, working, right now.

The architecture isn’t theoretical anymore. But it’s no longer just American.

Posted in ESAS, Lunar Exploration and Development, Orbital Dynamics | Leave a comment

EML2 Lunar Elevator Introduction

Posted on April 17, 2026 by Kirk Sorensen

The first two posts in this series made the case for staging lunar missions at EML2 and for building a permanent station there. This one describes where that architecture ultimately leads — and it’s more extraordinary than most people have imagined.

I’ve spent the last two posts arguing that EML2 is the right place to put a cislunar space station: the powered lunar swingby gets you there cheaply, global surface access and anytime return come for free, and the location sits on the edge of Earth’s gravity well in a way that makes it the natural node for missions anywhere in the solar system. If you haven’t read those posts, the short version is that EML2 solves, in a single architectural choice, most of the propulsive problems that have haunted lunar mission planning since Apollo.

But there’s a step beyond the station called the lunar space elevator, and it transforms EML2 from a waypoint into something far more fundamental: a fixed physical connection between the lunar surface and the threshold of interplanetary space.

The Concept

A lunar space elevator is a long tether — a ribbon of high-strength material — stretching from an anchor point on the lunar surface upward through space to a counterweight located beyond EML2. The Moon’s gravity pulls the lower portion of the ribbon toward the surface; the centrifugal effect of the counterweight in the Earth-Moon system pulls the upper portion away. At EML2, the two forces balance exactly. The result is a structure held in tension, hanging between the surface and deep space, along which cargo carriers can climb under their own power — no rockets required for the ascent itself.

This is not a new idea. Jerome Pearson, the engineer who first published the Earth space elevator concept in the international literature in 1975, extended the idea to the Moon in a 1979 paper. He recognized that the Lagrangian points serve the same structural role for a lunar elevator that geostationary orbit serves for an Earth elevator — they’re the natural balance points where the tether tension reverses direction. A lunar space elevator balanced about L1 would serve the near side; one balanced about L2 would serve the far side and provide the better geometry for launching material into cislunar space.

There is one structural property of the lunar elevator that is counterintuitive and worth stating clearly from the outset: the longer the elevator, the less massive the system becomes. This seems paradoxical — surely a longer ribbon weighs more? — but the explanation lies in the counterweight. If the counterweight is placed just barely beyond EML2, it needs to be very massive to keep the whole structure in tension. If the ribbon is extended much further beyond EML2, the excess centrifugal force on the extended ribbon does the work instead, and the required counterweight mass drops dramatically. The total system mass — ribbon plus counterweight — falls as the elevator gets longer. Building longer is cheaper. This is one of the most important design insights in the Pearson work.

Why the Moon and Not the Earth

The Earth space elevator has been discussed seriously since the 1970s and remains speculative for two reasons. First, it requires a material with a strength-to-density ratio far beyond anything commercially available — the breaking height requirement exceeds 4,800 km, achievable only by carbon nanotubes, which remain unavailable at manufacturing scale. Second, the structure would intersect every satellite and piece of debris in Earth orbit, eventually, at high relative velocity. The debris problem alone may make the Earth elevator permanently impractical.

The lunar elevator has neither problem. The Moon’s surface gravity is about 1.6 m/s² — one-sixth of Earth’s — which changes the material requirements completely. The taper ratio between the ribbon’s thickest point (at EML2) and its thinnest point (at the surface) is approximately 4 for available materials. The equivalent ratio for an Earth elevator using the same material would be around 6,000. Materials like T1000G carbon fiber, Zylon PBO, or M5 polymer — all commercially available today — are sufficient for the Moon. Not “sufficient with some engineering margin” — genuinely sufficient, right now, with off-the-shelf materials.

And the debris problem? Currently, there is nothing in lunar orbit. No satellites, no debris field, no collision hazard. Any future lunar orbital infrastructure can be designed from the start with the elevator in mind. The lunar environment is, in this respect, as clean a slate as you could hope for.

The Elevator as a Two-Way Highway

Most descriptions of the lunar elevator focus on moving material up — lunar resources ascending to EML2 for delivery into cislunar space. This is the most dramatic application, and I’ll get to the throughput numbers shortly. But the Pearson team is careful to emphasize that the elevator operates in both directions, and the downward direction matters just as much for making the architecture self-sustaining.

Payloads from LEO can be raised into elliptical orbits that just touch the top of the elevator, attached to a climbing vehicle, and carried down to the lunar surface. This means that supplies, equipment, and construction materials bound for a lunar base don’t need a dedicated landing rocket — they ride the elevator down. The same infrastructure that exports lunar resources to Earth orbit imports Earth supplies to the Moon. It’s a highway in both directions, operating continuously, and the cost per kilogram in each direction benefits from the same economies of scale.

There is also a fleet of orbital transfer vehicles in the system architecture: Pearson describes a fleet of roughly 50 ion-propulsion tugs cycling between LEO and the elevator top, each consuming 10–20 kW of solar power and moving 500-kg payloads in about two months transit time. Over the life of the system, 50 such tugs could move roughly a million kilograms per decade between LEO and the elevator. They would initially carry elevator construction material outbound and eventually carry lunar-derived propellants and building materials inbound to LEO, where they become available to any spacecraft in Earth orbit.

The picture that emerges is not a launcher but a logistics system: a slow, high-capacity pipeline connecting the lunar surface, EML2, and low Earth orbit in a continuous, bidirectional flow. Pearson uses the pipeline analogy explicitly, and it’s the right one. The throughput is not fast — climbers move at roughly 100 km/hour, and the trip from the surface to EML2 takes weeks — but it is continuous, it requires no propellant for the ascent phase, and it scales with the number of climbers rather than with launch vehicle capability.

What the Elevator Stabilizes

There’s a property of the elevator that I mentioned briefly in the previous post but want to expand on here, because it changes the picture for the EML2 station in a non-obvious way.

EML2 is an unstable equilibrium point. A station placed there requires station-keeping propellant to remain in its halo orbit — without active correction, perturbations will eventually cause it to drift away. The station-keeping cost is modest, as Farquhar established, but it’s a perpetual maintenance requirement. The EML2 station is never truly “parked” — it’s always being nudged back into position.

The lunar space elevator changes this. When a tether is anchored to the lunar surface and extended through EML2 to a counterweight beyond, the mechanical constraint of the tether stabilizes the entire structure at EML2. Eugene Levin, one of the co-authors of the Pearson papers and a specialist in tether dynamics, showed that the unstable collinear libration points L1 and L2 of the classical three-body problem become stable equilibria when a tether is “anchored” through them in this way. The elevator doesn’t just use EML2 — it tames it. The station at EML2 becomes genuinely stable, requiring minimal station-keeping, because the tether’s mechanical tension holds everything in place.

This is not an intuitive result. In the classical three-body problem, L4 and L5 are stable while L1, L2, and L3 are unstable. But add a tether, and the situation inverts: L4 and L5 become unstable for tethered systems while L1 and L2 become stable. The elevator earns its keep not only as a transportation system but as a structural anchor for the entire EML2 complex.

Connecting to the Poles

The baseline lunar space elevator is equatorial — the ribbon hangs straight from the lunar equator up through EML2. But the polar regions are where the most valuable resources are: permanently shadowed craters containing water ice, and adjacent permanently sunlit peaks providing continuous solar power. Connecting the elevator to the poles is therefore not a nice-to-have but a prerequisite for the full vision.

Pearson’s team worked through the physics carefully. The ribbon can be curved from the equatorial anchor point toward higher latitudes, but the material strength places a hard limit on achievable latitude. With M5 fiber at half its stress limit (preserving the other half for payload capacity), the ribbon can reach about 36 degrees latitude. With carbon nanotubes, about 76 degrees — still 426 km short of the pole. No material can reach the pole directly, regardless of strength.

The practical solution is a tramway: a catenary ribbon suspended from towers, extending from wherever the space elevator reaches down to the equator, across the surface to the poles. The numbers here are surprisingly manageable. Towers 1 km high, suspended in the Moon’s low gravity, can span 91 km of catenary. To traverse the full 2,700 km from the equatorial base to the poles requires only about 30 such towers. Many of them could be built on existing crater rims and mountain peaks, reducing the required tower height and increasing the span. The path of the tramway could also be routed to pass through mineral deposits, turning the transportation corridor into an access road for resource extraction along its entire length.

The same solar-powered climbing vehicles that ascend the vertical ribbon can travel horizontally along the catenary, carrying cargo all the way from a polar ice mine — through the tramway, up the elevator, past EML2 — and releasing it into Earth orbit, without changing vehicles or transferring cargo. One vehicle, one continuous journey, from polar crater to high Earth orbit. The power supply for the tramway segment comes from the same articulated solar panels the climbers carry for the elevator portion.

Building the towers doesn’t require material imported from Earth. The Pearson team proposes a robotic construction system based on two elementary shapes — small compression beads about 2 cm on a side, and tension wires connecting them — that can be fabricated from lunar regolith using sintering, cast basalt, or powder metallurgy. The same automated system that builds tramway towers can fabricate habitat components, structural beams, and landing pads. Indigenous lunar material, processed robotically, becomes the infrastructure. This is ISRU taken to its logical conclusion: not just extracting propellant from the Moon but building the Moon’s transportation network from the Moon’s own rocks.

The Bootstrapping Problem and the Lunar Sling

One of the most interesting engineering challenges in building the elevator is the counterweight. The system begins construction from the EML2 balance point, deploying ribbon in both directions simultaneously — toward the surface and away from the Moon — until the lower tip reaches the equator. The counterweight mass required to balance the ribbon during and after construction is substantial, particularly for the shorter initial configurations.

Pearson’s team proposes an elegant solution: a surface-based rotating tether, which they call a lunar sling, to bootstrap the counterweight into place. A tether 12 km long, rotating at 24 g tip acceleration on top of a 4 km mountain, can launch 500-kg payloads to low lunar orbit every 4 hours with a power requirement of about 100 kW — a throughput of 3 tonnes per day. Extended to 24 km, the same sling can reach lunar escape velocity, sending payloads directly to L1 or EML2 without any rocket propulsion at all.

The sling bootstraps the elevator. It delivers counterweight mass and additional ribbon material to EML2, allowing the elevator to grow without requiring every kilogram of construction material to be launched from Earth. Once the elevator is long enough that the counterweight problem eases — recall that longer elevators need smaller counterweights — the sling becomes the main feedstock mechanism for the elevator’s expansion. The two systems work together: the sling launches material to EML2, the elevator carries it back down or releases it into cislunar space, and the whole system grows in throughput with each additional ton of ribbon deployed.

The Numbers

It’s worth being concrete about what this system can deliver, because the numbers are large enough to change the economic framing of everything else.

A single initial elevator strand, with climbing vehicles maintaining 100 km/hour, can deliver approximately 385,000 to 584,000 kg of material per year into high Earth orbit — the range depending on which paper’s assumptions you use for climber spacing and payload fraction. At projected launch costs of $1,000/kg two decades from now, that material would be worth roughly half a billion dollars per year to the Earth orbital economy.

But this is only the first strand. Once the elevator is operational, additional ribbon strands can be added — built from indigenously processed lunar basalt fiber, if that proves viable — multiplying the throughput without proportionally increasing Earth-launch costs. The system is designed to grow on its own resources.

The total system cost Pearson estimates at a mean of approximately $10 billion — comparable to a large NASA program or a significant international mission. At the low end of launch cost projections, closer to $2.4 billion. This is not a trivial number, but it is in the range of things that space agencies and international consortia actually build when they are serious. The ISS cost roughly $150 billion. A lunar space elevator system at $10 billion, delivering hundreds of thousands of kilograms annually into cislunar space, is a bargain by comparison — if you are willing to take the long view.

Putting the Architecture Together

Step back and look at what this three-post series has described as a complete system.

It starts with trajectories: the powered lunar swingby that makes EML2 accessible for about 330 m/s each way, solving the delta-V problems that have plagued lunar architecture since Apollo. The right staging point, reached efficiently, enabling global surface access and anytime return as structural features rather than engineering compromises.

It moves to a station: a permanent facility at EML2, accumulating hardware over time — modules, propellant depots, spent ascent stages repurposed as additional volume — serving as operations hub, propellant cache, communications relay, and interplanetary departure node. Stabilized by the elevator tether once that structure exists, rather than requiring perpetual station-keeping propellant.

And it arrives at the elevator: a ribbon of existing commercial composite material, deployed from EML2 to the equatorial surface and extended via tramway to the polar ice deposits, carrying hundreds of thousands of kilograms of lunar resources annually into Earth orbit without rockets, while simultaneously delivering supplies from Earth orbit to the lunar surface. Bootstrapped via a surface-based rotating sling. Built incrementally from lunar materials once the initial strand is in place. Bidirectional, continuous, scalable.

The whole system is powered by sunlight. The climbers run on solar arrays. The sling runs on solar power. The ISRU electrolysis plants run on solar power from the permanently sunlit polar peaks. Earth provides the initial ribbon material, the first generation of climbers, and the seed crew — and then the Moon begins to pay its own way.

This is what a mature cislunar economy looks like. Not a surface base that must be continuously resupplied at rocket prices. Not a flags-and-footprints program that ends when the political will runs out. A self-reinforcing system, drawing energy from sunlight and material from the Moon itself, growing in capability with each mission, and opening the rest of the solar system at a cost structure that no Earth-launched rocket architecture can approach.

The physics is settled. The materials exist. The engineering has been analyzed. What’s missing is the program that starts at EML2 and works outward with the long view in mind, rather than one that optimizes for the next election cycle and wonders why nothing ever scales.

The Moon is not just a destination; it’s a resource base, a construction yard, and a launching point. The elevator is how you access it at the scale we’ll ultimately need.

 

Posted in Space Tethers | 2 Comments

Gateway is gone, do we need a lunar station?

Posted on April 15, 2026 by Kirk Sorensen

NASA just canceled the Lunar Gateway — the first space station ever planned for EML2. That’s worth reflecting on, because the case for a station at that location has never been stronger.

In my previous post I laid out the astrodynamic case for staging lunar missions from the Earth-Moon L2 libration point rather than from low lunar orbit. The short version: the powered lunar swingby trajectory gets you there for about 330 m/s each way, global access and anytime return come for free, and EML2 sits on the edge of Earth’s gravity well in a way that makes it the natural departure point for missions anywhere in the solar system.

That post was about trajectories and delta-V budgets. This one is about what you would actually do with a station at EML2 once you had one — and why the recent cancellation of NASA’s Lunar Gateway has regrettable aspects even if the Gateway itself had serious problems.

First, Some Credit Where It’s Due

The Lunar Gateway was to be placed in a Near-Rectilinear Halo Orbit — a specific, carefully chosen member of the family of halo orbits around EML2. After being rejected by ESAS in 2005, NASA eventually did build hardware for the “right” location. The Power and Propulsion Element was in final testing. The HALO habitation module had been delivered. ESA had completed habitation and logistics hardware. Canada, Japan, and the UAE all had contributions underway.

And then in March 2026, NASA Administrator Jared Isaacman canceled it, pivoting the agency toward a lunar surface base instead.

There are defensible reasons for that decision, and I’ll get to them. But before accepting the premise that a station in an EML2 halo orbit was somehow the wrong idea, it’s worth being specific about what would have been gained — and what is now lost.

A Node That Never Closes

The single most underappreciated property of EML2 as a station location is what might be called its permanent openness. From any point on the lunar surface, at any time, a vehicle can reach EML2. There is no waiting for orbital planes to align, no launch window that opens and closes on a lunar timescale, no plane-change penalty that varies depending on where you landed and how long you’ve been there.

This is not true of an arbitrary low lunar orbit. As I described in the previous post, a polar orbit precesses, the Moon rotates beneath it, and the combination means that for any given (non-polar) surface location there are only certain times when a coplanar ascent is possible. Miss that window and you either wait or pay a steep delta-V penalty. The ESAS architecture acknowledged this and budgeted 1,450 m/s in the Orion service module to cope with it. For the surface crews, it meant that “anytime abort” was a promise that couldn’t always be kept.

A station at EML2 dissolves this problem entirely. Many of the halo orbits maintain a roughly fixed geometry relative to the lunar surface below. The crew on the surface always has a known, reachable destination above them. And critically, this is true from any latitude — the equator, the poles, the far side. You could have a crew at the South Pole Aitken Basin and a crew on the Aristarchus Plateau and both of them would have the same abort option available at any moment: burn toward EML2, arrive in two to three days.

That’s not just operationally elegant. For a program trying to establish a permanent surface presence — multiple simultaneous crews, multiple surface sites, operations that span years — it’s a prerequisite for safety that low lunar orbit simply cannot provide.

The Accumulation Principle

One of the ideas I kept returning to in the old NSF thread was the notion of “aggregation” — the accumulation of hardware at EML2 over time as a natural byproduct of conducting missions there. Every LSAM ascent stage that rendezvoused at EML2 and was captured rather than discarded became a piece of infrastructure. Small, underpowered, battery-limited infrastructure — but real, paid-for hardware in a stable location that costs almost nothing to keep there.

This points to a general principle that EML2 enables and low lunar orbit prevents: things you put at EML2 tend to stay there. Low lunar orbit is gravitationally hostile. The Moon’s lumpy, irregular gravity field — dominated by mass concentrations, or mascons, left over from ancient impacts — causes initially circular orbits to grow eccentric over months until they impact the surface. Keeping hardware in LLO requires active station-keeping propellant. Leave something there unattended for long enough and it becomes a crater.

EML2 is different. A halo orbit around EML2 requires only modest station-keeping — on the order of 400 ft/s per year, as Farquhar established in 1972, and potentially as low as 100 ft/s per year for some orbit choices. Hardware parked there doesn’t decay. A propellant depot filled on one mission is still available for the next. A habitat module delivered unmanned remains habitable for the crew that follows. Infrastructure accumulates rather than decaying, and the station grows over time in a way that a facility in LLO would find very challenging.

This is the logical foundation for a reusable lunar transportation system. You build the station incrementally, mission by mission. Each flight deposits something — propellant, a module, a spent ascent stage repurposed as additional volume. Eventually the station becomes large enough and capable enough to support a permanently resident crew, whose job is partly to maintain and extend the facility and partly to coordinate surface operations below. The surface base and the orbital node grow together, each making the other more capable.

The Propellant Economy

If there is a single capability that transforms the economics of lunar exploration from “expensive government program” to “something approaching sustainability,” it is in-situ resource utilization — specifically, the extraction of oxygen from lunar regolith and its use as rocket propellant. The Moon is roughly 45% oxygen by mass. It’s locked up in oxides, but the chemistry to liberate it is well understood. The energy required is large but not unreasonable for a well-powered surface installation.

Lunar oxygen changes the architecture in a fundamental way, particularly if you have somewhere sensible to deliver it. This is where EML2 becomes so impressive.

Consider the alternatives. If you want to use lunar oxygen to fuel Earth-departure stages, you need to get it from the surface to wherever those stages are waiting. If they’re waiting in LEO, you have to boost the oxygen off the surface (~2,000 m/s) and then brake it into low Earth orbit (~3,000 m/s from a lunar trajectory) — burning most of your propellant just to move the propellant, which is exactly as wasteful as it sounds. If the stages are waiting in low lunar orbit, you save the Earth-braking cost but you’re still deep in the Moon’s gravity well with a propellant depot that needs active station-keeping.

If the stages are waiting at EML2, everything changes. Lifting oxygen from the surface to EML2 costs roughly the same as lifting it to LLO — about 2,000 m/s — but now the oxygen is sitting at a location from which Earth departure requires only a small additional burn. The orbital energy of the propellant is preserved for the mission it’s meant to fuel. A tanker flight from the surface to EML2 deposits propellant that can directly power a trans-Mars injection. Nothing is wasted on intermediate orbital maneuvering.

This is the endgame that makes EML2 so compelling as a long-term node. The Moon produces the propellant. A reusable lander — fueled by lunar oxygen, carrying lunar-extracted hydrogen or fuel shipped from Earth — shuttles between the surface and the station. The station accumulates propellant, stages crews, and dispatches missions. The Earth only has to supply the things the Moon can’t provide: crew, food, specialized equipment, and the hydrogen or carbon-based fuel that pairs with lunar oxygen. Over time, as ISRU matures, even some of those imports may shrink.

The Interplanetary Crossroads

I wrote in the previous post about EML2 as a Mars staging node, but I want to be more specific about why this matters and why no other location offers the same combination of advantages.

Staging a Mars mission from LEO has a fundamental problem that doesn’t get discussed enough: getting the nodal geometry right. A trans-Mars injection from LEO requires the departure asymptote — the direction and speed of departure — to align with the spacecraft’s orbital plane at the moment of the burn. Because the line of nodes of a LEO orbit drifts at only 2–4 degrees per day due to Earth’s oblateness, an orbit that is correctly aligned today may be completely wrong three weeks from now. For a vehicle assembled in LEO over multiple launches — which any serious Mars mission will require — there is no guarantee that assembly will finish exactly when the orbit is in the right position. Miss the window and you wait months for the next one, or pay an enormous penalty to reorient.

From EML2, this problem is greatly reduced. The Moon sweeps through all values of right ascension at about 12 degrees per day — fast enough that you can pick your departure time from L2 to hit any target in the inner solar system within a reasonable fraction of the launch window. You’re not at the mercy of slow nodal drift. A Mars vehicle assembled over months at EML2 can wait there without losing its departure geometry. When the window opens, you execute a small burn (~300 m/s) to re-enter a highly elliptical Earth orbit, coast to perigee, and fire the trans-Mars injection burn from deep in Earth’s gravity well where the Oberth effect makes every kilogram of propellant count for maximum energy.

For low-thrust propulsion — nuclear electric or solar electric — the advantages are even starker. EML2 is effectively already outside Earth’s gravity well. A low-thrust vehicle departing from L2 begins thrusting almost immediately toward its destination rather than spending weeks or months spiraling outward through radiation belts. The spiral-out time that makes low-thrust systems impractical from LEO becomes trivial from EML2.

And then there’s the question of what comes back. A vehicle returning from Mars or an asteroid carries enormous kinetic energy relative to Earth. Capturing it into LEO requires either a large propulsive burn or a precise aerobraking pass. Capturing it into EML2 is easier — the Moon’s gravity can assist — and once there, the returning vehicle’s propellant, hardware, and any payload can be transferred and reused. A Mars ship that returns to EML2 rather than LEO can be refueled with lunar oxygen and sent out again without descending into Earth’s gravity well at all.

The Far Side and the Science Case

There is one capability that EML2 provides and no other location does: continuous line-of-sight to both the Earth and the lunar far side simultaneously, from a stable quasi-stationary vantage point.

The far side of the Moon is one of the most scientifically valuable real estate in the inner solar system. Shielded from Earth’s radio noise by 3,500 kilometers of lunar rock, it offers radio-quiet conditions that simply don’t exist anywhere else in the vicinity of Earth. The proposals for low-frequency radio telescopes on the lunar far side have been circulating for decades — arrays that could detect signals from the cosmic dark ages, the era before the first stars formed, that are completely inaccessible from any Earth-based or Earth-orbiting facility. A robotic telescope on the far side, operated in real time from an EML2 station, would represent one of the most significant astronomical capabilities ever built.

Beyond radio astronomy, a station at EML2 could coordinate all far-side surface operations — rovers, sample return missions, drill sites — that today require relay satellites to communicate with Earth. Rather than a dedicated relay satellite with its own station-keeping requirements and limited lifespan, the EML2 station itself serves as the relay. One facility does the work of many.

The EML2 station also has nearly continuous line-of-sight to the lunar poles — which is where the ice is, where the permanent shadow regions are, and where the first serious ISRU operations will occur. A station at EML2 is therefore not just a relay for the far side but an operations hub for the poles. Communications, navigation support, emergency coordination — all of it flows through the station.

What the Gateway Got Wrong, and What It Got Right

I don’t want to be uncritical of the Gateway specifically just because I’ve spent twenty years arguing for EML2 generally. The Gateway had real problems.

The biggest was that it was never cleanly integrated into a coherent architecture. It was added to the Artemis program after the first lunar landings were already planned via direct LOR, which meant the landers had to reach EML2 from the surface on an ascent stage that wasn’t originally designed for that range. The delta-V from the surface to EML2 is manageable but not trivial — about 2,700 m/s with a realistic trajectory — and the Altair/LSAM design was optimized for LLO, not for EML2. The result was a lander perpetually constrained by its propellant budget and a station that felt like an add-on rather than the centerpiece of a coherent system.

The right way to design around EML2 is the way I described in the previous post: the crew transfer vehicle (Orion or its successor) goes to EML2 via the powered lunar swingby at low delta-V cost, and the lander is designed from the ground up to operate between EML2 and the surface. Instead, NASA designed the lander for LOR, then tried to stretch it to reach a station that was four times farther away. That’s not an EML2 architecture. That’s a LOR architecture with an expensive orbital waypoint bolted on.

The Gateway also suffered from being designed as a political object as much as an engineering one — sized to give international partners meaningful contributions, scheduled to keep SLS flights on the manifest, shaped to satisfy constituencies on Capitol Hill. These pressures produced a station that was too small to be genuinely useful as a long-duration habitat and too expensive to be politically safe when budgets tightened.

What the Gateway got right was the orbit. The Near-Rectilinear Halo Orbit is an excellent choice within the EML2 halo family — it provides good south polar coverage, manageable station-keeping costs, and reasonable transit times to the surface. The location was correct. The architecture around it was not.

What Should Come Next

NASA’s stated plan is now a lunar surface base, with the Gateway hardware repurposed where possible. That’s not necessarily wrong as a near-term priority — boots on the ground, ISRU demonstrations, understanding the surface environment — but it is incomplete as a long-term strategy if it doesn’t include a plan to eventually build the orbital node that makes the surface base part of a larger transportation system.

The good news is that the Power and Propulsion Element — the solar-electric propulsion heart of the Gateway — is apparently being redirected toward a Mars mission. That’s the most astrodynamically coherent use of that hardware imaginable: a high-power electric propulsion module operating in deep space, for exactly the kind of mission that EML2 staging is designed to support. If that program survives and succeeds, it may keep the institutional knowledge and hardware heritage of EML2 operations alive until the moment arrives to revive the station concept properly.

The case for an EML2 station hasn’t weakened. If anything, the experience of Artemis — the mass problems, the plane-change penalties, the inability to deliver true anytime return from arbitrary surface sites, the difficulty of making LOR scale beyond a handful of missions — has reinforced every argument that Farquhar made in 1972 and that I’ve been repeating since 2006. The orbit is right. The astrodynamics are right. The question is whether anyone will build the right architecture around them before the next round of cancellations makes the question moot.

I remain, as always, cautiously optimistic. The Moon will still be there. EML2 will still be there. The powered lunar swingby trajectory will still work. Somebody will eventually put a station in the right place and wonder why it took so long.

Posted in ESAS, Lunar Commerce, Lunar Exploration and Development, Orbital Dynamics, Propellant Depots, Reusable Lunar Landers, Space Development | Leave a comment

2006, a lunar architecture

Posted on April 13, 2026 by Kirk Sorensen

The astrodynamic case for Earth-Moon L2 rendezvous was made in 1972 and again in 2006. The math hasn’t changed. Neither has the conclusion.

Back in January 2006, I opened a thread on the NASASpaceFlight forum with a bit of a trick. I posted what I described as excerpts from a forthcoming article in Aerospace America by the renowned astrodynamicist Robert Farquhar, arguing that NASA’s new Exploration Systems Architecture Study should stage its lunar missions at the Earth-Moon L2 libration point rather than in low lunar orbit. I let readers absorb the argument for a few paragraphs before revealing the punchline: the article wasn’t coming out next month. It had come out in June 1972 — in Astronautics and Aeronautics — and every word of it applied just as well to 2006 as it had to the Apollo era.

That thread ran for over a hundred pages and several years. I’ve been meaning to write up the core argument in one place ever since. Here it is.

The Problem Nobody Wanted to Talk About

The ESAS architecture — the architecture that became Constellation — chose lunar orbit rendezvous as its mission mode. The CEV (later Orion) would wait in a low lunar polar orbit while the LSAM (later Altair) descended to the surface, ascended, and docked back up. Apollo did it this way, so ESAS did it this way.

The problem is that Apollo only ever landed near the equator. ESAS was explicitly required to provide global access to the lunar surface — including the poles, which is where the water ice and the scientifically interesting terrain are — combined with anytime return capability, meaning the crew could lift off from the surface at any time and get back to Earth without waiting for favorable orbital geometry.

Those two requirements together are what break LOR. Farquhar understood this in 1970. NASA’s own documents from that era said it plainly:

“For landings at higher lunar latitudes or longer stay times, rather large plane changes are often necessary, and the DV cost for these plane changes is significantly higher if lunar orbit rendezvous is employed. In some instances, this increased DV penalty is severe, and large portions of the lunar surface are not accessible with the present Apollo spacecraft. On the other hand, the DVs for the libration-point rendezvous technique are virtually identical for any lunar landing site or stay time because the necessary plane changes can be accomplished quite cheaply at the libration point.”

— Robert Farquhar, NASA TR-R-346, 1970

ESAS acknowledged the same problem in its own report. Page 184 states that the CEV service module requires 1,450 m/s of delta-V — 900 m/s for trans-Earth injection plus a worst-case 90-degree nodal plane change. Even with this, “loitering on the surface at the outpost for up to 14 days may be required” for non-polar, non-equatorial sites. A Flight International article from October 2007 later revealed that the LEO loiter before trans-lunar injection could itself stretch to six days just to get the nodal geometry right for a global-access mission.

So the architecture chosen in the name of “anytime return” and “global access” didn’t actually deliver either one cleanly — and it cost an enormous propulsive penalty to come close. The iron triangle of LOR is this: you cannot simultaneously achieve anytime return, global access, and minimal delta-V. ESAS tried to have all three and paid for it with 1,450 m/s of service module propellant and multi-day surface loiters. EML2 rendezvous delivers all three as structural features of the trajectory.

What Makes EML2 Different

The Earth-Moon L2 point sits about 61,500 km beyond the Moon on the far side, away from Earth. If you’re reading that and thinking “that sounds like it should be harder to get to than lunar orbit,” you’re having the intuition that everyone has — and the intuition is wrong.

The reason is a trajectory technique Farquhar called the powered lunar swingby. Rather than braking directly into L2 (which would cost about 1,100 m/s), you instead approach the Moon on a trajectory that swings around in front of it. At perilune — closest approach on the near side — you execute a relatively small burn of less than 200 m/s. That burn, placed at the most propulsively efficient point in the trajectory, sends you coasting outward to L2, where another small burn of about 150 m/s captures you into a halo orbit. Round trip, you’ve spent about 330 m/s going each way — roughly 660 m/s total for the CEV to get to L2 and come home.

Compare that to the ESAS baseline of 1,450 m/s. The table below puts the options side by side:

Architecture Staging Point CEV ΔV (m/s) Notes
ESAS LOR baseline Low Lunar Orbit 1,450 TEI + 90° plane change
Direct to EML1 Earth-Moon L1 1,400 ~700 m/s each way
Direct to EML2 Earth-Moon L2 1,100 ~550 m/s each way
Swingby to EML2 Earth-Moon L2 660 ~330 m/s each way via powered lunar swingby

That 800 m/s saving on the CEV, in propellant terms, means the service module gets about 60% smaller. That’s not a marginal improvement — that’s the difference between a vehicle that closes on its launch vehicle and one that doesn’t. Closing on the Ares I was exactly the problem that plagued Orion throughout Constellation’s existence.

From a halo orbit around EML2, the geometry to the lunar surface is essentially constant. The launch window from any point on the Moon to L2 is, for practical purposes, always open — you don’t have to wait for the orbital plane to precess over your landing site. You land when you want, you come back when you want. Global access and anytime return aren’t engineering compromises in this architecture; they’re structural features of it.

The other thing that changes is the mass coupling. In the ESAS LOR scenario, the LSAM descent stage had to brake itself and the CEV into low lunar orbit — every kilogram of Orion growth cascaded into LSAM descent propellant, which cascaded into descent stage mass, which cascaded into EDS propellant, which cascaded into Ares V performance. In the L2 architecture, the LSAM and CEV separate before lunar arrival and each handles its own burns independently. Mass growth in one vehicle no longer automatically punishes the other.

The ESAS Study’s Blind Spot

ESAS did look at libration point rendezvous and rejected it. The executive summary concludes: “LPR was eliminated early from the mission mode trade space. Recent studies performed by NASA mission designers concluded that equivalent landing site access and ‘anytime abort’ conditions could be met by rendezvous missions in LLO with less propulsive delta-V.”

This conclusion was wrong — and wrong for a specific, traceable reason. The ESAS team only considered direct trajectories to L1 and L2. On that basis, direct-to-L1 costs ~710 m/s each way and direct-to-L2 costs ~550 m/s each way — neither obviously superior to LOR when analyzed naively.

The powered lunar swingby was simply never evaluated. Once you include it, the L2 insertion cost drops from ~1,100 m/s to ~330 m/s each way, and the comparison reverses entirely. L2 rendezvous was never defeated on its merits. It was eliminated on a straw-man comparison against a trajectory that nobody would actually fly.

For the record, this result was not new information in 2005. T.N. Edelbaum’s 1970 NASA study (CR-86337) concluded explicitly:

“It was found that the lowest delta-V requirement to reach either libration point was to go to the L2 point behind the Moon via a powered lunar swingby… The mass requirement was found to be smaller than that for the standard lunar orbit rendezvous mode. Rendezvous at either the L2 point, or in a Halo Orbit about the L2 point, would also have operational advantages, including access to all points on the Moon and an infinite rendezvous launch window.”

— T.N. Edelbaum, NASA CR-86337, 1970

This analysis was done in 1970. It was confirmed by Farquhar in 1971 and 1972. Neither result was carried into ESAS thirty-five years later.

L1 vs. L2

People often ask why L2 rather than L1, since L1 — the point between the Earth and Moon — seems more “on the way.” The answer comes down to the powered swingby.

A direct trajectory to L1 costs ~710 m/s each way. There is a lunar swingby trajectory to L1, but it takes three weeks to fly and passes through L2 on the way. The best indirect trajectory to L1 still costs 535 m/s and takes 12+ days. L2 via powered swingby: 330 m/s, 8–9 days. For every purpose — crew transfer, cargo delivery, lander staging, Mars departure — L2 beats L1 on propellant.

L1 has a shorter direct transit time (four days vs. eight), which led some people to suggest it as a “training wheels” stepping stone. But putting a crewed architecture at EML1 would actually increase the DV burden on both Orion and the lander compared to ESAS LOR, so it’s not a stepping stone — it’s a step backward dressed up as prudence. Only the powered swingby to EML2 delivers the mass savings where they’re needed.

EML2 as an Interplanetary Node

The lunar applications of EML2 are compelling, but the interplanetary case is what I find truly exciting — and it’s where the non-intuitive nature of three-body astrodynamics pays the biggest dividend.

Every interplanetary mission from Earth requires targeting a departure asymptote: a specific direction and speed of departure that puts you on the right trajectory to your destination. From the Earth’s surface, you hit this by choosing your launch date, azimuth, and upper-stage performance. From LEO, it gets complicated. The ascending node of your orbit drifts at only 2–4 degrees per day due to Earth’s oblateness. If your orbit is in the wrong nodal orientation when the optimal launch window opens, you either miss the window or pay an enormous penalty to reorient. For an interplanetary vehicle assembled in LEO over multiple launches, this is a serious operational problem.

From EML2, the problem largely disappears. The Moon completes one orbit around Earth per month, sweeping through all 360 degrees of right ascension at about 12 degrees per day — three to six times faster than nodal regression in any Earth orbit. By choosing your departure time from L2, you can target essentially any right ascension. And the energy required to kick from L2 into a highly elliptical Earth orbit for a perigee burn is only ~300 m/s — after which the Oberth effect at perigee handles the rest of the trans-Mars injection efficiently.

For low-thrust propulsion systems — nuclear electric, solar electric — EML2 is even more attractive: it sits on the very edge of Earth’s gravity well, meaning spiral-out time before escaping to the solar system is measured in days rather than weeks or months.

And if you’re serious about using lunar resources to fuel interplanetary missions — lunar-derived oxygen launched off the surface and transferred to a Mars-bound vehicle — EML2 is the place to collect it. Bringing LOX to LEO would waste all its orbital energy just to capture it into a lower orbit, then burn it again later to escape. Keeping it at EML2 preserves every bit of that energy for trans-Mars injection.

What It Would Have Taken

The LSAM hadn’t been seriously designed when these arguments were being made in 2006 and 2007. The window was open. The required vehicle changes were not exotic:

  • Orion service module: ~60% smaller propellant load — solving the Ares I mass closure problem outright, with mass margin left over for radiation shielding.
  • LSAM descent stage: Smaller, because it no longer brakes the CEV into LLO — the CEV does its own swingby burn independently.
  • LSAM ascent stage: Larger, requiring ~670 m/s more to reach EML2 rather than LLO, plus extended power and life support for a 2–3 day transit.
  • Bonus: Each captured ascent stage at EML2 becomes a permanent orbital asset. Accumulated over multiple missions, they form the nucleus of an L2 gateway station at essentially zero additional cost.

None of this required a new launch vehicle. None of it required exotic propellants. The change was architectural, not technological. And yet it was never seriously evaluated.

Why It Matters Now

I’m writing this not to relitigate Constellation — that program is long gone — but because the underlying astrodynamics haven’t changed. Anyone designing a lunar architecture today faces exactly the same iron triangle: you cannot simultaneously achieve anytime return, global access, and low delta-V with lunar orbit rendezvous. Any architecture that commits to all three and uses LOR will either quietly fudge the requirements or be grotesquely over-massed.

EML2 is still there. The powered lunar swingby trajectory still works. The 330 m/s insertion cost is still real. The halo orbit still has an always-open launch window to any point on the lunar surface. None of the physics has changed.

If you’re building a sustainable human presence in the Earth-Moon system — with reusable landers, eventual lunar propellant production, and a pathway toward the rest of the solar system — there is no better place to put your staging node than EML2. Farquhar knew it in 1972. Edelbaum proved it in 1970. The math says it now.

I wrote a three-body astrodynamic simulation to convince myself. The trajectories are not intuitive, and I think that’s exactly why this good idea keeps getting left on the shelf. Astrodynamics are often non-intuitive. Three-body astrodynamics are extremely non-intuitive. That’s no excuse for not doing the math.

Posted in ESAS, Lunar Exploration and Development, Space Development, Space Exploration | 3 Comments

Starliner Reponendum Est? (Part II): Why Boeing’s Troubled Capsule Still Needs to be Replaced

Posted on March 4, 2026 by Jonathan Goff

[Note: I started writing this almost a year ago, but got stalled out. I wanted to finish writing this while it was still potentially relevant.]

I’ll be honest, I was hoping to not have to write the rest of this series. When Boeing’s new CEO started making noises about potentially selling off the parts of Boeing’s Space and Defense division, over a year ago now, I thought this series would be OBE1, allowing me to focus on funner, less spicy topics. Unfortunately, that so far hasn’t panned out, and in spite of details that emerged about how close of a call this last flight really was, NASA and Boeing started to roll out the beginnings of a plan for certifying Starliner without actually requiring it to successfully fly a complete crew flight test mission…

It was telling that NASA, under the previous administrator, was seriously thinking about “certifying” Starliner, and allowing its next mission to be a “post-certification flight”, while still talking about flying just cargo on-board for the next mission. Basically saying “We’ll do more testing and analyses and paperwork, say that you’re certified so you can bill for the remaining pre-certification milestones, then we’ll pay you something on the order of $320M to fly less cargo than a Dragon, because yeah right, who are we kidding thinking our astronauts would actually trust flying in this thing before it has at least one relatively flawless flight?”

That really felt like desperation on NASA’s part, or Boeing telling NASA that they’ll walk from the program if forced to fly another pre-certification mission on their own dime. Or maybe a bit of both.

The new administrator’s decision to label the CFT mission last year as a Type-A Mishap, and his indictment of how NASA handled this whole fiasco gives me some hope that maybe things will start heading in a reasonable direction. But I still think that almost everyone would be better off if NASA just “shoots the puppy”2 and uses the leftover money from canceling Starliner to fund development on a more reliable and more economically competitive replacement.

Why Almost Everyone Would Be Better Off if NASA Terminated Starliner (and Competed a Replacement)

If you think about the various groups that are impacted by NASA’s decision to stay the course with Starliner vs terminating it and replacing it, it’s striking that almost everyone (NASA, Boeing itself, CLD Developers, SpaceX, and even SpaceX’s competitors) would be better off if NASA pulled the plug on Starliner and started work on a replacement.

Why NASA Would Benefit

The main reason NASA hasn’t pulled the plug yet has been because they want assured access to ISS. They don’t want a situation where Dragon or Falcon 9 being grounded means they might have to temporarily abandon their part of the station. I think there’s some merit to the desire for dissimilar redundancy, but the reality is that the redundancy Starliner brings would be both temporary at best and potentially illusory.

First, the redundancy it provides will at best help for only the final few years of ISS operations. There’s many reasons to believe that even if Starliner flies through the end of ISS operations, that it won’t ever fly beyond that:

  • Right now, Starliner is designed to fly on Atlas V, which is no longer in production. After they fly the six Atlas V’s set aside for Starliner, they’d have to reintegrate Starliner with a new launch vehicle like Falcon 9 or Vulcan. Unlike Cygnus though, this will likely be a lot more complicated, since Starliner flies on the outside of the rocket instead of in a fairing, which likely means that adapting it to a new launch vehicle will require extensive analyses, testing, and potentially even some major structural redesigns.
  • Even without having to cover large NRE expenses to keep it flying post-ISS, Starliner still costs on the order of $320M per mission for four people, yielding a per-seat price that’s well over 50% higher than Dragon. CLD operators are way more cost sensitive than NASA was with ISS. Can any of them actually close their business plans if they have to regularly pay $80M/seat to get people to/from their station? Will they get any customers at that price?

If my hunch is right that Boeing wanted to back out of Starliner if forced to fly another test flight, that suggests how likely they think it is that anyone will buy another flight beyond the ISS missions they already have secured from NASA. Which basically means that once NASA transitions to being a major customer of station services in the CLD Era, they’ll be right back to a situation where they don’t have crew redundancy anymore, because nobody else can afford to keep Starliner on life support.

Second, I think that even if you’re only looking at the final few years of ISS operations, the redundancy value of Starliner is probably illusory at best. If Boeing and its employees and shareholders know that Starliner’s days are numbered, how well will they really support it through the end of operations? Given how much money they’ve lost already, will they really avoid the temptation to start cutting back on the team as soon as they can. Getting rid of or reassigning people who aren’t strictly necessary for operating those last few flights? Won’t their most talented and ambitious employees likely seek work elsewhere? If Dragon actually did have an issue, and they needed to accelerate Starliner’s availability to make up for that, would Boeing actually have the manpower to accelerate its schedule enough to matter? Especially as they’re trying to control costs on a program that has already cost them their shirt?

Finally, independent of arguments about redundancy, NASA would benefit from replacing Starliner because it increases its odds of the CLD facilities being successful, which will allow it to free up more of its resources to focus on beyond earth orbit human spaceflight3. But more on that in a moment.

Why Boeing Would Potentially Benefit

Frankly, Boeing has lost a lot of money on Starliner, and my guess is the only reason they haven’t walked away from it at this point is that NASA is probably putting pressure on them to stay in the game. They have a ton of house-cleaning they need to do to rebuild their capabilities and reputation as a reliable airframe manufacturer, and frankly Starliner is a distraction to that4. Honestly, I’m wondering if now that a Type-A Mishap5 has been declared, if Boeing isn’t regretting that they didn’t walk away from Starliner sooner. They now have an even worse black eye, and NASA almost certainly isn’t going to be able to let them skate by without properly and convincingly fixing Starliner’s problems. There’s a good chance that working through all of that could take another year of added costs, and cost them the opportunity for at least one Starliner flight to the ISS to offset them.

I’m not an expert in large government contracts6, so I’ve been confused about why Boeing hasn’t backed out yet, given how much Starliner continues to cost them. It could be that since they’ve started getting payments for milestones toward their first few operational flights, that backing out might require them to pay those milestones back? Or maybe they’re worried if they quit, it would give them severely negative grades on contract performance7? Or maybe NASA was cajoling them in some other way to stick with the program8. Maybe it’s something else, but if it is something like having to repay milestones, I think NASA would be better off if they gave Boeing a “golden bridge to retreat across”9 by waiving those fees if they walk away now.

CLD Operators

The CLD developers are probably some of the clearest beneficiaries of if Starliner was canceled and recompeted. As discussed in the previous post, CLD business plans live or die on affordable space transportation. They need not just a reliable second provider, but one that is economically competitive with Dragon. Right now NASA wants to scale back its LEO investment post ISS to ~$1B/yr, to free up the remaining ~$3B/yr for Artemis and other beyond LEO activities. If they wanted to keep two CLD providers alive, and they want continuous operations, and they limited crew rotations to six-month stints, Crew Dragon flights would take up almost that entire budget10. That’s not impossible to wrap a business model around, but very challenging. Having NASA invest in one or two more serious crew providers that are capable of competing with that price point would go a long way to ensuring CLD operators have a fighting chance of success.

And as discussed before, does anyone seriously think Starliner would keep flying beyond ISS? Given that it has proven to be far less reliable than Dragon, and costs almost twice as much per crew seat? And if they did by some miracle keep flying, how much will it cost to requalify Starliner to fly on a new launch vehicle once they run out of AtlasVs? There is practically zero value to the CLD providers to keeping Starliner on life support through the end of ISS. They would be unequivocally better off by having it canceled and replaced with something that actually would be available when they need it, especially if it’s economically competitive with Dragon.

Note, I haven’t said anything about Crew Starship, because in reality I don’t think it’s likely to be any more affordable for CLD operators, at least not without competition. Most of them aren’t building stations that need a dozen crew delivered at a time, and there’s no indication that SpaceX would actually lower their seat prices without real competition11.

SpaceX Competitors

This one is obvious. If Starliner gets canceled and replaced, someone is going to win that contract. There are several up-and-coming launch providers, many of whom have demonstrated more capability than SpaceX had when they won the initial COTS and CRS-1 contracts for Dragon, and a ton of groups developing cargo reentry vehicles that could potentially benefit from a recompete of Boeing’s Commercial Crew slot. With Falcon 9 being retired in favor of Starship, this would also likely lead to launches for whichever of the up-and-coming medium lift vehicles shows the most promise.

SpaceX

SpaceX would be an obvious beneficiary as well. Starting a new commercial crew provider from today would mean that NASA would likely need to tap Crew Dragon for the remaining ISS flights12, which would likely be another $800M-1B in revenue.

While a world where Crew Dragon or Crew Starship have real competition would mean lower margins per flight, a more successful CLD market is one that’s more likely to grow to the point where it could start using Starship services in the future. Also, a more successful CLD market is one that increases the odds of NASA being able to shift more resources from ISS to beyond-LEO human spaceflight, of which SpaceX would likely be a major beneficiary. Finally, a commercial LEO marketplace that is solid and growing probably helps make all of SpaceX’s long-term visions easier to execute — LEO Data Centers, Lunar settlement and ISRU mining, and large Mars missions are all easier to make happen if there’s a thriving LEO ecosystem that’s not 100% dependent on what SpaceX is spending money on.

There Are A Few Who Wouldn’t Be Better Off Though…

The group that loses out the most from Starliner going away is going to be Boeing’s Starliner employees, many of who would likely be at least temporarily out of a job if Starliner went away. This to me is sad, because frankly most of the worst issues with Starliner were driven by management decisions, not by the rank and file engineers. They did their best to try and solve problems within the constraints of a publicly traded engineering company that was driven by people at the top who only really knew how to do financial engineering. Many of these employees had good paying, stable jobs, with good work/life balance. Losing that is a real loss for them, and potentially their communities, and shouldn’t be glossed over, even if canceling Starliner would be the right decision for the industry.

There’s a chance ULA might be worse off as well, if they’re not able to repurpose and resell those remaining Atlas V tail numbers that were earmarked for Starliner.

The Precedent of the COTS RpK Cancellation and the Importance of “Opportunity Costs”

Pulling the plug on a non-performing commercial services provider is not without precedent. The initial COTS program selected SpaceX and Rocketplane Kistler to develop commercial cargo delivery solutions for station. RpK’s solution was more ambitious than SpaceX’s13, but was tied to a development approach that was way more expensive, and ultimately they were unable to raise the money needed for some of the financial milestones in their COTS Space Act Agreement. They did renegotiate milestones a few times, but they were never able to close the funding. Rather than keep RpK on the hook forever in the hopes of funding eventually coming in, or restructuring the contract in some way to give them additional help by say frontloading payment milestones or something like that, NASA decided to pull the plug on the contract in September of 2006. And they recompeted the slot, to get another commercial cargo provider — ultimately resulting in the Cygnus cargo vehicle, which provided NASA with a reliable, and reasonably affordable cargo system to go along with Cargo Dragon.

Canceling the RpK contract, and recompeting it was probably one of the smartest moves NASA made in the COTS program. It showed that there were consequences for not delivering. It gave another provider a chance to replace them, resulting in two reliable and reasonably affordable14 cargo providers. One of the best incentives to execute well is knowing you can be “fired for incompetence”15.

Opportunity Costs are Real Costs

Much like with the current situation with Starliner, in theory, NASA could’ve said that the RpK contract was a firm-fixed price contract, so there was no cost to NASA if they continued to give RpK time to try and make something happen. But the reality is that there’s a huge opportunity cost16 to leaving a non-delivering provider on contract like this, rather than cutting them off after they’ve had a reasonable chance to remedy things. Had NASA taken that “it doesn’t cost us anything to do nothing” approach with COTS, they almost certainly would’ve only ended up with Dragon as a commercial cargo provider, much like the situation we’re likely to see if they keep Boeing’s Starliner on life support. But more than that, the opportunity cost of not canceling RpK and recompeting includes not getting the Cygnus cargo vehicle17, or the Antares family of launch vehicles including the new Eclipse launch vehicle18 that’s about to have its debut flight. The opportunity cost would’ve likely included the ability to launch cubesats from the ISS before smallsat dedicated launch and eventually SpaceX rideshare came online19. Without those rides, would Planet Lab and Spire have been able to make it as companies? Without Cygnus, would Thales Alenia still have an active line building space station pressure vessels? There are a ton of things that wouldn’t have happened if NASA had chosen inaction on RpK. We can only really see most of those because NASA didn’t sit on its hands. What opportunity costs are we paying for NASA not terminating and recompeting the Starliner contract?

Basically, getting rid of non-performers frees up resources to shift them to more competent organizations, and also puts people on warning that there are real consequences to not delivering20.

But Isn’t It Too Late To Pull the Plug?

It’s true that the best time to pull the plug probably would’ve been right after the CFT debacle over a year ago. Or maybe even before then. But even with the ISS potentially retiring in 2030, it isn’t too late, because frankly the Commercial Crew program was never just about meeting NASA’s needs with ISS. Sure, it’s highly unlikely at this point that they could recompete the contract, get someone spooled up, and have them get all the way through development, NASA’s crew certification process21, and flight demonstrations before 2030. But as mentioned before, while NASA isn’t operating the CLD stations, its ability to shift resources and its focus to Artemis is strongly correlated with how successful the CLD stations are.

The best time to pull the plug on Starliner was over a year ago, the next best time is now.

[Next up: Considerations for how to recompete the Commercial Crew contract, who might be potential competitors, and other valuable lessons learned from other programs. Hopefully this last one won’t take as long for me to get around to writing.]

Posted in Commercial Crew, Commercial Space, COTS, Launch Vehicles, NASA, Space Safety, SpaceX, ULA | Tagged , , , , , | 1 Comment

Space-based Bitcoin Mining, Revisited for 2025

Posted on November 14, 2025 by Chris Stelter

Almost 9 years ago, I blogged about orbital data centers for bitcoin mining. https://selenianboondocks.com/2016/12/space-based-bitcoin-mining/
I wrote:

At 4 year replacement cycles (probably the power and thermal system would last a lot longer, but hardware would need to be updated), that’s about 11 cents per kWh of bitcoin processing, which is competitive (since it includes the mining hardware as well as the power). At longer intervals, even better.

But of course, this assumes bitcoin remains popular WHILE bitcoin mining hardware stagnates. The latter is not a terrible assumption, given that we’ve essentially reached the end of Moore’s Law (we’ll still see large improvements, but over longer timeframes, like 4 years instead of 2 years)…

And lo and behind, I think we’re just about there. At the time, Elon had just announced ITS “Interplanetary Transport System”, it was still carbon fiber, and had not morphed into the 9 meter diameter stainless steel Starship of today which has done 11 orbital (okay, okay, near-orbital) launches. Falcon 9 only did 8 launches that year (about 170 launches so far the last 365 days from writing), and it was before Falcon Heavy flew. Starlink had not launched at all (not even the 2 simple demo satellites, in 2018), let alone the ~100MW, ~10,000 satellite constellation of today.

I decided to double-check the efficiency of miners today to see if efficiency improvements have slowed. I checked a few sites. None had the release dates in a consistent format, so after checking a few websites manually, I cheated and used Gemini (Google’s AI) to make a list of the efficiency and release dates of the various Antminer ASICs. (The S1 is almost double that value, but if you undervolt it, you get ~1100J/TH. My spot-checks have confirmed these rough values for release date and efficiency.)

Model Release Date (approx.)Efficiency (J/TH)
Antminer S12013~1100
Antminer S3May 2014~770
Antminer S5Dec 2014~320
Antminer S7Aug 2015~250
Antminer S9May 2016~100
Antminer S15Nov 2018~50-60
Antminer S17Apr 2019~40+
Antminer S19 ProMay 2020~29.5
Antminer S19 XPQ3 2022~21.5
Antminer S21Jan 2024~17.5
Antminer S21 XP HydroQ4 2024~12
Antminer S23 HydroQ1 2026 (est)~9.5-9.7

I couldn’t get Gemini to plot these form me, so I had to manually enter it into Libre Calc to get:

dateJoules/Terahash
2013.01100.0
2014.3770.0
2014.9320.0
2015.6250.0
2016.3100.0
2018.855.0
2019.340.0
2020.329.5
2022.621.5
2024.017.5
2024.912.0
2026.1259.6

And from there, the graph:

I added two exponential trendlines to this logarithmic plot, and you can see there IS, in fact, a slowing of the efficiency improvement over time, as I had speculated in the previous post there would be.

Converting the exponential fit into a doubling time (take the reciprocal of the exponential constant and multiply by ln(2)… or just squint at the graph), and you get:

date rangedoubling time
2013-20181.26years
2019-20263.48years

Since I wrote that blog post, the doubling time for Hashes-per-joule went from around 15 months to 42 months. That’s pretty much at the 4 year hardware replacement cycle goal I had mentioned. Just as launch vehicles are developing which could reach the $100/kg launch costs (maybe even prices) that I mentioned, Moore’s Law (at least for mining) has slowed enough that orbital data centers for bitcoin mining are starting to make sense. Now I look at the $750 per Bitcoin price in that blogpost and cry… (As of writing, one Bitcoin is almost $100k… I have never owned any cryptocurrency.)

That doesn’t necessarily mean orbital datacenters make sense for everything. Bitcoin (at least slightly old hardware) has about a factor of 100 less cost per watt than high end AI training systems. That might change as NVidia’s margins get eaten by large companies realizing they don’t have to pay the Jensen tax forever, and the idiot factor for high end GPUs comes down a few pegs.

Posted in Uncategorized | 2 Comments

Ambition, Resources, Competence: Why SpaceX Doesn’t Yet Have Many Competitors

Posted on August 30, 2025 by Jonathan Goff

I read a tweet today that kind of rubbed me wrong, about SpaceX’s current market dominance:

At least one friend asked him if he really thought that nobody else had tried, and he pointed to how both ULA and Arianespace had written off reusable launch… while that’s true, it misses a much bigger picture, and it got me thinking about a framework for thinking about space companies that a mentor had once explained to me.

Ambition and Resources as Comparative Metrics

This mentor, Dylan Taylor, was explaining to me his hypothesis for a new roll-up company he was about to setup22. Basically, he said that you could compare different space companies on two axes: resources and ambition. Resources was a combination of team, facilities, skills, and access to capital. Ambition was about how focused they were on changing the world and bringing new capabilities to market.

The challenge as he put it is that most space companies tend to cluster in the upper left and bottom right corners of this chart. In the upper left, you have a lot of technical founder-led space startups with tons of ambitions, and sometimes great ideas, but frequently with very poor access to capital and resources for various reasons. Without resources it’s hard to convert good ideas into capabilities. On the opposite corner, you have a lot of the traditional aerospace firms, that tend to have huge teams, with a lot of talent and financial resources, capable of developing very complicated and exquisite technical solutions, but they have almost no ambition to do anything beyond winning the next government contract, and making sure their stock price goes up. There’s also technically some in the bottom left-hand corner, that don’t have a lot of ambition or resources. Some of these are good companies, that do valuable things for the economy, but they often get denigrated as SBIR farms or “lifestyle companies”23. I’ll use the term Small Businesses instead.

His point though was that there were very few companies in the upper right-hand corner that have both the resources to do big things, and also the ambition to seriously change the status quote and create new capabilities. SpaceX being one of the best examples in this category.

To complete the story, his hypothesis was that he could create a roll-up that could aggregate several founder-led startups, while preserving the entrepreneurial spark and ambition that made those startups worth buying in the first place. The goal would be to create an organization that had both the ambition and drive of those smaller startups, while gathering the resources needed to go after big, world-changing capabilities. It was a really compelling vision, and when he ultimately launched a company to do that, I sold my startup Altius Space Machines to them. While I wouldn’t say he’s achieved his goal yet, they did go public on the New York Stock Exchange24 earlier this year, and I sincerely hope they or someone else can make that model work, because I think the original hypothesis had a lot of value.

Competence as a Third Axis

But back to this framework. One thing I think he was missing from his framework was a third axis of competence, basically how effective is a company at turning what resources it has into results.

For instance, there are some companies that do an amazing job of turning modest resources into impressive results, punching way above their weight class, like my first startup Masten Space Systems, which I think in its whole existence raised less than $5-10M in investment. There’s a reason why Elon pivoted SpaceX from trying to do parachute splashdown recovery to powered landing after we posted our Xombie in-air relight video on Youtube. As he supposedly put it to his team25 “This was five @#%!ing guys in a shed in Mojave! If they can do this, we can do this.” They totally knocked it out of the park, but it was because of seeing what scrappy companies like Masten and Armadillo Aerospace26.

On the other hand, while I won’t name names, most of us can think of other space companies that have had a ton of resources, some big ambitions, and yet have very little results to show for it27. SpaceX isn’t the only company with high ambition and high access to resources, it’s just one of the few in that category that’s also shown a high degree of competence, at least until very recently.

Unfortunately, the ability to raise a lot of money, or the ability to earn a lot of money from previous businesses in totally unrelated fields, is often poorly correlated with being a good founder for an ambitious space startup. There are some people who are good at raising money, and good at putting that money to productive use, but a depressingly large amount of the money that’s available in the space community has landed in the hands of those who aren’t particularly effective at converting money into bringing new capabilities to market.

What All of This Means for SpaceX Competition

Ultimately, I think Edward’s original tweet misses some important points. There absolutely are companies that have tried to compete with SpaceX. The challenge is that until recently, almost none of them have had the full combination of ambition, resources, and competence. He correctly points out that there are those who have had the resources, but lacked any ambition. We’ve also seen a few that have had the resources and ambition, but frankly haven’t been very competent. What’s less appreciated has been the companies with ambition and competence who either haven’t or have only recently received enough resources to make a difference. It doesn’t matter how ambitious your goals are, or how effective you are at turning resources into results if you can’t get access to resources. Some groups are starting to crack this nut, most notably RocketLab, Stoke, Firefly, and a few others, but only very recently. And it takes a long time to go from resources to results when it comes to launch vehicles or other ambitious space projects28.

One strong contributor to this has been the very real attitude among VCs over the last ten years that you shouldn’t fund companies trying to compete with SpaceX. Talk about a way to create a self-fulfilling prophecy! If most VCs are too chicken to fund someone trying to compete in the same marketspace as SpaceX, it’s not surprising that SpaceX’s main competition so far has been mostly companies that already had access to large resources, either from legacy aerospace businesses, or from other industries, very few of which came with the necessary ambition or competence to effectively compete.

Another related contributor has been that government agencies also haven’t been very willing, until fairly recently, to take the kind of chance on other space startups that NASA did on SpaceX with the COTS and CRS programs. They totally made the right bet, but at the time SpaceX won COTS they hadn’t yet successfully flown anything to orbit, and by the time they won the >$1B CRS-1 contract, they had only flown successfully to orbit once as a company. Since COTS and CRS-1 there have been very few NASA or DOD efforts that have placed those sizes of bets on commercial companies. One of the few recent examples I can think of, where the government took bets of a similar scale on relatively new, but promising space startups has been SDA’s satellite constellation tranches. There have been smaller and medium bets with programs like STRATFIs29, VADR30, and CLPS31, but large bets where a company could get a shot at transformational levels of funding if they successfully deliver have also been extremely rare. Once again, when you refuse to retry the formula that helped give SpaceX a chance to become who they are today, government agencies shouldn’t be shocked that it’s taken a long time for others to follow in SpaceX’s footsteps.

As I said, this dynamic is finally starting to shift, and impressive ambitious and competent companies like RocketLab32, Stoke, Firefly, and others are starting to get the resources they need to actively compete in the launch arena. If a few of them start succeeding, hopefully it’ll show to investors that yes it is possible for other companies to develop reusable launch vehicles, and other advanced space capabilities, if actually given a chance.

Anyhow I thought it was worth sharing this framework for thinking about companies, and how it explains a lot of why SpaceX has only recently started getting competent and well-funded competitors. I think it’s critical for the long-term success of our industry that we have multiple successful startups, and real competition, and I’m glad that we’re finally seeing some other ambitious, competent startups get the access to capital and resources needed to truly compete.

Posted in Business, Commercial Space, Launch Vehicles, Space Transportation, SpaceX | Tagged , , , , | 5 Comments

The end of DRACO, the most recent NTR effort

Posted on June 30, 2025 by Kirk Sorensen

DARPA’s DRACO nuclear propulsion project ROARs no more

The Quiet Death of DRACO — and What It Tells Us About Nuclear Space Propulsion

DARPA has cancelled its DRACO program — the Demonstration Rocket for Agile Cislunar Operations — ending the most recent serious U.S. government effort to develop a nuclear thermal rocket (NTR) engine for in-space propulsion. The cancellation deserves more attention than it has received, because it reflects a genuine and important shift in the strategic calculus around space nuclear power.

What Was DRACO?

DRACO began life in 2020 with $10 million and the original name “Reactor on a Rocket,” or ROAR — a moniker agency scientists later decided might attract negative attention. The goal was development of a high-assay low-enriched uranium (HALEU) nuclear thermal propulsion system capable of demonstrating meaningful advantages over chemical propulsion for deep-space and cislunar missions. NASA joined the effort in 2023, with NASA taking responsibility for the engine development while DARPA focused on the space vehicle.

The case for NTR has always rested on straightforward physics. Nuclear propulsion offers thrust-to-weight ratios around 10,000 times greater than electric propulsion, and two to five times better efficiency than chemical propulsion systems. For missions requiring high delta-V — Mars transit, cislunar maneuvering, rapid repositioning in space — those numbers are compelling on paper.

Why It Was Cancelled

The explanation offered by DARPA deputy director Rob McHenry is worth quoting at length, because it’s unusually candid for a government official discussing a program cancellation. When DRACO was originally conceived, it predated the significant reduction in launch costs driven largely by SpaceX, and the analysis at the time showed nuclear thermal as likely to be the optimal solution for both national security missions and solar system exploration. Over the execution of the program, both of those assumptions grew weaker.

In other words: the efficiency advantage of NTR is real, but the value of that advantage is relative. If you can launch enough propellant cheaply enough with reusable chemical rockets, the enormous R&D investment required to develop and certify a flight-ready nuclear thermal engine starts to look like a losing proposition. McHenry noted that the national security operational interest in the technology was decreasing proportionally to the perception of its differentiated value.

This is a genuinely important insight, and one that cuts across many advanced propulsion discussions. Nuclear thermal propulsion doesn’t compete against chemical propulsion in the abstract — it competes against chemical propulsion at a given cost point. When launch costs fall dramatically, the economic and strategic case for exotic propulsion alternatives weakens accordingly.

The Pivot to Nuclear Electric

What’s perhaps most interesting about the DRACO cancellation is where the attention is now turning. Rather than abandoning nuclear space power entirely, DARPA and NASA have concluded that nuclear electric propulsion is probably a more optimal long-term solution — not only for deep space exploration but for national security missions, where power availability in the space domain may be as critical an enabler as propulsion efficiency.

This reframing makes sense. A compact fission reactor aboard a spacecraft eliminates the need for large solar panels — which are heavy, fragile, and increasingly a vulnerability in contested space environments — and can power electric propulsion systems, sensors, communications, and directed energy payloads simultaneously. The Air Force Research Laboratory’s Joint Emergent Technology Supplying On-orbit Nuclear Power (JETSON) program, awarded to Lockheed Martin, Westinghouse Government Services, and Intuitive Machines in 2023, is exploring exactly this concept.

For the national security space community, the appeal of abundant, reliable, solar-independent power in orbit is obvious. A Space Force asset with a nuclear electric power source can maneuver, communicate, and operate sensors far more aggressively than one dependent on solar arrays — and in a contested environment, that matters enormously.

A Pattern Worth Noticing

The death of DRACO fits into a longer pattern in nuclear propulsion history. The original NERVA program in the 1960s and early 1970s produced ground-tested engines of remarkable capability — and was cancelled in 1972 when the budget calculus and strategic priorities shifted. Every subsequent NTR effort has followed a similar arc: initial enthusiasm, serious technical work, then cancellation before any hardware ever flies in space.

The reasons vary from cycle to cycle, but the underlying dynamic is consistent: nuclear space propulsion requires a long, expensive development commitment, and that commitment has never survived contact with changing political, budgetary, and strategic environments.

Whether nuclear electric power will prove more durable remains to be seen. The mission case is arguably cleaner — a reactor powering spacecraft systems is easier to conceptualize and justify than one serving as the primary propulsion source — and the JETSON program represents a more modest initial step. But the history of nuclear space programs counsels humility about timelines and outcomes.

What’s clear is that the conversation has shifted. The question is no longer whether nuclear technology belongs in space, but which form of it makes the most strategic and economic sense. That’s a more productive question, and the answer may ultimately point toward something more transformative than either chemical rockets or nuclear thermal engines could provide on their own.

Posted in Uncategorized | 2 Comments

The Moon as a Preferred Off Earth Settlement

Posted on February 26, 2025 by johnhare

Many people think that Mars is the primary target for off Earth settlement and development. I think that is likely to prove wrong as people start the process of moving onward. Skipping the moon to go to Mars strikes me as similar to Europeans skipping the British isles because the Americas had so many more natural resources. There is a reason the isles were inhabited for millennia before Europeans crossed the oceans. It’s called logistics. Continent to isle was a few hours in a dugout canoe with very few supplies while Europe to the Americas was months with oceangoing equipment and equivalent supplies. It’s unlikely the anyone died of scurvy during the canoe day voyages.

This is not another NASA should or Elon must post. This is a possibility of what I suspect will work out over time. Let NASA, Elon, Jeff, and others do what they think is right. Let the ones that have a reason to do things do them in their own time and their own way.

The Mars advocates point out that Mars has an atmosphere that allows for aerobraking and is a resource for oxygen and carbon. There is also considerably more known water on Mars than on the moon. The search for possible life, or even evidence of it existing previously, pretty much rounds out the advantages Mars. Except for the projected second home of humanity, which can be accomplished in a number of other ways.

There are a number of advantages that the moon has over Mars in addition to the logistics. Far side radio telescopes away from Earth spectrum noise. 3 second lag time for remote operated equipment. Lighter gravity for retirement living as suggested by Dr. Plata. The possibility of those skilled retirees still working in an environment that is not stressing their bodies with a full gravity. Fully secure location for dangerous research biological, nuclear, and anything else that requires that mistakes occur away from Earth. A third data point on gravities effects on the body.

The moon is assumed to be poor in carbon and hydrogen in particular with nitrogen and other materials close behind. It would seem arrogant to assume that a land area larger than most continents has been properly prospected by a dozen or so sample returns. I suspect that most minerals and volatiles wanted will be found in some quantity with proper searching. One impactor could supply carbon for the next several decades.

In situ propellant production is often mentioned as the Mars advantage using the atmosphere and ice to make methane and oxygen. Locate one carbon source on the moon, and mine that along with oxygen and carbon monoxide/oxygen rocket propellant becomes feasible. Give up a hundred points in Isp compared to methane/oxygen, but still would only need a mass ratio of two and a half to reach Lunar escape. Even methane/oxygen takes considerably more than that to get from the Martian surface to Earth trajectory . Getting to Mars may take a little less deltaV than getting to the moon considering aerobraking, but the round trip takes less.

The logistics and capital investments are the key advantages for the moon. A company or government with one ship would have a turnaround of over 4 years for the Martian run (assuming Hohman transfers). A single ship going to the moon could make dozens to hundreds of runs in the same timeframe. Monthly runs would have a ship delivering 52 payloads to the moon while a similar ship is delivering 1 to Mars. And if there are not 52 payloads that need to go, the ship is available for other work.

For any off Earth development to occur, it must be assumed that transportation to LEO is a solved problem. If one assumes a mass ratio of 6 from LEO to the lunar surface, then perhaps 10 tons of mass to LEO to per ton on the Lunar surface. Oddly enough, it will be a similar ratio to get a ton to the Martian surface even with the advantages of aerobraking. The heat shields and parachutes are not free mass, nor are the redundancies required for a multi-month journey. Manned flight to the moon is about 4 days which can be done with forecasting and minimal shielding while a Martian trip of months will have to use advanced cleverness to have a storm shelter available.

Anyone that has been involved in development of equipment is familiar with the problems in new systems that require troubleshooting. Several of the last lunar attempts have been by companies with relatively limited resources launching small vehicles as rideshares. Some of them in the last couple of years failed and are being retried with upgraded equipment. This is a timeframe and investment that applies to the moon by multiple small operators. As the systems mature, reliability will improve, and costs will come down. More players can be expected to join the attempts and major players will be watching them in case opportunities become evident. The search for available water and other volatiles will be enhanced by many players with various ideas and focus.

Leaving the moon is an ongoing topic featuring railguns, tether slings, cannons, and a few other exotics. Lifting off in rockets with Lunar sourced propellant is likely to be the early method. Most people seem to assume the requirement to find a hydrogen source for either H2/O2 or Methalox engines. Something I recently noticed (again) was the concept of CO/O2 rockets. Carbon Monoxide and LOX give up about a hundred points in Isp compared to Methalox and almost twice that compared to H2/O2. However, even the lower Isp could reach Lunar orbit with a mass ratio of less than two, and TEI with a mass ratio of two and a half.. Find one carbon source and use that with the known oxygen for the rocket fuels.

The reasons for being on the moon, or Mars, or a beach in Tahiti will be as varied as the entities that go. As the attempted landers have failed and retried recently, so will the surface activities. Different groups will send different landers with different capabilities and goals. With the short travel distance and time, they can break things and move fast in the current buzzwords have it. You can’t do that with Mars due to trip times and communication lags. While trying to get a billion dollar project going on Mars, many other people will be attempting to get million dollar projects going on the moon. In the history of innovation, the multiple attempts by multiple players have found a lot of solutions. The logistics of the moon favor multiple players in a way that Mars cannot for the near future.

The Starship will try flight test 8 in a couple of days. How far along would the test program be if tests were restricted once to every 26 months when a launch window opened???

Posted in Uncategorized | 20 Comments