OrbitalHub

Where curiosity reaches escape velocity.

Domain is for sale. $50,000,000.00 USD. Direct any inquiries to contact@orbitalhub.com.

 

JWST NIRCam coronagraphic images of four target stars processed with the standard KLIP method (top row) and the new FLIP physics-based model (middle row), with zoomed detections below.

 

Somewhere in the archive of the James Webb Space Telescope sits a picture of a planet the standard analysis pipeline could not see. The planet is 51 Eridani b, a hot gas giant roughly three times the mass of Jupiter, about 96 light-years from Earth, orbiting a star only around twenty million years old (Chauvin et al. 2015). Webb observed it at 4.10 and 4.30 microns. In the longer band, where carbon dioxide in its atmosphere absorbs, the published analysis came back empty (Balmer et al. 2025). The planet’s light was in the detector data all along, buried in the scattered light of the star, and the subtraction method the field had used for two decades could not lift it out.

Then a doctoral student at Northwestern University built a physical model of the telescope and let it fit itself to the real observations. Subtract the synthetic starlight and 51 Eridani b was there, at 6.3 sigma, in a band where it had never been seen.

Direct imaging studies a planet’s own light rather than inferring it from a dimming star, and it is the hardest kind of observation in astronomy. Even the brightest young giants run about a hundred-thousandth as bright as their hosts in the near-infrared, a contrast of roughly 10^-5, and the next generation of space observatories aims for 10^-10, enough to photograph an Earth analog around a Sun-like star (Nature, Nov. 2026). Webb’s answer is a coronagraph, a mask inside the instrument that blocks starlight before it reaches the detector. Blocking is not removing. What leaks past is a field of speckles shaped by every micron of surface error, thermal wobble and misalignment in the optics.

The standard answer has been to observe a different star of similar type and brightness and use that image as a reference. Karhunen-Loeve image processing, known as KLIP and introduced by Soummer and colleagues in 2012, builds principal components from a library of reference frames and projects the science image onto them, discarding the directions that hold only star-like structure (Soummer et al. 2012). It has been the workhorse of direct imaging ever since, with one structural weakness: reference and science images are never identical. As Jason Wang, the Northwestern physicist who co-led the new work, put it, “We try to make telescopes extremely stable, but there are extreme engineering challenges that we haven’t quite solved yet” (Northwestern, Nov. 2026).

Those instabilities are mundane and unforgiving. Webb’s cryocoolers hold the near-infrared instrument near 40 kelvin, and their little motors shake the mirrors. A thermal change can expand a mirror by a picometer, enough to reorganize the speckle pattern completely. The telescope drifts in pointing too, so the star sits in a slightly different place behind the coronagraph mask each time. Such mismatches set the floor on how faint a planet you can detect, and that floor belongs to the subtraction method rather than to the detector or the optics.

The team, led by astronomy graduate student Rodrigo Ferrer-Chavez with Jason Wang, Katherine Bouman of Caltech, William Freeman and Brandon Feng of MIT, and Laurent Pueyo and Marshall Perrin of the Space Telescope Science Institute, named their method FLIP: Physics-based Learning for Imaging Planets (Nature, Nov. 2026). Where KLIP combines images taken at different times, FLIP simulates what the telescope did. It starts from STPSF, the open-source optical model the Space Telescope Science Institute maintains for predicting Webb’s point spread function, rewritten in PyTorch so that it is differentiable. That lets the software compute analytic gradients through the simulation and optimize it directly against real detector data.

The model carries roughly a million free parameters, in four families: the star’s flux and position behind the mask, the optical path difference before the mask, the optical path difference after it, and the response of each detector pixel. The first two capture pointing error and mirror misalignment, the third captures instrument-specific phase errors, the fourth soaks up pixel effects the optics cannot explain. The optimization runs in two stages. Stage one calibrates the full model against a reference star, where no planet can hide, repairing inaccuracies in the nominal optics. Stage two then tunes a reduced set of parameters on the science image itself, recovering whatever changed since calibration. On a single NVIDIA A30 GPU, stage one takes ten to fifteen minutes and stage two ten to thirty (Nature, Nov. 2026).

The gains were not marginal. On HIP 65426 b, the first exoplanet Webb ever photographed directly, the signal-to-noise ratio rose from 82.5 to 130.1, a factor of 1.6. On 51 Eridani b at 4.10 microns it went from 6.7 to 14.2, a factor of 2.1. In the 4.30-micron band it went from a non-detection to a 6.3-sigma detection, the first image of the planet where carbon dioxide absorbs. That band is out of reach from the ground because Earth’s own atmosphere absorbs there, so this is an independent look at a planet whose carbon dioxide was inferred from the ground (Northwestern, Nov. 2026). The clearest gain came on a star with no known planets. TYC 5899 was observed four months after the reference stars a self-referenced survey used to build its speckle library, a drift that should have wrecked the subtraction. FLIP recovered an injected fake planet KLIP could not see, and delivered roughly fifteen times better sensitivity at half an arcsecond and five times better at one arcsecond. It also beat the team’s estimate of the best KLIP could achieve given realistic drift.

Two details deserve attention. The first is how the optimization knows when to stop. Push a flexible model against real data long enough and it starts eating the planet you are hunting, so the team injects synthetic planets at several separations, tracks their signal-to-noise against iteration count, and halts when those curves plateau, around eleven thousand iterations (Nature, Nov. 2026). The second is the honest cost. FLIP subtracts some of the planet’s own signal along with the starlight, losing between 5 and 60 percent depending on separation, against a comparable range for KLIP. It has also not reached the floor set by photon noise alone. The authors are explicit that no observation hit that limit, that the models cannot yet attach uncertainties to the drift terms they solve for, and that the detector model needs work before it can handle the deepest exposures of bright stars.

The method does not need a new telescope to matter. It extends to Webb’s Mid-Infrared Instrument, where thermal background rather than contrast sets the floor, and to ground-based telescopes, where data volumes are severe but the photon limit is worth chasing at short wavelengths, where the extremely large telescopes could otherwise reach only for temperate terrestrial planets around low-mass stars. More consequential is the error budget of the Habitable Worlds Observatory, whose picometre-level stability requirement, one hundred to a thousand times tighter than Webb’s, is a main reason its design is so hard. If processing can absorb part of that instability, the requirement shrinks and the observatory gets cheaper to build. The Nancy Grace Roman Space Telescope, launched on 30 August 2026 with the Roman Coronagraph Instrument aboard, is the obvious place to test that: it was designed in part as a pathfinder for this class of hardware (JPL).

For thirty-five years the field has treated the stability limit as a property of the telescope. This result suggests it is substantially a property of the software, and that a fair amount of astronomy is sitting in archives waiting for better subtraction.

 

 

A pair of deployed roll-out solar arrays seen from the station's port truss with Earth below, photographed on March 18, 2026 during the spacewalk that installed mounting hardware for the 2A power channel (NASA/Jessica Meir)

 

A roll of solar cells sat outside the International Space Station for two weeks, strapped to a carrier on the port side of the truss, waiting for two people to collect it. On October 29, 2026, NASA astronaut Anil Menon and Canadian Space Agency astronaut Joshua Kutryk switched their suits to battery power at 7:57 in the morning and floated out through the Quest airlock to fetch it. Seven hours and eleven minutes later they were back inside, and the seventh of the station’s new roll-out solar arrays lay stretched in sunlight in front of one of the original wings. The eighth and final array goes up on November 5. What is unusual about the hardware is not the photovoltaic layer but the way the blanket opens: no motor, no winch, no hand crank. It unfurls itself, the way a tape measure snaps out of its case, driven by tension stored inside composite booms wound tight for the trip to orbit.

The station’s eight original Solar Array Wings were built for a 15-year service life, and the first pair went up in December 2000 on STS-97 (NASA). Twenty-six years on, radiation and micrometeoroid damage have dulled their photovoltaic blankets. Meanwhile the station kept growing: more modules, commercial crew, more research payloads, and NASA technology demonstrations in orbit for trips to the Moon and Mars. All of it wants electrons. The roll-out arrays exist because of that squeeze: the fix had to be light, installable by two people in a spacesuit, and large enough to matter.

Replacing the old wings was never on the table. Each hangs from a mast can at the end of a truss segment, driven by a beta gimbal assembly that tracks the Sun and wired into a power channel with its own switching. Swapping them would have meant new structures, new gimbals, new cabling and a launcher with shuttle-class cargo capacity. NASA took the cheaper route and mounted smaller arrays in front of the old ones. Each new blanket covers a little more than half the length of the wing behind it and feeds the same channel, so the new panel makes its own power, the old wing keeps making power from the strip that is not shaded, and the original hardware is never touched (NASA, Oct. 15, 2026). The price of that arrangement is arithmetic. A panel in front of a panel earns less than the sum of its parts.

The seventh array had been aboard since October 15, when a SpaceX Cargo Dragon launched on a Falcon 9 docked at Harmony with the final pair folded in half on a carrier in its unpressurized trunk (NASA). Menon and Kutryk unbolted it free. Menon then locked his boots into a foot restraint on the end of the 58-foot Canadarm2 and rode the array out along the truss while flight controllers Jessica Watkins and Luke Delaney flew the arm from inside. One practice changed since the earlier installs: the arm now pivots about the array’s center of mass instead of the astronaut’s, taking the load off the person holding 340 kilograms of it (Spaceflight Now, Oct. 28, 2026). The spacewalk began in orbital darkness, which is when this work has to happen. Connectors on a solar array run hot and cannot be handled with sunlight on the blanket, so the four cables tying the new array into power channel 2A were mated in the dark. Kutryk noted the conditions from the airlock: “the night is dark and full of wonders.” Four hours and 23 minutes in, the station crossed back into sunlight, Menon released the last restraint with a power tool, and the blanket began to unroll, reaching full length over about ten minutes with nobody pushing it.

It was the 100th U.S. spacewalk and the 285th supporting station assembly, maintenance and upgrades since 1998 (NASA, Oct. 29, 2026). Menon, an Air Force lieutenant colonel and emergency physician, had three spacewalks before this one. Kutryk, a Royal Canadian Air Force colonel and test pilot who boarded the station in October on Crew-13, made his first, the fifth Canadian to walk in space and the first since David Saint-Jacques in 2019. The eighth array is the harder placement. It goes on the 3B channel at the far end of the truss, past the reach of the robot arm, so on November 5 the two of them will pass it between foot restraints in a handoff the team calls the leapfrog, then repeat the sequence of unfolding, cabling in the dark and letting the booms go (Spaceflight Now).

The power arithmetic gives the upgrade its shape. An International Space Station Roll-Out Solar Array, or iROSA, is rated at about 28 kilowatts at beginning of life and weighs roughly 340 kilograms folded, about 10 feet long, unrolling to about 60 feet (NASA STMD; Spaceflight Now). Its net contribution is much smaller than 28 kilowatts, because the wing behind it makes less than it used to. NASA’s accounting puts the station’s orbital-daylight capacity at up to 160 kilowatts before the upgrade and about 215 kilowatts after all eight arrays are installed, a gain near 30 percent (NASA, Oct. 15, 2026). What the new hardware actually adds has been closer to 8 kilowatts per array: station output moved from roughly 152 kilowatts before the first roll-out array flew to about 200 kilowatts with six installed (CBS News, Oct. 29, 2026). Eight arrays that each make 28 kilowatts and each add 8 is the upgrade in one line.

The array’s spine is a pair of stiff composite booms, built like split tubes, flattened and rolled lengthwise. Rolled that way they hold their shape and store strain energy, the way a bent ruler holds its bend. Released, they want to straighten, and as they do they drag the photovoltaic blanket out of its canister like a measuring tape. No motor means no gearbox and no powered latches to jam at the wrong moment. Flight director Garrett Hehn compares the result to a snap bracelet that “just continues rolling out steadily” (CBS News, Oct. 29, 2026). The design came out of Deployable Space Systems, a startup that turned NASA Small Business Innovation Research money starting in 2009 into a roll-out array before Redwire absorbed it in 2021 (NASA STMD). The first flight test on the station in 2017 was meant to last a week. The array deployed on June 18, 2017, but the crew tried three times to fold it back and could not get the blanket to line up evenly, so the station released it rather than risk the hardware (NASA, June 26, 2017). Two roll-out arrays powered NASA’s DART spacecraft on its flight to a test asteroid in 2021, and a pair will power the Power and Propulsion Element that NASA’s Gateway lunar outpost depends on (NASA STMD).

What the extra power buys is time and margin. Roughly 55 kilowatts more than the station started with will cover life support, research payloads and visiting vehicles through the outpost’s remaining years, and some of it will still be doing work on the day the station is flown into the atmosphere in 2030 (Spaceflight Now). The oldest hardware in the station has been asked to outlast its own design life by more than a decade. The answer was to build something that opens itself, hang it in front of the part that is failing, and send two people in white suits out to bolt it on. On November 5 they will do it one final time.

 

 

An artist visualization shows one of NASA's twin TRACERS spacecraft leading the second into Earth's polar cusp, where solar wind particles stream down the funnel-shaped magnetic field.

 

At 590 kilometres above Earth, two identical satellites cruise in tandem through the polar cusp, one trailing the other by ten seconds to two minutes. Their orbit takes them over the magnetic poles several times a day, and every time they pass they do the same thing: fly into a funnel-shaped region of Earth’s magnetic field and watch what the solar wind does to it. What they record is not a picture but a signature, a spreading of ion energies that shows where the Sun’s magnetic field is being stitched onto Earth’s. On 29 October 2026, a team working from those measurements published the first statistical answer to a quarter-century-old question: when magnetic reconnection happens at Earth’s magnetopause, is the event driven mainly by where it occurs in space, or by how it changes with time? The answer, from 207 matched pairs of cusp crossings, is mostly where.

Reconnection is the process that lets the solar wind get inside Earth’s magnetic shield. The Sun’s field arrives embedded in a continuous stream of charged particles, and where it butts against the magnetopause, two field directions that point opposite ways have to resolve themselves. They do it by breaking and reconnecting, and the breaking dumps stored magnetic energy into plasma motion. The rate at which that happens is the reconnection electric field, E_rec = v_rec * B, and it drives the aurora, inflates the magnetotail during a geomagnetic storm, and pushes charged particles down into the polar upper atmosphere.

The hard part is that reconnection does not happen at one tidy point. It happens in patches, at different places along the magnetopause, at rates that shift as the solar wind pushes. And a single spacecraft cannot sort out which is which. Any trace a satellite records on its way through fits two stories: a structure evolving in place over the seconds the crossing took, or a structure that varies from one spot to the next while the spacecraft sweeps across it. Physicists call this the spatiotemporal ambiguity of reconnection, and for more than two decades the agreed way out has been to fly two spacecraft close together along the same track, so each acts as a control on the other.

TRACERS is that experiment. It began as a proposal from Craig Kletzing at the University of Iowa, and it carries him in its wiring: the instrument electronics box on each spacecraft has one of his purple guitar picks mounted inside (Miles et al., 2025). Kletzing died in 2023, before launch. NASA picked TRACERS for the Small Explorer programme in June 2019, and the pathfinder was a pair of sounding rockets, TRICE-2, launched on 8 December 2018 into the northern cusp with the ancestor of the same instrument package. The satellites rode a Falcon 9 out of Vandenberg on 23 July 2025, stacked as a two-high cake topper on the rideshare.

Two days after launch, one of the pair, Space Vehicle 1, stopped answering. The fault was in the power system, and the diagnosis was grim: the battery no longer held a charge. Without one the spacecraft could run only while its solar panels saw sunlight, which turned its orientation in orbit into a requirement. Contact was restored in September by pointing SV1 at the Sun, and the observing plan was rewritten around the constraint (NASA, December 2025). Where the mission had been designed to work in the northern cusp, alongside radar facilities on Svalbard, it would work in the southern cusp instead, so each spacecraft could charge on the dayside before reaching its target.

The first results landed on 29 October 2026. Researchers took every window from 7 November 2025 to 28 February 2026 in which both spacecraft recorded a cusp crossing, matched the pairs, and asked how the ion signatures compared between leader and follower. Of 207 paired encounters, 129 were dominated by spatial variability, 48 by temporal variability and 30 could not be classified cleanly (Fuselier et al., 2026). The structures TRACERS sees in the cusp are set mostly by where the spacecraft happens to be looking rather than by reconnection pulsing on and off at a fixed spot. That matters because the reconnection rate sets how much energy the solar wind hands to the magnetosphere at any moment.

Measuring it takes plasma instruments and patience. The Analyzer for Cusp Ions on each spacecraft is a top-hat electrostatic analyzer that samples 47 energy channels between 8 and 20,000 electron volts across the full sphere of directions, delivering a complete velocity distribution every 312 milliseconds. The electron instrument beside it works at 50 milliseconds. Compare that with the Defense Meteorological Satellite Program, whose ion and electron instrument sweeps once per second through a 90-degree field of view, smearing the pitch angle distribution. The background field comes from a fluxgate magnetometer sampling 128 times a second, and a two-axis electric field instrument on wire booms yields the E * B flow velocity.

The signature is old. When magnetosheath plasma is injected along newly opened field lines, the faster ions pull ahead of the slower ones and the population arrives spread out in energy. The lower edge of that spread, the energy at which the flux falls to about a tenth of the spectral peak, is what the team tracks; how fast it changes along the track stands in for the rate at which field lines open. Turning that into an electric field runs through Faraday’s law and a correction for how flux tubes widen between the magnetopause, where the field is around 50 nanotesla, and the spacecraft. For the 30 September 2025 storm, the first big one after launch, which bottomed out at a Dst index of minus 129 nanotesla, the team picked out 21 dispersion events and derived reconnection rates of 0.1 to 1.5 millivolts per metre (da Silva et al., 2026), a little below what DMSP has produced and in line with the in-situ measurements from Polar.

The pair is also separated on purpose, and the separation is treated as a measurement. TRACERS flies in a sun-synchronous polar orbit, the two spacecraft trailing each other by 10 to 120 seconds along the track, which at their 590-kilometre altitude and 96-minute period is roughly 75 to 900 kilometres. That range lets the team size the structures they are chasing: a signature that shows up on both craft after a short delay is one small structure the pair crossed twice, while one appearing on only one spacecraft is either bigger than their spacing or was changing while they waited. The spacecraft are plain spinning cylinders on a Millennium Space Systems bus, passively stabilised at 10 revolutions per minute, their spin axes steered along the local magnetic field by small torque-coil manoeuvres. Over twelve months they were expected to cross the cusp more than 3,000 times.

What comes next is largely a matter of waiting, and of geography. The extension approved in September 2026 carries TRACERS into spring 2027, and the team still wants the northern cusp, with its Svalbard radars, for whatever stretch the power constraint allows. What exists so far is a preliminary paper with 207 encounters behind it, in a mission built to deliver thousands. Eventually the archive opens up, since each day’s downlink is mirrored by the Science Operations Center at Iowa and released publicly, and other missions attack the same problem from the other end, with Magnetospheric Multiscale at the magnetopause and PUNCH and EZIE in low orbit. The more of those pieces that fit together, the closer a geomagnetic storm forecast comes to being a number rather than a probability.

 

 

An artist's impression of Juno's path through Jupiter's bow shock, showing the electron foreshock-like region, the ion foreshock-like region, the magnetosheath and the magnetosphere, with the outbound crossing on 10 December 2024 and the inbound crossing on 11 December 2024.

Figure 2 from the Nature Communications paper, reused under CC BY 4.0.

 

The solar wind is a planetary-scale leak from the Sun, a supersonic flow of protons and electrons that leaves the corona at several hundred kilometres per second and needs a little over three weeks to cross the 780 million kilometres to Jupiter. It cannot simply be stopped. Wherever it meets something immovable the flow has to decelerate, and conservation of energy does not care how gently anyone would prefer that to happen. In front of Jupiter the meeting point sits about 5.8 million kilometres from the planet, roughly 81 planetary radii, on the sunward side of the strongest magnetic field any planet generates. A University of Iowa team has now taken the most detailed look yet at that boundary, and their conclusion is that Jupiter does not meet the solar wind with a single shock but with a layered structure that keeps rebuilding itself, powered by plasma waves that are richer and far more violent than anything measured at Earth (Joseph et al., Nature Communications, 31 July 2026).

Why a shock has to be built out of waves explains why the instrument mattered more than the spacecraft. In an ordinary gas a shock dissipates energy through collisions: ordered motion becomes random motion, the flow slows, the pressure jumps, the gas heats. Plasma obeys the same conservation laws, but the relevant collisions are Coulomb collisions between ions and electrons, and in the solar wind they are far too rare to account for the heating that instruments actually measure across Earth’s bow shock. Something else has to do the work, and for three decades the leading candidate has been wave-particle interactions at ion scales. Earth measurements support that picture, with ion-acoustic waves, electrostatic solitary waves and waves driven by the electron cyclotron drift instability all implicated in heating the flow. What they do not support is the idea that the same handful of modes suffice everywhere. Jupiter’s shock is harder. The solar wind there runs at a higher multiple of the Alfvén speed, the plasma pressure-to-field ratio is lower, and the relevant length scales are larger. On the two crossings in question the Alfvénic Mach number was about 9, the jump in ion temperature across the shock was roughly 80 against about 18 at Earth, and the electron temperature ratio near 10 against roughly 5.

Eight spacecraft had crossed Jupiter’s bow shock before Juno, and each found the same thing: a step in the magnetic field where the solar wind was forced to change direction. Pioneers 10 and 11, Voyagers 1 and 2, Ulysses, Galileo, Cassini and New Horizons all left that step behind. Voyagers’ few high-resolution electric field snapshots came back with four bits of amplitude resolution, which flattened exactly the fine structure anyone was looking for. Juno’s Waves instrument, built by physicists at the University of Iowa, carried burst-mode waveform capture from the start of the mission but was not pointed at the bow shock until November 2024, because telemetry is a finite budget. By then the spacecraft had been flying its stretched 53-day polar orbit since July 2016, carrying two whip antennas that form an electric dipole and a search coil wound with ten thousand turns of fine copper wire on a 15-centimetre Mu-metal rod.

The payoff came on 10 and 11 December 2024. Juno left the magnetosheath at about 17:25 UT on the 10th, crossing outbound, and came back across around 03:05 UT on the 11th, inbound. Both crossings were at essentially the same distance, 81.44 Jupiter radii, and both sat on the dusk flank near magnetic local time 16, at southern magnetic latitudes of 61.5 and 59.7 degrees. The geometry was not the same, though. Applying the coplanarity theorem to the mean upstream and downstream fields, the outbound shock presented an angle between the upstream field and the shock normal of 16.5 degrees plus or minus 14.3, while the inbound shock presented 53.5 degrees plus or minus 7.9. That is the difference between a nearly parallel shock and a nearly perpendicular one, and it changes what a spacecraft sees on the way in. At a shallow angle, protons reflected off the shock escape upstream rather than coming back, filling a long ion foreshock with wave activity. At a steep angle they execute partial gyro-orbits and pile into a narrow foot just ahead of the ramp. The two cases duly looked different. The outbound pass crossed an extended ion foreshock-like region carrying ultra-low-frequency magnetic oscillations near 7 millihertz, with reflected solar wind protons whose count rate fluctuated by a factor of about two, followed by an electron foreshock-like region where electrons accelerated at the shock and reflected back drove strong Langmuir waves near 9 kilohertz. The inbound pass had only a compact foot, Langmuir waves beside it, and something new: magnetic field structures repeating every 8.3 seconds, with the ramp peaking near 10.5 nanotesla, two earlier foot peaks at about 7.1 and 5.6 nanotesla, and four downstream peaks above 8 nanotesla.

The 8.3-second repeat carries the most weight of all, because it comes down to what one spacecraft can and cannot distinguish. A repeating magnetic sequence in a single record could mean the shock surface is flapping under pushes from foreshock waves or solar wind fluctuations, or it could mean the shock is rebuilding itself. Flapping produces irregular spacing on timescales much longer than a fraction of a proton gyroperiod, and the inbound case showed no foreshock driving and no consistent sign of solar wind variability. What remained was shock self-reformation. The numbers agree: with an upstream field of about 2 nanotesla the proton gyroperiod is 32 seconds, so 8.3 seconds is 0.26 of a gyroperiod, the only timescale in the problem that could be responsible. The authors stay careful about the edge of their evidence, noting that Juno’s ion instrument samples only every 120 seconds, so direct confirmation of ion reflection at an eight-second cadence is out of reach.

The waves are the more striking half of the story. In the quasi-perpendicular crossing the ion-acoustic wave had a fundamental frequency of 5,610 hertz, and between 03:05:55.786 and 03:05:55.806 the electric field waveform was not a clean sinusoid but a distorted one with flattened tops, the signature of a steepened wave carrying trapped electrons. The spectrum showed the second harmonic at 11,130 hertz and the third at 16,680 hertz, and the amplitudes sat far below the converter’s saturation threshold, so this was physics rather than clipping. The local electron gyrofrequency was 196 hertz and the plasma frequency 9.6 kilohertz, putting the fundamental neatly between them. Trapping electrons in a wave’s potential well and dumping them at the crests heats plasma far more efficiently than simple nonlinear steepening, and each added harmonic widens the band of electrons able to resonate with the wave. The electron cyclotron drift instability comes out of the relative drift between incoming electrons and shock-reflected ions, and it grows when that drift speed exceeds the perpendicular electron thermal speed. Since the electron thermal speed scales with the magnetic field, and the field at 5 astronomical units is roughly a tenth of Earth’s, the same shock conditions produce a far more favourable ratio in the outer heliosphere. The measurements bear that out. Juno logged 18 to 20 electron cyclotron drift harmonics, while the Earth events tabulated beside them for comparison yielded 3, 5, 10 and 13. Scored as harmonics times gyrofrequency over plasma frequency, the Jupiter crossings came out at 0.56, 0.65 and 0.66 against 0.02 to 0.11 for the four terrestrial cases.

Scale is why any of this reaches past Jupiter. Shock waves with far more energy than a planetary bow shock sit around supernova remnants, and the same wave-particle microphysics governs how flow energy turns into heat in those expanding clouds. Jupiter offers the nearest place where that physics can be measured with instruments in hand. There are real limits to what has been established, though. Juno returns three-axis magnetic field data at 8 hertz and a single-axis electric field at 50 kilohertz in one burst out of eight, so estimates of how much power the waves actually dissipate stay qualitative, and a single probe cannot recover wave polarization or phase velocity. The team’s own conclusion is appropriately narrow: wave-particle interactions at moderately high Mach number and low plasma beta shocks in the outer planets resemble Earth’s, with nonlinear behaviour appearing more often, and more crossings are needed before anyone can say how often. Juno remains in good health and is still returning data, but its most recent extension expired in September 2025 and NASA has not said whether another will follow (Ars Technica, March 2026). The spacecraft that found the answer may not be the one to tell us how general it is.

 

 

The thorium-229 doped calcium fluoride crystal inside the vacuum chamber where the 148-nanometre laser interrogates it, photographed by Thorsten Schumm for TU Wien.

Photo: Thorsten Schumm for TU Wien, from the announcement.

 

In October 2026, two groups of physicists published a machine no clock had managed before: it keeps time with the nucleus of a single element rather than the electron cloud around it, and it steers itself. The Vienna prototype, built by Thorsten Schumm’s group at TU Wien with Ekkehard Peik’s group at PTB Braunschweig, held its frequency for a full day on its own feedback loop. A Chinese consortium at Tsinghua University and the Shanghai institutes of the Chinese Academy of Sciences published a rival system in the same issue of Nature (Schumm et al., Nature 658, 897-902, October 2026). What both machines lock a laser to is the low-lying isomer state of thorium-229, an excited nuclear level 8.4 electron volts above the ground state. No other nucleus has a gap that small, and that gap is the reason this clock can exist.

Getting to that number took fifty years of argument. A stream opened by Kroger and Reich in 1976 narrowed the energy from a guess of under 100 eV to minus 1 eV, then 3.5 eV, then 7.8 eV. Internal conversion proved the isomer existed in 2016 at 8.28 eV, and two years later its radiative decay was caught in the vacuum ultraviolet, fixing the value at 8.338(24) eV. In April 2024 the Vienna and Braunschweig teams excited the transition with a laser for the first time; by that autumn they were using the nuclei as a timekeeper, with the laser disciplined by an ordinary optical atomic clock. That arrangement borrowed stability from the electronic world. The new one does not.

The physics behind swapping atoms for nuclei is insensitivity. A nucleus is roughly 10,000 times smaller in linear size than the atom containing it and carries very little of the charge and magnetic moment that make electronic transitions so sensitive to stray fields, temperature and radiation. Peik and Tamm proposed in 2003 that this stubbornness could become a frequency standard. Doping micrograms of thorium-229 into a crystal gives you not one nucleus but around 10^15 of them, which buys signal and costs resolution: the solid broadens the resonance to about 100 kHz, against a potential quality factor of 10^19 for a bare nucleus whose excited state lives nearly an hour in vacuum.

Vienna’s clock begins with an external-cavity diode laser at 1,187 nm locked to a high-finesse optical cavity by the Pound-Drever-Hall method for short-term stability, plus a seed laser offset-locked to it that carries the readout modulation. The seed is frequency-quadrupled and then doubled in a strontium tetraborate crystal to reach 148 nm. The beam enters a vacuum chamber holding a thorium-doped calcium fluoride crystal and ends on a photomultiplier tube. The decisive choice was absorption instead of fluorescence: rather than waiting for a nucleus to decay and emit, the experiment counts the probe light the crystal removes. That yields three orders of magnitude more signal photons per second and lets the feedback loop close far faster than the ten-minute isomer lifetime (Morawetz et al., Nature, October 2026).

The numbers were modest but real. Sixty-five picowatts of 148-nm light reached the detector, the tube counted photons with 10 percent efficiency, the 5/2 to 3/2 transition absorbed 0.75 percent of them, and the signal-to-noise ratio reached 17 in one second of counting. The error signal came from modulating the probe by f_FWHM/sqrt(3) on either side of the line; inverting that curve gave the detuning, fed back to the electro-optical modulator in a step of about a second. Two modes were possible, correcting every 30 minutes or every 20 seconds, trading cavity drift against short-integration noise. Inside a run, fractional frequency instability sat in the low 1e-14 range and settled near 1e-15 over a day, an error of about one second in thirty million years (TU Wien, 8 October 2026). Reproducibility from one day to the next was far worse, around 5e-13. The Vienna team traced that to the crystal: shifting the beam across the sample by fractions of a millimetre moved the line centre by as much as 1.7 kHz, which they attribute to local strain.

The Chinese group solved the laser problem differently. Instead of crystal frequency doubling it used four-wave mixing in cadmium vapour, with two 375-nm photons driving a two-photon resonance in the metal and a 710-nm beam completing the sum-frequency process, yielding 10 microwatts of continuous 148.4-nm light with an estimated sub-Hertz linewidth (Huang et al., Nature, October 2026). Their crystal came from 1.4 micrograms of thorium-229, an activity of 10 kilobecquerels, grown in one run with a temperature-gradient technique that leaves the crucible stationary and the crystal bubble-free. The clock locked to the quadrupole component labelled line b, and the fractional frequency instability was 5e-13/sqrt(tau/s), with 15 percent of the cycle lost to feedback overhead. Two independently grown crystals and earlier comb measurements tested whether the transition frequency repeats from crystal to crystal, the question that decides whether a nuclear clock is a standard or a curiosity.

Why this matters beyond clockmaking comes from the origin of the small gap. The 8.4 eV separation is what remains when mega-electronvolt-scale Coulomb and nuclear contributions to the binding energies of the two states nearly cancel. Sensitive to what is left over means the transition frequency responds strongly to the constants that set that balance. The sensitivity factors for thorium-229 are roughly 10^3 for the fine-structure constant, 10^4 for the QCD scale and 10^5 for the quark masses, against about 1 for a ytterbium ion clock. Vienna’s team searched 23 hours of clock data with a Lomb-Scargle periodogram for the slow oscillation an ultralight scalar field would imprint on the ratio f_Th/f_Yb+, found nothing above its detection threshold, and set limits ten to a thousand times deeper into that parameter space than earlier atomic clock work. The floor is photon shot noise in the interrogation, which follows sigma_y(tau) = (Gamma/f_0)(1/SNR)(tau/s)^(-1/2), with Gamma near 100 kHz and f_0 = 2.0204 x 10^15 Hz. Better crystals and brighter lasers push it down.

For spaceflight the appeal is direct. Clocks in orbit run fast because gravity is weaker there, and a navigation satellite’s clock gains about 38 microseconds a day, 45 from the shallower potential and minus 7 from orbital speed (Ashby, Living Reviews in Relativity, 2003). ESA already flies a clock ensemble outside the International Space Station, where the Atomic Clock Ensemble in Space pairs a laser-cooled caesium clock with an active hydrogen maser on Columbus to test relativity, targeting an instability of 2 x 10^-16, about a second in 300 million years (ESA, ACES). A nuclear clock would suit that job better: a doped crystal tolerates vibration and temperature better than a magneto-optical trap, and the same frequency that keeps time also probes for new physics. That is a promise, not hardware: the lasers at 148 nm are still laboratory instruments, and the Vienna result loses to the best optical atomic clocks by orders of magnitude in instability.

What the two groups have built is a measurement chain with no weak link left to hide behind, and the remaining problems are materials problems: crystal growth has to become reproducible to a fraction of a kilohertz, dopant concentration has to rise without broadening the line, and the lasers have to get stronger. If that happens, the thorium nucleus stops being the most interesting part of a nuclear clock and becomes the part that makes it work. The end state is a time standard that no laboratory can argue with, running at room temperature, in a piece of crystal, on a wavelength that lies in the vacuum ultraviolet.

 

 

Artist's rendering of NASA's nuclear-powered PROMISE rover on the lunar surface, its mast camera head raised above a six-wheeled chassis built from spare Curiosity hardware.

 

On a Saturday morning at the MAX POWER aerospace expo at NASA’s Kennedy Space Center, the agency handed out the paperwork for a supply chain. Almost none of what NASA announced was about a spacecraft that flies soon. Eight companies were put on a list to fly cargo to the lunar south pole. Six task orders were awarded for launches and orbital transfer vehicles, and one company took home $94 million to deliver a nuclear-powered rover that will drill into the Moon (NASA, November 2026). That rover is PROMISE, the Polar Rover for Observation, Mapping, and In-Situ Exploration. Everything else that day was about how that box gets there, and what happens to the power and spare parts after it lands.

The program itself is young. In March 2026, at an event NASA called Ignition, administrator Jared Isaacman laid out a Moon Base: humanity’s first outpost, assembled one mission at a time rather than in a single leap. “The moon base will not appear overnight,” he said at the time (CNN, March 2026). Eight months later the program is issuing contracts, and the shape of those contracts says what NASA thinks a base is. Not a habitat first. A logistics chain first, with the habitat waiting at the far end of it. NASA’s phased plan is blunt about it: Phase One, through 2029, is titled “Gain Reliable Access, Experiment, and Learn,” and every asset listed under it flies without a crew aboard.

Getting to the south pole is not mainly a distance problem. It is a problem of arriving with power and staying alive. The sun never climbs high there, so sunlight is weak and arrives at a grazing angle. Shadowed craters near Shackleton reach roughly -242 degrees Celsius, cold enough to freeze tanks and kill batteries. A lunar night lasts about 14 Earth days. And the polar ice that would make a base self-sufficient sits under meters of dust, findable only by drilling or by watching neutrons scatter off hydrogen. The hard part is not the pressure vessel. It is landing repeatedly at a named spot on a world with no roads and no rescue, using hardware somebody else owns.

That is where the November 7 announcements land. NASA’s new CLPS 2 on-ramp admits eight companies: Blue Origin, Firefly Aerospace, Intuitive Machines, ispace-US, Lockheed Martin Space, Northrop Grumman Space Systems, SpaceX and Voyager Technologies. They compete for lunar delivery missions under a ten-year multiple-award contract with a ceiling of $10 billion. NASA also raised the value of the original CLPS 1.0 vehicle from $2.6 billion to $4.2 billion for Phase One, while deliveries already awarded must finish by November 2033. The trade has changed: the company builds the lander, operates it and flies it, and NASA arrives with a payload and a delivery ticket.

The legs in between are covered by a second contract. NASA’s Venture-Class Acquisition of Dedicated and Rideshare vehicle is a ten-year indefinite-delivery contract with a $5 billion ceiling. Six task orders worth roughly $1.3 billion went to Arrow Science and Technology, Blue Origin and Impulse Space, in three shapes (NASA, November 2026). One carries CAPSTONE 02, a pair of small spacecraft, into deep space to rehearse rendezvous and navigation around the Moon. One sends JPL’s MoonFall drones to scout terrain before astronauts arrive. One tests commercial orbital transfer vehicles in cislunar orbits. Launches will ride New Glenn, Relativity’s Terran R and Falcon 9, with transfer vehicles from Blue Origin, Firefly, Impulse Space and Quantum Space. SpaceX, separately, won a cislunar communications demonstration that will put satellites on a Starship in 2028, and NASA will buy parts for miniCAST towers: solar arrays, batteries and wireless power transfer, assembled at Marshall.

And then there is the rover, where the engineering gets specific. Blue Origin will fly PROMISE to the surface on its Blue Moon Mark 1 lander, under the $94 million award and $140 million in options. The rover is six wheels on a Curiosity-derived chassis. JPL is building it from the electronics, body structure and mobility system of the duplicate Curiosity vehicle used as a test bed for the Mars mission, supplementing that with flight spares and test models left over from Curiosity and Perseverance. That shortens the path from a parts bin to a landing site. Power comes from a flight-spare Multi-Mission Radioisotope Thermoelectric Generator, a unit that supplies about 110 watts of electricity at the start of a mission (NASA Radioisotope Power Systems). A hundred watts looks trivial next to a solar array, but it does not stop at sunset, and at the pole there is always a sunset.

The payload surveys everything the base will need to know. A rotary percussive drill called TRIDENT, designed by Honeybee Robotics, a Blue Origin subsidiary, pulls regolith up from a meter down in 10-centimeter increments while measuring how strong and how warm the material is. A mass spectrometer at Kennedy and a near-infrared spectrometer at Ames read its chemistry. A Lockheed Martin neutron spectrometer looks for hydrogen. A radiometer built at Johns Hopkins, using technology from the Lunar Reconnaissance Orbiter’s Diviner instrument, measures temperatures down toward -242 degrees Celsius inside permanently shadowed terrain. Everything runs on NASA’s High Performance Spaceflight Computing, which NASA says delivers more than 100 times the capability of current space processors. With light taking about 1.3 seconds to cross the Earth-Moon distance each way, the rover has to decide things on its own.

That chain is worth the arithmetic, because NASA has split the old lunar lander into three or four separately purchased pieces. From low Earth orbit to the lunar surface is roughly 5.6 kilometers per second of delta-v: about 3.1 for translunar injection, 0.9 for lunar orbit insertion and 1.7 for the powered descent. Tsiolkovsky’s rocket equation, Δv = v_e * ln(m_0 / m_f), explains why nobody flies that whole profile as one vehicle. A chemical stage with an exhaust velocity near 3.5 kilometers per second needs a mass ratio above two just for the departure burn. A Hall thruster needs far less propellant, but it thrusts too low to sprint, and solar flux at the Moon is a twenty-third of Earth’s. Electric transfers pay off when the cargo can carry a huge array, which is why the chain is layered. Impulse Space’s two awards show the logic: Helios, a chemical kick stage, takes the expensive first leg, while Mira and a throttleable lander handle the maneuvers and the descent. Blue Origin’s Blue Moon Mark 1 is scheduled to attempt a landing at the Shackleton Connecting Ridge, with SCALPSS cameras watching how engine plumes scour the ground. A second lander is booked to carry NASA’s VIPER rover in late 2027.

What to watch is not any single mission but the order they arrive in. Blue Moon’s Endurance landing attempt and CAPSTONE 02 both fly in 2027, followed by VIPER’s delivery, then the MoonFall drones, then PROMISE in late 2028 with its eighteen-month campaign. Every one of those dates assumes commercial hardware shows up on a schedule NASA does not control, at a place NASA does not own. The contracts do not make that harder. They make it financeable. An indefinite-delivery vehicle with a $10 billion ceiling turns a program of bespoke government spacecraft into a market where a lander operator has to show up repeatedly, not once, and where a failed delivery is somebody else’s problem. Whether the Moon Base ends up a settlement or a very expensive parking lot will be settled by how those eight companies perform over the next three years, long before anyone has to live there.