OrbitalHub

 

 

Abstract

This post explores what happens to proof-of-work cryptocurrency protocols once the network stretches across the solar system and beyond. Cryptocurrencies are here to stay, and it is plausible that future colonists will prefer a decentralized currency over a fiat currency issued by a distant central authority. Ubiquitous access, independence from exchange-rate and interest-rate policy set on another planet, and freedom from financial institutions serving foreign interests are strong arguments in an emerging exo-economy.

It is tempting to assume that relativistic time dilation would be the main threat to an interplanetary proof-of-work network. Running the numbers says otherwise: inside the solar system, time dilation is a rounding error, and the real enemy is the finite speed of light. This post works through both effects and sketches designs that could hold up.

Motivation

Today, information broadcast by a node reaches nearly every other node within a few seconds. The Internet spans the Earth’s surface and low Earth orbit, so the worst-case path is a fraction of a light-second long.

That stops being true the moment the network leaves Earth’s neighborhood. Approximate one-way light delays:

Link, One-way delay
Earth to LEO satellite: a few milliseconds
Earth to Moon: about 1.3 seconds
Earth to Mars: about 3 to 22 minutes, depending on orbital positions
Earth to Jupiter: about 33 to 53 minutes
Earth to TRAPPIST-1: about 40 years

Once delays approach the block interval, the network starts to suffer: more communication failures, more frequent and longer blockchain forks, and eventually segregation into local sub-networks (let us call them topological forks). For deep-space links, the plumbing is not TCP/IP either. NASA and others use Delay/Disruption-Tolerant Networking (the Bundle Protocol), which stores and forwards data across links that are slow and intermittently available.

Nullius in verba, as they say. Let’s go into the details and figure out what a truly interplanetary cryptocurrency would need.

Cryptocurrency 101

Think of a cryptocurrency as a digital money ecosystem. It combines decades of research in cryptography and distributed systems: a decentralized network of computers, a public transaction ledger (the blockchain), a set of transaction validation rules, and a decentralized consensus mechanism [ANTO].

The network. Computing power and data are spread across many nodes. Every full node keeps a copy of the ledger, so the network survives disruption, including physical damage to individual nodes.

The blockchain. A public, distributed ledger of every transaction the network has accepted. Each block holds as many transactions as fit within the network’s block size or weight limit. Transactions name the sender, the receiver and the amount or asset changing hands. Each block contains the cryptographic hash of the previous one, a backward link that runs all the way to the chain’s Genesis block. Tampering with any old block changes its hash and breaks every link after it.

Validation rules. Every node checks every transaction. The most common fraud attempt is double spending, so nodes verify that each transaction’s inputs exist and have not already been spent. Invalid transactions are rejected and never reach the chain.

Consensus. The consensus mechanism lets all nodes agree on which transactions go into the next block. It replaces the authority of a central clearing house with a network of nodes that do not trust each other, and do not need to, because the protocol enforces the trust.

The Proof-of-Work

Satoshi Nakamoto proposed proof-of-work consensus in 2008 [NAKA]. A proof-of-work is a piece of data that is costly and slow to produce but trivial to verify. Think of a semester project: it takes you a whole semester to finish, and your supervisor an hour to grade.

To add a block, a miner must find a nonce that makes the block’s hash fall below a target. The only way to find one is brute-force guessing, so a miner’s chance of winning the next block is proportional to its share of the network’s total hash rate. The network retunes the target so that blocks arrive at a fixed average interval. For Bitcoin that is 10 minutes, with difficulty recalculated every 2,016 blocks (about two weeks).

When two miners find blocks at nearly the same moment, the chain forks briefly. Nodes resolve it by following the chain with the most accumulated work. This is how Nakamoto consensus reaches agreement without a trusted authority.

Then and now. For years, proof-of-work was the de facto consensus mechanism. That is no longer true. Ethereum switched to proof-of-stake in September 2022 (“the Merge”), and most newer networks use proof-of-stake or other Byzantine-fault-tolerant designs. Bitcoin, the largest network by value, remains proof-of-work, and it is the reference case for the rest of this post.

Why latency breaks it. The longest-chain rule quietly assumes that a new block reaches everyone much faster than the block interval. If a block takes d seconds to reach the bulk of the hash rate, and blocks arrive every T seconds on average, the chance that a competing block appears in the meantime is roughly:

$$
P_{text{stale}} approx 1 – e^{-d/T}
$$

On Earth, d is a few seconds and T is 600 seconds, so stale blocks are rare. For a miner on Mars, d is 3 to 22 minutes, giving a stale probability of roughly 26% to 89%. Mars miners would also spend those minutes hashing on top of a block that Earth has already buried. Their effective reward collapses, they have no reason to stay honest participants, and the network segregates. That is the topological fork from the motivation, and it happens long before relativity enters the picture.

Enter the Ansible

We have no technology that communicates quickly and reliably across interplanetary, let alone interstellar, distances. Consider a mining corporation in the TRAPPIST-1 system, about 40 light-years away in the constellation Aquarius, paying a planetary engineering firm in the Sol system. The payment would take about 81 years for the round trip, plus the block confirmation time.

Science fiction already offers a fix. Readers of Rocannon’s World [LEGU], where Ursula K. Le Guin coined the word in 1966, and of Ender’s Game [CARD] will know the ansible: a device that sends and receives messages over any distance with no delay.

One honest caveat before we lean on it. Special relativity says there is no universal “now”. Two events that are simultaneous for one observer are not simultaneous for another moving relative to the first. An instantaneous link is therefore instantaneous only in some chosen frame, and combining such links with moving observers can deliver a message before it was sent. The ansible is not just beyond our engineering; it is in tension with the physics the rest of this post relies on. We will use it anyway, because it isolates a cleaner question: if latency vanished, would relativity still hurt proof-of-work?

The Special Theory of Relativity

Before upgrading the protocol, we need the physics of how time and distance depend on the observer’s frame of reference.

Einstein received the 1921 Nobel Prize in Physics, awarded in 1922, “for his services to Theoretical Physics, and especially for his discovery of the law of the photoelectric effect.” Relativity was famous by then but still controversial, and the committee chose not to cite it.

Einstein set out to reconcile two principles: the principle of relativity, and the constancy of the speed of light. His well-known thought experiment uses a train [EINS]. A passenger walks toward the front at speed w relative to the train, while the train moves at speed v relative to the embankment. Classical mechanics says an observer on the embankment measures the passenger’s speed as W = v + w. Replace the passenger with a beam of light and the same rule gives v + c, contradicting the principle that light travels at c in every inertial frame. Einstein resolved the contradiction by giving up absolute time.

The Lorentz transformation. Before relativity, time was absolute, independent of the observer’s motion. After it, every inertial frame has its own time, and a statement about time must say which frame it refers to. Requiring that light moves at c in every inertial frame leads to the Lorentz transformation between a frame K and a frame K′ moving at velocity v along the x axis:

$$
x’ = gamma,(x – v t), quad y’ = y, quad z’ = z, quad t’ = gammaleft(t – frac{v x}{c^{2}}right), qquad gamma = frac{1}{sqrt{1 – v^{2}/c^{2}}}
$$

Here γ is the Lorentz factor. Letting c go to infinity makes γ equal 1 and recovers the Galilean transformation of classical mechanics (x′ = x − vt, t′ = t), where both distance and time are absolute.

Time dilation. The corollary that matters here: a clock moving at speed v relative to an observer runs slow as measured by that observer. If Δτ is the interval between two ticks in the clock’s own rest frame (its proper time), the observer measures:

$$
Delta t = gamma,Deltatau
$$

Since γ ≥ 1, the observer always measures the longer interval.

Real frames are not inertial. A clock on Earth’s surface rides a rotating planet orbiting the Sun, so strictly it is not in an inertial frame. For most purposes the approximation works, but precise timekeeping must also account for gravity. General relativity adds gravitational time dilation: clocks deeper in a gravitational well tick slower. GPS is the everyday proof [ASHB]. Satellite clocks run about 7 microseconds per day slow because of their orbital speed, and about 45 microseconds per day fast because they sit higher in Earth’s gravity well. Without the net correction of roughly 38 microseconds per day, GPS positions would drift by kilometers within a day.

Relativistic Effects on Proof-of-Work

The twin paradox is the classic illustration of time dilation. One twin travels on a relativistic spaceship and returns home to find the other twin has aged far more. Fly fast enough or long enough, and several generations pass on Earth during the trip. Orson Scott Card used exactly this in Ender’s Game [CARD]: Mazer Rackham, the pilot who ended the Second Formic Invasion, is sent on a relativistic round trip so that, decades later, he can train the next commander while having aged only a few years.

Now put a mining sub-network on a fast-moving ship, connected to the rest of the network by an ansible. Measured in Earth’s frame, the ship’s hardware runs slow by the factor γ, so it computes fewer hashes per Earth-second.

The tempting fix. One could treat this as a fairness problem: lower the proof-of-work difficulty on the ship by the Lorentz factor, so relativistic miners solve an easier puzzle, and shorten their block confirmation time.

Why that does not work. Proof-of-work is not a race against the clock; it is a measure of computation. Chain weight is accumulated work, and the protocol never asks how long a miner took. A time-dilated miner is, to everyone else, simply a slower machine, exactly like an older ASIC on Earth. Bitcoin does not lower difficulty for slow hardware, and it should not lower it for fast-moving hardware either. A lower difficulty on the ship means cheaper blocks, which means either the ship’s blocks count for less chain weight (so the discount cancels out) or they count the same (so the ship can out-produce Earth at a fraction of the cost). Worse, the discount is triggered by a claim that nobody else can verify: any node that can fake a slower beacon gets cheaper blocks. Fairness in proof-of-work is measured in hashes, and the network should keep a single global difficulty.

How big is the effect, anyway? For realistic solar-system traffic, it is negligible:

Scenario, Clock rate difference vs Earth, Effect on hash rate
Mars miner (orbital speeds up to tens of km/s, plus gravity difference): on the order of 1 part in 100 million: about a millisecond per day
Ship at 0.5c: γ ≈ 1.15: about 13% fewer hashes per Earth-second
Ship at 0.87c: γ ≈ 2: half the hashes per Earth-second

Compare the first row with the 26% to 89% stale-block penalty that light delay imposes on the same Mars miner. Inside the solar system, latency is the whole problem and relativity is a footnote. Relativity only matters for ships moving at a large fraction of c, and even there the right response is to accept the lower effective hash rate, not to discount the difficulty.

Where clocks do matter: timestamps. Proof-of-work protocols do care about clocks in one place. Block timestamps must fall within a tolerance of network time, and difficulty retargeting is computed from them. A node whose local clock is dilated, or sits in a different gravity well, will drift away from everyone else’s notion of time. The fix is the one astronomers already use: define a single coordinate time for the network (for the solar system, something like Barycentric Coordinate Time, TCB) and have every node convert its local clock to it. A simple beacon does the job. If an Earth-side ansible broadcasts a tick every fixed interval of network time, a node that counts its own seconds between ticks learns its clock rate relative to the network and can correct its timestamps. (With ordinary light-speed beacons instead of an ansible, the measured interval would be dominated by the Doppler shift from the node’s motion, not by γ alone, and would need to be separated out.)

A More Down-to-Earth Solution

Let’s fork the storyline (pun intended) back at Enter the Ansible and drop the faster-than-light link. Light delay is now the constraint, and there are several ways to live with it.

1. Peripheral nodes relay, Earth mines. Mining stays on and around Earth, and nodes on Mars or elsewhere only relay transactions and verify blocks. A Martian payment confirms after roughly twice the one-way light delay plus the confirmation time: on the order of an hour or more. Its weakness is an eclipse attack. A local miner can feed Martian nodes a fabricated chain until the real one arrives minutes later. Martian merchants therefore cannot trust anything that has not been confirmed by blocks that demonstrably came from Earth’s hash rate, so local commerce waits on interplanetary round trips.

2. One currency, local settlement layers. Keep a single base chain for interplanetary settlement, and run fast local systems on each world: a Mars-side payment channel network, a sidechain, or a rollup that periodically anchors its state to the base chain. Everyday payments on Mars clear in seconds; only the net balances cross the gap. The catch is that today’s channel designs rely on time-locked dispute windows, which would need to be far longer than any interplanetary round trip.

3. Separate local chains, trustless exchange. Each world runs its own chain, with a block interval and consensus tuned to its own network. Value moves between them through atomic swaps or bridges. This accepts the topological fork as a feature rather than fighting it. It mirrors how separate economies already trade, but bridges have been among the most frequently exploited components in crypto, so this layer needs the most scrutiny.

4. Consensus with finality. Proof-of-stake and other Byzantine-fault-tolerant protocols finalize blocks once a supermajority of validators agrees. They avoid probabilistic forks, but each round of voting is bounded by the slowest validator, so an interplanetary validator set would finalize at the pace of light delay. A local validator set per world, with periodic cross-world checkpoints, is the more practical variant.

Whichever design wins, one incentive rule still holds: for every node, the reward for staying honest on the shared network must outweigh anything it could gain by cheating locally.

Concluding Remarks

The ansible is fiction; the rest is not. Running the numbers gives a clear answer: inside the solar system, the speed of light breaks proof-of-work long before time dilation does. Relativity’s real lesson for an interplanetary protocol is about clocks, not difficulty: pick one coordinate time and convert everything to it.

Thank you for working through this exercise with us. Cryptocurrencies have kept reshaping finance in recent years, and as a species we will go far beyond low Earth orbit. History suggests that decentralization is key to resilience and prosperity. The protocols that follow us out there will have to be designed for the delays of space from day one. Let us make that future happen together.

Image: A futuristic city on Mars, built under a vast geodesic glass dome around lakes and parks, with tower blocks at its centre and smaller domes scattered across the red terrain beyond. Credit: DALL-E 3, released into the public domain, via Wikimedia Commons.

References

• [ANTO] Antonopoulos, Andreas M., Mastering Bitcoin, Second Edition, O’Reilly Media, 2017.
• [ASHB] Ashby, Neil, “Relativity in the Global Positioning System,” Living Reviews in Relativity 6, 2003.
• [CARD] Card, Orson Scott, Ender’s Game, Tor Teen, 2014 (first published 1985).
• [EINS] Einstein, Albert, Relativity: The Special and the General Theory, Princeton University Press, 2015.
• [LEGU] Le Guin, Ursula K., Rocannon’s World, Harper & Row, 1977 (first published 1966).
• [NAKA] Nakamoto, Satoshi, Bitcoin: A Peer-to-Peer Electronic Cash System, 2008.

 

 

 

A single positron has been held inside a vacuum vessel in Evanston, Illinois, for months at a time, circling in a magnetic field with no container wall near it and no neighbouring particle to talk to. It sits in the lowest energy state its orbit allows, at a temperature of 300 millikelvin, in a trap barely a centimetre across. On 21 October 2026, Gerald Gabrielse’s group at Northwestern University described how they built that state and what they can now do with it (Myers et al., Physical Review Letters 137, 173001). The device is a one-positron quantum cyclotron, and the reason it matters is that a positron is about the cleanest referee available for a symmetry the Standard Model insists cannot break.

The asymmetry the Standard Model cannot explain is blunt. The visible universe is made of matter, and the antimatter that should have arrived with it is gone. The model predicts no imbalance at all, which is why physicists keep taking the simplest particles apart looking for a crack. The symmetry in question is CPT, which folds charge conjugation, parity and time reversal into a single operation. For any local Lorentz-invariant quantum field theory, the theorem says a particle and its antiparticle must have equal mass, equal lifetime and magnetic moments of equal magnitude and opposite sign. Electrons and positrons are supposed to be perfect mirrors, so a disagreement between their magnetic moments would be awkward to explain with physics we already have (Nature, 22 October 2026).

Magnetic moments became precision probes because of a sequence of accidents in 1947. Measurements of hydrogen hyperfine structure by Nafe, Nelson and Rabi, and then of gallium by Kusch and Foley, turned up a discrepancy with Dirac’s electron theory that the Lamb shift also confirmed. Julian Schwinger’s 1948 calculation of the leading quantum correction, a_e = (g_e - 2)/2 = α/2π, matched what the experiments were already seeing, and the agreement showed quantum electrodynamics worked. Theorists have since pushed the quantity to tenth order in the fine-structure constant, the fifth-order term alone requiring more than 12,000 Feynman diagrams (Crivelli, Physics 19, 132, October 2026).

The electron remains the benchmark: Gabrielse’s group pinned its magnetic moment to 0.13 parts per trillion in 2023, the most precise property ever measured for an elementary particle. The positron lagged badly. Its first direct free measurement, in 1966, fixed the g factor only to about ten parts per million, and the best direct electron-positron comparison, from 1987, established equality to 2.1 parts per trillion. Nothing about that gap was theoretical. It was logistics: a positron has to be born in radioactive decay, slowed by orders of magnitude and then isolated thoroughly enough that it neither touches a wall nor wanders into a stray field before anyone can measure it.

The Northwestern apparatus solves those logistics with a trick. Positrons emerge from sodium-22, whose beta decay endpoint is 546 keV, and are thermalised by scattering inside a single-crystal tungsten moderator. A positron that reaches the exit is pushed back out by the moderator’s work function, and it can pick up an electron from the tungsten surface to become a loosely bound, highly magnetised Rydberg positronium atom. Because that atom is neutral, it ignores the trap’s electric field and can slip through a 625-micrometre hole in a silver electrode into the trap’s interior. Only there, where the field is strong, does the atom ionise: the electron escapes and the positron stays. The source is deliberately feeble at 18 microcuries, and loading one positron takes eight to eleven hours, which sounds punishing until you notice that the particle then stays trapped for months.

Everything after loading is about making the trapped particle behave like a textbook. The trap hangs inside a dilution refrigerator that holds the electrodes and the sealed vacuum enclosure at 300 millikelvin, cold enough that photons capable of exciting the cyclotron states are essentially absent. A persistent 4.2 K superconducting solenoid supplies the 5.6 T field. The trap is a cylindrical Penning trap, combining that field with an electrostatic quadrupole, V ∝ z² - ρ²/2, which pins the orbit radially and adds a separate 206 MHz oscillation along the field axis. The cyclotron frequency is about 156 GHz and the spin precession frequency matches it to roughly a part in a thousand; their difference, ω_a = ω_s - ω_c, is the anomaly frequency, and the magnetic moment lives in the ratio a = ω_a / ω_c.

Reading that ratio without destroying it is the hard part. A cyclotron state radiates away on its own at a free-space rate of γ_c = (4/3)α(ℏω_c/mc²)ω_c, which would have the first excited state vanishing within 83 milliseconds at this field strength. The trap electrodes are therefore shaped into a high-Q microwave cavity with the particle suspended at its centre, and the cavity’s modes suppress spontaneous emission by a factor of two hundred or more, stretching the excited-state lifetime to roughly four seconds. Readout is subtler still. Saturated nickel rings encircle the trap to form a magnetic bottle, a gradient of 1500 tesla per square metre that shifts the axial frequency in proportion to the cyclotron and spin quantum numbers, Δω_z = β(n + m_s), with β/2π near 3.8 Hz for a single particle. The shift is picked up from a coaxial resonator of Q = 900 feeding a cryogenic HEMT amplifier, and since the measurement only watches axial motion it leaves the cyclotron and spin states untouched. That is quantum nondemolition: the experiment learns the state without spending it.

With that readout in place, the team drove the cyclotron motion at 156 GHz and watched the energy climb one step at a time and then fall back one step at a time, the signature of genuinely discrete orbital levels rather than a classical orbit. They flipped the spin up and back down by applying a drive near the anomaly frequency, and the same 3.8 Hz shift appeared on each flip. The size of the shift is itself proof that a single particle is in the trap, since two positrons sharing a centre of mass would halve it. None of this is a magnetic-moment measurement yet. The spectroscopy still has to be run and the systematics tamed, but the authors argue a quantum measurement at the electron’s precision is now in reach, which would make the lepton-sector CPT test more than thirty times stricter than anything before it (Myers et al., 2026).

A competing approach is worth watching. The Lepton Symmetry experiment at the Max Planck Institute for Nuclear Physics plans to trap a positron beside a hydrogen-like helium-4 ion, letting that ion’s single bound electron act as the electron reference, with both species held in the same field so field noise largely cancels instead of accumulating; the target is comparison at fourteen significant digits (Crivelli, October 2026). One approach reaches for a standalone positron measurement, the other for a side-by-side one.

The Northwestern result is candid about its limits. A positron vanishes on contact with anything, and the team had to thread a 546 keV particle into a millimetre-scale cryogenic cavity without spoiling the cavity before the physics could start. If the two magnetic moments eventually come out equal, the imbalance between matter and antimatter will still need explaining. If they come out unequal, the explanation may turn out to live inside the magnetic moment of one antiparticle, held still enough to read.

 

 

 

Every fusion machine begins with an awkward asymmetry. The nuclear reaction is easy to describe; sustaining it is not. Fuse two heavy hydrogen nuclei and you get back a sliver of mass as heat, and in a tokamak that heat has to arrive faster than the plasma leaks it to the walls, the radiation and the vacuum outside. Do that and the plasma heats itself, a state physicists call ignition. Miss it and you have built an extremely expensive space heater.

The puzzle organising fusion research since 1955 is not whether the reaction works, but how to reach a burning plasma without spending more energy getting there than the reaction returns. John Lawson’s answer, published in 1955, fixed three quantities a plasma must satisfy, temperature, density and confinement time (ITER). Lawson said where the finish line sat, not which road to it was cheapest. On 10 September 2026, Luis Delgado-Aparicio, Masayuki Ono and Jonathan Menard at the Department of Energy’s Princeton Plasma Physics Laboratory published a paper in Physical Review Letters that answers the second question, and not in the way most reactor designs assume.

In December 2022 the National Ignition Facility compressed a peppercorn-sized capsule hard enough to get more energy out of it than the lasers put in, 3.15 MJ from 2.05 MJ (LLNL), and by April 2025 that yield had reached 8.6 MJ at a target gain above four (LLNL). That is ignition, but it is inertial confinement, fuel squeezed by an imploding shell. A power plant needs the opposite, a burning plasma held by magnetic fields in a doughnut. No tokamak or stellarator has reached ignition.

Delgado-Aparicio and colleagues treat ignition as a problem of path rather than threshold. Their framework unifies four things fusion theory usually handles apart: the Lawson power balance, the Cordey construction that identifies the least demanding path to a burning state, the ignition ridge where a burning plasma tips into runaway, and Mills’ treatment of runaway dynamics (Delgado-Aparicio et al., Physical Review Letters, September 2026). Run as a zero-dimensional model, that turns ignition from a static line on a graph into a constrained pathway through density and temperature space.

The headline result reverses the usual operating order. Most designs compress first and heat second, squeezing a relatively cool plasma before pumping in enormous power. The Princeton calculations say the cheap route runs the other way: heat the plasma to fusion temperatures first, then raise its density. In the idealised limit, with clean fuel only, the least demanding crossing point, which the paper calls the Cordey saddle, sits where fusion power reaches roughly five times the injected heating power (PPPL, October 2026). Put real physics back in and that crossing point moves uphill.

What moves it uphill is the set of losses an ideal plasma lacks: the helium ash left behind by burnt fuel, light and heavy impurities sputtered off the reactor walls, synchrotron radiation from electrons spiralling in the magnetic field, and heat carried out by conduction and convection (Delgado-Aparicio et al., 2026). Fold all four in at once, as earlier work rarely did, and the ridge, the saddle and the entry points shift toward higher density and temperature. Masayuki Ono put it plainly: “When you leave these effects out, you say the design will work fine” (PPPL, October 2026).

The most alarming finding concerns tungsten. More than a dozen next-generation fusion machines have chosen it for their inner walls, yet contamination at one part in 10,000 can roughly double the pressure a machine needs to reach ignition (PPPL, October 2026). In a realistic three-dimensional reactor the required plasma pressure climbs further, along with the ratio of plasma pressure to magnetic pressure, where magnetohydrodynamic stability limits begin to bite (Delgado-Aparicio et al., 2026). A trace of eroded wall metal can push a reactor past the point where its plasma holds together.

The same losses that make ignition harder also brake it. A stabilizing feedback from transport, impurity and synchrotron damping, and from the unfavourable way energy confinement degrades as heating power climbs could hold a burning plasma in steady state, which suggests the thermal runaway that has worried designers for decades is self-limiting (Delgado-Aparicio et al., 2026).

Why does the order of operations matter so much? Because the gain a plasma can reach depends on where it sits on the density-temperature map, and that map is not flat. Fusion power scales as P_f = n_D n_T ⟨σv⟩ E_f, the bracketed term being the fusion reactivity. For deuterium-tritium fuel the Lawson requirement n T τ_E ≥ 3×10^21 keV·s·m⁻³ (LLNL-TR-676592) is minimised not at the highest temperature but in the low tens of keV, where reactivity is still climbing but radiation losses have not yet taken over.

Every route to ignition is also a budget. In deuterium-tritium burning, 20 percent of the released energy leaves as 3.5 MeV alpha particles that stay trapped in the plasma and heat it, so a self-sustaining burn needs P_α ≥ P_loss + P_aux. The losses come from bremsstrahlung, which scales roughly as n_e Σ n_i Z_i² √T; from synchrotron radiation, which grows with the square of the field; and from transport, which strips plasma energy at a rate W/τ_E, a confinement time that degrades as power rises. Compress a cold plasma first and you pay for density while the fusion power is still nearly zero, then buy heat against losses that scale with that density. Heat first and the injected power is spent where the triple product is easiest to satisfy.

Two escapes from the tungsten trap appear in the same work: coats of liquid lithium, which PPPL has pursued for years, to intercept sputtered metal before it reaches the core, and spin-polarized fuel, whose aligned nuclear spins raise the fusion reactivity and reopen part of the ignition window (PPPL, October 2026).

The caveats are real. The work is entirely computational, no experiment has reached the temperatures where the Cordey saddle sits, and the authors’ next step is numerical experiments rather than a machine (Delgado-Aparicio et al., 2026). It was funded by the Department of Energy’s Office of Science, so the modelling is public.

Before a reactor is built, somebody has to decide whether to line its walls with tungsten, how hard to compress, and when to start injecting heating power. Those choices run to billions, and they have usually been made with an ignition model that leaves out the four losses that matter most.

The other route to a burning plasma is being mapped just as hard. On 8 October 2026, General Fusion reported that its LM26 machine compressed a plasma to an electron temperature of 1.1 keV, measured by Thomson scattering with the UK Atomic Energy Authority, the first time a magnetised target machine crossed that threshold (General Fusion, October 2026). Its next milestone is 10 keV.

The lesson from both is that ignition is a route rather than a number. The physics of getting there is decided by losses most design studies have filed as footnotes, and the machines that might get us there are being specified at the moment those footnotes are promoted to the main text.

 

 

 

On October 26, 2026, OpenAI posted a preprint with a blunt title: The Exponential Time Hypothesis is False. The paper carried a date of October 17, a revision of October 25, a byline reading “OpenAI,” and a Lean formalization in the company’s public mathematics repository. Within hours, Anthropic published its own answer to the same problem, A 2^(o(n)) Time Algorithm for 3-SAT, with a Lean 4 development and build instructions. Both labs say a model found the result on October 16. The conjecture under attack was proposed in 1999, and much of complexity theory has leaned on it since.

The problem underneath is 3-SAT. An instance is a list of clauses, each one a disjunction of at most three literals over Boolean variables, and the question is whether some assignment of truth values satisfies every clause at once. Checking an assignment is trivial. Finding one by brute force means trying 2^n possibilities, and since 3-SAT is NP-complete, a polynomial-time algorithm for it would settle P versus NP the other way. The Exponential Time Hypothesis is a sharper bet. Put s₃ for the infimum of all exponents δ such that some algorithm decides 3-SAT on n variables in time O(2^(δn)). ETH says s₃ > 0: some fixed positive δ sits out of reach, so 3-SAT always costs a constant fraction of the full exponential. Russell Impagliazzo and Ramamohan Paturi proposed the hypothesis in 1999 and refined it with Francis Zane in 2001.

What makes it more than a guess about one problem is how much rests on it. ETH implies P ≠ NP, and its negation does not imply P = NP. Under ETH, k-Clique has no algorithm running in f(k) n^(o(k)) time, dynamic programming on bounded treewidth is optimal up to a log factor, and a long list of parameterized algorithms is optimal in the exponent. ETH also underwrites conditional circuit lower bounds: a theorem of Ryan Williams says that if ETH is false, the class E^NP has no linear-size Boolean circuits. Impagliazzo called the hypothesis a “principled educated guess” and put the odds at “a fair coin flip” in a 2012 lecture; Williams gave ETH about 70 percent odds of being true in 2019.

Both labs announced within hours of each other, and both shipped Lean proofs. OpenAI’s route was to enlist the people who know the terrain. Impagliazzo and Paturi at UC San Diego, who proposed the hypothesis, and Williams at MIT, who had already refuted strong versions of it, now lead an open examination of the claim, with a conference funded by OpenAI to work through the fallout. Williams told TechCrunch that in the OpenAI proof some parts are routine, “but then yes, there are two or three alien-looking parts,” and those are the pieces that convinced him the field needs to look at them together. His first question to OpenAI’s science lead Dan Roberts was whether any cryptographic system is now insecure; they had checked, he reports, and were reasonably confident nothing breaks. The release ran wider than 3-SAT: OpenAI also claimed the local smoothing conjecture in every dimension, which implies Kakeya, and existence for Yang-Mills theory in four dimensions, settling half of a Clay Millennium Prize problem. That manuscript stays with the group David Gross leads, a choice OpenAI’s advisory panel has publicly opposed.

Anthropic’s artifact is unusually well documented. Its paper, posted on the Hexagon preprint server under a CC BY licence, states the result in the abstract: a deterministic algorithm for 3-SAT running in 2^(o(n)) time, which also counts satisfying assignments, the harder #3-SAT problem. The provenance statement says the core algorithm was found on October 16 by an internal Claude research model using 14 million output tokens and no human input, after which Anthropic researchers worked with Claude on the write-up. The GitHub repository pins Lean 4.33.1 and the matching Mathlib, explains how to build the development, and notes that a second, independently written kernel called nanoda re-checks the proofs alongside Lean’s own. The README reports 1,050 modules, written by AI models and reviewed by other instances of the same models. Its strangest entry is a condensed description of the proof, enough, the README says, for a current Claude model to rederive the refutation on demand.

The mathematics behind Anthropic’s paper starts with the sparsification lemma, which lets you assume the formula is sparse, so its clause count is proportional to its variable count. The counting problem then becomes a sum of products over a graph whose nodes are clauses and whose edges carry the shared variables, and it can be split recursively as long as the bits crossing each cut stay few. The lever that makes it work is an identity two centuries old: Degen found it in 1818, Graves rediscovered it in 1843 through the octonions, and it says a match between two pairs of 3-bit vectors can be written as a product of sign functions, 2[x=v][y=u] = [x⊕y⊕u⊕v=0](σ(x,y)σ(v,u) + σ(x,u)σ(v,y)), where σ is a table of plus and minus signs. It spans 12 bits and can be checked at all 4,096 settings. Laying the two terms out as a grid turns each crossing into a choice between merging rows and merging columns, and iterating that, with pieces built at rate ρ = 1/√2, yields a runtime of 2^{O(min{km/√(log m), k²n log log n/√(log n)})}.

OpenAI’s paper reaches the same conclusion by a different road, giving a deterministic algorithm that counts satisfying assignments exactly, built from polynomial interpolation and sparse parity tests on low-degree forms, running in (n+m)^{O_k(1)} 2^{n/(log n)^{1/2+o(1)}} time with polynomial space. Both results are asymptotic in a way that matters: Anthropic’s bound, for a formula with the usual five clauses per variable, does not beat exhaustive search until the variable count passes 10^(10,000), and its bound in n alone does not bite until n passes 10^(100 million). Brute force is already unusable at 100 variables. Neither paper makes anything faster in practice.

What the Lean files do and do not establish is worth sitting with. Anthropic’s MainResults.lean holds no proofs at all, only definitions: def ETH : Prop := 0 < s 3 and def ETHIsFalse : Prop := ¬ ETH, followed by the main counting theorem. The proofs live in a separate file, and the README is blunt about the limit of machine checking: “Lean checks the proofs, but no machine checks that this file says what the paper says.” The formalization proves the stated proposition, not that the stated proposition is what the field means by ETH, and that gap is now a human responsibility. The paper’s appendix adds that the Lean development was written by AI models, reviewed by other models, and that no person is known to have checked it. OpenAI’s side has the same shape and a bigger pile: 721 manuscripts in 373 families, some formalized and some not, produced by posing roughly 4,000 problems to an unreleased model, each result averaging about three hours of thinking compute.

The consequences, if the work holds, are wide and uneven. The counting corollary gives s_k = 0 for every k ≥ 3, so s_∞ = 0 and both ETH and SETH are false. A literature scan carried out by Claude turned up more than 800 papers that prove a result of their own only under ETH, or under a hypothesis implying it. The Orthogonal Vectors hypothesis is untouched, though, and many fine-grained lower bounds rest on that rather than on ETH, so a large share of the lower-bound literature survives. P versus NP survives with it, with Williams describing the effect as a shift in priors rather than a solution.

What changed most is procedural. Two frontier labs have put the same kind of artifact in public view on the same day: a theorem the field spent decades formulating, a proof written by a model, and a machine-checked formalization that settles the steps but not the meaning. The checking is being organized as a community practice, with the proposers of the hypothesis running the examination. Credit is not, and the release policy is still being argued over in public, which is why the advisory panel OpenAI convened has been pressing for the physics manuscript to be released.

 

 

 

For an hour, the ceramic-lined furnace at Rex Heat Treat in Bedford, Pennsylvania, climbed from room temperature to 465 degrees Celsius. The vessel inside was titanium, about a meter across, the size of an exercise ball. The DAVINCI team put it through that profile six times. Six times it came out the same shape it went in. Each run mimicked the way the probe will heat up over the fifty-five to sixty minutes of its one-way trip through the atmosphere of Venus, a place whose surface sits at 740 kelvin and a pressure equivalent to nine hundred meters of seawater. The fact that the engineering development unit survived its own near-death preview is the news from NASA Goddard this week, and the reason scientists who study Venus are starting to plan around a real measurement instead of a hypothesis (NASA Science, 24 September 2026).

What DAVINCI is after is older than the program. Venus and Earth started as near twins, similar size, similar rock, similar distance from a young Sun. Today one is a temperate water world and the other is a planet whose surface would crush a submarine and whose clouds rain sulfuric acid. The question of when and how that divergence happened, and whether Venus ever had liquid oceans, has been the central question of inner-Solar-System science for two generations. The Pioneer Venus Multiprobe dropped four small entry probes in 1978, and Vega 2 carried the Soviet Vega program’s last successful lander to the surface in 1984. Venera 14 returned the only close-up images of Venus’s surface, and they are forty-four years old. Every chemical measurement below forty kilometers of altitude that exists today was made by instruments that fit inside a Soviet-era school bus and ran for a few minutes before the heat killed them (Garvin et al., 2022, Planetary Science Journal 3, 117).

The mission NASA picked to fill that gap is named, improbably, after Leonardo. Deep Atmosphere Venus Investigation of Noble gases, Chemistry, and Imaging started as a Discovery proposal in 2015, lost to Lucy and Psyche, came back as DAVINCI+ in 2019, and was selected in June 2021 alongside its companion orbiter VERITAS. The flight system is two pieces: a Lockheed Martin carrier that will perform two flybys of Venus to map cloud motions and surface composition, and a descent probe that will be released about two days before atmospheric entry. The current schedule puts launch in December 2030, atmospheric entry in 2034 or 2035, and a one-hour descent that ends at Alpha Regio, a tessera, a piece of the old continent-like crust that may hold the chemistry of whatever ocean Venus once had (NASA, Discovery Program selection, 2 June 2021; Wikipedia, DAVINCI).

The August furnace tests at Rex Heat Treat are about the inside of the probe, not the heat shield. The DAVINCI descent sphere, an oblate titanium pressure vessel about 0.98 meters by 0.85 meters and roughly two hundred kilograms, is what carries the science instruments down. The carbon-phenolic aeroshell that surrounds it for entry, a 2.24-meter cone designed to absorb peak heating above ten thousand degrees Celsius during the minutes of peak dynamic pressure, burns away and falls off below the cloud layer. After that the descent sphere is on its own, passively cooled by a layered thermal-soak design and by the buoyancy of the lower atmosphere, broadcasting measurements for an hour through a relay on the carrier spacecraft nine thousand kilometers overhead. The team put more than a hundred thermal sensors on instrument simulators inside the development unit and watched them through every cycle. Kristen Brown, the probe manager at Goddard, called the system a huge engineering challenge because the instruments have to be kept alive while the probe deliberately lets hot Venus gas in through inlet ports to feed the mass spectrometer and tunable laser spectrometer. The tests verified that the layered insulation, the inlet geometry, and the internal layout can do both at once (NASA Science, 24 September 2026; Garvin et al., 2022).

The science payload is the reason anyone cares about an hour of in-situ measurement. DAVINCI carries five primary instruments inside the descent sphere plus a context imager. The Venus Mass Spectrometer and the Venus Tunable Laser Spectrometer are tuned to read the ratios of noble gases, sulfur species, and water-related molecules with enough precision to test the leading hypothesis for Venus’s history. That hypothesis goes like this: early Venus was hot but had water; the water dissociated in the upper atmosphere, the hydrogen escaped to space, and a runaway greenhouse locked the planet into its present state. If the mass spec finds a primordial argon and neon inventory that matches a wet early Venus, the runaway model survives; if the ratios look like the mantle of a dry planet, the story changes. The Venus Descent Imager and the Venus Imaging System for Observational Reconnaissance, riding on the descent sphere and the carrier respectively, will produce the first sub-hundred-meter pixel-scale images of tessera terrain and the highest-resolution images of the surface since the Venera landers. Below the clouds the descent sphere’s pressure rises to roughly ninety-two bar and the temperature pins to four hundred and sixty-seven degrees Celsius, conditions the instruments are rated for, but only if the thermal design lets them keep measuring all the way down (Garvin et al., 2022; Wikipedia, Atmosphere of Venus).

Two engineering choices make that possible. First, the descent sphere is hermetically sealed: every gas path between the hot outside and the cold inside is broken by a long, narrow inlet that lets flow in without letting heat conduct in. Second, the relay link to the carrier operates at X-band through a high-gain antenna on the back shell, with the carrier recording the descent and playing it back to Earth later. Together the choices turn the descent sphere into something like a deep-ocean submersible, a thin titanium shell, internally cold, externally roasting, surviving because the thermal mass of the vessel is large compared to the heat leaking in through the instrument feed-throughs. The descent profile is also gentler than a Mars entry. Venus’s thick atmosphere is itself the brake, dropping entry velocity to about thirty meters per second by the time the parachute stage jettisons and the sphere free-falls the last thirty kilometers of cloudy, acidic air to the surface.

The test campaign is the second of two major engineering milestones the team targeted for 2026. The first, in June at Crater Island in Utah, was a VADIX helicopter demonstration of the descent imager concept under realistic lighting and dust; the second, just completed in Bedford, was the thermal survival test of the descent sphere itself. The team now turns to final design and fabrication, with a third milestone likely in 2027 covering full-up entry-vehicle qualification, then integration with the carrier and launch from Cape Canaveral in the December 2030 window. If the schedule holds, the first new measurements from below Venus’s cloud base in fifty years will arrive on Earth in the spring of 2035. They will not settle the runaway-greenhouse question on their own, but they will give planetary scientists a starting point that no present model of Venus has, measured chemistry, measured wind, measured surface composition, at every altitude from seventy kilometers down. That dataset, not the next spectacular image, is what will change the science.

 

 

 

For the first time, an engineered bacterium has built a useful polymer using only what Mars has on the surface. No Earth fertilizer, no pre-processed feedstock, no shipped-in sugar. The strain, called sPL.001, was announced in late September by Pioneer Labs, a small nonprofit in Emeryville, California, and the underlying paper landed on bioRxiv on 14 September 2026 (Martin et al., DOI 10.64898/2026.09.14.748363). It is the first of five microbes the team plans to field on Mars, and the first to clear a load-bearing question that has hung over every serious terraforming proposal for forty years: can Earth life actually eat Martian dirt?

The case for carrying biology to Mars goes back to the early days of in-situ resource utilization. The standard playbook for a future base assumes you will land everything, refine everything, and recycle almost nothing; biology changes that calculus by adding a self-replicating partner. Pioneer Labs was founded in 2024 by synthetic biologist Erika DeBenedictis, who had cut her teeth on interplanetary mission design as a JPL undergraduate, and the company has spent the last two years narrowing the most ambitious version of the idea down to a single concrete first step (Pioneer Labs, 28 September 2026). The first strain had to be useful on day one, on the path to terraforming, and provably unable to escape and survive in the wild. sPL.001 is the answer they settled on.

What the strain does is unglamorous and important. It pulls carbon, nitrogen, phosphorus, and trace metals out of a slurry of leached Mars regolith, fixes acetate the bioreactor makes from atmospheric carbon dioxide, and stores what it cannot immediately burn as polyhydroxyalkanoate (PHA), the same class of biodegradable polyester that already shows up in compostable cups and medical implants. C. necator H16, the parent, is one of the best-studied PHA factories in biotechnology, with gram-per-liter yields from sugar feedstocks on Earth that have kept industrial microbiologists employed since the 1970s. The trouble is that the Martian surface is a hostile place for it. The regolith is loaded with perchlorate salts, the water activity is negligible, the temperature swings are 80 °C, and the only carbon comes from a thin CO2 atmosphere measured in millibars.

The team’s first move was to screen sixteen candidate chassis organisms in a chemically defined Mars medium that mirrored the soluble fraction of regolith without shipping the actual dirt. C. necator H16 and Pseudomonas putida KT2440 both grew, badly; everything else either stalled or died. They then put the survivors through more than 10^12 cumulative cell divisions of adaptive laboratory evolution, serial passaging under progressively higher acetate and perchlorate stress until a single C. necator isolate, sPL.001, kept dividing. The headline result is a more than three-fold increase in PHA titer compared with the parent strain under the same simulated Mars conditions (bioRxiv preprint, 14 September 2026). That is the engineering win: not a new species, but a strain whose metabolism is rewired for a chemistry no Earth bacterium has ever seen concentrated in one place.

PHA itself is a microbial carbon-storage polymer, a polyester that the cell lays down as intracellular granules when carbon is plentiful and another nutrient, usually nitrogen, is not. The monomer in sPL.001’s case is overwhelmingly 3-hydroxybutyrate, so the material is poly-3-hydroxybutyrate (PHB), a thermoplastic with tensile properties in the same neighborhood as polypropylene and a melting point around 175 °C, both of which matter for habitat shells. In a Mars bioreactor the polymer is harvested from lysed cells, dried, and fed into an extruder or a 3D printer; the resulting panels can be radiation-shielded, pressurized, and stacked into a shell whose raw atoms are 100% Martian (Pioneer Labs, 28 September 2026). That is the second engineering win, and it is the one DeBenedictis leans on: the first rocket lands one bioreactor’s worth of kit, and four years later you have a settlement’s worth of plastic.

The physics of the bioreactor is what makes the system honest. The cells are kept at 30 °C in a stirred, heated vessel that shields them from Mars ambient temperature, vacuum, and UV; they are fed a continuous trickle of acetate from an electrochemical cell running on compressed CO2 and a small amount of water. Outside that controlled environment, the strain is engineered to be helpless. sPL.001 is not photosynthetic, so without the fed acetate it cannot multiply, and without multiplication it cannot evolve. In dry, irradiated, water-starved regolith, desiccation alone sterilizes the cells within hours. A separate arXiv paper, posted in late September 2026 under DOI 10.48550/arXiv.2609.12015, walks through the contamination math: even a catastrophic leak of an industrial-scale Mars bioreactor, the team argues, would clear the COSPAR planetary protection threshold by a factor of several hundred. The numbers do not yet constitute a regulatory verdict, but they are the first quantitative case that an early industrial bioprocess on Mars can coexist with the unfinished business of searching for extant life.

The reception inside astrobiology is measured. Chris McKay, a planetary scientist at NASA Ames, told Scientific American that the organism is “neither revolutionary nor particularly complex” (Scientific American, 2 October 2026). He meant it as a description, not a dismissal. Cupriavidus necator has been in industrial culture collections for half a century, and the pathway that converts acetate into PHA granules is the same one that runs in any overfed microbe on Earth. What is new is the substrate. Neveda Naz, an astrobiologist at Tufts who has grown microbes in material from actual Martian meteorites, called the work a “useful engineering technique” because the team “picked an organism, adapted it to a defined Mars-relevant chemical environment, and demonstrated improved production of a useful material” (Scientific American, 2 October 2026). In other words, the contribution is not the bacterium, but the proof that the bacterium can run on Mars dirt alone, at yields that scale.

That distinction is what makes this a story about engineering, not biology in the usual sense. The bioreactor is a chemical plant whose feedstocks are regolith, water, and air, and whose product is a structural material. The cells are the catalyst. The same logic, run in reverse, is what makes every modern biorefinery work: the organism is a black box that turns cheap local feed into a polymer you would otherwise have to import. The local feed here is just unusually inhospitable, and the polymer ends up holding up a habitat instead of going into a trash can. DeBenedictis has framed it in exactly that language: she called the project “a recipe for building the log-cabin equivalent for Mars” (Asimov Press, 22 September 2026), the first dwelling in a long chain of manufactured ones.

What comes next is the harder part. Pioneer Labs is already engineering a second strain, a perchlorate-eater that would let biology work on regolith that has not been rinsed, and three more chassis organisms are planned to handle nitrogen fixation, photosynthesis, and food-chain starter cultures respectively. None of those will be as photogenic as the first, because the first gets to do the simple thing, eat acetate, store plastic, and stop. The next four have to wrangle Martian chemistry in the open, and the planetary protection math gets tighter with each one. A first launch is years away, and an actual bioreactor on the surface is years past that, but the science question that used to sit on top of the program, can Earth life actually run on what Mars already has, has now been answered in the affirmative, in a peer-reviewable preprint, with a strain whose kill switch is the absence of a feed line.