Edition of October 23, 2026

When the universe was 470 million years old, a black hole was already feeding in a faint galaxy. Astronomers found it in public spectra from the James Webb Space Telescope.

Its tell-tale sign: a hydrogen line broadened by gas whirling at thousands of kilometres per second. It sits just past the edge of Webb's standard data processing, at 5.35 micrometres.

From that width, the team weighed the black hole: about 36 million Suns. Under another assumption, 2 million. Both are huge for so young a cosmos.

To grow that fast from the remains of a single star, it would have needed to swallow matter faster than the usual limit. Or it started out big. It is the most distant black hole weighed this way.

Source: Discovery of a Broad-Line Active Galactic Nucleus at z = 9.98, https://arxiv.org/abs/2610.28735

Each clock holds a single ytterbium ion in a trap. One sits in Teddington, United Kingdom, the other in Braunschweig, Germany. Two teams built them independently.

In under four weeks in 2025, a 2,220-km optical fibre through France compared their ticking, 12.6 days of usable data. Result: a difference of 1.0 × 10⁻¹⁸, with an uncertainty of 4.8 × 10⁻¹⁸. In other words, they agree.

According to the authors, it is the first comparison between two countries below 5 × 10⁻¹⁸. That target must be met before the second can be redefined with clocks like these.

The largest doubt is not the clocks. It is gravity: the doubt on the correction for each site's place in Earth's gravity field, worth about 3 cm of height. The same data also tightened one test of relativity — whether the speed of light depends on direction — about fivefold.

Source: International atomic clock comparison achieving 5×10⁻¹⁸ uncertainty, https://arxiv.org/abs/2610.29980

Solving a game means knowing who wins with perfect play. Small go boards were solved long ago, but under a simplified rule about repeated positions.

The rule real players use, superko, forbids recreating any earlier board. It makes solving brutal: a move's legality depends on the whole game's history.

Two Berlin researchers built Sanko, a solver that keeps its shortcuts honest under superko. It proves 5×5 go in 16 minutes on six threads. The previous best took 5 hours 15.

They solved 23 more boards, 11 never solved under any rule. And one answer changed: 4×7, a full-board win under the old rule, is worth only 12 points. They found the exact move where the published winning line breaks.

Source: Solving Go under Positional Superko: Sound Proof Transfer and New Board Values, https://arxiv.org/abs/2610.29414

China's giant FAST radio telescope found a pulsar spinning 21 times a second. It is not alone: another neutron star, of almost the same mass, orbits with it.

Their orbit lasts 79 days and is incredibly stretched: eccentricity 0.992, a record for any pair of stars. Most of the time they are far apart. Then, in just 46 minutes, they swing around each other at 1,140 km/s.

During the 2025 passage, the pulsar's signal vanished for 85 seconds behind its companion's magnetic surroundings. We see the orbit almost edge-on.

Such a shape requires a violent kick when the second star exploded as a supernova. Simulations give such systems odds below 1 in 200. Either we got lucky, or the theory needs revising. They will merge in 2.5 billion years.

Source: PSR J0653+0443: An Eclipsing Millisecond Pulsar in an Extremely Eccentric Double Neutron Star System, https://arxiv.org/abs/2610.29838

How few points does it take to stop a curve? Nagata's conjecture, from 1959, sets the limit for curves forced through 10 or more points, several times over.

In September, OpenAI released a manuscript, written by its AI model, that claims to settle it. Jenia Tevelev, a mathematician at UMass Amherst, studied it and published a guide for non-experts. The verdict: "remarkably simple".

The trick: let a smooth cubic curve pinch into a loop. The hard question then shrinks to filling in values on a triangle of dots, which is easy.

Tevelev also wonders about being the manuscripts' "first human reader". The notes call them a stunning achievement, but warn that checking such work falls on mathematicians, with little credit. They propose a fund paid by AI companies.

Source: On OpenAI's proof of the Nagata conjecture, https://arxiv.org/abs/2610.30050

Rooftop panels make power at noon. Electric cars charge in the evening. What if the parked cars stored the midday surplus and gave it back to the house at night?

Researchers copied the same Kyoto-sized city onto 5,000 places worldwide, changing only the climate. Rooftop solar plus car batteries covered 22% to 84% of electricity for buildings and driving. Adding the cars raised the median by 21 points.

Cairo and Mexico City stay near 80% all year: air conditioning peaks exactly when the sun shines. In Novosibirsk, winter falls to 20 to 30%.

Sunshine maps mislead. Among the world's 1,000 largest cities, 219 are badly underrated by solar resource alone. Most are in fast-growing regions of Asia, Africa and South America.

Source: Climate-Driven Supply-Demand Matching Reshapes Global SolarEV Decarbonization Potential, https://arxiv.org/abs/2610.29344

After some head surgeries, a hole remains in the skull. Specialists draw a custom implant by hand on a computer, which is slow and costly. Could ready-made AI models do it?

A Polish team tested three public 3D shape-completion models, two of them "foundation models", on 239 skulls. The bar was simple: beat doing nothing, and don't cut into the bone that's left.

None beat doing nothing on the 210 artificial holes. On 29 real cases, one did. But implantable results: zero out of 239, for all three. A model trained for this exact job managed 238.

The trap is that the outputs look convincing. The most skull-like one was no easier to fit. The authors' verdict: as released, these models are not usable in the clinic.

Source: Plausible but Unimplantable: A Trustworthiness Audit of Shape Completion Foundation Models for Cranial Defect Reconstruction, https://arxiv.org/abs/2610.30035

To check a silver coin, you can tap it and listen. Apps compare the ring with a library of known coins, which only works for coins already listed.

Sylvia Dong, a high-school student, and Takyiu Liu, of West Valley College in California, did better. From the ring, the coin's size and its volume, they compute a constant of the metal itself: how fast waves travel through it.

The trick for embossed coins: weigh them in water to get their true volume. Six copper rounds gave a stiffness of 119 GPa, right in copper's accepted range.

A fake can copy density, not stiffness. Gold and tungsten weigh almost the same per litre, yet waves cross them at speeds differing by a factor of 2.1. Tungsten itself stayed silent in their tests, so that check remains a prediction.

Source: An Absolute Acoustic Readout of Plate Wave Speed for Bullion Authentication, Using a Measured Fill-Factor Correction for Struck Coins, https://arxiv.org/abs/2610.29194

Some algal blooms give people on the coast breathing problems. In July 2025, on beaches in Saint-Jean-de-Luz, France, researchers caught whole cells of the toxic alga Ostreopsis in the air.

How do they fly? When a bubble bursts at the sea surface, a tiny jet shoots up and breaks into droplets. In dilute water, the first drop holds about ten times more of the small model alga than the water around it.

High-speed films show why: cells inside the jet rise faster than its tip and pile up in the first drop. No surface trick needed. A simple law predicts it without any fitted parameter.

In crowded water, algae act about ten times bigger than they are and pack less into the drop. Microplastics do not. Sea spray thus offers toxins a possible second way into the air.

Source: Hydrodynamics of microalgae in sea spray, https://arxiv.org/abs/2610.29719

A teacher eats what others hunt and has no more children. So why would evolution favour a stage of life spent teaching, especially other families' children?

Two researchers built a mathematical model. People hunt and have children, then "retire" to train young hunters. The food is shared by all, teachers included.

Result: teaching can evolve with no kinship at all, but only above a critical group size. A single early retiree gains nothing: there is no one of their kind to teach. Usually, a whole group must switch norms together, which is how culture changes, not genes.

Evolution also tends to produce fewer teachers than would feed the group best. In some conditions, longer-lived elders, cared for by others, close that gap.

Source: When to retire: the cultural evolution of post-reproductive oblique teaching, https://arxiv.org/abs/2610.30135