Edition of October 22, 2026

Some exploding stars brighten slowly, over about 100 days. Their model is supernova 1987A. In 2023, sky surveys caught the most extreme one ever seen: SN 2023iop.

It took at least 88 days to peak. It shone about four magnitudes brighter than 1987A and released at least 4 × 10⁵⁰ erg. An ordinary explosion driven by neutrinos cannot supply that.

One line in its late light may come from stable nickel, about 0.02 solar masses. Only the most massive "pair-instability" models make that much: a star of some 190 suns destroyed by a thermonuclear blast.

The authors call it the most plausible scenario. But it rests on a single spectral line, and no explosion model on its own explains the extra glow after day 150. The James Webb telescope will look again.

Source: SN 2023iop: An Extreme 1987A-like Supernova and the Death of Very Massive Stars, https://arxiv.org/abs/2610.27312

An Earth-like planet tugs its star back and forth at just 8.9 centimetres per second. Today's planet hunters reach that only with huge spectrographs that average thousands of spectral lines.

A team in Colorado borrowed the trick of optical clocks. A laser locks onto the centre of one iron line in sunlight, at 1,565 nanometres. A frequency comb counts its frequency against the official second.

In one second, they pin the line to 4.7 metres per second. After a few thousand seconds, about 30 centimetres per second: one part in a billion. That closely tracks the quantum limit set by the photons.

The setup even caught the Sun's five-minute oscillations and the eclipse of April 2024. Next target: nearby Sun-like stars, with a 30-metre telescope.

Source: A Solar Optical Clock, https://arxiv.org/abs/2610.28485

A strong ultrasonic wave can hold small objects floating in mid-air. But they often start spinning on their own, and nobody could predict which way.

In Austria, physicists 3D-printed tiny stars, one millimetre wide. Straight shapes never turn in a well-aligned field. Stars with curved arms spin at once, in the direction of the curve.

Then a surprise. Three or five arms: louder sound, faster spin. Ten arms: the star peaks near 7–8 volts, slows, stops, and reverses. Twelve arms reverse near 10 volts.

A simple two-term model explains every shape. It even let the team design a hybrid star that switches direction at the predicted voltage. Where the reversing force comes from is still open.

Source: Programmable nonlinear rotation of acoustically levitated chiral stars, https://arxiv.org/abs/2610.27927

Picture a frictionless ball bouncing forever on a table, like light between mirrors. Does it end up visiting every place, in every direction? Mathematicians call that "ergodic".

A "lemon" table is where two discs overlap: a small disc cut on one side by the arc of a huge circle. In 2013, four researchers conjectured it is ergodic when the two radii differ a lot.

Wentao Fan and Boris Hasselblatt have now proved it, in 58 pages. It is, as far as they know, the first ergodic billiard whose edge bulges outward everywhere. The big circle must be huge: at least 300,000 times the small one for the widest cut allowed, and more for a narrower cut.

Their next question: does such a table exist without any corners at all? They note: "No AI was used."

Source: Strictly convex ergodic billiards, https://arxiv.org/abs/2610.28247

In 2024, Korea's top baseball league became the first major league to let a machine call every ball and strike. The zone is a box scaled to each batter's registered height.

Two Korean economists studied over a million pitches from 2021 to 2026. Human umpires had habits: a wider zone when the batter was ahead in the count, a narrower one with two strikes, a little mercy for stars.

The machine erased them all, including the art of catchers who "frame" pitches into strikes. It also lifted the top of the zone for tall batters: 0.75 points more high strikes per centimetre.

Yet few pitches sit on those edges: at most one extra strike every 23 trips to the plate. Pitchers quickly aimed higher. Batters changed nothing. The shape of a rule decides who wins.

Source: Who Gains When the Umpire Is a Machine? Evidence from the Automated Ball-Strike System (ABS), https://arxiv.org/abs/2610.27576

An excavator cuts one buried conduit. A company loses its main internet link and its backup, bought from another provider. Both fibres ran through the same pipe.

Two researchers call this "illusory resilience". Routes that share no network on paper can share a building, a trench or a shoreline. They measured it with public data only.

Among 47 major operators, providers in the same region share far more interconnection sites. A single campus in Marseille hosts 153 operator pairs from their sample.

Undersea cables are worse. In two of twelve routes between continents, every pair of cables lands within 25 kilometres of each other. That simple rule flags ten of eleven cable pairs damaged together in 2024.

Source: Disjoint in Name Only: Illusory Resilience in the Internet, https://arxiv.org/abs/2610.27125

Headphone cables, garden hoses, phone cords: everything long and flexible ends up in a tangle. But many natural filaments are curly at rest. Does that help or hurt?

At EPFL in Lausanne, physicists cast silicone rods, straight or helical, from 0.6 to 4.5 metres long. They tumbled each one for 20 minutes in a small cylinder, 15 times per setting.

Straight rods knotted more as they got longer, every time beyond 3.5 metres. Curly rods knotted up to 73% less. For the 0.6- and 1.2-metre rods, knots vanished entirely above a critical curvature.

Getting stuck is another story. The chance that a rod locks when pulled depends on its length, not its curl. Two kinds of tangling, two different causes.

Source: Intrinsic Curvature Controls Spontaneous Entanglement of Elastic Rods, https://arxiv.org/abs/2610.27342

Chinese, English and French share few words and little grammar. Yet they can tell the same story. How does the brain keep that common meaning?

A team in Shenzhen and Hong Kong reanalysed brain scans of 227 people. Many listened to The Little Prince, each in one of the three languages. Their decoder turns brain activity back into text.

Activity patterns differ, but the geometry of meaning matches: ideas close in one language are close in the others. This match grows from sensory areas to the brain's high-level regions.

A decoder trained on one language beat chance on another it had never seen. With few recordings in a language, borrowing from the others helped. That could matter for languages with little brain data.

Source: Multilingual Brain Decoding Reveals Shared Cortical Semantic Geometry, https://arxiv.org/abs/2610.26927

Climate change mixes two things: the push of greenhouse gases and aerosols, and the climate's own natural swings. We only have one Earth, so they are hard to separate.

US researchers trained a neural network on hundreds of simulations from 15 climate models. On models it had never seen, it beat six other methods at isolating the human-driven part.

Applied to the real record: the human-driven part of Arctic warming runs about three times the global rate. Natural swings pushed that to the observed four. Since 1980, the eastern Pacific lagging behind the west and the drying US Southwest are mostly natural, too.

The comparison also points at the models. They warm the Northern Hemisphere too much and dry the land too little, the authors suggest.

Source: Disentangling forced and internal components in observed climate evolution using spatiotemporal deep learning, https://arxiv.org/abs/2610.28219

Removing a tongue tumour means cutting 5 mm around it, with an electric scalpel that heats the tissue. Too slow, and the heat burns healthy cells. Too fast, and the tissue snags and the cut drifts.

Autonomous robots usually cut at one fixed speed. At Johns Hopkins, engineers fixed a thermal camera to their robot. It measures the heated zone in real time and adjusts its speed accordingly.

On pig tongues, the adaptive robot spread heat damage about half as far as a slow fixed speed. Its margin error was well below that of a fast fixed speed. It cut in 22 seconds, against 52 at the slow speed.

A blinded expert rated its cuts 4 out of 5, but found the edges jagged. This is dead tissue at room temperature: living patients are still far off.

Source: Velocity Optimization via Thermal Imaging Enhances Autonomous Tumor Resection, https://arxiv.org/abs/2610.27562

The growing Earth was covered by an ocean of magma. Iron droplets rained through it to form the core, carrying dissolved gases with them.

Two astronomers in Copenhagen and Lund simulated this rain for two ways of building a planet: from small pebbles, which deliver their volatile elements early, or from larger bodies that keep delivering them.

The timing matters. With early delivery, about 70% of the hydrogen and carbon ends up in the core, against at most half otherwise. Nitrogen, which stays in the air, mostly escapes.

What remains is a crushing primitive atmosphere: about 760 bar, mostly carbon monoxide, with hydrogen and a little nitrogen. Uncertain lab data still prevent a verdict between the two scenarios.

Source: When the core ate the atmosphere, https://arxiv.org/abs/2610.27788