A 21-year-old conjecture about random networks just got proved — and it lets mathematicians borrow decades of theorems from one kind of graph to solve problems about another.
Mathematicians study two flavors of random graph: binomial random graphs, where every possible edge is flipped in independently like a coin toss, and random regular graphs, where every vertex is forced to have exactly the same number of connections — closer to how real networks actually look, but much harder to analyze. In 2004, Jeong Han Kim (then at Microsoft Research) and Van Ha Vu (then at UC San Diego) conjectured a shortcut: that a random regular graph could always be mathematically "sandwiched" between two binomial graphs of slightly different densities, so that any property shared by both binomial graphs automatically holds for the regular graph squeezed between them. The conjecture held for two decades until 2025, when Richard Montgomery, Natalie Behague, and Daniel Iľkovič, all at the University of Warwick, finally proved it. "The deep connection between the two seems almost too good to be true," Behague said, while Tel Aviv University's Michael Krivelevich called the conjecture "very natural" and admitted it had been "kind of annoying not to have it proven yet." The payoff is practical: a large body of existing theorems about binomial random graphs can now be carried over wholesale to regular graphs without re-proving them from scratch.
Why it's hereProgress in pure math is often not a flashy new object but a quiet "bridge" result that unlocks a whole literature at once — twenty-plus years of scattered effort resolved by a metaphor even non-mathematicians can picture.
A years-long staring contest between the Webb telescope and a swarm of impossibly compact red objects has split astronomers into two camps.
Since its earliest images, the James Webb Space Telescope has kept turning up "little red dots" — pinprick-bright, reddish objects that don't fit standard galaxy or black-hole models. In March 2025, three separate research teams independently proposed a radical explanation: some little red dots may be "black hole stars," supermassive black holes wrapped in hydrogen envelopes potentially dozens of times wider than our solar system. Anna de Graaff of the Max Planck Institute, whose team found two especially strange examples, put it bluntly: "In all the millions of [observations] we've taken with ground-based telescopes, there's nothing that looks like these sources." Not everyone agrees — Cambridge's Roberto Maiolino argues these are just ordinary supermassive black holes seen at unusual angles — and MIT's Anna-Christina Eilers says plainly that "the field has gotten very polarized." Both sides are now marshaling new evidence: an August 2026 census led by Colby College's Dale Kocevski found little red dots essentially vanish once the universe passes 2–3 billion years old, while a September 8 analysis by Rohan Naidu's team at the University of Hawai'i, probing the mass of the black hole "seeds" inside them, led Naidu to declare, "We are seeing the seed. This is the birth of potentially every massive black hole in the universe."
Why it's hereThis is a live, unresolved fight playing out among JWST observers in real time, not a single settled result — worth watching rather than filing away.
A rule chemistry students have memorized for nearly a century turns out to reach about a third as far as the textbooks claim.
Chemists have taught for decades that the "inductive effect" — the way an electron-withdrawing atom or group pulls electron density through a chain of bonds — fades out gradually over three or four bonds. A study in the Journal of Chemical Education, led by Cardiff University's Mark Elliott with Edwin Johnson (University of Newcastle, Australia), Kasimir Gregory (University of New England), and Colan Hughes, used modern computational methods to directly examine molecular electron distributions and argues that's wrong: in a neutral molecule, the inductive effect essentially doesn't extend past a single bond, and what looked like effects further out was actually a different phenomenon — through-space electrostatics — being misattributed. The paper consolidated decades of scattered experimental data rather than running one new experiment, but the practical fallout has been immediate: two UK A-level exam boards have announced reviews of how they teach the inductive effect and cited the research directly as the reason. Predicting reactivity and electron distribution in drugs, agrochemicals, and polymers often leans on inductive-effect reasoning taught at exactly this level of detail, so the correction reaches well past the classroom.
ScienceDaily, reporting Cardiff University / J. Chem. Educ. 2026;103(6):3156 · Sept 14, 2026 · sciencedaily.com
§
Why it's hereA rare case of a genuinely foundational chemistry teaching point being overturned by data rather than debate — and it's already visibly changing what students are taught, a concrete consequence rather than just an academic claim.
Your forebrain and hindbrain didn't develop from the same starting cell — they were stitched into one organ, and researchers can now watch the seam form.
Stanford Medicine researchers led by developmental biologist Kyle Loh, with graduate students Carolyn Dundes and Rayyan Jokhai as co-first authors, traced mouse embryos through early development and found that the brain isn't built from a single progenitor population. One group of cells, marked by the gene Otx2, gives rise to the forebrain and midbrain; a wholly separate population, marked by Gbx2, forms the hindbrain. "We've shown for the first time that the front of the brain arises from a totally different progenitor cell than the back," Loh said — the two lineages "never overlap; they are mutually exclusive from the earliest stages of development," carrying different chromatin packaging from the start. Using that insight, the team also grew functional human hindbrain motor neurons from pluripotent stem cells for the first time — tissue that's been hard to access because it sits deep at the base of the skull — and the lab-grown neurons fired action potentials and made proteins marking the segments that control facial and swallowing muscles.
Why it's hereIt reframes a textbook assumption — one brain, one origin — as an evolutionary merger of two lineages, with a concrete translational hook: lab-grown hindbrain neurons could open up research on diseases centered on hard-to-reach brainstem tissue, like ALS.
Brain scans of long COVID patients turned up a shortfall of dopamine nerve terminals — one that lines up strikingly well with which symptoms people actually report.
Researchers at Toronto's Centre for Addiction and Mental Health, led by senior scientist Jeffrey Meyer, used PET imaging to measure a marker of healthy dopamine nerve terminals in people with long COVID and found substantially lower levels of it compared with matched people who'd recovered fully, concentrated in three striatal brain regions. The shortfall wasn't uniform, and it tracked with specific symptoms: reductions in the ventral striatum lined up with greater loss of motivation, in the dorsal putamen with slower movement, and in the caudate putamen with memory difficulties. "Our findings provide compelling evidence that long COVID involves the loss of dopamine-releasing neurons," Meyer said — the first imaging work to tie long COVID's "brain fog" cluster to a specific, localized neural mechanism rather than a vague inflammation label. It's a small first study, and Meyer's team says it isn't yet known whether the loss is reversible; a follow-up trial testing dopamine-targeted treatment is planned.
Why it's hereLong COVID brain fog has been one of medicine's most stubborn "we know it's real but can't explain it" symptoms — this gives it a concrete, localized, and plausibly treatable mechanism instead of a diffuse inflammation story.
A bioethicist says autonomous AI will beat human doctors at core clinical tasks by 2030. The AMA's CEO says that's the wrong lesson to draw from the same evidence.
In a September 9 essay, bioethicist Ezekiel Emanuel and co-author Abe Baker-Butler argue that fully autonomous AI — not AI-assisted doctors — will outperform physicians on core clinical tasks within a few years. Their evidence: Google's AMIE was rated better than physicians at taking patient histories (97% of patients felt comfortable versus 65% with physicians); ChatGPT beat physicians at differential diagnosis 92% to 74%; Microsoft's AI Diagnostic Orchestrator reached correct diagnoses roughly four times more often than physicians while cutting testing costs; and a system called MIRA prescribed guideline-concordant treatment 35 percentage points more often than doctors did. Emanuel's sharpest claim is that when an AI system is good enough, human oversight can make outcomes worse, not safer, by adding errors and false corrections. The same day, STAT ran a rebuttal from American Medical Association CEO John Whyte, who accused Emanuel of "a leap in logic from 'AI can do many tasks physicians do' to 'AI can replace physicians.'" The two pieces, published side by side, leave the actual question — should AI ever practice without a human physician in the loop — genuinely open.
Why it's hereA sharper question than the usual "AI in medicine" hand-wringing — argued head-to-head by a leading bioethicist and the AMA's own CEO on the same day, with real benchmark numbers behind both sides.
Sources this issue
Quanta Magazine — math
Quanta Magazine — physics
ScienceDaily / Cardiff University / J. Chem. Educ. — chemistry
Stanford Medicine News / Nature Neuroscience — biology