A weekly notice from the margins of math and science
Marginalia
the good stuff, allegedly
Issue №1·7 September 2026
First issue. Marginalia is a weekly pull from mathematics, physics, chemistry, biology, and medicine — one item each, published every Tuesday. No ranking, no "top story" — just five things from five fields that were worth a second look this week, plus whatever's worth arguing about.
How fast does a random network go from mostly dry to mostly flooded?
Five mathematicians — Sahar Diskin, Philip Easo, Ritvik Ramanan Radhakrishnan, Benny Sudakov, and Vincent Tassion — proved the "supercritical sharpness" conjecture for every infinite transitive graph: once the probability of a connection crosses the critical threshold, the network doesn't slowly fill in, it floods almost all at once. The conjecture had stood for decades. What struck the field wasn't new machinery; one researcher described the proof as familiar pieces "put together" in a way nobody had managed before.
Two kinds of quantum particle that were never supposed to cooperate, doing it anyway.
Physicists at Monash University have predicted, on paper, that bosons and fermions — quantum particles that follow fundamentally different rules — can bind into a stable, self-supporting droplet with no container needed. It challenges decades of assumptions about strongly interacting quantum systems. Unlike a lot of theory, this one comes with a way to check it immediately: the predicted setup matches ultracold-atom experiments already running in labs today.
ScienceDaily · Monash University · Aug 20, 2026 · sciencedaily.com
§
Why it's hereA theoretical prediction earns space mainly because it's testable now — theory that outruns anyone's ability to check it is a different, less satisfying story.
A material class chemists have wanted in nanocrystal form for years — done by changing the solvent, not the chemistry.
Metal nitrides — gallium nitride, titanium nitride, niobium nitride — are useful, well-understood materials that chemists have never managed to shrink to nanocrystal scale. A team at the University of Chicago and Argonne National Laboratory got there using molten salts as the reaction medium, precisely controlling temperature and ammonia pressure to make nanocrystals from nearly a dozen nitride materials at once. The opening isn't just academic: flexible electronics and printable medical implants need materials in nanocrystal form that nitrides simply couldn't take before.
ScienceDaily · U Chicago / Argonne National Laboratory · Aug 27, 2026 · sciencedaily.com
§
Why it's hereOne technique unlocked roughly a dozen materials at once — a method-level result rather than a single discovery, sometimes the more useful kind.
Copy your whole genome and you usually die. Just often enough, you don't — and everything changes.
Whole-genome duplication — inheriting three or four full chromosome sets instead of two — is almost always fatal, and yet it's implicated in some of evolution's biggest leaps: vertebrate brains, spider silk, flowering plants. Researchers studying the New Zealand mud snail, which duplicated its genome less than a million years ago — a blink, evolutionarily — are watching in something close to real time how an organism survives what one biologist calls "the single biggest, most radical mutation" and reorganizes a genome it was never built to carry.
Why it's hereMost extinction-scale mutations only get studied after the fact, in species that already made it through. This is closer to catching the gamble mid-roll.
Novartis's antibody drops an inflammation marker tied to heart disease. Whether that becomes fewer heart attacks is still an open question.
Pacibekitug, Novartis's antibody against IL-6 — a protein central to immune signaling — produced sustained drops in inflammatory biomarkers in a Phase 2 trial, with higher doses working better. Inflammation has been a live suspect in cardiovascular disease for years. The honest caveat, stated plainly by the researchers presenting the data at a cardiology conference: moving a biomarker isn't the same as improving outcomes, and that study still has to happen.
A live panel at the International Congress of Mathematicians asks what AI actually threatens.
The consensus from the ICM 2026 panel: AI's gold-medal IMO run and its solved Erdős problems are real but narrow — "the first time a model answers a question, it's exciting. The seventh time... is far less interesting." The sharper worry wasn't capability, it was incentive: Ravi Vakil warned that publication-count pressure could push graduate students toward AI-solvable, low-hanging problems precisely when those problems are what trains a mathematician. One panelist called the bigger risk a "Trojan horse for anti-intellectualism" — funding cuts justified by AI claims that don't hold up.
Why it's hereEvery issue of this newsletter is itself a small entry in this argument. Worth reading the skeptics' case directly rather than only from a digest.
Sources this issue
Quanta Magazine — math, biology, and the ICM 2026 panel