Key points
- On February 10, 2025 (3:55 AM), a paper titled "A Counterexample to the Mizohata-Takeuchi Conjecture" appeared on arXiv with a single author: Hannah Cairo, then 17 years old, homeschooled, with no bachelor's degree and no high school diploma, living in Nassau, Bahamas.
- Cairo learned calculus by age 11 via Khan Academy, then studied under remote tutors Martin Magid (Wellesley) and Amir Aazaim (Clark). Aazaim recalled he was embarrassed to charge tuition because "I wasn't teaching her—she read and proved theorems on her own."
- In fall 2023 she began taking UC Berkeley graduate-level math courses; her family moved from the Bahamas to Davis, then Berkeley, so she could commute to classes—five days a week by spring 2024.
- 6 rejected her for lacking a bachelor's degree
- 2 initial admissions were overturned by school administrations
- Only University of Maryland and Johns Hopkins accepted her
- AI on Jacobian: brute-force search over millions of combinations—a victory of "trying more things."
- Cairo on Mizohata-Takeuchi: constructive intuition and aesthetics—a victory of "thinking deeper." She found an ugly-but-working construction and kept refining it until it was beautiful. "Beauty has no computable metric; this aesthetic cannot be encoded in a reward function."
- Quanta Magazine: At 17, Hannah Cairo Solved a Major Math Mystery
- arXiv:2502.06137: A Counterexample to the Mizohata-Takeuchi Conjecture
- IBM Think: AI cracked the conjecture. Humans called the play.
- Terence Tao's discussion on Mathstodon
The homework problem she didn't stop at
In the 2024–2025 academic year, Cairo took Ruixiang Zhang's course on Fourier restriction theory. Zhang, a 2008 IMO gold medalist, assigned a simplified version of the Mizohata-Takeuchi conjecture—a 40-year-old open problem that had been neither proven nor disproven. Cairo completed the assignment, then kept going:
> "Why should I stop?"
What the conjecture says
Harmonic analysis studies how to decompose complex functions into simple waves—the Fourier legacy underlying mobile communications and MRI. The Mizohata-Takeuchi conjecture asks: if a function's frequencies all lie on a surface (e.g., a sphere), how can its energy be distributed? The analogy given: in a room with sound-absorbing walls, acoustic energy may amplify and cancel in places, but the conjecture claims energy cannot concentrate into arbitrarily small regions no matter the geometry. The conjecture connects to the Stein conjecture, a "central hub" of the field, and for 40 years everyone believed it true—Carbery (Edinburgh) worked on it for 40 years; Oliveira (Birmingham) spent two years trying to prove it.
Constructing a function that shouldn't exist
Cairo built a function from waves with frequencies on a surface that, instead of canceling into uniform energy spread, produces a fractal-shaped energy distribution—concentrated in some regions, sparse in others, self-similar like a Koch snowflake—violating what the conjecture said was impossible.
> "I often run into situations where I think I've proved something, but I'm actually wrong."
Two things convinced her: she found a much simpler construction achieving the same result, and she persuaded Zhang. The simplification was not functional—the original already refuted the conjecture. It was for beauty. Oliveira called the paper "a model of elegance—natural, concise, striking at the heart."
The collapsed network
Mizohata-Takeuchi was considered a consequence of the Stein conjecture; if MT held, Stein would be more credible. Cairo's counterexample implies the Stein conjecture is false—the connections mathematicians hoped to build through it do not exist. Carbery: "I was completely blown away... even more impressed when I learned how young Cairo is. The paper is written with unusual elegance." Terence Tao shared the paper on Mathstodon.
Rejected by 6 of 10 graduate programs
After the paper, Cairo skipped undergrad and applied to 10 PhD programs:
She will pursue a PhD at Maryland—the PhD will be her first degree ever.
Contrast with AI: 'Why should I stop?'
The post contrasts Cairo with AI solving the Jacobian conjecture (open since 1939, allegedly overturned in July 2026 by Anthropic's Claude Fable 5, per an IBM Think commentary: "It's not that AI is smarter—it tried more things."). The two approaches are opposite:
The post's thesis: AI stops when reward signals stop. Cairo's drive was innocent curiosity—the problem was still there, so why stop?
Why refutation is more informative than proof
Proving a conjecture says the world is as you imagined. Refuting one says the world is stranger than you imagined. A single counterexample can collapse an entire network of hoped-for connections, forcing the field to redraw its map. Cairo's work collapsed Stein, Mizohata-Takeuchi, and many "hoped-for connections" in harmonic analysis—not failures, but a more precise map.
> "Mathematics is another world I can explore. An unbounded world I can enter anytime through thinking."
From a seaside house in the Bahamas, a 17-year-old entered a place no one had reached in 40 years—then told everyone: there's nothing there. An emptier, but truer map.