Ronald Graham posed a question in 1971 that stayed open for more than half a century: take any set of different nonzero whole numbers, and see whether you can rearrange them so that the running total after one number, after two, after three, and so on never repeats a value. The final piece of the answer arrived in February 2026, in a 27-page paper by Lisa Sauermann of the University of Bonn and Huy Tuan Pham of the University of Chicago, and it says the rearrangement always exists, Wired reports, reprinting the story from Quanta Magazine.

Graham was a mathematician who served as president of the American Mathematical Society and also as president of the International Jugglers' Association, a combination that turns out to be less random than it sounds. In a 1980 television interview he said many mathematicians and computer scientists are drawn to juggling, and put it down to "the search for patterns and structure." In juggling terms his question asks whether, given balls that each stay in the air for a different length of time, you can find a throwing order in which no two land on the same beat.

The problem gets hard in clock arithmetic. If all the numbers are positive the sums only grow, so every running total is new. If the set mixes positive and negative numbers the answer was already known to be yes. Graham asked about a finite world where numbers wrap around after a prime number p, so that adding 3 and 4 can return 0 when p is 7. He guessed the answer was yes there too.

Large sets, small sets, and a gap in the middle

The proof came as four papers rather than one. Alp Müyesser, then a graduate student at Oxford, found Graham's conjecture shortly after finishing a proof of his own in 2022 and realised the technique carried over. With his former adviser Alexey Pokrovskiy of University College London he took the case where the set contains almost every number up to p. A random ordering gets most of the way there, so the pair set aside a few numbers and scrambled the rest, then patched any stretch that added up to zero by swapping in a spare value. Their solution was published as a side result inside a paper about a broader problem.

Noah Kravitz, also at Oxford, came at it from the other end without knowing about that work. He found the conjecture in an online archive of unsolved problems and, with Benjamin Bedert, handled sets that are tiny next to p, on the scale of 100 numbers against a p of one billion. Their proof went up in September 2024. Müyesser saw it, got in touch, and the group extended his approach in a paper posted in August 2025.

That left medium sets, roughly half as many numbers as p. Müyesser said the existing methods did not work there and there were clear reasons they could not. Sauermann and Pham closed the gap.

Anti-concentration did the work

The two had met at Stanford in 2015, when Sauermann was a graduate student and Pham an undergraduate. A conference in Germany in September 2025 brought them back to the same chalkboard, and Pham followed Sauermann to Bonn for a short visit. They heard two talks by mathematicians who had tried and failed to bridge the gap. By the end of three days they had a plan.

It rested on a technique called anti-concentration, which proves that some event has a particularly low chance of happening. Kravitz said he and others knew the approach might succeed but had not had the guts to try it.

A repair rule with three failure modes Sauermann and Pham started the way their predecessors did, with a random ordering and a repair rule: whenever a stretch of consecutive numbers added up to zero, they swapped out its last entry.

Three kinds of bad event could break that rule. A zero-sum stretch might sit near the end, with nothing left to swap in. Several such stretches might cluster too tightly to fix one at a time. Or a repair might create a new zero-sum stretch further along. Using Fourier analysis to show that no single sum is especially likely to appear, the pair bounded the probability of each failure and showed the total chance of hitting one comes in under 100 percent.

The paper shows more than existence. A random ordering can be repaired to clear every bad event at least 90 percent of the time.

What is still open

All four papers assume p is enormous, on the order of 10 raised to the 100th power. Nobody has fixed an exact value. A proof for every p remains out of reach, which also means the result will not choreograph an actual juggling routine: the pattern would have to run far too long to perform.

Noga Alon of Princeton said the work fits a wider family of questions about designs and highly symmetric structures, and shows that some nice structures always exist even under tight constraints. Fan Chung of UC San Diego, Graham's wife, said posing a good problem is an art. Sauermann had met a related problem at the International Mathematical Olympiad as a schoolgirl and won a gold medal four times; Chung sat on the committee that wrote the questions and often drew on Graham's puzzles.

The pattern in all four proofs is the same bet: randomness is a way to find structure, not the opposite of it. That is a narrower claim than the one the labs make. Anthropic has said its own models lead 26 percent of its AI research, and OpenAI has been claiming routes into problems such as the Hodge Conjecture. This one was settled the older way, by four people and a chalkboard, and by a question a juggler left behind.