MathematicsPreprintTheory3 min read

A 1969 CONJECTURE ABOUT SUMS THAT CANCEL OUT TURNS OUT FALSE

Take a clock with six hours, and a list of moves: “turn the hand forward 1 hour”, “forward 4 hours”, and so on. With five moves of one hour each, no group of them ever brings the hand back exactly to the start. With six moves, whatever they are, some group always will. Mathematicians write this clock arithmetic as C₆, and the number six here is the clock’s Davenport constant.

Now put r such clocks side by side, and let each move turn every hand by its own amount. The Davenport constant of this set of clocks, written D(C₆^r), is the smallest number of moves that guarantees a group whose total brings every hand back to the start at once — a zero-sum subsequence, in the paper’s words. The constant is tied to problems in graph theory, finite geometry, combinatorics and factorization theory.

A formula from 1969

There is an easy lower bound. On each clock, take five one-hour moves that touch only that clock: with r clocks, that makes 5r moves among which no group cancels out. So at least 5r + 1 moves are needed. In general, for clocks with n hours, at least r(n − 1) + 1.

In 1969, van Emde Boas and Kruyswijk conjectured that this easy bound is always the exact answer:

D(C_n^r) = r(n − 1) + 1.

It was proved in the 1960s for some cases, such as clocks whose number of hours is a power of a prime, or at most two clocks. Beyond that, the exact value stayed out of reach: as late as 2018, even the case of four six-hour clocks was unknown. Still, the conjecture became “a prominent longstanding problem” of the field, and it was expected to hold.

116 moves that never cancel

Weidong Gao (Tianjin University), Alfred Geroldinger (University of Graz), Guoqing Wang (Tiangong University) and Xiangneng Zeng (Sun Yat-Sen University) now show it fails.

The team assembles the moves like building blocks:

  • a list of 24 moves on five six-hour clocks, built around a special position h, where hand 3 points to 2 and hand 5 to 4: three times h brings every hand back to 0. Whenever a group of these moves lands on 0, h or 2h, it lands on h;
  • two such lists that share the three positions 0, h and 2h: a group drawn from one list or both adds up there to h or 2h, never 0. That gives 48 moves with no group that cancels out;
  • a separate list of 20 moves in which no group cancels out either.

Placed side by side, these give 2 × 48 + 20 = 116 moves on 23 six-hour clocks, and no group of them brings every hand back to the start. The conjecture said 23 × 5 = 115 was the longest such list possible. So for 23 clocks, the Davenport constant is at least 117, not 116.

Copies of this block can be stacked: the general result is that D(C₆^r) ≥ 5r + ⌊r/23⌋ + 1, where ⌊r/23⌋ counts how many whole blocks of 23 fit into r. The conjecture is therefore false for six-hour clocks as soon as r reaches 23, and for every larger r. The authors note that the same theorem also disproves a conjectured result in a related problem.

What remains open

The proof does not give the true value of D(C₆^r), only a bound that the conjecture forbids. It says nothing new for fewer than 23 clocks, where the new bound equals the old formula, nor for clocks with a number of hours other than six. The note runs to five pages, and peer review is still to come.