AN ENDLESS LIST WRITTEN WITH SIX NUMBERS THAT NEVER BALANCES
Write a sequence of whole numbers, one after another. Now look for two neighbouring blocks with the same length and the same sum. In 1 4 2 3, the blocks 1 4 and 2 3 both have two numbers and both add up to 5: mathematicians call such a pair an additive square.
Can an infinite sequence, written with a finite set of numbers, avoid every additive square, forever? The question has been asked in different forms since 1972 — by Justin, then Brown and Freedman in 1987, Pirillo and Varricchio in 1994, Halbeisen and Hungerbühler in 2000. Brown and Freedman even conjectured the opposite: that every sequence of positive whole numbers that stays below some bound must contain one. The new note calls it “a problem that has been open for more than 50 years”.
Six numbers that never balance
Jeffrey Shallit, of the School of Computer Science at the University of Waterloo in Canada, now writes down an explicit infinite sequence that uses exactly six numbers and contains no additive square.
The six numbers — “letters”, in the paper’s words — are anything but small. With D = 2⁵⁸, L = 2 · 85¹¹, H = 2D + 1 and U = L³ + 1, they are:
1, U + 1, U + 2, U + L + 1, U + L² + 1 and (H + 2)U + 1.
Only the first one is small. The paper gives a rule for the number at any position, with checks: the number 1 first appears at position 1,152,921,504,466,744,130, counting from zero.
How the proof works
The construction has two steps.
- An auxiliary sequence. Start from one number and replace each number, again and again, by a block of 2⁶⁰ numbers. The blocks are designed so that if this sequence contained an additive square, it would also contain a square 2⁶⁰ times shorter. Pick the shortest square there is: an even shorter one would have to exist. That contradiction shows there is no square at all.
- A six-number code. The auxiliary sequence has no additive square, but it uses up to 423 different numbers. It is then encoded in the six numbers, using a four-symbol pattern found by Peltomäki and Whiteland, itself built on an 85-symbol substitution due to Veikko Keränen. The huge gaps between the six numbers make every sum readable, like digits in a very large base: any square in the final sequence would betray a square in the auxiliary one.
Four, five or six?
Six is not known to be the minimum. Another kind of repetition, an abelian square, is a block followed by a shuffled copy of itself, like 1 2 3 3 1 2. Its two halves contain the same numbers, so they always have the same sum: every abelian square is also an additive square. With only three different numbers, abelian squares cannot be avoided — so at least four numbers are needed. The smallest number of different numbers that can work is therefore 4, 5 or 6. “It remains to determine which one is correct,” the paper ends.
Written by a machine, signed by Jeffrey Shallit
The note closes with a “Statement of AI Usage”: “This paper was almost entirely written by GPT 6.2 Astra.” Shallit added introductory material, references and discussion, and writes: “I take full responsibility for the correctness.” A computer verification of the construction, written in the Lean language and rerun with a single command, is posted online — “also constructed by GPT 6.2 Astra.” The note is eight pages long and has not yet been peer reviewed.
