Proved theorem Unrefereed

Flammenkamp's twelve losing lengths

No dated manuscript catalogued; standalone verification archive linked above. On this site since · last revised · Verify & co-author
Kind
Theorem, computer-assisted (an exact computation treating k as a symbol)
Status
Proved, computer-assisted; five programs by five agents agree; no proof by hand; unrefereed
Prior work
Not found in the searches logged; conjectured on Flammenkamp's octal-games page and checked there for 3 < k < 32
On this page
  1. The statement
  2. Check it yourself
  3. Worked examples
  4. Status and who checked it
  5. Prior work
  6. Further reading

The statement

A row of nn pins stands in a line. A move takes away exactly kk pins that stand next to each other, leaving at most two shorter rows; the players alternate, and the player who cannot move loses. In octal notation this is the game .0k−17.0^{k-1}7. A row length n≥kn \ge k is a losing length when the player to move loses; rows of 00 to k−1k - 1 pins lose at once.

Achim Flammenkamp's web page "Sprague-Grundy Values of Octal Games" (printed between 2026-08-09 and 2026-11-17; his spelling kept):

We conjecture: For each k > 3 and for all i with k < i ≤ k+12 let j_i be the i-th lose position of the game, which always exists. Then there are 12 positive, stricly increasing integers α_{i-k} and 12 positive, increasing integers β_{i-k} such that holds j_i = α_{i-k} k - β_{i-k}. We have (α_h) = 3,5,9,13,17,21,29,33,41,49,65,69 and (β_h) = 1,1,3,5,5,7,9,11,11,13,19,19. We have confirmed this conjecture numerically for all 3 < k < 32.

Theorem. For every k≥4k \ge 4, the first twelve losing lengths n≥kn \ge k of the game .0k−17.0^{k-1}7 are

3k−1, 5k−1, 9k−3, 13k−5, 17k−5, 21k−7,3k-1,\ 5k-1,\ 9k-3,\ 13k-5,\ 17k-5,\ 21k-7, 29k−9, 33k−11, 41k−11, 49k−13, 65k−19, 69k−19.29k-9,\ 33k-11,\ 41k-11,\ 49k-13,\ 65k-19,\ 69k-19.

It fails for k=2k = 2 (Dawson's Kayles, losing lengths 5, 9, 15, …) and k=3k = 3 (8, 14, 24, …).

Check it yourself

  • 3k−13k - 1 by hand. Whatever kk pins the first player takes from a row of 3k−13k - 1, the two rows left add up to 2k−12k - 1: one has kk to 2k−12k - 1 pins and the other fewer than kk. The second player takes kk pins from the long one, leaving no row of kk or more. So 3k−13k - 1 is a losing length for every k≥2k \ge 2.
  • The values. Each row length gets a value: the least whole number that is not the xor of the values of the two rows left by some move. Gk(n)=0G_k(n) = 0 for n<kn < k and Gk(n)=mex{Gk(a)⊕Gk(n−k−a):0≤a≤n−k}G_k(n) = \mathrm{mex}\{G_k(a) \oplus G_k(n-k-a) : 0 \le a \le n-k\}. A losing length is an n≥kn \ge k with Gk(n)=0G_k(n) = 0. For k=4k = 4 the values for n=0n = 0 to 20 are 0 0 0 0 1 1 1 1 2 2 2 0 3 3 3 1 1 1 1 0 4.
  • k=4k = 4: the losing lengths begin 11, 19, 33, 47, 63, 77, 107, 121, 153, 183, 241, 257, and the next is 259. k=5k = 5: 14, 24, 42, 60, 80, 98, 136, 154, 194, 232, 306, 326.

Worked examples

Figure 1. (a) A row of 11=3k−111 = 3k - 1 pins with k=4k = 4: each of the first player's four moves (the other four are mirror images) and the second player's reply. (b) The value of every row length nn from 0 to 50 for k=4k = 4; the dark cells, value 0, are the losing lengths.Full size
Figure 2. (a) For k=4k = 4 to 12, the first twelve losing lengths, each in the column of the form it equals; the last column is the next one, where one exists up to 20,000. (b) Every losing length below 72k72k for k=4k = 4 to 40, drawn at x=n/kx = n/k: each form αk−β\alpha k - \beta closes in on the line x=αx = \alpha as kk grows.Full size

How a computer proves it for every kk. The values come in runs of equal value whose starts sit at places ak+bak + b with aa and bb the same for every large kk; for k=28k = 28 to 60 every game has exactly 175 runs below 70k70k. The proof runs the value rule with kk left as a symbol, comparing forms ak+bak + b as they compare for large kk, and records the least kk from which each comparison is true. For n<70kn < 70k that threshold is 27: above it the symbolic runs are the true values, and their only runs of value 0 are the trivial one and the twelve. The games with k=4k = 4 to 27 are computed one by one.

Status and who checked it

  • Proved, computer-assisted. The symbolic computation with its recorded threshold, plus direct checks of every kk at or below it. Five all-kk programs by five agents, none reading another's code, list the same twelve forms; three of them agree interval by interval. A reviewer agent confirmed the theorem and proved a short lemma that the threshold is never more than half the range.
  • Checked one game at a time for every kk from 4 to 120.
  • No proof by hand, and no human referee has read it.
  • Further, reviewed separately: the 13th losing length is 8255k−24418255k - 2441 for every k≥265k \ge 265, and the 14th is 11247k−347511247k - 3475 for every k≥433k \ge 433. An accompanying claim, that no losing length lies between 69k−1969k - 19 and k2k^2, holds for every k≤8254k \le 8254 and is false from k=8255k = 8255; it is not claimed here.

Prior work

Searches run. Five literature logs by falsifier, reviewer and follow-up agents pass the literature check (56, 33, 20, 44 and 55 keyed searches; 13, 8, 9, 14 and 13 candidates read), with OEIS searches by computed losing lengths and hand searches of Zenodo, DataCite, HAL, OSF and GitHub. Flammenkamp's page has the law as a conjecture, with no proof; Guy and Nowakowski's unsolved-problem lists (2015, 2025) have nothing on this family. A keyed CORE run on 2026-11-20 (6 requests, 31 records) found no record proving the twelve lengths. The proof was not found in the searches logged. On 2026-11-22, a Google Scholar sweep (5 queries) found no prior work stating this claim.

Nearest relatives. Brown et al., "Nimber Sequences of Node-Kayles Games" (2020), meet this game as Node-Kayles on a power of a path and turn to a variation; OEIS A071426, A071441 and A395248 list the values of single members; Grossman and Nowakowski's ruler regularity and the partition-games papers prove laws uniform in a parameter for other families.

Still owed: Winning Ways chapter 4 and Guy and Smith 1956 in print; Siegel's book, Flammenkamp's data files and the CGSuite database. Flammenkamp himself was not asked and may already hold a proof.

Further reading