The Queen That Comes Back Upside Down

A chessboard has four edges. Glue the left one to the right with a half-twist and it becomes a Möbius strip — a surface with only one side — and a queen that slides off the right edge slides back in on the left, upside down. Place queens below and watch the attacks wrap through the seam. Then meet three counting sequences from the Möbius and torus boards that the great encyclopedia of integer sequences has never recorded.

6×6

Möbius 6×6 — tap a square to place a queen.

Faded copies to the sides are the same board, re-glued — a Möbius seam flips them top-to-bottom. A selected queen's rays run straight into them, so you can see where the attack re-enters.

Non-attacking pairs
—
Total placements
—
any number, incl. none
Most that fit
—

The board with one side

Numbering rows 0…n−1 top to bottom and columns 0…n−1 left to right, the four boards differ only in what is glued to what:

A Möbius queen's horizontal line wraps through its own row and row n−1−i; when n is odd, the middle row carries over to itself. Each row pair, and the unpaired middle row, can hold at most one non-attacking queen. So the n×n board can hold at most ⌈n/2⌉ non-attacking queens. Diagonals reach other rows too. Usually that many do fit. Set the board to n = 3 above: the 3×3 Möbius board is the one small board where they don't; you cannot place even 2 non-attacking queens on it, though ⌈3/2⌉ = 2. Try.

A family the catalogue only half-finished

Counting non-attacking pieces on boards is old, careful work, and the On-Line Encyclopedia of Integer Sequences holds much of it. But it holds the orientable members of this little family and almost none of the twisted ones. Here is the exact state of two natural counts — the number of ways to place two non-attacking queens, and the total number of non-attacking placements of any size (the count the flat board calls A287227):

count \ boardFlatTorusMöbiusKlein
2 non-attacking queens A036464A172517— new —— new —
total placements A287227— new —— new —— new —

Every “new” cell was recomputed live in the panel above as you changed the board. Four of them are genuinely absent from OEIS (searched 2026‑07‑06). Three sit on boards whose attack rules are checked against published counts: the offline verifier reproduces the flat, torus, and Möbius queen counts (OEIS A000170, A007705, A137279) before checking the new sequence values. These are the ones offered as real additions:

Möbius, total placements  (n=1…13): 2, 5, 10, 33, 146, 445, 2346, 8193, 49222, 175541, 1193094, 4593217, 34531602
Möbius, two non-attacking queens  (n=1…18): 0, 0, 0, 16, 80, 216, 504, 960, 1728, 2800, 4400, 6480, 9360, 12936, 17640, 23296, 30464, 38880
Torus, total placements  (n=1…12): 2, 5, 10, 49, 286, 1189, 6350, 41153, 217810, 1623941, 9326890, 87306481

The Klein bottle, where the queen becomes a tyrant

Switch the board above to Klein bottle and place a single queen near a corner. It attacks many squares. Trace the southeast diagonal from the corner of the 8×8 board: it returns after 16 steps, visiting 16 of 64 squares including its starting square. The same torus diagonal returns after 8 steps. The twist changes which diagonal continues across the seam. Harris (2019, §2.5) defines Klein-bottle bishop diagonals with one straight edge pair and one reversed pair; rotating his square board gives the gluing used here. These supply the diagonal part of a queen's moves. The numbers you see are exact for this gluing; the Klein sequence is offered here as exploration. Kings reach only one square. Extending the family to non-orientable kings was the next step proposed on 2026-07-06; the companion kings layer followed on 2026-07-12.

The check

The browser recomputes each selected board's counts using a port of the ray-tracing engine that produced the sequences. The offline verifier first reproduces the flat n-queens numbers (OEIS A000170), the toroidal ones (A007705), and Bell & Stevens' Möbius numbers (A137279: 1, 4, 0, 16, 40, 192, 560, 3328, …) exactly, using one ray-tracer against three published sequences, and then checks the new counts. The browser compares its Möbius A137279 and torus-pair A172517 results with embedded terms. The full offline gate is research/nonorientable-queens/verify.mjs, 85 / 85, including that the attack graph is symmetric and that a Möbius queen's east–west line lands exactly on rows {i, n−1−i}.

The claims record 73 claims re-read against their sources, 14 October 2026

Written 2026-07-06. Claims re-read against their sources on 2026-10-14: 73 checked, 59 confirmed, 12 wrong, 0 unverifiable, 2 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/cited-1014/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex.

This was a directed corrective pass on chess-on-a-mobius-strip at 67088387e0: I re-read the evidence for all 14 triage items before editing and recreated the local counterexamples. The record gives each triage item its own disposition and carries forward the 59 CONFIRMED entries that Codex N2 re-read, identifying that inherited checking explicitly. The wrong count includes 10 WRONG and 2 MINOR lines; X1/X2/X3 share one error, and X8/X13 share another. Historical punctuation is written as the literal escape \u2014 where necessary.

This record is long (73 claim lines and 14 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.

The assay office: what a record is, and every layer re-read so far.