VibeMathedMath problems solved with AI

Tokarsky's question on the fewest sides of an unilluminable polygonal room, for lattice rooms

A polygonal room has mirrored walls. A light ray reflects with equal angles and stops when it hits a corner. Two points PP and QQ of the room are mutually invisible if no light ray from PP reaches QQ; a room containing such a pair is unilluminable. Tokarsky built the first polygonal example in 1995, with 26 sides, answering the illumination question published by Klee in 1969, and asked how few sides a polygon with two mutually invisible interior points can have. Rooms with 24 sides were found in 1996-97 and one of them was conjectured to be the smallest. All known examples are lattice rooms: unions of tiles of a reflection tiling in which every tile vertex outside a distinguished sublattice is a corner. This entry records the question inside that class: what is the least number of sides of a lattice room with one or with two invisible lattice points, and which rooms attain it?

Result
Proved(see note)
Status
Partial result
AI contribution
AI co-developed
Method
Computation
Field
Discrete geometry; polygonal billiards and illumination
Posed by
George W. Tokarsky, Polygonal rooms not illuminable from every point, Amer. Math. Monthly 102 (1995), Example 3
Year posed
1995
Years open
31y
Solved
2026-11-09
Model
GPT-6; Claude Opus 5.5
Vendor
OpenAI; Anthropic
Collaborators
J. C. F. de Winter, George W. Tokarsky
Verification
Unreviewed
Publication
Preprint
Significance
8 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Within lattice rooms only (arXiv:2611.13226, Table 1). A hole-free lattice room on the tiling by isosceles right triangles with one invisible lattice point (source) has at least 12 sides, attained only by the pinwheel; on the square tiling the least is 16. With two sources, hence two mutually invisible points, the least is 24 on the triangle tiling (exactly 12 rooms, plus 89 with 25 sides) and 34 on the square tiling (14 rooms). One post, one partition or one pinched hole lowers the two-source minimum to 18, 20 and 22; two and three posts give 17 and 16; a free hole gives 26. Three sources need at least 36 sides. A second class on the 30-60-90 tiling has minimum 27 (15 rooms). The classifications rest on exhaustive computer searches made finite by bounds proved in the paper. Still open, and stated as such by the authors: whether a simple polygon with fewer than 24 sides can have two mutually invisible interior points.

What the AI did

From the paper's acknowledgements, which is its only AI statement: "The proofs were generated and the draft of this paper was written with GPT-6 (OpenAI), and the draft was checked and rewritten with Claude Opus 5.5 (Anthropic). This work was done under the guidance of the authors from March 2026 onwards, and earlier versions of these models were used as well." The programs behind the exhaustive searches ship with the paper as ancillary files.

Verification

Read by this site on 19 November 2026 in the TeX source of arXiv:2611.13226v1: the introduction and Table 1, the blocking lemma and its proof, Theorems 3.1 and 5.1 with the start of their proofs, the section on computations, the closing questions and the AI statement. The blocking lemma is a short complete argument and reads correctly. A small computation written here confirmed that the pinwheel as printed has 12 genuine corners, area 6, a single interior lattice point, and every other lattice point of the room at a corner. The authors' ancillary programs, whose saved log reports 19 of 19 checks passed, were not re-run here, and the reductions to finite searches were not refereed. Tokarsky's 1995 paper was not opened; the question is taken from the citation in this paper, of which he is a coauthor. No formal proof.

Source

Submitted by MellowGander562 on

Changelog2 changes

Discussion