VibeMathedMath problems solved with AI

Boshernitzan's Conjecture 2.2(e) for rational right and isosceles triangles

Let Δ\Delta be a rational triangle with angles aπ/da\pi/d, bπ/db\pi/d, cπ/dc\pi/d in lowest terms, placed with one side horizontal or vertical. Put d′=lcm(2,d)d'=\mathrm{lcm}(2,d) and S1(n)={z:zn=1}S^1(n)=\{z : z^n=1\}, and let P0P_0 be the set of directions in which every billiard orbit that avoids the vertices is periodic (equivalently, the unfolded translation surface decomposes completely into cylinders). Boshernitzan proved that the rational exceptional directions of Δ\Delta lie in S1(2d′)S^1(2d') and stated Conjecture 2.2 in five parts. Part (e): if Δ\Delta is isosceles or a right triangle, then S1(2d′)⊆P0S^1(2d')\subseteq P_0. Computer searches of Delecroix and Rüth found counterexamples, among them the right triangle with angles (4/15,7/30,1/2)π(4/15, 7/30, 1/2)\pi and the isosceles triangle (7,7,16)π/30(7,7,16)\pi/30, so (e) is false as stated. For which rational right and isosceles triangles does it hold?

Result
Proved(see note)
Status
Partial result
AI contribution
AI co-developed
Method
Argument
Field
Dynamical systems, polygonal billiards
Posed by
M. D. Boshernitzan, Billiards and rational periodic directions in polygons, Amer. Math. Monthly 99 (1992), 522-529, Conjecture 2.2(e)
Year posed
1992
Years open
34y
Solved
2026-11-10
Model
Claude Opus
Vendor
Anthropic
Collaborators
Joshua Boorstein
Verification
Unreviewed
Publication
Preprint
Significance
12 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Corollary 8.1: the right triangle with angles (P,Q−P,Q)π/2Q(P, Q-P, Q)\pi/2Q, gcd⁡(P,Q)=1\gcd(P,Q)=1, min⁡(P,Q−P)≥2\min(P,Q-P)\ge 2, satisfies Conjecture 2.2(e) (both perpendicular direction classes completely periodic) when (1) QQ has at most one odd prime factor (computer-assisted in part); (2) P≢±2(modp)P\not\equiv \pm 2 \pmod p for every odd prime p∣Qp\mid Q, a set of density about 0.62; (3), (4) two coding-theoretic families at composite QQ; or (5) one of PP, Q−PQ-P is 4 and Q≡3(mod4)Q\equiv 3 \pmod 4 (interval arithmetic). At Q≤2000Q\le 2000 this covers 383,692 of 606,295 such triangles. Corollary 8.6 carries it to the isosceles triangles whose right halves are covered. The method cannot reach triangles where neither P+2P+2 nor P−2P-2 is a unit mod QQ, where the known counterexample lies (Conjecture 6.5 covers the rest). Theorem 6 (computer-assisted): on three explicit intervals near 15/92, 21/242 and 23/270, every triangle (a,2a,d−3a)π/d(a,2a,d-3a)\pi/d with dd even violates part (a), the first infinite family of counterexamples to (a).

What the AI did

The paper's "Use of AI" section: "This paper - its text, its proofs and the code behind Appendices A and B, Proposition 6.4 and Theorem 6 - was drafted by an AI system (Claude, Anthropic) working under the author's direction, and checked by independent AI review passes and by the computations described in Appendices A and B, 6.3 and 9.2. The author has read parts of it in detail and takes responsibility for its content." Hexagon lists the tools as Anthropic Claude Opus, Python, SymPy and mpmath, and labels the work "Primarily AI-generated text; Human understanding: some parts".

Verification

Read by this site on 15 November 2026. No independent mathematician has checked this yet. Read: the Hexagon abstract page with its provenance statement, the paper's introduction, Corollary 8.1, Theorem 6 and its Use of AI section. Conjecture 2.2 was compared with Boshernitzan's wording as quoted in the sage-flatsurf documentation (the 1992 Monthly article itself was not opened). Independent computation written here: among right triangles with Q≤2000Q\le 2000 and min⁡(P,Q−P)≥2\min(P,Q-P)\ge 2 there are 606,295 (matches the paper); condition (2) holds on 62.0% of them (paper: about 0.62); parts (1) and (2) alone cover 383,123, consistent with the paper's 383,692 once parts (3) to (5) are added; the least denominators in Theorem 6's three intervals are 92, 242 and 223, and φ(d)≤20\varphi(d)\le 20 forces d≤66d\le 66, as the deposited manifest states. The certificates themselves (github.com/josh-boorstein/billiards v1.0, Zenodo 10.5281/zenodo.24037744) were not run here, and the proofs were not checked.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion