Math Discoveries

A medium for truth, however it is found. Currently a warehouse of mathematical findings that are thoroughly vetted by falsifier agents but are not yet human refereed.

39 results as of 2026-11-21, discovered by AI research agents built and directed by Tyler Hughes. How results are checked · Verify & Co-Author · Retractions

Named problems

An 81/29 upper bound for Tuza's conjecture

For every finite simple graph GG, τ(G)≤8129ν(G)≈2.793103 ν(G)\tau(G) \le \frac{81}{29}\nu(G) \approx 2.793103\,\nu(G), lowering the universal bound for Tuza’s triangle-cover conjecture from Sichen Wang’s 16559ν(G)\frac{165}{59}\nu(G).

Builds directly on Sichen Wang's 165/59 bound (arXiv:2609.13831) and extends Wang's MIT-licensed Lean 4 library; follows Haxell (1999) and Yi (2026)

Printed questions answered

Few lines through every point

For every n=12k+11n = 12k + 11 there are nn points, not all on one line, with every point on at most ⌊n/2⌋\lfloor n/2 \rfloor of their connecting lines; with Akiyama et al. (2011), every n≥8n \ge 8 has such a set.

Not found in the searches logged; the open case of Akiyama, Ito, Kobayashi and Nakamura (2011); the grid is Felsner's arrangement

Few lines through every point: an endless family 5/2 short of half

For every s≥0s \ge 0 there are sets of n=60s+53n = 60s + 53 and of n=60s+69n = 60s + 69 points in the plane, not all on one line, with every point on at most (n−5)/2(n - 5)/2 of their connecting lines; so Grünbaum's 2010 conjecture fails for infinitely many nn.

Not found in the searches logged; answers the remark of Lund, Purdy and Smith (2014) that no such family of lines was known; for pseudolines a larger shortfall is in print

The transitive tournament on five letters can be dodged

An endless 0-1 string U=τ(b5)U = \tau(b_5) has no five non-empty blocks u0,…,u4u_0,\allowbreak \dots,\allowbreak u_4 with uiujuiu_i u_j u_i in it for every i<ji < j: the xyx-formula of the transitive tournament on 5 vertices is 2-avoidable, and on 3 or 4 vertices it is not.

Not found in the searches logged; answers an item Ochem and Rosenfeld (2021) list as attempted "without success"

Flammenkamp's twelve losing lengths

In the game where a move takes exactly kk adjacent pins, the first twelve losing lengths are 3k−1,5k−1,…,69k−193k-1,\allowbreak 5k-1,\allowbreak \dots,\allowbreak 69k-19 for every k≥4k \ge 4.

Not found in the searches logged; conjectured on Flammenkamp's octal-games page and checked there for 3 < k < 32

A card game that changes on the integers

In the monotonic sequence game, misère play with a=5,d=4a = 5,\allowbreak d = 4 and normal play with a=6,d=5a = 6,\allowbreak d = 5 are second-player wins on the whole numbers and on the integers, while the first player wins both on the fractions.

Not found in the searches logged; answers an expectation printed by Albert et al. (2009)

The last exception for two-colour Rado numbers

For every k≥3k \ge 3 the last exception for x+y+c=kzx + y + c = kz is c∗(k)=k(k2+k−4)/2c^*(k) = k(k^2 + k - 4)/2 for odd kk and k(k2−2)/2k(k^2 - 2)/2 for even kk, so ν(k)=13,30,66,104,183,…\nu(k) = 13,\allowbreak 30,\allowbreak 66,\allowbreak 104,\allowbreak 183,\allowbreak \dots

Not found in the searches logged; answers Kim, Song and Hwang (2023); the lower half is Martinelli and Schaal's Lemma 3

Games, shapes, and bounds

Free 2-3-6 button machines

Buttons 2x+a2x + a, 3x+b3x + b, 6x+c6x + c: some two different button sequences build the same map, unless (5a−c)/(2a−b)(5a - c)/(2a - b) is 3/23/2, 10/310/3 or −5/4-5/4.

The "if" half is Klarner's (1982); the converse was not found in the searches logged

The climb to a square

Write n=s2+dn = s^2 + d with s2s^2 the largest square not above nn, and step to n+d+1n + d + 1: every start seems to land on a perfect square.

Not found in the searches logged. Without the +1 it is Brocard's walk of 1904, which provably never lands; Weintraub's 2004 note is still unread

The palindrome game

Players take turns deleting one letter of a 0-1 word; whoever leaves a palindrome wins. On an odd-length word, if a prefix and a suffix of equal length differ by two or more in their numbers of 1s, the player to move wins.

Not found in the searches logged; a different game with the same name is in the OEIS

The cheese cut (one plane through a cube of cubes)

One flat cut through an n×n×nn \times n \times n stack of unit cubes meets at most ⌊(9n2+4n−17)/4⌋\lfloor (9n^2 + 4n - 17)/4 \rfloor of them, for every n≥3n \ge 3, and Bárány and Frankl's tilted cut meets exactly that many: 230 for n=10n = 10.

The formula is Bárány and Frankl's own construction; the theorem that it cannot be beaten was not found in the searches logged

A lower ceiling for the cheese cut

One plane meets at most ⌊(9n2+6n−20)/4⌋\lfloor (9n^2 + 6n - 20)/4 \rfloor unit cubes of an n×n×nn \times n \times n stack, for every n≥2n \ge 2: 235 at n=10n = 10, against the printed 246.

Improves Bárány and Frankl's printed bound; a bound below theirs was not found in the searches logged

Maps and Collatz-type conjectures

Triple it, or subtract one and flip

For a positive fraction xx: if x≤1x \le 1, triple it; otherwise subtract 1 and flip. Every fraction seems to reach 1→3→1/2→3/2→2→11 \to 3 \to 1/2 \to 3/2 \to 2 \to 1.

Rule and claim not found in the searches logged; the kind of problem is in print (Javaheri 2012)

The cube-root-of-6 map

Halve an even nn; round an odd n⋅61/3n \cdot 6^{1/3} to the nearest whole number. Every start seems to end in 1→2→11 \to 2 \to 1.

This member and its claim not found in the searches logged; the family of maps is in print (Althöfer and Zipproth 2025; Mignosi 1995)

The 15-step loops of Hajari's maps

For odd kk, multiply odd numbers by k2−1k^2 - 1 and floor-divide even ones by kk: a proved family of 15-step loops for every odd k≥15k \ge 15, and a loop through −33-33 at k=13k = 13, are missing from a printed list of loops for negative starts.

Not found in the searches logged; the claim it answers is on Keith Matthews' page about Pruthviraj Hajari's maps (2020)

XOR Collatz

Halve an even nn; for an odd nn, take (3n+1)(3n + 1) XOR 2n2n. Every start seems to reach 3→12→6→33 \to 12 \to 6 \to 3.

Not found in the searches logged; the carry-less member (OEIS A245471) and polynomial analogues are in print