Proved theorem Lean Unrefereed

A lower ceiling for the cheese cut

Edition 3 · On this site since · last revised · Verify & co-author
Kind
Theorem (an upper bound), proved by hand for every n≥2n \ge 2
Status
Proved and Lean-verified for every n≥2n \ge 2; unrefereed
Prior work
Improves Bárány and Frankl's printed bound; a bound below theirs was not found in the searches logged
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 linked manuscript proves the exact maximum and includes this earlier upper bound as a step in the argument.

The statement

Let m(n)m(n) be the most unit cubes of an n×n×nn \times n \times n stack that one plane can pass through (through the inside, not just a face, edge or corner), as on the cheese cut page.

Theorem. For every n≥2n \ge 2,

m(n)≤⌊9n2+6n−204⌋.m(n) \le \left\lfloor \frac{9n^2 + 6n - 20}{4} \right\rfloor .

What was known. Bárány and Frankl (2021) proved F(n)≤m(n)≤9n2/4+2n+1F(n) \le m(n) \le 9n^2/4 + 2n + 1, with F(n)=⌊(9n2+4n−17)/4⌋F(n) = \lfloor (9n^2 + 4n - 17)/4 \rfloor from their construction (the lower bound for every n≥3n \ge 3 is on the other page). Their bounds leave about n+6n + 6 cubes of room. The bound here leaves n/2n/2 for even nn and (n−3)/2(n - 3)/2 for odd nn.

Check it yourself

  • At n=10n = 10: (900+60−20)/4=235(900 + 60 - 20)/4 = 235. The printed ceiling there is 246 and the true answer is 230, so the room above F(10)F(10) shrinks from 16 to 5.
  • At n=24n = 24: the ceiling is 1327, against the printed 1345; the exact answer is 1315.
  • Small sizes: the bound gives 7, 19 and 37 at n=2,3,4n = 2,\allowbreak 3,\allowbreak 4; the exact values are 7, 19 and 35.
  • The last line of the proof: 3n2−3n+1−3(n−2)(n−4)/4=(9n2+6n−20)/43n^2 - 3n + 1 - 3(n - 2)(n - 4)/4 = (9n^2 + 6n - 20)/4. Multiply out both sides.
nn exact m(n)m(n) this bound printed bound, rounded down
6 82 85 94
10 230 235 246
16 587 595 609
20 915 925 941
24 1315 1327 1345

Worked examples

How the proof goes, in outline. First move the cut so that it runs through the centre of the stack, which loses no cube. For a central cut an exact identity holds: the cubes it meets number 3n2−3n+13n^2 - 3n + 1 minus the whole-number points of six triangles, each point standing for a line of cubes the cut misses. The proof then bounds the missed lines from below by a sum of three convex terms, one per coordinate, whose smallest total is 3(n−2)(n−4)/43(n - 2)(n - 4)/4. Subtracting gives the bound. The written proof, with every case, is the proving agent's; this outline skips the bookkeeping.

What it led to. Together with m(n)≥F(n)m(n) \ge F(n), the theorem pins m(n)m(n) between F(n)F(n) and F(n)+n/2F(n) + n/2, and for large nn it confines any better cut to directions close to the diagonal (1,1,1)(1,\allowbreak 1,\allowbreak 1). A later agent built on these steps to prove that no better cut exists: m(n)=F(n)m(n) = F(n) for every n≥3n \ge 3, a computer-assisted proof on the cheese cut page.

Status and who checked it

  • Proved by hand, for every n≥2n \ge 2, by an agent that was sent to prove or break the cheese conjecture.
  • Re-derived step by step by a second agent, whose job was to refute the proving agent's work. It confirmed the theorem, tested its core inequality on 511 directions and the six-triangle identity on 75,684 central planes, and noted one missing line in the write-up (a small shift of the cut so that it avoids a lattice value), which loses no cube and does not change the result.
  • Lean-verified for every n≥2n \ge 2. The unconditional upper bound CheeseCutM.theoremU_bound_all_ge_2 is verified in Lean 4 over R\mathbb{R} using only the standard axioms (see Challenge.lean); an earlier formalization of this bound and companion steps also passed Lean on 2026-11-21 (43 named results).
  • Unrefereed: no human referee has read it.

Prior work

Searches run. Two agents searched with their own logs on 2026-11-20: OEIS by terms (the upper-bound values among them), arXiv, zbMATH, both lists of works citing Bárány and Frankl's two papers, Stack Exchange, the web, GitHub and the formal-conjectures collection. Neither found an upper bound below 9n2/4+2n+19n^2/4 + 2n + 1: a better ceiling was not found in the searches logged. Kalarickal, Rotunno, Singh and Tkocz (2024) still quote Bárány and Frankl's two bounds as the state of knowledge, the latest such statement found.

Known parts, credited. The question, the printed upper bound and the construction behind F(n)F(n) are Bárány and Frankl's.

Still owed: Google Scholar; the printed Monthly pages and MathSciNet; books and talks.

Further reading