贝伦德球壳 · 球面往外鼓:同一层壳上任取两点 a、b,中点 (a + b)/2 总陷进球里,所以壳上的格点排不出 a, (a + b)/2, b 这样的三项等差Behrend sphere · a sphere bulges outward: for any two points a, b of one shell the midpoint (a + b)/2 sinks inside the ball, so the shell's lattice points never line up as a, (a + b)/2, b
当前壳SHELL—三项等差3-TERM APS—
壳上格点 ‖x‖² = Ron the shell ‖x‖² = Ra、b 与弦a, b and the chord中点 mmidpoint m其余格点other points
哪一层壳最胖?Which shell is fattest?
从 a 走到 b:一路都在壳内From a to b: inside the whole way
理论接口 · 这个页面在演示什么Theory interface · what this page shows
立方体 {0, …, M − 1}³ 的每个格点都有一个“半径平方” R = x² + y² + z²,球心在立方体的一个角(原点)。同一个 R 的点在同一层球壳上。页面画出一层壳与壳上的格点;读者点两个壳上的点 a、b,就能看见它们的中点陷进球里。本页是数值演示,不是证明。
Every lattice point of the cube {0, …, M − 1}³ has a squared radius R = x² + y² + z², measured from a corner of the cube, the origin. Points with the same R lie on the same spherical shell. The page draws one shell and its lattice points; click two points a, b of the shell and you see their midpoint sink inside the ball. The page is a numerical illustration, not a proof.
A sphere bulges outward. If a and b lie on one shell and a ≠ b, then ‖(a + b)/2‖² = R − ‖a − b‖²/4 < R: the midpoint is strictly inside the ball and cannot be on the shell. A 3-term progression a, c, b needs c = (a + b)/2, so a shell holds none (ARGUED; the page checks every shell for M = 3 … 6 one by one, NUM). And since ‖a − b‖² = 2R − 2a·b is even, a midpoint sinks by at least 1/2.
Pigeonholes. For M = 5 the 125 points lie on only 32 non-empty shells, so some shell holds at least ⌈125/32⌉ = 4 of them; the fattest, R = 17, holds 9. In higher dimensions the fattest shell still holds a not-too-small share of all the points: that is where Behrend's construction (1946) starts. He writes that shell as integers that never carry and obtains a subset of {1, …, N} with no 3-term progression and at least N·exp(−C√log N) elements (ARGUED: literature result, not formalized in this book).
How it enters the proof. As Chapter 6 tells it, the levels of the Iwasawa tower must avoid "arithmetic resonance", or cancellation errors pile up along one orbit. The formalization works with a more general kind of set, one with no small linear relation (the definition BehrendFree in Formalism/BehrendSets/Statement.lean: when α + β + γ = 0 and the coefficients are at most K, αx₁ + βx₂ + γx₃ = 0 has only trivial solutions; a 3-term progression is the case α = γ = 1, β = −2). From that part Chapter 6 cites select and trivStab (LEAN ✓), and the chain ends in everyCharacter (LEAN ✓). The page does not show that chain; it shows only the classical shell trick in a small cube.
Models and simplifications. The size of the cube, the colours of the walls and the glass and the camera angles are choices of the picture (MODEL); lattice points, squared radii, counts and midpoints are exact integer computations.