Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source: original paper, printed p. 543, Theorem 10.

Statement

No polygon with an odd number of positive-length edges, all the same length, has every vertex in Z2\mathbb Z^2. Self-intersections are allowed. In particular a finite square integer grid contains no congruent copy of an equilateral odd polygon, at any scale.

Full proof

Suppose a closed polygon has tt edges with common length d>0d>0. Write its integer edge vectors as (aj,bj)(a_j,b_j), so

aj2+bj2=d2,∑j=1taj=∑j=1tbj=0.a_j^2+b_j^2=d^2,\qquad \sum_{j=1}^t a_j=\sum_{j=1}^t b_j=0.

Let q≥0q\ge0 be the largest integer such that 2q2^q divides every coordinate of every edge vector. Such a largest qq exists because at least one coordinate is nonzero. After dividing all vectors by 2q2^q, at least one resulting vector has an odd coordinate, while all have the same squared length D=d2/4qD=d^2/4^q, an integer.

If one vector has exactly one odd coordinate, then D≡1(mod4)D\equiv1\pmod4. Every vector must then have exactly one odd coordinate, since squares modulo four are zero or one. Each sum aj/2q+bj/2qa_j/2^q+b_j/2^q is odd, so the total sum has parity tt. Its total is zero, hence tt is even.

Otherwise the vector with an odd coordinate has both coordinates odd, giving D≡2(mod4)D\equiv2\pmod4. Every vector then has both coordinates odd, so already ∑jaj/2q=0\sum_j a_j/2^q=0 forces tt even. Both cases contradict an odd number of edges.

The proof uses only the common edge length and the closed-walk sums; simplicity, convexity and a prescribed orientation are unnecessary.