GPT 6 Astra Ultra attempt on Math-5696

FORMAL RESTATEMENT

Erdős #502; a reconstruction of the quadratic bounds, 2026-09-24. Let $m(n)$ be the largest cardinality of a set in $\mathbb R^n$ having exactly two distinct nonzero pairwise distances. Determine its size.

QUICK LITERATURE AND CONTEXT CHECK

The problem page records the asymptotically matching bounds $\binom{n+1}{2}\leq m(n)\leq\binom{n+2}{2}$, with the upper bound due to Bannai--Bannai--Stanton. Petrov--Pohoata give a polynomial-method proof for sets with any fixed number of distances. Below is a direct positive-definite specialization for two distances. It proves the quadratic asymptotic, not an exact formula in every dimension.

OBJECTIVE AND ATTACK PLAN

Use a polynomial vanishing on every off-diagonal pair. If there are too many points, a nonzero signed measure annihilating all quadratic polynomials contradicts positivity on the diagonal. Construct the lower bound from edges of a simplex.

WORK

The upper bound. Suppose a finite set $A$ has distances $a,b>0$, and set \[ P(x,y)=(\|x-y\|^2-a^2)(\|x-y\|^2-b^2). \] The real vector space of polynomials in $n$ variables of total degree at most two has dimension $D=\binom{n+2}{2}$. If $|A|>D$, linear algebra supplies real numbers $c_x$, not all zero, such that \[ \sum_{x\in A}c_x x^\alpha=0\qquad (|\alpha|\leq2). \] Every monomial $x^\alpha y^\beta$ in $P$ has $|\alpha|+|\beta|\leq4$. At least one of $|\alpha|,|\beta|$ is therefore at most two, so \[ \sum_{x,y\in A}c_xc_yP(x,y)=0. \] But $P(x,y)=0$ when $x\ne y$ and $P(x,x)=a^2b^2$. The same sum equals $a^2b^2\sum_xc_x^2>0$, a contradiction. An infinite two-distance set would contain a finite subset of more than $D$ points, so the argument also excludes that possibility.

The lower bound. In $\mathbb R^{n+1}$ take the vectors \[ A=\{e_i+e_j:1\leq i<j\leq n+1\}. \] They lie in the affine hyperplane $\sum x_i=2$, which is isometric to $\mathbb R^n$. Two such vectors have distance $\sqrt2$ when the corresponding pairs share one index, and distance $2$ when the pairs are disjoint. For $n\geq3$, both possibilities occur, giving exactly two distances and $\binom{n+1}{2}$ points. Hence \[ \frac{n(n+1)}2\leq m(n)\leq\frac{(n+1)(n+2)}2, \qquad m(n)=\frac{n^2}{2}+O(n). \] For the small-dimensional boundary cases, $\{0,1,2\}$ gives $m(1)=3$ by the upper bound; a regular pentagon gives $m(2)\geq5$. No exact planar upper bound is needed for the asymptotic conclusion.

RESULT AND LIMITATIONS

Attempt status: unresolved in the literal exact-value formulation. Both finite bounds and the asymptotic $m(n)\sim n^2/2$ are completely proved. An exact value of $m(n)$ for every $n$ is not supplied. The completion estimate reflects this distinction; the database's solved classification concerns the established quadratic bound. No external extremal theorem is needed in the proof above.

SOURCES

https://www.erdosproblems.com/502; F. Petrov and C. Pohoata, A remark on sets with few distances in $\mathbb R^d$, https://arxiv.org/abs/1912.08181.

Completion rate estimate: 85%.