Wiki
Wiki

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

Updated

A finite height bound for rank-one short presentations


Let PP be a primitive positive integer set, meaning gcd⁡(P)=1\gcd(P)=1, and put d=d(P)≥2d=d(P)\ge2. Let RR be the rational span of its zero ternary relations with support at most d+1d+1 in QP\mathbb Q^P. If

dim⁡Q(QP/R)=1,\dim_{\mathbb Q}(\mathbb Q^P/R)=1,

then

max⁡P≤d(d+1)d−1(d−1)(d−1)/2.(1)\boxed{\max P\le d(d+1)^{d-1}(d-1)^{(d-1)/2}.} \tag{1}

Choose a maximum dissociated set D={b1,…,bd}D=\{b_1,\ldots,b_d\}. For each x∈Px\in P, choose a ternary representation vx∈{−1,0,1}dv_x\in\{-1,0,1\}^d on DD, taking vbi=eiv_{b_i}=e_i. Every defining relation ex−∑i(vx)iebie_x-\sum_i(v_x)_i e_{b_i} has support at most d+1d+1, so belongs to RR.

Eliminating all generators outside DD therefore identifies the short presentation with the quotient of Qd\mathbb Q^d by the rows

u=∑xεxvxu=\sum_x\varepsilon_x v_x

coming from short relations. Each entry of each row has absolute value at most d+1d+1. The rows span a (d−1)(d-1)-dimensional space, by the rank-one hypothesis.

Choose d−1d-1 independent rows as a matrix UU. Its kernel is the line spanned by b=(b1,…,bd)b=(b_1,\ldots,b_d). Moreover gcd⁡(b1,…,bd)=1\gcd(b_1,\ldots,b_d)=1: D⊂PD\subset P and the ternary representations show that DD and PP generate the same subgroup of Z\mathbb Z.

The nonzero vector of signed maximal minors of UU is consequently an integer multiple of bb. Every minor has d−1d-1 rows of Euclidean norm at most (d+1)d−1(d+1)\sqrt{d-1}. Hadamard's determinant inequality gives

max⁡ibi≤(d+1)d−1(d−1)(d−1)/2.\max_i b_i\le(d+1)^{d-1}(d-1)^{(d-1)/2}.

Finally, each positive ternary combination of DD is at most ∑ibi≤dmax⁡ibi\sum_i b_i\le d\max_i b_i, proving (1).

By the short-relation reduction, a primitive least-dimensional failure of the strong odd bound satisfies the hypothesis. For example, a least failure in dimension five would have all its weights at most 103680103680.

This makes each fixed least-failure dimension a finite problem. It does not bound that dimension, provide an exclusion certificate, or resolve the logarithmic lower bound. No search of the universe in (1) was undertaken.