Wiki
Wiki

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

Updated

Exact realization of rational signed-relation spaces


Theorem — proved

Let V≤QnV\leq\mathbb Q^n be a rational subspace containing none of

ei,ei−ej,ei+ej(i≠j).(1)e_i,\quad e_i-e_j,\quad e_i+e_j\qquad(i\ne j). \tag{1}

Let H(V)\mathcal H(V) be the hypergraph on [n][n] whose edges are the supports of nonzero vectors in V∩{−1,0,1}nV\cap\{-1,0,1\}^n. There are distinct positive integers a1,…,ana_1,\ldots,a_n whose signed-relation hypergraph is exactly H(V)\mathcal H(V).

Proof

In the quotient W=Qn/VW=\mathbb Q^n/V, put vi=ei+Vv_i=e_i+V. Consider the finite set of all nonzero vectors of the form ∑iϵivi\sum_i\epsilon_i v_i, ϵi∈{−1,0,1}\epsilon_i\in\{-1,0,1\}. Choose a rational linear functional ℓ:W→Q\ell:W\to\mathbb Q which is nonzero on each of these vectors. Such a functional exists: the excluded functionals form finitely many proper hyperplanes over the infinite field Q\mathbb Q. Multiplying ℓ\ell by a common positive denominator gives integers bi=ℓ(vi)b_i=\ell(v_i), and

∑iϵibi=0⟺(ϵi)i∈V(ϵi∈{−1,0,1}).(2)\sum_i\epsilon_i b_i=0\quad\Longleftrightarrow\quad (\epsilon_i)_i\in V \qquad(\epsilon_i\in\{-1,0,1\}). \tag{2}

By (1), all bib_i are nonzero and bi≠±bjb_i\ne\pm b_j for distinct indices. Set ai=∣bi∣a_i=|b_i| and σi=sgn⁡(bi)\sigma_i=\operatorname{sgn}(b_i). Then ∑iϵiai=0\sum_i\epsilon_i a_i=0 if and only if (σiϵi)i∈V(\sigma_i\epsilon_i)_i\in V. The diagonal sign change is a bijection of the ternary cube preserving every support, proving the theorem.

Fully explicit finite projection

Represent the viv_i as integer columns in Zd\mathbb Z^d, after clearing denominators. Let MM bound all absolute entries and choose an integer B>nM+1B>nM+1. The map

L(x0,…,xd−1)=∑j=0d−1xjBjL(x_0,\ldots,x_{d-1})=\sum_{j=0}^{d-1}x_jB^j

is nonzero on every nonzero signed sum of columns. Indeed, if its highest nonzero coordinate is ss, the leading term has absolute value at least BsB^s, while all lower terms together have absolute value at most nM(Bs−1)/(B−1)<BsnM(B^s-1)/(B-1)<B^s. Taking absolute values as above produces the positive integer realization. This is a finite encoding, not an infinite bounded-base encoding.

Signed closure is the essential difficulty

If R⊆{−1,0,1}n\mathcal R\subseteq\{-1,0,1\}^n is a list of planted relations, every realization necessarily also satisfies every vector in

span⁡Q(R)∩{−1,0,1}n.\operatorname{span}_{\mathbb Q}(\mathcal R)\cap\{-1,0,1\}^n.

Conversely, if that rational span obeys (1), the theorem realizes exactly this closure, up to signs. Thus an abstract high-chromatic hypergraph is not enough: one must control the ternary vectors in the entire rational span of its signed edges, not only the prescribed edges.

For any rational VV, replacing it by V′=span⁡Q(V∩{−1,0,1}n)V'=\operatorname{span}_{\mathbb Q}(V\cap\{-1,0,1\}^n) leaves its ternary intersection unchanged: V′⊆VV'\subseteq V gives one inclusion and the definition gives the other. A search may therefore restrict to spaces generated by their signed relations.

Consequence for E0774

It suffices to construct spaces VtV_t obeying (1) such that H(Vt)\mathcal H(V_t) has one uniform hereditary independence ratio and unbounded chromatic number. This is an exact alternative formulation of the finite-block obligation, not a construction satisfying it.