OctoInformBeta

OpenAI preprint, September 2026 · §2. A finite connection comparison

§2 (setting): Connections and cluster unions after suppressing hyperedges that meet a prescribed vertex set

Clusters formed by independently open hyperedges.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

A random hyperedge can connect several vertices at once. The library defines connectivity and source clusters for labelled hypergraphs, including suppression at a prescribed vertex set.

The Definition

Definition 1. A labelled hypergraph HH on a vertex set VV with label set EE is an indexed family of vertex subsets, written as in (1). H:E→𝒫(V),e↦He.H:E\longrightarrow\mathcal P(V),\qquad e\longmapsto H_e.(1)✦

A configuration ω∈{0,1}E\omega\in\{0,1\}^{E} is an assignment of a state to each label, with ωe=1\omega_e=1 meaning that label ee is open. For parameters pe∈[0,1]p_e\in[0,1], the Bernoulli product law ℙp\mathbb P_p on configurations is given by (2). ℙp=⨂e∈EBernoulli⁡(pe).\mathbb P_p=\bigotimes_{e\in E}\operatorname{Bernoulli}(p_e).(2)✦

For a suppression set D⊆VD\subseteq V, open adjacency x∼ω,DHyx\sim_{\omega,D}^{H}y in HH is the relation on vertices x,y∈Vx,y\in V that holds when they belong to a common open hyperedge disjoint from DD, as in (3). x∼ω,DHy⇔∃e∈E:ωe=1,He∩D=⌀,x,y∈He.x\sim_{\omega,D}^{H}y \quad\Longleftrightarrow\quad \exists e\in E:\quad \omega_e=1,\quad H_e\cap D=\varnothing,\quad x,y\in H_e.(3)✦

Open connectivity x↔ω,DHyx\leftrightarrow_{\omega,D}^{H}y in HH is the relation on vertices x,y∈Vx,y\in V that holds when they are joined by a finite adjacency path, allowing length zero, as in (4). x↔ω,DHy⇔∃n∈ℕ0,∃v0,…,vn∈V:v0=x,vn=y,vi∼ω,DHvi+1(0≤i<n).\begin{split} x\leftrightarrow_{\omega,D}^{H}y \quad\Longleftrightarrow\quad& \exists n\in\mathbb N_0,\ \exists v_0,\ldots,v_n\in V:\\ &v_0=x,\quad v_n=y,\quad v_i\sim_{\omega,D}^{H}v_{i+1}\quad(0\leq i<n). \end{split}(4)✦

For fixed H,ω,DH,\omega,D, the source cluster C(x)C(x) of a vertex x∈Vx\in V is the set of vertices connected to xx in (5). C(x)={y∈V:x↔ω,DHy}.C(x)=\{y\in V:x\leftrightarrow_{\omega,D}^{H}y\}.(5)✦

The cluster union C(S)C(S) of a source set S⊆VS\subseteq V is the union of the source clusters of vertices in SS, with HH, ω\omega and DD fixed, as in (6). C(S)=⋃x∈SC(x).C(S)=\bigcup_{x\in S}C(x).(6)✦

For a target set T⊆VT\subseteq V, connection x↔ω,DHTx\leftrightarrow_{\omega,D}^{H}T from x∈Vx\in V to TT is the relation that holds when xx is connected to at least one vertex of TT in HH with configuration ω\omega and suppression set DD, as in (7). x↔ω,DHT⇔∃y∈T:x↔ω,DHy.x\leftrightarrow_{\omega,D}^{H}T \quad\Longleftrightarrow\quad \exists y\in T:\ x\leftrightarrow_{\omega,D}^{H}y.(7)✦

In the paper, the result reads: a finite hypergraph has a finite vertex set VV and a finite labelled family (He)e∈E(H_e)_{e\in E} of nonempty vertex subsets, labels may repeat the same subset, open hyperedges join all their vertices, each cluster C(x)C(x) contains its source, C(S)C(S) is the union of the source clusters, and deleting DD removes its vertices and every hyperedge meeting DD.

What Is Checked

The hypergraph and connectivity definitions are stated over arbitrary vertex and label sets, with suppression and source or target subsets of the vertex set; the product law uses one parameter pe∈[0,1]p_e\in[0,1] for each label.