OctoInformBeta

OpenAI preprint, September 2026 · §2. Resampling a spread family

§2 (setting): indexed spread laws: Event probabilities count indices, with full support imposed separately

Spreadness for a weighted indexed family.

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

Let XX be a ground set, let BB be a finite index set, and let Ab⊆XA_b\subseteq X be a finite set for each b∈Bb\in B. Sampling an index gives a random subset of XX; spreadness bounds the probability that this subset contains a prescribed nonempty finite set.

The Definition

Definition 1. A finite probability law ν=(νb)b∈B\nu=(\nu_b)_{b\in B} on BB is a family of real weights satisfying (1): νb≥0(b∈B),∑b∈Bνb=1.\nu_b\geq 0\quad(b\in B),\qquad \sum_{b\in B}\nu_b=1.(1)✦ The probability of an event E⊆BE\subseteq B under ν\nu is the sum in (2): ν(E)=∑b∈Eνb.\nu(E)=\sum_{b\in E}\nu_b.(2)✦ A finite probability law ν\nu on BB has full support if it satisfies (3): νb>0(b∈B).\nu_b>0\qquad(b\in B).(3)✦ For a real parameter aa, a finite probability law ν\nu on BB is aa-spread for the family (Ab)b∈B(A_b)_{b\in B} if it satisfies (4): ν({b∈B:J⊆Ab})≤a|J|for every nonempty finite J⊆X.\nu\bigl(\{b\in B:J\subseteq A_b\}\bigr)\leq a^{|J|} \qquad\text{for every nonempty finite }J\subseteq X.(4)✦

In the paper, the result reads: for finite XX and a>0a>0, a distribution on subsets AA of XX is aa-spread if ℙ(J⊆A)≤a|J|\mathbb P(J\subseteq A)\leq a^{|J|} for every nonempty J⊆XJ\subseteq X, and it may be represented by a finite indexed family (Ab)(A_b) with strictly positive weights νb\nu_b summing to one, allowing equal and empty labels.

What Is Checked

These definitions concern real weights on a finite index set and finite subset labels in a ground set XX, with a real spread parameter aa.