OctoInformBeta

OpenAI preprint, September 2026 · Section 1. Introduction

Section 1.1: finite Ising laws, conditional probabilities, and mixtures: Conditional weights on observed set II, common pinned set FF, and mixtures of prescriptions on source set JJ

Conditioning finite Ising laws on source spins.

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 GG be a finite ferromagnetic weighted multigraph with vertex set VV and edge set ℰ\mathcal E. Each edge ee has distinct endpoints ue,ve∈Vu_e,v_e\in V and a coupling Ke≥0K_e\geq 0; parallel edges are allowed. Let hv∈ℝh_v\in\mathbb R be the field at vertex vv. A spin configuration σ∈Ω={−1,+1}V\sigma\in\Omega=\{-1,+1\}^{V} is a function assigning a physical spin σv\sigma_v to each vertex of VV.✦

The Ising interaction exponent EhE_h and its Boltzmann weight whw_h are real-valued functions on Ω\Omega given by Eh(σ)=∑e∈ℰKeσueσve+∑v∈Vhvσv,wh(σ)=exp⁡(Eh(σ)).E_h(\sigma) =\sum_{e\in\mathcal E}K_e\sigma_{u_e}\sigma_{v_e} +\sum_{v\in V}h_v\sigma_v, \qquad w_h(\sigma)=\exp\bigl(E_h(\sigma)\bigr).(1) For a constraint A⊆ΩA\subseteq\Omega, the constrained partition sum Zh(A)Z_h(A) is the sum of the weights in (1) over configurations satisfying AA: Zh(A)=∑σ∈Awh(σ).Z_h(A)=\sum_{\sigma\in A}w_h(\sigma).(2)✦

The observed vertices form a set I⊆VI\subseteq V, the source vertices a set J⊆VJ\subseteq V, and the common pinned vertices a set F⊆VF\subseteq V. For S⊆VS\subseteq V and η∈Ω\eta\in\Omega, the agreement constraint AS(η)A_S(\eta) is the subset of Ω\Omega consisting of configurations that agree with η\eta on SS: AS(η)={σ∈Ω:σv=ηv for every v∈S}.A_S(\eta)=\{\sigma\in\Omega:\sigma_v=\eta_v\text{ for every }v\in S\}.(3)✦ Let a,b,τ∈Ωa,b,\tau\in\Omega give the observed prescription, source prescription, and fixed pin signs, respectively. A probability mixture λ\lambda of source prescriptions in Ω\Omega is a nonnegative real weight function on Ω\Omega satisfying λ(b)≥0(b∈Ω),∑b∈Ωλ(b)=1.\lambda(b)\geq 0\quad(b\in\Omega), \qquad \sum_{b\in\Omega}\lambda(b)=1.(4)

The Definition

Definition 1. For observed vertices II, source vertices JJ, pinned vertices FF, and prescriptions a,b,τ∈Ωa,b,\tau\in\Omega, the conditional Ising probability μI∣J,Fb,τ(a)\mu_{I\mid J,F}^{b,\tau}(a) is the ratio of the constrained partition sums (2) for the agreement constraints (3): μI∣J,Fb,τ(a)=Zh(AI(a)∩AJ(b)∩AF(τ))Zh(AJ(b)∩AF(τ)).\mu_{I\mid J,F}^{b,\tau}(a) =\frac{Z_h\bigl(A_I(a)\cap A_J(b)\cap A_F(\tau)\bigr)} {Z_h\bigl(A_J(b)\cap A_F(\tau)\bigr)}.(5)✦ The quotient in (5) is assigned value zero when its denominator vanishes. For a real weight function λ:Ω→ℝ\lambda:\Omega\to\mathbb R, the source mixture probability μI∣J,Fλ,τ(a)\mu_{I\mid J,F}^{\lambda,\tau}(a) for observed prescription aa and pin signs τ\tau is the weighted sum of the conditional Ising probabilities: μI∣J,Fλ,τ(a)=∑b∈Ωλ(b)μI∣J,Fb,τ(a).\mu_{I\mid J,F}^{\lambda,\tau}(a) =\sum_{b\in\Omega}\lambda(b)\,\mu_{I\mid J,F}^{b,\tau}(a).(6)✦ Probability mixtures use the weights in (4) in (6).

In the paper, the result reads: the nn vertices and mm edges are separately indexed by {0,…,n−1}\{0,\ldots,n-1\} and {0,…,m−1}\{0,\ldots,m-1\}, spins are Boolean functions with true representing +1+1 and false representing −1-1, the graph has no original self-loops but admits parallel edges and zero couplings without a connectivity assumption, prescriptions are full configurations used through their restrictions to the relevant sets, and a mixture is a nonnegative real weight on full source configurations summing to one and representing every probability mixture on the JJ spins.

What Is Checked

The definitions are stated over finite ferromagnetic weighted multigraphs with real fields, arbitrary observed, source and pinned sets, full spin prescriptions, and real weights on source configurations.