OctoInformBeta

OpenAI preprint, September 2026 · Section 1. The theorem and the structure of the proof

Section 1 setting: the zero-field Gaussian SK law: Independent reference couplings WW and the scaling W↦βWW\mapsto\beta W for inverse temperature β\beta

The zero-field Gaussian Sherrington-Kirkpatrick Gibbs law.

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

Fix a nonnegative integer nn, with sites indexed by 1,…,n1,\ldots,n. The zero-field Sherrington-Kirkpatrick (SK) model is a finite Gibbs model with Gaussian pair interactions.

The Claim

Definition 1. The sign function σ\sigma on the two spin states ++ and −- is given by (1): σ(+)=+1,σ(−)=−1.\sigma(+) = +1, \qquad \sigma(-) = -1.(1)✦ A spin configuration xx on the sites 1,…,n1,\ldots,n is an assignment of a sign to each site; its configuration space Ωn\Omega_n is given by (2): Ωn={−1,+1}n.\Omega_n = \{-1,+1\}^{n}.(2)✦ The interaction edge set EnE_n consists of the pairs of distinct sites among 1,…,n1,\ldots,n listed in (3): En={(i,j):1≤i<j≤n}.E_n = \{(i,j): 1\leq i<j\leq n\}.(3)✦ A coupling array WW on the interaction edge set EnE_n is a real array indexed by EnE_n, as in (4): W=(Wij)(i,j)∈En∈ℝEn.W=(W_{ij})_{(i,j)\in E_n}\in\mathbb R^{E_n}.(4)✦ The zero-field Hamiltonian HW:Ωn→ℝH_W:\Omega_n\to\mathbb R is defined by (5): HW(x)=∑1≤i<j≤nWijxixj.H_W(x)=\sum_{1\leq i<j\leq n}W_{ij}x_i x_j.(5)✦ For a coupling array WW on EnE_n, the partition function ZWZ_W is the sum of the exponential weights of the configurations in Ωn\Omega_n, as given in (6): ZW=∑x∈Ωnexp⁡(HW(x)).Z_W=\sum_{x\in\Omega_n}\exp\bigl(H_W(x)\bigr).(6)✦ The Gibbs probability μW\mu_W on Ωn\Omega_n has the masses in (7): μW(x)=exp⁡(HW(x))ZW.\mu_W(x)=\frac{\exp\bigl(H_W(x)\bigr)}{Z_W}.(7)✦ Let δ⌀\delta_{\varnothing} denote the point mass on the empty coupling array. The reference Gaussian disorder law 𝖯n\mathsf P_n on ℝEn\mathbb R^{E_n} is the product measure (8): 𝖯n={⨂(i,j)∈EnN(0,1/n),n>0,δ⌀,n=0,\mathsf P_n= \begin{cases} \displaystyle\bigotimes_{(i,j)\in E_n}N(0,1/n), & n>0,\\ \delta_{\varnothing}, & n=0, \end{cases}(8)✦

For a real inverse temperature β\beta, the SK Gaussian disorder law 𝒢β,n\mathcal G_{\beta,n} on ℝEn\mathbb R^{E_n} is the probability measure with independent N(0,β2/n)N(0,\beta^2/n) coordinates when n>0n>0, and the point mass on the empty array when n=0n=0. Multiplication of the couplings by β\beta sends the reference law to this law, as in (9): (W↦βW)*𝖯n=𝒢β,n.(W\mapsto\beta W)_*\mathsf P_n=\mathcal G_{\beta,n}.(9)✦ The SK Hamiltonian with external field set to zero, HW,0SKH^{\mathrm{SK}}_{W,0}, satisfies (10): HW,0SK(x)=HW(x).H^{\mathrm{SK}}_{W,0}(x)=H_W(x).(10)✦ The corresponding zero-field Gibbs mass μW,0SK\mu^{\mathrm{SK}}_{W,0} satisfies (11): μW,0SK(x)=μW(x).\mu^{\mathrm{SK}}_{W,0}(x)=\mu_W(x).(11)✦

What Is Checked

The definitions and identities are stated for every finite system size nn, every real coupling array WW and every configuration xx, with the temperature identity (9) allowing any real β\beta.