OctoInformBeta

OpenAI preprint, September 2026 · §6. Full support and regularity

Lemma 6.2: For a minimizer γ\gamma, q(t)=tq(t)=t and 𝔼[Φγ,xx(t,Xt)2]≤1\mathbb{E}[\Phi_{\gamma,xx}(t,X_t)^2]\le 1 on the support of μγ\mu_\gamma, and the contact potential GG is nonnegative

The contact conditions hold at a Parisi minimizer.

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

At zero temperature, the Parisi formula for the Sherrington-Kirkpatrick model gives the ground-state energy as the minimum of the Parisi functional 𝒫\mathcal{P} over order parameters γ\gamma. The functional is the value at the origin of a stochastic control problem, less a term linear in γ\gamma; the gradient of the value function, weighted by γ\gamma, is the drift of a diffusion XX. Take a minimizer γ\gamma. What does minimality say about the second moment of that gradient along XX, at the times in the support of the Stieltjes measure μγ\mu_\gamma of γ\gamma, and about the integral, from a time ss to 11, of the excess of that second moment over the time itself?

Order Parameters and the Parisi Functional

An order parameter is a function γ\gamma on the time interval [0,1)[0,1) that is nonnegative, nondecreasing, right-continuous and Lebesgue integrable; outside [0,1)[0,1) it is extended by zero.✦ A Stieltjes measure of γ\gamma is a measure μγ\mu_\gamma on [0,1)[0,1) with μγ([0,t])=γ(t)\mu_\gamma([0,t])=\gamma(t) for every t∈[0,1)t\in[0,1).✦

A Brownian system WW on a measurable space Ω\Omega is a probability measure ℙ\mathbb{P} on Ω\Omega together with a real process B=(Bt)t≥0B=(B_t)_{t\ge 0}, measurable at each time, that is a standard Brownian motion under ℙ\mathbb{P}; 𝔼\mathbb{E} is the expectation under ℙ\mathbb{P}.✦ For a time t≥0t\ge 0, a bounded control from tt is a real process α=(αr)r≥0\alpha=(\alpha_r)_{r\ge 0} with |αr|≤1|\alpha_r|\le 1 at every time and outcome, progressively measurable for the filtration generated by the increments Bt+r−BtB_{t+r}-B_t, r≥0r\ge 0.✦

Fix a Brownian system WW. The value function of an order parameter γ\gamma is the function of time tt and position xx given by the supremum, over bounded controls α\alpha from tt, of the expected control payoff: Φγ(t,x)=supα𝔼[|x+B1−Bt+∫t1γ(s)αs−tds|−12∫t1γ(s)αs−t2ds].\Phi_\gamma(t,x)=\sup_\alpha\,\mathbb{E}\Big[\Big|x+B_1-B_t+\int_t^1\gamma(s)\,\alpha_{s-t}\,ds\Big| -\frac12\int_t^1\gamma(s)\,\alpha_{s-t}^2\,ds\Big].(1)✦ Its derivatives in xx are the value gradient uγ=Φγ,xu_\gamma=\Phi_{\gamma,x}✦ and the value curvature Φγ,xx=∂xuγ\Phi_{\gamma,xx}=\partial_x u_\gamma.✦

The Parisi functional at γ\gamma is 𝒫(γ)=Φγ(0,0)−12∫01tγ(t)dt,\mathcal{P}(\gamma)=\Phi_\gamma(0,0)-\frac12\int_0^1 t\,\gamma(t)\,dt,(2)✦ and a Parisi minimizer is an order parameter γ\gamma with 𝒫(γ)≤𝒫(η)\mathcal{P}(\gamma)\le\mathcal{P}(\eta) for every order parameter η\eta.✦

The Diffusion and the Contact Potential

A bounded Lipschitz drift is a jointly measurable function bb of time tt and position xx for which there are constants M,L≥0M,L\ge 0 with |b(t,x)|≤M|b(t,x)|\le M and |b(t,x)−b(t,y)|≤L|x−y||b(t,x)-b(t,y)|\le L\,|x-y| for all tt, xx, yy.✦ Given a process ZZ, the Picard solution of Xt=Zt+∫0tb(s,Xs)dsX_t=Z_t+\int_0^t b(s,X_s)\,ds is X=Z+YX=Z+Y✦, where YtY_t is the limit as k→∞k\to\infty of the iterates✦ Y0=0,Ytk+1=∫0tb(s,Zs+Ysk)ds.Y^0=0,\qquad Y^{k+1}_t=\int_0^t b\big(s,Z_s+Y^k_s\big)\,ds.(3)✦

The driving Brownian motion of WW is the process B‾t=B̃t−B̃0\bar B_t=\tilde B_t-\tilde B_0✦, where the progressive version B̃t\tilde B_t of BB is the limit as n→∞n\to\infty of B⌊(n+1)t⌋/(n+1)B_{\lfloor (n+1)t\rfloor/(n+1)}, the value of BB at the last point not after tt of the grid of mesh 1/(n+1)1/(n+1).✦

For n≥0n\ge 0 the horizon is Tn=1−1/(n+2)T_n=1-1/(n+2)✦, and the localized diffusion XnX^n is the Picard solution, driven by B‾\bar B, for the drift γ(s)uγ(s,x)\gamma(s)\,u_\gamma(s,x) with its time ss clamped to [0,Tn][0,T_n]: Xtn=B‾t+∫0tγ(s∧Tn)uγ(s∧Tn,Xsn)ds.X^n_t=\bar B_t+\int_0^t\gamma(s\wedge T_n)\,u_\gamma\big(s\wedge T_n,X^n_s\big)\,ds.(4)✦ With Xˆt=lim⁡n→∞Xtn\hat X_t=\lim_{n\to\infty}X^n_t✦, the diffusion of γ\gamma is Xt=B‾t+∫0tγ(s)uγ(s,Xˆs)ds,X_t=\bar B_t+\int_0^t\gamma(s)\,u_\gamma\big(s,\hat X_s\big)\,ds,(5)✦ where uγ(s,⋅)u_\gamma(s,\cdot) is taken to be 00 for s≥1s\ge 1✦.

Along XX, the second moment of the value gradient is q(t)=𝔼[uγ(t,Xt)2],q(t)=\mathbb{E}\big[u_\gamma(t,X_t)^2\big],(6)✦ that of the value curvature is 𝔼[Φγ,xx(t,Xt)2]\mathbb{E}[\Phi_{\gamma,xx}(t,X_t)^2]✦, and the contact potential of γ\gamma is the function of a real number ss G(s)=∫s1(q(t)−t)dt.G(s)=\int_s^1\big(q(t)-t\big)\,dt.(7)✦

The Claim

Theorem 1. Let WW be a Brownian system on a measurable space Ω\Omega, let γ\gamma be a Parisi minimizer for WW, and let μγ\mu_\gamma be a Stieltjes measure of γ\gamma. At every time tt in the support of μγ\mu_\gamma, q(t)=tand𝔼[Φγ,xx(t,Xt)2]≤1,q(t)=t\qquad\text{and}\qquad\mathbb{E}\big[\Phi_{\gamma,xx}(t,X_t)^2\big]\le 1,(8)✦ and at every time s∈[0,1)s\in[0,1) the contact potential is nonnegative: G(s)≥0.G(s)\ge 0.(9)✦

In the paper, the result reads: for a minimizer γ\gamma, the contact potential G(s)=∫s1(q(t)−t)dtG(s)=\int_s^1(q(t)-t)\,dt is nonnegative and vanishes on the support of μγ\mu_\gamma, and at every point ss of the support, q(s)=sq(s)=s and q′(s)≤1q'(s)\le 1.

What Is Checked

The claim is stated over a Brownian system WW, an order parameter γ\gamma that minimizes the Parisi functional (2), a Stieltjes measure μγ\mu_\gamma of γ\gamma and the times in its support, and the second moments of the value gradient and the value curvature along the diffusion (5), with the contact potential (7), all built from the value function (1). About these constructions, the library proves that each horizon TnT_n lies in [0,1)[0,1)✦ and that the clamped drift in (4) is a bounded Lipschitz drift with bound γ(Tn)\gamma(T_n).✦