OctoInformBeta

OpenAI preprint, September 2026 · §1. The variational problem and the main theorem

The Stieltjes measure and the controlled diffusion: For an order parameter γ\gamma: μγ([0,t])=γ(t)\mu_\gamma([0,t]) = \gamma(t), the diffusion dX=γ(t)uγ(t,X)dt+dBdX = \gamma(t)\,u_\gamma(t,X)\,dt + dB with X0=0X_0 = 0, and q(t)=𝔼[uγ(t,Xt)2]q(t) = \mathbb{E}[u_\gamma(t,X_t)^2]

The Stieltjes measure, the controlled diffusion and the moment qq..

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

In the zero-temperature Sherrington-Kirkpatrick model, the Parisi functional of an order parameter 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, where Φγ(t,x)\Phi_\gamma(t,x) is the value function of a stochastic control problem. Three objects are attached to γ\gamma beside 𝒫\mathcal{P}: the Stieltjes measure μγ\mu_\gamma of γ\gamma; the diffusion XX steered by the gradient of Φγ\Phi_\gamma in the position; and the square moment qq of that gradient along XX, a function of time.

Order Parameters and Brownian Systems

The extension by zero of a function ff on the time interval [0,1)[0,1) is the function on ℝ\mathbb{R} that equals ff on [0,1)[0,1) and vanishes outside it.✦ An order parameter is a function γ:[0,1)→ℝ\gamma \colon [0,1) \to \mathbb{R} that is nonnegative, nondecreasing and right-continuous, and whose extension by zero is Lebesgue integrable on ℝ\mathbb{R}.✦ In integrals over real times, γ(s)\gamma(s) stands for this extension.

A Brownian system WW on a measurable space Ω\Omega consists of a probability measure ℙ\mathbb{P} on Ω\Omega, the law of WW, and a real process B=(Bt)t≥0B = (B_t)_{t \ge 0} of random variables on Ω\Omega that is a standard Brownian motion under ℙ\mathbb{P}.✦ Expectation under ℙ\mathbb{P} is written 𝔼\mathbb{E}.

The Control Problem

Fix a Brownian system WW and an order parameter γ\gamma. For a time t≥0t \ge 0, the increment filtration after tt is the natural filtration of the increments of BB after tt, indexed by the elapsed time s≥0s \ge 0: ℱst=σ(Bt+r−Bt:0≤r≤s).\mathcal{F}^t_s = \sigma\bigl(B_{t+r} - B_t : 0 \le r \le s\bigr).(1)✦ A bounded control from tt is a real process α=(αs)s≥0\alpha = (\alpha_s)_{s \ge 0} on Ω\Omega, indexed by the time elapsed since tt, that is progressively measurable for the filtration (1) and satisfies |αs(ω)|≤1|\alpha_s(\omega)| \le 1 for every s≥0s \ge 0 and every ω∈Ω\omega \in \Omega.✦

The payoff of a bounded control α\alpha from tt, started from a point x∈ℝx \in \mathbb{R}, is the random variable Jt,x(α)=|x+B1−Bt+∫t1γ(s)αs−tds|−12∫t1γ(s)αs−t2ds.J_{t,x}(\alpha) = \Bigl| x + B_1 - B_t + \int_t^1 \gamma(s)\,\alpha_{s-t}\,ds \Bigr| - \frac12 \int_t^1 \gamma(s)\,\alpha_{s-t}^2\,ds.(2)✦ The value function of γ\gamma in WW is the function Φγ\Phi_\gamma of a time tt and a position xx given by Φγ(t,x)=supα𝔼[Jt,x(α)],\Phi_\gamma(t,x) = \sup_\alpha \mathbb{E}\bigl[J_{t,x}(\alpha)\bigr],(3)✦ the supremum running over all bounded controls α\alpha from tt. The value gradient uγu_\gamma of γ\gamma in WW is the derivative of the value function Φγ\Phi_\gamma in the position, uγ(t,x)=Φγ,x(t,x)u_\gamma(t,x) = \Phi_{\gamma,x}(t,x).✦

The Definition

Definition 1. Let γ\gamma be an order parameter. A Stieltjes measure of γ\gamma is a measure μγ\mu_\gamma on [0,1)[0,1) such that μγ([0,t])=γ(t)for every t∈[0,1).\mu_\gamma([0,t]) = \gamma(t) \qquad \text{for every } t \in [0,1).(4)✦

Definition 2. Let WW be a Brownian system on Ω\Omega, with law ℙ\mathbb{P} and Brownian motion BB, and let γ\gamma be an order parameter with value gradient uγu_\gamma. A controlled diffusion of γ\gamma in WW is a real process X=(Xt)t≥0X = (X_t)_{t \ge 0} on Ω\Omega, progressively measurable for the natural filtration of BB, such that for ℙ\mathbb{P}-almost every ω\omega the path t↦Xt(ω)t \mapsto X_t(\omega) is continuous on [0,1][0,1], X0(ω)=0X_0(\omega) = 0, and Xt(ω)=Bt(ω)+∫0tγ(s)uγ(s,Xs(ω))dsfor every t∈[0,1].X_t(\omega) = B_t(\omega) + \int_0^t \gamma(s)\,u_\gamma\bigl(s, X_s(\omega)\bigr)\,ds \qquad \text{for every } t \in [0,1].(5)✦

Definition 3. For a Brownian system WW on Ω\Omega, an order parameter γ\gamma with value gradient uγu_\gamma and a real process X=(Xt)t≥0X = (X_t)_{t \ge 0} on Ω\Omega, the square moment of the value gradient along XX is the function of time q(t)=𝔼[uγ(t,Xt)2].q(t) = \mathbb{E}\bigl[u_\gamma(t, X_t)^2\bigr].(6)✦

What Is Checked

The Stieltjes measure (4) is stated over an order parameter γ\gamma on the time interval [0,1)[0,1), and the controlled diffusion (5) and the square moment (6) over a real process XX, the extension of γ\gamma by zero, a Brownian system WW with its law ℙ\mathbb{P} and Brownian motion BB, and the value gradient uγu_\gamma of the value function (3), which is built from the payoff (2) of the bounded controls for the increment filtration (1).