OctoInformBeta

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

Proof of Theorem 1.1: Its Stieltjes measure μγ\mu_\gamma has support [0,1)[0,1), γ\gamma is strictly increasing, and q(t)=tq(t)=t, 𝔼Φγ,xx(t,Xt)2=1\mathbb{E}\,\Phi_{\gamma,xx}(t,X_t)^2=1 on [0,1)[0,1)

Every zero-temperature Parisi minimizer has full support.

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, with covariance ξ(t)=t2/2\xi(t)=t^2/2, the Parisi problem asks for an order parameter γ\gamma on [0,1)[0,1) minimizing the Parisi functional 𝒫\mathcal{P}. The question is how flat a minimizer can be. Is there an interval of [0,1)[0,1), perhaps one reaching up to 11, on which γ\gamma is constant, so that the support of its Stieltjes measure μγ\mu_\gamma has a gap? And does the second moment q(t)q(t) of the gradient along the diffusion XX of γ\gamma run along the diagonal q(t)=tq(t)=t, with slope one?

Order Parameters and Contact

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

Let γ\gamma be an order parameter with a Stieltjes measure μγ\mu_\gamma. A pair (q,q′)(q,q') of real functions satisfies the contact conditions on the support of μγ\mu_\gamma when q(t)=tq(t)=t and q′(t)≤1q'(t)\le 1 at every time tt of supp⁡μγ\operatorname{supp}\mu_\gamma. A terminal plateau of γ\gamma is an interval (a,1)(a,1), with 0≤a<10\le a<1, on which γ\gamma is constant. A flat interval of γ\gamma is an interval (a,b)(a,b), with 0≤a<b<10\le a<b<1, on which γ\gamma is constant.

The Parisi Problem

A Brownian system WW is a measurable space Ω\Omega with a probability measure ℙ\mathbb{P}, its law, and a standard Brownian motion B=(Bt)t≥0B=(B_t)_{t\ge 0} under ℙ\mathbb{P}; 𝔼\mathbb{E} is the expectation under ℙ\mathbb{P}.✦

A bounded control α\alpha from a time t≥0t\ge 0 is a real process (αs)s≥0(\alpha_s)_{s\ge 0} on Ω\Omega, bounded by one, |αs|≤1|\alpha_s|\le 1, and progressively measurable for the increment filtration after tt, in which ℱst\mathcal{F}^t_s is generated by the increments Bt+r−BtB_{t+r}-B_t, 0≤r≤s0\le r\le s.✦ The value function of an order parameter γ\gamma is the supremum of the expected control payoff over bounded controls α\alpha from tt: Φγ(t,x)=supα𝔼[|x+B1−Bt+∫t1γ(s)αs−tds|−12∫t1γ(s)αs−t2ds].\Phi_\gamma(t,x)=\sup_\alpha \mathbb{E}\Bigl[\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\Bigr].(2)✦ The gradient uγ=Φγ,xu_\gamma=\Phi_{\gamma,x} and the curvature Φγ,xx\Phi_{\gamma,xx} of the value function Φγ\Phi_\gamma are its first two derivatives in the position xx.

The Parisi functional of WW assigns to an order parameter γ\gamma the number 𝒫(γ)=Φγ(0,0)−12∫01tγ(t)dt,\mathcal{P}(\gamma)=\Phi_\gamma(0,0)-\frac12\int_0^1 t\,\gamma(t)\,dt,(3)✦ and a Parisi minimizer for WW 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

A bounded Lipschitz drift is a jointly measurable function b(t,x)b(t,x) of time and position 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 constants M,L≥0M,L\ge 0.✦ The Picard solution XX of the equation Xt=Bt+∫0tb(s,Xs)dsX_t=B_t+\int_0^t b(s,X_s)\,ds for a bounded Lipschitz drift bb, driven by a process BB, is X=B+YX=B+Y, where YY is the limit as k→∞k\to\infty of the iterates✦ Y0=0,Ytk+1=∫0tb(s,Bs+Ysk)ds.Y^0=0,\qquad Y^{k+1}_t=\int_0^t b\bigl(s,B_s+Y^k_s\bigr)\,ds.(4)✦

The diffusion of an order parameter γ\gamma is driven by the driving Brownian motion of WW: the progressive version of BB, the limit of B⌊(n+1)t⌋/(n+1)B_{\lfloor (n+1)t\rfloor/(n+1)} as n→∞n\to\infty, less its value at time 00, again written BB.✦ For a horizon TT in [0,1)[0,1), the strip drift of TT is the function (t,x)↦γ(t‾)uγ(t‾,x)(t,x)\mapsto\gamma(\bar t)\,u_\gamma(\bar t,x) of time tt and position xx, where t‾=max⁡(0,min⁡(T,t))\bar t=\max(0,\min(T,t)) is the time tt clamped to the strip [0,T][0,T].✦ For the horizons Tn=1−1/(n+2)T_n=1-1/(n+2), which increase from T0=1/2T_0=1/2 toward 11, the localized diffusion XnX^n is the Picard solution driven by BB for the strip drift of TnT_n: Xtn=Bt+∫0tγ(s∧Tn)uγ(s∧Tn,Xsn)ds.X^n_t=B_t+\int_0^t\gamma(s\wedge T_n)\,u_\gamma(s\wedge T_n,X^n_s)\,ds.(5)✦

With Xˆt=lim⁡n→∞Xtn\hat X_t=\lim_{n\to\infty}X^n_t, the diffusion of γ\gamma is Xt=Bt+∫0tγ(s)uγ(s,Xˆs)ds,X_t=B_t+\int_0^t\gamma(s)\,u_\gamma(s,\hat X_s)\,ds,(6)✦ with the feedback uγ(s,Xˆs)u_\gamma(s,\hat X_s) set to 00 from time 11 on: the integrated form of dX=γ(t)uγ(t,X)dt+dBdX=\gamma(t)\,u_\gamma(t,X)\,dt+dB, X0=0X_0=0. The second moment of the gradient along XX is q(t)=𝔼uγ(t,Xt)2,q(t)=\mathbb{E}\,u_\gamma(t,X_t)^2,(7)✦ and the second moment of the curvature is 𝔼Φγ,xx(t,Xt)2\mathbb{E}\,\Phi_{\gamma,xx}(t,X_t)^2.✦

The Claim

Theorem 1. Let γ\gamma be an order parameter with a Stieltjes measure μγ\mu_\gamma, and let qq and q′q' be real functions, continuous on [0,1)[0,1), with q(0)=0q(0)=0 and q′q' the derivative of qq on (0,1)(0,1). Suppose that (q,q′)(q,q') satisfies the contact conditions on the support of μγ\mu_\gamma, that γ\gamma has no terminal plateau, and that on every flat interval of γ\gamma one has q‴>0q'''>0 if the constant value of γ\gamma there is positive and q″>0q''>0 if it is 00. Then supp⁡μγ=[0,1),γis strictly increasing on[0,1),q(t)=t(0≤t<1).\operatorname{supp}\mu_\gamma=[0,1),\qquad \gamma\ \text{is strictly increasing on}\ [0,1), \qquad q(t)=t\quad(0\le t<1).(8)✦

Theorem 2. Let qq and q′q' be real functions with q(t)=tq(t)=t for 0≤t<10\le t<1, such that q′(t)q'(t) is the derivative of qq at every tt in (0,1)(0,1) and q′(0)q'(0) is its right derivative at 00. Then q′(t)=1,0≤t<1.q'(t)=1,\qquad 0\le t<1.(9)✦

Theorem 3. Let WW be a Brownian system, γ\gamma a Parisi minimizer for WW and μγ\mu_\gamma a Stieltjes measure of γ\gamma, with XX the diffusion of γ\gamma and qq the second moment (7). Then supp⁡μγ=[0,1),γstrictly increasing,q(t)=t,𝔼Φγ,xx(t,Xt)2=1(0≤t<1).\operatorname{supp}\mu_\gamma=[0,1),\qquad \gamma\ \text{strictly increasing},\qquad q(t)=t,\qquad \mathbb{E}\,\Phi_{\gamma,xx}(t,X_t)^2=1\quad(0\le t<1).(10)✦

For the minimizer, the slope q′(t)=1q'(t)=1 asked for in the question is stated in (10) as 𝔼Φγ,xx(t,Xt)2=1\mathbb{E}\,\Phi_{\gamma,xx}(t,X_t)^2=1.

What Is Checked

The criterion (8) is stated over an order parameter, a Stieltjes measure of it and two real functions qq and q′q'; the identity (9) over two real functions alone; the theorem (10) over a Brownian system WW, the value function Φγ\Phi_\gamma of bounded controls, the Parisi functional of WW and its minimizers, a Stieltjes measure of the minimizer, and the diffusion XX built from the localized diffusions XnX^n. Of these constructions the library proves that each horizon TnT_n lies in [0,1)[0,1)✦, and that the strip drift of a horizon TT in [0,1)[0,1) is a bounded Lipschitz drift with bound γ(T)\gamma(T).