OctoInformBeta

OpenAI preprint, September 2026 · §5. Uniform Gaussian survival · §5.1

§5.1 (setting): the stationary Ornstein-Uhlenbeck process: The state at step tt as a matrix on standard Gaussian noise, and covariance rt−ur^{t-u} between steps u≤t≤nu \le t \le n for 0≤r≤10 \le r \le 1

The stationary Gaussian autoregressive chain and its covariance.

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 nn be a natural number, let rr be a real number with −1≤r≤1-1 \le r \le 1, and let ξ0,ξ1,…,ξn\xi_0, \xi_1, \dots, \xi_n be independent standard Gaussian variables. The stationary Gaussian autoregressive chain with coefficient rr starts at X0=ξ0X_0 = \xi_0 and moves by Xt=rXt−1+1−r2ξt,1≤t≤n.X_t = r\,X_{t-1} + \sqrt{1 - r^2}\,\xi_t, \qquad 1 \le t \le n.(1)

Given coefficients rir_i for the coordinates ii of an index set ι\iota, the vector chain with coefficients rir_i is the family of states XtX_t, 0≤t≤n0 \le t \le n, whose coordinate ii is the stationary Gaussian autoregressive chain with coefficient rir_i, all driven by a noise array ξ=(ξk,i)\xi = (\xi_{k,i}) with 0≤k≤n0 \le k \le n and i∈ιi \in \iota, distributed as the standard Gaussian on ℝ{0,…,n}×ι\mathbb{R}^{\{0,\dots,n\} \times \iota}. Time is discrete: in the estimates on uniform Gaussian survival, this chain, observed at the steps 0,…,n0, \dots, n, takes the place of the stationary Ornstein-Uhlenbeck process.

The Autoregressive Row

The autoregressive row of a real coefficient rr is the array of reals ρr(t,k)\rho_r(t,k), indexed by natural numbers tt (the step) and kk (the noise index), defined by recursion on the step: ρr(0,k)=[k=0],ρr(t+1,k)=rρr(t,k)+1−r2[k=t+1],\rho_r(0,k) = [k = 0], \qquad \rho_r(t+1,k) = r\,\rho_r(t,k) + \sqrt{1 - r^2}\,[k = t+1],(2)✦ where [P][P] is 11 when PP holds and 00 otherwise. For −1≤r≤1-1 \le r \le 1, the entry ρr(t,k)\rho_r(t,k) is the weight of the noise ξk\xi_k in the state XtX_t of (1), so that Xt=∑k=0nρr(t,k)ξkX_t = \sum_{k=0}^{n} \rho_r(t,k)\,\xi_k for every t≤nt \le n.

The Claim

Definition 1. Let r=(ri)i∈ιr = (r_i)_{i \in \iota} be real coefficients, nn a natural number and tt a step with 0≤t≤n0 \le t \le n. The autoregressive evaluation at step tt is the real matrix EtE_t with rows indexed by ι\iota and columns indexed by the pairs (k,j)∈{0,…,n}×ι(k,j) \in \{0,\dots,n\} \times \iota, whose entries are (Et)i,(k,j)={ρri(t,k)if j=i,0otherwise.(E_t)_{i,(k,j)} = \begin{cases} \rho_{r_i}(t,k) & \text{if } j = i, \\ 0 & \text{otherwise.} \end{cases}(3)✦ Applied to a noise array ξ\xi, it gives the state EtξE_t\,\xi at step tt of the vector chain with coefficients rir_i.

Theorem 1. Let rr be a real number with 0≤r≤10 \le r \le 1, and let u≤t≤nu \le t \le n be natural numbers. Then the states XtX_t and XuX_u of the chain with coefficient rr satisfy Cov⁡(Xt,Xu)=∑k=0nρr(t,k)ρr(u,k)=rt−u.\operatorname{Cov}(X_t, X_u) = \sum_{k=0}^{n} \rho_r(t,k)\,\rho_r(u,k) = r^{t-u}.(4)✦

In the paper, the result reads: the stationary Ornstein-Uhlenbeck process UU solves dU=−QUdτ+2QdBdU = -QU\,d\tau + \sqrt{2Q}\,dB, where QQ is a positive definite matrix, with U(0)U(0) standard Gaussian and BB a Brownian motion, and Cov⁡(U(τ),U(σ))=e−Q|τ−σ|\operatorname{Cov}(U(\tau), U(\sigma)) = e^{-Q|\tau - \sigma|}.

What Is Checked

The definition (3) is stated over the autoregressive row (2) of each coefficient rir_i, and the covariance identity (4) over the row of a single coefficient rr with 0≤r≤10 \le r \le 1, at steps u≤t≤nu \le t \le n.