By OpenAISourceGitHub

Manuscripts from openai/math

Fork
Clone
git clone https://openversions.com/git/index/openai-math.git
Cite
Version link
https://openversions.com/index/openai-math/commit/dfd8c46870b48511e697cbb43a3c8c91d375fa2f
Version identifier (commit hash)
dfd8c46870b48511e697cbb43a3c8c91d375fa2f

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

All papers

Claims

Findings

Uniform real Lipschitz surfaces on the triangular lattice

There exist constants A>0A > 0, σ>0\sigma > 0, and λ∈(0,1/2]\lambda \in (0, 1/2], independent of D,a,bD, a, b, such that for uniformly sampled real nearest-neighbor Lipschitz heights hδh_\delta on triangular-lattice approximations DδD_\delta of a bounded smooth simply connected domain (D;a,b)(D; a, b) with boundary values ±λ\pm \lambda, the zero-height interface γδ\gamma_\delta converges in uniform curve distance modulo increasing reparametrization to chordal SLE4(D;a,b)\mathrm{SLE}_4(D; a, b), and (hδ−Ehδ)/σ(h_\delta - \mathbb{E}h_\delta)/\sigma converges as a random distribution to the Dirichlet Gaussian free field GFFD\mathrm{GFF}_D, with v=2/3v = 2/\sqrt{3}, σ2=(vA)−1\sigma^2 = (vA)^{-1}, λ2=π8vA\lambda^2 = \frac{\pi}{8vA}, and limiting mean mD(z)=λ[ωD(z,A+)−ωD(z,A−)]m_D(z) = \lambda[\omega_D(z, A_+) - \omega_D(z, A_-)].

For a sample hh from a bounded-data uniform Lipschitz law μb\mu_b (with ∣b∣≤B|b| \le B) and deterministic weights qxq_x satisfying ∑x∣qx∣≤C\sum_x |q_x| \le C, max⁡x∣qx∣≤CR−2\max_x |q_x| \le C R^{-2}, supp⁡q⊂Q\operatorname{supp} q \subset Q, diam⁡Q≤CR\operatorname{diam} Q \le C R, under either ∑xqx=0\sum_x q_x = 0 or a connected pin set KK of diameter at least cRcR in a CRCR-box with QQ, there exist a>0a > 0 and M<∞M < \infty depending only on c,C,Bc, C, B such that Eexp⁡(a∣∑xqxh(x)∣)≤M\mathbb{E}\exp(a|\sum_x q_x h(x)|) \le M.

In the stationary synchronous reflected comparison between two finite-volume Lipschitz fields with bounded Dirichlet data, one-sided sign walls, and annealed adapted exact-value observations, the good-edge and good-wall energy En(Qr)\mathcal{E}_n(Q_r) and step-cluster dissipation ∑C:C∩Qr≠∅DC\sum_{C: C \cap Q_r \neq \emptyset} D_C have uniformly bounded expectations, and the sparse-hole extension UnU_n satisfies E∥Un∥W1,p(n−1Qr)2≤C\mathbb{E}\|U_n\|_{W^{1,p}(n^{-1}Q_r)}^2 \le C and E∫n,Qr∣Un−u0∣→0\mathbb{E}\int_{n,Q_r} |U_n - u_0| \to 0 for 1<p<21 < p < 2.

Under the comparison and average estimates up to identical fixed boundaries of lattice domains converging to a bounded smooth domain DD, the tangent-flux scalar AA satisfies 0<A<∞0 < A < \infty, the mean response to relative drift n−2b(x/n)n^{-2}b(x/n) converges to A−1(−ΔD)−1bA^{-1}(-\Delta_D)^{-1}b, the untilted mean converges to the bounded harmonic extension of ±λ\pm\lambda, and the centered affinely interpolated field converges as a random distribution to σGFFD\sigma \mathrm{GFF}_D with σ2=(vA)−1\sigma^2 = (vA)^{-1}.

For deterministic oriented sign-string configurations sn∈Sns_n \in \mathcal{S}_n with common mesh thickness constants and Kn,sn→KK_{n,s_n} \to K in Hausdorff distance, log⁡En,snetXn,sn(f)→t2σ22⟨f,GQ∖Kf⟩\log \mathbb{E}_{n,s_n} e^{tX_{n,s_n}(f)} \to \frac{t^2\sigma^2}{2}\langle f, G_{Q\setminus K} f\rangle for tt near zero, and sup⁡s∈Snsup⁡f,g∈C∣Covn,s(Xn,s(f),Xn,s(g))−σ2⟨f,GVn,sg⟩∣→0\sup_{s\in\mathcal{S}_n} \sup_{f,g\in\mathcal{C}} |\mathrm{Cov}_{n,s}(X_{n,s}(f), X_{n,s}(g)) - \sigma^2\langle f, G_{V_{n,s}} g\rangle| \to 0.

Given a deterministic prescription fnf_n vanishing at grounded sites with ∣fx∣≤M|f_x| \le M and ∣fx−fy∣≤L/n|f_x - f_y| \le L/n, for any event EE with P(E)≥p0>0\mathbb{P}(E) \ge p_0 > 0, there is a measurable subevent E′⊂EE' \subset E with P(E′)≥p0/2\mathbb{P}(E') \ge p_0/2 on which the threshold-averaged shift T(h)=h+u(h)T(h) = h + u(h) is injective, has feasible image, satisfies sup⁡h∈E′∥T(h)−h−fn∥∞=o(1)\sup_{h\in E'}\|T(h) - h - f_n\|_\infty = o(1), and P(T(E′))≥c>0\mathbb{P}(T(E')) \ge c > 0.

For straight-attachment layouts of at most mm compatible (θ,R)(\theta, R)-barriers, the conditional probability that all continuations of the observed strands avoid all barriers is at least p>0p > 0 uniformly; furthermore, the harmonic-landing hookup, separation, survival, and nonreturn estimates (Lemmas 9.8, 9.9, 9.14, Proposition 9.15) hold uniformly.

Along every mesh subsequence, there is a further subsequence and a deterministic l∈(0,1/2]l \in (0, 1/2] such that all terminal conditional mean profiles in any prescribed finite collection of sign-compatible Dobrushin experiments have Brownian boundary trace ±l\pm l on positive/negative strand sides; the stopped harmonic observable satisfies sup⁡TE∣MTn(f)−∫f(z)Hn,Tl(z) dz∣→0\sup_T \mathbb{E}|M_T^n(f) - \int f(z) H_{n,T}^l(z)\,dz| \to 0; and every subsequential height gap satisfies l2=πσ2/8l^2 = \pi\sigma^2/8.

Methods

Triangular block filling-volume deficit via slice log-concavity (Prékopa-Leindler) and sparse-hole extension to prove macroscopic exponential moments for uniform Lipschitz heights.

Source excerpts

“Lemma 3.2 (Triangular block deficit). For these fixed parameters there is c∗ > 0 such that, if Q is convex-good for h, s = h|∂Q, χ : ∂Q → R, and both KQ(s + χ) and KQ(s − χ) have positive volume, then every two face centers v, w satisfy gQ := FQ(s) − 1/2(FQ(s + χ) + FQ(s − χ)) ≥ c∗|χ(v) − χ(w)|2. (3.5)” · Show in paper

Stationary synchronous reflected couplings on convex height polytopes with averaging-projection dissipation, uniform forcing on facet paths, and cutoff absorption of weighted square norms.

Source excerpts

“Lemma 4.4 (Averaging structure and cluster displacement). On a zero-drift step, normal reflection of a difference acts on each transmission cluster by compositions of averaging projections and their convex mixtures.” · Show in paper

“Lemma 4.5 (Facet paths and uniform forcing). ... For every central edge there are at most N + 1 genuine facet constraints joining its endpoints, or grounding both endpoints at genuine pins or walls.” · Show in paper

Stationary bulk cell problem and tangent-flux homogenization using volume-sampled tangent cocycles, zero-slope uniqueness, and small smooth exponential tilts to identify the GFF limit.

Source excerpts

“Proposition 5.8 (The cell flux and weak equation). Under the preceding inputs, j(q) is well defined on the vector space of attained slopes and is linear there. If any nonzero slope is attained, every slope in R2 is attained and j(q) = Aq for a finite scalar A.” · Show in paper

Quantitative averaged traces on 1/21/2-Frostman testing measures and Poisson uniqueness on varying thick planar continua to establish conditional Dirichlet covariance.

Source excerpts

“Theorem 6.8 (Poisson uniqueness from Frostman traces). Let K ⊂ Q contain ∂Q and have continuum thickness below r0. Let 3/2 < p < 2, A > 0, and b ∈ L∞(Q). Suppose u ∈ W1,p(Q∗) has zero 1/2-Frostman trace on K and −A∆u = b in V = Q \ K in the sense of distributions. Then, on Q, u = A−1GV b, (6.8)” · Show in paper

Threshold-averaged component shifts and a tangential-vector-field determinant identity on polytopes to control volume distortion under constraint-preserving perturbations.

Source excerpts

“Lemma 8.1 (Tangential vector fields). Let P ⊂ RN be a compact full-dimensional polytope, with normalized Lebesgue measure. Suppose that u : P → RN is Lipschitz and tangent to each facet of P. Then E div u = 0, E(div u)2 = E tr((Du)2) ≤ E∥Du∥2_HS. (8.3)” · Show in paper

Straight-attachment and circular-clearance barrier constructions, harmonic-landing records, and stopped harmonic observables to prove uniform SLE4\mathrm{SLE}_4 curve convergence and identify the height gap.

Source excerpts

“Lemma 9.8 (Hookup probability). For 9R/8 < R′ < 5R, with the observer and exterior boundary outside B6R, there is a universal upper bound for all R ≥ R0, where R0 > 1 is fixed: P(Zσ | ΦR′, ΘR) ≤ C/ log R. (G1)” · Show in paper