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

SLE3 universality for weak finite-range Ising interactions

For every J>0J > 0 and every finite-range, square-symmetric real potential UU, there is λ0=λ0(J,U)>0\lambda_0 = \lambda_0(J, U) > 0 such that for every ∣λ∣<λ0|\lambda| < \lambda_0 there is a positive inverse temperature βc=βc(J,U,λ)\beta_c = \beta_c(J, U, \lambda), independent of domains and boundary data, with βc(J,U,0)=β0:=log⁡(1+2)2J\beta_c(J, U, 0) = \beta_0 := \frac{\log(1 + \sqrt{2})}{2J}, at which the rescaled spin interface δnγ(Pn)\delta_n \gamma(P_n) in any uniformly approximated bounded Jordan domain DD with arbitrary admissible exterior contours Po,nP_{o,n} converges in distribution, in the oriented curve metric dcurvd_{\mathrm{curv}}, to chordal SLE3\mathrm{SLE}_3 in DD from aa to bb.

There is a constant C<∞C < \infty such that for each a>0a > 0, all sufficiently large fixed integers LL admit an integer s0s_0, linear functionals ℓs:Xs→R\ell_s : \mathcal{X}_s \to \mathbb{R}, and vectors es∈Xse_s \in \mathcal{X}_s for every s≥s0s \ge s_0 satisfying ∥ℓs∥≤C\|\ell_s\| \le C, ∥es∥q≤C\|e_s\|_q \le C, ℓs(es)=1\ell_s(e_s) = 1, ∥Asf∥q≤CL∣ℓs(f)∣+a∥f∥q\|A_s f\|_q \le C L |\ell_s(f)| + a \|f\|_q, and ∣ℓLs(Asf)−Lℓs(f)∣≤a∥f∥q|\ell_{Ls}(A_s f) - L \ell_s(f)| \le a \|f\|_q simultaneously for all s≥s0s \ge s_0 and all bounded even square-invariant functions f∈Xsf \in \mathcal{X}_s.

For every J>0J > 0 and finite-range square-symmetric real potential UU, for each sufficiently small ϵ>0\epsilon > 0 there is λ∗>0\lambda_* > 0 and a real analytic function βc:(−λ∗,λ∗)→(0,∞)\beta_c : (-\lambda_*, \lambda_*) \to (0, \infty) with βc(0)=β0\beta_c(0) = \beta_0, independent of domains and boundary data, for which the initial potential V0(β,λ)V_0(\beta, \lambda) has a scale-flow trajectory Vk+1=Φsk(Vk)V_{k+1} = \Phi_{s_k}(V_k) satisfying ∥Vk∥sk,d≤ϵ2−k\|V_k\|_{s_k, d} \le \epsilon 2^{-k} for all k≥0k \ge 0.

At the temperature βc(λ)\beta_c(\lambda) of Theorem 4.5, for all sufficiently large NN there is a coupling of σ(0)∼μ0,N\sigma^{(0)} \sim \mu_{0,N}, σ(λ)∼μλ,N\sigma^{(\lambda)} \sim \mu_{\lambda,N}, and mutually independent Bernoulli marks Mk,xM_{k,x} (independent of σ(0)\sigma^{(0)}) with P(Mk,x=1)≤Cϵ2−ksk−2\mathbb{P}(M_{k,x} = 1) \le C \epsilon 2^{-k} s_k^{-2} such that, except on an event of probability o(1)o(1) as N→∞N \to \infty, σ(λ)\sigma^{(\lambda)} is obtained from σ(0)\sigma^{(0)} by changing spins only inside the union of the marked squares BLsk(x)B_{L s_k}(x).

There is ϵ∗>0\epsilon_* > 0, independent of any finite family T\mathcal{T} of polygonal corridor crossings, polygonal-band circuits, and four-passage square annulus tests, such that if ϵ≤ϵ∗\epsilon \le \epsilon_*, then P(DT(σ0,M)=1 for some T∈T)→0\mathbb{P}(D_T(\sigma^0, M) = 1 \text{ for some } T \in \mathcal{T}) \to 0 as δ↓0\delta \downarrow 0; consequently, the pure and perturbed torus laws at βc(λ)\beta_c(\lambda) have asymptotically identical joint laws for every such finite family.

For sufficiently small ∣λ∣|\lambda|, the perturbed spin weight can be represented exactly as the spin marginal of ∏vWv(σv,bB(v))\prod_v W_v(\sigma_v, b_{\mathcal{B}(v)}) using finitely many private auxiliary bond bits, with Wv→Wv0W_v \to W_v^0 uniformly as λ→0\lambda \to 0; moreover, under strict boundary-incidence changes (free to pinned plus, pinned minus to free, or pinned minus to pinned plus), there is a monotone coupling on all bits touching the common free components and all physical spins in them.

For a fixed square Q⊂int⁡Q~Q \subset \operatorname{int} \widetilde{Q} in a physical torus chart, if every vertex in Q~\widetilde{Q} is free in a domain spin law νΩ,λτ\nu_{\Omega, \lambda}^{\tau}, then for all sufficiently small mesh there is a constant C<∞C < \infty independent of Ω,τ\Omega, \tau, and the mesh such that C−1≤νΩ,λτ(σQ∩δZ2=η)μλ,N(σQ∩δZ2=η)≤CC^{-1} \le \frac{\nu_{\Omega, \lambda}^{\tau}(\sigma_{Q \cap \delta\mathbb{Z}^2} = \eta)}{\mu_{\lambda, N}(\sigma_{Q \cap \delta\mathbb{Z}^2} = \eta)} \le C for every spin configuration η\eta.

For a fixed finite list of allowed plus strips S1,…,SmS_1, \dots, S_m in a displaced pure comparison domain AhA_h favoring minus, the vector VA∈{0,1}mV_A \in \{0, 1\}^m of plus crossings in AhA_h and the vector Vn+V_n^+ of plus crossings in the target domain (with portions outside DnD_n virtually filled with plus) can be coupled so that P(VA≤Vn+)=1−o(1)\mathbb{P}(V_A \le V_n^+) = 1 - o(1) as n→∞n \to \infty, uniformly over all admissible target exterior configurations.

If random simple curves gng_n converge in probability in Hausdorff distance and at their endpoints to a random simple chord Γ\Gamma, and the interior four-passage probability Bn(F;r,R)\mathcal{B}_n(F; r, R) vanishes as n→∞n \to \infty followed by r↓0r \downarrow 0 for every deterministic compact interior F⋐DF \Subset D, then dcurv(gn,Γ)→0d_{\mathrm{curv}}(g_n, \Gamma) \to 0 in probability.

Methods

Exact spin formulation of the contour-potential Ising model via disagreement indicators de(σ)=(1−σvσw)/2d_e(\sigma) = (1 - \sigma_v \sigma_w)/2 and fixed exterior spins τ\tau on Z2∖Ω\mathbb{Z}^2 \setminus \Omega, preserving all interactions across the domain boundary.

Source excerpts

“Lemma 1.2. For the boundary data above, every even completion P ∪ Po determines a unique spin configuration that is plus sufficiently far away and has disagreement set P ∪ Po. Its restriction τ to Z2 \ Ω is independent of P. The map P ↦ σ|Ω is a bijection between PξΩ and all spin assignments in Ω. Under this bijection, (1.1) is precisely the Gibbs conditional law for (1.3) with exterior spins τ .” · Show in paper

Gaussian observation smoothing Yx=Ms−7/8σx+ZxY_x = M s^{-7/8} \sigma_x + Z_x and coarse cell-projection to reduce arbitrary bounded spin functions to bounded continuous functions of finitely many smeared spin sums across a gap.

Source excerpts

“Conditional on the spins, in the selected core C observe independent Gaussian variables Yx = Ms−7/8σx + Zx, x ∈ C, (3.8) where the Zx are independent standard normal variables, independent of the spins.” · Show in paper

“Lemma 3.4 (Bounded replacement across a gap). Fix finitely many pairs of positively separated supports, each a square or a square band, with buffered enlargements that remain disjoint. For every t > 0, each lattice function v of norm at most one on any one support can be replaced, for its full-plane correlations against all functions of norm at most one on the paired supports, with error at most t, by a function of finitely many smeared spin sums in its enlargement.” · Show in paper

Osterwalder-Schrader radial reflection-positive Hilbert space construction combined with a complex determinant identity (A0dA0=4A22A_0^d A_0 = 4 A_2^2) to establish a spectral gap below degree 33 for even quarter-turn invariant spin tests.

Source excerpts

“We use the reflection-positive Hilbert-space construction of Osterwalder and Schrader [20, Section 4.1] in its radial form, verifying contraction and continuity below.” · Show in paper

“Then Ad0A0 = 4A22. (3.27)” · Show in paper

Exact scale flow of labeled potentials on connected covering lists, reversed by independent-rate local refreshes via the finite-state reverse generator identity, and coupled to the pure critical torus law via independent marked squares.

Source excerpts

“Lemma 4.1 (Reverse generator identity). Let µt be a differentiable family of strictly positive probability laws on a finite spin space, and let Ix be specified update sets.” · Show in paper

“During one step we hold the initial single-square function f fixed and solve, modulo constants, d/dt V (t) = −f + P V (t)Ls f, V (0) = V, 0 ≤ t ≤ 1. (4.8)” · Show in paper

Three-regime pivotal-square passage analysis (four bulk passages, two half-plane passages, one vertex passage) combined with a finite-size bootstrap over increasing mark radii to prove bulk stability of polygonal crossing, circuit, and four-passage tests.

Source excerpts

“Lemma 5.2 (Passages forced by a pivotal square). Let T be a fixed polygonal crossing, circuit, or one of the three annular tests in (5.1).” · Show in paper

“Proposition 5.8 (Uniform bootstrap of the three checks).” · Show in paper

Private auxiliary bond-bit representation of local interaction factors, yielding strict MTP2\mathrm{MTP}_2 and strict boundary-incidence comparisons without requiring boundary monotonicity of pinned physical spins.

Source excerpts

“For each complete spin pattern t ∈ {−1, +1}V (T), reserve two private bits on every edge of T, with the same coupling a.” · Show in paper

“Proposition 6.3 (Comparison at strict incidences).” · Show in paper

Displaced pure SLE3\mathrm{SLE}_3 comparison domains with finite deterministic lists of colored strip barriers to identify the Hausdorff trace limit, paired with progress coordinates and interior four-passage exclusion to upgrade trace convergence to dcurvd_{\mathrm{curv}} convergence.

Source excerpts

“We first construct a domain favoring minus, from which plus crossings can be transferred to the target. Move its plus boundary outward and its minus boundary inward.” · Show in paper

“Lemma 7.4 (Progress coordinates and the full curve metric).” · Show in paper