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

A Modulus Proof of Cannon’s Conjecture

Every hyperbolic group GG with boundary ∂G\partial G homeomorphic to S2\mathbb{S}^2 admits a homomorphism ρ:G→Isom⁡(H3)\rho : G \to \operatorname{Isom}(\mathbb{H}^3) for which the action on H3\mathbb{H}^3 is proper and cocompact, with finite kernel.

Assuming the macroscopic moduli M(n)M(n) are unbounded, along a subsequence of record indices, the regularized functions HNH_N converge locally uniformly on U=int⁡ZQ^U = \operatorname{int}_Z \widehat{\mathcal{Q}} to a continuous nonconstant function u:U→[0,1]u : U \to [0, 1] (locally Hölder continuous if λ>1\lambda > 1) and the dual-weight measures μN\mu_N converge weakly to a finite measure μ\mu satisfying (osc⁡B‾(x,s)u)2≤Cμ(B‾(x,Cμs))(\operatorname{osc}_{\overline{B}(x,s)} u)^2 \le C \mu(\overline{B}(x, C_\mu s)), with separating level continua Ct⊂QC_t \subset \mathcal{Q} of diameter at least d1d_1.

The contradiction setup with unbounded macroscopic moduli M(n)M(n) cannot have exponential record growth λ>1\lambda > 1.

The macroscopic moduli M(n)M(n) cannot be unbounded while satisfying lim sup⁡n→∞M(n)1/n≤1\limsup_{n\to\infty} M(n)^{1/n} \le 1.

If GG is a torsion-free hyperbolic group with ∂G\partial G homeomorphic to S2\mathbb{S}^2, then GG is the fundamental group of a closed hyperbolic three-manifold (possibly nonorientable), has a finite-index subgroup isomorphic to π1(Σ)⋊Z\pi_1(\Sigma) \rtimes \mathbb{Z} for a closed connected orientable surface Σ\Sigma, has positive virtual first Betti number, and is virtually compact special and residually finite.

Methods

Discrete crossing duality on a coordinate rectangle Q⊂Q^\mathcal{Q} \subset \widehat{\mathcal{Q}} combined with two-step path-envelope regularization FN(x)=inf⁡π:x→topmax⁡πUN0F_N(x) = \inf_{\pi:x\to\text{top}} \max_\pi U_N^0 and HN(x)=sup⁡π:x→bottominf⁡πFNH_N(x) = \sup_{\pi:x\to\text{bottom}} \inf_\pi F_N to produce continuous functions with separating level continua.

Source excerpts

“Lemma 3.3 (Crossing dual weights). For each retained N there are nonnegative weights ρq = ρ^(N)_q, q ∈ TN, admissible for top-to-bottom crossings of Q, such that ∑_{q∈TN} ρ_q^2 ≤ C/M(N).” · Show in paper

“Define FN(x) = inf_{π:x→top} max_π U_N^0, HN(x) = sup_{π:x→bottom} inf_π FN, where paths remain in Q̂ and end in the indicated band.” · Show in paper

Finite-cell level comparison integrating the oscillation of a second function over a family of level continua of a first function to bound ∣I∣|I| by local oscillation energies.

Source excerpts

“Lemma 3.8 (A finite-cell level comparison). Let f, w be continuous on an open set V ⊂ Z, let K ⊂ V be compact, and let I ⊂ R be an interval. Suppose that for every t ∈ I there is a continuum Dt ⊂ K with f = t on Dt and osc_{Dt} w ≥ b0 > 0. For sufficiently large n, b0|I| ≤ ∑_{q∈Tn, B(q,2rn)∩K≠∅} osc_{B(q,2rn)} f osc_{B(q,2rn)} w.” · Show in paper

Hölder magnification of local variation via Cayley-graph expansions hi=vi−1h_i = v_i^{-1} at scales rkir_{k_i} to extract a nonconstant locally Hölder limit on a punctured sphere.

Source excerpts

“Lemma 4.1 (Hölder magnification). Let f be a real-valued, locally Hölder continuous function on an open set Ω ⊂ Z. Suppose O is open, Ō ⊂ Ω, and O contains a compact path on which f is nonconstant.” · Show in paper

A probability flow on a tree of separated boundary balls carrying nested value intervals with linear and quadratic transition bounds to force good-node mass in the subexponential growth case.

Source excerpts

“The tree will record both where variation occurs and which values it carries. A node has a source center x, a scale R, and a value interval J. Its children have scale r = θR, where θ = a−Da^{-D} for one positive integer subdivision depth D used throughout the tree.” · Show in paper

“Proposition 5.4 (Properties of the flow). The construction gives a fixed D and a tree with generation-n centers Hrk0+Dn-separated and descendant displacement (5.19).” · Show in paper