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

Radius of comparison equals half the mean dimension

For every minimal homeomorphism hh of an infinite compact metrizable space XX, the radius of comparison of its crossed product satisfies rc⁡(C(X)⋊hZ)=12mdim⁡(X,h)\operatorname{rc}(C(X) \rtimes_h \mathbb{Z}) = \frac{1}{2}\operatorname{mdim}(X, h) in [0,∞][0, \infty].

For a minimal homeomorphism hh of an infinite compact metrizable space XX and A=C(X)⋊hZA = C(X) \rtimes_h \mathbb{Z}, the following are equivalent: (i) mdim⁡(X,h)=0\operatorname{mdim}(X, h) = 0; (ii) (X,h)(X, h) has the small boundary property; (iii) A≅A⊗ZA \cong A \otimes \mathcal{Z}; (iv) dim⁡nuc(A)<∞\dim_{\mathrm{nuc}}(A) < \infty; (v) dim⁡nuc(A)≤1\dim_{\mathrm{nuc}}(A) \le 1. In particular, if mdim⁡(X,h)∈(0,∞]\operatorname{mdim}(X, h) \in (0, \infty], then A≇A⊗ZA \not\cong A \otimes \mathcal{Z} and dim⁡nuc(A)=∞\dim_{\mathrm{nuc}}(A) = \infty.

For a minimal homeomorphism hh of an infinite compact metrizable space XX, A=C(X)⋊hZA = C(X) \rtimes_h \mathbb{Z}, every finite comparison bound rr for AA, and every finite open cover U\mathcal{U} of XX, lim⁡k→∞D(U[0,k))/k≤2r\lim_{k\to\infty} D(\mathcal{U}^{[0,k)})/k \le 2r.

Given integers k,H0k, H_0 with H0≥2H_0 \ge 2, e=kH0e = kH_0, b0>0b_0 > 0, γ>0\gamma > 0, and r≥0r \ge 0 with γ/k>r\gamma/k > r, there exist an open set B⊂XB \subset X, pairwise disjoint open sets O0,…,Ok−1⊂XO_0, \dots, O_{k-1} \subset X, a continuous function w:X→[0,1]w: X \to [0,1], a compact set K⊂BK \subset B, and J0≥1J_0 \ge 1 such that B,hB,…,hk−1BB, hB, \dots, h^{k-1}B are pairwise disjoint with γμ(B)>r\gamma\mu(B) > r for all μ∈Mh(X)\mu \in \mathcal{M}_h(X), hi+DkOi⊂Bh^{i+Dk}O_i \subset B, hi+Dk(supp⁡(w)∩Oi)⊂Kh^{i+Dk}(\operatorname{supp}(w) \cap O_i) \subset K, and w(hv+qex)<1w(h^{v+qe}x) < 1 occurs for at most b0Jb_0 J indices 0≤q<J0 \le q < J for all J≥J0J \ge J_0.

Under the comparison and parameter conditions of Proposition 3.4, for all sufficiently large JJ, there is a continuous map fJ:X→[−1,1]tnf_J: X \to [-1,1]^{tn} agreeing with Fj(hqe+i+Dkx)F_j(h^{qe+i+Dk}x) on a neighborhood of h−qe(supp⁡(w)∩Oi)h^{-qe}(\operatorname{supp}(w) \cap O_i) such that over the compact locus LJ\mathcal{L}_J, the direct sum of nn pullbacks of the rank-pp torus bundle HH embeds continuously in a trivial complex bundle of rank strictly less than 2pn2pn.

For a finite simplicial complex SS of dimension at most mm with vertices colored by 1,…,m+11, \dots, m+1 (distinct on each simplex) and t≥1t \ge 1, g=⌊(m+1)/(t+1)⌋g = \lfloor (m+1)/(t+1) \rfloor, there exist continuous maps Fj:S→Ct\mathcal{F}_j: S \to C_t (1≤j≤g1 \le j \le g) such that any continuous y:Y→Sy: Y \to S and continuous replacements vj:Y→Ctv_j: Y \to C_t agreeing with Fj∘y\mathcal{F}_j \circ y on ∂Ct\partial C_t admit a continuous reconstruction y~:Y→S\widetilde{y}: Y \to S preserving positive vertex coordinates and reducing the carrier dimension of y~(z)\widetilde{y}(z) to at most m−s(z)m - s(z), where s(z)=#{j:vj(z)∈∂Ct}s(z) = \#\{j : v_j(z) \in \partial C_t\}.

Methods

Uniform block allocation using Niu's open tower theorem, continuous cutoffs on invariant probability measures, and weak-convergence orbit estimates.

Source excerpts

“Apply Theorem 3.1 with height L and error η, and denote its base by V . Set B = ⋃_{a=0}^{L/k-1} hakh^{ak} V.” · Show in paper

“For completeness, the uniform orbit estimate for the closed set C is lim sup_{m→∞} sup_{x∈X} (1/m) ∑_{s=0}^{m-1} 1_C(h^s x) ≤ sup_{μ∈M_h(X)} μ(C).” · Show in paper

Fixed linear map for changing coordinate supports, avoiding discontinuous choices of projections across varying support patterns.

Source excerpts

“There is a complex linear map L : ⨁_{s∈E} C^b → CbqC^{bq} whose restriction to every coordinate subspace supported on at most q indices is injective.” · Show in paper

Encoding torus bundles into positive matrix functions in the crossed product and extracting a finite-propagation Cuntz comparison witness.

Source excerpts

“Define positive matrix functions P0 ∈ M2pgM_{2pg}(C(X))_+ and P1 ∈ M_R(C(X))_+ by P0(x) = ρ0(x) ⨁_{j=1}^g ψ(Fj(x)) Q̃(Fj(x)), P1(x) = ρ1(x) 1_R.” · Show in paper

“Taking the appropriate rectangular corner of a Cuntz witness for Equation (3.13), and then approximating its entries by finite Fourier sums, gives T0 ∈ MR,2pgM_{R,2pg}(A) such that ∥T0^* P1 T0 − P0∥ < 1/2, T0 = ∑_{ℓ=-M}^M Aℓ u^ℓ” · Show in paper

Colored subordinate nerve construction via barycentric subdivision and exact packet reconstruction.

Source excerpts

“Take S to be the barycentric subdivision of S0, and regard the same map as a map to S under the canonical identification of their realizations.” · Show in paper

“Partition the first (t + 1)g colors into packets Ij = {(j − 1)(t + 1) + 1, . . . , j(t + 1)}, 1 ≤ j ≤ g, and retain the remaining colors individually.” · Show in paper

Aligned and fallback join polyhedron construction to retain orbit-cover information while bounding the subcomplex dimension.

Source excerpts

“In the join, give the aligned branch weight w(h^{Tb,qT_{b,q}} x) and the fallback branch weight 1 − w(h^{Tb,qT_{b,q}} x).” · Show in paper