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

Cuntz comparison and Jiang–Su absorption

Let AA be a nonzero separable nuclear C∗C^*-algebra satisfying condition (C) (almost unperforation and full almost divisibility of Cu(A)\mathrm{Cu}(A)). For every fixed free ultrafilter ω\omega on N\mathbb{N}, there is a unital ∗*-homomorphism Z→Fω(A)\mathcal{Z} \to F_\omega(A), and consequently A≅A⊗min⁡ZA \cong A \otimes_{\min} \mathcal{Z}.

Let AA be a separable simple nuclear non-elementary C∗C^*-algebra. If AA has strict comparison in the extended-functional sense of Equation (1.1), then A≅A⊗min⁡ZA \cong A \otimes_{\min} \mathcal{Z}.

Let AA be a simple exact non-elementary C∗C^*-algebra with strict comparison as in Equation (1.1). For every x∈Cu(A)x \in \mathrm{Cu}(A) and integer N≥1N \geq 1, there exists u∈Cu(A)u \in \mathrm{Cu}(A) such that Nu≤x≤(N+1)uNu \leq x \leq (N+1)u.

Let AA be a separable nuclear C∗C^*-algebra, and let ι:A→A⊗min⁡Z\iota : A \to A \otimes_{\min} \mathcal{Z} be ι(a)=a⊗1Z\iota(a) = a \otimes 1_{\mathcal{Z}}. If Cu(ι):Cu(A)→Cu(A⊗min⁡Z)\mathrm{Cu}(\iota) : \mathrm{Cu}(A) \to \mathrm{Cu}(A \otimes_{\min} \mathcal{Z}) is an isomorphism, then A≅A⊗min⁡ZA \cong A \otimes_{\min} \mathcal{Z}.

Let AA be a separable nuclear C∗C^*-algebra. Then A≅A⊗min⁡Z  ⟺  Cu(A)≅Cu(A⊗min⁡Z)A \cong A \otimes_{\min} \mathcal{Z} \iff \mathrm{Cu}(A) \cong \mathrm{Cu}(A \otimes_{\min} \mathcal{Z}) in the category Cu\mathbf{Cu}.

Let AA be a unital simple infinite-dimensional C∗C^*-algebra with nonempty tracial state space T(A)T(A). For every p∈A+p \in A_+ and η>0\eta > 0, there are a finite integer LL and w∈ML(A)+w \in M_L(A)_+ such that 0≤τ(p)−dτ(w)≤η0 \leq \tau(p) - d_\tau(w) \leq \eta for all τ∈T(A)\tau \in T(A), where matrix traces are unnormalized.

Methods

Trace transfer and integer spectral rounding: transferring a positive function of an element vanishing near zero into a small hereditary support without changing any extended trace value, then approximating the trace by integer spectral cuts with error bounded by one support rank.

Source excerpts

“We approximate trace values by ranks in two steps. First, a positive function of an element that vanishes near zero is transferred into a small hereditary support without changing any trace value. Integer spectral cuts then approximate that value with an error bounded by the rank of the support.” · Show in paper

Two-support shrinking and pure-state word models: separating hereditary supports of pure-state word models via two-support shrinking while retaining all irreducible representations detecting a prescribed positive element, and tapering the row map to build norm-visible central matrix cones.

Source excerpts

“We will shrink two hereditary supports so that a prescribed cross-compression decreases by a fixed factor, while at least one new support remains nonzero in each irreducible representation where an original element has norm at least 1/2. Repeated use of this operation will separate the supports in the norm-cone construction of Section 8.” · Show in paper

“The construction in this section detects a prescribed positive element uniformly in the irreducible representations at each coordinate. Its building blocks are finite word spaces at pure states.” · Show in paper

Bounded-trace matrix cones via convex order-zero nuclear approximations: combining small matrix cones in the commutants of nuclear approximation maps using two support compressions and Lipschitz calculus in tracial 2-norm so that estimates are independent of the number of approximation summands.

Source excerpts

“The matrix cones in this section are approximately central in tracial 2-norm. The construction uses a convex nuclear approximation, and its estimates are independent of the number of maps in that approximation.” · Show in paper

Hereditary localization and relative trace ideal removal: joining matrix cones and removing a relative trace ideal JbJ_b specified by hereditary rank tests to turn tracial centrality into norm centrality while retaining relative trace mass.

Source excerpts

“Our goal is to retain the tracial mass from the previous section while obtaining norm-central maps. The trace-kernel ideal and relative commutant lifting methods have predecessors in Kirchberg–Rørdam [21, Section 4]. Here the relative ideal is specified by hereditary rank tests, including the traces which vanish on the controlling positive element.” · Show in paper

Adaptive rank-gap comparison and simultaneous domination: fixing norm-visible ideal multiplicities before choosing tracial cones with reciprocal relative error, bounding comparison witnesses independently of matrix size to obtain infinitely many central slots, and packing them via simultaneous domination for weighted inclusions.

Source excerpts

“Norm visibility supplies finite coordinate ideal multiplicities Ln, which need not be bounded. We fix the norm maps and these integers first, and then choose a tracial cone with relative defect error ηn/Ln, where ηn → 0.” · Show in paper

“We prove the weighted-inclusion case used here, so that the passage from ideal containment to one common finite list of compression coefficients is explicit.” · Show in paper