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

Tracial projection methods and uniform property Gamma

For a unital, separable, simple, infinite-dimensional, nuclear, stably finite C*-algebra A with nonempty tracial-state space T(A), uniform tracial completion B = \overline{A}^{T(A)}, and a fixed free ultrafilter \omega on \mathbb{N}, if B^\omega has real rank zero in its quotient C*-norm, then there is a projection p \in B^\omega \cap B' such that \lambda(px) = \frac{1}{2}\lambda(x) for all x \in B and \lambda \in \Lambda_\omega; in particular, A has uniform property \Gamma, answering Problem XXI of Schafhauser, Tikuisis and White affirmatively.

Under the hypotheses of Theorem 1.1, including real rank zero of B^\omega in its quotient C*-norm, every tracial state of B is the designated extension of a unique trace in T(A), restriction is an affine homeomorphism T(B) \to T(A) with inverse \tau \mapsto \bar{\tau}, and |\sigma(b)| \leq \|b\|_{2,T(A)} for all b \in B and \sigma \in T(B).

Let A \subseteq D be a unital inclusion of C*-algebras with A nuclear, D of real rank zero, and T(D) \neq \emptyset. If for every \delta > 0 there is a full projection g \in D with \sup_{\tau \in T(D)} \tau(g) < \delta, then for every finite K \subseteq \mathcal{U}(A) and \varepsilon > 0 there is a projection P \in D such that \sup_{\tau \in T(D)} \|[P, x]\|_{2,\tau} < \varepsilon and \sup_{\tau \in T(D)} |\tau(Px) - \frac{1}{2}\tau(x)| < \varepsilon for every x \in K.

Let A be a simple, separable, unital, infinite-dimensional nuclear C*-algebra with strict comparison. Then A \cong A \otimes \mathcal{Z}, resolving the unital Toms–Winter conjecture. When T(A) \neq \emptyset, the proof constructs uniform property \Gamma directly.

Let A be a nonzero simple, separable, nuclear C*-algebra such that A \otimes \mathcal{K} has no nonzero projections. Suppose that every trace in T_{\mathrm{lsc}}(A) is bounded on the positive unit ball and that T_1(A) is nonempty and weak-* compact. Assume that for a, b \in (A \otimes \mathcal{K})_+ with b \neq 0, d_\tau(a) < 1 for every \tau \in T_{\mathrm{lsc}}(A) with d_\tau(b) = 1 implies a \precsim b. Then A has uniform property \Gamma and A \cong A \otimes \mathcal{Z}.

Methods

Weighted matrix blocks and orthogonal packing: using Haagerup's convex virtual diagonal to choose nuclear averaging maps, compressing their outputs into weighted matrix blocks via scalar compressions, sampling subprojections inside each block, and packing them into an unoccupied corner via orthogonal insertion.

Source excerpts

“For the real-rank-zero route, Haagerup’s convex virtual diagonal [15, Theorem 3.1] provides nuclear averaging maps chosen before any input projection. Finite scalar compressions turn their outputs into weighted matrix blocks. A small proportion of a base projection can be sampled inside each block without making that subprojection central: scalar matrix compressions retain the coefficient traces and squared commutators. Orthogonal insertion then packs two labelled families with matching target traces.” · Show in paper

Rank approximation and projection supply with exact spectral support: transferring trace functions to a hereditary subalgebra of small support rank, taking integer spectral cuts, and using strict comparison to obtain projections in the uniform tracial ultrapower A^\omega with exact spectral support.

Source excerpts

“The construction transfers the trace function exactly to a hereditary subalgebra of small support rank, then takes integer spectral cuts. The numerical argument requires neither separability, nuclearity nor comparison. We will then add strict comparison to place these rank approximants in prescribed hereditary subalgebras and obtain projections in the uniform tracial ultrapower.” · Show in paper

Conditional central limit construction with tracial clipping: using centered symmetries in finite-dimensional commutants, Hirshberg–Kirchberg–White convex order-zero approximations, homomorphism replacement, and tracial clipping at a fixed radius to obtain commuting rows whose normalized sums converge in distribution to a standard Gaussian, yielding central projections via continuous approximations to equal-measure Gaussian intervals.

Source excerpts

“Scalar compressions and matrix-unit transport construct symmetries centered against any prescribed separable coefficient algebra. The convex order-zero approximations of Hirshberg, Kirchberg and White [16, Theorem 1.4] give a square-root-weighted sum of these symmetries with fourth moment at most three. Its norm may grow with the number of summands. Clipping at a fixed radius supplies bounded relative moment estimates; successively adjoining commuting variables gives normalized sums with a common Gaussian limit in coefficient traces. Continuous approximations to equal-measure Gaussian intervals then yield the required central projections.” · Show in paper

Cost-bounded projection selection and block-matrix truncation for stably projectionless algebras: selecting projections with prescribed traces and a one-sided cost bound controlling both overlap and model error, passing aggregate squared commutators to a truncated sum via a block-matrix functional-calculus estimate, and amplifying small splittings to uniform property \Gamma.

Source excerpts

“For the stably projectionless route, spectral allocation supplies small projections with a one-sided cost bound. A single cost records both overlap with earlier projections and error in a finite-dimensional model. A block-matrix functional-calculus estimate passes aggregate squared commutators to a truncated sum without a factor for the number of models. The resulting small splittings have trace and projection errors of order λ 3/2 . Reindexing and multiplication amplify them to exact splittings.” · Show in paper