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.

Latest version

Checks complete

Sync OpenAI Math manuscripts from 4eaee87548250b2682affda7d05d947c05554bae

Committed

All papers

Claims

Findings

A counterexample to Kaplansky's quasitrace conjecture and failure of tensor-product stable finiteness

There exist a separable unital complex C∗C^*-algebra AA, admitting a normalized 22-quasitrace, and positive contractions a,b∈Aa, b \in A such that every normalized 22-quasitrace τ\tau on AA satisfies τ(a+b)−τ(a)−τ(b)≥1144\tau(a+b) - \tau(a) - \tau(b) \ge \frac{1}{144}.

There is a separable unital complex C∗C^*-algebra B=C∗(1,x1,…,x24)B = C^*(1, x_1, \dots, x_{24}) such that ∑i=124xi∗xi=1\sum_{i=1}^{24} x_i^* x_i = 1, ∑i=124xixi∗≤231\sum_{i=1}^{24} x_i x_i^* \le \frac{2}{3} 1, and Mm(B)M_m(B) is not properly infinite for every integer m≥1m \ge 1.

For every pair of integers m,R≥1m, R \ge 1, with d0=64m4Rd_0 = 64m 4^R and dj=4jd0d_j = 4^j d_0 (0≤j≤R0 \le j \le R), there are a compact smooth manifold YY, Hermitian complex vector bundles Ej→YE_j \to Y for all j≥0j \ge 0, and continuous bundle maps Xi(j):Ej→Ej+1X_i^{(j)} : E_j \to E_{j+1} (1≤i≤24,j≥01 \le i \le 24, j \ge 0) such that E0E_0 is trivial of rank d0d_0, rank(Ej)=dj\mathrm{rank}(E_j) = d_j for 0≤j≤R0 \le j \le R, ∑i=124(Xi(j))∗Xi(j)=IEj\sum_{i=1}^{24} (X_i^{(j)})^* X_i^{(j)} = I_{E_j}, ∑i=124Xi(j)(Xi(j))∗≤23IEj+1\sum_{i=1}^{24} X_i^{(j)} (X_i^{(j)})^* \le \frac{2}{3} I_{E_{j+1}}, and (⨁j=0REj)⊕m(\bigoplus_{j=0}^R E_j)^{\oplus m} contains no trivial subbundle of rank 2md02md_0.

Let E,FE, F be smooth Hermitian complex vector bundles of ranks d,rd, r over a compact smooth manifold YY of real dimension 2h2h, and let q≥1q \ge 1. For any finite set A\mathcal{A} of integer pairs (s,t)(s, t) with 1≤s≤d1 \le s \le d, 0≤t≤r0 \le t \le r, and (qs−t)(r−t)−s(d−s)>h(qs - t)(r - t) - s(d - s) > h, there exist smooth bundle maps T1,…,Tq:E→FT_1, \dots, T_q : E \to F such that dim⁡C∑i=1qTi(y)S>t\dim_{\mathbb{C}} \sum_{i=1}^q T_i(y) S > t for every y∈Yy \in Y, (s,t)∈A(s, t) \in \mathcal{A}, and complex ss-plane S⊂EyS \subset E_y.

Let E,FE, F be Hermitian complex vector bundles of positive ranks over a compact manifold YY. If continuous bundle maps T1,…,Tn:E→FT_1, \dots, T_n : E \to F satisfy dim⁡C∑i=1nTi(y)S≥2dim⁡CS\dim_{\mathbb{C}} \sum_{i=1}^n T_i(y) S \ge 2 \dim_{\mathbb{C}} S for all y∈Yy \in Y and S⊂EyS \subset E_y, then there are continuous bundle maps X1,…,Xn:E→FX_1, \dots, X_n : E \to F with ∑i=1nXi∗Xi=IE\sum_{i=1}^n X_i^* X_i = I_E and ∑i=1nXiXi∗≤23IF\sum_{i=1}^n X_i X_i^* \le \frac{2}{3} I_F.

Let BB be a unital complex C∗C^*-algebra, let N≥1N \ge 1, and suppose x1,…,xN∈Bx_1, \dots, x_N \in B satisfy ∑i=1Nxi∗xi=1\sum_{i=1}^N x_i^* x_i = 1 and ∑i=1Nxixi∗≤c1\sum_{i=1}^N x_i x_i^* \le c 1 for some 0≤c<10 \le c < 1. Then there are positive contractions a,b∈M2N(B)a, b \in M_{2N}(B) such that every normalized 11-quasitrace τ\tau on M2N(B)M_{2N}(B) satisfies τ(a+b)−τ(a)−τ(b)≥1−c2N\tau(a+b) - \tau(a) - \tau(b) \ge \frac{1-c}{2N}.

There are two unital simple stably finite complex C∗C^*-algebras whose minimal tensor product is properly infinite, and one factor can be taken to be Cr∗(F2)C^*_r(\mathbb{F}_2).

There is a separable unital complex C∗C^*-algebra BB admitting a normalized 22-quasitrace such that B⊗min⁡Cr∗(F2)B \otimes_{\min} C^*_r(\mathbb{F}_2) is properly infinite and admits no normalized 22-quasitrace.

There is a separable unital stably finite complex C∗C^*-algebra with no tracial state.

Methods

Constructing a separable unital C∗C^*-algebra as a finitely generated subalgebra of a bounded C∗C^*-product of forward shifts between finite-rank vector bundles indexed by matrix size and polynomial degree (m,R)(m, R), where Chern-class obstructions rule out polynomial approximations to orthogonal isometries.

Source excerpts

“The construction below uses a bounded product and a generated subalgebra, with a separate test for every matrix size and polynomial degree. The fixed number of generators preserves the trace obstruction while all finite polynomial candidates are excluded.” · Show in paper

“Let D be the bounded C∗-product of the algebras B(Hm,R,y) over all these indices: its elements are the operator families of finite supremum norm. The uniform bound therefore defines a single tuple in D, xi = (S_im,Ri^{m,R}(y))_{m,R≥1, y∈Ym,R} (1 ≤ i ≤ 24).” · Show in paper

Continuous operator scaling of expanding bundle maps via fiberwise minimization of the determinant potential gy(P)=log⁡det⁡(IFy+Dy(P))−32log⁡det⁡Pg_y(P) = \log \det(I_{F_y} + D_y(P)) - \frac{3}{2} \log \det P over positive definite endomorphisms.

Source excerpts

“In each fiber define Dy(P) = ∑_{i=1}^n Ti(y)P Ti(y)^*, gy(P) = log det(IFy + Dy(P)) − 3/2 log det P for positive definite P ∈ End(Ey). Both determinants are intrinsic. We will prove that gy has a unique minimizer depending continuously on y.” · Show in paper

Dimension-counting on Grassmannian bundles combined with Sard's theorem to construct smooth bundle maps whose collective images avoid small-dimensional subspaces.

Source excerpts

“The image of the smooth projection Is,t → V consists exactly of tuples having some s-plane whose collective image lies in a t-plane. A smooth image of a manifold of smaller dimension is a Lebesgue null set, as in the lower-dimensional case of Sard’s theorem [11].” · Show in paper

Telescoping diagonal matrix construction in M2N(B)M_{2N}(B) from partial sums of xi∗xix_i^* x_i and xixi∗x_i x_i^* to witness a uniform additivity defect for every normalized 11-quasitrace using only abelian linearity.

Source excerpts

“We use these partial sums to define two diagonal matrices. In their additivity defect, the first N blocks will telescope the column partial sums, while the last N use complements to telescope the row partial sums with the opposite sign and keep the entries positive.” · Show in paper