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

The complete Crouzeix theorem: optimal similarity and a common positive boundary representation

The complete Crouzeix conjecture holds with sharp constant 22 for every bounded operator AA on a complex Hilbert space H\mathcal{H} (with no separability assumption) and every matrix-valued polynomial, and W(A)‾\overline{W(A)} is a complete 22-spectral set extending to finite matrix-valued functions holomorphic near W(A)‾\overline{W(A)}.

For every admissible domain Ω\Omega, every finite n≥1n \ge 1, every A∈Mn(C)A \in M_n(\mathbb{C}) with W(A)⊂ΩW(A) \subset \Omega, and the interior conformal coordinate f:Ω→Df: \Omega \to \mathbb{D}, the metric optimization problem κ2=min⁡{τ∈R:I⪯H⪯τI,T∗HT⪯H for some H=H∗}\kappa^2 = \min\{\tau \in \mathbb{R} : I \preceq H \preceq \tau I, T^*HT \preceq H \text{ for some } H = H^*\} (with T=f(A)T = f(A)) is strictly feasible and attains a minimum 1≤κ2≤41 \le \kappa^2 \le 4, and a minimizing metric HH gives S=H1/2S = H^{1/2}, A′=SAS−1A' = SAS^{-1}, D=Sf(A)S−1=f(A′)D = Sf(A)S^{-1} = f(A') satisfying D∗D⪯ID^*D \preceq I, ρ(D)<1\rho(D) < 1, and ∥S∥∥S−1∥=κ≤2\|S\|\|S^{-1}\| = \kappa \le 2.

Under the hypotheses of Theorem 1.2, for any chosen minimizing metric HH there is one continuous function Λ:T→Mn(C)\Lambda : \mathbb{T} \to M_n(\mathbb{C}) satisfying Λ(t)⪰0\Lambda(t) \succeq 0 and ∫TΛ(t) dσ(t)=In\int_{\mathbb{T}} \Lambda(t)\,d\sigma(t) = I_n such that for every positive integer mm and every Mm(C)M_m(\mathbb{C})-valued function vv holomorphic on a neighborhood of Ω‾\overline{\Omega}, v[A′]=∫TΛ(t)⊗v(G(t)) dσ(t)v[A'] = \int_{\mathbb{T}} \Lambda(t) \otimes v(G(t))\,d\sigma(t) and ∥v[A]∥≤κmax⁡Ω‾∥v∥\|v[A]\| \le \kappa \max_{\overline{\Omega}} \|v\|, with the same SS, Λ\Lambda, and κ\kappa working for all mm and vv.

For every pair of positive integers n,mn, m, every A∈Mn(C)A \in M_n(\mathbb{C}), every nonnegative integer dd, and every choice of matrices B0,…,Bd∈Mm(C)B_0, \dots, B_d \in M_m(\mathbb{C}), the polynomial P(z)=∑j=0dBjzjP(z) = \sum_{j=0}^d B_j z^j satisfies ∥P[A]∥=∥∑j=0dAj⊗Bj∥≤2max⁡z∈W(A)∥P(z)∥\|P[A]\| = \|\sum_{j=0}^d A^j \otimes B_j\| \le 2 \max_{z \in W(A)} \|P(z)\|, and no smaller constant holds uniformly over these choices.

For a bounded operator A∈B(H)A \in \mathcal{B}(\mathcal{H}) on a nonzero complex Hilbert space H\mathcal{H} (with no separability assumption), K=W(A)‾K = \overline{W(A)} is compact, convex, contains spec⁡(A)\operatorname{spec}(A), and is a complete 22-spectral set for AA: ∥P[A]∥≤2sup⁡z∈W(A)∥P(z)∥=2max⁡z∈K∥P(z)∥\|P[A]\| \le 2 \sup_{z \in W(A)} \|P(z)\| = 2 \max_{z \in K} \|P(z)\| for all matrix polynomials PP, ∥F[A]∥≤2max⁡z∈K∥F(z)∥\|F[A]\| \le 2 \max_{z \in K} \|F(z)\| for all Mm(C)M_m(\mathbb{C})-valued functions FF holomorphic on an open neighborhood of KK, and the same estimate holds for all matrix-valued rational functions with poles outside KK, with uniform optimality of the constant 22.

For every admissible domain Ω\Omega, there is a conformal bijection f:Ω→Df : \Omega \to \mathbb{D} such that ff and f−1f^{-1} extend univalently through Ω‾\overline{\Omega} and D‾\overline{\mathbb{D}}, respectively, and a conformal bijection G:{∣t∣>1}→C∖Ω‾G : \{|t| > 1\} \to \mathbb{C} \setminus \overline{\Omega} extending univalently to {∣t∣>r}∪{∞}\{|t| > r\} \cup \{\infty\} (0<r<10 < r < 1) with a simple pole at ∞\infty and G(T)=∂ΩG(\mathbb{T}) = \partial\Omega, such that ψ=f∘G\psi = f \circ G is an orientation-preserving smooth circle diffeomorphism.

If κ>1\kappa > 1 in Theorem 1.2, there exist X,Y∈Mn(C)X, Y \in M_n(\mathbb{C}) with tr⁡(XX∗)=tr⁡(YY∗)=1\operatorname{tr}(XX^*) = \operatorname{tr}(YY^*) = 1, SX=κXSX = \kappa X, SY=YSY = Y, X∗Y=0X^*Y = 0, and an MnM_n-valued rational function FDF_{\mathbb{D}} holomorphic on a neighborhood of D‾\overline{\mathbb{D}} and contractive on D‾\overline{\mathbb{D}} such that FD[D]x=yF_{\mathbb{D}}[D]x = y for x=vec⁡(X)x = \operatorname{vec}(X) and y=vec⁡(Y)y = \operatorname{vec}(Y).

Under the hypotheses on p,q,r∈Hkp, q, r \in \mathcal{H}_k and F=FΩ∘GF = F_\Omega \circ G with ∥F(t)∥≤1\|F(t)\| \le 1 on T\mathbb{T}, if the 2k2k-by-2k2k block Δ(t)=(pe(t)re(t)∗re(t)qe(t))⪰0\Delta(t) = \begin{pmatrix} p_e(t) & r_e(t)^* \\ r_e(t) & q_e(t) \end{pmatrix} \succeq 0 for almost every t∈Tt \in \mathbb{T}, then κ≤2\kappa \le 2.

For every nonempty compact convex K⊂R2K \subset \mathbb{R}^2 and every ϵ>0\epsilon > 0, there is an admissible domain Ωϵ\Omega_\epsilon such that K⊂ΩϵK \subset \Omega_\epsilon and Ωϵ‾⊂{x:dist⁡(x,K)<ϵ}\overline{\Omega_\epsilon} \subset \{x : \operatorname{dist}(x, K) < \epsilon\}.

Methods

Conformal coordinate construction and harmonic reflection at a regular real-analytic Jordan boundary, combined with local extension patching and inversion for the exterior coordinate.

Source excerpts

“We give the classical extremal proof of Riemann mapping and the harmonic-reflection argument [2, Chapter 6, Section 1.1; Chapter 4, Section 6.5], then prove the collar patching and global univalence needed in Lemma 1.1.” · Show in paper

Finite-dimensional convex separation in R⊕Herm⁡n3\mathbb{R} \oplus \operatorname{Herm}_n^3 and unitary colligation realization with transpose to construct an exact rational extremal function.

Source excerpts

“The construction passes from real separation to a dual balance, then to endpoint supports, and finally to a finite unitary realization.” · Show in paper

Common positive boundary density Λ(t)=j(t)ΦD(ψ(t))\Lambda(t) = j(t)\Phi_D(\psi(t)) and isometric embedding Vm:Cn⊗Cm→L2(T,dσ;Cn⊗Cm)V_m : \mathbb{C}^n \otimes \mathbb{C}^m \to L^2(\mathbb{T}, d\sigma; \mathbb{C}^n \otimes \mathbb{C}^m) representing holomorphic evaluations at all coefficient sizes.

Source excerpts

“For each m, define Vm : Cn ⊗ Cm −→ L2(T, dσ; Cn ⊗ Cm), (Vmb)(t) = (Λ(t)1/2 ⊗ Im)b.” · Show in paper

Exterior kernel m(w,t)=1+B(w,t)+B(w,t)‾m(w, t) = 1 + B(w, t) + \overline{B(w, t)}, Cauchy projection C\mathcal{C} and its Hilbert-space adjoint C†\mathcal{C}^\dagger, and ordered Hilbert-Schmidt L2L^2 density comparison against the original-matrix resolvent field EAE_A.

Source excerpts

“We next turn this kernel into a bounded Cauchy projection on boundary functions. For the density comparison we need estimates for its actual adjoint and a formula relating Laurent tests to the original-matrix field EA.” · Show in paper

Gaussian analytic outer approximation of compact convex sets in R2\mathbb{R}^2 via sublevel sets of hδ(x)=EdK(x−δZ)+δ∣x∣2h_\delta(x) = \mathbb{E} d_K(x - \delta Z) + \delta |x|^2.

Source excerpts

“Let Z be a standard Gaussian in R2 and put hδ(x) = EdK(x − δZ) + δ|x|2, c = E|Z|.” · Show in paper

Finite-dimensional compression onto E=span⁡{Akξj:0≤k≤d,1≤j≤m}E = \operatorname{span}\{A^k \xi_j : 0 \le k \le d, 1 \le j \le m\} and Runge's polynomial approximation theorem on KϵK_\epsilon to transfer finite-matrix bounds to bounded operators on arbitrary complex Hilbert spaces.

Source excerpts

“For ξ ̸= 0, define the nonzero finite-dimensional subspace E = span{Akξj : 0 ≤ k ≤ d, 1 ≤ j ≤ m}, AE = QEA|E, where QE is the orthogonal projection onto E.” · Show in paper