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

Universal strong Kadison–Kastler stability

For every ε>0\varepsilon > 0 there is δ=δ(ε)>0\delta = \delta(\varepsilon) > 0, independent of the algebras, their type, their representations, and the cardinalities of their preduals and Hilbert spaces, such that if HH is any complex Hilbert space and M,N⊆B(H)M, N \subseteq \mathcal{B}(H) are unital von Neumann algebras with identity IHI_H and d(M,N)<δd(M, N) < \delta, then there is a unitary u∈B(H)u \in \mathcal{B}(H) satisfying uMu∗=NuMu^* = N and ∥u−IH∥<ε\|u - I_H\| < \varepsilon.

There is an absolute constant CampC_{\mathrm{amp}} (for example Camp=600C_{\mathrm{amp}} = 600), independent of MM, its representation, and HH, such that for every type II1\mathrm{II}_1 factor M⊆B(H)M \subseteq \mathcal{B}(H) on a separable Hilbert space and every selfadjoint h∈B(H)h \in \mathcal{B}(H), ∥adh∣M∥cb≤Camp∥adh∣M∥\|\mathrm{ad}_h|_M\|_{\mathrm{cb}} \le C_{\mathrm{amp}} \|\mathrm{ad}_h|_M\|.

There are an absolute threshold tfin>0t_{\mathrm{fin}} > 0 and a nondecreasing function ηfin:(0,tfin)→(0,∞)\eta_{\mathrm{fin}} : (0, t_{\mathrm{fin}}) \to (0, \infty) with ηfin(t)→0\eta_{\mathrm{fin}}(t) \to 0 as t↓0t \downarrow 0, independent of MM, NN, and their representations, such that if HH is a separable nonzero Hilbert space, M⊆B(H)M \subseteq \mathcal{B}(H) is a unital finite factor, N⊆B(H)N \subseteq \mathcal{B}(H) is a unital von Neumann algebra, and d(M,N)<t<tfind(M, N) < t < t_{\mathrm{fin}}, then NN is a finite factor and there is a unitary u∈B(H)u \in \mathcal{B}(H) with uMu∗=NuMu^* = N and ∥u−IH∥≤ηfin(t)\|u - I_H\| \le \eta_{\mathrm{fin}}(t).

There are an absolute threshold tfac>0t_{\mathrm{fac}} > 0 and a nondecreasing function ηfac:(0,tfac)→(0,∞)\eta_{\mathrm{fac}} : (0, t_{\mathrm{fac}}) \to (0, \infty) with ηfac(t)→0\eta_{\mathrm{fac}}(t) \to 0 as t↓0t \downarrow 0, independent of the factors, their types, and their representations, such that for every separable complex Hilbert space HH and every pair of unital factors M0,N0⊆B(H)M_0, N_0 \subseteq \mathcal{B}(H) with the same identity and d(M0,N0)<t<tfacd(M_0, N_0) < t < t_{\mathrm{fac}}, there is a unitary u∈B(H)u \in \mathcal{B}(H) with uM0u∗=N0uM_0u^* = N_0 and ∥u−IH∥≤ηfac(t)\|u - I_H\| \le \eta_{\mathrm{fac}}(t).

For every a>0a > 0 there is δsep(a)>0\delta_{\mathrm{sep}}(a) > 0, independent of the algebras and the representation, such that whenever M,NM, N are unital von Neumann algebras on a separable Hilbert space and d(M,N)<δsep(a)d(M, N) < \delta_{\mathrm{sep}}(a), there is a unitary uu with uMu∗=NuMu^* = N and ∥u−I∥<a\|u - I\| < a.

For every ε>0\varepsilon > 0 there is δ>0\delta > 0, independent of the algebras, their representations, and their Hilbert spaces, such that unital von Neumann algebras M,N⊆B(H)M, N \subseteq \mathcal{B}(H) with the same identity and d(M,N)<δd(M, N) < \delta satisfy dcb(M,N)<εd_{\mathrm{cb}}(M, N) < \varepsilon and d(M′,N′)<εd(M', N') < \varepsilon.

If P,QP, Q are von Neumann algebras, ρ:P→B(H)\rho : P \to \mathcal{B}(H) is a faithful normal unital representation, Q⊆B(H)Q \subseteq \mathcal{B}(H) has the same identity, and F:P→QF : P \to Q is a (not necessarily normal) UCP map satisfying ∥F−ρ∥cb≤e<10−3\|F - \rho\|_{\mathrm{cb}} \le e < 10^{-3} and sup⁡y∈Q1dist(y,ρ(P)1)≤g\sup_{y \in Q_1} \mathrm{dist}(y, \rho(P)_1) \le g with g+e<1g + e < 1, then there is a normal unital ∗*-isomorphism θ:P→Q\theta : P \to Q and a unitary u∈B(H)u \in \mathcal{B}(H) such that ∥θ−ρ∥cb≤32e\|\theta - \rho\|_{\mathrm{cb}} \le 32e, uρ(x)u∗=θ(x)u\rho(x)u^* = \theta(x) for all x∈Px \in P, and ∥u−I∥≤128e\|u - I\| \le 128e.

There are absolute constants b0>0b_0 > 0 and Cst>0C_{\mathrm{st}} > 0 such that if M,N⊆B(K)M, N \subseteq \mathcal{B}(K) are type II1\mathrm{II}_1 factors on a separable Hilbert space in common standard tracial position with vector Ω\Omega, containing a common hyperfinite II1\mathrm{II}_1 subfactor RR irreducible in MM and generated by increasing matrix algebras with JMrJM=JNrJNJ_M r J_M = J_N r J_N for all r∈Rr \in R, and d(M,N)<b<b0d(M, N) < b < b_0, then there is a normal unital ∗*-isomorphism ϑ:M→N\vartheta : M \to N satisfying sup⁡x∈M1∥ϑ(x)−x∥≤Cstb\sup_{x \in M_1} \|\vartheta(x) - x\| \le C_{\mathrm{st}} b.

There are absolute constants c,C>0c, C > 0 such that if M0,N0⊆B(H)M_0, N_0 \subseteq \mathcal{B}(H) are factors on a separable Hilbert space, M0M_0 is infinite, and d(M0,N0)<t<cd(M_0, N_0) < t < c, then for the fixed hyperfinite factor D⊆B(K)D \subseteq \mathcal{B}(K) there are unitaries u0∈B(H)u_0 \in \mathcal{B}(H) and v∈B(H⊗K)v \in \mathcal{B}(H \otimes K) with ∥u0−1∥+∥v−1∥≤Ct\|u_0 - 1\| + \|v - 1\| \le Ct such that M=M0⊗‾DM = M_0 \mathbin{\overline{\otimes}} D and N=v((u0N0u0∗)⊗‾D)v∗N = v((u_0 N_0 u_0^*) \mathbin{\overline{\otimes}} D)v^* form a prepared pair and dcb(M,N)≤Ctd_{\mathrm{cb}}(M, N) \le Ct.

There is an absolute constant CC such that every prepared pair (M,N,Ω,R)(M, N, \Omega, R) with dcb(M,N)≤γd_{\mathrm{cb}}(M, N) \le \gamma satisfies sup⁡c>0∥EcM−EcN∥≤Cγ\sup_{c > 0} \|E_c^M - E_c^N\| \le C\gamma, where EcPE_c^P is the orthogonal projection onto the closure of {(xΩ,cx∗Ω‾):x∈P}\{(x\Omega, c\overline{x^*\Omega}) : x \in P\}.

There are absolute constants cmod,Cmod>0c_{\mathrm{mod}}, C_{\mathrm{mod}} > 0 such that if (M,N,Ω,R)(M, N, \Omega, R) is a prepared pair acting on a separable Hilbert space HH with dcb(M,N)<γ<cmodd_{\mathrm{cb}}(M, N) < \gamma < c_{\mathrm{mod}}, then there is a unitary q∈B(H)q \in \mathcal{B}(H) such that ∥q−I∥≤Cmodγ\|q - I\| \le C_{\mathrm{mod}} \gamma, dcb(qMq∗,N)≤Cmodγd_{\mathrm{cb}}(qMq^*, N) \le C_{\mathrm{mod}} \gamma, and ΔNit(qMq∗)ΔN−it=qMq∗\Delta_N^{it}(qMq^*)\Delta_N^{-it} = qMq^* for all t∈Rt \in \mathbb{R}.

There are an absolute γ0>0\gamma_0 > 0 and a nondecreasing function ζ:(0,γ0)→(0,∞)\zeta : (0, \gamma_0) \to (0, \infty) with ζ(γ)→0\zeta(\gamma) \to 0 as γ↓0\gamma \downarrow 0 such that if P,Q⊆B(H)P, Q \subseteq \mathcal{B}(H) are unital von Neumann algebras on a separable Hilbert space, dcb(P,Q)<γ<γ0d_{\mathrm{cb}}(P, Q) < \gamma < \gamma_0, a strongly continuous unitary group UU normalizes both, and C(Q,U)C(Q, U) is a type II∞\mathrm{II}_\infty factor, then a unitary a∈B(H)a \in \mathcal{B}(H) satisfies aPa∗=QaPa^* = Q and ∥a−I∥≤ζ(γ)\|a - I\| \le \zeta(\gamma).

Methods

Converting a unital completely positive (UCP) map that is close in completely bounded norm to a faithful normal representation into a small unitary conjugacy using Ricard–Roydor's multiplication perturbation theorem and close-representation intertwining.

Source excerpts

“We use the completely bounded version supplied by Ricard–Roydor’s multiplication perturbation theorem [38, Proposition 3.2]: on a von Neumann algebra P, an associative product m with ∥m−mP ∥cb ≤ 1/11 admits a cb linear isomorphism L with L(m(x, y)) = L(x)L(y), ∥L − id∥cb ≤ 10∥m − mP ∥cb.” · Show in paper

“Transport multiplication from Q to P: m(x, y) = F−1(F(x)F(y)). ... The cited theorem gives an involution-preserving L with ∥L − id∥cb ≤ λ := 30e/(1 − e) < 1. Then θ = F L−1 is a bijective ∗-homomorphism ... Apply Theorem 2.3, since 32e < 1/4.” · Show in paper

Bounding completely bounded norms of inner-like derivations on type II1 factors by their ordinary operator norms via Haagerup's cyclic similarity theorem, Popa's tracial ultraproduct independence theorem, and free-word randomization.

Source excerpts

“We first control columns by the cyclic similarity theorem. A free-word estimate then permits us to apply this column bound to randomized matrix coefficients. The randomization also recovers a fixed positive multiple of the identity in a weak limit; this will detect every matrix commutator.” · Show in paper

Correcting approximate unitary representations on a countable group to exact measurable cocycles over an amenable nonsingular boundary action and extracting a nearby isomorphism via double ergodicity with unitary coefficients.

Source excerpts

“For their countable indexing group G, we construct a probability space (D, ν) with a nonsingular G-action and an equivariant mean that corrects these multiplication errors. On D2 two actions have different roles. The separate action (g, h)(x, y) = (gx, hy) permits averaging of cocycles. The diagonal action g(x, y) = (gx, gy) has a rigidity property: bounded measurable vector fields transforming under a unitary representation are constant.” · Show in paper

Comparing rescaled Tomita graphs uniformly by randomizing and clipping graph vectors into bounded rows, then using Arveson's nest distance formula, character averaging, and upper half-plane analytic extension to align modular actions.

Source excerpts

“Bounding x in the ordinary operator norm directly would lose this uniformity. We instead flatten a graph vector and then replicate it into a row, where cb closeness supplies the needed estimate.” · Show in paper

“A uniform infinitesimal estimate controls norm-continuous paths of approximate normalizers. Analytic smoothing puts the modular comparison into that setting, and a completely positive limit recovers exact normalizers at each time. Finally an averaging iteration corrects their multiplication law and produces the near-identity change of position.” · Show in paper

Descending unitary conjugacy from a common type II∞ modular crossed product to the original infinite factors by conjugating finite corners, inflating via matrix units, averaging over the dual character action, and slicing by a normal vector state.

Source excerpts

“We now pass to its crossed products, compare finite corners there, and descend the resulting conjugacy twice: first to the adjusted tensor pair, and then through the auxiliary tensor factor to the original Hilbert space.” · Show in paper

Assembling factor stability on separable Hilbert spaces via central decomposition and Jankov–von Neumann uniformization, and passing to arbitrary Hilbert spaces via WOT cluster points of UCP maps from separable restrictions.

Source excerpts

“On a separable Hilbert space, aligning the centers permits a measurable assembly of the factor implementers. On an arbitrary Hilbert space, separable restrictions instead supply a completely positive map into the neighboring algebra. The UCP criterion then gives a unitary on the original Hilbert space.” · Show in paper