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

Unitarizability implies amenability for discrete groups

A discrete group is amenable if and only if it is unitarizable; moreover, for every discrete nonamenable group GG and every ε>0\varepsilon > 0, there exist a complex Hilbert space K\mathcal{K} and a representation π:G→GL(K)\pi : G \to \mathrm{GL}(\mathcal{K}) with ∣π∣≤1+ε|\pi| \le 1 + \varepsilon that is not similar to a unitary representation, and if GG is countable, K\mathcal{K} may be chosen separable.

For a countable group GG, if A,T∈B(Hk)A, T \in \mathcal{B}(\mathcal{H}_k) with A∈Uk(G)′A \in U_k(G)', A[x,y]=W(x−1y)A[x,y] = W(x^{-1}y), E=∑s∈G∥W(s)∥HS2E = \sum_{s \in G} \|W(s)\|_{\mathrm{HS}}^2, ∥rowe(T)∥≤K\|\mathrm{row}_e(T)\| \le K, and ∥cole(A−T)∥≤K\|\mathrm{col}_e(A-T)\| \le K, then EE is finite and dist(T,Uk(G)′)≥12E/k−K\mathrm{dist}(T, U_k(G)') \ge \frac{1}{2}\sqrt{E/k} - K.

For a countable group GG, given sequences Aj,Tj∈B(Hj)A_j, T_j \in \mathcal{B}(\mathcal{H}_j) with Aj∈Uj(G)′A_j \in U_j(G)' such that ∥rowe(Tj)∥≤K\|\mathrm{row}_e(T_j)\| \le K, ∥cole(Aj−Tj)∥≤K\|\mathrm{col}_e(A_j - T_j)\| \le K, ∥Tj−Uj(g)TjUj(g)−1∥≤K\|T_j - U_j(g)T_jU_j(g)^{-1}\| \le K, and Ej/kj→∞E_j/k_j \to \infty, GG has a representation on a separable complex Hilbert space with sup⁡g∈G∥π(g)∥≤1+K\sup_{g \in G} \|\pi(g)\| \le 1 + K that is not similar to a unitary representation.

For a countable nonamenable group GG with finite S⊂GS \subset G, d=∣S∣≥2d = |S| \ge 2, and 0<ρ<10 < \rho < 1 satisfying ∥1d∑t∈SRt∥≤ρ\|\frac{1}{d}\sum_{t\in S} R_t\| \le \rho, for every ℓ≥1\ell \ge 1 with n=dℓn = d^\ell and r=⌈nρℓ⌉r = \lceil n\rho^\ell \rceil, there is an assignment a:G×{1,…,n}→{0,1}a : G \times \{1,\dots,n\} \to \{0,1\} such that #{i:a(x,i)=1}≤r\#\{i : a(x,i)=1\} \le r for all x∈Gx \in G and #{i:a(ysi−1,i)=0}≤r\#\{i : a(ys_i^{-1},i)=0\} \le r for all y∈Gy \in G.

For integers n≥2n \ge 2 and r≥1r \ge 1, with p=min⁡(2r,n)p = \min(2r, n), k=⌈2rlog⁡(2n)⌉k = \lceil 2r\log(2n) \rceil, and b0=10b_0 = 10, there exist unit vectors v1,…,vn∈Ckv_1, \dots, v_n \in \mathbb{C}^k such that for every J⊆{1,…,n}J \subseteq \{1,\dots,n\} with ∣J∣≤p|J| \le p, the synthesis operator VJ:ℓ2(J)→CkV_J : \ell^2(J) \to \mathbb{C}^k satisfies ∥VJ∥≤b0\|V_J\| \le b_0.

In the setting of Section 4.1, if ∥VJ∥≤b0\|V_J\| \le b_0 for all ∣J∣≤p|J| \le p and a mask c:G×{1,…,n}→Cc : G \times \{1,\dots,n\} \to \mathbb{C} with ∣c(x,i)∣≤1|c(x,i)| \le 1 has at most pp nonzero entries in every row Rx\mathcal{R}_x and every column Cy\mathcal{C}_y, then ∥Tc∥≤b02\|T_c\| \le b_0^2.

For every nonamenable discrete group GG and every ε>0\varepsilon > 0, there exists a uniformly bounded representation π:G→GL(H)\pi : G \to \mathrm{GL}(H) with ∣π∣≤1+ε|\pi| \le 1 + \varepsilon such that the bounded ℓ1(G)\ell^1(G)-homomorphism Tπ(f)=∑g∈Gf(g)π(g)T_\pi(f) = \sum_{g \in G} f(g)\pi(g) restricted to C[G]\mathbb{C}[G] has no bounded complex-linear unital algebra-homomorphism extension to either the full group C∗C^*-algebra C∗(G)C^*(G) or the reduced group C∗C^*-algebra Cr∗(G)C_r^*(G).

Methods

Invariant-kernel obstruction: bounding the distance of an operator TT from the translation commutant Uk(G)′U_k(G)' from below by 12E/k−K\frac{1}{2}\sqrt{E/k} - K using the Hilbert–Schmidt energy EE of a comparison operator A∈Uk(G)′A \in U_k(G)' and operator norms of one block row of TT and one block column of A−TA - T.

Source excerpts

“Consequently, dist(T, Uk(G)′) ≥ 1/2 √(E/k) − K, (2.2)” · Show in paper

Direct-sum cocycle obstruction: assembling a sequence of operators TjT_j with uniformly bounded row, column, and conjugation-difference norms and diverging Ej/kjE_j/k_j into a single uniformly bounded triangular representation on H⊕H\mathcal{H} \oplus \mathcal{H} that is not unitarizable.

Source excerpts

“The next proposition turns an unbounded sequence of these distances into one uniformly bounded representation that cannot be unitarized.” · Show in paper

Sparse endpoint assignment via Hall's marriage theorem and König's infinity lemma: converting an averaging-norm bound ∥∑i=1nRsi∥≤nρℓ\|\sum_{i=1}^n R_{s_i}\| \le n\rho^\ell on a nonamenable group into an assignment of labelled convolution edges to their row or column endpoints with load at most r=⌈nρℓ⌉r = \lceil n\rho^\ell \rceil.

Source excerpts

“The next argument converts an operator-norm estimate into an assignment of edges with bounded loads at their endpoints.” · Show in paper

Random-sign synthesis bound and indexed-edge factorization: constructing nn unit vectors in Ck\mathbb{C}^k (k=⌈2rlog⁡(2n)⌉k = \lceil 2r\log(2n)\rceil) with bounded synthesis operators on all subfamilies of size ≤2r\le 2r, and factoring masked rank-one weighted convolution operators Tc=SMcFT_c = S M_c F through ℓ2(Ec)\ell^2(E_c).

Source excerpts

“Keep the index i as part of each edge and set Ec = {(x, i) : c(x, i) ≠ 0}, Ec = ℓ2(Ec).” · Show in paper