OctoInformBeta

OpenAI preprint, September 2026 · Section 2. An invariant-kernel obstruction

Lemma 2.1: Kernel energy on a finite-dimensional fibre over an arbitrary discrete group

Row and column bounds constrain the commutant distance.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let GG be a discrete group with identity ee, and let KK be a finite set of cardinality kk. The question is how an operator row and an operator column constrain approximation by operators commuting with left translation.

The fibre ℱK=ℂK\mathcal F_K=\mathbb C^K is the Euclidean space with its standard orthonormal basis (ea)a∈K(e_a)_{a\in K}.✦ The Hilbert space HKH_K in (1) consists of square-summable functions from GG to ℱK\mathcal F_K: HK=ℓ2(G;ℱK),∥ξ∥2=∑x∈G∥ξ(x)∥2.H_K=\ell^2(G;\mathcal F_K), \qquad \|\xi\|^2=\sum_{x\in G}\|\xi(x)\|^2.(1)✦ Write ℬ(HK)\mathcal B(H_K) for the algebra of bounded complex-linear operators on HKH_K.✦

For a bijection σ:G→G\sigma:G\to G, the coordinate permutation Pσ:HK→HKP_\sigma:H_K\to H_K is the complex-linear map defined by (2): (Pσξ)(x)=ξ(σ(x)).(P_\sigma\xi)(x)=\xi(\sigma(x)).(2)✦ The fibre translation representation U:G→ℬ(HK)U:G\to\mathcal B(H_K) is the monoid homomorphism given by (3): (U(g)ξ)(x)=ξ(g−1x).(U(g)\xi)(x)=\xi(g^{-1}x).(3)✦ The translation commutant ℳ=U(G)′\mathcal M=U(G)' is the set of bounded operators on HKH_K commuting with each translation, as in (4): ℳ={F∈ℬ(HK):FU(g)=U(g)F for every g∈G}.\mathcal M=\{F\in\mathcal B(H_K):FU(g)=U(g)F\text{ for every }g\in G\}.(4)✦

Rows, Columns, and Energy

For x,y∈Gx,y\in G, coordinate evaluation εx:HK→ℱK\varepsilon_x:H_K\to\mathcal F_K and coordinate inclusion ιy:ℱK→HK\iota_y:\mathcal F_K\to H_K are the bounded complex-linear maps in (5): εxξ=ξ(x),(ιyv)(z)={v,z=y,0,z≠y.\varepsilon_x\xi=\xi(x), \qquad (\iota_y v)(z)= \begin{cases} v,&z=y,\\ 0,&z\ne y. \end{cases}(5) For F∈ℬ(HK)F\in\mathcal B(H_K), its row rx(F):HK→ℱKr_x(F):H_K\to\mathcal F_K and column cy(F):ℱK→HKc_y(F):\mathcal F_K\to H_K are the bounded maps in (6); its block Fx,y:ℱK→ℱKF_{x,y}:\mathcal F_K\to\mathcal F_K is their coordinate matrix entry: rx(F)=εxF,cy(F)=Fιy,Fx,y=εxFιy.r_x(F)=\varepsilon_xF, \qquad c_y(F)=F\iota_y, \qquad F_{x,y}=\varepsilon_xF\iota_y.(6) The norms of rows and columns are operator norms between these Hilbert spaces.

The coefficient matrix coef⁡(B)∈ℂK×K\operatorname{coef}(B)\in\mathbb C^{K\times K} of B∈ℬ(ℱK)B\in\mathcal B(\mathcal F_K) has entries (Beb)a(Be_b)_a.✦ Its Euclidean norm is the unnormalized Hilbert-Schmidt norm used in (7): ∥coef⁡(B)∥2=∥B∥HS2=∑a,b∈K|(Beb)a|2=Tr⁡(B*B).\|\operatorname{coef}(B)\|^2 =\|B\|_{\mathrm{HS}}^2 =\sum_{a,b\in K}|(Be_b)_a|^2 =\operatorname{Tr}(B^*B).(7) The kernel energy E(A)E(A) of a bounded operator A∈ℬ(HK)A\in\mathcal B(H_K) is the sum of the squared Hilbert-Schmidt norms in its identity row, defined by (8): E(A)=∑s∈G∥Ae,s∥HS2.E(A)=\sum_{s\in G}\|A_{e,s}\|_{\mathrm{HS}}^2.(8)✦

The distance dist⁡(T,ℳ)\operatorname{dist}(T,\mathcal M) from T∈ℬ(HK)T\in\mathcal B(H_K) to the translation commutant ℳ\mathcal M is the operator norm infimum in (9): dist⁡(T,ℳ)=infC∈ℳ∥T−C∥.\operatorname{dist}(T,\mathcal M)=\inf_{C\in\mathcal M}\|T-C\|.(9)

The Claim

Theorem 1. Every bounded operator A∈ℬ(HK)A\in\mathcal B(H_K) has the row summability property (10): ∑s∈G∥Ae,s∥HS2<∞.\sum_{s\in G}\|A_{e,s}\|_{\mathrm{HS}}^2<\infty.(10)✦ Let A∈ℳA\in\mathcal M and T∈ℬ(HK)T\in\mathcal B(H_K), and suppose that a real number κ≥0\kappa\ge0 satisfies the row and column bounds (11): ∥re(T)∥≤κ,∥ce(A−T)∥≤κ.\|r_e(T)\|\le\kappa, \qquad \|c_e(A-T)\|\le\kappa.(11) For every C∈ℳC\in\mathcal M, the energy estimate is (12): E(A)≤4k(κ+∥T−C∥)2.E(A)\le4k\bigl(\kappa+\|T-C\|\bigr)^2.(12)✦ When KK is nonempty, the pointwise bound (13) holds for each such CC, and the commutant distance satisfies (14): 12E(A)k−κ≤∥T−C∥.\frac12\sqrt{\frac{E(A)}{k}}-\kappa\le\|T-C\|.(13)✦ 12E(A)k−κ≤dist⁡(T,ℳ).\frac12\sqrt{\frac{E(A)}{k}}-\kappa \le\operatorname{dist}(T,\mathcal M).(14)✦

In the paper, the result reads: for countable GG and a positive integer kk, let A,T∈ℬ(HK)A,T\in\mathcal B(H_K) with |K|=k|K|=k and A∈ℳA\in\mathcal M, write Ax,y=W(x−1y)A_{x,y}=W(x^{-1}y) and E=∑s∈G∥W(s)∥HS2E=\sum_{s\in G}\|W(s)\|_{\mathrm{HS}}^2, and suppose that κ≥0\kappa\ge0, ∥re(T)∥≤κ\|r_e(T)\|\le\kappa and ∥ce(A−T)∥≤κ\|c_e(A-T)\|\le\kappa; then EE is finite and every C∈ℳC\in\mathcal M satisfies E≤4k(κ+∥T−C∥)2E\le4k(\kappa+\|T-C\|)^2, with dist⁡(T,ℳ)≥12E/k−κ\operatorname{dist}(T,\mathcal M)\ge\frac12\sqrt{E/k}-\kappa.

What Is Checked

The statements concern bounded complex-linear operators on HKH_K for an arbitrary discrete group GG and finite fibre ℱK\mathcal F_K, with translation commutation and the row and column hypotheses specified above; the two distance bounds require KK to be nonempty.

Coordinate permutations are complex-linear isometries, with the norm identity (15) for every bijection σ:G→G\sigma:G\to G and ξ∈HK\xi\in H_K: ∥Pσξ∥=∥ξ∥.\|P_\sigma\xi\|=\|\xi\|.(15)