OctoInformBeta

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

Proposition 2.2: Common operator bounds and divergent energy per fibre dimension on a concrete Hilbert direct sum

An invariant kernel obstruction to unitarizability.

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

The obstruction compares the kernel energy of translation-invariant operators with bounds on a row, a complementary column and changes under conjugation. Let GG be a discrete group with identity ee. For a finite set KK, the Euclidean fibre FKF_K is the space of complex-valued functions on KK with its sum-of-squares norm.✦ The associated Hilbert space ℋK\mathcal H_K consists of square-summable FKF_K-valued families, as in (1); its bounded-operator algebra ℬ(ℋK)\mathcal B(\mathcal H_K) consists of all bounded complex-linear maps on this space. FK=ℂK,∥v∥2=∑a∈K|va|2,ℋK=ℓ2(G;FK),∥ξ∥2=∑x∈G∥ξ(x)∥2.F_K=\mathbb C^K,\qquad \|v\|^2=\sum_{a\in K}|v_a|^2,\qquad \mathcal H_K=\ell^2(G;F_K),\qquad \|\xi\|^2=\sum_{x\in G}\|\xi(x)\|^2.(1)

For a permutation σ\sigma of GG, the coordinate permutation PσP_\sigma is a bounded operator on ℋK\mathcal H_K, and UKU_K is the left regular representation of GG on ℋK\mathcal H_K, with actions given by (2). The translation commutant 𝒞K\mathcal C_K is the set of bounded complex-linear operators on ℋK\mathcal H_K that commute with every UK(g)U_K(g), as in (3).✦ (Pσξ)(x)=ξ(σ(x)),(UK(g)ξ)(x)=ξ(g−1x).(P_\sigma\xi)(x)=\xi(\sigma(x)),\qquad (U_K(g)\xi)(x)=\xi(g^{-1}x).(2) 𝒞K={A∈ℬ(ℋK):AUK(g)=UK(g)A for every g∈G}.\mathcal C_K=\{A\in\mathcal B(\mathcal H_K): AU_K(g)=U_K(g)A\text{ for every }g\in G\}.(3)

Coordinate evaluation px:ℋK→FKp_x:\mathcal H_K\to F_K, for x,y∈Gx,y\in G, reads the coordinate at xx; coordinate inclusion iy:FK→ℋKi_y:F_K\to\mathcal H_K places a vector at yy and zero elsewhere. For F∈ℬ(ℋK)F\in\mathcal B(\mathcal H_K) and x,y∈Gx,y\in G, the row row⁡x(F)\operatorname{row}_x(F) is a bounded map from ℋK\mathcal H_K to FKF_K, the column col⁡y(F)\operatorname{col}_y(F) is a bounded map from FKF_K to ℋK\mathcal H_K, and the matrix block F[x,y]F[x,y] is a bounded operator on FKF_K, given by (4). row⁡x(F)=pxF,col⁡y(F)=Fiy,F[x,y]=pxFiy.\operatorname{row}_x(F)=p_xF,\qquad \operatorname{col}_y(F)=Fi_y,\qquad F[x,y]=p_xFi_y.(4)

The matrix coefficient vector c(B)∈ℂK×Kc(B)\in\mathbb C^{K\times K} of B∈ℬ(FK)B\in\mathcal B(F_K) has the entries in (5), where (ea)a∈K(e_a)_{a\in K} is the standard orthonormal basis.✦ Its Euclidean norm is the unnormalized Hilbert-Schmidt norm. The kernel energy E(A)E(A) of an operator A∈ℬ(ℋK)A\in\mathcal B(\mathcal H_K) is the nonnegative number given by the sum in (6).✦ c(B)a,b=(Beb)a,∥c(B)∥2=∥B∥HS2=Tr⁡(B*B).c(B)_{a,b}=(Be_b)_a,\qquad \|c(B)\|^2=\|B\|_{\mathrm{HS}}^2=\operatorname{Tr}(B^*B).(5) E(A)=∑s∈G∥c(A[e,s])∥2.E(A)=\sum_{s\in G}\|c(A[e,s])\|^2.(6)

A representation VV on a complex Hilbert space ℰ\mathcal E is a monoid homomorphism from GG to ℬ(ℰ)\mathcal B(\mathcal E); its operator inverse at gg is V(g−1)V(g^{-1}). For a representation VV of GG on a complex Hilbert space ℰ\mathcal E, conjugation Ad⁡V(g)\operatorname{Ad}_{V(g)} is a map from ℬ(ℰ)\mathcal B(\mathcal E) to itself for each g∈Gg\in G, and the coboundary δVT\delta_VT of T∈ℬ(ℰ)T\in\mathcal B(\mathcal E) is a function from GG to ℬ(ℰ)\mathcal B(\mathcal E), defined by (7). Ad⁡V(g)(T)=V(g)TV(g)−1,(δVT)(g)=T−Ad⁡V(g)(T).\operatorname{Ad}_{V(g)}(T)=V(g)TV(g)^{-1},\qquad (\delta_VT)(g)=T-\operatorname{Ad}_{V(g)}(T).(7) A representation π:G→ℬ(ℰ)\pi:G\to\mathcal B(\mathcal E) on a complex Hilbert space ℰ\mathcal E is similar to a unitary representation when there is a bounded invertible complex-linear map S:ℰ→ℰS:\mathcal E\to\mathcal E satisfying (8); no bound is imposed on ∥S∥∥S−1∥\|S\|\|S^{-1}\|.✦ ∥Sπ(g)S−1ξ∥=∥ξ∥(g∈G,ξ∈ℰ).\|S\pi(g)S^{-1}\xi\|=\|\xi\| \qquad(g\in G,\ \xi\in\mathcal E).(8)

For complex Hilbert spaces (Ei)i∈I(E_i)_{i\in I}, the Hilbert direct sum ⨁i∈IEi\bigoplus_{i\in I}E_i is the space of families (ξi)i∈I(\xi_i)_{i\in I} with ξi∈Ei\xi_i\in E_i and ∑i∥ξi∥2<∞\sum_i\|\xi_i\|^2<\infty, with squared norm equal to that sum.✦ For a sequence (kj)j≥0(k_j)_{j\geq0} of nonnegative integers, the stages HjH_j are complex Hilbert spaces, HH is their Hilbert direct sum, and the obstruction space XX is the Hilbert direct sum of two copies of HH, as specified in (9). Hj=ℓ2(G;ℂkj),H=⨁j≥0Hj,X=H⊕H,∥(ξ,η)∥X2=∥ξ∥H2+∥η∥H2.H_j=\ell^2(G;\mathbb C^{k_j}),\qquad H=\bigoplus_{j\geq0}H_j,\qquad X=H\oplus H,\qquad \|(\xi,\eta)\|_X^2=\|\xi\|_H^2+\|\eta\|_H^2.(9) Write UjU_j for left translation on HjH_j and 𝒞j\mathcal C_j for its commutant.

The Claim

Theorem 1. Let GG be a discrete group and (kj)j≥0(k_j)_{j\geq0} a sequence of nonnegative integers. For countable GG, the carrier has the property in (10).✦ X is separable.X\text{ is separable}.(10)✦ Suppose that every kjk_j is positive. For each jj, let Aj,Tj∈ℬ(Hj)A_j,T_j\in\mathcal B(H_j) with Aj∈𝒞jA_j\in\mathcal C_j. Assume that a common real number κ≥0\kappa\geq0 satisfies the bounds and limit in (11). ∥row⁡e(Tj)∥≤κ,∥col⁡e(Aj−Tj)∥≤κ,∥(δUjTj)(g)∥≤κ(j≥0,g∈G),E(Aj)kj→+∞(j→∞).\begin{gathered} \|\operatorname{row}_e(T_j)\|\leq\kappa,\qquad \|\operatorname{col}_e(A_j-T_j)\|\leq\kappa,\qquad \|(\delta_{U_j}T_j)(g)\|\leq\kappa \quad(j\geq0,\ g\in G),\\ \frac{E(A_j)}{k_j}\longrightarrow+\infty\qquad(j\to\infty). \end{gathered}(11) There is a representation π:G→ℬ(X)\pi:G\to\mathcal B(X) satisfying (12).✦ ∥π(g)∥≤1+κ(g∈G),π is not similar to a unitary representation.\|\pi(g)\|\leq1+\kappa\quad(g\in G),\qquad \pi\text{ is not similar to a unitary representation}.(12)✦ For countable GG, there is a representation on the separable space XX as in (13).✦ $$X\text{ is separable, and there exists a representation on }X \text{ satisfying \eqref{eq:claim1}}.$$(13)✦

In the paper, the result reads: for countable GG, positive integers kjk_j indexed by j≥1j\geq1, Hj=ℓ2(G;ℂkj)H_j=\ell^2(G;\mathbb C^{k_j}) with left translations UjU_j, and bounded operators Aj,TjA_j,T_j on HjH_j with AjA_j in the translation commutant and matrix-valued kernels WjW_j on GG satisfying Aj[x,y]=Wj(x−1y)A_j[x,y]=W_j(x^{-1}y), if Ej=∑s∥Wj(s)∥HS2E_j=\sum_s\|W_j(s)\|_{\mathrm{HS}}^2 satisfies Ej/kj→∞E_j/k_j\to\infty and one κ≥0\kappa\geq0 bounds ∥row⁡e(Tj)∥\|\operatorname{row}_e(T_j)\|, ∥col⁡e(Aj−Tj)∥\|\operatorname{col}_e(A_j-T_j)\| and ∥Tj−Uj(g)TjUj(g)−1∥\|T_j-U_j(g)T_jU_j(g)^{-1}\| for every j,gj,g, then GG has a representation on a separable complex Hilbert space by bounded invertible operators, bounded by 1+κ1+\kappa at every group element and not similar to a unitary representation, with no uniform bound assumed on ∥Tj∥\|T_j\| or ∥Aj∥\|A_j\|.

What Is Checked

The obstruction (12) is stated over the concrete stages and carrier in (9), with invariant AjA_j, the common bounds in (11) and divergent energy per fibre dimension.

For every permutation σ\sigma of GG, the coordinate permutation is complex-linear and satisfies (14). ∥Pσξ∥=∥ξ∥(ξ∈ℋK).\|P_\sigma\xi\|=\|\xi\|\qquad(\xi\in\mathcal H_K).(14)

For bounded operators Ti∈ℬ(Ei)T_i\in\mathcal B(E_i) with a common bound c≥0c\geq0, their direct sum ⨁iTi\bigoplus_iT_i on ℰ=⨁i∈IEi\mathcal E=\bigoplus_{i\in I}E_i is the bounded operator given by (15).✦ The direct sum representation VV of unitary representations ViV_i on EiE_i acts by (16).✦ Functions Di:G→ℬ(Ei)D_i:G\to\mathcal B(E_i) satisfying ∥Di(g)∥≤c\|D_i(g)\|\leq c for every i,gi,g have the bounded direct sum D(g)∈ℬ(ℰ)D(g)\in\mathcal B(\mathcal E) at g∈Gg\in G given by (17). ((⨁iTi)ξ)i=Tiξi.\left(\left(\bigoplus_iT_i\right)\xi\right)_i=T_i\xi_i.(15) (V(g)ξ)i=Vi(g)ξi.(V(g)\xi)_i=V_i(g)\xi_i.(16) D(g)=⨁iDi(g).D(g)=\bigoplus_iD_i(g).(17)

The summand inclusion Ji:Ei→ℰJ_i:E_i\to\mathcal E places a vector in summand ii and zero in every other summand; the summand evaluation Qi:ℰ→EiQ_i:\mathcal E\to E_i takes its iith coordinate. For B∈ℬ(ℰ)B\in\mathcal B(\mathcal E), the diagonal compression BiiB_{ii} to the summand EiE_i is the bounded operator given by (18).✦ Bii=QiBJi∈ℬ(Ei).B_{ii}=Q_iBJ_i\in\mathcal B(E_i).(18)

Theorem 2. Let ViV_i be unitary representations on complex Hilbert spaces EiE_i, and let Di:G→ℬ(Ei)D_i:G\to\mathcal B(E_i) have a common bound c≥0c\geq0 for every i,gi,g. Suppose B∈ℬ(ℰ)B\in\mathcal B(\mathcal E) satisfies D(g)=(δVB)(g)D(g)=(\delta_VB)(g) for every g∈Gg\in G. Its diagonal compressions implement the summands as in (19).✦ Di(g)=(δViBii)(g)(i∈I,g∈G).D_i(g)=(\delta_{V_i}B_{ii})(g)\qquad(i\in I,\ g\in G).(19)✦