betaFeedback

WorksMathematics

Quantum Gromov–Hausdorff Convergence of Spectral Truncations for Amenable Discrete Quantum Groups with Modified Rapid Decay

Abstract

Let GG be an amenable discrete quantum group with the modified rapid decay property relative to a proper central length ℓ\ell. We construct a modularly corrected commutator seminorm on Cr∗(G)C_r^*(G) and prove its right invariance. Averaging over the coproduct yields a regular bi-invariant Lip-norm. For the spectral projections PR=χ[0,R](Dℓ)P_R=χ_{[0,R]}(D_\ell), we equip the finite-dimensional operator systems AR=PRCr∗(G)PRA_R=P_RC_r^*(G)P_R with the Lip-norms induced by the compressed left and right coactions. We prove quantum Gromov–Hausdorff convergence and obtain an explicit approximation estimate. These spectral compressions are length-filtered Peter–Weyl truncations of the compact dual. The resulting regular bi-invariant Lip-norm also places the family within Leimbach's operator Gromov–Hausdorff convergence theorem. The construction applies to non-unimodular discrete quantum groups, including discrete duals of standard deformations of compact semisimple Lie groups.

Introduction

Compact quantum metric spaces provide a noncommutative counterpart of compact metric spaces [9]. Their finite-dimensional approximations have been studied via Fourier, Peter–Weyl and spectral truncations; see, for example, [2, 10, 6, 3, 5]. In particular, Leimbach [5] established convergence of Peter–Weyl compressions for coamenable compact quantum groups under a regular bi-invariant Lip-norm hypothesis. In our previous work [8], we studied spectral truncations of compact quantum groups and proved their convergence in quantum Gromov–Hausdorff distance.

We now turn to amenable discrete quantum groups satisfying the modified rapid decay property. Our aim is to construct suitable Lip-norms and study the convergence of the associated spectral truncations. The modular correction in the RD estimate plays an essential role, particularly for non-unimodular quantum groups. We use the construction of Bhowmick, Voigt and Zacharias [1]. Given a proper central length ℓ\ell, let DℓD_\ell be the associated length operator. Set

PR=χ[0,R](Dℓ),AR=PRCr∗(G)PR.P_R=\chi_{[0,R]}(D_\ell),\qquad A_R=P_RC_r^*(G)P_R.

By properness of ℓ\ell, each ARA_R is a finite-dimensional operator system.

It is useful to make explicit the relation with Peter–Weyl truncations. If G^\widehat{G} denotes the compact dual of GG and pαp_\alpha is the projection onto the coefficient space associated with α∈Irr⁡(G^)\alpha\in\operatorname{Irr}(\widehat{G}), then

L2(G)=⨁α∈Irr⁡(G^)Hα,Dℓ=∑αℓ(α)pα.L^2(G)=\bigoplus_{\alpha\in\operatorname{Irr}(\widehat{G})}H_\alpha,\qquad D_\ell=\sum_\alpha\ell(\alpha)p_\alpha.

Consequently,

PR=∑ℓ(α)≤Rpα.P_R=\sum_{\ell(\alpha)\le R}p_\alpha.

Thus ARA_R is precisely the compression associated with the length-filtered finite set

ΛR={α∈Irr⁡(G^):ℓ(α)≤R}.\Lambda_R=\{\alpha\in\operatorname{Irr}(\widehat{G}):\ell(\alpha)\le R\}.

In particular, these are length-filtered Peter–Weyl truncations; we do not regard spectral truncation as a different class of compression. The point here is to construct the metric structure required for this particular family directly from the modified RD estimates on GG. More precisely, the modular correction is described by a positive self-adjoint operator TT, which is generally unbounded and whose eigenvalues depend on the row index of an irreducible corepresentation. The associated seminorm

LTk(a)=∥Tkad⁡(Dℓ)k(a)Tk∥,a∈C[G]sa,L_T^k(a)=\left\|T^k\operatorname{ad}(D_\ell)^k(a)T^k\right\|,\qquad a\in\mathbb{C}[G]_{\mathrm{sa}},

is right-invariant. Since the modular weights need not be scalar on irreducible blocks, we do not assume that this original seminorm is bi-invariant. Instead, we form its bi-invariant envelope

L~Tk(a)=sup⁡ω∈S(A)LTk((ω⊗id⁡)Δ^(a)).\widetilde{L}_T^k(a)=\sup_{\omega\in S(A)}L_T^k\bigl((\omega\otimes\operatorname{id})\widehat{\Delta}(a)\bigr).

We then use both compressed coactions to define

LR(x)=max⁡{LRr(x),LRℓ(x)}.L_R(x)=\max\{L_R^r(x),L_R^\ell(x)\}.

The two terms are needed separately: the compression map is controlled by the bi-invariance of L~Tk\widetilde{L}_T^k, whereas the symbol map is controlled by the left coaction.

Our contribution is the construction and verification of this modularly corrected regular bi-invariant Lip-norm from modified RD, together with the resulting quantitative approximation estimates for the length-filtered compressions. This is not a new abstract Peter–Weyl convergence mechanism. Rather, it supplies a concrete class of Lip-norms on compact duals of amenable discrete quantum groups, including non-unimodular examples, to which the general convergence theory of Leimbach applies. We also give a direct order-unit quantum Gromov–Hausdorff estimate.

The paper is organized as follows. Section 2 recalls the modified RD estimates and the order-unit approximation criterion. Section 3 constructs the invariant seminorms and the compressed coactions, and proves convergence. Section 4 records examples.

Preliminaries

Discrete quantum groups, lengths and RD

Let GG be a discrete quantum group with compact dual G^\widehat{G}. Write

C0(G)=⨁α∈Irr⁡(G^)c0B(Hα),A=Cr∗(G)=C(G^),C_0(G)=\bigoplus_{\alpha\in\operatorname{Irr}(\widehat{G})}^{c_0}B(H_\alpha),\qquad A=C_r^*(G)=C(\widehat{G}),

with coproducts Δ\Delta and Δ^\widehat{\Delta}, respectively. The canonical dense Hopf ∗*-algebra of AA is denoted by C[G]\mathbb{C}[G]; the algebraic direct sum of the matrix blocks of C0(G)C_0(G) is denoted by Cc(G)C_c(G). The Fourier transform gives inverse linear isomorphisms

F:Cc(G)⟶C[G],G:C[G]⟶Cc(G).\mathcal{F}:C_c(G)\longrightarrow\mathbb{C}[G],\qquad\mathcal{G}:\mathbb{C}[G]\longrightarrow C_c(G).

The left Haar weight on Cc(G)C_c(G) gives the Hilbert space L2(G)L^2(G). We use its Fourier identification with the GNS space of the Haar state of AA and follow the conventions of [1, 12]. The quotient map Cf∗(G)→AC_f^*(G) \to A is an isomorphism precisely when GG is amenable. In this case the algebraic counit extends to a state ϵ\epsilon on AA.

A length function is a map ℓ:Irr⁡(G^)→[0,∞)\ell: \operatorname{Irr}(\widehat{G}) \to[0,\infty) such that ℓ(1)=0\ell(1)=0, ℓ(α‾)=ℓ(α)\ell(\overline{\alpha})=\ell(\alpha), and ℓ(γ)≤ℓ(α)+ℓ(β)\ell(\gamma)\leq\ell(\alpha)+\ell(\beta) whenever γ\gamma is contained in α⊗β\alpha\otimes\beta. It is proper if each set {α:ℓ(α)≤R}\{\alpha:\ell(\alpha)\leq R\} is finite and ℓ(α)=0\ell(\alpha)=0 only for the trivial corepresentation. The corresponding central length is L=∑αℓ(α)pαL=\sum_\alpha\ell(\alpha)p_\alpha. Properness is precisely the condition ensuring that each bounded spectral subspace of the associated length operator is finite dimensional. In particular, the spectral projections PRP_R used below have finite rank and increase strongly to the identity.

Let h^\widehat{h} be the Haar state of AA and let Λ^:A→L2(G)\widehat{\Lambda}:A\to L^2(G) be its GNS map. The associated length operator is

Dℓ=⨁α∈Irr⁡(G^)ℓ(α)IHα.D_\ell=\bigoplus_{\alpha\in\operatorname{Irr}(\widehat{G})}\ell(\alpha)I_{H_\alpha}.

where HαH_\alpha is the coefficient subspace spanned by Λ^(uijα)\widehat{\Lambda}(u_{ij}^{\alpha}). Properness makes its spectral projections finite rank.

Let F=⨁αFαF=\bigoplus_\alpha F^\alpha be the modular element of GG. Each FαF^\alpha is positive and invertible, with

Tr⁡(Fα)=Tr⁡((Fα)−1)=dim⁡q(α).\operatorname{Tr}(F^\alpha)=\operatorname{Tr}((F^\alpha)^{-1})=\dim_q(\alpha).

Choose an orthonormal basis which diagonalizes every FαF^\alpha, and write

Fαeiα=fiαeiα,fiα>0.F^\alpha e_i^\alpha=f_i^\alpha e_i^\alpha,\qquad f_i^\alpha>0.

Put

ξijα=Λ^(uijα),Hα,i=span⁡‾{ξijα:1≤j≤dα}.\xi_{ij}^\alpha=\widehat{\Lambda}(u_{ij}^\alpha),\qquad H_{\alpha,i}=\overline{\operatorname{span}}\{\xi_{ij}^\alpha:1\leq j\leq d_\alpha\}.

The GNS Hilbert space decomposes orthogonally as

L2(G)=⨁α∈Irr⁡(G^)⨁i=1dαHα,i.L^2(G)=\bigoplus_{\alpha\in\operatorname{Irr}(\widehat{G})}\bigoplus_{i=1}^{d_\alpha}H_{\alpha,i}.

Set

cα,i=dim⁡q(α)dαfiα,dα=dim⁡(α).c_{\alpha,i}=\frac{\dim_q(\alpha)}{d_\alpha}f_i^\alpha,\qquad d_\alpha=\dim(\alpha).

On the coefficient space associated with α\alpha, the modular correction is the matrix

Cα=dim⁡q(α)dαFα.C^\alpha=\frac{\dim_q(\alpha)}{d_\alpha}F^\alpha.

Define the positive self-adjoint direct-sum operators F,C,TF,C,T by

F∣Hα,i=fiαI,C∣Hα,i=cα,iI,T=C1/(2k)∣Hα,i=cα,i1/(2k)I.F|_{H_{\alpha,i}}=f_i^\alpha I,\qquad C|_{H_{\alpha,i}}=c_{\alpha,i}I,\qquad T=C^{1/(2k)}|_{H_{\alpha,i}}=c_{\alpha,i}^{1/(2k)}I.

Their domains are determined by the corresponding square-summability conditions; explicitly,

Dom⁡(F)={ξ=⨁α,iξα,i:∑α,i(fiα)2∥ξα,i∥2<∞},\operatorname{Dom}(F)=\left\{\xi=\bigoplus_{\alpha,i}\xi_{\alpha,i}:\sum_{\alpha,i}(f_i^\alpha)^2\|\xi_{\alpha,i}\|^2<\infty\right\},
Dom⁡(C)={ξ=⨁α,iξα,i:∑α,icα,i2∥ξα,i∥2<∞},\operatorname{Dom}(C)=\left\{\xi=\bigoplus_{\alpha,i}\xi_{\alpha,i}:\sum_{\alpha,i}c_{\alpha,i}^2\|\xi_{\alpha,i}\|^2<\infty\right\},
Dom⁡(T)={ξ=⨁α,iξα,i:∑α,icα,i1/k∥ξα,i∥2<∞}.\operatorname{Dom}(T)=\left\{\xi=\bigoplus_{\alpha,i}\xi_{\alpha,i}:\sum_{\alpha,i}c_{\alpha,i}^{1/k}\|\xi_{\alpha,i}\|^2<\infty\right\}.

These operators are positive and injective, and may be unbounded. Their finite-support vectors form a common invariant core, and T∗=C1/2T^{*}=C^{1/2}. Since DℓD_{\ell} acts as ℓ(α)I\ell(\alpha)I on each Hα,iH_{\alpha,i}, the operators F,C,TF,C,T and DℓD_{\ell} strongly commute. The unimodular case is characterized by Fα=IF^{\alpha}=I and hence C=T=IC=T=I.

For x=∑α,i,jxijαuijα∈C[G]x=\sum_{\alpha,i,j}x_{ij}^{\alpha}u_{ij}^{\alpha}\in\mathbb{C}[G], the modified Sobolev norm of [1] is

∥x∥2,s2=∑α,i,j(1+ℓ(α))2sdα∣xijα∣2.(1)\|x\|_{2,s}^{2}=\sum_{\alpha,i,j}\frac{(1+\ell(\alpha))^{2s}}{d_{\alpha}}|x_{ij}^{\alpha}|^{2}. \tag*{(1)}

Equivalently, it is the norm of (1+L)sG(C1/2∗x)(1+L)^{s}\mathcal{G}(C^{1/2}*x) in the left-Haar GNS space. Thus, in the non-unimodular case, it is not the uncorrected Haar-GNS norm of xx. We say that GG has the modified property RD with respect to ℓ\ell if there are constants c,s>0c,s>0 such that

∥F(f)∥A≤c∥f∥2,s,f∈Cc(G).(2)\|\mathcal{F}(f)\|_{A}\leq c\|f\|_{2,s},\qquad f\in C_{c}(G). \tag*{(2)}

By the Fourier isometry, this is equivalent to

∥x∥A≤c∥x∥2,s,x∈C[G].\|x\|_{A}\leq c\|x\|_{2,s},\qquad x\in\mathbb{C}[G].

For a∈C[G]a\in\mathbb{C}[G], define

ad⁡(Dℓ)0(a)=a,ad⁡(Dℓ)j+1(a)=[Dℓ,ad⁡(Dℓ)j(a)].\operatorname{ad}(D_{\ell})^{0}(a)=a,\qquad\operatorname{ad}(D_{\ell})^{j+1}(a)=[D_{\ell},\operatorname{ad}(D_{\ell})^{j}(a)].

The modularly corrected commutator seminorm is

LTk(a)=∥Tkad⁡(Dℓ)k(a)Tk∥.(3)L_{T}^{k}(a)=\left\|T^{k}\operatorname{ad}(D_{\ell})^{k}(a)T^{k}\right\|. \tag*{(3)}

The product in (3) is initially defined on the common finite-support core. By the estimates in [1 Lemma 6.5], it extends to a bounded operator for a∈C[G]a\in\mathbb{C}[G]. We use the same notation for this bounded extension. No boundedness of CC or TT is assumed.

Theorem 2.1 (Proper-length form of the BVZ criterion). Suppose GG has a proper central length ℓ\ell and satisfies (2). If k>sk>s is a positive integer, then LTkL_{T}^{k}, with domain C[G]sa\mathbb{C}[G]_{\mathrm{sa}}, is a Lip-norm on AsaA_{\mathrm{sa}} in the sense that its Monge–Kantorovich metric induces the weak topology on S(A)S(A).*

Proof. By [1 Lemma 6.5], LTkL_{T}^{k} is a Lipschitz seminorm with scalar kernel. The coefficient estimate of [1 Lemma 6.6] gives

∑α,i,jℓ(α)2kdα∣aijα∣2≤LTk(a)2.(4)\sum_{\alpha,i,j}\frac{\ell(\alpha)^{2k}}{d_{\alpha}}|a_{ij}^{\alpha}|^{2}\leq L_{T}^{k}(a)^{2}. \tag*{(4)}

For a>N=∑ℓ(α)>N,i,jaijαuijαa_{>N}=\sum_{\ell(\alpha)>N,i,j}a_{ij}^{\alpha}u_{ij}^{\alpha}, the RD estimate and (1) imply, for N≥1N\geq1,

∥a>N∥≤c∥a>N∥2,s≤c2sNs−k(∑α,i,jℓ(α)2kdα∣aijα∣2)1/2≤c2sNs−kLTk(a).(5)\begin{aligned} \|a_{>N}\|\leq c\|a_{>N}\|_{2,s} \\ &\leq c2^{s}N^{s-k}\left(\sum_{\alpha,i,j}\frac{\ell(\alpha)^{2k}}{d_{\alpha}}|a_{ij}^{\alpha}|^{2}\right)^{1/2} \\ &\leq c2^{s}N^{s-k}L_{T}^{k}(a). \tag*{(5)} \end{aligned}

Fix the Haar state h^\widehat{h} and consider the centered unit ball

E={a∈C[G]sa:LTk(a)≤1, h^(a)=0}.E=\{a\in\mathbb{C}[G]_{\mathrm{sa}}:L_{T}^{k}(a)\leq1,\ \widehat{h}(a)=0\}.

This is the centering required in the Ozawa–Rieffel criterion. Write

a≤N=∑ℓ(α)≤N,i,jaijαuijα,a>N=a−a≤N.a_{\le N}=\sum_{\ell(\alpha)\le N,i,j}a_{ij}^{\alpha}u_{ij}^{\alpha},\qquad a_{>N}=a-a_{\le N}.

For fixed NN, properness implies that only finitely many irreducibles occur in a≤Na_{\le N}. Since h^(a)=0\widehat{h}(a)=0, the coefficient of the trivial corepresentation vanishes. For every remaining coefficient with ℓ(α)≤N\ell(\alpha)\le N, the number ℓ(α)2k/dα\ell(\alpha)^{2k}/d_{\alpha} is strictly positive and fixed. Thus (4) bounds every coefficient of a≤Na_{\le N} uniformly for a∈Ea\in E. It follows that

{a≤N:a∈E}\{a_{\le N}:a\in E\}

is bounded in a finite-dimensional vector space and hence is totally bounded in the C∗C^*-norm. On the other hand, (5) gives

sup⁡a∈E∥a−a≤N∥≤c2sNs−k⟶0.\sup_{a\in E}\lVert a-a_{\le N}\rVert\le c2^{s}N^{s-k}\longrightarrow0.

A finite net for the low-frequency truncations, enlarged by this uniform tail bound, is therefore a finite net for EE. Hence EE is totally bounded. The Ozawa–Rieffel criterion [1 Proposition 6.2] proves the Lip-norm property.

Bhowmick, Voigt and Zacharias state [1 Theorem 7.4] for finitely generated discrete quantum groups equipped with word length. The argument above uses only the finiteness of bounded length balls, which is exactly the properness assumption. Thus the same low-frequency and high-frequency estimates establish the Lip-norm property for an arbitrary proper central length. □

Compact quantum metric spaces

Let XX be a unital C∗C^*-algebra and XsaX_{\mathrm{sa}} its self-adjoint part. A seminorm LL on a dense real subspace of XsaX_{\mathrm{sa}} is a Lip-norm if L(1X)=0L(1_X)=0, ker⁡L=R1X\ker L=\mathbb{R}1_X, and the metric

dL(μ,ν)=sup⁡{∣μ(x)−ν(x)∣:x∈Xsa, L(x)≤1}d_L(\mu,\nu)=\sup\{|\mu(x)-\nu(x)|:x\in X_{\mathrm{sa}},\ L(x)\le1\}

induces the weak∗^* topology on S(X)S(X). We write

diam⁡(X,L)=sup⁡μ,ν∈S(X)dL(μ,ν).\operatorname{diam}(X,L)=\sup_{\mu,\nu\in S(X)}d_L(\mu,\nu).

We use Rieffel’s equivalent criterion: for one (hence every) state ω\omega,

{x∈Xsa:L(x)≤1, ω(x)=0}\{x\in X_{\mathrm{sa}}:L(x)\le1,\ \omega(x)=0\}

is norm totally bounded [9]. The same definitions apply to the self-adjoint order-unit space of an operator system. In particular, ARA_R is treated as an operator system, not as a C∗C^*-subalgebra; its seminorms are defined on (AR)sa(A_R)_{\mathrm{sa}} and have kernel RPR\mathbb{R}P_R.

For compact quantum metric order-unit spaces, an admissible Lip-norm on A⊕BA\oplus B is a Lip-norm whose coordinate quotient seminorms are LAL_A and LBL_B. The quantum Gromov–Hausdorff distance is the infimum of the Hausdorff distances between the two state spaces in the metric induced by such an admissible Lip-norm; see also [4] for background on quantum metric structures. We recall an order-unit approximation criterion for quantum Gromov–Hausdorff convergence, a convenient formulation of the standard contractive approximation argument; see, e.g., [4, 11].

Proposition 2.2 (Contractive approximation criterion). Let (A,LA)(A,L_A) and (B,LB)(B,L_B) be compact quantum metric order-unit spaces. Suppose unital positive maps Φ:A→B\Phi:A\to B and Ψ:B→A\Psi:B\to A satisfy

LB(Φ(a))≤LA(a),LA(Ψ(b))≤LB(b),L_B(\Phi(a))\le L_A(a),\qquad L_A(\Psi(b))\le L_B(b),

and, for some η≥0\eta\ge0,

∥ΨΦ(a)−a∥≤ηLA(a),∥ΦΨ(b)−b∥≤ηLB(b)\lVert\Psi\Phi(a)-a\rVert\le\eta L_A(a),\qquad\lVert\Phi\Psi(b)-b\rVert\le\eta L_B(b)

on the self-adjoint Lip-domains. Then

dist⁡q((A,LA),(B,LB))≤η.\operatorname{dist}_{q}((A,L_A),(B,L_B))\le\eta.

Proof. Fix r>ηr>\eta and put

Mr(a,b)=max⁡{LA(a),LB(b),r−1∥b−Φ(a)∥,r−1∥a−Ψ(b)∥}.M_r(a,b)=\max\{L_A(a),L_B(b),r^{-1}\lVert b-\Phi(a)\rVert,r^{-1}\lVert a-\Psi(b)\rVert\}.

The quotient seminorms are LAL_A and LBL_B: the lower bounds follow from the first two terms, while the reverse bounds follow by testing (a,Φ(a))(a,\Phi(a)) and (Ψ(b),b)(\Psi(b),b) and using η<r\eta<r. The kernel of MrM_r is the common scalar line.

Choose states ωA∈S(A)\omega_A\in S(A) and ωB∈S(B)\omega_B\in S(B). If Mr(a,b)≤1M_r(a,b)\le1, then the centered elements

a∘=a−ωA(a)1A,b∘=b−ωB(b)1Ba^\circ=a-\omega_A(a)1_A,\qquad b^\circ=b-\omega_B(b)1_B

lie in the respective centered Lip-unit balls, which are totally bounded. Moreover,

∣ωB(b)−ωA(a)∣≤∥b−Φ(a)∥+∣ωB(Φ(a))−ωA(a)∣≤r+dLA(ωB∘Φ,ωA)≤r+diam⁡(A,LA).\begin{aligned} |\omega_B(b)-\omega_A(a)| &\le\lVert b-\Phi(a)\rVert+|\omega_B(\Phi(a))-\omega_A(a)|\\ &\le r+d_{L_A}(\omega_B\circ\Phi,\omega_A)\\ &\le r+\operatorname{diam}(A,L_A). \end{aligned}

Thus, after subtracting the common scalar ωA(a)\omega_A(a), the pairs in the MrM_r-unit ball lie in a totally bounded subset of Asa⊕BsaA_{\mathrm{sa}}\oplus B_{\mathrm{sa}}. Hence MrM_r is an admissible Lip-norm. For μ∈S(A)\mu\in S(A), pair μ\mu with μ∘Ψ∈S(B)\mu\circ\Psi\in S(B). Whenever Mr(a,b)≤1M_r(a,b)\le1,

∣μ(a)−(μ∘Ψ)(b)∣≤∥a−Ψ(b)∥≤r.|\mu(a)-(\mu\circ\Psi)(b)|\le\lVert a-\Psi(b)\rVert\le r.

The analogous estimate pairs each state of BB with its pullback through Φ\Phi. The Hausdorff distance is at most rr; letting r↓ηr\downarrow\eta proves the claim. □

Spectral compressions and quantum distance

Let λ:A→B(L2(G))\lambda:A\to B(L^2(G)) be the left regular representation, and suppress λ\lambda when no confusion can arise. For R>0R>0, set

PR=χ[0,R](Dℓ),HR=PRL2(G),P_R=\chi_{[0,R]}(D_\ell),\qquad\mathcal{H}_R=P_RL^2(G),
AR=PRλ(A)PR⊆B(HR),τR(a)=PRλ(a)PR.A_R=P_R\lambda(A)P_R\subseteq B(\mathcal{H}_R),\qquad\tau_R(a)=P_R\lambda(a)P_R.

Lemma 2.3. For every R>0R>0, the projection PRP_R has finite rank,

rank⁡(PR)=∑ℓ(α)≤Rdα2\operatorname{rank}(P_R)=\sum_{\ell(\alpha)\le R}d_\alpha^2

and ARA_R is a finite-dimensional operator system with order unit PRP_R. The compression τR:A→AR\tau_R:A\to A_R is ucp and surjective. Moreover,

τR(C[G])=AR.\tau_R(\mathbb{C}[G])=A_R.

Proof. Only finitely many irreducibles satisfy ℓ(α)≤R\ell(\alpha) \le R. Since the coefficient space of uαu^\alpha has dimension dα2d_\alpha^2,

rank⁡(PR)=∑ℓ(α)≤Rdα2.\operatorname{rank}(P_R)=\sum_{\ell(\alpha)\le R}d_\alpha^2.

Thus ARA_R is finite dimensional. Compression is completely positive and τR(1A)=PR\tau_R(1_A)=P_R, so τR:A→AR\tau_R:A\to A_R is ucp and is onto by the definition of ARA_R. Density of C[G]\mathbb{C}[G] in AA implies that τR(C[G])\tau_R(\mathbb{C}[G]) is dense in ARA_R; being a linear subspace of a finite-dimensional space, it is closed. Hence τR(C[G])=AR\tau_R(\mathbb{C}[G])=A_R. □

We use the order-unit version of Rieffel’s quantum Gromov–Hausdorff distance described above. The contractive approximation criterion in Proposition 2.2 is the only form needed below.

Spectral truncation convergence

Throughout this section GG is amenable, and hence the counit ϵ\epsilon is a state on AA. Let W,VW,V be the standard implementing unitaries on L2(G)⊗L2(G)L^2(G)\otimes L^2(G), so that

Δ^(a)=W∗(1⊗a)W=V(a⊗1)V∗.\widehat{\Delta}(a)=W^*(1\otimes a)W=V(a\otimes1)V^*.

Invariance of the modularly corrected seminorm

Choose bases so that Fα=diag⁡(f1α,…,fdαα)F^\alpha=\operatorname{diag}(f^\alpha_1,\ldots,f^\alpha_{d_\alpha}). Then

TΛ^(uijα)=τiαΛ^(uijα),τiα=(dim⁡q(α)dαfiα)1/(2k).T\widehat{\Lambda}(u^\alpha_{ij})=\tau^\alpha_i\widehat{\Lambda}(u^\alpha_{ij}),\qquad \tau^\alpha_i=\left(\frac{\dim_q(\alpha)}{d_\alpha}f^\alpha_i\right)^{1/(2k)}.

The dependence on the row index is relevant when GG is non-unimodular.

Lemma 3.1. On the algebraic core Λ^(C[G])⊙Λ^(C[G])\widehat{\Lambda}(\mathbb{C}[G])\odot\widehat{\Lambda}(\mathbb{C}[G]), the unitaries W,W∗,V,V∗W,W^*,V,V^* preserve the core and strongly commute with the spectral projections of 1⊗Dℓ1\otimes D_\ell (for W,W∗W,W^*) and Dℓ⊗1D_\ell\otimes1 (for V,V∗V,V^*). Moreover,

[1⊗Dℓ,W]=[1⊗Dℓ,W∗]=0,[Dℓ⊗1,V]=[Dℓ⊗1,V∗]=0,[T⊗1,V]=[T⊗1,V∗]=0.[1\otimes D_\ell,W]=[1\otimes D_\ell,W^*]=0,\qquad [D_\ell\otimes1,V]=[D_\ell\otimes1,V^*]=0,\qquad [T\otimes1,V]=[T\otimes1,V^*]=0.

The commutation relations with T⊗1T\otimes1 are strong commutation relations.

Proof. Put ξijα=Λ^(uijα)\xi^\alpha_{ij}=\widehat{\Lambda}(u^\alpha_{ij}) and

Hα,i=span⁡{ξijα:1≤j≤dα},Hβ=span⁡{ξpqβ:1≤p,q≤dβ}.H_{\alpha,i}=\operatorname{span}\{\xi^\alpha_{ij}:1\le j\le d_\alpha\},\qquad H_\beta=\operatorname{span}\{\xi^\beta_{pq}:1\le p,q\le d_\beta\}.

With our convention for the implementing unitaries, the algebraic formulas are

W∗(ξijα⊗ξpqβ)=∑rΛ^(uprβuijα)⊗ξrqβ,W^*(\xi^\alpha_{ij}\otimes\xi^\beta_{pq}) =\sum_r\widehat{\Lambda}(u^\beta_{pr}u^\alpha_{ij})\otimes\xi^\beta_{rq},
V(ξijα⊗ξpqβ)=∑rξirα⊗Λ^(urjαupqβ).V(\xi^\alpha_{ij}\otimes\xi^\beta_{pq}) =\sum_r\xi^\alpha_{ir}\otimes\widehat{\Lambda}(u^\alpha_{rj}u^\beta_{pq}).

Products of matrix coefficients are finite linear combinations of coefficients. The corresponding formulas for the inverse unitaries give

W±1(L2(G)⊗Hβ)=L2(G)⊗Hβ,V±1(Hα,i⊗L2(G))=Hα,i⊗L2(G).W^{\pm1}(L^2(G)\otimes H_\beta)=L^2(G)\otimes H_\beta,\qquad V^{\pm1}(H_{\alpha,i}\otimes L^2(G))=H_{\alpha,i}\otimes L^2(G).

In particular, the relevant subspaces are reducing. Since Dℓ∣Hβ=ℓ(β)ID_\ell|_{H_\beta}=\ell(\beta)I and Dℓ∣Hα,i=ℓ(α)ID_\ell|_{H_{\alpha,i}}=\ell(\alpha)I, we obtain

[1⊗Dℓ,W]=[1⊗Dℓ,W∗]=0,[Dℓ⊗1,V]=[Dℓ⊗1,V∗]=0.[1\otimes D_\ell,W]=[1\otimes D_\ell,W^*]=0,\qquad [D_\ell\otimes1,V]=[D_\ell\otimes1,V^*]=0.

Moreover, T∣Hα,i=τiqIT|_{\mathcal{H}_{\alpha,i}}=\tau_i^q I, so the same reducing subspaces give

[T⊗1,V]=[T⊗1,V∗]=0.[T\otimes1,V]=[T\otimes1,V^*]=0.

These are strong commutation relations: the unitaries commute with the spectral projections of the indicated self-adjoint operators. □

Remark 3.2. For non-Kac GG, one cannot in general infer [1⊗T,W]=0[1\otimes T,W]=0: the second row index changes from pp to rr, and the corresponding TT-eigenvalues may differ. We therefore use only the right invariance of LTkL_T^k and obtain the other invariance by averaging in (3.1).

Proposition 3.3. For a∈C[G]a\in\mathbb{C}[G] and φ∈S(A)\varphi\in S(A),

LTk((id⁡⊗φ)Δ^(a))≤LTk(a).L_T^k((\operatorname{id}\otimes\varphi)\widehat{\Delta}(a))\leq L_T^k(a).

Thus LTkL_T^k is right-invariant.

Proof. Lemma 3.1 gives, on the algebraic core,

ad⁡(Dℓ⊗1)k(Δ^(a))=V(ad⁡(Dℓ)k(a)⊗1)V∗.\operatorname{ad}(D_\ell\otimes1)^k(\widehat{\Delta}(a))=V(\operatorname{ad}(D_\ell)^k(a)\otimes1)V^*.

The strong commutation with T⊗1T\otimes1 gives

(Tk⊗1)ad⁡(Dℓ⊗1)k(Δ^(a))(Tk⊗1)=V(Tkad⁡(Dℓ)k(a)Tk⊗1)V∗.(T^k\otimes1)\operatorname{ad}(D_\ell\otimes1)^k(\widehat{\Delta}(a))(T^k\otimes1)=V(T^k\operatorname{ad}(D_\ell)^k(a)T^k\otimes1)V^*.

Write Δ^(a)=∑j=1maj⊗bj\widehat{\Delta}(a)=\sum_{j=1}^{m}a_j\otimes b_j with aj,bj∈C[G]a_j,b_j\in\mathbb{C}[G]. The weighted commutator is then

∑j=1mTkad⁡(Dℓ)k(aj)Tk⊗bj∈B(L2(G))⊙A,\sum_{j=1}^{m}T^k\operatorname{ad}(D_\ell)^k(a_j)T^k\otimes b_j\in B(L^2(G))\odot A,

so the slice id⁡⊗φ\operatorname{id}\otimes\varphi is defined and contractive. Since it acts on the second leg, it commutes with the first-leg commutators and with multiplication by TkT^k. Consequently,

LTk((id⁡⊗φ)Δ^(a))=∥(id⁡⊗φ)(V(Tkad⁡(Dℓ)k(a)Tk⊗1)V∗)∥≤∥φ∥∥Tkad⁡(Dℓ)k(a)Tk∥=LTk(a),\begin{aligned} L_T^k((\operatorname{id}\otimes\varphi)\widehat{\Delta}(a)) &=\|(\operatorname{id}\otimes\varphi)(V(T^k\operatorname{ad}(D_\ell)^k(a)T^k\otimes1)V^*)\|\\ &\leq\|\varphi\|\|T^k\operatorname{ad}(D_\ell)^k(a)T^k\|\\ &=L_T^k(a), \end{aligned}

where the slice map is contractive because φ\varphi is a state. □

Define, for a∈C[G]saa\in\mathbb{C}[G]_{\mathrm{sa}},

L~Tk(a)=sup⁡ω∈S(A)LTk((ω⊗id⁡)Δ^(a)).(6)\widetilde{L}_T^k(a)=\sup_{\omega\in S(A)}L_T^k((\omega\otimes\operatorname{id})\widehat{\Delta}(a)). \tag*{(6)}

Theorem 3.4. The seminorm L~Tk\widetilde{L}_T^k is finite on C[G]sa\mathbb{C}[G]_{\mathrm{sa}}, is bi-invariant, and is a Lip-norm on AA.

Proof. For a∈C[G]a\in\mathbb{C}[G], write the finite algebraic coproduct as

Δ^(a)=∑j=1maj⊗bj\widehat{\Delta}(a)=\sum_{j=1}^{m}a_j\otimes b_j

with aj,bj∈C[G]a_j,b_j\in\mathbb{C}[G]. For every state ω\omega,

(ω⊗id⁡)Δ^(a)=∑jω(aj)bj∈C[G].(\omega\otimes\operatorname{id})\widehat{\Delta}(a)=\sum_j\omega(a_j)b_j\in\mathbb{C}[G].

The triangle inequality for L~Tk\widetilde{L}_{T}^{k} gives

L~Tk(a)≤∑j∥aj∥L~Tk(bj)<∞.\widetilde{L}_{T}^{k}(a) \leq\sum_{j} \lVert a_j\rVert\widetilde{L}_{T}^{k}(b_j) < \infty.

Thus L~Tk\widetilde{L}_{T}^{k} is a seminorm. Since the counit is a state and (ϵ⊗id⁡)Δ^=id⁡(\epsilon\otimes\operatorname{id})\widehat{\Delta}=\operatorname{id}, we have

LTk(a)≤L~Tk(a).(7)L_{T}^{k}(a) \leq\widetilde{L}_{T}^{k}(a). \tag*{(7)}

For ω,μ∈S(A)\omega,\mu\in S(A), put

αω=(ω⊗id⁡)Δ^,βμ=(id⁡⊗μ)Δ^.\alpha_{\omega}=(\omega\otimes\operatorname{id})\widehat{\Delta},\qquad\beta_{\mu}=(\operatorname{id}\otimes\mu)\widehat{\Delta}.

Coassociativity gives

αωαμ=αμ∗ω,αωβμ=βμαω,\alpha_{\omega}\alpha_{\mu}=\alpha_{\mu*\omega},\qquad\alpha_{\omega}\beta_{\mu}=\beta_{\mu}\alpha_{\omega},

where

μ∗ω=(μ⊗ω)Δ^.\mu*\omega=(\mu\otimes\omega)\widehat{\Delta}.

Consequently,

L~Tk(αμ(a))=sup⁡ω∈S(A)LTk(αμ∗ω(a))≤sup⁡ν∈S(A)LTk(αν(a))=L~Tk(a),\begin{aligned} \widetilde{L}_{T}^{k}(\alpha_{\mu}(a)) &= \sup_{\omega\in S(A)} L_{T}^{k}(\alpha_{\mu*\omega}(a))\\ &\leq\sup_{\nu\in S(A)} L_{T}^{k}(\alpha_{\nu}(a))\\ &=\widetilde{L}_{T}^{k}(a), \end{aligned}

since μ∗ω∈S(A)\mu*\omega\in S(A) for every ω∈S(A)\omega\in S(A). On the other hand, right invariance of LTkL_{T}^{k} gives

L~Tk(βμ(a))=sup⁡ω∈S(A)LTk(αωβμ(a))=sup⁡ω∈S(A)LTk(βμαω(a))≤sup⁡ω∈S(A)LTk(αω(a))=L~Tk(a).\begin{aligned} \widetilde{L}_{T}^{k}(\beta_{\mu}(a)) &= \sup_{\omega\in S(A)} L_{T}^{k}(\alpha_{\omega}\beta_{\mu}(a))\\ &=\sup_{\omega\in S(A)} L_{T}^{k}(\beta_{\mu}\alpha_{\omega}(a))\\ &\leq\sup_{\omega\in S(A)} L_{T}^{k}(\alpha_{\omega}(a))\\ &=\widetilde{L}_{T}^{k}(a). \end{aligned}

Thus L~Tk\widetilde{L}_{T}^{k} is bi-invariant.

Moreover, (3.2) gives

ker⁡L~Tk⊆ker⁡LTk=R1.\ker\widetilde{L}_{T}^{k}\subseteq\ker L_{T}^{k}=\mathbb{R}1.

Since L~Tk(1)=0\widetilde{L}_{T}^{k}(1)=0, its kernel is exactly R1\mathbb{R}1.

Fix a state ω0\omega_{0} and consider the centered unit ball

EL~={a∈C[G]sa:L~Tk(a)≤1, ω0(a)=0}.E_{\widetilde{L}}=\{a\in\mathbb{C}[G]_{\mathrm{sa}}:\widetilde{L}_{T}^{k}(a)\leq1,\ \omega_{0}(a)=0\}.

By (3.2), this set is contained in the corresponding centered unit ball for LTkL_{T}^{k}, which is norm totally bounded by Theorem 2.1. Rieffel’s criterion therefore proves that L~Tk\widetilde{L}_{T}^{k} is a Lip-norm. □

The coactions on the compressed operator systems

We use the standard operator-system definition: a right coaction is a unital completely isometric map α:X→X⊗A\alpha: X \to X \otimes A satisfying coassociativity and span⁡{(1X⊗A)α(X)}=X⊗A\operatorname{span}\{(1_X \otimes A)\alpha(X)\} = X \otimes A; the left-handed definition is analogous. Two coactions cocommute when the two iterated maps agree.

All tensor products below are the concrete minimal operator-system tensor products. The fixed-point space of a right coaction is

Xα={x∈X:α(x)=x⊗1};X^\alpha= \{x \in X : \alpha(x) = x \otimes1\};

for a left coaction β:X→A⊗X\beta: X \to A \otimes X it is defined by β(x)=1⊗x\beta(x) = 1 \otimes x. Ergodicity means that the relevant fixed-point space consists only of scalar multiples of the order unit.

Proposition 3.5 (Compressed coactions). For every R>0R > 0, the formulas

αR(x)=V(x⊗1)V∗,βR(x)=W∗(1⊗x)W,x∈AR,\alpha_R(x) = V(x \otimes1)V^*, \qquad\beta_R(x) = W^*(1 \otimes x)W, \qquad x \in A_R,

define completely isometric right and left coactions

αR:AR→AR⊗A,βR:AR→A⊗AR.\alpha_R : A_R \to A_R \otimes A, \qquad\beta_R : A_R \to A \otimes A_R.

They satisfy the Podleś density conditions, are ergodic, and cocommute:

(βR⊗id⁡)αR=(id⁡⊗αR)βR.(8)(\beta_R \otimes\operatorname{id})\alpha_R = (\operatorname{id} \otimes\alpha_R)\beta_R. \tag*{(8)}

Proof. The projection PRP_R is the Peter–Weyl projection associated with ΛR={α:ℓ(α)≤R}\Lambda_R = \{\alpha: \ell(\alpha) \le R\}. The general compression construction gives the unital completely isometric coactions αR\alpha_R and βR\beta_R, their coassociativity, and the Podleś density conditions; see [5 Proposition 7.4]. We verify the identities and additional properties needed here.

By Lemma 3.1 and the spectral theorem,

[PR⊗1,V]=0,[1⊗PR,W]=0.[P_R \otimes1,V] = 0, \qquad[1 \otimes P_R,W] = 0.

Hence the formulas for αR\alpha_R and βR\beta_R take values in AR⊗AA_R \otimes A and A⊗ARA \otimes A_R, respectively. For a∈Aa \in A,

αR(τR(a))=(τR⊗id⁡)Δ^(a),βR(τR(a))=(id⁡⊗τR)Δ^(a).(9)\alpha_R(\tau_R(a)) = (\tau_R \otimes\operatorname{id})\widehat{\Delta}(a), \qquad\beta_R(\tau_R(a)) = (\operatorname{id} \otimes\tau_R)\widehat{\Delta}(a). \tag*{(9)}

These identities are independent of the choice of lift because τR\tau_R is surjective and the compressed coproducts are well-defined.

Let h^\widehat{h} be the Haar state of AA. If αR(x)=x⊗1\alpha_R(x) = x \otimes1, write x=τR(a)x = \tau_R(a). Applying id⁡⊗h^\operatorname{id} \otimes\widehat{h} to (9) gives

x=(id⁡⊗h^)αR(x)=τR((id⁡⊗h^)Δ^(a))=h^(a)PR.x = (\operatorname{id} \otimes\widehat{h})\alpha_R(x) = \tau_R((\operatorname{id} \otimes\widehat{h})\widehat{\Delta}(a)) = \widehat{h}(a)P_R.

Thus ARαR=CPRA_R^{\alpha_R} = \mathbb{C}P_R. The same argument, applying h^⊗id⁡\widehat{h} \otimes\operatorname{id}, gives ARβR=CPRA_R^{\beta_R} = \mathbb{C}P_R.

Finally, for x=τR(a)x = \tau_R(a), compatibility gives

(βR⊗id⁡)αR(x)=(id⁡⊗τR⊗id⁡)(Δ^⊗id⁡)Δ^(a),(\beta_R \otimes\operatorname{id})\alpha_R(x) = (\operatorname{id} \otimes\tau_R \otimes\operatorname{id})(\widehat{\Delta} \otimes\operatorname{id})\widehat{\Delta}(a),
(id⁡⊗αR)βR(x)=(id⁡⊗τR⊗id⁡)(id⁡⊗Δ^)Δ^(a).(\operatorname{id} \otimes\alpha_R)\beta_R(x) = (\operatorname{id} \otimes\tau_R \otimes\operatorname{id})(\operatorname{id} \otimes\widehat{\Delta})\widehat{\Delta}(a).

The two expressions agree by coassociativity of Δ^\widehat{\Delta}. This proves ergodicity and cocommutativity. □

Remark 3.6. The existence of the compressed coactions is part of the general Peter–Weyl compression framework of [5]. The points used specifically below are the compatibility identities (3.4), the ergodicity of the two coactions, and their cocommutativity for the length-filtered family (PR)R>0(P_R)_{R>0}. Properness ensures that PRP_R has finite rank and that τR(C[G])=AR\tau_R(\mathbb{C}[G]) = A_R.

Lemma 3.7. Let ϵ\epsilon be the bounded counit of AA. If γ:X→X⊗A\gamma: X \to X \otimes A is an injective right coaction, then

(id⁡X⊗ϵ)γ=id⁡X.(\operatorname{id}_X \otimes\epsilon)\gamma= \operatorname{id}_X.

If η:X→A⊗X\eta: X \to A \otimes X is an injective left coaction, then

(ϵ⊗id⁡X)η=id⁡X.(\epsilon\otimes\operatorname{id}_X)\eta= \operatorname{id}_X.

Proof. For a right coaction, coassociativity gives

γ((id⁡X⊗ϵ)γ(x))=(id⁡X⊗id⁡A⊗ϵ)(id⁡X⊗Δ^)γ(x)=γ(x),\gamma((\operatorname{id}_X \otimes\epsilon)\gamma(x)) = (\operatorname{id}_X \otimes\operatorname{id}_A \otimes\epsilon)(\operatorname{id}_X \otimes\widehat{\Delta})\gamma(x) = \gamma(x),

where the last equality uses (id⁡A⊗ϵ)Δ^=id⁡A(\operatorname{id}_A \otimes\epsilon)\widehat{\Delta} = \operatorname{id}_A. Injectivity of γ\gamma proves the first identity. For a left coaction, the same argument, using (ϵ⊗id⁡A)Δ^=id⁡A(\epsilon\otimes\operatorname{id}_A)\widehat{\Delta} = \operatorname{id}_A, proves the second. □

The truncation Lip-norm and convergence

For x∈(AR)sax \in(A_R)_{\mathrm{sa}} define

LRr(x)=sup⁡ψ∈S(AR)L~Tk((ψ⊗id⁡)αR(x)),L_R^r(x) = \sup_{\psi\in S(A_R)} \widetilde{L}_T^k((\psi\otimes\operatorname{id})\alpha_R(x)),
LRℓ(x)=sup⁡ϕ∈S(AR)L~Tk((id⁡⊗ϕ)βR(x)),L_R^\ell(x) = \sup_{\phi\in S(A_R)} \widetilde{L}_T^k((\operatorname{id} \otimes\phi)\beta_R(x)),
LR(x)=max⁡{LRr(x),LRℓ(x)}.(10)L_R(x) = \max\{L_R^r(x), L_R^\ell(x)\}. \tag*{(10)}

For ϕ∈S(AR)\phi\in S(A_R), let

ρRϕ(x)=(id⁡⊗ϕ)βR(x),x∈AR.(11)\rho_R^\phi(x) = (\operatorname{id} \otimes\phi)\beta_R(x), \qquad x \in A_R. \tag*{(11)}

Proposition 3.8. LRL_R is a Lip-norm on the finite-dimensional operator system ARA_R. Moreover, for a∈C[G]saa \in\mathbb{C}[G]_{\mathrm{sa}} and x∈(AR)sax \in(A_R)_{\mathrm{sa}},

LR(τR(a))≤L~Tk(a),L~Tk(ρRϕ(x))≤LR(x).(12)L_R(\tau_R(a)) \leq\widetilde{L}_T^k(a), \qquad\widetilde{L}_T^k(\rho_R^\phi(x)) \leq L_R(x). \tag*{(12)}

Proof. Choose a∈C[G]a \in\mathbb{C}[G] with x=τR(a)x = \tau_R(a) and write Δ^(a)=∑j=1maj⊗bj\widehat{\Delta}(a) = \sum_{j=1}^{m} a_j \otimes b_j with aj,bj∈C[G]a_j,b_j \in\mathbb{C}[G]. By (3.4),

(ψ⊗id⁡)αR(x)=∑j=1mψ(τR(aj))bj,(id⁡⊗ϕ)βR(x)=∑j=1mϕ(τR(bj))aj.(\psi\otimes\operatorname{id})\alpha_R(x) = \sum_{j=1}^{m}\psi(\tau_R(a_j))b_j,\qquad (\operatorname{id} \otimes\phi)\beta_R(x) = \sum_{j=1}^{m}\phi(\tau_R(b_j))a_j.

Consequently,

LRr(x)≤∑j=1m∥aj∥L~Tk(bj),L_R^r(x) \leq\sum_{j=1}^{m}\lVert a_j\rVert\widetilde{L}_T^k(b_j),
LRℓ(x)≤∑j=1m∥bj∥L~Tk(aj).L_R^\ell(x) \leq\sum_{j=1}^{m}\lVert b_j\rVert\widetilde{L}_T^k(a_j).

In particular, both suprema in (3.5) are finite. If LRr(x)=0L_R^r(x) = 0, then

L~Tk((ψ⊗id⁡)αR(x))=0for all ψ∈S(AR).\widetilde{L}_T^k((\psi\otimes\operatorname{id})\alpha_R(x)) = 0 \quad\text{for all } \psi\in S(A_R).

Since ker⁡L~Tk=R1\ker\widetilde{L}_T^k = \mathbb{R}1, every such slice is scalar. States separate points of ARA_R, so αR(x)∈AR⊗C1\alpha_R(x) \in A_R \otimes\mathbb{C}1. The right counit identity from Lemma 3.7 gives αR(x)=x⊗1\alpha_R(x) = x \otimes1, and ergodicity gives x∈RPRx \in\mathbb{R}P_R. Conversely, unitality shows that every scalar multiple of PRP_R lies in the kernel. The same argument for βR\beta_R gives

ker⁡LRr=ker⁡LRℓ=RPR,ker⁡LR=RPR.\ker L_R^r = \ker L_R^\ell= \mathbb{R}P_R,\qquad\ker L_R = \mathbb{R}P_R.

A finite-dimensional seminorm with scalar kernel is a Lip-norm, so the first assertion follows. For a∈C[G]saa \in\mathbb{C}[G]_{\mathrm{sa}}, the compatibility identities for the coactions and the bi-invariance of L~Tk\widetilde{L}_{T}^{k} give

LR(τR(a))≤L~Tk(a),LRℓ(τR(a))≤L~Tk(a).L_{R}(\tau_{R}(a)) \leq\widetilde{L}_{T}^{k}(a), \qquad L_{R}^{\ell}(\tau_{R}(a)) \leq\widetilde{L}_{T}^{k}(a).

Taking the maximum proves the first inequality in (12). The second follows immediately from the definition of LRℓL_{R}^{\ell} and (11). The symbol map is ucp, as it is a composition of a coaction and a state slice. □

Lemma 3.9. For a∈C[G]saa \in\mathbb{C}[G]_{\mathrm{sa}}, x∈(AR)sax \in(A_{R})_{\mathrm{sa}} and μ,ν∈S(A)\mu,\nu\in S(A),

∥(id⁡⊗μ)Δ^(a)−(id⁡⊗ν)Δ^(a)∥≤dL~Tk(μ,ν)L~Tk(a),(13)\left\|(\operatorname{id} \otimes\mu)\widehat{\Delta}(a)-(\operatorname{id} \otimes\nu)\widehat{\Delta}(a)\right\| \leq d_{\widetilde{L}_{T}^{k}}(\mu,\nu)\widetilde{L}_{T}^{k}(a), \tag*{(13)}
∥(id⁡⊗μ)αR(x)−(id⁡⊗ν)αR(x)∥≤dL~Tk(μ,ν)LRr(x).(14)\left\|(\operatorname{id} \otimes\mu)\alpha_{R}(x)-(\operatorname{id} \otimes\nu)\alpha_{R}(x)\right\| \leq d_{\widetilde{L}_{T}^{k}}(\mu,\nu)L_{R}^{r}(x). \tag*{(14)}

Proof. For (13), set z=(id⁡⊗μ)Δ^(a)−(id⁡⊗ν)Δ^(a)z=(\operatorname{id}\otimes\mu)\widehat{\Delta}(a)-(\operatorname{id}\otimes\nu)\widehat{\Delta}(a). Since z=z∗z=z^{*},

∥z∥=sup⁡ω∈S(A)∣ω(z)∣.\|z\|=\sup_{\omega\in S(A)}|\omega(z)|.

For each ω\omega, put yω=(ω⊗id⁡)Δ^(a)y_{\omega}=(\omega\otimes\operatorname{id})\widehat{\Delta}(a). Then yω=yω∗y_{\omega}=y_{\omega}^{*} and left invariance gives L~Tk(yω)≤L~Tk(a)\widetilde{L}_{T}^{k}(y_{\omega})\leq\widetilde{L}_{T}^{k}(a). Hence

∣ω(z)∣=∣μ(yω)−ν(yω)∣≤dL~Tk(μ,ν)L~Tk(a).|\omega(z)|=|\mu(y_{\omega})-\nu(y_{\omega})|\leq d_{\widetilde{L}_{T}^{k}}(\mu,\nu)\widetilde{L}_{T}^{k}(a).

Taking the supremum gives (13). For (14), put zR=(id⁡⊗μ)αR(x)−(id⁡⊗ν)αR(x)z_{R}=(\operatorname{id}\otimes\mu)\alpha_{R}(x)-(\operatorname{id}\otimes\nu)\alpha_{R}(x). For each ψ∈S(AR)\psi\in S(A_{R}), let yψ=(ψ⊗id⁡)αR(x)y_{\psi}=(\psi\otimes\operatorname{id})\alpha_{R}(x). Then L~Tk(yψ)≤LRr(x)\widetilde{L}_{T}^{k}(y_{\psi})\leq L_{R}^{r}(x), and the same argument, now using ∥zR∥=sup⁡ψ∈S(AR)∣ψ(zR)∣\|z_{R}\|=\sup_{\psi\in S(A_{R})}|\psi(z_{R})|, proves (14). □

Theorem 3.10 (Quantum Gromov–Hausdorff convergence). Let GG be an amenable discrete quantum group with modified property RD for a proper central length ℓ\ell, and choose an integer k>sk>s as in Theorem 2.1. Equip A=C∗(G)A=C^{*}(G) with L~Tk\widetilde{L}_{T}^{k} and AR=PRAPRA_{R}=P_{R}AP_{R} with LRL_{R}. Then

dist⁡q((A,L~Tk),(AR,LR))⟶0(R→∞).\operatorname{dist}_{q}\left((A,\widetilde{L}_{T}^{k}),(A_{R},L_{R})\right)\longrightarrow0 \qquad(R\to\infty).

Proof. We first construct states ϕR∈S(AR)\phi_{R}\in S(A_{R}) such that

μR:=ϕR∘τR⟶ϵin the weak∗ topology.\mu_{R}:=\phi_{R}\circ\tau_{R}\longrightarrow\epsilon\quad\text{in the weak}^{*}\text{ topology}.

Since λ\lambda is faithful, the convex hull of its vector states is weak∗^{*} dense in S(A)S(A). The metric dL~Tkd_{\widetilde{L}_{T}^{k}} induces the weak∗^{*} topology. Hence, for each nn, choose a finite convex combination ω(n)\omega^{(n)} of vector states with

dL~Tk(ω(n),ϵ)<12n.d_{\widetilde{L}_{T}^{k}}(\omega^{(n)},\epsilon)<\frac{1}{2n}.

Write the vectors implementing this convex combination as ξ1,…,ξm\xi_{1},\ldots,\xi_{m}. Since PR→IP_{R}\to I strongly, the normalized vectors

ξj(R)=PRξj∥PRξj∥\xi_{j}^{(R)}=\frac{P_{R}\xi_{j}}{\|P_{R}\xi_{j}\|}

are defined for all sufficiently large RR and converge to ξj\xi_{j}. The associated vector states converge weak∗^{*}, hence in dL~Tkd_{\widetilde{L}_{T}^{k}}, to the original vector states. Their finite convex combination ωR(n)\omega_{R}^{(n)} therefore converges to ω(n)\omega^{(n)} in the same metric. More explicitly, if ωR(n)\omega_{R}^{(n)} is implemented by ξ1(R),…,ξm(R)\xi_{1}^{(R)},\ldots,\xi_{m}^{(R)} with weights t1,…,tmt_{1},\ldots,t_{m}, then

ωR(n)(a)=∑j=1mtj⟨λ(a)ξj(R),ξj(R)⟩=ϕR(n)(τR(a))\omega_{R}^{(n)}(a)=\sum_{j=1}^{m}t_{j}\langle\lambda(a)\xi_{j}^{(R)},\xi_{j}^{(R)}\rangle=\phi_{R}^{(n)}(\tau_{R}(a))

for a state ϕR(n)\phi_R^{(n)} on ARA_R. Choose an increasing sequence Rn→∞R_n \to\infty such that

dL~Tk(ωR(n),ω(n))<12nd_{\widetilde{L}_T^k}(\omega_R^{(n)},\omega^{(n)}) < \frac{1}{2n}

whenever R≥RnR \ge R_n.

For Rn≤R<Rn+1R_n \le R < R_{n+1}, set ϕR=ϕR(n)\phi_R=\phi_R^{(n)}; the choice for R<R1R<R_1 is immaterial. Then

dR:=dL~Tk(ϕR∘τR,ϵ)<1n(Rn≤R<Rn+1),d_R:=d_{\widetilde{L}_T^k}(\phi_R\circ\tau_R,\epsilon)<\frac{1}{n}\qquad(R_n\le R<R_{n+1}),

so dR→0d_R\to0. Put ρR=ρRϕR\rho_R=\rho_R^{\phi_R}. By (12), τR\tau_R and ρR\rho_R are unital positive Lip-contractions.

For a∈C[G]saa\in C[G]_{\mathrm{sa}}, coaction compatibility gives

ρRτR(a)=(id⊗μR)Δ^(a).\rho_R\tau_R(a)=({\rm id}\otimes\mu_R)\widehat{\Delta}(a).

Since a=(id⊗ϵ)Δ^(a)a=({\rm id}\otimes\epsilon)\widehat{\Delta}(a), (13) yields

∥ρRτR(a)−a∥≤dRL~Tk(a).\|\rho_R\tau_R(a)-a\|\le d_R\widetilde{L}_T^k(a).

For x∈(AR)sax\in(A_R)_{\mathrm{sa}}, the two induced coactions satisfy (τR⊗id)βR=(id⊗τR)αR(\tau_R\otimes{\rm id})\beta_R=({\rm id}\otimes\tau_R)\alpha_R, and hence

τRρR(x)=(id⊗μR)αR(x).\tau_R\rho_R(x)=({\rm id}\otimes\mu_R)\alpha_R(x).

Using the right counit identity and (14), we obtain

∥τRρR(x)−x∥≤dRLR′(x)≤dRLR(x).\|\tau_R\rho_R(x)-x\|\le d_RL_R^{\prime}(x)\le d_RL_R(x).

Apply Proposition 2.2 with η=dR\eta=d_R. It follows that

dist⁡q((A,L~Tk),(AR,LR))≤dR→0.\operatorname{dist}_{q}\left((A,\widetilde{L}_T^k),(A_R,L_R)\right)\le d_R\to0.

Remark 3.11 (Comparison with Leimbach’s operator-Gromov–Hausdorff theorem). The convergence mechanism in Theorem 3.10 is closely related to, but should be distinguished from, the general Peter–Weyl convergence theorem of Leimbach [5 Theorem 7.13]. We verify that its hypotheses are satisfied for the family at hand.

First, amenability of GG is equivalent to coamenability of its compact dual G^\widehat{G}. Properness of ℓ\ell implies that

Irr⁡(G^)=⋃n=1∞{α:ℓ(α)≤n},\operatorname{Irr}(\widehat{G})=\bigcup_{n=1}^{\infty}\{\alpha:\ell(\alpha)\le n\},

a countable union of finite sets. Hence A=C(G^)A=C(\widehat{G}) is separable. Moreover, if R≤SR\le S, then

PR≤PS,PR∨PS=PS,P_R\le P_S,\qquad P_R\vee P_S=P_S,

and

PR⟶1strongly as R⟶∞.P_R\longrightarrow1\quad\text{strongly as }R\longrightarrow\infty.

Thus (PR)R>0(P_R)_{R>0} is an increasing Peter–Weyl net of the type required in [5 Theorem 7.13].

Second, Theorem 3.4 shows that L~Tk\widetilde{L}_T^k is a bi-invariant Lip-norm. It is regular in the sense used there, since its domain contains the dense Hopf ∗*-algebra C[G]C[G]. Finally, the seminorm LRL_R in (10) is precisely the maximum of the seminorms induced by the compressed right and left coactions, which is the operator-system Lip-norm appearing in [5].

Consequently, all hypotheses of [5 Theorem 7.13] are satisfied, and we obtain the stronger convergence

dist⁡op((AR,LR),(A,L~Tk))⟶0.\operatorname{dist}_{\mathrm{op}}\left((A_R,L_R),(A,\widetilde{L}_T^k)\right)\longrightarrow0.

The role of the present paper is to construct the required regular bi-invariant Lip-norm from the modified RD property of an amenable discrete quantum group, with the modular correction retained in the non-unimodular case. We also give a proper-length version of the BVZ Lip-norm criterion, so that finite generation is not imposed on the discrete quantum group. The abstract Peter–Weyl convergence principle itself is not claimed as a new result here. The direct estimates in Theorem 3.10 additionally give the explicit order-unit quantum Gromov–Hausdorff bound dRd_R.

Examples

Example 4.1 (Groups of polynomial growth). Let Γ\Gamma be a finitely generated discrete group of polynomial growth, with a proper word length ℓ\ell. Such a group is amenable and has RD; if its growth degree is dd, one may take any RD exponent s>d/2s > d/2. Since Γ\Gamma is unimodular, T=1T = 1. Thus the theorem applies to A=Cr∗(Γ)A = C_r^*(\Gamma), including Γ=Zn\Gamma= \mathbb{Z}^n and the discrete Heisenberg group

H3(Z)={(1ac01b001):a,b,c∈Z},H_3(\mathbb{Z}) = \left\{ \begin{pmatrix} 1 & a & c \\ 0 & 1 & b \\ 0 & 0 & 1 \end{pmatrix} : a,b,c \in\mathbb{Z} \right\},

which has polynomial growth of degree four and admits an RD exponent s>2s > 2.

Example 4.2 (Non-Kac examples). Let KK be a simply connected compact semisimple Lie group and KqK_q its standard deformation, 0<q<10 < q < 1. By [1 Proposition 4.3], the discrete dual G=Kq^G = \widehat{K_q} is amenable and has the modified property RD. Since KqK_q is non-Kac for q≠1q \ne1, this gives non-unimodular examples covered by the theorem.

For Kq=SUq(2)K_q = SU_q(2) the irreducibles are indexed by n∈N0n \in\mathbb{N}_0, with ordinary dimension n+1n + 1, quantum dimension [n+1]q[n + 1]_q, and standard proper length ℓ(n)=n\ell(n) = n. In a suitable orthonormal basis the modular matrix is diagonal with eigenvalues q−n,q−n+2,…,qnq^{-n}, q^{-n+2}, \ldots, q^n (up to reversing the basis order). Thus

Cn=[n+1]qn+1diag⁡(q−n,q−n+2,…,qn),Tn=Cn1/(2k).C_n = \frac{[n+1]_q}{n+1}\operatorname{diag}(q^{-n},q^{-n+2},\ldots,q^n), \qquad T_n = C_n^{1/(2k)}.

The modular correction is nontrivial for q≠1q \ne1, so this example tests the genuinely non-Kac part of the construction. The same argument applies to SUq(3)SU_q(3) and, more generally, to the standard deformations covered by [1 Proposition 4.3].

Remark 4.3. The group examples above are unimodular, so F=C=T=1F = C = T = 1 and the construction reduces to the higher commutator seminorm associated with the length operator. The compact semisimple qq-deformations show that the argument is not restricted to the tracial case: the row-dependent modular weights are retained in the right-invariance calculation and are handled by the two-sided truncation seminorm.

References

  1. [1]J. Bhowmick, C. Voigt and J. Zacharias, Compact quantum metric spaces from quantum groups of rapid decay, J. Noncommut. Geom. 9 (2015), no. 4, 1175–1200.arxiv.org/abs/1406.0771
  2. [2]A. Connes and W. D. van Suijlekom, Spectral truncations in noncommutative geometry and operator systems, Commun. Math. Phys. 383 (2021), 2021–2067.DOI
  3. [3]Y. Gaudillot-Estrada and W. D. van Suijlekom, Convergence of spectral truncations for compact metric groups, Int. Math. Res. Not. IMRN (2025), no. 13, rnaf197.
  4. [4]J. Kaad and D. Kyed, The Quantum Metric Structure of Quantum SU(2), Memoirs of the European Mathematical Society, vol. 18, EMS Press, 2025.arxiv.org/abs/2205.06043
  5. [5]M. Leimbach, Convergence of Petrov–Weyl truncations of compact quantum groups, J. Noncommut. Geom. (2025), doi:10.4171/JNCG/634.DOI
  6. [6]M. Leimbach and W. D. van Suijlekom, Gromov–Hausdorff convergence of spectral truncations for tori, Adv. Math. 439 (2024), 109496.DOI
  7. [8]X. Peng and Q. Wang, Gromov–Hausdorff Convergence of Spectral Truncations for Quantum Groups, arXiv:2604.27475 [math.OA], 2026.arxiv.org/abs/2604.27475
  8. [9]M. A. Rieffel, Compact quantum metric spaces, in Operator Algebras, Quantization, and Noncommutative Geometry, Contemp. Math., vol. 365, Amer. Math. Soc., 2004, pp. 315–330.
  9. [10]M. A. Rieffel, Convergence of Fourier truncations for compact quantum groups and finitely generated groups, J. Geom. Phys. 192 (2023), 104921.DOI
  10. [11]W. D. van Suijlekom, Gromov–Hausdorff convergence of state spaces for spectral truncations, J. Geom. Phys. 162 (2021), 104075.DOI
  11. [12]R. Vergnioux, The property of rapid decay for discrete quantum groups, J. Operator Theory 57 (2007), no. 2, 303–324.

Paper details

vlnc:W_y6mfe3g6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents