Plectis Atlas
Search

↑ and ↓ move, Enter opens, Esc closes.

Reading view

Close Separable C*-Algebras Without Spatial Conjugacy

OpenAI, 5 October 2026

PDFTeX source

Part of family 289: Strong Kadison–Kastler stability and its spatial boundaries. Read it with its family

This page is converted from the paper’s TeX source. Atlas reconstructs statement numbers from the paper’s typesetting source (TeX). Use the original PDF if they differ. One figure is shown as a placeholder; see the PDF for it.

Abstract

We disprove the separable C^*-algebraic spatial form of the Kadison–Kastler conjecture. For every \varepsilon>0, we construct unital, norm-separable C^*-algebras on a common separable complex Hilbert space, with the same identity and Kadison–Kastler distance less than \varepsilon, that are not conjugate by any unitary. The two algebras have the same von Neumann closure.

1 Introduction

For C^*-algebras A,B\subseteq\B(H), write A_1 and B_1 for their closed unit balls. Their Kadison–Kastler distance is

\KK(A,B)=\max\left\{ \sup_{a\in A_1}\inf_{b\in B_1}\norm{a-b},\, \sup_{b\in B_1}\inf_{a\in A_1}\norm{b-a}\right\}. \tag{1}

The spatial form of the perturbation problem asks whether sufficiently close algebras must be related by uAu^*=B for a unitary u\in\B(H). Kadison and Kastler initiated this problem in [6]. In its separable C^*-algebraic form, one asks for a constant \delta>0 valid for all norm-separable C^*-algebras on separable Hilbert spaces; no bound on \norm{u-\id} is required. The following theorem resolves negatively this separable C^*-algebraic spatial form of the Kadison–Kastler conjecture, even for unital algebras with a common identity.

Theorem 1.1. For every \varepsilon>0 there are a separable complex Hilbert space H and unital, norm-separable C^*-algebras A,B\subseteq\B(H) with common identity \id_H such that

\KK(A,B)<\varepsilon,\qquad A''=B'',

but no unitary u\in\B(H) satisfies uAu^*=B.

The scope of this statement is important. Johnson constructed arbitrarily close separable nuclear algebras for which implementing unitaries cannot be chosen close to the identity [5]; those examples are nevertheless spatially conjugate. Choi and Christensen constructed arbitrarily close nonisomorphic algebras [2], but norm nonseparability is essential to those examples. These distinctions, and the resulting separable spatial formulation, are discussed in [3]. In the positive direction, Christensen, Sinclair, Smith, White, and Winter proved spatial conjugacy for sufficiently close separable nuclear C^*-algebras on a separable Hilbert space [3, Theorem 4]; see [4, Theorem 5.4] for the full proof. Theorem 1.1 imposes no nuclearity assumption. It makes no assertion of abstract nonisomorphism and does not give a counterexample for von Neumann algebras.

The construction. We first isolate a way to turn an obstruction for inclusions into an obstruction to arbitrary spatial conjugacy. Let D\subseteq\B(L) be unital, norm-separable, and have scalar center. For unital subalgebras P,Q\subseteq D, consider the algebra

\E_D(P)=\{(x_n):x_n\in D,\ x_n\longrightarrow p\in P \text{ in norm}\}

acting diagonally on \bigoplus_{n\ge1}L. Changing P changes only the condition at the limit. The coordinate projections are intrinsic minimal central projections of this algebra. Consequently, a unitary conjugating \E_D(P) onto \E_D(Q) must permute the coordinates and normalize D on every coordinate. Pulling back constant sequences then forces each finite family in Q to be a simultaneous norm limit of families in unitary conjugates of P by normalizers of D.

Write D' for the commutant of D on L, consisting of the bounded operators commuting with every element of D. The obstruction we use is the equality

\norm{\sum_i p_i z_i}_{\B(L)} =\norm{\sum_i p_i\otimes z_i}_{\B(L\otimes L)}, \qquad p_i\in P,\quad z_i\in D'. \tag{2}

This equality is invariant under normalizers of D and passes to simultaneous finite-family norm limits. We construct very close subalgebras P_0,P_t\subseteq D for which it holds for P_0 but fails for P_t. In fact, P_0 and P_t themselves are conjugate by a unitary close to the identity. The obstruction concerns their positions inside D, and the sequence construction makes it an obstruction to every spatial conjugacy of the resulting algebras.

For the close inclusions we use a real-parameter deformation of the regular representation of the free group on countably many generators. This representation-theoretic ingredient belongs to the Pytlik–Szwarc family [7]; its uniform-topology behavior at the regular representation is discussed by Burger, Ozawa, and Thom [1, Theorem 4.6]. We give a direct tree construction and prove the uniform estimate needed here. The additional mechanism is the combination of the relative tensor-norm test with the convergent-sequence reduction. The reduction is stated separately, so it can be used for other pairs of inclusions.

Section 2 proves the reduction. Section 3 constructs the representations, and Section 4 establishes the tensor-norm obstruction. Section 5 assembles the counterexamples. All Hilbert spaces are complex, and all operator tensor products below act on the indicated Hilbert-space tensor products. No abstract choice of C^*-tensor norm is needed. Unital subalgebras are understood to contain the identity of their ambient operator algebra.

2 A convergent-sequence obstruction

We explain how the limit condition records an inclusion P\subseteq D. The resulting algebras preserve the distance between limit algebras, but their spatial conjugacies are constrained by the coordinate algebra D.

Proposition 2.1. Let L be a nonzero separable Hilbert space, let D\subseteq\B(L) be a unital norm-separable C^*-algebra with Z(D)=\C\id_L, and let P,Q\subseteq D be unital C^*-subalgebras containing \id_L. On H=\bigoplus_{n\ge1}L, define

\E_D(P)=\{(x_n)_{n\ge1}:x_n\in D,\ x_n\to p\in P\text{ in norm}\}, \tag{3}

and define \E_D(Q) similarly. Then:

  1. 1.

    These are unital norm-separable C^*-algebras with common identity, and

    \KK(\E_D(P),\E_D(Q))=\KK(P,Q). \tag{4}
  2. 2.

    If a unitary conjugates \E_D(P) onto \E_D(Q), there are unitaries U_n\in\B(L) with U_nDU_n^*=D such that, for every b\in Q, some p_b\in P satisfies

    \norm{b-U_np_bU_n^*}\longrightarrow0. \tag{5}

    The same sequence (U_n) works for every b.

  3. 3.

    The two algebras have the same commutant, and hence the same von Neumann closure.

Proof. A norm-convergent sequence is bounded, so (3) defines a diagonal operator on H, with norm \sup_n\norm{x_n}. Algebra operations and adjoints preserve the limit condition, and the constant identity belongs to the algebra. To check norm closedness, suppose x^{(k)}\to x in the supremum norm, with coordinate limits p_k\in P. Then

\norm{p_k-p_j}\le\sup_n\norm{x_n^{(k)}-x_n^{(j)}}.

Thus p_k\to p\in P. The estimate

\norm{x_n-p}\le\norm{x_n-x_n^{(k)}} +\norm{x_n^{(k)}-p_k}+\norm{p_k-p}

shows that x_n\to p. Moreover,

\E_D(P)=c_0(\N,D)+\{(p,p,\ldots):p\in P\}. \tag{6}

The two summands are norm-separable: P is a subspace of the separable metric space D, and finite-support sequences from a countable dense subset of D are dense in c_0(\N,D). Their sum is norm-separable.

For the metric assertion, write \alpha(P,Q)=\sup_{p\in P_1}\dist(p,Q_1). Given a contraction x\in\E_D(P) with limit p and \eta>0, choose q\in Q_1 with \norm{p-q}<\alpha(P,Q)+\eta, and choose N so that \sup_{n>N}\norm{x_n-p}<\eta. Keeping the first N entries of x and putting q thereafter gives a contraction y\in\E_D(Q) with \norm{x-y}<\alpha(P,Q)+2\eta. This proves the upper bound for the directed distance. Conversely, for a constant contraction (p,p,\ldots) and any contraction y\in\E_D(Q) with limit q\in Q_1,

\norm{(p,p,\ldots)-y}=\sup_n\norm{p-y_n}\ge\norm{p-q} \ge\dist(p,Q_1).

Taking infima and suprema proves equality of the directed distances. Interchanging P,Q gives (4).

Let e_n be the projection onto the nth copy of L. Every e_n is a central element of \E_D(P). Conversely, a central element (x_n) commutes with all finite-support D-valued sequences, so each x_n lies in Z(D)=\C\id_L. Thus the center consists exactly of convergent scalar sequences. Every nonzero central projection contains an e_n, and hence the minimal nonzero central projections are precisely the e_n. The same holds for \E_D(Q).

If S\E_D(P)S^*=\E_D(Q), conjugation by S therefore permutes the projections e_n. Its restriction from the nth coordinate to its image is a unitary U_n:L\to L. The corresponding coordinate corners of both algebras are exactly D, since every finitely supported D-valued sequence is allowed. Consequently U_nDU_n^*=D. For b\in Q, pull back the constant sequence (b,b,\ldots) under S. Its nth entry is U_n^*bU_n, regardless of the coordinate permutation. It belongs to \E_D(P), so its entries converge in norm to some p_b\in P. This is equivalent to (5) and proves (ii).

Finally, an operator commuting with \E_D(P) commutes with all e_n, so it is block diagonal. Commutation with the finite-support D corners means that each block lies in D'. These conditions are also sufficient, and do not depend on P. Thus both commutants equal

\{(z_n):z_n\in D',\ \sup_n\norm{z_n}<\infty\}.

This proves (iii).

We next give a concrete condition preserved by the approximation in Proposition 2.1(ii).

Definition 2.2. For a unital inclusion P\subseteq D\subseteq\B(L), say that P has the spatial tensor-norm identity relative to D if

\norm{\sum_{i=1}^m p_i z_i}_{\B(L)} =\norm{\sum_{i=1}^m p_i\otimes z_i}_{\B(L\otimes L)} \tag{7}

for every positive integer m, all p_1,\ldots,p_m\in P, and all z_1,\ldots,z_m\in D'.

Lemma 2.3. Let P,Q\subseteq D\subseteq\B(L) be unital C^*-algebras. The identity in Definition 2.2 is preserved by conjugating P with any unitary normalizing D. If (U_n) is a sequence of unitaries in \B(L) satisfying U_nDU_n^*=D and every b\in Q admits the approximation (5), then the identity for P implies the identity for Q.

Proof. If UDU^*=D, then UD'U^*=D'. For q_i=Up_iU^* and z_i\in D', apply (7) to p_i and U^*z_iU. Conjugate its product sum by U and its tensor sum by U\otimes U. This gives (7) for q_i,z_i.

Now fix finite families b_i\in Q and z_i\in D'. Put q_{i,n}=U_np_{b_i}U_n^*. For each n, the first part gives the identity for the entire list (q_{i,n})_i, since its entries all lie in U_nPU_n^*. Both errors

\norm{\sum_i(q_{i,n}-b_i)z_i},\qquad \norm{\sum_i(q_{i,n}-b_i)\otimes z_i}

are at most \sum_i\norm{q_{i,n}-b_i}\norm{z_i}, which tends to zero. Taking the limit proves the identity for the prescribed families in Q and D'.

Corollary 2.4. Under the hypotheses of Proposition 2.1, suppose that P has the spatial tensor-norm identity relative to D, whereas Q does not. Then \E_D(P) and \E_D(Q) are not spatially conjugate. If W\in\B(L) is unitary and WPW^*=Q, then nevertheless

\KK(\E_D(P),\E_D(Q))\le2\norm{W-\id_L}.

Proof. A spatial conjugacy would give the unitaries in Proposition 2.1(ii), contradicting Lemma 2.3. For the distance bound, p\mapsto WpW^* bijects the closed unit balls and

\norm{WpW^*-p}\le2\norm{p}\norm{W-\id_L}.

Apply Proposition 2.1(i).

The unitary W used for the distance estimate is not required to normalize D. Indeed, under the stated tensor-norm obstruction, no unitary normalizing D can conjugate P onto Q.

3 A uniformly small free-group deformation

We now construct the representations used to produce the inclusions in Corollary 2.4. Let G be the free group on the countable set \{s_1,s_2,\ldots\}, let e denote its identity, and put K=\ell^2(G). On its usual orthonormal basis define

\lambda(g)\delta_x=\delta_{gx},\qquad \rho(g)\delta_x=\delta_{xg^{-1}}. \tag{8}

Both are unitary representations, and every left translation commutes with every right translation. The group G is countable, so K is separable.

The following elementary version of the real Pytlik–Szwarc deformation [7, 1] records exactly the coefficient and uniform estimate that the proof needs.

Lemma 3.1. For every 0<t<1, there is a unitary representation \pi_t:G\to\B(K) such that, with r=\sqrt{1-t^2},

\begin{align*} \inner{\pi_t(s_i)\delta_e}{\delta_e}&=t\quad(i\ge1),\tag{9}\\ \sup_{g\in G}\norm{\pi_t(g)-\lambda(g)}&\le c(t),\qquad c(t)=2(1-r)+\frac{t}{1-t}+t. \tag{10}\end{align*}

In particular, c(t)\to0 as t\downarrow0.

Proof. Use the tree of reduced words with edges given by right multiplication by a generator or its inverse, rooted at e. Its distance is d(x,y)=|x^{-1}y|, so left translations preserve distance. For x\ne e, let p(x) be the neighbor of x toward e, obtained by deleting the last letter of its reduced word. Recursively define

\xi_e=\delta_e,\qquad \xi_x=t\xi_{p(x)}+r\delta_x\quad(x\ne e). \tag{11}

These are unit vectors supported on their root paths. If z is the last common vertex of the root paths to x and y, the common-support components are t^{d(x,z)}\xi_z and t^{d(y,z)}\xi_z, respectively. The remaining supports are disjoint. Hence

\inner{\xi_x}{\xi_y}=t^{d(x,y)}. \tag{12}

Also \delta_e=\xi_e and \delta_x=(\xi_x-t\xi_{p(x)})/r for x\ne e, so their span is dense. The assignment \xi_x\mapsto\xi_{gx} preserves every finite Gram matrix by (12). It therefore defines an isometry on the span, and the inverse assignment for g^{-1} makes the extension unitary. Denote this unitary by \pi_t(g). The representation law holds on the dense span and hence on K. Since \pi_t(s_i)\delta_e=\xi_{s_i}=t\delta_e+r\delta_{s_i}, (9) follows.

It remains to prove that the perturbation is uniformly small over all group elements, rather than just over the generators. Fix g\ne e, write h_0=e,h_1,\ldots,h_l=g for its root path, and put T=\pi_t(g)\lambda(g)^*. For y\ne g, (11) gives

T\delta_y=\frac{\xi_y-t\xi_{g p(g^{-1}y)}}{r}. \tag{13}

The vertex g p(g^{-1}y) is the neighbor of y toward g. If y is off the path [e,g], this neighbor is also p(y), the neighbor toward e; see Figure 1. Thus T\delta_y=\delta_y off that path. On the path, the neighbor toward g of h_k, for k<l, is h_{k+1}, so

T\delta_{h_k}=r\xi_{h_k}-t\delta_{h_{k+1}}\quad(0\le k<l), \qquad T\delta_{h_l}=\xi_{h_l}. \tag{14}

Figure 1: Off the geodesic [e,g], the first edges toward e and g coincide. Hence \pi_t(g)\lambda(g)^* fixes all basis vectors outside the finite path span. Only a finite part of the tree is shown.

Let M_g=\spanop\{\delta_{h_k}:0\le k\le l\}. On this space define the truncated shift S\delta_{h_k}=\delta_{h_{k+1}} for k<l and S\delta_{h_l}=0, and put X\delta_{h_k}=\xi_{h_k}. For R=\operatorname{diag}(1,r,\ldots,r), the recursion gives

X=R(\id-tS^*)^{-1}. \tag{15}

Indeed, its kth column is t^k\delta_e+r\sum_{j=1}^k t^{k-j}\delta_{h_j}. Since \norm{S}\le1 and \norm{R}=1,

\norm{X-\id}\le\norm{R-\id} +\norm{R}\sum_{j=1}^{\infty}t^j \le1-r+\frac{t}{1-t}. \tag{16}

Writing R'=\operatorname{diag}(r,\ldots,r,1), (14) becomes T|_{M_g}=XR'-tS. Therefore

\begin{align*} \norm{T|_{M_g}-\id} &\le\norm{(X-\id)R'}+\norm{R'-\id}+t\norm{S}\\ &\le2(1-r)+\frac{t}{1-t}+t=c(t).\end{align*}

The same bound holds on K, since T is the identity on M_g^\perp. Multiplication by \lambda(g) proves (10). For g=e the difference is zero, so the supremum includes this case as well.

Remark 3.2. Only the operator X on a single finite path is used in the estimate. On this infinitely branching tree, the global map \delta_x\mapsto\xi_x need not be bounded: the vectors \xi_{s_i} all have root coefficient t. The Gram-matrix construction of \pi_t does not require such a global bounded map.

4 Close inclusions distinguished by a tensor norm

Fix 0<t<1, let \pi_t be given by Lemma 3.1, and write \pi_0=\lambda. We construct a common coordinate algebra D and subalgebras P_0,P_t satisfying Corollary 2.4. First we record the absorption calculation that will be used both for closeness and for the tensor-norm estimate.

Lemma 4.1 (Absorption). Let \sigma be a unitary representation of G on a nonzero Hilbert space J. The operator

W_\sigma(\delta_x\otimes v)=\delta_x\otimes\sigma(x)v \quad(x\in G,\ v\in J)

is unitary on K\otimes J and satisfies

W_\sigma(\lambda(g)\otimes\id_J)W_\sigma^* =\lambda(g)\otimes\sigma(g)\quad(g\in G). \tag{17}

Proof. The operator is the orthogonal direct sum of the unitaries \sigma(x). On \delta_x\otimes v, the left side of (17) equals

\delta_{gx}\otimes\sigma(gx)\sigma(x)^*v =\delta_{gx}\otimes\sigma(g)v.

On L=K\otimes K, put

Q_a(g)=\lambda(g)\otimes\pi_a(g),\qquad P_a=C^*(Q_a(g):g\in G)\quad(a=0,t). \tag{18}

For the matrix units E_{xy}\delta_z=\mathbf1_{\{z=y\}}\delta_x on K, define D\subseteq\B(L) to be generated by

\lambda(g)\otimes\id_K,\quad \id_K\otimes\pi_0(g),\quad \id_K\otimes\pi_t(g)\quad(g\in G), \qquad \id_K\otimes E_{xy}\quad(x,y\in G). \tag{19}

The generators include the identity, and P_0,P_t\subseteq D. All three algebras are norm-separable, since their generating sets are countable.

Lemma 4.2. The algebra D in (19) has scalar center, and

D'=\{Y\otimes\id_K:Y\in\lambda(G)'\}. \tag{20}

There is a unitary W\in\B(L) such that

WP_0W^*=P_t,\qquad \norm{W-\id_L}\le c(t). \tag{21}

Proof. Commutation with all \id_K\otimes E_{xy} forces a bounded operator to be Y\otimes\id_K: commutation with the diagonal matrix units removes off-diagonal blocks, and commutation with the remaining units makes the diagonal blocks identical. The first-factor generators then impose exactly Y\in\lambda(G)'. Such operators commute with all the second-factor generators, proving (20). In particular, \rho(g)\otimes\id_K\in D'.

If Y\otimes\id_K\in Z(D), then Y commutes with \lambda(G) by (20). It also commutes with \rho(G), since Y\otimes\id_K belongs to D and the right translations belong to D'. Thus Y\delta_e is invariant under \lambda(g)\rho(g), which acts on the basis by conjugation x\mapsto gxg^{-1}. Every nonidentity conjugacy class in G is infinite: conjugate a fixed nonidentity reduced word by distinct generators absent from that word. A square-summable vector invariant under conjugation must therefore be supported at e. Write Y\delta_e=\alpha\delta_e. Commuting with left translations gives Y\delta_x=\alpha\delta_x for every x, proving Z(D)=\C\id_L.

For the final assertion, Lemma 4.1 shows that W=W_{\pi_t}W_{\pi_0}^* sends every Q_0(g) to Q_t(g) by conjugation. Moreover, the direct-sum description of these unitaries gives

\norm{W-\id_L}=\norm{W_{\pi_t}-W_{\pi_0}} =\sup_{x\in G}\norm{\pi_t(x)-\lambda(x)}\le c(t).

Lemma 4.3. The inclusion P_0\subseteq D has the spatial tensor-norm identity.

Proof. Define a unitary on L by

V(v\otimes\delta_x)=\lambda(x)^*v\otimes\delta_x.

It commutes with every element of D' by (20). On elementary vectors one computes

VQ_0(g)V^*(v\otimes\delta_x)=v\otimes\delta_{gx}.

Thus VP_0V^*=\id_K\otimes C^*(\lambda(G)). Given finite families p_i\in P_0 and z_i\in D', write

Vp_iV^*=\id_K\otimes X_i,\qquad z_i=Y_i\otimes\id_K.

Conjugation by V gives

\norm{\sum_i p_i z_i}=\norm{\sum_i Y_i\otimes X_i}.

On L\otimes L=K^{\otimes4}, conjugation by V\otimes\id_L gives

\norm{\sum_i p_i\otimes z_i} =\norm{\sum_i\id_K\otimes X_i\otimes Y_i\otimes\id_K} =\norm{\sum_iY_i\otimes X_i}.

The last equality permutes the four tensor factors and uses the fact that tensoring with a nonzero identity operator does not change norm. This proves (7).

For P_t, the identity will fail because the product norm has a lower bound linear in the number of generators, while absorption gives a square-root upper bound on the tensor norm.

Lemma 4.4. For every positive integer m,

\norm{\sum_{i=1}^m\lambda(s_i)}\le2\sqrt m. \tag{22}

Proof. Let F_i be the orthogonal projection onto the closed span of words beginning with s_i^{-1}. The F_i are pairwise orthogonal. Write \lambda(s_i)=A_i+B_i, where

A_i=\lambda(s_i)(\id_K-F_i),\qquad B_i=\lambda(s_i)F_i.

There is no cancellation when s_i is prepended to a word not beginning with s_i^{-1}. Thus the range of A_i consists of vectors supported on words beginning with s_i, and these ranges are mutually orthogonal. For every v\in K,

\norm{\sum_i A_iv}^2=\sum_i\norm{(\id_K-F_i)v}^2 \le m\norm{v}^2,

while

\norm{\sum_i B_iv}\le\sum_i\norm{F_iv} \le\sqrt m\left(\sum_i\norm{F_iv}^2\right)^{1/2} \le\sqrt m\norm{v}.

Adding the two bounds proves (22).

Lemma 4.5. For every 0<t<1, the inclusion P_t\subseteq D fails the spatial tensor-norm identity.

Proof. Choose an integer m>4/t^2, so mt>2\sqrt m, and take

p_i=Q_t(s_i),\qquad z_i=\rho(s_i)\otimes\id_K, \qquad 1\le i\le m.

These belong to P_t and D', respectively. Since \lambda(s_i)\rho(s_i)\delta_e=\delta_e, (9) implies

\norm{\sum_i p_i z_i} \ge\left|\inner{\left(\sum_i p_i z_i\right) (\delta_e\otimes\delta_e)}{\delta_e\otimes\delta_e}\right| =mt. \tag{23}

On the other hand, the tensor sum on L\otimes L is

\sum_{i=1}^m\lambda(s_i)\otimes\pi_t(s_i) \otimes\rho(s_i)\otimes\id_K.

Apply Lemma 4.1 to the representation \sigma=\pi_t\otimes\rho\otimes1 on K^{\otimes3}, where 1 is the trivial representation on the last copy of K. Together with Lemma 4.4, it gives

\norm{\sum_i p_i\otimes z_i} =\norm{\sum_{i=1}^m\lambda(s_i)}\le2\sqrt m<mt. \tag{24}

Equations (23) and (24) contradict the identity for this finite list.

5 The counterexamples

Proof of Theorem 1.1. Given \varepsilon>0, choose 0<t<1 so small that 2c(t)<\varepsilon. Use the algebras P_0,P_t\subseteq D\subseteq\B(L) from Section 4, and put

H=\bigoplus_{n\ge1}L,\qquad A=\E_D(P_0),\qquad B=\E_D(P_t).

The space H is separable. Lemma 4.2 supplies the scalar center of D, so Proposition 2.1 shows that A,B are unital norm-separable C^*-algebras with common identity and that A''=B''. The same lemma supplies a unitary W with WP_0W^*=P_t and \norm{W-\id_L}\le c(t). By Corollary 2.4,

\KK(A,B)\le2c(t)<\varepsilon.

Finally, Lemmas 4.3 and 4.5 show that the relative tensor-norm identity holds for P_0 but not for P_t. Corollary 2.4 therefore excludes every unitary conjugating A onto B.

The common ideal c_0(\N,D) makes the von Neumann closures agree, while the norm-limit condition retains the distinction between the two inclusions.

References

  1. [1]M. Burger, N. Ozawa, and A. Thom, On Ulam stability, Israel J. Math. 193 (2013), 109–129. doi:10.1007/s11856-012-0050-z.
  2. [2]M.-D. Choi and E. Christensen, Completely order isomorphic and close C^*-algebras need not be *-isomorphic, Bull. London Math. Soc. 15 (1983), no. 6, 604–610. doi:10.1112/blms/15.6.604.
  3. [3]E. Christensen, A. M. Sinclair, R. R. Smith, S. A. White, and W. Winter, The spatial isomorphism problem for close separable nuclear C^*-algebras, Proc. Natl. Acad. Sci. USA 107 (2010), no. 2, 587–591. doi:10.1073/pnas.0913281107.
  4. [4]E. Christensen, A. M. Sinclair, R. R. Smith, S. A. White, and W. Winter, Perturbations of nuclear C^*-algebras, Acta Math. 208 (2012), no. 1, 93–150. doi:10.1007/s11511-012-0075-5.
  5. [5]B. E. Johnson, A counterexample in the perturbation theory of C^*-algebras, Canad. Math. Bull. 25 (1982), no. 3, 311–316. doi:10.4153/CMB-1982-043-4.
  6. [6]R. V. Kadison and D. Kastler, Perturbations of von Neumann algebras. I. Stability of type, Amer. J. Math. 94 (1972), no. 1, 38–54. doi:10.2307/2373592.
  7. [7]T. Pytlik and R. Szwarc, An analytic family of uniformly bounded representations of free groups, Acta Math. 157 (1986), 287–309. doi:10.1007/BF02392596.