The MF property for amalgamated free products

Tatiana Shulman

Abstract

A C*-algebra (or a group) is called MF (matricial field) if it admits finite dimensional approximate unitary representations which are approximately injective, where approximately is meant with respect to the operator norm.

It is proved that for any MF C*-algebra $A$ and its C*-subalgebra $C$ , $A\ast_{C}A$ is MF. For general amalgamated free products, $A\ast_{C}B$ , a necessary and sufficient condition for being MF is given. It is shown that the following groups – amalgamated free products of amenable groups, semidirect products of amenable groups by free groups, and $\mathbb{Z}^{2}\rtimes SL_{2}(\mathbb{Z})$ – all have MF full group C*-algebra. It is shown that the class of MF C*-algebras is closed under maximal tensor products with $C^{*}(\mathbb{F}_{n})$ . We also show that for any Hilbert-Schmidt stable C*-algebra $A$ and any its C*-subalgebra $C$ , all hyperlinear traces on $A\ast_{C}A$ are MF.

1 Introduction

Three approximation conjectures for groups – the sofic conjecture, the Connes Embedding conjecture for groups, and the MF-conjecture – state that every countable group can be approximated by ”nice” ones, namely by the permutation groups $Sym(n)$ and by the unitary groups $\mathcal{U}(n)$ of $n\times n$ matrices.

More specifically, the MF conjecture states that every countable group is MF, meaning that there is an approximate representation $\phi_{n}\colon G\to\mathcal{U}(n)$ which is separating in the sense that $\|\phi_{n}(g)-1_{n}\|$ is bounded away from zero for all $g\neq 1_{G}$ .

The Connes Embedding conjecture for groups (as well as the sofic conjecture) is formulated similarly, by replacing the operator norm by the normalized Hilbert-Schmidt norm (by replacing $\mathcal{U}(n)$ by $Sym(n)$ and the operator norm by the Hamming distance, respectively).

Not so much is known about MF groups. Amenable groups – which had long been known to be sofic and hyperlinear – were shown to be MF only relatively recently in the breakthrough paper [30].

MF groups were introduced in [6] as a group-theoretic analogue of MF C*-algebras. The latter are C*-algebras that can be approximated, in an appropriate sense, by finite-dimensional ones. For C*-algebras, the MF property is closely related to several other important properties: it is implied by quasidiagonality and, in turn, implies stable finiteness. It also has an interesting connection with the Ext semigroup of Brown, Douglas and Fillmore (see e.g. [10]).

Since a discrete group $G$ embeds into $C^{*}(G)$ and $C^{*}_{r}(G)$ , there is the following connection between the C*-algebraic MF property and the group theoretic one: if either $C^{*}(G)$ or $C^{*}_{r}(G)$ is MF, so is $G$ .

There are no examples of groups whose full or reduced C*-algebra is not MF, and at the same time there are not so many examples of non-amenable groups whose C*-algebras are known to be MF, at least in the case of full group C*-algebras.

In the reduced case, $C^{*}_{r}(\mathbb{F}_{2})$ is MF by the Haagerup and Thorbjornsen theorem [10], and new very interesting examples are found recently – see [23] and references therein.

The main source of groups with MF full C*-algebras is the class of amenable groups ([30]) and the class of groups with residually finite dimensional full group C*-algebras (see e.g. [18], [29] and [28] for examples of such). Outside those classes, it is not clear when full group C*-algebras are MF or at least stably finite.

There are however some constructions that preserve the MF property of C*-algebras. Hadwin, Li and Shen proved that free products of MF C*-algebras are MF [11]. Later Li and Shen discovered that if $A$ is a separable MF C*-algebra and $F$ is its finite-dimensional C*-subalgebra, then $A\ast_{F}A$ is MF. We prove here that in fact the amalgamated C*-subalgebra can be arbitrary.

Theorem (Th. 10) Let $A$ be a separable MF C*-algebra and let $C$ be any C*-subalgebra of $A$ . Then $A\ast_{C}A$ is MF.

In particular, we conclude that if $G$ is an MF group, then any its double, $G\ast_{H}G$ , is MF.

For amalgamated free products $A\ast_{F}B$ , where $A$ and $B$ are separable C*-algebras and $F$ is their common finite-dimensional C*-subalgebra, Li and Shen found a necessary and sufficient condition for being MF. We prove here that the same condition as Li and Shen found stays necessary and sufficient for arbitrary amalgamated free products.

Theorem (Th. 20) Let $A,B,C$ be separable C*-algebras and $\theta_{A}:C\to A$ , $\theta_{B}:C\to B$ be inclusions. Then $A\ast_{C}B$ is MF if and only if there exist embeddings $\phi_{A}:A\to\prod M_{n}/\oplus M_{n}$ and $\phi_{B}:B\to\prod M_{n}/\oplus M_{n}$ such that $\phi_{A}\circ\theta_{A}=\phi_{B}\circ\theta_{B}$ .

We use the theorem above to find a new class of groups that have MF full group C*-algebras.

Theorem (Th. 23) If $G_{1}$ and $G_{2}$ are amenable groups and $H$ is their common subgroup, then $C^{*}(G_{1}\ast_{H}G_{2})$ is MF.

In [22] Rainone and Schafhauser proved that if $G$ is amenable, then for any action of $\mathbb{F}_{n}$ on $G$ , $C^{*}_{r}(G\rtimes\mathbb{F}_{n})$ is MF. Here we use our results on amalgamated free products to obtain a similar result for full group C*-algebras.

Theorem (Cor. 35) For any semidirect product $G\rtimes\mathbb{F}_{n}$ of an amenable group $G$ by a free group $\mathbb{F}_{n}$ , $C^{*}(G\rtimes\mathbb{F}_{n})$ is MF.

We also prove that $C^{*}(\mathbb{Z}^{2}\rtimes SL_{2}(\mathbb{Z}))$ is MF (Cor. 34).

The same technique that we used for amalgamated free products applies to prove that the class of MF C*-algebras is closed under maximal tensor products with $C^{*}(\mathbb{F}_{n})$ (Th. 28) and under central HNN-extensions (Th. 29).

Approximation properties, such as hyperlinearity, quasidiagonality and the MF property, are of central importance in the C*-theory. Similar approximation properties for traces were introduced in [5], and turned out to play an important role in classification of nuclear simple C*-algebras ([30], [24]). It is an open problem whether all hyperlinear traces are MF.

Examples of C*-algebras with all hyperlinear traces being MF include all nuclear simple C*-algebras satisfying UCT ([30]), many crossed product C*-algebras ([22]) and all cone C*-algebras ([26]). Here we find more examples:

Theorem (Cor. 15) Let $A$ be a separable C*-algebra and let $C$ be any C*-subalgebra of $A$ . If $A$ is Hilbert-Schmidt stable, then every hyperlinear trace on $A\ast_{C}A$ is MF.

As an easy consequence of the theorem above we obtain that all hyperlinear traces on $C^{*}(\mathbb{F}_{2}\times\mathbb{F}_{2})$ are MF.

The main tool used for proving the above results is the lifting characterization of the MF property found by the author in [27].

Acknowledgments. The author is grateful to Don Hadwin for a useful discussion. The author is partially supported by a grant from the Swedish Research Council.

2 Preliminaries

2.1 Amalgamated free products

Definition 1.

Let $A,B,C$ be (unital) C*-algebras with (unital) embeddings $\theta_{A}:C\to A$ and $\theta_{B}:C\to B$ . The (unital) amalgamated free product of $A$ and $B$ over $C$ is the C*-algebra $D$ , equipped with (unital) embeddings $i_{A}:A\to D$ and $i_{B}:B\to D$ satisfying $i_{A}\circ\theta_{A}=i_{B}\circ\theta_{B}$ , such that $D$ is generated by $i_{A}(A)\bigcup i_{B}(B)$ and satisfies the following universal property:

whenever $\mathcal{E}$ is a (unital) C*-algebra and $\pi_{A}:A\to\mathcal{E}$ and $\pi_{B}:B\to\mathcal{E}$ are (unital) $\ast$ -homomorphisms satisfying $\pi_{A}\circ\theta_{A}=\pi_{B}\circ\theta_{B}$ , there is a (unital) $\ast$ -homomorphism $\pi:D\to\mathcal{E}$ such that $\pi\circ i_{A}=\pi_{A}$ and $\pi\circ i_{B}=\pi_{B}$ .

Standardly the (unital) amalgamated free product $C$ is denoted by $A\ast_{C}B$ . Sometimes, when it is important to emphasize how $C$ is embedded into $A$ and $B$ respectively, we will write $A\ast_{\theta_{A},\theta_{B}}B$ rather than $A\ast_{C}B$ . When we write $A\ast_{C}A$ we always mean that $C$ is embedded into both copies of $A$ identically.

Amalgamated free product of more than two factors is defined analogously.

C*-algebraic definition of amalgamated free product agrees with the group-theoretical definition of amalgamated free product in the following sense:

C^{*}(G_{1}\ast_{H}G_{2})=C^{*}(G_{1})\ast_{C^{*}(H)}C^{*}(G_{2}).

2.2 MF C*-algebras and groups

We use notation $M_{n}$ for the C*-algebra of all $n\times n$ matrices. Let

\prod M_{n}=\{(T_{n})_{n\in\mathbb{N}}\;|\;T_{n}\in M_{n},\;\sup_{n}\|T_{n}\|<\infty\},

\bigoplus M_{n}=\{(T_{n})_{n\in\mathbb{N}}\;|\;\lim_{n\to\infty}\|T_{n}\|=0\}.

Definition 2.

([4]) A C*-algebra is matricial field (MF) if it embeds into $\prod M_{n}/\oplus M_{n}$ .

Equivalently, $A$ is MF if there exist maps $\phi_{n}:A\to M_{k_{n}}$ , for some $k_{n}\in\mathbb{N}$ , which are approximately multiplicative, approximately linear, approximately self-adjoint, and approximately injective. Reformulating it ”locally”, $A$ is MF if for any finite $F\subset A$ and $\epsilon>0$ there is $k$ and a map $\phi_{k}:A\to M_{k}$ such that

\|\phi_{(}a)\|>\|a\|-\epsilon,\;\|\phi(ab)-\phi(a)\phi(b)\|\leq\epsilon,\\ \|\phi(a+b)-\phi(a)-\phi(b)\|\leq\epsilon,\;\|\phi(a^{*})-\phi(a)^{*}\|\leq\epsilon,

for any $a,b\in F$ .

Definition 3.

([6]) A group is MF if it embeds into the unitary group $\mathcal{U}\left(\prod M_{n}/\oplus M_{n}\right)$ of $\prod M_{n}/\oplus M_{n}$ .¹¹ 1 A different, more strong, property of a group was also called MF in [23], see also [19].

A discrete group $G$ embeds into $C^{*}(G)$ and $C^{*}_{r}(G)$ and therefore, if either $C^{*}(G)$ or $C^{*}_{r}(G)$ is MF, so is $G$ .

2.3 Asymptotic homomorphisms

Definition ([7]). An asymptotic homomorphism from $A$ to $B$ is a family of maps $(f_{\lambda})_{\lambda\in[0,\infty)}:A\to B$ satisfying the following properties:

(i) for any $a\in A$ , the mapping $[0,\infty)\to B$ defined by the rule $\lambda\to f_{\lambda}(a)$ is continuous;

(ii) for any $a,b\in A$ and $\mu_{1},\mu_{2}\in\mathbb{C}$ , we have

•

$\lim_{\lambda\to\infty}\|f_{\lambda}(a^{*})-f_{\lambda}(a)^{*}\|=0;$
•

$\lim_{\lambda\to\infty}\|f_{\lambda}(\mu_{1}a+\mu_{2}b)-\mu_{1}f_{\lambda}(a)-\mu_{2}f_{\lambda}(b)\|=0;$
•

$\lim{\lambda\to\infty}\|f_{\lambda}(ab)-f_{\lambda}(a)f_{\lambda}(b)\|=0.$

We will call $(f_{\lambda})_{\lambda\in\Lambda}:A\to B$ , where $\Lambda$ is a directed set, a discrete asymptotic homomorphism if the condition (ii) above is satisfied and for each $a\in A$ one has $\sup_{\lambda}\|f_{\lambda}(a)\|<\infty$ . (For usual asymptotic homomorphisms the last condition holds automatically, see [7] or [8]). In this paper discrete asymptotic homomorphisms will be indexed by $\Lambda=\mathbb{N}$ .

Let $q:B\to B/I$ be a quotient map and $f:A\to B/I$ a $\ast$ -homomorphism. Following Manuilov and Thomsen ([21], [20]), we say that $f$ lifts to a discrete asymptotic homomorphism $\phi_{n}$ , $n\in\mathbb{N}$ , if $q\circ\phi_{n}=f$ , for each $n\in\mathbb{N}$ .

3 Amalgamated free products $A\ast_{C}A$

. Let $H$ be a Hilbert space and let $P_{n}$ , $n\in\mathbb{N}$ , be an increasing sequence of projections of dimension $n$ that $\ast$ -strongly converges to $1_{B(H)}$ . We will identify $M_{n}$ with $P_{n}B(H)P_{n}$ . Let $\mathcal{D}\subset\prod_{n\in\mathbb{N}}M_{n}$ be the C*-algebra of all $\ast$ -strongly convergent sequences of matrices. Let $q:\mathcal{D}\to B(H)$ be the surjection that sends each sequence to its $\ast$ -strong limit.

Our main tool for exploring the MF-property of amalgamated free products will be the following lifting characterization of the MF-property obtained in [27].

Theorem 4.

(Shulman [27, Th. 14]) Let $A$ be a separable C*-algebra. TFAE:

(i)

$A$ is MF,
(ii)

every $\ast$ -homomorphism from $A$ to $B(H)$ lifts to a discrete asymptotic homomorphism from $A$ to $\mathcal{D}$ ,
(iii)

there exists an embedding of $A$ into $B(H)$ that lifts to a discrete asymptotic homomorphism from $A$ to $\mathcal{D}$ .

Remark 5.

In the above characterization of the MF property $\mathcal{D}$ and $B(H)$ can be replaced by $M_{N}(\mathcal{D})$ and $B(H^{\oplus N})$ respectively, for any $N\in\mathbb{N}$ .

Remark 6.

In Theorem 4 one can replace liftability to a discrete asymptotic homomorphism by liftability to a contractive discrete asymptotic homomorphism. To arrange this, in the proof of Theorem 4 one uses the existence of a contractive continuous section proved in [26].

Lemma 7.

Let $A,B$ and $D$ be C*-algebras and let $C$ be a common C*-subalgebra of $A$ and $B$ . Let $\phi_{\lambda}^{A}:A\to D$ and $\phi_{\lambda}^{B}:B\to D$ , $\lambda\in\Lambda$ , be contractive (discrete) asymptotic homomorphisms such that

\lim_{{\lambda}\to\infty}\|\phi_{\lambda}^{A}(c)-\phi_{\lambda}^{B}(c)\|=0,

for each $c\in C$ . Then there exists a contractive (discrete) asymptotic homomorphism $\Phi_{\lambda}:A\ast_{C}B\to D$ , ${\lambda}\in\Lambda$ , such that

\lim_{{\lambda}\to\infty}\|\Phi_{\lambda}^{A}(a)-\phi_{\lambda}^{A}(a)\|=0,

\lim_{{\lambda}\to\infty}\|\Phi_{\lambda}^{B}(b)-\phi_{\lambda}^{B}(b)\|=0,

for each $a\in A,b\in B$ .

Proof.

The asymptotic homomorphisms $\phi_{\lambda}^{A}:A\to D$ and $\phi_{\lambda}^{B}:B\to D$ , $\lambda\in\Lambda$ , induce $\ast$ -homomorphisms $\phi^{A}:A\to C_{b}(\Lambda,D)/C_{0}(\Lambda,D)$ and $\phi^{B}:B\to C_{b}(\Lambda,D)/C_{0}(\Lambda,D)$ by formula

\phi_{A}(a)=\left(\phi_{\lambda}^{A}(a)\right)_{\lambda\in\Lambda}+C_{0}(\Lambda,D),

\phi_{B}(b)=\left(\phi_{\lambda}^{B}(b)\right)_{\lambda\in\Lambda}+C_{0}(\Lambda,D).

Since for each $c\in C$ , $\lim_{{\lambda}\to\infty}\|\phi_{\lambda}^{A}(c)-\phi_{\lambda}^{B}(c)\|=0,$ we have

\phi^{A}(c)=\phi^{B}(c),

$c\in C$ . Therefore $\phi^{A}$ and $\phi^{B}$ induce a $\ast$ -homomorphism

\Phi=\phi^{A}\ast\phi^{B}:A\ast_{C}B\to C_{b}(\Lambda,D)/C_{0}(\Lambda,D).

By [26, Th. 3] there is a contractive continuous section $s:C_{b}(\Lambda,D)/C_{0}(\Lambda,D)\to C_{b}(\Lambda,D)$ . Then

\Phi_{\lambda}=ev_{\lambda}\circ s\circ\Phi:A\ast_{C}B\to D

is a contractive (discrete) asymptotic homomorphism. Since $\left(\Phi_{\lambda}(a)\right)_{\lambda\in\Lambda}$ and $\left(\phi_{\lambda}^{A}(a)\right)_{\lambda\in\Lambda}$ are both representatives of $\phi^{A}(a)$ , we conclude that

\lim_{{\lambda}\to\infty}\|\Phi_{\lambda}^{A}(a)-\phi_{\lambda}^{A}(a)\|=0,

for each $a\in A$ . Similarly for $B$ . ∎

Lemma 8.

Let $q:B\to B/I$ be a quotient map and let $\{i_{\lambda}\}_{\lambda\in\Lambda}$ be a quasicentral approximate unit for $I\lhd B$ . Then for any $x,y\in B$ and $N\in\mathbb{N}$

\limsup_{\lambda}\|[(x(1-i_{\lambda}))^{N},y]\|=\|[q(x)^{N},q(y)]\|.

Proof.

Since $\{i_{\lambda}\}_{\lambda\in\Lambda}$ is quasicentral,

\limsup_{\lambda}\|[(x(1-i_{\lambda}))^{N},y]\|=\limsup_{\lambda}\|(1-i_{\lambda})^{N}[x^{N},y]\|.

(1)

Let $u_{\lambda}=1-(1-i_{\lambda})^{N}.$ Then $\{u_{\lambda}\}_{\lambda\in\Lambda}$ is itself a (quasicentral) approximate unit. It is well-known that for any approximate unit, hence for $\{u_{\lambda}\}_{\lambda\in\Lambda}$ , and any $z\in B$

\limsup_{\lambda}\|(1-u_{\lambda})z\|=\|q(z)\|

(see e.g. [17]). This and (1) imply the statement. ∎

Lemma 9.

Let $u,a_{1},\ldots,a_{N}\in B/I,$ where $u$ is a unitary and $[a_{i},u]=0,$ $i=1,\ldots,N$ . Let $A_{1},\ldots,A_{N}\in B$ be lifts of $a_{1},\ldots,a_{N}$ , $i=1,\ldots,N$ , and let $\epsilon>0$ . Then $\left(\begin{array}[]{cc}u&\\ &-u^{*}\end{array}\right)$ lifts to a unitary $V\in M_{2}(B)$ such that

\left\|\left[\left(\begin{array}[]{cc}A_{i}&\\ &A_{i}\end{array}\right),V\right]\right\|\leq\epsilon,

$i=1,\ldots,N$ .

Proof.

First we lift $u$ to a contraction $X\in B$ . Let $\{i_{\lambda}\},\lambda\in\Lambda,$ be a quasicentral approximate unit for $I\lhd B$ . By Lemma 8 there exists $\lambda$ such that for $\tilde{X}=X(1-i_{\lambda})$ we have

\|[\tilde{X},A_{i}]\|\leq\frac{\epsilon}{2},

\|[\tilde{X}^{*},A_{i}]\|\leq\frac{\epsilon}{2},

\|[\tilde{X}\tilde{X}^{*},A_{i}]\|\leq\frac{\epsilon^{2}}{4},

\|[\tilde{X}^{*}\tilde{X},A_{i}]\|\leq\frac{\epsilon^{2}}{4},

$i=1,\ldots,N$ . We define a unitary $V\in M_{2}(B)$ by

V=\left(\begin{array}[]{cc}\tilde{X}&\sqrt{1-\tilde{X}\tilde{X}^{*}}\\ \sqrt{1-\tilde{X}^{*}\tilde{X}}&-\tilde{X}^{*}\end{array}\right).

Then

\left[V,\left(\begin{array}[]{cc}A_{i}&0\\ 0&A_{i}\end{array}\right)\right]=\left(\begin{array}[]{cc}[\tilde{X},A_{i}]&[(1-\tilde{X}{\tilde{X}}^{*})^{1/2},A_{i}]\\ [(1-\tilde{X}^{*}\tilde{X})^{1/2},A_{i}]&[-\tilde{X}^{*},A_{i}]\end{array}\right),

$i=1,\ldots,N$ .

By Pedersen’s inequality (which says that $\|[A^{1/2},B]\|\leq\frac{5}{4}\|[A,B]\|^{1/2}$ when $0\leq A\leq 1$ ) we have

\|[(1-\tilde{X}{\tilde{X}}^{*})^{1/2},A_{i}]\|\leq\frac{5}{4}\|[1-\tilde{X}\tilde{X}^{*},A_{i}]\|^{1/2}=\frac{5}{4}\|[\tilde{X}\tilde{X}^{*},A_{i}]\|^{1/2}<\epsilon/2,

$i=1,\ldots,N$ , and the same for $[(1-\tilde{X}{\tilde{X}}^{*})^{1/2},A_{i}]\|$ . Therefore

\left\|\left[V,\left(\begin{array}[]{cc}A_{i}&0\\ 0&A_{i}\end{array}\right)\right]\right\|\leq\\ \left\|\left(\begin{array}[]{cc}[\tilde{X},A_{i}]&0\\ 0&[-\tilde{X}^{*},A_{i}]\end{array}\right)\right\|+\\ \left\|\left(\begin{array}[]{cc}0&[(1-\tilde{X}{\tilde{X}}^{*})^{1/2},A_{i}]\\ [(1-\tilde{X}^{*}\tilde{X})^{1/2},A_{i}]&0\end{array}\right)\right\|<\epsilon/2+\epsilon/2=\epsilon,

(2)

$i=1,\ldots,N$ .

∎

Theorem 10.

Let $A$ be a separable MF C*-algebra and let $C$ be any C*-subalgebra of $A$ . Then $A\ast_{C}A$ is MF.

Proof.

Let $\pi=\pi_{1}\ast\pi_{2}:A\ast_{C}A\to B(H)$ be an embedding. Let

\rho_{1}=\pi_{1}\oplus\pi_{2},\;\rho_{2}=\pi_{2}\oplus\pi_{1}.

Then $(\rho_{1}\oplus\rho_{1})\ast(\rho_{2}\oplus\rho_{2}):A\ast_{C}A\to B\left(H^{\oplus 4}\right)$ is also an embedding. We will show that it asymptotically lifts to a discrete asymptotic homomorphism from $A\ast_{C}A$ to $M_{4}(\mathcal{D})$ . Then Theorem 4 would finish the proof.

Since $\rho_{1}=u^{*}\rho_{2}u$ , where $u=\left(\begin{array}[]{cc}0&1\\ 1&0\end{array}\right)$ , and since $\rho_{1}|_{C}=\rho_{2}|_{C}$ , we have

\rho_{2}(c)=u^{*}\rho_{2}(c)u,

$c\in C$ , or, equivalently,

[\rho_{2}(c),u]=0,

$c\in C$ . Since $A$ is MF, by Theorem 4 and Remark 6 the representation $\rho_{2}$ lifts to a contractive discrete asymptotic homomorphism $\phi_{n}:A\to M_{2}(\mathcal{D})$ , $n\in\mathbb{N}$ . Let $\{c_{1},c_{2},\ldots\}$ be a dense subset of $C$ and let $F_{k}=\{c_{1},\ldots,c_{k}\}$ . By Lemma 9 there exists a lift $V_{k}\in M_{4}(\mathcal{D})$ of $\left(\begin{array}[]{cc}u&\\ &-u^{*}\end{array}\right)$ such that

\|\left[V_{k},\left(\begin{array}[]{cc}\phi_{k}(c)&0\\ 0&\phi_{k}(c)\end{array}\right)\right]\|\leq\frac{1}{k},

(3)

$c\in F_{k}$ .

Define asymptotic homomorphisms $\phi_{k}^{(1)},\phi_{k}^{(2)}:A\to M_{4}(\mathcal{D})$ , $k\in\mathbb{N}$ , by

\phi_{k}^{(2)}=\phi_{k}\oplus\phi_{k},\;\phi_{k}^{(1)}=V_{k}^{*}\phi_{k}^{(2)}V_{k}.

Then

q\circ\phi_{k}^{(2)}=\rho_{2}\oplus\rho_{2},

q\circ\phi_{k}^{(1)}=(u\oplus(-u^{*}))^{*}(\rho_{2}\oplus\rho_{2})(u\oplus(-u^{*})=(u\oplus u)^{*}(\rho_{2}\oplus\rho_{2})(u\oplus u)=\rho_{1}\oplus\rho_{1}

(we used here that $u^{*}=u$ ).

Now we will show that the asymptotic homomorphisms $\phi_{k}^{(1)},\phi_{k}^{(2)}$ asymptotically agree on $C$ . Let $c\in C,\epsilon>0$ . There exists $k_{0}\in\mathbb{N}$ and $c_{0}\in F_{k_{0}}$ such that $\frac{1}{k_{0}}<\epsilon$ and $\|c-c_{0}\|<\epsilon.$ Since $\phi_{k}^{(1)},\phi_{k}^{(2)}$ are contractive asymptotic homomorphisms, there exists $k_{1}>k_{0}$ such that for any $k>k_{1}$

\|\phi_{k}^{(i)}(c)-\phi_{k}^{(i)}(c_{0})\|\leq\|\phi_{k}^{(i)}(c-c_{0})\|+\epsilon\leq\|c-c_{0}\|+\epsilon<2\epsilon,

$i=1,2$ . Then, using (3), for any $k>k_{1}$ we obtain

\|\phi_{k}^{(1)}(c)-\phi_{k}^{(2)}(c)\|<4\epsilon+\|\phi_{k}^{(1)}(c_{0})-\phi_{k}^{(2)}(c_{0})\|\\ =4\epsilon+\|V_{k}^{*}\phi_{k}^{(2)}(c_{0})V_{k}-\phi_{k}^{(2)}(c_{0}))\|=4\epsilon+\|[V_{k},\phi_{k}^{(2)}(c_{0})]\|\leq 4\epsilon+\frac{1}{k_{0}}<5\epsilon.

Thus, for each $c\in C$ ,

\lim_{k\to\infty}\|\phi_{k}^{(1)}(c)-\phi_{k}^{(2)}(c)\|=0.

By Lemma 7 there exists an asymptotic homomorphism $\Phi_{k}:A\ast_{C}A\to M_{4}(\mathcal{D})$ , $k\in\mathbb{N}$ , such that

\lim_{k\to\infty}\|\Phi_{k}(i_{1}(a))-\phi_{k}^{(1)}(a)\|=0,

\lim_{k\to\infty}\|\Phi_{k}(i_{2}(a))-\phi_{k}^{(2)}(a)\|=0,

$a\in A$ . Applying the quotient map $q$ to these equalities we obtain that $\Phi_{k}|_{i_{1}(A)}$ , $k\in\mathbb{N}$ , is an asymptotic lift of $\rho_{1}\oplus\rho_{1}$ and $\Phi_{k}|_{i_{2}(A)}$ , $k\in\mathbb{N}$ , is an asymptotic lift of $\rho_{2}\oplus\rho_{2}$ . Then for $x$ in the dense subset

E=span\{i_{1}(a_{1})i_{2}(a_{2})\ldots i_{1}(a_{n})i_{2}(a_{n+1})\;|\;n\in\mathbb{N},a_{i}\in A\}

of $A\ast_{C}A$ , $\|q\circ\Phi_{k}(x)-(\rho_{1}\oplus\rho_{1})\ast(\rho_{2}\oplus\rho_{2})(x)\|\to 0$ as $k\to\infty$ . It implies that $\Phi_{k}$ , $k\in\mathbb{N}$ , is an asymptotic lift of $(\rho_{1}\oplus\rho_{1})\ast(\rho_{2}\oplus\rho_{2})$ . By Theorem 4, $A\ast_{C}A$ is MF. ∎

Example 11.

In [14] Higman proved that the group

G=\left<a,b,c\;|\;a^{-1}ca=c^{2}=b^{-1}cb\right>

is not residually finite. Since it is finitely generated, it is also non-MAP by Maltsev’s Theorem. This group can be written as the group double

G=BS(1,2)\ast_{\mathbb{Z}}BS(1,2),

where $BS(1,2)=\left<a,c\;|\;a^{-1}ca=c^{2}\right>$ is the Baumslag-Solitar group and $\mathbb{Z}=\left<c\right>$ . Then $C^{*}(G)=C^{*}(BS(1,2))\ast_{C(\mathbb{T})}C^{*}(BS(1,2)).$ Since $BS(1,2)$ is amenable and therefore has MF full group C*-algebra ([30]), by Theorem 10 $C^{*}(G)$ is MF.

Example 12.

A group double of a residually finite (RF) group need not be RF even when amalgamation is done over a central subgroup ([28]). Namely for the Abels group

\Gamma=\left\{\left(\begin{array}[]{cccc}1&x_{12}&x_{13}&x_{14}\\ 0&p^{k}&x_{23}&x_{24}\\ 0&0&p^{n}&x_{34}\\ 0&0&0&1\end{array}\right):x_{ij}\in\mathbb{Z}\left[\frac{1}{p}\right],k,n\in\mathbb{Z}\right\}

and its central subgroup

N=\left\{\left(\begin{array}[]{cccc}1&0&0&x\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&0&1\end{array}\right):x\in\mathbb{Z}\right\}\cong\mathbb{Z},

the amalgamated free product $\Gamma\ast_{N}\Gamma$ is not maximally almost periodic, hence is not RF ([28, Prop.8.]). Again, by Theorem 10 we see that $C^{*}(\Gamma\ast_{N}\Gamma)$ is MF.

3.1 MF traces

A trace $\tau$ on a C*-algebra $A$ is hyperlinear if for any $\epsilon>0$ and a finite $G\subset A$ there is $n\in\mathbb{N}$ and a map $\phi:A\to M_{n}$ such that

\|\phi(ab)-\phi(a)\phi(b)\|_{2}<\epsilon\;\;\text{and}\;\;|\tau(a)-tr\phi(a)|<\epsilon,

for any $a\in G$ .

(Equivalently, one can say that $\tau$ is hyperlinear if $\tau=tr\circ f$ , for some $\ast$ -homomorphism $f:A\to\prod M_{n}/\oplus_{2}M_{n}$ , where $\oplus_{2}M_{k_{n}}$ is the ideal of all sequences that converge to zero in the 2-norm.)

A trace $\tau$ is MF if for any $\epsilon>0$ and a finite $G\subset A$ there is $n\in\mathbb{N}$ and a map $\phi:A\to M_{n}$ such that

\|\phi(ab)-\phi(a)\phi(b)\|<\epsilon\;\;\text{and}\;\;|\tau(a)-tr\phi(a)|<\epsilon,

for any $a\in G$ .

The property of a trace to be MF is another property that can be formulated in terms of asymptotic lifting [27].

Proposition 13.

([27, Prop. 22]) Let $\tau$ be a trace on a C*-algebra $A$ . TFAE:

(i)

$\tau$ is an MF trace,
(ii)

there exists a $\ast$ -homomorphism $f:A\to\prod M_{m_{n}}/\oplus_{2,\omega}M_{m_{n}}$ such that $\tau=trf$ and $f$ lifts to a discrete asymptotic homomorphism from $A$ to $\prod M_{n}$ ,

In particular Proposition 13 implies that if all $\ast$ -homomorphisms from $A$ to $\prod M_{m_{n}}/\oplus_{2,\omega}M_{m_{n}}$ lift to discrete asymptotic homomorphisms to $\prod M_{n}$ , then all hyperlinear traces on $A$ are MF.

The same arguments as in the proof of Theorem 10 apply to obtain the following result.

Theorem 14.

Let $A$ be a separable C*-algebra and let $C$ be any C*-subalgebra of $A$ . If every $\ast$ -homomorphism from $A$ to $\prod M_{m_{n}}/\oplus_{2,\omega}M_{m_{n}}$ lifts to a discrete asymptotic homomorphism to $\prod M_{n}$ , then the same holds for $A\ast_{C}A$ . In particular, every hyperlinear trace on $A\ast_{C}A$ is MF.

A C*-algebra is called Hilbert-Schmidt stable (HS-stable) if every $\ast$ -homomorphism from $A$ to $\prod M_{m_{n}}/\oplus_{2,\omega}M_{m_{n}}$ lifts to a $\ast$ -homomorphism to $\prod M_{n}$ .

Corollary 15.

Let $A$ be a separable C*-algebra and let $C$ be any C*-subalgebra of $A$ . If $A$ is Hilbert-Schmidt stable, then every hyperlinear trace on $A\ast_{C}A$ is MF.

Example 16.

$\mathbb{F}_{2}\times\mathbb{F}_{2}$ can be written as amalgamated free product

\mathbb{F}_{2}\times\mathbb{F}_{2}\cong(\mathbb{F}_{2}\times\mathbb{Z})\ast_{\mathbb{F}_{2}}(\mathbb{F}_{2}\times Z).

By [12, Th. 2.7] (also [15]) $C^{*}(\mathbb{F}_{2}\times\mathbb{Z})$ is Hilbert-Schmidt stable. Therefore every hyperlinear trace on $C^{*}(\mathbb{F}_{2}\times\mathbb{F}_{2})$ is MF.

3.2 General amalgamated free products

Here we find necessary and sufficient conditions for general amalgamated free products to be MF.

Theorem 17.

([9, Th. 4.11]) Let $A,B,D$ be unital C*-algebras and $C$ be a separable unital C*-algebra. Let ${\theta}_{A}:C\to A$ , ${\theta}_{B}:C\to B$ , and $\phi_{A}:A\to D$ , $\phi_{B}:B\to D$ be unital inclusions such that $\phi_{A}\circ{\theta}_{A}=\phi_{B}\circ{\theta}_{B}$ . Then $A\ast_{C}B$ embeds into $D\ast_{C}D$ .

Remark 18.

In fact the embedding in the theorem above is given by the $\ast$ -homomorphism $\phi_{A}\ast\phi_{B}$ that sends $A$ to the first copy of $D$ via $\phi_{A}$ and $B$ to the second copy of $D$ via $\phi_{B}$ . This $\ast$ -homomorphism is well-defined since $\phi_{A}\circ\theta_{A}=\phi_{B}\circ\theta_{B}$ .

Lemma 19.

([9, Lemma. 4.4]) Let $A^{+}$ denote the forced unitization of $A$ . Then for any $A,B,C$

A^{+}\ast_{C^{+}}B^{+}=(A\ast_{C}B)^{+}.

Theorem 20.

Let $A,B,C$ be separable C*-algebras and $\theta_{A}:C\to A$ , $\theta_{B}:C\to B$ be inclusions. Then $A\ast_{C}B$ is MF if and only if there exist embeddings $\phi_{A}:A\to\prod M_{n}/\oplus M_{n}$ and $\phi_{B}:B\to\prod M_{n}/\oplus M_{n}$ such that $\phi_{A}\circ\theta_{A}=\phi_{B}\circ\theta_{B}$ .

Proof.

”Only if” is clear. We prove the ”if” part. At first we consider the unital case. Let

D=C^{*}(\phi_{A}(A),\phi_{B}(B))\subset\prod M_{n}/\oplus M_{n}.

By Theorem 17 $A\ast_{C}B\subset D\ast_{C}D$ . Since $D$ is MF, by Theorem 10 $D\ast_{C}D$ is MF. Therefore $A\ast_{C}B$ is MF.

In the non-unital case (that is, when any of the C*-algebras or the inclusions is non-unital) we consider the forced unitizations of $A,B,C$ and the corresponding maps. By what is proved above, $A^{+}\ast_{C^{+}}B^{+}$ is MF. Lemma 19 finishes the proof. ∎

Now we consider amalgamated free products of more than two factors. If $\theta_{j}:C\to A_{j}$ , $j=1,\ldots,N$ , are inclusions, then the corresponding amalgamated free product will be denoted by $\ast_{j=1}^{N}(A_{j},\theta_{j})$ . Let $i_{j}:A_{j}\to\ast_{j=1}^{k}(A_{j},\theta_{j})$ denote the canonical inclusion. Using the universal property of amalgamated free products, it is straightforward to check that

\ast_{j=1}^{N}(A_{j},\theta_{j})=\left(\ast_{j=1}^{N-1}(A_{j},\theta_{j})\right)\ast_{i_{i}\circ\theta_{1},\theta_{N}}A_{N}.

Lemma 21.

If $A$ is a separable MF C*-algebra and $\theta:C\to A$ is an inclusion, then $\ast_{j=1}^{N}(A,\theta)$ is MF.

Proof.

By induction. Suppose the statement is proved for $k<N$ . We write

\ast_{j=1}^{N}(A,\theta)=\left(\ast_{j=1}^{N-1}(A,\theta)\right)\ast_{i_{i}\circ\theta,\theta}A.

By the induction assumption there is an embedding $\phi:\ast_{j=1}^{N-1}(A,\theta)\to\prod M_{n}/\oplus M_{n}$ . We define $\phi_{N}:A\to\prod M_{n}/\oplus M_{n}$ by

\phi_{N}=\phi\circ i_{1}.

Then $\phi_{N}\circ\theta=\phi\circ\char 16\relax_{1}\circ\theta$ . By Theorem 20 $\ast_{j=1}^{N}(A,\theta)$ is MF. ∎

Theorem 22.

Let $A_{j},$ $j=1,\ldots,N$ , and $C$ be separable C*-algebras and $\theta_{j}:C\to A_{j}$ , $j=1,\ldots,N$ , be inclusions. Then $\ast_{j=1}^{N}(A_{j},\theta_{j})$ is MF if and only if there exist embeddings $\phi_{j}:A_{j}\to\prod M_{n}/\oplus M_{n}$ , $j=1,\ldots,N$ , such that $\phi_{j}\circ\theta_{j}=\phi_{k}\circ\theta_{k}$ , for all $j,k\leq N$ .

Proof.

”Only if” is clear. We prove the ”if” part by induction. Lemma 19 holds for more factors than two and therefore, similar to the proof of Theorem 20, it is sufficient to consider the unital case. To avoid cumbersome notation, we will show how to prove the statement for three factors. For four and more it is similar. We have

\ast_{j=1}^{3}(A_{j},\theta_{j})=(A_{1}\ast_{\theta_{1},\theta_{2}}A_{2})\ast_{i_{1}\circ\theta_{1},\theta_{3}}A_{3}.

By Remark 18, $\phi_{1}\ast\phi_{2}:A_{1}\ast_{\theta_{1},\theta_{2}}A_{2}\to\left(\prod M_{n}/\oplus M_{n}\right)\ast_{\phi_{1}\circ\theta_{1}(C)}\left(\prod M_{n}/\oplus M_{n}\right)$ is an inclusion. We define $\tilde{\phi}_{3}:A_{3}\to\left(\prod M_{n}/\oplus M_{n}\right)\ast_{\phi_{1}\circ\theta_{1}(C)}\left(\prod M_{n}/\oplus M_{n}\right)$ by

\tilde{\phi}_{3}=i_{1}\circ\phi_{3}.

Then

\tilde{\phi}_{3}\circ\theta_{3}(c)=i_{1}(\phi_{3}(\theta_{3}(c)))=i_{1}(\phi_{1}(\theta_{1}(c)))=(\phi_{1}\ast\phi_{2})\circ(i_{1}\circ\theta_{1})(c).

By Lemma 21 there is an embedding

\alpha:\left(\prod M_{n}/\oplus M_{n}\right)\ast_{\phi_{1}\circ\theta_{1}(C)}\left(\prod M_{n}/\oplus M_{n}\right)\to\prod M_{n}/\oplus M_{n}.

Then $\alpha\circ\tilde{\phi}_{3}$ and $\alpha\circ(\phi_{1}\ast\phi_{2})$ satisfy the necessary and sufficient condition of Theorem 20, and therefore $\ast_{j=1}^{3}(A_{j},\theta_{j})$ is MF. ∎

4 Amalgamated free products of amenable groups

The next theorem shows that for amalgamated free products of amenable groups the necessary and sufficient condition of Theorem 20 holds automatically.

Theorem 23.

If $G_{1}$ and $G_{2}$ are amenable groups, then $C^{*}(G_{1}\ast_{H}G_{2})$ is MF.

Proof.

Let $\mathcal{Q}$ denote the UHF-algebra $\otimes_{d=1}^{\infty}M_{d}$ . By [23, Th. 2.5] there exist unital embeddings $\rho_{i}:C^{*}(G_{i})\to\mathcal{Q}$ , $i=1,2$ , such that

tr_{\mathcal{Q}}\rho_{1}(g_{1})=0=tr_{\mathcal{Q}}\rho_{2}(g_{2}),

for any $g_{1}\in G_{1}\setminus\{1\},g_{2}\in G_{2}\setminus\{1\}$ . In particular

tr_{\mathcal{Q}}\rho_{1}(\theta_{1}(h))=0=tr_{\mathcal{Q}}\rho_{2}(\theta_{2}(h)),

for any $h\in H\setminus\{1\}$ . By [23, Th. 2.5] there exist unitaries $u_{n}\in\mathcal{Q}$ such that

\rho_{1}(\theta_{1}(h))=\lim_{n\to\infty}u_{n}^{*}\rho_{2}(\theta_{2}(h))u_{n},

for any $h\in H$ . Let $\mathcal{Q}_{\omega}$ be the norm ultraproduct of $\mathcal{Q}$ and let $q:\prod\mathcal{Q}\to\mathcal{Q}_{\omega}$ be the canonical surjection. We define embeddings $\tilde{\rho}_{i}:C^{*}(G_{i})\to\mathcal{Q}_{\omega}$ , $i=1,2$ , by

\tilde{\rho}_{1}=q\circ\left(\rho_{1}\right)_{n\in\mathbb{N}},

\tilde{\rho}_{2}=q\circ\left(u_{n}^{*}\rho_{2}u_{n}\right)_{n\in\mathbb{N}}.

Then

\tilde{\rho}_{1}\circ\theta_{1}(h)=\tilde{\rho}_{2}(\theta_{2}(h)),

$h\in H$ . Since MF is a local porperty, $\mathcal{Q}_{\omega}$ is an MF-algebra [22, Prop. 3.2 (3)]. Let $\alpha:\mathcal{Q}_{\omega}\to\prod M_{n}/\oplus M_{N}$ be an embedding. Let $\phi_{i}=\alpha\circ\tilde{\rho}_{i},$ $i=1,2$ . Then $\phi_{1},\phi_{2}$ satisfy the necessary and sufficient condition of Theorem 20. Therefore $C^{*}(G_{1}\ast_{H}G_{2})$ is MF. ∎

Remark 24.

By the same arguments and Theorem 22, Theorem 23 generalizes to amalgamated free products of arbitrarily many amenable groups.

Example 25.

Let us consider the Baumslag-Solitar group $BS(1,n)=\left<a,t|t^{-1}at=a^{n}\right>.$ Out of two copies of $BS(1,n)$ one can form three natural examples of free products amalgamated over an infinite cyclic group:

G_{1}=BS(1,n)\ast_{\left<t_{1}=t_{2}\right>}BS(1,n),

G_{2}=BS(1,n)\ast_{\left<a_{1}=a_{2}\right>}BS(1,n),

G_{3}=BS(1,n)\ast_{\left<a_{1}=t_{2}\right>}BS(1,n).

All three C*-algebras $C^{*}(G_{i})$ , $i=1,2,3$ , are MF by Theorem 23 (for $C^{*}(G_{1})$ and $C^{*}(G_{2})$ this follows already from Theorem 10). In fact $C^{*}(G_{1})$ is RFD [29, Prop. 5.9].

The group $G_{3}$ is a building block of the famous Higman’s group $H\cong G_{3}\ast_{\mathbb{F}_{2}}G_{3}.$ Higman’s group has always been considered as a candidate for counterexample to the approximation conjectures. Now we at least we can see that its building blocks are MF.

Below we find some sufficient condition for amalgamated free products of groups to have RFD full group C*-algebra.

A C*-algebra is residually finite-dimensional (RFD) if it has a separating family of finite-dimensional representations.

A group is called maximally almost periodic (MAP) if it has a separating family of finite-dimensional representations. E.g. any residually finite group is MAP.

The full group C*-algebra of a MAP group might not be RFD as discovered by Bekka [3]. In [28] it was proved that for a large class of amenable groups (e.g. all polycyclic-by-finite groups) free products with amalgamation over central subgroups are RFD. Here we show that amalgamated free products of amenable groups with amalgamation over finite subgroups are RFD.

Below, given a homomorphism $\theta:H\to G$ , the induced $\ast$ -homomorphism $C^{*}(H)\to C^{*}(G)$ will also be denoted by $\theta$ .

Proposition 26.

Suppose $G_{1}$ and $G_{2}$ are amenable, $\theta_{k}:H\to G_{k}$ , $k=1,2$ , are embeddings, and the corresponding amalgamated free product $G_{1}\ast_{H}G_{2}$ is MAP. Then there exist embeddings $j_{k}:C^{*}(G_{k})\hookrightarrow\prod M_{n}$ such that $j_{1}\circ\theta_{1}=j_{2}\circ\theta_{2}$ .

Proof.

Since $G_{1}\ast_{H}G_{2}$ is MAP, delta-function on it can be written as

\delta_{e}=\lim_{n\to\infty}tr\;\pi_{n},

for some finite-dimensional representations $\pi_{n}$ of $G_{1}\ast_{H}G_{2}$ ([13, Prop.3]). Then

\delta_{e}|_{i_{k}(G_{k})}=\lim_{n\to\infty}tr\;\pi_{n}|_{i_{k}(G_{k})},

(4)

$k=1,2$ . Since $G_{k}$ is amenable, delta-function on it extends to a faithful trace on $C^{*}(G_{k})$ . Therefore, (4) implies that

j_{k}:=\prod_{n}\pi_{n}|_{i_{k}(G_{k})}:C^{*}(G_{k})\to\prod M_{n}

is injective. We have

j_{1}\circ\theta_{1}(h)=\prod_{n}\pi_{n}(i_{1}(\theta_{1}(h)))=\prod_{n}\pi_{n}(i_{2}(\theta_{2}(h)))=j_{2}\circ\theta_{2}(h),

$h\in H$ , and therefore $j_{1}\circ\theta_{1}(x)=j_{2}\circ\theta_{2}(x)$ , for any $x\in C^{*}(H)$ . ∎

Corollary 27.

If $G_{1},G_{2}$ are amenable residually finite groups and $H$ is a finite subgroup of them, then $C^{*}(G_{1}\ast_{H}G_{2})$ is RFD.

Proof.

If $G_{1},G_{2}$ are amenable residually finite and $H$ is finite, then $G_{1}\ast_{H}G_{2}$ is RF ([2]), hence is MAP. By Proposition 26 there exist embeddings $j_{k}:C^{*}(G_{k})\hookrightarrow\prod M_{n}$ such that $j_{1}\circ\theta_{1}=j_{2}\circ\theta_{2}$ . Since $C^{*}(H)$ is finite-dimensional, by [16] $C^{*}(G_{1}\ast_{H}G_{2})$ is RFD. ∎

5 Central HNN-extensions and maximal tensor product with $C^{*}(F_{n})$

Below $\otimes$ stands for the maximal tensor product.

Theorem 28.

Let $N\in\mathbb{N}\cup\{\infty\}$ . If a C*-algebra $A$ is MF, then $A\otimes C^{*}(F_{N})$ is MF.

Proof.

Let $\pi:A\otimes C^{*}(F_{N})\to B(H)$ be an embedding. We define a representation $\pi^{\prime}:A\otimes C^{*}(F_{N})\to B(H)$ by

\pi^{\prime}(a)=\pi(a),\;a\in A,

\pi^{\prime}(u_{i})=-\pi(u_{i})^{*},

for each generator $u_{i}$ of $C^{*}(F_{N})$ . Let $\tilde{\pi}=\pi\oplus\pi^{\prime}$ . Since $A$ is MF, by Theorem 4 $\pi|_{A}$ lifts to a discrete asymptotic homomorphism $\phi_{k}:A\to\mathcal{D}$ , $k\in\mathbb{N}$ . Then $\tilde{\pi}|_{A}$ lifts to the discrete asymptotic homomorphism $\tilde{\phi}_{k}=\phi_{k}\oplus\phi_{k},$ $k\in\mathbb{N}$ . Let $\{a_{1},a_{2},\ldots\}$ be a dense subset of $A$ . By Lemma 9 for each $k\in\mathbb{N}$ there is a lift $V_{i,k}$ of the generator $u_{i}$ of $F_{N}$ , $i\leq k$ , such that

\|[V_{i,k},\tilde{\phi}_{k}(a_{l})]\|<\frac{1}{k},

$l\leq k$ . Thus $V_{i,k}$ ’s and $\tilde{\phi}_{k}(a)$ ’s, $a\in A$ , asymptotically satisfy the relations of $A\otimes C^{*}(F_{N})$ . By [26, Lemma 7 + Rem. 8], there exists a discrete asymptotic homomorphism from $A\otimes C^{*}(F_{N})$ to $M_{2}(\mathcal{D})$ that lifts $\tilde{\pi}$ . By Theorem 4, $A\otimes C^{*}(F_{N})$ is MF. ∎

Let $A$ be a unital $C^{*}$ -algebra, $B$ and $C$ its C*-subalgebras, and $\phi:B\to C$ an isomorphism. The corresponding HNN-extension $\langle A,t\;|\;t^{-1}Bt=C,\;\phi\rangle$ is a unital C*-algebra with the following properties:

1) There exists a unital $\ast$ -homomorphism $i_{A}:A\to\langle A,t\;|\;t^{-1}Bt=C,\;\phi\rangle$ and a unitary $t\in\langle A,t\;|\;t^{-1}Bt=C,\;\phi\rangle$ such that

t^{-1}i_{A}(b)t=i_{A}(\phi(b)),

for any $b\in B$ .

2) For any unital C*-algebra $D$ and any unitary $u\in D$ and unital $\ast$ -homomorphism $\pi:A\to D$ such that $u^{-1}\pi(b)u=\pi(\phi(b))$ , for any $b\in B$ , there is a unique $\ast$ -homomorphism $\sigma:\langle A,t\;|\;t^{-1}Bt=C,\;\phi\rangle\to D$ such that $\sigma\circ i_{A}=\pi,\;\sigma(t)=u.$

Such C*-algebra exists and is unique up to isomorphism. An HNN-extension is central if $B=C$ and $\phi=id$ . We will show that the MF-property passes to central HNN-extensions.

Below we identify $A$ with $i_{A}(A)$ to simplify notation.

Theorem 29.

Let $A$ be an MF-algebra and $B$ its C*-subalgebra. Then the HNN-extension $\langle A,t\;|\;t^{-1}Bt=B,\;id\rangle$ is MF.

Proof.

Similar to the proof of Th. 28. Let $\pi:\langle A,t\;|\;t^{-1}Bt=B,\;id\rangle\to B(H)$ be an embedding. We define a representation $\pi^{\prime}:\langle A,t\;|\;t^{-1}Bt=B,\;id\rangle\to B(H)$ by

\pi^{\prime}(a)=\pi(a),\;a\in A,

\pi^{\prime}(t)=-\pi(t)^{*}.

Let $\tilde{\pi}=\pi\oplus\pi^{\prime}$ . Since $A$ is MF, by Theorem 4 $\pi|_{A}$ lifts to a discrete asymptotic homomorphism $\phi_{n}:A\to\mathcal{D}$ , $n\in\mathbb{N}$ . Then $\tilde{\pi}|_{A}$ lifts to the discrete asymptotic homomorphism $\tilde{\phi}_{n}=\phi_{n}\oplus\phi_{n}$ , $n\in\mathbb{N}$ . Let $\{b_{1},b_{2},\ldots\}$ be a dense subset of $B$ . By Lemma 9 $\tilde{\pi}(t)$ lifts to a unitary $V_{n}$ of such that

\|[V_{n},\tilde{\phi}_{n}(b_{i})]\|<\frac{1}{n},

$i\leq n$ . By [26, Lemma 7 + Rem. 8], there exists a discrete asymptotic homomorphism from $\langle A,t\;|\;t^{-1}Bt=B,\;id\rangle$ to $M_{2}(\mathcal{D})$ that lifts $\tilde{\pi}$ . By Theorem 4 $\langle A,t\;|\;t^{-1}Bt=B,\;id\rangle$ is MF. ∎

Corollary 30.

If $C^{*}(H)$ is MF and $G$ is a right-angled Artin group, then $C^{*}(H\times G)$ is MF.

Proof.

Induction on the number $k$ of vertices of the defining graph $\Gamma$ of $G$ . If $k=1$ , $C^{*}(G)=C^{*}(\mathbb{Z})$ and the statement holds by Theorem 28. Suppose it is proved for $k\leq n-1$ . Let $v$ be one of vertices of $\Gamma$ , let $\Gamma_{1}$ be the graph obtained from $\Gamma$ by removing $v$ and all the edges coming from $v$ , and let $G_{1}$ be the right-angled Artin group defined by $\Gamma_{1}$ . Then

C^{*}(H\times G)=\langle C^{*}(H\times G_{1}),t\;|\;t^{-1}Bt=B,\;id\rangle,

where $B$ is the C*-subalgebra of $C^{*}(H\times G_{1})$ generated by $C^{*}(H)$ and $\{w\in V(\Gamma_{1})\;|\;[w,v]=0\}$ . Theorem 29 finishes the proof. ∎

Remark 31.

It is known that if $G$ is a right-angled Artin group, then $C^{*}(G)$ is not only MF, but is even quasidiagonal [1, Th. 4.8].

6 Crossed products

Let $A$ be a unital C*-algebra and $G$ a group that acts on $A$ by automorphisms. We will identify $A$ and $G$ with their canonical images in $A\rtimes G$ . We will identify $G$ and $H$ with their canonical images in $G\ast_{F}H$ .

Proposition 32.

$A\rtimes(G\ast_{F}H)\cong(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)$ .

Proof.

Let

i_{1}:A\rtimes G\to(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H),

i_{2}:A\rtimes H\to(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)

be the canonical inclusions. We define $\alpha_{A}:A\to(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)$ by

\alpha_{A}(a)=i_{1}(a)\equiv i_{2}(a),

$a\in A$ . We define $\alpha_{G\ast_{F}H}:G\ast_{F}H\to(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)$ by

\alpha_{G\ast_{F}H}(g_{1}h_{1}\ldots g_{N}h_{N})=i_{1}(g_{1})i_{2}(h_{1})\ldots i_{1}(g_{N})i_{2}(h_{N}),

$g_{1},\ldots,g_{N}\in G,h_{1},\ldots h_{N}\in H,N\in\mathbb{N}$ . Since the amalgamated C*-subalgebra in $(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)$ contains both $A$ and $F$ , both $\alpha_{A}$ and $\alpha_{G\ast_{F}H}$ are well-defined. We have

\alpha_{G\ast_{F}H}(g_{1}h_{1}\ldots g_{N}h_{N})\alpha_{A}(a)\alpha_{G\ast_{F}H}(g_{1}h_{1}\ldots g_{N}h_{N})^{-1}\\ =i_{1}(g_{1})i_{2}(h_{1})\ldots i_{1}(g_{N})i_{2}(h_{N})i_{2}(a)i_{2}(h_{N}^{-1})i_{1}(g_{N}^{-1})\ldots i_{2}(h_{1}^{-1})i_{1}(g_{1}^{-1})\\ =i_{1}(g_{1})i_{2}(h_{1})\ldots i_{1}(g_{N})i_{1}(h_{N}a)i_{1}(g_{N}^{-1})\ldots i_{2}(h_{1}^{-1})i_{1}(g_{1}^{-1})\\ =\ldots=i_{1}(g_{1}h_{1}\ldots g_{N}h_{n}a)=\alpha_{A}(g_{1}h_{1}\ldots g_{N}h_{N}a).

Thus $\alpha_{A}$ , $\alpha_{G\ast_{F}H}$ is a covariant pair and therefore defines a $\ast$ -homomorphism

\alpha:A\rtimes(G\ast_{F}H)\to(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)

such that

\alpha|_{A}=\alpha_{A},\;\alpha|_{G\ast_{F}H}=\alpha_{G\ast_{F}H}.

We define

\beta:(A\rtimes G)\ast_{A\rtimes F}(A\rtimes H)\to A\rtimes(G\ast_{F}H)

\beta(i_{1}(a))=a,

\beta(i_{1}(g))=g,

\beta(i_{2}(a))=a,

\beta(i_{2}(h))=h,

$a\in A,g\in G,h\in H$ . Since $\beta(i_{1}(a))=\beta(i_{2}(a))$ and $\beta(i_{1}(f))=\beta(i_{2}(f))$ , $\beta$ is a well-defined $\ast$ -homomorphism. It is straightforward to check that $\alpha\circ\beta=id$ and $\beta\circ\alpha=id$ .

∎

Remark 33.

In the proposition above the amalgamated free product of two factors can be replaced by amalgamated free product of more than two factors.

Corollary 34.

$C^{*}(\mathbb{Z}^{2}\rtimes SL_{2}(\mathbb{Z}))$ is MF.

Proof.

$SL_{2}(\mathbb{Z})\cong\mathbb{Z}_{4}\ast_{\mathbb{Z}_{2}}\mathbb{Z}_{6}$ ([25]). Therefore by Proposition 32

\mathbb{Z}^{2}\rtimes SL_{2}(\mathbb{Z})\cong\left(\mathbb{Z}^{2}\rtimes\mathbb{Z}_{4}\right)\ast_{\mathbb{Z}^{2}\rtimes\mathbb{Z}_{2}}\left(\mathbb{Z}^{2}\rtimes\mathbb{Z}_{6}\right).

Since $Z^{2}\rtimes\mathbb{Z}_{4}$ and $\mathbb{Z}^{2}\rtimes\mathbb{Z}_{6}$ are amenable, by Theorem 23 $C^{*}(\mathbb{Z}^{2}\rtimes SL_{2}(\mathbb{Z}))$ is MF. ∎

Corollary 35.

For any semidirect product $G\rtimes\mathbb{F}_{n}$ of an amenable group $G$ by a free group $\mathbb{F}_{n}$ , $C^{*}(G\rtimes\mathbb{F}_{n})$ is MF.

Proof.

By Proposition 32 and Remark 33, $G\rtimes\mathbb{F}_{n}$ is a free product of $G\rtimes\mathbb{Z}$ amalgamated over $G$ . Since $G\rtimes\mathbb{Z}$ is amenable, the statement follows from Theorem 23 and Remark 24. ∎

References

[1] S. Atkinson, Some results on tracial stability and graph products, Indiana University Mathematics Journal 70(3): 1167-1187, 2021.
[2] G. Baumslag, On the residual finiteness of generalized free products of nilpotent groups, TAMS 106:193-209, 1963.
[3] B. Bekka, On the full C*-algebras of arithmetic groups and the congruence subgroup problem, Forum Math. 11:705-715, 1999.
[4] B. Blackadar and E. Kirchberg, Generalized inductive limits of finite dimensional C*-algebras, Math. Ann. 307:343-380, 1997.
[5] N. Brown, Invariant Means and Finite Representation Theory of C*-algebras, Memoirs of the American Mathematical Society 184(865), 2013.
[6] J. Carrion, M. Dadarlat, and C. Eckhardt, On groups with quasidiagonal C*-algebras, J. Funct. Anal. 265(1): 135-152, 2013.
[7] A. Connes and N. Higson, Deformations, morphismes asymptotiques et K-theorie bivariante, C. R. Acad. Sci. Paris, I. Ser., Math. 311: 101-106, 1990.
[8] M. Dadarlat, A note on asymptotic homomorphisms, K-Theory 8: 465–482, 1994.
[9] D. Enders and T. Shulman, On the (Local) Lifting Property, arXiv:2403.12224.
[10] U. Haagerup and S. Thorbjornsen, A new application of random matrices: $Ext(C^{*}_{red}(\mathbb{F}_{2}))$ is not a group, Ann. of Math. (2), 162(2):711-775, 2005.
[11] D. Hadwin, Q. Li and J. Shen, Topological free entropy dimensions in Nuclear C*-algebras and in Full Free Products of Unital C*-algebras, Canad. J. Math. 63, 2011.
[12] D. Hadwin and T. Shulman, Tracial stability for C*-algebras, Int. Eq. Op. Th. 90:1, 2018.
[13] D. Hadwin and T. Shulman, Stability of group relations under small Hilbert-Schmidt perturbations, J. Funct. Anal. 275 (4):761-792, 2018.
[14] G. Higman, A finitely related group with an isomorphic proper factor group, J. London Math. Soc. 26: 59-61, 1951.
[15] A. Ioana and P. Spaas: $II_{1}$ factors with exotic central sequence algebras, J. Inst. Math. Jussieu 20: 1671-1696, 2021.
[16] Q. Li and J. Shen, A note on unital full amalgamated free products of RFD $C^{*}$ -algebras, Illinois J. Math. 56:647-659, 2012.
[17] T. Loring, Lifting solutions to perturbing problems in C*-algebras, volume 8 of Fields Institute Monographs. AMS, Providence, RI, 1997.
[18] A. Lubotzky and Y. Shalom, Finite representations in the unitary dual and Ramanujan groups, Discrete geometric analysis: proceedings of the first JAMS Symposium on Discrete Geometric Analysis, December 12-20, 2002, Sendai, Japan, Contemporary Mathematics 347:173-189, 2004.
[19] M. Magee and M. de la Salle, $SL_{4}(\mathbb{Z})$ is not purely matricial field, Comptes Rendus Mathématique 362: 903-910, 2024.
[20] V. Manuilov, Asymptotic homomorphisms of C*-algebras and C*-extensions, J. Math. Sciences 123(4), 2004.
[21] V. Manuilov and K. Thomsen, The Connes-Higson construction is an isomorphism, Journal of Functional Analysis 213: 154-175, 2004.
[22] T. Rainone and C. Schafhauser, Crossed products of nuclear C*-algebras by free groups and their traces, Adv. Math., 347: 105-149, 2019.
[23] C. Schafhauser, Finite dimensional approximations for certain amalgamated free products of groups, to appear in Groups, Geom. Dym., 2023.
[24] C. Schafhauser, A new proof of the Tikuisis-White-Winter Theorem, J. Reine Angew. Math. 759: 291-304, 2020.
[25] J-P. Serre, Trees, Springer-Verlag, 1980.
[26] T. Shulman, Sections and cones, arXiv:2507.22783.
[27] T. Shulman, Homotopy lifting, asymptotic homomorphisms, and traces, arXiv:2508.00125.
[28] T. Shulman, Central amalgamation of groups and the RFD property, Advances in Mathematics 394 (22), 2022.
[29] T. Shulman and A. Skalski, RFD property for groupoid C*-algebras of amenable groupoids and for crossed products by amenable actions, arXiv:2305.12122.
[30] A. Tikuisis, S. White, and W. Winter, Quasidiagonality of nuclear C*-algebras, Annals of Math. (2), 185: 229-284, 2017.

The MF property for amalgamated free products

Abstract

1 Introduction

2 Preliminaries

2.1 Amalgamated free products

Definition 1.

2.2 MF C*-algebras and groups

Definition 2.

Definition 3.

2.3 Asymptotic homomorphisms

3 Amalgamated free products A∗CAA\ast_{C}A

Theorem 4.

Remark 5.

Remark 6.

Lemma 7.

Proof.

Lemma 8.

Proof.

Lemma 9.

Proof.

Theorem 10.

Proof.

Example 11.

Example 12.

3.1 MF traces

Proposition 13.

Theorem 14.

Corollary 15.

Example 16.

3.2 General amalgamated free products

Theorem 17.

Remark 18.

Lemma 19.

Theorem 20.

Proof.

Lemma 21.

Proof.

Theorem 22.

Proof.

4 Amalgamated free products of amenable groups

Theorem 23.

Proof.

Remark 24.

Example 25.

Proposition 26.

Proof.

Corollary 27.

Proof.

5 Central HNN-extensions and maximal tensor product with C∗​(Fn)C^{*}(F_{n})

Theorem 28.

Proof.

Theorem 29.

Proof.

Corollary 30.

Proof.

Remark 31.

6 Crossed products

Proposition 32.

Proof.

Remark 33.

Corollary 34.

Proof.

Corollary 35.

Proof.

References

3 Amalgamated free products $A\ast_{C}A$

5 Central HNN-extensions and maximal tensor product with $C^{*}(F_{n})$