A note on strong similarity and the Connes embedding problem ORCID 0000-0002-3091-2049

by · 2026

Read A note on strong similarity and the Connes embedding problem ORCID 0000-0002-3091-2049 by by (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-10654.txt.

Introduction

A short note in operator algebra takes up the Connes embedding problem, a long-running question about von Neumann algebras. The note shows that a certain completely bounded map exists. The map runs from a C*-algebra with the lifting property into a QWEP von Neumann algebra. The key point is what the map cannot be.

The map is not strongly similar to a *-homomorphism, the nicest kind of map between such algebras. That gap between bounded maps and star-shaped maps carries the result. Notes of this kind sort out which algebra maps can be cleaned up and which cannot.

Catalog record

CREATORby
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE12 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-10654

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A note on strong similarity and the Connes embedding problem ORCID 0000-0002-3091-2049 by Abstract We show that there exists a completely bounded (c.b. in short) homomorphism u u from a C ∗ C^{*} -algebra C C with the lifting property (in short LP) into a QWEP von Neumann algebra N N that is not strongly similar to a ∗ * -homomorphism, i.e. the similarities that “orthogonalize” u u (which exist since u u is c.b.) cannot belong to the von Neumann algebra N N . Moreover, the map u u does not admit any c.b. lifting up into the WEP C ∗ C^{*} -algebra of which N N is a quotient. We can take C = C ∗ ​ ( 𝔽 ) C=C^{*}(\mathbb{F}) (full C ∗ C^{*} -algebra) where 𝔽 \mathbb{F} is any nonabelian free group and N = B ⁡ ( H ) ​ ⊗ ¯ ​ M N=B(H)\bar{\otimes}M where M M is the von Neumann algebra generated by the reduced C ∗ C^{*} -algebra of 𝔽 \mathbb{F} . In this note (which is not intended for publication) we produce a c.b. homomorphism u from a C^{*} -algebra C with LP (which we could choose to be C^{*}(\mathbb{F}_{n}) if we wish using free Haar unitaries in place of (x_{j}) ) into a QWEP von Neumann algebra N which is NOT strongly similar to a * -homomorphism, here by strongly similar we mean that the similarity can be chosen in the von Neumann algebra N . This means that a certain statement from [ 6 ] crucially used in [ 1 ] is not valid, as well as the application proposed in [ 1 ] to the Connes embedding problem. If N=B/I with B WEP ( I closed two sided ideal in B ) the map u does not admit any c.b. lifting. Indeed the paper [ 1 ] shows that if it did then u would be strongly similar to a * -homomorphism. Remark 0.1 . By known results (see [ 2 ] ) there are c.b. maps u:C\to B/I that do not admit any c.b. lifting. Apparently the existence of a c.b. homomorphism is new. Let \ell_{j} (resp. r_{j} ) denote the left (resp. right) creators on the full Fock space H so that x_{j}=\ell_{j}+\ell_{j}^{*}\quad(1\leq j\leq n) are free semicirculars generating the von Neumann algebra M\subset B(H) and their commutant M^{\prime} is generated by y_{j}=r_{j}+r_{j}^{*}. Let C=\ast_{j}C^{*}(x_{j}) (full free product). Let \sigma:C\to M be the natural representation, so that M=\sigma(C)^{\prime\prime} , and let \pi:C\to M\bar{\otimes}B(H) be defined by \pi(x)=\sigma(x)\otimes 1 . Let N=B(H)\bar{\otimes}M . We will produce a c.b. derivation \delta:C\to N relative to \pi with \|\delta\|_{cb}\leq 2 but such that any T in N such that \delta(x)=[T,\pi(x)] satisfies \|T\|\geq\sqrt{n}/4 . Then we will show that the associated homomorphism u:C\to M_{2}(N) defined by C\ni x\mapsto\left(\begin{matrix}x\otimes 1&\ \delta(x)\\ 0&\ x\otimes 1\end{matrix}\right) is also c.b. but if n=\infty it is not strongly similar to a * -homomorphism. An estimate will also be given for large finite n ’s. The construction we use is a particular case of the one described in [ 4 ] . Let S=\sum-r_{j}\otimes\ell_{j}^{*}+r_{j}^{*}\otimes\ell_{j}\in B(H\otimes H) Clearly \|\sum\ell_{j}\ell_{j}^{*}\|=\|\sum r_{j}r_{j}^{*}\|\leq 1 and hence \|S\|\leq 2. Let C=\ast_{j}C^{*}(x_{j}) (full free product) We denote by \pi:C\to M\otimes 1 the representation from the full free product to M taking x_{j} to x_{j}\otimes 1 . Note that C has the LP. Let \delta:C\to B(H\otimes H) be defined by \delta(x)=[x\otimes 1,S]. Then \delta is inner with (0.1) \|\delta\|_{cb}\leq 2. Since [x_{i},y_{j}]=0 we have \delta(x)=[x\otimes 1,\sum r_{j}^{*}\otimes\ell_{j}^{*}+r_{j}^{*}\otimes\ell_{j}]=[x\otimes 1,\sum r_{j}^{*}\otimes x_{j}]=\sum[x,r_{j}^{*}]\otimes x_{j}=-\sum[x,r_{j}]\otimes x_{j}. Clearly the von Neumann algebra generated by \pi(C) and \delta(C) is included in B(H)\bar{\otimes}\{x_{j}\}^{\prime\prime}=B(H)\bar{\otimes}M . Let T\in B(H)\bar{\otimes}M be such that for all x\in C \delta(x)=[x\otimes 1,T]. Since we have a projection P on M with range the span of (x_{j}) with \|P\|_{cb}\leq 2 we must have \delta(x)=(Id\otimes P)\delta(x)=[x\otimes 1,T_{1}] with T_{1}=(Id\otimes P)(T). It follows that T_{1}=\sum z_{j}\otimes x_{j} and (0.2) \max\{\|\sum z_{j}z_{j}^{*}\|^{1/2},\|\sum z_{j}^{*}z_{j}\|^{1/2}\}\leq\|T_{1}\|\leq 2\|T\|. Moreover since \delta(x)=\sum[x,r_{j}^{*}]\otimes x_{j}=-\sum[x,r_{j}]\otimes x_{j}=\sum[x,z_{j}]\otimes x_{j} we must have r_{j}^{*}-z_{j}\in M^{\prime}\text{ and }r_{j}+z_{j}\in M^{\prime} y_{j}=(r_{j}+z_{j})+(r_{j}^{*}-z_{j}) and hence \sqrt{n}=(\sum\|y_{j}\Omega\|^{2})^{1/2}\leq(\sum\|(r_{j}+z_{j})\Omega\|^{2})^{1/2}+(\sum\|(r_{j}^{*}-z_{j})\Omega\|^{2})^{1/2} but by the trace identity in M^{\prime} (\sum\|(r_{j}+z_{j})\Omega\|^{2})^{1/2}=(\sum\|(r_{j}+z_{j})^{*}\Omega\|^{2})^{1/2}=(\sum\|z_{j}^{*}\Omega\|^{2})^{1/2} and clearly (\sum\|(r_{j}^{*}-z_{j})\Omega\|^{2})^{1/2}=(\sum\|z_{j}\Omega\|^{2})^{1/2} so that we obtain by ( 0.2 ) (0.3) \sqrt{n}\leq\|\sum z_{j}z_{j}^{*}\|^{1/2}+\|\sum z_{j}^{*}z_{j}\|^{1/2}\leq 2\|T_{1}\|\leq 4\|T\|. Now it remains to check that the von Neumann algebra generated N by \pi(C) (or M\otimes 1 ) and \delta(C) is QWEP. We will show that N=B(H)\bar{\otimes}M , which is clearly QWEP since M being QWEP is exactly the original observation of Connes in the 1976 paper where he raised what is now known as the Connes embedding problem. Recall \delta(x)=\sum[x,r_{j}^{*}]\otimes x_{j}=-\sum[x,r_{j}]\otimes x_{j} Observe that [l_{k},r_{j}]=[l_{k}^{*},r_{j}^{*}]=0 for all k,j and also [l_{k},r_{j}^{*}]=0 if k\not=j and [l_{k},r_{k}^{*}]=P_{\Omega} . Thus \delta(x_{k})=P_{\Omega}\otimes x_{k} which implies that N contains all elements of the form aP_{\Omega}b\otimes x_{k} ( a,b\in M ), i.e. all elements in F(H)\otimes x_{k} where F(H) denotes the space of finite rank operators on H . It follows that N contains F(H)\otimes M and hence N\supset B(H)\bar{\otimes}M . Since we already noted that the converse is obvious, this proves as announced N=B(H)\bar{\otimes}M . The arguments in [ 1 ] give the following: Proposition 0.2 . Consider u:C\to N a unital cb homomorphism with values in a von Neumann algebra N in B(H) . Let v:N^{\prime}\to B(H) be the inclusion map. Then u is strongly similar to a * -homomorphism iff the map u.v:C\otimes N^{\prime}\to B(H) (defined by u.v(a\otimes b)=u(a)v(b) extends to a c.b. map w from C\otimes_{\max}N^{\prime} to B(H) . Moreover \|w:C\otimes_{\max}N^{\prime}\to B(H)\|_{cb}=\inf\{\|T\|\|T^{-1}\|\} where the (attained) inf runs over all invertible T\in N such that Tu(.)T^{-1} is a * -homomorphism. Proof. Assume w c.b. Clearly w is still a homomorphism, and hence if it is c.b. it must be similar to a * -homomorphism whence the existence of an invertible T on H (with \|T\|\|T^{-1}\|=\|w\|_{cb} ) such that x\mapsto Tw(x)T^{-1} is a * -homomorphism. But restricting this to 1\otimes N^{\prime} we find that N^{\prime}\ni y\mapsto TyT^{-1} is a * -homomorphism which implies by a simple calculation ( TyT^{-1}=(Ty^{*}T^{-1})^{*} and hence T^{*}Ty=yT^{*}T ) that |T|\in N^{\prime\prime}=N . Now T=U|T| with U unitary and the map C\ni x\mapsto U(|T|u(x)|T|^{-1})U^{-1} is a * -homomorphism. Since U is unitary C\ni x\mapsto(|T|u(x)|T|^{-1}) is a * -homomorphism, proving the strong similarity of u . Conversely, if T\in N is such that C\ni x\mapsto(Tu(x)T^{-1}) is a * -homomorphism denoted by \rho:C\to N , then obviously \rho.v is a (completely contractive) * -homomorphism from C\otimes_{\max}N^{\prime} to B(H) . Then a\otimes b\mapsto T^{-1}\rho.v(a\otimes b)T has cb-norm at most \|T\|\|T^{-1}\| on C\otimes_{\max}N^{\prime} and since T commutes with v(b) , we have T^{-1}\rho.v(a\otimes b)T=u.v(a\otimes b) . Thus we conclude that \|u.v:C\otimes_{\max}N^{\prime}\to B(H)\|_{cb}\leq\|T\|\|T^{-1}\| . The converse inequality was proved in the first part. ∎ Let \delta be as before. Then C\ni x\mapsto u(x)=\left(\begin{matrix}x\otimes 1&\ \delta(x)\\ 0&\ x\otimes 1\end{matrix}\right) is a c.b. homomorphism u into M_{2}(N) . We have \|u\|_{cb}\leq 1+\|\delta\|_{cb}\leq 3 by ( 0.1 ) (and replacing \delta by (\varepsilon/2)\delta we could get \|u\|_{cb}\leq 1+\varepsilon if desired). We know that M_{2}(N) is QWEP, so we have M_{2}(N)=B/I with B WEP. Lemma 0.3 . Let S be an invertible in M_{2}(N) such that x\mapsto Su(x)S^{-1} is a * -homomorphism. Then there is T in N with \|T\|\leq(\|S\|\|S^{-1}\|)^{2} such that \delta(x)=[x,T] for any x\in C . Proof. See [ 3

Read the complete text