Introduction to sofic and hyperlinear groups and Connes' embedding conjecture. With an appendix by Vladimir Pestov (Q2348139)

Let \(n\in \mathbb{N}\) and \(S_{n}\) be the group of permutations on \(n\) symbols. The function \(\ell _{S_{n}}\,:\,S_{n}\,\longrightarrow\,[0,\,1]\) defined by \(\ell _{S_{n}}(\sigma )\,=\,\frac{1}{n}\mid \{\,i\,:\,\sigma (i)\,\not =\,i\,\}\mid\) is called the \textit{Hamming invariant length function} on \(S_{n}\). Definition 1. A countable discrete group \(\Gamma\) is called \textit{sofic} if for every \(\varepsilon > 0\) and every finite subset \(F\) of \(\Gamma \setminus\{\,1_{\Gamma }\,\}\) there is a natural number \(n\) and a function \(\Phi :\Gamma \,\longrightarrow\, S_{n}\) such that \(\Phi (1_{\Gamma })\,=\,1_{S{n}}\) and for every \(g,\,h\in F\) the following inequalities are satisfied. \( \bullet \quad d_{S_{n}}(\Phi (gh),\,\Phi (g) \Phi (h))<\varepsilon\); \( \bullet \quad \ell _{S_{n}}(\Phi (g))> r(g)\) \quad where \(r(g)\) is a positive constant depending only on \(g\). Let \(M_{n}(\mathbb{C})\) be the tracial von Neumann algebra of \(n\times n\) matrices over the complex numbers. If \(U_{n}\) is the group of the unitary elements of \(M_{n}(\mathbb{C})\) (i.e. the elements \(x\in M_{n}(\mathbb{C})\) with \(x^{\ast}x\,=\,xx^{\ast}\,=\,1\)), then the \(\ell _{U_{n}}\,:\,U_{n}\,\longrightarrow\,[0,\,1]\) defined by \(\ell _{U_{n}}(u )\,=\,\frac{1}{2}\parallel u-1\parallel_{2}\) is called the \textit{Hilbert-Schmidt invariant length function} on \(U_{n}\). Here, \(\parallel\cdot \parallel _{2}\) denotes the \textit{Hilbert-Schmidt norm} on \(M_{n}(\mathbb{C})\) defined by \(\parallel x \parallel _{2}\,=\,\tau (x^{\ast }x)^{1/2}\), where \(\tau \) is the normalized trace on \(M_{n}(\mathbb{C})\) defined by \(\tau ((a_{ij}))\,=\,\frac{1}{n}\sum_{i\,=\,1}^{n}a_{ii}\). If in Definition 1 above, the symmetric group \(S_{n}\) is replaced by the group \(U_{n}\) of the unitary elements of \(M_{n}(\mathbb{C})\) we have the definition of the \textit{hyperlinear} groups. Definition 2. A countable discrete group \(\Gamma\) is called \textit{hyperbolic} if for every \(\varepsilon>0\) and every finite subset \(F\) of \(\Gamma \setminus\{\,1_{\Gamma }\,\}\) there is a natural number \(n\) and a function \(\Phi :\Gamma \,\longrightarrow\, U_{n}\) such that \(\Phi (1_{\Gamma })\,=\,1_{U{n}}\) and for every \(g,\,h\in F\) the following inequalities are satisfied. \( \bullet \quad d_{U_{n}}(\Phi (gh),\,\Phi (g) \Phi (h))< \varepsilon\); \( \bullet \quad \ell _{U_{n}}(\Phi (g))>r(g)\) \quad where \(r(g)\) is a positive constant depending only on \(g\) . In this monograph, the authors develop the theory of sofic and hyperlinear groups. These groups can be approximated, in a sense, by finite symmetric groups and by groups of unitary matrices respectively. These approximation properties initially introduced, almost before 20 years, by Gromov and Rádulescu gave a new boost in the study of these groups and mainly to prove that several long-standing conjectures, that are still open for arbitrary groups, are valid in the case of the sofic and hyperlinear groups. In the first chapter, (the Introduction) Connes', Kaplansky's and Gottschalk's conjectures are presented. A short discussion and some comments concerning the history of these conjectures are very useful for the rest. A tracial von Neumann algebra is called a \textit{factor} if its center contains only the scalar multiplies of the identity. An infinite dimensional factor is called a II\(_{1}\) \textit{factor}. (In Paragraph 2.11 of the text, a discussion on tracial von Neumann algebras takes place and characterizations for factors and II\(_{1}\) factors are given.) Definition 3: Suppose that \(M\) and \(N\) are two factors. If \(F\) is a subset of \(M\) and \(\varepsilon\) is a positive real number, then an \((F,\varepsilon )\)-approximate morphism from \(M\) to \(N\) is a function \(\Phi : M\to N\) such that \( \Phi (1)\,=\,1 \) and for every \(x,\,y\in F\) the following inequalities are satisfied. \( \bullet \quad \parallel\Phi (x+y)-(\Phi (x)+ \Phi (y))\parallel _{2}< \varepsilon\); \( \bullet \quad \parallel\Phi (xy)-\Phi (x) \Phi (y)\parallel _{2}< \varepsilon\); \(\bullet \quad \mid \tau _{M}(x)-\tau _{N}(\Phi (x))\mid<\varepsilon\). A II\(_{1}\) factor \(M\) satisfies Connes' embedding conjecture if for every finite \(F\) subset of \(M\) and every positive real number \(\varepsilon\) there is a natural number \(n\) and an \((F,\varepsilon )\)-approximate morphism \(\Phi :M\to M_{n}(\mathbb{C})\). A discrete group \(\Gamma \) satisfies Gottschalk's conjecure if for every finite set \(A\) and every continuous injective equivariant function \(f: A^{\Gamma}\,\longrightarrow\, A^{\Gamma}\) implied that \(F\) is surjective. Let \(K\) be a field, a discrete countable group \(\Gamma\) satisfies Kaplansky's conjecture if the group algebra \(K\Gamma \) is directly finite i.e. every one-side invertible element of \(K\Gamma\) is invertible. The second chapter begins with the definitions of sofic and hyperlinear groups and establishes the equivalence of these notions with the existence of embeddings of sofic and hyperlinear groups into the ultrafilters \(\prod_{\mathcal{U}}S_{n}\) and \(\prod_{\mathcal{U}}U_{n}\), respectively. This equivalence allows to see that large classes of groups are sofic or hyperlinear groups and that any countable discrete sofic group is hyperlinear. For example, finite groups, amenable groups and, more generally, residually finite groups are sofic groups. Also, it is proved that the class of sofic groups is closed with respect to a list of operations (Proposition 2.4.1). In a sequel, more examples of sofic groups are demonstrated. On the other hand, there is not any example of a countable discrete group, which is not hyperlinear or not sofic group. The following interesting question is still open. Open Problem. Is there any countable discrete group that is not sofic or hyperlinear? Next, a short introduction of logic for metric structures is written to describe and study groups endowed with invariant length functions. With this, it is shown that the classes of sofic and hyperlinear groups are axiomatizable in the logic for invariant metric groups. Equivalently, the properties of being sofic or hyperlinear are elementary. Let \(\Gamma\) be a countable discrete group and the monomial \(w(x)\,=\,x^{n_{1}}g_{1}\,\cdots\,x^{n_{m}}g_{m}\) with \(g_{i}\in \Gamma \) and \(n_{i}\in \mathbb{Z}\). A conjecture attributed to Kervaire and Laudenbach asserts that, if the sum \(\sum _{i\,=\,1}^{m}n_{i} \) is not zero, then the equation \(w(x)\,=\,1\) has a solution in some group extension of \(\Gamma\). This conjecture holds for hyperlinear groups. Sofic groups satisfy Kaplansky's conjecture. Here, this is shown within the framework of rank rings. Very roughly, it is obtained as follows. A sofic group \(\Gamma\) embeds into \(\prod_{\mathcal{U}}S_{n}\). This induces a ring homomorphism from \(K\Gamma\) into \(K(\prod_{\mathcal{U}}S_{n})\). Another homomorphism is defined from \(K(\prod_{\mathcal{U}}S_{n})\) into \(\prod_{\mathcal{U}}M_{n}(K)\). The composition of these homomorphisms is one to one. Since the ring \(\prod_{\mathcal{U}}M_{n}(K)\) is directly finite, the result follows. Let \(\Gamma \) be a countable discrete group. The natural action of the group algebra \(\mathbb{C} \Gamma\) on the Hilbert space \(\ell ^{2}\Gamma \) defines an inclusion of \(\mathbb{C} \Gamma\) into \(B(\ell ^{2}\Gamma )\). The algebraic eigenvalues conjecture asserts that the elements of the group ring \(\mathbb{Z} \Gamma\), regarded as linear operations on \(\ell ^{2}\Gamma\), have algebraic integers as eigenvalues. If \(\Gamma \) is a sofic group, the embedding of \(\Gamma \) into \(\prod_{\mathcal{U}}S_{n}\) allows to see that the group ring is \(\mathbb{Z} \Gamma\) embedded into \(\prod_{\mathcal{U}}M_{n}(\mathbb{Z})\). It is proved that the elements of \(\prod_{\mathcal{U}}M_{n}(\mathbb{Z})\) have algebraic eigenvalues. Therefore, the algebraic eigenvalues conjecture is satisfied for the sofic groups. The chapter concludes with the proof of Gottschalk's conjecure for sofic groups. It is proved firstly by Gromov using graph-theoretical methods. Here, the proof of \textit{D. Kerr} and \textit{H. Li} [Am. J. Math. 135, No. 3, 721--761 (2013; Zbl 1282.37011)] is developed, which is obtained as an application of the theory of entropy for actions of sofic groups. In the third chapter, with a motivation of the discussion of the relationship between hyperlinear groups and Connes' embedding conjecture, equivalent reformulations of Connes' embedding conjecture are developed. More precisely, Kirchberg's theorem is developed. Theorem 1. The following statments are equivalent: \(1.\)\quad Connes' embedding conjecture holds true. \(2.\)\quad \(C^{\ast}(\mathbb{F}_{\infty})\,\otimes _{\max}\,C^{\ast}(\mathbb{F}_{\infty})\,=\, C^{\ast}(\mathbb{F}_{\infty})\,\otimes _{\max}\,C^{\ast}(\mathbb{F}_{\infty}))\). (In Appendix A, the necessary material to understand the statement of Kirchberg's theorem is quoted.) This theorem is an astonishing link between the theory of von Neumann algebras and the theory of \(C^{\ast}\)-algebras. Kirchberg proceeded further. Referring to the terminology in the book he proved: Theorem 2. Let \(M\) be a separable II\(_{1}\) factor. The following statements are equivalent \(1.\)\quad \(M\) is embeddable into some \(\mathcal{R}^{\mathcal{U}}\). \(2.\)\quad \(M\) has the quotient weak expectation property (QWEP). From this theorem, the natural question is raised whether there is a direct relation between Connes' embedding conjecture and WEP. Here, Brown's affirmative answer is developed: Theorem 3. Let \(M\) be a separable II\(_{1}\) factor and \(\mathcal{U}\) be a free ultrafilter on the natural numbers. The following conditions are equivalent \(1.\)\quad \(M\) is embeddable into \(\mathcal{R}^{\mathcal{U}}\). \(2.\)\quad \(M\) has the weak expectation property (WEP) relative to some weakly dense subalgebra. An purely algebraic approach of Connes' embedding conjecture is also descriptively developed. This is initiated by Rádulesku and carried over by Klep-Schweighofer and Juschenko-Popovich. The purpose of this section is to introduce the reader to a new field of research. Here, we merely quote the statements of three theorems (referring for the terminology in the text ) concerning equivalent reformulations of Connes' embedding conjecture Theorem 4. The following statements are equivalent. {\parindent=0.5 cm \begin{itemize}\item[1.] Connes' embedding conjecture is true; \item[2.] The non-commutative analogue of Hilbert's 17th problem has a positive answer. \end{itemize}} Theorem 5. The following statements are equivalent. {\parindent=0.5cm \begin{itemize}\item[1.]Connes' embedding conjecture is true; \item[2.] For every \(f\in V^{+}\) and for every \(\varepsilon>0\), there exists \(q\in Q\) such that \(f+\varepsilon \) is equivalent to \(q\), in the sense that \(f+\varepsilon -q\) is a sum of commutators. \end{itemize}} Theorem 6. The following statements are equivalent. {\parindent=0.5 cm \begin{itemize}\item[1.] Connes' embedding conjecture is true; \item[2.] For every self-adjoint positive semidefinite \(f\in \mathcal{F}\) and for every \(\varepsilon>0\), one has \(f+\varepsilon=g+c\), where \(g\) is a sum of squares (elements of the form \(u^{\ast}u\)) and \(c\) is a sum of commutators. \end{itemize}} Finally, another point of view is discussed. Recently, N. Brown assumed that a fixed separable II\(_{1}\) factor \(M\) satisfies Connes' embedding conjecture and, associated an invariant to \(M\), tried to investigate interesting properties of \(M\). The book concludes with two Appendixes. Especially, Appendix B (by V. Pestov) provides a comprehensible introduction to the concepts of ultrafilters, ultralimits, and ultraproducts and it is recommended to be read before the (non familial ) reader goes into the main body of the book. As this monograph originated from an intensive course, it is dense and many results are ``offered'' as exercises. It is addressed to the (really) interested readers on the topic giving them the incentive and the opportunity for proceeding.