A new application of random matrices: \(\operatorname{Ext} (C_{\text{red}}^*(F_2))\) is not a group (Q2496969)

Given a separable unital C\(^*\)-algebra \(\mathcal{A}\), the invariant Ext\((\mathcal{A})\) was introduced by \textit{L.~G.\ Brown, R.~G.\ Douglas} and \textit{P.~A.\ Fillmore} in [Lect.\ Notes Math.\ 345, 58--128 (1973; Zbl 0277.46053) and Ann.\ Math.\ (2) 105, 265--324 (1977; Zbl 0376.46036)]. It is defined as the set of essential unitary equivalence classes \([\pi]\) of unital \(*\)-monomorphisms \(\pi\) of \(\mathcal{A}\) into the Calkin algebra \(\mathcal{C}(\mathcal{H})\) of the Hilbert space \(\mathcal{H}=\ell^2(\mathbb{N})\). Namely, if \(\rho:\mathcal{B}(\mathcal{H})\rightarrow\mathcal{C}(\mathcal{H})\) is the quotient map, then \[ \pi_1\sim\pi_2\;\iff \;\exists u\in\mathcal{U}(\mathcal{B}(\mathcal{H})), \pi_2=\text{Ad}\,\rho(u)\,\circ\,\pi_1. \] From the fact that \(\mathcal{H}\oplus\mathcal{H}\simeq\mathcal{H}\), one can define a semigroup structure on Ext\((\mathcal{A})\) by \([\pi_1]+[\pi_2]=[\pi_1\oplus\pi_2]\). \textit{D.~Voiculescu} [Rev.\ Roum.\ Math.\ Pures Appl.\ 21, 97--113 (1976; Zbl 0335.46039)] showed that when \(\mathcal{A}\) is separable and unital, Ext\((\mathcal{A})\) has a neutral element; in the same year, \textit{M.--D.\ Choi} and \textit{E.~G.\ Effros} [Ann.\ Math.\ (2) 104, 585--609 (1976; Zbl 0361.46067)] showed that Ext\((\mathcal{A})\) is a group when \(\mathcal{A}\) is nuclear. \textit{J.~Anderson} [Ann.\ Math.\ (2) 107, 455--458 (1978; Zbl 0363.46053)] found the first example of a unital C\(^*\)-algebra with Ext\((\mathcal{A})\) not a group. Later, \textit{S.~Wassermann} [Ann.\ Math.\ (2) 134, No.~2, 423--431 (1991; Zbl 0754.46040)] provided more examples of C\(^*\)-algebras \(\mathcal{A}\) for which Ext\((\mathcal{A})\) is not a group. \textit{D.~Voiculescu} pointed out in \S\ 5.11 of [Integral Equations Oper.\ Theory 17, No.~1, 137--149 (1993; Zbl 1048.47501)] that \(C^*_{\text{red}}(\mathbb{F}_n)\) -- the reduced C\(^*\)-algebra of a free group -- is a natural example to consider when looking for further instances where Ext fails to be a group. Let \(\lambda\) denote the left regular representation of \(\mathbb{F}_n\). In \S\ 5.12 and \S\ 5.14 from the same paper, it was shown that if for each \(k\in\mathbb{N}\), there exist unitary representations \(\rho_k\) of the group algebra \(\mathbb{C}(\mathbb{F}_n)\) into \(M_k(\mathbb{C})\) such that \[ \lim_{k\rightarrow\infty}\| \rho_k(X)\| =\| \lambda(X)\| , \tag{1} \] then Ext\((C_{\text{red}}^*(\mathbb{F}_n))\) is not a group (the argument is nicely explained in Remark 8.6 of the paper under review). The main result can be stated as follows. Let \((\Omega,{\mathcal F}, P)\) be a probability space, \(m\in\mathbb{N} \cup\{\infty\}\), \((X_k^{(n)})_{k=1}^m\) independent \(n\times n\) Gaussian self-adjoint random matrices (in random matrix language, \(X_k^{(m)}\in\text{SGRM}(n,\frac1n)\)), let \((x_i)_{i=1}^m\) be a semi-circular family in a C\(^*\)-probability space \((\mathcal{B},\tau)\), and let \(p\) be a noncommutative complex polynomial in \(m\) variables. Then, \[ \lim_{n\rightarrow\infty}\,\| p((X_k^{(n)}(\omega))_{k=1}^m)\| =\| p((x_i)_{i=1}^m)\| , \text{ \(P\)-a.e. in }\omega.\tag{2} \] The proof of this theorem appears in Section 7, requiring many prerequisites developed in Sections 2--6. In Section 8, it is shown how (2) implies (1) (as a matter of fact, just one point \(\omega\) of sure convergence is enough, as pointed out by the authors). In Section 9, several applications are given, including an extension of (2) to \(\mathcal{A}\)-valued polynomials (with \(\mathcal{A}\) a unital exact C\(^*\)-algebra), and the determination of the numbers \(C(r)\) introduced by \textit{M.~Junge} and \textit{G.~Pisier} in [Geom.\ Funct.\ Anal.\ 5, No.~2, 329--363 (1995; Zbl 0832.46052)].