Fuglede-Kadison determinants and sofic entropy (Q2628931)

Let \(\Gamma\) be a countable group and \(\mathcal{R}\) a commutative ring. We denote by \(\mathcal{R}(\Gamma)\) the ring of finite formal linear combinations of elements of \(\Gamma\) with addition defined naturally and the multiplication defined by \[ \Big(\sum_{g \in \Gamma}a_gg\Big) \Big(\sum_{g \in \Gamma}b_gg\Big)= \sum_{g \in \Gamma}\Big(\sum_{g \in \Gamma}a_{h}b_{h^{-1}g}\Big)g. \] If \(\mathcal{R}(\Gamma)\) is the ring of complex numbers \(\mathbb{C}\). Then, we also define a conjugate-linear involution on \(\mathcal{R}(\Gamma)\) by \(\Big(\sum_{g \in \Gamma}a_gg\Big)^{*}= \Big(\sum_{g \in \Gamma}\overline{a_g}g\Big).\) Let us denote by \(S_n\) the symmetric group on \(n\) letters. A sofic approximation of \(\Gamma\) is given by a sequence of natural numbers \((d_i)\) and a sequence of maps \(\sigma_i~:~\Gamma \longrightarrow S_{d_i}\), \(i \in \mathbb{N}\), such that for all \(g,h \in \Gamma\), \[ \lim_{k \longrightarrow +\infty}\frac{1}{d_k}\big|\big\{i \in [1,n]:\sigma_k(g)\sigma_k(h)i=\sigma_k(gh)i\big\}\big|=1, \] and for every \(g,h \in \Gamma\), \(g \neq h\), \[ \lim_{k \longrightarrow +\infty}\frac{1}{d_k}\big|\big\{i \in [1,n]:\sigma_k(g)i \neq \sigma_k(h)i\big\}\big|=1. \] A group is \textit{sofic} if it admits a sofic approximation. We extend the sofic approximation to the ring \(\mathcal{R}(\Gamma)\) as follows: Given a sofic approximation \((d_i,\sigma_i)_{i \in \mathbb{N}}\) and \(f=\sum_{g \in \Gamma}f_g g \in \mathcal{R}(\Gamma)\). For each \(i \in \mathbb{N}\), we define \(\sigma_i(f) \in \mathcal{M}_{d_i}(\mathcal{R})\) by \(\sigma_i(f)=\sum_{g \in \Gamma}f_g \sigma_i(g)\). We denote by \(\lambda\) and \(r\) the left, respectively right, regular representation of \(\Gamma\) in \(\ell^2 (\Gamma)\), i.e., for all \(g, h \in \Gamma\), \(\lambda(g)\delta_h=\delta_{gh}\), \(r(g)\delta_h=\delta_{hg^{-1}}\), where \(\delta_x\) denotes the indicator function of \(\{x\}\). Notice that \((\delta_g)_{g \in \Gamma}\) is the natural orthonormal basis of \(\ell^2 (\Gamma)\). The maps \(\lambda,r\) can be extended in the usual way to maps from \({\mathbb{C}}(\Gamma)\) to \(\mathcal{B}(\ell^2(\Gamma))\), and maps from \(\mathcal{M}_{m,n}(\mathbb{C}(\Gamma))\) to \(B(\ell^2(\Gamma)^{\oplus n},\ell^2(\Gamma)^{\oplus m})\). We further have that every finitely-presented \(\mathbb{Z}(\Gamma)\)-module is of the form \(\mathbb{Z}(\Gamma)^{\oplus n}/r(f)\mathbb{Z}(\Gamma)^{\oplus n}\), \(f \in \mathcal{M}_{m,n}(\mathbb{Z}(\Gamma))\). We denote by \(X_f\) the Pontryagin dual of \(\mathbb{Z}(\Gamma)^{\oplus n}/r(f)\mathbb{Z}(\Gamma)^{\oplus n}\). We let \(L(\Gamma)\) be the closure of \(\lambda(\mathbb{C}(\Gamma))\) in the weak operator topology, this is called the group von Neumann algebra of \(\Gamma\). If \(A\) is a \(\mathbb{Z}(\Gamma)\)-module, then the algebraic action of \(\Gamma\) is the action of \(\Gamma\) on \(\widehat{A}\), the Pontryagin dual of \(A\) given by \[ (g\chi)(a)=\chi(g^{-1}a), \;g \in \Gamma, \;\chi \in \widehat{A}, \;a \in A. \] The topological entropy of the action of a sofic group \(\Gamma\) on \(X_f\) will be denoted by \(h_{\Sigma}(X_f,\Gamma)\), where \(\Sigma\) is the sofic approximation. \(X_f\) is equipped with the Haar measure probability measure \(m_{X_f}\) and its metric entropy is denoted by \(h_{\Sigma, m_{X_f}}(X_f,\Gamma)\). Now, let \(f \in \mathcal{M}_{n}(L(\Gamma))\). The Fuglede-Kadison determinant of \(f\) is given by \[ {\det}_{L(\Gamma)}(f)=\exp\Big(\int \log(t) d\mu_{|f|}(t)\Big), \] where \(|f|=(f* f)^{\frac12},\) and \(\mu_{|f|}\) is the spectral measure of \(|f|\). In this setting, the author proves the following theorem. Theorem. Let \(\Gamma\) be a countable discrete sofic group with sofic approximation \(\Sigma\). Let \(f \in \mathcal{M}_{m,n} (\mathbb{Z}(\Gamma))\). Then, the topological entropy of the action of \(\Gamma\) on \(X_f\) is finite if and only if \(\lambda(f)\) is injective as an operator on \(\ell^2(\Gamma)^{\oplus n}\). Furthermore, if \(m=n\) and \(\lambda(f)\) is injective, then \(h_{\Sigma}(X_f,\Gamma)= h_{\Sigma,m_{X_f}}(X_f,\Gamma) =\log({\det}_{L(\Gamma)}^{+}(f)).\) If \(m \neq n\) and \(\lambda(f)\) is injective, then \(h_{\Sigma,m_{X_f}}(X_f,\Gamma) \leq h_{\Sigma}(X_f,\;\Gamma) \leq \log({\det}_{L(\Gamma)}^{+}(f))\).