Unions of rank/trace complete preorders (Q2418999)

Let \(f: D\to M_n(\mathbb{C})\) be an analytic matrix-valued function defined on a Cauchy domain \(D\subseteq\mathbb{C}\). If \(f(\lambda)\) is invertible when \(\lambda\) is restricted to \(\partial D\), a positively oriented boundary of \(D\), then the matrices \[ \frac{1}{2\pi i}\int_{\partial D}f'(\lambda)f(\lambda)^{-1}\,d\lambda\quad \text{and}\quad\frac{1}{2\pi i}\int_{\partial D}f(\lambda)^{-1}f'(\lambda)\,d\lambda \] are called the left and right logarithmic residues of \(f\) and it is known [see \textit{B. Harm}, \textit{E. Torsten}, and \textit{S. Bernd}, Linear Algebra Appl. 498, 262--316 (2016; Zbl 1334.15039)] that both of them are sums of (in general different) idempotents. However, when \(f\) maps to a subalgebra \(\mathcal{A}\subseteq M_n(\mathbb{C})\) the idempotents whose sum equals the left/right logarithmic residue may not belong to \(\mathcal{A}\) and it may happen that within the subalgebra \(\mathcal{A}\) the logarithmic residues are no longer the sum of idempotents. The paper under review, a continuation of the above mentioned paper, seeks for sufficient conditions that left/right logarithmic residues in a subalgebra are always the sums of idempotents. The authors concentrate on unital zero-pattern subalgebras \(\mathcal{A}\subseteq M_n(\mathbb{C})\), that is, subalgebras which consist of all matrices whose entries at specific positions vanish identically. Those positions can be seen as the non-edges of a unique digraph \(\Gamma(\mathcal{A})\) on \(n\) nodes. The fact that \(\mathcal{A}\) is a unital subalgebra rather than merely a vector subspace is equivalent to the fact that a digraph \(\Gamma(\mathcal{A})\) is a preorder relation on \(n\) nodes. In such algebras \(\mathcal{A}\), a necessary condition was found in [the authors, Linear Algebra Appl. 498, 262--316 (2016; Zbl 1334.15039)]. It says that if an element \(A\in\mathcal{A}\) is a sum of idempotents within \(\mathcal{A}\), then \[ \operatorname{rank}\Psi^{(S)}(A)\le \operatorname{Trace}\Psi^{(S)}(A)\in \{0,1,2,\dots\} \] holds for every multiplicative compression \(\Psi^{(S)}:\mathcal{A}\to\Psi^{(S)}(\mathcal{A})\) to the span of collections \(S\) of standard basis vectors in \(\mathbb{C}^n\). The authors call the algebra \(\mathcal{A}\) to be \textit{rank/trace complete} when this condition is also sufficient. The main result then states the following. Assume that finitely many preorders \(\Gamma_1,\dots,\Gamma_k\) on nodes \(\Omega_1,\dots,\Omega_k\), respectively, are such that either every node from \(\Omega_i\cap \Omega_j\) is maximal with respect to both preorders \(\Gamma_i\) and \(\Gamma_j\) or every such node is minimal with respect to \(\Gamma_i\) and \(\Gamma_j\) (\(i\neq j\)). Then the zero-pattern algebra \(\mathcal{A}(\Gamma_1\cup\dots\cup\Gamma_k)\) corresponding to a union of digraphs \(\Gamma_1\cup\dots\cup\Gamma_k\) is rank/trace complete if and only if the zero-pattern algebras \(\mathcal{A}(\Gamma_1),\dots,\mathcal{A}(\Gamma_k)\) are all rank/trace complete. Applications to several concrete preorders are also given.