Nonconcavity of the spectral radius in Levinger's theorem (Q2199353)

Let \(\mathbb{R}^{n\times n}\) be the set of all \(n\times n\)-matrices whose entries are real numbers. Let \(\textbf{A}\in\mathbb{R}^{n\times n}\) be a nonnegative irreducible square matrix and let \[r(t):=r((1-t)\textbf{A}+ t\textbf{A}^T)\] be the spectral radius over the convex combinations of the nonnegative matrix \(\textbf{A}\) and its transpose \(\textbf{A}^T\). In [An inequality for nonnegative matrices, Notices Am. Math. Soc. 17, 260 (1970)], \textit{B. W. Levinger} asserted that \(r(t):=r((1-t)\textbf{A}+ t\textbf{A}^T)\) increases over \(\left[0,\frac{1}{2}\right]\) and decreases over \(\left[\frac{1}{2},1\right]\). This result has been proved by several authors, together with the fact that \(r(t)\) is concave over \((0,1)\). The authors of this paper provide a number of counterexamples to show that the last claim about concavity is false in general. They prove that such a concavity claim is true for \(2\times2\) and other special families of matrices.