A characterization of sets realizable by compensation in the SNIEP (Q6536726)

Given a list \(\Lambda :=\{\lambda _{1},\dots,\lambda _{n}\}\) of \(n\) real numbers, is it possible to determine if \(\Lambda \) is the spectrum of an \(n\times n\) matrix \(A\) with nonnegative entries? There are simple necessary conditions; for example, if \(\Lambda \) is the spectrum of a nonnegative matrix \(A\) then \(\max \Lambda \geq \left\vert \lambda _{i}\right\vert \) for each \(i\) and \(\sum_{i}\lambda _{i}=\mathrm{Tr} A\geq 0\). The problem has been solved for \(n\leq 5\) but the general problem remains open. The case when \(A\) is required to be symmetric is known as the symmetric nonnegative inverse eigenvalue problem (SNIEP) and further progress has also been made in this case for so-called \(C\)-realizable lists. In [Linear Algebra Appl. 615, 42--76 (2021; Zbl 1458.15030)] the authors of the present paper solved the problem for \(C\)-realizable lists when \(\sum_{i}\lambda _{i}=0\).\N\NIn the present paper, the authors extend their results to require only that \(\sum_{i}\lambda _{i}\geq 0\). There is considerable overlap between the two papers and the conclusions are not very different, although the new paper uses an approach from [\textit{R. Ellard} and \textit{H. Šmigoc}, Linear Algebra Appl. 498, 521--552 (2016; Zbl 1334.15028)]. The authors point out that their arguments also show that if \(\Lambda \) is \(C\)-realisable then any list \( \Lambda ^{\prime }\) obtained by increasing the positive entries is also \(C\)-realisable.