Power series with positive coefficients arising from the characteristic polynomials of positive matrices (Q2634889)

Editorial remark: For this item inadvertently two reviews were received: \textbf{1st review by Alexander Kovačec (Coimbra):} The main result of this very interesting paper is: Theorem 5. Let \(f(t)=\prod_{i=1}^n (1-\lambda_i t)\in\mathbb{C}[t].\) Assume the list \(\sigma=(\lambda_1,\dots,\lambda_n)\) is such that \(s_k(\sigma)=\sum_{i=1}^n \lambda_i^k\) is real and positive for all \(k=1,2,\dots\) and \(\lambda_1=|\lambda_1|> |\lambda_i|\), for \(i=2,\dots,n.\) Then for all sufficiently large integers \(N,\) the power series \(1- f(t)^{1/N}\) has positive coefficients. The authors derive this result as an offspin of research on the nonnegative inverse eigenvalue problem (NIEP) which asks for a characterization of the possible spectra of entrywise nonnegative matrices. The influence is mutual. As a corollary one has for example that if \(A\) is a positive \(n\times n\) matrix, then for all large enough integers \(N,\) \(1- (\det (I-tA))^{1/N}\) is a power series with only positive coefficients. Laffey had found in [\textit{T. J. Laffey}, Linear Algebra Appl. 436, No. 6, 1701--1709 (2012; Zbl 1241.15007)] a constructive proof of \textit{M. Boyle} and \textit{D. Handelman}'s characterization [Ann. Math. (2) 133, No. 2, 249--316 (1991; Zbl 0735.15005)] of the nonzero part of spectra of nonnegative matrices. He used for this matrices \(X_n\) which in the diagonal starting at adress \((i,1)\) are filled with \(x_i,\) \(i=1,\dots,n\); whose first superdiagonal has the entries \(1,2,\dots,n-1\); and all whose other entries are 0. (So the \(X_n\) are lower Hessenberg almost Töplitz). He showed: a list \(\sigma\) of nonzero complexes that satisfies (essentially) the conditions announced above, will be equal to the nonzero spectrum of some matrix \(X_{n+N}.\) Using this and the following Theorem 4, the paper gets above Theorem 5. Theorem 4. The polynomial \(f_n(t)=\det(I-tX_n)\) gives rise to a monomially positive power series \(1-f_n(t)^{1/n}\): that is the coefficients of all \(t^j\) are polynomials in \(\mathbb{R}_{\geq 0}[x_1,\dots,x_n].\) The largest part of the paper (Sections 5, 6) is occupied with proving Theorem 4. To this aim recurrence relations for \(F_n(t)=\det(tI -X_n)\) and hence \(f_n\) are established; note \(f_n(t)=t^n F_n(1/t)\). It is then shown that the sequence \(F_n\) is essentially Appell and this fact then implies that the \(n\)-th standard vector \(e_n\) is a trace vector (see [\textit{R.J. Pereira}, Trace vectors in matrix analysis. Toronto: University of Toronto (PhD Thesis) (2003)]) for \(X_n;\) in particular \(e_n^T X_n^k e_n =\frac{1}{n} \operatorname{trace} X_n^k,\) \(k\in \mathbb{Z}_{\geq 0}\). From this one finds that, defining \(t_k(n)=\frac{1}{n}\text{trace} X_n^k\), one has \(t_k(n)-t_k(n-1)\in \mathbb{R}_{\geq 0}[x_1,\dots,x_n]\) which is one of the ingredients for showing monomial positivity of various other power series. These together yield an expression for \(1-f_n(t)^{1/n}\) and thus Theorem 4. In Section 7 it is shown that the coefficients of \(f_n(t)^{1/n}=1-\sum_j \gamma_j(n) t^j\) can be expressed as \(\gamma_j(n)=(-1)^{j-1} \det T_m(n)/j!,\) where the matrix \(T_m(n),\) incredibly, arises from \(X_n\) by replacing \(x_i\) by \(t_i(n).\) No proof free of matrix theory is known for Theorem 5. For interesting observations of the implications of condition \(s_k(\sigma) \geq 0\) on other relevant properties of \(\sigma\) for NIEP and a discussion nearer to complex analysis, see \textit{F. Holland}'s paper [Math. Proc. R. Ir. Acad. 113A, No. 2, 81--96 (2013; Zbl 1297.15011)]. \textbf{2nd review by Frank Uhlig (Auburn):} This paper studies entrywise positive matrices \(A\in\mathbb R^{n,n}\) and power series expansions of \(k(t) = 1-f (t)^{1/N}\) and their coefficients for the characteristic polynomial \(f (t)\) of \(A\). The main, new and surprising result here is that there always exists an \(N > 0\) so that \(k(t)\) has positive coefficients in its power series expansion around \(0\). The paper designs a new family of multivariable functions with positive coefficients from the determinants of Kac-Toeplitz type Heisenberg matrices \(X_n\) that are built on the entries of \(A\). The method of proof is entirely matrix theoretical and touches on the nonnegative inverse eigenvalue problem using the same \(X_n\) that was originally introduced by \textit{M. Boyle} and \textit{D. Handelman} [Ann. Math. (2) 133, No. 2, 249--316 (1991; Zbl 0735.15005)].