An analog of the Poincaré separation theorem for normal matrices and the Gauss-Lucas theorem (Q1434040)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 2077836
Language Label Description Also known as
default for all languages
No label defined
    English
    An analog of the Poincaré separation theorem for normal matrices and the Gauss-Lucas theorem
    scientific article; zbMATH DE number 2077836

      Statements

      An analog of the Poincaré separation theorem for normal matrices and the Gauss-Lucas theorem (English)
      0 references
      0 references
      1 July 2004
      0 references
      The author states a number of theorems on inequalities of which the following are typical (proofs appear elsewhere). Let \(q(\lambda)\) be a monic polynomial of degree \(m\) over \(\mathbb{C}\), and denote the roots of \(q(\lambda)\) by \(\lambda_{1},\dots,\lambda_{m}\) and the roots of its derived polynomial \(q^{\prime}(\lambda)\) by \(\mu_{1},\dots,\mu_{m-1}\). (1) If \(q_{1}(\lambda)\) is another monic polynomial of degree \(m-1\). Then there exists a normal \(m\times m\) matrix \(A\) such that \(A\) has characteristic polynomial \(q(\lambda)\) and its principal minor \(A_{m-1}\) of size \((m-1)\times(m-1)\) has characteristic polynomial \(q_{1}(\lambda)\) if and only if \(q_{1}(\lambda_{k})\) lies in the convex hull of \(0\) and \(q^{\prime} (\lambda_{k})\) for each \(\lambda_{k}\). This can be recognized as a generalization of the Cauchy-Poincaré interlacing theorem for Hermitian matrices. (2) For all convex functions \(f:\mathbb{C}\to\mathbb{R}\), all \(\alpha\in\mathbb{C}\) and all positive integers \(p\leq m-1\) we have: \[ \binom{m-1}{p}^{-1}\sum f \left(\prod_{s=1}^{p}(\mu_{i_{s}}-\alpha)\right)\leq \binom{m}{p}^{-1}\sum f\left(\prod_{s=1}^{p}(\lambda_{i_{s}}-\alpha)\right) \] where the first sum is over all indices \(1\leq i_{1}<\dots<i_{p}\leq m-1\) and the second is over all indices \(1\leq i_{1}<\dots<i_{p}\leq m\). This is a generalization of an inequality conjectured by \textit{N. G. de Bruijn} and \textit{T. A. Springer} [Proc. Akad. Wet. Amsterdam 50, 458--464 (1947; Zbl 0029.19801)]. (3) If \(\lambda_{1}+\dots+\lambda_{m}=0\), then \(m\sum_{j=1}^{m-1}| \mu _{j}| ^{2}\leq(m-2)\sum_{j=1}^{m}| \lambda_{j}| ^{2}\); and equality holds if and only if all the roots \(\lambda_{k}\) lie on the same line. This verifies a conjecture of [\textit{I. Schoenberg, Am. Math. Mon. 93, 8--13 (1986; Zbl 0627.30001)]. Editorial remark: Other proofs of the aforementioned conjectures have appeared by \textit{R.~Pereira} [J. Math. Anal. Appl. 285, 336--348 (2003; Zbl 1046.47002)].}
      0 references
      0 references
      normal matrix
      0 references
      inverse problem
      0 references
      Gauss-Lucas theorem
      0 references
      majorization
      0 references
      zeros of polynomials
      0 references
      Cauchy-Poincaré separation theorem
      0 references

      Identifiers