Stability of multivariate polynomials. IV: Conic sets (Q1398037): Difference between revisions

This is the fourth part of a series of interesting papers [for Parts I--III see ibid. 10, 7-20 (1999; Zbl 0918.93034); ibid. 10, 21-32 (1999; Zbl 0951.93061); ibid. 11, 213-231 (2000; Zbl 1030.93039)] on multivariate polynomials, algebraic objects arising in systems theory, control theory and signal processing. The authors investigate to what extent the stability theory developed in the mid 1990s for univariate polynomials with conic uncertainty can be extended to multivariate polynomials. As pointed out by the authors, the stability theory of conic sets of polynomials is more general than that of compact sets (such as polytopes). Various notions of stability are recalled in section 3.1, following the notation and the concepts introduced in the first part of the series, published in the same journal four years ago. A multivariate polynomial is called stable if, in addition to having all its roots lying in the closed left half-plane, its coefficients satisfy additional algebraic conditions, summarized in definition 4. This class of stable polynomials is the largest class of polynomials preserving the stability property under small coefficient perturbations. Another, different notion of stability is semi-stability, when the roots of a multivariate polynomial belong to the open left half-plane. The concept of value set, instrumental to the reduction of robust stability analysis of polynomial families to the study of simple geometric objects in the complex plane, is also extended to multivariate polynomials. For a conic polynomial \(p+K\) which is the sum of a (nominal) polynomial \(p\) with positive coefficients with a polynomial \(K\) with non-negative coefficients within a convex cone, it is shown that a necessary condition of stability of \(p+K\) is that \(p\) is stable and \(K\) is semi-stable. This motivates the study in section 3.2 of semi-stable cones of polynomials. The main result of this section is the multivariate counterpart of the edge theorem, establishing the equivalence between checking stability of a polyhedral cone of polynomials and checking stability of its edges only. Segments of polynomials are studied in section 3.3, and extensions to multivariate polynomials of the concept of convex direction and Rantzer's phase growth condition are also proposed. Finally, in section 3.4, an extension of the celebrated Kharitonov's theorem on interval polynomials (proposed by the first author in 1978) is described in theorem 36 for multivariate polynomials, establishing the equivalence between checking stability of an interval polynomial (a family of conic polynomials where all the coefficients belong to given, independent intervals) and checking stability of a finite (yet large) number of polynomials.