Interpolation theory and \(\lambda\)-matrices (Q583332): Difference between revisions

Let \(A(\lambda)\) be a monic matrix polynomial with right solvents \(R_1,\ldots,R_ s\) of order \(m_1,\ldots,m_ s\) (i.e. \((\lambda I-R_ i)^{m_ i}\) is a right divisor of \(A(\lambda)\) and \((\lambda I-R_ i)^{m_ i+1}\) isn't). Assume that (a): degree \(A(\lambda) = m = m_1+ \cdots + m_ s\) and (b): the generalized block Vandermonde matrix \[ V_ R := \text{row} \left\{ \text{row} \left[ \text{col}\left( \binom{k-1}\ell R_ i^{k-\ell -1} \right)^ m_{k=1} \right]_{\ell =0}^{m_ i-1} \right\}^ s_{i=1} \] (where \(\binom pq = 0\), \(q>p\)) is invertible. Define \(m_ i \times m_ i\)-block-Jordan-matrices \(J_ i\) and the \(m \times m\)-block-matrix \(J\) by \[ J_ i = \begin{pmatrix} R_ i & I & & & 0 \\ & R_ i & \ddots \\ && \ddots & \ddots \\ & 0 && R_ i & I \\ &&&& R_ i \end{pmatrix}, \quad J = J_1 \oplus \cdots \oplus J_ s \] and let \(C = \text{row}\{\text{row}[\delta_{0\ell}I]_{\ell =0}^{m_ i-1}\}^ s_{i=1 }\) denote the first block row of \(V_ R\), \(B=\text{col}\{\text{col}[M_1^{(i,\;ell)}]_{\ell =0}^{m_ i-1}\}^ s_{i=1}\) the final block column of \(V_ R^{-1}\). Then \((C,J,B)\) is a standard triple for \(A(\lambda)\), and hence \[ A(\lambda)^{-1} = C(\lambda I-J)^{-1} B = \sum^{s}_{i=1} \sum^{m_ i-1}_{\ell =1} (\lambda I-R_ i)^{-\ell -1} M_ 1^{(i,\ell)} \tag{\(*\)} \] [cf. \textit{I. Gohberg}, \textit{P. Lancaster} and \textit{L. Rodman}, Matrix polynomials (1982; Zbl 0482.15001), Theorem 2.4]. The invertibility of \(V_ R\) is automatically satisfied if (a) holds and (c): \(\sigma (R_ i) \cap \sigma (R_ j) =\emptyset\), \(i\neq j\) is met (op. cit., {\S} 1.8). Observe that (d): \(\sigma (A) = \{\lambda | \det A(\lambda) = 0\}= \bigcup^{s}_{i=1} \sigma (R_ i)\), if (a), (b) hold. In the present paper formula (\(*\)) is obtained via ``interpolating polynomials'' \(M_{i,\ell}(\lambda)\), constructed from the block rows of \(V_ R^{- 1}\). The conditions for the construction are not correctly formulated, as (a) is ignored, and (b), (c), (d) are required where (a), (c) would suffice. Using Theorem 2.4, op. cit., for the standard pair \((C,J)\) one has \[ A(\lambda) = \lambda^ m I - \sum^{s}_{i=1} \sum^{m_ i-1}_{\ell=1} \binom m\ell R_ i^{m-\ell} M_{i,\ell}(\lambda), \] but this obvious representation is not obtained by the authors. Needless to say that the standard reference given here is absent in the list of references in the paper.