General theory of regular matrix polynomials and band Toeplitz operators (Q1119714)

Consider the \(m\times m\) matrix polynomial \(L(\lambda)=\lambda^{\ell}A_{\ell}+\lambda^{\ell -1}A_{\ell - 1}+...+\lambda A_ 1+A_ 0\) which is regular, i.e., \(\det L(\lambda)\neq 0\) for all but a finite number of points (called the zeros of \(L(\lambda)\)) in \({\mathbb{C}}\). For the monic case (that is, when \(A_{\ell}=I\), the identity matrix) the theory is well-developed [see \textit{I. Gohberg}, \textit{P. Lancaster}, \textit{L. Rodman}: Matrix Polynomials (1982; Zbl 0482.15001)] Ch. I]; the essential tool is the description of \(L(\lambda)\) via standard pairs (Q,T) and standard triplets (Q,T,R), where, among other representations, one has \(L(\lambda)^{-1}=Q(\lambda I-T)^{-1}R\); it is not difficult to obtain a standard triplet for the product \(L_ 1L_ 2\) of monic polynomials from the standard triplets of the factors, and this leads to an extensive theory on the (monic) divisors of monic polynomials. In the above mentioned monograph the theory of non-monic regular polynomials was also presented, following the outline provided by \textit{N. Cohen} [Integral Equations Oper. Theory 6, 161-183 (1983; Zbl 0521.15015)], but this theory was considered somewhat unsatisfactory: Division, e.g., is described in terms of the spectral data of the polynomial and its divisor, and hence a formula for the quotient is absent. In the present paper the approach allows for an ample amount of analogy to the monic case: Given a regular \(m\times m\) matrix polynomial \(L(\lambda)=\lambda^{\ell}A_{\ell}+\lambda^{\ell -1}A_{\ell - 1}+...+\lambda A_ 1+A_ 0,\) degree \(L\leq \ell\), one describes L(\(\lambda)\) by means of right-\(\ell\)-pencil pair (C,\(\lambda\) G-A) for L(\(\lambda)\), where \(\lambda\) G-A is a regular \(m\ell \times m\ell\) matrix pencil, C is an \(m\times m\ell\) matrix, such that \(C(\lambda G- A)^{-1}x\neq 0\) for some \(\lambda\) if \(x\neq 0\) and \(L(\lambda)C(\lambda G-A)^{-1}=M_ 0+\lambda M_ 1+...\lambda^{\ell -1}M_{\ell -1}\) is a polynomial of degree \(\leq \ell -1\). It turns out that \((C,\lambda G-A)\) is a right-\(\ell\)-pencil pair for \(L(\lambda)\) if and only if it is also a right-\(\ell\)-pencil pair for the associated polynomial \(L^{\#}(\lambda):=\lambda^{\ell}L(\lambda^{-1})\); an important example of such a right-\(\ell\)-pencil pair for L(\(\lambda)\) is provided by \(C=(0...0\) I), \[ (\lambda G-A) = C_{2,L}(\lambda) = \lambda \begin{pmatrix} I \\ & I && 0 \\ && \ddots \\ & 0 && I \\ &&&& a_\ell \end{pmatrix} - \begin{pmatrix} 0 &&&& A_0 \\ I & \ddots &&& \vdots \\ & \ddots & \ddots && \vdots \\ && \ddots & 0 & -A_{\ell-2} \\ &&& I & -A_{\ell-1} \end{pmatrix}, \] the second companion pencil of L(\(\lambda)\). Two right-\(\ell\)-pencil pairs \((C_ i,\lambda G_ i-A_ i)\) for L(\(\lambda)\), \(i=1,2\), are strictly equivalent, i.e., there exist invertible matrices E,F such that \(C_ 1=C_ 2F\), \(\lambda G_ 1-A_ 1=E(\lambda G_ 2-A_ 2)F.\) There are representations for L(\(\lambda)\) connected with (C,\(\lambda\) G- A): the matrix \(E_{\theta}=col(M_ i)_{i=0}^{\ell -1}\), formed with the coefficients of the polynomial \(L(\lambda)C(\lambda G-A)^{- 1}\), is invertible; writing \((B_ 1...B_{\ell})=E_{\theta}^{-1}\) one has \(\lambda^{i-1}L(\lambda)^{-1}=C(\lambda G-A)^{-1}B_ i,\) \(i=1,2,...,\ell\), whereas a representation for L(\(\lambda)\) is obtained by choosing \(\alpha\) such that (\(\alpha\) G-A) is invertible. Setting \((V_ 1...V_{\ell})=[col(C[(\alpha G-A)^{-1}G]^ i)_{i=1}^{\ell -1}]^{\quad -1}\) one has \[ L(\lambda)=L(\alpha)+(\lambda - \alpha)L(\alpha)C[(\alpha G-A)^{-1}G]^{\ell}\sum^{\ell - 1}_{j=0}(\alpha -\lambda)^ jV_{\ell -j}. \] Both these representations are in close analogy to the monic case. An import tool for the description of a right pencil pair for a product if regular matrix polynomials in terms of a right pencil pair of the factors is the \(\Gamma\)-complement. If \(L(\lambda)\) is as before, and \(\Gamma\) is a Cauchy contour with inner domain \(\Delta_+\) containing 0 and outer domain \(\Delta_-\) containing \(\infty\) such that \(L(\infty)\) (and hence \(\lambda G-A\)) is invertible for all \(\lambda\in \Gamma\), then the \(m\ell \times m\) matrix R is called the \(\Gamma\)-complement of \((C,\lambda G-A)\) if \(L(\lambda)^{-1}-C(\lambda G-A)^{-1}R\) is analytic on \(\Delta_+\cup \Gamma\) and \(\lambda^{\ell -1}L(\lambda)^{-1}- C(\lambda G-A)^{-1}R\) is analytic on \(\Delta_-\cup \Gamma\) with value 0 at \(\infty\). It turns out that R is given by \(R=QB_ 1+(I- Q)B_{\ell}\), with \(B_ 1,B_{\ell}\) as above and the projection Q given by \(Q=\frac{1}{2\pi i}\int_{\Gamma}G(\lambda G-A)^{- 1}d\lambda\). Now, given two regular matrix polynomials \(L_ i(\lambda)\) with \(right\)- \(\ell_ i\)-pencil pairs \((C_ i,\lambda G_ i-A_ i)\) and \(\Gamma\)- complements \(R_ i\), \(i=1,2\), then one obtains a \(right\)-\(\ell_ 1+\ell_ 2\)-pencil pair \((C,\lambda G-A)\) with \(\Gamma\)-complement R via \[ G=\begin{pmatrix} G_ 1 & (I-Q_ 1)R_ 1C_ 2(I-P_ 2)\\ 0 & G_ 2 \end{pmatrix},\quad A=\begin{pmatrix} A_ 1 & Q_ 1R_ 1C_ 2P_ 2\\ 0 & A_ 2 \end{pmatrix} \] \[ C=(C_ 1\quad W+X),\quad R=\begin{pmatrix} Y+Z\\ R_ 2 \end{pmatrix}, \] where \(Q_ 1=\frac{1}{2\pi i}\int_{\Gamma}G_ 1(\lambda G_ 1-A_ 1)^{-1}d\lambda\), \(P_ 2=\frac{1}{2\pi i}\int_{\Gamma}(\lambda G_ 2-A_ 2)^{-1}G_ 2d\lambda\), and W,X,Y,Z are defined by somewhat messy formulae involving \(L_ 1,L_ 2\), both pencil pairs, complements and the projections \(Q_ 1,P_ 2.\) Once the problem of multiplication is mastered it is possible to obtain a powerful division theorem. To this end one considers the notion of a restriction \((C',\lambda G'-A')\) of the right-\(\ell\)-pencil pair \((C,\lambda G-A)\) for \(L(\lambda)\): such a restriction is defined by \(C'=CE\), \(F(\lambda G'-A')=(\lambda G-A)E\) where F,E are left invertible matrices. The division theorem then reads as follows: Given regular matrix polynomials \(L(\lambda,L_ 1(\lambda)\) with right-\(\ell\)-(resp. \(\ell\)-) pencil pairs \((C,\lambda G-A)\), resp. \((C_ 1,\lambda G_ 1-A_ 1)\) then \(L_ 1(\lambda)\) is a right divisor of \(L(\lambda)\) with degree \(L(\lambda)L_ 1(\lambda)^{-1}\leq \ell -\ell_ 1\) if and only if \((C_ 1,\lambda G_ 1-A_ 1)\) is a restriction of \((C,\lambda G-A)\). In order to describe the quotient \(L(\lambda)L_ 1(\lambda)^{-1}\) one chosen any \(\Gamma\)-complement R; then \((\lambda G-A,R)\) is a left- \(\ell\)-pencil pair for \(L(\lambda)\). Let \(C_ 1=CE_ 1\), \(F_ 1(\lambda G_ 1-A_ 1)=(\lambda G-A)E_ 1\) with \(F_ 1,E_ 1\) left invertible. Choosing \(F_ 2,E_ 2\) right invertible with \(KerE_ 2=ImE_ 1\), \(KerF_ 2=ImF_ 1\) one forms the compression \((\lambda G_ 2-A_ 2,R_ 2)\) by \(R_ 2=F_ 2R\), \((\lambda G_ 2-A_ 2)E_ 2=F_ 2(\lambda G-A)\). Then \((\lambda G_ 2-A_ 2,R_ 2)\) is a left- \((\ell-\ell_ 1)\)-pencil pair for \(L(\lambda)L_ 1(\lambda)^{-1}.\) On the strength of this theorem several explicit formulae for both the divisor \(L_ 1(\lambda)\) and the quotient \(L(\lambda)L_ 1(\lambda)^{-1}\) are derived. Special attention is paid to spectral divisors. As an example of possible applications the inversion problem for band- block-Toeplitz operators is considered, and formulae for the inverse are presented. The (lengthy) paper is well-written, and it supersedes Ch. 7. in the above mentioned book of Gohberg et al.