Decomposable matrix polynomials (Q1260957)

Let \(L(z)=zI-A\), where \(A\) is an \(n\times n\) complex matrix. It is well known that \(\mathbb{C}^ n\) is the sum of subspaces of the form \(X(z_ 0)=\{x\in\mathbb{C}^ n| L(z)^{-1}x\) has singularity only at \(z_ 0\}\), where \(z_ 0\in\sigma(A)\). If this property holds for a monic matrix polynomial \(L(z)\) (which now can have degree \(l\geq 1)\), then \(L(z)\) is called decomposable. (A matrix polynomial is called monic if its leading coefficient is \(I.)\) Characterizations of decomposable matrix polynomials are given: one for the general case, another for the case when the given matrix polynomial is represented as a product of linear factors. The first characterization is given in terms that directly generalize the corresponding well-known result for the linear \((l=1)\) case. The second characterization reads as follows: A matrix polynomial \(L(z)=(zI-A_ 1)\cdots(zI-A_ l)\) is decomposable if and only if the \(n\times n\) matrices \(A_ 1,\ldots,A_ l\) have the same set of eigenvalues, and for every \(z_ 0\in\sigma(A_ 1)=\cdots=\sigma(A_ l)\) the spectral projections \(P(A_ k,z_ 0)\) \((k=1,\ldots,l)\) are identical. An example is given showing that not every decomposable matrix polynomial of degree \(l>1\) can be factored into a product of \(l\) linear factors.