Perturbed marked reduced forms of invariant subspaces (Q1794328)

A nilpotent linear transformation \(f\) of an \(N\)-dimensional vector space \(E\) into itself is determined up to equivalence by its Segre characteristic, namely, the partition \(p\) of \(N\) whose parts are the sizes of the Jordan blocks of \(f\) (the nilpotence index \(n\) of \(f\) is the size of the largest part of the partition \(p\)). Let \(V\) be a \(d\)-dimensional subspace of \(E\). It is a classical problem to classify the nilpotent \(f\) on \(E\) for which \(V\) is an \(f\)-invariant subspace and the Segre characteristics of \(f\) and of the restriction \(f_{|V}\) are prescribed. \textit{C. M. Ringel} and \textit{M. Schmidmeier} [J. Reine Angew. Math. 614, 1--52 (2008; Zbl 1145.16005)] have shown that for \(n\leq5\) this problem is of finite representation type, but that for \(n\geq7\) the representation problem is ``wild'' and there is no prospect of a general classification (for \(n=6\) the problem is tame but the classification is of infinite representation type). Let \(f\in {\mathrm{End}}(E)\) be nilpotent of Segre characteristic \(p=(p_{1},\dots,p_{n})\). Then \(E\) has a Jordan basis for \(f\) of the form \(e_{1},f(e_{1}),\dots,f^{p_{1} -1}(e_{1}),\dots,e_{n},f(e_{n}),\dots,f^{p_{n}-1}(e_{n})\) (this basis is by no means unique). An \(f\)-invariant subspace \(V\)\ is called a marked subspace if it has a Jordan basis which can be extended to a Jordan basis for \(E\). Let \({\mathrm{Inv}}(p,q)\) be the set of all \(f\)-invariant subspaces \(V\) for which \(f_{|V}\) has Segre characteristic \(q=(q_{1},\dots,q_{m})\). There are finitely many equivalence classes of marked subspaces in \({\mathrm{Inv}}(p,q)\) and these classes can be uniquely labelled by the permutations \(p^{\prime}=(p_{j_{1}},\dots,p_{j_{n}})\) of \(p\) for which \(q_{i}\leq p_{j_{i}}\) for each \(i\leq m\) and \(p_{j_{m}+1}\geq p_{j_{m+2}}\geq\dots\geq p_{j_{n}}\) (see [\textit{J. Ferrer} et al., Linear Algebra Appl. 235, 15--34 (1996; Zbl 0852.15007)]). The object of the paper is to partition the equivalence classes of \({\mathrm{Inv}}(p,q)\) into a finite number of families where each family contains exactly one equivalence class of marked subspaces; this subspace is called the marked type of the family. In order to do this the authors first represent each equivalence class of \({\mathrm{Inv}}(p,q)\) by an \(N\times d\) matrix (called the vectorial representation) whose rows and columns are partitioned into blocks according to the partitions \(p\) and \(q\), respectively. For each family of a specified marked type, there is a matrix of this form with a certain number of parameters such that when the parameters are set to \(0\) the matrix represents the marked subspace, and for the other specializations gives a matrix representing one of the other equivalence classes in the family (some equivalence classes may be obtained more than once). An algorithm is provided for computing with these block matrices. The paper concludes with a number of examples and applications including a new proof of the classification of \({\mathrm{Inv}}(p,q)\) where \(m=1\) (the monogenic case) (see [\textit{A. Compta} and \textit{J. Ferrer}, Linear Multilinear Algebra 63, No. 9, 1768--1785 (2015; Zbl 1322.15006)]) and analysis of the non-monogenic cases where \(n\leq5\) and \(n=6\).