The linear preservers of non-singularity in a large space of matrices (Q417489)

It is a classical result due to \textit{J. Dieudonné} [Arch. Math., Oberwolfach 1, 282--287 (1949; Zbl 0032.10601)] that linear bijections on \(M_n({\mathbb K})\), the algebra of \(n\)-by-\(n\) matrices with coefficients from a field \({\mathbb K}\), which leave invariant the general linear group are a change of basis possibly composed by transposition, i.e., \[ X\mapsto PXQ\quad \text{ or }\quad X\mapsto PX^tQ\eqno(1) \] for some invertible matrices \(P\) and \(Q\). The paper under review is an important extension of the above result. The main theorem proves that if \(\mathbb K\) is an infinite field and \(V\subseteq M_{n}({\mathbb K})\) is a linear space of co-dimension at most \((n-2)\), then any linear injection \(f:V\to M_n({\mathbb K})\) which maps invertible matrices from \(V\) to invertible matrices from \(M_{n}({\mathbb K})\) takes the standard form (1). It turns out that the bound \((n-2)\) on co-dimension is exact: the author shows that a linear subspace of \(n\)-by-\(n\) matrices vanishing at the off-diagonal entries in the last row admits a linear bijection which maps the set of invertible matrices onto itself (that is, preserves invertibility in both directions) but is not of the standard form. If \(f\) preserves invertibility in both directions so that invertible matrices from \(V\) are mapped onto invertible matrices from \(f(V)\), then similar conclusions hold also for finite fields, with the sole exception of pathological cases in \(M_3({\mathbb Z}_2)\). Actually, up to a change of basis and simultaneous transposition, there exists a single exception: a hyperplane \({\mathcal V}_1\) that contains all \(3\)-by-\(3\) matrices over the field \({\mathbb Z}_2=\{0,1\}\) vanishing at the \((3,3)\) entry. It is shown that \({\mathcal V}_1\) admits a linear bijection \(f:{\mathcal V}_1\to {\mathcal V}_1\) which preserves invertibility in both directions but is not of the form (1). The bound \((n-2)\) on co-dimension is void for \(2\)-by-\(2\) matrices. In this case the author shows that both two main results hold for hyperplanes in \(2\)-by-\(2\) matrices regardless of the underlying field. It is also shown that both main results hold for the hyperplane of trace-zero matrices, regardless of their size or of the underlying field. Interestingly, when a field is infinite, the proof of the main theorem uses a similar reduction trick which \textit{P. Botta} [Linear Multilinear Algebra 20, 197--201 (1987; Zbl 0619.15002)] used for algebraically closed fields to show that if a linear injection preserves invertibility it does so in both directions. In fact, by fixing a basis of \(V\), the author observes that if \(|{\mathbb K}|=\infty\), then the restriction \(\det|_V\) of a determinant function to space \(V\) is an irreducible homogeneous polynomial in \(d=(\dim V)\) variables, and any polynomial function \(p:V\to{\mathbb K}\), which vanishes on singular matrices from \(V\) is divisible by \(\det|_V\). As a corollary, each linear subspace in \(M_n({\mathbb K})\) of co-dimension at most \((n-2)\) is spanned by its invertible elements and also by its singular elements (this fact was observed by the author already in his previous research).