Let K be an arbitrary field, and a,b,c,d be elements of K such that the
polynomials t^2-at-b and t^2-ct-d are split in K[t]. Given a square matrix M
with entries in K, we give necessary and sufficient conditions for the
existence of two matrices A and B such that M=A+B, A^2=a A+bI_n and B^2=c
B+dI_n. Prior to this paper, such conditions were known in the case b=d=0, a<>0
and c<>0, and also in the case a=b=c=d=0. Here, we complete the study, which
essentially amounts to determining when a matrix is the sum of an idempotent
and a square-zero matrix.
Given an endomorphism u of a finite-dimensional vector space (over an
arbitrary field), we give necessary and sufficient conditions for the existence
of a regular quadratic form (resp. a symplectic form) for which u is orthogonal
(resp. symplectic). When the field of scalars has characteristic 2, we also
give necessary and sufficient conditions for the existence of a regular
symmetric bilinear form for which u is orthogonal. For the field of real
numbers and for finite fields, we characterize the existence of a regular
quadratic form in a given equivalence class for which u is orthogonal.
Using a recent result of Bogdanov and Guterman on the linear preservers of
pairs of simultaneously diagonalizable matrices, we determine all the
automorphisms of the vector space M_n(R) which stabilize the set of
diagonalizable matrices. To do so, we investigate the structure of linear
subspaces of diagonalizable matrices of M_n(R) with maximal dimension.
Given an arbitrary field K, let V be a linear subspace of M_n(K) consisting
of matrices of rank lesser or equal to some r<n. A theorem of Atkinson and
Lloyd states that, if dim V>nr-r+1 and #K>r, then either all the matrices of V
vanish on some common (n-r)-dimensional subspace of K^n, or it is true of the
matrices of its transpose V^t. Following some arguments of our recent proof of
the Flanders theorem for an arbitrary field, we show that this result holds for
any field.
Let K denote a field. Given an arbitrary linear subspace V of M_n(K) of
codimension lesser than n-1, a classical result states that V generates the
K-algebra M_n(K). Here, we strengthen this in three ways: we show that M_n(K)
is actually generated as a linear space by products of the form AB with A and B
in V; we prove that every matrix in M_n(K) can be decomposed into a product of
elements of V; finally, when V is a linear hyperplane of M_n(K) and n>2, we
show that every matrix in M_n(K) is a product of two elements of V.
When K is an arbitrary field, we study the affine automorphisms of M_n(K)
that stabilize GL_n(K). Using a theorem of Dieudonn\'e on maximal affine
subspaces of singular matrices, this is easily reduced to the known case of
linear preservers when n>2 or #K>2. We include a short new proof of the more
general Flanders' theorem for affine subspaces of M_{p,q}(K) with bounded rank.
We also find that the group of affine transformations of M_2(F_2) that
stabilize GL_2(F_2) does not consist solely of linear maps.
In this paper, we reduce the determination of the singular endomorphisms $f$
of M_n(K) that stabilize GL_n(K) to the classification of n-dimensional
division algebras over K. Our method, which is based upon Dieudonn\'e's theorem
on singular subspaces of M_n(K), also yields a proof for the classical
non-singular case.
Given an integer n greater of equal to 3, we investigate the minimal
dimension for a subalgebra of square matrices of order n with a trivial
centralizer. It is shown that this dimension is 5 when n is even and 4 when it
is odd. In this latter case, all 4-dimensional subalgebras with a trivial
centralizer are explicitely computed.
When $\mathbb{K}$ is a field, and $\mathcal{A}$ and $\mathcal{B}$ denote
commuting subspaces of $\text{M}_n(\K)$ each of which contains a non-scalar
matrix, we prove that $\dim \mathcal{A} +\dim \mathcal{B} \leq (n-1)^2+3$. We
also give a complete description of the cases when equality holds.