The linear preservers of real diagonalizable matrices (Q551279)

Let a non-singular matrix \(P\in GL_n(K)\), a linear form \(\lambda: M_n(K)\to K\) and a scalar \(\mu\in K\) be given. The maps \(\varphi_{\lambda,P,\mu} :M\to \lambda(M) I_n+\mu PMP^{-1}\) applying possibly together with the transposition transformation are endomorphisms of the vector space \(M_n(K)\) which preserve diagonalizability of matrices. These maps are automorphisms iff \(\mu\neq 0\) and \(\lambda (I_n)\neq -\mu\). It is known that over an algebraically closed field of characteristic zero these maps are the only automorphisms of the vector space \(M_n(K)\) which preserve diagonalizability. The main result of the paper under review is that for \(n\geq 2\) and if the field \(K\) is such that any symmetric matrix in \(M_n(K)\) is diagonalizable, then any vector space automorphism which stabilizes the set of diagonal matrices is some operator \(\varphi_{\lambda,P,\mu}\) possibly together with the transposition map. In order to obtain this result the authors investigate the structure of vector subspaces of matrices formed by the diagonalizable matrices. In particular, it is shown that if there exists a subspace of the maximal dimension \({n+1 \choose 2}\) formed only by diagonalizable matrices then {\parindent6.5mm \begin{itemize}\item[(a)] the ground field \(K\) is formally real and Pythagorean; \item[(b)] every maximal (with respect to the dimension) diagonalizable subspace \(V\) of \(M_n(K)\) is conjugate to the space of symmetric matrices \(S_n(K)\), i.e., there exists an invertible matrix \(T\) such that \(V=TS_n(K)T^{-1}\); \item[(c)] every symmetric matrix in \(M_n(K)\) is diagonalizable. \end{itemize}}