Stability and equivariant maps (Q1121957)

Let G be a reductive algebraic group acting (linearizably) on projective varieties X and Y, and let \(\pi:\quad Y\to X\) be a G-equivariant morphism. The author defines a suitable linearization of the G-action on Y and then compares stability in X and Y. His most general result is that \(q\in Y\) is unstable (resp. properly stable) if \(\pi\) (q) is unstable (resp. properly stable). The most important result of the paper is a relative Hilbert-Mumford theorem: it states that stability of \(q\in Y\) and \(\pi\) (q)\(\in X\) can be tested simultaneously by the weights of 1- parameter subgroups of G. If \(\pi\) : \(Y\to X\) is a blowing up the author completely describes the unstable and properly stable loci in Y in terms of \(\pi\) (outside the exceptional divisor in general, and under certain smoothness hypothesis even everywhere). The last theorems of the paper deal with the ``strictly semi-stable'' locus \(X^{sss}\) in X, i.e. the semi-stable but not properly stable points: it is shown that \(X^{sss}\) is singular or empty if the semi-stable locus \(X^{ss}\) is smooth and that a resolution of singularities of \(X^{sss}\) yields a ``stable resolution'' of X, i.e. a G-variety \(\tilde X\) where each point is either unstable or properly stable.