Towards non-reductive geometric invariant theory (Q930777)

Classical geometric invariant theory is an effective tool in the study of actions by a reductive algebraic group. For example, if the reductive group \(G\) over a characteristic zero field \(k\) acts on an affine variety \(X\) then the ring \(k[ X] ^{G}\) of invariants is finitely generated; furthermore there is a good categorical quotient \(\pi:X\to \)Spec\(( k[ X] ^{G}) \) together with a distinguished open subset \(X^{s}\) of stable orbits whose image under \(\pi\) is a geometric quotient. More generally, if \(X\) is a variety which is equipped with a linearization of the group action one can still construct a categorical quotient which gives rise to a geometric quotient of \(X^{s}.\) The work under review is an attempt to create a geometric invariant theory in the case where \(G\) is not reductive, and to do this the authors focus on the unipotent component of the group. Let \(X\) be a projective variety endowed with an ample line bundle \(L\), and let \(H\) be a connected unipotent group which acts linearly on \(L\), and let \(G\) be a reductive group containing \(H\) as a closed subgroup. Then a nested series of stable sets in the set of naively semistable points \(X^{nss}\) is described as are their quotients. Some of the stable sets turn out to be equal, and if the reductive envelope of \(G\times_{H}X\) is normal then more of the subsets become equalities. One of the difficulties in considering non-reductive groups is that the invariant ring \(k[ X] ^{H}\) may not be finitely generated. Here a necessary and sufficient condition is given for when \(k[ X] ^{H}\) is in fact finitely generated. Finite generation depends on the ability for certain \(G\)-invariant sections of a positive tensor power of a line bundle to vanish on the codimension one components in the boundary of \(G\times_{H}X\) in \(\overline{G\times_{H}X}.\)