The incidence class and the hierarchy of orbits (Q1040220)

The results of this paper are mostly stated in terms of representation theory of complex algebraic groups, but the motivation comes from singularity theory. By a singularity the authors mean an orbit of the contact group \({\mathcal K}(n,p)\) on the space \({\mathcal E}(n,p)\) of holomorphic map germs \(({\mathbb C}^n,0)\to ({\mathbb C}^p,0)\). Both \(\mathcal K\) and \(\mathcal E\) are infinite-dimensional and it makes sense to restrict to finite codimension singularities. These are finitely determined so one may look at the \(k\)-jets of both \(\mathcal K\) and \(\mathcal E\) instead. The aim is to study the inclusion relation among orbit closures. If \(\eta\) and \(\xi\) are singularities and \([\eta]\) denotes the Thom polynomial, i.e., the Poincaré dual of \(\bar\eta\) in equivariant cohomology, then the incidence class is defined (by Rimányi) to be \([\eta]|_\xi\). This, of course, makes sense for any algebraic group representation \(\rho: G \to GL(V)\) of an algebraic group over \(\mathbb C\). Rimányi conjectured (wrongly: see below) that \([\eta]|_\xi\) should be nonzero if and only if \(\xi\subset \bar\eta\): the authors call this the incidence property for \(\xi\). Precisely, a \(G\)-orbit \(\xi\subset V\) is said to have the incidence property if, for any \(G\)-invariant subvariety \(Y\subset V\), we have \([Y]|_\xi\neq 0 \) if and only if \(\xi\subset [Y]\). This is too much to expect. At the very least, one needs the stabiliser of \(\xi\) to contain a \(U(1)\), since if the stabiliser is trivial (it often is), then the restrictions will be trivial. In terms of singularities, we have to confine ourselves to weighted homogeneous singularities. The main result is that, if the convex hull of the weights of an algebraic group representation \(\rho\) does not contain 0, then any nonempty \(G\)-invariant subvariety of \(V\) has nonzero Poincaré dual. The authors call such a \(\rho\) positive. It follows that, if the normal representation of the stabiliser of each \(x\in V\) is positive, then \(\rho\) has the incidence property for every orbit. ``After this'', the authors explain ``we faced the problem, familiar to many mathematicians, that our theory doesn't apply to the original conjecture.'' Indeed, they give several examples (some due to M.~Kazarian) to show that positivity often fails and that Rimányi's original conjecture, as well as a natural modification of it, is false. From one point of view this is not surprising: positivity is the same as requiring every orbit to be GIT unstable, which is clearly not to be expected. Nevertheless, there are interesting cases when it does hold: some not too complicated singularities, and the quiver representations of Dynkin type.