TITLE: Is there a Chern-Gauss-Bonnet theorem for orbifolds?
QUESTION [26 upvotes]: There's a Gauss-Bonnet theorem for compact 2-orbifolds(due to Satake, I think), which gives a relation between the curvature of a Riemannian orbifold and the orbifold topology(i.e. taking into account not just the structure of O as a Hausdorff topological space, but also the structure of the singular points.) The only proof that I've seen was very much like the classical one of the Gauss-Bonnet theorem, with geodesic triangulations, angle defects and so on. This approach doesn't generalise to higher dimensions to prove the Chern-Gauss-Bonnet theorem, and I haven't even found a conjectured Chern-Gauss-Bonnet formula. I'm especially interested in the four-dimensional case, where the integrand is amenable. One reason I'm interested is just to get a feel for how much more complicated orbifolds are than manifolds; on the one hand, many basic definitions seem to go through from the manifold world to the orbifold world, but the generalisation leads to significant complications in practice. On the other hand, Kleiner and Lott posted a paper on the arxiv in which they use Ricci flow to geometrise 3-orbifolds, so orbifolds certainly seem like a good arena in which to generalise differential geometry. Chern-Gauss-Bonnet is a bit of a benchmark for higher-dimensional Riemannian geometry, and I'd like to know the state of the art is in this case.

REPLY [7 votes]: One could probably show that the Gauss-Bonnet integral is a topological invariant of the orbifold (i.e. independent of the Riemannian metric), by doing local deformations of the metric within orbifold atlases. If you take an atlas of the orbifold which is the quotient of a ball by a finite group, then for any two Riemannian metrics which agree in a neighborhood of the boundary, the Gauss-Bonnet integral should be the same (since one may work equivariantly in the ball cover). Any two Riemannian metrics may be connected by a sequence of such local deformations by using a partition of unity, and so the total Gauss-Bonnet integral is a topological invariant. The question is whether this invariant equals the Euler characteristic. 
I think this may be shown by reduction to the good orbifold case by a cut and paste trick. If you have a codim-1 suborbifold, one may assume that the Riemannian metric is reflection symmetric in a neighborhood of the codim-1 suborbifold.  Cut along the suborbifold,  then mirror the resulting boundary to get a new orbifold. Clearly the Gauss-Bonnet integral and the Euler characteristic of the resulting orbifold is the same (the orbifold euler characteristic is additive upon gluing along mirrored suborbifolds). I think one should be able to iterate this  to reduce to the case of a good orbifold, which would look like a ball quotient with a right-angled boundary pattern (like in Thurston and Davis' reflection trick). 
Take a covering by ball-quotients, and assume that their boundaries intersect transversally. One may choose a metric on the orbifold so that the metric is locally a product metric near the boundary of each ball quotient. When several of these intersect, the metric will look locally like a product of a metric on a cube times the metric on the intersection (so that the boundaries meet orthogonally). Then cut along these codim-1 suborbifolds, and mirror the boundary. Davis' argument shows that these orbifolds are good, and thus the Gauss-Bonnet theorem applies. Since each region is contained in a ball quotient, it is good and you can take a manifold cover. Then reflect iteratively in the boundary by doubling along each subset of the boundary coming from a single boundary of a ball quotient. You end up with a closed manifold, so the orbifold is good.