The moduli space of curves, double Hurwitz numbers, and Faber's intersection number conjecture (Q659646)

The celebrated Faber conjectures give a description of the tautological ring in the cohomology of the moduli space of smooth genus \(g\) curves. They assert, roughly speaking, that it behaves like a cohomology ring of a \((g-2)\)--dimensional complex projective manifold, in particular its structure is completely determined by the top intersections of the Mumford-Morita-Miller \(\kappa\)-classes. Furthermore, Faber's Intersection Number Conjecture gives a combinatorial description of these top intersections. Graber and Pandharipande showed that the latter follows from the degree \(0\) part of the Virasoro conjecture for the projective plane, proved by Givental. However, the proof via Virasoro conceals the combinatorial structure behind the intersection numbers, so a direct combinatorial proof is desirable. This paper outlines a strategy for such a proof and partially implements it. The idea is to express the cohomology of the moduli space of stable \(n\)--pointed genus \(g\) maps with rational tails \(\mathcal{M}_{g,n}^{rt}\) in terms of combinatorics of the double Hurwitz numbers. A stable map with rational tails consists of a single smooth genus \(g\) component with chains of spheres attached, the dual graph is therefore a tree. The paper relies on the observation that an analog of the ELSV formula links the cohomology of \(\mathcal{M}_{g,n}^{rt}\) to the genus \(0\) double Hurwitz numbers, just as the ELSV formula itself links the cohomology of the Deligne-Mumford compactification \(\overline{\mathcal{M}}_{g,n}\) to the Hurwitz numbers. The dual graphs being trees simplifies the link considerably. The double Hurwitz numbers enumerate the branched covers of a sphere by another sphere with prescribed ramifications over \(0\) and \(\infty\) and simplest non-trivial branchings at other points. The number \(n\) of marked points corresponds to the maximal number of parts in the partitions describing the ramifications. Degeneration of the target yields a join-cut recursion for the tautological classes of \(\mathcal{M}_{g,n}^{rt}\), the Faber-Hurwitz classes, which can be solved explicitly. On the other hand, localization expresses them as sums over decorated trees in terms of the double Hurwitz numbers and the top intersections of the \(\kappa\)--classes. As with the ELSV formula, this relation can be explicitly inverted expressing the top intersections in terms of the Faber-Hurwitz classes and the double Hurwitz numbers. The results are merged into a single generating series in genus, which contains all the necessary pieces for proving Faber's Intersection Number Conjecture. The proof is completed for \(n\leq3\), where genus \(0\) double Hurwitz numbers are known explicitly, and the same approach works for \(n=4,5\). Unfortunately, for \(n>5\) such explicit expressions are lacking so far.