Hurwitz numbers and intersections on moduli spaces of curves. (Q1608542)

The authors exhibit a spectacular, very fundamental link between two superficially unrelated concepts in mathematics, namely in combinatorics, on the one hand, and in algebraic geometry on the other. The amazing connection described here has utmost significant consequences for further research in both moduli theory of complex algebraic curves and mathematical physics. In fact, the authors' discovery has already initiated quite an avalanche of subsequent research activities, in this direction, since its first announcement in 1999 [cf.: \textit{T. Ekedahl}, \textit{S. K.Lando}, \textit{M. Shapiro} and \textit{A. Vainshtein}, C. R. Acad. Sci., Paris, Sér. I, Math. 328, No. 12, 1175--1180 (9991; Zbl 0953.14006)]. The present paper is a fully elaborated version of that first announcement, with somewhat different technicalities and with decisive improvements. Basically, the main result of this work is a numerical identity relating the classical Hurwitz numbers (in combinatorics) to certain numerical invariants in the enumerative geometry of moduli spaces of Riemann surfaces (in complex geometry). In the meantime, the authors' fundamental identity has already become well-known as the so-called ``ELSV formula'', where its possible applications and generalizations have rapidly grown into a subject of intensive current research. As to the precise statement of the ELSV formula, recall that Hurwitz numbers are classical combinatorial quantities which suitably count factorizations of permutations into transpositions. Equivalently, for given tuples \((g,n)\in\mathbb{N}^2\) and \((k_1,\dots, k_n)\in\mathbb{N}^n\), the Hurwitz number \(h_{g; k_1,\dots, k_n}\) is the number of topological equivalence classes of ramified coverings \(f: C\to\mathbb{P}^1\), where \(C\) is a genus-\(g\) compact Riemann surface, \(f\) is branched over the point at infinity with branching vector \((k_1,\dots, k_n)\), and \(f\) has \(r= d +n +2g -2\) fixed simple branch points otherwise Here the number \(d\) denotes the degree of the covering map \(f\), that is \(d= k_1+\cdots +k_n\). A. Hurwitz, who first studied these numbers (from the latter point of view) as early as in 1891, once posed the problem of computing \(h_{g;k_1,\dots, k_n}\) by an explicit formula. Nearly 100 years later, physicists made some major progress in understanding the geometric structure underlying the Hurwitz numbers, and the authors of the present paper finally gave the first complete, explicit, and mathematically rigorous answer to Hurwitz's classical problem, by expressing Hurwitz numbers in terms of intersection numbers for the Chern classes of certain vector bundles on the Deligne-Mumford compactification \(\overline M_{g,n}\) of the moduli space of complex curves of genus-\(g\) with \(n\) marked points on them. The ELSV formula for Hurwitz numbers, derived in the present paper, reads as follows: \[ h_{g; k_1,\dots, k_n}= {(d+ n+ 2g- 2)!\over \#\Aut(k_1,\dots, k_n)}\,\prod^n_{i=1} {k^{k_i}_i\over k_1!}\, \int_{\overline M_{g,n}} {c(\Lambda^\lor_{g,n})\over \prod^n_{i=1} (1- k_i\psi_i)}. \] Here \(\Lambda_{g,n}\) denotes the Hodge bundle over the moduli space \(\overline M_{g,n}\), \(c(\Lambda^\lor_{g,n})\) is the total Chern class of its dual, and the forms \(\psi_i\) are first Chern classes of other natural line bundles on \(\overline M_{g,n}\). The proof of this formula is based on several new ideas and techniques which appear to be crucial in other contexts, too. First of all, and most importantly, the so-called Lyashko-Looijenga mapping, which is defined on the Hurwitz spaces \(H_{g;k_1,\dots, k_n}\) and essentially gives the branch points of a map from a curve to the projective line, is extended to the completed Hurwitzes spaces. These completed Hurwitz spaces are here appropriately defined, and their main feature is that they admit a natural structure of a cone over the compactified moduli space \(\overline M_{g,n}\). Furthermore, the authors extend the theory of Segre classes in algebraic intersection theory (à la W. Fulton) to cones, which allows them to interpret the Hurwitz numbers in different ways, namely as multiplicities of the Lyashko-Looijenga mapping, on the one hand, and as combinatorial multiplicities of Segre classes of completed Hurwitz spaces on the other. Combining this with a careful local analysis of the Lyashko-Looijenga mapping, the final outcome is indeed the integral on the right-hand side of the ELSV formula mentioned above. The general significance of the ELSV formula becomes immediately evident in view of the very fact that numerous earlier results on both combinatorial Hurwitz numbers and intersection numbers for moduli spaces of \(n\)-pointed stable curves appear as special cases of it. The authors explain this aspect in an extra section of their paper, entitled ``Previous Research''. More evidence of the pioneering character of the ELSV formula, and of its tremendous impact likewise, becomes manifest from the many subsequent research articles (by various authors) that were following the present paper hard on its heels, including further generalizations and applications of just as great importance. For instance, another proof of the ELSV formula has recently been obtained by \textit{T. Graber} and \textit{R. Vakil} [Compos. Math. 135, No. 1, 25--36 (2003; Zbl 1063.14032)], and a generalization to the geometry of double Hurwitz numbers has been undertaken by \textit{I. P.Goulden}, \textit{D. M. Jackson} and \textit{R. Vakil} [Towards the geometry of double Hurwitz numbers, preprint, \url{http://arviv.org/abs/math.AG/0309440}] shortly after that. On the other hand, as the ELSV formula involves top intersections in the tautological ring of the moduli space \(\overline M_{g,n}\), its relation to the mathematical physics around Witten's conjecture (or Kontsevich's theorem) is apparent. Actually, this connection becomes transparent in the recent proof of Witten's conjecture by \textit{A. Okounkov} and \textit{R. Pandharipande} [Gromov-Witten theory, Hurwitz numbers, and matrix models. I: preprint, \url{http://arxiv.org/abs/math.AG/0101147}]. Altogether, the significance of the results of the paper under review can barely be overestimated.