Computing linear Hodge integrals (Q2479707)

The author proves that all linear Hodge integrals, i.e., Hodge integrals with at most one \(\lambda\)-class, can be expresed as polynomials in terms of lower dimensional linear Hodge integrals. This can be algorithmically implemented to compute any given linear Hodge integral; some computational examples are presended. More precisely, let \(\overline{\mathcal M}_{g,n}\) be the Deligne-Mumford moduli stack of genus \(g\) \(n\)-marked stable curves. A Hodge integral is an integral of the form \[ \int_{\overline{\mathcal M}_{g,n}}\psi_1^{j_1}\cdots\psi_n^{j_n}\lambda_1^{k_1}\cdots \lambda_g^{k_g}, \] where \(\psi_i\) is the first Chern class of the cotangent line bundle at the \(i\)-th marked point, and \(\lambda_i\) is the \(i\)-th Chern class of the Hodge bundle. Explicit evaluation of Hodge integrals is a nontrivial problem. When \(\lambda\)-classes are not involved, Hodge integrals can be recursively computed via their relation to the string equation and the KdV hierarchy [\textit{E.~Witten}, Surv. Differ. Geom., Suppl. J. Diff. Geom. 1, 243--310 (1991; Zbl 0757.53049); \textit{M.~Kontsevich}, Commun. Math. Phys. 147, No. 1, 1--23 (1992; Zbl 0756.35081)]. The case of one top-degree \(\lambda\)-class is known as the \(\lambda_g\)-conjecture, and has been proved in [\textit{C.~Faber} and \textit{R.~Pandharipande}, Ann. Math. (2) 157, No. 1, 97--124 (2003; Zbl 1058.14046)]: \[ \int_{\overline{\mathcal M}_{g,n}}\psi_1^{j_1}\cdots\psi_n^{j_n}\lambda_g =\binom{2g+n-3}{j_1,\dots,j_n}b_g. \] where \(b_g=\frac{| (2^{2g-1}-1)B_{2g}| }{2^{2g-1}(2g)!}\). The case of \(\lambda_{g-1}\) with one marked point can be computed from the Mariño-Vafa formula, obtaining \[ \int_{\overline{\mathcal M}_{g,1}}\psi_1^{2g-1}\lambda_{g-1} =b_g\sum_{i=1}^{2g-1}\frac{1}{i}-\frac{1}{2} \sum_{g_1+g_2=g} \frac{(2g_1-1)!(2g_2-1)!}{(2g-1)!}b_{g_1}b_{g_2}, \] and the case of more than one marked points can then be computed via the cut-and-join equation [\textit{M.~Mariño} and \textit{C.~Vafa}, Contemp. Math. 310, 185--204 (2002; Zbl 1042.81071); \textit{C.-C.~M.~Liu, K.~Liu} and \textit{J.~Zhou}, J. Differ. Geom. 65, No. 2, 289--340 (2003; Zbl 1077.14084), J. Am. Math. Soc. 20, No. 1, 149--184 (2007; Zbl 1111.14053)]. For a fixed partition \(\mu\) of a positive integer \(d\), the author introduces a moduli space \(\overline{\mathcal M}^{\bullet}_{\chi,n}({\mathbb P}^1,\mu)\) of morphisms from a possibly disconnected prestable curve of Euler number \(\chi\) and with \(l(\mu)+n\) marked points to a chain of \({\mathbb P}^1\)'s. This moduli space is naturally equipped with a \({\mathbb C}^*\)-action and as a result of the localization formula, the author obtains the identity \[ \text{ Coeff}_{\lambda^{l(\mu)-\chi}}\left(\sum_{| \nu| =| \mu| }\Phi^\bullet_{\mu,\nu}(-\lambda)z_\nu{\mathcal D}^\bullet_{\nu,e}(\lambda)\right)=0, \] for any partitions \(\mu\) and \(e\) with \(| e| <| \mu| +l(\mu)-\chi\). In the above formula \(\Phi^\bullet(\lambda)\) is a generating series of double Hurwitz numbers and \({\mathcal D}^\bullet(\lambda)\) is a certain generating series of Hodge integrals. In particular, the above formula implies that any linear Hodge integral can be explicitly expressed as a polynomial in terms of lower-dimensional linear Hodge integrals. This can be algorithmically implemented to recursively compute all linear Hodge integrals. In the final part of the paper, the author uses this algorithm to compute the 16 possible linear Hodge integrals on 4-dimensional moduli spaces, showing that the resulting values agree with the known ones.