A proof of a conjecture of Mariño-Vafa on Hodge integrals (Q1885378)

Let \(\pi\colon \overline{\mathcal M}_{g,n+1}\to\overline{\mathcal M}_{g,n}\) be the universal curve over the Deligne-Mumford stack of genus \(g\) stable curves with \(n\) marked points, and let \(\omega_\pi\) be the relative dualizing sheaf. The Hodge bundle is the rank \(g\) vector bundle on \(\overline{\mathcal M}_{g,n}\) defined by \({\mathbb E}=\pi_*\omega_\pi\); the line bundles \({\mathbb L}_i\) on \(\overline{\mathcal M}_{g,n}\) are defined as \({\mathbb L}_i=s_i^*\omega_\pi\), where \(s_i\colon \overline{\mathcal M}_{g,n}\to\overline{\mathcal M}_{g,n+1}\) is the section of \(\pi\) corresponding to the \(i\)-th marked point in the fiber of the universal curve. It is customary to denote by \(\psi_i=c_1({\mathbb L}_i)\) the first Chern class of \({\mathbb L}_i\) and by \(\lambda_j=c_j({\mathbb E})\) the \(j\)-th Chern class of the Hodge bundle. A Hodge integral is an integral of the form \[ \int_{\overline{\mathcal M}_{g,n}}\psi_1^{k_1}\cdots \psi_n^{k_n}\lambda_1^{l_1}\cdots \lambda_g^{l_g}. \] Computing Hodge integrals is a difficult problem in general. \textit{M.~Mariño} and \textit{C.~Vafa} [in: Orbifolds in mathematics and physics. Contemp. Math. 310, 185--204 (2002; Zbl 1042.81071)], obtained a closed formula for a generating function of certain open Gromov-Witten invariants. By localization techniques, and comparing it with the calculations in [\textit{S.~Katz} and \textit{C.-C.~M.~Liu}, Adv. Theor. Math. Phys. 5, No.~1, 1--49 (2001; Zbl 1026.32028)], Mariño and Vafa's formula can be reduced to a formula for Hodge integrals. The proof of the Mariño-Vafa formula relies on large \(N\) duality, a tool which is not fully clarified from a mathematical point of view at the moment. Yet, the resulting formula for Hodge integrals is a purely mathematical statement and in this paper the authors give a purely algebro-geometric proof of it. The strategy of the proof is the following. The authors denote by \({\mathcal C}(\lambda;\tau;p)\) the generating function of Hodge integrals occurring in the Mariño-Vafa formula, and by \(R(\lambda;\tau;p)\) the value of \({\mathcal C}(\lambda;\tau;p)\) predicted by Mariño and Vafa: it is an explicit expression written in terms of characters of symmetric groups. Then they show that both \({\mathcal C}(\lambda;\tau;p)\) and \(R(\lambda;\tau;p)\) satisfy the same differential equation (the cut-and-join equation), and have the same initial value: \({\mathcal C}(\lambda;0;p)=R(\lambda;0;p)\), hence \({\mathcal C}(\lambda;\tau;p)\) and \(R(\lambda;\tau;p)\) coincide. The hard part of the proof consists in showing that \({\mathcal C}\) and \(R\) satisfy the cut-and-join equation. As far as concerns \(R\), this fact is a consequence of algebro-combinatorial relations on symmetric groups. The proof that \({\mathcal C}\) satisfies the cut-and-join equation, instead, is based on functorial virtual localization on the moduli spaces of relative stable morphisms to \({\mathbb P}^1\) [\textit{J.~Li}, J. Differ. Geom. 57, No.~3, 509--578 (2001; Zbl 1076.14540)]; \textit{J.~Li} and \textit{Y.~S.~Song}, Adv. Theor. Math. Phys. 5, No.~1, 67--91 (2001; Zbl 1015.14026)].