Localization and gluing of topological amplitudes (Q2575348)

The authors develop a gluing algorithm for Gromov-Witten invariants of toric Calabi-Yau threefolds, based on localization and gluing graphs and involving Hodge integrals on moduli spaces of stable curves. They also conjecture a precise relation between this construction and the topological vertex, and provide computational evidence to support it. More precisely, let \(X\) be a (non-compact) toric Calabi-Yau threefold. The torus action on \(X\) can be encoded into a (at most) trivalent graph \(\Gamma\) whose vertices are the fixed points of the action, and whose edges are invariant rational curves joining fixed points. The torus action on \(X\) induces a torus action on the moduli space of stable maps to \(X\), and the components of the fixed locus in \(\overline{M}_{g,0}(X,\beta)\) are indexed by maps of graphs \(\Upsilon\to \Gamma\). As a consequence of localization formulæ\ for equivariant integrals, the Gromov-Witten potential of \(X\) can be written as a sum over graphs \(\Upsilon\to \Gamma\). The main goal of the paper is to find a formula for the Gromov-Witten invariants of \(X\) based on a decomposition of \(\Gamma\) (and consequently of the maps \(\Upsilon\to \Gamma\)) into smaller units. Ultimately, one is reduced to determine the contribution of a single trivalent vertex and a gluing rule to recover the total contribution of the graph \(\Gamma\) from the contributions of its vertices. Geometrically, the operation of ``cutting out a trivalent vertex'' in the graph \(\Gamma\) can be realized by cutting the toric threefold \(X\) by means of three Lagrangian cycles. Each of these cycles cuts an invariant rational curve along a circle, so one is left with a collection of three disks in \({\mathbb C}^3\) with common origin; the boundaries of these disks are contained in the given Lagrangian cycles. The local contribution from such a vertex is then expected to be the open string free energy associated to a collection of three Lagrangian cycles in \({\mathbb C}^3\). Unfortunately, no completely rigorous mathematical treatment of open string Gromov-Witten invariants is available at the moment. So far, two different strategies have been proposed to compute the vertex contribution. The first one relies on large \(N\) duality and Chern-Simons theory and led to the theory of the topological vertex [\textit{M.~Aganagic}, \textit{A.~Klemm}, \textit{M.~Mariño} and \textit{C.~Vafa}, Commun. Math. Phys. 254, No.2, 425--478 (2005; Zbl 1114.81076)]; the second approach consists in applying localization techniques for equivariant integrals, and is the approach adopted by the authors. They are able to determine an explicit formula for the contribution of a trivalent vertex in terms of integrals over the Deligne-Mumford moduli spaces \(\overline{M}_{g,s}\) of stable curves of a combination of Hodge and Mumford classes, and to descibe the algorithm to recover the Gromov-Witten invariants of \(X\) from gluing of the local contributions. Also, the authors provide nontrivial computational evidence for the following conjecture: the Chern-Simons topological vertex computations coincide with the localization ones when the framing variables \(n_i\) in the Chern-Simons formulation are related to the torus weights \(\rho_i\) of the localization formula by \(n_i=\rho_{i+1}/\rho_{i}\), for \(i=1,2,3\) mod \(3\). The occurrence of non-integral framing variables in the context of large \(N\) duality had been already pointed out in [\textit{D.-E.~Diaconescu} and \textit{B.~Florea}, Large \(N\)-duality for compact Calabi-Yau threefolds, preprint, \texttt{hep-th/0302076}]. Moreover, a proof of the authors' conjecture in the case of univalent vertices has been proved in [\textit{C.-C.~M.~Liu}, \textit{K.~Liu} and \textit{J.~Zhou}, J. Differ. Geom. 65, No.~2, 289--340 (2003; Zbl 1077.14084) and Math. Res. Lett. 11, No.~2--3, 259--272 (2004; Zbl 1094.14043); \textit{A.~Okounkov} and \textit{R.~Pandharipande}, Geom. Topol. 8, 675--699 (2004; Zbl 1062.14035)].