Wall-crossings in toric Gromov-Witten theory. I: Crepant examples (Q834392)

Many recent conjectures in algebraic geometry come from string theory. Invariably, their original formulations turn out to be too naive and are modified to accomodate new examples. In this paper the authors put forth such a modification for the Crepant Resolution Conjecture of Ruan and Bryan-Graber based on detailed analysis of \(\mathbb{P}(1,1,2)\) and \(\mathbb{P}(1,1,1,3)\) examples. Informally, the Crepant Resolution Conjecture claims that the Chern-Ruan quantum cohomology of an orbifold coincides with the usual quantum cohomology of its crepant resolution under analytic continuation in quantum parameters. The authors describe how the conjecture fits into the larger picture from the string theory. Presumably, there is a stringy Kähler moduli \(\mathcal{M}\) with a bundle of algebras over it, which become isomorphic to quantum cohomology algebras near the cusps of \(\mathcal{M}\). The conjecture comes from identification of the cohomologies with the isomorphic algebras at two different cusps. However, as the authors point out, there is additional flexibility built into this picture, because the isomorphism depends on a choice of flat trivializations near the cusps. Namely, the quantum parameters are the exponentiated flat coordinates, and one should allow for a change of variables in addition to analytic continuation. This idea is developed for the two orbifolds mentioned above. Following Givental and Barannikov, the authors deal with the mirror moduli \(\mathcal{M}_B\) which, unlike \(\mathcal{M}\), can be rigorously constructed from toric data. They construct a refinement of the bundle of algebras over it, called the variation of semi-infinite Hodge structure, which along with some extra data determines the flat trivializations canonically. For \(\mathbb{P}(1,1,2)\) and its resolution the flat structures are the same, and the isomorphism of quantum cohomologies extends to the corresponding Frobenius manifolds. However, this is not the case for \(\mathbb{P}(1,1,1,3)\), where the isomorphism itself depends on the quantum parameters. Although not a counterexample, this argument takes away the main reason for believing in the original Crepant Resolution Conjecture. Proofs are based on mirror theorems for toric orbifolds. Those relate the variation of semi-infinite Hodge structure over \(\mathcal{M}_B\) to a family of tangent spaces to the Givental Lagrangian submanifold, which encodes the genus-zero Gromov-Witten invariants. Accordingly, the authors replace the Crepant Resolution Conjecture with a claim that the Givental submanifolds of an orbifold and its crepant resolution are related by a linear symplectic isomorphism. To support it, they prove that it holds for both \(\mathbb{P}(1,1,2)\) and \(\mathbb{P}(1,1,1,3)\), and implies the Bryan-Graber (and hence Ruan) Crepant Resolution Conjecture provided the orbifold cohomology is hard Lefschetz. In general however, they predict that even the weaker Ruan conjecture is false. As a nice application, the methods of this paper allow to single out a canonical germ of a Frobenius manifold from the Landau-Ginzburg model for a singularity. It was known previously that many germs exist, but the authors show that only one satisfies extra conditions of global monodromy and homogeneity.