The logarithmic gauged linear sigma model (Q2049910)

The paper continues the authors' project to develop mathematical foundations and computational methods for the gauged linear sigma model (GLSM) introduced by Witten in 1993. Compared to the Gromov-Witten theory, there is an additional torus action on the target (the ``\(R\)-charge'') that creates difficulties with the properness of the moduli stacks carrying the virtual cycles. The paper focuses on GLSM with the so-called hybrid targets, such as non-GIT quotients. The main results include relating several natural virtual fundamental cycles to each other, establishing a localization formula for GLSM invariants, and developing a new GLSM based technique for computing higher genus Gromov-Witten invariants of complete intersections. The authors' approach is based on \(R\)-maps, logarithmic \(R\)-maps that factor through a proper substack with the trivial logarithmic structure, which generalize pre-stable maps and whose moduli give a mathematical description of spaces on which the general A-twist localizes. A subtle \(\infty\)-stability type condition allows proving that the category of stable log \(R\)-maps is represented by a proper Deligne-Mumford stack, resolving a major technical hurdle of GLSM theory. The moduli of log \(R\)-maps admit two different perfect obstruction theories. The first, canonical one, is obtained from Olsson's logarithmic cotangent complex. With the help of a superpotential, it allows to define, in addition to the canonical virtual cycle, the cosection localized virtual cycle supported on the proper substack parametrizing \(R\)-maps to the critical locus of the superpotential. However, the canonical cosection over the substack category does not have a nice extension to the target stack category. A proper morphism, called modular principalization, from the authors' previous paper is used to get around that, and extend cosections to the boundary of logarithmic compactification. This leads to the second, reduced, perfect obstruction theory and the corresponding virtual cycle. The paper then establishes correspondences among the three virtual cycles and sketches how they can be applied to computing invariants. The reduced cycle is proved to be just the pullback of the cosection localized cycle by the inclusion map, which recovers the FJRW theory and Clader's hybrid model. As a result, the Gromov-Witten invariants of complete intersections involving only cohomology cycles from the ambient space can be computed in terms of log GLSM invariants, which is especially effective for higher genera. The canonical cycle is proved to be the sum of the reduced cycle and an additional virtual cycle, which can be further decomposed in terms of torus localizations. This correspondence can be used to prove structural properties of the Gromov-Witten invariants of quintics, for example. Finally, the authors relate virtual cycles for different twisting choices, which translates into relating GLSM invariants for different targets with isomorphic infinity hyperplanes. The applications include proving the Landau-Ginzburg/Calabi-Yau correspondence for quintics, and the formula for the cycle of holomorphic differentials with specified zeros conjectured by the second author.