Gromov-Witten gauge theory (Q900850)

The paper constructs gauge-theoretic analogs of Gromov-Witten invariants for the quotient stack \([\mathrm{pt}/\mathbb{C}^{\times}]\). The invariants are obtained by pushing forward products of evaluation classes along the forgetful morphism \(F:\widetilde{\mathcal{M}}_{g,I}([\mathrm{pt}/\mathbb{C}^{\times}])\to\overline{\mathcal{M}}_{g,I}\). Here \(\overline{\mathcal{M}}_{g,I}\) is the stack of stable genus \(g\), \(I\) marked curves, but \(\widetilde{\mathcal{M}}_{g,I}([\mathrm{pt}/\mathbb{C}^{\times}])\) is rather unconventional, accounting for many technical difficulties dealt with in the paper. It is a stack of certain principal \(\mathbb{C}^{\times}\) bundles on marked nodal curves, but without the stability conditions imposed. As a result, it is only Artin, not Deligne-Mumford, stack (points may have continuous stabilizers), which trivializes the usual approach of integrating cohomology classes along the fibers of \(F\) for degree reasons. To get non-trivial invariants the authors use \(K\)-theory instead of cohomology, where there is a non-trivial pushforward based on the functor of \(\mathbb{C}^{\times}\) invariants from \(\mathbb{C}^{\times}\) representations to vector spaces. The second difficulty is topological: \(\widetilde{\mathcal{M}}_{g,I}([\mathrm{pt}/\mathbb{C}^{\times}])\) is complete but not proper, neither separated (limits of bundle families may not be unique) nor of finite type, so not all \(K\)-theory classes have well-defined indices. The authors single out a well-behaved subring of \(K\big(\widetilde{\mathcal{M}}_{g,I}([\mathrm{pt}/\mathbb{C}^{\times}])\big)\) consisting of \textit{admissible classes}. The main result of the paper is that those do have indices, and the derived pushforward of an admissible complex along \(F\) is a bounded complex of coherent sheaves, thus yielding the well-defined invariants. In contrast to the traditional case, these invariants are always ``twisted'' in the sense of Coates-Givental. At the end of the paper the authors construct an analogous stack, but not invariants, for general GIT quotients. They expect to recover the ordinary Gromov-Witten invariants by applying the Chern character to certain limits of theirs, which reflect more information. Physically, these gauge invariants are expected to be the correlation functions of a TQFT, which is also a gauge theory. The construction can be interpreted as a topological version of Feynman's path integral over the space of \(U(1)\) connections. The idea that the path integral in gauged non-linear sigma-models might compute invariants of the moduli spaces of connections \(A\) and \(\overline{\partial}_A\) holomorphic sections goes back to \textit{E. Witten}'s paper [Commun. Math. Phys. 118, No. 3, 411--449 (1988; Zbl 0674.58047)] on topological sigma models.