Higher genus quasimap wall-crossing for semipositive targets (Q2359445)

For a complex affine algebraic variety $W$, acted upon by a reductive group $G$, a choice of character $\theta$ determines a GIT quotient $W/\!\!/G$. For $\varepsilon\in\mathbb{Q}_{>0}\cup\{0,\infty\}$ stability conditions produce relatively proper Deligne-Mumford moduli stacks of $\varepsilon$-stable quasimaps from pointed curves of genus $g$ to $W/\!\!/G$, which determine $\varepsilon$-quasimap descendant invariants when the target is projective or quasi-projective with a nice torus action. For $\varepsilon\in(2,\infty)$ they coincide with the Gromov-Witten invariants of $W/\!\!/G$. \par The paper extends the authors' previous conjectures on the wall-crossing formulas for these invariants to higher genus for semi-positive triples $(W,G,\theta)$ (i.e. with nef anti-canonical class), and proves them for toric varieties, including the Calabi-Yau case. The proof is inspired by Marian-Oprea-Pandharipande proof [\textit{A. Marian} et al., Geom. Topol. 15, No. 3, 1651--1706 (2011; Zbl 1256.14057)] of the conjectures for $W/\!\!/G$ being the Grassmannian, but introduces several new ideas. \par Let $J_0^\varepsilon$, $J_1^\varepsilon$ be the $q$-truncations of the Givental's functions $I_0$, $I_1$; $t(\psi)=t_0+t_1\psi+t_2\psi^2+\dots$, where $t_j\in H^*(W/\!\!/G,\mathbb{Q})$ are the even cohomology classes, and $F_g^\varepsilon=\sum q^\beta/m!\langle t(\psi_1),\dots,t(\psi_m)\rangle_{g,m,\beta}^\varepsilon$, $\beta\in\mathrm{Eff}(W,G,\theta)$, be the descendant potential. Then a wall-crossing formula asserts that $(J_0^\varepsilon)^{2g-2}F_g^\varepsilon(J_0^\varepsilon t(\psi)-J_1^\varepsilon)$ remains unchanged for all $\varepsilon$. The conjecture is proven first for $g=0$ when $W$ admits a torus action commuting with the action of $G$, and such that the fixed points of the induced action on $W/\!\!/G$ are isolated. \par The main result is that the conjecture holds for non-singular quasi-projective toric semi-positive GIT quotients $X=\mathbb{C}^{n+r}/\!\!/(\mathbb{C}^*)^r$, and for the total spaces of the canonical bundles over type A partial flag manifolds. This implies, in particular, that the invariants of non-singular projective semi-positive Fano varieties are independent of $\varepsilon$. For the toric Calabi-Yau, $F_g^{0+}|_{t(\psi)=0}$ is the $A$-model genus $g$ potential after applying the mirror map. Assuming mirror symmetry, it should match the B-model potential expanded around the large complex structure point.