Counting sheaves on Calabi-Yau 4-folds. I (Q6046465)

The paper under review concerns the enumerative geometry of sheaves on Calabi-Yau manifolds of dimension 4. The analogous problem in dimension 3, known as Donaldson-Thomas (DT) theory, is an active subject of research. When \(X\) is a Calabi-Yau threefold, the moduli space \(Sh(X)\) of sheaves, with appropriate stability and numerical conditions, has a remarkable property: it is virtually smooth. In particular, it can locally be expressed as the zero locus of a section of a vector bundle \(\mathcal E\) on a smooth space \(\mathcal A\). Moreover, the difference between the rank of \(\mathcal E\) and the dimension of \(\mathcal A\) is \(0\); this is a consequence of the Calabi-Yau assumption. The local descriptions are sufficiently compatible to enusre that there is a canonically defined class in Chow homology of \(Sh(X)\), morally given by the Chern class \(c_{top}(\mathcal E)\). This virtual class gives rise to a well-behaved enumerative geometry. When \(X\) is a Calabi-Yau fourfold, the picture above does not immediately generalize. One reason to hope for a generalization is that a sister theory to DT theory, Gromov-Witten (GW) theory, is defined in all dimensions. Moreover, for threefolds, the relationship between GW and DT theory is an important line of inquiry. The purpose of this paper (and related earlier work of Borisov-Joyce) is to establish this hoped-for theory. The main geometric insight is the following. The moduli space \(Sh(X)\) of sheaves on a Calabi-Yau fourfold locally looks like the zero locus of a section \(s\) of a vector bundle \(\mathcal E\) on a smooth space \(\mathcal A\) but with two important new features -- the fibers of the vector bundle \(\mathcal E\) are naturally equipped with a quadratic form and the section \(s\) is isotropic. Given this, the authors construct a class in the Chow homology of \(Sh(X)\), morally given by a square root of the top Chern class of \(\mathcal E\). This ``square root'' class essentially comes from the fact that because \(SO(r,\mathbb C)\) is homotopy equivalent to \(SO(r,\mathbb R)\), so one expects the real bundle underlying \(\mathcal E\). However it should be noted that there is interesting geometry needed to access the class in Chow homology. After the dust has settled, the virtual class on \(Sh(X)\) is shown to satisfy torus localization and cosection localization formulas. This makes the theory computable in practice, and this is a important feature of the authors' proposal for the theory. It leads to an important and rich new direction in enumerative geometry.