Sheaves on abelian surfaces and strange duality (Q957890)

For a polarized abelian surface \(A\) with a polarization \(H\), let \(\hat{A} = \text{Pic}^o(A)\) be its dual, and let \({\mathcal M}_v\) be the moduli space of Gieseker \(H\)-semistable sheaves \(E\) on \(A\) with Mukai vector \(v\). Fixing a line bundle \(\Lambda\) on \(A\) with \(c_{1}(\Lambda)=c_{1}(v)\) yields a determinant morphism \(\alpha^+: {\mathcal M}_v \rightarrow \hat{A}, E \mapsto \det(E) \otimes \Lambda^{-1}\); and let \(M^+_v\) be the fiber of \(\alpha^+\) at \(0\). After applying a Fourier-Mukai transform \(RS\) one gets dually a map \(\alpha^-_{\Lambda}: {\mathcal M}_v \rightarrow A, E \mapsto \det\;RS(E) \otimes \det\;RS(\Lambda)^{-1}\); and let \(M^-_v\) be the fiber of \(\alpha_{\Lambda}^-\) at the origin. The elements \(w \in v^{\perp}\) define (non-uniquely) line bundles \(\Theta_w\) on \({\mathcal M}_v\). The authors compute, under certain mild conditions on \(v\) and \(w\) (see Assumptions 1), the characteristics \(\chi (M^+_v, \Theta_w)\) (Theorem 1) and \(\chi(M^-_w, \Theta_v)\) (Theorem 2), which turns out to be the same. This coincidence answers a question raised in [\textit{L. Göttsche, N. Nakajima, K. Yoshioka}, K-theoretic Donaldson invariants via instanton counting, \url{arXiv:math/0611945}] and can be regarded as an analog of the strange duality for curves [see e.g. \textit{A. Beauville}, Math. Sci. Res. Inst. Publ. 28, 17--33 (1995; Zbl 0846.14024)]. Under the same assumptions, a similar coincidence is obtained for \(\chi({\mathcal M}_w, \Theta_v)\) and \(\chi({K}_v, \Theta_w)\), where \(K_v\) is the fiber over 0 of the Albanese map \(a_v = (\alpha^+_{\Lambda}, \alpha^-_{\Lambda}): {\mathcal M}_v \rightarrow \check{A} \times A\), see Theorem 3. Under the additional assumption that \(h^2(E \otimes F) = 0\) for any \(E\) and \(F\) with Mukai vectors respectively \(v\) and \(w\) (see Assumption 2) the authors show that there exist naturally defined ``strange duality'' morphisms \(SD^{\pm}: H^0(M^{\pm}_v, \Theta_w)^\vee\rightarrow H^0(M^{\pm}_w, \Theta_v)\) (see \S 2); and it is conjectured that the maps \(SD^{\pm}\) should be either zero or isomorphisms.