A rank 2 Dijkgraaf-Moore-Verlinde-Verlinde formula (Q2418749)

The paper addresses the problem of finding closed formulae for virtual elliptic genera and virtual cobordism classes of moduli spaces of rank \(2\) sheaves on smooth projective surfaces with holomorphic 2-forms. Such formulae refine the ones proposed by the authors in [Commun. Math. Phys. 376, No. 1, 1--49 (2020; Zbl 1440.14253)], where they address the case of \( \chi_{-y}\)-genus and virtual Euler characteristic. Let \(S\) be a smooth projective surface with \(b_1=0\) and polarization \(H\). Denote by \[M:=M_S^H(r,c_1,c_2) \] the moduli space of rank \(r\) Gieseker \(H\)-stable torsion-free sheaves on \(S\) with Chern classes \(c_1\in H^2(S, \mathbb{Z})\), \(c_2\in H^4(S, \mathbb{Z})\). The authors assume, throughout the paper, that no rank \(r\) Gieseker \(H\)-semistable torsion-free sheaves on \(S\) with Chern classes \(c_1, c_2\) exist; then \(M_S^H(r,c_1,c_2) \) is a projective scheme. \(M\) is endowed with a perfect obstruction theory in the sense of Behrend-Fantechi, therefore carries a virtual cycle \([M]^{vir}\); virtual elliptic genera are defined by integrating against such class. The authors propose that, if \(S\) is a smooth projective surface with \(b_1(S)=0\), \(p_g(S)> 0\) with only Seiberg-Witten basic classes \(0\) and \(K_S\neq 0\), then \(\mathrm{Ell}^{\mathrm{vir}}(M)\) is the coefficient of \(p^{\mathrm{vdim}(M)}\) of \[ \psi_S(p,q,y):=8\left(\frac{1}{2 \mathsf{L}_2(\phi_{0,1})} \right)^{\chi(\mathcal{O}_S)}\left( \frac{2\mathsf{L}_4(2\phi_{0,\frac{1}{2}}\phi_{0,\frac{3}{2}})\mathsf{L}(-2\phi_{0,\frac{1}{2}})}{\mathsf{L}_2(-2\phi^{\mathrm{ev}}_{0,\frac{1}{2}}|_{(q^{1/2},y)}-\phi_{0,\frac{1}{2}}|_{(q^2,y^2)} +2\phi^2_{0,\frac{1}{2}}) } \right)^{K_S^2} \] All ingredients in these formula are expressed in terms of the Igusa cusp form \(\chi_{10}\) and Borcherds type lifts (denoted by \(\mathsf{L}\)) of three quasi-Jacobi forms, related to the Weierstrass elliptic function. The motivation for looking at Borcherds lifts of modular forms comes from the celebrated formula in string theory of Dijkgraaf-Moore-Verlinde-Verlinde \[ \sum_{n=0}^{\infty}\mathrm{Ell}(K3^{[n]})p^n=\frac{1}{\mathsf{L}( \mathrm{Ell}(K3))}, \] relating the virtual elliptic genera of Hilbert schemes of points on a \(K3\) surface to the elliptic genus of the \(K3\) surface at hand. Next, the authors shift their attention to virtual cobordism classes. By a result of \textit{J. Shen} [J. Lond. Math. Soc., II. Ser. 94, No. 2, 427--446 (2016; Zbl 1375.14184)], the perfect obstruction theory on \(M\) induces a virtual cobordism class \([M]^{vir}_{\Omega_{*}}\in \Omega_{vd}(M)\) in the algebraic cobordism of \(M\); invariants are defined by pushing this class to the algebraic cobordism of the point \(\pi_*[M]^{vir}_{\Omega_{*}}\in \Omega_{vd}(pt)\). Under the same hypotheses as before, the authors propose that \(\pi_*[M]^{vir}_{\Omega_{*}}\) is the coefficient of \(p^{\mathrm{vdim}(M)}\) of \[ \psi_S(p,\mathsf{v}):=8\left(\frac{1}{2} \left(\sum_{n=0}^\infty[K3^{[n]}]p^{2n}\right)^{1/2} \right)^{\chi(\mathcal{O}_S)}\bigg( 2L(p,\mathsf{v}) \bigg)^{K_S^2} \] where \(L(p,\mathsf{v})\) is a (non-explicit) universal power series. The authors generalize the conjectures also to arbitrary smooth projective surface with \(b_1=0\) and \(p_g(S)>0\), with no restrictions on its Seiberg-Witten classes, providing a more complicated closed formula. This formula has a very deep understanding in the realm of string theory. In fact, the conjectured formula refines part of a formula proposed by Vafa-Witten, related to \(S\)-duality, and by Dijkgraaf-Park-Schroers present in the physics literature. Amazingly, the unrefined missing part of Vafa-Witten formula was recently recovered by Tanaka-Thomas by studying moduli spaces of Higgs bundles on \(S\). The approach to the conjectures is the following. Via Hirzebruch-Riemann-Roch the invariants are related to certain descendent Donaldson invariants. These are expressed via Mochizuki's formula in terms of Seiberg-Witten invariants of \(S\) and certain tautological integrals over \(S^{[n_1]}\times S^{[n_2]}\). Universality of these integrals reduce the computation to seven universal series, which are entirely determined by their values on \(S=\mathbb{P}^2\) and \(\mathbb{P}^1\times \mathbb{P}^1\). At this point, using Atiyah-Bott localization the first terms of these universal series are computed. The authors present also a list of surfaces where the conjecture has been verified, up to some orders.