Twisted sheaves and $\mathrm{SU}(r) / \mathbb{Z}_r$ Vafa-Witten theory
Y. Jiang, M. Kool
TL;DR
The paper addresses a mathematical formulation of S-duality between $SU(r)$ and $SU(r)/\mathbb{Z}_r$ Vafa-Witten partition functions on smooth projective surfaces by employing Yoshioka's moduli of twisted sheaves and Chern characters twisted by a rational $B$-field. It develops a rigorous moduli framework with twisted Higgs pairs, establishes symmetric obstruction theories, and defines $\mathsf Z_{w}(q)$ for all Brauer classes, proving independence from choices in the prime rank setting. For $K3$ surfaces, it derives explicit closed forms and demonstrates that the $S$-duality relation holds as a modular transformation, incorporating flux-sum identities on the $K3$ lattice. The results bridge physical S-duality with algebro-geometric constructions, providing a robust mathematical foundation for twisted Vafa-Witten invariants and their modular properties. The work highlights how Brauer-twisted moduli and B-field twists yield modular generating functions tied to Hilbert schemes and discriminant modular forms, with potential implications for broader geometric and physical contexts.
Abstract
The $\mathrm{SU}(r)$ Vafa-Witten partition function, which virtually counts Higgs pairs on a projective surface $S$, was mathematically defined by Tanaka-Thomas. On the Langlands dual side, the first-named author recently introduced virtual counts of Higgs pairs on $μ_r$-gerbes. In this paper, we instead use Yoshioka's moduli spaces of twisted sheaves. Using Chern character twisted by rational $B$-field, we give a new mathematical definition of the $\mathrm{SU}(r) / \mathbb{Z}_r$ Vafa-Witten partition function when $r$ is prime. Our definition uses the period-index theorem of de Jong. $S$-duality, a concept from physics, predicts that the $\mathrm{SU}(r)$ and $\mathrm{SU}(r) / \mathbb{Z}_r$ partitions functions are related by a modular transformation. We turn this into a mathematical conjecture, which we prove for all $K3$ surfaces and prime numbers $r$.
