The level-rank duality for non-abelian theta functions (Q879933)

Let \(C\) be a smooth complex projective curve of genus \(g\geq 1\) and let \(\text{U}(k,d)\) be the moduli space of semistable vector bundles of rank \(k\) and degree \(d\). It is well known that, when \(d=k(g-1)\), the moduli space \(\text{U}(k,k(g-1))\) carries a canonical theta divisor \(\Theta_k\). Then let \(\text{SU}(k,\Lambda)\) be the moduli space of semistable bundles with fixed determinant \(\Lambda\). Even in this case we have a theta divisor \(\vartheta_k\). There exists a ``strange'' level-rank duality between these moduli space given by the tensor product map: \[ \tau: \text{SU}(r,\mathcal{O}) \times \text{U}(k,k(g-1)) \longrightarrow \text{U}(kr,kr(g-1)), \] \[ (E,f) \mapsto E \otimes F. \] The main result of the paper under review is that the induced map \[ \text{SD}:H^0(\text{SU} (r,\mathcal{O}),\vartheta^k_r)^* \rightarrow H^0(\text{U} (k,k(g-1)),\Theta^r_k) \] is an isomorphism. This was conjectured by Beauville and Donagi-Tu in the middle 90s and was proved by Belkale for generic curves. The proof is based on a more extended level-rank duality, which is interesting in its own right, from which the standard level-rank duality follows by restriction.