Rank-level duality and conformal block divisors (Q895546)

The setting in this article fixes a simple complex Lie algebra \({\mathfrak g}\), a Cartan subalgebra \({\mathfrak h}\), a root system \(Q\) and a Killing form, normalized such that the longest root is of length 2. The set of dominant integral weights of level \(\ell\) is denoted \(P_{\ell}({\mathfrak g})\). For an \(n\)-tuple \(\vec{\lambda}\) of weights, the corresponding bundle of conformal blocks on the moduli stack \(\overline{\mathrm M}_{g,n}\) of genus \(g\) curves with \(n\) marked points is denoted \({\mathbb V}_{\vec{\lambda}}({\mathfrak g},\ell)\). Rank-level duality is a duality in the special case \(g=0\) between certain conformal blocks of \({\mathfrak s}{\mathfrak l}(r)\) at level \(s\) and certain conformal blocks of \({\mathfrak s}{\mathfrak l}(s)\) at level \(r\). There exist similar results for the symplectic and odd orthogonal Lie algebras. The goal of the article under review is to relate explicitely the conformal block divisors, i.e. the Chern classes \(c_1({\mathbb V}_{\vec{\lambda}}({\mathfrak g},\ell)) \), on \(\overline{\mathrm M}_{g,n}\) using the rank-level duality isomorphisms. Under some technical conditions on \(\vec{\Lambda}\in P_1({\mathfrak g})^n\), \(\vec{\lambda}\in P_{\ell_1}({\mathfrak g}_1)^n\) and \(\vec{\mu}\in P_{\ell_2}({\mathfrak g}_2)^n\) which are known to be satisfied in the case of conformal embeddings \({\mathfrak s}{\mathfrak l}(r)\oplus{\mathfrak s}{\mathfrak l}(s)\to {\mathfrak s}{\mathfrak l}(rs)\) and \({\mathfrak s}{\mathfrak p}(2r)\oplus{\mathfrak s}{\mathfrak p}(2s)\to {\mathfrak s}{\mathfrak o}(4rs)\), the main rsult of the article (Theorem 1.2) asserts the following relation among conformal block divisors in the Picard group \(\mathrm{Pic}(\overline{\mathrm M}_{0,n})\): \[ c_1({\mathbb V}_{\vec{\lambda}}({\mathfrak g}_1,\ell_1))+ c_1({\mathbb V}_{\vec{\mu}}({\mathfrak g}_2,\ell_2)) = \mathrm{rk}{\mathbb V}_{\vec{\lambda}}({\mathfrak g}_1,\ell_1).\left\{ c_1({\mathbb V}_{\vec{\Lambda}}({\mathfrak g},1))+\sum_{j=1}^{n} n_{\lambda_j,\mu_j}^{\Lambda_j}\psi_j\right\} \] \[ - \sum_{i=2}^{\;lfloor\frac{n}{2}\rfloor}\epsilon_i\left\{\sum_{A\subset\{1,\ldots,n\},|A|=i} b_{A,A^c}[D_{A,A^c}]\right\}, \] where \([D_{A,A^c}]\) denotes the class of the boundary divisor corresponding to the partition \(A\cup A^c=\{1,\ldots,n\}\), \(\epsilon_i=\frac{1}{2}\) if \(i=n/2\) and \(\epsilon_i=1\) otherwise, \(\psi_j\) is the \(j\)th psi class, \(n_{\lambda_j,\mu_j}^{\Lambda_j}\) is an integer related to the conformal embedding (a sum/difference of trace anomalies), and \(b_{A,A^c}\) is another integer (sum of products of ranks of conformal blocks with some \(n_{\lambda,\mu}^{\Lambda}\) as coefficients). The author proposes two proofs of the above formula, one using a geometric approach (vertex algebra techniques) and one using Fakhruddin's Chern class formula.