The moduli space of stable vector bundles over a real algebraic curve (Q2268150)

The moduli spaces of vector bundles on a complex projective algebraic curve (or compact Riemann surface) have been extensively studied for more than half a century. The corresponding problem for real algebraic curves has however received little attention. The purpose of this paper is to lay some foundations for the theory of vector bundles on a real algebraic curve, to relate the stability of such bundles to the representation theory of a certain group related to the fundamental group and to discuss, in the spirit of \textit{M. F. Atiyah} and \textit{R. Bott} [Philos. Trans. R. Soc. Lond., A 308, 523--615 (1983; Zbl 0509.14014)], the topology of the corresponding moduli spaces. The viewpoint taken here is that a real algebraic curve is a compact Riemann surface \(X\) equipped with an anti-holomorphic involution \(\sigma\); we write \(X({\mathbb R})\) for the set of fixed points of \(\sigma\). There are then three types of such curves: \(X({\mathbb R})=\emptyset\) (type 0), \(X({\mathbb R})\neq\emptyset\) and \(X\setminus X({\mathbb R})\) not connected (type I) and \(X({\mathbb R})\neq\emptyset\) and \(X\setminus X({\mathbb R})\) connected (type II); the different types have different cell structures. Next, let \(E\) be a holomorphic vector bundle on \(X\) of rank \(n\) equipped with an anti-holomorphic isomorphism \(\widetilde{\sigma}:E\to E\), which is anti-linear on the fibres. The pair \((E,\widetilde{\sigma})\) will be called a real (resp. quaternionic) vector bundle on \(X\) if \(\widetilde{\sigma}^2=Id_E\) (resp. \(-Id_E\)) (Definition 3.3). These bundles can be classified topologically using classical techniques. If \(X\) is of type I or II, the real bundles are classified by the first Stiefel-Whitney classes \(w_1(E_{i,{\mathbb R}})\) of \(E_{\mathbb R}\) on the components \(X_{i,{\mathbb R}}\) of \(X_{\mathbb R}\) and the first Chern class \(c_1(E)\) of \(E\), subject to the restriction \(c_1(E)\equiv\sum_iw_i(E_{i,{\mathbb R}})\bmod 2\) (Proposition 4.1). If \(X\) is of type 0, one needs only \(c_1(E)\), which must be even (Proposition 4.2). For quaternionic bundles, one again needs only \(c_1(E)\), which must satisfy \(k+n(g-1)\equiv 0\bmod 2\) (Proposition 4.3). All values of these invariants satisfying the stated restrictions occur. The next step is to consider the fundamental group, its unitary representations and flat bundles. For type 0, one can take the usual fundamental group of the smooth space \(X/\sigma\); for the other types, however, where there are fixed points for the action of \(\sigma\), one must take the orbifold fundamental group (also known as the equivariant fundamental group of \((X,\sigma)\)). Denoting the relevant fundamental group by \(\Gamma\), one has in all cases an exact sequence \(0\to\pi_1(X)\to \Gamma\to{\mathbb Z}/2\to0\). The authors give explicit presentations of \(\Gamma\) in section 5 and define real and quaternionic unitary representations of \(\Gamma\) (the target for such representations is the natural semi-direct product of the unitary group \({\mathbb U}_n\) and the Galois group of \({\mathbb C}/{\mathbb R}\) (which is of course isomorphic to \({\mathbb Z}/2\))). One can then identify the equivalence classes of representations (real or quaternionic) with the isomorphism classes of (real or quaternionic) unitary flat vector bundles (Theorem 5.2). One can extend this to bundles which are not flat by using a central extension \(0\to{\mathbb Z}\to\Gamma'\to\Gamma\to0\) as in the complex case. The authors now introduce real and quaternionic connections and obtain a version of the theorem of \textit{M. S. Narasimhan} and \textit{C. S. Seshadri} [Ann. Math. (2) 82, 540--567 (1965; Zbl 0171.04803)] (Theorem 6.1), stating that the moduli space of semistable real (resp. quaternionic) bundles \((E,\widetilde{\sigma})\) of rank \(n\) with \(c_1(E)=k\) is diffeomorphic to the space of equivalence classes of real (resp. quaternionic) unitary representations of \(\Gamma'\) which map the generator of the centre to \(\exp(\pi ik/n)\). From this, one can proceed to discuss the topology of the moduli spaces using the action of a gauge group on an appropriate space of connections (as in the paper of Atiyah and Bott). The authors first compute the first and second homotopy groups of the classifying spaces for each type of real or quaternionic representations in both the variable and fixed determinant cases. They then show that there exist connected moduli spaces of real and quaternionic semistable bundles for each allowable topological type (Theorem 6.6) and compute the first and second homotopy groups of these moduli spaces (Theorem 6.7).