Singular equivariant asymptotics and the momentum map. Residue formulae in equivariant cohomology (Q326640)

Let \(\mathbf{X}\) be a symplectic manifold, \(G\) the Lie group acting on \(M\) as Hamiltonian action. Let \(\mathbb{J}:\mathbf{X}\to \mathfrak{g}^\ast\), where \(\mathfrak{g}\) is the Lie algebra of \(G\), be the corresponding momentum map. If \(\mathbf{X}\) is compact and 0 a regular value of \(\mathbb{J}\), then the cohomology of the Marsden-Weinstein reduced space \(\mathbf{X}_{\mathrm{red}}=\mathbb{J}^{-1}(0)/G\) was expressed in terms of the equivariant cohomology of \(\mathbf{X}\) via a certain residue formula [\textit{L. C. Jeffrey} and \textit{F. C. Kirwan}, Topology 34, No. 2, 291--327 (1995; Zbl 0833.55009)]. The residue formula when 0 is not a regular value is also derived if \(\mathbf{X}\) is a nonsingular projective variety [\textit{L. C. Jeffrey} et al., Transform. Groups 8, No. 3, 217--259 (2003; Zbl 1080.14537)]. In this paper, as an extension of these results, a residue formula in the case \(\mathbf{X}=T^\ast M\), \(M\) a Riemannian manifold and \(G\) acts as isometries, is proved via desingularization of \(\mathbf{X}_{\mathrm{red}}\) at 0 (\S10, Theorem 7). For this purpose, the equivariant cohomology \(H_G^\ast(\mathbf{X})\) and cohomology of \(\mathbf{X}_{\mathrm{red}}\) are related by a Duistermaat-Heckman type integral [\textit{J. J. Duistermaat} and \textit{G. J. Heckman}, Invent. Math. 69, 259--268 (1982; Zbl 0503.58015)] \[ X\to L_\alpha(X)=\int_{\mathbf X}e^{iJX}\alpha, \, X\in\mathfrak{g}, \, \alpha\in\Lambda(\mathbf{X}), \] where \(J\) is the almost complex structure of \(\mathbf{X}\) compatible with the symplectic structure. Applying the localization formula in equivariant cohomology (\S2, Theorem 1. [\textit{N. Berline} et al., Heat kernels and Dirac operators. Berlin etc.: Springer-Verlag (1992; Zbl 0744.58001)], Theorem 7.13), it is shown that \[ L_{e^{-i\omega}\rho(Y)}=\sum_{F\in\mathcal{F}}u_F(Y). \] Here, \(\mathcal{F}\) is a connected components of fixed points of a \(T\)-action, where \(T\) is a fixed maximal torus of \(G\) (\S2, Corollary 2. The explicit form of \(u_F\) is given as (5)). Let \(\Phi(Y)=\prod_{\gamma\in\Delta_+}\gamma(Y)\), \(\Delta_+\) be the set of positive roots, and set \[ \mathcal{F}_t(u_F\Phi^k)=U_F^{\Phi^k}, \] where \(\mathcal{F}\) is the (\(\mathfrak{g}\)-)Fourier transform. The residue treated in this paper is defined by \[ \mathrm{Res}^{\Lambda,\zeta}(u_F\Phi^k)=\lim_{t\to 0}U^{\Phi^k}_F(t\zeta), \] (\S2, Definition 1). Then it is shown that to understand the residue, it is necessary to study the asymptotic behavior of \[ I_\zeta(\mu)=\int_{\mathfrak g}\int_{\mathbf X}e^{i\psi_\zeta(\eta,X)/\mu}a(\eta,X)d\eta d\mathbf{X}. \] Here, \(\psi_\zeta(\eta,X)=\mathbb{J}(\eta)(X)-\zeta(X), \zeta\in\mathfrak{g}^\ast\). If the critical set of the phase function is smooth, the asymptotic behavior of \(I_\zeta\) is computed by the stationary phase theorem (\S3, Theorem A. Proved in Appendix A, cf. [\textit{M. Combescure} et al., Commun. Math. Phys. 202, No. 2, 463--480 (1999; Zbl 0939.58031)]). To get the asymptotic expansion of \(I_\zeta\), a desingularization of the critical set is performed following [\textit{K. Kawakubo}, The theory of transformation groups. Oxford etc.: Oxford University Press (1991; Zbl 0744.57001)]. It is also shown that a repeating use of the stationary phase method is allowed. These are the main parts of the paper. After these preparations, \[ I(\mu)=(2\pi\mu)^\kappa L_0+O(\mu^{k+1}(\log\mu)^{\Lambda-1}) \] is shown with an explicit form of the leading coefficient \(L_0\) (\S9. Theorem 6. The explicit form of \(L_0\) is given as (56)). The author says that Theorem 6 is the main theorem of the paper. Let \(\tilde{\mathcal{K}}:\Lambda^{\ast+\kappa}(T^\ast M)\to\Lambda^{\ast-d}(\mathrm{Reg}\mathbf{X}_{\mathrm{red}})\) be the composition of the maps \[ \begin{aligned} \frac{r}{\mathrm{vol}\mathcal{O}}:\lambda^{\ast+\kappa}(T^\ast M)\to \Lambda^\ast(\mathrm{Reg}\mathbb{J}^{-1}(0)), \\ \pi_\ast:\Lambda^\ast(\mathrm{Reg}\mathbb{J}^{-1}(0))\to \Lambda^{\ast-d}(\mathrm{Reg}\mathbf{X}_{\mathrm{red}}).\end{aligned} \] Let \(\rho\in H^\ast_G(T^\ast M)\) be \(\rho(X)=\alpha+D\nu(X)\), where \(\alpha\) is a closed basic differential form on \(T^\ast M\) of compact support and \(\nu\) an equivariant differential form of compact support, and the dimension of a principal \(G\)-orbit is \(d=\dim\mathfrak{g}\). Then from Theorem 6, \[ (2\pi)^d\int_{\mathrm{Reg}\mathbf{X}_{\mathrm{red}}}\tilde{\mathcal{K}}(e^{-i\bar{\omega}}\alpha)=\frac{|H|}{|W|\mathrm{vol} T}\mathrm{Res}\bigl(\Phi^2\sum_{F\in\mathcal{F}}u_F\bigr), \] is derived (\S10. Theorem 7). Here, \(W\) is the Weyl group and \(H\) is the principal isotropy group of the \(G\)-action. This is the final result of the paper. The author remarks that the assumption on \(\rho\) is not so restrictive, because if \(G\) acts locally freely, then \(H_G^\ast(\mathbf{X})\cong H^\ast(\mathbf{X}/G)\).