Index theory on locally homogeneous spaces (Q1176116)

Let \(G\) be a linear, connected, semisimple Lie group, \(\Gamma\) a torsion- free discrete subgroup, having finite covolume in \(G\). Assume \(G\) and \(\Gamma\) satisfy the assumption 2.1 of \textit{A. Borel} and \textit{H. Garland} [Am. J. Math. 105, 309-335 (1983; Zbl 0572.22007)]. Consider a compact subgroup \(H\) of \(G\), a properly supported \(G\)-invariant, elliptic pseudo-differential operator \(\tilde P\) of nonnegative order between sections of homogeneous bundles \({\mathcal W}_ 1\) and \({\mathcal W}_ 2\) over \(H\backslash G\), descending to an elliptic operator \(P: L^ 2({\mathcal W}_ 1)\to L^ 2({\mathcal W}_ 2)\) over \(H\backslash G/\Gamma\), \(P^*\) the Hilbert space adjoint of \(P\). Although \(P\) and \(P^*\) need not be Fredholm but the kernel of \(P\) and \(P^*\) are finite dimensional, and therefore one can define the \(L^ 2\)-index of \(P\) to be \[ \hbox{index} P=\dim(\hbox{kernel }P)-\dim(\hbox{kernel }P^*). \] The \(G\)-equivariant \(K\)-homology \(KK_ G(C_ 0(H\backslash G),\mathbb{C})\) shares several properties with the better known \(K\) homology groups of compact manifold: --- The \(G\)-invariant elliptic pseudodifferential operator \(\tilde P: L^ 2({\mathcal W}_ 1)\to L^ 2({\mathcal W}_ 2)\) defines an element \([\tilde P]\) of \(KK_ G(C_ 0(H\backslash G),\mathbb{C})\). --- Such an element \([\tilde P]\) is determined by the homotopy class (respecting the \(G\)-action) of the principal symbol. --- If \(H_ 1\hookrightarrow H_ 2\) are compact subgroups of \(G\), then the map \(f: H_ 1\backslash G\to H_ 2\backslash G\) induced from the identity map on \(G\) determines a map \(f_ *: KK_ G(C_ 0(H_ 1\backslash G),\mathbb{C})\to KK_ G(C_ 0(H_ 2\backslash G),\mathbb{C})\). The connection between equivariant \(K\) homology of homogeneous spaces and \(L^ 2\) index theory of locally homogeneous space is provided by the following three steps: 1) The analytic index \(Ind_ a(\tilde P)\in K_ 0(C^*(G))\cong KK(\mathbb{C},C^*(G))\) can be recovered from \([\tilde P]\in KK_ G(C_ 0(H\backslash G),\mathbb{C})\). 2) \(\hbox{kernel}(P)=\hbox{kernel}(P_ d)\), \(\hbox{kernel}(P^*)=\hbox{kernel}(P^*_ d)\) of a truely Fredholm operator \[ P_ d:=P\mid_{\hbox{discretespectrum}}: L^ 2_ d({\mathcal W}_ 1/\Gamma)\to L^ 2_ d({\mathcal W}_ 2/\Gamma). \] This Fredholm operator determines a class \([P_ d]\in KK(\mathbb{C},\mathbb{C})\cong\mathbb{Z}\), identifying \([P_ d]\) with \(L^ 2-\hbox{index}(P)\). 3) The discrete spectrum part \(L^ 2_ d(G/\Gamma)\) of \(L^ 2(G/\Gamma)\) defines an element \([\Gamma]\in KK(C^*(G),\mathbb{C})\) and the Kasparov product \(\hbox{Ind}_ a(\tilde P)\otimes_{C^*(G)}[\Gamma]=[P^ d]\in KK(\mathbb{C},\mathbb{C})\). The construction of \([\Gamma]\) and the calculation of this product are the contributions of the authors. Every compact subgroup \(H\) can be included into some maximal compact subgroup \(K\), and \(I_ H: KK_ G(C_ 0(H\backslash G),\mathbb{C})\to K_ 0(C^*(G))\), satisfying \(I_ H([\tilde P])=\hbox{Ind}_ a(\tilde P)\). Then \[ I_ H([\tilde P])=I_ K(f_ *([\tilde P])). \] Thus the technique developed for calculating the \(L^ 2\)-indices on locally symmetric spaces are reduced to calculating \(L^ 2\)-indices on locally homogeneous spaces. From this, the authors deduce an interesting connection with the Langlands conjecture about multiplicities of discrete series representations: Let us denote by \(K\) a maximal compact subgroup of a connected \(G\) and assume that \(G\) has a compact Cartan subgroup \(T\subseteq K\), \(\tilde D_ 1\) is a Dolbeault operator on \(T\backslash G\) with kernel the discrete series \(\pi\), and with \(\hbox{kernel}(\tilde D^*_ 1)=(0)\), \(f: T\backslash G\to K\backslash G\) the induced from identity map on \(G\), \(\tilde D_ 2\) a Dirac operator on \(K\backslash G\) with kernel \(\pi\) and with \(\hbox{kernel}(\tilde D^*_ 2)=(0)\). The authors show that \(\hbox{Ind}_ a(\tilde D_ 1)=\hbox{Ind}_ a(\tilde D_ 2)\). But it is well-known from the William theorem that \(\hbox{mult}(\pi,L^ 2(G/\Gamma))=L^ 2-\hbox{index}(D_ 2)\). Thus one has a proof of the Langlands conjecture'' \(\hbox{mult}(\pi,L^ 2(G/\Gamma))=L^ 2-\hbox{index}(D_ 1)\).