\(L^ 2\) Riemann-Roch theorem for elliptic operators (Q1902008)

The generalization of the classical Riemann-Roch theorem to the case of the general elliptic operators and point singularities was first considered by \textit{M. Gromov} and the author in the paper: ``The Riemann-Roch theorem for elliptic operators'', in Adv. Sov. Math. 16, No. 1, 211-241 (1993; Zbl 0802.58051); the case of distributed singularities is described by \textit{M. Gromov} and the author in `The Riemann-Roch theorem for elliptic operators and solvability of elliptic equations with additional conditions on compact subsets', Preprint ETH Zürich (1993); see also Invent. Math. 117, No. 1, 165-180 (1994; Zbl 0822.58046). This paper is a continuation and extension of these papers where only compact or similar to compact situations were considered. Namely, the present generalization is the \(L^2\)-Riemann-Roch theorem on regular coverings of compact manifolds, i.e. manifolds with a free action of a discrete group \(\Gamma\) such that \(X/\Gamma\) is a compact manifold without boundary. Since \(X\) is allowed to be non-compact (this is the case when \(\Gamma\) is infinite), some conditions at infinity are needed. They are provided by imposing the condition of finiteness of the \(L^2\)-norm (or an appropriate uniform Sobolev norm). Still the corresponding spaces of solutions can be infinite-dimensional but they prove to be finite-dimensional in the von Neumann sense. This means that they have finite \(\Gamma\)-dimensions. Here, the \(\Gamma\)-dimension is a function with values in \([0, \infty]\) which is defined on so called Hilbert \(\Gamma\)-modules which are just Hilbert spaces with an unitary action of \(\Gamma\) such that they are \(\Gamma\)-invariant subspaces in the Hilbert tensor products \(L^2\Gamma\times {\mathcal H}\), where \(L^2\Gamma\) is the Hilbert space of all square-integrable functions on \(\Gamma\) (with respect to the canonical discrete measure) and \(\mathcal H\) is an arbitrary Hilbert space. Given an elliptic \(\Gamma\)-invariant differential operator \(A\) on \(X\), the permitted singularities can be situated on a \(\Gamma\)-invariant closed nowhere dense set \(D^+\subset X\) and are described in terms of a given \(\Gamma\)-invariant distribution space \(L^+\) such that \(\text{supp } f\subset D^+\) for all \(f\in L^+\), \(L^+\) is a subspace in a uniform negative Sobolev space and \(\dim_\Gamma L^+< \infty\) where \(\dim_\Gamma\) is the von Neumann \(\Gamma\)-dimension. Namely, the permitted singularities are singularities of solutions \(u\) of the equation \(Au= 0\) defined on \(X- D^+\) such that \(u\) can be extended to a distribution \(\widehat u\) on \(X\) with \(A\widehat u\in L^+\). Also \(u\) should be in \(L^2\) in a generalized sense which can be formulated e.g. by saying that the extension \(\widehat u\) is in a uniform negative Sobolev space. Similarly the required orthogonality conditions say that \(u\) should be orthogonal to a \(\Gamma\)-invariant distribution space \(L^-\) such that \(L^-\) belongs to a uniform negative Sobolev space, \(\dim_\Gamma L^-< \infty\) and all elements of \(L^-\) are supported in a \(\Gamma\)-invariant closed nowhere dense set \(D^-\) such that \(D^+\cap D^-= \emptyset\). All these data are encoded into a notion of a rigged divisor which is a tuple \(\mu= (D^+, L^+; D^-, L^-)\). Denote the corresponding space of solutions with permitted singularities and required orthogonality conditions by \(L(\mu, A)\); denote also \(r(\mu, A)= \dim_\Gamma L(\mu, A)\). The symmetry of the description of singularities and orthogonality conditions allows us to interchange the pairs \((D^+, L^+)\) and \((D^-, L^-)\) to form the inverse divisor \(\mu^{- 1}= (D^-, L^-; D^+, L^+)\) which is naturally associated with the adjoint operator \(A^*\). The main result of this paper is the following Riemann-Roch type formula: \[ r(\mu, A)= \text{ind}_\Gamma A+ \deg_A(\mu)+ r(\mu^{- 1}, A^*). \] Here \(\text{ind}_\Gamma A= \dim_\Gamma\text{Ker }A- \dim_\Gamma\text{Ker } A^*\) is the \(\Gamma\)-index of Atiyah, \(\deg_A(\mu)\) is a number which is expressed in terms of \(\Gamma\)-dimensions \(\ell^\pm= \dim_\Gamma L^\pm\) and \(\Gamma\)-dimensions \(\widetilde\ell^\pm= \dim_\Gamma \widetilde L^\pm\) of ``secondary spaces'' \(\widetilde L^\pm\) as follows \[ \deg_A(\mu)= (\ell^+- \widetilde\ell^+)- (\ell^-- \widetilde\ell^-); \] The spaces \(\widetilde L^\pm\) are defined as follows \[ \begin{aligned} \widetilde L^+ & = \{u\mid \text{supp } u\subset D^+,\;u\in W^{- \infty},\;Au\in L^+\},\\ \widetilde L^- & = \{v\mid \text{supp } v\subset D^-,\;v\in W^{- \infty},\;A^* v\in L^-\},\end{aligned} \] where \(W^{- \infty}= \bigcup_{s\in Z} W^s\), \(W^s\) is the uniform Sobolev space on \(X\). Some applications of this formula are given: the existence of non-trivial \(L^2\)-solutions of the Dirac equation and similar equations which provide spaces where discrete series of representations of semi-simple Lie groups can be constructed.