Cusp forms for character varieties (Q1326537)

The trace formula of Selberg for a cofinite but not cocompact discrete group \(\Gamma\) acting on the upper half plane \({\mathcal H}\) implies that \(N_ c(R) + M(R) \sim \text{vol} (\Gamma \backslash {\mathcal H}) R^ 2/4 \pi\) as \(R \to \infty\). Here \(N_ c(R)\) is the sum of the dimensions of the spaces of cuspidal Maass forms with eigenvalue \(\leq {1 \over 4} + R^ 2\), and \(M(R)\) is a multiple of the logarithm of the scattering matrix integrated along the critical line between the levels \(-R\) and \(R\). The question arises which of the terms \(N_ c(R)\) and \(M(R)\) is more important. For congruence subgroups and trivial character it is known that \(N_ c (R) \sim \text{vol} (\Gamma \backslash {\mathcal H}) R^ 2/4 \pi\). The authors have put forward the conjecture that \(M(R)\) is the main term in the other cases. In this paper they present evidence for this conjecture by considering the principal congruence subgroup \(\Gamma = \Gamma (2)\) with a one- parameter family of unitary characters \(\eta \mapsto \chi_ \eta\). They show that \(N_ c(R) \sim \text{vol} (\Gamma \backslash {\mathcal H}) R^ 2/4 \pi\) holds for at most countably many \(\eta \in (0,1)\) under an assumption concerning the case of the trivial character \(\chi_ 0\); for the other values of \(\eta\) we have \(M(R)>cR^ 2\) with some positive number \(c\). For \(\eta = 0\) each space of cuspidal Maass forms has a decomposition in old forms and new forms. The assumption is that these spaces of new forms have dimension one, or weaker, that dimension one occurs with positive density. The proof is based on various methods. Among these are perturbation methods, present in previous work of the authors [see e.g. Commun. Pure Appl. Math. 38, 853-866 (1985; Zbl 0614.10027)], of \textit{Y. Colin de Verdière} [Ann. Inst. Fourier 33, No. 2, 87-113 (1983; Zbl 0496.58016)], and others. The proof also uses properties of \(L\)-functions attached to automorphic forms, and the theory of old and new forms. For \(\Gamma (2)\) and the family \(\eta \mapsto \chi_ \eta\) there are two special circumstances, discussed below, that make the proof work. The family of characters is defined by \(\chi_ \eta ({1\;2 \choose 0 \;1}) = 1\) and \(\chi_ \eta ({\;1\;0 \choose - 2\;1}) = e^{2 \pi i \eta}\). So two of the three cusps of \(\Gamma(2)\) are regular for general values of \(\eta\), whereas the cusp \(\infty\) stays singular. Much of the proof concerns itself with automorphic forms that are invariant under the reflection in the line \(\text{Re} z= {1 \over 2}\). It turns out that for this combination of regular and singular cusps one can prove that each family \((0,1) \to [{1 \over 4}, \infty) : \eta \mapsto \lambda (\eta)\) of eigenvalues that corresponds to a family of symmetric cusp forms for \((\Gamma (2), \chi_ \eta)\) has a bounded limit as \(\eta \downarrow 0\), that these limiting values form a discrete set and that each limit value corresponds to only a finite number of families. The other crucial point concerns the \(L\)-function that gives information on the ``destruction of cusp forms''. A cuspidal Maass form \(F\) for \(\chi_ 0\) can occur as the limit of an analytic family of cuspidal Maass forms for \(\chi_ \eta\) with \(\eta>0\) if a certain \(L\)-function vanishes in a point determined by the value of the spectral parameter of \(F\). In the present example the value of this \(L\)-function turns out to have an expression in which occur two values of the usual \(L\)-function associated to \(F\) at points on the boundary of absolute convergence of the \(L\)-series. This makes it possible to show that these \(L\)-values do not vanish.