Modular functions and the uniform distribution of CM points (Q818547)

These two articles under review (Duke and B/J/O (= J. H. Bruinier, P. Jenkins, K. Ono, Hilbert class polynomials and traces of singular moduli. ibid. 334, No. 2, 373--393 (2006; Zbl 1091.11011)) have much in common -- in subject matter, in the strength and intrinsic interest of their mathematics and, not least, in clarity of organization and writing style. Both begin with properties of Klein's familiar invariant: \(j(z)=q^{-1}+744+ \sum^\infty_{n=1}a_n q^n\). Here, \(q=\exp(2\pi iz)\), \(\text{Im}\,z>0\) and the \(a_n\) are in \(\mathbb Z\). These properties concern the values of \(j(z)\) for imaginary quadratic \(z\) such that \(\text{Im}\,z>0\). Let \(-d\) be a negative fundamental discriminant, and put \(z_d=i\sqrt d/2\), if \(d\equiv 0\pmod 4;=(-1+i \sqrt d)/2\), if \(d\equiv 3\pmod 4)\). Then \(j(z_d)\) is an algebraic integer of degree \(h(-d)\), the class number of \(\mathbb Q(z_d)\), so the trace of \(j(z_d)\) (Duke denotes this \(\text{Tr}\,j(z_d)\); B/J/O denote it simply \(\text{Tr}(d))\) is an ordinary integer. There is the relatively easy (acording to Duke) asymptotic estimate \[ \text{Tr}\,j(z_d)=(-1)^de^{\pi\sqrt d}+O (e^{\alpha\pi\sqrt d})\tag{*} \] for any \(\alpha>1/2\), as \(d\to\infty\), with \(-d\) a fundamental discriminant. The main result in both papers deals with the trace of \(f(z_d)\), for \(f\in\mathbb{C}[j]\). In B/J/O \(f\) is restricted to \(I_m(j)\), where \(I_m\) is the unique polynomial of degree \(m\) such that \(\{j(z)-744\}/T_m=m^{-1}I_m(j(z))\), and \(T_m\) \((m\in \mathbb Z^+)\) is the usual weight 0 Hecke operator on \(\Gamma(1)\). Zagier has defined the trace in this more general setting by appealing to the theory of reduced positive-definite integral quadratic forms and their associated complex multiplication (CM) points. The main result of B/J/O is Theorem 1.2, which presents an exact formula for the trace of \(I_m (j(z_d))\) as an infinite sum involving generalized Kloosterman sums (half-integral weights) and half-integral index Bessel functions. The (somewhat surprising) exact formula is, of course, reminiscent of Rademacher's formula for the partition function \(p(n)\), the formulae of \textit{H. Rademacher} and \textit{H. S. Zuckerman} for general modular forms (holomorphic in the upper half-plane) of negative weight, on \(\Gamma(1)\) [Ann. Math. (2) 39, 433--462 (1938; Zbl 0019.02201)] and a number of extensions of this work by Rademacher, Zuckerman, Lehner and others in the Rademacher school, all of these results, obtained by the circle method. It is worth noting that Petersson derived these same results (more or less) by completely different methods involving Poincaré series (and the Riemann-Roch theorem). The proof of Theorem 1.2 probably could be carried out by an application of the circle method, but the authors follow the more interesting (and more natural, in my opinion) Petersson approach, introducing appropriate Poincaré series holomorphic in the upper half-plane. (It seems possible that Theorem 1.2 can be generalized to arbitrary \(f\in\mathbb{C}[j]\), but most likely with difficulty.) On the basis of this exact formula the authors were able to predict a far more precise version of the asymptotic formula (*). Indeed, after they informed Duke of their findings (including some relevant exact calculations), the latter proved his Theorem 1 (the main result of this paper). This is both a generalization of (*), in which \(j\) is replaced by an arbitrary \(f\in\mathbb{C}[j]\), and an impressively exact improvement of (*) in this general context. A fair notion of the strength of this improvement is the corollary obtained by applying Theorem 1 to \(f=j\) itself: \[ \frac{1}{h(-d)}\left\{\text{Tr}\, j(z_d)-\sum_{0<c<2 \sqrt d}\frac 12 S_d(c)\exp\left(\frac{4\pi\sqrt d} {c}\right)\right\}\to 720, \tag{**} \] as \(d\to\infty\), with \(-d\) a fundamental discriminant. In (**), \(S_c(d)=\sum_{x^2\equiv-d(c)}\exp (2x/c)\), an exponential sum related to Kloosterman sums. (See Proposition 5 of Duke's paper.) According to the author, the proof of Theorem 1 is ``based directly on methods used [by him] in [Invent. Math. 79, 1--56 (1995; Zbl 0838.11058)] to establish the uniform distribution of CM points''. These methods, in turn, are founded in part upon results of Iwaniec, Zagier and Duke/Friedlander/Iwaniec.