Heegner divisors in generalized Jacobians and traces of singular moduli (Q313950)

This paper shows how two seemingly unrelated results, the assertions of both of which end up having similar nature, are incarnations of a more general phenomenon. The first result is the Gross-Kohnen-Zagier Theorem, stating that the images of Heegner divisors inside Jacobians of modular curves behave like the coefficients of a modular form of weight \(\frac{3}{2}\). The second one is Zagier's result about traces of singular moduli (i.e., the traces of the values of the normalized \(j\)-invariant on CM points) becoming, up to the appropriate principal part, the coefficients of a weakly holomorphic modular form of weight \(\frac{3}{2}\). Both results have been generalized (the latter to other functions and higher levels, the former to other varieties like Shimura curves), but they were mainly still considered as independent phenomena. The paper in question proves an analogue of the Gross-Kohnen-Zagier Theorem in quotients that are finer than the Jacobian (named generalized Jacobians). The resulting modular forms (of weight \(\frac{3}{2}\)) are now weakly holomorphic rather than holomorphic (the principal part vanishes when mapped to the Jacobian itself, whence the difference). Moreover, modular functions produce maps from these generalized Jacobians to the complex numbers, and using these maps the results of Zagier and others about singular moduli can also be proven. In more detail, the Introduction in Section 1 mentions the basic notions and presents the main results. Section 2 introduces the generalized Jacobian \(J_{\mathfrak{m}}(X)\) associated with an effective divisor \(\mathfrak{m}\) on a smooth projective curve \(X\), which can be seen as the quotient of the divisors of degree 0 on \(X\) modulo the (principal) divisors of rational functions on \(X\) whose divisors, roughly speaking, satisfy a certain congruence condition modulo \(\mathfrak{m}\). These generalized Jacobians form an inverse system with respect to the associated effective divisors, all lying over the classical Jacobian \(J(X)\) of \(X\). The kernel of the map \(J_{\mathfrak{m}}(X) \to J(X)\) consists of an additive part with a nice basis and a multiplicative part modulo global constants. A canonical image of line bundles (with specific trivializations and a choice of base point) in \(J_{\mathfrak{m}}(X)\) is also presented. Section 3 skims through the construction of modular curves as the orthogonal Shimura varieties of isotropic even lattices of signature \((1,2)\), with congruence subgroups of \(\mathrm{SL}_{2}(\mathbb{Z})\) as discriminant kernels, and gives some basic information about the associated Weil representations of the metaplectic double cover of \(\mathrm{SL}_{2}(\mathbb{Z})\) and the corresponding spaces of vector-valued modular forms. The Heegner divisors \(Y(d,\varphi)\) are, in this case, sums of classes of positive definite binary quadratic forms of negative discriminant \(-d\) (normalized by the size of the stabilizer), which satisfy some congruence conditions given in terms of the Schwartz function \(\varphi\) on the (finite) discriminant group \(S_{L}\) of the lattice \(L\) with which one works. By choosing a cusp to as the base point, the divisors \(Y(d,\varphi)\) yield normalized divisors \(Z(d,\varphi)\) of degree 0. Note that \(d\) need not be an integer, but a rational number whose denominator is universally bounded in terms of \(L\). Section 4 proves the main result of the paper for degree 0 divisors, which is the modularity of the degree 0 Heegner divisors \(Z(d,\varphi)\) in the generalized Jacobian associated with a cuspidal effective divisor, once the correct principal part (which one may consider as consisting of Heegner divisors with non-negative discriminants) is added. The terms with \(d=0\) are given in terms of the class of the line bundle of modular forms of weight \(-2\). The term for negative \(d\) can be non-trivial only if \(d\) is minus an integral square, where it is described explicitly in terms of appropriate isotropic lines in the lattice \(L\) and meromorphic functions resembling the ones appearing in the basis for the additive part of the kernel of the map \(J_{\mathfrak{m}}(X) \to J(X)\). In fact, the result is proved not in \(J_{\mathfrak{m}}(X)\) but in a quotient of it, denoted \(J_{\mathfrak{m}}^{\mathrm{add}}(X)\), the map from which to \(J(X)\) is just the additive part of the kernel from above (this is, in some sense, dual to replacing \(\mathfrak{m}\) by a smaller divisor). One also gets rid of the dependence on \(\varphi\) by tensoring \(J_{\mathfrak{m}}^{\mathrm{add}}(X)\) with the space dual to \(\mathbb{C}[S_{L}]\). The proof uses, like Borcherds' proof of the Gross-Kohnen-Zagier Theorem and its generalizations, Serre duality and the construction of Borcherds products. The classical Gross-Kohnen-Zagier Theorem is obtained either as the case \(\mathfrak{m}=0\) of that theorem, or after projecting all the coefficients to \(J(X)\), where all those with non-positive indices vanish. Note, however, that while the classical result only requires the weight and divisor of the Borcherds product (its product formula is just used as means of construction), the current theorem with the generalized Jacobian does require a deeper analysis of the formula defining the Borcherds product. Section 5 considers harmonic weak Maaß\ forms of weight 0, with meromorphic principal parts. If \(F\) is such a modular function, having no constant term at any cusp, and \(\mathfrak{m}\) is an effective cuspidal divisor that is strictly larger than the polar divisor of \(F\) at any cusp, then the paper constructs a trace map \(tr_{F}\) from \(J_{\mathfrak{m}}(X)\) to \(\mathbb{C}\), and shows that it factors through \(J_{\mathfrak{m}}^{\mathrm{add}}(X)\). Applying this map to the coefficients of the series from Section 4 thus produces a modularity result for singular moduli and its generalizations (of level 1, but for vector-valued generating series). Indeed, the terms with positive \(d\) reduce to the usual traces (sums of values at points) with values in the dual space of \(\mathbb{C}[S_{L}]\), and the non-positive index terms are also evaluated explicitly (an expression for the 0th term is proven only in the case where the modular curve maps to \(X(1)\), since the proof there uses the weight 12 modular form \(\Delta\)). A scalar-valued version, with higher level, is also given, which encompasses the known results about modularity of generating series of traces of modular functions. Finally, Section 6 investigates some generalization in several directions. The first one is concerned with discarding the condition that the divisors have degree 0. Then a similar modularity phenomenon for the divisors \(Y(d,\varphi)\) is shown, but involving a certain non-holomorphic Eisenstein series of weight \(\frac{3}{2}\). The second direction involves twisting the Heegner divisors with genus characters (which in particular annihilates the 0th coefficient), where modularity is obtained also in this setting, which in particular allows the authors to reproduce and generalize in this manner some known results about twisted traces of singular moduli. Applicability of this method for higher-dimensional Shimura varieties is also mentioned, with some details about the 2-dimensional case of \(X(1) \times X(1)\) being given explicitly.