A simple extension of Ramanujan-Serre derivative map and some applications (Q6164982): Difference between revisions

If \(E_2\) is the quasimodular form of weight \(2\) and depth \(1\) with constant Fourier coefficient at infinity being \(1\), \textit{D. Zagier} mentions that if \(\sum_{M\mid N}c_M/M=0\), then \(\sum_{M\mid N}c_ME_2(Mz)\) is a modular form of weight \(2\) and level \(N\) [in: From number theory to physics. Lectures of a meeting on number theory and physics held at the Centre de Physique, Les Houches (France), March 7-16, 1989. Berlin: Springer-Verlag. 238--291 (1992; Zbl 0791.11022)]. On the other hand, denoting \(D=\frac{1}{2i\pi}\frac{\partial}{\partial z}\), if \(f\) is a modular form of weight \(k\), then its Serre derivative \(S(f)=D f-\frac{k}{12}fE_2\) is a modular form of weight \(k+2\). These two pieces of information imply that if \(f\) is a modular form of weight \(k\) and level \(N/M_0\), then \[ c_{M_0}(Df)(M_0z)+\frac{k}{12}\sum_{\substack{M\mid N\\ M\neq M_0}}c_ME_2(Mz)f(M_0z) \] is modular of weight \(k+2\) and level \(N\). This is the starting point of this article, which introduces the differential operator \(\vartheta_{k;(a,b)}\) (generalized Serre derivative) of the space \(M_k(M,\chi)\) of modular forms of weight \(k\), level \(M\), and character \(\chi\) into \(M_{k+2}(\mathop{\mathrm{lcm}}(a,Mb),\chi)\) defined by \[ \vartheta_{k;(a,b)}(f)(z)=b(Df)(bz)-\frac{ka}{12}E_2(az)f(bz). \] The quasimodular forms \(DE_2\) and \(E_2^2\) have depth \(2\), and the non-existence of modular forms of weight \(2\) and level \(1\) implies that their linear combinations are of depth \(2\) or \(0\). Thus, \(DE_2-\frac{1}{12}E_2^2\) is modular. The authors generalize this result by showing that \[ F_{a,b}(z)=\frac{6a}{b}(DE_2)(az)+\frac{6b}{a}(DE_2)(bz)-E_2(az)E_2(bz) \] is a modular form of weight \(4\) and level \(\mathop{\mathrm{lcm}}(a,b)\) for all integers \(a\) and \(b\). The authors use these results to establish linearizations of divisor sums \[ W_{a,b}^{r,s}(n)=\sum_{\substack{(\ell,m)\in\mathbb{Z}_{>0}^2\\ a\ell+bm=n}}\sigma_r(\ell)\sigma_s(m)\qquad \left(\sigma_t(q)=\sum_{d\mid q}d^t\right). \] By applying the generalized Serre derivative to the Eisenstein series \(E_k\) (for \(k\geq 4\)), the authors show that for any integers \(a\) and \(b\), then \[ W_{a,b}^{1,k-1}(n)=\frac{B_k}{2k}\sigma_1(n/a)+\left(\frac{1}{24}-\frac{n}{2ka}\right)\sigma_{k-1}(n/b)-\frac{B_k}{4ak^2}\sum_{f\in\mathcal{B}_{k+2}(a,b)}\alpha_f\widehat{f}(n). \] In this formula, the coefficients \(B_k\) are defined by the exponential generating series \(\sum_kB_kt^k/k!=t/(e^t-1)\), the set \(\mathcal{B}_{k+2}(a,b)\) is a basis of \(M_{k+2}(\mathop{\mathrm{lcm}}(a,b))\), each element \(f\) of this basis has a Fourier expansion whose \(n\)-th coefficient is \(\widehat{f}(n)\), and \(\alpha_f\) is the coordinate of \(\nu_{k;(a,b)}(E_k)\) relative to \(f\). All these elements can be explicitly calculated for each value of \(k\), for example, using the software PARI/GP (\url{https://zbmath.org/software/680}). Similarly, denoting \(\beta_f\) the coordinate of \(F_{a,b}\) relative to any form \(f\) from a basis of \(\mathcal{B}_{4}(a,b)\), the authors show that \[ W_{a,b}^{1,1}(n)=\frac{1}{24}\left(1-\frac{6n}{b}\right)\sigma_1\left(\frac{n}{a}\right)+\frac{1}{24}\left(1-\frac{6n}{a}\right)\sigma_1\left(\frac{n}{b}\right)-\frac{1}{576}\sum_{f\in\mathcal{B}_{4}(a,b)}\beta_f\widehat{f}(n). \] Finally, formulas are given for the linearizations of \[ \sum_{a\ell+bm=n}\ell\sigma_1(\ell)\sigma_{k-1}(m) \] using the modularity of \(\nu_{k+2;(a,1)}(F_{a,b})\) and then of \[ \sum_{a\ell+bm=n}\ell\sigma_1(\ell)\sigma_{1}(m) \] using the modularity of \(\nu_{4;(a,1)}(F_{a,b})\).