Moment of Derivatives of Quadratic Twists of Modular L L -functions
Zijie Zhou · 2025
Read Moment of Derivatives of Quadratic Twists of Modular L L -functions by Zijie Zhou (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-14680.txt.
Introduction
Zijie Zhou studies averages of derivatives of L-functions, the central objects of analytic number theory, after they are twisted by quadratic characters. Two distinct modular L-functions are twisted at once, and the paper computes an asymptotic formula for the resulting moment. Ian Petrow had reached a similar result earlier, but only by assuming the generalized Riemann hypothesis. Zhou's proof removes that assumption, making the asymptotic unconditional.
Catalog record
| CREATOR | Zijie Zhou |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 95 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-14680
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Moment of Derivatives of Quadratic Twists of Modular L L -functions Zijie Zhou Abstract . We prove an asymptotic for the moment of derivatives of quadratic twists of two distinct modular L L -functions. This was previously known conditionally on GRH by the work of Ian Petrow [ 7 ] . 1. Introduction Moments of L -functions are central objects in the study of analytic number theory. Motivated by the Birch-Swinnerton-Dyer (BSD) conjecture, we want to understand the behavior of the modular L -functions at the central value. Analytic methods have been used successfully to study the behavior of the central value in some families of modular L -functions; one particular interesting family is that of quadratic twists of modular L -functions. According to the BSD conjecture, moments of quadratic twists of modular L -functions at the central value can provide crucial information about the distribution of ranks of associated twists of elliptic curves. Before stating our main result, we introduce some notation and recall some standard facts. Let f be a Hecke new eigenform for the group \Gamma_{0}(q) of even weight \kappa with trivial central character. It has Fourier expansion \displaystyle f(z)=\sum_{n}\lambda_{f}(n)n^{(\kappa-1)/2}\exp(2\pi inz). The L -function associated with f is given by \displaystyle L(s,f):=\sum_{n}\frac{\lambda_{f}(n)}{n^{s}}=\prod_{p}\Bigg(1-\frac{\lambda_{f}(p)}{p^{s}}+\frac{1}{p^{2s}}\Bigg)^{-1}, for \text{Re}(s)>1 . Here, the Hecke eigenvalues of f are all real, and hence f is self-dual. Let d be a fundamental discriminant coprime to q and let \chi_{d}(\cdot)=\mathopen{}\mathclose{{\left(\frac{d}{\cdot}}}\right) be a primitive quadratic character of conductor |d| . Then f\otimes\chi_{d} is a newform on \Gamma(q|d|^{2}) and the twisted L -function is defined by \displaystyle L(s,f\otimes\chi_{d}):=\sum_{n}\frac{\lambda_{f}(n)\chi_{d}(n)}{n^{s}}=\prod_{p\nmid qd}\mathopen{}\mathclose{{\left(1-\frac{\lambda_{f}(p)\chi_{d}(p)}{p^{s}}+\frac{1}{p^{2s}}}}\right)^{-1}\prod_{p\mid q}\mathopen{}\mathclose{{\left(1-\frac{\lambda_{f}(p)\chi_{d}(p)}{p^{s}}}}\right)^{-1}, for \text{Re}(s)>1 . It satisfies the functional equation \Lambda(s,f\otimes\chi_{d})=i^{\kappa}\eta\chi_{d}(-q)\Lambda(1-s,f\otimes\chi_{d}) with \Lambda(s,f\otimes\chi_{d}):=\mathopen{}\mathclose{{\left(\frac{|d|\sqrt{q}}{2\pi}}}\right)^{s}\Gamma\mathopen{}\mathclose{{\left(s+\frac{\kappa-1}{2}}}\right)L(s,f\otimes\chi_{d}). Here, \eta is given by the eigenvalues of the Fricke involution, which is independent of d and always \pm 1 . We also denote the root number by \omega(f\otimes\chi_{d}):=i^{\kappa}\eta\chi_{d}(-q) . When \omega(f\otimes\chi_{d})=1 , an asymptotic formula for the second moment of L(1/2,f\otimes\chi_{d}) was computed assuming the Generalized Riemann hypothesis (GRH) by Soundararajan and Young [ 9 ] , which was proved unconditionally in Li’s recent work [ 6 ] . Here, we apply their techniques to our problem when the sign of the functional equation is -1 and the derivative L^{\prime}(1/2,f\otimes\chi_{d}) is the correct object to study instead. For \text{Re}(s)>1 , the derivative of the twisted L -function is L^{\prime}(s,f\otimes\chi_{d}):=-\sum_{n}\frac{\lambda_{f}(n)\chi_{d}(n)\log(n)}{n^{s}}. It also has the functional equation \Lambda^{\prime}(s,f\otimes\chi_{d})=-i^{\kappa}\eta\chi_{d}(-q)\Lambda^{\prime}(1-s,f\otimes\chi_{d}). Based on Li’s work [ 6 ] , Kumar, Mallesham, Sharma and Singh [ 5 ] proved the asymptotic of second moment of L^{\prime}(1/2,f\otimes\chi_{d}) unconditionally, which was previously shown by Petrow [ 7 , Theorem 1] assuming GRH. In this paper, we study the moment of the derivatives of quadratic twists of two distinct modular L -functions, which was shown by Petrow [ 7 , Theorem 2] assuming GRH. We let \operatornamewithlimits{\sum\nolimits^{*}} denote a sum over squarefree integers and \sideset{}{{}^{d}}{\sum}_{N} denote a dyadic sum over N=2^{\ell} for integer \ell\geq 1 . For convenience, we restrict the modulus to be of the form 8d where d is odd and squarefree. Now we state the main result. Theorem 1.1 . Let f,g be distinct normalized cuspidal Hecke newforms with trivial central character, odd levels q_{1},q_{2} and even weights \kappa_{1},\kappa_{2} (we let Q=q_{1}q_{2} ). Let F be a smooth, nonnegative function compactly supported on [1/2,2] , we have (1.1) \displaystyle\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ \omega(f\otimes\chi_{8d})=-1\\ \omega(g\otimes\chi_{8d})=-1\end{subarray}}L^{\prime}(1/2,f\otimes\chi_{8d})L^{\prime}(1/2,g\otimes\chi_{8d})F\mathopen{}\mathclose{{\left(\frac{8d}{X}}}\right)=C_{0}X(\log X)^{2}+C^{\prime}X\log X+O(X(\log\log X)^{6}). In the above, we have \displaystyle C_{0} \displaystyle=\frac{\check{F}(0)L(1,\text{sym}^{2}f)L(1,f\otimes g)L(1,\text{sym}^{2}g)Z^{*}(0,0)}{2\pi^{2}}, and C^{\prime} is some explicit constant that depends on f,g and F . Here, L(s,f\otimes g) is the Rankin-Selberg convolution of two modular forms f,g and Z^{*}(u,v) is a holomorphic function ( 5.3 ) defined in \text{Re}(u),\text{Re}(v)\geq-1/4+\epsilon , depending on f and g , given by a sum of four absolutely convergent Euler products and uniformly bounded in u,v where it converges, and \check{F} is a Fourier-type transform of F defined as in ( 3.1 ). Remark 1 . The combination of work of Kolyvagin [ 4 ] , Gross and Zagier [ 2 ] , and others, implies that if the analytic rank of an elliptic curve is equal to one, then so is the rank of that elliptic curve. By the modularity theorem, any elliptic curve E over \mathbb{Q} of conductor N , there is a weight 2 , level N Hecke eigenform f with L(s,E)=L(s,f) , where L(s,E) is the Hasse-Weil L -function of the elliptic curve E . Our result then implies that for any fixed elliptic curves E_{1} and E_{2} over \mathbb{Q} such that L(s,E_{1})=L(s,f) and L(s,E_{2})=L(s,g) , there exist infinitely many fundamental discriminants d such that the twists of the elliptic curves E_{1}(d) and E_{2}(d) are exactly of rank 1 . Independently, S. Choi, B. Kim, H. Kim, H. Kim and W. Lee claimed the following in their work [ 1 , Theorem 2.1] \displaystyle\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ \omega(f\otimes\chi_{8d})=-1\\ \omega(g\otimes\chi_{8d})=-1\end{subarray}}L^{\prime}(1/2,f\otimes\chi_{8d})L^{\prime}(1/2,g\otimes\chi_{8d})F\mathopen{}\mathclose{{\left(\frac{8d}{X}}}\right)=C_{0}X(\log X)^{2}+O(X(\log X)^{2}(\log\log X)^{-A}). for any 0<A<2 . However, there appears to be unresolved issues in their proof of [ 1 , Proposition 6.1] . Specifically, the condition on A [ 1 , p.28] in the first range should give the bound of X(\log X)^{2+\epsilon} at the end; and the number of dyadic intervals from the second range [ 1 , p.29] is about \log X which implies that the contribution of the second range is also \ll X(\log X)^{2+\epsilon} after applying Cauchy-Schwarz. Both issues result in a larger error term compared to the main term. We provide some details in the outline below on the differences in our approach compared to previous works. 2. Outline of Proof of Theorem 1.1 Let M=X/(\log X)^{100} , we define \displaystyle\mathcal{A}_{f} \displaystyle\approx\sum_{n\leq M}\frac{\lambda_{f}(n)\chi_{8d}(n)}{\sqrt{n}}W_{f}\mathopen{}\mathclose{{\left(\frac{n}{M}}}\right), \displaystyle\mathcal{B}_{f} \displaystyle\approx\sum_{M<n\leq X}\frac{\lambda_{f}(n)\chi_{8d}(n)}{\sqrt{n}}\Bigg(W_{f}\mathopen{}\mathclose{{\left(\frac{n}{X}}}\right)-W_{f}\mathopen{}\mathclose{{\left(\frac{n}{M}}}\right)\Bigg). where W_{f}(x) is the smooth cut-off function defined as in ( 4.2 ). With the notation above, the approximate functional equation gives us the decomposition \displaystyle\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}L^{\prime}(1/2,f\otimes\chi_{8d})L^{\prime}(1/2,g\otimes\chi_{8d})=\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}\mathcal{A}_{f}\mathcal{A}_{g}+\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}\mathcal{A}_{f}\mathcal{B}_{g}+\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}\mathcal{B}_{f}\mathcal{A}_{g}+\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}\mathcal{B}_{f}\mathcal{B}_{g}. When one of the n_{i} sums is small, we can apply Poisson summation to obtain both diagonal and off-diagonal terms. The diagonal term gives us the main contributions while we need to bound the off-diagonal terms with Lemma 3.5 from [ 6 ] . The main challenge is to bound the tail \displaystyle\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}\mathcal{B}_{f}\mathcal{B}_{g} since that is when both n_{i} sums are large. Our idea behind bounding the tail is to insert a smooth partition of unity for each n_{i} sum which leads us to analyze the following (2.1) \displaystyle\sideset{}{{}^{d}}{\sum}_{N_{1},N_{2}}\operatornamewithlimits{\sum\nolimits^{*}}_{\begin{subarray}{c}(d,2Q)=1\\ d\asymp X\end{subarray}}\operatorname*{\sum\sum}_{n_{1},n_{2}}\frac{\lambda_{f}(n_{1})\lambda_{g}(n_{2})}{\sqrt{n_{1}n_{2}}}\chi_{8d}(n_{1}n_{2})\Bigg(W_{f}\mathopen{}\mathclose{{\left(\frac{n_{1}}{X}}}\right)-W_{f}\mathopen{}\mathclose{{\left(\frac{n_{1}}{M}}}\right)\Bigg)\Bigg(W_{g}\mathopen{}\mathclose{{\left(\frac{n_{2}}{X}}}\right)-W_{g}\mathopen{}\mathclose{{\left(\frac{n_{2}}{M}}}\right)\Bigg)G\mathopen{}\mathclose{{\left(\frac{n_{1}}{N_{1}}}}\right)G\mathopen{}\mathclose{{\left(\frac{n_{2}}{N_{2}}}}\right), where G(x) is a smooth real-valued function that satisfies ( 3.3 ). Next we truncate each dyadic sum N_{i} in three different ranges: N_{i}\leq M,M<N_{i}\leq X and X<N_{i} . In the work of [ 5 ] , the authors applied Cauchy-Schwarz for all of the ranges which caused them to lose an extra power of (\log X)^{2} in the bound for \operatornamewithlimits{\sum\nolimits^{*}}_{d\asymp X}\mathcal{B}_{f}^{2} . But such loss is acceptable given the main term is X(\log X)^{3} in their work. This is illustrated in the table below. \displaystyle\begin{array}[]{|c|c|c|c|}\hline\cr&N_{1}\leq M&M<N_{1}\leq X&X<N_{1}\\ \hline\cr N_{2}\leq M&\text{C.S.}&\text{C.S.}&\text{C.S.}\\ \hline\cr M<N_{2}\leq X&\text{C.S.}&\text{C.S.}&\text{C.S.}\\ \hline\cr X<N_{2}&\text{C.S.}&\text{C.S.}&\text{C.S.}\\ \hline\cr\end{array} . In the work of [ 1 ] , the authors have attempted to apply Cauchy-Schwarz for all of the ranges but with respect to a different partition on the dyadic sums: N_{i}\leq\mathcal{P},\mathcal{P}<N_{i}\leq X and X<N_{i} with \mathcal{P}=X^{{(\log\log X)}^{-1+A/2}} for some 0<A<2 \displaystyle\begin{array}[]{|c|c|c|c|}\hline\cr&N_{1}\leq\mathcal{P}&\mathcal{P}<N_{1}\leq X&X<N_{1}\\ \hline\cr N_{2}\leq\mathcal{P}&\text{C.S.}&\text{C.S.}&\text{C.S.}\\ \hline\cr\mathcal{P}<N_{2}\leq