The discrete second moment of mixed derivatives of the Riemann zeta function
Benjamin Durkan, Christopher Hughes, Andrew Pearce-Crump · 2026
Read The discrete second moment of mixed derivatives of the Riemann zeta function by Benjamin Durkan, Christopher Hughes, Andrew Pearce-Crump (2026) 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-2601-06292.txt.
Introduction
Benjamin Durkan and 2 co-authors compute a full asymptotic formula in a 2026 number theory paper about the Riemann zeta function. The zeta function sits at the heart of prime number theory. Its derivatives, evaluated at the function's zeros, carry information that researchers average to understand its behavior.
Earlier work by Gonek gave only the leading term for the discrete second moment of mixed derivatives at the zeros. Conrey, Snaith, and Milinovich later added lower-order terms for the first derivative alone. Durkan and colleagues extend the analysis to all derivatives at once. Their result supplies the complete asymptotic expansion, with error terms given both unconditionally and under standard assumptions. The mixed-derivative case closes a gap left by the single-derivative treatments.
Catalog record
| CREATOR | Benjamin Durkan, Christopher Hughes, Andrew Pearce-Crump |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-06292
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The discrete second moment of mixed derivatives of the Riemann zeta function Benjamin Durkan, Christopher Hughes, Andrew Pearce-Crump Abstract . We establish the full asymptotic for the discrete second moment of the Riemann zeta function of mixed derivatives evaluated at the zeta zeros, providing both unconditional and conditional error terms. This was first studied by Gonek, where only the leading order asymptotic was given, later extended by Conrey–Snaith and Milinovich to include the lower order terms for the first derivative. We extend the case of the first derivative to all derivatives. 1. Introduction The central object of this paper is the discrete second moment of mixed derivatives of the Riemann zeta function, given by the sum (1.1) I(\mu,\nu):=I(\mu,\nu;T)=\sum_{0<\gamma\leq T}\zeta^{(\mu)}(\rho)\zeta^{(\nu)}(1-\rho), as T\to\infty , where the sum is over non-trivial zeros of zeta \rho=\beta+i\gamma , and where \zeta^{(\mu)}(s) denotes the \mu^{\text{th}} derivative of the Riemann zeta function. We will establish a full asymptotic expansion for this sum, with a power-saving error term under the Riemann Hypothesis. This type of discrete moment was introduced by Gonek [ 9 ] , who proved for positive integers \mu,\nu , a leading order asymptotic of the form (1.2) \sum_{0<\gamma\leq T}\zeta^{(\mu)}(\rho)\zeta^{(\nu)}(1-\rho)\\ =(-1)^{\mu+\nu}\left(\frac{1}{\mu+\nu+1}-\frac{1}{(\mu+1)(\nu+1)}\right)\frac{T}{2\pi}\left(\log\frac{T}{2\pi}\right)^{\mu+\nu+2}+O\left(T(\log T)^{\mu+\nu+1}\right) as T\to\infty . The special case \mu=\nu , called the discrete second moment of zeta, is given under the Riemann Hypothesis by (1.3) \sum_{0<\gamma\leq T}\left|\zeta^{(\nu)}\left(\frac{1}{2}+i\gamma\right)\right|^{2}=\frac{\nu^{2}}{(2\nu+1)(\nu+1)^{2}}\frac{T}{2\pi}\left(\log\frac{T}{2\pi}\right)^{2\nu+2}+O\left(T(\log T)^{2\nu+1}\right) as T\rightarrow\infty . Conrey and Snaith [ 3 ] conjectured under the Riemann Hypothesis that for \varepsilon>0 arbitrary and L=\log t/2\pi , \displaystyle\sum_{0<\gamma\leq T}|\zeta^{\prime}(\rho)|^{2}=\frac{1}{2\pi}\int_{1}^{T}\Bigg( \displaystyle\frac{1}{12}L^{4}+\frac{2\gamma_{0}}{3}L^{3}+\left(\gamma_{0}^{2}-2\gamma_{1}\right)L^{2}-\left(2\gamma_{0}^{3}+2\gamma_{0}\gamma_{1}+\gamma_{2}\right)L \displaystyle+\left(2\gamma_{0}^{4}+2\gamma_{0}^{2}\gamma_{1}+14\gamma_{1}^{2}+8\gamma_{0}\gamma_{2}+\frac{10\gamma_{3}}{3}\right)\Bigg)\ dt+O\left(T^{1/2+\varepsilon}\right), as T\rightarrow\infty , where the \gamma_{n} are the Stieltjes coefficients from the expansion of \zeta(s) around s=1 , \zeta(s)=\frac{1}{s-1}+\gamma_{0}-\gamma_{1}(s-1)+\frac{\gamma_{2}}{2!}(s-1)^{2}+\cdots+(-1)^{n}\frac{\gamma_{n}}{n!}(s-1)^{n}+\cdots. Milinovich [ 18 ] proved this conjecture under the assumption of the Riemann Hypothesis, writing the asymptotic as (1.4) \sum_{0<\gamma\leq T}\left|\zeta^{\prime}\left(\frac{1}{2}+i\gamma\right)\right|^{2}=\frac{T}{2\pi}P_{4}\left(\log\frac{T}{2\pi}\right)+O\left(T^{\frac{1}{2}+\varepsilon}\right) as T\rightarrow\infty , where P_{4}(x) is a degree four polynomial given by \displaystyle P_{4}(x) \displaystyle=\frac{1}{12}x^{4} \displaystyle+\left(\frac{2\gamma_{0}-1}{3}\right)x^{3} \displaystyle+\left(1-2\gamma_{0}+\gamma_{0}^{2}-2\gamma_{1}\right)x^{2} \displaystyle+\left(-2+4\gamma_{0}-2\gamma_{0}^{2}-2\gamma_{0}^{3}-10\gamma_{0}\gamma_{1}+4\gamma_{1}-\gamma_{2}\right)x \displaystyle+\left(\frac{6+6\gamma_{0}(5\gamma_{1}+4\gamma_{2}-2)+6\gamma_{0}^{2}(\gamma_{0}+\gamma_{0}^{2}+6\gamma_{1}+1)-12\gamma_{1}+42\gamma_{1}^{2}+3\gamma_{2}+10\gamma_{3}}{3}\right) Remark . The equivalence of Milinovich’s result and the conjecture of Conrey and Snaith follows by performing the integral. Some examples of other generalisations of the results discussed above are shifted second moment results [ 8 , 7 ] , higher moment conjectures [ 10 , 12 , 13 , 20 ] , extreme values of derivatives of zeta [ 21 ] , upper and lower bounds [ 17 , 1 ] on moments, and negative moments [ 10 , 6 , 11 , 5 , 2 ] . 2. Statement of Results We generalise the results of both ( 1.2 ) and ( 1.4 ) by proving a full asymptotic expansion for the discrete second moment of mixed derivatives of the Riemann zeta function ( 1.1 ), both unconditionally and conditionally under the Riemann Hypothesis. Theorem 1 . For positive integers \mu,\nu , we have (2.1) \sum_{0<\gamma\leq T}\zeta^{(\mu)}(\rho)\zeta^{(\nu)}(1-\rho)=\frac{T}{2\pi}\mathcal{P}_{\mu,\nu}\left(\log\frac{T}{2\pi}\right)+O\left(Te^{-C\sqrt{\log T}}\right) as T\to\infty , where C is a positive constant and where \mathcal{P}_{\mu,\nu}(x) is the polynomial of degree \mu+\nu+2 given by (2.2) \mathcal{P}_{\mu,\nu}(x)=\sum_{m=0}^{\mu+\nu+2}\sum_{k=0}^{\nu}(-1)^{\nu}\binom{\nu}{k}C_{1}^{(\mu,\nu)}(m,k)x^{m}\\ +\sum_{m=0}^{\mu+\nu+2}\sum_{k=0}^{\mu}(-1)^{\mu}\binom{\mu}{k}\left(C_{1}^{(\nu,\mu)}(m,k)+C_{2}^{(\mu,\nu)}(m,k)\right)x^{m}, where C_{1}^{(\mu,\nu)}(m,k)=\\ \begin{cases}\displaystyle\frac{(-1)^{m}(\nu-k)}{m!}\sum_{j=0}^{\mu+\nu+1-m}(-1)^{\mu+\nu-j}\frac{(\mu+\nu+1-j)!}{(\mu+k+2-j)!}c^{(\mu,k)}_{j}+\\ \displaystyle\qquad+\frac{c^{(\mu,k)}_{\mu+\nu+2-m}}{(k+m-\nu)!}&m\geq\nu-k\\[12.91663pt] \displaystyle\frac{(-1)^{m}(\nu-k)}{m!}\sum_{j=0}^{\mu+k+2}(-1)^{\mu+\nu-j}\frac{(\mu+\nu+1-j)!}{(\mu+k+2-j)!}c^{(\mu,k)}_{j}&m\leq\nu-k-1\end{cases} where c^{(\mu,k)}_{j} are the Laurent series coefficients around s=1 of (2.3) \frac{\zeta^{\prime}}{\zeta}(s)\zeta^{(\mu)}(s)\zeta^{(k)}(s)\frac{1}{s}=\sum_{j=0}^{\infty}c^{(\mu,k)}_{j}(s-1)^{-\mu-k-3+j}, and where C_{2}^{(\mu,\nu)}(m,k)=\\ \begin{cases}\displaystyle\frac{(-1)^{m}(\mu-k+1)}{m!}\sum_{j=0}^{\mu+\nu+1-m}(-1)^{\mu+\nu-j}\frac{(\mu+\nu+1-j)!}{(\nu+k+1-j)!}d_{j}^{(\nu,k)}+\\ \displaystyle\qquad+\frac{d_{\mu+\nu+2-m}^{(\nu,k)}}{(k+m-\mu-1)!}&m\geq\mu-k+1\\[12.91663pt] \displaystyle\frac{(-1)^{m}(\mu-k+1)}{m!}\sum_{j=0}^{\nu+k+1}(-1)^{\mu+\nu-j}\frac{(\mu+\nu+1-j)!}{(\nu+k+1-j)!}d_{j}^{(\nu,k)}&m\leq\mu-k\end{cases} where d_{j}^{(\nu,k)} are the Laurent series coefficients around s=1 of \zeta^{(\nu)}(s)\zeta^{(k)}(s)\frac{1}{s}=\sum_{j=0}^{\infty}d_{j}^{(\nu,k)}(s-1)^{-\nu-k-2+j}. If one assumes the Riemann Hypothesis, the error term may be replaced with O\left(T^{\frac{1}{2}+\varepsilon}\right) for arbitrary \varepsilon>0 . We now state several corollaries of this result, with Corollary 1 clearly following immediately from this theorem and with Corollaries 2 and 3 proved in Section 6 . Corollary 1 . Assume the Riemann Hypothesis. For \nu a positive integer, the discrete second moment of zeta for all derivatives is given by \sum_{0<\gamma\leq T}\left|\zeta^{(\nu)}\left(\frac{1}{2}+i\gamma\right)\right|^{2}\\ =\frac{T}{2\pi}\sum_{m=0}^{2\nu+2}\left(\sum_{k=0}^{\nu}(-1)^{\nu}\binom{\nu}{k}\left(2C_{1}^{(\nu,\nu)}(m,k)+C_{2}^{(\nu,\nu)}(m,k)\right)\right)\left(\log\frac{T}{2\pi}\right)^{m}+O\left(T^{\frac{1}{2}+\varepsilon}\right), where C_{1}^{(\nu,\nu)}(m,k) and C_{2}^{(\nu,\nu)}(m,k) are defined in Theorem 1 . Corollary 2 . Assume the Riemann Hypothesis. In the case \mu=\nu=1 , ( 2.1 ) recovers the polynomial P_{4}(x) in ( 1.4 ). Corollary 3 . For \mu,\nu positive integers, the leading order coefficient of \mathcal{P}_{\mu,\nu}(x) agrees with the leading order given in ( 1.2 ). As we see in Milinovich’s polynomial P_{4}(x) in ( 1.4 ), the coefficients quickly become unwieldy. We give an example of our result for discrete second moment of the second derivative in Appendix A by explicitly writing out the polynomial for the second derivative. Finally, in Appendix B we demonstrate the theorem by plotting the graphs for the first and second derivatives. 3. Brief outline of the proof For c=1+\frac{1}{\log T} and \mathcal{R} the rectangular contour with vertices c+i , c+iT , 1-c+iT , and 1-c+i , we may use Cauchy’s theorem to write \displaystyle I(\mu,\nu) \displaystyle=\sum_{0<\gamma\leq T}\zeta^{(\mu)}(\rho)\zeta^{(\nu)}(1-\rho) \displaystyle=\frac{1}{2\pi i}\oint_{\mathcal{R}}\frac{\zeta^{\prime}}{\zeta}(s)\zeta^{(\mu)}(s)\zeta^{(\nu)}(1-s)\ ds \displaystyle=\frac{1}{2\pi i}\left(\int_{c+i}^{c+iT}+\int_{c+iT}^{1-c+iT}+\int_{1-c+iT}^{1-c+i}+\int_{1-c+i}^{c+i}\right)\frac{\zeta^{\prime}}{\zeta}(s)\zeta^{(\mu)}(s)\zeta^{(\nu)}(1-s)\ ds (3.1) \displaystyle=I_{1}(\mu,\nu)+I_{2}(\mu,\nu)+I_{3}(\mu,\nu)+I_{4}(\mu,\nu). It is immediate that I_{4}(\mu,\nu)=O(1) since the integral has finite length and bounded integrand. Furthermore, as shown by Gonek [ 9 ] , we have I_{2}(\mu,\nu)=O\left(T^{\frac{1}{2}+\varepsilon}\right) with the harmless restriction that T lies a distance \gg 1/\log T from a zero ordinate \gamma . We shall use this restriction without loss of generality throughout our proof. From this it follows that (3.2) I(\mu,\nu)=I_{1}(\mu,\nu)+I_{3}(\mu,\nu)+O\left(T^{\frac{1}{2}+\varepsilon}\right). In Section 4 we deal with the right-hand vertical segment I_{1}(\mu,\nu) . The starting point is by introducing a functional equation which allows us to write \zeta^{(\nu)}(1-s) in terms of \zeta^{(k)}(s) for k between 0 and \nu . As a consequence we may write I_{1}(\mu,\nu) as an integral of an absolutely convergent Dirichlet series. We then use the method of stationary phase to rewrite this integral as the sum \sum_{n_{1}n_{2}n_{3}\leq\frac{T}{2\pi}}\Lambda(n_{1})(\log n_{2})^{\mu}(\log n_{3})^{k}(\log n_{1}n_{2}n_{3})^{\nu-k}, plus a small error that is subsumed by our error term written above. We evaluate this sum without the (\log n_{1}n_{2}n_{3})^{\nu-k} factor through Perron’s formula. This gives the sum as the residue of a Dirichlet series at s=1 , which in turn is a polynomial of degree \mu+\nu+2 . Finally we reinsert the logarithmic term via partial summation. In Section 5 we handle the left-hand segment I_{3}(\mu,\nu) . We use another functional equation to relate the left vertical segment of our integral to the right vertical segment, and then follow similar methods to those used in Section 4 to establish the asymptotics for the left vertical segment. Combining these two contours, together with the error term that we carry through the proof, gives Theorem 1 . Finally, as mentioned in the introduction, in Section 6 we prove some corollaries of our result. 4. The right-hand vertical segment 4.1. Initial manipulations We start by writing I_{1}(\mu,\nu) in a form more amenable to analysis. Recall that I_{1}(\mu,\nu)=\frac{1}{2\pi}\int_{1}^{T}\frac{\zeta^{\prime}}{\zeta}(c+it)\zeta^{(\mu)}(c+it)\zeta^{(\nu)}(1-c-it)\ dt. We apply the functional equation for \zeta^{(\nu)}(1-c-it) so we can write the resulting expression in terms of convergent Dirichlet series. Lemma 1 . [ 14 , Lemma 4] For s=\sigma+it , with \sigma\geq 1 and t\geq 1 we have (4.1) \zeta^{(\nu)}(1-s)=(-1)^{\nu}\chi(1-s)\sum_{k=0}^{\nu}\binom{\nu}{k}\left(\log\frac{t}{2\pi}\right)^{\nu-k}\zeta^{(k)}(s)+O\left(t^{\sigma-\frac{3}{2}}(\log t)^{\nu}\right), where \chi(s) is the factor from the functional equation \zeta(s)=\chi(s)\zeta(1-s) . Substituting ( 4.1 ) into our expression for I_{1}(\mu,\nu) gives I_{1}(\mu,\nu)\\ =\frac{(-1)^{\nu}}{2\pi}\sum_{k=0}^{\nu}\binom{\nu}{k}\int_{1}^{T}\frac{\zeta^{\prime}}{\zeta}(c+it)\zeta^{(\mu)}(c+it)\zeta^{(k)}(c+it)\chi(1-c-it)\left(\log\frac{t}{2\pi}\right)^{\nu-k}\ dt\\ +O\left(T^{\frac{1}{2}+\varepsilon}\right). Each term in the integrand can be expressed as a Dirichlet series since \Re(s)=c>1 , namely \frac{\zeta^{\prime}}{\zeta}(s)=-\sum_{n=1}^{\infty}\frac{\Lambda(n)}{n^{s}}\text{ and }\zeta^{(\mu)}(s)=(-1)^{\mu}\sum_{n=1}^{\infty}\frac{(\log n)^{\mu}}{n^{s}}, where \Lambda(n) is the von Mangoldt function. It follows that, by multiplying the Dirichlet series in the integral above together, we can write \displaystyle\frac{\zeta^{\prime}}{\zeta}(s)\zeta^{(\mu)}(s)\zeta^{(k)}(s) \displaystyle=(-1)^{\mu+k+1}\left(\sum_{n_{1}=1}^{\infty}\frac{\Lambda(n_{1})}{n_{1}^{s}}\right)\left(\sum_{n_{2}=1}^{\infty}\frac{(\log n_{2})^{\mu}}{n_{2}^{s}}\right)\left(\sum_{n_{3}=1}^{\infty}\frac{(\log n_{3})^{k}}{n_{3}^{s}}\right) \displaystyle=\sum_{n=1}^{\infty}\frac{A_{n}^{(\mu,k)}}{n^{s}} where (4.2) A_{n}^{(\mu,k)}:=(-1)^{\mu+k+1}\sum_{n_{1}n_{2}n_{3}=n}\Lambda(n_{1})(\log n_{2})^{\mu}(\log n_{3})^{k}. The integral I_{1}(\mu,\nu) can then be written as I_{1}(\mu,\nu)\\ =\frac{(-1)^{\nu}}{2\pi}\sum_{k=0}^{\nu}\binom{\nu}{k}\int_{1}^{T}\chi(1-c-it)\left(\log\frac{t}{2\pi}\right)^{\nu-k}\left(\sum_{n=1}^{\infty}\frac{A_{n}^{(\mu,k)}}{n^{c+it}}\right)\ dt+O\left(T^{\frac{1}{2}}(\log T)^{\nu}\right). We now use a lemma due to Gonek [ 9 , Lemma 5] which comes from