1

Posting this question here as the community has addressed the Riemann Zeta function with questions such as, i.e. with Simpler zeta zeros and Zeta function zeros and analytic continuation.

Define the following Dirichlet series, i.e. partial sum of the Riemann Zeta function as \begin{equation} \zeta_{n}(s)=\zeta_{n}(\sigma+it)=\sum_{k=1}^{n}\frac{1}{k^{\sigma+it}} \end{equation}

I take it known that in the critical strip that using the Euler-Maclaurin formula, Hardy and Littlewood showed that for a large $n\in\mathbb{N}$: \begin{align} \zeta(s)=\zeta_{n}(s)-\frac{n^{1-s}}{1-s}+O(n^{-\sigma}) \end{align}

The expression $\frac{n^{1-s}}{1-s}$ describes the specific logarithmic spiral that the polylines of $\zeta_{n}(s)$ approximate, but always circumnavigates the origin. It follows that if and only if $\sigma+it$ is a non-trivial zeta zero, the polylines formed from $\zeta_{n}(\sigma+it)$ spiral the origin, this has been noted elsewhere, see the paper "A Geometric Perspective on the Riemann Zeta Function’s Partial Sums" by Carl Erickson (https://dl.icdst.org/pdfs/files3/46b9a82eac445746e69d8e6d76c17749.pdf) which has been referenced by members of these boards.

The Riemann Zeta function can be expressed by isolating the real and complex parts of the function as \begin{equation} \zeta(\sigma+it)=\sum_{n=1}^{\infty}\frac{\cos(t\ln(n))}{n^\sigma}-i\sum_{n=1}^{\infty}\frac{\sin(t\ln(n))}{n^\sigma} \end{equation}

Take $\theta+it$ as some non-trivial zeta zero, then the modulus (distance from the origin, $\zeta(\theta+it)$, and $\zeta_{n}(\theta+it)$) can be computed as:

\begin{align} |\zeta_{n}(\sigma+it)|&=\sqrt{\left(\sum_{1}^{n}\frac{\cos(t\ln(n))}{n^\sigma}-0\right)^2+\left(\sum_{1}^{n}\frac{\sin(t\ln(n))}{n^\sigma}-0\right)^2}\\ &=\sqrt{\sum_{j=1}^{n}\sum_{k=1}^{n}\frac{\cos(t\ln(j))\cos(t\ln(k))}{(jk)^\sigma}+\sum_{j=1}^{n}\sum_{k=1}^{n}\frac{\sin(t\ln(j))\sin(t\ln(k))}{(jk)^\sigma}}\\ \normalsize &=\sqrt{\sum_{j=1}^{n}\sum_{k=1}^{n}\frac{\cos(t(\ln(j)-\ln(k)))}{(jk)^\sigma}}\\ &=\sqrt{\sum_{j=1}^{n}\sum_{k=1}^{n}\frac{\cos(t\ln(\frac{j}{k}))}{(jk)^\sigma}} \end{align}

An similar/ alternating version of this last equation is noted in "Asymptotic formulas for harmonic series in terms of anon-trivial zero on the critical line" by Arhur Kawalec (https://arxiv.org/pdf/1911.06115.pdf).

As a side note before my question, this is a beautiful equation, it describes something unique to the zeta function by combining the natural logarithm, trigonometry, and the set of rationals (i.e. the collection of $\frac{i}{j}$).

My first question, have I made any mistakes yet?

Take $\theta+it$ as some non-trivial zeta zero, for any $\varepsilon>0$, an $n\in\mathbb{N}$ can be chosen large enough such that the logarithmic spiral above has the same modulus (within $\varepsilon$) of $|\zeta_{n}(\sigma+it)|$, for any and all integers greater than $n$. [Given the curvature of said logarithmic spiral in the critical strip,] I believe that I can now claim that for any $\varepsilon>0$, $||\zeta_{n+1}(\sigma+it)|-|\zeta_{n}(\sigma+it)||<\varepsilon$, for any integers greater than $n$. Note that this claim cannot be made if $\theta+it$ is not a zeta zero as the partial sums of $\zeta_{n}$ won't spiral the origin, see the two images here to help visualize geometrically what is happening:

log spiral approximating (and not approximating) the partial sums of the zeta function

My second question: is this previous assertion correct?

jwren
  • 29

1 Answers1

0

Note that if $n \ge 2\pi C|t|, C>1$ so with $s=\sigma+it, |t| \ge 1$ $$\begin{align} \zeta(s)=\zeta_{n}(s)-\frac{n^{1-s}}{1-s}+O_{C, \delta}(n^{-\sigma}) \end{align}$$ holds uniformly in $0< \delta<\sigma <2$ say, then:

$$\zeta_{n+1}(s)-\zeta_{n}(s)=\frac{(n+1)^{1-s}-n^{1-s}}{1-s}+O(n^{-\sigma})$$ But $$\frac{(n+1)^{1-s}-n^{1-s}}{1-s}=\int_n^{n+1}x^{-s}dx$$ hence $|\frac{(n+1)^{1-s}-n^{1-s}}{1-s}|=|\int_n^{n+1}x^{-s}dx|\le \int_n^{n+1}x^{-\sigma}dx \le n^{-\sigma}$ so $$|\zeta_{n+1}(s)-\zeta_{n}(s)| =O(n^{-\sigma})$$. In particular $$||\zeta_{n+1}(\sigma+it)|-|\zeta_{n}(\sigma+it)|| \le|\zeta_{n+1}(s)-\zeta_{n}(s)|=O(n^{-\sigma}) $$ regardless if $\sigma+it$ is a zero or not, so the assertion "Note that this claim cannot be made if $θ+it$ is not a zeta zero" is false

Conrad
  • 27,433
  • Thank you Conrad, yes, that is correct, and not what I was trying to express.

    Geometrically (see image), it seems to me that a distinction can be made though, while both are within $O(n^{-\sigma})$ when $s$ is a zeta zero, and when $s$ is not a zero. Geometrically (& and even through computation), the series $|\zeta_{n}(s)|, |\zeta_{n+1}(s)|, |\zeta_{n+2}(s)| $ is monotonic if and only if $s$ is a zero. Put another way, while both are less than $O(n^{-\sigma})$, if $s$ is a zero then the modulus would be significantly smaller than just $O(n^{-\sigma})$ (after $n$ is large enough).

    – jwren Nov 09 '23 at 17:31
  • why should $|\zeta_n(s)|$ be monotonic? That's not clear to me since regardless of $s$ being zero or not, the amplitude of $\zeta_n(s)$ increases with $n$ so in the long run for fixed $s$, sure $|\zeta_n(s)| \sim n^{1-\sigma} \to \infty$ regardless of what $\zeta(s)$ is but in the short run there is no reason for the difference between two terms to be monotonic as the difference is close to the error $O(n^{-\sigma})$ regardless of again of what $\zeta(s)$ is – Conrad Nov 09 '23 at 17:44
  • Right, what you say about approaching infinity is correct. In the top graph of the image $s$ is a zero, the line connecting the origin to the next $\zeta_{n}$ does approach infinity, but it is monotonic (strictly I believe). This is in contrast to the next graph where, since the circumnavigation isn't around the origin, not a zeta zero, the next $\zeta_{n}$ will always not be monotonic, think about the distance from the origin in this case to the next $n+1^{st}$ point. – jwren Nov 09 '23 at 19:27
  • as noted I highly doubt that monotonicity since $\zeta_n$ is oscillating so the next term moving to $\zeta_{n+1}$ has argument $-t\ log (n+1)$ which can go in any direction; note that a zero where $\zeta_n$ is about $\frac{n^{1-s}}{1-s}$ the modulus is about $n^{1-\sigma}/|1-s|$ and the difference between two consecutive such is within the error of the approximation – Conrad Nov 09 '23 at 19:58
  • I know, again to my original point of the double summation equation being pretty interesting. Go try it out though, when sigma is 1/2, not only is it positive, it is also monotonic, then ideas and theorems around bounded monotonic functions can be used to work out some interesting ideas with these equations. Assuming it could be proved, how would one go about such a formal proof? – jwren Nov 10 '23 at 17:47