19
$\begingroup$

While playing around, I encountered the following very curious and cool identity. Consider the exponential integral $\text{Ei}(x)$ and the $n$th nontrivial zero of the Riemann Zeta function $p_n$.

Now, look at the first few imaginary parts of the following function: $$f(x)=\sum_{n=1}^x \text{Ei}(p_n)$$

$$\Im \ \ f(1)=3.13732$$ $$\Im \ \ f(10)=31.3169$$ $$\Im \ \ f(100)=314.097$$ $$\Im \ \ f(1000)=3141.54$$ $$\Im \ \ f(10000)=31415.9$$

As you can see, it is each time adding a digit of pi.

Question: Is this a known result that can be proved easily? Does this pattern even continue?

$\endgroup$
  • 3
    $\begingroup$ hint: consider asymptotic expansion of Ei, you'll see it's not about zeros, you can take $p_n$ to be any growing function, e.g. $p_n = e^n$ $\endgroup$ – nik Sep 14 '16 at 20:27
  • 1
    $\begingroup$ The less 'numerological' version of this is that evidently $\Im \ \ f(n)\sim \pi n$ as $n\to \infty$. Alternatively, you can divide both sides of this by $n$, and you're asserting that $\frac{1}{n} \sum_{k=1}^n \text{Ei}(p_k)\to \pi$ i.e. the average value of $\text{Ei}(p_n)$ converges to $\pi$ as you use more and more zeta zeros. $\endgroup$ – Semiclassical Sep 14 '16 at 21:09
  • 1
    $\begingroup$ As nik says this has very little to with the zeros of the $\zeta$ functions and the same behavior happens for a wide class of sequences $\rho_n$. For example if $\rho_n = 1 + i n^2$ then $\Im\frac{f(1000)}{1000} = 3.1410$ and $\Im\frac{f(1000)}{1000} = 3.14158$ which also converges to $\pi$. $\endgroup$ – Winther Sep 15 '16 at 9:30
14
$\begingroup$

The result found by OP turns out to be quite generic; it holds for a wide range of sequences $\rho_n$ and not just the zeros of the $\zeta$-function. A precise formulation of what is observed is the following:

$$\lim_{N\to\infty} \frac{1}{N}\sum_{n=1}^N \text{Ei}(\rho_n) = \pi i \tag{1}$$

The sum above is the Cesaro mean of the sequence $a_n = \text{Ei}(\rho_n)$. If a sequence $a_n$ converges to $a$ then the Cesaro mean of $a_n$ also converges to $a$ (see e.g. this question) so $(1)$ would follow if we could show that $\lim_{n\to\infty}\text{Ei}(\rho_n) = \pi i$.

This is indeed true and the reason for this is as follows:

  1. We can write $\text{Ei}(z)$ on the form $\text{Ei}(z) = -\Gamma(0,z) + \pi i$ where the incomplete $\Gamma$-function has the asymptotical expansion $\Gamma(0, z) = \frac{e^{-z}}{z} + \mathcal{O}(z^{-2})$. This implies that $\Gamma(0, z_n) \to 0$ for any sequence $z_n = x_n + i y_n$ where $y_n\to \infty$ and where $x_n$ is bounded.
  2. The non-trivial zeros of the $\zeta$-function satisfy the conditions on the sequence above since they all lie in the strip $0<\Re[z]< 1$ in the complex plane and $\lim_{n\to\infty}\Im[\rho_n] = \infty$. This means that $\lim_{n\to\infty}\Gamma(0,\rho_n) = 0$ so $\lim_{n\to\infty} \text{Ei}(\rho_n) = \pi i$ and $(1)$ follows.

Note that the conditions in point 1. above are far from being very restrictive so there are infinitely many sequences for which $(1)$ holds (random examples are $\rho_n = 7 + n^2i$ and $\rho_n = \frac{2n}{1+n} + n\log(n)i$).

$\endgroup$
4
$\begingroup$

This is not a proof, just some tips that can may be leads to a proof

(I really hope someone manage to find a proof for this result)


You can consider the other form for the exponential integral:

$$Ei(z)=\gamma +\ln(x)+\sum_{k=1}^\infty\frac {x^k}{k\cdot k!},$$

where $\gamma$ is the Euler-Mascheroni's constant.

Call the $n$-th non trivial zero

$$p_n=:\frac 12 +i \alpha_n.$$

So you get

$$Ei(p_n)=\gamma +\ln\left(\frac 12 +i \alpha_n\right)+\sum_{k=1}^\infty\frac {\left(\frac 12 +i \alpha_n\right)^k}{k\cdot k!}.$$

Use the principal determination of the complex logarithm: $\ln(x+iy)=\ln(x)+i\arg(x+iy)$, where

$$\arg(x+iy)=\arctan\left(\frac yx\right).$$

So

$$Ei(p_n)=\gamma +\ln\left(\frac 12\right) +i\arctan(2\alpha_n)+\sum_{k=1}^\infty\frac {\left(\frac 12 +i \alpha_n\right)^k}{k\cdot k!}.$$

So

$$\mathfrak I(Ei)(p_n)=\arctan(2\alpha_n)+\sum_{k=1}^\infty\left(\sum_{\substack{m=0 \\ m\equiv 1 (4)}}^{k}\frac {\binom km\alpha_n^m}{2^{k-m}k\cdot k!}-\sum_{\substack{m=0 \\ m\equiv 3 (4)}}^{k}\frac {\binom km\alpha_n^m}{2^{k-m}k\cdot k!}\right),$$

using

$$\binom km=\frac{k!}{m!(k-m)!}$$

you get

$$\mathfrak I(Ei)(p_n)=\arctan(2\alpha_n)+\sum_{k=1}^\infty\left(\sum_{\substack{m=0 \\ m\equiv 1 (4)}}^{k}\frac {\alpha_n^m}{2^{k-m}k\cdot m!(k-m)!}-\sum_{\substack{m=0 \\ m\equiv 3 (4)}}^{k}\frac {\alpha_n^m}{2^{k-m}k\cdot m!(k-m)!}\right).$$

The key is to use now the approximation:

$$\alpha_n\approx \frac{2\pi n}{\ln(n)}$$

to put $\pi$ in the place.

To conclude, you have to prove that

$$\lim_{x\to \infty}\frac 1x\sum_{n=1}^x\mathfrak I(Ei)(p_n)=\pi.$$

For the first term, you can say that

$$\frac 1x\sum_{n=1}^x \arctan(2\alpha_n)=\frac 1x\left(\text{a finite number of terms}\right)+\frac 1x(x-K)\left(\text{something close to $\frac\pi2$}\right),$$

so

$$\lim_{x\to \infty}\frac 1x\sum_{n=1}^x \arctan(2\alpha_n)=\frac\pi 2.$$

Now, you need to deal with the second term, you can use Stirling formula

$$y!\sim \sqrt{2\pi y}\frac{y^y}{e^y}$$

to get

$$\frac {\alpha_n^m}{2^{k-m}k\cdot m!(k-m)!}=\frac{(2\pi n)^m e^m e^{k-m}}{\ln(n)^m 2^{k-m}k\sqrt{2\pi m}m^m \sqrt{2\pi(k-m)}(k-m)^{k-m}}$$

$$=k\left(\frac{2\pi n}{m\ln(n)}\right)^m\left(\frac e{2(k-m)}\right)^{k-m}\sqrt{2\pi k}.$$

$\endgroup$
  • $\begingroup$ "The key is to use now the approximation ... to put $\pi$ in place". I don't think this is key as we find the same kind of result even if $\alpha_n$ has no $\pi$ in it (e.g. if we take $\alpha_n = \frac{7n}{\log^2(n)}$). $\endgroup$ – Winther Sep 15 '16 at 9:56
  • $\begingroup$ Sorry if I'm missing something here, but I don't see how you would end up with the same result... $\endgroup$ – E. Joseph Sep 15 '16 at 14:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.