I'm currently trying to evaluate the following limit: $$ \lambda=\lim_{z\to\infty}{\left[2^z-\left(\frac{4}{3}\right)^z-\zeta{(\zeta{(z)})}\right]} $$ A look at numerical approximations suggests, that $\lambda=1-\gamma$, but I have no idea how to tackle the iterated zeta function. I know that $\zeta(x)$ is asymptotical to $\frac{1}{1-x}$ when $x\to 1$, but this seems not to help here.

Any help will be highly appreciated!

  • 1
    $\begingroup$ If you let $z$ approach $\infty$ in an arbitrary manner, the limit doesn't exist. Do you assume that $z$ is a positive real number, or maybe that $\operatorname{Re} z \to +\infty$? $\endgroup$ – Daniel Fischer Aug 18 '15 at 17:57

Assuming we're looking at the limit for $\operatorname{Re} z \to +\infty$, we have $\zeta(z) \to 1$, so the Laurent expansion

$$\zeta(s) = \frac{1}{s-1} + \gamma + O(s-1)\tag{1}$$

is relevant. We thus need to look at $\dfrac{1}{\zeta(z)-1}$ for large $\operatorname{Re} z$.

For $\operatorname{Re} z > 1$ and $k \geqslant 2$, define

$$R_k(z) = \sum_{n = k}^\infty \frac{1}{n^z}.$$

We will need bounds for $\lvert R_k(z)\rvert$: Letting $\sigma = \operatorname{Re} z$, we have

\begin{align} \lvert R_k(z)\rvert &\leqslant \sum_{n = k}^\infty \frac{1}{n^\sigma}\\ &< \frac{1}{k^\sigma} + \int_k^\infty \frac{dt}{t^\sigma}\\ &= \frac{1}{k^\sigma} + \frac{1}{(\sigma-1)k^{\sigma-1}}\\ &= \frac{1}{k^\sigma}\biggl(1 + \frac{k}{\sigma-1}\biggr). \end{align}

For $\lvert w\rvert \leqslant \frac{1}{2}$ we have

$$\biggl\lvert \frac{1}{1+w} - (1-w)\biggr\rvert = \frac{\lvert w\rvert^2}{\lvert 1+w\rvert} \leqslant 2\lvert w\rvert^2,$$

so with $\dfrac{1}{\zeta(z)-1} = \dfrac{2^z}{1 + 2^zR_3(z)}$ we obtain that

$$\biggl\lvert \frac{1}{\zeta(z)-1} - 2^z(1-2^zR_3(z))\biggr\rvert \leqslant 2\lvert 2^z\rvert\cdot\lvert 2^zR_3(z)\rvert^2 \leqslant 2\frac{2^{3\sigma}}{3^{2\sigma}}\biggl(1+\frac{3}{\sigma-1}\biggr)^2 \leqslant 8\cdot \biggl(\frac{8}{9}\biggr)^\sigma$$

for $\sigma = \operatorname{Re} z \geqslant 4$. Hence, using $R_3(z) = 3^{-z} + 4^{-z} + R_5(z)$,

\begin{align} \zeta(\zeta(z)) &= \frac{1}{\zeta(z)-1} + \gamma + O(\zeta(z)-1)\\ &= 2^z(1-2^zR_3(z)) + O\bigl((8/9)^{\operatorname{Re} z}\bigr) + \gamma + O\bigl(2^{-\operatorname{Re} z}\bigr)\\ &= 2^z - 2^{2z}\cdot 3^{-z} - 2^{2z}\cdot4^{-z} - 2^{2z}\cdot O\bigl(5^{-\operatorname{Re} z}\bigr) + \gamma + O\bigl((8/9)^{\operatorname{Re} z}\bigr)\\ &= 2^z - \biggl(\frac{4}{3}\biggr)^z - 1 + \gamma + O\bigl((8/9)^{\operatorname{Re} z}\bigr) \end{align}

for $\operatorname{Re} z \geqslant 4$, and therefore

$$\lim_{\operatorname{Re} z \to +\infty} \Biggl[ 2^z - \biggl(\frac{4}{3}\biggr)^z - \zeta(\zeta(z))\Biggr] = 1-\gamma.$$

If on the other hand we let $z_k = -2k$ for $k\in\mathbb{N}\setminus \{0\}$, then we see that $2^{z_k} \to 0$, $(4/3)^{z_k} \to 0$, and $\zeta(\zeta(z_k)) = \zeta(0)$, so

$$\lim_{k\to+\infty} \Biggl[2^{z_k} - \biggl(\frac{4}{3}\biggr)^{z_k} - \zeta(\zeta(z_k))\Biggr] = -\zeta(0) = \frac{1}{2} \neq 1-\gamma,$$

which shows that the limit as $z\to\infty$ of the expression doesn't exist.


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.