On the Wolfram page about the Euler-Mascheroni Constant $\gamma $, the following amazing limit is given without proof (referring to "personal communication"):

$$\lim_{z\to\infty}\left[\zeta(\zeta(z))-2^z+\Bigl(\frac43\Bigr)^z\right]=\gamma -1$$ I have checked to thousands of decimals that likewise $$\lim_{z\to\infty}\left[\zeta\Bigl(\frac1{\zeta(z)}\Bigr)+2^z-\Bigl(\frac43\Bigr)^z\right]=\gamma $$ and started wondering what happens if we combine both, e.g. by fixing a weight $0\leqslant\lambda\leqslant1$ and looking at the behaviour of $$ f_\lambda(z):=\zeta \Bigl(\lambda{\zeta(z)+\frac{1-\lambda}{\zeta(z)}}\Bigr) $$ for again growing $z$ on the real axis. So the argument in the big parenthesis still tends to $1$, such that the pole at $\zeta(1)$ is approached in different ways depending on $\lambda$. And something cute happens: Numerically it looks like $$\boxed{\color{blue}{ \lim_{z\to\infty}\left[f_{\lambda}(z)- \frac1{2\lambda-1}\left(2^z- \Bigl(\frac43\Bigr)^z\right) \right]=\gamma -\frac{\lambda}{(2\lambda-1)^2}}}$$ for $\lambda\ne\frac12$ (even outside the interval $[0,1]$), while, completely off, $$ f_{1/2}(z)\sim 2\cdot4^z-4\cdot\Bigl(\frac83\Bigr)^z -2\cdot 2^z -\text {(5 other exp terms)}-4+\gamma. $$

I am wondering if such a formula has appeared before and how to prove it!

  • 9
    $\begingroup$ A nice consequence of your conjectured formula is $\displaystyle{\lim_{z\to\infty}(f_{\lambda}(z)+f_{1-\lambda}(z))=\gamma-\frac{1}{(2\lambda-1)^2}}$. $\endgroup$ Dec 1 '19 at 20:38

The paper entitled Euler constant as a renormalized value of Riemann zeta function at its pole by Andrei Vieru contains a derivation of the first formula in the OP (Benoȋt Cloitre's formula), and a method to obtain variations thereof, such as

as well as similar iterates in terms of the Dirichlet function $\lambda(s)=\sum_{n=0}^\infty(2n+1)^{-s}$,


The first formula is trivial. $$f(s)= \frac1{s-1}+\gamma +O(s-1)$$ $$g(z)=1+2^{-z}+3^{-z}+4^{-z}+O(5^{-z})=1+2^{-z}(1+(3/2)^{-z}+(4/2)^{-z}+O(5/2)^{-z})$$

$$f(g(z)) = \frac1{2^{-z}(1+(3/2)^{-z}+(4/2)^{-z}+O(5/2)^{-z})}+\gamma+O(2^{-z})$$ $$=2^z \left(1-(3/2)^{-z}-(4/2)^{-z}+O(5/2)^{-z}+(O(3/2)^{-z})^2\right)+\gamma + O(2^{-z})$$

$$=2^z- (3/4)^{-z}-1+O(9/8)^{-z}+\gamma$$

  • 17
    $\begingroup$ I would not call it exactly "trivial"... Pulling out the $2^{-z}$ is rather clever than trivial. Could we agree on "straightforward"? $\endgroup$
    – Wolfgang
    Dec 2 '19 at 9:02
  • 1
    $\begingroup$ It doesn't contain any non-trivial information about $\zeta(s)$ (for example the abscissa of convergence of $\frac1{\zeta(s)-1}$ is non-trivial) $\endgroup$
    – reuns
    Dec 2 '19 at 10:40
  • 4
    $\begingroup$ yes I agree. But sometimes one can make sophisticated proofs out of trivial information/axioms, think of some of the proofs of $\zeta(2)=\frac{\pi^2}6$. :) :) $\endgroup$
    – Wolfgang
    Dec 2 '19 at 10:49
  • $\begingroup$ For $\lambda=1/2$, this approach shows that there are altogether $8$ exponential terms. I have not included in the edit those with $(\frac{16}{9})^z,(\frac{16}{10})^z,(\frac{16}{12})^z,(\frac{16}{14})^z,(\frac{16}{15})^z$. Doing their coefficients manually is too error-prone... $\endgroup$
    – Wolfgang
    Dec 2 '19 at 12:34
  • $\begingroup$ Yes I tried in pari-gp but I didn't success in manipulating $2^{-z}+O(3^{-z})$ (whereas $x+O(x^2)$ is a built-in function) $\endgroup$
    – reuns
    Dec 4 '19 at 14:00

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.