This question expands on this one from MSE.

In the literature about Dirichlet $L$-series, I found that their Euler products:

$$L(s, \chi) =\prod_p \bigg(\frac {1}{1-\frac{\chi(p)}{p^s}} \bigg)$$

are typically considered to be only converging for $\Re(s)>1$.

However, there seems to be an exception to this rule since Euler proved that:

$$L(1, \chi_4) =\prod_p \bigg(\frac {p}{p-\chi_4(p)} \bigg)=\prod_p \bigg(\frac {p}{p-\sin\left(\frac{p \,\pi}{2}\right)} \bigg)=\frac{3}{4}\cdot\frac{5}{4}\cdot\frac{7}{8}\cdot\frac{11}{12}\cdot\frac{13}{12}\dots=\beta(1)=\frac{\pi}{4}$$

does converge (albeit slowly).

I then decided to explore values for $\Re(s) \lt 1$ and numerical evidence suggests that the Euler-product:

$$\prod_p \bigg(\frac {p^s}{p^s-\sin\left(\frac{p \,\pi}{2}\right)} \bigg)$$

also (slowly) converges in the domain $\frac12 < \Re(s) \le 1$.


1) Is the Euler product for $L(s,\chi_4)$ the only one known to converge for $s=1$?

2) Does this particular Euler product indeed converge in the right half of the strip?


  • 10
    $\begingroup$ The Euler product for every Dirichlet series (for a nonprincipal character) converges at $s=1$, but that's more-or-less equivalent with the Prime Number Theorem in arithmetic progressions, and in particular Euler's heuristic argument for $\chi_4$ does not readily yield a rigorous proof for that $L$-function. Convergence for real part $> 1/2$ is equivalent to the Riemann hypothesis for the same $L$-function. (While I was editing this ABCDveve wrote similarly in his/her answer.) $\endgroup$ – Noam D. Elkies Jun 17 '15 at 21:40
  • 5
    $\begingroup$ To see the equivalence that Noam mentions, see Theorem 3.3 (and the two lemmas preceding it) in math.uconn.edu/~kconrad/articles/eulerprod.pdf, setting $d = 1$ and $\alpha_{p,1} = \chi(p)$. Note that for infinite products like an Euler product, the term "converges" means "converges and is not $0$." Since nobody has ever proved the $L$-function of a nontrivial Dirichlet character has no zeros in any vertical strip $1-\varepsilon < {\rm Re}(s) < 1$, it is basically hopeless at present to expect anyone to prove the Euler product converges at any $s$ with ${\rm Re}(s) < 1$. $\endgroup$ – KConrad Jun 17 '15 at 23:14
  • $\begingroup$ Related question here: mathoverflow.net/questions/63714/… $\endgroup$ – GH from MO Jun 17 '15 at 23:57
  • $\begingroup$ @NoamD.Elkies Why is the trivial character corresponding to zeta exception to convergence? $\endgroup$ – joro Jun 18 '15 at 13:41
  • 1
    $\begingroup$ Good question. The answer is basically the pole at $s=1$. The connection with the $L$-function zeros is via the logarithmic derivative $L'/L$, which is singular when $L$ has either a zero or a pole. Dirichlet $L$-functions other than the Riemann zeta function have no poles, so only their zeros affect the analysis; but when you try to do the same for $\zeta$ you run into the $s=1$ pole before any zero shows up. $\endgroup$ – Noam D. Elkies Jun 18 '15 at 15:08

By taking the logarithm (or log derivative) of $L(s)$, you get a Dirichlet series whose convergence relates directly to the zeros. Your question about convergence in the right half of the critical strip is equivalent to the Riemann hypothesis for $\chi$. Similarly, (conditional) convergence of the Euler product (again as the "Euler sum" after taking the logs) on the 1-line is equivalent to the lack of zeros on this line.

  • $\begingroup$ If $\chi$ is kronecker character it is nonnegative. Let $t=0$. Every term is real greater than one. I don't see how this will converge to the $L$ function (computing a single value above the correct is counterexample). $\endgroup$ – joro Jun 18 '15 at 6:36
  • 1
    $\begingroup$ @joro: The only character modulo $n$ that is nonnegative is the trivial character modulo $n$. $\endgroup$ – GH from MO Jun 18 '15 at 6:37
  • $\begingroup$ @GHfromMO Thank you. My mistake. So why the answer doesn't apply to the trivial character which is zeta AFAICT? $\endgroup$ – joro Jun 18 '15 at 7:03
  • $\begingroup$ @joro: Roughly speaking, the reason is the oscillation of $\chi$ (when $\chi$ is nontrivial). More precisely, while $\sum p^{-s}$ over the primes clearly diverges unless $\Re s>1$, the oscillating sum $\sum\chi(p)p^{-s}$ has a chance to converge for smaller values of $\Re s$ assuming the primes are sufficiently well-distributed mod $n$ (where $n$ is the conductor of $\chi$). And, indeed, GRH implies that the latter sum converges for $\Re s>1/2$. $\endgroup$ – GH from MO Jun 18 '15 at 7:41
  • $\begingroup$ @joro: Two related posts are mathoverflow.net/questions/63714/… and mathoverflow.net/questions/181241/… $\endgroup$ – GH from MO Jun 18 '15 at 7:43

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.