I first asked the following question on Mathematics StackExchange (a few weeks ago), since the content of MathOverflow is mostly above my academic level. I didn't want to bother people on this forum with a maybe trivial question. But since I got no answer, I'm trying my luck here.

I'd like some help to prove the following theorem:

Let $\sum_{n \geq 1}\frac{f(n)}{n^s}$ and $\sum_{n \geq 1}\frac{g(n)}{n^s}$ be two Dirichlet series with respective abscissas of absolute convergence $\alpha_f$ and $\alpha_g$ ($\alpha_f, \alpha_g \neq -\infty$). Then the abscissa of convergence of $\sum_{n \geq 1}\frac{f*g(n)}{n^s}$ is:

  • $\max(\alpha_f, \alpha_g)$ if $\alpha_f \neq \alpha_g$ ;
  • less than or equal to $\alpha$ if $\alpha_f = \alpha_g = \alpha$

where $f*g$ refers to the Dirichlet convolution of $f$ and $g$.

The result is to be proved in an exercise from this book: exercise 9, page 259. I need help to prove the theorem in the case $\alpha_f \neq \alpha_g$.

After searching the literature, I found similar theorems stated in a few sources, but never saw any proof.

What I have managed to prove up to now:

  • If $f$ and $g$ are positive real-valued functions, the result is obvious.
  • Suppose $\alpha_f < \alpha_g$. Then, if $$\sum_{n=2}^{+\infty}\frac{|f(n)|}{n^{\alpha_g}} < |f(1)|$$ the result is also true.

I tried to experiment with particular functions not satisfying either of these strong conditions, but I fail to see why the result is true in the general case.

  • $\begingroup$ Here is a link to the Math.StackExchange question: math.stackexchange.com/questions/1043825/… $\endgroup$ – Ricardo Andrade Dec 26 '14 at 23:45
  • 2
    $\begingroup$ Why don't you try relating $\alpha_f$ to a growth condition on $f(n)$? $\endgroup$ – Kimball Dec 27 '14 at 4:11
  • $\begingroup$ If $\alpha_f > 0$ (if not, we can still multiply both $f$ ang $g$ by a power function to come down to this case, since abscissae of absolute convergence are finite), then $$\alpha_f = \limsup\frac{\ln\sum_{k=1}^{n}|f(k)|}{\ln(n)}$$ Is is what you refer to as "growth condition" ? $\endgroup$ – edouard.gilles Dec 27 '14 at 16:21

I think it is false:

Consider a simple series with a zero at $s=2$. For example $$E(s)=1-\frac{1}{2^s}-\frac{12}{4^s}=P(2^{-s}),\quad \text{with} \quad P(x)=1-x-12x^2.$$ We have $E(2)=0$ and $E(1)=-5/2$. Since $$\frac{1}{1-x-12x^2}=\frac{1}{7}\sum_{k=0}^\infty (4^{k+1}+(-1)^k 3^{k+1})x^k,\qquad |x|<\frac14.$$

We have $$E(s)^{-1}=1+\frac{1}{7}\sum_{k=1}^\infty (4^{k+1}+(-1)^k 3^{k+1})\frac{1}{2^{ks}},\qquad \sigma>2.$$ Consider $$A(s)=\zeta(s) E(s)$$ The Dirichlet series $A(s)$ is absolutely convergent for $\sigma>1$, because it is the product of two absolutely convergent Dirichlet series is absolutely convergent. The abscissa of absolute convergence of $A(s)$ is $a=1$ because the function $A(s)$ extends to a meromorphic function and has a simple pole at $s=1$.

Let now $B(s)=E(s)^{-1} \zeta(s)^{-1}$. The Dirichlet series $B(s)$ is absolutely convergent for $\sigma>2$ as product of two absolutely convergent Dirichlet series. But the function $B(s)$ extends to a meromorphic function with a pole at $\sigma=2$. Therefore $b=2$ is is abscissa of absolute convergence.

The product $A(s)B(s)=1$ and its abscissa of absolute convergence is $-\infty$.


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.