0

Let's consider: $$\tau(x,a,b)=\sum_{1 \le d \le x \\ (d,x)=d^a \\} d^b$$

Where $(q,r)$ denotes the gcd of $q$ and $r$.

I think this could be interesting thing to look at because it's somehow a type of bridge between the sum of the divisor function $\sigma_k(x)=\tau(x,1,k)$ and Euler's totient function $\phi(x)=\tau(x,0,0)$.

Now, the average order of these functions is fairly well understood.

For example,

$$\sum_{n=1}^x \tau(n,1,1) \approx \frac{\pi^2}{12}x^2$$

$$\sum_{n=1}^x \tau(n,0,1) \approx \frac{1}{\pi^2}x^3$$

$$\sum_{n=1}^x \tau(n,1,0) \approx x\log(x)+(2\gamma+1)x$$

$$\sum_{n=1}^x \tau(n,0,0) \approx \frac{3}{\pi^2}x^2$$

And these can be argued using the standard techniques which are Abel Summation formula and Dirichlet convolutions. Where $\gamma$ is the Euler Macheroni constant.

Is it possible to achieve similar results for non integer values $a$?

For example, what is the average order of $\tau(x,\frac{1}{2},1)$? What is the average order $\tau(x,\frac{1}{2},0)$?

Mason
  • 3,792
  • Here's the first 100 values of the functions. – Mason Dec 08 '18 at 19:01
  • I just chose $\tau$ because it's in between $\sigma$ and $\phi$ in the Greek alphabet. It shouldn't be confused with $\tau$ as $\sigma_0$ which is sometimes how it used in number theory. If this is confusing to anyone I guess I could change it to an upsilon. – Mason Dec 08 '18 at 19:10
  • Why asking for the asymptotic of many weird arithmetic functions instead of studying the famous ones (in particular the prime number theorem) ? Here and to all your questions the methods of the PNT and the Dirichlet divisor problem (poles and bounds for the Dirichlet series, Mellin inversion, tauberian theorem) apply. – reuns Dec 08 '18 at 20:43
  • @reuns. Undoubtedly, I should study the classics. But I think $\tau$ is interesting in that is a type of bridge between important functions. – Mason Dec 08 '18 at 20:53
  • Then can you at least define the Dirichlet series corresponding to your question and why there is a chance we can find the asymptotic of the partial sums of their coefficients – reuns Dec 08 '18 at 22:56
  • like this – Mason Dec 08 '18 at 23:30
  • The generating functions for $\sigma_k$ and $\phi$ are presented in the wiki on Lambert Series – Mason Dec 09 '18 at 02:25
  • This also seems relevant – Mason Dec 09 '18 at 02:42
  • It looks like we can (sometimes) move from LG to DGFs according to this article. – Mason Dec 09 '18 at 02:53
  • 1
    Since $\sum_{d | n}\phi(d)=n $ then $\sum_{n=1}^\infty \phi(n) n^{-s}= \frac{\zeta(s-1)}{\zeta(s)}$. The Riemann zeta function has a lot of nice properties from which we can say a lot on $\sum_{n=1}^\infty \phi(n) n^{-s}$ and $\sum_{n=1}^N \phi(n)$. Can you do the same with your above functions ? – reuns Dec 09 '18 at 02:57
  • I dunno but I can try – Mason Dec 09 '18 at 02:58
  • $\tau$ (or from my other question $\phi_k$) isn't multiplicative so I'm not sure that this approach is going to work for a non-multiplicative function. – Mason Dec 09 '18 at 19:44

0 Answers0