I'm trying to find the ordinary generating function $f(n)$ such that $$f(n) = \sum_{k=0}^\infty\mu^2(n)\, x^k$$

I have a nasty looking answer that involves Hadamard products, and I was hoping there was a cleaner, canonical answer.

The Hadamard solution, for those interested:

\begin{align*} e_{n}(x) & = \sum_{k=0}^\infty \mu(x) \, \mathbf{1}_{k \leq n} x^k \\ f_n(x) & = \sum_{k=0}^\infty \mu(x) \, \mathbf{1}_{k \leq n} x^k \\ g_n(x) & = \sum_{k=0}^\infty \mu^2(x) \,\mathbf{1}_{k \leq n} x^k \\ & = \sum_{k=0}^\infty\bigg(\mu(x) \, \mathbf{1}_{k \leq n}\bigg)\bigg( \mu(x)\, \mathbf{1}_{k \leq n} \bigg)x^k \end{align*}

where we recognize the last line as the Hadamard product of the power series $e_n, \, f_n$ (i.e. the term-wise product).

According to the Wikipedia page above, if closed forms are known for $e_n, f_n$ then

$$ g_n(x) =\frac1{2\pi} \int_{0}^{2\pi} e_n\big(\sqrt{x} e^{xt}\big) \, f_n\big(\sqrt{x} e^{-xt}\big) dt$$ where

$$e_n(x) = f_n(x) = \sum \limits_{n=1}^{\infty} \mu(n)x^n = x - \frac{x^{2}}{1-x} + \sum \limits_{a=2}^{\infty} \frac{x^{2a}}{1-x^{a}} - \sum \limits_{b=2}^{\infty} \sum \limits_{a=2}^{\infty} \frac{x^{2ab}}{1-x^{ab}} + \sum \limits_{c=2}^{\infty} \sum \limits_{b=2}^{\infty} \sum \limits_{a=2}^{\infty} \frac{x^{2abc}}{1-x^{abc}} - \sum \limits_{d=2}^{\infty} \sum \limits_{c=2}^{\infty} \sum \limits_{b=2}^{\infty} \sum \limits_{a=2}^{\infty} \frac{x^{2abcd}}{1-x^{abcd}} + ...,$$ a known result.


Here are some ideas which I got, starting with Lambert series.

This first part is in https://en.wikipedia.org/wiki/Lambert_series

A zeta series of a sequence is $Z_f(s) =\sum_{n=1}^{\infty} \dfrac{f_n}{n^s} $.

A Lambert series is of the form $S_a(q) =\sum_{n=1}^{\infty} a_n\dfrac{q^n}{1-q^n} =\sum_{m=1}^{\infty} b_mq^m $ where $b_m =\sum_{n|m} a_n $.

Examples are for the Mobius function, $S_{\mu}(q) =\sum_{n=1}^{\infty} \mu(n)\dfrac{q^n}{1-q^n} =q $ and for Liouville's $\lambda$ function, $S_{\lambda}(q) =\sum_{n=1}^{\infty} \lambda(n)\dfrac{q^n}{1-q^n} =\sum_{m=1}^{\infty} q^{m^2} $ (a theta function).

The reason I mentioned the last series is because of this which is in https://en.wikipedia.org/wiki/Liouville_function:

"The Liouville function's Dirichlet inverse is the absolute value of the Möbius function."

If you use another result in the article above, $\dfrac{\zeta(2s)}{\zeta(s)} =\sum_{n=1}^{\infty} \dfrac{\lambda(n)}{n^s} $, this means that, if $Z_{\mu^2}(s) =\sum_{n=1}^{\infty} \dfrac{\mu^2(n)}{n^s} $, then $Z_{\mu^2}(s)\dfrac{\zeta(2s)}{\zeta(s)} =1 $ so $Z_{\mu^2}(s) =\dfrac{\zeta(s)}{\zeta(2s)} $.

Since $\mu^2(n) = |\mu(n)|$, this is the zeta series for your function.

Combining all these, you might be able to get what you want.

I'll stop here.

  • $\begingroup$ Is there a way to convert a zeta series to an OGF? $\endgroup$ – Tiwa Aina May 3 '18 at 0:36
  • $\begingroup$ There might be a way using a Lambert series, but I am not clear on how to do it. That is why I stopped where I did. $\endgroup$ – marty cohen May 3 '18 at 2:16

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.