Let $a_1,a_2,\ldots$ be the positive squarefree numbers in strictly increasing order and consider the sequence $x_i \equiv a_i \pmod{2}$ where $x_i \in \{0,1\}$. Is the sequence $\{x_i\}$ aperiodic? That is, does there not exist $T$ and $n_0$ such that for all $n \geq n_0$, $x_{n+T} = x_n$?

The sequence starts out as $$1,0,1,1,0,1,0,1,1,0,1,1,1,1,0,1,0,1,0,1,1,0,1,1,0,1,1,0,1,0,1,1,1,1,1,0,\ldots.$$ How could we show, if it is, aperiodic?


Let $Q_1(x)$ and $Q_2(x)$ count the number of odd and even squarefree numbers up to $x$ respectively. Notice that if the sequence you proposed is periodic, then at every period $T$ we always get the same number of odd and of even squarefree numbers. Assume that the ratio between the count of odd and even numbers in the period is $q$. Then it will follow that $|Q_1(x) - qQ_2(x)|$ is bounded (since at every period we get exactly q times as many odd squarefrees as evens). We will show that is too regular to be true.

First remember that:

$$f(s) = \sum_{squarefree \,n} \frac{1}{n^s} = \prod_p(1+\frac{1}{p^s}) = \prod_p\frac{(1-\frac{1}{p^{2s}})}{(1-\frac{1}{p^s})} = \frac{\zeta(s)}{\zeta(2s)}$$

Now define analogous functions $f_1$, $f_2$, but restricted only to sums of odd and eben squarefree numbers respectively.

Notice that:

$$f_2(s) = \sum_{even\,squarefree \,n} \frac{1}{n^s} = \sum_{odd\,squarefree \,n} \frac{1}{(2n)^s} = 2^{-s} f_1(s)$$

While $f_1(s)+f_2(s) = f(s)$. From that it is easy to get:

$$f_1(s) = \frac{2^s}{1+2^s}\frac{\zeta(s)}{\zeta(2s)}\,\,\,,\,\,\, f_2(s) = \frac{1}{1+2^s}\frac{\zeta(s)}{\zeta(2s)}$$

Now we consider:

$$f_1(s) - qf_2(s) = \frac{2^s-q}{1+2^s}\frac{\zeta(s)}{\zeta(2s)}$$

We can write this as a series:

$$f_1(s) - qf_2(s) = \sum_{n\, squarefree} \frac{A(n)}{n^s}$$

Where $A(n)$ is $1$ if $n$ is odd or $-q$ if $n$ is even. But notice that the partial sums of $A$ satisfy:

$$A(1)+...+A(x) = Q_1(x)-qQ_2(x)$$

Hence from Abel's lemma (or summing by parts) we obtain that the series $\frac{A(n)}{n^s}$ converges and defines an analytic function for $\Re(s) > 0$.

This provides an analytic extension for:


All the way to $\Re(s)> 0$. But this contradicts the fact that $\zeta(s)$ has at least one zero with real part $\frac{1}{2}$ (so our function must have a pole at that zero). So we get a pole for or function for some $s$ with real part $1/4$ - one can verify that the numerator doesn't vanish.

Indeed, one can show that $Q_1(x) \sim 2Q_2(x)$, and this same method shows that

$$|Q_1(x) - 2Q_2(x)| = O(x^{1/4-\epsilon})$$

cannot hold for $\epsilon>0$.

  • $\begingroup$ The idea is there. I would say that if $x_m$ is $q$-periodic and $C = \frac{1}{q}\sum_{m=1}^q x_m$ then $\sum_{m=1}^\infty (x_m-C)m^{-s}$ is analytic for $\Re(s) > 0$. Here $\sum_{m=1}^\infty (x_m-C)m^{-s} = \frac{(1-2^{-s}) \zeta(s)}{(1-2^{-2s})\zeta(2s)}-C \zeta(s)$ which we know isn't analytic for $\Re(s) > 0$ since $\frac{1}{\zeta(2s)}$ has a pole at $2s = 1/2+i 14.15...$ not cancelled by the other terms $\endgroup$ – reuns Jul 12 '17 at 22:47
  • $\begingroup$ Indeed the function that appears is $\frac{\zeta(s)}{\zeta(2s)}$ rather than $\frac{\zeta(2s)}{\zeta(s)}$ - I have edited the solution accordingly - thanks for the correction. $\endgroup$ – Rodrigo Jul 12 '17 at 23:04

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.