Constructive proof of Kronecker’s density theorem

A couple months ago I shared a blog post on Dirichlet’s approximation theorem which we may use to prove that integer angles are dense in the unit circle. Now, I tried to go further and show that $\forall k \in \mathbb{N}, e^{i k\mathbb{N}}$ is dense in the unit circle. This particular result is interesting because it leads one to deduce that $\forall N \in \mathbb{N}, {\bigcap}_{n=0}^N sin( p_n\mathbb{N})$ is dense in $[-1,1]$ where $p_n$ is the nth prime.

This week I learned about the Weyl equidistribution theorem from which this particular result follows as an immediate corollary. However, I wondered whether there might be a simpler proof and after searching I came across a short paper on Kronecker’s density theorem which had exactly the result I wanted and a bit more. The purpose of this blog post is to present the proof contained in this paper.

Kronecker’s density theorem:
If $\theta \in \mathbb{R} \setminus \pi \mathbb{Q}$, then $\{ e^{i \theta n} | n \in \mathbb{Z} \}$ is dense in the unit circle.

Proof:

1. $\mathbb{Q} \pi$ is dense in $\mathbb{R}$ so it’s enough to prove that:

\begin{aligned} \forall t \in \mathbb{R} \forall \epsilon \in \pi \mathbb{Q} \exists n \in \mathbb{Z}, |e^{it}-e^{in\theta}|<\epsilon \end{aligned}

Equivalently we have:

\begin{aligned} \forall t \in \mathbb{R} \forall \epsilon \in \mathbb{Q} \pi \exists p,q \in \mathbb{Z}, |p\theta -t+2q\pi|<\epsilon \end{aligned}

2. Without loss of generality, we may assume that $0<\theta< 2 \pi$ since given that $\theta \notin \mathbb{Q} \pi \exists k \in \mathbb{Z}$,

\begin{aligned} 0< \theta - 2k \pi < 2 \pi \end{aligned}

It’s sufficient to prove the case $t=0$ since if we have found $p,q \in \mathbb{Z}$ such that

\begin{aligned} |p\theta +2q\pi|<\epsilon \end{aligned}

for any $t \in \mathbb{R}$ we can find $k \in \mathbb{Z}$ such that:

\begin{aligned} |k - \frac{t}{p\theta+2q \pi}| <1 \end{aligned}

To obtain $k$ it’s sufficient to take the integer part of the fraction in the inequality.

Now, it follows that we have:

\begin{aligned} kp\theta -t+2kq\pi|= |p\theta+2q\pi| |k - \frac{t}{p\theta+2q \pi}|< |p\theta -t+2q\pi|<\epsilon \end{aligned}

3. For fixed $\epsilon \in \mathbb{Q} \pi$ the proof strategy is as follows:

a) starting at $e^{i\theta}$ we move anti-clockwise around the unit circle in steps of arc length $\theta$ until we cross the positive x-axis.

b) $\theta \notin \mathbb{Q} \pi$ so we reach $e^{i\theta_1}$ where $0< \theta_1<\theta$ and

\begin{aligned} \exists k \in \mathbb{N}, k\theta=2\pi+\theta_1 \end{aligned}

c) if $\theta_1-\theta > \epsilon$ we’re done. Otherwise, we repeat the procedure with $\theta := \theta_1$.

d) The remainder of this proof will show that we can find an upper-bound for the number of iterations of $\theta_n$ required.

5. Before proceeding, it’s useful to note that there exists $\{k_n \}_{n=1}^{\infty} \subset \mathbb{N}$ where at the first iteration we have:

\begin{aligned} k_1\theta = \theta_1 + 2\pi \implies \theta_1 = k_1 \theta mod 2\pi < \theta \end{aligned}

So at the nth iteration we have:

\begin{aligned} \theta_n = \theta \prod_{i=1}^{n} k_i mod 2 \pi \end{aligned}

6. If $\theta_i - \theta_{i-1} > -\epsilon$, then we’re done as we have:

\begin{aligned} 0< \theta_{i-1}-\theta_{i}= (k_i - 1) \theta \prod_{j=1}^{i-1} k_i mod 2 \pi < \epsilon \end{aligned}

Otherwise, let’s define the smallest $M \in \mathbb{N}$ such that:

\begin{aligned} M \epsilon > \theta \end{aligned}

If the process continued after the Mth iteration we would have:

\begin{aligned} 0< \theta_M < \theta_{M-1} -\epsilon < \theta_{M-2}-\epsilon-\epsilon < ...< \theta-M\epsilon < 0 \end{aligned}