Today I will be sharing my proof that the decimal representations of rational numbers must repeat. This proof is based on a simple idea, but turned out to be pretty involved, with several parts.
The fundamental idea can be illustrated using long division. Consider the rational number $\frac{8}{7}$. We can find its decimal representation using the long division calculation below.
Each step of this calculation begins with the remainder from the previous step (circled in red). Since the remainder on division by 7 must be an integer between 0 and 6, there are only 7 possible remainders. Thus, the remainder must repeat within no more than 8 steps. Once it repeats, the calculation will begin to cycle, causing the decimal representation it produces to repeat, as this one does.
Now let’s prepare for the more formal proof by proving two lemmas. I think the first of these is quite interesting and not immediately obvious. The second one is obvious, but I wanted to be assured of it, since it is vital to the main proof.
Lemma 1: Given a rational number $x$, if there exist two distinct digits $d_1$ and $d_2$ in the decimal representation of $x$ such that the sequence of digits following $d_1$ is the same as the sequence of digits following $d_2$, then the decimal representation of $x$ repeats.
Assume that two such digits exist and let the one which appears first be called $d_1$. Then the decimal representation of $x$ begins as follows: $s_1.s_2d_1s_3d_2$, where $s_1$ is a sequence of at least one digit and $s_2$ and $s_3$ are sequences of zero or more digits.
Now, since the sequence of digits following $d_2$ is the same as the sequence of digits following $d_1$, the decimal representation must continue: $s_1.s_2d_1s_3d_2s_3d_2$. Yet this further specifies the sequence of digits following $d_1$, so the same digits must appear again in the sequence following the initial appearance of $d_2$, yielding $s_1.s_2d_1s_3d_2s_3d_2s_3d_2$. This process (illustrated below) will continue infinitely, so $x=s_1.s_2d_1\overline{s_3d_2}$.
Thus, the lemma holds.
(I realize the notation was a little sloppy here, with $d_2$ representing both a digit and a digit with its position, but I couldn’t find a good way to fix it.)
Lemma 2: Let $x$ be a non-negative rational number such that $x=\frac{n}{m}$, where $n$ and $m$ are non-negative integers and $m\neq 0$. There is no more than one way to represent $x$ as a mixed number $q\frac{r}{m}$ such that $q$ and $r$ are non-negative integers and $r<m$.
Suppose $\frac{n}{m}=q_1\frac{r_1}{m}$ and $\frac{n}{m}=q_2\frac{r_2}{m}$ where $q_1$, $q_2$, $r_1$, and $r_2$ are non-negative integers and $r_1,r_2<m$.
It follows that $q_1+\frac{r_1}{m}=q_2+\frac{r_2}{m}$ and thence that $q_1-q_2=\frac{r_1-r_2}{m}$.
Since $0<r_1<m$ and $0<r_2<m$, $-1<\frac{r_1-r_2}{m}<1$.
Thus $-1<q_1-q_2<1$, from which it follows that $q_1=q_2$, since $q_1$ and $q_2$ are integers.
Hence $\frac{r_1-r_2}{m}=0$, so $r_1=r_2$.
Therefore, both mixed numbers are the same, and the theorem holds.
Finally, we come to the main theorem.
Proposition: If $x$ is a rational number, then the decimal representation of $x$ repeats.
For any $x$, if the decimal representation of $x$ repeats, then the decimal representation of $-x$ repeats. Thus, it is sufficient to show that the theorem holds for all non-negative $x$.
Let $x$ be a non-negative rational number. Then $x=\frac{n}{m}$ for some non-negative integers $n$ and $m$, where $m\neq 0$.
The fraction $\frac{n}{m}$ can be expressed as a mixed number $q_1\frac{r_1}{m}$ such that $q_1$ and $r_1$ are non-negative integers and $r_1<m$. By Lemma 2, there is only one way to do this.
Using algebra we have: $$q_1\frac{r_1}{m}=q_1+\frac{r_1}{m}=q_1+\frac{r_1}{m}\left(\frac{10}{10}\right)=q_1+\frac{10r_1}{m}\left(\frac{1}{10}\right)$$
As above, the fraction $\frac{10r_1}{m}$ can be expressed uniquely as a mixed number $q_2\frac{r_2}{m}$ such that $q_2$ and $r_2$ are non-negative integers and $r_2<m$.
This yields $$q_1+\frac{10r_1}{m}\left(\frac{1}{10}\right)=q_1+q_2\left(\frac{1}{10}\right)+\frac{r_2}{m}\left(\frac{1}{10}\right)\text{.}$$
Notice that $q_2$ has only one digit, since $0\leq\frac{r_1}{m}<1$ and $0\leq q_2\leq\frac{10r_1}{m}$. Notice also that $0\leq\frac{r_2}{m}\left(\frac{1}{10}\right)<\frac{1}{10}$, since $0\leq\frac{r_2}{m}<1$. This means that $q_2$ is the tenths-place digit of the decimal expansion of $x$.
The algorithm can be continued as shown below, $$q_1+q_2\left(\frac{1}{10}\right)+\frac{r_2}{m}\left(\frac{1}{10}\right)=q_1+q_2\left(\frac{1}{10}\right)+\frac{r_2}{m}\left(\frac{1}{10}\right)\left(\frac{10}{10}\right)$$ $$=q_1+q_2\left(\frac{1}{10}\right)+\frac{10r_2}{m}\left(\frac{1}{100}\right)=q_1+q_2\left(\frac{1}{10}\right)+q_3\left(\frac{1}{100}\right)+\frac{r_3}{m}\left(\frac{1}{100}\right)$$ $$=q_1+q_2\left(\frac{1}{10}\right)+q_3\left(\frac{1}{100}\right)+\frac{r_3}{m}\left(\frac{1}{100}\right)\left(\frac{10}{10}\right)=q_1+q_2\left(\frac{1}{10}\right)+q_3\left(\frac{1}{100}\right)+q_4\left(\frac{1}{1000}\right)+\frac{r_4}{m}\left(\frac{1}{1000}\right)…$$ with $q_i\frac{r_i}{m}$ for $i>1$ always the unique mixed number representation of the fraction $\frac{10r_{i-1}}{m}$ such that $q_i$ and $r_i$ are non-negative integers and $r_i<m$. By logic similar to that above, when $i>1$, $q_i$ will always be a digit of the decimal representation of $x$, with its value given by the power of $\frac{1}{10}$ by which it is multiplied.
Since, for all $i$, $r_i$ is a non-negative integer less than $m$, there are only $m$ possible values of $r_i$. This means that, among the infinite values of $r_i$, there must exist distinct $r_j$ and $r_k$ such that $r_j=r_k$. (ETA: By the Pigeonhole Principle!)
Since the mixed number representations used by the algorithm are unique, if $r_j=r_k$, then $q_{j+1}=q_{k+1}$ and $r_{j+1}=r_{k+1}$. By induction, it therefore follows that $q_{j+h}=q_{k+h}$ for all $h$. Since, when $i>1$, each $q_i$ is a digit of the decimal representation of $x$, it follows by Lemma 1 that decimal representation of $x$ repeats.
Therefore the theorem holds.
I hope this makes some measure of sense. I had a lot of trouble making this proof lucid, and eventually I had spent too much time working on. Those whose eyes did not glaze over will note that the algorithm I employed here is essentially long division. In developing it, I did not set out specifically to reproduce the long division algorithm, and I was intrigued when I noticed they are basically the same. Here is our long division problem annotated to show the relationship:
