If $G:[0,1] \rightarrow [0,1]$ is the Gauss map which is defined as $$G(x) = \left\{\frac{1}{x}\right\} = \frac{1}{x} - \left\lfloor\frac{1}{x}\right\rfloor,$$ show that if $x$ is periodic of order $n$ for $G$ then it must be an quadratic irrational.
I am looking for a quadratic equation I believe. One can solve $G(x)=x$ for $n=1$ quite easily and with some more work can solve $G^2(x)=x$. However, this method does not generalise well to general $n$, due to the number of terms. Moreover, I could not see a formula that induction could be used upon that may make the problem easier.
Can anyone solve this problem?
Ok, I'll assume you are aware that the Gauss map is the same, after a change of 'encoding', as the left shift map on the continued fraction expansion of a real number. Stated in these terms, the (slightly stronger version of the) theorem we want to prove is
First let's prove a Lemma. Let $x=[q_0,q_1,\ldots]$ be a real number with infinite continued fraction expansion and let $$\begin{array}{ccc} a_0=q_0 & a_1=q_0q_1+1 & a_{n+2}=a_{n+1}q_{n+2}+a_n\\ b_0=1 & b_1=q_1 & b_{n+2}=b_{n+1}q_{n+2}+b_n. \end{array}$$ The rational $a_n/b_n$ is the usual $n$th convergent to $x$.
Proof: We'll prove this by induction. The $n=1$ case is some simple algebra to see that $$x=[q_0,q_1,x_1]=\frac{x_1q_0q_1+x_1+q_0}{x_1q_1+1}=\frac{x_1a_1 + a_0}{x_1b_1 + b_0}.$$ Suppose the Lemma is true for some $n\geq 1$, then $$\begin{array}{rcl} x=[q_0,\ldots, q_{n}, q_{n+1}, x_{n+1}] & = & [q_0,\ldots,q_{n},q_{n+1}+\frac{1}{x_{n+1}}]\\ & = & \dfrac{(q_{n+1} + \frac{1}{x_{n+1}})a_n + a_{n-1}}{(q_{n+1} + \frac{1}{x_{n+1}})b_n + b_{n-1}}\\ & = & \dfrac{x_{n+1}a_{n+1} + a_n}{x_{n+1} b_{n+1} + b_n} \end{array}$$ where the inductive hypothesis was used to get from the first to second line because $x_n = q_{n+1} + \frac{1}{x_{n+1}}$. We used the recursive definition of the $a_i$ and $b_i$ to get to the last line, along with some simple algebra (multiply through by $x_{n+1}$).
Now we can prove the theorem.
Proof: Let $x=[q_0,q_1,\ldots, q_{k-1}, \overline{q_{k},\ldots, q_{k+n}}]$ so that by the lemma we have $$x=\frac{x_k a_k + a_{k-1}}{x_k b_k + b_{k-1}}$$ and $$x_k=[q_k,\ldots, q_{k+n-1},x_k]=\frac{x_k a'_n + a'_{n-1}}{x_k b'_n + b'_{n-1}}$$ where $a'_n/b'_n$ is the $n$th convergent to $x_k$. So we can write $x_k(x_k b'_{n} + b'_{n-1})=x_k a'_n + a'_{n-1}$ which is a quadratic in $x_k$ with rational coefficients so $x_k$ is a quadratic irrational. If $x_k$ is the root of a quadratic, then it follows quickly from the first equation above that $x$ must also be a quadratic irrational and so we are done.