2

Let $f(x) = \frac{ax+b}{cx+d}$ with $x \in \mathbb{R} \setminus \left \{ -\frac{d}{c} \right \} $ be the Möbius Transformation. Find all $a,b,c,d$ such that $$ \underbrace{f \circ f \circ \dots \circ f}_{n \text{ times}} = \text{id}.$$ My professor and I came to the conclusion that you probably can't write down nice forms depending on $n$. (The case $n=3$ gives a complicated solution.)

However, can you think of any "good" condition or any meaningful progress towards this problem? Feel free to change it a little - it should just be approximately this problem!

(Note that it's easy to find a solution $a,b,c,d$ that works for all $n$. The problem is to find all solutions.)

Qi Zhu
  • 9,413
  • 3
    I am going to say something very basic, forgive me if you already know. Möbius transformations can be written as projective transformations of $P^1(\mathbb C)$ and as such, represented as matrices. Therefore the composition of two of them should be represented as the product of two matrices. This would reduce your problem to the algebraic one of finding all idempotent matrices in $GL_2(\mathbb C)$. – Giuseppe Negro Nov 06 '17 at 18:23
  • https://en.wikipedia.org/wiki/M%C3%B6bius_transformation#Projective_matrix_representations – Giuseppe Negro Nov 06 '17 at 18:28
  • 1
    Exact same question here (the answer there seems to be weaker than the present one) and here – Giuseppe Negro Nov 07 '17 at 18:03

1 Answers1

2

NOTATION. We denote $f^{\circ n}:=f\circ f\circ \ldots \circ f$.

I cannot find a "pointwise" condition on $a,b,c,d$. (EDIT: see bottom of the post for such a condition, due to Will Sawin). However if we regard the matrix $$ A=\begin{bmatrix} a & b \\ c & d\end{bmatrix},$$ then interpreting the Möbius transformation as a projective transformation we see that the given Möbius transformation sastisfies $f^{\circ n}=\mathrm{id}$ if and only if $A^n$ is a scalar multiple of $I=\begin{bmatrix} 1 & 0 \\ 0 & 1\end{bmatrix}$. As shown here such matrices are necessarily diagonalizable, and the equation $$ A^n=zI$$ shows that the eigenvalues $\lambda_1, \lambda_2$ of $A$ must satisfy $$\lambda_1^n=\lambda_2^n=z.$$

Conclusion. The condition $f^{\circ n} =\mathrm{id}$ for the Möbius transformation represented by the invertible matrix $A=\begin{bmatrix} a & b \\ c & d \end{bmatrix}$ is equivalent to $A$ being diagonalizable and the eigenvalues $\lambda_1, \lambda_2$ of $A$ satisfying $\lambda_1^n=\lambda_2^n$.

EDIT. A necessary condition for $f^{\circ n}=\mathrm{id}$ is the following: $$\tag{1} \frac{(a+d)^2}{ad-bc}=2(1+\cos\theta),\quad \theta=\frac{2\pi}{n}k,\ k\in\mathbb Z.$$ If $k\ne 0 \mod n$, the condition is also sufficient. Otherwise, it is not, and the necessary and sufficient condition is that (1) holds and $A$ is diagonalizable.

Proof. This is due to Will Sawin. Letting $\lambda_1, \lambda_2$ denote the eigenvalues of $A$, we have $$ \frac{(a+d)^2}{ad-bc} = \frac{\lambda_1}{\lambda_2}+\frac{\lambda_2}{\lambda_1}+2,$$ which equals $2+2\cos\theta$ precisely when $\lambda_1/\lambda_2 = e^{2\pi i \frac k n}$, that is, $\lambda_1^n=\lambda_2^n$. If $k\ne 0 \mod n$, then $\lambda_1\ne \lambda_2$ and this automatically implies that $A$ is diagonalizable. Therefore $f^{\circ n}=\mathrm{id}$ because of the characterization proven above.

On the other hand, the matrix $A=\begin{bmatrix} 1 & 1 \\ 0 & 1\end{bmatrix}$ satisfies (1) with $k=0$ and $$A^n= \begin{bmatrix} 1 & n\\ 0 & 1\end{bmatrix},$$ so it cannot satisfy $A^n=A$ for any $n>1$. Notice that $A$ is not diagonalizable.

To conclude, we note that in the case $k\ne 0\mod n$, we need to manually add to (1) the assumption that $A$ is diagonalizable to ensure that $f^{\circ n}=\mathrm{id}$ holds. $\square$

  • 2
    We have $(a+d)^2/(ad-bc)= (\lambda_1+\lambda_2)^2 / (\lambda_1\lambda_2) = 2 + \lambda_1 \lambda_2^{-2} + \lambda_2 \lambda_1^{-1}$, which is equal to $2 + 2 \cos \theta$ for some $\theta$ of the form $2 \pi k /n$, $k$ an integer, if and only if $\lambda_1/\lambda_2= e^{2\pi k i /n}$ for some integer $k$. – Will Sawin Nov 16 '17 at 15:07
  • @WillSawin: Thank you, I have incorporated your nice remark into the answer. – Giuseppe Negro Nov 21 '17 at 16:21