This question is closely related to this one. Again consider the group $G := PSL(2, \mathbb F)$ over some field $\mathbb F$.

Then as written in the other post, there are two natural actions associated with $G$, one is by considering the points $\mathbb F^2$ and look at the induced action of $SL(2,\mathbb F)$ (as $PSL(2,\mathbb F)$ is just a quotient of $SL(2, \mathbb F)$), another by viewing $G$ as consisting of "Möbius"-like maps of the form $$ x \mapsto \frac{ax + bc}{cx + d} $$ where $ad - bc = 1$ (remark: you get the same set of mappings by just asserting that $ad - bc$ must be a square) and let them act on $\mathbb F_{\infty} = \mathbb F \cup \{\infty\}$ (the extended field, or the projective line). Now I want to show that both actions are equivalent, i.e. we have essentially the same actions by identifiying the points $\mathbb F^2$ with the projective line $\mathbb F_{\infty}$.

I started to give such an indentification, and then started to compute if the actions are equivalent, I give all these computations. In case iv) and v) I got stuck. So I am asking if anybody could help me resolving these points?

Now the computations: Given two actions of $G := PSL(2, \mathbb F)$ on two sets, $\Omega := \mathbb F^2$ and $\Gamma := \mathbb F_{\infty}$. Then an operation of a group $G$ on two sets $\Omega, \Gamma$ is called equivalent if there exists a bijection $\lambda : \Omega \to \Gamma$ such that $$ \lambda(\alpha^g) = \lambda(\alpha)^g $$ for all $g \in G$. The bijection is $\lambda((x,y)^T) = x/y$ if $y \ne 1$ and $\lambda((x,0)^T) := \infty$ otherwise. An element of $g \in PSL(2,\mathbb F)$ is represented by four numbers $a,b,c,d \in \mathbb F$ such that $ad - bc = 1$. This could be viewed as a matrix modulo $\{I, -I\}$, or as a mapping as written at the beginning. So if $(x,y)^T \in \mathbb F^2$ we define the action by matrix multiplication, and for $x \in \mathbb F_{\infty}$ by applying the mapping. Now let $g \in PSL(2, \mathbb F)$ and let $a,b,c,d$ the corresponding numbers (up to a common constant).

i) $y \ne 0, cx + dy \ne 0$

Then \begin{align*} \lambda\left(\begin{pmatrix} a & b \\ c & d \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} \right) & = \lambda\left( \begin{pmatrix} ax + by \\ cx + dy \end{pmatrix} \right) = \frac{ax + by}{cx + dy} = \frac{a(x/y) + b}{c(x/y) + d} = \lambda((x,y)^T)^g. \end{align*}

ii) $y = 0, cx + dy \ne 0$. Then $x \ne 0$ and $c \ne 0$ is implied. \begin{align*} \lambda\left(\begin{pmatrix} a & b \\ c & d \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} \right) & = \lambda\left( \begin{pmatrix} ax \\ cx \end{pmatrix} \right) = \frac{ax}{cx} = \frac{a}{c} = \infty^g = \lambda((x,y)^T)^g. \end{align*}

iii) $y = 0, cx + dy = 0, c = 0$. \begin{align*} \lambda\left(\begin{pmatrix} a & b \\ c & d \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} \right) = \infty = \infty^g = \lambda((x,y)^T)^g \end{align*}

iv) $y = 0, cx + dy = 0, c \ne 0$

Then we must have $x = 0$. Hence $\lambda((0,0)^g) = \lambda((0,0))$ and this should equal $\lambda((0,0))^g$, hence $\lambda((0,0))^g$ must be a fixed point of each $g$ with the above conditions. But I do not see what this fixed point should be? The only reasonable choice would be $\lambda((0,0)) = \infty$, but that does not works out as for $c \ne 0$ we have $\infty^g = a/c$.

v) $y \ne 0, cx + dy = 0$.

\begin{align*} \lambda\left(\begin{pmatrix} a & b \\ c & d \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} \right) = \infty \end{align*} Here $\lambda((x,y)) = x/y$ and $(x/y)^g = (a(x/y) + b)/(c(x/y) + d)$, and this must equal $\infty$, but it does only if $y = 0$ (as then $\lambda((x,y)) = \infty)$ or $(-d/c)^g = \infty$ is also another value that gives $\infty$, but in general we do not have $x = -d, y = c$ in this situation. So how to go on?

EDIT: With Derek's answer I am able to resolve my issues:

iv) As this gives $x = 0, y = 0$ the argument does not desribe a line, therefore this case is excluded.

v) Here $cx + dy = 0$ implies $\langle (x,y)^T \rangle = \langle (-d, c)^T \rangle$ if $(-d, c) \ne (0,0)$ (note $(-d,c) = (0,0)$ is not possible anyway). Now $$ \lambda(\langle (x,y)^T \rangle^g) = \lambda(\langle (ax + by,0)^T \rangle) = \infty. $$ On the other side we have $\lambda(\langle (x,y)^T \rangle) = \lambda(\langle (-d, c)^T \rangle) = -d/c$ if $c \ne 0$, and $\lambda(\langle (-d, c)^T\rangle) = \infty$ if $c = 0$. An now by convention $(-d/c)^g = \infty$ for $c \ne 0$ and $\infty^g = \infty$ (see wikipedia:Möbius transformation). So that the equality asserted for equivalent actions holds.

Now suppose $c = 0$, then by $ad - bc = 1$ we must have $d \ne 0$ and by convention $\infty^g = \infty$ then. Further as above $\lambda(\langle (x,y)^T \rangle^g) = \infty$ and $\lambda(\langle (x,y)^T \rangle) = \infty = \infty^g$ the equation for equivalent actions hold.

If $d = 0$, then $c \ne 0$ and $b \ne 0$, and $y \ne 0$ together with $cx + dy = 0$ implies $x = 0$. This gives $$ \lambda(\langle (0,y)^T \rangle^g) = \lambda(\langle (by, dy)^T \rangle) = \lambda(\langle (b, 0)^T \rangle) = \infty. $$ On the other side $\lambda(\langle (0,y) \rangle) = 0$, so we must have $0^g = \infty$ if $d = 0$ and the argument ist $0$? I guess this would be natural convention by looking at the mapping side, as then $g$ would be the mapping $$ \frac{ax + b}{cx} $$ and for $x = 0$ setting the value of this "division by zero" to $\infty$ might be a natural definition (anyway, the convention $0^g = \infty$ for $d = 0$ is included in $(-d/c)^g = \infty$).

  • 1
    $\begingroup$ I don't understand how you are defining the action on $F^2$. There is a natural action of ${\rm SL}(2,F)$ on $F^2$, but this does not induce a action of ${\rm PSL}(2,F)$, because the scalars are acting non-trivially. $\endgroup$ – Derek Holt Dec 15 '15 at 20:24
  • $\begingroup$ For some $g \in \mbox{PSL}(2, F)$ and $(x,y) \in F^2$ I take the matrix $\begin{pmatrix} a & b \\ c & d \end{pmatrix}$ and multiply by it, it this not well-defined? But this is exactly how I derived that the image of $\infty$ should be $a/c$ in the other post I linked, and they answered that this is the right way to do it... $\endgroup$ – StefanH Dec 15 '15 at 20:41
  • $\begingroup$ It is not well defined because the matrix is not uniquely determined by $g$. $\endgroup$ – Derek Holt Dec 15 '15 at 21:02
  • $\begingroup$ Okay, then is there such an action on points possible? And why they told me in the other post that my computations where fine, is there a construction closely similar to mine? But thank you for pointing out, one misconception less in my mind. $\endgroup$ – StefanH Dec 15 '15 at 21:14

For all $n>1$, the natural action by matrix multiplication of ${\rm GL}(n,F)$ on $F^n$ induces a (doubly transitive) action on the set $\Omega$ of $1$-dimensional subspaces of $F^n$. This induced action has the scalar matrices in its kernel, so there is an induced action of ${\rm PGL}(n,F)$ on $\Omega$. When $n=2$, this is equivalent to the action on $F \cup \{ \infty \}$ that you are describing.

The equivalence $\lambda$ between the actions is as you defined it, but with $\Omega$ equal to the $1$-dimensional subspaces rather than the vectors of $F^2$. So $\lambda(\langle (x,y)^{\rm T}\rangle)=x/y$ if $y \ne 0$, and $\lambda (\langle (x,0)^{\rm T}\rangle)= \infty$.

All of this is also true with $\rm (P)SL$ in place of $\rm (P)GL$.

  • $\begingroup$ Thank you, that clarified it. I put an edit in my question where everything is resolved now. $\endgroup$ – StefanH Dec 16 '15 at 11:48

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.