For every freely acting affine transformation, its linear part has 1 as an eigenvalue.

183 Views Asked by At

I was reading a math paper and got stuck on a fact that was considered to be elementary by the author.

The title is pretty much all about my question. More detailed question is as the following:

Let $A$ be a finite dimensional affine space, and let $g$ be an affine transformation on $A$ which acts freely, i.e. it does not fix any element of $A$. Then the linear part of $A$, which is a linear transformation on the vector space associated to $A$, has 1 as an eigenvalue.

I tried to verify this statement, but I do not have any idea to start with. I will appreciate any helps or hints.

Thanks in advance.

1

There are 1 best solutions below

0
On BEST ANSWER

$\newcommand\mysetminus{\raise.3ex\smallsetminus}$ I don't know if a replay to your post is still useful to you, as it has been a long time. But, in doubt, I post it anyway.
A simple proof is in https://it.wikipedia.org/wiki/Trasformazione_affine#Punti_fissi (in italian, sorry). This proof use coordinate systems, matrices,... I want to propose here a proof that does not make use of coordinates, let us say synthetic; a little less easy, to be honest.

Let $\mathcal A(E)$ be an $n$-dimensional affine space over the vector space $E$, and

$$ g\;:\; \mathcal A(E) \longrightarrow \mathcal A(E) $$

an affine function without fixed points. Suppose by absurd that the linear part $dg$ of $g$ does not admit the eigenvalue $1$. This means that $dg(x) \neq x$ for all $x\in E\mysetminus\{0\}$, i.e.

$$ (dg-1_E)(x) \neq 0\qquad \forall x\in E\mysetminus\{0\}, $$

where $1_E$ is the identity function of $E$. It follows that the kernel of $dg-1_E$ is $\{0\}$, so the linear function $dg-1_E\;:\;E \to E$ is invertible [remember the rank-nullity Theorem].

Let us now consider the function

$$ g-1_{\mathcal A(E)}\;:\;\mathcal A(E) \longrightarrow E^{\text{aff}}(E) $$ $$ (g-1_{\mathcal A(E)})(P) := g(P)-1_{\mathcal A(E)}(P) = g(P)-P $$

where $E^{\text{aff}}$ is the vector space $E$ equipped with the canonical structure of affine space (the points are the vectors,...) over the vector space $E$, and $1_{\mathcal A(E)}$ is the identity function of $\mathcal A(E)$. It is an affine function because:

\begin{align} \big(g-1_{\mathcal A(E)}\big)(&P)-\big(g-1_{\mathcal A(E)}\big)(Q) = \big(g(P)-P\big)-\big(g(Q)-Q\big)\\[2ex] &= g(P)-g(Q)-(P-Q)\\[2ex] &= dg(P-Q)-1_E(P-Q)\\[2ex] &= (dg-1_E)(P-Q), \end{align}

and this also proves that

$$ d(g-1_{\mathcal A(E)}) = dg-1_E $$

and therefore that $g-1_{\mathcal A(E)}$ is invertible, being invertible its linear part. Taking then

$$ Q := (g-1_{\mathcal A(E)})^{-1}(\vec 0), $$

we deduce that

$$ g(Q)-Q = (g-1_{\mathcal A(E)})(Q) = \big(g-1_{\mathcal A(E)}\big)\big((g-1_{\mathcal A(E)})^{-1}(\vec 0)\big) = \vec 0, $$

so $g(Q)=Q$, a contradiction.