4
$\begingroup$

In Yves Benoist and Jean-François Quint's notes Introduction to random walks on homogeneous spaces (top of page 11), the following is listed as a step in the non-Fourier analytic proof of ergodicity of hyperbolic toral automorphisms.

Let ${\mathbb T}^d={\mathbb R }^d/{\mathbb Z }^d$. Let $g$ be a matrix in $\operatorname{SL}(d,\mathbb{Z})$ with no eigenvalue which is a root of unity. Then the subgroup of the torus given by vectors $v+{\mathbb Z}^d \in {\mathbb T}^d$ such that $\lim_{n \to \infty} g^n v = 0$ is dense in ${\mathbb T}^d$.

Benoist's notes merely say this is proved by "induction on $d$", but I do not understand why this is true. Can anyone elucidate this?

$\endgroup$
4
  • 2
    $\begingroup$ I’m not sure about the induction part, but the fact seems straightforward enough: there is no rational relationship between the coordinates of any eigenvector (or real and imaginary parts of such a vector). Hence the set of real multiples of a stable eigenvector is dense in the torus. $\endgroup$ Commented Nov 26, 2022 at 20:32
  • $\begingroup$ @AnthonyQuas Why is this the case: there is no rational relationship between the coordinates of any eigenvector (or real and imaginary parts of such a vector). This is pretty clear in dimension 2 but why in higher dimensions? $\endgroup$ Commented Nov 26, 2022 at 21:01
  • 2
    $\begingroup$ Here is the eigenvector case assuming also irreducibility of the characteristic polynomial. Suppose $v$ is an eigenvector and suppose $u$ is a primitive integer vector such that $u\cdot v=0$. Using duality and the fact that $v$ is an eigenvector, $(A^T)^nu$ is orthogonal to $v$ for all $n$. So the span of these vectors is a sub-lattice of $Z^d$ dimn between 1 and $d-1$ of that is invariant under $A$. Now if you restrict $A$ to the corresponding real subspace you get a matrix whose minimal polynomial is a factor of the original minimal poly. But the original minimal polynomial was irreducible. $\endgroup$ Commented Nov 27, 2022 at 4:58
  • 2
    $\begingroup$ If the characteristic polynomial is reducible, you should be able to work separately on the reducible pieces. $\endgroup$ Commented Nov 27, 2022 at 5:05

1 Answer 1

2
$\begingroup$

It's even more direct. Fix $g\in\mathrm{GL}_d(\mathbf{Z})$, acting on $T=\mathbf{R}^d/\mathbf{Z}^d$, with no eigenvalue that is a root of unity.

Let $T$ be the given torus, $T_{\mathrm{cont}}$ the subgroup of elements contracted by $g$. Define $S$ as the quotient torus $T/\overline{T_{\mathrm{cont}}}$. Then $g$ acts on the torus $S$ with no eigenvalue (on the tangent space) of modulus $<1$. Since the determinant of $g$ on (the tangent space of) $S$ is $\pm 1$, we deduce that all eigenvalues of $g|_S$ have modulus 1, and are not roots of unity. By the next lemma (with $h^{-1}=g|_S$), this forces $\dim(S)=0$, i.e. $\overline{T_{\mathrm{cont}}}$ is dense in $T$.

Lemma. Let $h$ be a matrix in $\mathrm{GL}_n(\mathbf{Z})$. If all eigenvalues have modulus $\le 1$, then all are roots of unity.

Proof: up to conjugate $h$ in $\mathrm{GL}_n(\mathbf{Z})$, we can suppose that $h$ is upper block-triangular matrix with $\mathbf{Q}$-irreducible diagonal blocks. If $s$ is a diagonal block, then $(s^m)_{m\ge 0}$ is bounded and hence achieves finitely many values. So the eigenvalues are roots of unity.

$\endgroup$
2
  • 1
    $\begingroup$ What do you mean by `algebraic units of modulus 1 have modulus ≠1'? I believe that we may have roots with modulus 1 but not roots of unity. $\endgroup$ Commented Nov 28, 2022 at 9:59
  • 1
    $\begingroup$ @ChristopheLeuridan you're right, the argument is now fixed. $\endgroup$
    – YCor
    Commented Nov 28, 2022 at 11:00

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .