2
$\begingroup$

The recurrence theorem of Halmos is well known in the case of a non-singular endomorphism $T$ of a measured space $(X,\mathcal B,\mu)$. A measurable subset $A$ is contained in the conservative part (mod $\mu$) if and only if $$ \sum_{n \geq 0} 1_B \circ T^n = \infty $$ holds a.e. in $B$ (where $1_B$ stands for the characteristic function) for any measurable $B \subset A$.

See e.g. Aaronson's book. Now I have never seen a proof of this theorem in the case of a non-singular group operation (where the group is locally compact and second countable). In fact I have never seen it stated in this form but I believe it must be true.

Does someone know of a reference?

A related theorem is Hopf's theorem which states that the conservative part is (mod $\mu$) the set of all $x$ such that $$ \int f(hx) d\eta(h) = \infty $$ where $f \in L^1(\mu)$ is $>0$ and $\eta$ is some Haar measure on $H$. I would be equally happy with a proof of Hopf's theorem instead.

Thank you.

$\endgroup$

2 Answers 2

1
$\begingroup$

Assume that $A$ is contained in the conservative part and $\mu(A)>0$.

We argue by contradiction so let us pretend there is some measurable subset $A' \subset A$ such that for any $x \in A'$, $\eta(H(x))$ is finite, where $H(x)=\{h \in H ; hx \in A \}$. We may also assume that there is some constant $C>0$ such that $\eta(H(x) < C$ for any $x \in A'$. Then there must be some relatively compact neighbourhood $V$ of the identity element of $G$ such that $\mu(H(x) \cap V) > \frac{1}{2} \mu(H(x))$ ($x \in A'$).

Now let $x\in A',h \in H$ be such that $hx \in A'$. We have $$ \frac{\mu(H(hx) \cap V)}{\mu(H(hx))} > 1/2$$ and $H(hx)=H(x)h^{-1}$ so we get $$ \frac{\mu(H(x) \cap Vh)}{\mu(H(x))} > 1/2 $$ (thanks to relative right invariance of left Haar measure). Since we also have the same inequality with $Vh$ replaced by $V$, we deduce that $Vh \cap V \neq \varnothing$, that is, $h \in V^{-1} V$: $h$ belongs to some compact set. This contradicts the fact that $A$ is contained in the conservative part.

$\endgroup$
0
$\begingroup$

Regarding what you wrote I would be equally happy with a proof of Hopf's theorem instead, it seems that the very fact that there is "conservative part" for a nonsingular action is unavailable for general lcsc groups. However, there is a "version" of the Hopf Decomposition for the purpose of checking the divergence $$(*)\quad \intop_{G}\frac{d\mu\circ g^{-1}}{d\mu}f\left(gx\right)dm\left(g\right)=\infty.$$

This can be done as follows. Pick a lattice in $G$ (if there exists any!) and take the Hopf Decomposition $D\cup C$ w.r.t. this lattice (regarding this lattice as an acting group on its own right, and using the Hopf Decomposition for the countable case as in [1, Proposition 1.6.2]). It is then true that $(*)$ holds for $x\in D$ and, moreover, the set $D$ is independent on the lattice up to null set. See [1, Theorem 1.6.4] and [2, Corollary 2.2].

References

[1] Aaronson, Jon, An introduction to infinite ergodic theory, Mathematical Surveys and Monographs. 50. Providence, RI: American Mathematical Society (AMS). xii, 284 p. (1997). ZBL0882.28013.

[2] Roy, Parthanil, Nonsingular group actions and stationary S(\alpha )S random fields, Proc. Am. Math. Soc. 138, No. 6, 2195-2202 (2010). ZBL1196.60093.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.