Values appearing with density in an ergodic system

Let $(X,\mu)$ be a probability space with invertible, measure preserving, totally-ergodic map $T:X \to X$. ($(X,\mu,T)$ is a $\mathbb{Z}$ dynamical system). Suppose $f:X \to \mathbb{R}$ is a measurable function such that $\mu(x \colon f(x) = a) = 0$ for all $a \in \mathbb{R}$.


$$ D(x) = \left\{ a \in \mathbb{R} \colon \liminf_{n \to \infty} \frac{ \# \{0 \leq k < n \colon f(T^{k}x) = a \}}{n} > 0 \right\} $$

be the set of numbers that appear with density. Is $D(x) = \emptyset$ with probability $1$? Are there standard conditions on $(X,\mu,T)$ like mixing, positive entropy and so on that guarantee that $D(x) = \emptyset$ with probability $1$? One such is the example where $X = [0,1]^\mathbb{Z}$ $\mu = \mu_0^{\otimes \mathbb{Z}}$ where $\mu_0$ is the uniform measure on $[0,1]$, and $T$ is the shift map.

Speculation: Obviously this is true for all the rationals; i.e., $D(x) \cap \mathbb{Q} = \emptyset$ with probability 1. So maybe I need some sort of continuity for this to have a chance to be true? I'm in the fairly general setting of a shift system, and do have some flexibility in the way I formulate my problem. My only constraint is that $f$ should have the right "finite dimensional distributions" that are specified, and $X$ is a factor of a Bernoulli system ($\mu$ is a product measure). So perhaps I can construct $X$ and $f$, and put a topology on it such that it is continuous?

  • $\begingroup$ I suppose you wanted $\#\{0\le k<n\colon f(T^{k}x)=a\}$ instead of $\{0\le k<n\colon f(T^{k}x)=a\}$. $\endgroup$ – Iosif Pinelis Feb 14 at 18:50
  • $\begingroup$ @IosifPinelis yup, I fixed, thanks! $\endgroup$ – arjun Feb 14 at 19:28

Indeed $D(x)$ is empty for almost all $x$. It suffices to show that for all integers $b,\ell>0$, the set $$ D_b(x,\ell) = \left\{ a \in [b,b+1) \colon \liminf_{n \to \infty} \frac{ \# \{0 \leq k < n \colon f(T^{k}x) = a \}}{n} > \frac{1}{\ell} \right\} $$ is empty for almost all $x$, and then take a countable union over $b,\ell$. Fix $m$ large enough so that $$\max_{1 \le j \le m} \mu(f^{-1}[b+(j-1)/m,b+j/m)) <1/\ell \,.$$ (Such $m$ exists by the assumption that $\mu(x \colon f(x) = a) = 0$ for all $a \in \mathbb{R}$.)

write $$ E_b (\ell,j) = \left\{ x \in X \colon \liminf_{n \to \infty} \frac{ \# \{0 \leq k < n \colon f(T^{k}x)\in [b+(j-1)/m,b+j/m) \}}{n} > \frac{1}{\ell} \right\} \,.$$ the Birkhoff Ergodic Theorem implies that $\mu[ E_b (\ell,j)]=0$ for all $b,\ell,j$. Since $$\{x :\, D_b(x,\ell)\ne \emptyset\} \subset \bigcup_{j=1}^m E_b(\ell,j) \,, $$ we conclude that $\mu\{x :\, D_b(x,\ell)\ne \emptyset\} =0$.


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.