Asymptotic measures and distributions of Birkhoff averages with respect to Lebesgue measure (Q1865962)

Let \(A=\{x\in X\mid \overline\varphi(x)\in\mathbb{Q}\}\), \(B=\{x\in X\mid \overline\varphi(x)\in \mathbb{R}\backslash \mathbb{Q}\}\), \(C=\{x\in X\mid \overline\varphi(x)\text{ is not a point}\}\). Let for an open subinterval \(J\) \(\tau_{x,n}(J)={1\over n}\#\{i\leq n\mid \varphi_i(x)\in J\}.\) Then \(x\) is said to have a time distribution when \(\limsup_{n\to\infty}\tau_{x,n}(J)=\liminf_{n\to\infty}\tau_{x,n}(J)\) for all open subintervals \(J\). Define \(D\) is the set of points with time distribution. Theorem 2.6. \(D^c=C\). The purpose of this paper is to study a general theory including the cases that \(m(c)\) is not necessarily equal to zero. Question 1. Are predictable and statistically predictable equivalent? Question 2. If \({1\over i}\sum_{j=0}^{i-1}f^j_*m\) is weakly convergent, does it imply statitically predictability? One gives a flow on a plane which have heteroclinic attractors, and shows: Theorem 3.1. It has an asymptotic measure with respect to Lebesgue measure. There is an open, dense set of smooth measures for which the system has no asymptotic measure. There are smooth measures for which the system has an asymptotic measure, but is not statistically predictable. Thus this system is a counterexample of both Question 1 and 2. Let \[ \nu(J,x)=\begin{cases} 1, &\text{if }x\in A\cup B \text{ and }\overline\varphi(x)\in J,\\ 0, &\text{if }x\in A\cup B\text{ and }\overline\varphi(x)\not\in J,\\ |J\cap\overline\varphi(x)|/ |\overline\varphi(x)|, &x\in C, \end{cases} \] \(\overline\nu_\varphi(J)=\int\nu(J,x) dm,\) and \(\nu_\varphi\) is a measure constructed by \(\overline\nu_\varphi\). Let also \(\mu_\varphi=\overline\varphi_*m\). Then there are conjectures: Conjecture 1. Whenever \(\mu_\varphi\) exists it equals \(\nu_\varphi\). Conjecture 2. If \((X,f,m)\) is statistically predictable, then for any \(\varphi\) \(\mu_\varphi\equiv\nu_\varphi\). One also poses several questions.