Deviations of ergodic sums for toral translations. I: Convex bodies (Q2453447): Difference between revisions

It is a known result that many deterministic dynamical systems with non-zero Lyapunov exponents satisfy the same limit theorems as the sums of independent random variables. In contrast to these systems, the limiting distributions depend on many facts like the dynamical systems and the observables, etc. In the present paper, the authors study systems of toral translations which are isometries and have zero Lyapunov exponents. To state the main contributions, let us recall some classical results of \textit{H. Kesten} [Ann. Math. (2) 71, 445--471 (1960; Zbl 0101.03803); Acta Arith. 7, 355--380 (1962; Zbl 0113.27004)] for the circle rotations. Any \(\alpha \in {\mathbb T} = {\mathbb R}/{\mathbb Z}\) defines a rotation \(x\to x+\alpha\) on \({\mathbb T}\). For \(0<r<1\), consider the indicator function \(\chi_{[0,r]}\) as an observable. One has the discrepancies \[ D_N(r,x,\alpha) =\sum_{n=0}^{N-1} \chi_{[0,r]}(x+n\alpha) - N r, \quad N\in {\mathbb N}. \] The results of Kesten are as follows. { Theorem 1. Given \(0<r<1\), there exists a constant \(\rho=\rho(r)\) such that if \((x,\alpha)\) is uniformly distributed on \({\mathbb T}^2\), then \[ \lim_{N\to\infty} \text{mes}\left\{ (x,\alpha): \frac{D_N(r,x,\alpha)}{r\ln N}\leq z\right\}= \frac{\tan^{-1} z}{\pi} + \frac{1}{2}, \quad z\in \mathbb R, \] a standard Cauchy distribution. Moreover, \(\rho(r) = \rho_0\) is independent of \(r\) if \(r\not \in {\mathbb Q}\), and \(\rho(r)\) depends on \(r\in {\mathbb Q}\) in a non-trivial way. } One of the main results of the present paper is the following extension of Theorem 1 to higher-dimensional cases. For the translations \(x\to x+\alpha\) on \({\mathbb T}^d\), \(d\geq 2\), by considering a strictly convex analytic body \({\mathcal C}\) which fits inside the unit cube of \({\mathbb R}^d\) and its scaled bodies \({\mathcal C}_r\), where \(0<r < r_0\) is such that \({\mathcal C}_r\) still fits inside the unit cube of \({\mathbb R}^d\), the observables are the indicator functions \(\chi_{{\mathcal C}_r}\) of \({\mathcal C}_r\) and the discrepancies are \[ D_{\mathcal C}(r,\alpha,x,N) = \sum_{n=0}^{N-1}\chi_{{\mathcal C}_r} (x+n\alpha) - N \text{Vol}({\mathcal C}_r). \] {Theorem 2. There exists a distribution function \({\mathcal D}_{\mathcal C}(z): {\mathbb R}\to [0,1]\) such that for any \(b>a>0\), there holds \[ \lim_{N\to\infty} \lambda\left\{(r,\alpha,x) \in [a,b]\times {\mathbb T}^d \times {\mathbb T}^d: \frac{D_{\mathcal C}(r,\alpha,x,N)}{r^{\frac{d-1}{2}} N^{\frac{d-1}{2d}}}\leq z\right\}={\mathcal D}_{\mathcal C}(z). \] Here \(\lambda\) is the Lebesgue measure on \([a,b]\times {\mathbb T}^d \times {\mathbb T}^d\).} Compared with Theorem 1, the parameter \(r\) in Theorem 2 is also random so that the irregular dependence of the limiting distributions can be suppressed. The paper is a systematic study to this topic in the following aspects: \(\bullet\) The limiting distributions \({\mathcal D}_{\mathcal C}(z)\) have been analyzed. \(\bullet\) The extensions of Theorem 2 to continuous time translations on \({\mathbb T}^d\) are given. \(\bullet\) Different scaling regimes for the discrepancies are discussed. \(\bullet\) Applications to counting lattice points in slanted cylinders and to time spent in a given ball by random geodesic on the flat torus are also given. The proof is mainly to establish some correspondence between the small divisors in the Fourier series of the discrepancy function and the lattices with short vectors, and mixing of diagonal flows on the space of lattices. For Part II, see [the authors, Publ. Math., Inst. Hautes Étud. Sci. 132, 293--352 (2020; Zbl 1473.37012)].