The full diapason of convergence rates

of Birkhoff averages for ergodic flows · 2026

Read The full diapason of convergence rates by of Birkhoff averages for ergodic flows (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-21404.txt.

Introduction

A study of ergodic flows asks how fast long-run averages settle down. For a flow that mixes a space thoroughly, time averages of a quantity should approach its overall mean. The speed of that approach is called the rate of convergence of Birkhoff averages.

The work shows that every speed, from the fastest possible down to slower than any named rate, can occur. The trick is the choice of the quantity being averaged. For flows that wind around a torus, the averaging functions can still be kept continuous. The result maps the full range of calm and wild behavior these averages can show.

Catalog record

CREATORof Birkhoff averages for ergodic flows
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-21404

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The full diapason of convergence rates of Birkhoff averages for ergodic flows Abstract To Alexander Grigorievich Kachurovskii on the occasion of his 65th birthday Abstract: For an ergodic flow, a range of rates of convergence of Birkhoff averages from the maximum rate to an arbitrarily slow rate is realized by choosing the averaging function. For torus windings, the continuity of the averaging functions is ensured. This complements Krengel’s classical result on the slow rates of convergence of means for ergodic automorphisms. Keywords: rates of convergence in ergodic theorems, special flow, odometer, torus winding MSC: 37A30, 37A10 1 Introduction For an ergodic measure preserving flow T_{t} on a probability space (X,m) and a function f\in L_{1}(X,m) , Birkhoff’s ergodic theorem states that the time averages A(f,t,x):=\frac{1}{t}\int_{0}^{t}f(T_{s}x)ds converge for almost all {x\in X} to the spatial average \int fdm . For some functions, this convergence may be uniform. This type of convergence is the subject of this article. Our goal is to demonstrate how to control the rate of convergence of averages by selecting the function to be averaged. The function is sought as a functional series. As a result, for a given flow, a range of speeds is realized, from the so-called maximum rate to an arbitrarily slow one. Continuous realizations of the functions being averaged are found for ergodic torus windings. We also consider the general case of ergodic flows without periodic trajectories. Here, Rudolph’s theorem on a special representation of a flow when the return function differs little from a constant is applied. Rudolph’s theorem can be viewed as a continuous analogue of the Rokhlin–Halmos lemma. It should be noted that Krengel used this lemma in [ 1 ] , and in [ 2 ] , some delicate generalization of the Rokhlin–Halmos lemma due to Alpern [ 3 ] was used for slow convergences. The effect of slowing down the convergence rate is also possible in the case of weighted averages, as shown in [ 4 ] . Remarkably, in certain situations, weighted averages yield ultrafast convergence rates of type o(\frac{1}{t}) , see [ 5 ] . Our paper considers only classical Birkhoff averages, for which the maximum rate of convergence of the averages cannot be about o(\frac{1}{t}) in the case of a nonzero averaged function. 2 Convergence with maximum rate Let X=[0,1) . Consider the rotation of a circle as the flow, i.e. T_{t}x=\{x+t\} (the fractional part of the sum x+t ). Obviously, for a function {f\in L_{1}(X,m)} with zero mean for all x , {A(f,t,x)=0} holds for {t\in\mathbb{Z}} and |A(f,t,x)|\leq\frac{\|f\|_{1}}{t},\ \ t>0. Thus, convergence in Birkhoff’s theorem is uniform and occurs at the maximum possible rate (a rate of the form o(1/t) can only occur in the case of a zero function) [ 6 ] . Note that discrete Birkhoff sums may behave quite differently; see [ 7 , 8 ] for example. Above, we mentioned the degenerate case where the entire phase space of the flow is a periodic trajectory. The maximum rate of convergence of averages also arises in the non-ergodic case, when each point of the phase space has a bounded (periodic) trajectory. Specifically, let \mathcal{P}(x) be the period of the point {x\in X} with respect to the flow {T_{t}} , i.e., the minimum number t>0 such that {T_{t}x=x} ; we set for non-periodic points {\mathcal{P}(x)=\infty} . If the function {\mathcal{P}\in L_{\infty}(X,m)} , then the convergence of ergodic averages will also be maximum, but, generally speaking, non-uniform. Indeed, assuming 0<t=N\mathcal{P}(x)+r,\,0\leq r<\mathcal{P}(x),N\in\mathbb{N}\cup\{0\}, we obtain \left|A(f,t,x)-\frac{1}{\mathcal{P}(x)}\int_{0}^{\mathcal{P}(x)}f(T_{s}x)\,ds\right|=\\ =\left|\frac{1}{t}\sum_{k=0}^{N-1}\int_{k\mathcal{P}(x)}^{(k+1)\mathcal{P}(x)}\!\!f(T_{s}x)\,ds+\frac{1}{t}\int_{N\mathcal{P}(x)}^{t}\!\!f(T_{s}x)\,ds-\frac{1}{\mathcal{P}(x)}\int_{0}^{\mathcal{P}(x)}\!\!f(T_{s}x)\,ds\right|=\\ =\left|\left(\frac{N}{t}-\frac{1}{\mathcal{P}(x)}\right)\int_{0}^{\mathcal{P}(x)}f(T_{s}x)\,ds+\frac{1}{t}\int_{0}^{r}f(T_{s}x)\,ds\right|\leq\\ \leq\frac{2}{t}\int_{0}^{\mathcal{P}(x)}|f(T_{s}x)|\,ds. Thus, the presence of slow convergence rates in Birkhoff’s ergodic theorem is related to the unboundedness (i.e., non-periodicity) of the flow trajectories. Below, we will consider only such cases. Example of a special ergodic flow. Let S:[0,1)\to[0,1) be a bijection preserving the Lebesgue measure \mu on [0,1) . We define a flow T_{t} on [0,1)\times[0,1) whose action is verbally described as follows: all points (x,y) from {[0,1)\times[0,1)} move vertically upward with unit velocity until they reach the boundary of the square, i.e. {T_{t}(x,y)=(x,y+t)} . Upon reaching the boundary point, {(x,1)} jump to {(Sx,0)} and then continue moving upward until they end up in (S^{2}x,0) , and so on. We glued the point {(x,1)} to {(Sx,0)} , and the points move with constant velocity, and this motion preserves area (i.e., the measure m=\mu\times\mu on a square). The resulting flow is ergodic with respect to the measure m only if the transformation S is ergodic with respect to the Lebesgue measure \mu on the interval. Such a flow is called a special flow (or suspension flow) over the automorphism S with constant return function. Note that for a function independent of x on [0,1)\times[0,1) with zero mean, the assertion of uniform convergence with maximum rate in Birkhoff’s ergodic theorem is obvious (the arguments are no different from the case of a rotating circle). Next, we consider a suspension flow over a 2-odometer with a constant return function. Dyadic odometer. Consider a 2-odometer S ; it operates on the interval [0,1) as follows. For every partition of the interval [0,1) into 2^{n} intervals of the form {[\frac{m}{n}\,,\frac{m+1}{n})} , the transformation S cyclically permutes these small intervals. Moreover, each such interval is mapped to itself under the action of the power S^{2^{n}} . And the restriction of this power to a small interval is similar to the original odometer. Let us give an explicit definition of such a transformation S . We represent the number x\in[0,1) in binary notation: x=\sum\limits^{\infty}_{i=1}\frac{x_{i}}{2^{i}} , x_{i}\in\{0,1\} . We set S(0,0\dots)=(0,1\dots),\ \ S(0,10\dots)=(0,01\dots), S(0,\underbrace{11\dots 11}_{m}0\dots)=(0,\underbrace{00\dots 00}_{m}1\dots), where the remaining digits, denoted by ellipses, remain unchanged. The odometer is also called an adic shift, an addition machine, a von Neumann transform, or an automorphism with a dyadic-rational spectrum. Fig. 1 shows its (self-similar) graph. Special flow over the odometer. To study the flow over the 2-odometer, for every n\in\mathbb{N} we will identify the space M=[0,1)\times[0,1) with the rectangle {M_{n}=[0,\frac{1}{2^{n}})\times[0,2^{n})} . Рис. 1: Self-similar graph of 2-odometer (left); Realization of M_{n},n=2 using the flow over the 2-odometer (right) The square M is the union of pieces of trajectories of length 1 for points belonging to the lower boundary. But M can be represented as M_{n} — the union of trajectories of length 2^{n} for points from the interval [0,2^{-n})\times\{0\} . In Fig. 1 we depicted this for n=2 . We introduce coordinates x_{n},y_{n} on such a rectangle M_{n} . Let us consider functions {f\in L_{1}(M)} satisfying the following conditions: (1) {f(x,y)=\sum\limits_{n=1}^{\infty}f_{n}(x_{n},y_{n})}, where {f_{n}\in L_{1}(M_{n})} such that (2) {\int_{0}^{2^{n}}f_{n}(x_{n},y_{n})\,dy_{n}=0} for all x_{n}\in[0,2^{-n}) and (3) {\sum_{n=1}^{\infty}\|f_{n}\|_{\infty,1}<\infty}, where {\|f_{n}\|_{\infty,1}=\sup\limits_{x_{n}\in[0,2^{-n})}\int_{0}^{2^{n}}|f_{n}(x_{n},y_{n})|\,dy_{n}}. Теорема 2.1 . Let the function f satisfy conditions (1)-(3). Then the average A(f,t,(x,y)) for the flow over the 2-odometer converge uniformly to 0 as t\to\infty with maximum rate. Proof. It is easy to see that condition (2) implies that the mean value of f is equal to zero. We will show that for all t>0 and all {(x,y)\in M} |A(f,t,(x,y))|\leq\frac{2\sum_{n=1}^{\infty}\|f_{n}\|_{\infty,1}}{t}. Indeed, let {2^{N-1}<t\leq 2^{N}}, then |A(f,t,(x,y))|\leq\left|A\left(\sum_{n=1}^{N}f_{n},\ t,(x,y)\right)\right|+\left|A\left(\sum_{n=N+1}^{\infty}f_{n},\ t,(x,y)\right)\right|\leq\\ \leq\sum_{n=1}^{N}\left|A\left(f_{n},\ t,(x,y)\right)\right|+\sum_{n=N+1}^{\infty}A(|f_{n}|,\ t,(x,y)). For each term of the second sum we immediately obtain the estimate A(|f_{n}|,\ t,(x,y))=\frac{1}{t}\int_{0}^{t}|f_{n}(T_{s}(x_{n},y_{n}))|\,ds=\\ =\frac{1}{t}\int_{0}^{2^{n}-y_{n}}|f_{n}(x_{n},s+y_{n})|\,ds+\frac{1}{t}\int_{0}^{t+y_{n}-2^{n}}|f_{n}(x^{\prime}_{n},s)|\,ds\leq\\ \leq\frac{1}{t}\int_{0}^{2^{n}}|f_{n}(x_{n},s)|\,ds+\frac{1}{t}\int_{0}^{2^{n}}|f_{n}(x^{\prime}_{n},s)|\,ds\leq\frac{2\|f_{n}\|_{\infty,1}}{t}. Here we assumed that {t+y_{n}>2^{n}}, otherwise the factor 2 would not appear in the final estimate. For the terms of the first sum we

Read the complete text