Weyl's equidistribution theorem

[12/75%]
Show
LaTeX

We use the notation δ(x)=x−[x]\delta (x)=x-\left[x\right] for the fractional part of a real number xx. Let f∈C([0,1])f \in C(\left[0,1\right]) and x∈R\Qx \in \mathbb {R} \backslash \mathbb {Q}. Prove that

1n∑1⩽k⩽nf(δ(kx))→∫01f(t)dt as n→+∞. \frac{1}{n} \sum _{1 \leqslant k \leqslant n} f(\delta (k x)) \rightarrow \int _{0}^{1} f(t) d t \quad \text{ as } \quad n \rightarrow +\infty .
?

For which p>0p>0 do the series

?
(a)

∑cos⁡(2πn2)3np\sum \frac{\sqrt[3]{\cos (2 \pi n \sqrt{2})}}{n^{p}};

(b)

∑(−1)n∣sin⁡(2πn2)∣nP\sum (-1)^{n} \frac{\left|\sin (2 \pi n \sqrt{2})\right|}{n^{P}}

converge?

*) For a proof not using Fourier series, see the hint to problem I.2.40. Various generalizations of the Wirtinger inequality can be found in [KhLP] on pp. 221, 384.

Prove the convergence of the series

?
(a)

∑cos⁡(π2n2)3n\sum \frac{\sqrt[3]{\cos \left(\pi \sqrt{2} n^{2}\right)}}{n};

(b)

∑(−1)n∣sin⁡(π2n3)∣n\sum (-1)^{n} \frac{\left|\sin \left(\pi \sqrt{2} n^{3}\right)\right|}{n}.

The measure μ\mu is assumed to be finite and to satisfy the normalization condition μ(X)=1\mu (X)=1 (such measures are called probability measures). Let TT be a transformation of the set XX. A set A⊂XA \subset X is called invariant if T−1(A)=AT^{-1}(A)=A. A transformation TT preserving the measure μ\mu is called ergodic if for every invariant set A⊂XA \subset X one of the following two holds: either μ(A)=0\mu (A)=0, or μ(X\A)=0\mu (X \backslash A)=0.

Consider the irrational shift

S(x)=(x+α) mod 1(x∈[0,1)=X,α∈R\Q). S(x)=(x+\alpha ) \bmod 1 \quad (x \in [0,1)=X, \alpha \in \mathbb {R} \backslash \mathbb {Q}) .

It is ergodic by 2.21a). By the ergodic theorem, for any summable function ff on XX and for almost all x∈Xx \in X the equality

lim⁡1n∑0⩽k<nf(Sk(x))=∫01f(t)dt(1) \lim \frac{1}{n} \sum _{0 \leqslant k<n} f\left(S^{k}(x)\right)=\int _{0}^{1} f(t) d t \tag {1}

holds.

?
(a)

Prove that if f∈C([0,1])f \in C(\left[0,1\right]), then for the shift SS one can assert more, namely that equality (1) holds for every x∈[0,1)x \in [0,1). (First assume that f(0)=f(1)f(0)=f(1), and then consider the general case.)

(b)

Prove that if Δ=⟨a,b⟩⊂[0,1),χΔ(x)=1\Delta =\langle a, b\rangle \subset [0,1), \chi_{\Delta }(x)=1 for x∈Δx \in \Delta, χΔ(x)=0\chi_{\Delta }(x)=0 for x∉Δx \notin \Delta, then

lim⁡1n∑0⩽k<nχΔ(Sk(x))=b−a(2) \lim \frac{1}{n} \sum _{0 \leqslant k<n} \chi _{\Delta }\left(S^{k}(x)\right)=b-a \tag {2}

for all x∈[0,1)x \in [0,1).

If {Sk(x)}\left\{ S^{k}(x)\right\} is replaced by an arbitrary sequence {xk}\left\{ x_{k}\right\}, then property b) is called uniform distribution of this sequence (see [PS], [KN]).

For a uniformly distributed sequence {xk}\left\{ x_{k}\right\} and a function ff continuous on [0,1]\left[0,1\right], a relation analogous to (1) holds:

1n∑k=1nf(xk)⟶n→∞∫01f(t)dt \frac{1}{n} \sum _{k=1}^{n} f\left(x_{k}\right) \underset {n \rightarrow \infty }{\longrightarrow } \int _{0}^{1} f(t) d t

In particular,

1n∑k=1ne2πimxk⟶n→∞0 for any integer m≠0.  \frac{1}{n} \sum _{k=1}^{n} e^{2 \pi i m x_{k}} \underset {n \rightarrow \infty }{\longrightarrow } 0 \text{ for any integer } m \neq 0 \text{. }

This condition is not only necessary but also sufficient for uniform distribution. The latter statement is called Weyl's criterion (cf. problem I.3.9).

Remark.
?

The Birkhoff--Khinchin ergodic theorem (see [Bi], [KSF]): let TT be a transformation of a set XX preserving a probability measure μ\mu, and let ff be a measurable function such that ∫X∣f(x)∣dμ(x)<+∞\int_{X}\left|f(x)\right| d \mu (x)<+\infty. Then for almost every (with respect to the measure μ\mu) x∈Xx \in X the limit

lim⁡1n∑0⩽k<nf(Tk(x))=fˉ(x) \lim \frac{1}{n} \sum _{0 \leqslant k<n} f\left(T^{k}(x)\right)=\bar{f}(x)

exists, called the mean value of the function ff along the trajectory of the point xx. Moreover the function fˉ\bar{f} is invariant and ∫Xfˉ(x)dμ(x)=∫Xf(x)dμ(x)\int_{X} \bar{f}(x) d \mu (x)=\int_{X} f(x) d \mu (x). If TT is ergodic, then fˉ\bar{f} turns out to be a constant (see problem 2.20), equal to ∫Xf(x)dμ(x)\int_{X} f(x) d \mu (x) (i.e. the mean value of ff on XX).

A sequence x1,x2,x3,…,xn,…x_{1}, x_{2}, x_{3}, \ldots , x_{n}, \ldots, 0≤xn≤10 \leq x_{n} \leq 1, is equidistributed on [0,1]\left[0,1\right] if and only if the two equations

lim⁡n→∞cos⁡2πkx1+cos⁡2πkx2+⋯+cos⁡2πkxnn=0, \lim _{n \to \infty } \frac{\cos 2\pi k x_{1} + \cos 2\pi k x_{2} + \cdots + \cos 2\pi k x_{n}}{n} = 0, lim⁡n→∞sin⁡2πkx1+sin⁡2πkx2+⋯+sin⁡2πkxnn=0 \lim _{n \to \infty } \frac{\sin 2\pi k x_{1} + \sin 2\pi k x_{2} + \cdots + \sin 2\pi k x_{n}}{n} = 0

hold for every positive integer kk.

?
Remark.
?

Problem 137.

Let θ\theta be an irrational number. The numbers

xn=θn−[θn] x_{n} = \theta n - \left[\theta n\right]

are equidistributed on the interval [0,1]\left[0,1\right].

?

Let θ\theta denote an irrational number. Put εn=1\varepsilon_{n}=1 or εn=0\varepsilon_{n}=0 according as the integer next to nθn\theta is larger or smaller than nθn\theta. Let aa and dd, a≥0a \geq 0, d>0d > 0, stand for two integers. Then we find that

lim⁡n→∞∑k=0n−1εa+kdn=12. \lim _{n \to \infty } \frac{\sum _{k=0}^{n-1} \varepsilon _{a+kd}}{n} = \frac{1}{2} .
?

Let θ\theta be an irrational number, xn=nθ−[nθ]x_{n} = n\theta - \left[n\theta \right], n=1,2,3,…n = 1, 2, 3, \ldots and let α1,α2,α3,…,αn,…\alpha_{1}, \alpha_{2}, \alpha_{3}, \ldots , \alpha_{n}, \ldots be a monotone decreasing sequence of positive numbers whose sum diverges. Then we find for any properly integrable function f(x)f(x) on [0,1]\left[0,1\right] that

lim⁡n→∞α1f(x1)+α2f(x2)+⋯+αnf(xn)α1+α2+⋯+αn=∫01f(x) dx. \lim _{n \to \infty } \frac{\alpha _{1}f(x_{1}) + \alpha _{2}f(x_{2}) + \cdots + \alpha _{n}f(x_{n})}{\alpha _{1}+\alpha _{2}+\cdots +\alpha _{n}} = \int _{0}^{1} f(x) \, dx .
?

Suppose ff is a continuous function on R1,f(x+2π)=f(x)\mathbb {R}^{1}, f(x+2 \pi ) = f(x), and α/π\alpha / \pi is irrational. Prove that

lim⁡N→∞1N∑n=1Nf(x+nα)=12π∫−ππf(t)dt \lim _{N \rightarrow \infty } \frac{1}{N} \sum _{n = 1}^{N} f(x+n \alpha ) = \frac{1}{2 \pi } \int _{-\pi }^{\pi } f(t) d t

for every xx. Hint: Do it first for f(x)=eikxf(x) = e^{i k x}.

?

Suppose ff is a continuous function on R1\mathbb {R}^{1}, with period 1. Prove that

lim⁡N→∞1N∑n=1Nf(nα)=∫01f(t)dt \lim _{N \rightarrow \infty } \frac{1}{N} \sum _{n = 1}^{N} f(n \alpha ) = \int _{0}^{1} f(t) d t

for every irrational real number α\alpha. Hint : Do it first for

f(t)=exp⁡(2πikt),k=0,±1,±2,… f(t) = \exp (2 \pi i k t), \quad k = 0, \pm 1, \pm 2, \ldots
?

(Weyl's equi-distribution property). Let 0<α<10 < \alpha < 1 be an irrational number. Let μn(⋅)\mu_{n}(\cdot ) be the measure defined by μn(A)≡1n∑j=0n−1IA(jα mod 1),A∈B([0,1])\mu_{n}(A) \equiv \frac{1}{n} \sum_{j = 0}^{n-1} I_{A}(j \alpha \bmod 1), A \in \mathcal{B}([0,1]). Show that μn⟶d\mu_{n} \longrightarrow^{d} Uniform (0,1)(0,1).

?

Let g:R→Rg: \mathbb {R} \rightarrow \mathbb {R} be periodic with period 11, and uniformly continuous and integrable over [0,1][0,1]. Define Zn=g(X+(n−1)α),n≥1Z_{n} = g(X+(n-1) \alpha ), n \geq 1, where XX is uniform on [0,1][0,1] and α\alpha is irrational. Show that, as n→∞n \rightarrow \infty,

1n∑j=1nZj→∫01g(u)du a.s.  \frac{1}{n} \sum _{j = 1}^{n} Z_{j} \rightarrow \int _{0}^{1} g(u) d u \quad \text{ a.s. }
?

References

[МГЛ+04] Макаров Б.М., Голузина М.Г., Лодкин А.А., Подкорытов А.Н. Избранные задачи по вещественному анализу. 2-е изд., Невский Диалект; БХВ-Петербург, 2004.
[PóSz98] Pólya G., Szegő G. Problems and Theorems in Analysis. Springer, 1998.
[Rud76] Rudin W. Principles of Mathematical Analysis. 3rd ed., McGraw-Hill, 1976.
[Rud87] Rudin W. Real and Complex Analysis. 3rd ed., McGraw-Hill, 1987.
[AtLa06] Athreya K.B., Lahiri S.N. Measure Theory and Probability Theory. Springer, 2006.
[GrSt20] Grimmett G.R., Stirzaker D.R. Probability and Random Processes. 4th ed., Oxford University Press, 2020.