-2

I want to prove that the sequences $\{\frac{1}{2\pi}\sum_{k=1}^{n}\frac{1}{k}\}$ is equidistributed modulo $1$.

This question is for the following question: Accumulation points form a circle. It need to show that $\{\sum_{k=1}^{n}\frac{1}{k}\}$ is equidistributed modulo $2\pi$. I think Weyl’s criterion will work! Weyl's equidistributed criterion. The following are equivalent: $$\{x_n\}\quad\text{is equidistributed modulo 1};$$

$$\forall~ \text{continuous & 1-peridic} f: \quad\frac{1}{N}\sum_{n=1}^Nf(x_n)\rightarrow\int_0^1f ;$$

$$\forall~ k\in \mathbb Z^*:\quad \frac{1}{N}\sum_{n=1}^Ne^{2πikx_n}\rightarrow 0.$$

Riemann
  • 7,203
  • What is "equivalent modulo $1$"? – Gary Jan 08 '22 at 12:32
  • @Gary Means ${\frac{1}{2\pi}\sum_{k=1}^{n}\frac{1}{k}}$mod $1$ is dense in $[0,1)$ – Riemann Jan 08 '22 at 12:33
  • I mean "equivalent" not "equidistributed". The one you wrote in the Weyl criterion. – Gary Jan 08 '22 at 12:34
  • @Gary Just a typo, adn fixed it! – Riemann Jan 08 '22 at 12:36
  • Harmonic series is asymptotic to $\log{n}$ and ${\log{n}}$ is dense in $[0,1]$. – rtybase Jan 08 '22 at 13:00
  • 1
    @rtybase Thank you very much! – Riemann Jan 08 '22 at 13:06
  • 2
    @Riemann What?????????? No, equidistributed mod 1 does not mean just "dense mod 1". – David C. Ullrich Jan 08 '22 at 14:59
  • 1
    @DavidC.Ullrich is right. Dense is weaker than equidistributed according to most definitions. See for example the Weyl equidistribution theorem to see the usual definition of ``equidistributed.'' – Shannon Starr Jan 08 '22 at 15:35
  • 1
    @ShannonStarr According to "most" definitions? Can you give us a plausible reference where the two are the same? – David C. Ullrich Jan 08 '22 at 15:41
  • Why the plural "sequences"? That's only one sequence. – jjagmath Jan 08 '22 at 18:27
  • @DavidC.Ullrich I am agreeing with you. I am merely seconding your point. A couple references for Weyl's equidistribution theorem are:Dym and McKean "Fourier Series and Integrals, Sec 1.6, pp.54-56; Stein and Shakarchi, "Fourier Analysis: An Introduction," Sec4.2, pp.105-113.The Wikipedia article on equidistribution also agrees with you: https://en.wikipedia.org/wiki/Equidistributed_sequence – Shannon Starr Jan 08 '22 at 19:12
  • One more comment is that density versus equi-distribution roughly follows the difference between Poincare's recurrence theorem versus Liouville's theorem on volume preserving measures. A good reference for that is V.I Arnol'd's book on "Mathematical Methods of Classical Mechanics," for example. – Shannon Starr Jan 08 '22 at 20:53
  • @rtybase Yes, $s_n\sim\log(n)$ and ${\log(n)}$ is dense. How does that imply anything about ${s_n}$??? (I'm assuming you were using ${\cdot}$ for fractional part...) – David C. Ullrich Jan 08 '22 at 23:44
  • @rtybase Thm: Knowing that $a_j\sim b_j$ and knowing everything about ${a_j}$ tells you nothing whatever about ${b_j}$.

    Ok, that's a little colloquial. A precise statement that says the same thing: IOW If $(\alpha_j)$ and $(\beta_j)$ are any two sequences in $[0,1)$ then there exist $a_j$ and $b_j$ with $a_j\sim b_j$, ${a_j}=\alpha_j$ and ${b_j}=\beta_j$. Proof: $a_j=j+\alpha_j$, $b_j=j+\beta_j$.

    – David C. Ullrich Jan 09 '22 at 00:20
  • @rtybase (Cor $s_n\sim\log n$ does not imply that ${s_n}$ is dense.) – David C. Ullrich Jan 09 '22 at 00:26
  • @DavidC.Ullrich I guess I should have not used the "asymptotic" word. Harmonic series and $\log{n}$ are more than asymptotic. Here is another link. – rtybase Jan 09 '22 at 12:03

1 Answers1

-2

Have a look at this general fact:

If $\{a_n\}$ is a sequence such that $a_n\ge0$ with $a_n\to0$ and $\sum a_n\to\infty$ then $\{\sum a_n\}$ is equidistributed modulo 1.

Proof:

Let $a,b\in [0,1]$ assume $a< b$ then there exist $n_0\in \mathbb{N}$ such that $0\le a_n< b-a$ for all $n\ge n_0$.

Now let us define partial sum $s_n=\sum_{i=1}^na_i$ then $s_n\to \infty$ (increasing sequence).

Choose $M$ such that it is the least positive integer greater or equal to $s_{n_0}$, i.e. $M\in \mathbb{N}$ such that $(M-1)<s_{n_0}\le M$.

We know there exist $k\in \mathbb{N}$ such that $s_k > (M+a)$.Hence the set $\{k\in\mathbb{N}|s_k > (M+a)\}$ is non empty subset of $\mathbb{N}$

Choose $n_1=min\{k\in\mathbb{N}|s_k > (M+a)\}$, (existance by well ordering principle).

Now observe that $(M+a)<s_{n_1}<(M+b)$ and hence $a<\{s_{n_1}\}<b$ . (here ${s_{n_1}}$ denotes fractional part of $s_{n_1}$ ).

So the fractional part of $s_n$ is dense in $[0,1]$


Apply this to solve the above problem.

  • 1
    Nice fact! Thank you! – Riemann Jan 08 '22 at 14:12
  • 1
    This is not equidistribution – Bananach Jan 08 '22 at 14:30
  • Dear @Bananach, the definition is given here for 'equidistribution modulo 1'-https://en.m.wikipedia.org/wiki/Equidistributed_sequence#Equidistribution_modulo_1 – Noob mathematician Jan 08 '22 at 14:38
  • Dear @Bananach I agree that it is not equidistribution, but it is equiditribution modulo 1 which is defined differently . Please read the discussion in comments. And if a down vote makes you happy then it is fine. :) – Noob mathematician Jan 08 '22 at 14:41
  • 2
    NO, saying it's dense does not say it's equidistributed mod 1. That comment is wrong. Can you give us the actual definition? In fact not only is the proof incomplete, the general result you state is false. – David C. Ullrich Jan 08 '22 at 15:00
  • 1
    @Riemann nice or not, it's a false fact; the opinion expressed in another comment notwithstanding, "equidistributed mod 1" is not the same as "dense mod 1". – David C. Ullrich Jan 08 '22 at 15:11
  • 1
    @Noobmathematician Yes, the definition is given at that link! I don't see how you get the idea from that wikipedia article that equidistribution mod 1 is the same as density mod 1. Its says clearly that $(x_n)$ is equidistributed mod 1 if $(frac(x_n))$ is equidistributed on $[0,1)$; otoh the sequence is dense mod 1 if the fractional parts are dense in $[0,1)$. – David C. Ullrich Jan 08 '22 at 15:15
  • 1
    @Noobmathematician Let $L=[0,1/2)$, $R=[1/2,1)$. Say $(x_n)\subset[0,1)$ is a sequence dense in $[0,1)$ consisting of one element of $L$ followed by two elements of $R$, then one element of $L$ and two elements of $R$ etc.Then $(x_n)$ is dense in$[0,1)$ but not equidistributed, since (speaking loosely) only a third of terms lie in $L$. Since $frac(x_n)=x_n$ this says that the sequence is dense mod 1 but not equidistributed mod 1. – David C. Ullrich Jan 08 '22 at 15:28
  • 1
    (that wasn't intended as a counterexample to the result you state, just a counterexample to the notion that dense mod 1 implies equidistributed mod 1.) – David C. Ullrich Jan 08 '22 at 15:40
  • @DavidC.Ullrich I assumed the fact equidistribution mod 1 means density mod 1and said 'it is done!'. Apologies if that is wrong . – Noob mathematician Jan 09 '22 at 08:33
  • @DavidC.Ullrich why is this general fact false? Is the proof wrong? – Noob mathematician Jan 09 '22 at 08:33
  • yes, the proof is wrong, simply because you state equidistribution and prove just density – David C. Ullrich Jan 09 '22 at 11:13
  • Exercise: construct a counterexample to the claim. (The same sort of thing as in the example I gave, but a little more complicated...) – David C. Ullrich Jan 09 '22 at 11:14