3
$\begingroup$

I noticed an interesting numerical phenomennon \begin{equation} \dfrac{1}{n}\sum_{k=0}^{n-1} w_k \alpha_n^{km} \rightarrow 0 \text{ uniformly for all $m\in \{1, \dots, n-1\}$ when }n\rightarrow \infty, \, w_k\overset{iid}{\sim}\mathrm{Exp}(1). \end{equation} where $\alpha_n = e^{\tfrac{2\pi i}{n}}$.

This possibly should hold in probability, or in maybe even 'almost sure'. At best, I would be interested in rates, at worst just to understand how to give reasonable bounds.

Bellow I attach simulations for $n=100, n=1000$ sampling $N=10000$ times. From this data I check distribution of maximum of absolute value of the sum over $k$. Plots clearly show some simultaneous contraction for all $k$.

enter image description here enter image description here

$\endgroup$
3
  • $\begingroup$ I have just understood that if one considers imaginary or real part, then one gets the sum of independent random variables. It is easy to see that variance of real/imaginary part of the sum equals (2n)^{-1}. This is for fixed m, I wonder if it is difficult to obtain a joint bound. $\endgroup$ Commented yesterday
  • $\begingroup$ What are your plots, exactly -- formally described, as, say, (possibly pseudo-random) sets? $\endgroup$ Commented yesterday
  • $\begingroup$ Plots show distribution of maximum of absolute value of the above sum, where maximum is taken over m. I also tried to scale not with n^{-1} but with n^{-1/2} and it looked like 'almost correct rate' of contraction. $\endgroup$ Commented yesterday

1 Answer 1

5
$\begingroup$

$\newcommand\ep\varepsilon$Yes, these random variables (r.v.'s) converge to $0$ uniformly in $m\in[n-1]:=\{1,\dots,n-1\}$ almost surely (a.s.).

Indeed, the r.v. in question is a special case of $C_{n,m}+i S_{n,m}$, where $$C_{n,m}:=\frac1n\,\sum_{k=0}^{n-1}X_k\cos\frac{2\pi km}n,\quad S_{n,m}:=\frac1n\,\sum_{k=0}^{n-1}X_k\sin\frac{2\pi km}n,$$ where $X_1,X_2,\dots$ are i.i.d. r.v.'s with finite absolute fifth moment. Let also $Y_k:=X_k-EX_k$. Then $$C_{n,m}=\frac1n\,\sum_{k=0}^{n-1}Y_k\cos\frac{2\pi km}n,$$ because $\sum_{k=0}^{n-1}\cos\frac{2\pi km}n=0$ for $m\in[n-1]$. So, $$EC_{n,m}^2=\frac{EY_1^2}{n^2}\,\sum_{k=0}^{n-1}\cos^2\frac{2\pi km}n=O(1/n);$$ the constants in $O(\cdot)$ here and in what follows depend only on $E|Y_1|^5$.
So, by Rosenthal's inequality (see e.g. this), $$E|C_{n,m}|^5=O(1/n^{5/2}).$$ So, by the union and Markov inequalities, for all real $\ep>0$, $$P(\max_{m\in[n-1]}|C_{n,m}|>\ep) \le\sum_{m\in[n-1]} P(|C_{n,m}|>\ep) \le\sum_{m\in[n-1]} \frac{E|C_{n,m}|^5}{\ep^5} =O\Big(\frac{n^{-3/2}}{\ep^5}\Big).$$ So, by the Borel--Cantelli lemma, a.s. the events $\{\max_{m\in[n-1]}|C_{n,m}|>\ep\}$ occur only for finitely many $n$. That is, a.s. $$\max_{m\in[n-1]}|C_{n,m}|\to0$$ (as $n\to\infty$).

Quite similarly, a.s. $$\max_{m\in[n-1]}|S_{n,m}|\to0.$$

So, a.s. $$\max_{m\in[n-1]}|C_{n,m}+i S_{n,m}|\to0.\quad\Box$$

Remark: Actually, it would be enough to assume that $E|X_1|^p<\infty$ for some real $p>4$.

$\endgroup$
2
  • $\begingroup$ Top ! Thanks, I've never heard before of Rosenthal's inequality! Could you give a link maybe to a book, where I could find it ? I like your proof as it also gives an upper bound on the rate of convergence to zero, though the exact rate seems more difficult to get. $\endgroup$ Commented yesterday
  • 1
    $\begingroup$ @FedorGoncharov : The best link on the state of the art concerning Rosenthal's inequality for sums of independent zero-mean real-valued random variables that I can give is the second link given in this previous answer; see also references there. Strangely enough, I don't know a book reference to it. $\endgroup$ Commented yesterday

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .