I have been proving some facts about Rademacher functions:

I know that the Rademacher functions are: $r_n(t)=\operatorname{sign}(\sin{2^n\pi t})$

or, equivalently: $$r_m(t):=\left\{\begin{array}{cl}-1 & \mbox{if }t\in\displaystyle\bigcup_{k=1}^{2^{m-1}} \left(\frac{2k-1}{2^m},\frac{2k}{2^m}\right)\\&\\0,&\mbox{if }t\in\left\{\displaystyle\frac{k}{2^m}\,:\,k=1,2,\ldots,2^m\right\}\\&\\1,&\mbox{if }t\in\displaystyle\bigcup_{k=1}^{2^{m-1}}\left(\frac{2k-2}{2^m},\frac{2k-1}{2^m}\right)\end{array}\right.$$

I know the next properties:

$\triangleright$ If $n>m$ then $\displaystyle\int_{I_j^m} r_n=0$, where $I_j^m=\left[\displaystyle\frac{j}{2^m},\frac{j+1}{2^m}\right],0\leq j<2^m$.

$\triangleright$ If $n_1<n_2<\ldots<n_k$, then $\displaystyle\int r_{n_1}\ldots r_{n_k}=0$

$\triangleright$ $(r_n)_{n\geq 0}$ is an orthonormal system in $L_2([0,1])$ that is not basis.

I have just proved this inequality: $$\displaystyle\int_0^1\left|\displaystyle\sum_{i=1}^n a_i r_i(t)\right|^4\leq 3\left(\displaystyle\sum_{i=1}^n |a_i|^2\right)^2$$ where $a_1,\ldots, a_n$ are scalars (real or complex).

Now I need to show that the sequence $\left(\displaystyle\sum_{i=1}^n \frac{r_i(t)}{n}\right)$ converges t0 $0$ almost everywhere in $[0,1]$ using the last inequality (is assumed that is immediate, but I don't control the convergence a.e.). I thought to use $a_i=\frac{1}{n}$ or something similar, but I don't know.

Many thx.


Let $a_1, \ldots, a_n = n^{-1}$ then we have

$$\int_0^1 \left |\sum_{i = 1}^n \frac{r_i(t)}{n} \right |^4 \, \textrm{d}t \leqslant \frac{3}{n^2}.$$

Hence taking the limit $n \to \infty$ we obtain $$\int_0^1 \left |\sum_{i = 1}^n \frac{r_i(t)}{n} \right |^4 \, \textrm{d}t \to 0.$$

Now we can apply Fatou's lemma to deduce that

$$\int_0^1 \lim_{n \to \infty} \left| \sum_{i = 1}^n \frac{r_i(t)}{n} \right|^4 \, \textrm{d}t = 0.$$


$$\lim_{n \to \infty} \left| \sum_{i = 1}^n \frac{r_i(t)}{n} \right|^4 = 0 \text{ a.e..}$$

  • 1
    $\begingroup$ @ Jonas Teuwen: For $f_n(t)=|n^{-1}\sum_{i=1}^n r_i(t)|^4$, Fatou's lemma only gives $\int_0^1\liminf_{\,n\to\infty}f_n(t){\kern.4mm\rm d\kern.4mm}t = 0$ from which you can conclude that $\liminf_{\,n\to\infty}f_n = 0$ almost everywhere. Wherefrom do you conclude that $\liminf_{\,n\to\infty}f_n(t)< \limsup_{\,n\to\infty}f_n(t)$ holds only for $t$ in a set of measure zero? $\endgroup$ – The-unKnowN Sep 22 '13 at 15:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.