1

Suppose that f is a 2π-periodic function that satisfies the estimate \begin{equation} |f(x)-f(y)|\leqslant M|x-y|^\alpha \end{equation} for an 0< $\alpha$ <1

Show that $S_n(x)$ converges uniformly to f (x) for all real x. \begin{equation} S_n(x)=\int_0^{2\pi} f(y)\frac{sin((N+\frac{1}{2})(x-y))}{2\pi ~sin(\frac{1}{2}(x-y))} dy. \end{equation}

My confuse is, $S_n(x)$ should $\longrightarrow$ 0 when n $\longrightarrow$ $\infty$ by using Riemann–Lebesgue lemma. So what mistake that I made?

LYN
  • 149

1 Answers1

1

$$ f(x)-S_n(x) = \frac{1}{2\pi}\int_{x-\pi}^{x+\pi}\left[\frac{f(x)-f(y)}{\sin{\frac{1}{2}(x-y)}}\right]\sin((n+1/2)(x-y))dy $$ Because $|f(x)-f(y)| \le M|x-y|^{\alpha}$, the expression enclosed by square brackets is absolutely integrable. So $S_n(x)\rightarrow f(x)$ for all $x$ by the Riemann-Lebesgue Lemma. The uniformity of the Lipshitz condition leads to uniform convergence.

To show how the above gives uniform convergence of $S_n$ to $f$, let $\epsilon > 0$ be given. Then, for small enough $\delta$, the following estimate holds independent of $x$: $$ \left|\frac{1}{2\pi}\int_{x-\delta}^{x+\delta}\frac{f(x)-f(y)}{\sin(\frac{1}{2}(x-y))}\sin((n+1/2)(x-y))dy\right| \\ \le \frac{1}{2\pi}\int_{x-\delta}^{x+\delta}M\frac{|x-y|^{\alpha}}{|\sin\frac{1}{2}(x-y)|}dy \\ \le \frac{2M}{2\pi}\int_{x-\delta}^{x+\delta}|x-y|^{\alpha-1}dy \\ = \frac{4M}{2\pi}\int_{0}^{\delta}t^{\alpha-1}dt = \frac{4M}{2\pi\alpha}\delta $$ So, by choosing $\delta$ small enough, the above is bounded by $\frac{\epsilon}{2}$ for all $x$. To bound the remaining piece by $\epsilon/2$ for large $n$, choose a function $g$ that is $2\pi$ periodic, continuously differentiable and satisfies $$ \frac{1}{2\pi\sin(\delta/2)}\int_{\delta \le |y-x| \le \pi}|f(y)-g(y)|dy < \frac{\epsilon}{6} \tag{$\dagger$} $$ Then \begin{align} &\left|\frac{1}{2\pi}\int_{\delta \le |x-y| \le \pi}\left[\frac{f(x)-f(y)}{\sin\frac{1}{2}(x-y)}\right]\sin((n+1/2)(x-y))dy\right| \\ & \le \left|f(x)\frac{1}{2\pi}\int_{\delta\le|u| \le \pi}\frac{\sin((n+1/2)u}{\sin\frac{1}{2}u}du\right| \\ & + \left|\frac{1}{2\pi}\int_{\delta \le |x-y|\le \pi}(g(y)-f(y))\frac{\sin((n+1/2)(x-y))}{\sin\frac{1}{2}(x-y)}dy\right| \\ & + \left|\frac{1}{2\pi}\int_{\delta \le |x-y| \le \pi}\frac{g(y)}{\sin\frac{1}{2}(x-y)}\sin((n+1/2)(x-y)dy\right| \end{align} The function $f$ is uniformly bounded because of the Lipschitz condition. Therefore, by the Riemann-Lebesgue lemma, there exists $N_1$ large enough that the first term on the right is bounded uniformly in $x$ by $\frac{\epsilon}{6}$ for $n \ge N_1$. The second term is automatically bounded by $\epsilon/6$ because of $(\dagger)$. The third term can be uniformly bounded for large $N$ by $\epsilon/6$ after a single integration by parts, which is permitted because $g$ is a smooth approximating function: $$ \frac{1}{2\pi}\int_{\delta \le |x-y| \le \pi}\frac{g(y)}{\sin\frac{1}{2}(x-y)}\sin((n+1/2)(x-y)dy. \tag{$\dagger\dagger$} $$ (Integrating the $\sin$ term gives $1/(n+1/2)$, leading to a uniform bound on the above for large enough $n$.) For example, the integral over $\delta < |x-y| \le \pi$ is over $y \in [x+\delta,x+\pi]$ and $y\in [x-\pi,x-\delta]$. The part of the integral over $[x+\delta,x+\pi]$ becomes \begin{align} &\int_{x+\delta}^{x+\pi}\frac{g(y)}{\sin\frac{1}{2}(x-y)}\sin((n+1/2)(x-y))dy\\ = &-\left.\frac{g(y)}{\sin\frac{1}{2}(x-y)}\frac{\cos((n+1/2)(x-y))}{(n+1/2))}\right|_{y=x+\delta}^{x+\pi} \\ +&\frac{1}{n+1/2}\int_{x+\delta}^{x+\pi}\frac{\partial}{\partial y}\left(\frac{g(y)}{\sin\frac{1}{2}(x-y)}\right)\cos((n+1/2)(x-y))dy \end{align} Now you can see that $(\dagger\dagger)$ tends to $0$ as $n\rightarrow\infty$. The evaluation terms tend to $0$ because of the presence of $1/(n+1/2)$; the integral term on the right tends to $0$ even without the $1/(n+1/2)$. And the convergence is uniform in $x$. Using a smooth $g$ to approximate $f$ in the integral sense simplifies the argument.

Disintegrating By Parts
  • 87,459
  • 5
  • 65
  • 149
  • not sure about your last sentence, how do you show that the Fourier coefs of $h_x(y) = \frac{f(x)-f(y)}{\sin((x-y)/2)}$ tend $\to 0$ uniformly ? – reuns Jul 22 '16 at 17:08
  • @user1952009 : For $\epsilon > 0$, there exists $\delta > 0$ such that $\int_{x-\delta}^{x+\delta}\left|\frac{f(x)-f(y)}{\sin(\frac{1}{2}(x-y))}\right|dy <\epsilon$ holds uniformly in $x$. That's due to your Liptschitz condition, and is a uniform Dini condition. – Disintegrating By Parts Jul 23 '16 at 00:11
  • @user1952009 : I added more details for you about the uniform convergence. – Disintegrating By Parts Jul 23 '16 at 02:49
  • Thank you for your answers. But I think there might be sth wrong with your

    $$ S_n(x)-f(x) = \frac{1}{\pi}\int_{x-\pi}^{x+\pi}\left[\frac{f(x)-f(y)}{\sin{\frac{1}{2}(x-y)}}\right]\sin((n+1/2)(x-y))dy $$ (I don't know why it cannot display as a formula here.) and your $\frac{1/2|x-y|}{1/2 sin|x-y| } \le 1(but~ you\geqslant 1)$

    – LYN Jul 23 '16 at 16:36
  • @Gatsby : doesn't display because you are using sin with {} to enclose the argument. If you are referring to $(\dagger)$, then I'm integrating over $\pi \ge |x-y| \ge \delta$, where $\sin(\frac{1}{2}(x-y)) \ge \sin(\delta/2)$. If it's in the Lipschitz estimate, I added a 2, and there is a $\delta$ small enough so that $1/|\sin(\cdots)| \le 2/|x-y|$ for $|x-y| < \delta$ and $\delta$ small enough. – Disintegrating By Parts Jul 23 '16 at 18:37
  • I mean the first row of your formula: you take f(x) from the outside to the numerator of the integral but this operation might not true. Second, from $\frac{1}{\pi}\int_{x-\delta}^{x+\delta}M\frac{|x-y|^{\alpha}}{|\sin\frac{1}{2}(x-y)|}dy$ to $\frac{2M}{\pi}\int_{x-\delta}^{x+\delta}|x-y|^{\alpha-1}dy$, you assume $sin(x)\ge x$ but in fact, $sin(x) \le x$ – LYN Jul 24 '16 at 01:19
  • I think the right operation in the first row of your answer should be this: $S_n(x)-f(x) = \frac{1}{2\pi}\int_{x-\pi}^{x+\pi}\frac{f(x)\sin(\frac{1}{2}(x-y))-f(y)\sin((n+1/2)(x-y))dy}{\sin(\frac{1}{2}(x-y))}$ – LYN Jul 24 '16 at 01:28
  • @Gatsby. This is the standard trick: $\frac{1}{2\pi}\int_{x-\pi}^{x+\pi}\frac{\sin((n+1/2)(x-y))}{\sin(\frac{1}{2}(x-y)}dy=1$. So you can multiply by $f(x)$ and move it inside along with $f(y)$. I just noticed I have $1/\pi$ outside all the integrals and that should be $1/2\pi$, but that error does not change the arguments. – Disintegrating By Parts Jul 24 '16 at 01:41
  • @Gatsby : I changed the $\frac{1}{\pi}$ factors in my arguments to $\frac{1}{2\pi}$. I should have caught that earlier. If you let $D_n(x-y)=\frac{1}{2\pi}\frac{\sin((n+1/2)(x-y)}{\sin\frac{1}{2}(x-y)}$ you have $\frac{1}{2\pi}\int_{x-\pi}^{x+\pi}D_N(x-y)dy = 1$ which gives $$ S_n(x)-f(x)=\int_{x-\pi}^{x+\pi}f(y)D_n(x-y)dy-f(x)\int_{x-\pi}^{x+\pi}D_n(x-y)dy \=\int_{x-\pi}^{x+\pi}D_n(x-y){f(y)-f(x)}dy.$$ That's the standard trick for convolution type kernels with unit mass. – Disintegrating By Parts Jul 24 '16 at 01:52
  • @TrialAndError Thank you so much for your detailed replies! And now I understand much better, but I still have a doubt on the g(y) that you choose, because I think f(y) is continous due to Lipshitz Condition and thus has primitive function F(y). So do you need to choose a new g(y)? Another doubt is that I cannot get 1/ (n+1/2) after integrations by part from your last equation. So could you add more steps on it? – LYN Jul 26 '16 at 03:54
  • @Gatsby : I add more for you. This should be enough to address your concerns. – Disintegrating By Parts Jul 26 '16 at 04:24
  • @TrialAndError Now I totally understand your answer, thank you so much for your helps and supports!! – LYN Jul 26 '16 at 15:46
  • @Gatsby : Good, glad I could help. You can check the checkmark next to my post to accept it as an answer. – Disintegrating By Parts Jul 26 '16 at 16:47
  • @TrialAndError OK!! I have done it – LYN Jul 27 '16 at 02:42