3

My question stems from Exercise $8$, in Chapter $2$ of Stein and Shakarchi's Fourier Analysis. I have verified that $$\frac{1}{2i} \sum_{n\ne 0} \frac{e^{inx}}{n} = \sum_{n\ge 1} \frac{\sin nx}{n}$$ is the Fourier series of the $2\pi$-periodic saw-tooth function defined by $f(0) = 0$, and $$f(x) = \begin{cases} \frac{-\pi-x}{2} & -\pi < x < 0\\ \frac{\pi - x}{2} & 0 < x < \pi\end{cases}$$ enter image description here

The book says:

Note that this function is not continuous. Show that nevertheless, the series converges for every $x$ (by which we mean, as usual, that the symmetric partial sums of the series converge). In particular, the value of the series at the origin, namely $0$, is the average of the values of $f(x)$ as $x$ approaches the origin from the left and the right.

By Dirichlet's test for convergence, I have shown that $\sum_{n\ge 1} \frac{\sin nx}{n}$ converges for all $x\in (-\pi,\pi) \setminus\{0\}$. At $x = 0$, $\sum_{n\ge 1} \frac{\sin nx}{n}$ clearly converges to $0$. So, $\sum_{n\ge 1} \frac{\sin nx}{n}$ converges for all $x\in (-\pi,\pi)$.

Question: The series $\sum_{n\ge 1} \frac{\sin nx}{n}$ converges to $f$, at $0$. Does it also converge to $f$ at other points in $(-\pi,\pi)$? If yes, how can we show this with elementary methods?

Please note that this is only the second chapter of Stein and Shakarchi's Fourier Analysis, so the machinery we can use is limited. In particular, we can't use Dirichlet conditions, etc.


Related questions: Post 1.


Edit(s):

  1. Integrating the identity suggested by @mathcounterexamples.net from $0$ to $x$, we get $$\sum_{n=1}^N \frac{\sin nx}{n} = -\frac{x}{2} + \frac{\sin Nx}{N} + \int_0^x \sin Nx \cot\frac{x}{2}\, dx$$ I have trouble evaluating the last integral, and don't know how to proceed. For $0 < x < \pi$, we want $$\left|\sum_{n=1}^N \frac{\sin nx}{n} + \frac{x-\pi}{2}\right| =\left|-\frac{\pi}{2} + \frac{\sin Nx}{N} + \int_0^x \sin Nx \cot\frac{x}{2}\, dx\right| \xrightarrow{N\to\infty} 0$$ and for $-\pi < x < 0$, we want $$\left|\sum_{n=1}^N \frac{\sin nx}{n} + \frac{x+\pi}{2}\right| =\left|\frac{\pi}{2} + \frac{\sin Nx}{N} + \int_0^x \sin Nx \cot\frac{x}{2}\, dx\right| \xrightarrow{N\to\infty} 0$$
  • I don't understand the premise of your question. You have shown the series converges everywhere, by Dirichlet's test, answering the question in Stein and Shakarchi. If you want to show that it actually converges to f, then that's a different question that I don't think is easy to answer just with the tools in chapter 1. – operatorerror Dec 16 '21 at 22:45
  • @operatorerror Oh, okay, I see! I wanted to show convergence of the Fourier series to $f$, and I know that it follows from the fact that $f$ is differentiable - but I was hoping to see a more elementary approach; perhaps just appealing to introductory real analysis. – stoic-santiago Dec 17 '21 at 06:03
  • Functions don't need to be bounded to be integrable, fyi. Example: $x^{-\frac 1 2}$ on $[0,1]$ is unbounded but the (improper) integral converges. – Alan Dec 27 '21 at 06:31
  • @Alan I beg to differ. Kindly see: https://math.stackexchange.com/questions/610054/if-a-function-fx-is-riemann-integrable-on-a-b-is-fx-bounded-on-a – stoic-santiago Dec 27 '21 at 06:33

3 Answers3

3

One way is to start from Lagrange's trigonometric identity

$$\sum_{n=1}^N \cos nx = -\frac{1}{2} + \frac{\sin \left(\frac{(N+1)x}{2}\right)}{\sin \left(\frac{x}{2}\right)}$$ to integrate it and to justify the swap of $\int$ and $\sum$ symbols as $N \to \infty$.

For instance, for $x \in (0, \pi]$ we have

$$\begin{aligned} \sum_{n=1}^N \frac{\sin nx }{n} &= \int_{\pi}^x\sum_{n=1}^N \cos nx \ dx= \frac{\pi - x}{2} + \int_{\pi}^x\frac{\sin \left(\frac{(N+1)x}{2}\right)}{\sin \left(\frac{x}{2}\right)} \ dx\\ &=\frac{\pi - x}{2} + 2\int_{\frac{\pi}{2}}^{\frac{x}{2}}\frac{\sin \left((N+1)u\right)}{\sin u} \ du \end{aligned}$$

Now using the Riemann-Lebesgue Lemma you get

$$\lim\limits_{N \to \infty} \int_{\frac{\pi}{2}}^{\frac{x}{2}}\frac{\sin \left((N+1)u\right)}{\sin u} \ du = 0$$ and therefore the desired result.

Note: see at the bottom of the reference for easy proof of Riemann-Lebesgue lemma using integration by parts which is justified in our case.

  • 2
    Please see my edit - I am not able to complete the solution with this hint. – stoic-santiago Dec 16 '21 at 06:05
  • @delta-divine See an update of my answer using Riemann-Lebesgue lemma. – mathcounterexamples.net Dec 21 '21 at 10:15
  • To apply the Riemann-Lebesgue lemma, you need $\frac{1}{\sin x}$ to be integrable in $(0, \pi/2)$. However, $\frac{1}{\sin x}$ is not even bounded in this interval, let alone integrable. The lemma does not apply, from what I can see. Please clarify. – stoic-santiago Dec 23 '21 at 07:35
  • @delta-divine No. I need $\frac{1}{\sin x}$ to be integrable on $[x/2,\pi/2]$ with $x \in (0,\pi]$ as I mentioned. Which is fine as $\sin x$ doesn’t vanish on $[x/2, \pi/2]$. See the interval of integration that I use. – mathcounterexamples.net Dec 23 '21 at 08:06
  • This may be too elementary, but how does this agree with the statement of the Riemann Lebesgue lemma in my post? Could you please elaborate on how $\hat f(n) \xrightarrow{n\to\infty} 0$ implies what we need? – stoic-santiago Dec 23 '21 at 08:18
  • @delta-divine Just have a look at the last equality of my answer and what happens when $N \to \infty$. – mathcounterexamples.net Dec 23 '21 at 08:40
1

Hoping that this could help.

Similar to what you wrote in your own answer, using $$\sum_{n=1}^p \cos(nt)=\csc \left(\frac{t}{2}\right) \sin \left(\frac{p }{2}t\right) \cos \left(\frac{p+1}{2} t\right)$$ the integration gives $$-\frac 12t+\frac{i\,e^{-i p t}}{2 p (p+1)}\Big[(p+1) \, _2F_1\left(1,-p;1-p;e^{i t}\right)+p e^{i (2 p t+t)} \, _2F_1\left(1,p+1;p+2;e^{i t}\right) \Big]$$ which can also write in terms of the incomplete beta function $$-\frac 12t-\frac{1}{2} i \left(B_{e^{i t}}(-p,0)-B_{e^{i t}}(p+1,0)\right)$$ So, the definite integral (from $t=0$ to $t=x$) is $$-\frac 12x-\frac{1}{2} i\Big[\psi(-p)+B_{e^{i x}}(-p,0)-\psi (p+1)-B_{e^{i x}}(p+1,0) \Big]$$

1

The @mathcounterexamples.net's answer is good, but there is a little bug in the first formula. Actually, $$ \frac 12 + \sum\limits_{k=1}^n \cos kx = \frac{\sin\big(\big(n+\frac 12\big)x\big)}{2\sin \frac x2}, $$ so for $0<x<\pi$ we have $$ \sum\limits_{k=1}^n \frac{\sin kx}k = \frac{\pi - x}2 - \int\limits_{x/2}^{\pi / 2} \frac{\sin(2n+1)u}{\sin u}\,du. $$ The solution could be made more elementary if we avoid using Riemann-Lebesgue Lemma. In order to show that the last integral tends to $0$ as $n\to\infty$, it is enough to integrate by parts: $$ \int\limits_{x/2}^{\pi / 2} \frac{\sin(2n+1)u}{\sin u}\,du = -\frac 1{2n+1} \cdot\left.\frac{\cos(2n+1)u}{\sin u}\right|^{\pi/2}_{x/2} - \frac 1{2n+1}\int\limits_{x/2}^{\pi / 2} \frac{\cos(2n+1)u}{\sin^2 u}\cos u\,du. $$ All functions here are bounded since $0 < \frac1{\sin u} \leq \frac 2x$, so the limit is $0$.

  • Won't it be $$ \frac 12 + \sum\limits_{k=1}^n \cos kx = \frac{\sin\big(\big(n+\frac 12\big)x\big)}{2\sin \frac x2}, $$ instead? – stoic-santiago Dec 26 '21 at 20:24
  • Another typo: $$\int\limits_{x/2}^{\pi / 2} \frac{\sin(2n+1)u}{\sin u},du = \frac 1{2n+1} \cdot\left.\frac{\cos(2n+1)u}{\sin u}\right|{\pi/2}^{x/2} + \frac 1{2n+1}\int\limits{x/2}^{\pi / 2} \frac{\cos(2n+1)u}{\sin^2 u}\cos u,du$$ Other than that - looks good! I will accept your answer once you make the necessary edits. – stoic-santiago Dec 26 '21 at 20:31
  • I think the last one is not a typo. Three minuses produce minus. – Sergei Lytkin Dec 27 '21 at 06:29
  • Except there are only two minuses. You have written the first term after reversing the limits of integration. – stoic-santiago Dec 27 '21 at 06:31
  • Actually, it doesn't matter. Anyway, now I rewrote it in usual form without reversing. – Sergei Lytkin Dec 27 '21 at 06:33