1

This question is asking for a proof of a claim I don't understand (namely the pointwise convergence of the partial sums to a Fourier series). Alternatively, a completely different approach that's not too complicated is also fine (i.e. it doesn't use things like the Dirichlet kernel or the Lebesgue dominated convergence theorem or fatou's lemma, etc.).

Let $f : [-\pi, \pi] \to\mathbb{C}$ be continuous be given by $f(x) = 1-|x/\pi|$ for each $x\in [-\pi, \pi]$. Let $A_n = \frac{1}{2\pi} \int_{-\pi}^{\pi} f(x)e^{-inx}dx$. Show that $\sum_{|n|\leq N} A_n e^{in\theta}$ converges uniformly to $f$.

I know how to show that $S_n(\theta) := \sum_{|n|\leq N}a_ne^{in\theta}$ converges uniformly, and this is by showing that $\{S_n\}_{n\ge 1}$ is Cauchy under the supremum metric. Since each $S_n(\theta) \in C([-\pi, \pi],\mathbb{C})$ (i.e. is a continuous function from $[-\pi,\pi]$ to $\mathbb{C}$), it must converge to some $g \in C([-\pi, \pi],\mathbb{C})$ by the completeness of the metric space $(C([-\pi, \pi],\mathbb{C}), \lVert \cdot \rVert_\infty)$.

If I could just show that $S_n$ converges pointwise to $f$, then since $S_n$ converges uniformly to $g$, it converges pointwise to $g$, and since pointwise limits are unique, this would show that $f = g$.

More details are shown below. Feel free to point out any errors I could have made.

First, we compute the Fourier coefficients of $f'$ using integration by parts. Suppose the Fourier coefficients of $f$ are $a_n :=\frac1\pi \int_{-\pi}^{\pi} f(x)\cos (nx)dx$ and $b_n := \frac1\pi\int_{-\pi}^{\pi}f(x)\sin (nx)dx$. Then $\frac1\pi\int_{-\pi}^\pi f'(x)\cos (nx)dx = \frac1\pi([f(x)\cos (nx)]_{-\pi}^\pi + n\int_{-\pi}^\pi f(x)\sin (nx)dx) = nb_n$ and $\frac1\pi\int_{-\pi}^{\pi} f'(x)\sin (nx)dx = \frac1\pi([f(x)\sin (nx)]_{-\pi}^\pi -n\int_{-\pi}^\pi f(x)\cos(nx)dx) = -na_n.$ So the Fourier series of $f'$ is $\sum_{n\ge 0} (nb_n \cos (nx) - na_n\sin (nx)).$

Note that the Fourier coefficients of $f'$ are square-summable (this follows for arbitrary functions from Bessel's inequality). That is, if $a_n = \frac{1}{\pi}\int_{-\pi}^{\pi} f'(x)\cos nx dx = \frac{1}{\pi} \langle f'(x), \cos nx\rangle$ and $b_n = \frac{1}{\pi} \int_{-\pi}^{\pi} f'(x)\sin nx dx= \frac{1}{\pi} \langle f'(x), \cos nx\rangle$, then $\sum_{n\ge 0} |a_n|^2 + |b_n|^2 \leq \frac{1}{\pi} \lVert f'\rVert_2^2 < \infty$, where $\langle f, g\rangle := \int_{-\pi}^{\pi} f(x)\overline{g(x)}dx$ for complex valued functions f and g and $\lVert f\rVert_2 := \sqrt{\langle f, f\rangle}$ for a function f.

By the Cauchy Schwarz inequality, we have that $(\sum_{n\ge 1} |n\cdot a_n\cdot \frac{1}n|)^2\leq \sum_{n\ge 1} (n^2 a_n^2)\cdot \sum_{n\ge 1} \frac{1}{n^2}<\infty,$ so $\sum_{n\ge 1} |a_n| < \infty.$ Similarly, $\sum_{n\ge 1} |b_n| < \infty$.

Note that if $\frac{c_0}2 + \sum_{n\ge 1}(c_n \cos (nx) + d_n \sin(nx)),d_0 := 0,$ is the Fourier series of a function $f$ and $\sum_{n\ge 1} (|c_n|+|d_n|) <\infty$, then by the Weierstrass M-test, $\sum_{|n|\leq N} (c_n \cos (nx) + d_n \sin(nx))\to \sum_{n\in\mathbb{Z}} (c_n \cos (nx) + d_n \sin(nx))$ uniformly. This can be applied to $f'$ by setting $c_n = a_n$ and $d_n=b_n$ for all n.

user3472
  • 1,195
  • There is a general theorem stating that if $f$ is a function on $[-\pi,\pi]$ such that $f(-\pi)=f(\pi)$ and $f$ satisfies the Lipschitz condition ($|f(x_1)-f(x_2)|\le c|x_1-x_2|$), then the Fourier series is uniformly convergent to $f.$ The Lipschitz condition means that the slope of the graph of $f$ is bounded. For your function $a_n$ behaves like $O(n^{-2})$ and $b_n=0.$ So the uniform convergence follows straightforward. – Ryszard Szwarc May 13 '22 at 22:54
  • @RyszardSzwarc where can I find a proof of that general theorem? Also, is there a simpler proof for my question? – user3472 May 14 '22 at 21:54
  • 1
    Fourier series of continuous functions cannot converge pointwise except at the function (they may diverge at various points sure, but where they converge the sum is the function) this is basic result appwaring early in any book on Fourier series and easily proven with the Dirichlet kernel – Conrad May 16 '22 at 20:16
  • More concretely this follows from the fact that convergence at a point easily implies that the Caesaro limit is same as the limit and Feijer theorem gives the uniform convergence (to the function) of the Caesaro sums of the fourier series for a continuous function - so actually one uses the Feijer kernel – Conrad May 16 '22 at 20:31

0 Answers0