[Math] Prove uniform convergence of a Fourier series

fourier analysisfourier seriesintegrationreal-analysisuniform-convergence

Suppose $f$ is continuous and $f \sim a_0 + \sum_{n=1}^{\infty}a_n \cos(nx) + b_n \sin(nx)$. Furthermore suppose the series $a_0 + \sum_{n=1}^{\infty}a_n \cos(nx) + b_n \sin(nx)$ converges uniformly. Show that $f$ is the uniform limit of this fourier series.

My attempt: Suppose $g$ is the uniform limit of $\sum_{n=1}^{\infty}a_n \cos(nx) + b_n \sin(nx)$, and it suffices to show that $f$ and $g$ has the same fourier coefficient, i.e. $a_n(f) = a_n(g)$ and $b_n(f) = b_n(g)$. If we define $T_N = a_0 + \sum_{n=1}^{N}a_n \cos(nx) + b_n\sin(nx)$, it is easy to verify that for any $n \le N$, the $n$-th Fourier coefficient of $T_N$ agrees with $a_n(f)$ and $b_n(f)$, but how to use the uniformly convergence condition to show that $a_n(g) = a_n(f), b_n(g) = b_n(f)$? Thanks!

Best Answer

The easy part is to show that $f$ and $g$ have the same Fourier coefficients. The hard part is to show that $f$ and $g$ are identically equal.

First show that $f$ and $g$ have the same Fourier series.

From periodicity of the series, we can consider any interval $[\alpha, \alpha+ 2\pi]$, and adopting standard notation replace $a_0$ with $a_0/2$.

We have

$$\tag{1}g(x) = a_0/2 + \sum_{n=1}^\infty (a_n \cos nx + b_n \sin nx)$$

where the series is uniformly convergent on the interval, and, since each term in the series is continuous, it follows from uniform convergence that $g$ is continuous.

Multiplying by $\sin mx$ and integrating we get

$$\tag{2}\frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}g(x) \sin mx \,dx \\= \frac{a_0}{2\pi}\int_{\alpha}^{\alpha + 2\pi}\sin mx \, dx +\frac{1}{\pi}\int_{\alpha}^{\alpha + 2\pi}\sum_{n=1}^\infty(a_n \cos nx \sin mx + b_n \sin nx \sin mx) \, dx.$$

Since (1) is uniformly convergent and $\sin mx$ is bounded, the series on the RHS of (2) is uniformly convergent and can be integrated term by term to obtain

$$\tag{3}\frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}g(x) \sin mx \,dx \\= \frac{a_0}{2\pi}\int_{\alpha}^{\alpha + 2\pi}\sin mx \, dx +\sum_{n=1}^\infty\frac{a_n}{\pi}\left(\int_{\alpha}^{\alpha + 2\pi}\cos nx \sin mx \, dx \right) + \frac{b_n}{\pi} \left(\int_{\alpha}^{\alpha + 2\pi} \sin nx \sin mx \, dx \right)$$

All the integrals on the RHS of (3) vanish except the integral of $\sin nx \sin mx$ when $n = m$.

Hence,

$$\frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}g(x) \sin mx \,dx = b_m = \frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}f(x) \sin mx \,dx $$.

Similarly we can show

$$\frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}g(x)\,dx = a_0 = \frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}f(x)\,dx , \\ \frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}g(x) \cos mx \,dx = a_m = \frac{1}{\pi} \int_{\alpha}^{\alpha + 2\pi}f(x) \cos mx \,dx. $$

Thus $f$ and $g$ have the same Fourier series.

It still remains to prove that $f = g$.

This follows by showing that if two continuous functions differ at just one point, $f(c) \neq g(c)$, then they cannot have the same Fourier series.

Take $h = f -g$. If $f$ and $g$ have identical Fourier series, then all Fourier coefficients of $h$ vanish, and for any trigonometric polynomial $T_m(x) = A_0/2 + \sum_{n=1}^{m} (A_n \cos nx + B_n \sin nx)$ we have for any $\alpha \in \mathbb{R}$,

$$\tag{4}\int_{\alpha}^{\alpha + 2\pi} h(x) T_m(x) \, dx = 0.$$

We also have $h(c) = f(c) - g(c) \neq 0$ and WLOG can assume that $h(c) > 0$. Since $h$ is continuous , there exists $K > 0$ and $\delta > 0$ such that $h(x) \geqslant K > 0$ when $x \in [c - \delta,c + \delta]$.

It can be shown that $T_m(x) = [1 + \cos(x-c) - \cos \delta]^m$ is a trigonometric polynomial satisfying

$$T_m(x) \geqslant 1 \text{ for } x \in [c-\delta,c+\delta] \\ \lim_{m \to \infty} T_m(x) = \infty \text{ uniformly on } [c - \delta/2, c+\delta/2] \\ |T_m(x)| \leqslant 1 \text{ for } x \in [c + \delta , c - \delta + 2\pi] $$

From these properties it follows that

$$\tag{5} \int_{c - \delta}^{c + \delta} h(x) T_m(x) \, dx \geqslant \int_{c - \delta/2}^{c + \delta/2} h(x) T_m(x) \, dx \geqslant \delta \, K \, \inf T_m(x) $$

and $\int_{c + \delta}^{c - \delta + 2\pi} h(x) T_m(x) \, dx$ is bounded.

Since the RHS of (5) tends to $\infty$ as $m \to \infty$ it follows for sufficiently large $m$

$$\int_{c - \delta}^{c - \delta + 2\pi} h(x) T_m(x) \, dx \neq 0,$$

contradicting (4) and leading to the conclusion that if $f$ and $g$ differ at one point, they cannot have the same Fourier series.

Related Question