1
$\begingroup$

I am trying to prove that if $f\in\mathcal{C}^m([0,1])$, then the Fourier coefficients of $f$, $$\hat{f}(k)=\int_{0}^1f(x)e^{2\pi i k x}dx,$$ satisfy $$\hat{f}(k)=o\left(\frac{1}{k^m}\right).$$ I have tried to do it via integration by parts but I don't get anywhere.

  • 1
    It looks like you want to integrate by parts $m$ times, passing the derivative onto $f^{(l)}$ and integrating the exponential every time. Have you tried this?2017-01-09
  • 0
    Do you have a periodic boundary condition?2017-01-09

1 Answers 1

2

Integrating by parts with $\displaystyle u=f(x)$ and $\displaystyle v=\frac{e^{i2\pi kx}}{i2\pi k}$ yields

$$\int_0^1 f(x)e^{i2\pi kx}\,dx=\frac{f(1)-f(0)}{i2\pi k}-\frac{1}{i2\pi k}\int_0^1 f'(x)e^{i2\pi kx}\,dx $$


Repeated integration by parts with $\displaystyle u=f^{(n)}(x)$ and $\displaystyle v=\frac{e^{i2\pi kx}}{i2\pi k}$ reveals

$$\int_0^1 f(x)e^{i2\pi kx}\,dx=\sum_{n=0}^{m-1} (-1)^n\frac{f^{(n)}(1)-f^{(n)}(0)}{(i2\pi k)^{n+1}}+(-1)^m\frac{1}{(i2\pi k)^m}\int_0^1 f^{(m)}(x)e^{i2\pi kx}\,dx \tag 1$$


If $f$ is periodic, then all of the terms in the summation in $(1)$ are zero and we have

$$\int_0^1 f(x)e^{i2\pi kx}\,dx=(-1)^m\frac{1}{(i2\pi k)^m}\int_0^1 f^{(m)}(x)e^{i2\pi kx}\,dx$$


Since the Riemann-Lebesgue Lemma guarantees that $\lim_{k\to \infty}\int_0^1 f^{(m)}(x)e^{i2\pi kx}\,dx=0$, then

$$\int_0^1 f^{(m)}(x)e^{i2\pi kx}\,dx=o(1)$$

as $k\to \infty$ and we have the desired result.