Let $\varepsilon>0$. I was interested in understanding the justification of defining the following function $\phi$ via its Fourier transform, satisfying the following properties:

(1) $\widehat{\phi}\in C^\infty(\mathbb{R})$

(2) $\widehat{\phi}(\xi)=1$ for $\xi\in[-\pi+\varepsilon,\pi-\varepsilon]$

(3) supp$(\widehat{\phi})\subset[-\pi-\varepsilon,\pi+\varepsilon]$

(4) $\widehat{\phi}$ goes from 1 to 0 monotonically on the intervals $[-\pi-\varepsilon,-\pi+\varepsilon]$ and $[\pi-\varepsilon,\pi+\varepsilon]$.

(5) $\underset{n\in\mathbb{Z}}\sum\widehat{\phi}(\xi-2\pi n)=1$ for every $\xi\in\mathbb{R}$.

The Fourier Transform convention used was $$\widehat{f}(\xi)=\int_\mathbb{R}f(x)e^{-ix\xi}dx.$$ I am sure there is no problem with it, but I don't quite understand how the construction goes. I would appreciate any explanation. Thank you very much!

  • $\begingroup$ Do you understand how to do the construction if you do not have to satisfy property 5 necessarily? $\endgroup$ – Ian Dec 13 '15 at 22:18
  • $\begingroup$ No, I am not quite sure how that construction works either... $\endgroup$ – Johnny T. Dec 14 '15 at 1:45
  • 1
    $\begingroup$ Look up bump functions and approximate identities. Basically, the function that you want (property 5 aside) is the convolution of an indicator function with an appropriate bump function. $\endgroup$ – Ian Dec 14 '15 at 2:01
  • $\begingroup$ Related $\endgroup$ – Giuseppe Negro Dec 16 '15 at 16:40

In this answer I will carry out in detail the suggestions in Ian's comments.

Instead of constructing $\phi$, construct its Fourier transform $\Phi$. Let $$ b(\xi)=\begin{cases}c\,e^{-\tfrac{1}{1-\xi^2}} &\text{if }|\xi|\le1,\\0&\text{if }|\xi|>1, \end{cases}$$ where $c>0$ is chosen so that $\int_{-1}^1b(\xi)\,d\xi=1$. Next, let $\psi$ be the convolution of the characteristic function of the interval $[-\pi,\pi]$ with $\dfrac1\epsilon\,b\Bigl(\dfrac{\xi}{\epsilon}\Bigr)$. Then $\psi$ satisfies properties (1)-(4) as a function of $\xi$. Finally, observe that $\sum_{n\in\mathbb{Z}}\psi(\xi-2\pi n)$ is well defined, since only a finite number of summands are $\ne0$, and $>0$. Then define $$ \Phi(\xi)=\frac{\psi(\xi)}{\sum_{n\in\mathbb{Z}}\psi(\xi-2\pi n)} $$ and $\phi(x)$ as the inverse Fourier transform of $\Phi$.

  • $\begingroup$ Thank you very much for your answer. There are two things I am still wondering about. (1) I am still wondering how to show $\psi$ is smooth? (2) $\Phi$ no longer has value $1$ in the interval $[-\pi+\epsilon, \pi - \epsilon]$. How can we get around that? Thank you ! $\endgroup$ – Johnny T. Dec 17 '15 at 6:22
  • $\begingroup$ 1) $\psi$ is the convolution of an integrable function with a $C^\infty$ function of compact support. By the properties of convolution, $\psi$ is smooth. 2) If $|\xi|<\pi-\epsilon$, then $\psi(\xi)=1$ and $\psi(\xi-2\,n\,\pi)=0$ for $n\ne0$. $\endgroup$ – Julián Aguirre Dec 17 '15 at 10:13
  • $\begingroup$ Thank you very much for the explanation! I get it now! $\endgroup$ – Johnny T. Dec 19 '15 at 18:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.