2
$\begingroup$

I am now studying Quantum Mechanics, and I am facing some issue about the convergence of Fourier Transform. I want to know when I can have the following equality: $$ \lim_{n \to \infty}(\mathcal{F}f_n)(\omega)= \mathcal{F}(\lim_{n \to \infty}f_n)(\omega) $$ Where $\mathcal{F}f(\omega)=\int_\mathbb{R}f(x)e^{-2\pi ix\omega}$. Both the limit on the LHS and RHS means convergent pointwise.

Of course, I can apply Lebesgue Dominated convergence theorem here when $f(x)e^{-2\pi ix\omega}$ is integrable. However, even if it is not integrable, the fourier transformation could exist. I try to find a theorem about whether we can swap lim and $\mathcal{F}$, but I have not get anything.

Let's take an example. Take $f_n(x)=\cos{x}$ when $x\in [-\frac{(2n-1)\pi}{2},\frac{(2n-1)\pi}{2}]$, and $f_n(x)=0$ otherwise. Then $f_n \to \cos$ pointwise. The fourier transform of $f_n$ is something like $(-1)^n\frac{2\cos(\frac{(2n-1)\pi}{2}x)}{x^2-1}$. But the Fourier transform of $\cos x$ is $\sqrt{\frac{\pi}{2}}\delta(\omega-1)+\sqrt{\frac{\pi}{2}}\delta(\omega+1)$. Clearly $\lim_{n \to \infty}(\mathcal{F}f_n)(\omega)= \mathcal{F}(\lim_{n \to \infty}f_n)(\omega)$ is not true in this case. Why it is not true?

Could anyone list some general theorems about this topic?

$\endgroup$
  • $\begingroup$ You might be able to argue it from distribution arguments coupled with dominated convergence. $\endgroup$ – Cameron Williams Aug 10 '18 at 1:58
  • $\begingroup$ You should specify what kind of limits you mean. And what is $t$ in your Fourier transform of $f_n$? $\endgroup$ – amsmath Aug 10 '18 at 2:36
  • $\begingroup$ @amsmath I have now revised my question. $\endgroup$ – Jethro Aug 12 '18 at 4:29
1
$\begingroup$

Let $g_R(x) = \chi_{[-R,R]}(x) \cos x.$ Then the Fourier transform can easily be calculated: $$ \widehat{g_R}(\xi) = \int_{-\infty}^{\infty} g_R(x) \, e^{-i \xi x} \, dx = \int_{-R}^{R} \cos x \, e^{-i \xi x} \, dx = \int_{-R}^{R} \frac12 \left( e^{ix} + e^{-ix} \right) \, e^{-i \xi x} \, dx \\ = \frac12 \int_{-R}^{R} \left( e^{i(1-\xi)x} + e^{-i(1+\xi)x} \right) \, dx = \frac12 \left[ \frac{1}{i(1-\xi)} e^{i(1-\xi)x} + \frac{1}{-i(1+\xi)} e^{-i(1+\xi)x} \right]_{-R}^{R} \\ = \frac12 \left[ \frac{1}{i(1-\xi)} e^{i(1-\xi)R} + \frac{1}{-i(1+\xi)} e^{-i(1+\xi)R} - \frac{1}{i(1-\xi)} e^{i(1-\xi)(-R)} - \frac{1}{-i(1+\xi)} e^{-i(1+\xi)(-R)} \right] \\ = \frac{1}{1-\xi} \frac{e^{i(1-\xi)R} - e^{-i(1-\xi)R}}{2i} + \frac{1}{1+\xi} \frac{e^{i(1+\xi)R} - e^{-i(1+\xi)R}}{2i} \\ = \frac{\sin (1-\xi)R}{1-\xi} + \frac{\sin (1+\xi)R}{1+\xi} $$ After a small rewrite we end up with $$\widehat{g_R}(\xi) = R \frac{\sin (\xi-1)R}{(\xi-1)R} + R \frac{\sin (\xi+1)R}{(\xi+1)R} = K_R(\xi-1) + K_R(\xi+1),$$ where $$K_R(\xi) = R \frac{\sin \xi R}{\xi R}.$$

As a function, $K_R$ does not converge as $R \to \infty,$ neither pointwise nor in $L^p.$ But as a distribution it does. For any $\phi \in C_c^\infty(\mathbb R)$ or $\mathscr S(\mathbb R)$ we have $$ \langle K_R, \phi \rangle = \int_{-\infty}^{\infty} R \frac{\sin \xi R}{\xi R} \phi(\xi) \, d\xi = \{ \text{set } \eta = \xi R \} = \int_{-\infty}^{\infty} \frac{\sin\eta}{\eta} \phi(\eta/R) \, d\eta \\ \to \int_{-\infty}^{\infty} \frac{\sin\eta}{\eta} \phi(0) \, d\eta = \int_{-\infty}^{\infty} \frac{\sin\eta}{\eta} \, d\eta \, \phi(0) = \pi \, \phi(0) = \langle \pi \, \delta, \phi \rangle, $$ i.e. $K_R \to \pi \, \delta$ as a distribution. Thus, $\widehat{g_R}(\xi) = K_R(\xi-1) + K_R(\xi+1) \to \pi \, \delta(\xi-1) + \pi \, \delta(\xi+1)$ as a distribution.

And since $$ \cos x = \frac12 (e^{ix} + e^{-ix}) = \frac{1}{2\pi} \int_{-\infty}^{\infty} \left( \pi \, \delta(\xi-1) + \pi \, \delta(\xi+1) \right) e^{i \xi x} \, d\xi $$ we have $$\widehat{\cos}(\xi) = \pi \, \delta(\xi-1) + \pi \, \delta(\xi+1).$$

Thus, as a distribution, $g_R \to \cos$ and $\widehat{g_R} \to \widehat{\cos},$ i.e. limit commutes with Fourier transform.

$\endgroup$
1
$\begingroup$

One simple criteria would be dominated convergence. That is, if $\{ f_n \}$ are functions such that $|f_n| \le g$ with $g\in L^1$, and if $\{ f_n \}$ converges pointwise to $f$, then what you wrote would hold. The Lebesgue dominated convergence theorem would give it.

Just having $f\in L^1$ is not enough to have dominated convergence if $\{ f_n \}$ only converges poinwise, is it?

$\endgroup$

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.