3

Consider the Fourier transform $\mathcal F: L^p(\Bbb R) \to L^q(\Bbb R)$ for $1 < p < 2$. Is this map surjective? I suspect it isn't; if one had to extrapolate the behavior on $L^1(\Bbb R)$ (see this post) but I couldn't find any explanations after looking around for a bit. Can we use the non-surjectivity of $\mathcal F: L^1(\Bbb R) \to C_0(\Bbb R)$ somehow?

Another direction uses the open-mapping theorem: if $\mathcal F: L^p(\Bbb R) \to L^q(\Bbb R)$ is surjective, then it is an open map. We want to show that this is not possible.


Edit: Thanks to @Severin, I came across this answer on MO. I'm not sure how they show $||\hat f_\lambda||_{q} \leq c \lambda^{1/q-1/2}$ though.

1 Answers1

2

One way to see it is to look at the formula for the Fourier transform $\hat{f_\lambda}(\xi)$: $$ \hat{f_\lambda}(\xi) = \int f_\lambda(x) e^{-2\pi ix\xi}\,dx = \int e^{-\pi x^2} e^{i\pi(\lambda x^2 -2x\xi)}\,dx. $$ The phase function $\phi_\xi(x) = \lambda x^2-2x\xi$ has a critical point at $x = \frac{\xi}{\lambda}$. Expanding $\phi_\xi(\frac{\xi}{\lambda}+h) = \phi_\xi(\frac{\xi}{\lambda}) + \frac{h^2}{2}\phi_\xi''(\frac{\xi}{\lambda}) = -\frac{\xi^2}{\lambda}+\lambda h^2$, we have $$ \hat{f_\lambda}(\xi) = e^{-i\pi \frac{\xi^2}{\lambda}}\int e^{-\pi (\frac{\xi}{\lambda}+h)^2} e^{i\pi\lambda h^2}\,dh. $$ If $|h|\gg \frac{1}{\sqrt \lambda}$, then $e^{i\pi\lambda h^2}$ will be oscillating very fast, so the contribution to the integral appearing above from $|h|\gg \frac{1}{\sqrt \lambda}$ will be very small. When $|\xi|\gg\lambda$ and $|h|\lesssim \frac{1}{\sqrt{\lambda}}$, the amplitude $e^{-\pi(\frac{\xi}{\lambda}+h)^2}$ will be very small. While when $|\xi|\lesssim \lambda$ and $|h|\lesssim \frac{1}{\sqrt{\lambda}}$, the integrand $e^{-\pi (\frac{\xi}{\lambda}+h)^2} e^{i\pi\lambda h^2}$ will be approximately $1$.

Putting these observations together, the integral $\int|\hat{f_\lambda}(\xi)|^{q}\,d\xi$ is approximately \begin{align*} \int|\hat{f_\lambda}(\xi)|^{q}\,d\xi\approx \int_{|\xi|\lesssim \lambda}|\int_{|h|\lesssim\frac{1}{\sqrt\lambda}}1\,dh|^{q}\,d\xi \lesssim \lambda^{1-\frac{q}{2}}. \end{align*} Taking the $q$th root gives $\|\hat{f_\lambda}\|_{q} \le c \lambda^{\frac{1}{q}-\frac{1}{2}}$.

Alex Ortiz
  • 26,211
  • I appreciate the answer - thanks! Is there perhaps an elementary way that uses lesser knowledge of oscillatory integrals? – stoic-santiago Dec 01 '23 at 05:51
  • 1
    @stoic-santiago, since the function $f_\lambda$ is a Gaussian, its Fourier transform is also a Gaussian (albeit a complex one). You could try playing with the formula of the Fourier transform of a Gaussian to estimate the $L^q$ norm. That said, to fill in this proof sketch, I think it wouldn’t take much more than integration by parts. You can look up the principles of stationary and nonstationary phase. – Alex Ortiz Dec 01 '23 at 15:20