Let $h_a=\frac{\Gamma(\frac{a}{2})}{\pi^{\frac{a}{2}}}|x|^{-a}, x \in R^d$ Then $\hat{h_a}=h_{d-a}$ in the sence of $L^1+L^2-$Fourier transforms if $\frac{d}{2}<Re(a)<d,$ and in the sence of distributional Fourier transforms if $0<Re(a)<d$.

This is Lemma $4.1$ in these lecture notes: http://www.math.ubc.ca/~ilaba/wolff/notes_march2002.pdf

I only managed to complete the details of the proof for $d>2$

How can i prove this for general $d>0$ with the same arguments i used for $d>2$?

Here is my complete proof for $d>2$

$\underline{\text{Proof}}$ Let $h(x)=|x|^{-a}$ where $a\in (\frac{d}{2},d),$ then $h$ is radial and\ $h \in L^1(R^d)+L^2(R^d)$ because $h(x)=h_1+h_2$ where $$h_1(x)=h(x)1_{\{|x|<1\}}(x) \in L^1(R^d)$$ $$h_2(x)=h(x)1_{\{|x| \geq 1\}}(x) \in L^2(R^d)$$\

Now recall that if $f \in L^2(R^d)$ and $\phi_n \in \mathrm{S}$ such that $\phi_n \to^{L^2}f$, then $\hat{\phi_n} \to ^{L^2} \mathcal{F}(f)$

Using this fact and simple changes of variables we can easily deduce that the $L^2-$Fourier transform $\mathcal{F}$ of a radial function $f \in L^2(R^d)$ is radial. Thus the $L^1+L^2-$Fourier transform of $h$ is radial.

Also using the previous fact, again we have $\hat{h}(M\xi)=M^{-(d-a)}\hat{h}(\xi).$ If $\xi \in R^d$ then $$\hat{h}(\xi)=\hat{h}(|\xi| \frac{\xi}{|\xi|})=|\xi|^{-(n-a)}\hat{h}( \frac{\xi}{|\xi|})$$ So $\hat{h}(\xi)=c|\xi|^{-(d-a)}$ where $c=\hat{h}(x), \forall x \in \mathbb{S}^{d-1}$

Using the Duality relation(which is true for $L^2$ functions by approximation)we have $$\int_{R^d}|x|^{-a}e^{-\pi |x|^2}dx=c\int_{R^d}|x|^{-(d-a)}e^{-\pi|x|^2}dx \text{ }(1)$$

Here we used the identity: $\widehat{e^{-\pi|x|^2}}(\xi)=e^{-\pi|\xi|^2}$ where $e^{-\pi|x|^2} \in \mathrm{S}$

Polar Coordinates formula and appropriate changes of variables to on both sides of the equation $(1)$ gives us: $$c=\frac{\Gamma(\frac{d-a}{2})\pi^{\frac{a}{2}}}{\Gamma(\frac{a}{2})\pi^{\frac{d-a}{2}}}$$

Hence $\hat{h_a}=h_{d-a}.$

Now for the general case, let $\phi \in \mathrm{S},d>2$ and $$A(z)=\int_{R^d}h_z\hat{\phi}$$ $$B(z)=\int_{R^d}h_{d-z}\phi$$\We will show that $A(z),B(z)$ are holomorphic in the strip\ $I=\{z: 0<Re(z)<d-1\}$ and agree everywhere on $I.$ Note that $\frac{d}{2} < d-1$

Recall that $\Gamma(z)$ is holomorphic define on the region $\Omega=\{z:Re(z)>0\} $ and has no zeroes, so the reciprocal gamma function $\frac{1}{\Gamma}(z)$ \ is holomorphic in $\Omega.$ So it suffices to show the holomorphy of the function $G:I \to \Bbb{C}$ where $$G(z)=\int_{R^d}|x|^{-z}\phi(x)dx $$

Let $z \in I$ and define $F(x,z)=|x|^{-z}\phi(x)$ and let $h_n \in \Bbb{C}$ such that $h_n \to 0$ and $|h_n| <\min\{1,\frac{d-Rez-1}{2}\}, \forall n \in \Bbb{N}$. Then

$$\frac{F(x,z+h_n)-F(x,z)}{h_n} \to \frac{-\ln|x|}{x^z}\phi(x), \text{ }a.e$$ and also\ $$|\frac{F(x,z+h_n)-F(x,z)}{h_n}|= |\frac{1}{|h_n||x|^z}\bigg(\frac{1}{|x|^{h_n}}-1 \bigg)||\phi(x)|= \frac{|e^{-h_n \ln{|x|}}-1|}{|h_n||x|^{Re(z)}}|\phi(x)|$$ $$=\frac{|e^{-h_n \ln{|x|}}-1|}{|h_n||x|^{Re(z)}}|\phi(x)|1_{\{|x| \leq 1\}}+\frac{|e^{-h_n \ln{|x|}}-1|}{|h_n||x|^{Re(z)}}|\phi(x)|1_{\{|x|>1\}}$$

$\textbf{(1)}$ If $|x| \leq 1$ then,

$$\frac{|e^{-h_n \ln{|x|}}-1|}{|h_n||x|^{Re(z)}}1_{\{|x| \leq 1\}}|\phi(x)| \leq \frac{|\ln{|x|}||e^{|h_n ||\ln{|x|}|}}{|x|^{Re(z)}}1_{\{|x| \leq 1\}}|\phi(x)|=\frac{\ln{\frac{1}{|x|}}e^{|h_n| \ln{\frac{1}{|x|}}}}{|x|^{Re(z)}}1_{\{|x| \leq 1\}}|\phi(x)|$$ $$\leq \frac{1}{|x|^{Re(z)+1+|h_n|}}1_{\{|x| \leq 1\}}|\phi(x)|\leq \frac{1}{|x|^{Re(z)+1+\frac{d-1-Re(z)}{2}}}1_{\{|x| \leq 1\}}|\phi(x)| \in L^1(R^d)$$ since $\phi$ is bounded everywhere.

$\textbf{(2)}$ If $|x|>1$ then $$\frac{|e^{-h_n \ln{|x|}}-1|}{|h_n||x|^{Re(z)}}|\phi(x)|1_{\{|x|>1\}}\leq |x|^2|\phi(x)|1_{\{|x|>1\}} \in L^1(R^d)$$ since $\phi \in \mathrm{S}.$

Using $\textbf{(1),(2)}$ and the dominated convergence theorem we have that $A(z),B(z)$ are differentiable at $z$. So $A,B$ are differentiable at every $z \in I$ thus holomorphic in $I$ By $\textbf{Proposition}$, $$A(z)=B(z), \forall z \in (\frac{d}{2},d) \supset (\frac{d}{2},d-1)$$ Thus by identity theorem $A(z)=B(z), \forall z \in I $ If $Re(z)>\frac{d}{2}$ then $h_a \in L^1(R^d)+L^2(R^d),$ so its $L^1+L^2$ and distributional Fourier transforms coincide. $\text{ }\blacksquare.$


Here's a simple way to determine the form of $\hat{h}_a$:

The function $h_a$ satisfies $x\cdot\nabla h_a = -a h_a$. Taking the Fourier transform gives the equation $i\nabla\cdot(ix\hat{h}_a) = -a \hat{h}_a.$ The left hand side can be rewritten as $-(\nabla\cdot x)\hat{h}_a - (x\cdot\nabla)\hat{h}_a.$ Here, $\nabla\cdot x = d$ so we have a differential equation $x\cdot\nabla\hat{h}_a = -(d-a)\hat{h}_a.$ The solutions of this are $\hat{h}_a = C_{a,d} |x|^{-(d-a)}.$

But how to determine the constant $C_{a,d}$? Some thoughts:

Taking the Fourier transform again we get $\hat{\hat{h}}_a = C_{d-a,a} C_{a,d} |x|^{-a}.$ By the Fourier inversion theorem, $\hat{\hat{h}}_a(x) = (2\pi)^d h(-x),$ so $C_{d-a,d}C_{a,d} = (2\pi)^d.$

  • $\begingroup$ +1...Thank you for your answer..your idea is slick..but can do you know how can i prove with the same arguments the theorem for d<2?..(I forgot to mention that i want a proof for $d<2$ with the same arguments i used for $d>2$.) $\endgroup$ – Marios Gretsas Aug 17 at 17:56

This is too complicated. There is a much simpler proof of the distributional Fourier transform with $0<\Re(a)<d$ in this previous answer I gave: How to calculate $c_a$ where $\left(f\mapsto\int_{\mathbb{R}}\frac{f(t)-f(0)}{|t|^{a}}dt\right)=c_a\mathcal{F}_x(|x|^{-1+a})$

It is based on representing $\frac{1}{|x|^a}$ as a continuous superposition of Gaussians.

  • $\begingroup$ +1 Thank you for your answer. $\endgroup$ – Marios Gretsas Aug 21 at 1:45

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.