We know that the Fourier transform of the Poisson kernel $P(x,t)$ \begin{equation} \frac{\Gamma(\frac{(n+1)}{2})}{\pi^{\frac{(n+1)}{2}}}\frac{t}{(t^2+\lvert x\rvert^2)^{\frac{(n+1)}{2}}} \end{equation} is the Abel kernel $K(x,t)$ \begin{equation} e^{-2\pi t \lvert \xi \rvert}. \end{equation} However, I have just seen one method of proving it from Stein's Introduction to Fourier Analysis on Euclidean Spaces. The key of the proof is to use \begin{equation} e^{-\beta}=\frac{1}{\sqrt \pi} \int_0^\infty \frac{e^{-u}}{\sqrt u} e^{-\frac{\beta^2}{4u}} \, \mathrm{d} u. \end{equation} And it start with the Abel kernel to Poisson kernel. But I feel that this proof is a little trick. So is there any other proof of it?

Thank you very much!


1 Answer 1


Here is a proof that uses PDE theory to bypass the extensive computation.

First note that for $f$ a Schwartz function on $\mathbb{R}^n$ there is a unique $u\in C^{\infty}(\mathbb{R}^n\times \mathbb{R}_+)$ satisfying $$ \begin{cases} \frac{\partial^2 u}{\partial t^2}+\Delta_x u=0\\ \lim_{t^2+|x|^2\to \infty} u=0\\ \lim_{t\to 0}||u(\cdot, t)-f(\cdot)||_{\infty}=0. \end{cases} $$ Uniqueness of the above solution follows from the maximum principle, and existence follows from choosing $u(x,t)=P(x,t)*f(x)$.

One can also show existence and uniqueness for the following problem. $$ \begin{cases} \frac{\partial^2 u}{\partial t^2}-4\pi|\xi|^2 u=0\\ \lim_{t^2+|\xi|^2\to \infty} u=0\\ \lim_{t\to 0}||u(\cdot, t)-f(\cdot)||_{\infty}=0. \end{cases} $$ In this case, uniquness follows from the fact that the first two conditions imply for each fixed $\xi_0$, $u(t,\xi_0)=Ce^{-2\pi t |\xi_0|}$. The last condition then implies that these constants are determined by $f$, so we have $u(t,\xi)=e^{-2\pi t |\xi|}\cdot f(\xi)$, and this formula gives existence.

Finally, if $u$ solves the first system with initial data $f$, then $\hat{u}$ solves the second system with initial data $\hat{f}$, so we have that by uniqueness of solutions, $$ \hat{f}\cdot e^{-2\pi t |\xi|}=(f*P(x,t))^\wedge=\hat{f}\cdot \hat{P}(x,t). $$ As this holds for all $f$ Schwartz functions, $\hat{P}(x,t)=e^{-2\pi t |\xi|}$.


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.