Deriving (using Fourier transform) the Poisson kernel for solving the Dirichlet problem on unit balls

534 Views Asked by At

Let's first consider the following Dirichlet problem on the upper half-space $\mathbb H^n=\{(x_1,\ldots, x_n)\in \mathbb R^n:x_n>0\}$. $$ \Delta u =0, u|_{x_n=0}=g(x). $$ Performing Fourier transform on $\Delta u=0$, changing the variables as $x_i\mapsto \xi_i, 1\leq i\leq n-1$, we get $$ i^{2} (\xi_1^2+ \xi_2^2+\ldots +\xi_{n-1}^2) (\mathcal F_{\partial\mathbb H_n} u)=-\partial^2_{x_{n}x_n}(\mathcal F_{\partial\mathbb H_n} u). $$ If we let $\xi=(\xi_1, \xi_2,\ldots ,\xi_{n-1}) ,|\xi|=\sqrt{\xi_1^2+ \xi_2^2+\ldots +\xi_{n-1}^2}$, then this is very easy to solve: $$ (\mathcal F_{\partial\mathbb H_n} u) (\xi,x_n)=A_+(\xi) \exp{(|\xi|x_n)}+A_-(\xi) \exp{(-|\xi|x_n)} $$ We now do the same transform on the boundary data. We have $$ \mathcal F_{\partial\mathbb H_n} u|_{x_n=0}=\mathcal F_{\partial\mathbb H_n} g(\xi_1,\ldots,\xi_{n-1}). $$ So, for even dimensions, the solution is unique if we limit ourselves to bounded solutions. In this case, $$ (\mathcal F_{\partial\mathbb H_n} u) (\xi,x_n)=(\mathcal F_{\partial\mathbb H_n} g)(\xi)\exp{(-|\xi|x_n)}. $$ To find an expression of $u$ in terms of an integral, we just evaluate the fourier transform of $\exp{(-|\xi|x_n)}$, and we get the solution $u$ in terms of a convolution: $$ u(x)=(g* P_{x_n}) (x_1\ldots, x_{n-1}), $$ where $P_{x_n}(x_1\ldots, x_{n-1})$ is the Poisson kernel.

Now, my question is, could I deduce in a similar fashion an explicit expression for the Poisson kernel for the unit ball? In other words, I am aiming to find $P_x(y)$ such that for all $y\in B(0,1)$, the function $$ u(y)=\int_{\partial B(0,1)} f(x) P_x(y) d\sigma $$ solves the Dirichlet problem $\Delta u =0, u|_{\partial B(0,1)}=f(x)$.

I have considered a conformal isomorphism $\phi:B(0,1)\to \mathbb H^n$, but the composition of a harmonic function with such a conformal map is not necessarily harmonic in higher dimensions.

Also, in odd dimensions, even the condition that $u$ is bounded could not gaurantee a unique solution, so I could use this idea in this case.

Edit: I already know other ways of reaching the Poisson Kernel; I just wish to see if we could use Fourier transform to derive it.

1

There are 1 best solutions below

3
On

First, FT wrt variables $x_i$, $1\leq i\leq n-1 $, gives sum $-\xi_1^2-\ldots-\xi_{n-1}^2$, not a product. Second, For a ball usually one gets a formula for the Green function $G$ of the Dirichlet problem in a ball and then the integral representation of a harmonic function in a ball gives that the kernel you seek is $P_x(y)=\frac{\partial G(y,x)}{\partial r_x}$. All this is written in PDE textbooks.