Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I have a Laplace equation with some data along the $y$-axis: $$ \begin{cases} u_{xx} + u_{yy} &= 0 \\ u(0,y) &= f(y) \\ u_x(0,y) &= g(y). \end{cases} $$ There is no information of any region, properties of the solution or properties of $f$ and $g$ other than that we may assume $f$ and $g$ to have well-defined Fourier transforms.

So, Fourier transforming everything with respect to the $y$-variable we obtain: $$ \begin{cases} \hat{u}_{xx} -\xi ^2 \hat{u} &= 0 \\ \hat{u}(0,\xi ) &= \hat{f}(\xi ) \\ \hat{u}_x(0,\xi ) &= \hat{g}(\xi ). \end{cases} $$ Thus $\hat{u}(x,\xi ) = A(\xi )e^{-\xi x} + B(\xi ) e^{\xi x}$.

I suppose this is where I become puzzled. Clearly $\hat{u}$ behaves very badly as $\xi \to \pm \infty $, depending on the sign of $x$. To fix this I first restricted to the case $x>0$ and simply assumed that $A(\xi ) = 0$ for $\xi <0$ and $B(\xi ) = 0$ for $\xi > 0$ so that $$ \hat{u}(x,\xi ) = C(\xi )e^{-|\xi | x} = \hat{f}(\xi )e^{-|\xi | x}. $$ This is easy to invert since $\mathcal{F}^{-1}[e^{-|\xi |x}](x,y)$ is well-known. But now it seems there is no room for $g$ (or $\hat{g}$) to fit in anymore.

I suppose I must have gone wrong when I restricted to $x>0$ and singled out certain solutions, but how can I avoid this?

share|improve this question
add comment

3 Answers 3

up vote 2 down vote accepted

You cannot put $A(\xi)=0$ given that boundary conditions. So, $$ \hat u(0,\xi)=A(\xi)+B(\xi)=\hat f(\xi) $$ and $$ \hat u_x(0,\xi)=-\xi A(\xi)+\xi B(\xi)=\hat g(\xi). $$ This is a simple system you have to solve to get $A$ and $B$.

share|improve this answer
    
Thank you, I have done that, but does it make sense to inverse transform something that grows exponentially? –  flavio Dec 6 '12 at 10:41
    
Do the computations and you will realize that your inverse Fourier transform is not that trivial. –  Jon Dec 6 '12 at 10:50
    
I see, I obtain convolutions with Dirac deltas and some characteristic function. Well, thanks, I suppose I was thinking too classically. –  flavio Dec 6 '12 at 13:40
add comment

In fact you don't need to solve this problem by using Fourier transform, since you can just seckilling this problem by using D’Alembert’s formula:

$u(x,y)=\dfrac{f(y+ix)+f(y-ix)}{2}-\dfrac{i}{2}\int_{y-ix}^{y+ix}g(t)~dt$

Even you have to solve this problem by using Fourier transform:

$\mathcal{F}_{y\to\xi}\{u_{xx}(x,y)\}+\mathcal{F}_{y\to\xi}\{u_{yy}(x,y)\}=0$

$\hat{u}_{xx}(x,\xi)-\xi^2\hat{u}(x,\xi)=0$

$\hat{u}(x,\xi)=A(\xi)e^{-x\xi}+B(\xi)e^{x\xi}$

$u(x,y)=\mathcal{F}^{-1}_{\xi\to y}\{A(\xi)e^{-x\xi}\}+\mathcal{F}^{-1}_{\xi\to y}\{B(\xi)e^{x\xi}\}$

$u(x,y)=\dfrac{1}{2\pi}\int_{-\infty}^\infty A(\xi)e^{-x\xi}e^{iy\xi}~d\xi+\dfrac{1}{2\pi}\int_{-\infty}^\infty B(\xi)e^{x\xi}e^{iy\xi}~d\xi$

$u(x,y)=\dfrac{1}{2\pi}\int_{-\infty}^\infty A(\xi)e^{i(y+ix)\xi}~d\xi+\dfrac{1}{2\pi}\int_{-\infty}^\infty B(\xi)e^{i(y-ix)\xi}~d\xi$

$u(x,y)=C_1(y+ix)+C_2(y-ix)$

Substitute $u(0,y)=f(y)$ and $u_x(0,y)=g(y)$ to the above equation and follow the procedure that similar in http://en.wikipedia.org/wiki/D%27Alembert%27s_formula, you will get $u(x,y)=\dfrac{f(y+ix)+f(y-ix)}{2}-\dfrac{i}{2}\int_{y-ix}^{y+ix}g(t)~dt$ .

share|improve this answer
    
Yes, but the problem asks specifically to use the Fourier transform. –  flavio Dec 6 '12 at 12:01
    
Thank you doraemonpaul, but I figured it out. –  flavio Dec 7 '12 at 8:04
add comment

What might get you out of trouble here is that existence of the Fourier transforms of $f,g$ will imply decay conditions at infinity ($|\xi|\to\infty$) on those Fourier transforms $\hat{f}, \hat{g}$. This is the essential content of the Riemann-Lebesgue lemma: see here and page 308 here (from Applied Analysis by Hunter and Nachtergaele). You probably need to say more about the functions $f,g$ in order to check that (i) the lemma applies and (ii) the decay is sufficiently rapid that you can invert $\hat{u}=\hat{f}\cosh(\xi x) +\xi^{-1}\hat{g}\sinh(\xi x)$.

Just to note: your problem is a Cauchy initial value problem for the Laplace equation. Such problems are not generally well-posed. See Hadamard's example e.g. here.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.