0
$\begingroup$

The following is a question from Strauss:

Let $u:[0,1]^3\to\mathbb{R}$ satisfy the Laplace equation $\Delta u = u_{xx} + u_{yy} + u_{zz} = 0 $ and the Neumann boundary condition $u_z(x,y,1) = g(x,y)$ with the normal derivatives on all other faces equal to zero, and $$\int_0^1\!\!\!\!\!\int_0^1 g(x,y)\,\mathrm{d}x\,\mathrm{d}y = 0.$$ Solve for $u$.

My immediate approach is try attempt separation of variables, and write $u(x,y,z) = X(x)Y(y)Z(z)$, so that we have $$X''-\lambda_1 X = 0,\ \ Y'' - \lambda_2 Y = 0,\ \ Z'' - \lambda_3 Z = 0,\ \ \mathrm{and}\ \ \lambda_1 + \lambda_2 + \lambda_3 = 0$$ with boundary conditions $$X'(0) = X'(1) = 0,\ \ Y'(0) = Y'(1) = 0,\ Z'(0) = 0$$ so we solve to get $$\lambda_1 = \beta_1^2 = k_1^2\pi^2,\ \ \lambda_2 = \beta_2^2 = k_2^2\pi^2$$ which gives us eigenfunctions $$X(x) = \cos\beta_1 x,\ \ Y(y) = \cos\beta_2 x$$ which leaves us with the equation $Z'' = \pi^2(k_1^2 + k_2^2)Z, Z'(0)=0$.

Now, my suspicion is that the next step is to write $g$ as a fourier series $$g(x,y) = \sum_{k_1 = 0}^\infty\sum_{k_2 = 0}^\infty a_{k_1,k_2}\cos\left(\pi k_1 x\right)\cos\left(\pi k_2 y\right)$$ to obtain a solution of the form $$u(x,y,z) = \sum_{k_1 = 0}^\infty\sum_{k_2 = 0}^\infty a_{k_1,k_2}\cos\left(\pi k_1 x\right)\cos\left(\pi k_2 y\right)\frac{\cosh\left(\pi z\sqrt{k_1^2+k_2^2} \right)}{\pi\sqrt{k_1^2+k_2^2}\sinh\left(\pi\sqrt{k_1^2+k_2^2} \right)}.$$

Now, assuming sufficient regularity of $g$, and assuming that $g$ has fourier series in this form (which is probably not true in general), it is easy to see that this satisfies the boundary conditions, and has vanishing laplacian. However, is my Fourier expansion for $g$ justified, and can the solution be expressed in a nicer way?

1 Answers 1

1

Yes, the separation of variables is the expected approach. The book by Strauss does not go deeply into regularity issues (it's written for undergraduates, and the rigorous analysis of regularity for solutions of PDE is typically a graduate topic).

If $g$ is in $L^2$, the series converges in $L^2$, and so does the series for the solution - notice that $\cosh/\sinh$ ratio tends to $1$, so the fraction containing it is $O(1/\|k\|)$. If $g$ is nice enough to have uniformly convergent Fourier series, the series for the solution converges uniformly.

There is no better way to express the solution; this series is the expected answer.