0
$\begingroup$

I have a PDE that looks like this

$$\Delta T=\frac{\partial^{2}T}{\partial x^{2}}+\frac{\partial^{2}T}{\partial y^{2}}=0,\quad (x,y)\in \Omega$$

with boundary conditions

$$T(0,y)=20,\quad 0\leq y\leq 2, \\ T(5,y)=100,\quad 0\leq y\leq 2, \\ \frac{\partial T}{\partial y}(x,0)=0,\quad 0\leq x\leq 5, \\ \frac{\partial T}{\partial y}(x,2)=0,\quad 0\leq x\leq 5.$$

In short I have a two dimensions Laplace equation with mixed Dirichlet and Neumann boundary conditions on a rectangular domain $\Omega$. I believe I have found a way to solve it by separation of variables and I will get an infinite sum, but is this necessary? All boundary conditions are exactly specified. Also, when I solve the PDE this way I get a (particular?) solution which is

$$T(x,y)=16x+20,$$

and this solution satisfies all boundary conditions. Can I conclude that this is the only unique solution that exists?

Moreover I obtain something like this (quick solution sketch):

Let $T(x,y)=X(x)Y(y)$. Inserted into the Laplace equation, it yields

$$\frac{X''}{X}=-\frac{Y''}{Y}=\pm \lambda^{2},\quad \text{choose positive sign}.$$

Then we first get for $Y$

$$Y''+\lambda^{2}Y=0\implies Y(y)=A\sin(\lambda y)+B\cos(\lambda y),\\ \text{with BCs}\quad \frac{\partial Y}{\partial y}(0)=0=\frac{\partial Y}{\partial y}(2).$$

Taking into account the BCs we get the solution $Y$ to be

$$Y_{n}(y)=\cos(\lambda_{n}y),\quad \lambda_{n}=\frac{n\pi}{2}\quad n=0,1,....$$

Then we finally get for $X$

$$\lambda_{n}\neq 0: X''-\lambda^{2}X=0\implies X_{n}(x)=C\sinh(\lambda_{n}x)+D\cosh(\lambda_{n}x) \\ \text{with BCs}\quad X(0)=20,\quad X(5)=100.$$

Taking the BCs into account we get the solution

$$X(x)=\frac{100-20\cosh(5\lambda_{n})}{\sinh(5\lambda_{n})}\sinh(\lambda_{n}x)+20\cosh(\lambda_{n}x).$$

For $\lambda_{n}=0$ we get

$$X''=0\implies X(x)=cx+k \\ \text{with BCs}\quad X(0)=20,\quad X(5)=100.$$

Therefore,

$$X_{0}=16x+20$$

Furthermore the complete solution becomes (at least I think it becomes a series like this!)

$$T(x,y)=X(x)Y(y)=c_{0}(16x+20)+\sum_{n=1}^{\infty}c_{n}[\frac{100-20\cosh(5\lambda_{n})}{\sinh(5\lambda_{n})}\sinh(\lambda_{n}x)+20\cosh(\lambda_{n}x)]\cos(\lambda_{n}y).$$

(For some this might have been to long to read, but) is this the correct solution, or is $16x+20$ the correct solution? If the former, how do I determine the $c_{n}$ in this case? If none is correct (might as well be), how do I solve it?

Best regards//

$\endgroup$
1
$\begingroup$

Solutions to your boundary value problem are unique. Let me give you a quick proof, which uses the divergence theorem (see e.g. https://en.wikipedia.org/wiki/Divergence_theorem): Let $T_1,T_2$ be two solutions. Then $T:=T_1-T_2$ satisfies the differential equation and additionally $$T(0,y)=0,\quad 0\leq y\leq 2, \\ T(5,y)=0,\quad 0\leq y\leq 2, \\ \frac{\partial T}{\partial y}(x,0)=0,\quad 0\leq x\leq 5, \\ \frac{\partial T}{\partial y}(x,2)=0,\quad 0\leq x\leq 5.$$

Let $\Omega:=(0,5)\times(0,2)$ the open set, on which you are looking for your solutions. With $n$ being the outer unit normal of $\Omega$, we have $$0=\int_\Omega T\cdot\Delta T\, d(x,y) = \int_{\partial\Omega}\langle n,\nabla T\rangle T\, dS - \int|\nabla T|^2\, d(x,y).$$ The boundary term is now as follows: $$\int_{\partial\Omega}\langle n,\nabla T\rangle T\, dS = \int_0^2 T(0,y)(-1)\frac{\partial T}{\partial x}(0,y)\, dy+\int_0^2 T(5,y)\frac{\partial T}{\partial x}(5,y)\, dy\\ + \int_0^5 T(x,0)(-1)\frac{\partial T}{\partial y}(x,0)\, dx + \int_0^5 T(x,2)\frac{\partial T}{\partial y}(x,2)\, dx$$ Since the integrands vanish, the boundary term is zero. Hence $$0=\int_\Omega |\nabla T|^2\, d(x,y).$$ This yields $T$ to be constant and since we have $T(0,0)=0$, $T$ has to be zero on the whole of $\Omega$. Hence $T_1=T_2$.

The solution $T(x,y)=16x+20$ is therefore the only solution. You can check it easily by just plugging it into your problem.

EDIT:

Where might be a problem with your solution? The only thing, that comes to my mind, is that it is not clear a-priori that your series converges uniformely, i.e. that you can interchange derivatives and summation. If you only add finitely many, then the uniqueness result really yields $c_n=0$ for $n\geq 1$.

$\endgroup$
  • $\begingroup$ Thank you for the extensive explanation! I am just wondering, was my solution process wrong? Or does it lead to that c_0 = 1 and the rest of c_n = 0 ? $\endgroup$ – SimpleProgrammer Nov 6 at 16:18
  • $\begingroup$ @SimpleProgrammer: Answered in an edit $\endgroup$ – humanStampedist Nov 7 at 15:37

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.