A question in my PDE class is

Suppose $a(x) = a_{ij}(x)$ takes values in the class of symmetric, positive definite $n \times n$ matrices. Consider the PDE

$$u_t = \sum_{i,j=1}^n \frac{\partial}{\partial x_i}\left( a_{ij}(x) \frac{\partial u}{\partial x_j} \right)$$

in a bounded domain $\Omega$, with intial condition $u=u_0(x)$ at $t=0$ and a Dirichlet boundary condition $u=g$ at $\partial \Omega$. Use the "energy method" to show there can be at most one solution.

In the past, we have used the energy method to show uniqueness by multiplying both sides by $u$ and integrating:

$$\int_{\Omega}uu_t= \frac12 \frac{d}{dt}\int_{\Omega}u^2 = \int_{\Omega} u \sum_{i,j=1}^n \frac{\partial}{\partial x_i}\left( a_{ij}(x) \frac{\partial u}{\partial x_j} \right),$$

and attempting to show somehow that the right hand side must be negative, therefore if there are two solutions $u_1$ and $u_2$, their difference has zero initial state, and $\frac{d}{dt}\int_{\Omega}u^2 dx \leq 0$ shows that it must remain there, so their difference is the zero solution, and therefore the solution is unique.

Here I am trying to use the positive definiteness of $a(x)$ to show that the right hand side must be negative. If we can write it in the form $\int_{\Omega} -\langle a(x)\vec{v}, \vec{v}\rangle$ we will be done. But so far I have not been able to do so.

If we apply the chain rule we get

$$\frac12 \frac{d}{dt}\int_{\Omega}u^2 = \int_{\Omega} u \sum_{i,j=1}^n \left( \frac{\partial a_{ij}}{\partial x_i} \frac{\partial u}{\partial x_j} + a_{ij}(x)\frac{\partial^2 u}{\partial x_i \partial x_j}\right),$$

but I'm not getting anywhere with that. Any ideas?

  • 1
    $\begingroup$ Aren't you missing an integral in the second equation? $\endgroup$ – dls Sep 15 '13 at 4:52

As dls mentioned you're lacking an integral on the RHS. With this notice that

$$ \sum_{i,j=1}^n \partial_i(a_{ij} \partial_j u) =\text{div}(a \nabla u), $$

so that, integrating by parts and using the zero boundary conditions, we get that the RHS is equal to

$$ -\int_{\Omega} \langle \nabla u, a \nabla u \rangle dx \leq 0. $$

Edit: Notice that you want to prove that if $u(x,0)=0$ for every $x\in \Omega$ and $u(y,t)=0$ for every $y\in \partial \Omega$ and $t>0$ then $u\equiv 0$. So take such an $u$, then your argument gives

$$ \frac{1}{2}\frac{d}{dt} \int_{\Omega} |u|^2 dx = \int_{\Omega} u \text{div}(a\nabla u) dx. $$

Call $v=a\nabla u$. The divergence theorem, applied to the vector field $uv$, gives

$$ \int_{\Omega}u \text{div}(v)dx= -\int_{\Omega}\langle \nabla u, v \rangle dx + \int_{\partial \Omega}\langle uv,\nu \rangle d \sigma. $$

Since we have that $u(y,t)=0$ for every $y\in \partial \Omega$ the boundary integral is zero for every $t$. Therefore we have

$$ \frac{1}{2}\frac{d}{dt}\int_{\Omega} |u|^2 dx =- \int_{\Omega} \langle \nabla u,a\nabla u\rangle dx \leq 0. $$

  • $\begingroup$ I'm sorry, can you clarify the integration by parts? We have $$\int_{\Omega}\operatorname{div}{(a\nabla u)} = \int_{\partial \Omega}(a\nabla u \cdot \hat{n}dS),$$ and then what? Normally I would try to use $$\nabla \cdot (f \nabla g) = \langle \nabla f, \nabla g \rangle + f \Delta g,$$ but here $a$ is a matrix valued function... $\endgroup$ – Eric Auld Sep 15 '13 at 16:01
  • $\begingroup$ @EricAuld: I've edited, let me know if you still have doubts on the argument. $\endgroup$ – Jose27 Sep 15 '13 at 18:44
  • $\begingroup$ Thank you very much! I forgot about the $u$. $\endgroup$ – Eric Auld Sep 15 '13 at 18:49

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.