I have a second-order PDE $$ \mathcal{L} u = f \text{ + homogenous Dirichlet boundary conditions} $$ and I can prove that the bilinear form of the weak form of the differential equation fulfills all prerequisites of the Lax-Milgram Theorem, such that there exists a unique weak solution $u \in H^{1}$ for every data $f \in H^{-1}$. My question now is, whether there can be said anything about less regular data $f \in H^{-2}$? Can it be shown that there exists a unique solution $u \in H^0 = L^2$?
1 Answers
How do you define $H^{-2}?$ As the dual of $H^2$? Take any $\varphi \in H^{-2} \setminus H^{-1}$ and then the problem
$$
L(u,v)=\varphi(u)
$$
is...
1)not defined for $u,v \in L^2$ since the bilinearform requires $u,v \in H^1$. What is the (weak) gradient of an $L^2$ function anyway. As an example: How do you define
$$
L(u,v)=\int \nabla u \cdot \nabla v dx
$$
for $u,v$ in $L^2$?
2)The RHS $\varphi(u) \in H^{-2} \setminus H^{-1}$ is not a continuous functional for input functions $u \in H^1$ or $u \in L^2$ and hence Lax-Milgram does not apply.
The only remedy you have is to switch to some space of distributions and formulate your problem there, in which case existence gets a lot harder. A good sample problem to test your idea is $$ u''=\delta'(x) $$ on the unit intervall $I=[0,1]$.
- 4,335
- 1
- 12
- 21
-
Thank you. I see that I was just not thinking of whether all the operations make sense or not. – bheinzek Oct 17 '23 at 13:35
-
Its a good question nontheless to see that we cannot further relax the assumptions on the right hand side. There might be a correct setting/notion for your questions, but I don't think its Lax-Milgram. – F. Conrad Oct 17 '23 at 16:32