2
$\begingroup$

Suppose that $\Omega=[0,1]^2$. I will say that a real valued function $u$ on $\Omega$ satisfies periodic boundary values if $$u(x,0)=u(x,1), \;u(0,y)=u(1,y),\;\;\;\text{ for all }x,y\in[0,1].$$

Now let $f\in L^2(\Omega)$, and suppose that I want to look for a solution of the following problem $$ \begin{cases}-\Delta u+u=f, \;\;\text{ on }\Omega=(0,1)^2,\\ u,\; u_x,\text{ and }u_y\text{ satisfy periodic boundary values.}\end{cases}$$

The question is how to show existence (if possible) of $u$.


My input to the problem: The weak formulation of the problem with test space taken to be $$X=\{v\in H^1(\Omega):\text{ $v$ satisfies periodic boundary values}\},$$ is the same as the $H^1_0$ formulation, ie $$a(u,v):=\int_\Omega\nabla u\cdot \nabla v+\int_\Omega uv =\int_\Omega fv=: F(v),\;\;\;\text{ for all }v\in X.$$ This is because the boundary integral vanishes due to $u_x$, $u_y$ and $v$ being in $X$. On the other hand, $X$ is a closed subspace of $H^1(\Omega)$ and therefore it is a Hilbert space with the $\|\cdot\|_{H^1}$ norm. I think we can use the Lax-Milgram theorem on $X$, because $a$ seems coercive on $X$, $a$ and $F$ are continuous on $X$.

However, the Lax-Milgram theorem gives uniqueness, which is weird because unlike $H^1_0(\Omega)$, the space $X$ does not exactly prescribe the boundary values (it just says that they are periodic). In any case, even if I do get the solution $u\in X$ and then by regularity $u\in X\cap H^2(\Omega)$, I should still show that $u_x$ and $u_y$ satisfy periodic boundary values, and this I don't know how to.

$\endgroup$

1 Answer 1

1
$\begingroup$

You are using the wrong space, that's all. The correct space is $$ X=\{u \in H^1_{\textrm{loc}}(\mathbb R^2): u(\cdot+n)=u(\cdot) \quad \forall n=(n_1,n_2)\in \mathbb Z^2\}. $$ Then you apply Lax-Milgram and you are good to go. Look at Asymptotic Analysis for Periodic Structures, by Bensoussan, Lions, Papanicolaou (1979) chapter 1, for example.


Answering your comment. $f$ is likewise extended periodically, namely $f$ is extended by zero into $f_0$, and then identified with $\sum_{n\in \mathbb Z^2} f_0(\cdot + n)$ which is periodic on $\mathbb R^2$ and $L^2_{\textrm{loc}} (\mathbb R^2)$.

Now, on say $(-2,2)^2$, $-\Delta u + u=f$ as a weak solution, and therefore interior regularity shows $u\in H^2((-\frac32,\frac32)^2)$. All in all, $u\in H^2_{\textrm{loc}} (\mathbb R^2)$. Consequently, $\partial_x u \in X$, and $\partial_y u \in X$.

Incidentally, because all coefficients are constant, you can write write $u$ explicitely as a (double) Fourier series, its coefficients being that of $f$, divided by $n_1^2+n_2^2+1$. It saves you the trouble of using Lax-Milgram.

$\endgroup$
6
  • $\begingroup$ Thank you for the reference. However, even if I do get $u\in X$ as you defined it, I also need that $u_x,u_y\in X$, which doesn't seem that it follows trivially from the weak formulation! $\endgroup$
    – demlevi33
    Nov 14, 2021 at 19:57
  • $\begingroup$ Wait so $f$ should also be periodic on the boundary? (the way I defined periodic) $\endgroup$
    – demlevi33
    Nov 14, 2021 at 20:18
  • $\begingroup$ Also, I don't see how $u\in H^2_{\text{loc}}(\mathbb{R}^2)$ directly implies that $u_x,u_y\in X$. $\endgroup$
    – demlevi33
    Nov 14, 2021 at 20:20
  • $\begingroup$ @demlev You are asking why if $f(\cdot +n)=f(\cdot)$ for every $n$ then the same hold for $Df$? $\endgroup$
    – username
    Nov 14, 2021 at 20:24
  • $\begingroup$ Yes. How did you deduce this? $\endgroup$
    – demlevi33
    Nov 14, 2021 at 20:27

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.