I have a sequence $(u_k) \in L^2_{loc}(\mathbb{R}^+; H^1_0(\Omega) )$ and $u \in L^2_{loc}(\mathbb{R}^+\times \Omega )$ such that for any $T >0$ and any compact $K \subset \Omega$ we have : $\int_{[0;T]\times K} |u_k(t,x)-u(t,x)|^2dtdx \rightarrow 0$ when $k \rightarrow \infty$

The book I am reading wants to prove that for any $\Psi \in L^2([0;T], H^1_0(\Omega))$ and any $\phi \in L^2([0;T] \times \Omega )$;

$\int_{[0;T]\times \Omega} (\nabla u_k(t,x) - \nabla u(t,x)) \cdot \nabla \Psi(t,x) dtdx \rightarrow 0$

$\int_{[0;T]\times \Omega} (u_k(t,x) - u(t,x))\cdot \phi(t,x) dtdx \rightarrow 0$^

It says the following:

$u_k$ converges towards $u$ in the space $ \mathcal{D}'(]0;T[ \times \Omega)$, the sequence $(u_k)$ is bounded in $L^2([0;T], H^1_0(\Omega))$, so as $ \mathcal{D}(]0;T[ \times \Omega)$ is dense both in $L^2([0;T], H^1_0(\Omega))$ and in $L^2([0;T]\times \Omega)$ the weak convergence holds both in the space $L^2([0;T], H^1_0(\Omega))$ and $L^2([0;T]\times \Omega)$.

I don't understand this argument, I don't see how we could pass to the limit both for k tending to $\infty$ and for a sequence $\phi_n \in \mathcal{D}(]0;T[ \times \Omega)$ tending towards $\phi$ or $\Psi$, does any of you find some clarity in this ?


I guess the only delicate point is to retrieve (weak) convergence for the whole sequence, not only along some subsequence (which is trivial by the Banach Alaoglu theorem since $u_k$ is bounded in $L^2(0,T;H^1_0)$). A classical separation argument would do the trick here: if any converging subsequence converges to the same limit, then in fact the whole sequence converges. Here is how I think the author argues.

Fix $\Phi,\Psi\in \mathcal{D}((0,T)\times\Omega)$. Your original assumption $\int_{[0;T]\times K} |u_k(t,x)-u(t,x)|^2dtdx \rightarrow 0$ implies in particular $u_k\to u$ in $\mathcal{D}'((0,T)\times\Omega)$, so by definition of the convergence in the sense of distributions $$ \int_{[0;T]\times \Omega} u_k(t,x) \Phi(t,x) dtdx =\left< u_k, \Phi\right>_{\mathcal{D}',\mathcal{D}}\to \left< u, \Phi\right>_{\mathcal{D}',\mathcal{D}}\hspace{2cm}(1) $$ and $$ \int_{[0;T]\times \Omega} \nabla u_k(t,x) \cdot \nabla \Psi(t,x) dtdx =\left<\nabla u_k,\nabla \Psi\right>_{\mathcal{D}',\mathcal{D}}\to \left<\nabla u,\nabla \Psi\right>_{\mathcal{D}',\mathcal{D}}\hspace{2cm}(2) $$ (no need for extraction or Banach Alaoglu here, this is merely the continuity of differentiation in the sense of distributions). Since $u_k$ is bounded in $L^2(0,T;L^2)$ and $L^2(0,T;H^1_0)$ you see that both left-hand sides in (1)-(2) are controlled by $|\Phi|_{L^2(0,T;L^2)}$ and $|\nabla\Psi|_{L^2(0,T;L^2)}$ respectively, thus $$ \left|\left< u, \Phi\right>_{\mathcal{D}',\mathcal{D}}\right|\leq C |\Phi|_{L^2(0,T;L^2)} $$ and $$ \left|\left<\nabla u,\nabla \Psi\right>_{\mathcal{D}',\mathcal{D}}\right|\leq C |\nabla\Psi|_{L^2(0,T;L^2)} $$ for all $\Phi,\Psi\in \mathcal{D}((0,T)\times\Omega)$. Now by density of $\mathcal{D}((0,T)\times\Omega)$ in $L^2(0,T;L^2)$ and $L^2(0,T;H^1_0)$ one concludes that the distributions $u,\nabla u$ can be regarded as $L^2$ functions, i-e $$ \left< u, \Phi\right>_{\mathcal{D}',\mathcal{D}}=\int_{[0;T]\times \Omega} u_k(t,x) \Phi(t,x) dtdx $$ and $$ \left< \nabla u, \nabla\Psi\right>_{\mathcal{D}',\mathcal{D}}=\int_{[0;T]\times \Omega} \nabla u_k(t,x) \cdot\nabla\Psi(t,x) dtdx $$ for all $\Phi,\Psi\in \mathcal{D}((0,T)\times\Omega)$. Now by (1) and (2) you get $$ \int_{[0;T]\times \Omega} (u_k(t,x) - u(t,x))\cdot \Phi(t,x) dtdx \rightarrow 0 $$ $$ \int_{[0;T]\times \Omega} (\nabla u_k(t,x) - \nabla u(t,x)) \cdot \nabla \Psi(t,x) dtdx \rightarrow 0 $$ for all $\Phi,\Psi\in \mathcal{D}(0,T\times\Omega)$. Using one last time the density of $\mathcal{D}((0,T)\times\Omega)$ you deduce that this holds for all $\Phi\in L^2(0,T;L^2)$ and $\Psi\in L^2(0,T;H^1_0)$ (take $\Phi_n\to\Phi$ and $\Psi_n\to \Psi$ in these spaces), QED.

  • $\begingroup$ The fact that $\nabla u$ belong to $L^2$ wouldn't rather be coming from the inequality $|<\nabla u , \phi>|_{\mathcal{D}',\mathcal{D}}\leq C |\phi|_{L^2(0,T;L^2(\Omega))}$ which is true because the sequence $(u_k)_k$ is bounded in $L^2(0,T; H^1_0)$. $\endgroup$ – incas Aug 1 '14 at 12:11
  • $\begingroup$ The inequality $|<\nabla u; \nabla \Psi>|_{\mathcal{D}',\mathcal{D}} \leq C |\nabla \Psi|_{L^2(0,T;L^2(\Omega))}\leq C | \Psi|_{L^2(0,T;H^1_0(\Omega))}$ giving us rather that $\Delta u \in L^2(0,T; H^1_0)$ $\endgroup$ – incas Aug 1 '14 at 12:17
  • $\begingroup$ Indeed, my argument really shows that $\nabla u_k\rightharpoonup \nabla u$ in $L^2L^2$, I just kept using $\nabla \Psi$ as test functions in order to stay consistent with the OP's setting. But I disagree with $\Delta u \in L^2(0,T; H^1_0)$, the above inequality gives $\Delta u \in L^2(0,T; H^{-1})$! $\endgroup$ – leo monsaingeon Aug 1 '14 at 15:41
  • $\begingroup$ You are right $\Delta u \in L^2(0,T; H^{-1})$, Thank you your answers have been really helpfull. $\endgroup$ – incas Aug 1 '14 at 16:53

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.