Consider the inner product by $\langle f,g \rangle_{H^1} = \langle f, g \rangle_{L^2} + \sum_{|\alpha|=1} \langle D^\alpha f, D^\alpha g \rangle_{L^2}$ where $\alpha$ is a multi-index and $D$ denotes the weak derivative. Define $H^1(\Omega)$ as the space of functions that are finite under the norm induced from this inner product. It can be shown that $H^1(\Omega)$ is a Hilbert space.

Now, suppose there exists a sequence of functions $\{ f_n \} \subset H^1(\Omega)$ that converges weakly in $H^1$ to some limit $f \in H^1(\Omega)$. Can I then say that this sequence converges weakly to the same limit under the $L^2$ inner product?

By an application of the Banach-Alaoglu Theorem, I know that weak convergence of this sequence in $H^1$ will imply strong convergence of a subsequence in $H^1$. And then I think strong convergence of this subsequence in $H^1$ will imply strong convergence in $L^2$ as well.

However, I'm not sure if anything can be said about the entire sequence under the $L^2$ inner product and its weak/strong convergence properties.


$L^2(\Omega)$ is a subspace of $H^{-1}(\Omega)$, the dual of $H^1(\Omega)$. Thus, a sequence converging weakly in $H^1$ converges weakly in $L^2$ to the same limit.

In general, nothing can be said about strong convergence in $L^2$.


I don't see how a simple statement '$L^2(\Omega)$ is a subspace of $H^{-1}(\Omega)$' proves that the weakly converging sequence in $H^1$ converges to the same limit in $L^2$ because the inner products on $L^2$ and $H^1$ are different. Here is my attempt:

Let $(u_n)_n$ be a sequence converging weakly to $u$ in $H^1$, hence $(u_n)_n$ is bounded in $L^2$, and we can extract a subsequence $(u_{\varphi(n)})_n$ that converges weakly to $v \in L^2$. $\forall \phi \in C^1_c(\Omega)$, $\left|\int_\Omega u_{\varphi(n)} \phi'\right| = \left|\int_\Omega u_{\varphi(n)}' \phi\right|\leq C ||\phi||_{L^2}\Rightarrow \left|\int_\Omega v\phi'\right|\leq C ||\phi||_{L^2}$, hence $v$ is weakly differentiable. Furthermore, $\left|\int_\Omega (u_{\varphi(n)}' - v')\phi\right| = \left|\int_\Omega (u_{\varphi(n)} - v)\phi'\right| \xrightarrow{n\rightarrow \infty} 0$ by the weak convergence of $(u_{\varphi(n)})_n$. Thus we have that $(u_{\varphi(n)}')_n$ converges weakly to $v'$ in $L^2$ by the density of $C^1_c$ in $L^2$. To conclude, $\forall f\in H^1, \int_\Omega u_{\varphi(n)} f + \int_\Omega u_{\varphi(n)}' f' \xrightarrow{n\rightarrow \infty} \int_\Omega v f + \int_\Omega v' f'$ by the respective weak convergence of $u_{\varphi(n)}$ and $u_{\varphi(n)}'$. The uniqueness of weak limit then implies $v=u$ in $H^1\subset L^2$.

  • 1
    $\begingroup$ Weak convergence in either function space implies convergence in the sense of distributions: that is, $\int u_n\phi \to \int u \phi$ for any $\phi\in C^\infty_c(\Omega)$. The distributional limit is unique. Therefore, the limits in $H^1$ and in $L^2$ represent the same distribution, i.e., they are the same. // Distributions are a convenient common denominator of function spaces. $\endgroup$ – user53153 Feb 28 '13 at 1:52
  • $\begingroup$ The statement "$L^2(\Omega)$ is a subspace of $H^{-1}(\Omega)$, the dual of $H^1(\Omega)$" should IMHO be interpreted as follows: for each $g \in L^2(\Omega)$, the map $f \mapsto \int_{\Omega} fg$ is a continuous linear functional on $H^1(\Omega)$. Indeed, $\lvert \int_\Omega fg \rvert \le \|f\|_{H^1} \|g\|_{L^2}$. This is easy to verify and immediately implies the desired statement about weak convergence. $\endgroup$ – Nate Eldredge Mar 9 '13 at 15:28

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.