Let $a$ be a non negative (positive almost everywhere) weight in $L_{loc}^1(\Omega)$, $\Omega\subseteq\mathbb{R}^n$ is open. For $\varphi\in C_c^{\infty}(\Omega)$ define $$ \Vert\varphi\Vert_a^2=\int_{\Omega} a(x)|\nabla\varphi(x)|^2\,\mathrm{d}x. $$ Then $\Vert\cdot\Vert_a^2$ is a norm on $C_c^{\infty}(\Omega)$. Let $D_0^1(\Omega;a)$ denotes the closure of $C_c^{\infty}(\Omega)$ with respect to the $\Vert\cdot\Vert_a^2$-norm.

Let $\{\varphi_n\}\in C_c^{\infty}(\Omega)$ and $\varphi_n\to \varphi$ in $D_0^1(\Omega;a)$ and also we have $\varphi_n\to \psi$ in some space $L^p(\Omega),\,1\leqslant p\leqslant\infty$.

Can we say that $\varphi=\psi$? And why?

A similar situation happens in the proof of Sobolev embedding theorem. Where we have a smooth sequence $\{\varphi_n\}$ tends to a $u\in W^{1,p}(\mathbb{R}^n)$ and then we show $\varphi_n$ tends to some $v\in L^{p^*}(\mathbb{R}^n)$ and then we say $u=v$ a.e. In this situation, my understanding is : since $\Vert\varphi_n-u\Vert_p\to 0$ and $\Vert\varphi_n-v\Vert_{p^*}\to 0$ there is a subsequence $\{\varphi_{n_{k_i}}\}$ converges almost everywhere to both $u$ and $v$, then $u=v$ a.e.

But in the $D_0^1(\Omega;a)$ case, does there also a subsequence converges to $\varphi$ a.e.? Or some other approachs?

Any advice will be appreciated!


Here is some of my thought:

  1. For $u_k\in L^p(\Omega), 1\leqslant p\leqslant\infty$, $u_k\to u$ in $L^p(\Omega)$, define $$ \langle u,\varphi\rangle=\int_{\Omega}u\varphi $$ for all $\varphi\in C_c^{\infty}(\Omega)$. We see this integral makes sense since $\varphi$ has compact support and $u$ is locally integrable. And it is a continuous linear functional on $C_c^{\infty}(\Omega)$. Furthermore $$ |\langle u_k,\varphi\rangle-\langle u,\varphi\rangle|\leqslant \int_{\Omega}|u_k-u||\varphi|\to 0 $$ by Holder's inequality. Thus we have $u_k\to u$ in $\mathcal{D}'(\Omega)$.

In Sobolev case we have $u_k$ tends to both $u$ and $v$ in $\mathcal{D}'(\Omega)$, so $u=v$ a.e.

  1. For $u_k\in D_0^1(\Omega;a)$, $u_k\to v$ in $D_0^1(\Omega;a)$, define $$ \langle u,\varphi\rangle=\int_{\Omega}a^{1/2}(x)|\nabla u|\varphi. $$ for all $\varphi\in C_c^{\infty}(\Omega)$. Then the integral makes sense by applying Holder's inequality. And it is also a linear continuous functional on $\mathcal{D}'(\Omega)$. Furthermore $$ |\langle u_k,\varphi\rangle-\langle v,\varphi\rangle|\leqslant \int_{\Omega}a^{1/2}(x)|\nabla u_k-\nabla u||\varphi|\to 0 $$ by Holder's inequality. So we have $u_k \to v$ in $\mathcal{D}'(\Omega)$, which implies $u=v$ a.e. And this answers my question why $\psi=\varphi$ a.e.

Is there something wrong in my "proof"? Or some better approaches?



$$ |\langle u_k,\varphi\rangle-\langle v,\varphi\rangle|\leqslant \int_{\Omega}a^{1/2}(x)|\nabla u_k-\nabla u||\varphi|\to 0 $$ seems only mean $a^{1/2}(x)|\nabla u|\to a^{1/2}(x)|\nabla v|$ in $\mathcal{D}'(\Omega)$, and this says nothing about my question I think!

Could any one help me?

  • $\begingroup$ The answer probably depends on $a$. Can you prove that elements of $D_0^1(\Omega;a)$ actually belong to some $L^p(\Omega)$? $\endgroup$
    – Siminore
    May 4, 2012 at 11:08
  • $\begingroup$ @DavideGiraudo: Thanks! Fixed. $\endgroup$
    – Y.Z
    May 4, 2012 at 11:19
  • $\begingroup$ @Siminore: Acturally this question appears in the proof of $D_0^1(\Omega;a)\hookrightarrow L^p(\Omega)$ for some $p$. $\endgroup$
    – Y.Z
    May 4, 2012 at 11:22

1 Answer 1


Your answer lies in the theory of distributions. Both your space $D^1_0(\Omega,a)$ and $L^p(\Omega)$ are subspaces of the space of distributions, that is to say, bounded linear functionals acting on $C^\infty_c(\Omega)$.

Now, without an embedding theorem, it is not given that just because $\varphi_n$ converges in one space, it needs to converge to another - in fact, for properly ugly cases, you can construct sequences that converge in one but not the other. Consider, for an example, the case of $\frac{\cos(nx)}{n}$ on some bounded interval. This sequence tends to 0 in any $L^p$ space as $p \rightarrow \infty$, but does not have a strong limit in $H^1$ (taking, in this case, $a(x) \equiv 1$).

However, if the same sequence converges in both spaces, then the properties of distributions say that they must converge to the same distribution.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .