MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let the weighted sobolev space $M^p_{s,\delta}$ be the completion of $C_0^\infty(\mathbb{R}^n)$ in the norm \begin{equation} \sum_{\left\vert\alpha\right\vert\leq s}\left\Vert(1+\left\vert x\right\vert^2)^{(\delta+\left\vert\alpha\right\vert)/2}D^\alpha \varPhi\right\Vert_{L^p}. \end{equation} In dimension $4$ one can state that if $p>1,\delta\in(-\frac{4}{p},\frac{4}{p})$ and we pose $L^p_\delta=M^p_{0,\delta}$ then we have that the operator $\Delta^2$ is an isomorphism from $M^p_{4,\delta}$ into $\Lambda$, where \begin{equation} \Lambda=\left\{f\in L^p_{4+\delta},\int_{\mathbb{R}^4}f\, dx=0\right\}. \end{equation} More precisely, if $p>1,\delta>-\frac{4}{p}$ and $s>\frac{4}{p}$, then $M^p_{s,\delta}$ is compactly embedded in $C_0(\mathbb{R}^4)$.

Question: does anyone know how to prove the compactness of the embedding? Does this result hold also in higher dimension? Thanks

share|cite|improve this question
How do you prove the embedding itself? – timur Sep 3 '13 at 13:56
Could you add references for your statements? It helps. – András Bátkai Sep 3 '13 at 14:28

Let $B_R\subset\mathbb{R}^n$ denote the $n$-dimensional open ball of radius $R$, centred at the origin. Then we have $$ \|\partial^\alpha u\|_{L^p(B_R)} \leq \max\{1,R^{-\delta-s}\} \|\langle x\rangle^{\delta+s}\partial^\alpha u\|_{L^p(\mathbb{R}^n)} \qquad\textrm{for}\quad |\alpha|=s, $$ and $$ \|u\|_{L^p(B_R)} \leq \max\{1,R^{-\delta}\}\|\langle x\rangle^{\delta}u\|_{L^p(\mathbb{R}^n)}, $$ where $\langle x\rangle=\sqrt{1+x^2}$. Hence, for any $R>0$, the restriction to $B_R$ of any $u\in M^p_{s,\delta}$ is in $W^{s,p}(B_R)$. In other words, the restriction map $J_R:u\mapsto u|_{B_R}$ is well defined and bounded as a map $J_R:M^p_{s,\delta}\to W^{s,p}(B_R)$. By the embedding result for the usual Sobolev spaces, this implies that $M^p_{s,\delta}\subset C(\mathbb{R}^n)$ for $s>\frac np$. Moreover, by a simple scaling argument it is easy to see that $$ \|u\|_{L^\infty(A_R)} \leq c \left(R^{-n/p}\|u\|_{L^{p}(A_R)} + R^{s-n/p}|u|_{W^{s,p}(A_R)}\right), $$ for $s>\frac np$, where $A_R=B_{2R}\setminus \overline{B}_R$ is an annulus. On $A_R$ with $R>1$, we have $\langle x\rangle\sim R$, which implies that $$ \|u\|_{L^{p}(A_R)} \leq c R^{-\delta}\|u\|_{M^p_{s,\delta}}, \qquad\textrm{and}\qquad |u|_{W^{s,p}(A_R)} \leq c R^{-\delta-s}\|u\|_{M^p_{s,\delta}}, $$ for $R>1$, and thus $$ \|u\|_{L^\infty(A_R)} \leq c R^{-\delta-n/p}\|u\|_{M^p_{s,\delta}}. \qquad\qquad(*) $$ It shows that if $u\in M^p_{s,\delta}$ with $s>\frac np$ and $\delta>-\frac np$, then $|u(x)|\to0$ uniformly as $x\to\infty$. In particular, we have the continuous embedding $M^p_{s,\delta}\hookrightarrow C_0(\mathbb{R}^n)$.

Let us now discuss the compactness of $M^p_{s,\delta}\hookrightarrow C_0(\mathbb{R}^n)$. By the Rellich-Kondrashov theorem, the restriction map $J_R:M^p_{s,\delta}\to C_b(B_R)$ is compact, with $C_b(B_R)$ denoting the bounded continuous functions on $B_R$ equipped with the uniform norm. Let $\{u_k\}$ be a bounded sequence in $M^p_{s,\delta}$. Then we can extract a subsequence $$ u_{11},\ldots,u_{1m},\ldots, $$ such that $\|u_{1m}-u_{1m'}\|_{L^\infty(B_1)}\to0$ as $m,m'\to\infty$. From this subsequence, we can further extract a subsequence $$ u_{21},\ldots,u_{2m},\ldots, $$ such that $\|u_{2m}-u_{2m'}\|_{L^\infty(B_2)}\to0$ as $m,m'\to\infty$. We repeat this process to get the doubly-infinite sequence $$ \begin{split} u_{11},&\ldots,u_{1m},\ldots,\\ \ldots&\ldots \ldots\\ u_{j1},&\ldots,u_{jm},\ldots,\\ \ldots&\ldots \ldots \end{split} \qquad\qquad(**) $$ which has the property that $\|u_{jm}-u_{jm'}\|_{L^\infty(B_j)}\to0$ as $m,m'\to\infty$. Now as usual, we look at the diagonal $\{u_{jj}\}$. Let $\varepsilon>0$ be given. Then we choose (an integer) $R$ so large that $$ \|u_k\|_{L^\infty(\mathbb{R}^n\setminus B_R)}\leq\varepsilon, $$ for all $k$. This is possible because of the uniform decay $(*)$. Hence $$ \|u_{jj}-u_{mm}\|_{L^\infty(\mathbb{R}^n)} \leq \|u_{jj}-u_{mm}\|_{L^\infty(B_R)} + \|u_{jj}-u_{mm}\|_{L^\infty(\mathbb{R}^n\setminus B_R)} \leq \|u_{jj}-u_{mm}\|_{L^\infty(B_R)} + 2\varepsilon. $$ If $j\geq R$ and $m\geq R$ then both $u_{jj}$ and $u_{mm}$ are elements of the $R$-th row of $(**)$, and since this row is a Cauchy sequence in $L^\infty(B_R)$, we conclude that $\{u_{jj}\}$ is Cauchy in $L^\infty(\mathbb{R}^n)$.

share|cite|improve this answer
Thank you, this seems to work. Can I ask you for some references on this topic? – gin111 Sep 4 '13 at 10:18
@gin111: I don't know any references on weighted spaces, but I know references on Sobolev spaces, such as Adams and Triebel. – timur Sep 4 '13 at 13:08
@gin111: Please consider accepting this answer if you think it was helpful. – timur Aug 27 '14 at 17:44

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.