In the book Sobolev Spaces with Application of Maz'ya, $\mathring {L^k_p}(\Omega)$ is defined to be the completion of $\mathcal D(\Bbb R^n)$ under the norm $||\nabla^ku||_{L_p(\Omega)}$.

For nice domain (and correct values of $k,p,q,n$), Sobolev's embedding theory tell us that $\mathring {L^k_p}(\Omega)\hookrightarrow L_q(\Omega)$ continuously. However, I am having a trouble with understanding the space for more general cases. Actually, in the same book we have this theorem:

Let $\Omega=\Bbb R^n$, $n<kp$, $p>1$. For a multi-index $\alpha$ of order $|\alpha|\le l-n/p$, there exists a sequence $u_{m}\in\mathcal D(\Bbb R^n)$ such that $u_{m}\to 0$ in $\mathring {L^k_p}(\Bbb R^n)$ while $u_{m}\to x^{\alpha}$ in $\mathcal D'(\Bbb R^n)$.

The way I interpreted this is that for these particular values of $n,p,k$; some "polynomials" (that are obviously not in any $L_q(\Bbb R^n)$) are arbitrary close to $0$ of $\mathring {L^k_p}(\Omega)$.

Here are some of my questions:

  1. How does a typical element of $\mathring {L^k_p}(\Omega)$ look like?

  2. We may define, for $u\in\mathring {L^k_p}(\Omega)$, $$|u|_{\mathring {L^k_p}}:=\lim_{m\to\infty} ||\nabla^ku_m||_{L_p}$$ where $u_m\in\mathcal D(\Omega)$ converges to $u$ in the above sense. Since $\nabla^ku_m$ actually converges in $L_p$, we may even write its limit suggestively as $\nabla^ku\in L_p$ but what does it mean when $u\notin \mathcal D'(\Omega)$? Surely it cannot be the weak derivative of $u$ right?

  3. $|u|_{\mathring {L^k_p}}$ as defined above is just a seminorm right (generally)? Is this space even Hausdorff?

  • $\begingroup$ By the Poincaré inequality, we get $|\cdot|_{\mathring L^k_p(\Omega)}$ is a norm on $D(\Omega).$ I suspect you can prove a variant of the Hardy inequalities which will give a good description of these spaces, but I don't know about the specifics. $\endgroup$ – user17597 Feb 6 '18 at 17:09
  • $\begingroup$ @user17597 My question aimed pretty much at the cases where the values of $n,p$ is such that Sobolev embedding does not hold. How would we use Poincare inequality in those settings? $\endgroup$ – BigbearZzz Feb 6 '18 at 17:19
  • $\begingroup$ The application of Poincaré works in complete generality: if $u \in D(\Omega),$ then it is supported in some open bounded subset $U \subset \Omega.$ Writing $u(x) = \int_{-\infty}^{x_1} \partial_{x_1}u(t,x_2,\dots,x_n)\,\mathrm{d}t$ and using the Hölder inequality, we get a bound of the form $\lVert u \rVert_{L^p(U)} \leq C\lVert Du\rVert_{L^p(U)}$ where $C$ depends on $U$ and $p.$ $\endgroup$ – user17597 Feb 6 '18 at 17:37
  • $\begingroup$ @user17597 But an element of $\mathring L^k_p(\Omega)$ can be a limit of $\phi_n\in\mathcal D(\Omega)$ where each $\phi_n$ is supported in $U_n$. There's nothing preventing $U_n$ from expanding without bound so the constant from Poincare is not really a constant. $\endgroup$ – BigbearZzz Feb 6 '18 at 20:40
  • $\begingroup$ Of course this estimate will not hold in the limit, but the fact that it's a norm on $D(\Omega)$ means its completion will be a Banach space. I was mentioning this to answer your third question. $\endgroup$ – user17597 Feb 6 '18 at 21:00

The answer presented here is copied from the paper [2]. Many similar results (sometimes with more complicated proofs) can be found in [1]. Let the space $L^{k,p}$ be defined by: $$ L^{k,p}(\mathbb{R}^n)=\{ f\in \mathcal{D}'(\mathbb{R}^n):\, \nabla^kf\in L^p(\mathbb{R}^n)\}, \quad \Vert f\Vert_{L^{k,p}}=\Vert \nabla^kf\Vert_p. $$

Theorem (Theorem 4 in [2]). Let $1\leq p<\infty$ and $k=1,2,\ldots$ Then $\mathcal{D}(\mathbb{R}^n)$ is dense in $L^{k,p}$ if and only if $n>1$ or $p>1$.

Here density is with respect to the seminorm $\Vert \nabla^kf\Vert_p$. That is for any $f\in L^{k,p}(\mathbb{R}^n)$ there is a sequence $f_i\in \mathcal{D}(\mathbb{R}^n)$ such that $\Vert \nabla^k(f-f_i)\Vert_p\to 0$. Therefore the space coincides with the completion that is described in the question i.e., $\mathring {L^k_p}(\mathbb{R}^n)=L^{k,p}(\mathbb{R}^n)$. Formally, the spaces $L^{k,p}$ are defined as distributions and only their $k$-th order derivatives are functions. However we have $$ L^{k,p}(\mathbb{R}^n)\subset W^{k,p}_{\rm loc}(\mathbb{R}^n)=\{f\in L^p_{\rm loc}:\, \nabla^\ell f\in L^p_{\rm loc} \text{ for all $0\leq\ell\leq k$\}}. $$ This is a classical result, but a short and self-contained proof is given in this post https://mathoverflow.net/a/296464/121665.

This gives an answer to questions 2 and a partial answer to question 1:

$$ \mathring {L^k_p}(\mathbb{R}^n)=L^{k,p}(\mathbb{R}^n)\subset W^{k,p}_{\rm loc}(\mathbb{R}^n). $$

Clearly $u$ is a function and $\nabla^k u$ defined as a limit of approximations (as in question 2) is a weak derivative of $u$.

So how does a typical element of $L^{k,p}$ look like?

If $\Omega$ is a bounded domain, then by the Poincare inequality $\Vert u\Vert_{W^{k,p}}\leq \Vert \nabla^k u\Vert_{p}$ for $u\in\mathcal{D}(\mathbb{R}^n)$. This gives

If $\Omega$ is a bounded domain, then $\mathring {L^k_p}(\Omega)=W^{k,p}_0(\Omega)$.

A precise description of the space $\mathring {L^k_p}(\mathbb{R}^n)$ is available when $kp<n$.

If $kp<n$ or $k=n$, $p=1$ we define the homogeneous Sobolev space by $$ \mathring W^{k,p}(\mathbb{R}^n)=\left\{f:\, \nabla^\ell f\in L^{p_{\ell}^*}(\mathbb{R}^n),\ \ p_\ell^*=\frac{np}{n-(k-\ell)p},\ \ \ 0\leq\ell\leq k\right\}. $$ This space is equipped with the norm $$ \Vert f\Vert_{\mathring W^{k,p}} =\sum_{\ell=0}^k \Vert\nabla^\ell f\Vert_{p_\ell^*}. $$

It follows from the Sobolev embedding that $W^{k,p}\subset \mathring W^{k,p}$, but the later space is larger. The next result provides a complete answer to question 1 in the case when $kp<n$ and $1\leq p<\infty$.

Theorem. (Theorem 5 in [2]). Let $kp<n$ and $1\leq p<\infty$. The for every $f\in L^{k,p}(\mathbb{R}^n)$ there exists a unique polynomial $P^{k-1}f\in \mathcal{P}^{k-1}$ of degree $\leq k-1$ such that $f-P^{k-1}f\in \mathring W^{k,p}(\mathbb{R}^n)$. Morover $$ \Vert f- P^{k-1}f\Vert_{ \mathring W^{k,p}}\leq C\Vert\nabla^k f\Vert_p. $$

In other words $L^{k,p}= \mathring W^{k,p}+\mathcal{P}^{k-1}$.

For other related results, see [1], [2] and [3].

[1] O. V. Besov, V. P. Ilʹin, S. M. Nikolʹskiĭ, Integral representations of functions and imbedding theorems. Vol. I & II. Scripta Series in Mathematics. Edited by Mitchell H. Taibleson. V. H. Winston & Sons, Washington, D.C.; Halsted Press [John Wiley & Sons], New York-Toronto, Ont.-London, 1979.

[2] P. Hajłasz, A. Kałamajska, Polynomial asymptotics and approximation of Sobolev functions, Studia Math. 113 (1995), 55-64.

[3] P. J. Rabier, $L^p$ regularity of homogeneous elliptic differential operators with constant coefficients on $\mathbb{R}^N$. Rev. Mat. Iberoam. 34 (2018), 423–454.

  • $\begingroup$ I've just seen your edit to the answer. This seems interesting, I'll read it in detail soon, thank you very much. $\endgroup$ – BigbearZzz Apr 10 '18 at 13:29
  • $\begingroup$ @BigbearZzz You can also check mathoverflow.net/a/296464/121665 since it is related. While I consider the first order derivatives there, the higher order case follows by induction. $\endgroup$ – Piotr Hajlasz Apr 10 '18 at 13:32
  • $\begingroup$ Right! I completely forgot about that. Thank you for reminding me. $\endgroup$ – BigbearZzz Jun 15 '18 at 22:55

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.