Let $\Omega\subset\mathbb{R}^N$ be a bounded regular domain and consider the Sobolec space $W^{1,p}(\Omega)$ for $p\in (1,\infty)$. I have some doubts with th trace theorem.

Roughly speaking, the Trace theorem states that there is a linear functional $$T:W^{1,p}(\Omega)\to L^p(\partial \Omega)$$

such that $Tu=u_{|\partial\Omega}$ for every $u\in C^\infty(\overline{\Omega})$.

My questions is:

The only functions for which $Tu=u_{|\partial\Omega}$ are the $C^\infty(\overline{\Omega})$ functions, or can we find, for instance, functions not continuous in $W^{1,p}(\Omega)$ such that the same equality holds?

Thank you

  • $\begingroup$ You can relax the requirement of $u \in C^{\infty}(\bar{\Omega})$ to be $C(\bar{\Omega})$, but your question is essentially the same. $\endgroup$
    – al0
    May 28, 2013 at 12:38
  • $\begingroup$ Out of personal curiosity: What measure is used on $\partial\Omega$? $N-1$-dimensional Hausdorff measure? Surely it can't be $N$-dimensional Lebesgue measure, because $\partial\Omega$ will be a Lebesgue(-$N$) null set for most "nice" $\Omega$ making $L^p(\partial\Omega)$ rather boring. $\endgroup$
    – kahen
    May 28, 2013 at 13:37
  • $\begingroup$ @kahen, yes it is surface measure. $\endgroup$
    – Tomás
    May 28, 2013 at 14:39

1 Answer 1


The reason that the equality $Tu=u_{|\partial \Omega}$ is stated for functions in $C(\overline{\Omega})$ is that for other functions it is not clear what $u_{|\partial \Omega}$ means. Of course, we can take any function $u\in W^{1,p}(\Omega)$ and extend its domain to $\overline{\Omega}$ by letting $u$ be equal to $Tu$ on the boundary. This will be an instance of $Tu=u_{|\partial \Omega}$ being true despite $u$ not necessarily being in $C(\overline{\Omega})$. But this is merely a tautology that does not teach us anything new.

But there are useful and nontrivial results along the lines of your question: starting with $\phi\in L^p(\partial \Omega)$, one can extend $\phi$ to a function in $\Omega$ (by solving the Dirichlet problem for some elliptic operator) and then recover $\phi$ from $u$ via nontangential limits. This does not require $\phi$ to be continuous. For a simple example, take $\Omega=\{z\in\mathbb C: |z|<1\}$ and $u=\log|z-1|$. Then $u\in W^{1,p}(\Omega)$ for $p<2$. The trace operator agrees with the boundary values of $u$ understood in the sense of nontangential limits.

  • $\begingroup$ Interesting. So if $\Omega=B(0,1)=B$, can I conclude from here math.stackexchange.com/questions/404142/… that for almost $t\in (0,1]$ the trace of $u$ in $W^{1,p}(B_t)$ is the restriction of $u$ to $\partial B_t$? $\endgroup$
    – Tomás
    May 28, 2013 at 14:48
  • $\begingroup$ @Tomás For almost every $t$, almost every point of $\partial B_t$ is a Lebesgue point of $u$. Does the trace agree with the values at Lebesgue points on the boundary? I'm not sure. $\endgroup$ May 28, 2013 at 23:44
  • $\begingroup$ This is really a annoying thing and I am having a hard time in trying to understand it. If you have access to this paper, please take a look on it Proposition 1: Zhikov, V. V.; Pastukhova, S. E. The compensated compactness principle. (Russian) Dokl. Akad. Nauk 433 (2010), no. 5, 590--595; translation in Dokl. Math. 82 (2010), no. 1, 590–595 $\endgroup$
    – Tomás
    May 29, 2013 at 11:16

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.