6
$\begingroup$

Let $\Omega \subset \mathbb{R}^n$ be open, then $H_0^m(\Omega) := \overline{C^{\infty}_c(\Omega)}$ with respect to the Sobolev norm $\|\|_{H^m(\Omega)}$. The problem is that I don't really see what kind of functions are in this space. Apparently for $m=0$ we just have $H_0^0 (\Omega)=L^2(\Omega).$

Despite, from the definition I don't see a defining property for all the functions in $H_0^m$?- I mean it is intuitively clear to me what $H^m$ is (weakly differentiable functions that are square integrable in every derivative) , but this closure definition is pretty unclear to me in the sense that I don't understand what these functions have in common.

What is so special about this space?

In particular, what is the difference between $H^m(\Omega)$ and $H_0^m(\Omega)$? It looks to me as if they are quite often the same, as $C_c^{\infty}$ should be dense in $H^m$

$\endgroup$
  • 1
    $\begingroup$ $C_c^\infty$ is not dense in $H^m(\Omega)$ unless $\Omega$ is all or almost all of $\mathbb{R}^n $. $\endgroup$ – user147263 Feb 25 '15 at 20:26
10
$\begingroup$

It may help to observe that in general, $C_c^\infty(\Omega)$ is not dense in $H^m(\Omega)$, so the closure $\overline{C_c^\infty(\Omega)}$ is a proper subspace, and this definition actually says something nontrivial.

Take for example $n=1$ and $\Omega = (0,1)$, the open unit interval. Observe that the constant function 2 is in $H^1(\Omega)$. However, for any $f \in C_c^\infty(\Omega)$, $\|2-f\|_{H^1} \ge 1$.

Consider two cases. If $f \le 1 $ everywhere, then clearly $\|(2-f)\|_{L^2} \ge 1$. Otherwise, suppose there exists $x$ such that $f(x) \ge 1$. Since $f$ is compactly supported inside $(0,1)$, we have $f(0)=0$, and so $$\begin{align*} 1 \le |f(x)| &= \left|\int_0^x f'(t)\,dt\right| && \text{FTC} \\ &\le \int_0^x |f'(t)|\,dt && \text{triangle inequality} \\ &\le \int_0^1 |f'(t)|\,dt && \text{integrating over a larger set}\\ &\le \left(\int_0^1 |f'(t)|^2\,dt\right)^{1/2} && \text{Jensen or Cauchy--Schwarz} \end{align*}$$ and in particular $\|(2-f)'\|_{L^2} \ge 1$. In either case we have $\|2-f\|_{H^1} \ge 1$. So $2 \notin H^1_0(\Omega)$.

By a similar argument, if you know that $H^1((0,1))$ consists (up to almost everywhere equivalence) of the absolutely continuous functions on $(0,1)$ whose derivatives are in $L^2$, then you can show that $H^1_0((0,1))$ consists precisely of those $g \in H^1((0,1))$ satisfying $\lim_{x \to 0^+} g(x) = \lim_{x \to 1^-} g(x) = 0$. In other words, all those $g$ satisfying Dirichlet boundary conditions.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ can we generalize this Dirichlet boundary condition property or is this a special case in 1-dimension? $\endgroup$ – Jacobian Feb 25 '15 at 20:50
  • $\begingroup$ @Jacobian: You can generalize it, but it gets more complicated since in higher dimensions, $H^m$ functions do not generally have continuous representatives. The appropriate generalization is the trace theorem mentioned by Ian: if $T : H^1(\Omega) \to L^2(\partial \Omega)$ is the operator he describes, you are supposed to think of $Tf$ intuitively as "take the continuous extension of $f$ to $\overline{\Omega}$ and restrict to $\partial \Omega$". (For some $f$ there may not actually be such a continuous extension so you have to interpret this in a sort of "almost everywhere" sense.) $\endgroup$ – Nate Eldredge Feb 25 '15 at 20:55
  • $\begingroup$ @Jacobian: Then $H^1_0(\Omega)$ is those functions whose trace is 0; i.e. intuitively, those which vanish on the boundary $\partial \Omega$. $\endgroup$ – Nate Eldredge Feb 25 '15 at 20:56
  • $\begingroup$ @Jacobian: But don't take this too seriously; in some cases the boundary can be "too small" to impose Dirichlet boundary conditions. For instance, if $\Omega = \mathbb{R}^3 \setminus \{0\}$ is $\mathbb{R}^3$ minus the origin, you can check that in fact $H^1_0(\Omega) = H^1(\Omega) = H^1(\mathbb{R}^3)$, and in particular contains the constants. This has to do with the fact that points in $\mathbb{R}^3$ have zero capacity. $\endgroup$ – Nate Eldredge Feb 25 '15 at 20:59
3
$\begingroup$

First of all, $C^\infty_c$ is not dense in all of $W^{k,p}$ ($H^k=W^{k,2}$) in general. For instance, take a bounded domain. Then a constant function is in $W^{1,p}$. But you typically can't approximate it in $W^{1,p}$ by compactly supported functions. The function values will converge easily enough, but the constant function has a zero gradient, while the gradients of the compactly supported functions are bounded away from zero, in the sense of $L^p$. Since constant functions are smooth, we've actually shown that $C^\infty_c$ isn't even dense in $C^\infty$, if we equip both with the $W^{1,p}$ metric.

I think to properly understand this phenomenon, you need the trace theorem. For functions which are smooth up to the boundary, we have a well-defined linear operator which restricts them to the boundary. The trace theorem says that this operator has an extension to all of $W^{1,p}$, and that the extension is continuous as a function from $ W^{1, p}(\Omega) $ into $ L^p (\partial \Omega)$. This operator is called the trace, and commonly denoted by $T$. (Note that the trace theorem requires a certain amount of boundary regularity.)

Where the trace theorem applies, $ W^{1, p}_0$ consists of those Sobolev functions whose trace is zero. (This is easy to prove, because the restriction of a compactly supported function to the boundary is zero, and then you use continuity to conclude.)

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ so it has nothing to do with Dirichlet boundary conditions, let's say? $\endgroup$ – Jacobian Feb 25 '15 at 20:28
  • $\begingroup$ @Jacobian Yes it does. A smooth function in $W^{1,p}_0$ satisfies a homogeneous Dirichlet boundary condition. A nonsmooth function in $W^{1,p}_0$ satisfies the appropriate generalization of this idea (since it has no pointwise restriction). $\endgroup$ – Ian Feb 25 '15 at 21:27

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.