I'm trying to combine two ways of looking at the Laplacian $\Delta$ on $\mathbb R^n$ (and on other domains).

Firstly, it is well known that this operator is essentially self-adjoint on $C_c^\infty(\mathbb R^n)$. Secondly, I know that for $f,g \in C^\infty_c(\mathbb R^n)$ it holds that $\langle u, \Delta u \rangle = \langle \nabla u, \nabla v \rangle $, where the inner-product is the $L^2$ one.

Now the domain $\mathcal D (\Delta) \subset L^2$ on which the Laplacian is self-adjoint must contain $C_c^\infty (\mathbb R^n)$, but I can't seem to find much more information on this. By the previous identity I feel like $\Delta u \in L^2$ should imply that $u \in H^1(\mathbb R^n)$, but I can't seem to find a way to prove this. I know it is possible to prove that $\mathcal D(\Delta) \subset H^2(\mathbb R^n$) via elliptic regularity etc, but it seems to me like there "should" be an elementary argument for showing $\mathcal D(\Delta) \subset H^1$ (preferably via the non-Fourier-transform characterisation of weak derivatives) but I can't seem to find it anywhere.

Edit: To clarify my question: I am looking for a simple proof of $\Delta u = f \in L^2$ implies $u \in H^1$ where $\Delta u$ is defined as the self-adjoint extension of the Laplacian $\Delta$ on $C^\infty_c$, preferrably one which also works on domains other than $\mathbb R^n$.

  • 2
    $\begingroup$ Are you familiar with the Friedrichs extension theorem? $\endgroup$ – Nate Eldredge Jul 11 at 21:14
  • 1
    $\begingroup$ I am not, but this seems like it would apply, thank you for the reference! $\endgroup$ – Nathanael Schilling Jul 11 at 21:25
  • $\begingroup$ You want to state the question a little differently for other domains $\Omega$, because the Laplacian is typically not essentially self-adjoint on $C^\infty_c(\Omega)$ in that case. $\endgroup$ – Nate Eldredge Jul 12 at 3:56
  • $\begingroup$ I was thinking of the (geodesically) complete Riemannian manifold case, where we do have essential self-adjointness. $\endgroup$ – Nathanael Schilling Jul 12 at 7:20
  • $\begingroup$ If you want simple proofs, I don't think it's wise to ban the Fourier transform. After taking FTs, your operator becomes a multiplication operator, and for these the questions you ask are extremely easy to analyze. $\endgroup$ – Christian Remling Jul 12 at 23:52

Maybe this is the argument you're looking for.

Suppose you know that $\Delta$ is essentially self-adjoint on $C^\infty_c(\mathbb{R}^n)$. This means $C^\infty_c$ is a core for $\Delta$, so for any $u \in D(\Delta)$, there is a sequence $u_n \in C^\infty_c$ such that $u_n \to u$ and $\Delta u_n \to \Delta u$ in $L^2$. In particular, we have $$|\langle \Delta(u_n - u_m), u_n - u_m \rangle_{L^2} | \le \|\Delta u_n - \Delta u_m\|_{L^2} \|u_n - u_m\|_{L^2} \to 0$$ using Cauchy-Schwarz. On the other hand, since $u_n, u_m \in C^\infty_c$, we are perfectly justified in integrating by parts to say that $\langle \Delta(u_n - u_m), u_n - u_m \rangle = \|\nabla (u_n - u_m)\|^2_{L^2}$. So, since $\|u_n - u_m\|_{L^2} \to 0$ and $\|\nabla (u_n - u_m)\|_{L^2} \to 0$, we have shown that $\|u_n - u_m\|_{H^1} \to 0$; in other words, $u_n$ is Cauchy in $H^1$. Now $H^1$ is a Hilbert space, so $u_n$ converges in $H^1$ to some $v \in H^1$. But we already know $u_n \to u$ in $L^2$, which is weaker than $H^1$ convergence, so we must have $u = v$, thus $u \in H^1$.


The Laplace operator is essentially self-adjoint: define $$ \mathcal D(∆)=\{u\in L^2, ∆u\in L^2\}=H^2. $$ Then for $u,v\in \mathcal D(∆)$, $\langle ∆ u,v\rangle=\langle u,∆v\rangle$: to prove this, consider $\lim u_k=u, \lim v_k=v \text{ in $H^2$}$ with $u_k, v_k$ smooth compactly supported. Then you have $$ \langle ∆ u,v\rangle=\lim_k\langle ∆ u,v_k\rangle=\lim_k\lim_l\langle ∆ u_l,v_k\rangle=\lim_k\lim_l\langle u_l,∆v_k\rangle=\lim_k\langle u,∆v_k\rangle=\langle u,∆v\rangle, \text{qed}. $$ Moreover you define for the symmetric operator $∆$, $$ \mathcal D(∆^*)=\{v\in L^2,\exists C, \forall u\in \mathcal D(∆), \vert\langle ∆u,v\rangle\vert\le C\Vert u\Vert_{L^2}\}. $$ Then as for all symmetric operators we have $\mathcal D(∆^*)\supset \mathcal D(∆)$. To get self-adjointness, you must prove $\mathcal D(∆^*)\subset \mathcal D(∆)$. Let $v\in \mathcal D(∆^*)$. Let $u$ be a smooth compactly supported function: then we have $$ \vert\langle ∆u,v\rangle\vert\le C\Vert u\Vert_{L^2}, \quad \text{i.e.} \quad \vert\langle u,∆v\rangle\vert\le C\Vert u\Vert_{L^2}, $$ proving that $∆ v\in L^2$ and since $v\in L^2$, the ellipticity of the Laplace operator implies that $v\in H^2=\mathcal D(∆)$. The above argument works for any symmetric elliptic operator.

  • $\begingroup$ I know the Laplacian is essentially self-adjoint, I was maybe a bit unclear but my question was why its domain of definition is a subset of $H^1$, not why the operator is essentially self-adjoint. The definition of $H^2$ you give isn't the standard one (existence of all weak partial derivatives of order 1 and 2 in $L^2$), I would like to know why $\mathcal D(\Delta) \subset H^1$ with the standard definition. $\endgroup$ – Nathanael Schilling Jul 11 at 20:26
  • $\begingroup$ Maybe you might ask yourself what the relation of $H^2$ and $H^1$ is. $\endgroup$ – Fabian Wirth Jul 11 at 20:31
  • $\begingroup$ I think I misunderstood the post above in that I thought it was a definition for $H^2$, whereas the definition was for $\mathcal D(\Delta)$ in terms of $H^2$. Nevertheless, my question still stands as the point I was asking about is contained in the reference to the ellipticity of the Laplace operator; which is nontrivial on general domains as far as I remember. This is why I would like a simple proof of the weaker statement that the domain is contained in $H^1$. $\endgroup$ – Nathanael Schilling Jul 11 at 20:49
  • $\begingroup$ In this situation, essential self-adjointness is exactly that the domain of the unique self-adjoint extension is $H^2$. (More generally, for semi-bounded operators, we can define analogous $H^2$, Friedrichs' extensions, and so on.) $\endgroup$ – paul garrett Jul 12 at 2:07
  • $\begingroup$ @WillieWong Yes, thanks, I did correct that. $\endgroup$ – Bazin Jul 12 at 19:17

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.