Set $ -\Delta: H^2(\mathbb{R}^3) \subseteq L^2(\mathbb{R}^3) \to L^2(\mathbb{R}^3) $. Then $ \mathcal{R}(-\Delta) $ is non-closed?

Sorry if this question is trivial. I am not familiar with theory of linear partial differential operators. This is a result I may use as a "black box".

Thank you in advance!


The range of the Laplace operator $-\Delta: L^2(\mathbb{R^d}) \supseteq H^2(\mathbb{R^d}) \to L^2(\mathbb{R^d})$ is not closed (for any dimension $d \ge 1$).

To see this, one can for instance use the following observations:

  • $-\Delta$ has empty point spectrum, so $0$ is not an eigenvalue of $-\Delta$.

  • $0$ is a spectral value of $-\Delta$, and the spectrum of $-\Delta$ has empty interior (within the complex plane) since $-\Delta$ is self-adjoint and its spectrum is thus real.

Those (and more) spectral properties of $-\Delta$ are very well-known in PDE theory and in Mathematical Physics; see for instance Theorem 7.17 in "G. Teschl: Mathematical Methods in Quantum Mechanics - With Applications to Schrödinger Operators (2014)".

Now the claim follows from the following general result:

Proposition. Let $A: E \supseteq D(A) \to E$ be a closed linear operator on a (complex) Banach space $E$. Assume that a given number $\lambda \in \mathbb{C}$ is not an eigenvalue of $A$, but contained in the topological boundary of the spectrum of $A$. Then $\lambda - A$ has non-closed range.

Proof. The operator $\lambda - A$ is an injective and continuous linear operator from $D(A)$ to $E$ (where $D(A)$ is endowed with the graph norm $\|\cdot\|_{D(A)}$). Hence, if $\lambda - A$ had closed range, it would be a linear homeomorphism from the Banach space $D(A)$ to the range of $\lambda - A$. In particular, $\lambda - A$ would be bounded below in the sense that there exists a constant $c > 0$ such that $\|(\lambda - A)x\|_E \ge c \|x\|_{D(A)}$ for all $x \in D(A)$.

However, as $\lambda$ is a value in the topological boundary of the spectrum of $A$, it follows that $\lambda$ is an approximate eigenvalue of $A$, meaning that there exists a sequence $(x_n) \subseteq D(A)$, normalized in $E$, such that $(\lambda - A)x_n \to 0$ in $E$. Note that $\|x_n\|_{D(A)} \ge \|x_n\|_E = 1$ for all $n$, so we obtain a contradiction to the fact that $\lambda - A$ is bounded below.

Remark. The fact that every $\lambda$ in the boundary of the spectrum $\sigma(A)$ is an approximate eigenvalue of $A$ is a simple consequence of standard properties of the resolvent of $A$; see for instance Lemma IV.1.9 in [Engel, Nagel: One-Parameter Semigroups for Linear Evolution Equations (2000)].

  • $\begingroup$ This is a really clear answer! However, I still need some appropriate references since I am not familiar with the details of spectrum of $ -\Delta $ and concepts: "topological boundary of spectrum", " approximate eigenvalue". $\endgroup$ – Yidong Luo Sep 12 '19 at 6:51
  • 1
    $\begingroup$ @YidongLuo: The topological boundary of the spectrum is no special concept, but just the boundary of the spectrum with respect to the usual topology in $\mathbb{C}$. I added a reference to the fact that every point in the boundary of the spectrum is an approximate eigenvalue (and I slightly changed the properties of the sequence $(x_n)$ in the above proof as to be consistent with this reference). The spectral properties of the Laplace operator on $\mathbb{R}^d$ are standard and can be probably be found in many books or manuscripts about PDE or matematical physics. $\endgroup$ – Jochen Glueck Sep 12 '19 at 8:22
  • $\begingroup$ I failed in searching the appropriate materials for the spectral properties shown above. Could you help recommend some materials? $\endgroup$ – Yidong Luo Sep 12 '19 at 9:49
  • 1
    $\begingroup$ @YidongLuo: I added a reference. $\endgroup$ – Jochen Glueck Sep 12 '19 at 10:18
  • $\begingroup$ @JochenGlueck Is it in general true that a continuous operator has a closed range if, and only if, 0 is not an eigenvalue and does not lie on the boundary of the spectrum? $\endgroup$ – Ivan Nov 10 '20 at 12:32

After comprehension of the answer by Jochen Glueck, we give an answer with our background.

Use the following observation:

  • $ 0 $ is a spectral value of $ -\Delta $, and further since there exist no point spectrum and residual spectrum of $ -\Delta $, we know $ 0 $ locates in the continuous spectrum of $ -\Delta $.

This yields that \begin{equation*} \mathcal{N}(-\Delta) = \{ 0 \}, \ \ \overline{\mathcal{R}(-\Delta)} = L^2(\mathbb{R}^3) \\ (-\Delta)^{-1} \ \textrm{unbounded}. \end{equation*}

Now assume that $ \mathcal{R}(-\Delta) $ is closed, by

Proposition: If $ A \in \mathcal{C}(X,Y) $, then \begin{equation} A|_{\mathcal{C}(A)} \ \textrm{has a bounded inverse} \Longleftrightarrow \mathcal{R}(A) \ \textrm{closed}, \ \textrm{where} \ \mathcal{C}(A) := \mathcal{D}(A)\cap \mathcal{N}(A)^\perp \end{equation} (See Chapter 9.3, 2 (M) in "A. Israel, T. Greville, Generalized Inverses Theory and Applications. Second Edition, Springer-Verlag, New York, 2003.")

we have $ (-\Delta)^{-1} $ bounded. This is a contradiction.


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.