6
$\begingroup$

According to the answers in the the following questions: How to prove the spectrum of the Laplace operator? and What is spectrum for Laplacian in $\mathbb{R}^n$ , the spectrum of the Laplace operator $\Delta :H^2(\mathbb{R}^2)\subset L^2(\mathbb{R}^2)\to L^2(\mathbb{R}^2)$ is in fact $\sigma(\Delta)=(-\infty,0].$ However, I was not able to find a discussion on the eigenvalues of $\Delta$.

The set of eigenvalues $\sigma_p(\Delta)$ (also called point spectrum) is known to be contained in $\sigma(\Delta)$ and one can have $\sigma_p(\Delta)\subsetneq \sigma(\Delta)$. Indeed, by taking the Fourier transform $\mathcal{F}:L^2(\mathbb{R}^2)\to L^2(\mathbb{R}^2)$ of the eigenvalue problem one has $$\Delta u(x) = \lambda u(x),\;\;\forall x\in \Bbb R^2 \;\;\;\overset{\mathcal F}{\longrightarrow}\;\;\;\;-4\pi^2|\xi|^2\hat u(\xi) =\lambda \hat u(\xi), \;\;\;\forall\xi\in \Bbb R^2,$$ and this can only be satisfied by $\hat u=u=0$. This means that the only eigenvalue-eigenvector pair in this setting is $(\lambda,u)=(0,0)$ . Also, the same argument applies when $\Delta$ is seen as $\Delta:W^{m,p}(\mathbb{R}^2)\subset L^p(\mathbb{R}^2)\to L^p(\mathbb{R}^2)$ with $p\in [1,2)$ and $\mathcal F:L^p(\mathbb{R}^2)\to L^{p^*}(\mathbb{R}^2)$ with $1/p+1/p^*=1$.

Question 1. What happens when $p>2$ and the Fourier transform becomes distribution valued, so that the above elementary argument cannot be applied directly?

It seems if $u\in C^2(\mathbb{R}^2)$ is in fact an eigenvalue of $\Delta$, then it cannot be in $L^p(\mathbb{R}^2)$ for any $p\in [1,2]$. Depending on the answer to Q1, this might also hold for $p>2$. In any case, it seems that the $L^p$ framework is not suitable for this problem.

Question 2. On what space(s) could one define the domain of $\Delta$ to obtain non-trivial eigenvalues?


Edit. The crossed out sentence should be replaced by: "The point spectrum $\sigma_p(\Delta)$ is therefore empty."

$\endgroup$

1 Answer 1

8
$\begingroup$

The point spectrum coincides with the spectrum minus 0 if $p>2n/(n-1)$ and it is empty in the remaining cases ($n$ is the dimension). This is proved in G. Talenti: "Spectrum of the Laplace operator acting in $L^p(R^n)$", Indam, Symposia Mathematica vol VII, Academic Press 1971.

$\endgroup$
17
  • $\begingroup$ Thank you very much for the reference. However, it remains to show that for $p>2n/(n-2)$ (so in our case $p>4$) and $\Delta:W^{m,p}\to L^p$ , the spectrum is still given by $\sigma(\Delta)=(\infty,0]$. I have a proof of this using the Fourier transform that works, but only for $1\leq p\leq 2$. However, in our case, we need to study the spectrum for $p>4$! $\endgroup$
    – UserA
    Jan 29, 2022 at 10:31
  • 2
    $\begingroup$ @JochenGlueck Jochen you are right, the argument works for the spectrum minus 0. Of course, there is a way of avoiding semigroups. For $l\lambda$ not a negative number (and not 0), write formally the inverse through the Fourier transform and then use Mikhlin to check the boundedness. If $\lambda<0$ one shows it is an approximate eigenvlaue (the functions $e^{iax}$ should be an approximate eigenvector). For the point spectrum, one first show that if there is an eigenfuntions, then there is a radial one, by averaging, and then ends with a Bessel equation whose asymptotic is known. $\endgroup$ Jan 29, 2022 at 12:26
  • 1
    $\begingroup$ @JochenGlueck neither was I and this is actually a remarkable property! On another another, I think I made a mistake by calling the trivial pair $(\lambda, \hat u)=(0,0)$ an eingenpair. $\endgroup$
    – UserA
    Jan 29, 2022 at 12:51
  • 2
    $\begingroup$ @JochenGlueck Since you are interested I go on. If $\Delta u+k^2 u=0$ ($K \neq 0$) and $u$ is radial, then setting $u(r)=r^{(1-n)/2}w(r)$ you get a Bessel equation whose solutions at infinity oscillate lile $\sin$. Then the asympotics for $v$ at inifinity is like $r^{(1-n)/2}$ which is in $L^p$ for $p>2n(n-1)$. $\endgroup$ Jan 29, 2022 at 12:52
  • 1
    $\begingroup$ I checked better the book. Some papers are in italian but this one is in English. If you need I can scan and send a copy via mail. Then you should write to me via mail to [email protected] $\endgroup$ Jan 29, 2022 at 18:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.