2
$\begingroup$

Maybe the simplest non-trivial settings in which the spectrum of the Laplacian be can be computed is on the round sphere $\mathbf{S}^n$, and for products of manifolds. I want to use the two as examples to situate my question, which goes as follows.

Question. Can the first few eigenvalues of the manifold \begin{equation} \mathbf{S}^n \quad \text{or} \quad (M \times N,g \oplus h) \end{equation} say $\lambda_1 \leq \cdots \leq \lambda_D$ be computed 'directly', using an argument that does not involve the other, higher eigenspaces?

I would be happy with an argument that produces the first $D = 5$ eigenvalues for example in either setting. To be clear, in the latter case one would assume knowledge of the spectra of $M$ and $N$. The question arose in a more complicated setting, in which the larger eigenvalues seem inaccessible even a posteriori.

To explain potential difficulties, let me point out a step that the two computations have in common. (I am following the book of Berger, Gauduchon and Mazet [1].) Both rely on the Stone-Weierstrass theorem, combined with the following lemma [Lemma A.II.1 on p.143,1].

Lemma. Let $(M,g)$ be a compact manifold and $(V_i \mid i \in \mathbf{N})$ be a collection of vector subspaces of $C^\infty(M)$ so that for all $i \in \mathbf{N}$ there is $\lambda_i \in \mathbf{R}$ for which $\Delta \varphi = \lambda_i \varphi$ for all $\varphi \in V_i$. If $\sum_i V_i$ is $L^2$-dense in $C^\infty(M)$ then the spectrum of $(M,g)$ is $(\lambda_i \mid i \in \mathbf{N})$ and the $V_i$ are its eigenspaces.

I want to avoid going into too much detail, but just quickly illustrate how this is used in the latter example, to compute the spectrum of the product manifold $(M \times N,g \oplus h)$ [Prop. A.II.3 on p. 144,1]. Basically this states that its eigenspaces all factor into \begin{equation} \mathcal{E}_\nu(M \times N) = \mathcal{E}_\lambda(M) \otimes \mathcal{E}_\mu(N) \quad \text{with $\nu = \lambda + \mu$.} \end{equation} The first inclusion $\mathcal{E}_\nu(M \times N) \supset \mathcal{E}_\lambda(M) \otimes \mathcal{E}_\mu(N)$ is easy. It is the reverse inclusion that uses Stone-Weierstrass and the lemma cited above. A by-product of this is that the relation above is simultaneously established for all eigenvalues.

[1] Berger, M., Gauduchon, P., Mazet, E. Le spectre d'une variete riemannienne. Lect. Notes Math. 194 (1971).

$\endgroup$
11
  • 2
    $\begingroup$ Normally, eigenvalues and eigenspaces are computed using a min-max argument with the Rayleigh quotient. This gives you one eigenvalue at a time. $\endgroup$ Commented Aug 26, 2021 at 12:23
  • $\begingroup$ @RomainGicquaud While this is true, I'm not sure it's applicable here. Do you know of a way to use the Rayleigh quotients to calculate the spherical harmonics, for example? $\endgroup$
    – Leo Moos
    Commented Aug 26, 2021 at 12:35
  • 1
    $\begingroup$ So you mean you want to find the exact first eigenvalues of some manifold? You should be more precise regarding the manifolds you are considering. I don't think that there is a procedure to find the eigenvalues in general. Rayleigh quotient is useful to get estimates. $\endgroup$ Commented Aug 26, 2021 at 12:41
  • 1
    $\begingroup$ For spheres, tori and more generally for (compact) homogeneous spaces, you can rely on representation theory by the fact that the Laplacian commutes with the action of the symmetry group. This is how the spectrum of the sphere (and e.g. the complex projective space) are computed, see e.g. mathoverflow.net/questions/85481/… $\endgroup$ Commented Aug 26, 2021 at 13:00
  • 1
    $\begingroup$ It also gives a reference to the book "Harmonic analysis on homogeneous spaces" by Nolan Wallach where you will find more details. As quoted in an answer below, the Laplacian is (up to some constant) the Casimir operator of the representation. $\endgroup$ Commented Aug 26, 2021 at 13:35

0

You must log in to answer this question.