Consider a surface $\Gamma$ in $R^3$. The surface $\Gamma$ is a graph, i.e. $\Gamma = (x,y, h(x,y))$, for $x \in R^2$ and some smooth function $h$, where $h$ and all its derivatives are periodic on [0,1]^2.

I'm trying to estimate the eigenfunctions and eigenvalues of the Laplace-Beltrami operator for this surface. Of course, the laplace beltrami operator can be very easily be expressed in local coordinates on $[0,1]^2$, and so the problem is effectively approximating the spectra of this second order elliptic operator on $L^2[0,1]^2$ with periodic boundary conditions. This is what I am doing, however I still cannot obtain any useful estimates.

Is anyone aware of any references, or has had any experience with this problem?

  • $\begingroup$ Are you interested in the low lying eigenvalues or the asymptotics of the eigenvalues? The second one should follow like Weyl's law .. $\endgroup$
    – Helge
    Mar 12, 2011 at 18:18
  • $\begingroup$ Hi, thanks for your answer. It's mainly the low lying eigenvalues. Even a spectral gap result would be a great start. $\endgroup$ Mar 13, 2011 at 9:33
  • $\begingroup$ Are you not interested also in the spectrum of the non-compact surface $\Gamma$. Periodic BC gives the (discrete) spectrum of the (non flat) torus $\Gamma/Z^2$. $\endgroup$
    – BS.
    Mar 13, 2011 at 13:16
  • 1
    $\begingroup$ @BS: There is more than one operator involved for the non-compact case. It's still "estimating eigenvalues". $\endgroup$
    – Helge
    Mar 14, 2011 at 3:41
  • $\begingroup$ I'm interested in the discrete spectrum and the corresponding eigenfunctions. My ultimate aim is to estimate the H1 inner product of a function ($x\cdot k$, where $k$ is a constant vector) against the eigenfunctions. $\endgroup$ Mar 14, 2011 at 12:23

1 Answer 1


Some suggestions:

  1. To show there is a spectral gap, one can use the following approach: The variational characterization of the first eigenvalue $$ E_1 = \inf_{\|\psi\| = 1} \langle \psi, H \psi \rangle $$ can be used to obtain an upper bound on the first eigenvalue. Then Temple's inequality (I think it's in Reed-Simon IV) can be used to obtain a lower bound on the second eigenvalue. Note in order to show the existence of a spectral gap, one needs to study all the operators $H(k)$ of the Floquet-Bloch decomposition.

  2. But I would guess that at least as long as $h$ is "sufficiently small", there is no spectral gap (without having done any of the computations). Denote by $E_1(k)$ and $E_2(k)$ the first and second eigenvalue of $H_0(k)$. Here $H_0$ denotes the usual Laplacian and the Floquet--Bloch decomposition is done with respect to $[0,1]^2$. I believe that one has that $$ \sup_{k} E_1(k) > \inf_{k} E_2(k). $$ Then using that your operator will be a small perturbation of the Laplacian (in an appropriate sense), one should obtain that there is no spectral gap.

    2b. (added in edit) It is clear that for fixed $k$, one has $E_1(k) < E_2(k)$. This follows from the first eigenfunction being positive.

  3. There is what is called "Bohr-Sommerfeld conjecture", which is related to higher eigenvalues.

  • $\begingroup$ The first eigenvalue will always be zero, with corresponding eigenfunction being constant. By spectral gap I mean the gap between first and second eigenvalues. Numerically I have computed the spectrum of my operator for various $h$ and have observed that a spectral gap exists and appears to be inversely proportional to the surface area of my surface (over [0,1]^2), though I can't prove this. Temple's inequality seems to be a useful tool for this type of problem. Thanks for telling me about it. I admit I know nothing about Floquet-Bloch decompositions, I'm reading about this now. Thanks. $\endgroup$ Mar 14, 2011 at 12:38
  • $\begingroup$ The first eigenvalue of $H_0(k)$ is not always $0$ nor is the eigenfunction constant. Look up Floquet-Bloch decomposition. You need to consider multiple operators. $\endgroup$
    – Helge
    Mar 14, 2011 at 19:20
  • $\begingroup$ I am reading about the Floquet-Bloch decomposition and I am realising that rather that do all the estimates above, I simply need to compute the Hessian of the first bloch eigenvalue. I suspect this might be just as hard as what I was doing previously, but I will attempt it. $\endgroup$ Mar 15, 2011 at 10:43

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.