Let $M$ be a manifold with smooth boundary. We can consider the Dirichlet or the Neumann problem on $M$. Let $(\phi_k)$ be an orthonormal basis of eigenfunctions to the Dirichlet problem and let $(\psi_k)$ and orthonormal basis of eigenfunctions to the Neumann problem.

What can be said about the functions $\partial_{\mathbf{n}}\phi_k|_{\partial M}$ and $\psi_k|_{\partial M}$? For example, does a subsequence of them constitute an orthonormal basis of $L^2(\partial M)$? Are they solutions to some PDE?

  • $\begingroup$ I think you meant $\psi_k|_{\partial M}$ and $\partial_n\phi_k|_{\partial M}$ $\endgroup$ Apr 9, 2015 at 13:41
  • 1
    $\begingroup$ You are essentially asking about the image of a Poincaré-Steklov operator. I don't know the answer, but that term may help you do a more thorough search. $\endgroup$ Apr 9, 2015 at 20:50
  • $\begingroup$ There cannot be a geometric (differential) operator naturally associated with the (nontrivial) traces of the Dirichlet or Neumann eigenfunctions, simply because there are too many of these traces. $\endgroup$
    – ifw
    Apr 10, 2015 at 10:59
  • $\begingroup$ Hey ifw, could you elaborate? $\endgroup$ Apr 10, 2015 at 15:04
  • 1
    $\begingroup$ For the last point, one can also look at the oscillations of the $u_k$. This rules out the possibility of getting the same $u_k$ many times for any $M$ (not just the generic ones). $\endgroup$
    – ifw
    Apr 10, 2015 at 16:38

1 Answer 1


Consider the Dirichlet eigenfunctions. Let $\Delta \phi_k + \lambda_k^2\phi_k=0$, where $0<\lambda_1\leq \lambda_2\leq \lambda_3\dots$, and $\displaystyle u_k = \frac{\partial \phi_k}{\partial n}\bigr|_{\,\partial M}$. Here is a quotation from Bäcker, Fürstberger, Schubert, and Steiner when $\dim M=2$:

Our results show that the mean behaviour of the normalized boundary functions is very similar to the mean behaviour of eigenfunctions. The crucial difference between the two sequences of functions $\{\phi_k\}_{k\in\mathbb N}$, $\{u_k\}_{k\in\mathbb N}$ is that the eigenfunctions live on a two–dimensional space whereas the boundary functions live on a one–dimensional space. Since both $u_k$ and $\phi_k$ oscillate roughly with the same de Broglie wave length $2π/\lambda_k$, this leads to an overcompleteness of the set $\{u_k\}_{k\in\mathbb N}$. This statement can be made more explicit $\ldots$

More precisely, this means that for every $w \in C^\infty(\partial M)$ $$ w(s) = \sum_{k\in\mathbb N} \rho(\lambda-\lambda_k) w_k u_k(s) + O\left(1/\lambda\right) \tag{54} $$ holds with coefficients $$ w_k := \frac{\pi}{2\lambda_k^2}\int_{\partial M} w(s)u_k(s)\,ds. \tag{55} $$ This follows from ($\ldots$) by the method of stationary phase (see $\ldots$). Since $\rho$ is a rapidly decreasing function, this means that the boundary functions with spectral parameter $\lambda_k$ in an interval of fixed width around $\lambda$ form a complete set in the limit $\lambda\to\infty$. The number of these states grows like $\lambda$, in contrast to the number of all states up to energy $\lambda^2$, which, according to the Weyl formula, grows like $\lambda^2$. Therefore this result gives a quantitative measure of the overcompleteness of the set $\{u_k\}$.

In formulas (54), (55), $s$ is the generic point in the boundary $\partial M$, while $\rho\in S(\mathbb R)$ has Fourier transform which is even and compactly supported in a small neighborhood of $0$. Also the notation has been slightly adjusted (e.g., $\lambda_k$ in place of $k_n$ which is used by BFSS)

Similar results hold for the Neumann eigenfunctions and in higher dimensions, see e.g. Hassell and Barnett.


Your Answer

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

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