3

1

What's the most natural way to establish the asymptotics of $\Delta$ on a compact Riemannian manifold M of dimension N? The asymptotics should be $$\#\{v < A^2\} = \mathrm{const}*\mathrm{vol}(M)*A^n + o(\mathrm{something})$$ (Perhaps one could consider first a case of Kahler manifold? The Laplacian is partiularly simple there).

flag
FYI, the most covariant "Laplacian" when the metric is not constant is the Laplace-Beltrami operator |g|^{-1/2} \partial_i |g|^{1/2} g^{ij} \partial_j (see en.wikipedia.org/wiki/…). I think that this agrees with yours, which depends on your choice of coordinates (but is the one I actually use) up to a first-order differential operator, and so the asymptotics should agree on compact manifolds. Oh, also, I don't think you should have a wedge? g^{ij} is symmetric in i,j, whereas \partial_i \wedge \partial_j looks antisymmetric? – Theo Johnson-Freyd Oct 19 2009 at 6:33
Ah, yes, it looks like the one I wrote in many, but not all, metrics. And I've written a corresponding symplectic form. My bad. – Ilya Nikokoshev Oct 19 2009 at 17:34

1 Answer

4

The most natural way is to study the short-time asymptotics of the heat or wave kernel on M. For example, you can use the heat kernel $p_t(x,y) = \sum_i e^{-lambda_i t} f_i(x) \overline{f_i(y)}$ where $f_i$ are the eigenfunctions with eigenvalues $\lambda_i$. This is a fundamental solution to the heat equation.

When $t$ is small then you can construct a good approximation to $p_t$ near any particular $x$ by hand, using Fourier analysis in local co-ordinates. The end result is that that $p_t(x,x) \approx C t^{-n/2}$. Now integrate this estimate $dx$, noting that $\int_M p_t(x,x)dx$ basically counts eigenvalues with $\lambda_i \leq 1/t$.

link|flag
For this proof a good reference (which is somewhat more general) is Chapter 2 of Heat Kernels and Dirac Operators by Berline, Getzler and Verne. See thm 2.41, which is a consequence of the asympotics of the heat kernel, thm. 2.30. – Simon Pepin Lehalleur Aug 9 2010 at 23:01

Your Answer

Get an OpenID
or

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