6
$\begingroup$

I am looking for a proof the following theorem.

Let $U \subset \mathbb{R}^n$ be a bounded domain with $C^2$ boundary and $p(x,y,t)$ be the Neumann heat kernel. Then there exist a constant $C>0$ such that \begin{equation} p(x,y,t)\geq \frac{1}{Ct^{n/2}} e^{-\frac{Cd(x,y)^2}{t}} \end{equation} where $d(x,y)$ is the geodesic distance in $U$ between $x$ and $y$.

I could not find any references that gives the theorem in the form stated above. Other similar results I've found until now and the reasons why the do not satisfy my needs are listed below.

  • In [1] the result is given for a complete Riemannian manifold.
  • In [2] the estimate near the boundary is worser.
  • In [3] the result is given for convex domains.
  • In [4] the result is given for inner uniform domains.
  • In [5] the result is given for the Dirichlet heat kernel on inner uniform domains.

Furthermore, in [5] the inner uniform domains considered are assumed to have a smooth boundary, but this has not been proven. So if we know the exact proof of this, We could use the result of [4] by changing from smooth to $C^2$. Therefore, I would like to know not only where I can find a proof of the above inequality but also matters related to its application on an inner uniform domain.

Thanks.

References:

[1] Sharp explicit lower bounds of heat kernels, by Feng-Yu Wang.
[2] Heat kernel estimates for general boundary problems, by Liangpan Li and Alexander Strohmaier.
[3] On domain monotonicity of the Neumann heat kernel, by Richard F. Bass and Krzysztof Burdzy.
[4] The heat kernel and its estimates, by Laurent Saloff-Coste.
[5] The Dirichlet heat kernel in inner uniform domains: Local results, compact domains and non-symmetric forms, by Janna Lierl and Laurent Saloff-Coste.

$\endgroup$
4
  • 1
    $\begingroup$ Saloff-Coste says in [4] that a Lipschitz domain is inner uniform, without a proof. I never tried, but I have the feeling that for $C^1$ domain the proof of the lower bound should be similar to that in the whole space. Having the upper bound and conservation of probability one obtains the lower bound $t^{-d/2}$ on the diagonal. Then one needs the same bound for close points $|x-y| \leq c\sqrt t$ and this could follow from gradient estimates of the kernel (and some regularity of the domain). Then the chain argument gives the global lower bound. Sorry for being vague. $\endgroup$ Commented Aug 26, 2023 at 17:26
  • 1
    $\begingroup$ It's straightforward to show that a (connected, bounded) Lipschitz domain is uniform by hand. Locally, assuming the boundary is a graph nearby (say $x_n = g(x')$, you can connect any two points $x, y$ by going away from the boundary from $x$ while keeping $x'$ constant until you are a distance $~|x - y|$, then sideways moving from $x'$ to $y'$, and then back towards the boundary and $y$. This kind of curve satisfies the "Harnack chain" condition. If you have two points which are not near enough to each other that you are a graph, you instead move inwards a finite distance and use connectivity. $\endgroup$
    – user378654
    Commented Aug 27, 2023 at 2:39
  • 1
    $\begingroup$ In the articles you link they are careful about the definition because they are potentially interested in unbounded domains, slit domains, etc. If you are looking for references about $C^2$ implies uniform/inner uniform/whatever for bounded domains, I would suggest searching the term "NTA domain," which is a slightly stronger notion more commonly used by less sophisticated PDE people such as myself. $\endgroup$
    – user378654
    Commented Aug 27, 2023 at 2:45
  • $\begingroup$ Thank you very much. I will check out paper of NTA domain first. If there are no good results, I will try to prove it myself. $\endgroup$
    – mark
    Commented Aug 28, 2023 at 12:31

0

You must log in to answer this question.