1
$\begingroup$

Consider the Laplace operator $\Delta:=\sum_{j=1}^{n}\partial_{x_{j}}^{2}$ on $\mathbb{R}^n$. Let $E_{\lambda}$ be the spectral resolution of $\Delta$, and $$ H_{t}[f]:=\cos{(t\sqrt{-\Delta})}f=\int_{0}^{\infty}\cos{t\sqrt{\lambda}}dE_{\lambda}f\qquad \forall f\in L^2(\mathbb{R}^n)$$ be the Huygens operator.

For any $u_{0}\in C_{0}^{\infty}(\mathbb{R}^n)$, the function $$ u(x,t)=(H_{t}[u_{0}])(x)=\int_{0}^{\infty}\cos(t\sqrt{\lambda})dE_{\lambda}u_{0}$$ solves \begin{equation}\label{1-7} \left\{ \begin{array}{ll} \frac{\partial^{2}u}{\partial t^{2}}-\Delta u=0 , & \hbox{on $\mathbb{R}^n \times (0,+\infty)$;} \\ u_{t}(x,0)=0, & \hbox{$\forall x\in \mathbb{R}^n$;} \\ u(x,0)=u_{0}, & \hbox{$\forall x\in \mathbb{R}^n$.} \end{array} \right. \end{equation} The Huygens operator $H_t$ admits a kernel given by \begin{equation} w(x,y,t)=\int_{0}^{\infty}\cos(t\sqrt{\lambda})de(x,y,\lambda) \end{equation} Here, $e(x,y,\lambda)$ is the spectral function on $\mathbb{R}^n$. It follows that, for any fixed $y\in \mathbb{R}^n$, $w(x,y,t)$ satisfies the equation

\begin{equation} \left\{ \begin{array}{ll} (\frac{\partial^{2}}{\partial t^{2}}-\Delta_{x})w(x,y,t) =0 , & \hbox{on $\mathbb{R}^n \times (0,+\infty)$;} \\ w_{t}(x,y,0)=0, & \hbox{$\forall x\in \mathbb{R}^n$;} \\ w(x,y,0)=\delta_{y}(x), & \hbox{$\forall x\in \mathbb{R}^n$.} \end{array} \right. \end{equation}

Let $\Omega$ be an open bounded domain in $\mathbb{R}^n$. Denote by $\Delta_{\Omega}$ be the Dirichlet Laplacian operator on $\Omega$, and $E_{\lambda,\Omega}$ the corresponding spectral resolution. We can also defined the Huygens operator $$ H_{t}^{\Omega}[f]:=\cos{(t\sqrt{-\Delta_{\Omega}})}f=\int_{0}^{\infty}\cos{t\sqrt{\lambda}}dE_{\lambda,\Omega}f\qquad \forall f\in L^2(\Omega)$$ We denote by $w_{\Omega}(x,y,t)$ the kernel of $H_{t}^{\Omega}$: $$ w_{\Omega}(x,y,t)=\int_{0}^{\infty}\cos(t\sqrt{\lambda})de_{\Omega}(x,y,\lambda), $$ where $e_{\Omega}(x,y,\lambda)$ is the spectral function of $-\Delta_{\Omega}$. Similarly, for fixed $y\in \Omega$, $w_{\Omega}(x,y,t)$ satisfies the equations: \begin{equation} \left\{ \begin{array}{ll} (\frac{\partial^{2}}{\partial t^{2}}-\Delta_{x})w_{\Omega}(x,y,t) =0 , & \hbox{on $\Omega \times (0,+\infty)$;} \\ w_{\Omega}(x,y,t)=0, & \hbox{$\forall x\in \partial\Omega$;} \\ (w_{\Omega})_{t}(x,y,0)=0, & \hbox{$\forall x\in \Omega$;} \\ w_{\Omega}(x,y,0)=\delta_{y}(x), & \hbox{$\forall x\in \Omega$.} \end{array} \right. \end{equation}

In (1.6) of the paper Seeley, R., A sharp asymptotic remainder estimate for the eigenvalues of the Laplacian in a domain of (R^3), Adv. Math. 29, 244-269 (1978). ZBL0382.35043. , it claims the following fact that $$ w(x,y,t)=w_{\Omega}(x,y,t) \tag{*}$$ holds for all $|t|\leq d(y,\partial\Omega)$ and all $x\in \Omega$.

My Question: How can we prove the conclusion $(*)$ by means of the finite speed of propagation property of wave equation?

Remark: I review many other papers, but all of them just write "this is due to the "Finite Speed of Propagation Property". I feel very confuse about the details in applying the finite speed of propagation property. Can someone help me with above question? Thank you very much!

Recall that the Finite Speed of Propagation Property states that:

Let $u(x,t)$ solves $\frac{\partial^2 u}{\partial t^2}-\Delta u=0$. If $u(x,t)=u_{t}(x,t)=0$ on the set $B(x_0,t_0)\times\{t=0\}$, then $u(x,t)=0$ on $K(x_0,t_0):=\{(x,t)|0\leq t\leq t_0, |x-x_0|\leq t_0-t\}$.

$\endgroup$
6
  • 1
    $\begingroup$ Apply the finite speed of propagation property to $u = w - w_\Omega$ (where $w_\Omega$ is extended arbitrarily to the entire Euclidean space), and note that $\Omega$ contains $B(y,t)$ whenever $y \in \Omega$ and $t \leq d(y,\partial \Omega)$. (If one is worried about the discontinuities of the extension of $w_\Omega$, one can convolve $u$ by some mollifier (approximation to the identity) if desired, though with the theory of distributions this is not really necessary.) $\endgroup$
    – Terry Tao
    Commented Aug 15, 2023 at 14:51
  • $\begingroup$ (Strictly speaking, one needs a version of the finite speed of propagation property that also allows for an inhomogeneous term, but this can be deduced from the version you stated via Duhamel's formula.) $\endgroup$
    – Terry Tao
    Commented Aug 15, 2023 at 14:54
  • $\begingroup$ @TerryTao Thank you for your comments! But I am still confuse about the restriction $t\leq d(y,\partial\Omega)$. Here is my understanding: After extended $w_{\Omega}$ to the entire Euclidean space, for any fixed $y\in \Omega$, $u=w-w_{\Omega}$ satisfies $\partial_{t}^{2}u-\Delta u=0$ in $\mathbb{R}^n\times (0,+\infty)$, and for any $x_0\in \Omega$, we can find $B(x_0,r_0)$ ($r_0$ large enough) such that $\Omega\times [0,d(y,\partial\Omega)] \subset K(x_0,r_0)$. Then applying the finite speed of propagation property to $u$ in the set $B(x_0,r_0)\times\{t=0\}$. Am I misunderstanding something? $\endgroup$
    – pxchg1200
    Commented Aug 16, 2023 at 2:01
  • $\begingroup$ @TerryTao However, according to the arguments above, it seems that we can choose $r_0$ as large as possible, such that $\Omega\times [0, m\cdot d(y,\partial\Omega)]\subset K(x_0,r_0)$ as $m\to+\infty$. This seems very strange to me. $\endgroup$
    – pxchg1200
    Commented Aug 16, 2023 at 2:03
  • $\begingroup$ $w - w_\Omega$ only solves the wave equation in the interior and exterior of $\Omega$; on the boundary $\partial \Omega$ there will be a distributional forcing term. $\endgroup$
    – Terry Tao
    Commented Aug 16, 2023 at 4:32

0

You must log in to answer this question.