Relation between Stein-Tomas adjoint restriction estimate and the Helmholtz equation

111 Views Asked by At

Let $d\sigma$ denote the surface measure on $\mathbb{S}^2$. For each function $f\in L^2(\mathbb S^2)$, the Fourier transform $\widehat{fd\sigma}$ is defined as the integral $$ \int_{\mathbb S^2} f(\xi)e^{ix\cdot \xi}\, d\sigma(\xi), \qquad x\in \mathbb R^3,$$ and as Stein and Tomas proved, it satisfies the inequality $$\tag{1} \lVert \widehat{f d\sigma}\rVert_{L^4(\mathbb R^3)}\le C\lVert f\rVert_{L^2(\mathbb S^2)}.$$

Question. The function $u=\widehat{fd\sigma}$ satisfies the Helmholtz equation $\Delta u + u =0$ in $\mathbb R^3$. Is there a corresponding PDE interpretation of the estimate (1)?

A nice PDE interpretation is available for the Stein-Tomas estimate on the paraboloid $$\mathbb P^2=\{(\tau, \xi)\in \mathbb R\times \mathbb R^2\ :\ \tau=\lvert \xi\rvert^2\}.$$ Indeed, letting $$ d\mu:=\frac{\delta(\tau-\lvert \xi\rvert^2)}{(2\pi)^2}\, d\tau d\xi$$ we define a measure supported on $\mathbb P^2$, and the estimate analogous to (1) reads $$\tag{2} \lVert \widehat{f d\mu}\rVert_{L^4(\mathbb R^3)}\le C\lVert f\rVert_{L^2(\mathbb R^2)}.$$ If we denote $u(t, x)=\widehat{fd\mu}$, we see that it satisfies the initial value problem for the Schrödinger equation $$ \begin{cases}  i\partial_t u = \Delta u, & t\in\mathbb R, x\in \mathbb R^2, \\ u|_{t=0}=\check{f}, \end{cases} $$ where $\check{f}$ denotes the inverse Fourier transform of $f$. By Plancherel's theorem, $\lVert f\rVert_{L^2}=\lVert \check{f}\rVert_{L^2}$, up to an irrelevant constant. Thus, (2) reads $$ \lVert u\rVert_{L^4(\mathbb{R}\times \mathbb{R}^2)}\le C\lVert f\rVert_{L^2(\mathbb R^2)}, $$ which is the celebrated Strichartz estimate.

I wonder if, similarly, the inequality (1) can be written as an estimate of the solution to the Helmholtz equation in terms of some kind of boundary values.

1

There are 1 best solutions below

6
On BEST ANSWER

The answer is that, if $u$ solves $\Delta u + u=0$ on $\mathbb{R}^3$, writing $$ (x_1, x_2, x_3)=(\bar x, t), \qquad \bar x\in \mathbb R^2, t\in \mathbb R, $$ and $$ \bar \Delta = \partial_{x_1}^2+ \partial_{x_2}^2, $$ then the adjoint restriction inequality (1) above is equivalent to $$\tag{3} \lVert u\rVert_{L^4(\mathbb R^3)}^2\le 2C^2 \left(\int_{\lvert \bar \xi\rvert\le 1}\lvert \mathcal F(u|_{t=0})(\bar \xi)\rvert^2 (1-\lvert\bar\xi\rvert^2)^{1/2}\,d\bar\xi + \int_{\lvert\bar\xi\rvert\le 1} \lvert \mathcal F(u_t|_{t=0})(\bar \xi)\rvert^2 (1-\lvert\bar\xi\rvert^2)^{-1/2}\,d\bar\xi\right). $$ Unfortunately, this is not the prettiest thing around, but that's what we get. Here, and in the following, $\mathcal F$ refers to the Fourier transform in the variables $\bar x\to\bar \xi$, that is
$$\tag{4} \mathcal F(u|_{t=0})(\bar\xi)=\int_{\mathbb R^2} u(\bar x, 0)e^{-i\bar x\cdot \bar \xi}\, d\bar x,\quad \mathcal F(u_t|_{t=0})(\bar\xi)=\int_{\mathbb R^2} u_t(\bar x, 0)e^{-i\bar x\cdot \bar \xi}\, d\bar x. $$ Proof: each $f\in L^2(\mathbb S^2)$ can be written as $$f=f_+ + f_-,$$ where $f_+$ is supported in the upper hemisphere ($t>0$), while $f_-$ is supported in the lower hemisphere ($t<0$). We can regard $f_\pm$ as functions defined on the unit disk $\lvert \bar \xi \rvert\le 1$. Then

$$ u(\bar x, t)= \widehat{f d\sigma}= \int_{|\bar \xi|\le 1} e^{i \bar x \cdot \bar \xi + t\sqrt{1-\lvert \bar \xi \rvert^2}} f_+\, d\sigma + \int_{|\bar \xi|\le 1} e^{i \bar x \cdot \bar \xi - t\sqrt{1-\lvert \bar \xi \rvert^2}} f_-\, d\sigma. $$ Recall that the surface element on $\mathbb S^2$ reads $d\sigma=\pm(1-\lvert\bar\xi\rvert^2)^{-\frac12}d\bar \xi$. Thus, the above formula can be rewritten as $$\tag{5} u(\bar x, t)= e^{it\sqrt{1+\bar{\Delta}}}g_+ + e^{-it\sqrt{1+\bar{\Delta}}}g_-, $$ where $$\tag{6} \mathcal F g_{\pm}(\bar \xi) = f_{\pm}(\bar \xi)(1-\lvert\bar \xi \rvert^2)^{-\frac12} \mathbf 1_{\lvert\bar \xi\rvert\le 1}.$$ In particular, $$ \tag{7} u(\bar x, 0)=g_++g_-,\quad u_t(\bar x, 0)=i\sqrt(1+\bar{\Delta})(g_+-g_-).$$ To conclude, we observe that the Tomás-Stein estimate reads, in our notation, $$ \lVert u\rVert_{L^4(\mathbb R^3)}^2\le C^2\int_{\lvert\bar\xi\rvert\le 1} \lvert f_+(\bar \xi)\rvert^2 +\lvert f_-(\bar \xi)\rvert^2 \frac{d\bar \xi}{(1-\lvert\bar \xi\rvert^2)^{\frac12}},$$ and plugging (6) and (7) into it, we obtain (4), as we wanted. $\Box$