Let $f\in\mathscr{S}(\mathbb{R}^3)$ (the Schwartz space), $k>0$ and consider the integral $$\varphi(k)=\int_{\mathbb{R}^3}\frac{e^{ik|x|}}{|x|}f(x)dx$$Is the estimation $|\varphi(k)|\leq\frac{M}{k^2}$ right?

I can write it as $$\int_{\mathbb{R}^3}\frac{e^{ik|x|}}{|x|}f(x)dx=\lim_{\varepsilon\to 0}\frac{1}{k^2}\int_{\mathbb{R}^3\setminus B_\varepsilon(0)}k^2\frac{e^{ik|x|}}{|x|}f(x)dx$$ (the limit equals the given integral because $f$ is in the Schwartz space using the dominated convergence) Now I can use the fact that if $x\neq 0$ $$-\Delta\bigg(\frac{e^{ik|x|}}{|x|}\bigg)=k^2\frac{e^{ik|x|}}{|x|}$$ so $$\lim_{\varepsilon\to 0}\frac{1}{k^2}\int_{\mathbb{R}^3\setminus B_\varepsilon(0)}k^2\frac{e^{ik|x|}}{|x|}f(x)dx=\lim_{\varepsilon\to 0}\int_{\mathbb{R}^3\setminus B_\varepsilon(0)}\bigg(-\Delta\frac{e^{ik|x|}}{|x|}\bigg)f(x)dx$$ At this point I can use the integration by parts to have $$=\frac{1}{k^2}\lim_{\varepsilon\to 0}\bigg\{-\int_{\mathbb{R}^3\setminus B_0(\varepsilon)}dx\frac{e^{ik\vert x\vert}}{\vert x\vert}(\Delta f)(x)-\int_{\partial B_0(\varepsilon)}d\sigma f\nabla\bigg(\frac{e^{ik\vert x\vert}}{\vert x\vert}\bigg)\cdot\nu+\int_{\partial B_0(\varepsilon)}d\sigma\frac{e^{ik\vert x\vert}}{\vert x\vert}\nabla f\cdot\nu\bigg\}$$ Using the regularity of $f$ (and of all its derivatives) the integral in the last equation are bounded, so I obtain that $$|\varphi(k)|\leq\frac{M}{k^2}$$.

  • $\begingroup$ I'm just curoius, but is this true that $$ -\Delta \left( \frac{e^{ik|x|}}{|x|} \right) = k^2 \frac{e^{ik|x|}}{|x|} $$ holds for $x \neq 0$? $\endgroup$ – Sangchul Lee Apr 26 '13 at 9:43
  • $\begingroup$ If my calculation are correct, I think yes. $\endgroup$ – Mat Apr 26 '13 at 9:48
  • $\begingroup$ Oh, I forgot that we are working on $n = 3$. Indeed, my calculation also yields $$ -\Delta \left( \frac{e^{ikr}}{r} \right) = \left( \frac{n-3}{r^3} - \frac{ik(n-3)}{r^2} + \frac{k^2}{r} \right)e^{ikr}. $$ $\endgroup$ – Sangchul Lee Apr 26 '13 at 9:56
  • $\begingroup$ Yes your calculations are true. $\endgroup$ – Mat Apr 26 '13 at 10:08

Your argument is correct. But here is a different approach which gives more precise information. First, observe that for any orthogonal matrix $M\in O(3)$ the function $ f(Mx)$ is also in the Schwartz class. Consider $ g(x)=\int_{O(3)}f(Mx)\,d\mu(M)$ where $\mu$ is the Haar measure on $O(3)$. The function $g$ is still in the Schwarz class, and is constant on every sphere with center at the origin. All this is a fancy way of saying that the average of $f$ over the sphere $|x|=r$ is a smooth function of $\mathbb r$ which extends (by even reflection) to a Schwartz function on $\mathbb R$. Let $F(r)$ be this average, and observe that $$\varphi(k)=4\pi\int_0^\infty rF(r)e^{ikr}\,dr \tag0$$ The problem has become one-dimensional.

Naturally, we want to extend the integral in (0) to all of $\mathbb R$. Recall that $F$ has a smooth even extension, $F(r)=F(-r)$. Thus, $$\operatorname{Re}\varphi(k)=4\pi\int_0^\infty rF(r)\cos kr \,dr = 2\pi \int_{-\infty}^\infty |r| F(r)\cos kr \,dr = 2\pi \widehat{(|r|F)}(k)\tag1$$ and $$\operatorname{Im}\varphi(k)=4\pi\int_0^\infty rF(r)\sin kr \,dr = 2\pi \int_{-\infty}^\infty r F(r)\sin kr \,dr = -2\pi i \widehat{( r F)}(k)\tag2$$ Since $rF $ is a Schwartz function, so is $\operatorname{Im}\varphi$, according to (2). Unfortunately, $\operatorname{Re}\varphi$ is not as nice since the absolute value in (1) spoils the regularity. Well, at least $|r|F$ is a continuous function with globally integrable, piecewise continuous derivative. It is a shown here that the Fourier transform of such a function is $O(1/k^2)$. In fact, a remark in the linked answer implies that $k^2 \varphi(k)\to 4\pi f(0)$ as $k\to\infty$, if I managed to keep the constants right.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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