3
$\begingroup$

Consider a function $u\in L^2(\mathbb R^N)$, and another function $\varphi$ which is the unique solution to the Poisson equation $\Delta \varphi = u$ vanishing at $\infty.$ We know that the radial part of $u$, defined by $$ \tilde u(r):=\frac{1}{S_{N-1}}\int_{\partial B(0,1)} u(ry)d\sigma_y $$ is radially symmetric and we have the inequality $\| \nabla\tilde u\|_{L^2} \leq \|\nabla u\|_{L^2}$, if $\nabla u$ is defined in the weak sense. Intuitively we have reduced the fluctuation by taking an average. In terms of Fourier transform, we have $\||\xi|\hat{\tilde u}\|_{L^2} \leq \||\xi|\hat{u}\|_{L^2}$.

Now, my question is, do we have a similar inequality for comparing other powers of $\xi,$ that is, and inequality for $s>0$ of the form $$ \||\xi|^s\hat{\tilde u}\|_{L^2} \leq \||\xi|^s\hat{u}\|_{L^2} \space? \label{1} $$

And also, let $\Delta \tilde\varphi=u$ be solved by a function that vanish at $\infty$. For $s=-1,$ the above inequality becomes $\||\xi|^{-1}\hat{\tilde u}\|_{L^2} \geq \||\xi|^{-1}\hat{u}\|_{L^2}$ (I have reverse the direction of the inequality, because this makes more sense for an "antiderivative"). This corresponds to the inequality $$ \|\nabla \tilde\varphi\|_{L^2} \leq\|\nabla \varphi\|_{L^2}. $$

Is this inequality true?

$\endgroup$

1 Answer 1

1
$\begingroup$

Yes, it is true. Take a function $u \in L^2(R^N)$ and expand for $x=r\omega$ $$ u(r\omega)=\sum_{k=0}^\infty u_k(r)P_k(\omega)$$ where $(P_k)$ an orthonormal basis of spherical harmonics in $L^2(S^{N-1})$ and $u_0$ is your $\tilde u$. Then for $\xi=s\eta$ $$\hat u(s\eta) =\sum_{k=0}^\infty U_k(s)P_k(\eta),$$ see Stein-Weiss "Fourier Analysis....IV.4" and $U_0=\hat{\tilde u}$. If you multiply by a radial weight $g(s)$, then $$\|g \hat u\|_2^2=\sum_k \int_0^\infty |g(s)U_k(s)|^2 s^{N-1}ds \geq \int_0^\infty |g(s)U_0(s)|^2 s^{N-1}ds=\|g\hat{\tilde u}\|^2_2. $$

$\endgroup$
1
  • $\begingroup$ A simplification of the above proof is the following. Write $u=\tilde u+v$ and check that $\hat{\tilde u}$ is radial and $\hat v(s\eta)$ has mean zero on $S^{N-1}$ for every $s$, since this is true for $v$ (you need that $$\int_{S^{N-1}} e^{irs \omega \cdot \eta} d\eta$$ is radial). Then $g\hat {\tilde u}$ and $g\hat v$ are orthogonal when $g$ is radial. $\endgroup$ Jan 3, 2021 at 8:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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