6
$\begingroup$

In the paper of Bourgain, the way equation (3.78) is deduced from (3.69) and (3.76) seems via the following interpolation result. Let $(X,\mu)$ and $(Y,\nu)$ be two measure spaces and let $T$ be a linear operator from $L^1(X,\mu)+L^2(X,\mu)$ to $L^\infty(Y,\nu)+L^2(Y,\nu)$ such that

\begin{align} \|Tf\|_{L^\infty}&\leq A\|f\|_{L^1}, \\ \|Tf\|_{L^2}&\leq B\|f\|_{L^2}+D\|f\|_{L^1}. \end{align}

For $0<\theta<1$, let $p=\frac{1-\theta}{1}+\frac{\theta}{2}$, $\frac{1}{p}+\frac{1}{q}=1$. Then for some constant $C$ independent of $A,B,D$, we have $$\|Tf\|_{L^{q}}\leq C(A^{1-\theta}B^\theta\|f\|_{L^p}+A^{1-\theta}D^\theta\|f\|_{L^1}).$$

By scaling the measures $\mu$ and $\nu$ as well as the linear operator $T$, one can reduce it to the case when \begin{align} A=B=D=1. \end{align}

I tried to prove it by looking for the complex interpolation space between $L^1$ and $L^2\cap L^1$, as well as to directly prove it as to prove the Marcinkiewicz's interpolation inequality. But I could not find the correct proof. Any help is appreciated. Thank you!

$\endgroup$
8
  • $\begingroup$ So generous with the bounty! $\endgroup$
    – Fan Zheng
    May 13, 2018 at 4:42
  • 1
    $\begingroup$ Erm... The conditions make way more sense if $T:L^1\cap L^2\to L^\infty\cap L^2$. Are you sure that $+$ is not a misprint? $\endgroup$
    – fedja
    May 13, 2018 at 18:17
  • $\begingroup$ @fedja No, it should be +. Otherwise the condition $\|Tf\|_{L^\infty}\leq A\|f\|_{L^2}$ makes no sense. $\endgroup$ May 13, 2018 at 21:07
  • 1
    $\begingroup$ Theorem 3 (and, generally, Section 3) in this article discusses interpolation between $L^q \cap L^2$ for $q \geqslant 2$. Theorem 4 might provide the right tool to prove the desired bound for $\|T f\|_q$. $\endgroup$ May 13, 2018 at 22:36
  • $\begingroup$ @MateuszKwaśnicki I think it actually works! Except that Theorem 4 there is stated only for domains in a Euclidean space, but here I am looking for at least as general as smooth compact manifolds. But I believe it can be generalized into such cases by a standard procedure. $\endgroup$ May 14, 2018 at 0:14

2 Answers 2

3
+100
$\begingroup$

As requested, I post my comment as an answer (although this is not a true answer, just a possibly useful reference; feel free to edit it if this approach works out).

In Section 3 of the article Interpolation of sum and intersection spaces of $L^q$-type and applications to the Stokes problem in general unbounded domains, P.F. Riechwald studies interpolation between spaces $L^2 \cap L^q$ for $q \geqslant 2$ (and also $L^2 + L^q$ for $q \leqslant 2$). His main tool is Theorem 4, which I reproduce below.

Theorem. Let $\Omega \subseteq \mathbb{R}^n$ be a domain and let $f \in L^1(\Omega) + L^\infty(\Omega)$ be a given and fixed function. Then there exist linear maps $$ S_1 : L^1(\Omega) + L^\infty(\Omega) \to L^1((0, 1)) , \qquad S_2 : L^1(\Omega) + L^\infty(\Omega) \to \ell^\infty $$ and $$ T_1 : L^1((0, 1)) \to L^1(\Omega) + L^\infty(\Omega) , \qquad T_2 : \ell^\infty \to L^1(\Omega) + L^\infty(\Omega) $$ satisfying the equality $$ f = T_1 S_1 f + T_2 S_2 f $$ almost everywhere. Moreover, these maps satisfy the estimates $$ \|S_1 u\|_{L^p((0, 1))} \leqslant \|u\|_{L^p(\Omega)} , \qquad \|S_2 u\|_{\ell^p} \leqslant \|u\|_{L^p(\Omega)} $$ and $$ \|T_1 u\|_{L^p(\Omega)} \leqslant \|u\|_{L^p((0, 1))} , \qquad \|T_2 u\|_{L^p(\Omega)} \leqslant \|u\|_{\ell^p} $$ for all $1 \leqslant p \leqslant \infty$ and all $u$ in the respective $L^p$-spaces.

I suppose the argument used in the proof of Theorem 3 implies that the complex interpolation space between two spaces $L^1(\Omega) \cap L^p(\Omega)$ is again a space of this form. This, in turn, should imply the desired bound on $\|T f\|_p$.

$\endgroup$
1
  • 1
    $\begingroup$ Using this method, it can be shown that for all $1\leq p\leq p_0\leq p_1\leq\infty$ or $1\leq p_0\leq p_1\leq p\leq\infty$, for any underlying $\sigma$-finite measure space, the complex interpolation space of parameter $\theta$ between $L^{p_0}\cap L^{p}$ and $L^{p_1}\cap L^p$ is $L^{p_\theta}\cap L^p$, and the space between $L^{p_0}+ L^{p}$ and $L^{p_1}+ L^p$ is $L^{p_\theta}+L^p$. $\endgroup$ May 14, 2018 at 20:24
3
$\begingroup$

In the case of a set of finite measure (Bourgain in the quoted paper deals with the case of finite measure (torus)) we have that $\Vert f\Vert_1\leq C\Vert f\Vert_2$ so we actually have $(\infty,1)$ and $(2,2)$ estimate and therefore we have $\Vert Tf\Vert_{p'}\leq C\Vert f\Vert_p$, $1<p<2$, by Marcinkiewicz or Riesz-Thorin.

Edit: This is an answer to an earlier version of the question which was not so detailed and only after a discussion with the author of the question details regarding the dependence of the constant were added so you should take it into account before you decide to downvote my answer.

$\endgroup$
0

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.