Pseudo-localization of singular integrals and noncommutative Calderón-Zygmund theory (Q1004371)

The present paper treats a non-commutative scalar-valued Calderón-Zygmund theory. Inspired by an open problem about the weak type \((1,1)\) boundedness of singular integrals acting on matrix-valued functions, the author proposes two main tools for its solution. The first one is a pseudo-localization principle, and the second one is the construction of a non-commutative form of Calderón-Zygmund decomposition by means of the recent theory of non-commutative martingales, cf. \textit{M.\,Junge} [J. Reine Angew.\ Math.\ 549, 149--190 (2002; Zbl 1004.46043)]. As application, the author obtains the sharp asymptotic behavior of the constants for the strong \(L_p\) inequalities. The following is the first main result on the pseudo-localization principle. Let \(s\) be a positive integer. For a given function \(f\) in \(L_2({\mathbb R}^n; dx)\) and any integer \(k\), the level set is defined by \[ E_\lambda= \{x\in\mathbb R^n\mid M_d f(x)> \lambda\} \] with the dyadic Hardy-Littlewood maximal function \(M_d f\), and write \(E_\lambda=\bigcup_j Q_j\) as a disjoint union of maximal dyadic cubes. Let \({\mathcal R}_k\) be the class of sets in \({\mathbb R}^n\) being the union of a family of cubes in \(Q_k\). When \(f_k\) (resp., \(d f_k\)) denotes, respectively, the \(k\)-th conditional expectation of \(f\) with respect to the standard dyadic filtration (resp., its corresponding \(k\)-th martingale difference), we define \(\Omega_k\) to be the smallest \({\mathcal R}_k\)-set containing the support of \(d f_{k+s}\), and set \(\Sigma_{f,s}=\bigcup_{k \in {\mathbb Z} }\Omega_k\). Then the author shows the localization estimate \[ \left(\int_{{\mathbb R}^n\setminus\Sigma_{f,s}} |T f(x)|^2\,dx\right)^{1/2} \leq c_{n,\gamma} s2^{-\gamma s/4} \left(\int_{{\mathbb R}^n} |f(x)|^2\,dx\right)^{1/2} \tag{1} \] for any \(L_2\)-normalized Calderón-Zygmund operator \(T\) with Lipschitz parameter \(\gamma\). One of the other main results about the weak type and strong \(L_p\) inequalities is given as follows. Let \(T\) be a generalized Calderón-Zygmund operator associated to an operator-valued kernel \(k: {\mathbb R}^{2n} \setminus \varDelta\to{\mathcal M}\) (= semifinite von Neumann algebra with a normal semifinite faithful trace \(\tau\)) satisfying size and smoothness conditions. If \(T\) is an \({\mathcal M}\)-bimodule map bounded on \(L_q({\mathcal A})\) for some \(1 < q < \infty\), then there is some constant \(c_{n,\gamma}\) depending only on the dimension \(n\) and the Lipschitz smoothness parameter \(\gamma\) such that \[ \sup_{\lambda>0} \lambda\varphi\{|Tf|> \lambda\}\leq c_{n,\gamma} \|f\|_1 \tag{2} \] holds for a von Neumann algebra \({\mathcal A}\) obtained as weak-operator closure of \({\mathcal A}_B\) with the normal semifinite faithful trace \(\varphi(f)=\int_{ {\mathbb R}^n } \tau( f(x)) dx\), where \[ {\mathcal A}_B= \Big\{f : \mathbb R^n\to{\mathcal M}\mid f\text{ strongly measurable such that } \underset{x \in{\mathbb R}^n} {\text{ess\,sup}} \|f(x)\|_{\mathcal M}< \infty\Big\}. \tag{3} \] Moreover, in particular, given \(1<p<\infty\) and \(f\in L_p({\mathcal A})\), \[ \|Tf\|_p\leq c_{n,\gamma} \frac{ p^2}{p-1} \|f\|_p \tag{4} \] holds.