The Walsh model for \(M_2^*\) Carleson (Q1001588)

Let \(C\) be the Carleson-Hunt operator \[ Cf(x, \theta)= p.v. \int_{R} \frac{f(x-y)}{y}e^{i \theta y} dy, \] which controls the convergence of the Fourier series. The authors [Duke Math. J. 143, No. 2, 281--355 (2008; Zbl 1213.42064)] pointed out striking connections between this operator and the return times theorem by \textit{J. Bourgain} [C. R. Acad. Sci., Paris, Ser. I 306, No. 12, 483--485 (1988; Zbl 0655.58022)]. Let \((X, \sum, \mu, \tau)\) be a dynamical system and let \(1 \leq p,q, \leq \infty\) satisfy \(1/p+1/q \leq 1\). For each function \(f \in L^p(X)\) there is a universal set \(X_0 \subset X\) with \(\mu (X_0) =1\), such that for each second dynamical system \((Y, {\mathcal F}, \nu, \sigma)\), each \(g \in L^q(Y)\) and each \(x \in X_0\), the averages \[ \frac{1}{N}\sum_{n=0}^{N-1} f(\tau^n x)g( \sigma^n y) \] converge \(\nu\)-almost everywhere. In the paper under review the authors consider the Walsh analogue of the Carleson-Hunt operator, and develop Walsh time-frequency analysis. Let \(W_{l}\) be \(l\)-the Walsh function. A tile \(P\) is a rectangle \(I_P \times \omega_P\) area one, such that \(I_P\) and \(\omega_P\) are dyadic intervals. The Walsh wave packet \(w_P\) is defined by \[ w_P (x) = 2^{-i/2}W_{l}(2^{-i}x-n) \quad \text{where} \quad P=[2^i n,2^i(n+1)) \times [2^{-i}l, 2^{-i}(l+1)). \] A bitile \(P\) is a rectangle \(I_P \times \omega_P\) area two, such that \(I_P\) and \(\omega_P\) are dyadic intervals. For any bitile \(P=[2^i n,2^i(n+1)) \times [2^{-i+1}l, 2^{-i+1}(l+1))\), let \(P_1=[2^i n,2^i(n+1)) \times [2^{-i+1}l, 2^{-i}(2l+1))\) and \(P_2=[2^i n,2^i(n+1)) \times [2^{-i}(2l+1), 2^{-i+1}(l+1))\). They concern with the operator \[ Wf(x)= \left\| \left( \sum_{P: \text{bitile}, | I_P | < 2^k} \langle f, w_{P_1} \rangle w_{P_1}(x)1_{\omega_{P_2}}(\theta) \right)_{k \in \mathbb{Z}} \right\|_{M^{*}_2(\theta)}, \] where the \(M_2^{*}\)-norm of a family of Walsh multipliers \(m_k\) is defined as \[ \| (m_k (\theta))_{k \in Z}\|_{M_2^{*}(\theta)} = \sup_{\| g \|_{2} =1} \| \sup_{k} | (\hat{g}m_k)\check{\;} (x) | \|_{L^2_x}. \] They prove the following. \[ \| Wf \|_{L^p} \leq C_p \| f \|_{L^p} \quad \text{where} \quad 1<p< \infty. \] Their argument relies on first splitting the collection of bitiles into structured collections called ``trees''. The operator \(W\) restricted to each such a tree is estimated by using Calderón-Zygmund type estimates. \(Wf(x)\) is estimated pointwise, and it is shown that for each \(x\) the contribution to \(Wf(x)\) comes from one stack of trees. To estimate this contribution they consider a weighted version of a maximal multiplier result due to Bourgain.