The vector valued quartile operator (Q379980): Difference between revisions

In [\textit{T. P. Hytönen} and \textit{M. T. Lacey}, Math. Ann. 357, No. 4, 1329--1361 (2013; Zbl 1285.42005)], the authors started the development of time-frequency harmonic analysis for Banach space valued functions by proving a central result: the almost everywhere pointwise convergence of Fourier series of Banach space valued functions. Their result holds for Banach spaces that are complex interpolation spaces between a Hilbert space and a UMD Banach space (note that it is unknown whether or not all UMD spaces belong to this class). More precisely, Hytönen-Lacey's Banach space valued Carleson theorem relies on the UMD Banach space \(X\) having a non-trivial tile type, a property that naturally arises when trying to extend time-frequency arguments to the Banach space valued setting, and that holds for interpolation spaces between a Hilbert space and a UMD space. As is often the case in time-frequency analysis (thanks to the pioneering work of Thiele), the authors first work in the Walsh setting, i.e. starting with the Walsh basis of \(L^2(0,1)\) instead of the trigonometric basis. The idea is that the estimates proven in this setting can usually be transferred to the Fourier setting, but are cleaner as one can use compactly supported functions that have a compactly supported Walsh-Fourier transform (while a function and its classical Fourier transform can only be jointly well localised in a weaker sense). The paper under review establishes an important result, that can be seen as the Walsh analogue of the boundedness of the bilinear Hilbert transform as explained in [\textit{C. Thiele}, Trans. Am. Math. Soc. 352, No. 12, 5745--5766 (2000; Zbl 0976.42017)]: the boundedness of the trilinear form defined by \[ \Lambda(f_{1},f_{2},f_{3}) = \sum \limits _{p \in P}|I_{p}|^{-\frac{1}{2}}\Pi(\langle f_{1},w_{p_1} \rangle, \langle f_{2},w_{p_2} \rangle, \langle f_{3},w_{p_3} \rangle) \] where \(P\) is a set of quartiles (which are products of dyadic rectangles of the form \(\bigcup \limits _{v=0} ^{3} I_{p}\times \frac{1}{|I_{p}|}[4n+v,4n+v+1]\)), \(w_{p_{v}}\) is a wave packet supported in \(I_{p}\times \frac{1}{|I_{p}|}[4n+v,4n+v+1]\) (i.e. an appropriate dilate and translate of an element of the Walsh basis), and \(\Pi: X_{1}\times X_{2} \times X_{3} \to \mathbb{C}\) is a fixed trilinear form on the Banach spaces \(X_{1},X_{2},X_{3}\) (replacing the product used in the scalar case). The result is proven in \(L^{\frac{1}{\beta_{1}}}(\mathbb{R}_{+};X_{1}) \times L^{\frac{1}{\beta_{2}}}(\mathbb{R}_{+};X_{2})\times L^{\frac{1}{\beta_{3}}}(\mathbb{R}_{+};X_{3})\) under some intricate conditions relating \(\beta_{1},\beta_{2},\beta_{3}\) and the quartile types (defined in the paper) of \(X_{1},X_{2},X_{3}\). The proof uses Lacey-Thiele's approach by obtaining a result for special families of quartiles first (the trees), then decomposing arbitrary families into trees in an appropriate manner. This is combined with subtle ideas from UMD-valued harmonic analysis to exploit the orthogonality available in \(L^2(\mathbb{R}_{+};H)\) where \(H\) is a Hilbert space, when working in an \(L^{p}(\mathbb{R}_{+};[X,H]_{\theta})\) space where \(X\) is a UMD Banach space. The paper is thus relevant both for harmonic analysts interested in the intricate relationship between time-frequency analysis and martingale estimates, and for functional analysts interested in the geometric properties that Banach spaces need to possess to ensure the boundedness of the extension of classical integral operators of Fourier analysis to the Banach space valued setting.