The Bellman functions of dyadic-like maximal operators and related inequalities (Q1771323): Difference between revisions

The author provides explicit computations of Bellman functions associated with maximal operators and with Carleson embedding theorems. The setting is a nonatomic probability space \((X,\mu)\). In this context one calls a {\textit{tree}} a family \(\mathcal{T}\) of \(\mu\)-measurable subsets of \(X\) such that (i) \(\mu(I)>0\) for each \(I\in\mathcal{T}\), (ii) each \(I\in\mathcal{T}\) is a finite or countable union of at least two pairwise disjoint (a.e. \(\mu\)) elements \(C(I)\) of \(\mathcal{T}\), (iii) \(\mathcal{T}=\cup_{m\geq 0} \mathcal{T}_m\) with \(\mathcal{T}_0=\{X\}\) and \(\mathcal{T}_{m+1}=\cup_{I\in \mathcal{T}_m} C(I)\) and (iv) \(\lim_{m\to\infty}\sup_{I\in \mathcal{T}_m}\mu(I)=0\). This tree structure generalizes the dyadic cubes in \(\mathbb{R}^n\). One defines the following analog of the dyadic maximal operator: \[ M_{\mathcal{T}} \phi(x) =\sup_{x\in I\in \mathcal{T}} \frac{1}{\mu(I)}\int_{I} | \phi| \, d\mu. \] One then associates to \(\mathcal{T}\) the Bellman function \[ \mathcal{B}^{\mathcal{T}} (F,f,L)=\sup\biggl\{ \int_X (\max (M_{\mathcal{T}}\phi,L)^p\,d\mu:\, \phi\geq 0, \int_X \phi^p\, d\mu =F, \int_X \phi\, d\mu =f \biggr\} \] For \(p>1\) and \(z\in [1,p/(p-1)]\), one sets \(H_p(z)=(1-p)z^p+pz^{p-1}\) and denotes by \(\omega_p\) the inverse function of \(H_p\). It is shown that (for fixed \(p>1\)), if \(L<\frac{p}{p-1}f \) then \[ \mathcal{B}^{\mathcal{T}} (F,f,L) = F\omega_p \Bigl(\frac{p L^{p-1} f-(p-1)L^p}{F}\Bigr)^p \] while, if \( L\geq \frac{p}{p-1} f\) then \[ \mathcal{B}^{\mathcal{T}} (F,f,L) = L^p+\Bigl(\frac{p}{p-1}\Bigr)^p(F-f^p). \] This enables a sharp bound for the maximal operator \(M_{\mathcal{T}}\) on \(L^2\). A corresponding result is proved for the Bellman function associated with Carleson embeddings. This function is defined by replacing the supremum over \(\phi\) of \(\int_X (\max (M_{\mathcal{T}}\phi,L)^p\,d\mu\) in the definition of \(\mathcal{B}^{\mathcal{T}} (F,f,L)\) by a supremum over \(\phi\) of \(\int_K (\max (M_{\mathcal{T}}\phi)^p\,d\mu\) with \(\mu(K)=k\) fixed, to define a function \(\widetilde{\mathcal{B}}^{\mathcal{T}} (F,f,k)\).