A Bellman function proof of the \(L^2\) bump conjecture (Q393443)

The original (still) open problem concerning two weight estimates for singular integrals is to find necessary and sufficient condition on the weights \(u\) and \(v\) such that a Calderón-Zygmund operator \(T: L^p(u) \to L^p(v)\) is bounded, i.e., \[ \int_{\mathbb R^n} | Tf |^p v \, dx \leq C \int_{\mathbb R^n} | f |^p u \, dx. \] We can rewrite the problem as follows. Describe all weights \(u,v\) such that the operator \(M_{v^{1/p}} T M_{u^{1/p'}}\) is bounded on non-weighted \(L^p\), where \(M_a\) is the multiplication operator \(M_a f= af.\) The \(A_p\) condition \[ \sup_{I} \Big( \frac{1}{ | I |} \int_I v \, dx \Big)^{1/p} \Big( \frac{1}{ | I |} \int_I u \, dx \Big)^{1/p'} < \infty \] is necessary for the boundedness of the operator \(M_{v^{1/p}} T M_{u^{1/p'}}\), but simple counterexamples show that this condition is not sufficient. So a natural way to obtain a sufficient condition is to replace the \(L^1\) norm of \(u\) and \(v\) with some stronger Orlicz norms. Namely, given a Young function \(\Phi\) and a cube \(I\), one can consider the Orlicz space \(L^{\Phi}(I)\) with norm given by \[ \| f \|_{L^{\Phi}(I)} := \inf \left\{ \lambda>0 : \;\int_I \Phi \left( \frac{f(x)}{ \lambda} \right) \frac{dx}{ | I |} \leq 1 \right\}. \] It has been conjectured (for \(p=2\)) that if the Young functions \(\Phi_1\) and \(\Phi_2\) satisfy some suitable conditions, then \[ \sup_{I} \| v \|_{\Psi_1 (I)} \| u \|_{\Psi_2 (I)} < \infty \] implies that \(M_{v^{1/2}} T M_{u^{1/2}}\) is bounded on \(L^2\). \textit{A. K. Lerner} [J. Anal. Math. 121, 141--161 (2013; Zbl 1285.42015)] solved this problem. In this paper the authors give another proof by using a Bellman function. Define the normalized distribution function \(N\) by \[ N_I^{w}(t) = \frac{1}{ | I |} \left| \{ x \in I : w(x) >t\} \right|. \] Let \(\Psi : (0,1] \to \mathbb R_{+}\) be a decreasing function such that the function \(s \to s \Psi (s)\) is increasing. Let \(\Phi\) be a Young function and \(\Psi (s) \leq C \Phi'(t) \) where \(s = 1/\Psi(t) \Psi'(t)\) for all sufficiently large \(t\). Then \[ n_{\Psi}(N_I^{w}):= \int_0^{\infty} N_I^{w}(t) \Psi (N_I^{w}(t))dt \leq C \| w \|_{L^{\Phi}(I)}. \] They prove the following: If \[ \sup_{I} n_{\Psi_1}(N_I^{v}) n_{\Psi_2}(N_I^{u}) < \infty, \] then \(M_{v^{1/2}} T M_{u^{1/2}}\) is bounded on \(L^2\). For the proof they use the ideas of dyadic shifts and paraproducts by \textit{T. Hytönen, C. Pérez, S. Treil} and \textit{A. Volberg} [``Sharp weighted estimates for dyadic shifts and the \(A_2\) conjecture'', J. Reine Angew. Math. (2014)].