Bloom's inequality: commutators in a two-weight setting (Q2634770)

The authors give an alternative proof of Bloom's inequality with a view to a later extension of the result. Let \(\mu(x)\) be a function on \(R\) that is positive almost everywhere and define \(L^2(\mu)\) in the obvious way with the norm \[ \| f\|^2_{L^2(\mu)} \equiv \int_R | f(x) |^2 \mu(x) \, dx. \] Let \(H\) be the Hilbert transform. It is easiest to understand Bloom's inequality starting from a special case, \(\mu = \lambda \in A_2\), where \(A_2\) is the Muckenhoupt class. Then it is known that the commutator \([b, H]\) satisfies \([b, H]: L^2(\mu) \to L^2(\mu)\) if and only if \(b \in BMO\). Bloom characterizeses when the commutator satisfies \([b, H]: L^p(\mu) \to L^p(\lambda)\), by showing that \(b\) must be in a \(BMO\) space adapted to the weights, and Bloom's inequality says that \[ \| [b, H]: L^p(\mu) \to L^p(\lambda) \| \approx \| b \|_{BMO_{\rho} }, \] where \(\rho = (\frac{\mu}{\lambda})^{1/p}\) and the \(BMO\) space is defined as usual except for dividing the \(L^2\)-average over an interval \(I\) by \(\rho(I) = \int_I \rho(x) \, dx = \langle \rho \rangle_I\) instead of by the measure of the interval, \[ \| b \|_{BMO_{\rho} } \approx \sup_I \left( \frac{1}{\rho(I)} \int_I | b(x) - \langle b \rangle_I |^2 \, dx \right)^{1/2}. \] The authors prove this in the case \(p = 2\) using dyadic intervals and writing the Hilbert transform as an average of Haar shifts. In the course of the proof, equivalent characterizations of the space \(BMO_{\rho}\) are used. They propose in future papers to extend the result from the Hilbert transform in one dimension to general Calderón-Zygmund operators in all dimensions.