BMO-boundedness of the maximal operator for arbitrary measures (Q995394)

Let \(\mu\) be a nonnegative Radon measure on \(\mathbb R^n\). The Hardy-Littlewood and Fefferman-Stein maximal operators with respect to \(\mu\) are defined by \[ M_{\mu}f(x) = \sup_{Q \ni x} \frac{1}{\mu (E)} \int_E | f(y) |\, d\mu \quad \text{and} \quad f_{\mu}^{\sharp}(x) = \sup_{Q \ni x} \inf_{c} \frac{1}{\mu (E)} \int_E | f(y) - c |\, d\mu, \] respectively. A function \(f\) is said to belong to \(\text{BMO}(\mu)\) if \[ \| f \|_{\text{BMO}(\mu)} \equiv \| f_{\mu}^{\sharp} \|_{\infty} < \infty. \] When \(\mu\) is a doubling measure, \textit{C. Bennett} [Proc. Am. Math. Soc. 85, 552--556 (1982; Zbl 0512.42022)] proved the following. \[ \| M_{\mu}f \|_{\text{BLO}(\mu)} \leq C_{\mu} \| f \|_{\text{BMO}(\mu)}, \quad \text{where BLO is a subspace of BMO}. \tag{1} \] The main result of this paper is somewhat surprising since it says that in one-dimensional case (1) holds for any nonnegative Radon measure. The proof is essentially new. To prove this theorem the author obtains the following pointwise estimate. (2) is new even in the case of Lebesgue measure. \[ M_{\mu}( M_{\mu} f )(x) \leq 40 M_{\mu} (f_{\mu}^{\sharp})(x) + M_{\mu} f(x). \tag{2} \] The author also proves that for \(n=2\), there is an absolutely continuous measure \(\mu\) such that \(M_{\mu}\) is not bounded on \(\text{BMO}(\mu)\).