We investigate the behaviour of the classical (non-smooth) Hardy-Littlewood maximal operator in the context of Banach lattices. We are mainly concerned with end-point results for p = oo. Naturally, the main role is played by the space BMO. We analyze the range of the maximal operator in BMO X . This turns out to depend strongly on the convexity of the Banach lattice X. We apply these results to study the behaviour of the commutators associated to the maximal operator. We also consider the parallel results for the maximal fractional integral operator.1991 Mathematics subject classification: 42A28, 46B20, 43A25.
IntroductionIn our paper [8] we have introduced the Hardy-Littlewood property (sometimes abbreviated as H.L. property) of a Banach lattice and have characterized it by several boundedness properties of the lattice version M of the Hardy-Littlewood maximal operator or its smooth variants. We refer the reader to this paper and also to [12] for the basic notation and terminology concerning Banach lattices and function spaces.As we explain in [8], the Hardy-Littlewood property is intimately connected to the U.M.D. property, which plays a central role in the development of Vector-Valued Fourier Analysis and, in spite of having been extensively studied (see [6] and [3]), still has some aspects that need to be completely understood. By studying systematically the Hardy-Littlewood property, we hoped to cast some light on the behaviour of Banach lattices regarding the U.M.D. property.The key idea in [8] is that the maximal operator M has a smooth version, obtained by replacing the averages over balls by the convolutions with an appropriate approximate identity. This smooth version can be viewed as a vector-valued singular integral, and its boundedness properties are a consequence of the general theory which was originally set up in [1] and then fully developed in [14]. This latter paper will also be a basic reference for the present work. In particular, the notation for all our function spaces will be taken from it.When one is interested in estimates on the Bochner-Lebesgue spaces L X (K"), only the size matters, and it is irrelevant whether we look at M or to its smooth version