Unions of cubes in \(\mathbb{R}^n\), combinatorics in \(\mathbb{Z}^n\) and the John-Nirenberg and John-Strömberg inequalities (Q2196728)

The author deals with a concrete question intimately connected with the fact that the constants in the John-Nirenberg inequality for BMO functions can be chosen independently of the dimension. The paper is a continuation of the work by the author et al. [J. Funct. Anal. 263, No. 1, 129--166 (2012; Zbl 1254.46033)]. The problem called Question $A(1/2)$ consists in knowing whether there exists an absolute constant \(s>0\) such that, for any positive integer \(d\), whenever \(E_+\) and \(E_-\) are two disjoint sets given by finite unions of dyadic cubes in \([0,1]^d\) satisfying that \(\min\{\lambda(E_+), \lambda(E_-)\}>\frac{1}{2} (1-\lambda(E_+)-\lambda(E_-))\), then there exists a cube \(W\) with sides parallel to the axes contained in \([0,1]^d\) for which \(\min\{\lambda(E_+\cap W), \lambda(E_-\cap W)\}\ge s \lambda(W)\). A similar and a priori more restrictive problem, called Question $A(\tau,s)$, consists in asking whether there exist two absolute constants \(0<\tau<1/2\) and \(s>0\) such that, for any positive integer \(d\), whenever \(E_+\) and \(E_-\) are two disjoint sets given by finite unions of dyadic cubes in any cube \(Q\subset \mathbb R^d\) such that \(\min\{\lambda(E_+), \lambda(E_-)\}>\tau (\lambda(Q)-\lambda(E_+)-\lambda(E_-))\), then there exists a cube with sides parallel to the axes \(W\subset Q\) for which \(\min\{\lambda(E_+\cap W), \lambda(E_-\cap W)\}\ge s \lambda(W)\). Such pairs \((\tau,s)\) are called John-Strömberg pairs for the family of cubes and the author denotes by \(JS(d)\) the collection of all John-Strömberg pairs. It is known from the preceding work [loc.\,cit.]\ that an affirmative answer to Question \(A(\tau\),s) will solve the independence of dimension in John-Nirenberg's inequality. In the paper under review, the author shows that actually that, if Question $A(1/2)$ is answered affirmatively, then Question $ A(\tau,s)$ is also answered affirmatively for some pairs of \((\tau,s)\). In particular, the main result of the paper establishes that, if \((\tau,s)\in JS(d)\), then \(s\in (0,1/2]\) and \(((1-\theta)\tau, s-\theta(1-s))\in JS(d)\) for any \(\theta\in (0,s/(1-s))\). The paper also contains some equivalent formulation of John-Strömberg pairs and some additional results about them, for instance, that \((1/2, 1/4+\varepsilon)\notin JS(d)\) for every \(d\in \mathbb N\) and \(\varepsilon>0\) or that \((\tau,\sigma(\tau,d))\in JS(d)\), where \(\sigma(x,d)=\sup\{s>0: (x,s)\in JS(d)\}\).