Oscillation in ergodic theory: Higher dimensional results (Q1424026)

This paper extends the studies of [\textit{R. L. Jones, R. Kaufmann, J. M. Rosenblatt}, and \textit{M. Wierdl}, Ergodic Theory Dyn. Syst. 18, No.~4, 889--935 (1998; Zbl 0924.28009)] of square function inequalities to multi-dimensional averages. The basic tool is a comparison of the ergodic averages with various dyadic reversed martingales, but in contrast to the 1-dimensional case, the Fourier transform arguments are now replaced by more geometric almost orthogonality arguments. Let \(X = \mathbb{Z}^d\), and let \(\sigma_n\) be the \(n\)-th dyadic \(\tau\)-algebra, generated by the dyadic cubes \([u,v)\), for which all coordinates of \(u\) and \(v\) are divisible by \((2^n)\). For \(A\subset X\) let \(M_A f(x)\) denote the average of all \(f(x + y)\) with \(y\in A\). Let \(H_n = [-2^{(n+1)}, 2^{(n+1)})^d\), and let \({\mathcal A}_n\) contain all cubes with side length between \(2^{(n-1)}\) and \(2^n\), contained in \(H_n\). If \({\mathcal E}_n\) is the conditional expectation with respect to \(\sigma_n\), it is shown that \(\|(\sum_n\sup_{(A\in{\mathcal A}_n)}| M_Af -\varepsilon_nf|^2)^\frac12 \;\| p\leq c_p\| f\|_p\) \((1 < p \leq 2)\) (the norms are in \(l_p\).) There is also a similar \(l_1\)-inequality. These estimates have several interesting consequences: Hardy-Littlewood inequalities, square function inequalities, oscillation inequalities, jump inequalities, and variation inequalities. The paper also discusses other averages than those over cubes, and one section deals with different results.