Variation in probability, ergodic theory and analysis (Q1129411)

The role of square functions is considered for pointwise convergent averaging operators in analysis, ergodic theory and probability. Let \(X=[0,1)\) with Lebesgue measure \(\mu \), let \(E_k\) denote conditional expectation with respect to the dyadic \(\sigma\)-field \({\mathcal F}_k\) with \(2^k\) atoms, \[ V_qf(x) =\left(\sum^\infty_{k=1} \bigl| E_{n_k} f(x)- E_{n_{k-1}} f(x) \bigr |^q \right)^{1/q}, \] where \(\{n_k, k\geq 1\}\) is a fixed increasing sequence of positive integers. It is known that \(V_2\) is a bounded operator on all \(L^p(x)\), \(1<p <\infty\). It is proved that there is a function \(f\in L^\infty (x)\) with \(\| f\|_\infty\leq 1\) such that \(V_qf(x) =\infty\) \(\mu\)-a.e. for all \(q\in [1,2)\). Similar properties are proved for operators \[ V_{q,C} f(x)= \left( \sum^\infty_{k =1} \bigl| D_{2^{-k}} f(x)- E_kf(x) \bigr|^q \right)^{1/q} \] and \[ V_{q,D} f(x)= \left( \sum^\infty_{k=1} \bigl| D_{l_k} f(x)- D_{l_{k-1}} f(x)\bigr |^q \right)^{1/q}, \] where \(D_l f(x)= {1\over l} \int^l_0 f(x+t)dt\), \(\{l_k, k\geq 1\}\) denotes a decreasing sequence of numbers from \([0, 1)\), satisfying \(\liminf_{k\to \infty} (l_k/l_{k-1}) <1\). Let now \((X,\Sigma,m,\tau)\) denote an ergodic dynamical system, \[ A_nf(x) ={1\over n} \sum^{n-1}_{k=0} f(\tau^kx), \quad V_{q,E} f(x)= \left(\sum^\infty_{k=1} \bigl| A_{n_k} f(x)- A_{n_{k-1}} f(x)\bigr |^q \right)^{1/q}, \] where \(\{n_k, k\geq 1\}\) is a fixed increasing sequence of positive integers. It is known that \(V_{2,E}\) is a bounded operator on all \(L^p(X)\), \(1<p< \infty\), and that it is weak type (1,1). It is proved that for \(q\in [1,2)\) and \(\liminf_{k\to\infty} (n_{k-1}/n_k)<1\) there is an \(f\in L^\infty (X)\) such that \(\| f\|_\infty \leq 1\), but \(V_{q,E} f(x)= \infty\) \(m\)-a.e. Finally, it is proved that there is a function \(F\), which is analytic in the open unit disc and continuous on the closed unit disc, and a sequence of numbers \(\{r_k, k\geq 1\}\), \(0<r_k <r_{k+1} <1\), such that \[ \sum^\infty_{k=1} \bigl| F (r_{k+1} e^{2\pi it}) -F(r_ke^{2\pi it}) \bigr|^q =\infty \] for each \(q\in [1,2)\) and for \(\mu\)-a.e. \(t\) in the unit circle. On the other side, for each \(f\in L^2 [0,1)\) we have that \(\| Sf\|_2 \leq \| f\|_2\), where \[ Sf(\theta)= \left(\sum^\infty_{k=1} \left| \sum^\infty_{k= -\infty} a_n (r^n_{k+1} -r^n_k) e^{2\pi in\theta} \right|^2 \right)^{1/2}, \quad f(\theta)= \sum^\infty_{n= -\infty} a_ne^{2\pi in\theta}. \]