On the UMD constants for a class of iterated \(L_p(L_q)\) spaces (Q1760163)

A Banach space is called UMD if all the difference sequences of martingales in it are unconditional. More precisely, a Banach space \(X\) is UMD if for (all or some) \(1 < s < \infty\) there is a constant \(C > 0\), depending only on \(s\) and \(X\), such that \[ \sup_{\varepsilon_k = \pm 1} \bigg\| \sum_{k = 0}^n \varepsilon_k d f_k \bigg\|_{L^s (X)} \leq C \, \bigg\| \sum_{k = 0}^n d f_k \bigg\|_{L^s (X)} \] for all \(n \geq 0\) and all \(X\)-valued martingale difference sequences \((d f_k)_{0 \leq k \leq n}\). The infimum of such constants \(C\) is denoted by \(C_s (X)\) and is called the UMD constant of \(X\). It is known that every UMD space is super-reflexive. \textit{G. Pisier} [Sémin. Maurey-Schwartz 1974-1975, Espaces \(L^p\), Appl. radonif., Géom. Espaces de Banach, Annexe 2 (1975; Zbl 0315.46022)] constructed a super-reflexive Banach space which is not UMD. Later, \textit{J. Bourgain} [Ark. Mat. 21, 163--168 (1983; Zbl 0533.46008); Math. Nachr. 119, 41--53 (1984; Zbl 0568.42010)] constructed super-reflexive Banach lattices which are not UMD. In this paper, the author gives an elementary construction of super-reflexive Banach lattices \(X (p, q)\), \(1 < p \neq q < \infty\), which are not UMD. The space \(X (p, q)\) is the inductive limit of a sequence of spaces \(Y_n\) defined by iteration as follows: \(Y_0 (p, q) = {\mathbb C}\), \(Y_{n + 1} = L^p [D, \mu (L^q (D, \mu, Y_n)]\), where \(D = \{ - 1, 1\}\) and \(\mu = \frac{1}{2} (\delta_{- 1} + \delta_1)\). Actually, \(X (p, q)\) is \(\theta\)-Hilbertian for some \(0 < \theta <1\), that is, the \(\theta\)-interpolation space of a Banach space with a Hilbert space. To prove that this space is not UMD, the author shows that, if we define \(X_0 (p, q) = {\mathbb C}\), \(X_{n + 1} = \ell_p^2 [\ell_q (X_n)]\), where \(\ell_r^2\) is the two-dimensional space with the norm \(\| \, . \, \|_r\), there exist two constants \(c_1 = c_1 (p, q) > 1\), depending only on \(p, q\), and \(c_2 = c_2 (p, q, s)\), depending only on \(p, q, s\), such that the UMD constant of \(X_n\) satisfies \(c_1^n \leq C_s (X_n) \leq c_2^n\). The upper estimate is not difficult. The main tool to prove the lower estimate is the introduction of a new constant. Given a fixed family of vectors \(\{x_i\}_{i \in I}\) in a Banach space \(X\), \(S( X; \{x_i\})\) is the smallest constant \(C\) (if there is any), such that \[ \bigg\| \sum_{k = 0}^N {\mathbb E}^{{\mathcal A}_k} (\theta_k) x_{i_k} \bigg\|_{L^1 (\Omega, {\mathbb P}; X)} \leq C \, \bigg\| \sum_{k = 0}^N \theta_k x_{i_k} \bigg\|_{L^\infty (\Omega, {\mathbb P}; X)} \] holds for every probability space \((\Omega, {\mathcal T}, {\mathbb P})\), any filtration \({\mathcal A}_0 \subseteq {\mathcal A}_1 \subseteq \cdots \subseteq {\mathcal A}_n \subseteq \cdots \subseteq {\mathcal T}\), every integer \(N\), every family of \(N + 1\) distinct indices \(i_0, \dots, i_N \in I\), and every family of \(N + 1\) functions \(\theta_0, \dots, \theta_N \in L^\infty (\Omega, {\mathcal T}, {\mathbb P})\). Then, if \(X\) is UMD and \(\{x_i\}_{i \in I}\) is a \(1\)-unconditional basic sequence in \(X\), one has \(S(X; \{x_i\}) \leq C_s (X)\) for all \(1 < s < \infty\). Moreover, if \(E\) is a Banach space with an unconditional basis and \(F\) is a Banach space, one can define a space \(E (F)\) in a natural way, and one has \(S [E (F)] \geq S (E) \, S (F)\). Hence, for the canonical basis of the \(X_n\)'s, one has \(S(X_{n + 1}) \geq S [\ell_p^2 ( \ell_q^2)] \, S (X_n)\), so \(S (X_n) \geq c_1 (p, q)^n\), with \(c_1 (p, q) = S [\ell_p^2 ( \ell_q^2)]\), which is \(> 1\). Similar results for the analytic UMD constants are obtained.