Duality theorems for noncommutative quasi-martingale spaces (Q519948)

Let \(({\mathcal M}, \tau)\) be a non-commutative probability space, where \({\mathcal M}\) is a von Neumann algebra acting on a Hilbert space \(H\) and \(\tau\) is a normal faithful trace on \({\mathcal M}\) with \(\tau(1) =1\). For \( 1 \leqslant p \leqslant \infty\), \(L_p({\mathcal M})\) denotes the associated non-commutative \(L_p\)-space with the norm \[ \| x \|_p = \tau(| x |^p)^{1/p}, \qquad x \in L_p({\mathcal M}), \tag{1} \] where \(| x | = (x^* x)^{1/2}\) is the usual modulus of \(x\). Let \(({\mathcal M}_n)_{n \geq 1}\) be an increasing filtration of von Neumann subalgebras of \({\mathcal M}\) such that the union of \({\mathcal M}_n\)'s is weak\(^*\)-dense in \({\mathcal M}\), and \({\mathcal E}_n\) the conditional expectation with respect to \({\mathcal M}_n\). An adapted sequence \(x\) \(=\) \((x_n)_{n \geq 1}\) in \(L_1({\mathcal M})\) is called a \(p\)-quasi-martingale with respect to \(({\mathcal M}_n)_{n \geq 1}\) (or simply a quasi-martingale for \(p=1\)) if \[ \sum_{n=1}^{\infty} \| {\mathcal E}_{n-1} (d x_n) \|_p < \infty. \tag{2} \] If in addition \(x\) \(=\) \((x_n)_{n \geq 1}\) \(\subset\) \(L_p({\mathcal M})\), then \(x\) is called an \(L_p({\mathcal M})\)-quasi-martingale. The non-commutative quasi-martingale space \(\tilde{L}_p({\mathcal M})\) is defined as the space of all bounded \(L_p({\mathcal M})\)-quasi-martingales, equipped with the norm \(\| \cdot \|_p\). In this paper, the focus is put on duality theorems for non-commutative quasi-martingales. Let \(1 < p < \infty\) and \(p^{-1} + q^{-1} = 1\). Note that the dual space of all bounded \(L_p({\mathcal M})\)-martingales is the space of all bounded \(L_q({\mathcal M})\)-martingales, which is derived from the fact that the space of all bounded \(L_p({\mathcal M})\)-martingales is isometric to the \(L_p\)-space \(L_p({\mathcal M})\) of operators. However, the case of non-commutative quasi-martingales is quite different. In particular, the space \(\tilde{L}_p({\mathcal M})\) and the Hardy space \(\tilde{H}_p({\mathcal M})\) of quasi-martingales are not isomorphic to the non-commutative \(L_p({\mathcal M})\). Hence their dual spaces cannot come from that of \(L_p({\mathcal M})\). The authors prove that the dual space of \(\tilde{L}_p({\mathcal M})\) admits the following decomposition \[ \tilde{L}_p({\mathcal M})^* = L_q({\mathcal M}) \oplus BD_q({\mathcal M}), \tag{3} \] with equivalent norms, where \(BD_q({\mathcal M})\) is the space of all predictable sequences \(x = (x_n)_{n \geq 1}\) such that \(d x\) \(=\) \((d x_n)_{n \geq 1}\) \(\in\) \(\ell_{\infty}(L_q({\mathcal M}))\) and \(x_1 = 0\), \(d x_n = x_n - x_{n-1}\). Moreover, it can be proved that the dual space of \(\tilde{H}_p({\mathcal M})\) is \(S_q({\mathcal M})\), where \(S_q({\mathcal M})\) is a kind of space which is like but bigger than \(\tilde{H}_q({\mathcal M})\). For the case of \(p=1\), the dual results are also obtained. A basic fact is that each \(p\)-quasi-martingale can be decomposed as a sum of a martingale and a predictable quasi-martingale. This is called Doob's decomposition, which plays an important role in this paper. For other related works, see e.g. [\textit{G. Pisier} and \textit{Q. Xu}, Commun. Math. Phys. 189, No. 3, 667--698 (1997; Zbl 0898.46056)], where the duality relationship \(H_1({\mathcal M})^*\) \(=\) \(BMO({\mathcal M})\) is stated for the case of \(p=1\).