Martingale inequalities in noncommutative symmetric spaces (Q5894513): Difference between revisions

The classical Burkholder inequalities express the equivalence between the \(L^p\) norm of an \(L^p\)-bounded discrete-time martingale and the \(L^p\) norm of its square function, where \(1 < p < \infty\). Pisier and Xu generalised this to the non-commutative context, where the probability space is replaced by a finite von~Neumann algebra \(\mathcal{M}\) with a faithful normal trace \(\tau\). Their result employs row and column versions of the square function, and takes a different form depending on whether \(p \geqslant 2\) or \(p < 2\). The work under review extends the result of Pisier and Xu to symmetric operator spaces. Let \(E\) be a rearrangement-invariant Banach function space on \([ 0, 1 ]\) and let \(E( \mathcal{M} )\) be the corresponding symmetric operator space; if \(E = L^p[ 0, 1 ]\) then \(E( \mathcal{M} )\) is the usual non-commutative \(L^p\) space associated with \(( \mathcal{M}, \tau )\). Suppose \(x = ( x_n )_{n \geqslant 1}\) is a bounded \(E( \mathcal{M} )\)-martingale, so that \[ \| x \|_{E( \mathcal{M} )} := \sup_{n \geqslant 1} \| x_n \|_{E( \mathcal{M} )} < \infty. \] With \(p_E\) and \(q_E\) denoting the lower and upper Boyd indices of \(E\), the main result is as follows. (a) If \(2 < p_E \leqslant q_E < \infty\) then \(\| x \|_{E( \mathcal{M} )}\) is equivalent to \[ \max\{ \| S_c( x ) \|_{E( \mathcal{M} )}, \| S_r( x ) \|_{E( \mathcal{M} )} \}, \] where \(S_c\) and \(S_r\) are the column and row versions of the square function. (b) If \(1 < p_E \leqslant q_E < 2\) then \(\| x \|_{E( \mathcal{M} )}\) is equivalent to \[ \inf\{ \| S_c( y ) \|_{E( \mathcal{M} )} + \| S_r( z ) \|_{E( \mathcal{M} )} \}, \] where the infimum is taken over all decompositions \(d x_n = d y_n + d z_n\) with \(( d y_n )\) and \(( d z_n )\) martingale difference sequences. This result is applied to give Burkholder-Gundy inequalities for martingales in non-commutative weak \(L^p\)~spaces and non-commutative weak Orlicz spaces.