The complex convexity of quasi-normed linear spaces (Q801510)

In this remarkable paper the authors introduce a complex analogue of the well-known concept of uniform convexity for Banach spaces by, roughly speaking, measuring subharmonicity rather than convexity of the unit ball. This is obtained by replacing norms of midpoints of segments in the moduli of convexity by norms of centers of complex disks in the space E under consideration. Correspondingly, complex moduli of convexity are defined by \(H_ p^ E(\epsilon):=\inf \{(1/2\pi)\int^{2\pi}_{0}\| x+e^{i\theta}y\|^ pd\theta)^{1/p}-1| \| x\| =1,\quad \| y\| =\epsilon \},\) for \(0<p<\infty\). Here E is actually only required to be a continuously quasi-normed space, i.e. a space with a quasi-norm which is uniformly continuous on bounded sets. Such a space is then called uniformly PL-convex (henceforth abbreviated uPLc) if \(H^ E_ p(\epsilon)>0\forall \epsilon >0,\) for some \(0<p<\infty\). This is equivalent with \(H^ E_ p(\epsilon)>0\) \(\forall \epsilon >0\), \(\forall 0<p<\infty\). (To the case \(p=\infty\) corresponds the concept of uniform c-convexity introduced by \textit{J. Globevnik} [Proc. Am. Math. Soc. 47, 175-178 (1975; Zbl 0307.46015)], but it is not clear what the precise relations are to uPLc.) It is clear that the behaviour of \(H^ E_ p\) near zero is more important than its actual values. This leads to a parametrization of uPLc through comparing appropriately \(H^ E_ p\) with the functions \(\epsilon^ r\), for \(2\leq r<\infty\). E is defined to be r-uPLc if \((1/2\pi)\int^{2\pi}_{0}\| x+e^{i\theta}y\|^ pd\theta)^{1/p}\geq (\| x\|^ r+\lambda \cdot \| y\|^ r)^{1/r}\) \(\forall x,y\in E\) hols for some \(\lambda >0\). The largest possible \(\lambda\), denoted by \(I_{r,p}(E),\) can be used to describe the behaviour of \(r\cdot H(\epsilon)/\epsilon^ r\) as \(\epsilon\) \(\to 0\). If E is r-uPLc, then so are all Bochner spaces \(L_ p(E)\) for \(0<p\leq r\), and \(I_{r,p}(L_ p(E))=I_{r,p}(E)\) holds. \(L_ p\)- spaces are q-uPLc for all \(0<p<\infty\), with \(q=\max (2,p).\) Further, if E is the dual of any \(C^*\)-algebra, then E is 2-uPLc with \(I_{2,2}(E)\geq\) (result due to U. Haagerup). The existence of an equivalent uniformly convex norm on a Banach space E is known to be closely connected with the behaviour of certain martingale difference sequences in E, cf. \textit{G. Pisier} [Isr. J. Math. 20, 326-350 (1975; Zbl 0344.46030)]. Complex analogues of such results are obtained in this paper by considering martingales which take into account in a particular manner the complex structure of the spaces under consideration. They can be applied to show that a Banach space has finite cotype (cotype q) if it is uPLc (q-uPLc). In particular, Haagerup's result mentioned above yiels again that the dual of a \(C^*\)-algebra has cotype 2 [cf. \textit{N. Tomczak-Jaegermann}, Stud. Math. 50, 163-182 (1974; Zbl 0282.46016)]. On the other hand, q-uPLc is stronger than cotype q: According to G. Pisier, \(L_ 1/H^ 0_ 1\) is not uniformly PL-convexifiable, but it has cotype 2 as was shown by J. Bourgain. From this it follows for example that the usual disc algebra cannot occur as the continuous image of a \(C^*\)- algebra. The difference between uPLc and finite cotype vanishes if one deals with complex Banach lattices. It is shown, among others, that a complex Banach lattice E admits an equivalent norm under which it is a q- uPLc Banach lattice, \(2\leq q<\infty\), if and only if E is of cotype q.