Nonpositive curvature is not coarsely universal (Q2316827)

The paper is devoted to the non-Hilbertian part of the following problem suggested by \textit{M. Gromov} on page 218 of his famous essay [Geometric group theory. Volume 2: Asymptotic invariants of infinite groups. Proceedings of the symposium held at the Sussex University, Brighton, July 14--19, 1991. Cambridge: Cambridge University Press (1993; Zbl 0841.20039)]: ``There is no known geometric obstruction for uniform embeddings into infinite dimensional spaces. In particular, it is unclear whether every separable metric space can be uniformly embedded into the Hilbert space \(\mathbb{R}^\infty\).'' The context of the essay allows to conclude that Gromov meant nonpositively curved infinite dimensional spaces. Following the established tradition, the authors use the term ``coarse embeddings'' for embeddings which Gromov named ``uniform''. The case of Hilbert spaces was a subject of extensive research and some geometric obstructions for coarse embeddings into Hilbert spaces (and some other Banach spaces) are known by now. Answering the non-Hilbertian part of Gromov's problem, the authors prove (Theorem 1): ``There is a metric space \(Y\) that does not embed coarsely into any nonpositively curved Alexandrov space \(X\).'' The proof uses the notion of barycentric metric spaces, which is introduced in the following way: ``For a set \(\Omega\), let \(\mathscr{P}_{\Omega}^{<\infty}\) be the space of all finitely supported probability measures on \(\Omega\). A function \(\mathfrak{B}:\mathscr{P}_\Omega^{<\infty}\to X\) is said to be a barycenter map if \(\mathfrak{B}(\delta_x)=x\) for every \(x\in \Omega\), where \(\delta_x\) is the point mass at \(x\).'' ``If \(q\in [1,\infty)\), then a metric space \((X,d_X)\) is said to be \(q\)-barycentric with constant \(\beta\in (0,\infty)\) if there exists a barycenter map \(\mathfrak{B}:\mathscr{P}_X^{<\infty}\to X\) such that every \(x\in X\) and \(\mu\in\mathscr{P}_X^{<\infty}\) satisfy \[ d_X\big(\mathfrak{B}(\mu),x\big)^q + \frac{1}{\beta^q} \int_X d_X\big(\mathfrak{B}(\mu),y\big)^q \mathrm{d}\mu(y) \leq \int_X d_X(x,y)^q\mathrm{d}\mu(y). \] A metric space is said to be barycentric if it is \(q\)-barycentric for some \(q\in [1,\infty)\).'' It is known that any Alexandrov space of nonpositive curvature is \(2\)-barycentric with constant \(\beta=1\). For this reason, Theorem 1 (stated above) follows from the next result of the authors (Theorem 3): ``If \(p,q\in [1,\infty)\) satisfy \(p>q\), then \(\ell_p\) does not embed coarsely into any \(q\)-barycentric metric space.'' Theorem 3 is derived from Theorem 5 stating that \(q\)-barycentric spaces have sharp metric cotype \(q\). The term \textit{sharp metric cotype} is introduced in this paper as metric cotype with ``good inequalities'' for the so-called scaling parameter. (The authors decided to give a name to the notion which was already used in the foundational paper [\textit{M. Mendel} and \textit{A. Naor}, Ann. Math. (2) 168, No. 1, 247--298 (2008; Zbl 1187.46014)].) To prove Theorem 5, the authors prove some inequalities for nonlinear martingales with values in barycentric metric spaces. These inequalities also allow the authors to evaluate distortion of embeddings of grids of \(\ell_\infty^n\) into \(\ell_q\) for \(q\in(2,\infty)\) and to improve previously known results for such embeddings. The paper also contains a very interesting Section 2 entitled ``Limitations of previous approaches and further open questions''. This section contains a survey of other tools which one could try to use in order to prove Theorem 1. This section contains a very useful and readable survey of modern metric geometry and Banach space theory in their relation with the Gromov's problem which is the subject of this paper, with many motivated open questions suggested throughout this survey.