Saturating constructions for normed spaces (Q1774019): Difference between revisions

The present paper deals with saturating properties of high-dimensional normed spaces. The authors investigate, in a sense, how much information can be lost when one passes from a space to its subspaces or quotients. They discover the following saturation phenomenon: ``passing to large subspaces or quotients cannot, in general, erase \(k\)-dimensional features of a space'' if \(k\) is small enough. More precisely, the authors prove the following. Theorem. There exists an absolute constant \(c_1>0\) such that for every \(1\leq m\leq n\) and every finite-dimensional normed space \(Y\) with \(\dim Y \leq c_1 \min \{ m/\sqrt{n}\), \(m^2/(n \ln n )\},\) there exists an \(n\)-dimensional normed space \(X\) such that every quotient \(Z\) of \(X\) with \(\dim Z \geq m\) contains a \(1\)-complemented subspace isometric to \(Y\). This theorem, in particular, implies a negative answer to the problem posed by \textit{V.~D.\ Milman} in his 1986 ICM talk [Proc.\ Int.\ Congr.\ Math., Berkeley/Calif.\ 1986, No.~2, 961--975 (1987; Zbl 0698.46009)], namely: ``Does every \((2n)\)-dimensional normed space admit a quotient of dimension at least \(n\) whose cotype 2 constant is bounded by a universal numerical constant?'' The authors also solve (in the negative) two other problems of similar nature posed by V.~Milman in the same talk. The proofs use probabilistic arguments. The most basic idea of applying probabilistic constructions of normed spaces goes back to \textit{E.~D.\ Gluskin} [Funct.\ Anal.\ Appl.\ 15, 57--58 (1981); translation from Funkts.\ Anal.\ Prilozh.\ 15, No.~1, 72--73 (1981; Zbl 0469.46017)]. After Gluskin's work, such constructions have been used in many proofs, see, e.g., the survey by \textit{P.~Mankiewicz} and the second author [in: W.~B.\ Johnson, J.~Lindenstrauss (eds.), ``Handbook of the geometry of Banach spaces. Volume 2'' (North--Holland, Amsterdam), 1201--1246 (2003; Zbl 1057.46010)]. The authors of the present paper develop probabilistic techniques and introduce new ingredients, such as a decoupling-type lemma which allows decoupling of otherwise dependent events and which is of independent interest. The study started in this paper was continued by the authors in [J.\ Funct.\ Anal.\ 221, No.~2, 407--438 (2005; Zbl 1090.46008), see the following review].