Random Euclidean sections of some classical Banach spaces (Q4674832)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 2168741
Language Label Description Also known as
default for all languages
No label defined
    English
    Random Euclidean sections of some classical Banach spaces
    scientific article; zbMATH DE number 2168741

      Statements

      Random Euclidean sections of some classical Banach spaces (English)
      0 references
      0 references
      0 references
      0 references
      0 references
      19 May 2005
      0 references
      isomorphic Dvoretzky theorem
      0 references
      volume ratios
      0 references
      random subspaces
      0 references
      The first part of the paper deals with the so-called isomorphic Dvoretzky theorem for some classes of normed spaces. The isomorphic Dvoretzky theorem by \textit{V.~D.~Milman and G.~Schechtman} [see Math. Sci. Res. Inst. Publ. 34, 159--164 (1999; Zbl 0942.46012) and C. R. Acad. Sci., Paris, Sér. I 321, No. 5, 541--544 (1995; Zbl 0836.46007); see also \textit{A.~E.~Litvak, P.~Mankiewicz} and \textit{N.~Tomczak-Jaegermann}, C. R., Math., Acad. Sci. Paris 335, No. 4, 345--350 (2002; Zbl 1016.52002) and references therein] states that any \(n\)-dimensional normed space has for every \(k\leq n\) a \(k\)-dimensional subspace \(E\) such that the Banach--Mazur distance between \(E\) and \(\ell _2^k\) is bounded by \(c (1 + \sqrt{(k / \ln (2n /k))})\), where \(c\) is a positive absolute numerical constant and the Banach--Mazur distance between two normed spaces \(X\) and \(Y\) is defined by \(d(X, Y) = \inf \{ \| T \| \cdot \| T^{-1}\| \}\), where the infimum is taken over all linear invertible operators \(T: X \to Y\). The authors introduce the notion of the \((C, s)\)-estimate, \(C>0\), \(s>0\), for a given set of vectors. Namely, a set of vectors \(u_1,\dots, u_N\) satisfies a \((C, s)\)-estimate if for every \(t_1,\dots, t_N \in \mathbb R\) and every \(m\leq N\) one has \[ C m^{-1/s} \biggl(\sum _{i=1}^m (t_i^*)^{2} \biggr)^{1/2} \leq \biggl\| \sum _{i=1}^N t_i u_i \biggr\| \leq \biggl(\sum _{i=1}^N t_i^2 \biggr)^{1/2}, \] where \(\{t_i^*\}_{i=1}^N\) denotes the non-increasing rearrangement of \(\{| t_i| \}_{i=1}^N\). Clearly, the canonical basis of \(\ell _q^n\), \(q\geq 2\), satisfies a \((1, s)\)-estimate with \(1/s = (1/2)-1/q\). Moreover, by a result of \textit{J.~Bourgain} and \textit{S.~J.~Szarek} [Isr. J. Math. 62, No. 3, 169--180 (1988; Zbl 0654.46018)], there exists an absolute positive constant \(C\) such that every \(n\)-dimensional space has a set of \(N\geq n/2\) vectors satisfying a \((C, 2)\)-estimate. The authors improve the estimate of the distance in the isomorphic Dvoretzky theorem for the \(n\)-dimensional spaces containing a set of \(N\geq n/2\) vectors satisfying a \((C, s)\)-estimate, \(s>2\). As a corollary they obtain sharp (in \(n\), \(k\) and \(q\)) estimates for \(\ell _q^n\), \(q\geq 2\) (see also the work by Milman and Schechtman mentioned above and \textit{A.~E.~Litvak and V.~D.~Milman} [Can. Math. Bull. 46, 242--251 (2003; Zbl 1074.46005)]). Namely, they prove that there are absolute positive constants \(c_1\), \(c_2\), \(c_3\), \(C_1\), \(C_2\) such that the Banach--Mazur distance from a ``random'' \(k\)-dimensional subspace of \(\ell _q^n\) to the Euclidean space of corresponding dimension is bounded by (i) 3 for \(k\leq c_1 q n^{2/q}\); (ii) \(C_1 n^{-1/q} \sqrt{k/q}\) for \(c_1 q n^{2/q} \leq k\leq c_2 q e^{-q} n\); (iii) \(C_2 k^{(1/2)-1/q}/ \ln (2n/k)\) otherwise. Then the Schatten classes are investigated in a similar way. The second part of the paper is devoted to the study of volume ratios with respect to quotients of subspaces of \(L_q\), started by \textit{Y.~ Gordon} and \textit{M.~Junge} in [Positivity 1, No. 1, 7--43 (1997; Zbl 0911.46010) and Stud. Math. 136, No. 2, 147--182 (1999; Zbl 0952.46008)]. The authors investigate volume ratios of ``random'' subspaces of normed spaces. Finally, the authors obtain estimates for Gelfand numbers of operators acting from \(X\) to \(Y\), where \(Y\) is an arbitrary normed space and (i) \(X\) is a space with a basis satisfying a \((C, s)\)-estimate, or (ii) \(X\) is a space of real matrices with some special norm.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references