RANDOM SECTIONS OF ELLIPSOIDS AND THE POWER OF RANDOM INFORMATION

被引:22
作者
Hinrichs, Aicke [1 ]
Krieg, David [1 ]
Novak, Erich [2 ]
Prochno, Joscha [3 ,4 ]
Ullrich, Mario [1 ,5 ]
机构
[1] Johannes Kepler Univ Linz, Inst Anal, Altenbergerstr 69, A-4040 Linz, Austria
[2] Friedrich Schiller Univ Jena, Inst Math, Ernst Abbe Pl 2, D-07743 Jena, Germany
[3] Karl Franzens Univ Graz, Inst Math & Wissensch Rechnen, Heinrichstr 36, A-8010 Graz, Austria
[4] Univ Passau, Fac Comp Sci & Math, Passau, Germany
[5] Lomonosov Moscow State Univ, Moscow Ctr Fundamental & Appl Math, Moscow, Russia
基金
奥地利科学基金会;
关键词
Random intersection; random information; L-2; approximation; high dimensional convexity; Gaussian random matrix; comparison principles for Gaussian processes; least squares; SMALLEST SINGULAR-VALUE; RANDOM MATRICES; PROPORTIONAL SECTIONS; RANDOM SUBSPACES; DIAMETER; INVERTIBILITY; DIMENSION; NORMS;
D O I
10.1090/tran/8502
中图分类号
O1 [数学];
学科分类号
0701 ; 070101 ;
摘要
We study the circumradius of the intersection of an rn-dimensional ellipsoid epsilon with semi-axes sigma(1) >= ... >= sigma(m) with random subspaces of codimension n, where n can be much smaller than m. We find that, under certain assumptions on a, this random radius R-n = R-n (sigma) is of the same order as the minimal such radius sigma(n+1) with high probability. In other situations R-n is close to the maximum sigma(1). The random variable R-n naturally corresponds to the worst-case error of the best algorithm based on random information for L-2-approximation of functions from a compactly embedded Hilbert space H with unit ball epsilon. In particular, sigma(k) is the kth largest singular value of the embedding H hooked right arrow L-2. In this formulation, one can also consider the case m = infinity and we prove that random information behaves very differently depending on whether a sigma is an element of l(2) or not. For a sigma is not an element of l(2) we get E[R-n] = sigma(1) and random information is completely useless. For sigma is an element of l(2) the expected radius tends to zero at least at rate o(1/root n,) as n -> infinity. In the important case sigma(k) asymptotic to k(-alpha) ln(-beta) (k + 1), where alpha > 0 and beta is an element of R (which corresponds to various Sobolev embeddings we prove E[R-n(sigma)] asymptotic to {sigma 1 if alpha < 1/2 or beta <= alpha = 1/2, sigma(n+1) root ln(n+1) if beta > alpha = 1/2, sigma(n+1) if alpha > 1/2. In the proofs we use a comparison result for Gaussian processes a la Gordon, exponential estimates for sums of chi-squared random variables, and estimates for the extreme singular values of (structured) Gaussian random matrices. The upper bound is constructive. It is proven for the worst case error of a least squares estimator.
引用
收藏
页码:8691 / 8713
页数:23
相关论文
共 41 条
[1]   Smallest singular value of random matrices with independent columns [J].
Adamczak, Radoslaw ;
Guedon, Olivier ;
Litvak, Alexander ;
Pajor, Alain ;
Tomczak-Jaegermann, Nicole .
COMPTES RENDUS MATHEMATIQUE, 2008, 346 (15-16) :853-856
[2]  
Alain Pajor, 1985, C R ACAD SCI PARIS 1, V301, P743
[3]   SHARP NONASYMPTOTIC BOUNDS ON THE NORM OF RANDOM MATRICES WITH INDEPENDENT ENTRIES [J].
Bandeira, Afonso S. ;
van Handel, Ramon .
ANNALS OF PROBABILITY, 2016, 44 (04) :2479-2506
[4]  
Bernd Carl, 1990, CAMBRIDGE TRACTS MAT, V98, DOI [10.1017/ CBO9780511897467, DOI 10.1017/CBO9780511897467]
[5]  
Boult T.t., 1988, INFORM BASED COMPLEX
[6]   Linear vs. nonlinear algorithms for linear problems [J].
Creutzig, J ;
Wojtaszczyk, P .
JOURNAL OF COMPLEXITY, 2004, 20 (06) :807-820
[7]  
Davidson KR, 2001, HANDBOOK OF THE GEOMETRY OF BANACH SPACES, VOL 1, P317, DOI 10.1016/S1874-5849(01)80010-3
[8]   EIGENVALUES AND CONDITION NUMBERS OF RANDOM MATRICES [J].
EDELMAN, A .
SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS, 1988, 9 (04) :543-560
[9]   Asymptotic formulas for the diameter of sections of symmetric convex bodies [J].
Giannopoulos, A ;
Milman, VD ;
Tsolomitis, A .
JOURNAL OF FUNCTIONAL ANALYSIS, 2005, 223 (01) :86-108
[10]  
Giannopoulos AA, 1998, J REINE ANGEW MATH, V497, P113