zbMATH — the first resource for mathematics

Examples
Geometry Search for the term Geometry in any field. Queries are case-independent.
Funct* Wildcard queries are specified by * (e.g. functions, functorial, etc.). Otherwise the search is exact.
"Topological group" Phrases (multi-words) should be set in "straight quotation marks".
au: Bourbaki & ti: Algebra Search for author and title. The and-operator & is default and can be omitted.
Chebyshev | Tschebyscheff The or-operator | allows to search for Chebyshev or Tschebyscheff.
"Quasi* map*" py: 1989 The resulting documents have publication year 1989.
so: Eur* J* Mat* Soc* cc: 14 Search for publications in a particular source with a Mathematics Subject Classification code (cc) in 14.
"Partial diff* eq*" ! elliptic The not-operator ! eliminates all results containing the word elliptic.
dt: b & au: Hilbert The document type is set to books; alternatively: j for journal articles, a for book articles.
py: 2000-2015 cc: (94A | 11T) Number ranges are accepted. Terms can be grouped within (parentheses).
la: chinese Find documents in a given language. ISO 639-1 language codes can also be used.

Operators
a & b logic and
a | b logic or
!ab logic not
abc* right wildcard
"ab c" phrase
(ab c) parentheses
Fields
any anywhere an internal document identifier
au author, editor ai internal author identifier
ti title la language
so source ab review, abstract
py publication year rv reviewer
cc MSC code ut uncontrolled term
dt document type (j: journal article; b: book; a: book article)
For most large underdetermined systems of linear equations the minimal 1 -norm solution is also the sparsest solution. (English) Zbl 1113.15004

The author studies sparse solutions to underdetermined linear systems y=Φx where y n is a given vector and Φ is a given n×m matrix with n<mτn.

In the last years, a large body of research has focused on the use of overcomplete signal representations. If we have more atoms φ i Φ than observation dimensions, m>n, then there are many possible representations Φx=y for a given Φ and y. The sparse representation problem is then to find the representation x with the fewest possible non-zero components,

minx 0 subjecttoΦx=y,(P 0 )

where 0 is the l 0 norm of a vector x, i.e., the number of its non-zero coefficients. This is well known to be a NP-hard problem [cf. B. K. Natarajan, SIAM J. Comput. 24, No. 2, 227–234 (1995; Zbl 0827.68054)].

In the signal processing community, S. S. Chen, D. L. Donoho and M. A. Saunders [SIAM J. Sci. Comput. 20, No. 1, 33–61 (1999; Zbl 0919.94002)] proposed to approximate (P 0 ) with the ‘relaxed’ l 1 problem

minx 1 subjecttoΦx=y,(P 1 )

where 1 = i=1 m |x i | is the l 1 norm of a vector x. Problem (P 1 ), which they called basis pursuit (BP), can be formulated as a linear programming (LP) problem, which can be solved using well known optimization methods such as the simplex method or interior point methods [cf. M. H. Wright, Bull. Am. Math. Soc., New Ser. 42, No. 1, 39–56 (2005; Zbl 1114.90153)]. Surprisingly, when the answer to (P 0 ) is sparse, it can be the same as the answer to (P 1 ).

The equivalence breakdown point of a matrix Φ, EBP(Φ), is the maximal number N such that, for every x 0 with fewer than N nonzeros, the corresponding vector y=Φx 0 generates a linear system y=Φx for which problems (P 0 ) and (P 1 ) have identical unique solutions, both equal to x 0 .

The main result of the paper is: For each τ>1, there exists a constant ρ * (τ)>0 so that for every sequence (m n ) with m n τn,

Prob{n -1 EBP(Φ n,m n )ρ * (τ)}1,n·

In other words, for the overwhelming majority of n×m matrices Φ, the sparsest possible solution and the minimal l 1 -solution of y=Φx coincide whenever a solution with at most ρ * n nonzeros exists.

The author uses techniques from Banach space theory, in particular quantitative extensions of the almost spherical sections theorem [cf. A. Dvoretsky, Proc. Int. Symp. Linear Spaces, Jerusalem 1960, 123–160 (1961; Zbl 0119.31803)], and other related tools exploiting randomness in highdimensional spaces, including properties of the minimum eigenvalue of Wishart matrices.

The paper finishes showing as two standard approaches, simple thresholding and greedy subset selection algorithms, in contrast to the l 1 -minimization, which are poorly performed in the setting of the paper.

MSC:
15A06Linear equations (linear algebra)
65F20Overdetermined systems, pseudoinverses (numerical linear algebra)
65F50Sparse matrices (numerical linear algebra)
60D05Geometric probability and stochastic geometry
41A65Abstract approximation theory
52A41Convex functions and convex programs (convex geometry)
90C20Quadratic programming
Software:
lars