Approximating the little Grothendieck problem over the orthogonal and unitary groups

Afonso Bandeira, Christopher Kennedy, Amit Singer

Research output: Contribution to journalArticle

Abstract

The little Grothendieck problem consists of maximizing ∑ i jCi jxixj for a positive semidefinite matrix C, over binary variables xi∈ { ± 1 }. In this paper we focus on a natural generalization of this problem, the little Grothendieck problem over the orthogonal group. Given C∈ Rd n × d n a positive semidefinite matrix, the objective is to maximize ∑ijtr(CijTOiOjT) restricting Oi to take values in the group of orthogonal matrices Od, where Ci j denotes the (ij)-th d× d block of C. We propose an approximation algorithm, which we refer to as Orthogonal-Cut, to solve the little Grothendieck problem over the group of orthogonal matrices Od and show a constant approximation ratio. Our method is based on semidefinite programming. For a given d≥ 1 , we show a constant approximation ratio of αR(d) 2, where αR(d) is the expected average singular value of a d× d matrix with random Gaussian N(0,1d) i.i.d. entries. For d= 1 we recover the known αR(1) 2= 2 / π approximation guarantee for the classical little Grothendieck problem. Our algorithm and analysis naturally extends to the complex valued case also providing a constant approximation ratio for the analogous little Grothendieck problem over the Unitary Group Ud. Orthogonal-Cut also serves as an approximation algorithm for several applications, including the Procrustes problem where it improves over the best previously known approximation ratio of 122. The little Grothendieck problem falls under the larger class of problems approximated by a recent algorithm proposed in the context of the non-commutative Grothendieck inequality. Nonetheless, our approach is simpler and provides better approximation with matching integrality gaps. Finally, we also provide an improved approximation algorithm for the more general little Grothendieck problem over the orthogonal (or unitary) group with rank constraints, recovering, when d= 1 , the sharp, known ratios.

Original languageEnglish (US)
Pages (from-to)433-475
Number of pages43
JournalMathematical Programming
Volume160
Issue number1-2
DOIs
StatePublished - Nov 1 2016

Fingerprint

Orthogonal Group
Unitary group
Approximation algorithms
Approximation
Approximation Algorithms
Orthogonal matrix
Positive Semidefinite Matrix
Procrustes Problem
Binary Variables
Integrality
Semidefinite Programming
Singular Values
Maximise
Denote

Keywords

  • Approximation algorithms
  • Procrustes problem
  • Semidefinite programming

ASJC Scopus subject areas

  • Software
  • Mathematics(all)

Cite this

Approximating the little Grothendieck problem over the orthogonal and unitary groups. / Bandeira, Afonso; Kennedy, Christopher; Singer, Amit.

In: Mathematical Programming, Vol. 160, No. 1-2, 01.11.2016, p. 433-475.

Research output: Contribution to journalArticle

Bandeira, Afonso ; Kennedy, Christopher ; Singer, Amit. / Approximating the little Grothendieck problem over the orthogonal and unitary groups. In: Mathematical Programming. 2016 ; Vol. 160, No. 1-2. pp. 433-475.
@article{7425a21c1a4843b7bb1514886c9ece03,
title = "Approximating the little Grothendieck problem over the orthogonal and unitary groups",
abstract = "The little Grothendieck problem consists of maximizing ∑ i jCi jxixj for a positive semidefinite matrix C, over binary variables xi∈ { {\^A}± 1 }. In this paper we focus on a natural generalization of this problem, the little Grothendieck problem over the orthogonal group. Given C∈ Rd n {\~A}— d n a positive semidefinite matrix, the objective is to maximize ∑ijtr(CijTOiOjT) restricting Oi to take values in the group of orthogonal matrices Od, where Ci j denotes the (ij)-th d{\~A}— d block of C. We propose an approximation algorithm, which we refer to as Orthogonal-Cut, to solve the little Grothendieck problem over the group of orthogonal matrices Od and show a constant approximation ratio. Our method is based on semidefinite programming. For a given d≥ 1 , we show a constant approximation ratio of αR(d) 2, where αR(d) is the expected average singular value of a d{\~A}— d matrix with random Gaussian N(0,1d) i.i.d. entries. For d= 1 we recover the known αR(1) 2= 2 / π approximation guarantee for the classical little Grothendieck problem. Our algorithm and analysis naturally extends to the complex valued case also providing a constant approximation ratio for the analogous little Grothendieck problem over the Unitary Group Ud. Orthogonal-Cut also serves as an approximation algorithm for several applications, including the Procrustes problem where it improves over the best previously known approximation ratio of{\^A} 122. The little Grothendieck problem falls under the larger class of problems approximated by a recent algorithm proposed in the context of the non-commutative Grothendieck inequality. Nonetheless, our approach is simpler and provides better approximation with matching integrality gaps. Finally, we also provide an improved approximation algorithm for the more general little Grothendieck problem over the orthogonal (or unitary) group with rank constraints, recovering, when d= 1 , the sharp, known ratios.",
keywords = "Approximation algorithms, Procrustes problem, Semidefinite programming",
author = "Afonso Bandeira and Christopher Kennedy and Amit Singer",
year = "2016",
month = "11",
day = "1",
doi = "10.1007/s10107-016-0993-7",
language = "English (US)",
volume = "160",
pages = "433--475",
journal = "Mathematical Programming",
issn = "0025-5610",
publisher = "Springer-Verlag GmbH and Co. KG",
number = "1-2",

}

TY - JOUR

T1 - Approximating the little Grothendieck problem over the orthogonal and unitary groups

AU - Bandeira, Afonso

AU - Kennedy, Christopher

AU - Singer, Amit

PY - 2016/11/1

Y1 - 2016/11/1

N2 - The little Grothendieck problem consists of maximizing ∑ i jCi jxixj for a positive semidefinite matrix C, over binary variables xi∈ { ± 1 }. In this paper we focus on a natural generalization of this problem, the little Grothendieck problem over the orthogonal group. Given C∈ Rd n × d n a positive semidefinite matrix, the objective is to maximize ∑ijtr(CijTOiOjT) restricting Oi to take values in the group of orthogonal matrices Od, where Ci j denotes the (ij)-th d× d block of C. We propose an approximation algorithm, which we refer to as Orthogonal-Cut, to solve the little Grothendieck problem over the group of orthogonal matrices Od and show a constant approximation ratio. Our method is based on semidefinite programming. For a given d≥ 1 , we show a constant approximation ratio of αR(d) 2, where αR(d) is the expected average singular value of a d× d matrix with random Gaussian N(0,1d) i.i.d. entries. For d= 1 we recover the known αR(1) 2= 2 / π approximation guarantee for the classical little Grothendieck problem. Our algorithm and analysis naturally extends to the complex valued case also providing a constant approximation ratio for the analogous little Grothendieck problem over the Unitary Group Ud. Orthogonal-Cut also serves as an approximation algorithm for several applications, including the Procrustes problem where it improves over the best previously known approximation ratio of 122. The little Grothendieck problem falls under the larger class of problems approximated by a recent algorithm proposed in the context of the non-commutative Grothendieck inequality. Nonetheless, our approach is simpler and provides better approximation with matching integrality gaps. Finally, we also provide an improved approximation algorithm for the more general little Grothendieck problem over the orthogonal (or unitary) group with rank constraints, recovering, when d= 1 , the sharp, known ratios.

AB - The little Grothendieck problem consists of maximizing ∑ i jCi jxixj for a positive semidefinite matrix C, over binary variables xi∈ { ± 1 }. In this paper we focus on a natural generalization of this problem, the little Grothendieck problem over the orthogonal group. Given C∈ Rd n × d n a positive semidefinite matrix, the objective is to maximize ∑ijtr(CijTOiOjT) restricting Oi to take values in the group of orthogonal matrices Od, where Ci j denotes the (ij)-th d× d block of C. We propose an approximation algorithm, which we refer to as Orthogonal-Cut, to solve the little Grothendieck problem over the group of orthogonal matrices Od and show a constant approximation ratio. Our method is based on semidefinite programming. For a given d≥ 1 , we show a constant approximation ratio of αR(d) 2, where αR(d) is the expected average singular value of a d× d matrix with random Gaussian N(0,1d) i.i.d. entries. For d= 1 we recover the known αR(1) 2= 2 / π approximation guarantee for the classical little Grothendieck problem. Our algorithm and analysis naturally extends to the complex valued case also providing a constant approximation ratio for the analogous little Grothendieck problem over the Unitary Group Ud. Orthogonal-Cut also serves as an approximation algorithm for several applications, including the Procrustes problem where it improves over the best previously known approximation ratio of 122. The little Grothendieck problem falls under the larger class of problems approximated by a recent algorithm proposed in the context of the non-commutative Grothendieck inequality. Nonetheless, our approach is simpler and provides better approximation with matching integrality gaps. Finally, we also provide an improved approximation algorithm for the more general little Grothendieck problem over the orthogonal (or unitary) group with rank constraints, recovering, when d= 1 , the sharp, known ratios.

KW - Approximation algorithms

KW - Procrustes problem

KW - Semidefinite programming

UR - http://www.scopus.com/inward/record.url?scp=84961665160&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84961665160&partnerID=8YFLogxK

U2 - 10.1007/s10107-016-0993-7

DO - 10.1007/s10107-016-0993-7

M3 - Article

VL - 160

SP - 433

EP - 475

JO - Mathematical Programming

JF - Mathematical Programming

SN - 0025-5610

IS - 1-2

ER -