Deterministic discrepancy minimization

Nikhil Bansal, Joel Spencer

Research output: Contribution to journalArticle

Abstract

We derandomize a recent algorithmic approach due to Bansal (Foundations of Computer Science, FOCS, pp. 3-10, 2010) to efficiently compute low discrepancy colorings for several problems, for which only existential results were previously known. In particular, we give an efficient deterministic algorithm for Spencer's six standard deviations result (Spencer in Trans. Am. Math. Soc. 289:679-706, 1985), and to find a low discrepancy coloring for a set system with low hereditary discrepancy. The main new idea is to add certain extra constraints to the natural semidefinite programming formulation for discrepancy, which allow us to argue about the existence of a good deterministic move at each step of the algorithm. The non-constructive entropy method is used to argue the feasibility of this enhanced SDP.

Original languageEnglish (US)
Pages (from-to)451-471
Number of pages21
JournalAlgorithmica (New York)
Volume67
Issue number4
DOIs
StatePublished - 2013

Fingerprint

Coloring
Discrepancy
Computer science
Colouring
Entropy
Entropy Method
Set Systems
Semidefinite Programming
Deterministic Algorithm
Standard deviation
Computer Science
Efficient Algorithms
Formulation

Keywords

  • Algorithms
  • Derandomization
  • Discrepancy
  • Semi-definite programming

ASJC Scopus subject areas

  • Computer Science(all)
  • Computer Science Applications
  • Applied Mathematics

Cite this

Deterministic discrepancy minimization. / Bansal, Nikhil; Spencer, Joel.

In: Algorithmica (New York), Vol. 67, No. 4, 2013, p. 451-471.

Research output: Contribution to journalArticle

Bansal, Nikhil ; Spencer, Joel. / Deterministic discrepancy minimization. In: Algorithmica (New York). 2013 ; Vol. 67, No. 4. pp. 451-471.
@article{006aa6e96a92472784a3a934e7f1d5fc,
title = "Deterministic discrepancy minimization",
abstract = "We derandomize a recent algorithmic approach due to Bansal (Foundations of Computer Science, FOCS, pp. 3-10, 2010) to efficiently compute low discrepancy colorings for several problems, for which only existential results were previously known. In particular, we give an efficient deterministic algorithm for Spencer's six standard deviations result (Spencer in Trans. Am. Math. Soc. 289:679-706, 1985), and to find a low discrepancy coloring for a set system with low hereditary discrepancy. The main new idea is to add certain extra constraints to the natural semidefinite programming formulation for discrepancy, which allow us to argue about the existence of a good deterministic move at each step of the algorithm. The non-constructive entropy method is used to argue the feasibility of this enhanced SDP.",
keywords = "Algorithms, Derandomization, Discrepancy, Semi-definite programming",
author = "Nikhil Bansal and Joel Spencer",
year = "2013",
doi = "10.1007/s00453-012-9728-1",
language = "English (US)",
volume = "67",
pages = "451--471",
journal = "Algorithmica",
issn = "0178-4617",
publisher = "Springer New York",
number = "4",

}

TY - JOUR

T1 - Deterministic discrepancy minimization

AU - Bansal, Nikhil

AU - Spencer, Joel

PY - 2013

Y1 - 2013

N2 - We derandomize a recent algorithmic approach due to Bansal (Foundations of Computer Science, FOCS, pp. 3-10, 2010) to efficiently compute low discrepancy colorings for several problems, for which only existential results were previously known. In particular, we give an efficient deterministic algorithm for Spencer's six standard deviations result (Spencer in Trans. Am. Math. Soc. 289:679-706, 1985), and to find a low discrepancy coloring for a set system with low hereditary discrepancy. The main new idea is to add certain extra constraints to the natural semidefinite programming formulation for discrepancy, which allow us to argue about the existence of a good deterministic move at each step of the algorithm. The non-constructive entropy method is used to argue the feasibility of this enhanced SDP.

AB - We derandomize a recent algorithmic approach due to Bansal (Foundations of Computer Science, FOCS, pp. 3-10, 2010) to efficiently compute low discrepancy colorings for several problems, for which only existential results were previously known. In particular, we give an efficient deterministic algorithm for Spencer's six standard deviations result (Spencer in Trans. Am. Math. Soc. 289:679-706, 1985), and to find a low discrepancy coloring for a set system with low hereditary discrepancy. The main new idea is to add certain extra constraints to the natural semidefinite programming formulation for discrepancy, which allow us to argue about the existence of a good deterministic move at each step of the algorithm. The non-constructive entropy method is used to argue the feasibility of this enhanced SDP.

KW - Algorithms

KW - Derandomization

KW - Discrepancy

KW - Semi-definite programming

UR - http://www.scopus.com/inward/record.url?scp=84885330935&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84885330935&partnerID=8YFLogxK

U2 - 10.1007/s00453-012-9728-1

DO - 10.1007/s00453-012-9728-1

M3 - Article

AN - SCOPUS:84885330935

VL - 67

SP - 451

EP - 471

JO - Algorithmica

JF - Algorithmica

SN - 0178-4617

IS - 4

ER -