Randomized Algorithms to Locate the Sink in Low Dimensional Unique Sink Orientations of Cubes


Author / Producer

Date

2004-09-18

Publication Type

Student Paper

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

An orientation of the edges of an n-dimensional cube graph such that every face has a unique sink is called a unique sink orientation. These orientations provide a powerful framework for optimization problems, since, for instance, linear and strictly-convex quadratic programming, as well as many other problems, can be reduced to the problem of finding the unique sink of a unique sink orientation. We assume the orientation is implicitly available by evaluating vertices, that is an algorithm can access the orientations by choosing vertices for which the orientation of the incident edges is disclosed. In this setting, an algorithm requiring a polynomial amount of vertex evaluations (in the dimension n) for finding the sink could be used for solving linear programs with running time polynomial in the dimension and in the number of constraints, hence solving a challenging, and up to now still open, problem. On the other hand, game theory, and more precisely, the theory of zero- sum games, provides very powerful tools for the analysis of randomized algorithms. Making use of this machinery, some relevant results about optimal randomized algorithms for finding sinks can be shown, and some insight into this problem can be gained. For instance, through the use of these techniques, Rote managed to reveal the existence of a randomized algorithm which requires 1.438^n expected vertex evaluations in the worst case. This semester project is an independent verification of Rote’s result, which is stated above. We formalize the game-theoretic approach for modeling the problem of finding the sink of a unique sink orientation. In particular, we prove some properties, mainly related to symmetries, for optimal algorithms in any dimension, making use of the sequence form of zero-sum two-player games. These findings allow us to obtain a linear program, whose solutions are the optimal algorithms for finding the sink in 3-dimensional cubes. This LP is much smaller than the one provided by Rote. Moreover, we represent two distinct optimal strategies, hence showing that the optimal strategy for finding the sink is not unique. A computer program generating the LP has also been implemented.

Publication status

published

External links

Editor

Contributors

Examiner : Szabó, Tibor
Examiner : Schurr, Ingo

Book title

Journal / series

Volume

Pages / Article No.

Publisher

ETH Zurich

Event

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

unique sink orientation; optimization; randomized algorithms; zero-sum games

Organisational unit

02643 - Institut für Theoretische Informatik / Inst. Theoretical Computer Science

Notes

Funding

Related publications and datasets