Universal approximations for tsp steiner tree and set cover
Download
1 / 18

Universal Approximations for TSP, Steiner Tree, and Set Cover - PowerPoint PPT Presentation


  • 69 Views
  • Uploaded on

Universal Approximations for TSP, Steiner Tree, and Set Cover. Authors: Lujun Jia, Guolong Lin, Guevara Noubir, Rajmohan Rajaraman, Ravi Sundaram Presented by Songjian Lu. contents. Introduction Definition of UTSP,UST,USC Detail of UTSP Other result. Introduction.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Universal Approximations for TSP, Steiner Tree, and Set Cover' - beyla


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
Universal approximations for tsp steiner tree and set cover

Universal Approximations forTSP, Steiner Tree, and Set Cover

Authors: Lujun Jia, Guolong Lin, Guevara Noubir,

Rajmohan Rajaraman, Ravi Sundaram

Presented by Songjian Lu


Contents
contents

  • Introduction

  • Definition of UTSP,UST,USC

  • Detail of UTSP

  • Other result


Introduction
Introduction

  • Consider a courier who delivers packages to different houses and businesses in a city every day. One challenge faced by the courier is to determine a suitable route every day, given the packages to be delivered that day.


UTSP

  • Universal traveling salesman problem

  • DEFINITION:(UTSP).An instance of UTSP is a metric space (V,d). For any cycle(tour) C containing all the vertices in V and a subset S of V, Let CS denote the unique cycle over S in which the ordering of vertices in S is consistent with their ordering in V. The stretch of C is defined as maxSV ||Cs||/||OptTrS||, where OptTrS denotes the minimum cost tour on set S. The universal traveling salesman problem is to find a tour on V with minimum stretch.


UST

  • Universal steiner trees

  • Definition:(UST) An instance of the Universal steiner tree problem is a triple <V,d,r>, where (V,d) forms a metric space, and r is a distinguished vertex in V that we refer to as the root. For any spanning tree T of V, define the stretch of T as maxSV||Ts+r||/||OptSts+r, The goal of the UST problem is to determine a spanning tree with minimum stretch.


USC

  • Universal Set Cover

  • Definition:(USC). An instance of USC is a triple <U,C,c>, where U={e1,e2,…,en} is a ground set of elements, C={S1,S2,…,Sm} is a collection of sets, and c is a cost function mapping C to Q+. We define an assignment f as a function from U to C that satisfies ef(e) for all e in U. We extend the definition of f to apply to any SU as follows: f(S)={f(e):e  S}. We next define the cost of f(S), ||f(S)||, as  SU c(x) for all x in f(S), and the stretch of an assignment f as max SU c(f(S))/Opt(S), where Opt(S) is the cost of the optimal set cover solution to S. And the goal is to compute an assignment f with minimum stretch.


Utsp detail 1
UTSP detail-1

  • Definition: ((r,,I)-PARTITION). A (r,,I)-partition of a metric space (V,d) is a partition {Si} of V such that (i) the diameter of every set Si in the partition is at most r and (ii) for every node vV, the ball Br(v) intersects at most I set in the partition, where Br(v)={uV|d(u,v)r}.

  • A (,I)-partition scheme is a procedure that computes a (r,,I)-partition for any r>0.


Utsp detail 2
UTSP detail-2

  • Lemma 1: The general metric spaces has a (r,log(n),log(n))-partition that can be computed efficiently for any r>0.

  • Lemma 2: For a doubling metric space (V,d) with doubling constant , a (r,1,3)-partition can be computed efficiently for any r>0.


Utsp detail 3
UTSP detail-3

  • Lemma 3: If the points are in k-dimensional Euclidean space, a (r,k1/2(2+),2k)-partition can be computed efficiently for any r>0, >0.

  • Lemma 5: For a growth-restricted metric space (V,d) with expansion rate c, a (r,4,c3)-partition can be computed efficiently for any r>0.


Utsp detail 5
UTSP detail-5

  • Theorem 1: Given a metric space (V,d) with n vertices and a (,I)-partitioning scheme, UTSP-ALG returns a tour with stretch O(2I log n) in polynomial time.


Utsp detail 6
UTSP detail-6

  • Corollary 1: For any metric space over n vertices, the algorithm UTSP-ALG returns a tour with stretch O(log4(n)/log(log(n))).

  • Corollary 2: For any doubling. Euclidean, or growth-restricted metric space over n vertices, UTSP-ALG return a tree with stretch O(log(n)).


Utsp detail 61
UTSP detail-6

  • UTSP-ALG runing in p time.

  • For j 1, let PSj be a maxima subset of vertices S that are pairwise separated by distance at least j-1.

  • Lemma 6: For j2, the cost of edges in Ts at level j is at most |PSj-1|jI.

  • Lemma 7: If mi 1, i=1,…,k,c 2, then 1 i  kmici  (logc(max1 i  k mi)+3) max1 i  k (mi ci)


Other result 1
Other result-1

  • Universal steiner trees

  • Theorem 2: Given a metric space (V,d) with n vertices, a root rV, and a (,I)-partitioning scheme for (V,d), a spanning tree with stretch O(2I log(n)) can be constructed in polynomial time.

  • Corollary 3: For any metric space over n vertices, UST-ALG returns a tree with stretch O(log4(n)/loglog(n)).

  • Corollary 4: For any doubling, Euclidean, or growth-restricted metric space over n vertices, UST-ALG returns a tree with stretch O(log(n)).


Other result 2
Other result-2

  • Theorem 3: There exists a set of n vertices in two-dimensional Euclidean space, for which every spanning tree has an (log(n)/loglog(n)) stretch.

  • Theorem 4: No on-line algorithm can achieve a competitive ratio which is better than (log(n)/loglog(n)) for the Steiner tree problem of n vertices in vertices in the plane, or even for n vertices in the n by n grid.


Other result 3
Other result-3

  • Theorem 5: For any USC instance with n elements, USC-ALG has a stretch of O((nln(n))1/2).

  • Theorem 6: There exists an n-element instance of USC with uniform costs for which the best stretch achievable is ( n1/2).


Homework 2
Homework-2

  • Try to find a natural problem and define it to be a universal approximation problem.



ad