Integrality-Integrality of the Min-cut Polytope-Approximations Algorithms-Home Work 03-Computer Science, Study notes of Approximation Algorithms

Integrality of the Min-cut Polytope, Prize-collecting Steiner Tree, Steiner Tree, Minimum-cut Linear Arrangement Problem, Approximations Algorithms, Shuchi Chawla, Home Work, University of Wisconsin, United States of America

Typology: Study notes

2011/2012

Uploaded on 02/15/2012

alexey
alexey 🇺🇸

4.7

(20)

325 documents

1 / 2

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
CS 880: Approximation Algorithms Homework 3
Out: 3/15/07 Due: 4/12/07
1. (Integrality gaps.) In both of the following cases try to find as large a gap as you can.
(a) Give an integrality gap example for the facility location LP that we discussed in class. (That is, give an
example for which the optimal LP solution is much smaller than the optimal integral solution.)
(b) Given an integrality gap example for the Steiner tree LP that we discussed in class.
2. (Integrality of the Min-cut polytope.) Recall the following LP for the s-tmin-cut problem from class:
min X
eE
cexesubject to
X
eP
xe1s-tpaths Pin G= (V, E )
xe0eE
Prove that all basic solutions to this LP are integral.
(Hint: Show that any optimal fractional solution can be written as a convex combination of integral cuts.)
3. (Cuts and `1metrics.) A metric dover a set Vis said to be an `1metric if the points can be mapped to points
in Rkfor some k, such that the distance between any two points according to dis the `1distance between their
mappings: d(x, y) = Pi|xiyi|.
Also, a linear combination over cuts {αS}SVdefines the following metric µα(verify that this is indeed a
metric):
µα(x, y) = X
SV:|S∩{x,y}|=1
αSx6=yV
(Note that PSVαSis not necessarily equal to 1.)
In this problem you will show that the above two classes of metrics—`1metrics and linear combinations of
cuts—are in fact equivalent.
(a) Prove that any metric defined by a linear combination of cuts is an `1metric.
(b) Prove that any `1metric can be expressed as a linear combination of cuts. (Hint: Prove this statement for
a unit-dimensional `1metric first, that is, k= 1. Then extend it to multiple dimensions.)
4. (Prize-collecting Steiner tree.) The prize-collecting Steiner tree problem (PCST) is a variant of Steiner tree in
which there are prizes πvon nodes and costs ceon edges, and a special node rcalled the root. The goal is to
construct a Steiner tree containing rthat minimizes the cost of the edges in the tree plus the value of the nodes
not in the tree.
(a) Give an LP relaxation for this problem using xeas an indicator of the extent to which an edge is included
in the solution, and yvas an indicator of the extent to which a node is covered. (It is okay to have an
exponential number of constraints, as for the Steiner forest LP we studied in class.)
(b) Write the dual of the above LP.
(c) Give a primal dual algorithm for this problem based on the one for Steiner tree (forest). (Don’t forget the
pruning step!)
1
pf2

Partial preview of the text

Download Integrality-Integrality of the Min-cut Polytope-Approximations Algorithms-Home Work 03-Computer Science and more Study notes Approximation Algorithms in PDF only on Docsity!

CS 880: Approximation Algorithms Homework 3

Out: 3/15/07 Due: 4/12/

  1. (Integrality gaps.) In both of the following cases try to find as large a gap as you can.

(a) Give an integrality gap example for the facility location LP that we discussed in class. (That is, give an example for which the optimal LP solution is much smaller than the optimal integral solution.) (b) Given an integrality gap example for the Steiner tree LP that we discussed in class.

  1. (Integrality of the Min-cut polytope.) Recall the following LP for the s-t min-cut problem from class:

min

e∈E

cexe subject to ∑

e∈P

xe ≥ 1 ∀ s-t paths P in G = (V, E)

xe ≥ 0 ∀ e ∈ E

Prove that all basic solutions to this LP are integral. (Hint: Show that any optimal fractional solution can be written as a convex combination of integral cuts.)

  1. (Cuts and 1 **metrics.)** A metric **d** over a set V is said to be an 1 metric if the points can be mapped to points in Rk^ for some k, such that the distance between any two points according to d is the ` 1 distance between their mappings: d (x, y) =

i |xi^ −^ yi|. Also, a linear combination over cuts {αS }S⊂V defines the following metric μα (verify that this is indeed a metric): μα(x, y) =

S⊂V :|S∩{x,y}|=

αS ∀x 6 = y ∈ V

(Note that

S⊂V αS^ is not necessarily equal to^1 .) In this problem you will show that the above two classes of metrics—` 1 metrics and linear combinations of cuts—are in fact equivalent.

(a) Prove that any metric defined by a linear combination of cuts is an 1 metric. (b) Prove that any 1 metric can be expressed as a linear combination of cuts. (Hint: Prove this statement for a unit-dimensional ` 1 metric first, that is, k = 1_. Then extend it to multiple dimensions.)_

  1. (Prize-collecting Steiner tree.) The prize-collecting Steiner tree problem (PCST) is a variant of Steiner tree in which there are prizes πv on nodes and costs ce on edges, and a special node r called the root. The goal is to construct a Steiner tree containing r that minimizes the cost of the edges in the tree plus the value of the nodes not in the tree.

(a) Give an LP relaxation for this problem using xe as an indicator of the extent to which an edge is included in the solution, and yv as an indicator of the extent to which a node is covered. (It is okay to have an exponential number of constraints, as for the Steiner forest LP we studied in class.) (b) Write the dual of the above LP. (c) Give a primal dual algorithm for this problem based on the one for Steiner tree (forest). (Don’t forget the pruning step!)

(d) Prove that your algorithm achieves a 2-approximation. (If you cannot get the 2, try to get a slightly larger constant factor.) (Hint: Handle the component of the tree connected to the root separately from the other components. In each case, charge the cost of your solution to the value of the dual for that component.)

  1. (Minimum-Cut linear arrangement.)

In the minimum-cut linear arrangement problem, we are given an unweighted graph G = (V, E). Our goal is to find a one-to-one map from the n vertices in V to integers from 1 to n, such that the largest of the cuts C 1 , · · · , Cn− 1 is minimized, where the cut Ci is defined by the set of i nodes mapped to integers 1 through i. For example, the picture below shows a linear arrangement with value 3. Give a poly-log approximation for this problem. (Hint: Use the Sparsest Cut or Balanced Cut algorithm from class.)

Graph Linear Arrangement

Cut C 2

2

3 5

1 4

1 2 3 4 5