reconnect ‘04 a few topics in polyhedral combinatorics
DESCRIPTION
Reconnect ‘04 A Few Topics in Polyhedral Combinatorics. Cynthia Phillips, Sandia National Laboratories. Strengthen Linear Program with Cutting Planes. Make LP polytope closer to integer polytope Use families of constraints too large to explicitly list - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/1.jpg)
Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company,for the United States Department of Energy under contract DE-AC04-94AL85000.
Reconnect ‘04A Few Topics in Polyhedral Combinatorics
Cynthia Phillips, Sandia National Laboratories
![Page 2: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/2.jpg)
Slide 2
Strengthen Linear Program with Cutting Planes
Original LP
Feasible region
LP optimal solution
Cutting plane
(valid inequality)
Integer optimal
• Make LP polytope closer to integer polytope• Use families of constraints too large to explicitly list
– Exponential, pseudopolynomial, polynomial (n4, n5)
![Page 3: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/3.jpg)
Slide 3
Separation
• Consider a family of cutting planes
– Abbreviate as (ai,bi)
• A separation algorithm takes this family and an x* and in polynomial time
either
– Returns a member of the family (ai,bi) such that
• x* violates (ai,bi)
– Says (truthfully) that x* violates no member of the family
• If we iteratively add the cuts returned by the separation algorithm, in
polynomial time, we will have an optimal LP solution that satisfies the
whole family (Ellipsoid algorithm)
€
aiT x ≤ bi
€
aiT x* > bi
![Page 4: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/4.jpg)
Slide 4
Example: Traveling Salesman Problem
• Input: a set of n cities , , distance dij between cities i and j
– Can travel between any pair (automatic with the triangle inequality)• Goal: Visit each city exactly once so as to minimize the total distance
Variable xij = 1 if edge (i,j) in the tour, 0 otherwise
Undirected formulation (xij has i < j)
Enforce all nodes have degree 2 in the tour
€
min c ij x ij
i< j
∑
st x ji
i> j
∑ + x ij
i< j
∑ = 2 ∀ j
€
1K n ≡ N
![Page 5: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/5.jpg)
Slide 5
Subtours
• Degree constraints aren’t sufficient for an IP formulation because we can
have disconnected cycles
![Page 6: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/6.jpg)
Slide 6
Eliminate Subtours
• Force 2 edges to cross every (nontrivial) cut
![Page 7: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/7.jpg)
Slide 7
Subtour Elimination Constraints
• If we give each edge e weight xe*, then the separation algorithm is looking
for a cut of capacity less than 2• Just run a standard minimum cut code
€
x ij
i∈S, j∉S orj∈S,i∉S
∑ + x ji
i∈S, j∉S orj∈S,i∉S
∑ ≥ 2 ∀S ⊆N, 2 ≤ S ≤ n − 2
![Page 8: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/8.jpg)
Slide 8
The Power of Separation
For 300 cities, there are over 1090 subtour elimination constraints!
But we can enforce them all for instances with thousands of cities.
Adding classes of cutting planes can provably reduce the integrality gap
(ratio between best IP solution and best LP solution)
![Page 9: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/9.jpg)
Slide 9
Compact Formulations
• When the separation algorithm is itself an LP we can sometimes represent
the entire separation process as a single LP (with polynomially more
variables)
![Page 10: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/10.jpg)
Slide 10
Valid Inequality
An inequality is valid for a polytope if it contains the whole polytope
feasible
€
aT x = b
€
aT x ≥ b
![Page 11: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/11.jpg)
Slide 11
Facet
Let be a valid inequality for polyhedron P
Then is a face of the polyhedron
If , then F supports P
If F is exactly one dimension smaller than P, then it is a facet
Families of facet-defining inequalities are optimal in a sense
feasible
€
aT x = b
€
aT x ≤ b
€
F = x ∈ P | aT x = b{ }
€
F ≠ ∅
![Page 12: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/12.jpg)
Slide 12
Convex Combinations
• A point x is a convex combination of two others x1 and x2 if
(componentwise)
€
x = λ1x1 + λ 2x2, with
λ1 + λ 2 =1
λ1,λ 2 ≥ 0
x1x
x2
![Page 13: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/13.jpg)
Slide 13
Extreme Points
Another definition of an extreme point (corner of a polyhedron):
is an extreme point if and only if there are no
such that x is a convex combination of x1 and x2
€
x ∈ P
€
x1,x2 ∈ P
x
x2
x1
![Page 14: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/14.jpg)
Slide 14
Convex Decomposition
x* = optimal solution to the LP relaxation
Find feasible integer solutions
Convex combination:
• Implies one of the Si has cost
at most * LP optimal
(one is as good as group average)
gradient
LP
* LP
Integer polytope
LP opt ≠0
€
0 ≤ λ j ≤1, λ j =1∑
€
S0,S1,K ,Sm such that λ jS j = ρx*
j
∑
![Page 15: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/15.jpg)
Slide 15
Convex Decomposition
We have
Order the Si such that
Since
Suppose
Then
Contradiction
€
, 0 ≤ λ j ≤1, λ j =1∑
€
cT S0 > ρcT x*
€
cT S0 ≤ cT S1 ≤K cT Sm
€
S0,S1,K ,Sm such that λ jS j = ρx*
j
∑
€
λ jS j = ρx*
j
∑ , then λ jcT S j = ρcT x*
j
∑
€
λ jcT S j ≥
j
∑ λ jcT S0
j
∑ = cT S0 λ i
i
∑ = cT S0 > ρcT x*
![Page 16: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/16.jpg)
Slide 16
Decomposition Precisely Defines Integrality Gap
An IP has a solution within times the LP bound if and only if x* can be
decomposed into a convex combination of feasible solutions.
Definition: A -approximation algorithm for a minimization problem
guarantees a solution no more than times the optimal solution for all
instances.
![Page 17: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/17.jpg)
Slide 17
LP-Relaxation-Based Approximation for IP
• Compute LP relaxation (lower bound).• Common technique:
– Use structural information from LP solution to find feasible IP solution (use parallelism if possible)
– Bound quality using LP bound
• Integrality gap = maxI(IP (I))/(LP(I))
• This technique cannot prove anything better than integrality gap
![Page 18: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/18.jpg)
Slide 18
Example: Vertex Cover
Find a minimum-size set of vertices such that each edge has at least one
endpoint in the set.
2
1 4
7
5
3
6
![Page 19: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/19.jpg)
Slide 19
Example: Vertex Cover
2
1 4
7
5
3
6
vi =1 if vertex i is in the VC
0 otherwise ⎧ ⎨ ⎩
min vi∑s.t. vi +vj ≥1 ∀ i, j( )∈E
vi ∈ 0,1{ }
![Page 20: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/20.jpg)
Slide 20
2-Approximation algorithm for Vertex Cover
• Solve the LP relaxation for vertex cover:
• Select all vertices i such that vi ≥ 1/2.
• This covers all edges: at least one endpoint will have value at least 1/2.• Each such vertex contributed as least 1/2 to the optimal LP solution, so
rounding to 1 at most doubles cost.
min vi∑s.t. vi +vj ≥1 ∀ i, j( )∈E
0≤vi ≤1
![Page 21: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/21.jpg)
Slide 21
Capacitated Network Design
• Each pair (vi, vj) has a demand (required connectivity) dij
– Min cut separating vi and vj is at least dij
• Choose min-cost subgraph s.t. all pairwise demands satisified• All/none decision for each edge.
12 (4)
s
1427 (6)
9 (15)2 (7)
5 (1)5 (17)
3 (10)t
2
3
58 (2)
20 (30)
2 (8)
1 (4)
Capacity ue (cost ce)
![Page 22: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/22.jpg)
Slide 22
Network Reinforcement - Communication Network
message packets take “all” paths, must capture all packets to compromise (Franklin)• Capacity = attacker cost to compromise edge• Min cut = attacker cost to eavesdrop
Pay to protect all communication at desired level.
12 (4)
s
1427 (6)
9 (15)2 (7)
5 (1)5 (17)
3 (10)t
2
3
58 (2)
20 (30)
2 (8)
1 (4)
Capacity ue (cost ce)
![Page 23: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/23.jpg)
Slide 23
Special Case - Minimum Knapsack Problem
Given: Set of objects: Object i has cost ci, value vi
Required value V
Find: minimum-cost set of objects with total value at least V
s t
v1 (c1)
v2 (c2)
vm (cm)
.
.
.
![Page 24: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/24.jpg)
Slide 24
Generalization - Capacitated Covering
• All entries of c,U,d are nonnegative.
€
min cT x
st. Ux ≥ d
0 ≤ x ≤ b
x ∈ Z +
![Page 25: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/25.jpg)
Slide 25
Definition: Bond
A bond is a minimal set of edges whose removal disconnects a pair with
positive demand. Count multiedges as 1.
Max bond of graph G, (G) is max cardinality of any bond in G
s
14
t
2
3
5
Card(Bond) = 4
![Page 26: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/26.jpg)
Slide 26
Integer Program (IP) for capacitated network design
Where d(C) is the maximum demand dij for any pair that crosses cut C
xe = 1 if edge e is selected
min cexeΣe ∈C
uexe ≥d(C) ∀ cutsetCΣe ∈C
xe ∈ 0,1
![Page 27: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/27.jpg)
Slide 27
Simple Network Reinforcement IP has Bad Integrality Gap
u=D-1 c=0
s tu=D c=1
s t
u=D-1
u=D
c=0
c=1
s tu=D c=1
IP cost = 1
xe = 1
xe = 1/D
LP cost = 1/D
Ratio OPT(IP)/OPT(LP) = D
![Page 28: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/28.jpg)
Slide 28
Effective Capacities
Can assume
C is a cut,
€
ue ≤ maxe∈C D(C)
€
D(C) = maxi, j split by C dij
i j
![Page 29: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/29.jpg)
Slide 29
Inhibiting One Form of Cheating
• New problem with remaining edges and residual Demand D - (D-1) = 1
u=D-1
u=D c=1
u=D-1
s tu=D
c=0
Demand D
u=D-1
u=D c=1
u=D-1
s tu=D c=1
c=0
ResidualDemand 1
u=1
xe = 1
![Page 30: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/30.jpg)
Slide 30
Knapsack Cover (KC) Inequalities
€
u(A) = uA < D(C)e∈A
∑
residual D(A) = D − u(A)
uA (e) = min(ue,D(A))
KC : uA(e)xe ≥ D(A)e∈C−A
∑
AC
![Page 31: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/31.jpg)
Slide 31
Knapsack Cover (KC) Cuts for General Graphs
€
min ce
e∈E
∑ xe
st uA(e)xe ≥ D(A)e∈E−A
∑ for A ⊂ E,u(A) < D
xe ∈ 0,1{ }
![Page 32: Reconnect ‘04 A Few Topics in Polyhedral Combinatorics](https://reader036.vdocuments.us/reader036/viewer/2022062314/56812c6b550346895d910122/html5/thumbnails/32.jpg)
Slide 32
New Integrality Gaps
• 2 for Knapsack (G) + 1 for general graphs
Proof: Find feasible integer solution with cost 2 (or (G) + 1) times LP
optimal