IP Cutting Planes10082014
-
Upload
amanda-green -
Category
Documents
-
view
215 -
download
0
Transcript of IP Cutting Planes10082014
-
8/9/2019 IP Cutting Planes10082014
1/26
Using Cutting Planes
Example. Minimize x + 10ysubject to x, y are in P
x, y integer
P
x
y
Optimumfractional
(i.e. infeasible)solution
Optimum(integer)solution
1IEOR160-2014
-
8/9/2019 IP Cutting Planes10082014
2/26
Using Cutting Planes
Example. Minimize x + 10ysubject to x, y are in P
x, y integer
P
x
yIdea: add
constraints thateliminate fractionalsolutions to the LPwithout eliminating
any integersolutions.
add y x 1
add y 1
These constraintswere obtained byinspection. There areformal techniques.2IEOR160-2014
-
8/9/2019 IP Cutting Planes10082014
3/26
Using Cutting Planes
Example. Minimize x + 10ysubject to x, y are in P
x, y integer
x
y
Optimum(integer)solution
If we add exactlythe rightinequalities, thenevery corner pointof the LP will beinteger, and the IPcan be solved by
solving the LP
PWe call thisminimal LP, the
convex hull of the IPsolutions.
For large problems,these constraints
are hard to find. 3IEOR160-2014
-
8/9/2019 IP Cutting Planes10082014
4/26
IEOR160-20144
More on adding constraints
The tightest possible constraints are veryuseful, and are called facets .Suppose that we are maximizing, and z LP is the opt for the LP relaxation, and z IP is
the opt for the IP. Then z IP zLPIdeally, we want z IP to be close to z LP . Thisis GREAT for branch and bound.
Adding lots of valid inequalities can bevery helpful.
It has no effect on z IP . It can reduce z LP significantly.
-
8/9/2019 IP Cutting Planes10082014
5/26
IEOR160-20145
Pure Cutting Plane Technique
Instead of parti tioning the feasible region, the(pure) cutting plane technique works with a singleLP
It adds cutting planes (valid linear programminginequalit ies) to this LP iteratively.
At each iteration the feasible region is successivelyreduced until an integer optimal is found by solvingthe LP.
In practice, it is also used as part of branch andbound. The essential idea is finding valid cuts orinequalities.
-
8/9/2019 IP Cutting Planes10082014
6/26
-
8/9/2019 IP Cutting Planes10082014
7/26IEOR160-2014
7
Problem Specific
The capital budgeting (knapsack) problem
maximize 16x 1 + 22x 2 + 12x 3 + 8x 4 +11x 5 + 19x 6
subject to 5x 1 + 7x 2 + 4x 3 + 3x 4 +4x 5 + 6x 6 14
x j binary for j = 1 to 6
-
8/9/2019 IP Cutting Planes10082014
8/26IEOR160-2014
8
The LP Relaxation
The capital budgeting (knapsack) problem
maximize 16x 1 + 22x 2 + 12x 3 + 8x 4 +11x 5 + 19x 6
subject to 5x 1 + 7x 2 + 4x 3 + 3x 4 +4x 5 + 6x 6 14
0 x j 1 for j = 1 to 6
The optimal solution: x 1 = 1, x 2 = 3/7, x 3 = 0x
4 = 0, x
5 = 0, x
6 = 1
We say that a subset S is a cover if the sum of its weights ismore than 14 (more than the budget)
What are some covers of this capital budgeting problem?
-
8/9/2019 IP Cutting Planes10082014
9/26IEOR160-2014
9
Getting constraints from covers
5x 1 + 7x 2 + 4x 3 + 3x 4 +4x 5 + 6x 6 14
Consider the cover S = {1, 2, 3}
We have the constraint 5x 1 + 7x 2 + 4x 3 14We can obtain the constraint x1 + x 2 + x 3 2. This is a cover constraint This constraint is stronger It does not eliminate any IP solutions, but it cuts off
fractional LP solutions.
In general, for each cover S,we obtain the constraint
j S x
j |S| - 1
-
8/9/2019 IP Cutting Planes10082014
10/26IEOR160-2014
10
The LP Relaxation
The capital budgeting (knapsack) problem
maximize 16x 1 + 22x 2 + 12x 3 + 8x 4 +11x 5 + 19x 6
subject to 5x 1 + 7x 2 + 4x 3 + 3x 4 +4x 5 + 6x 6 14
0 x j 1 for j = 1 to 6
5x 1 + 7x 2 + 6x 6 14
The optimal solution: x 1 = 1, x 2 = 3/7, x 3 = 0x4 = 0, x 5 = 0, x 6 = 1 z = 44 3/7
x1 + x 2 + x 6 2
A violated cover is {1, 2, 6}
Our approach: Given the linear solution, try to find aviolated cover constraint.
-
8/9/2019 IP Cutting Planes10082014
11/26
-
8/9/2019 IP Cutting Planes10082014
12/26
-
8/9/2019 IP Cutting Planes10082014
13/26
-
8/9/2019 IP Cutting Planes10082014
14/26
IEOR160-201414
After three cutsmaximize 16x 1 + 22x 2 + 12x 3 + 8x 4 +11x 5 + 19x 6
subject to 5x 1 + 7x 2 + 4x 3 + 3x 4 +4x 5 + 6x 6 14
x1 + x 2 + x 6
2x2 + x 3 + x 6 2
x1 + x 2 + x 3 + x 6 2
0 x j 1 for j = 1 to 6
Note: the new cuts dominates the other cuts.
-
8/9/2019 IP Cutting Planes10082014
15/26
IEOR160-201415
We eliminate the redundant constraints
maximize 16x 1 + 22x 2 + 12x 3 + 8x 4 +11x 5 + 19x 6
subject to 5x 1 + 7x 2 + 4x 3 + 3x 4 +4x 5 + 6x 6 14
x1 + x 2 + x 3 + x 6
20 x j 1 for j = 1 to 6
The optimal solution: x 1 = 0, x 2 = 1, x 3 = 0, x 4 = 0,x5 = 1/4, x 6 = 1 z = 43 3/4
So, z* 43
-
8/9/2019 IP Cutting Planes10082014
16/26
IEOR160-201416
Summary for knapsack problemWe could find some simple valid inequalit ies that
showed that z* 43. This is the optimal objectivevalue.
It took 3 cuts Had we been smarter it would have taken 1 cut
We had a simple approach for finding cuts. This does not find all of the cuts.
Recall, it took 14 nodes of a branch and bound tree
In fact, researchers have found cutting planetechniques to be necessary to solve large integerprograms (usually as a way of getting better bounds.)
-
8/9/2019 IP Cutting Planes10082014
17/26
IEOR160-201417
Traveling Salesman Problem (TSP)
What is a minimum length tour that visits each point?
-
8/9/2019 IP Cutting Planes10082014
18/26
IEOR160-201418
Comments on the TSP
Very well studied problem
Its often the problem for testing out newalgorithmic ideas
NP-complete (it is intrinsically difficult in sometechnical sense)
Large instances have been solved optimally (5000cities and larger)
Very large instances have been solved
approximately (10 million cities to within a coupleof percent of optimum.)
We will formulate it by adding constraints that looklike cuts
-
8/9/2019 IP Cutting Planes10082014
19/26
IEOR160-201419
The TSP as an IP, almost
Let x e = 1 if arc e is in the tourx e = 0 otherwise
Let A(i) = arcs incident to node i
Minimize e c e x e
subject to e A(i) x e = 2
x e is binary
Are theseconstraintsenough?
-
8/9/2019 IP Cutting Planes10082014
20/26
IEOR160-201420
Subtour : a cycle that passesthrough a strict subset of cities
7
2
Fact: Any integersolution withexactly two arcsincident to everynode is the unionof cycles.Why?
9
3
4
Improved IP/LP: for each possible subtour, add a constraintsthat makes the subtour infeasible for the IP. These are calledsubtour breaking constraints.
-
8/9/2019 IP Cutting Planes10082014
21/26
IEOR160-201421
A subtour breaking constraintLet S be any proper subset ofnodes, e.g., S = {2, 3, 4, 7, 9}.
This ensures that the set S will not have a subtour goingthrough all five nodes.
7
2
9
3
4
2. A tour for the entirenetwork has at most |S| - 1
arcs with two endpoints in S.
Observations:
1. A subtour that includes allnodes of S has |S| arcs
1
,
ij
i S j S
x S
-
8/9/2019 IP Cutting Planes10082014
22/26
IEOR160-201422
A traveling salesman tour
There are at most 4 arcs of any tour incident to 2, 3, 4, 7, 9
7
2
9
34
-
8/9/2019 IP Cutting Planes10082014
23/26
-
8/9/2019 IP Cutting Planes10082014
24/26
IEOR160-201424
More on the TSP
The IP formulation is a very good formulationgood in the following sense: for practical problemthe LP bound is usually 1% to 2% from theoptimal TSP tour length. (The LP bound is close.)
One can add even more complex constraints toget a better LP formulation, and people do. (and ithelps)
-
8/9/2019 IP Cutting Planes10082014
25/26
-
8/9/2019 IP Cutting Planes10082014
26/26
IEOR160 2014
26
Other detailsMonday 6-8pm, Oct 13, 2014Location: Room 2040 Valley LifeSciences BuildingCan bring one (1) cheat sheet with yourname on it (cannot share)No calculator needed, or allowedNo class 10-11 Monday Oct 13.