15.082J and 6.855J and ESD.

78J October 26, 2010

Introduction to Minimum Cost Flows

Overview of lecture

Quick review of min cost flow problem An application of min cost flows The residual network, again The cycle canceling algorithm for solving the min cost flow problem Reduced costs and optimality conditions

The Minimum Cost Flow Problem

uij = capacity of arc (i,j). cij = unit cost of shipping flow from node i to node j on (i,j). xij = amount shipped on arc (i,j)

Minimize
j xij -

(i,j)A cijxij
j xji = bi for all i N.

and 0 xij uij for all (i,j) A.

Find the shortest path from node 1 to node 6

0 2 0 4 4

2
b(1) = 1 1 4 3 0 3 5 0 1 2 3

2
6 b(6) = -1 2

The optimal flow is to send one unit of flow along 1-2-5-6. This transformation works so long as there are no negative cost cycles in G. (What if there are negative cost cycles?)

Find the Maximum Flow from s to t

b(i) = 0 for all i;
t
10,6

10, 8 s 6, 5

1 1,1 2 13

8,7

add arc (t,s) with a cost of -1 and large capacity. The cost of every other arc is 0.

The optimal solution in the corresponding minimum cost flow problem will send as much flow in (t,s) as possible.
5

Transshipment Problems
Plants with given production capabilities for a product. One can ship directly from the plants to retailers, or from plants to warehouses, and then from warehouses to retailers. There is a given demand for each retailer. Costs of shipment are given. What is the minimum cost method for satisfying demands?
6

A Network Representation
4

Demands

190 1 6
310 2 7 180 400

Retailers 100 3 5 Plants Warehouses

The Caterer Problem

Demand for di napkins on day i for i = 1 to 7 (so, j [1..7]). Cost of new napkins: a cents each,

2-day laundry: b cents per napkin

1-day laundry: c cents per napkin. Minimize the cost of meeting demand.

Assumptions: there is no initial inventory of napkins at the end, no clean napkins remain

Application to airplane maintenance.

An example with a feasible solution

new napkins: 1 day cleaning $10 $2 Demand: M-F 100 napkins, Sa-Su 125 napkins 200

2-day cleaning
0

$1

new napkins 100

2 3

100
1

clean napkins
4 5 25 6 7

100
1

125
50 6

100 2

100 3

100 4

100 75 5

75

125
n+1
9

dirty napkins

Nodes and Arcs

Node j : Clean napkins on day j Node j : Dirty napkins on day j Clean napkins on day 5 come from purchases (0, 5) leftover clean napkins (4, 5)
0

Node 0: Where napkins originate

Node n+1: Where napkins go to stay dirty

2-day laundry (3, 5)

Clean napkins

0
4

a
5

1-day laundry (4, 5) 0

6

dirty napkins

flow = 5 d5

Clean napkins on day 5 go to

being used (5, 5) (5, 6)10 being stored

n+1

Nodes and Arcs

Clean napkins
5 6

Dirty napkins

Flow = d5
5

0
n+1

Dirty napkins on day 5 go to Dirty napkins on day 5 come from being used on day 5 (5, 5)

never cleaned
1-day cleaning 2- day cleaning

(5, n+1)
(5, 6) (5, 7)
11

The network for the caterer problem

0

n+1

12

Find a minimum cost circulation such that the flow on (j, j) = dj on arcs (j, j) for j = 1 to n. Lower bound = upper bound = dj

Arc (n+1, 0) : each purchased napkin ends up dirty.

The minimum cost flow problem

Simplifying assumptions Finding a feasible flow The residual network The cycle canceling algorithm Reduced costs and optimality conditions

13

Some Assumptions
1. All data is integral. (Needed for some proofs, and some running time analysis). 2. The network is directed and connected

3. i=1 to n b(i) = 0.
(Otherwise, there cannot be a feasible solution.)

14

Finding a feasible solution

One can find a feasible solution when all lower bounds are 0 by solving a single max flow problem.
1. 2.

If b(j) > 0, create arc (s, j) with usj = b(j), If b(j) < 0, create arc (j, t) with ujt = - b(j)

2 s

2 2

-4 4

5 1
3 3 5

-4 6

4 4 2
15

-2

Finding an artificial feasible solution

One can start with an artificial feasible solution with large cost. The flow in these arcs will be 0 at the end.
1. 2. 3.

Add nodes s and t with b(s) = b(t) = 0 If b(j) > 0, create arc (j, t) with ujt = b(j) and cjt = 0 If b(j) < 0, create arc (s, j) with usj = - b(j) and cjt = 0

4.

Add an arc (t, s) with uts = mumax and cts = mcmax

2 2

-4 4

5 1 3 3 3 5 -2

-4

4 4 2 cost = - M
16

Finding an artificial feasible solution

2 2 2 -4 4 -4 4 4 2 s

5 1 3 3 5

3
Initial solution. If b(j) > 0, then xjt =
b(j) If b(j) < 0, then xsj = - b(j) xts = b(j)>0 b(j) xij = 0 otherwise

-2
(supplies are satisfied) (demands are satisfied) (flow into t = flow out of t)

In an optimal feasible solution, xts = 0. There is no flow in any of the artificial arcs)

17

Mental Break
Why is the word ring part of boxing ring? They used to be round.

How many dimples are in a regulation golf ball? 336

Was tug of war ever an Olympic event? Yes. Between 1900 and 1920.
18

Mental Break
Where did karate originate?

In India.
What do the following nicknames for sports teams have in common: the Miami Heat, the Minnesota Wild, the Utah Jazz, the Boston Red Sox, and the Chicago White Sox ? None of them ends in the letter s.

In which country is kite flying a professional sport? Thailand.

19

The Residual Network G(x)

uij cij

$10
Suppose x12 = 3

uij - xij cij

i j 1

5 $10
2 3 -$10

xij -cij

Reducing the flow in (i, j) by 1 is equivalent to sending one unit of flow from j to i. It reduces the cost by cij.
20

Negative cost cycles and augmentations

-4 2 7 1 -7 -5 6 3 2 -2 5 5 2 -2 -1 4

Note: each arc of G(x) has a cost and a capacity.

Typically, we will only show one of them. A negative cost cycle refers to a directed cycle in G(x) whose total cost is negative, e.g., 1-3-2-1 and 3-5-4-3
21

A residual network G(x) and its arc costs.

Negative cost cycles and augmentations

10 2 5 1 25 15 10 10 10 20 20 3 5 5 20 4

To augment around a cycle C is to send units of flow around C, where = min {rij : (i, j) C}
The cycle 1-3-2-1 had negative cost. Its capacity is 15.

The capacities of the residual network G(x).

22

The impact of an augmentation

10 2 5 1 25 15 10 10 10 20 4 20 2 10 4 10 25 15 20 3 5 5 10 20

10

20

5
20 3 5 5

The capacities of G(x) before the augmentation.

The capacities after the augmentation.

23

Negative Cycle Algorithm

(also known as the cycle canceling algorithm)
Algorithm NEGATIVE CYCLE; establish a feasible flow x in the network; while G(x) contains a negative cost cycle do use some algorithm to identify a negative cost cycle C in G(x); let : = min { rij : (i, j) C }; augment units of flow in cycle C;

update G(x);
Negative Cycle Algorithm
24

More on the Negative Cycle Algorithm

Suppose that all supplies/demands are integral, and capacities are integral. Then the negative cycle algorithm maintains an integral solution at each iteration.

25

Finiteness
Theorem. The Negative Cycle Algorithm is finite if all data are finite and integral. Proof. By flow decomposition, we can express the min cost flow as the sum of n+m paths and cycles. Each path and cycle has a cost bounded by nC, where C = max (|cij| : (i,j) A). The cost of the flow is at most (nC)(n+m)U, where U is the largest capacity. At each iteration of cycle canceling, the cost improves by at least one.
26

Optimality
Theorem. The Negative Cycle Algorithm terminates with an optimal flow. Proof. Consider the final residual network G(x*). Suppose that x* is not optimal. Equivalently, the flow y = 0 is not optimal for the circulation problem in G(x*). Let y* be a minimum cost circulation in G(x*). Then y* can be decomposed into flows around cycles. At least one of these cycles (say C) has negative cost. But this contradicts that the algorithm terminated.
27

Reduced costs and optimality conditions

Reduced costs

recall replacing cij i + j for the shortest path

problem. The same transformation is very useful

for min cost flow algorithms.

Optimality conditions

Most iterative optimization algorithms stop when optimality conditions are satisfied. We describe optimality conditions for the min cost flow problem.
28

Reduced Costs in G(x)

Let i denote the node potential for node i.
cij cij i j

For unit of flow out of node i, subtract i from the cost. For unit of flow into node j, add j to the cost. $3 1 + 2

1 -$5

$3

2 $1

-$5 3 + 1 3

$1 2 + 3

29

More on Reduced Costs

Lemma. The reduced cost of a cycle is the cost of a cycle. Proof:

c (C ) ( i , j )C cij i j ( i , j )C cij c(C )

$3 2 $1 3 1
$3 1 + 2

1 -$5

2
$1 2 + 3

-$5 3 + 1

Corollary. Optimizing with respect to reduced costs is equivalent to optimizing with respect to the original costs.
30

Optimality Conditions
Theorem. A flow x* is optimal if and only if there is a vector * so that c*ij 0 for all (i, j) G(x*). Proof. We already know that x* is optimal if and only if there is no negative cost cycle in G(x*). It remains to show that there is no negative cycle in G(x*) if * so that c*ij 0 for all (i, j) G(x*). Suppose first that there is such a vector *. Then the reduced cost of every cycle in G(x*) must be non-negative
31

Optimality Conditions
Proof. Continued. Suppose now that there is no negative cycle cycle in G(x*). Let d(i) be the shortest path length from node 1 to node i in G(x*) using original costs. (Assume that such a path exists). Then for all (i, j) in G(x*), d(j) d(i) + cij (i, j) G(x*) Let *i = - d(i) for all i. Then c*ij = cij + d(i) d(j) 0 for all (i, j) G(x*). QED
32

Optimality conditions for the original network

If c*ij 0 for all (i, j) G(x*), what is true about the original network?
i

ui
j c*ij

Optimality conditions.
1.

uij - xij c*ij i xij j

2.

If x*ij = 0, then (j, i) G(x*) and c*ij 0.

If x*ij = uij, then (i, j) G(x*) and c*ij 0 (so that c*ji 0)
If 0 < x*ij < uij, then (i, j) G(x*) and (j, i) G(x*) and c*ij = 0.

-c*ij

3.

33

Optimality conditions again

Optimality conditions 1.
1.

If x*ij = 0, then c*ij 0.

2.
3.

If x*ij = uij, then c*ij 0.

If 0 < x*ij < uij, then c*ij = 0. Opt conditions 1 are equivalent to optimality conditions 2.

Optimality conditions 2.
1.

If c*ij > 0 then x*ij = 0,

2.
3.

If c*ij < 0 then x*ij = uij,

If c*ij = 0, then 0 x*ij uij.
34

Review of Cycle Canceling

Given a flow x, we look for negative cost cycles in G(x).

If we find a negative cost cycle, we sent flow around the cycle

If we dont find a negative cost cycle, we establish optimality.

It is a very generic algorithm for solving minimum cost flows. Key subroutine: finding a negative cost cycle. It can be done in different ways.
35

How to Find a Negative Cycle

POSSIBILITY 1. Use a shortest path algorithm to determine a negative cost cycle.

POSSIBILITY 2. Find the most negative cost cycle.

POSSIBILITY 3. Augment along the cycle that minimizes COST(C)/|C|. (The cost divided by the number of arcs.)

36

Summary

Some applications of the minimum cost flow problem Cycle Canceling Algorithm

Integrality Property for Minimum Cost Flows

Optimality Conditions

37

MITOpenCourseWare http://ocw.mit.edu

15.082J / 6.855J / ESD.78J Network Optimization

Fall 2010

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.