Chapter 6: Network Optimization

1. Maximum Flows
¨ LP Model
¨ Ford-Fulkerson Algorithm

2. Minimum-Cost Network Flow Problems (MCNFP)

¨ LP Model
¨ Feasible Solution
¨ Cycle-Canceling Algorithm

3. The Multicommodity Flow Problem (reading)

1
 1. Maximum Flows
¨ Many types of networks: transportation network, information
network, electricity network, manufacturing, distributing, planning,
positioning facilities,…
¨ Given a graph with “Source” and “Sink” nodes we want to compute:
 The maximum rate at which fluid can flow from Source to Sink
 The rate of flow through each edge of the graph
¨ Given a graph:
 Edges have flow capacities
Min-
 First value is flow capacity a cut
 Second value is current flow rate 5, 4,
 Edges behave like pipes 2 4
 Nodes are junctions of pipes s 6, t
2
8, 5,
7 5
b
 Max Flows Problem (cont.)
 Flow f(u,v) is a real-valued function
defined for every edge (u,v) in the graph
 The flow needs to satisfy the
following 3 properties:
 f(u,v) ≤ c(u,v) i.e. capacity of edge
(u,v)
 f(u,v) = - f(v,u)
 Flow coming into node v = Flow
leaving v
a
5, 2

4, 4
8, 5,
7 5
b
s
 Max Flows Problem (cont.)
 Residual Graph:
 A Graph that contains edges that can admit
more flow
 Define a Graph G(V,E’) for G(V,E)
 We define residual flow r(u,v) = c(u,v) – f(u,v)
 If r(u,v) > 0 then (u,v) is in E’
Original a Residual
Graph a 2 Graph
5, 4, 4
2 4 3
s 6, t s 4 t
2 2
1
8, 5, 5
7
7 5
b Preflow Push
b 4
Algorithm
 1.1. LP Model 1 2

¨ LP model of the Maximum-Flow Problem:

s t
Max z = x0 (1)
Subject to 3 4

(3)
(2)

(4)
0 ≤ xij ≤ cij (5)
¨ Where:
¨ x = volume of flows pass through arc (i, j)
ij
¨ x = total flow entering the sink
0
¨ For a flow to be feasible, it must have two characteristics:
¨ 0≤ flow through each arc ≤ arc capacity
¨ Flow into node i = Flow out of node i
 Example
¨ Sunco Oil wants to ship the maximum
possible amount of oil (per hour) via
pipeline from node so to node si in the
following figure. On its way from node so
to node si, oil must pass through some or
all of stations 1, 2, and 3. The various
arcs represent pipelines of different
diameters. The maximum number of
barrels of oil (millions of barrels per hour)
that can be pumped through each arc is
shown in the following table. Each
number is called an arc capacity.
Formulate an LP that can be used to
determine the maximum number of
barrels of oil per hour that can be sent
from so to si.
Solution
 1.2. Ford-Fulkerson Algorithm
¨ Questions:
¨ Q1: Given a feasible flow, how can we tell if it is an optimal flow (that is,
maximizes x0)?
¨ Q2: If a feasible flow is nonoptimal, how can we modify the flow to obtain a
new feasible flow that has a larger flow from the source to the sink?
¨ Properties:
¨ P1: The flow through arc (i, j) is below the capacity of arc (i, j). In this case,
the flow through arc (i, j) can be increased. For this reason, we let I
represent the set of arcs with this property.
¨ P2: The flow in arc (i, j) is positive. In this case, the flow through arc (i, j)
can be reduced. For this reason, we let R be the set of arcs with this
property.
 Example: Set I and R

Capacity
The Ford-Fulkerson Labeling Procedure
¨ Purpose: to modify a feasible flow in an effort to increase the flow from the source to
the sink.
¨ Step 1: Label the source.
¨ Step 2: Label nodes and arcs (except for arc a0) according to the following rules:
1. If node x is labeled, then node y is unlabeled and arc (x, y) is a member of I; then
label node y and arc (x, y). In this case, arc (x, y) is called a forward arc.
2. If node y is unlabeled, node x is labeled and arc (y, x) is a member of R; label
node y and arc (y, x). In this case, (y, x) is called a backward arc.
¨ Step 3: Continue this labeling process until the sink has been labeled or until no more
vertices can be labeled. If the sink cannot be labeled current flow is optimal.
¨ Output: a chain of labeled arcs (call it C) leading from the source to the sink 
adjusting the flow of the arcs in C, we can maintain a feasible flow and increase the
total flow from source to sink.
¨ Case 1 C consists entirely of forward arcs 
¨ Case 2 C contains both forward and backward arcs  To increase the flow from
source to sink (while maintaining a feasible flow), decrease the flow in all of C’s
backward arcs by min (k1, k2) and increase the flow in all of C’s forward arcs by
min(k1, k2).
Capacity
Example for Case 1
Example for Case 2
 The Ford-Fulkerson Algorithm
¨ Step 1 Find a feasible flow (setting each arc’s flow
to zero will do).
¨ Step 2 Using the labeling procedure, try to label
the sink. If the sink cannot be labeled, then the
current feasible flow is a maximum flow; if the sink
is labeled, then go on to step 3.
¨ Step 3 Using the method previously described,
adjust the feasible flow and increase the flow from
the source to the sink. Return to step 2.
Example
Practice
Solution
 2. Minimum-Cost Network Flow
Problems (MCNFP)
¨ Let:
¨ xij = number of units of flow sent from node i to node j through arc
(i, j)
¨ bi = net supply (outflow or inflow) at node i
¨ cij = cost of transporting 1 unit of flow from node i to node j via arc (i,
j)
¨ Lij = lower bound on flow through arc (i, j). If there is no lower bound,
let Lij = 0
¨ Uij = upper bound on flow through arc (i, j). If there is no upper
bound, let Uij = ∞
 2.1. The LP Model of the MCNFP

(3)

(4)

¨ Constraint (3) is the flow balance constraint

¨ The transportation, assignment, transshipment, shortest-path,
maximum flow, and CPM problems are all special cases of the
minimum-cost network flow problem (MCNFP).

18
 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.)

19
 Example: The Traffic Assignment
¨ Each hour, an average of 900 cars enter the network in the following figure
at node 1 and seek to travel to node 6. The time it takes a car to traverse
each arc is shown in the following table. In the figure, the number above
each arc is the maximum number of cars that can pass by any point on the
arc during a one-hour period. Formulate an MCNFP that minimizes the
total time required for all cars to travel from node 1 to node 6.
 Solution

21
Example: 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.

22
A Network Representation
4
Demands
190 1

6 400

310 2
7 180

Retailers
100 3
5
Plants
Warehouses

23
 2.2. Finding a feasible solution
One can find a feasible solution when all lower
bounds are 0 by solving a single max flow
problem.
1. If b(j) > 0, create arc (s, j) with usj = b(j),
2. If b(j) < 0, create arc (j, t) with ujt = - b(j)

2 -4
2
2 4 4
5 -4
5 4
s 1 6 t
2
3 5
3
3 -2 24
 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. Add nodes s and t with b(s) = b(t) = 0
2. If b(j) > 0, create arc (j, t) with ujt = b(j) and cjt = 0
3. 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 -4
2
2 4 4
5 -4
5 4
t 1 6 s
2
3 5
3
3 -2 cost = - M 25
 Finding an artificial feasible solution
2 -4
2
2 4 4
5 -4
5 4
t 1 6 s
2
3 5
3
3 -2

Initial solution.
• If b(j) > 0, then xjt = b(j) (supplies are satisfied)
• If b(j) < 0, then xsj = - b(j) (demands are satisfied)
• xts = ∑b(j)>0 b(j) (flow into t = flow out of t)
• xij = 0 otherwise

In an optimal feasible solution, xts = 0. There is no flow in any

26
Find the shortest path from node 1 to
node 6
0 0
2 4 4
2 2
1 2 3 6 b(6) = -1
b(1) = 1 1
4 2

3 3 5
0 0

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?)
27
Find the Maximum Flow from s to t
1 8,7
b(i) = 0 for all i;
10, 8
add arc (t,s) with a
s 1,1 t cost of -1 and large
6, 5 10,6
capacity.
2
The cost of every
13 other arc is 0.

The optimal solution in the corresponding minimum cost

flow problem will send as much flow in (t,s) as possible.
28
 The Residual Network G(x)
uij 8
i j 1 2
cij $10
Suppose x12 = 3

uij - xij 5
cij $10
i j 1 2
xij 3
-cij -$10
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.
29
Negative cost cycles & augmentations
-4
2 4
7 Note: each arc of G(x) has
-7 a cost and a capacity.
1 2
5 -2
-5 -1 Typically, we will only show
one of them.
6
2
3 5
-2 A negative cost cycle refers
to a directed cycle in G(x)
A residual network G(x) whose total cost is
and its arc costs. negative, e.g., 1-3-2-1 and
3-5-4-3

30
Negative cost cycles & augmentations
10
2 4
5
To augment around a cycle
25 10
1 15 10 C is to send δ units of flow
20
10 around C, where
δ = min {rij : (i, j) ∈ C}
20
20
3 5
5 The cycle 1-3-2-1 had
negative cost. Its capacity
The capacities of the is 15.
residual network G(x).

31
 The impact of an augmentation
10
2 4 10
2 4
5
20
25 10
1 10 10
15 10 20 1
10 25 20
10
15
20
20 5
3 5 20
3 5
5
5

The capacities of G(x) The capacities after the

before the augmentation. augmentation.

32
 2.3. Cycle-Canceling Algorithm
¨ The algorithm first establishes a feasible flow x in the network.
¨ Then it iteratively finds negative cost-directed cycles in the
residual network and augments flows on these cycles.
¨ The algorithm terminates when the residual network contains no
negative cost-directed cycle.
 A Negative Cycle
¨ A negative cycle is a directed cycle such that the sum of its
edge weights is negative
Example

(i , j) (j , i)
Cost
Residual
capacity
3. The Multicommodity Flow Problem
¨ Network G = (N, A)
¨ K origin-destination pairs of nodes
(s1, t1), (s2, t2), …, (sK, tK)
¨ dk = amount of flow that must be sent from s k to tk.
¨ uij = capacity on (i,j) shared by all commodities
¨ cijk
= cost of sending 1 unit of commodity k in (i,j)
k
x
¨ ij = flow of commodity k in (i,j)
36
A Linear Multicommodity Flow Problem

5 units $5 5 units
good 1 1 4 good 1

$1 $1
$1
2 5
u25 = 5
$1
$1
2 units $6 2 units
good 2 3 6 good 2

37
A Linear Multicommodity Flow Problem

5 units $5 5 units
good 1 1 4 good 1

$1 $1 2 2 2
x32  x25  x56 2
$1
1 1 1
2 5 x12  x25  x54 3
u25 = 5 1
x14 2
$1
$1
2 units $6 2 units
good 2 3 6 good 2

38
 The Multicommodity Flow LP
Min   ij x ij
c k k

( i , j )A k
 d k if i  sk Supply/
 demand
j ij j ji   d k if i  t k
x k
 x k

constraints
 0 otherwise

Bundle
 ij uij
x
k
k
for all (i , j )  A
constraints

x ijk 0 (i , j )  A, k  K

39
 Assumptions
¨ Homogeneous goods. Each unit flow of commodity k
on (i, j) uses up one unit of capacity on (i,j).

¨ No congestion. Cost is linear in the flow on (i, j) until

capacity is totally used up.

¨ Fractional flows. Flows are permitted to be fractional.

¨ OD pairs. Usually a commodity has a single origin and

single destination. 40
 Application areas
Type
Nodes Arcs Flow
of Network

Communic. O-D pairs Transmission message

Networks for messages lines routing

Computer storage dev. Transmission data,

Networks or computers lines messages

Railway yard and

Tracks Trains
Networks junction pts.

highways
Distribution plants trucks,
railway tracks
Networks warehouses,... trains, etc
etc.
41
 On Fractional Flows

¨ In general, linear multicommodity flow problems

have fractional flows, even if all data is integral.

¨ The integer multicommodity flow problem is

difficult to solve to optimality.

42
 A Fractional Multicommodity Flow
uij = 1 for all arcs 1 unit of flow must be sent
from si to ti for i = 1, 2, 3.
cij = 0 except as listed.
s1 t3
$2 $2
t1 1 s3

2 3

s2 t2
$2 43
 A Fractional Multicommodity Flow
uij = 1 for all arcs 1 unit of flow must be sent
from si to ti for i = 1, 2, 3.
cij = 0 except as listed.
s1 t3
$2
$2
t1 1 s3

Optimal solution:
2 3
send ½ unit of flow
in each of these
15 arcs.
s2 t2 Total cost = $3.
$2 44
 Decomposition Based Approaches
¨ Price directed decomposition.
¨ Focus on prices or tolls on the arcs. Then solve the problem
while ignoring the capacities on arcs.
¨ Resource directive decomposition.
¨ Allocate flow capacity among commodities and solve
¨ Simplex based approaches
¨ Try to speed up the simplex method by exploiting the structure of
the MCF problem.

45
 A Formulation Without OD Pairs

Minimize 
1k K
ck x k (1a)

subject to 
1k K
x ijk uij for all (i , j )  A (1b)

Nx k b k for k 1, 2, ..., k (1c)

for all (i , j )  A
0  x u k k
(1d)
ij ij
for k 1, 2, ..., k

46
Optimality Conditions: Partial Dualization
The multicommodity flow x = (xk) is an optimal
multicommodity flow for (1) if there exists non-negative
prices w = (wij) on the arcs so that the following is true

1. If w ij  0, then  k ij uij
x k

2. The flow xk is optimal for the k-th commodity

if ck is replaced by cw,k, where
cijw ,k cijk  w ij

Note: xk is optimal for the k-th commodity if there is no

negative cost cycle in the kth residual network.
47
A Linear Multicommodity Flow Problem
5 units $5 5 units
good 1 1 4 good 1

$1 $1 2 2 2
x32  x25  x56 2
$1
1 1 1
2 5 x12  x25  x54 3
u25 = 5 1
x14 2
$1
$1
2 units $6 2 units
good 2 3 6 good 2

Create the residual

Set w2,5 = 2
networks
48
The Residual Network for Commodity 1
$5
1 4
-$5
$1 -$1 -$1 $1
$3
2 5
$-3
$1
$1
$6
3 6

Set w2,5 = $2 There is no negative cost cycle.

49
The Residual Network for Commodity 2
$5
1 4
$1 $1
$3
2 5
$-3
-$1 -$1 $1
$1
$6
3 6

Set w2,5 = $2 There is no negative cost cycle.

50