Design and Analysis of Algorithms

CS345

Lecture 20
Analysis of Ford Fulkerson algorithm
(Max-Flow Min-Cut Theorem)

1
FORD FULKERSON ALGORITHM

A simple path based algorithm

+
Residual Network

2
 Formal Description of Flow
2
𝒗 4 𝒙 =(, ) : a directed graph with
5 : source,
5 11 : sink
6 1 15
8 6 𝒕: 
𝒔 7 1
(,) = capacity of edge (,)
7
7 14 7 17

Flow :  such that

𝒖 4 𝒚 • For each edge (,)
flow leaving (,) (,)

• For each vertex

3
 Formal Description of Flow
2
𝒗 𝒙 =(, ) : a directed graph with
5 : source,
: sink
6 15
8 6 𝒕: 
𝒔 (,) = capacity of edge (,)

7 14 7 17

𝒖 4 𝒚
Max-Flow Problem:
flow leaving Given a network =(,),
and two vertices ,,
find a flow from to of maximum value.

4
 Residual network

Let =(, ) be a network

: any valid - flow.

The residual network

Vertices of = ? 𝑽
Edges of = 𝑬 ?
𝒇

For each edge (,)

– If (,) < (,),
 edge (,) with capacity = ?
Forward edge (,) - (,)
– If (,) >
 edge (,) with capacity = ?
Backward edge (,)

5
 Ford Fulkerson algorithm
Ford-Fulkerson-algo(,,)
{ ;
While (there is - path in ) do
{ Let be an - path in
Let be bottleneck capacity of ; Correctness of
the algorithm ?
For each (,) do
{ If (,) is a forward edge then
(,)  (,) + ;
Else
(,)  (,) - ;
}
} return ;
}

6
-CUTS

7
 -cut
, with
,

𝒔 𝒕

𝑨 𝑨
𝑮

8
 -cut
, with
,
=
{ | or }

𝒔 𝒕 Capacity of a cut:
=

𝑨 𝑨
𝑮
Note: capacity of only outgoing edges is
considered in capacity of a cut.
9
CUTS AND FLOWS

10
 Useful Generalizations

conservation constraint:
For each vertex

𝒇 𝒊𝒏 (𝒖) 𝒇 𝒐𝒖𝒕 (𝒖)

𝒖
Mathematically,

𝒇 𝒊𝒏 (𝒖) 𝒇 𝒐𝒖𝒕 (𝒖)

Lemma: = 0 if
if
11
if
 Useful Generalizations

Consider any valid flow . Question: How would you define ?

Question: What would be ?

Answer:
𝒔
Proof ?

12
 Proof for =

Consider any valid flow . Proof :

= 𝒇 𝒐𝒖𝒕 ( 𝒔 ) − 𝒇 𝒊𝒏 ( 𝒔 )

𝒙
𝒔
𝒗

13
 A SIMPLE RELATION

between
Flows and capacity of cuts

14
 Flows and capacity of cuts

Consider any valid flow . Lemma: .

Proof:

𝒔 𝒕

𝑨 Theorem: the maximum value of -flow is

𝑨
bounded by the capacity of every -cut.
𝑮

15
 A DEEP RELATION

between
Flows and capacity of cuts

16
 Max-Flow Min-Cut Theorem

Theorem: the maximum value of -flow is equal to the capacity of min -cut.

Based on
Ford Fulkerson Algorithm

17
 Ford Fulkerson algorithm
Ford-Fulkerson-algo(,,)
{ ;
While (there is - path in ) do
{ Let be an - path in
Let be bottleneck capacity of ;
For each (,) do
Which is this cut defined by
{ If (,) is a forward edge then Ford Fulkerson algorithm?
(,)  (,) + ;
Else
(,)  (,) - ;
}
} return ;
}

18
 Proof of max-flow min-cut Theorem
How does
: flow computed by Ford-Fulkerson algorithm. look like ?

𝒔 𝒕 𝒔 𝒕

𝑨 𝑨 𝑨 𝑨
𝑮 𝑮𝒇
Claim: = .
: set of vertices reachable from in

19
 Proof of max-flow min-cut Theorem

: flow computed by Ford-Fulkerson algorithm.

𝒔 𝒕 𝒔 𝒕

𝑨 𝑨 𝑨 𝑨
𝑮 𝑮𝒇
=
Question: For to be equal to = , what must happen ?
Answer: 1. All out going edges must be fully saturated.
2. Every incoming edge must have zero flow. How will you
20
show this ?
 Proof of max-flow min-cut Theorem

: flow computed by Ford-Fulkerson algorithm.

𝒙 𝒙
𝒔 𝒚 𝒕 𝒔 𝒚 𝒕

𝑨 𝑨 𝑨 𝑨
𝑮 𝑮𝒇
If <
 must appear as forward edge in.
 is reachable from .
A contradiction. 21
 Proof of max-flow min-cut Theorem

Consider the flow computed by Ford-Fulkerson algorithm.

𝒔 𝒙 𝒕 𝒔 𝒙 𝒕
𝒚 𝒚

𝑨 𝑨 𝑨 𝑨
𝑮 𝑮𝒇
If >
 appears as a backward edge in .
 is reachable from .
A contradiction. 22
 Ford Fulkerson algorithm
Ford-Fulkerson-algo(,,)
{ ; Theorem: Ford Fulkerson algorithm
While (there is - path in ) do indeed computes the maximum - flow
upon termination of the While loop..
{ Let be an - path in
Let be bottleneck capacity of ;
For each (,) do
{ If (,) is a forward edge then
(,)  (,) + ; Time complexity ?
Else
(,)  (,) - ;
}
} return ;
}

23
 Ford Fulkerson algorithm
Ford-Fulkerson-algo(,,)
{ ; This algorithm does not say anything about
While (there is - path in ) do the way has to be selected.
{ Let be an - path in
Let be bottleneck capacity of ; 
For each (,) do it is up to us (or the adversary) to select so
as to force the execution of the algorithm to
{ If (,) is a forward edge then
take huge time 
(,)  (,) + ;
Else
We shall use the above idea
(,)  (,) - ;
to show a bad example of a network with
} integer edge capacities.
} return ;
}

24
 A worst case example for
networks with integer edge weights
𝑮 𝑮𝒇
𝒙 𝒙

1
1000 1000
1000 1000

𝒔 1 1 𝒕 𝒔 1 𝒕
1
1000 1000 1000 1000

𝒚 𝒚

25
 A worst case example for
networks with integer edge weights
𝑮 𝑮𝒇
𝒙 𝒙
1
1 1
1000 999 1000
1000

𝒔 1 1 𝒕 𝒔 1 𝒕
1 1
1000 1000 1000 999

𝒚 𝒚

26
 A worst case example for
networks with integer edge weights
𝑮 𝑮𝒇
𝒙 𝒙
1 1
1
2 1
1000 999
1000 999

𝒔 1 1 𝒕 𝒔 1 𝒕
1 12
1000 1000 999 999

1
1
𝒚 𝒚
The algorithm will run for 2000
iterations to compute max-flow ! 27
Theorem:
There are networks with integer edge capacities on which Ford-Fulkerson algo
may take () time, where is max-edge capacity.

No. of bits required to store capacity = ? log 2 𝑪

 Ford-Fulkerson algorithm is not a polynomial time algorithm

even for networks with integer edge capacities 

28
 INTEGRALITY OF MAX-FLOW

A useful tool for many applications of Max-Flow

29
 Integrality of max-flow

Theorem: Given a network =(,), and two vertices ,,

if edge capacities are integers, then there exists a maximum -flow
which is “integral”.
for each

1.67 1
.67
1 1
1
𝒔 .33
0 .33
0
1
𝒕
1 1

30
 Proof for Integrality theorem
Ford-Fulkerson-algo(,,) {
; Claim:
While (there is - path in ) do Ford Fulkerson algorithm computes a maximum
flow which is integral.
{ Let be an - path in Proof: (By induction on the no. of iteration)
Let be bottleneck capacity of ; Inductive Assertion:
For each (,) do At the end of th iteration, is integral.
{ If (,) is a forward edge then
(,)  (,) + ; [Homework: give all details of the proof].
Else
(,)  (,) - ; In fact, the statement can be shown to hold for
any flow (not necessarily max flow) whose
} value is an integer.
} return ;
}

31
 A motivating application of
Integrality theorem Max Flow

ROUNDING OF A MATRIX

32
 Rounding of a matrix
𝟏𝟏 𝟏𝟏. 𝟕 𝟏𝟐

Design an algorithm to round a matrix, if possible ?

Make sincere attempt to solve this problem. We shall discuss it after midsem. 33
 A useful exercise
to internalize Ford Fulkerson algorithm
Execute Ford Fulkerson algorithm on
𝒙 this example when
• The first path selected for sending
the flow is <, , , >.
10 3 9
5
Show the execution of the iterations
of the algorithm along with the
𝒔 𝒕 residual network.
8
𝒖 10 𝒗 10

3 3
• What if the first path selected is
6
5 <, , , >.
Internalize the entire algorithm fully
through this example.
𝒚

34