Lecture 27.

Maximum Flow / P or NP
Introduction to Algorithms
Sungkyunkwan University

Hyunseung Choo
[email protected]

Superintelligence Laboratory Copyright 2000-2022 Superintelligence

H.CHOO
Laboratory
1/39
 Max Flow Problem
 𝑮 = 𝐺(𝑉, 𝐸)
 𝒙𝑖𝑗 = flow on arc (𝑖, 𝑗)
 𝒖𝑖𝑗 = capacity of flow in arc (𝑖, 𝑗)
 𝒔 = source node
 𝒕 = sink node
 Maximize 𝒗
Subject to: σ𝒋 𝒙𝒊𝒋 − σ𝒌 𝒙𝒌𝒊 = 𝟎 for each 𝒊  𝒔, 𝒕
σ𝒋 𝒙𝒔𝒋 = 𝒗
𝟎 ≤ 𝒙𝒊𝒋 ≤ 𝒖𝒊𝒋 for all (𝒊, 𝒋) ∈ 𝑬

Superintelligence Laboratory H.CHOO 2/39

 Maximum Flows
 We refer to a flow x as maximum if it is feasible and
maximizes v
 Our objective in the max flow problem is to find a
maximum flow

1
10, 8 8,7

s 1,1
t

6, 5 10,6
2

A max flow problem

Capacities and a non-optimum flow
Superintelligence Laboratory H.CHOO 3/39
 Feasibility Problem: Find a Feasible Flow

warehouses retailers

6 1

6 6
5 2

7 7
4 3

8 6
5 4

9 5
4 5

Is there a way of shipping from the warehouses to

the retailers to satisfy demand?
Superintelligence Laboratory H.CHOO 4/39
 Transformation to a Max Flow Problem
warehouses retailers

6 1
6
6 6
5 5 2 6

4 7 77
s 4 3 t
6
5
8 6
5 4 4 5

9 5
4 5

There is a 1-1 correspondence with flows from s to t

with 24 units (why 24?) and feasible flows for the
transportation problem
Superintelligence Laboratory H.CHOO 5/39
 Feasibility Problem: Find a Matching

persons tasks

1 5

2 6

3 7

4 8

Is there a way of assigning persons to tasks so that each

person is assigned a task, and each task has a person
assigned to it?
Superintelligence Laboratory H.CHOO 6/39
 Transformation to a Max Flow Problem

persons tasks

1 5
1 1

1 2 6 1
s 1 1 t
3 7
1
1
4 8

Does the maximum flow from s to t have 4 units?

Superintelligence Laboratory H.CHOO 7/39

 Residual Networks

1 8,7
10, 8
1,1 uij ,xij
s t i j

6, 5 10,6
2

1
2 1
rij = uij - xij
8 7
s 1 t i j
1 4
xij
5 6
2
Let rij denote the
The Residual Network G(x) residual capacity of
arc (i,j)
Superintelligence Laboratory H.CHOO 8/39
 Augmenting Paths

 An augmenting path is a path from s to t in the residual network

 The residual capacity of the augmenting path P

 d(P) = min{rij : (i,j)  P}

 To augment along P, send d(P) units of flow along each arc of the
path, then, modify x and the residual capacities appropriately

 rij := rij - d(P) and rji := rji + d(P) for (i,j)  P

1 1
2 1 2
8 7 8 8
s 1 t s 1 t
1 4 4
5 6 6 6
2 2

Superintelligence Laboratory H.CHOO 9/39

 Ford-Fulkerson Max-Flow (1/16)

 Begin
 x := 0
 create the residual network G(x)
 while there is some directed path from s to t in G(x) do
 begin
 let P be a path from s to t in G(x)
  := d(P)
 send  units of flow along P
 update the r's
 end
 End {the flow x is now maximum}

Superintelligence Laboratory H.CHOO 10/39

 Ford-Fulkerson Max-Flow (2/16)

4
2 5
3 1 1 1

2 2
s 4 t
3 1 2

This is the original network, plus reversals of the arcs

Superintelligence Laboratory H.CHOO 11/39

 Ford-Fulkerson Max-Flow (3/16)

4
2 5
3 1 1 1

2 2
s 4 t
3 1 2

3
This is the original network, and
the original residual network

Superintelligence Laboratory H.CHOO 12/39

 Ford-Fulkerson Max-Flow (4/16)

4
2 5
3 1 1 1

2 2
s 4 t
3 2
1

Find any s-t path in G(x)

Superintelligence Laboratory H.CHOO 13/39

 Ford-Fulkerson Max-Flow (5/16)

4
2 5
3 1 1 1

2 1
2
s 4 t
2 1
3 1 2
1 1
3
Determine residual capacity  of the path
Send  units of flow in the path
Update residual capacities
Superintelligence Laboratory H.CHOO 14/39
 Ford-Fulkerson Max-Flow (6/16)

4
2 5
3 1 1 1

2 1
2
s 4 t
1
2
3 2
1
1
3

Find any s-t path

Superintelligence Laboratory H.CHOO 15/39

 Ford-Fulkerson Max-Flow (7/16)

4
2 5
3 1 1 1
2
1 11
s 4 t
1 1
2
3 1 2
1
1
1 1
3
Determine the residual capacity  of the path
Send  units of flow in the path
Update residual capacities
Superintelligence Laboratory H.CHOO 16/39
 Ford-Fulkerson Max-Flow (8/16)

4
2 5
3 1 1 1
2
1 11
s 4 1
t
1
2
3 2
1
1
1 1
3

Find any s-t path

Superintelligence Laboratory H.CHOO 17/39

 Ford-Fulkerson Max-Flow (9/16)

4
2 5
3 1 1 1
1 11
s 2
1 4 1
2
t
32 1
1
1
3 1

Determine the residual capacity  of the path

Send  units of flow in the path
Update residual capacities
Superintelligence Laboratory H.CHOO 18/39
 Ford-Fulkerson Max-Flow (10/16)

4
2 5
3 1 1 1
1
2 11
s 21 4 1
2
t
2
3 1
1
1
3 1

Find any s-t path

Superintelligence Laboratory H.CHOO 19/39

 Ford-Fulkerson Max-Flow (11/16)

4
2 5
3 1 1 1
1
2 11
s 2 4 2 t
1 1
2 1
1
11
2
1 2
1
3
Determine the residual capacity  of the path
Send  units of flow in the path
Update residual capacities
Superintelligence Laboratory H.CHOO 20/39
 Ford-Fulkerson Max-Flow (12/16)

4
2 5
3 1 1 1
1
2 11
s 21
4 1
2
t
1
2 1
12 2
1
3

Find any s-t path

Superintelligence Laboratory H.CHOO 21/39

 Ford-Fulkerson Max-Flow (13/16)

4
3
2 1
5
2
3 1 1 1
1 1

s 4 t
1 21
2 1
2
1
2
1 2
1
3
Determine the residual capacity  of the path
Send  units of flow in the path
Update residual capacities
Superintelligence Laboratory H.CHOO 22/39
 Ford-Fulkerson Max-Flow (14/16)

4
3
2 5
1
23 1 1 1
1 1

s 21
4 1
2
t
1
2 1
2
1 2
1
3

There is no s-t path in the residual network

This flow is optimal
Superintelligence Laboratory H.CHOO 23/39
 Ford-Fulkerson Max-Flow (15/16)

4
3
2 1
5
3
2 1 1 1
1 1

s 1
2 4 1
2 t
1
2
1
2
1 2
1
3

These are the nodes that are reachable from node s

Superintelligence Laboratory H.CHOO 24/39

 Ford-Fulkerson Max-Flow (16/16)

1
2 5
1 1

2 2
s 4 t
2 2

Here is the optimal flow

Superintelligence Laboratory H.CHOO 25/39

 How Do We Know
When a Flow Is Optimal?
1 8,8
10, 9

s 1,1
t

6,6 10,7
2
 METHOD
 There is no augmenting path in the residual network

1 reachable
1 8
from s in G(x)
9
s 1 t
3 not reachable
6
2
7 from s in G(x)

Superintelligence Laboratory H.CHOO 26/39

 A Simple and Very Bad Example

1
M M

s 1 t
M M

Superintelligence Laboratory H.CHOO 27/39

 After 1 Augmentation

1
M-1 M
1
s 1 t
M M-1
1
2

Superintelligence Laboratory H.CHOO 28/39

 After Two Augmentations

1
M-1 M-1
1 1
s 1 t
M-1 M-1

1 1
2

Superintelligence Laboratory H.CHOO 29/39

 After Three Augmentations

1
M-2 M-1
2 1
s 1 t
M-1 M-2

1 2
2

And so on
Superintelligence Laboratory H.CHOO 30/39
 After 2M Augmentations

1
M M

s 1 t
M M

Superintelligence Laboratory H.CHOO 31/39

 P or NP

I couldn’t find a
polynomial time
algorithm. I guess I’m
too dumb.

Superintelligence Laboratory H.CHOO 32/39

 P or NP
I couldn’t find a
polynomial time
algorithm, because no
such algorithm exists !

Superintelligence Laboratory H.CHOO 33/39

 P or NP

I couldn’t find a polynomial

time algorithm, but neither
could all these other smart
people.

Superintelligence Laboratory H.CHOO 34/39

 P or NP

Superintelligence Laboratory H.CHOO 35/39

 NP-Completeness
 Some problems are intractable
 They grow large, we are unable to solve them in reasonable
time
 What constitutes reasonable time?
Standard working definition: polynomial time
 For an input of size n, the worst-case running time is O(nk)
for some constant k
 Polynomial time: O(1), O(n lg n), O(n2), O(n3)
 Not in polynomial time: O(2n), O(nn), O(n!)

Superintelligence Laboratory H.CHOO 36/39

 Polynomial-Time Algorithms
 Are some problems solvable in polynomial time?
 Every algorithm we’ve studied provides polynomial-time
solution to some problems
 We define P to be the class of problems solvable in polynomial
time
 Are all problems solvable in polynomial time?
 No: Turing’s “Halting Problem” is not solvable by any computer,
no matter how much time is given
 Such problems are clearly intractable, not in P

Superintelligence Laboratory H.CHOO 37/39

 NP-Complete Problems
 Definition
 The class P (or NP) denotes the family of all problems that can
be solved by deterministic (nondeterministic) polynomial time
algorithms
All decision problems only answer YES or NO
 Nondeterministic polynomial time algorithm exist?

Superintelligence Laboratory H.CHOO 38/39

 NP-Complete Problems
 The NP-Complete problems are an interesting class of
problems whose status is unknown
 No polynomial-time algorithm has been discovered for an NP-
Complete problem
 We call this the P = NP?
 The biggest open problem in CS
 Examples of NP-Complete problems:
 Vertex cover problem
 Traveling salesman problem
 Job scheduling problem
 …

Superintelligence Laboratory H.CHOO 39/39

 Thanks to contributors
Mr. Pham Van Nguyen (2022)
Dr. Thien-Binh Dang (2017 – 2022)
Prof. Hyunseung Choo (2001 – 2022)

Superintelligence Laboratory Copyright 2000-2022 Superintelligence

H.CHOOLaboratory
40/39