Graph Theory

Video Series
 Max Flow
Ford-Fulkerson
method
William Fiset
 Max flow
Q: With an infinite input source, how much “flow” can we push
through the network given that each edge has a certain capacity?

0/2 0/7
Source 0/7 0
0/4
3 6 Sink
0/3 0/1
0/4 0/1
0/2 0/3
s 1 0/5
4 0/8 7 t
0/4
0/8 0/3
0/1 2 0/6 5 8
0/3
 Max flow
Q: With an infinite input source, how much “flow” can we push
through the network given that each edge has a certain capacity?

0/2 0/7
Source 0/7 0
0/4
3 6 Sink
0/3 0/1
0/4 0/1
0/2 0/3
s 1 0/5
4 0/8 7 t
0/4
0/8 0/3
0/1 2 0/6 5 8
0/3

Suppose the edges are roads with cars, pipes with water
or wires with electric current. Flow represents the
volume of water allowed to flow through the pipes, the
number of cars the roads can sustain in traffic and net
electric current. Effectively, it’s the “bottleneck”
value for the amount of flow that can pass through the
network from source to sink under all the constraints.
 Max flow
Q: With an infinite input source, how much “flow” can we push
through the network given that each edge has a certain capacity?

2/2 1/7
Source 4/7 0
0/4
3 6 Sink
0/3 1/1
2/4 1/1
2/2 2/3
s 1 1/5
4 1/8 7 t
4/4
2/8 1/3
1/1 2 1/6 5 8
3/3

In this flow graph, the max flow is 7 since 1+2+4 = 7

is the sum of the flows entering the sink node.
That’s the most that we can push through the network.
 0 2

s t

1 3

A flow graph (flow network) is a directed

graph where each edge (also called an arc)
has a certain capacity which can receive a
certain amount of flow. The flow running
through an edge must be less than or equal
to the capacity.
 0/10 0 0/6 0/25
2

0/10

s t
0/10

0/10
1 0/15
3

Each edge in the flow graph has a certain

flow and capacity specified by the fraction
adjacent to each edge. Initially, the flow
through each edge is 0 and the capacity is
a non negative value.
 0/10 0 0/6 0/25
2

0/10

s t
0/10

0/10
1 0/15
3

To find the maximum flow (and min-cut as a by

product), the Ford-Fulkerson method repeatedly
finds augmenting paths through the residual
graph and augments the flow until no more
augmenting paths can be found.
 0/10 0 0/6 0/25
2

0/10

s t
0/10

0/10
1 0/15
3

An augmenting path is a path of edges in the

residual graph with unused capacity greater
than zero from the source s to the sink t.
 0/10 0 0/6 0/25
2

0/10

s t
0/10
Bottleneck of 6

0/10
1 0/15
3
In the augmenting path above, the bottleneck
is the “smallest” edge on the path. We can
use the bottleneck value to augment the flow
along the path.

The smallest remaining capacity of any edge

along the augmenting path is:
min(10-0, 15-0, 6-0, 25-0, 10-0) = min(10,15,6,25,10) = 6
 0/10 0 6/6 6/25
2

6/10

s t
0/10

6/10
1 6/15
3

Augmenting the flow means updating the flow

values of the edges along the augmenting path.

For forward edges, this means increasing

the flow by the bottleneck value.
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3

When augmenting the flow along the augmenting

path, you also need to decrease the flow along
each residual edge by the bottleneck value.
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3

Residual edges exist to “undo” bad augmenting

paths which do not lead to a maximum flow.
 -6/0

0/10 0 6/6 6/25

2 -6/0

0/0 6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3 0/0

You can think of every edge in the original graph

as having a residual edge with a flow/capacity of
0/0 which is not usually shown.

The residual graph is the graph which

also contains residual edges. In general,
when I mention the flow graph, I mean the
residual graph.
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3
Q: Residual edges have a capacity of 0? Isn’t
that forbidden? How does that work?
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3
Q: Residual edges have a capacity of 0? Isn’t
that forbidden? How does that work?

A: Think of the remaining capacity of an edge

e (residual or not) as: e.capacity - e.flow,
This ensures that the remaining capacity of an
edge is always non-negative (even if the flow
can be negative).
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3

The Ford-Fulkerson method continues finding

augmenting paths and augments the flow until
no more augmenting paths from s -> t exist.

A key realization to make at this point is

that the sum of the bottlenecks found in each
augmenting path is equal to the max-flow!
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -6/0
t
0/10

-6/0 -6/0
6/10
1 6/15
3

Bottleneck is 4 since the minimum of the

remaining capacities on the augmenting path is 4:

min(10 - 6, 15 - 6, 10 - 0) = min(4, 9, 10) = 4

 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -10/0
t
4/10

-10/0 -6/0
10/10
1 10/15
3 -4/0

Augment the edges with bottleneck value of 4

 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -10/0
t
4/10

-10/0 -6/0
10/10
1 10/15
3 -4/0
 -6/0

0/10 0 6/6 6/25

2 -6/0

6/10

s -10/0
t
4/10

-10/0 -6/0
10/10
1 10/15
3 -4/0

The bottleneck value is 6 since the smallest

remaining capacity amongst the edges of our
augmenting path is:

min(10 - 0, 0 - -6, 10 - 4) = min(10, 6, 6) = 6

 -6/0

6/10 0 0/6 6/25

2 -6/0

-6/0 6/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0

Augment the edges with bottleneck value of 6

 -6/0

6/10 0 0/6 6/25

2 -6/0

-6/0 6/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0
 -6/0

6/10 0 0/6 6/25

2 -6/0

-6/0 6/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0
 -6/0

6/10 0 0/6 6/25

2 -6/0

-6/0 6/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0

The bottleneck value is 4 since the smallest

remaining capacity amongst the edges of our
augmenting path is:

min(10 - 6, 25 - 6, 10 - 6) = min(4, 19, 4) = 4

 -10/0

10/10 0 0/6 10/25

2 -10/0

-10/0 10/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0

Augment the edges with bottleneck value of 4

 -10/0

10/10 0 0/6 10/25

2 -10/0

-10/0 10/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0
 -10/0

10/10 0 0/6 10/25

2 -10/0

-10/0 10/10

s -10/0
t
10/10

-10/0 0/0
10/10
1 10/15
3 -10/0

No more augmenting paths can be found,

so the algorithm terminates!

maximum flow = 6 + 4 + 6 + 4 = 20

bottleneck values
 The time complexity of the Ford-Fulkerson
method depends on the algorithm being used
to find the augmenting paths, which is left
unspecified.
 Assuming the method of finding augmenting
paths is by using a Depth First Search (DFS),
the algorithm runs in O(fE), where f is the
maximum flow and E is the number of edges.
 Assuming the method of finding augmenting
paths is by using a Depth First Search (DFS),
the algorithm runs in O(fE), where f is the
maximum flow and E is the number of edges.

0/100 0
0/0
0/0

0/100

0/0
s t
0/100

0/0

0/100 0/1

0/0
1
 Recall that a DFS traversal chooses edges in
a random order, so it is possible to pick the
middle edge every time when finding an
augmenting path.

1/100 0
0/0
-1/0

0/100

-1/0
s t
1/100

0/0

0/100 1/1

-1/0
1
 Recall that a DFS traversal chooses edges in
a random order, so it is possible to pick the
middle edge every time when finding an
augmenting path.

1/100 0
-1/0
0/0

1/100

-1/0
s t
1/100

-1/0

1/100 0/1

-1/0
1
This results in flipping back and forth
between the same two alternating paths
for 200 iterations…

2/100 0
-1/0
-1/0

1/100

-2/0
s t
2/100

-1/0

1/100 1/1

-2/0
1
This results in flipping back and forth
between the same two alternating paths
for 200 iterations…

2/100 0
-2/0
0/0

2/100

-2/0
s t
2/100

-2/0

2/100 0/1

-2/0
1
This results in flipping back and forth
between the same two alternating paths
for 200 iterations…

3/100 0
-2/0
-1/0

2/100

-3/0
s t
3/100

-2/0

2/100 1/1

-3/0
1
 Other much faster algorithms and heuristics
exist to find the maximum flow.

Edmonds-Karp: Uses a BFS as a method of

finding augmenting paths, O(E²V)
Capacity scaling: Adds a heuristic on top of Ford-
Fulkerson to pick larger paths first, O(E²log(U))
Dinic’s algorithm: Uses combination of BFS +
DFS to find augmenting paths, O(V²E)
Push Relabel: Uses a concept of maintaining a
"preflow" instead of finding augmenting paths to
achieve a max-flow solution, O(V²E) or O(V²√E)
variant.
NOTE: Be mindful that the time complexities for
flow algorithms are very pessimistic. In practice,
they tend to operate much faster, making it hard
to compare the performance of flow algorithms
solely based on complexity.
Next Video: Source Code
Graph Theory
Video Series
 Max Flow
Ford-Fulkerson
source code
William Fiset
Previous video explaining
Ford-Fulkerson method:
 Let’s setup this flow graph:

Source 0/10 0/15 Sink

0/10 0 3 6
0/3
0/15
0/20 0/10
0/5 0/10
s 1 0/15
4 7 t
0/4
0/10
0/10
0/7

0/10
2 0/15
5 0/10
8
 Then let’s code Ford-Fulkerson and hope to
achieve the following graph (or similar)
with max-flow of 23:

Source 10/10 13/15 Sink

10/10 0 3 6
3/3
13/15
0/20 0/10
5/5 0/10
s 1 0/15
4 7 t
3/4
10/10
5/10
0/7

8/10
2 13/15
5 10/10
8
Source Code Link
Implementation source code can
be found at the following link:
github.com/williamfiset/algorithms

Link in the description:

 0/10 0/15
0/10 0 3 6
0/3
0/15
0/20 0/10
0/5 0/10
s 1 0/15
4 7 t
0/4
0/10
0/10
0/7

0/10
2 0/15
5 0/10
8
 0/10 0/15
0/10 0 3 6
0/3
0/15
0/20 0/10
0/5 0/10
s 1 0/15
4 7 t
0/4
0/10
0/10
0/7

0/10
2 0/15
5 0/10
8
 0/10 0/15
0/10 0 3 6
0/3
0/15
0/20 0/10
0/5 0/10
s 1 0/15
4 7 t
0/4
0/10
0/10
0/7

0/10
2 0/15
5 0/10
8
Graph Theory
Video Series
 Network Flow:
unweighted bipartite
graph matching
William Fiset
 Bipartite Graph
A bipartite graph is one whose vertices can be
split into two independent groups U, V such
that every edge connects between U and V.

Other definitions exist such as: The

graph is two colourable or there is no
cycle with an odd length.
Maximum Cardinality Matching
Generally we’re interested in what’s called a
Maximum Cardinality Bipartite Matching (MCBM).
This is when we’ve maximized the pairs that
can be matched with each other.
There exists several applications such as
matching candidates to jobs, chairs to desks,
surfers to surf boards, etc…
 Common matching variations
Easier Harder

Easier Bipartite Non-Bipartite

Unweighted
Edges

Weighted
Edges

Harder
Reference: Competitive Programming 3 The New Lower
Bound of Programming Contests. P 349
 Common matching variations
Easier Harder

Easier Bipartite Non-Bipartite

• Max flow algorithms

Unweighted • Repeated augmenting
Edges paths with dfs
• Hopcroft-Karp

Weighted
Edges

Harder
Reference: Competitive Programming 3 The New Lower
Bound of Programming Contests. P 349
 Common matching variations
Easier Harder

Easier Bipartite Non-Bipartite

• Max flow algorithms

Unweighted • Repeated augmenting
Edges paths with dfs
• Hopcroft-Karp

• Min cost max flow

algorithms
Weighted
• Hungarian algorithm
Edges
(perfect matching)
• LP network simplex

Harder
Reference: Competitive Programming 3 The New Lower
Bound of Programming Contests. P 349
 Common matching variations
Easier Harder

Easier Bipartite Non-Bipartite

• Max flow algorithms

Unweighted • Repeated augmenting • Edmond’s blossom
Edges paths with dfs algorithm
• Hopcroft-Karp

• Min cost max flow

algorithms
Weighted
• Hungarian algorithm
Edges
(perfect matching)
• LP network simplex

Harder
Reference: Competitive Programming 3 The New Lower
Bound of Programming Contests. P 349
 Common matching variations
Easier Harder

Easier Bipartite Non-Bipartite

• Max flow algorithms

Unweighted • Repeated augmenting • Edmond’s blossom
Edges paths with dfs algorithm
• Hopcroft-Karp

• Min cost max flow

algorithms
Weighted • DP solution for
• Hungarian algorithm
Edges small graphs
(perfect matching)
• LP network simplex

Harder
Reference: Competitive Programming 3 The New Lower
Bound of Programming Contests. P 349
Suppose there are 5 people + 5 books and some people
express interest in some books, who gets what book?
Suppose there are 5 people + 5 books and some people
express interest in some books, who gets what book?
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution

This book is
already matched!
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution
Let’s try a greedy matching solution

This book is
already matched!
Let’s try a greedy matching solution
Let’s try a greedy matching solution

This book is
already matched!
Let’s try a greedy matching solution
Let’s try a greedy matching solution
The greedy approach found 4 matchings,
but can we do better?
 Yes! We can turn this matching problem into a
network flow problem by adding a sink s and a source
t and making each edge have unit capacities.
First make the edges directed and add unit capacities
to each edge. The 0/1 besides each edge means 0 flow
and a maximum capacity of 1.

0/1

0/1
0/1
0/1

0/1

0/1
0/1
0/1
0/1

0/1

0/1

0/1

0/1
 Then attach the source s and the sink t nodes
allowing for one unit of capacity.

0/1 0/1

0/1
0/1
0/1 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1 0/1 t
0/1
0/1 0/1
0/1

0/1
0/1
0/1

0/1
0/1

0/1
 Once the flow graph is set up, use any max-flow
algorithm to push flow through the network.

0/1 0/1

0/1
0/1
0/1 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1 0/1 t
0/1
0/1 0/1
0/1

0/1
0/1
0/1

0/1
0/1

0/1
Pushing flow through the network to achieve the maximum
flow changes the edge capacities. We can find which
edges were taken by checking if the flow is 1

1/1 1/1

1/1
0/1
1/1 0/1

1/1
0/1
1/1
0/1
s 1/1 0/1 1/1 t
1/1
0/1 1/1
1/1

1/1
1/1
0/1

1/1
0/1

1/1

NOTE: Residual edges are not shown

Pushing flow through the network to achieve the maximum
flow changes the edge capacities. We can find which
edges were taken by checking if the flow is 1

1/1 1/1

1/1
0/1
1/1 0/1

1/1
0/1
1/1
0/1
s 1/1 0/1 1/1 t
1/1
0/1 1/1
1/1

1/1
1/1
0/1

1/1
0/1

1/1

NOTE: Residual edges are not shown

 Q: We originally set the capacity of each edge
from the source to each person to be 1, what
constraint does this enforce?

0/1 0/1

0/1
0/1
0/1 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1 0/1 t
0/1
0/1 0/1
0/1

0/1
0/1
0/1

0/1
0/1

0/1

NOTE: Residual edges are not shown

 A: The capacity of 1 ensures that each person
can get up to one book and no more.

0/1 0/1

0/1
0/1
0/1 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1 0/1 t
0/1
0/1 0/1
0/1

0/1
0/1
0/1

0/1
0/1

0/1

NOTE: Residual edges are not shown

 Let’s increase some of these values to allow some
people to possibly pick up more than one book.

0/1
0/1

0/1
0/1

0/2 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1 t
0/3
0/1
0/1 0/1
0/1

0/1

0/1 0/1

0/2 0/1

0/1

NOTE: Residual edges are not shown

If we run max-flow again through the network we see that
it’s now possible for one person to have multiple books.

1/1
1/1

1/1
0/1

2/2 0/1

1/1
0/1
1/1
1/1
s 0/1 1/1 t
2/3
0/1
0/1 1/1
1/1

1/1

0/1 0/1

0/2 1/1

0/1

NOTE: Residual edges are not shown

 Q: What do we need to change in the flow network to
allow a book to be selected multiple times (e.g there
are multiple copies of the book in the library)?

0/1
0/1

0/1
0/1

0/2 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1 t
0/3
0/1
0/1 0/1
0/1

0/1

0/1 0/1

0/2 0/1

0/1

NOTE: Residual edges are not shown

A: The number of “copies” of a book is determined by
the capacity of the edges leading to the sink, t.

0/1
0/1

0/1
0/1

0/2 0/1

0/1
0/1
0/1
0/1
s 0/1 0/1
t
0/3
0/1
0/1 0/1
0/1

0/1

0/1 0/1

0/2 0/1

0/1

NOTE: Residual edges are not shown

 Let’s change the capacity of some of the edges
leading to the sink to allow having multiple
copies of the same book.

0/1
0/1

0/1
0/1

0/2 0/1

0/5
0/1
0/1
0/1
s 0/1 0/2
t
0/3
0/1
0/1 0/1
0/3

0/1

0/1 0/1

0/2 0/1

0/1

NOTE: Residual edges are not shown

 If we re-run the algorithm on the network, we see
that we now have people matched with the same
books because multiple copies exist.

1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

 1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

 1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

 1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

 1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

 1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

The flow value on the edge from the book to the sink
indicates how many copies of that book were given out
to various people.

1/1
1/1

1/1
0/1

2/2 0/1

1/5
0/1
1/1
1/1
s 1/1 2/2
t
3/3
0/1
0/1 1/1
2/3

1/1

0/1 0/1

1/2 1/1

1/1

NOTE: Residual edges are not shown

Q: Currently, each person is only allowed to pick up one copy
of each book (even though there are multiple copies of each
book). How do we modify the flow network to support this?

0/1
0/1

0/1
0/1

0/2 0/1

0/5
0/1
0/1
0/1
s 0/1 0/2
t
0/3
0/1
0/1 0/1
0/3

0/1

0/1 0/1

0/2 0/1

0/1

NOTE: Residual edges are not shown

 A: Modify the edge capacity between a person and
a book to allow that person to pick up multiple
copies of that book.

0/1
0/1

0/1
0/1

0/2 0/1

0/5
0/1
0/1
0/1
s 0/1 0/2
t
0/3
0/1
0/1 0/1
0/3

0/1

0/1 0/1

0/2 0/1

0/1

NOTE: Residual edges are not shown

Graph Theory
Video Series
Network Flow:
mice and owls
William Fiset
 Suppose M mice are out on a field and
there’s a hungry owl about to make a move.
 Further suppose there are H holes scattered
across the ground that the mice can hide in
(each having a certain capacity).

1 1

2
3

2
1

2
 Assume every mouse is capable of running a
radius of r before being caught by the owl.
What is the maximum number of mice that can
hide safely?

1 1

2
3

2
1

2
 The first step is to figure out which holes
each mouse can reach.

1 1

2
3

2
1

2
 1 1

2
3

2
1

2
 Which mice should go in which holes to
maximize the overall safety of the group?

1 1

2
3

2
1

2
 Which mice should go in which holes to
maximize the overall safety of the group?

1 1

2
3

2
1

2
 The key realization to make is that this
graph is bipartite.

1 1

2
3

2
1

2
 Flow Graph Setup

Mice Holes
0

… …
M-3

M-2

M-1

Label (index) the mice nodes [0, M)

 Flow Graph Setup

Mice Holes
0

1 M

2 M+1

… …
M-3 M+H-2

M-2 M+H-1

M-1

Label (index) the hole nodes [M, M+H)

 Flow Graph Setup

Mice Holes
0

1 0/1 M
0/1
0/1

2 0/1 M+1
0/1

… …
0/1
M-3 M+H-2
0/1
0/1
M-2 M+H-1
0/1

M-1

Place an edge with a capacity of 1 between the mouse and

the hole if the mouse can reach the particular hole.
 Flow Graph Setup

Mice Holes
0
0/1

1 0/1 M
0/1 0/1
0/1

2 0/1 M+1
0/1 0/1

s … …
0/1
0/1 M-3 M+H-2
0/1
0/1
0/1 M-2 M+H-1
0/1

0/1 M-1

Connect an edge with capacity 1 from the source

to each mouse node to indicate that each node
can have at most 1 mouse.
 Flow Graph Setup

Mice Holes
0
0/1

1 0/1 M
0/1 0/1
0/1

2 0/1 M+1
0/1 0/1 0/h1

s … … 0/h2
t
0/h3
0/1
M-3 0/h4
0/1 M+H-2
0/1
0/1
0/1 M-2 M+H-1
0/1

0/1 M-1

Connect an edge from each hole node to the sink

with the capacity of the hole.
 Flow Graph Setup

Mice Holes
0
0/1

1 0/1 M
0/1 0/1
0/1

2 0/1 M+1
0/1 0/1 0/h1

s … … 0/h2
t
0/h3
0/1
M-3 0/h4
0/1 M+H-2
0/1
0/1
0/1 M-2 M+H-1
0/1

0/1 M-1

The problem has now been transformed into a maximum flow

problem. Run any max-flow algorithm and the most mice
that are safe is equal to the value of the max-flow.
Graph Theory
Video Series
 Network Flow:
elementary math
William Fiset
 Elementary Math

Problem Statement: Ellen is a math teacher who is

preparing n (1 ≤ n ≤ 2500) questions for her math
exam. In each question, the students have to add
(+), subtract (−) or multiply (*) a pair of numbers.

Ellen has already chosen the n pairs of numbers.

All that remains is to decide for each pair which
of the three possible operations the students
should perform. To avoid students getting bored,
Ellen wants to make sure that the n correct
answers to her exam are all different.

Problem Link (in description):

https://open.kattis.com/problems/elementarymath
 Elementary Math
For each pair of numbers (a,b) in the same order
as in the input, output a line containing a valid
equation. Each equation should consist of five
parts: a, one of the three operators, b, an equals
sign (=), and the result of the expression. All
the n expression results must be different.

If there are multiple valid answers, output any of

them. If there is no valid answer, output a single
line with the string “impossible” instead.

Problem Link (in description):

https://open.kattis.com/problems/elementarymath
Elementary Math

1 5 =

3 3 =

4 5 =

-1 -6 =
Elementary Math

1 * 5 = 5

3 + 3 = 6

4 - 5 = -1

-1 * -6 = 6
 Elementary Math

1 * 5 = 5

3 + 3 = 6
Answers must
be unique!
4 - 5 = -1

-1 * -6 = 6
Elementary Math

1 - 5 = -4

3 - 3 = 0

4 + 5 = 9

-1 * -6 = 6
Elementary Math

1 - 5 = -4

3 - 3 = 0

4 + 5 = 9

-1 * -6 = 6
Elementary Math

1 2 =

2 1 =

1 2 =

2 1 =

1 2 =
 Elementary Math

1 2 =

2 1 =
No solution
exists! 1 2 =

2 1 =

1 2 =
 Elementary Math
This problem presents itself as a network flow
problem even though it might not be obvious at
first. Take a moment and attempt to set up a
flow graph that can solve this problem.
 Elementary Math
This problem presents itself as a network flow
problem even though it might not be obvious at
first. Take a moment and attempt to set up a
flow graph that can solve this problem.

Questions to ask yourself:

Q: Is there a way that this problem can be

simplified as a bipartite graph?
 Elementary Math
This problem presents itself as a network flow
problem even though it might not be obvious at
first. Take a moment and attempt to set up a
flow graph that can solve this problem.

Questions to ask yourself:

Q: Is there a way that this problem can be

simplified as a bipartite graph?

Q: How do I detect an impossible set of pairs?

 Elementary Math
This problem presents itself as a network flow
problem even though it might not be obvious at
first. Take a moment and attempt to set up a
flow graph that can solve this problem.

Questions to ask yourself:

Q: Is there a way that this problem can be

simplified as a bipartite graph?

Q: How do I detect an impossible set of pairs?

Q: How do I handle multiple repeated pairs?

 Elementary Math
This problem presents itself as a network flow
problem even though it might not be obvious at
first. Take a moment and attempt to set up a
flow graph that can solve this problem.

Questions to ask yourself:

Q: Is there a way that this problem can be

simplified as a bipartite graph?

Q: How do I detect an impossible set of pairs?

Q: How do I handle multiple repeated pairs?

A: This slide deck explains the first

two, the third is left as an exercise.
 Elementary Math
Key realization: For every pair, at most
three unique solutions are produced.

2 + 3 = 5

2 3 = 2 - 3 = -1

2 * 3 = 6
This makes it easy to formulate the flow graph
as a bipartite graph with input pairs on one
side and solutions on the other.
 Elementary Math
Let’s see an example of how to set up
the flow graph for the following pairs:

1 5 =

3 3 =

-1 -6 =

2 2 =
(a,b) pairs Answers
 -4
1,5
6

5
1 - 5 = -4
1 + 5 = 6
1 * 5 = 5

(a,b) pairs Answers

 -4
1,5
6

5
3,3
3 - 3 = 0
9
3 + 3 = 6
3 * 3 = 9
0

(a,b) pairs Answers

 -4
1,5
6

5
3,3
-1 - -6 = 5
9
-1 + -6 = -7
-1 * -6 = 6
-1,-6 0

-7

(a,b) pairs Answers

 -4
1,5
6

5
3,3
2 - 2 = 0
9
2 + 2 = 4
2 * 2 = 4
-1,-6 0

-7

2,2
4

(a,b) pairs Answers

 -4
1,5
6

5
3,3
9

-1,-6 0

-7

2,2
4

(a,b) pairs Answers

 -4
1,5
6

5
3,3
9
s t
-1,-6 0

-7

2,2
4

(a,b) pairs Answers

 Capacities from answer nodes to sink should
have a capacity of one since answers need
to be unique.
-4
1,5
6

5 0/1
3,3 0/1

0/1
9
s 0/1 t
0/1
-1,-6 0 0/1
0/1

-7

2,2
4

(a,b) pairs Answers

 Capacities from input pairs to answers also
have capacity one since only one of ‘+’, ‘-‘
or ‘*' can be matched with an answer.
-4
0/1

1,5
0/1 6
0/1
0/1

0/1
5 0/1
3,3 0/1 0/1

0/1
9
s
0/1
0/1 t
0/1 0/1
-1,-6 0 0/1
0/1 0/1

0/1
-7

2,2
0/1 4

(a,b) pairs Answers

 Capacities from source to input pairs should reflect the
frequency of the input pair. In this example, all frequencies
are 1, but as we know, this isn’t always the case.

-4
0/1

1,5
0/1
0/1 6
0/1
0/1

0/1
5 0/1

0/1
3,3 0/1 0/1

0/1
9
s
0/1
0/1 t
0/1 0/1
0/1 -1,-6 0 0/1
0/1 0/1

0/1
0/1
-7

2,2
0/1 4

(a,b) pairs Answers

 Once the flow graph is set up, run a
max-flow algorithm on it!

-4
0/1

1,5
0/1
0/1 6
0/1
0/1

0/1
5 0/1

0/1
3,3 0/1 0/1

0/1
9
s
0/1
0/1 t
0/1 0/1
0/1 -1,-6 0 0/1
0/1 0/1

0/1
0/1
-7

2,2
0/1 4

(a,b) pairs Answers

 Once the flow graph is set up, run a
max-flow algorithm on it!

-4
0/1

1,5
1/1
1/1 6
0/1
0/1

0/1
5 0/1

1/1
3,3 0/1 1/1

0/1
9
s
0/1
0/1 t
1/1 1/1
1/1 -1,-6 0 1/1
0/1 1/1

1/1
1/1
-7

2,2
1/1 4

(a,b) pairs Answers

NOTE: For simplicity, the residual edges are not shown
 Every edge in the middle with one unit of
flow represents a matching from a pair
(a,b) to its answer.
-4
0/1

1,5
1/1
1/1 6
0/1
0/1

0/1
5 0/1

1/1
3,3 0/1 1/1

0/1
9
s
0/1
0/1 t
1/1 1/1
1/1 -1,-6 0 1/1
0/1 1/1

1/1
1/1
-7

2,2
1/1 4

(a,b) pairs Answers

NOTE: For simplicity, the residual edges are not shown
We can even deduce the operator used for each matching
(needed for final output) by trying which of (+, -, *)
results in the answer matched the one on the right.

-4
0/1

1,5
1/1
1/1 6
0/1
0/1

0/1
5 0/1

1/1
3,3 0/1 1/1

0/1
9
s
0/1
0/1 t
1/1 1/1
1/1 -1,-6 0 1/1
0/1 1/1

1/1
1/1
-7

2,2
1/1 4

(a,b) pairs Answers

NOTE: For simplicity, the residual edges are not shown
 -4
1,5
6 1 5 =
5
3,3 3 3 =
9

-1,-6 0 -1 -6 =
-7
2 2 =
2,2
4
 -4
1,5
6 1 5 =
5
3,3 3 3 =
9

-1,-6 0 -1 -6 =
-7
2 2 =
2,2
4

1 + 5 equals 6?
1 - 5 equals 6?
1 * 5 equals 6?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 3 =
9

-1,-6 0 -1 -6 =
-7
2 2 =
2,2
4

1 + 5 equals 6?
1 - 5 equals 6?
1 * 5 equals 6?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 3 =
9

-1,-6 0 -1 -6 =
-7
2 2 =
2,2
4

3 + 3 equals 0?
3 - 3 equals 0?
3 * 3 equals 0?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 -6 =
-7
2 2 =
2,2
4

3 + 3 equals 0?
3 - 3 equals 0?
3 * 3 equals 0?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 -6 =
-7
2 2 =
2,2
4

-1 + -6 equals -7?
-1 - -6 equals -7?
-1 * -6 equals -7?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 + -6 = -7
-7
2 2 =
2,2
4

-1 + -6 equals -7?
-1 - -6 equals -7?
-1 * -6 equals -7?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 + -6 = -7
-7
2 2 =
2,2
4

2 + 2 equals 4?
2 - 2 equals 4?
2 * 2 equals 4?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 + -6 = -7
-7
2 2 =
2,2
4

2 + 2 equals 4?
2 - 2 equals 4?
2 * 2 equals 4?
 You can pick either operator if
multiple of them work.

-4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 + -6 = -7
-7
2 * 2 = 4
2,2
4

2 + 2 equals 4?
2 - 2 equals 4?
2 * 2 equals 4?
 -4
1,5
6 1 + 5 = 6
5
3,3 3 - 3 = 0
9

-1,-6 0 -1 + -6 = -7
-7
2 * 2 = 4
2,2
4
 1 + 5 = 6

3 - 3 = 0

-1 + -6 = -7

2 * 2 = 4
Graph Theory
Video Series
Network Flow:
Edmonds-Karp
Algorithm
William Fiset
Ford-Fulkerson overview

The Ford-Fulkerson method is a common

technique used to find the maximum flow.

0/2 0/7
Source 0/7 0
0/4
3 6 Sink
0/3 0/1
0/4 0/1
0/2 0/3
s 1 0/5
4 0/8 7 t
0/4
0/8 0/3
0/1 2 0/6 5 8
0/3
 Ford-Fulkerson overview
At a high level, Ford-Fulkerson says that all
we want to do is repeatedly find augmenting
paths from s -> t in the flow graph, augment
the flow and repeat until no more paths exist.

0/2 0/7
Source 0/7 0
0/4
3 6 Sink
0/3 0/1
0/4 0/1
0/2 0/3
s 1 0/5
4 0/8 7 t
0/4
0/8 0/3
0/1 2 0/6 5 8
0/3
 Ford-Fulkerson overview
The key takeaway here is that the Ford-Fulkerson
method does not specify how to actually find
augmenting paths. This is where optimizations
come into play.

0/2 0/7
Source 0/7 0
0/4
3 6 Sink
0/3 0/1
0/4 0/1
0/2 0/3
s 1 0/5
4 0/8 7 t
0/4
0/8 0/3
0/1 2 0/6 5 8
0/3
 Edmonds-Karp
The Ford-Fulkerson method using a DFS to find
augmenting paths takes O(Ef) where E is the
number of edges and f is the max flow.

ANIMATION HERE
 Edmonds-Karp
The Ford-Fulkerson method using a DFS to find
augmenting paths takes O(Ef) where E is the
number of edges and f is the max flow.

The Edmonds-Karp algorithm uses a Breadth

First Search (BFS) to find augmenting
paths which yields an arguably better
time complexity of O(VE²). The major
difference in this approach is that the
time complexity no longer depends on the
capacity value of any edge!
 Shortest augmenting path

The Edmonds-Karp algorithm can also be

thought of as a method of augmentation which
repeatedly finds the shortest augmenting path
from s -> t in terms of the number of edges
used each iteration.

Using a BFS to find augmenting paths

ensures that the shortest path from s -> t
is found every iteration.
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
Using a DFS can lead to zigzagging through the flow graph
to find the sink which can cause longer augmenting paths.
Longer paths are generally undesirable because the longer
the path, the higher the chance for a small bottleneck
value which results in a longer runtime.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 Using the shortest path (in terms of the number of
edges) as an augmenting path is a great approach to
avoid the DFS worse case and reduce the length of
augmenting paths.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
Do a Breadth First Search (BFS) starting at the source
and ending at the sink. While exploring the flow graph,
remember that we can only reach a node if the capacity
of the edge to get there is greater than 0.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 When a complete path from s -> t is found, get
the augmenting path.

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 Find the bottleneck value:
min(5-0, 10-0, 10-0, 10-0) = 5

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 Update (augment) the flow of the forward and residual
edges along the path by the bottleneck value (5).

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Do a Breadth First Search (BFS) starting at the source
and ending at the sink. While exploring the flow graph
remember that we can only reach a node if the capacity
of the edge to get there is greater than 0
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…
The bottom left node here is unreachable because the
edge from the source leading to it has a capacity of 0.

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Reconstruct the augmenting path

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Find the bottleneck value:
min(10-0, 20-0, 30-0, 15-0) = 10

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Update (augment) the flow of the forward and residual
edges along the path by the bottleneck value (10).

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Do a Breadth First Search (BFS) starting at the source
and ending at the sink. While exploring the flow graph
remember that we can only reach a node if the capacity
of the edge to get there is greater than 0
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Add all reachable neighbours to the queue and proceed…

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Reconstruct the augmenting path

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Find the bottleneck value:
min(5-0, 10-0, 10-0, 5-0) = 5

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Update (augment) the flow of the forward and residual
edges along the path by the bottleneck value (5).

5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 There are no more augmenting paths left to be found
because all the edges leading outwards from the source
have a remaining capacity of 0. However, more generally
we know to stop when there are no more augmenting paths
from s -> t
5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 The max flow from running Edmonds-Karp is the sum
of the bottleneck values:

Max flow = 5 + 10 + 5 = 20

5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 You can also get the max flow by summing the
capacity values going into the sink.

Max flow = 5 + 10 + 5 = 20

5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 10/15
S -10/0 -10/0 -10/0 -10/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Summary
A depth first search will sometimes find long
windy paths from the source to the sink.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 Summary
This is usually undesirable because the longer
the path, the smaller the bottleneck value,
which results in a longer runtime.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
0/10 0/0
 Summary
Edmonds-Karp tries to resolve this problem
by finding the shortest path from the source
to the sink using a breadth first search.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
 Summary
More importantly, the big achievement of
Edmonds-Karp is that its time complexity of
O(VE²) is independent of the max flow.
0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/5
0/0
-5/0 0/0 5/10

-5/0
5/5 0/5 0/5
-5/0 0/0
5/10
5/10 -5/0
Next Video: Edmonds-Karp Source Code
Graph Theory
Video Series
Network Flow:
Edmonds-Karp
source code
William Fiset
Previous video explaining
Edmonds-Karp
Source Code Link
Implementation source code can
be found at the following link:
github.com/williamfiset/algorithms

Link in the description:

Graph Theory
Video Series
 Network Flow:
Capacity Scaling

William Fiset
Let’s quickly revisit finding the maximum
flow using a DFS.

0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 Find the bottleneck (edge with smallest
remaining capacity) along the augmenting path.
min(6-0,1-0,7-0,11-0) = 1

0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10

bottleneck is 1 unit of flow!

Augment (update) the flow by adding the bottleneck value
to the flow along the forward edges and subtracting flow
by the bottleneck value along the residual edges.

0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10

Unfortunately, we are only able to augment

the flow along the path by one unit :/
 0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 Recall that the remaining capacity of an edge is
calculated as: capacity - flow. This allows residual edges
with negative flow to have a positive remaining capacity.
min(14-0,0-(-1),5-0,11-1) = 1

0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10

The bottleneck is still one unit of flow

going through the residual edge!
Augment (update) the flow by adding the bottleneck value
to the flow along the forward edges and subtracting flow
by the bottleneck value along the residual edges.

1/5

1/6 0 2
-1/0

0/0 1/7 -2/0

0/0
2/11
-1/0
S T
-1/0

0/12
-1/0
0/1 0/0
0/1
0/0
1/14
1 3
0/10

Unfortunately, we are only able to augment

the flow along the path by one unit :/
 1/5

1/6 0 2
-1/0

0/0 1/7 -2/0

0/0
2/11
-1/0
S T
-1/0

0/12
-1/0
0/1 0/0
0/1
0/0
1/14
1 3
0/10
You could imagine a DFS algorithm repeatedly taking an
edge with a capacity of 1, ultimately limiting how
much flow we can push through the network with each
iteration.
1/5

2/6 0 2
-1/0

-1/0 2/7 -3/0

0/0
3/11
-2/0
S T
-1/0

0/12
-2/0
1/1 0/0
0/1
0/0
1/14
1 3
0/10
You could imagine a DFS algorithm repeatedly taking an
edge with a capacity of 1, ultimately limiting how
much flow we can push through the network with each
iteration.
2/5

2/6 0 2
-2/0

0/0 2/7 -4/0

0/0
4/11
-2/0
S T
-2/0

0/12
-2/0
0/1 0/0
0/1
0/0
2/14
1 3
0/10
You could imagine a DFS algorithm repeatedly taking an
edge with a capacity of 1, ultimately limiting how
much flow we can push through the network with each
iteration.
2/5

3/6 0 2
-2/0

-1/0 3/7 -5/0

0/0
5/11
-3/0
S T
-2/0

0/12
-3/0
1/1 0/0
0/1
0/0
2/14
1 3
0/10
You could imagine a DFS algorithm repeatedly taking an
edge with a capacity of 1, ultimately limiting how
much flow we can push through the network with each
iteration.
3/5

3/6 0 2
-3/0

0/0 3/7 -6/0

0/0
6/11
-3/0
S T
-3/0

0/12
-3/0
0/1 0/0
0/1
0/0
3/14
1 3
0/10
Capacity scaling is the idea that we should prioritize
taking edges with larger capacities to avoid ending up
with a path that has a small bottleneck.

0/5

0/6 0 2
0/0

0/0 0/7 0/0

0/0
0/11
0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14
1 3
0/10
 If we adjust the size of each edge based on its
capacity value, then we can more easily visualize
which edges we should give more attention to.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
 Capacity Scaling Algorithm

Let U = the value of the largest edge

capacity in the initial flow graph.

Let Δ = the largest power of 2 less

than or equal to U.

The capacity scaling heuristic says

that you should only take edges whose
remaining capacity is ≥ Δ in order to
achieve a better runtime.
 Capacity Scaling Algorithm
The algorithm repeatedly finds augmenting
paths with remaining capacity ≥ Δ until no
more paths satisfy this criteria, then
decrease the value of delta by dividing it
by 2 (Δ = Δ / 2) and repeat while Δ > 0.

The capacity scaling works very well in

practice. In terms of time complexity,
capacity scaling with a DFS is bounded by
O(E²log(U)) or O(EVlog(U)) if the shortest
augmenting path (like Edmonds-Karp) is used
(although I've found that the latter seems
much slower in practice).
Let’s use capacity scaling to find the max flow
of the following flow graph!

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
First, compute U as the maximum of all initial capacities.

U = max(6,14,1,5,7,10,1,11,12) = 14

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
 Next, compute the starting value for Δ which is the
smallest power of 2 less than or equal to U (=14).
Δ = 2³ = 8, since the next power of 2 is 2⁴ = 16 > U

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

The bottleneck value is 10, because min(14-0,10-0,12-0) =

10 is the smallest remaining capacity along the path.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10

The bottleneck value will always be greater than or equal to Δ.

Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 8.

The bottleneck value is 10, because min(14-0,10-0,12-0) =

10 is the smallest remaining capacity along the path.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
 0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
There are no more augmenting paths from s -> t which have
a remaining capacity greater than or equal to Δ (=8), so
the new Δ is: Δ = Δ / 2 = 4.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

The bottleneck value is 5, because min(6-0,5-0,11-0) = 5

is the smallest remaining capacity along the path.

0/5
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

The bottleneck value is 5, because min(6-0,5-0,11-0) = 5

is the smallest remaining capacity along the path.

5/5
5/6 0 2
-5/0

0/0 0/7 -5/0

0/0 5/11

-5/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

5/5
5/6 0 2
-5/0

0/0 0/7 -5/0

0/0 5/11

-5/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

5/5
5/6 0 2
-5/0

0/0 0/7 -5/0

0/0 5/11

-5/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

5/5
5/6 0 2
-5/0

0/0 0/7 -5/0

0/0 5/11

-5/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

5/5
5/6 0 2
-5/0

0/0 0/7 -5/0

0/0 5/11

-5/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

The bottleneck value is 4, because min(14-10,7-0,11-5) =

4 is the smallest remaining capacity along the path.

5/5
5/6 0 2
-5/0

0/0 0/7 -5/0

0/0 5/11

-5/0
S T
-10/0

10/12
0/0
0/1 -10/0
0/1
-10/0
10/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 4.

The bottleneck value is 4, because min(14-10,7-0,11-5) =

4 is the smallest remaining capacity along the path.

5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
 5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
There are no more augmenting paths from s -> t which have
a remaining capacity greater than or equal to Δ (=4), so
the new Δ is: Δ = Δ / 2 = 2.
However, there are also no more augmenting paths from
s -> t which have a remaining capacity greater than or
equal to Δ (=2), so the new Δ is: Δ = Δ / 2 = 1.
5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.

5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.

5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.

5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.

5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.

5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.
The bottleneck value is 1, because min(6-5, 1-0, 7-4,
11-9) = 1 is the smallest remaining capacity along the
path.
5/5
5/6 0 2
-5/0

0/0 4/7 -9/0

0/0 9/11

-5/0
S T
-14/0

10/12
-4/0
0/1 -10/0
0/1
-10/0
14/14
1 3
10/10
Let’s find an augmenting path which has all edges with a
remaining capacity greater than or equal to Δ = 1.
The bottleneck value is 1, because min(6-5, 1-0, 7-4,
11-9) = 1 is the smallest remaining capacity along the
path.
5/5
6/6 0 2
-5/0

-1/0 5/7 -10/0

0/0 10/11

-6/0
S T
-14/0

10/12
-5/0
1/1 -10/0
0/1
-10/0
14/14
1 3
10/10
 There are no more augmenting paths from s -> t
which have a remaining capacity greater than or
equal to Δ (=1), so the new Δ is: Δ = Δ / 2 = 0,
which terminates the algorithm.

5/5
6/6 0 2
-5/0

-1/0 5/7 -10/0

0/0 10/11

-6/0
S T
-14/0

10/12
-5/0
1/1 -10/0
0/1
-10/0
14/14
1 3
10/10
The max flow value is the sum of the bottleneck
values found during each iteration:

Max flow = 10 + 5 + 4 + 1 = 20

5/5
6/6 0 2
-5/0

-1/0 5/7 -10/0

0/0 10/11

-6/0
S T
-14/0

10/12
-5/0
1/1 -10/0
0/1
-10/0
14/14
1 3
10/10
You can also get the max flow by summing the
flow values going into the sink

Max flow = 10 + 10 = 20

5/5
6/6 0 2
-5/0

-1/0 5/7 -10/0

0/0 10/11

-6/0
S T
-14/0

10/12
-5/0
1/1 -10/0
0/1
-10/0
14/14
1 3
10/10
 Summary
Ford-Fulkerson implemented with a DFS can
result in having a bottleneck value of 1 every
iteration for a complexity of O(fE).
0/5

1/6 0 2
0/0

-1/0 1/7 -1/0

0/0
1/11
-1/0
S T
0/0

0/12
-1/0
1/1 0/0
0/1
0/0
0/14
1 3
0/10
 Summary
Capacity scaling is when we push flow through
the larger edges first to achieve a better
runtime.

0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
 Summary
One approach to capacity scaling is to maintain
a decreasing parameter Δ (i.e by halving) which
acts as a threshold for which edges should be
accepted or rejected based on their remaining
capacity.
0/6 0 2
0/0

0/0 0/7 0/0

0/0 0/11

0/0
S T
0/0

0/12
0/0
0/1 0/0
0/1
0/0
0/14 1 3
0/10
Next Video: Capacity Scaling
Source Code
Graph Theory
Video Series
 Network Flow
Capacity Scaling
source code
William Fiset
Previous video explaining
capacity scaling:
Source Code Link
Implementation source code can
be found at the following link:
github.com/williamfiset/algorithms

Link in the description:

Graph Theory
Video Series
 Network Flow
Dinic’s Algorithm

William Fiset
 Dinic’s Algorithm
Dinic’s is a strongly polynomial maximum flow
algorithm with a runtime of O(V2E).

It is extremely fast in practice and works

even better on bipartite graphs giving a time
complexity of O(√VE) due to the algorithm’s
reduction to Hopcroft-Karp.

The algorithm was originally invented by

Yefim Dinitz in 1969 and published in 1970.
The algorithm was later modified slightly and
popularized by Shimon Even mispronouncing
“Dinitz’s algorithm” as “Dinic’s algorithm”.
 Dinic’s Intuition Analogy

Suppose you and a friend planned to meet up at

a coffee shop a few streets east of where you
are. You have never been to this coffee shop
and you don’t exactly know where it is, but
you know it’s somewhere east, how would you
get there?

You Coffee shop

Dinic’s Intuition Analogy

Q: With the information you have, would

it make sense to head south? What about
north west?

You Coffee shop

Dinic’s Intuition Analogy
The only sensible directions are: east,
north east and south east, this is because
you know that those directions guarantee
positive progress towards the coffee shop.

You Coffee shop

Dinic’s Intuition Analogy
This form of heuristic ensures that we
continuously make progress towards whatever
place of interest we desire to go, how can
we apply this to solving maximum flow?

You Coffee shop

 Dinic’s Intuition Analogy
In this analogy, you are the source and the
coffee shop is the sink. The main idea behind
Dinic’s algorithm is to guide augmenting paths
from s -> t using a level graph, and in doing
so greatly reducing the runtime.

You Coffee shop

The way Dinic’s determines which edges make progress
towards the sink T and which do not is by building a
level graph.

S T
 The levels of the graph are those obtained by
doing a BFS from the source.

L = 0 L = 1 L = 2 L = 3 L = 4
1 2 3

S 1 2 3 T

1 2 3
 Furthermore, an edge is only part of the level graph if it
makes progress towards the sink. That is, the edge must go
from a node at level L to another at level L+1.

L = 0 L = 1 L = 2 L = 3 L = 4
1 2 3

S 1 2 3 T

1 2 3
 The requirement that edges must go from L to
L+1 prunes backwards and “sideways” edges
(grey edges in this slide).

L = 0 L = 1 L = 2 L = 3 L = 4
1 2 3

S 1 2 3 T

1 2 3
 Q: If we're trying to go from s -> t as quickly
as possible, does it make sense to take the red
edge going in the backwards direction?

S T
Take this edge?
A: No, taking the red edge doesn’t bring you closer
to the sink, so it should only be taken if a detour
is required. This is why backwards edges are omitted
in the level graph.

S T
Take this edge?
 The same thing can be said about edges
which cut across sideways to the sink as
no progress is made.

S T
It’s also worth mentioning that residual (dotted) edges
can be made part of the level graph, but they must have
a remaining capacity (capacity - flow) > 0.
L = 0 L = 1 L = 2 L = 3 L = 4
1 2 3

S 1 2 3 T

1 2 3
 Dinic’s Algorithm Steps
Step 1: Construct a level graph by doing a
BFS from the source to label all the levels
of the current flow graph.

Step 2: If the sink was never reached while

building the level graph, then stop and
return the max flow.

Step 3: Using only valid edges in the level

graph, do multiple DFSs from s -> t until a
blocking flow is reached, and sum over the
bottleneck values of all the augmenting
paths found to calculate the max flow.

Repeat Steps 1 to 3
 Let’s run through an example of using Dinic’s
algorithm to find the maximum flow on the
following flow graph.

0/10 0/0
0 3 6
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
4 7 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
2 0/25
5 8
0/0
 Begin by constructing the level graph for
the current flow graph.

0/10 0/0

0/5 0/0 0/10

0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 0/0 0/0 0/0
T
0/20
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
0/25 0/0
 Begin by constructing the level graph for
the current flow graph.

0/10 0/0
1
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0 0/0 0/0
T
0/20
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25 0/0
 Begin by constructing the level graph for
the current flow graph.

0/10 0/0
1 2
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2
0/0
 Begin by constructing the level graph for
the current flow graph.

0/10 0/0
1 2 3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 Begin by constructing the level graph for
the current flow graph.

0/10 0/0
0
1 3
2 6
3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
4
2 7
3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
2
1 0/25
5
2 8
3
0/0
 The level graph consists of all edges which go from
L to L+1 in level and have remaining capacity > 0.

0/10 0/0
0
1 3
2 6
3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
4
2 7
3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
2
1 0/25
5
2 8
3
0/0
 Find paths from s -> t until a blocking flow is
reached (i.e. you cannot find any more paths).

0/10 0/0
0
1 3
2 6
3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
4
2 7
3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
2
1 0/25
5
2 8
3
0/0
 Find paths from s -> t until a blocking flow is
reached (i.e. you cannot find any more paths).

0/10 0/0
0
1 3
2 6
3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
4
2 7
3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
2
1 0/25
5
2 8
3
0/0
 Find paths from s -> t until a blocking flow is
reached (i.e. you cannot find any more paths).

0/10 0/0
1 2 3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 Find paths from s -> t until a blocking flow is
reached (i.e. you cannot find any more paths).

0/10 0/0
1 2 3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 The current path has a bottleneck value of 5
since min(5-0,10-0,10-0,5-0) = 5.

0/10 0/0
1 2 3
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 Augment the flow values along the path by 5.

5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 The blocking flow has not yet been reached since
there still exists paths from s -> t.

5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 The current path has a bottleneck value of 15
since min(15-0,25-0,20-0,15-0) = 15.

5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
2 3 T
0/0 0/20 0/0
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0 0/0
0/10
1 0/25
2 3
0/0
 Augment the flow values along the path by 15.

5/10 -5/0
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 15/15
0/20 0/30
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
1 2 3
15/25 0/0
 Let’s try and find another path from s -> t.

5/10
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 15/15
0/20 0/30
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
1 2 3
15/25 0/0
 5/10
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 15/15
0/20 0/30
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
1 2 3
15/25 0/0
 5/10
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 15/15
0/20 0/30
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
1 2 3
15/25 0/0
 5/10
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 15/15
0/20 0/30
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
1 2 3
15/25 0/0
 We get stuck finding a path, so the blocking flow has
been reached.

5/10
1 2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 15/15
0/20 0/30
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
1 2 3
15/25 0/0
 5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 0/0 0/0 -15/0
T
15/20
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
15/25 0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 0/0 0/0 -15/0
T
15/20
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
15/25 0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0
0 3 6
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
4 7 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
2 15/25
5 8
0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0

5/5 -5/0 5/10

0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0 0/0 -15/0
T
15/20
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
15/25 0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0
2
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
15/25 0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0
2 3
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25 0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Rebuild the level graph. This time it should look
different because the remaining capacities of
multiple edges has changed.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 The level graph consists of all edges which go from
L to L+1 in level and have remaining capacity > 0.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Let’s try and find a path from s -> t.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Let’s try and find a path from s -> t.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Let’s try and find a path from s -> t.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Let’s try and find a path from s -> t.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Oops, we reached a dead end in our DFS, so backtrack
and keep going until we reach the sink.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 The current path has a bottleneck value of 10
since min(10-0,20-0,30-0,15-0,10-0) = 10.

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 0/10 0/20 0/30 15/15
S 0/0 1 0/0
2 3 T
0/0 15/20 -15/0
0/0
-15/0 0/0 0/10

15/15 0/5 0/0

0/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 Augment the flow values along the path by 10

5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 15/15
S -10/0 1 2 3 T
-10/0 -10/0 15/20 -15/0
0/0 -10/0
-15/0 10/10

-10/0
15/15 0/5 10/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 15/15
S -10/0 1 2 3 T
-10/0 -10/0 15/20 -15/0
0/0 -10/0
-15/0 10/10

-10/0
15/15 0/5 10/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 The blocking flow has been reached and no
more flow can be pushed through the
network, so the algorithm terminates.
5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 15/15
S -10/0 1 2 3 T
-10/0 -10/0 15/20 -15/0
0/0 -10/0
-15/0 10/10

-10/0
15/15 0/5 10/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
The max flow is the sum of the bottleneck values:

Max flow = 5 + 15 + 10 = 30
5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 15/15
S -10/0 1 2 3 T
-10/0 -10/0 15/20 -15/0
0/0 -10/0
-15/0 10/10

-10/0
15/15 0/5 10/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 The max flow can also be calculated by looking at
the flow values of the edges leading into the sink.
Max flow = 5 + 15 + 10 = 30
5/10 -5/0
2 3 4
5/5 -5/0 5/10
0/0 -5/0
0/15 0/15

5/5
0/0 0/25 0/0
-5/0 10/10 10/20 10/30 15/15
S -10/0 1 2 3 T
-10/0 -10/0 15/20 -15/0
0/0 -10/0
-15/0 10/10

-10/0
15/15 0/5 10/15
-15/0 -15/0
0/10
3 15/25
4 4
0/0
 History & Optimizations
One of the pitfalls of the current
implementation of Dinic’s algorithm at the
moment is that the algorithm may encounter
multiple dead ends during the DFS phase.

2 3 4

S 1 2 3 T

3 4 4
 History & Optimizations
Dead ends are highly undesirable because they
extend the length of the journey from the
source to the sink, especially if the same
dead end is taken multiple times during a
blocking flow iteration.

2 3 4

S 1 2 3 T

3 4 4
 History & Optimizations

In Dinitz’s original paper, he suggested

“cleaning” the level graph ridding it of all
dead ends before each blocking flow phase.

In 1975, Shimon Even and Alon Itai

suggested pruning dead ends when
backtracking during the DFS phase.

This trick greatly speeds up

(and simplifies) the algorithm
because dead ends are only ever
encountered once.
 Summary
A good pathfinding heuristic which ensures
that we continuously make progress towards a
goal will greatly speed up an algorithm.

Coffee shop
You
 Summary
Dinic’s algorithm uses a BFS to construct
a level graph which directs edges towards
the sink.

0/10 0/0

0/5
0
1 0/0
3
2 6
3
0/10
0/0 0/0
0/15 0/15

0/5
0/25 0/0
0/0
0/0 0/10
0/20 0/30 0/15
S 0/0 1 0/0
4
2 0/0
7
3 0/0
T
0/20
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0
0/0 0/10
2
1 0/25
5
2 8
3
0/0
 Summary
In the next phase of the algorithm,
multiple DFSs are run on the level graph
until a blocking flow is reached.
0/10 0/0

0/5
0
1 0/0
3
2 6
3
0/10
0/0 0/0
0/15 0/15

0/5
0/25 0/0
0/0
0/0 0/10
0/20 0/30 0/15
S 0/0 1 0/0
4
2 0/0
7
3 0/0
T
0/20
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0
0/0 0/10
2
1 0/25
5
2 8
3
0/0
 Summary
The process of rebuilding the level graph
and finding the blocking flow is repeated
until no more augmenting paths exist and the
max flow is found.
0/10 0/0

0/5
0
1 0/0
3
2 6
3
0/10
0/0 0/0
0/15 0/15

0/5
0/25 0/0
0/0
0/0 0/10
0/20 0/30 0/15
S 0/0 1 0/0
4
2 0/0
7
3 0/0
T
0/20
0/0
0/0 0/0 0/10

0/0
0/15 0/5 0/15
0/0
0/0 0/10
2
1 0/25
5
2 8
3
0/0
 Summary
A critical optimization of Dinitz’s
algorithm is pruning dead ends so that you
do not encounter them again.

2 3 4

S 1 2 3 T

3 4 4
Graph Theory
Video Series
 Network Flow:
Dinic’s Algorithm
source code
William Fiset
Previous video explaining
Dinic’s algorithm:
Source Code Link
Implementation source code can
be found at the following link:
github.com/williamfiset/algorithms

Link in the description:

 Let’s run through an example of using Dinic’s
algorithm to find the maximum flow on the
following flow graph.

0/10
0 3 6
0/5 0/10

0/15 0/15

0/5
0/25
0/10 0/15
0/20 0/30
S 1 4 7 T
0/20
0/10

0/15 0/5 0/15

0/10
2 0/25
5 8
Graph Theory
Video Series
 Network Flow:
Mincost Maxflow

William Fiset
 What’s
TODO list
up with negative cycles?

How to draw an edge with cap and cost?

Mention residual edges have negative cost

figure out which examples to use.

Implement dummy S->T max flow + negative

cycle cancelation.

Mention most of the time edge cost are > 0

and the problem is solved with SSSP
algorithms.

What is the bigger picture here? What

general class of problems is MCMF solving?
We have these graphs with cap and cost on
the edges. rn I know how to solve this
when the cost is >= 0 because it’s a
reduction to SSSP which is awesome and
really what we care about most of the
time. But, when cost < 0 shit goes crazy
because now we can have negative cycles —
oh shit right. To handle these we need to
use Klien’s cycle canceling method
(analogous to FF for max flow). Basically
Kliens finds cycles and augments the flow
as per my understanding. Now finding these
negative cycles is a hard problem
(computationally intensive that is) so
some fancy Network Simply algorithm can
find and cancel negative cycles on the
fly. This is the boss of all graph
algorithms.
 A mincost flow graph cannot contain a
directed negative cycle, otherwise the
optimal cost is unbounded.

It is safe to assume all edge costs are

non-negative because we don’t lose
anything by doing so, only need to
tranform. Also edges must be integral for
this to work.
Klien’s Cycle Canceling
Method

illustrate that there can be multiple

graphs with the same “flow” but
different costs?
 TODO list

illustrate that there can be multiple

graphs with the same “flow” but
different costs?
 Negative Cycles

"Because of negative edge costs, these

networks can have negative-cost cycles. The
concept of negative cycles, which seemed
artificial when we first considered it in
the context of shortest-path algorithms,
plays a critical role in mincost-flow
algorithms we now see."

Algorithms In C++ third Edition, Part 5 Graph

Algorithms, P446. Robert Sedgewick.
 Negative Cycles

"A maxflow is a mincost maxflow if and only

if its residual network contains no
negative-cost (directed) cycles."

Algorithms In C++ third Edition, Part 5 Graph

Algorithms, P446. Robert Sedgewick.
Graph Theory
Video Series
 Network Flow:
Mincost Maxflow
Source Code
William Fiset
 TODO list

Make sure to show changes to the

NetworkFlowSolverBase. In particular
mention the addEdge method and the
negative cost edge there and the new cost
property added to edges.
Next Video: Dinic’s Source Code
Unused slides
 Max Flow
0/4 0/3
0 7
5
0/12
0/5 0/3 0/14 0/8
0/7

6 0/9

0/4
0/20 1 0/15 t
s
0/14 0/22
8
4
0/60
0/5
0/10 0/11
0/3

9
3
2 0/4
0/11
Would have been a good example but it’s too
large and might confuse some folks

1 4 7 T
88/88 88/88

88/88
88/88

88/88
0 3 0/0
6 9

S 2 5 8
 0/10 0/0
0 3 6
0/5 0/0 0/10
0/0 0/0
0/15 0/15

0/5
0/0 0/25 0/0
0/0 0/10 0/20 0/30 0/15
S 0/0 1 0/0
4 7 T
0/0 0/0
0/5
0/0
0/0 0/0 0/10

0/0
0/5 0/5 0/5
0/0 0/0
0/10
2 0/10
5 8
0/0