Maximum Flow

Overview

• Network flow
– Use a graph to model material that flows through conduits
– Each edge represents one conduit, and has a capacity ,which is an uppe
r bound on the flow rate = units /time
– Can think of edges as pipes of different sizes. But flows don’t have to b
e of liquids. Book has an example where a flow is how many trucks per
day can ship hockey pucks between cities.
– Want to compute max rate that we can ship material from a designated
source to a designated sink
 Flow networks

G = (V , E) directed.
Each edge (u , v) has a capacity c(u,v)  0
If (u , v)  E ,then c (u , v) = 0.
Source vertex s, sink vertex t, assume s v t for all v V
Example: [Edge are labeled with capacities.]

2
3
w x 3

s 1 3 1 3 2 t
2
y 3
z 2
• Positive flow :A function p : V × V → R satisfying
– Capacity constraint: For all u, v V ,0  p (u, v )  c(u, v )
– Flow conservation: For all u V  {s, t}, 
p( v, u ) 
vV

p ( u, v )
vV
flow into u flow out of u

Equivalently,  p ( u, v )   p ( v , u )  0
vV vV
 2/2
w x 2 /3
1/ 3

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

2 /2
 y 2/3
z 1 /2

• Note that all positive flow are  capacities.

• Verify flow conservation by adding up flow at a couple of vertices
• Note that all positive flow = 0 is legitimate
Cancellation with positive flows

• Without loss of generality, can say positive flow goes either from u to v or
from v to u ,but not both.( Because if not true, can transform by cancellatio
n to be true.)
• In the above example, we can “cancel” 1 unit of flow in each direction bet
ween x and z
1 unit x  z 0 unit x  z

2units z  x 1 unit z  x
• In both case, ”net flow” is 1 unit z → x
• Capacity constraint is still satisfied (because flow only decrease)
• Flow conservation is still satisfied (flow in and flow out are both reduced b
y the same amount)
Here’s a concept similar to positive flow:
Netflow: A function f : V × V → R satisfying
• Capacity constraint: For all u, v V , f (u, v )  c(u, v )
• Skew symmetry: For all u, v V , f (u, v )   f ( v, u )
• Flow conservation: For all u V  {s, t},  f ( u, v )  0
vV
Another way to think of flow conservation:

 f ( v , u )   f ( u, v )
vV : f ( v ,u ) 0 vV : f ( u ,v ) 0
total positive total positive
flow entering u flow leaving u

“flow in =flow out”

The differences between positive flow p and net flow f
– p ( u, v )  0
– f satisfies skew symmetry
Equivalence of positive flow and net flow definitions

Define net flow in terms of positive flow:

• Define f( u , v) = p( u ,v) - p( v ,u)
• Argue, given definition of p ,that this definition of f satisfies capacity const
raint and flow conservation
Capacity constraint:
p(u, v )  c(u, v ) and p( v, u )  0  p(u, v )  p( v, u )  c(u, v )
Flow conservation:
 f (u, v )   ( p(u, v )  p(v, u))
vV vV

  p ( u, v )   p ( v , u )
vV vV

0
• Skew symmetry is trivially satisfied by the definition of
f ( u, v )  p ( u, v )  p ( v , u )
 ( p( v, u )  p(u, v ))
  f ( v, u )
Define positive flow in terms of net flow:

p(u,v )  {0 f (u,v ) if f (u, v )  0,

if f (u, v )  0.
• Argue, given definition of f , that this definition of p satisfies capacity
constraint and flow conservation.
Capacity constraint:
If f (u, v )  0 :
f ( u, v )  c ( u, v )  0  p ( u, v )  c ( u, v )
If f (u, v )  0 :
0  p (u, v )  c(u, v ).
Flow conservation :
 p (u, v )   p ( v , u )
vV vV
(  p (u, v )   p (u, v ))
vV : f ( u ,v ) 0 vV : f ( u ,v ) 0
 (  p ( v, u )   p ( v, u ))
vV : f ( u ,v ) 0 vV : f ( u ,v ) 0
  p (u, v )   p (u, v )
vV : f ( u ,v ) 0 vV : f ( u ,v ) 0
  p( v, u )   p( v, u )
vV : f ( v ,u )0 vV : f ( v ,u ) 0
  p (u, v )  0  0   p ( v , u )
vV : f ( u ,v ) 0 vV : f ( v ,u ) 0
  f ( u, v )   f ( v , u )
vV : f ( u ,v ) 0 vV : f ( v ,u ) 0
  f (u, v )   (  f (u, v ))
vV : f ( u ,v ) 0 vV : f ( u ,v ) 0

  f ( u, v )
vV

0
Value of flow f  f   f ( s, v )  total flow out of source
vV

Consider the example blow. [The cancellation possible in the previous

example has been made here. Also, showing only flows that are positive.]
2/2
w x 2 /3
1 /3

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

2 /2
y 2/3
z 1 /2

Value of flow f  f  3
Cancellation with flow

If we have edges (u , v) and (v , u) ,skew symmetry makes it so that at most \

one of these edges has positive flow.
Say f (u ,v) = 5 .If we “ship” 2 units v → u. we lower f (u ,v) to 3.The 2 units
v → u cancel 2 of the u → v units.
Due to cancellation, a flow only gives us this “net” effect. We cannot
reconstruct actual shipments from a flow
5 units u  v 8 units u  v
same as
0 units v  u 3 units v  u
We could add another 3 units of flow u → v and another 3 units
v → u ,maintaining flow conservation.
The flow from u to v would remain f (u ,v) = 5 ,and f (v ,u) = -5.
Maximum-flow problem: Given G,s,t and c, find a flow whose value is
maximum.

Implicit summation notation

We work with function, like f ,that take pairs of vertices as arguments. Extend
to take sets of vertices, with interpretation of summing over all vertices in the
set.
Example: If X and Y are sets of vertices,
f ( X , Y )   f ( x, y )
xX yY

Therefore, can express flow conservation as f (u,V )  0, for all u V  s, t

Notation: Omit braces in sets with implicit summations.
Example: f ( s,V  s )  f ( s,V ). Here, f ( s,V  s ) means f ( s,V  {s})
Lemma
For any flow f in G  (V , E )
1.For all X  V , f ( X , X )  0,
2.For all X , Y  V , f ( X , Y )   f (Y , X ),
3.For all X , Y , Z  V such that X  Y   ,
f ( X  Y , Z )  f ( X , Z )  f (Y , Z ) and
f ( Z , X  Y )  f ( Z , X )  f (Z ,Y )
Proof

2. f ( X , Y )   f ( x, y )
xX yY
   f ( y , x ) (skew symmetry)
xX yY
   f ( y , x )
xX yY
  f (Y , X )
Proof
1. f ( X , X )   f ( X , X ) (part(2))
 f (X , X )  0
3. f ( X  Y , Z )    f ( v, z )
v  X Y zZ
   
    f ( v, z )      f ( v, z )  ( X  Y )  
vX  zZ  vY  zZ 
 f ( X , Z )  f (Y , Z )
Lemma f  f (V , t )
Proof First, show that f (V ,V  s  t )  0
f (u,V )  0 for all u V  {s, t}
 f (V  s  t ,V )  0 (add up f (u,V ) for all u V  {s, t})
 f (V ,V  s  t )  0 (by lemma, part (2)).

Thus, f  f ( s ,V ) (definition of f )
 f (V ,V )  f (V  s,V ) (lemma, part(3))
  f (V  s,V ) (lemma, part(1))
 f (V ,V  s ) (lemma, part(2))
 f (V , t )  f (V ,V  s  t ) (lemma, part(3))
 f (V , t ) (from above)
Cuts

A cut (S,T) of flow network G = (V,E) is a partition of V into S and T=V-S

such that s  S , t  T
• Similar to cut used in minimum spanning trees, except that here the graph
is directed, and we require s  S , t  T
For flow f, the net flow across cut (S , T) is f (S , T).
Capacity of cut (S , T) is c (S , T).
A minimum cut of G is a cut whose capacity is minimum over all cuts of G.
For our example: 2/2
w x 2/3
1/ 3

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

2/ 2
y 2/3
z 1/2

Consider the cut S  s, w, t, T  x, z, t

f ( S , T )  f ( w, x )  f ( y , x )  f ( y , z )
 2  ( 1)  2
3
c( S , T )  c( w, x )  c( y , z )
23
5

Same flow as previous cut, higher capacity.

Note the difference between capacity and flow.
• Flow obeys skew symmetry, so f (y , x) = -f (x , y) = -1.
• Capacity does not: c (y , x) = 0 ,but c (x , y) = 1.
So include flows going both ways across the cut, but capacity going only S to
T.
Now consider the cut
f ( S , T )  f ( x, z )  f ( x, t )  f ( y , z )
 ( 1)  2  2
3
c( S , T )  c( x, z )  c( x, t )  c( y , z )
 2  3 3
8
Lemma
For any cut (S,T), f ( S , T )  f
Proof First, show that f ( S  s,V )  0
S  s V  s, t
Therefore,
f ( S  s ,V )   f (u,V )
uS  s
 0 (flow conservation and S - s V  s, t)
uS  s
0
So
f ( S , T )  f ( S ,V )  f ( S , S ) (lemma, part(3), and
S  T  V, S  T   )
 f ( S ,V ) (lemma, part(1))
 f ( s,V )  f ( S  s,V ) (lemma, part(3))
 f ( s ,V ) ( f ( S  s , V )  0)

 f
Corollary
The value of any flow  capacity of any cut.
Proof Let (S,T) be any cut, f be any flow.
f  f (S,T ) (lemma)
  f (u, v )
uS vT
  c(u, v ) (capacity constraints)
uS vT
 c(S,T)

Therefore, maximum flow  capacity of minimum cut. Will see a little later
that this is in fact an equality.
The Ford-Fulkerson Method

Residual network
Given a flow f in network G = (V , E)
Consider a pair of vertices u , v  V.
How much additional flow can we push directly from u to v ?
That’s the residual capacity.
cf ( u , v )  c ( u , v )  f ( u , v )
0 (since f (u, v )  c(u, v ))
Residual network: Gf  (V , Ef )
Ef  (u, v ) V  V : cf (u, v )  0

Each edge of the residual network can admit a positive flow.

For our example:
2
2
w x 1

2
1
s 1 2 1 2 3 t
1
1
3
y z 1
2

Every edge (u , v)  Ef corresponds to an edge (u , v) E or (u , v) E (or both)

Therefore, Ef  2 E
Given flows f1 + f2 ,the flow sum f1 + f2 is defined by
( f1 + f2 ) (u , v) = f1 (u , v) + f2 (u , v)
For all u , v  V
Lemma
Given a flow network G, a flow f in G, and the residual network Gf ,let f’ be
any flow in Gf .Then the flow sum f + f’ is a flow in G with value
f  f'  f  f'

[See book for proof]

Augmenting path
A path s t in Gf .
• Admits more flow along each edge
• Like a sequence of pipes through which we can squirt more flow from s to
t
How
cf ( pmuch
)  mincf (flow
more u, v )can
: (uwe
, v )push p s to t along augmenting path p?
is onfrom
 For our example, consider the augmenting path p  s, w, y , z, x, t
Minimum residual capacity is 1.
After we push 1 additional unit along p:

2/2
w
2/2
x 2/3
w x 3/3
1/3 2/3

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

2/2 2/2
y 2/3
z 1/2
y 3/3
z 1/2
Observe that Gf now has no augmenting path. Why? No edges cross the cut

({ s , w },{ x , y , z , t }) in the forward direction in Gf .So no path can get fro

m
s to t. Claim that the flow shown in G is a maximum flow.
2
1
w x 3

s 3 1 1 4 t
1

3 1
y 3
z
Lemma
Given flow network G, flow f in G, residual network Gf. Let p be an
augmenting path in Gf . Define fp : V × V → R:
cf ( p ) if (u, v ) is on p,

fp (u, v )   cf ( p ) if ( v, u ) is on p,
0 otherwise.


fp  cf ( p )  0
Then fp is a flow in Gf with value
Corollary
Given flow network G, flow f in G, and a augmenting path p in Gf ,define fp as
in lemma, and defin f’ : V × V → R by f’ = f + fp. Then f’ is a flow in G with
Value f '  f  cf ( p )  f
Theorem (Max-flow min-cut theorem)
The following are equivalent:
1. f is a maximum flow.
2. f admits no augmenting path
3. |f| = c (S , T) for some cut (S , T)

Proof
(1)  ( 2)
If f admits an augmenting path p ,then (by above corollary) would get a flow
with value |f| + cf(p) > |f| , so f wasn’t a max flow to start with.
( 2)  (3)
Suppose f admits no augmenting path. Define
S  v V : there exists a path s  v in Gf 
T V  S
Must have t  T ;otherwise there is an augmenting path.
Therefore, ( S , T ) is a cut.
For each u  S and v  T , must have f ( u , v) = c (u , v) , since otherwise
(u , v) Ef and the v S
By lemma (f ( S , T ) = |f| ), |f| = f ( S , T ) = c ( S , T ).
(3)  (1)
By corollary, f  c( S , T )
f  c( S , T )  f is a max flow
Ford-Fulkerson algorithm

FORD-FULKERSON (V , E , s , t)
for all (u , v)  E
do f [u , v] ← f [v , u] ← 0
while there is an augmenting path p in Gf
do augment f by cf (p)

Subtle different between f [u , v] and f (u , v):

• f (u , v) is a function, defined on all u ,v V
• f [u , v] is a value computed by algorithm.

– f [u , v] = f (u , v) where (u , v) E or (v , u) E

– f [u , v] is undefined if neither (u , v) E nor (v , u)  E
Analysis: If capacities are all integer, then each augmenting path raises f by  1
If max flow is f* , then need  |f*| iterations  time is O  E f 
*

 
Now that this running time is not polynomial in input size. It depends on
|f*| ,which is not a function of |V| and |E|.
If capacities are rational, can scale them to integers.
If irrational, FORD-FULKERSON might never terminate.
Edmonds-Karp algorithm

Do FORD-FULKERSON, but compute paths by BFS of Gf , Augmenting

paths are shortest path s t in Gf . With all edge weights = 1.
Edmonds-Karp runs in O(V E2) time.
To prove, need to look at distance to vertices in Gf.
Let δf (u , v) = shortest path distance u to v in Gf , with unit edge weights.
Lemma
For all v V  s, t, f s, v  increases monotonically with each flow
augmentation.
Proof
Suppose there exists v V  s, t such that there is a flow augmentation that
cause δf (s , v) to decrease. Will derive a contradiction.
Let f be the flow before the first augmentation that causes a shortest-path
distance to decrease, f’ be the flow afterward.
Let v be a vertex with minimum δf’ (s , v) whose distance was decreased by
the augmentation, so δf’ (s , v) < δf (s , v) .

Let a shortest path s to v in Gf’ be s u → v, so (u , v) Ef ’ and
δf ‘(s , v) =δf ’ (s , u) + 1. (Or δf’ (s , u) = δf’ (s , v) - 1 .)
Since δf ‘ (s, u) < δf ‘ (s , v) and how we chose v, we have
δf ‘ (s , u) δf (s , v) .
Claim (u, v )  Ef
Proof If (u, v )  Ef ,then
f ( s, v )  f ( s, u )  1 ( triangle inequality)
 f ' ( s , u )  1
 f ' ( s , v )
contradicting f ' ( s, v )  f ( s, v )

How can (u, v )  Ef and (u, v )  Ef ' ?

The augmentation must increase flow v to u. Since Edmonds-Karp augment
s
along shortest paths, the shortest path s to u in Gf has v → u as its last edge.
Therefore,
f ( s , v )  f ( s , u )  1
 f ' ( s , u )  1
 f ' ( s , v )  2
contradict ing f ' ( s, v )  f ( s, v )

Therefore, v cannot exist.

Theorem
Edmond-Karp performs O(V E) augmentations.
Proof Suppose p is an augmenting path and cf (u , v) = cf (p). Then call (u , v) a
critical edge in Gf , And it disappears from the residual network after an
augmentation along p .
 1 edge on any augmenting path is critical.
Will show that each of the |E| edges can become critical  |V|/2 – 1 times.
Consider u,v V such that either (u , v) E or (v , u) E or both. Since
augmenting paths are shortest paths, when (u , v) becomes critical first time,
δf (s , v) = δf (s , u) + 1
Augment flow, so that (u , v) disappears from the residual network. The edge
cannot reappear in the residual network until flow from u to v decreases,
which happens only if (v , u) is on an augmenting path in
Gf ‘ :δf ‘ (s , u) = δf ‘ (s , v) +1 ( f ‘ is flow when this occurs)
By Lemma, f ( s, v )  f ' ( s, v ) 
f ' ( s , u )  f ' ( s , v )  1
 f ( s , v )  1
 f ( s , u )  2
Therefore, from the time (u , v) becomes critical to the next time, distance of
u from s increases by  2.Initially, distance to u is  0, and augmenting path
can’t have s, u, and t as intermediate vertices.
Therefore, until u becomes unreachable from source, its distance is  V  2  u
can become critical  V  2 / 2  V / 2  1 times
Since O(E ) pairs of vertices can have an edge between then in residual graph,
total # of critical edges is O(V E). Since each augmenting path has  critical
1
edge, have O(V E) augmentations.
Use BFS to find each augmenting path in O(E ) time  O(V E2) time.
Can get better bounds.
Push-relabel algorithms in Sections 26.4-26.5 give O(V3).
Can do even better.
Maximum bipartite matching

Example of a problem that can be solved by turning it into a flow problem.

G = (V , E) (undirected) is bipartite if we can partition V = L ∪ R such that all
edges in E go between L and R.

L R L R
matching maximum matching
A matching is a subset of edges M  E such that for all v V ,  1 edge of M
is incident on v .(Vertex v is matched of an edge M is incident on it, otherwise
unmatched).

Maximum matching: a matching of maximum cardinality. (M is a maximum

matching if |M|  |M’| for all the matchings M’.)
Problem:
Given a bipartite graph (with the partition), find a maximum matching.
Application:
Matching planes to routes.
• L = set of planes.
• R = set pf routes.
• (u , v) E if plane u can fly route v .
• Want maximum # of routes to be served by planes.
Given G, define flow network G’ = (V’ , E’)
• V’ = V ∪ {s , t}
• E '  s, u  : u  L
 u, v  : u  L, v  R, u, v  E 
 u, t  : v  R

• c(u, v )  1 for all (u, v )  E