Minimum Cost Flows

Primal Dual Algorithm

and the Out-of-Kilter Algorithm

Overview of lecture
• Quick review of min cost flow and optimality conditions

• Primal Dual algorithm

• Out-of-kilter algorithm

2
Optimality Conditions
Theorem. Suppose that x* is flow and π* is a vector of node
potential. Then x* and π* are optimal if
1. the flow x* is feasible and
2. cπ*ij ≥ 0 for all (i, j) ∈ G(x*).
The second property is often called dual feasibility.

3
Optimality Conditions
The cycle canceling algorithm starts with a feasible
flow and maintains a feasible flow at each iteration.
It terminates with dual feasibility.

The successive shortest path algorithm starts with

(and maintains) an infeasible flow x and node
potentials π that are dual feasible. It stops when it
obtains a flow that is feasible.

It’s main subroutine involves sending flow along a 4

path from a node with excess to a node with deficit.

 b(i) b(j)
Primal –Dual Algorithm   (, )
i j
• min-cost  max-flow
• single excess node and single deficit
node.
2 -2
  (, )
u w
  (,
) )
(  ,
v z
  (, )
2 -2

5
 b(i) b(j)
Primal –Dual Algorithm   (, )
i j
• min-cost  max-flow
• single excess node and single deficit
node.
2 -2
 Add nodes s and t.
  (, )
u w
s   (, t
) )
(  ,
v z
  (, )
2 -2

6
 b(i) b(j)
Primal –Dual Algorithm   (, )
i j
•  min-cost  max-flow
• single excess node and single deficit
node.
2 -2
 Add nodes s and t.
  (, )
4 ,) u w
 For every v such that b(v)>0:  (
Add an edge (s,v) such that u(s,v)=b(v) s   (, t
) )
 b(s) =   (, (  ,
) v z
  (, )
2 -2

7
 b(i) b(j)
Primal –Dual Algorithm   (, )
i j
•  min-cost  max-flow
• single excess node and single deficit
node.
2 -2
  (, ) -4
 Add nodes s and t.
4 ,) u  
w (,
( )
 For every v such that b(v)>0:  
Add an edge (s,v) such that u(s,v)=b(v) s   (, t
) )
 b(s) =   (,   (, )
) v z  ( ,
 For every v such that b(v)<0:
Add an edge (v,t) such that u(v,t)= -b(v)
  (, )
 b(t) =
2 -2

8
 b(i) b(j)
Primal –Dual Algorithm   (, )
i j
•  min-cost  max-flow
• single excess node and single deficit
node.
0 0
  (, ) -4
 Add nodes s and t.
4 ,) u w   (,
( )
 For every v such that b(v)>0:  
Add an edge (s,v) such that u(s,v)=b(v) s   (, t
) )
 b(s) =   (,   (, )
) v z  ( ,
 For every v such that b(v)<0:
Add an edge (v,t) such that u(v,t)= -b(v)
  (, )
 b(t) =
0 0
 For all
9
 b(i) b(j)
Primal –Dual Algorithm   (, )
i j
• min-cost  max-flow
• single excess node and single deficit
node.
0 0
• a minimum cost flow in the
  (, )   -4
4 ,) u w (,)
transformed network gives a  (
minimum cost flow in the original s   (, t
) )
network   (,   (, )
) v z  ( ,
• There exists a feasible solution iff all   (, )
the edges from s are saturated. 0 0

10
Primal –Dual Algorithm
• Similar to the previous SSP :
• Start with optimal (RC) feasible pseudoflow
• At every step keeps optimality (RC) and tries to build a feasible
flow.

• Instead of sending flow along one shortest path at a time it solve

maximum flow that send flow along all shortest paths.

11
Primal –Dual Algorithm
•  (x) Admissible network :

• will contain only arcs in G(x) with =0

• =

• Every directed path from node s to node t in is a shortest path in

G(x) between the same pair of nodes.

12
Primal –Dual Algorithm
• algorithm
  primal-dual;
• begin
• x: = 0 and : = 0;
• e(s) : = b(s) and e(t) : = b(t);
• while e(s) > 0 do
• begin
• determine shortest path distances d(·) from node s to all other nodes
in G(x) with respect to the reduced costs ;
• update : = - d;
• define the admissible network ;
• establish a maximum flow from node s to node t in ;
• update e(s) , e(t), and G(x);
• end; 13
• end;
 b(i) b(j)
Primal –Dual Algorithm i
  (, )
j

• Example network:

2 -2
  (, )
u w
  (,
) )
(  ,
v z
  (, )
2 -2

14
 b(i) b(j)
  (, )
Primal –Dual Algorithm i j

• Transformed network

2 -2 0 0
  (, )   (, )
u w ) u w   (,
)
  (,
  (, 4 s   (, t -4
) ) ,) )
(  ,   (,  ( )
v z ) v z (  ,
  (, )   (, )
2 -2 0 0

15
 b(i) b(j)
  (, )
Primal –Dual Algorithm i j

• We start with x=0, so G(x) = G

0 0
  (, )
) u w   (,
)
  (,
4 s   (, t -4
,) )
  (,  ( )
) v z (  ,
  (, )
0 0

16
 b(i) b(j)
  (, )
Primal –Dual Algorithm i j

•  =0, so = (i) + (j) =

0 0
  (, )
) u w   (,
)
  (,
4 s   (, t -4
,) )
  (,  ( )
) v z (  ,
  (, )
0 0

17
 b(i) b(j)
  (, )
Primal –Dual Algorithm i j

• Running shortest path from s yield vector

d = (0,0,0,1,2,1)

0 0
  (, )
) u w   (,
)
  (,
4 s   (, t -4
,) )
  (,  ( )
) v z (  ,
  (, )
0 0

18
 b(i) b(j)
  (, )
Primal –Dual Algorithm i j

• Running shortest path from s yield vector

d = (0,0,0,1,2,1)
1
0
0 0
0   (, ) 1
) u w   (,
)
  (,
4 s   (, t -4
,) )
  (,  ( )
) v z (  ,
  (, )
0 0
0 2
19
   (, )
Primal –Dual Algorithm𝑐 𝜋
i
=𝑐 𝑢 , 𝑣 − 𝜋 ( 𝑢 ) + 𝜋 ( 𝑣 )
j
  𝑢, 𝑣

•  G(x) after updating (= -d) :

 (w) = 1 1
 (u) = 0 0
 (s) = 0  e(s)= -4 0 0
e(s) = 4   (, )   (, (t) = 1   (, )
) u w
) 0 ) u w   (,
  (, ) 1
  (,   (,
s ) ) t 4 s   (, t -4
  (, (, ) )
) v  
, )   (,
)   (, )
z  ( v z (  ,
  (, )   (, )
0 0
 (v) = 0  (z) = 2 0 2
20
 𝑟  ❑
Primal –Dual Algorithm i
𝑖𝑗
j
𝑐  𝜋𝑢 , 𝑣 =𝑐 𝑢 , 𝑣 − 𝜋 ( 𝑣 )+ 𝜋 ( 𝑢 )
•  The admissible network ( = 0)

 (w) = 1
 (s) = 0  (u) = 0
 e(s)= -4
e(s) = 4
u
  (, )
w   (, (t) = 1  1
) ) u w  2
  (, 2 
  (,
s ) ) t s  1 t
  (,   (, )  2
1 
) v z  ( ,
v z
  (, )  2
 (v) = 0  (z) = 2
21
 𝑟  ❑
Primal –Dual Algorithm i
𝑖𝑗
j

• Run max flow from s to t:

 1
u w  2
2 
s  1 t
1 
 2
v z
 2

22
 𝑟  ❑
Primal –Dual Algorithm i
𝑖𝑗
j

• Run max flow from s to t:

 0
u w  0
1 
s  1 t
0 
 1
v z
 2

23
Primal –Dual Algorithm
• Correctness + Runtime:
• The primal-dual algorithm sends flow along all shortest paths at
once; therefore, proof of correctness is similar to the successive
shortest path one

• Excess of the node s decreases at every iteration

• Since everything is integral the distance between s

and t increases by at least one unit.

24
Primal –Dual Algorithm
•  Correctness + Runtime:
• Let U be an upper bound to the supply and C an upper bound to
the costs.

• Number of iteration bounded by min{nU,nC}

e(s)
The value of no node potential can fall below –nC

• O(min{nU,nC}* {S(n,m,nC) + M(n,m,U)} where S(n,m,nC) is the

running time for shortest path and M(n,m,U) is the running time 25
for Max Flow.
Optimality Conditions
(Reminder)
• Complementary
  Slackness:
Theorem: A feasible solution x* is an optimal solution of the MCF problem if
and only if for some set of node potentials , the reduced costs and flow values
satisfy the following complementary slackness conditions for every edge (u,v) in
the network:
 If then
 If then
 If then

Out-Of-Kilter Algorithm

26
The Out of Kilter Algorithm
• An edge can be in one of 2 states:

• In-kilter : satisfy the complementary slackness

• Out-of-kilter : don’t satisfy the complementary slackness

28
Kilter Diagram
  The kilter diagram represents
the conditions:

 If then
 If then
 If then

29
 Kilter Number
•• kilter
  number of each edge (i, j) is the
magnitude of the change in required
to make the edge an in-kilter arc
while keeping fixed.

• A kilter number can be thought of as

the change required to bring a flow
into feasibility and eventually
optimality.

• we can add up all the kilter numbers

to find how far from optimality we
are at any given time.

30
• An in-kilter edge has a kilter number
of zero.
Kilter Number
 
 

 < 0 -|   If then
 If then
 =0  If then



31
Kilter Number – residual
network
   

 We must send units of flow

in the edge (I,j) so that it
drops out of the residual
network and satisfy the RC

32
The Out of Kilter Algorithm
  Start with a feasible flow and = 0


 Iterate while we still have out-of-kilter edges

 Look at G(x) and select and out-of-kilter (p,q)
 Edges are given a cost of the higher of zero or the reduced cost.
 Run shortest path and update the potential function same as before
 If after the update the reduce cost of the edge(p,q) is negative
 We use a shortest path to find the cheapest cycle that includes the out of
kilter edge using the residual graph.
 Having found the cycle, we check the amount of flow that can be
augmented (min in the cycle ) and update the flow variables (x) and the
residual graphs (G(x)).

33
The Out of Kilter Algorithm
 Correctness:
 Kilter numbers of the edges are non-increasing
 Two operations in the algorithm affect the kilter numbers of arcs:
 updating node potentials

 augmenting flow along the cycle W

 At each iteration the algorithm selects and edge (p,q)

 Makes it in-kilter during potential update
 decrease it by at least 1 by the flow augmentation

34
The Out of Kilter Algorithm
 
 Runtime:
 The algorithm terminate within O(mU) iterations

 Dominant computation is shortest path computation

  total runtime is : O(

35