Duality and Complementary Slackness

Theorem

U Dinesh Kumar
This session…
 Relationship between Primal and Dual Linear
Programming.

 Complementary Slackness Theorem.

 Linear
Programming

Allocation of Valuation of
limited resource Resource

Primal Problem Dual Problem

Primal and Dual Linear Programming
 Associated with any LP is another LP called Dual.

 The formulation that solves resource allocation

problem is called the Primal (primal linear
programming).

 The problem that finds the marginal value of the

limited resource is called dual (dual linear
programming)
Applications of Dual LP
 In some cases it is easy to solve dual than primal.

 Provides better insight to the original LPP.

Converting Primal to Dual
 If primal is a maximization problem, the dual will be a
minimization problem, and vice versa.

 Before finding the dual of a LP, we convert the original

problem in its standard form.
Standard Form of LP (Maximization)
 Standard Maximization Problem (n variables and m
constraints)

M aximize Z  C1 x1  C2 x2  ...  Cn xn
Subject
a11x1  a12 x2  ...  a1n xn  b1
a 21x1  a22 x2  ...  a2 n xn  b2
...........................................
a m1 x1  am 2 x2  ...  amn xn  bm
x1 , x2 ,..., xn  0
Standard Form of LP (Minimization)
 Standard Minimization Problem

M inimize Z  C1 x1  C2 x2  ...  Cn xn
Subject
a11x1  a12 x2  ...  a1n xn  b1
a 21x1  a22 x2  ...  a2 n xn  b2
...........................................
a m1 x1  am 2 x2  ...  amn xn  bm
x1 , x2 ,..., xn  0
Primal and Dual

Primal Problem Dual Problem

M ax Z  CX M in W  Yb
Subject to : Subject to :
AX  b yA  c
and and
X0 y0
Manvi Motors (Primal and Dual)

Primal Dual
M ax 3500 C  5000 T M in 12000 y1 10000 y2
S.t. S.t.
20 C  40 T  12000 20 y1 25 y2  3500
25 C  10 T  10000 40 y1 10 y2  5000
C, T  0 y1, y2  0
Primal-Dual relationship
 If the primal has n variables and m resource constraints (standard
form), the dual will have m variables and n constraints.

 There is one-to-one correspondence between the primal

constraints and dual variables. For every primal constraint there is
a dual variable. For every primal variable, there is a dual
constraint.

 The value of dual variables are shadow prices and indicate

marginal value of each resource.
 Weak Duality Property
If x is a feasible solution for the primal and y
is a feasible solution for the dual problem,
then:

cx  yb

That is, the value of dual objective function is

always greater than or equal to the primal
objective function value for any pair of feasible
solutions.
Applications of weak duality
 The value of primal objective cx is a lower bound to the
value of dual optimal value y*b.

 The value of dual objective yb is upper bound to any

primal objective cx*.
 20C + 40T  12000

Manvi Motors – Primal Problem

D
C
Cars
E

45C + 50T  14000 25C + 10T  10000

A
O
Isoprofit line
Trucks
 40Y1 + 10Y2  5000
Manvi Motors – Dual Problem

C
Y2

B
A 20Y1 + 25Y2  3500
O

Y1
Strong Duality Property
 If x* is an optimal solution for the primal and y* is an
optimal solution for the dual problem, then:

cx* = y*b
Applications of strong duality
 If X* and Y* are feasible solutions of primal and dual and if their
corresponding objective functions are equal, then X* and Y* are
optimal solutions.

 If X* is an optimal solution to primal, then a finite optimal

solution exists for dual (vice versa).
Symmetry Property
1. If one problem has a feasible solutions and bounded
objective function (and so an optimal solution), then so
does the other problem.

2. If one problem has a feasible and an unbounded objective

function (and so no optimal solution), then the other
problem has no feasible solutions.

3. If one problem has no feasible solution, then the other

problem has either no feasible solutions or an unbounded
objective function.
Finding Dual of Non-standard LP (Max Problem)
 Step 1: Multiply each  constraint by – 1. Then convert
 into  constraint.

 Step 2: Replace each equality constraint by two

inequality constraint.

 Replace unrestricted variables with two non-negative

variables.
Example 1

Primal
Max Z  2X1  X2 M ax Z  2x1  X3  X4
Subject to : S.t.
X1  X3  X4  2
X1  X2  2
- X1 - X3  X4  -2
2X1 - X2  3 - 2X1  X3  X4  - 3
X1 - X2  1 X1 - X3  X4  1
Where
X1  0, X2 is urs X2  X3 - X4
 Example 1

Primal Dual
M ax Z  2x1  X3  X4
S.t. M in 2y1- 2y2- 3y3 y4
X1  X3  X4  2
- X1 - X3  X4  -2 y1- y2- 2y3 y4  2
- 2X1  X3  X4  - 3
X1 - X3  X4  1 y1- y2  y3- y4  1
Where
X2  X3 - X4
- y1 y2- y3 y4  -1
Y1 – y2 + y3 – y4 = 1
 Primal – Dual Relationship

 If the primal variable is unrestricted then the

corresponding dual constraint will be an equality
constraint.

 If the primal constraint is an equality constraint then the

corresponding dual variable will be unrestricted.
Dual of non-standard Min problem
 Step 1: Change each  constraint to a  constraint by
multiplying each  by -1.

 Step 2: Replace each equality constraint by a  and 

constraint. Then transform  into a  constraint.

 Replace any unrestricted variables using two non-

negative variables.
Example 2
 Primal Dual
Min w = 2Y1 + 3Y2 + Y3 M ax Z  2X1  X2
Subject to:
S.t.
Y1 + 2Y2 + Y3  2
Y1 – Y2 – Y3 = 1 X1  X2  2
Y1 is urs, Y2 and Y3  0 2X1  X2  3
X1  X2  1
X2 is urs, X1  0
Complementary Slackness Theorem
Complementary Slackness Theorem

 Let x = (x1, x2, . . ., xn) is primal feasible and that y = (y1, y2,
. . . , ym) is dual feasible.

 Let (w1, w2, . . ., wm) denote the corresponding primal

slack variables, and let (z1, z2, . . . , zn) denote the
corresponding dual slack variables.

 Then x and y are optimal for their respective problems if

and only if xjzj = 0, for j = 1, 2, . . . , n, wiyi = 0, for i = 1, 2, . .
. , m.
Complementary Slackness
 If a constraint in either the primal (or dual problem) is
non-binding then the corresponding variable in the
other (complementary problem) must equal to zero.

 If a constraint in a primal problem is non-binding

then the corresponding variable in dual will be
zero.

 If a variable in the optimal solution is non-zero

then the corresponding dual constraint is binding.
 Complementary Slackness

A pair of feasible solutions (x* and y*) for the primal and dual,
respectively, to be optimal for the primal and dual a necessary and
sufficient conditions is:

1. yk* = 0 whenever kth constraint is slack at x*.

2. xk* = 0 whenever kth constraint is slack at y*.
 Complementary Slackness

 If x* and y* are feasible and satisfy the complementary

slackness conditions, then they are optimal.
 Application of Duality Theorem and
Complementary Slackness Theorem

M anvi M otors Primal M anvi M otors Dual

M ax 3500C  5000T M in 12000y1 10000y2- 14000y3
20C 40T  12000  500y4  200y5
25C  10T  10000 20y1 25y2- 45y3 y4  3500
45C  50T  14000 40y1 10y2 - 50y3  y5  5000
C  500 y1, y2  0
T  250
C, T  0
Example 1: Strong Duality Example
 If C = 350 and T = 125 is the optimal solution, then find
the shadow price.

 C = 350 and T = 125 means, constraints 1 and 2 in the

primal problems are binding and the remaining
constraints are non-binding. That is, y3 = y4 = y5 = 0

 Using complementary slackness theorem, 20y1 + 25 y2 =

3500 and 40y1 + 10y2 = 5000. That is, y1 = 112.5 and y2
= 50.
Example 2: Complementary slackness
Theorem and weak duality
M ax 30x1  6x 2  5x 3  18x 4
S.t.
x1  2x 3  x 4  20
 2x1  x 2  x 4  15
6x1  2x 2  3x 3  54
x i  0, i  1,2,3,4
Check whether (0, 25, 5, 10) is an optimal solution for the above problem
 Dual problem

M in w  20y1  15y 2  54y 3

s.t.
Using CST,
y1  2y 2  6y 3  30 y 2  2 y3  6
y 2  2y 3  6 2 y1  3 y3  5
2y1  3y 3  5
y1  y2  18
y1  y 2  18
y i  0, i  1,2,3,4 (4,14,1) which is not feasibe
so (0,25,5,10) is not optimal
 Primal Dual Relationship

Primal Solution Item Dual Solution Item

Objective: Maximization Objective: Minimization

Shadow Prices Variable Values

Slacks Reduced Cost

Variable Values Shadow Prices

Reduced costs Slacks

Recommended Reading
 Wayne L Winston, “Operations Research: Applications and
Algorithms”, Cengage Learning, 2009.

 Hillier and Hillier, “Introduction to Management Science”,

2009.