MODM

Quadratic Programming
A Quadratic Programming problem is a non-linear programming problem
with a quadratic objective function and linear constraints.

The General Structure of Quadratic Programming Problem is as follows,

Maximize Z = C1x1 + C2x2 + --- + Cnxn + --- +

Cn+1 x12 + Cn+2x22 + --- +
C2n+1 x1x2 + C2n+2 x1x3 + ---
S.T. ai1 x1 + ai2 x2 + --- ain xn ≤ bi
and x1, x2, --- xn ≥ 0
 MODM
Matrix Form Representation
Maximize Z = Cx + xT A x
S.T. Dx ≤ b
x≥0
where A is a symmetric matrix and b and C are real vectors.
𝑥1
𝑎11 𝑎12 ⋯ 𝑎1𝑛
𝑥2
𝐶 = 𝐶1 𝐶2 ⋯ 𝐶𝑛 𝑋= ⋮ 𝐴= ⋮ ⋮ ⋱ ⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛
𝑥𝑛

𝑏1
𝑑11 𝑑12 ⋯ 𝑑1𝑛
𝑏2
𝐷= ⋮ ⋮ ⋱ ⋮ b=
⋮
𝑑𝑚1 𝑑𝑚2 ⋯ 𝑑𝑚𝑛
𝑏𝑚
 MODM

Example – f(x) = x12 - 2x22 + x1x2 - 2x2x3 + 5x1x3

1 0.5 2.5 𝑥1
f(x) = 𝑥1 𝑥2 𝑥3 0.5 −2 −1 𝑥2
2.5 −1 0 𝑥3

4
 MODM

Example – f(x) = - 3x12 + 2x22 - 3x32

−3 0 0 𝑥1
f(x) = 𝑥1 𝑥2 𝑥3 0 2 0 𝑥2
0 0 −3 𝑥3

Example – f(x) = x12 + x1 - 2x2 + 10

1 0 𝑥1 𝑥1
f(x) = 𝑥1 𝑥2
𝑥2 + 1 −2 𝑥2 + 10
0 0

5
 MODM
To check the definiteness of a function
The definiteness of a function f(x) depends on the nature of A in xT A x
We can use the matrix minor test or Eigen value test to check the definiteness
of a function.

The principal minors decides the nature of xT A x

𝑎11 𝑎12 𝑎13

𝑎11 𝑎12
𝐷1 = 𝑎11 𝐷2 = 𝐷3 = 𝑎21 𝑎22 𝑎23
𝑎21 𝑎22
𝑎31 𝑎32 𝑎33

6
 MODM

A function f(x) = xT A x is

i) Positive definite D1 > 0 ; D2 > 0 ; D3 > 0

ii) Positive semi-definite D1 > 0 ; D2 ≥ 0 ; D3 ≥ 0
(At least one of Di = 0 , for i = 2,3,---,n)
iii) Negative definite D1 ˂ 0 ; D2 > 0 ; D3 ˂ 0
iv) Negative semi-definite D1 ˂ 0 ; D2 ≥ 0 ; D3 ≤ 0
(At least one of Di = , for i = 2,3,---,n)
v) Indefinite All other cases are indefinite

7
 MODM
Decide the definiteness of the function
f(x) = -3 x12 + 2 x22 – 3x32 – 10 x1x2 + 4 x2x3 + 6 x1x3

Solution –
Write the function as f(x) = xT A x
−3 −5 3 𝑥1
f(x) = 𝑥1 𝑥2 𝑥3 −5 2 2 𝑥2
3 2 −3 𝑥3
The Principal minors are,
𝐷1 = −3 = −3 ˂ 0

−3 −5
𝐷2 = = −31 ˂ 0
−5 2

−3 −5 3
Thus D1 < 0, D2 < 0, D3 > 0.
𝐷3 = −5 2 2 = 27 > 0
3 2 −3
This ensures that the function is indefinite.
8
 MODM
Decide the definiteness of the function
f(x) = 2 + 2x1 + 3 x2 - x12 - x22
Solution –
Write the function in the form,
−1 0 𝑥1 𝑥1
f(x) = 𝑥1 𝑥2 𝑥 + 2 3 𝑥 +2
0 −1 2 2
By matrix minor test – D1 = -1 ˂ 0 ; D2 = 1 > 0
Alternate sign with first sign negative ⸫ Negative definite.
By Eigen value test – Eigen values are -1 , -1
Both are negative ⸫ Negative definite.
9
 MODM
Decide the definiteness of the function
f(x) = x12 - 2x22
Solution –
Write the function as f(x) = xT A x
1 0 𝑥1
f(x) = 𝑥1 𝑥2
0 −2 𝑥2
By matrix minor test – D1 = 1 > 0 ; D2 = -1 ˂ 0
This ensures the function is indefinite.
By Eigen value test – Eigen values are 1 , -2
Positive and negative ⸫ Function is indefinite.
10
 MODM

Quadratic programming method can be obtained by

i) Wolfe’s method
ii) Beale’s method

11
 MODM
4 0
Find the quadratic form of matrix A =
0 3
Solution –
Quadratic form, f(x) = xT A x
4 0 𝑥1
𝑥1 𝑥2
0 3 𝑥2

𝑥1 𝑥2 4𝑥1 + 0𝑥2
(0𝑥1 + 3𝑥2)

𝑥1 𝑥2 4𝑥1
3𝑥2

4x12 + 3x22
12
 MODM
4 −5 7
Find the quadratic form of matrix A = −5 −6 8
7 8 −9
Solution –
Quadratic form, f(x) = xT A x
4 −5 7 𝑥1
f(x) = 𝑥1 𝑥2 𝑥3 −5 −6 8 𝑥2
7 8 −9 𝑥3
(4𝑥1 − 5𝑥2 + 7𝑥3 )
f(x) = 𝑥1 𝑥2 𝑥3 (−5𝑥1 − 6𝑥2 + 8𝑥3 )
(7𝑥1 + 8𝑥3 − 9𝑥3 )
f(x)=(4𝑥12 − 5𝑥1 𝑥2 + 7𝑥1 𝑥3 )+(−5𝑥1 𝑥2 − 6𝑥22 + 8𝑥2 𝑥3 )+(7𝑥1 𝑥3 + 8𝑥2 𝑥3 − 9𝑥32 )
f(x)=4𝑥12 − 6𝑥22 − 9𝑥32 − 10𝑥1 𝑥2 + 14𝑥1 𝑥3 + 16𝑥2 𝑥3
13
 MODM
4 −5 7
Find the quadratic form of matrix A = −5 −6 8
7 8 −9
Solution – 𝑥1 𝑥2 𝑥3
4 −5 7
A = −5 −6 8
7 8 −9

1) For diagonal elements square it

4𝑥12 − 6𝑥22 − 9𝑥32
2)For 𝑎ij and 𝑎ji add it
(-5-5) = -10 { 𝑎21 and 𝑎12 (7+7) = 14 { 𝑎13 and 𝑎31 (8+8) = 16 { 𝑎23 and 𝑎32
⸫ −10𝑥1 𝑥2 ⸫14𝑥1 𝑥3 ⸫16𝑥2 𝑥3
⸫ Quadratic form → 4𝑥12 − 6𝑥22 − 9𝑥32 − 10𝑥1 𝑥2 + 14𝑥1 𝑥3 + 16𝑥2 𝑥3
14
 MODM
Geometric Programming
• G. P. was developed by Duffin, Peterson & Zener.

• G. P. is used to minimize the functions that are in the form of POSYNOMIALS subject to
constraints of the same type.

• It differs from other optimization techniques in the way that it finds the optimal value of
the objective function first, instead of finding optimal values of the design variables first.

• It is advantageous in situations where the optimal value of the objective function may be
all that is of interest and the calculation of the optimum design vectors may be omitted.
15
 MODM
The major disadvantage of G. P. is that it requires objective function and constraints in the
form of Posynomials.
Posynomials –
Consider for example the total cost as the objective function f(x) which is given by the sum
of several component costs Ui(x) as,
f(x) = U1 + U2 + --- + UN
In many cases, the component cost Ui can be expressed as power function of the type,
Ui = Ci x1a1i x2a2i --- xnani
where, coefficients Ci are positive constants
design parameters x1, x2, --- xn are positive variables
exponents aij are real constants (positive, negative or zero)
Function f(x), because of positive coefficients and variables and real exponents are called Posynomials.
16
 MODM

G. P. technique is based on the arithmetic mean - geometric mean

inequality and therefore called geometric programming.

Let, u1, u2,… un, be any n non-negative numbers.

𝛿1 , 𝛿2 , … 𝛿𝑛 be such that
𝛿𝑖 > 0 {say 𝛿 are weights}
i =1, 2, 3,…, n
𝛿1 +𝛿2 + … +𝛿𝑛 = 1

17
 MODM
Now,
Arithmetic mean ≥ Geometric mean

𝛿1 𝑢1 +𝛿2 𝑢2 +⋯𝛿𝑛 𝑢𝑛 𝛿 𝛿 𝛿
≥ 𝑢1 1 𝑢2 2 …𝑢𝑛𝑛
𝛿1 +𝛿2 +⋯+𝛿𝑛

𝛿
σ𝑛𝑖=1 𝛿𝑖 𝑢𝑖 ≥ ς𝑛𝑖=1 𝑢𝑖 𝑖 {𝛿1 + 𝛿2 + … +𝛿𝑛 = 1

Let, 𝑈𝑖 = 𝛿𝑖 𝑢𝑖 {Ɐ i

𝑈𝑖 𝛿𝑖 𝑈𝑖
σ𝑛𝑖=1 𝑈𝑖 ≥ 𝑛
ς𝑖=1 {⸪𝑢𝑖 =
𝛿𝑖 𝛿𝑖
18
 MODM
𝑈1 𝑈2 𝑈𝑛
The equality hold when , = =… = K (say)
𝛿1 𝛿12 𝛿𝑛

⸫ 𝑈1 = 𝐾𝛿1 ; 𝑈2 = 𝐾𝛿2 ; … 𝑈𝑛 = 𝐾𝛿𝑛

From
𝑛
LHS
෍ 𝑈𝑖 = 𝐾( 𝛿1 + 𝛿2 + … +𝛿𝑛 ) = 𝐾 {⸪ 𝛿1 + 𝛿2 + … +𝛿𝑛 = 1
𝑖=1

From RHS
𝑛 𝑈𝑖 𝛿𝑖
ς𝑖=1 = 𝐾 𝛿1 𝐾 𝛿2 … 𝐾 𝛿𝑛 = 𝐾 (𝛿1 +𝛿2 + …+𝛿𝑛 ) = 𝐾
𝛿𝑖

LHS = RHS = K
⸫ Equality holds 19
 MODM
1
Minimize 𝑓 𝑥 = 𝑥1 + 𝑥2 + ; 𝑥1 , 𝑥2 > 0
𝑥1 𝑥2

𝑓 𝑥 = 𝑈1 + 𝑈2 + 𝑈3

𝑈1 𝛿1 𝑈2 𝛿2 𝑈3 𝛿3 𝑈𝑖 𝛿𝑖
≥ {σ𝑛𝑖=1 𝑈𝑖 ≥ 𝑛
ς𝑖=1
𝛿1 𝛿2 𝛿3 𝛿𝑖

𝛿1 𝛿2 𝛿3
𝑥1 𝑥2 1
=
𝛿1 𝛿2 𝑥1 𝑥2 𝛿3

𝛿1 𝛿2 𝛿3
𝛿1 −𝛿3 𝛿2 −𝛿3
1 1 1
= 𝑥1 𝑥2
𝛿1 𝛿2 𝛿3

In order to minimize f(x) we need to maximize above equation.

Now the variables are unknown, so let us make above equation free from variables.
20
 MODM
To make it free from variables,
Let 𝛿1 − 𝛿3 = 0; 𝛿2 − 𝛿3 = 0; and 𝛿1 + 𝛿2 + 𝛿3 = 1
𝛿1 = 𝛿3 ; 𝛿2 = 𝛿3 𝛿1 = 𝛿2 = 𝛿3 = 1/3

There are three unknowns 𝛿1 , 𝛿2 , 𝛿3 and three equations

So we get a fix solution

Substituting the values of 𝛿1 = 𝛿2 = 𝛿3 = 1/3 in above RHS equation

𝛿1 −𝛿3 𝛿2 −𝛿3 1 𝛿1 1 𝛿2 1 𝛿3 1/3 1/3 1/3
𝑥1 𝑥2 = (1)(1) 3 3 3
𝛿1 𝛿2 𝛿3
=3
⸫Min f(x) = 3
21
 MODM
Now Values of 𝑥1 , 𝑥2 = ?
𝑈1 𝑈2 𝑈𝑛
Applying the equality condition = =
𝛿1 𝛿12 𝛿𝑛

𝑥1 𝑥2 1
= =
1/3 1/3 (𝑥1 𝑥2 )1/3

𝑥1 = 𝑥2 and 𝑥1 𝑥2 2 = 1 𝑥1 3 =1 𝑥1 = 𝑥2 =1

22
 MODM
Minimize 𝑓 𝑥 = 5x1 + 20x2 + 10x1-1x2-1 ; x1 , x2 ≥ 0
𝑓 𝑥 = 𝑈1 + 𝑈2 + 𝑈3

𝑈1 𝛿1 𝑈2 𝛿2 𝑈3 𝛿3 𝑈 𝑖 𝛿𝑖
≥ {σ𝑛𝑖=1 𝑈𝑖 ≥ 𝑛
ς𝑖=1
𝛿1 𝛿2 𝛿3 𝛿𝑖

𝛿1 𝛿2 𝛿3
5𝑥1 20𝑥2 10x1−1x2−1
=
𝛿1 𝛿2 𝛿3

𝛿1 𝛿2 𝛿3
𝛿1 −𝛿3 𝛿2 −𝛿3
5 20 10
= 𝑥1 𝑥2
𝛿1 𝛿2 𝛿3

In order to minimize f(x) we need to maximize above equation.

Now the variables are unknown, so let us make above equation free from variables.
23
 MODM
To make it free from variables,
Let 𝛿1 − 𝛿3 = 0; 𝛿2 − 𝛿3 = 0; and 𝛿1 + 𝛿2 + 𝛿3 = 1
𝛿1 = 𝛿3 ; 𝛿2 = 𝛿3 𝛿1 = 𝛿2 = 𝛿3 = 1/3

There are three unknowns 𝛿1 , 𝛿2 , 𝛿3 and three equations

So we get a fix solution

Substituting the values of 𝛿1 = 𝛿2 = 𝛿3 = 1/3 in above RHS equation

𝛿1 −𝛿3 𝛿2 −𝛿3 5 𝛿1 20 𝛿2 10 𝛿3 1/3 1/3 1/3
𝑥1 𝑥2 = (1)(1) 15 60 30
1/3 1/3 1/3
= 30
⸫Min f(x) = 30
24
 MODM
Now Values of 𝑥1 , 𝑥2 = ?
𝑈1 𝑈2 𝑈𝑛
Applying the equality condition = =
𝛿1 𝛿2 𝛿𝑛

5𝑥1 20𝑥2 10𝑥1−1 𝑥2−1

= =
1/3 1/3 1/3

1
𝑥1 = 4𝑥2 2𝑥2 = → 2𝑥1 𝑥2 2 = 1 8𝑥2 3 =1
𝑥1 𝑥2
⸫ 𝑥1 =2 𝑎𝑛𝑑 𝑥2 =1/2

25
 MODM

• Goal Programming set some estimated targets for each goal and assign
priorities to them, i.e. to rank them in order of importance.

• GoalProgramming tries to minimize the deviations from the targets that

were set. It begins with the most important goal and continues in such a
way that a less important goal is considered only after the more important
ones are satisfied or have reached the point beyond which no further
improvement are desired.

• In
the final solution all the goals may not be fulfilled to the fullest extent
however, the deviations will be the minimum possible.
 MODM
A factory can manufacture 2 products A & B. The profit on one unit of A is
Rs. 80 and one unit of B is Rs. 40. The maximum demand of A is 6 units per
week and of B it is 8 units per week. The manufacturer has set up a goal of
achieving a profit of Rs. 640 per week. Formulate the problem as goal
programing and solve it.

Solution -
Let, x1 and x2 be the no. of units of A & B product per week respectively.

Let, Z be the profit

Then, Z = 80x1 + 40x2
Goal is 640
Z may exceed the goal of 640 or fall short of it.
 MODM
Let, u ≥ 0 denote the shortfall (underachieved) and
v ≥ 0 denote the excess (overachieved) in profit from goal

(u & v are deviational variables)

3 possibilities exists.
Goal will be …
i) Exactly achieved (u = 0 and v = 0)
ii) Overachieved (u = 0 and v > 0)
iii) Underachieved (u > 0 and v = 0)

So only those solutions are acceptable in which at least one of the

variable, u or v is zero or both are zero.
 MODM
So, Either Z = 640 – u
Or Z = 640 + v

Combining the above two conditions,

Z + u – v = 640 {where at least one variable u or v is zero

To achieve the goal as closely as possible, the objective should be to minimize

the deviation from the goal.
 MODM
Problem Formulation – 𝑥2
18
Minimize Z' = u + v 𝑥1 = 6
16
S.T. 80 x1 + 40 x2 + u – v = 640
14
x1 ≤ 6
12
x2 ≤ 8
10
x1 , x2 , u, v ≥ 0 𝐶 𝑥2 = 8
8
𝐵 𝐸
such that either u or v or both equal to zero. 6

4
𝐷
2 80𝑥1 + 40𝑥2 =640
𝑂 𝐴
0
0 1 2 3 4 5 6 7 8 9 10 𝑥1
 MODM

Considering Z = 80 x1 + 40 x2 on the line DE, Z = 640.

⸫ Goal can be met exactly if x1 and x2 lies on DE.
For points on DE, u = v = 0, giving minimum value of Z'

If (x1, x2) falls in DEC, then profit is over and above 640. (u = 0 and v > 0)

If (x1, x2) falls in OADEB, then profit is less than 640. (u > 0 and v = 0)
 MODM
Simplex Solution

Adding the slack variables the problem becomes,

Minimize Z' = u + v
S.T. 80 x1 + 40 x2 + u – v = 640
x1 + x3 = 6
x2 + x4 = 8
x1 , x2 , u, v ≥ 0
such that either u or v or both equal to zero
Cj 0 0 -1 -1 0 0
CB XB X1 X2 u v X3 X4 RHS Ɵ
-1 u 80 40 1 -1 0 0 640 8
0 X3 1 0 0 0 1 0 6 6
0 X4 0 1 0 0 0 1 8 ---
Cj – Zj 80 40 0 -2 0 0 -640
 MODM
Cj 0 0 -1 -1 0 0
CB XB X1 X2 u v X3 X4 RHS Ɵ
-1 u 80 40 1 -1 0 0 640 8
0 X3 1 0 0 0 1 0 6 6
0 X4 0 1 0 0 0 1 8 ---
Cj – Zj 80 40 0 -2 0 0 -640
X1 enters ; X3 leaves ; u = u – 80 X1 ; X4 = X4
-1 u 0 40 1 -1 -80 0 160 4
0 X1 1 0 0 0 1 0 6 ---
0 X4 0 1 0 0 0 1 8 8
Cj – Zj 0 40 0 -2 -80 0 -160
X2 enters ; u leaves ; Divide u by 40 to get X2 ; X1 = X1 ; X4 = X4 – X2
0 X2 0 1 1/40 -1/40 -2 0 4
0 X1 1 0 0 0 1 0 6
0 X4 0 0 -1/40 1/40 2 1 4
Cj – Zj 0 0 -1 -1 0 0 0
No positive values in Cj – Zj ⸫ The solution is X1 = 6 and X2 = 4
 MODM

0 X2 0 1 1/40 -1/40 -2 0 4 ---

0 X1 1 0 0 0 1 0 6 6
0 X4 0 0 -1/40 1/40 2 1 4 2
Cj – Z j 0 0 -1 -1 0 0 0
X3 enters ; X4 leaves ; Divide X4 by 2 to get X3 ; X2 = X2 + 2 X3 ; X1 = X1 – X3
0 X2 0 1 0 0 0 1 8
0 X1 1 0 1/80 -1/80 0 -1/2 4
0 X3 0 0 -1/80 1/80 1 1/2 2
Cj – Zj 0 0 -1 -1 0 0 0

Now the solution is X1 = 4 ; X2 = 8

⸫ The problem has multiple solutions with all the points on D and E
where u = v = 0
⸫ The goal is exactly achieved on line DE.
 MODM
Goal Programming Formulation (In General)

Minimize Z' = u + v
S. T. Z + u – v = Goal
σ𝑛𝑗=1 aij xj = bi
u, v, xj ≥ 0
with either u or v or both equal to zero

where, u is the underachievement

v is the overachievement
 MODM
A company manufactures 2 products, radios and transistors, which must be processed
through assembly & finishing departments. Assembly has 90 hours available, finishing
can handle up to 72 hours of work. Manufacturing one radio requires 6 hours in assembly
and 3 hours in finishing. Each transistor requires 3 hours in assembly and 6 hours in
finishing. If profit is Rs. 120 per radio and Rs. 90 per transistor, determine the best
combination of radios and transistors to realize profit of Rs. 2100.

Solution –
Expressing the given problem in goal problem Adding slack variables, the problem is expressed as
Minimize Z' = u Minimize Z' = u
S.T. 120 x1 + 90 x2 + u – v = 2100 S.T. 120 x1 + 90 x2 + u – v = 2100
6x1 + 3x2 ≤ 90 6x1 + 3x2 + x3 = 90
3x1 + 6x2 ≤ 72 3x1 + 6x2 + x4 = 72
x1, x2, u, v ≥ 0 x1, x2, u, v, x3, x4 ≥ 0
Such that either u or v or both equal to zero. Such that either u or v or both equal to zero.
 MODM
Cj 0 0 -1 0 0 0
CB XB X1 X2 u v X3 X4 RHS Ɵ
-1 u 120 90 1 -1 0 0 2100 17.5
0 X3 6 3 0 0 1 0 90 15
0 X4 3 6 0 0 0 1 72 24
Cj – Z j 120 90 0 -1 0 0 -2100
X1 enters ; X3 leaves ; Divide X3 by 6 to get X1 ; u = u – 120 X1 ; X4 = X4 - 3X1
-1 u 0 30 1 -1 -20 0 300 10
0 X1 1 ½ 0 0 1/6 0 15 30
0 X4 0 4.5 0 0 -0.5 1 27 6
Cj – Z j 0 30 0 -1 -20 0 -300
X2 enters ; X4 leaves ; Divide X4 by 4.5 to get X2 ; u = u – 30 X2 ; X1 = X1 – 1/2 X2
-1 u 0 0 1 -1 -16.67 -6.66 120
0 X1 1 0 0 0 0.222 -0.111 12
0 X2 0 1 0 0 -0.111 0.222 6
Cj – Z j 0 0 0 -1 -16.67 -6.66 -120
No positive values in Cj – Zj ⸫ The algorithm terminates. The solution is X1 = 12 ; X2 = 6 and u = 120
 MODM
In the previous example, the company sets two equally ranked goals, one to reach a profit goal of Rs. 1500
and other to meet a radio goal of 10. Find the optimal solution.
Solution –
Let, u1 = amount by which the profit goal is underachieved
v1 = amount by which the profit goal is overachieved
u2 = amount by which the radio goal is underachieved
v2 = amount by which the radio goal is overachieved
Goal programming model can be formulated as,
Minimize Z' = u1 + u2
S.T. 120 x1 + 90 x2 + u1 – v1 = 1500
x1 + u2 – v2 = 10
6x1 + 3x2 ≤ 90
3x1 + 6x2 ≤ 72
x1, x2, u1, v1, u2, v2 ≥ 0
Such that either u1 or v1 or both equal to zero
and either u2 or v2 or both equal to zero
 MODM
Adding slack variables to assembly & finishing department constraints.
Minimize Z' = u1 + u2
S.T. 120 x1 + 90 x2 + u1 – v1 = 1500
x1 + u2 – v2 = 10
6x1 + 3x2 + x3 = 90
3x1 + 6x2 + x4 = 72
x1, x2, u1, v1, u2, v2, x3, x4 ≥ 0
Such that either u1 or v1 or both equal to zero
and either u2 or v2 or both equal to zero

Cj 0 0 -1 0 -1 0 0 0
CB XB X1 X2 u1 v1 u2 v2 X3 X4 RHS Ɵ
-1 u1 120 90 1 -1 0 0 0 0 1500 12.5
-1 u2 1 0 0 0 1 -1 0 0 10 10
0 X3 6 3 0 0 0 0 1 0 90 15
0 X4 3 6 0 0 0 0 0 1 72 24
Cj – Zj 121 90 0 -1 0 -1 0 0 -1510
X1 enters ; u2 leaves ; Divide u2 by key element to get X1 ; u1 = u1 – 120 X1 ; X3 = X3 - 6X1 ; X4 = X4 - 3X1
 MODM
Cj 0 0 -1 0 -1 0 0 0
CB XB X1 X2 u1 v1 u2 v2 X3 X4 RHS Ɵ
-1 u1 0 90 1 -1 -120 120 0 0 300 2.5
0 X1 1 0 0 0 1 -1 0 0 10 ---
0 X3 0 3 0 0 -6 6 1 0 30 5
0 X4 0 6 0 0 -3 3 0 1 42 14
Cj – Zj 0 90 0 -1 -121 120 0 0 -300
v2 enters ; u1 leaves ; Divide u1 by key element to get v2 ; X1 = X1 + v2 ; X3 = X3 - 6v2 ; X4 = X4 - 3v2
0 v2 0 3/4 1/120 -1/120 -1 1 0 0 5/2
0 X1 1 3/4 1/120 -1/120 0 0 0 0 25/2
0 X3 0 -3/2 -1/20 1/20 0 0 1 0 15
0 X4 0 15/4 -1/40 1/40 0 0 0 1 69/2
Cj – Zj 0 0 -1 0 -1 0 0 0 0

No positive values in Cj – Zj ⸫ The algorithm terminates. The solution is X1 = 25/2 ; X2 = 0 and v2 = 5/2
 MODM
A company manufactures tables & chairs. Each chair requires 4 man hours of labour
while each table requires 5 man hours of labour. If only 40 man hours are available each
week and the owner of the company would neither hire additional labour nor utilise
overtime, formulate the linear goal programming problem and solve it. Both the table &
chair fetch a profit of Rs. 100 each. The owner has a target to earn a profit of Rs. 2000
per week. Also he would like to supply 10 chairs if possible to a sister concern.
Solution –
The goal of the company and their assigned priorities are,

Priority Goal
1 To avoid hiring extra labour or utilise overtime
2 To reach a profit goal of Rs. 2000 a week
3 To supply 10 chairs a week to the sister concern
 MODM

Let x1 and x2 denote the number of chairs and tables to be produced per week
If ui and vi represent the amount by which the ith goal is underachieved or overachieved

Then the goals (expressed as constraints in the order of priority)

G1 → 4x1 + 5x2 + u1 – v1 = 40 (Man hour goal)

G2 → 100x1 + 100x2 + u2 – v2 = 2000 (Profit goal)
G3 → x1 + u3 – v3 = 10 (Chair goal)
x1, x2, u1, v1, u2, v2, u3, v3 ≥ 0
 MODM
𝑥2
22

20

18

16

14

12 𝑣3
10
𝐺3
8
𝑢3 𝑣2
x1 = 10 ; x2 = 0
6 Zmin = (0, 1000, 0)
𝑣1
4
𝑢2
2
𝑢1
𝐺1 𝐺2
0
0 2 4 6 8 10 12 14 16 18 20 22 24 𝑥1
 MODM

Dynamic Programming (Richard Bellman – 1950)

D. P. is used to solve a wide variety of optimization problem such as inventory

management, scheduling, etc.

The main concept is to break the problems into sub-problems (called stages) and
combine their solutions to get the solution of larger problem.
 MODM

Bellman’s Principle of Optimality

The solution of D. P. is based upon Bellman’s principle of optimality

which states that,

The optimal policy must be one such that, regardless of how a particular
state is reached, all later decisions preceding from that state must be
optimal.

45
 MODM
Use D.P. to solve the following problem Stage 2 - f2(K2) = min {x22 + f1(K1)}

Minimize Z = x12 + x22 + x32 = min {x22 + K12}

= min {x22 + (K2 – x2)2}
S.T. x1 + x2 + x3 = 15
Differentiate w.r.t. x2 and equate it to zero
x1, x2, x3 ≥ 0
2x2 – 2 (K2 – x2) = 0 → x2 = K2 / 2
Substituting the value of x2 in the equation of f2(K2)
Solution –
⸫ f2(K2) = K22 / 4 + K22 / 4 = K22 / 2
State Variables K3 = x1 + x2 + x3 = 15
K2 = x 1 + x 2 = K3 – x3 Stage 3 - f3(K3) = min {x32 + f2(K2)}
= min {x32 + K22 / 2}
K1 = x 1 = K2 – x2 = min {x32 + (K3 – x3)2 / 2}
Stage 1 -
Differentiate w.r.t. x3 and equate it to zero
f1(K1) = min {x12} = K12 2x3 – 2(K3 – x3) / 2 = 0 → x3 = K3 / 3

Substituting the value of x3 in the equation of f3(K3)

⸫ f3(K3) = {(K3 / 3)2 + (K3 - K3 / 3) 2 / 2 } = K32 / 3
46
 MODM
Now, we know that K3 = 15

⸫ f3(K3) = K32 / 3 = 152 / 3 = 75

x3 = K3 / 3 = 15 / 3 = 5

K2 = K3 – x3 = 15 – 5 = 10

x2 = K2 / 2 = 10 / 2 = 5

K1 = K2 – x2 = 10 – 5 = 5

x1 = K1 = 5

⸫ x1 = x2 = x3 = 5 & Z = f3(K3) = 75

47
 MODM
Use D.P. to solve the following problem Stage 2 - f2(K2) = max {y2 f1(K1)}
Maximize Z = y1y2y3 = max {y2 K1}
S.T. y1 + y2 + y3 = 5 = max {y2 (K2 – y2)}
y1, y2, y3 ≥ 0 Differentiate w.r.t. y2 and equate it to zero
y2 (–1) + (1) (K2 – y2) = 0 → y2 = K2 / 2
Solution – Substituting the value of y2 in the equation of f2(K2)
State Variables K3 = y1 + y2 + y3 = 5 ⸫ f2(K2) = K2 / 2 (K2 - K2 / 2 ) = K22 / 4
K2 = y1 + y2 = K3 – y3
Stage 3 - f3(K3) = max {y3 f2(K2)}
K1 = y1 = K2 – y2 = max {y3 K22 / 4}
Stage 1 - = max{y3 (K3 – y3)2 / 4}
f1(K1) = max {y1} = K1 Differentiate w.r.t. y3 and equate it to zero
¼ {1 (K3 – y3)2 – y3 2 (K3 – y3)} = 0 → y3 = K3/3

48
 MODM
Now, we know that K3 = 5

y3 = K3 / 3 = 5 / 3

K2 = K3 – y3 = 5 – 5 / 3 = 10 / 3

y2 = K2 / 2 = (10 / 3) / 2 = 5 / 3

K1 = K2 – y2 = 10 / 3 – 5 / 3 = 5 / 3

y1 = K1 = 5 / 3

⸫ y1 = y2 = y3 = 5 / 3 & Z = Z = y1y2y3 = 125 / 27

49