Post on 10-Apr-2015
description
©Prof. R N Bhattacharya
1
Linear Programming ModelsLinear Programming Models
©Prof. R N Bhattacharya
2
ContentsContents
1. Steps in Developing a Linear Programming (LP) Model
2. Properties of LP Models
3. Basic Assumptions in LP
4. LP Solutions: Four Cases
5. Graphical Method
6. Special Situation in LP
7. Simplex Method
8. Big M Method
9. Two Phase Method
10. Duality in Linear Programming Problems
11. Introduction to Sensitivity Analysis
©Prof. R N Bhattacharya
3
Steps in Developing a Linear Programming (LP) ModelSteps in Developing a Linear Programming (LP) Model
1) Formulation
2) Solution
3) Interpretation and Sensitivity Analysis
©Prof. R N Bhattacharya
4
Properties of LP ModelsProperties of LP Models
1) A single objective - minimize or maximize
2) Include “constraints” or limitations
3) There must be alternatives available
4) All equations are linear, all variables are continuous
©Prof. R N Bhattacharya
5
Basic Assumptions in LPBasic Assumptions in LP
o Proportionality
o The contribution to the objective function from each decision variable is proportional to the value of the decision variable
o The contribution of each decision variable to the LHS of each constraint is proportional to the value of the decision variable
o Additivity
o The contribution to the objective function for any decision variable is independent of the values of the other decision variables
o The contribution of a decision variable to LHS of each constraint is independent of the values of other decision variables
o Divisibility
o Each decision variable is allowed to assume fractional values. If we actually can not produce a fractional number of decision variables, we use IP
o Certainty
o Each parameter is known with certainty
©Prof. R N Bhattacharya
6
LP Solutions: Four CasesLP Solutions: Four Cases
When an LP is solved, one of the following four cases will occur:1. The LP has a unique optimal solution.2. The LP has alternative (multiple) optimal solutions. It has more than one
(actually an infinite number of) optimal solutions3. The LP is infeasible. It has no feasible solutions (The feasible region contains
no points).4. The LP is unbounded. In the feasible region there are points with arbitrarily
large (in a max problem) objective function values.
Graphically,o In the unique optimal solution case, isoprofit line last hits a point (vertex -
corner) before leaving the feasible region.o In the alternative optimal solutions case, isoprofit line last hits an entire line
(representing one of the constraints) before leaving the feasible region.o In the infeasible case, there is no feasible region.o In the unbounded case, isoprofit line never lose contact with the feasible
region
©Prof. R N Bhattacharya
7
Linear Programming- Linear Programming- Graphical Method
©Prof. R N Bhattacharya
8
Example: A Furniture Manufacturing CompanyExample: A Furniture Manufacturing Company
• A furniture company manufactures chairs tables. Each type requires different amounts of construction time and painting time, as given in the table below. The company wants to know how many of each type must be produced in order to maximize profit.
Other Limitations:• Make no more than 450 chairs• Make at least 100 tables
Tables (X1)
(per table)
Chairs (X2)
(per chair)
Hours Available
Profit Contribution Rs.7 Rs.5
Carpentry 3 hrs 4 hrs 2400
Painting 2 hrs 1 hr 1000
©Prof. R N Bhattacharya
9
Decision Variables:Let
X1 = Number of tables to make
X2 = Number of chairs to make
Objective Function: Maximize Profit
Maximize 7 X1 + 5 X2
Constraints
• Have 2400 hours of carpentry time available
3 X1 + 4 X2 < 2400(hours)
• Have 1000 hours of painting time available
2 X1 + 1 X2 < 1000 (hours)
More Constraints:• Make no more than 450 chairs
X2 < 450• Make at least 100 tables
X1 > 100
Non negativity:Cannot make a negative number of chairs or tables
X1 > 0
X2 > 0
©Prof. R N Bhattacharya
10
Model Summary
Maximize 7 X1 + 5 X2 (profit)
Subject to the constraints:
3 X1 + 4 X2 < 2400 (carpentry hrs)
2 X1 + 1 X2 < 1000 (painting hrs)
X2 < 450 (max # chairs)
X1 > 100 (min # tables)
X1, X2 > 0 (non negativity)
©Prof. R N Bhattacharya
11
Steps
• Step 1: Plot the constraints on a graph
• Step 2 : Locate all the corner points on the graph
• Step 3 : Test all corner points to see which yields maximum profit.
©Prof. R N Bhattacharya
12
Carpentry Constraint Line 3 X1 + 4 X2 < 2400
Intercepts
(X1 = 0, X2 = 600)
(X1 = 800, X2 = 0)
0 800 X1
X2
600
0
Feasible
< 2400 hrs
Infeasible
> 2400 hrs3X
1 + 4X2 = 2400
©Prof. R N Bhattacharya
13
Painting Constraint Line
2 X1 + 1 X2 = 1000
Intercepts
(X1 = 0, X2 = 1000)
(X1 = 500, X2 = 0)
0 500 800 X1
X2
1000
600
0
2 X1 + 1 X
2 = 1000
©Prof. R N Bhattacharya
14
0 100 500 800 X1
X2
1000
600
450
0
Max Chair Line
X2 = 450
Min Table Line
X1 = 100
Feasible
Region
0 100 200 300 400 500 X1
X2
500
400
300
200
100
0
Objective Function Line
7 X1 + 5 X2 = Profit
7 X1 + 5 X
2 = 2,100
7 X1 + 5 X
2 = 4,040
Optimal Point(X1 = 320, X2 =
360)
7 X1 + 5 X
2 = 2,800
0 100 200 300 400 500 X1
X2
500
400
300
200
100
0
Additional Constraint
Need at least 75 more chairs than tables
X2 > X1 + 75
Or
X2 – X1 > 75
X1 = 320
X2 = 360
Now infeasible
New optimal pointX1 = 300, X2 = 375
©Prof. R N Bhattacharya
17
LP CharacteristicsLP Characteristics
• Feasible Region: The set of points that satisfies all constraints
• Corner Point Property: An optimal solution must lie at one or more corner points
• Optimal Solution: The corner point with the best objective function value is
optimal
©Prof. R N Bhattacharya
18
Special Situation in LPSpecial Situation in LP
1. Redundant Constraints - do not affect the feasible region
Example: x < 10
x < 12
The second constraint is redundant because it is less restrictive. 2. Infeasibility – when no feasible solution exists (there is no feasible
region)
Example: x < 10
x > 15
Special Situation in LPSpecial Situation in LP
3. Alternate Optimal Solutions – when there is more than one optimal solution. Note that if there are more than one optimal solution, there will be infinite number of optimal solutions.
Max 2X1 + 2X2
Subject to:X1 + X2 < 10X1 < 5 X2 < 6
X1, X2 > 0
0 5 10 X1
X2
10
6
0
2X1 + 2X
2 = 20 All points on Blue segment are optimal
Special Situation in LP
4. Unbounded Solutions – when nothing prevents the solution from becoming
infinitely large
Max 2X1 + 2X2
Subject to: 2X1 + 3X2 > 6
X1, X2 > 0
0 1 2 3 X1
X2
2
1
0
Directi
on
of solutio
n
©Prof. R N Bhattacharya
21
Exercise
Alpha ltd. produces two products X and Y each requiring same production
capacity. The total installed production capacity is 9 tones per days. Alpha Ltd. Is
a supplier of Beta Ltd. Which must supply at least 2 tons of X & 3 tons of Y to
Beta Ltd. Every day. The production time for X and Y is 20 machine hour pr units
& 50 machine hour per unit respectively the daily maximum possible machine
hours is 360 profit margin for X & Y is Rs. 80 per ton and Rs. 120 per ton
respectively. Formulate as a LP model and use the graphical method of
generating the optimal solution for determining the maximum number of units of
X & Y, which can be produced by Alpha Limited.
©Prof. R N Bhattacharya
22
Note that the region of feasible solution shown in the following figure is bounded by the graphs of the Linear equalities:
x 1+ x2 =9, x 1=2, x2 = 3 and 20x1 + 50x2 =360, and by the coordinates axes.
©Prof. R N Bhattacharya
23
©Prof. R N Bhattacharya
24
Simplex MethodSimplex Method
©Prof. R N Bhattacharya
25Simplex Method – Introduction
o Note that in the example considered of the graphical method, the optimal solution occurred at a vertex (corner) of the feasible region. In fact it is true that for any LP the optimal solution occurs at a vertex of the feasible region.
o This fact fact is the key to the simplex algorithm for solving LP's.
o The Simplex algorithm for solving LPP was developed by Dantzig in the late 1940’s and since then a number of different versions of the algorithm have been developed. One of these later versions, called the revised simplex algorithm (sometimes known as the "product form of the inverse" simplex algorithm) forms the basis of most modern computer packages for solving LP's.
o Steps:
1. Convert the LP to standard form
2. Obtain a basic feasible solution (bfs) from the standard form
3. Determine whether the current bfs is optimal. If it is optimal, stop.
4. If the current bfs is not optimal, determine which nonbasic variable should become a basic variable and which basic variable should become a nonbasic variable to find a new bfs with a better objective function value
5. Go back to Step 3.
©Prof. R N Bhattacharya
26
Simplex Method – Introduction
• Related Concepts:
Standard form: all constraints are equations and all variables are nonnegative
bfs: any basic solution where all variables are nonnegative
Nonbasic variable: a chosen set of variables where variables equal to 0
Basic variable: the remaining variables that satisfy the system of equations at the standard form
©Prof. R N Bhattacharya
27
Simplex Method - Example
• A furniture company manufactures 2 types of dining tables. Each type requires different amounts of construction time and painting time, as given below. The company wants to know how many tables of each type must be produced in order to maximize profit.
Type 1 Type 2 Available
(hrs/month)
Decision variable x1 X2
Construction time (hrs/table)
6 12 120
Painting time (hrs/table)
8 4 64
Profit (Rupees/ table)
200 240
©Prof. R N Bhattacharya
28
Simplex Method - Example
• The Model:
Max z = 200 x1 + 240 x2 (Objective function)Subject to:6 x1 +12 x2 < 120
8 x1 + 4 x2 < 64
X1,x2 > 0
©Prof. R N Bhattacharya
29
Simplex method – Converting to standard form
Introducing slack variables s1 and s2 and modifying objective function suitably:
Max z = 200 x1 + 240 x2 + 0 s1 + 0 s2
S.T
6 x1 + 12 x2 + s1 + 0 s2 = 120
8 x1 + 4 x2 + 0 s1 + s2 = 64
©Prof. R N Bhattacharya
30Starting simplex table
Label BV x1 x2 s1 s2 Value Value/Coefficient Formula
a0 s1 6 12 1 0 120 120/12 = 10
b0 s2 8 4 0 1 64 64/4 = 16
O0 z -200 -240 0 0 0
Interpretation of coefficients in row O:
Each coefficient represents the increase (for –ve coeff.) or decrease (for + ve coeff.) in Z with a unit increase of the associated non-basic variable.
©Prof. R N Bhattacharya
31Simplex table - Pivoting
Label BV x1 x2 s1 s2 Value Value/Coefficient Formula
a0 s1 6 12 1 0 120 120/12 = 10
b0 s2 8 4 0 1 64 64/4 = 16
O0 z -200 -240 0 0 0
Pivot column = lowest value in z row
Pivot row = lowest value in V/C
Pivot Number Entering Variable
Leaving variable
©Prof. R N Bhattacharya
32Simplex table – Iteration 1
Label BV x1 x2 s1 s2 Value V/C
Formula
a1 x2 6/12 = 1/2
12/12 = 1
1/12 0/12 = 0
120/12 =10
- a1=a0/12
b1 s2 8-4x(6/12)=
6
4-4x1=
0
0-4x(1/12
)=
-4/12
1-4x (0) =
1
64- 4x 10 = 24
- b1= b0 – 4a1
O1 z -200+(240x ½) =
-80
-240+ (240x1)
=
0
0+240x 1/12 =
20
0+ 240x0
=
0
0 +240x10
=
2400
O1= O0+240a1
Make the pivot number = 1 and the vales of all other cells in pivot column = zero through elementary operations
New Pivot col. For Iteration 2 (most –ve in z row)
©Prof. R N Bhattacharya
33
Simplex table – Iteration 2
Label BV x1 x2 s1 s2 Value V/C Formula
a2 x2½-(½)x1 =
01-(½)x0
= 1
1/12 –
(½) x (-1/18)0- ½)x (1/6)
10-(½)x4=
8
a2=a1 – (½)b2
b2 x1
6/6 = 1
0/6 =
0
(-1/3)
x(1/6)=
- 1/18
1/6 24/6 =
4
b2= b1/6
O2 z -80+80x1= 0
0+80x0=
0
20 + 80x
(-1/18)=
280/18
0+80x
(1/6)=
80/6
2400+
80x4=
2720
O2= O1+80b2
Make the pivot number = 1 and the vales of all other cells in pivot column = zero through elementary operations
All values in Z row are non-negative, hence this is optimal solution.
Make 4 units of x1 and 8 units of x2, max. profit = 2720
©Prof. R N Bhattacharya
34
Big M MethodBig M Method
(Charnes Method of Penalties)
©Prof. R N Bhattacharya
35
Big M Method (Charnes method of penalties)
The original problem:
Max z = 3x1 – x2
ST: 2 x1 + x2 > 2
x1 + 3 x2 < 3
x2 <4
x1,x2 > 0
Adding slack, surplus and artificial variables and writing the problem in standard form, we get:
Max z = 3 x1 – x2 + 0 s1 + 0 s2 + 0 s3 – M A
ST: 2 x1 + x2 – s1 + A = 2
x1 + 3 x2 + s2 = 3
x2 + s3 = 4
Let:
A = an artificial variable
s1 = a surplus variable
s2 and s3 = slack variable
Big M
©Prof. R N Bhattacharya
36
Big M Method – Initial Operation – Eliminate M
Label BV x1 x2 s1 s2 s3 A Val V/C Formula
a0 A 2 1 - 1 0 0 1 2
b0 s2 1 3 0 1 0 0 3
c0 s3 0 1 0 0 1 0 4
O0* z -3 1 0 0 0 M 0
O0 z -3 – 2M 1 - M M 0 0 0 -2MO0 = O0* - M a0
The initial operations removes M from the Z row of artificial variable vector using formula O0 = O0* - M a0. This gives us the starting table, as shown in next slide.
©Prof. R N Bhattacharya
37
Big M Method – Pivoting and solving
Label BV x1 x2 s1 s2 s3 A Val V/C Formula
a0 A2
1 - 1 0 0 1 2 2/2=1
b0 s2 1 3 0 1 0 0 3 3/1=3
c0 s3 0 1 0 0 1 0 44/0
Ignore
O0 z -3 – 2M 1 - M M 0 0 0 -2M
Now, proceed as in normal simplex method. The most – ve Z row value is
–3 – 2M, hence it becomes the pivot column and V/C=1 is minimum so it is pivot row.
Solving, after two iterations, the optimal solution is found to be:
X1 = 3, x2 = 0, Max Z = 9 . (In the final table, s1=4, s3 = 4)
Most negative
MinimumPivot number
©Prof. R N Bhattacharya
38
Some Notes on artificial variables
• Artificial Variable:The idea of using artificial variable calls for adding a non negative variable to the left side of each equation that has no obvious starting basic variable.
• However, since such artificial variables have no physical meaning from the standpoint of the original problem (hence the name ‘artificial’), the procedure will be valid only if we force these variables to be zero when the optimum is reached.
• Hence, in the optimal table (optimal table = all coefficients in the objective function are non negative), if the artificial variable remained in the ‘basis’ (basis = the variables appearing in the BV column as Basic Variable) with a positive value, it means there is no feasible solution to the problem.
• There are two methods based on the idea of artificial variables. The first is the Big M method, and the second is called the Two Phase Technique.
©Prof. R N Bhattacharya
39
Some Notes on artificial variables
Artificial Variables are also used when we have a constraint equation with equality sign. For example:
Min Z = 4 x1 + x2
Subj.To. : 3x1 + x2 = 3
x1 + 2 x2 =< 4
x1, x2 > 0
Is written as :
Min Z = 4 x1 + x2 + 0 S1 – M A
Subj.To. 3x1 + x2 + A = 3
x1 + 2 x2 + s1 = 4
x1, x2 > 0
©Prof. R N Bhattacharya
40
Two Phase MethodTwo Phase Method
©Prof. R N Bhattacharya
41
Two Phase MethodTwo Phase Method
• A drawback of the Big-M method is the possible computational error that could result from assigning a very large value to the constant M while solving on a digital computer.
• For example suppose M = 1,00,000. Then if in the objective function row we have –4+M it would be (-4+7,00,000).
• The effect of the original coefficient ‘4’ is now too small relative to the large number created by the multiple of M.
• Because of round-off error, which is inherent in any digital computer, the computation may become insensitive to the relative value of the original objective function coefficient.
• The dangerous outcome is that the variables x1, x2 etc. may be treated as having zero coefficients in the objective function.
• The Two Phase Method is designed to alleviate this difficulty. Although the artificial variables are added in the same manner, the use of the constant M is eliminated by solving the problem in two phases.
©Prof. R N Bhattacharya
42
Two Phase MethodTwo Phase Method
Phase 1
1. Augment the artificial variables as necessary to secure a starting solution. 2. Form a new objective function which seeks the minimization of the sum of the artificial variables subject to the constraints of the original problem, modified by the artificial variables.
3. Go to Phase 2. 4. Otherwise, if the minimum is positive (I.e. max. is –ve), the problem has no feasible solution, hence STOP.
Phase 2
Use the optimum basic solution of phase 1 as a starting point for the original problem.
If the minimum value of the new objective function is zero (meaning that all artificials are zero, whether in the basis or otherwise), the problem has a feasible solution space.
©Prof. R N Bhattacharya
43
Two Phase Method - ExampleTwo Phase Method - Example
Min Z = 4 x1 + x2
Subj To 3 x1 + x2 = 3 ……(1)
4 x1 + 3 x2 > 6 …(2)
x1 + 2 x2 < 4 ……(3)
x1, x2 > 0
We have :
3 x1 + x2 + A1 =3 ……………….(1)
4 x1 + 3 x2 – s1 + A2 = 6 ……….(2)
x1 + 2 x2 + s2 = 4 ……………..(3)
From (1) A1 = 3 – 3 x1 – x2
From (2) A2 = 6 – 4 x1 – 3 x2 + s1
Substituting, we have the new objective function is
Min A = 9 – 7 x1 – 4 x2 + s1
Introduce Artificial Variables A1 into equation 1, Artificial variable A2 and surplus variable s1 into equation 2 and slack variable s2 into equation 3.
The new objective function to be MINIMIZED is the sum of the artificial variables.
Thus Min A = A1 + A2
©Prof. R N Bhattacharya
44
Two Phase Method - ExampleTwo Phase Method - Example
Thus the problem in Phase 1 is :
Min A = 9 – 7 x1 – 4 x2 + s1
Subj To 3 x1 + x2 + A1 =3 ……………….(1)
4 x1 + 3 x2 – s1 + A2 = 6 ……….(2)
x1 + 2 x2 + s2 = 4 ……………..(3)
x1, x2,s1,s2,A1,A2 > 0
Converting to maximization (by multiplying the RHS of objective function with –1) we have Max A* = - 9 + 7 x1 + 4 x2 - s1 , or
Max A* – 7 x1 – 4 x2 + s1 = -9
Subj To 3 x1 + x2 + A1 =3 ……………….(1)
4 x1 + 3 x2 – s1 + A2 = 6 ……….(2)
x1 + 2 x2 + s2 = 4 ……………..(3)
x1, x2,s1,s2,A1,A2 > 0
The starting solution is:
A1 = 3, A2 =6, s2 =4
and starting value of objective function A* = - 9
©Prof. R N Bhattacharya
45
Two Phase Method - ExampleTwo Phase Method - Example
Using Simplex method to solve this Phase 1 problem, after two iterations we get the following optimal simplex table:
Label BV x1
x2 A1 A2 s1 s2 Val
a2
x11 0 3/5 -1/5 1/5 0 3/5
b2
x20 1 -4/5 3/5 -3/5 0 6/5
c2
s20 0 1 -1 1 1 1
O2 A* 0 0 1 1 0 0 0
Opt. Solution to Phase 1:x1= 3/5, x2=6/5, s2= 1
Max A* = 0 = (-Min A) = 0
Since Min A = 0, the problem has a feasible solution and we can move to phase 2.
©Prof. R N Bhattacharya
46
Phase 2Phase 2
• Step 1With the help of the optimal table of phase 1, write the equivalent constraint equations, noting that the optimal table is nothing but equations written in detached coefficient form. Thus we have:
1x1 +0 x2 +(3/5)A1 – (1/5)A2 + (1/5)s1 + 0 s2 = 3/50x1 +1 x2 -(4/5)A1 + (3/5)A2 - (3/5)s1 + 0 s2 = 6/50x1 +0 x2 +(1)A1 - (1)A2 + (1)s1 + (1) s2 = 1
• Step 2Since artificial variables have served their purpose in phase 1, rewrite the constraints neglecting the artificial variables. This gives:x1 + (1/5)s1 = 3/5x2 - (3/5)s1 = 6/5s1 + s2 = 1
• Our starting basis is x1 = 3/5, x2 = 6/5, s2 = 1
• The Original Obj. Function is Min Z = 4 x1 + x2
©Prof. R N Bhattacharya
47
Phase 2Phase 2
• Step 3Since x1 and x2 are in the basis, eliminate them from objective function by using the constraint equations.The constraint equations give:x1 = 3/5 – (1/5)s1 and x2 = 6/5 + (3/5)s1
Substituting these in objective function, we have the following modified obj.fn.:
Min Z = 18/5 – (1/5)S1 or Max Z* = (1/5) S1 – (18/5)
• The LPP can thus be written as:
Max Z* = (1/5) S1 – (18/5)
Subj.to x1 + (1/5)s1 = 3/5x2 - (3/5)s1 = 6/5s1 + s2 = 1
©Prof. R N Bhattacharya
48Phase 2 Phase 2 simplex table
Label BV x1 x2 s1 s2 Value V/C
a0
x11 0 1/5 0 3/5
(3/5) / (1/5)
= 3
b0 x2 0 1 -3/5 0 6/5(6/5) / (-3/5) =
-ve value so neglect it
c0 s2 0 0 1 1 1 1/1 = 1
O0 Z* 0 0 -1/5 0 -18/15
a1 x1 1 0 0 -1/5 2/5 a1=a0 - (1/5) c1
b1 x2 0 1 0 3/5 9/5 b1=b0 + (3/5) c1
c1 s1 0 0 1 1 1 C1 = c0
O1 Z* 0 0 0 1/5 -17/5 O1 = O0 + (1/5) c2
Most -ve
All values in row O1 are non-negative, hence the optimal solution is x1= 2/5, x2=9/5, s1=1,s2=0, Min Z = - Max Z* = 17/5
©Prof. R N Bhattacharya
49
Duality in Linear Programming Duality in Linear Programming ProblemsProblems
The Duality TheoremThe Duality Theorem ““For every maximization (or, minimization) problem in For every maximization (or, minimization) problem in linear programming, there is a unique similar problem linear programming, there is a unique similar problem of minimization (or, maximization) involving the same of minimization (or, maximization) involving the same data which describes the original problem.”data which describes the original problem.”
©Prof. R N Bhattacharya
50
Duality in Linear Programming Problems
• For every LPP there is a corresponding LPP called the dual.
• If the original problem is a maximization problem then the dual problem is minimization problem and if the original problem is a minimization problem then the dual problem is maximization problem.
• In either case the final table of the dual problem will contain both the solution to the dual problem and the solution to the original problem.
• The solution of the dual problem is readily obtained from the original problem solution if the simplex method is used.
• The formulation of the dual problem, also sometimes referred as the concept of duality, is helpful for the understanding of the linear programming. The variables of the dual problem are known as the dual variables or shadow prices of the various resources.
©Prof. R N Bhattacharya
51
Rules for writing Duals 1. Define a Dual Variable for each Primal (Constraint) equation. 2. Define a Dual Constraint for each Primal Variable. 3. The constraint (column) coefficients of a primal variable defines the left
hand side coefficients of the dual constraint and its objective coefficient define the right-hand side.
4. The objective coefficients of the dual equal the right-hand side of the primal constraint equations.
Considerations:• If a constraint in Primal has an equal sign, the corresponding Dual
Variable is Unrestricted.• If a Primal is a Maximization (or Minimization) Problem, the
corresponding Dual is Minimization (or Maximization).• The sign of a Dual Variable is same as its Primal Constraint if it is a Dual
Maximization.• The sign of a Dual Variable is opposite of its Primal Constraint if it is a
Dual Minimization.
©Prof. R N Bhattacharya
52
Rules for writing Duals
• For Minimization Primal The constraints should be having sign ‘≥’ If it is ‘<’ then make it ‘≥’ by multiplying the constraint by ‘-1’ on both sides Then form the dual in normal way. After that, replace the variable multiplied with
‘-1’. Example: Y1=-Y2.• For Maximization Primal
The constraints should be having sign ‘<’ Follow the same process to change the sign if there is ‘>’ in the constraint.
• If there is ‘=’ sign then form 2 constraints one with ‘≥’ and the other with ‘<’ Example: 2X1+3X2+5X3=120 form two constraints as follows:
o 2X1+3X2+5X3 < 120o 2X1+3X2+5X3 > 120
The 2 constraints thus formed would gives the ‘=’ constraint. There will be 2 dual variables now. They will be of form Y1-Y2. Replace
them with a single variable Y3 and since Y1-Y2 can give any value, Y3 will be unrestricted.
©Prof. R N Bhattacharya
53
Sl.No.
Primal Dual
Dual Primal
Maximization Minimization
1 Constraints Variables
2 ≥ ≤ 0
3 ≤ ≥ 0
4 = Unrestricted
5 Variables Constraints
6 ≥ 0 ≥
7 ≤ 0 ≤
8 Unrestricted =
Rules for writing Duals
©Prof. R N Bhattacharya
54
Some Examples of Primal-Dual
Primal-1
Max Z = 200 X1 + 240 X2
S.T. 6 X1 + 12 X2 < 120
8 X1 + 4 X2 < 64
X1, X2 > 0
Dual-1
Min C = 120 y1 + 64 y2
S.T 6 y1 + 8 y2 > 200
12 y1 + 4 y2 > 240
y1, y2 > 0
Primal-2
Max Z = 40 X1 + 60 X2
S.T. 3 X1 + 2 X2 < 2000
X1 + 2 X2 < 1000
X1, X2 > 0
Dual-2
Min C = 2000 y1 + 1000 y2
S.T 3 y1 + y2 > 40
2 y1 + 2 y2 > 60
y1, y2 > 0
©Prof. R N Bhattacharya
55
Understanding Duality
Consider the furniture shop example with the following formulation:
Max Z = 200 X1 + 240 X2
S.T. 6 X1 + 12 X2 < 120 ….. Construction constraint … (y1)
8 X1 + 4 X2 < 64 …….. Painting constraint ……..(y2)
X1, X2 > 0
Dual formulation:
Min C = 120 y1 + 64 y2
S.T 6 y1 + 8 y2 > 200
12 y1 + 4 y2 > 240
y1, y2 > 0
Suppose our goal was to know the base minimum prices at which the company could rent out these two departments to outside parties.
Let y1 and y2 be the rents per hour at which the company is willing to rent out. Then the total monthly rental is:
C = 120 y1 + 64 y2.
The lowest value of this is the minimum that the company should charge, hence the obj. fn. Is Min (C)
©Prof. R N Bhattacharya
56
Understanding Duality – contd.
• The rents y1 and y2 should be competitive with available alternatives. The available alternative is to produce X1 and X2 units of furniture using the construction and painting resources.
• For example, since 6 hrs of construction and 8 hrs of painting time are needed to produce one unit of X1, the valuation of these two resources, if they are to be rented out, would be 6 y1 + 8 y2, and this total rental value must be > 200 (which is the profit obtained from Type 1 furniture).
• Hence one of the constraints is 6 y1 + 8 y2 > 200. Using similar reasoning for the resources of type 2 furniture, we get 12 y1 + 4 y2 > 240
• Also, we can not have any negative rents per hour, hence y1, y2 >0
©Prof. R N Bhattacharya
57
Shadow Costs and Shadow Prices
• Shadow CostsThe coefficients of the Z row in the optimal simplex table of the original variables are called shadow costs.
• Shadow PricesIn the optimal table, the Z row coefficients of the slack variables are called shadow prices for the corresponding scarce resource.
• Example: Below is the optimal simplex table:
BV X1 X2 X3 X4 S1 S2 S3 Val
X1 1 5/7 0 -5/7 10/7 0 -1/7 50/7
S2 0 -6/7 0 13/7 -61/7 1 4/7 325/7
X3 0 2/7 1 12/7 -3/7 0 1/7 55/7
Z 0 3/7 0 11/7 13/7 0 5/7 695/7
Shadow Costs Shadow Prices
©Prof. R N Bhattacharya
58
Understanding Shadow costs and prices
BV X1 X2 X3 X4 S1 S2 S3 Val
X1 1 5/7 0 -5/7 10/7 0 -1/7 50/7
S2 0 -6/7 0 13/7 -61/7 1 4/7 325/7
X3 0 2/7 1 12/7 -3/7 0 1/7 55/7
Z 0 3/7 0 11/7 13/7 0 5/7 695/7
Suppose we decide to set X2= 1. The shadow cost (3/7) in the Z row under the variable X2 tells us that the value of Z would decrease by 3/7
Again, the shadow price of S1 is 13/7. It indicates that if the first scarce resource (I.e. the first constraint) can somehow be improved by one unit (i.e. the RHS of the first constraint is increased by 1 unit) this will improve the profit level by 13/7.
Note however that since S2 is in the basis of the opt. table, any improvement in the second resource (second constraint) will not improve the profit. This is evident from the value of the shadow cost for S2 being equal to zero as above.
©Prof. R N Bhattacharya
59
Finding solution of Dual from the solution of the primal
Find the solution of the dual model from the solution of the primal model when the primal is:
Max Z = 5 X1 + 70 X2
Subj. to 3 X1 + 2 X2 < 2000
x1 + 2 x2 < 1000 , X1, X2 > 0
The final simplex table is:
BV X1 X2 S1 S2 Value
S1 2 0 1 -1 1000
X2 1/2 1 0 1/2 500
Z 30 0 0 35 35000
The opt. value of C, y1 and y2 are found as follows:
y1 = Coeff. Of S1 in Z row = 0
y2 = Coeff. Of S2 in Z row = 35
Min cost C = Value of Z = 35,000
©Prof. R N Bhattacharya
60
Introduction to Sensitivity AnalysisIntroduction to Sensitivity Analysis
©Prof. R N Bhattacharya
61
Sensitivity Analysis
• Sensitivity analysis is concerned with studying possible changes in the available optimal solution as a result of making changes in the original model.
• There could be five basic types of changes in the original model:1. Changes in the right hand side constants bi – It affects feasibility
2. Changes in the cost coefficient of objective function c j – It affects optimality
3. Changes in the coefficients of the technology matrix cij – It affects optimality
4. Addition of new variable – It affects both optimality and feasibility
5. Addition of new constraint – It affects feasibility
• In General, when a parameter is changed, it results in one of the following three cases:
The optimal solution remains unchanged The basic variables remain the same but their values are changed The basic variables as well as their values are changed.
©Prof. R N Bhattacharya
62
Sensitivity Analysis – An illustration
Max Z = 3 X1 + 4 X2 + X3 + 7 X4
Subj. to 8 X1 + 3 X2 + 4 X3 + X4 < 7
2 X1 + 6 X2 + X3 + 5 X4 < 3
X1 + 4 X2 + 5 X3 + 2 X4 < 8
Xi > 0 I = 1,2,3,4
Find the limits between which the elements of the requirement vector (7,3,8) can be changed one at a time so that an optimal solution may still be obtained
Requirement Vectors
©Prof. R N Bhattacharya
63
Sensitivity Analysis – An illustration
BV X1 X2 X3 X4 S1 S2 S3 Val
S1 8 3 4 1 1 0 0 7
S2 2 6 1 5 0 1 0 3
S3 1 4 5 2 0 0 1 8
Z -3 -4 -1 -7 0 0 0 0
Starting Table
BV X1 X2 X3 X4 S1 S2 S3 Val
X1 1 9/38 1/2 0 5/38 -1/38 0 16/19
X4 0 21/19 0 1 -1/19 8/38 0 5/19
S3 0 59/38 9/2 0 -1/38 -15/38 1 126/19
Z 0 169/38 1/2 0 1/38 53/38 0 83/19
Final Optimal Table
©Prof. R N Bhattacharya
64
Sensitivity Analysis – An illustration
The current optimal values are X1=16/19, X4=5/19 and S3=126/19.
Let the RHS of the first constraint be changed from 7 to 7 + b1.
Notice that S1 is associated with the first constraint, hence look up the coefficient under the S1 column in the final table.
These are (5/38,-1/19,-1/38).
So, a change of b1 in RHS of first constraint will change the values of the current basic variables (optimal) as follows:
X1 will be changed from 16/19 to (16/19) + (5/38) b1
X4 will be changed from 5/19 to (5/19) - (1/19) b1
S3 will be changed from 126/19 to (126/19) - (1/38) b1
Imposing non-negativity constraints for X1,X4,S3 we get three inequalities as discussed in next slide.
©Prof. R N Bhattacharya
65
Sensitivity Analysis – An illustration
• (16/19) + (5/38) b1 > 0 b1 > -32/5
(5/19) - (1/19) b1 > 0 b1 < 5
(126/19) - (1/38) b1 > 0 b1 < 252
The third inequality is redundant as b1 < 5 implies b1 < 252
Hence:
If the RHS of the first constraint is changed within the range
-32/5 < b1 < 5
the present basic variable will remain feasible.
• Similarly, the range of the others may be found.