DSME 6620

Decision Models and Applications MGT578

Lecture Two

Linear Programming I
ecision Making

ptimization – Part I:
1
ion, and Math Software
DSME 6620
Decision Models and Applications MGT578

Agenda

• Motivation example
– Microcomputer problem
– Problem formulation
– Graphical method
• Learning Excel Solver
– Installing Solver
– Solving the microcomputer problem
– Reading report
• Sensitivity analysis
– Discussions of microcomputer problem
– Reading sensitivity report
– Another view of shadow

2
DSME 6620
Decision Models and Applications MGT578

The Potential Benefits of Mathematical

Models

• Help you develop intuition (“managerial

insights”)
– Sharper understanding of trade-offs
– See how actions influence performance
– Explore “what if” questions (scenario analysis)
• Help you develop a disciplined approach to
decision-making
– Makes you think about key metrics
– You need to be able to formulate
cause-and-effect relationship
– Leads you to better understanding of
constraints
– Leads you to look for and analyze relevant
data
• Disciplined approach to decision-making in
organizations
– Decision are based on intelligent use of
data, rather than on the basis of “feelings,”
“who shouts louder,” “who complains the
most,” etc.
• Help you improve performance of complex
systems
3
– Optimization! Automation!
DSME 6620
Decision Models and Applications MGT578

What is Linear Programming?

• The term linear refers to a characteristic of

the objective function and the constraints.
• A linear program is a mathematical
optimization model in which
– The objective function is a linear function of
the decision variables, and
– The constraints are given by bounds on linear
functions of the decision variables.
– For example, suppose that we have three
decision variables: X, Y, and Z. Then, the
following is a linear program:

4
DSME 6620
Decision Models and Applications MGT578

More on Linearity
OMM5705 Quantitative Decision Making Prof. Aviv Olin Business School, Washington University

More on Linearity

• Linear objective function and constraints

• How does the feasible region look like, when all constraints are
linear?
弧线
not convex

• The feasible region imposed by a set of linear constraints is called a

convex polytope

OMM5705 Quantitative Decision Making Prof. Aviv Olin Business School, Washington University

5
DSME 6620
Decision Models and Applications MGT578

Some More Definitions

• A feasible region is the range of all alternative

courses of actions (i.e., decision variables) that
satisfy the constraints.

• A feasible solution is any point in the feasible

region.
• An optimal solution – a feasible solution that
achieves the best value for the objective
function.
• – It is useful to keep in mind the difference
between “the optimal solution” and “an
optimal solution.” If you got an optimal
solution, and you are not sure if it is the only
one, say “I got an optimal solution.”
• • The optimal value is the best achievable
value of the objective function.

6
DSME 6620
Decision Models and Applications MGT578

Why Study Linear Programming?

• A wide variety of applications.

• Easy to understand.
• Provides you with basic vocabulary relevant
to constrained optimization.
• Can be solved optimally and efficiently
(i.e., not only that you can get to the
optimal value, but it also doesn’t take
forever).
– This is particularly true in spreadsheet
applications.

7
DSME 6620
Decision Models and Applications MGT578

Microcomputer Problem

A manufacturer that assembles computers is about

to start production of two new microcomputers,
Type 1 and Type 2 Microcomputers. Either type of
microcomputer will require assembly time,
inspection time, and storage space. The amount of
each of these resources that can be devoted to the
production of microcomputers is limited. The
resource requirement by either type per unit, as well
as the amount availability, is summarized in the
table below:

Resources Type 1 Type 2 Amount

available
Assembly time 4 hours 10 hours 100 hours
Inspection time 2 hours 1 hour 22 hours
Storage space 3 cubic feet 3 cubic feet 39 cubic feet

The manager of the firm would like to determine the

quantity of either microcomputer to produce in
order to maximize the total profit. The unit profit of
Type 1 Microcomputer is $60, while that of Type 2 is
$50. There is ample demand in the market.
8
DSME 6620
Decision Models and Applications MGT578

Formulation of
Microcomputer Problem
Typically, there are three basic steps in the
formulation process of an optimization problem:
• decision variables,
• objective function, and
• constraints.

1) Decision variables: What decisions will be made?

2) Objective function: What metric will the decision

maker want to optimize?

3) Constraints: What restrictions will limit the

decisions?

9
DSME 6620
Decision Models and Applications MGT578

Formulation of
Microcomputer Problem

Write down the entire formulation:

Max p = 60x1 + 50x2

subject to
4x1 + 10x2 ≤ 100
2x1 + x2 ≤ 22
3x1 + 3x2 ≤ 39
x1 ≥ 0, x2 ≥ 0

Discussions:
1) Do we need to restrict the two decision
variables to be integers?

2) We probably have taken calculus, in which we

use differentiation to solve optimization
problems. But a linear program is an
optimization problem that is solved by methods
other than differentiation.
10
DSME 6620
Decision Models and Applications MGT578

Linear Programming (LP)

• We just formulated the microcomputer

problem as a linear program.

• The word “linear” means that relationships

are linear, and the word “program” means
plan or schedule.

• A diversity of problems can be posed as

linear programs: they are routinely used in
the business world for planning and
managing decision making.

• Linear programs are important!

11
DSME 6620
Decision Models and Applications MGT578

Graphical Method

Since there are only two variables, we can

use a graph to represent the problem. To find
the optimal solution, we can do the following:

Ø We can draw a line for each constraint

and determine which side of the line satisfies
the corresponding constraint.

Ø The lines thus define a feasible region

where any point in this region satisfies all the
constraints.

Ø Represent the objective function by a

movable line that depends on the profit
target.

12
DSME 6620
Decision Models and Applications MGT578

Graphical Method

Line 1: constraint on assembly time availability

Line 2: constraint on inspection time availability
Line 3: constraint on storage space availability

Line 2

objective function line

Line 1
Line 3

From the graph, we find that the optimal

solution corresponds to (x1, x2) = ( , )
with the optimal profit p =
13
DSME 6620
Decision Models and Applications MGT578

Installing Solver in Excel 2010+

• If using Excel 2010 or 2013, click ‘File’ in the menu

and then ‘Options’ on the left. In the new window,
choose Add-ins on the left and select Excel Add-
ins in the Manage box, followed by clicking Go. In
the third window, select the Solver Add-in check
box, and then click OK.

• We can use Solver by clicking ‘Data’ in the menu.

DSME 6620
Decision Models and Applications MGT578

Inputting Data to Excel

Enter parameters and descriptions as shown below:

(see the next slide for formula)

Decision variable cells Constraint cells

B5:C5 E11:G13

Objective function cell

E8
15
DSME 6620
Decision Models and Applications MGT578

Inputting Data to Excel

• Decision variable cells (cells B5:C5):

– Fill zeros or other numbers as initial values

• Objective function cell (cell E8):

– Input “E8 = SUMPRODUCT(B5:C5,B8:C8)”

• Constraint cells (cell E11:G13):

– Input “E11 = SUMPRODUCT(B$5:C$5,B11:C11)”
– Copy E11 to E12:G13

Why place a dollar sign “$” here?

16
DSME 6620
Decision Models and Applications MGT578

Setting Solver Parameters

(Old Version)

Set Target Cell (Objective function): E8 Max

By Changing Cells (Decision variables): B5:C5
Subject to the Constraints: E11 <= G11
E12 <= G12
E13 <= G13 17
DSME 6620
Decision Models and Applications MGT578

Setting Solver Parameters

(New Version)

Set Target Cell (Objective function): E8 Max

By Changing Cells (Decision variables): B5:C5
Subject to the Constraints: E11 <= G11
E12 <= G12
E13 <= G13 18
DSME 6620
Decision Models and Applications MGT578

Setting Solver Options

(Old Version)

Assume Linear Model

Assume Non-Negative
19
DSME 6620
Decision Models and Applications MGT578

Setting Solver Options

(New Version)

Make Unconstrained Variables Non-Negative

Select a Solving Method: Simplex LP 20
DSME 6620
Decision Models and Applications MGT578

Reading Answer Report

Optimal decision
variable values
Optimal objective
(optimal solution)
function value

Constraints:
Binding means LHS = RHS, hence Slack = 0
Not Binding means LHS ≠ RHS, hence Slack ≠ 0

21
DSME 6620
Decision Models and Applications MGT578

Remarks on Solver

• Excel Solver is a useful tool for solving

optimization problems, including linear
programs, integer programs, and nonlinear
programs.

• Pay attention to the following two issues

when using Solver to solve linear programs:
– How to inform Solver that we want to solve
linear programs (as opposed to nonlinear
ones)
– How to inform Solver that some decision
variables are nonnegative

22
DSME 6620
Decision Models and Applications MGT578

Sensitivity Analysis

Discussion: On the supply side, there can be some

variations on the resource availability. Consider a
case in which the inspection time can be
increased (for instance, by asking the inspectors to
work overtime). How much is the firm willing to pay
for the increase in the inspection time?

23
DSME 6620
Decision Models and Applications MGT578

Sensitivity Analysis

Discussion: On the demand side, there can be

some variations on the selling prices of the
microcomputers. Consider a case in which the
profit of Type 1 Microcomputer may change due
to the probable variation in its selling price. What is
range of the profit change within which there is NO
need to vary the production quantities?

24
DSME 6620
Decision Models and Applications MGT578

Reading Sensitivity Report

Ranges for objective coefficient variations

within which the optimal solution keeps same

Shadow prices Ranges for

of constraints’ RHS shadow prices
(such as resources)
25
DSME 6620
Decision Models and Applications MGT578

Another View of Shadow Price

- Primal and Dual Problems

• Another way to understand the shadow

price is to study the so-called primal and
dual. We can view the manufacturer’s
problem as the primal problem:
Max p = 60x1 + 50x2
subject to
4x1 + 10x2 ≤ 100
2x1 + x2 ≤ 22
3x1 + 3x2 ≤ 39
x1 ≥ 0, x2 ≥ 0

• Then its dual problem is defined as

Min w = 100y1 + 22y2 + 39y3
subject to
4y1 + 2y2 + 3y3 ≥ 60
10y1 + y2 + 3y3 ≥ 50
y1 ≥ 0, y2 ≥ 0, y3 ≥ 0

26
DSME 6620
Decision Models and Applications MGT578

Another View of Shadow Price

- Primal and Dual Problems

• Here is an economic explanation of the

dual problem. The setting is that a
businessman wants to acquire the
manufacturer’s business and thus needs to
make a (cash) offer to the manufacturer.
The businessman wonders the minimum offer
price that is acceptable to the
manufacturer.
– y1, y2, y3 :

– Min w = 100y1 + 22y2 + 39y3

– 4y1 + 2y2 + 3y3 ≥ 60

– 10y1 + y2 + 3y3 ≥ 50

• The optimal solution to the dual problem is

(y1, y2, y3) = ( , , ) and w = 27
DSME 6620
Decision Models and Applications MGT578

Summary

• Linear programming is an important and useful

tool for decision making.

• Formulating a decision making problem as

linear programming involves three steps:
– Decision variables
– Objective function
– Constraints

• Excel Solver is a convenient tool for solving

linear programming problems.

28