DSME 4020B

Decision Modeling and Analytics

Spring 2022
Lecture 5

Sean ZHOU
CUHK Business School
Chinese University of Hong Kong

1
 Lecture outline
• The Geometry of Linear Programming

• Solving Linear Programming in Excel (Solver)

2
The Geometry of Linear Programming

3
 Geometry of Linear Programming
• How to visualize a linear program geometrically?
– Decision variables?
– Constraints?
– Objective?
• How to find the optimal solution geometrically (for simple
problems)?

• An example:

4
 Geometry of LP – Decision Variables
• Each dimension represents a decision variable
n decision variables ⇒ an n-dimensional space
• Each point in the space represents a particular solution
x2

1 (1, 1)

0 1 x1 5
 Geometry of LP – Constraints
• Constraint: x1 ≥ 0
• Inequality constraint x1 ≥ 0 defines a half space

X2 X1 ≥ 0

x1 = 0

0 X1
6
 Geometry of LP – Constraints
• Constraint: x2 ≥ 0
• Inequality constraint x2 ≥ 0 defines a half space

X2 X2 ≥ 0

0 x2 = 0 X1
7
 Geometry of LP – Constraints
• Constraint: 2x1 + x2 ≤ 3
• Inequality constraint 2x1 + x2 ≤ 3 defines a half space

X2
2x1+x2 = 3
(0, 3)

0 (1.5, 0) X1
2x1 + x2 ≤ 3
8
 Geometry of LP – Constraints
• Constraint: x1 + 2x2 ≤ 3
• Inequality constraint x1 + 2x2 ≤ 3 defines a half space

X2

x1+2x2 = 3

(0, 1.5)

0 (3, 0) X1
x1 + 2x2 ≤ 3 9
 Geometry of LP – Constraints
• The intersection of all the constraints ⇒ Feasible Region

X2

0 X1
10
 Geometry of LP – Constraints
• Feasible Region: The set of all the allowed solutions; a region
(polygon) bounded by the constraints
– Each equality constraint defines a line
– Each inequality constraint defines a half-space
• Extreme Points: Corner points on the boundary of the feasible
region. E.g., (0, 0), (1.5, 0), (0, 1.5), and (1, 1)
• Infeasible problem: A problem with an empty feasible region
• Redundant constraint: Adding or removing the constraint does
not affect the feasible region, e.g., X2
𝑥1 + 𝑥2 ≤ 5

11
0 X1
 Geometry of LP – Objective
• Objective: maximize x1 + x2
• Isoquant: A line on which all points have the same objective
value; all points are equally good on the objective function.
X2

x1 + x 2 = 0

0 X1
12
 Geometry of LP – Objective
• Isoquant: x1 + x2 = 1

X2

x1 + x 2 = 1

(0, 1)

0 (1, 0) X1
 Geometry of LP – Objective
• Isoquant: x1 + x2 = 1.5

X2

x1 + x2 = 1.5

0 X1
 Geometry of LP – Objective
• Isoquant: x1 + x2 = 3
x1 + x 2 = 3
X2
No feasible
solution can
attain this value

0 X1
15
Geometry of LP – Optimal Solution
• Optimal Solution: The best feasible solution
• Theorem: For any feasible LP with a finite optimal solution
and at lest one extreme point, there exists an optimal
solution that is an extreme point
Max x1 + x2 X2
Optimal Solution:
(1, 1)

0 X1
x1 + x 2 = 2 16
Geometry of LP – Optimal Solution
• Optimal solutions may NOT be unique

X2
Max 2x1 + x2

0 X1
17
Geometry of LP – Optimal Solution
• Optimal solution may NOT be finite

-X1 + 2X2 ≤ 2
X2
Max x1 + x2

X1 - 2X2 ≤ 2

(0, 1)

0 (2, 0) X1
18
Geometry of LP – Optimal Solution
• Binding (or active) constraints: The constraints that are
satisfied as equality at the optimal solution
– All equality constraints are binding by definition

• Non-binding (or inactive) constraints are satisfied as

strict inequality at the optimal solution

• The inequality level (= RHS – LHS (for ``≤ ′′constraint)) is

known as the slack
– Binding constraints have zero slack by definition

19
 Geometry of LP – Optimal Solution
• Active constraints are the ones that pass through the optimal solution.
• Inactive constraints are the ones that do not pass through the optimal
solution.
• Why are binding/active constraints important?

Constraints Binding? Slack

Optimal X1 + 2X2 ≤ 3 Yes 0
(1, 1) 2X1 + X2 ≤ 3 Yes 0
X1 ≥ 0 No 1
X2 ≥ 0 No 1

20
 Geometry of LP – Summary
• Decision variables
– Each decision variable forms a dimension
– n decision variables define an n-dimensional space
– A solution is a point in the space
• Constraints
– Each equality constraint defines a hyper-plane (line in a 2-D space)
– Each inequality constraint defines a half-space
– All constraints collectively define the feasible region
• Objective
– The objective function defines isoquants and a direction in the space
• To find the optimal solution, push along the direction defined by the
objective until we reach the boundary of the feasible region
• At the optimal solution, some constraints are binding (or active) while
others are not

21
 Recap of Lecture 5
• Visualize the three components of a Linear Program
– Decision variables  Dimensions of space
– Constraints  Hyperplanes or half-spaces
– Objective  Isoquant and the direction
• Feasible region and extreme (or corner) points
• Binding and non-binding constraints (at optimal solution)
• Slack of a constraint

22
 Geometry of LP – Benefits
• An infinite number of feasible solutions

• A finite number of extreme points

– Number of decision variables n, and number of
constraints (excluding sign constraints) m
– How many? ≤ 𝑛+𝑚 𝑛
• An extreme point is the optimal solution
for a finite LP
• The Simplex algorithm (there are other methods)
developed by Dantzig

23
Geometry of LP – 3-Dimension

24
Solving Linear Programming
in Excel (Solver)

25
• Add Solver in Excel: File  Option  Add-Ins  Manage  Go

26
 The Lego Game Revisited
• Products: Chair ($40) and Table ($50)
Chair Table Inventory
(C) (T)
Big
Brick 1 2 10
Small
Brick 4 3 30

MAX 40C + 50T

s.t. 1C + 2T ≤ 10 (Big Bricks)
4C + 3T ≤ 30 (Small Bricks)
C, T ≥ 0
27
First step : input data MAX 40C + 50T
s.t. 1C + 2T ≤ 10 (Big Bricks)
from the problem to the
4C + 3T ≤ 30 (Small Bricks)
spreadsheet. C, T ≥ 0

28
• Initial input about decision variables: B5=0, C5=0

• LHS of the constraint:

– D8=B5*B8+C5*C8;
– D9=B5*B9+C5*C9;

• Objective function:
– D12=B5*B12+C5*C12;

• All the other inputs are numbers

29
 SUMPRODUCT
• SUMPRODUCT (array1, array2, array3,…): return the sum of the products
of corresponding range or arrays.
• All arrays must have same dimensions.
• Example: E2=SUMPRODUCT(A2:A5,B2:B5,C2:C5)
=A2*B2*C2 + A3*B3*C3 + A4*B4*C4 + A5*B5*C5

• In The Lego Game, we can rewrite D8, D9, D12 as

– D8= SUMPRODUCT(B5:C5,B8:C8), D9= SUMPRODUCT(B5:C5,B9:C9),
D12= SUMPRODUCT(B5:C5,B12:C12) 30
 Set
objective

Choose
decision
variable

Add
constrain

Check if all
variables
are non-
negative

31
32
 Example: Blending Model
Investment Problem of Pension Planners
The portfolio manager of Pension Planners has been asked to invest
$1 million of a large pension fund in six mutual funds with varying
potential returns and risks, as summarized in the Table below.

Fund
1 2 3 4 5 6
Price ($/share) 45 76 110 17 23 22
Exp Return (%) 30 20 15 12 10 7
Risk Category High High High Med Med Low

33
LP Model
Max 0.30F1 + 0.20F2 + 0.15F3 + 0.12F4 + 0.10F5 + 0.07F6
st
F1 + F2 + F3 ≥ 0.50
F1 + F2 + F3 ≤ 0.75
F4 + F5 ≥ 0.20
F4 + F5 ≤ 0.30
F6 ≥ 0.05
2F1 − F2 = 0 Optional :
3F1 − F3 = 0 remove one of
3F2 − 2F3 = 0
these constraints
2F4 − F5 = 0
F1 + F2 + F3 + F4 + F5 + F6 = 1

Fi ≥ 0 i = 1, 2, . . . , 6
34
35
 Answer Report
Cell Name Original Value Final Value
$H$17 Objective LHS 0 0.168583333

Cell Name Original Value Final Value Integer

$B$5 F1 0 0.125Contin
$C$5 F2 0 0.25Contin
$D$5 F3 0 0.375Contin
$E$5 F4 0 0.066666667Contin
$F$5 F5 0 0.133333333Contin
$G$5 F6 0 0.05Contin

Cell Name Cell Value Formula Status Slack

$H$10 LHS 0.2$H$10<=$J$10 Not Binding 0.1
$H$11 LHS 0.05$H$11>=$J$11 Binding 0
$H$12 LHS 0$H$12=$J$12 Binding 0
$H$13 LHS 0$H$13=$J$13 Binding 0
$H$14 LHS 0$H$14=$J$14 Binding 0
$H$15 LHS 1$H$15=$J$15 Binding 0
$H$7 LHS 0.75$H$7>=$J$7 Not Binding 0.25
$H$8 LHS 0.75$H$8<=$J$8 Binding 0
$H$9 LHS 0.2$H$9>=$J$9 Binding 0

36
 Example: Nurse Scheduling
• University hospital requires different number of nurses
at different times
• Each nurse works for 8 consecutive hours per day
Period Time Min Number of Nurses Needed
1 12am – 4am 10
2 4am – 8am 25
3 8am – 12pm 35
4 12pm – 4pm 30
5 4pm – 8pm 40
6 8pm – 12am 20

• Find the minimum number of nurses used per day

37
 Example: Nurse Scheduling

8pm – 4am
12am – 8am
4am – 12pm
8am – 4pm
12pm – 8pm
4pm – 12am
8pm – 4am
12am – 4am 4am – 8am 8am – 12pm 12pm – 4pm 4pm – 8pm 8pm – 12am

10 25 35 30 40 20
Monday Tuesday Wednesday

38
 Example: Nurse Scheduling
• Decision variables?
– 𝑥𝑖 : number of nurses who start work in period 𝑖 , 𝑖 =
1,2, … , 6
• Objective?
– Minimize total number of nurses
• Constraints?
– Total staff requirement for each period

39
 Example: Nurse Scheduling
𝑥6 + 𝑥1 𝑥1 + 𝑥2 𝑥2 + 𝑥3 𝑥3 + 𝑥4 𝑥4 + 𝑥5 𝑥5 + 𝑥6

8pm – 4am
𝑥6
12am – 8am
𝑥1
4am – 12pm
𝑥2
8am – 4pm
𝑥3
12pm – 8pm
𝑥4
4pm – 12am
𝑥5
8pm – 4am
12am – 4am 4am – 8am 8am – 12pm 12pm – 4pm
𝑥6
4pm – 8pm 8pm – 12am

10 25 35 30 40 20
Monday Tuesday Wednesday

40
Example: Nurse Scheduling

min x1  x2  x3  x4  x5  x6
subject to
x1  x6  10
x1  x2  25
x2  x3  35
x3  x4  30
x4  x5  40
x5  x6  20
xi  0, for all i  1,2,...,6
41
42
Nurse Scheduling: Optimal Solution
10 25 35 40 40 20

8pm – 4am
0
12am – 8am
10
4am – 12pm
15 Total no. = 85
8am – 4pm
20
12pm – 8pm
20
4pm – 12am
20
8pm – 4am
12am – 4am
0
4am – 8am 8am – 12pm 12pm – 4pm 4pm – 8pm 8pm – 12am

10 25 35 30 40 20
Monday Tuesday Wednesday

43
44
Nurse Scheduling: Alternative Solution

10 35 35 30 40 20

8pm – 4am
0
12am – 8am
10
4am – 12pm
25 Total no. = 85
8am – 4pm
10
12pm – 8pm
20
4pm – 12am
20
8pm – 4am
12am – 4am
0
4am – 8am 8am – 12pm 12pm – 4pm 4pm – 8pm 8pm – 12am

10 25 35 30 40 20
Monday Tuesday Wednesday

45