LP Modeling I (SS 2022)

Introduction to LP Modeling

Johanna Mlekusch
[email protected]
What is Linear Programming?

Linear Programming

Problem solving approach for minimizing (or maximizing) a linear objective

function subject to linear equality and inequality constraints

 Linear: All mathematical functions in the model are required to be linear

functions

 Programming: Not in the sense of computer programming; here rather a

synonym for planning → planning of activities to obtain an optimal result
What is Linear Programming Used For?

Assignment problem
Packing problems

Lot-Sizing

Max flow Transportation problem

Scheduling

Shortest path Traveling salesman problem

and many, many more…

General LPs

The following is a LP problem:

Objective

Inequality
constraints

Equality
constraints

Non-negativity
constraint
Modeling

• Usually the LP is not given and has to be extracted from

information

• Translation from a verbal description of the problem into a

mathematical one

 LP Modeling

• Often there are different ways to model a problem

• The mathematical formulation of the problem decides which

solution methods can be applied
Modeling

Practical modeling Mathematical

problem model

compatible? methodologies

Practical implementation Mathematical

solution solution
The Burglar Problem

• Pete, the burglar, sees 10 items of different weight and different

value.
• He can not carry all of the items, only up to a certain total ????
weight.
• Which items should he take?
Model Formulation

• Understand the problem thoroughly

 What does the decision maker want to achieve with the

present model?
 What instruments does the decision maker have?

 What are the objectives?

 What are the constraints?

 What data is available?

Decision Variables and Parameters

Decision Variables

• Quantities a decision maker would like to determine

• Examples: production quantity, current size of the inventory, starting time of activity, …
• Decision variables might be measured in different units → compatibility?

• Let us assume is a decision variable. can for example be either

• a real number ()
• an integer () : all-integer problems (IP LPs), mixed-integer problems (MIP LPs)
• a binary variable with

• Size of decision variable can be restricted

Parameter
• Exogenously given (e.g. value of an item, capacity of an inventory, …), known and constant
General LPs

Decision variables (unknown)

Parameters (known)

s.t.

0
0

Notation is important,
it can make working with LPs more efficient!
General LPs

s.t.
s.t.

0
0 0
0

Notation is important,
it can make working with LPs more efficient!
The Burglar Problem

• A burglar sees 10 items of different weight and different value.

• He can not carry all of the items, bag holds only up to a certain
total weight.
????
• Which items should he take?

 Step 1: Determine the decision variables and parameters

… set of items

… (binary) decision variable (take item?),

… value of item i,
… weight of item i,
… maximum weight that can be transported in bag
The Burglar Problem

• Step 2: Determine the objective

max ∑ 𝑣 𝑖 ∙ 𝑥 𝑖 ????
❑ 𝑖∈𝐼 … maximize the profit

• Step 3: Determine the constraints

∑ 𝑤 𝑖 𝑥 𝑖❑ ≤ 𝑊 … sum of the weight of the stolen items must be

𝑖∈𝐼
lower than maximum capacity W
for all
𝑥𝑖 ∈ { 0 , 1 } , ∀ 𝑖 ∈ 𝐼 … item i is either stolen () or it is not ()
( must be binary)

• Further steps: Implement the model, solve the problem,…

Why are we discussing this?

• Please be careful when you write down a model:

• A good notation saves time, it helps you to read ????

(and implement) the problem more easily
• Be consistent and precise!

• This model is actually relevant in supply chain

management / logistics.
Model Formulation & Interpretation

• The burglar problem is a so-called knapsack (or rucksack)

problem
!!!!!
• Different application: A firm needs to decide which items it
wants to transport in a certain truck.

 Mathematical models can often be interpreted in various

manners / are often similar from a formal point of view

 Understanding one problem can often help to solve others

Model Formulation

 One can also write down the same problem in different manners, e.g.

 Maximization problem as minimization problem (and vice versa)

 Inequality constraint as equality constraint (and vice versa)

 Approximate a non-linear function by linear functions

…

 A well-considered model formulation can make solving a problem easier

and/or more efficient!
General LPs

s.t.
s.t.

0
0 0
0

Notation is important,
it can make working with LPs more efficient!
General LPs

( )
𝑥1
′
𝑥 =( 𝑥 1 , … , 𝑥 𝑛 ) ⇔ 𝑥 = ⋮
𝑥𝑛

(e.g. is the set of inequalities with ,

is the set of equality constraints here)

LP in general form…

∀
∀
∀
∀
∀
General LPs

Consider the LP problem from before: LP in general form…

∀
∀
∀
∀
∀

s.t. 𝑥=( 𝑥 1 , 𝑥 2 , 𝑥 3 , 𝑥 4 ) ′ 𝑐= ( 2, −1 , 4 , 0 ) ′
→ ={3}, ={1}, ={2},
0 ={1}, ={3},
0
Constraints

How to express an equality constraint in terms of inequality constraints?

Create two inequalities, which have to hold at the same time:

∙ ( −1)

To only have inequalities, where the l.h.s. is larger than the right hand side,
multiply second constraint by -1:
Matrix notation

Matrix notation of an LP problem, e.g.

Matrix notation

Example: transform the previous LP to a form such that

you only have inequalities with ≥
Matrix notation

Example: transform the previous LP to a form such that

you only have inequalities with ≥
Matrix notation

Example: transform the previous LP to a form such that

you only have inequalities with ≥

min 2 𝑥1 − 𝑥 2+ 4 𝑥 3

− 𝑥 1 − 𝑥 2 − 𝑥4 ≥− 2
3 𝑥 2 − 𝑥 3 ≥5
−3 𝑥 2 + 𝑥3 ≥ −5
𝑥 3+ 𝑥 4 ≥ 3
𝑥1 ≥ 0
− 𝑥3 ≥ 0
Matrix notation

Example: transform the previous LP to a form such that

you only have inequalities with ≥

min 2 𝑥1 − 𝑥 2+ 4 𝑥 3

− 𝑥 1 − 𝑥 2 − 𝑥4 ≥− 2
3 𝑥 2 − 𝑥 3 ≥5
−3 𝑥 2 + 𝑥3 ≥ −5
𝑥 3+ 𝑥 4 ≥ 3
𝑥1 ≥ 0
− 𝑥3 ≥ 0
Matrix notation

Example: transform the previous LP to a form such that

you only have inequalities with ≥

min 2 𝑥1 − 𝑥 2+ 4 𝑥 3 ‘

− 𝑥 1 − 𝑥 2 − 𝑥4 ≥− 2 𝑐= ( 2, −1 , 4 , 0 ) ′
3 𝑥 2 − 𝑥 3 ≥5
𝑏=( −2 , 5 , −5 , 3 , 0 ,0 ) ′
−3 𝑥 2 + 𝑥3 ≥ −5

( )
𝑥 3+ 𝑥 4 ≥ 3 −1 −1 0 −1
𝑥1 ≥ 0 0 3 −1 0
− 𝑥3 ≥ 0 𝐴= 0 −3 1 0
0 0 1 1
1 0 0 0
0 0 −1 0
Standard form of an LP problem (in matrix notation)

Standard form…
• only equality constraints
• only non-negative variables
Reduction to Standard Form

General form: Standard form:

Standard form

Eliminating free variables…

free variables are neither in

N1 nor in N2
Standard form

Eliminating free variables…

Essentially, you redefine the decision variable

Example: Imagine is the balance of your bank account (which can be

negative).

 replace by the amount of money you own minus the debt .

Balance
100 100 0
-200 0 200
Standard form

Eliminating inequality constraints…

Introduce slack/surplus variables which tell you how much smaller/larger is the left hand side of the
inequality compared to the right hand side
1. Slack variables

2. Surplus variables

Slack and surplus variables are decision variables!

Standard form

Example: transform the LP

below to standard form
Standard form

Example: transform the LP

below to standard form
−

𝑥 2= 𝑥+2 ¿ − 𝑥 2 ¿

𝑥 =¿
𝑐= ( 2, 4 , − 4 , 0 )

𝐴=
(13 1
2
−1
−2
−1
0 )
𝑏=( 3 , 14 )′
Standard form

Get rid of non-positive decision variables:

𝑥3≤ 0 −
𝑥 3 =− 𝑥3
𝑥−
3 ≥0
Minimization vs. maximization

Any minimization problem can be transformed into a

maximization problem and vice versa.

minimize

maximize -
Absolute values

Absolute values

Can be written as:

Or:

¿
Absolute values

Idea: Find the smallest number z that is larger than x as well as larger than –x
(i.e. z satisfies z ≥ x and z ≥ -x ) . |x| is equal to z.

min z s.t. z ≥ x and z ≥ -x

|x| z

x x
Absolute values

Can be written as:

Absolute values

Core idea:
¿
Redefine decision variables (see bank account example)
Balance
100 100 100 0
-200 200 0 200

 =-
=
Absolute values

Example. Reformulate:

−
−𝑥
𝑥1 = 𝑥+¿ 1 ¿
1

min 2 𝑧 + 𝑥 2
−
+ ¿ +2 𝑥 + 𝑥 2 ¿
min 2 𝑥1 1

𝑥1 + 𝑥 2 ≥ 4
𝑥1
−
+ ¿ − 𝑥 1 + 𝑥2 ≥ 4 ¿ 𝑧 ≥ 𝑥1
𝑧 ≥ − 𝑥1
−
¿, 𝑥
𝑥+ 1 ≥0 ¿
1
Example: Production Problem

A firm produces n different goods using m different raw materials

Let bi , i=1…m, be the available amount of the ith raw material

The jth good, j=1...n, requires aij units of the ith raw material

The jth good, j=1...n, results in a revenue of cj per unit produced.

How much of each good to produce to maximize total revenue?

Example: Production Problem
Example: Production Problem

Product A Product B Available

Raw material I 1 3 24
Raw material II 1 1 10
Raw material III 2 1 18
Revenue 100 200
Example: Production Problem – Graphical Solution
 How to draw a line / inequalities

The area for which the inequality holds is bounded by

Approach 1:

Find the points where the line intersects with the axes and connect the points. (I.e. set , then you
get ; Point 1: (0, ), set y, then you get Point 2: ( ), connect points 1 and 2.

Now you can draw the line.

()
On which side of the line is the feasible area?? 𝑥
How steep is the line?? (0, )
How to draw a line / inequalities

Approach 2:
(better approach for the graphical solution of LPs)

Rewrite the inequality as or

𝑦
→ Only values of which are greater than are admissible.

The line bounds the feasible area.

𝑥
Slope of the line:

Intersection of the line with y-axis:

Feasibility

Solution
Any specification of values for all decision variables, regardless of whether it is a
desirable or even allowable choice

Feasibility

• Feasible (admissible) solution: all constraints are fulfilled

• Infeasible solution: at least one constraint is violated

Feasible region:
Feasible
region
• Collection of all feasible solutions

 Obviously: An optimal solution must be feasible, but a feasible solution does not
have to be optimal
 Example: Production Problem – Graphical Solution

= gives a line with profit Z

Consider value of the objective at the

corners of the feasible region
Optimal solution:
,
𝑍 =1600
𝑍 =1700

Feasible
region 𝑍 =1200
𝑍 =0
𝑍 =900
Slope =
()
Example: Production Problem – Graphical Solution

𝑍 =1700
binding at (3,7)

Feasible
region Optimal solution:
,

non-binding at (3,7)

A constraint can be
• binding / active at the optimal solution point
• non-binding / in-active at the optimal solution point
• redundant e.g.
Example: Production Problem – Graphical Solution

Consider value of the objective at the

corners of the feasible region

𝑍 =100 𝑥1 +200 𝑥 2

Slope =

Slope = 𝑍 =1700
𝑍 =1600
Slope =

Feasible 𝑍 =1200
region
𝑍 =0
𝑍 =900
Feasible
region At the optimal solution the line =
𝑍 =0 does not enter the feasible region, it just touches
it

Slope =
Example: Production Problem – Graphical Solution

Consider value of the objective at the

corners of the feasible region

Slope = Corner point can only be optimal if the slope is

between the two slopes of the constraint
Slope = restricting the feasible region of the point

Slope =
E.g. , is optimal, one can observe at this point --
1

If we consider another objective with and , then

Feasible =-1.5, comparing the slopes of the constraints
region we find --2 implying that the intersection point
𝑍 =0 between the yellowish and the black constraint ,
2 is optimal

Slope =
Example: Production Problem – Graphical Solution

Consider a different objective 𝑍 =100 𝑥1 +100 𝑥 2

𝑍 ∗ =1000

Feasible
region

 Any point on the red line has the same objective value and is optimal
LP solutions

The following can happen when solving a LP:

(a) Unique optimal solution

(b) Multiple optimal solutions (bounded or unbounded set)
(c) Objective value is ∞ and no feasible solution is optimal
(d) The feasible set is empty (the LP is then infeasible)
Infeasibility

6
6

2
𝑥1 ≥ 4
4 𝑥1 +2 𝑥 2 ≤ 8
2 4 6 8

No point fulfills all of the constraints

 No feasible solution
Objective Value is Infinite

s.t. 6

0
4

2
𝑥1 ≤ 4

2 4 6 8

One can always increase the profit by increasing

A Bin Packing Example

We want to find the minimum number of vehicles required to transport all the
parcels demanded by customers.
Set of V parcels
weight of parcel i V
Vehicle capacity Q
Upper bound on number of vehicles M
A Bin Packing Example

Set of V parcels
weight of parcel i W
Vehicle capacity Q
Upper bound on number of vehicles M

Decision variable {0,1} j {1..M} Is vehicle j used or not

Decision variable {0,1} i V, j {1..M} is parcel i in vehicle j

j {1..M}

V, j {1..M}
A Scheduling Example

A hospital wants to make a weekly night shift schedule for its nurses

The demand for nurses for the night shift on day j is an integer dj, j=1…7:

d1 5
Every nurse works 4 days in a row on the night shift
d2 4
(and then off 3 days)
d3 5
What is the minimal number of nurses
d4 3
the hospital needs to hire?
d5 5
d6 7
d7 8
Nurse Scheduling Example

Decision variable: …. Number of nurses starting their shift on day i, i=1,…,7

Outlook

• Homework 1
• Due March 10th 945 (Hard deadline)
• Solution will be uploaded on March 10th at 1300
????
• Install Xpress