Sc0x w3l1 Clean Final

Integer
and Mixed Integer

Linear Programs
MIT Center for

Transportation & Logistics
ctl.mit.edu
Numbers, Numbers, Everywhere!

N = Natural, Whole, or Counting Numbers = 0, 1, 2, 3, . . .
Z = Integers = . . . -3, -2, -1, 0, 1, 2, 3, . . .
Q = Rational Numbers = any fraction of Integers, 1/2 , -5/9, 0/22, . . . etc.
R = Real Numbers = all Rational and Irrational numbers, i.e, , 2, e, . . . etc.
B = Binary Integers = {0,1}

B
Why the heck do we care?

MIT Center for
By Mortalmoth (Own work) [Public domain], via Wikimedia Commons
Integer Variables
Why use them?
n
When its physically impossible to have fractional solutions

w For example; number of people to hire, number of ships to make
w However, if dealing with large numbers, continuous is fine
Allows for modeling logical conditions (Binary)

w If Then:
If we have product leaving plant A then we must open it

w Either Or:
We can either produce 1000 units or none at all.
w Select From:
We must select 4 DCs to open from the 10 possible
We must select 5 products to make from the 15 available
Why do we have to treat them differently?

MIT Center for
Banner Chemicals II: IP Example
MIT Center for

Motivating Problem Banner Chemicals II

Situation
n
Banner Chemicals manufactures specialty chemicals. One of their products

comes in two grades, high and supreme. The capacity at the plant is 110
barrels per week.
The high and supreme grade products use the same basic raw materials but
require different ratios of additives. The high grade requires 3 gallons of
additive A and 1 gallon of additive B per barrel while the supreme grade
requires 2 gallons of additive A and 3 gallons of additive B per barrel.
The supply of both of these additives is quite limited. Each week, this product
line is allocated only 300 gallons of additive A per week and 280 gallons of
additive B.
A barrel of the high grade has a profit margin of $80 per barrel while the
supreme grade has a profit margin of $200 per barrel.
Question
n
How many barrels of High and Supreme grade should Banner Chemicals
produce each week assuming you can only produce in 10 barrel lots?
MIT Center for

Banner Chemicals
Optimal LP Solution
XH= 25 barrels
XS = 85 barrels
Max Profit = 19,000
Max z(XH, XS) = 80XH + 200XS

s.t.
Plant
XH + XS 110
Add. A
3XH + 2XS 300
Add. B
XH + 3XS 280
XS
XH 0
XS 0
What is the optimal solution
with 10 barrel lots?
XH
Notes:
Feasible region becomes a collection of points, no longer a convex hull

MIT Center for
We cannot rely on corner solutions anymore solution space is much bigger!
How to find solution to IP?

Lets try rounding the solution to
the closest acceptable integer values?

n

s.t.
Plant
XH + XS 110
Add. A
3XH + 2XS 300
Add. B
XH + 3XS 280
LP Solution:
w XH=25 barrels XS= 85 barrels
3.
XS 0
Rounding to closest 10 barrel solution for (XH, XS):

1.
2.
XH 0
zLOT(30, 90) = $20,400 but it is infeasible (Plant constraint)

zLOT(30, 80) = $18,400 feasible
zLOT(20, 90) = $19,600 but it is infeasible (Additive B constraint)
So, using this approach z*LOT= $18,400 with XH=30, XS=80

But, is it the best?
Lets solve all of the points to make sure!

This approach is called Mass Enumeration.
MIT Center for
Mass Enumeration
of Banner Chemical
Optimal IP Solution
XH= 10 barrels
XS = 90 barrels
Max Profit = 18,800
Optimal LP Solution
XH= 25 barrels
XS = 85 barrels
Max Profit = 19,000

s.t.
Closest rounded
LP Solution
XH= 30 barrels
XS = 80 barrels
Max Profit = 18,400
Plant
XH + XS 110
Add. A
3XH + 2XS 300
Add. B
XH + 3XS 280
XH 0
XS 0
Each cell shows z = 80XH + 200XS

x indicates infeasible solution
Notes:
Rounding the optimal LP solution will not always lead to an optimal IP solution
Mass enumeration is very time consuming not always possible for real problems!
IP solution can never be better than the LP solution!
MIT
Center
IPs forare much, much, much harder to solve than LPs!
Formulation Changes . . . not much!

s.t.
Plant
XH + XS 110
Add. A
3XH + 2XS 300
Add. B
XH + 3XS 280
XH 0
XS 0
In order to solve in integer values of

lots of ten, we need to:
n
Convert Decision Variables

w XHL = XH /10 XSL = XS /10
Scale the coefficients and constraint RHS

w e.g. 110 barrels becomes 11 lots of ten
Indicate that the new DVs are Integers

Max z(XHL, XSL) = 800XHL + 2000XSL
s.t.
Plant
XHL + XSL 11
Add. A
3XHL + 2XSL 30
Add. B
XHL + 3XSL 28
XHL, XSL 0 Integers
MIT Center for

GoNuts Juice Company: Model 1
MIT Center for

10
GoNuts Juice Company
GoNuts manufactures different juices made entirely of various exotic nuts. Their
primary market is China and they operate three plants located in Ethiopia,
Tanzania, and Nigeria. You have been asked to help them determine where to
manufacture the two newest juices they offer, Gingko Nut and Kola Nut. Each
plant has a different variable cost structure and capacity for manufacturing the
different juices. Also, each juice has an expected demand.
Cost/Unit
Ethiopia
Tanzania
Nigeria
Ginko
21.00
22.50
23.00
Kola
22.50
24.50
25.50
Capacity Units/Month
Ethiopia
425
Tanzania
400
Nigeria
750
Demand Units/Month
Ginko
550
Kola
450
How much of each juice should be made at each plant in order to minimize total
cost while meeting demand and adhering to plant capacity?
MIT Center for
Image CC0 Public Domain from https://pixabay.com
11
Formulating GoNuts
Step 1. Determine Decision Variables

xG,E = Number of Ginko Juice units made in Ethiopia plant
xK,E = Number of Kola Juice units made in Ethiopia plant
xG,T = Number of Ginko Juice units made in Tanzania plant
xK,T = Number of Kola Juice units made in Tanzania plant
xG,N = Number of Ginko Juice units made in Nigeria plant
xK,N = Number of Kola Juice units made in Nigeria plant
xij = Number of units of product i made in plant j
xij 0
for all i,j
Step 2. Formulate Objective Function

Minimize z = Cost = 21xG,E + 22.5xK,E + 22.5xG,T + 24.5xK,T + 23xG,N + 25.5xK,N
Min
where:
MIT Center for
z =
cij xij

cij = Cost per unit of product i made at plant j
12
Formulating GoNuts
Step 3. Formulate Constraints

Plant Capacity
xG,E + xK,E 425
xG,T + xK,T 400
xG,N + xK,N 750
x
i
ij
Cj
where:
Cj = Capacity in units at plant j
Product Demand
xG,E + xG,T + xG,N 550
xK,E + xK,T + xK,N 450
xij Di
where:
Di = Demand for product i in units
MIT Center for
13
Formulating GoNuts
Minimize z = Cost = 21xG,E + 22.5xK,E + 22.5xG,T + 24.5xK,T + 23xG,N + 25.5xK,N

subject
xG,E + xK,E 425
to
Capacity
xG,T + xK,T 400
j=1
xG,N + xK,N 750
cij
i=1
i=2
j=2
j=1 21.00 22.50
j=3
j=2 22.50 24.50
xG,E, xK,E, xG,T, xK,T, xG,N, xK,N 0
j=3 23.00 25.50
Min
z =
Cj
425
400
750
Demand Di
i=1 550
i=2 450
cij xij
s.t.
x
x
i
ij
Cj
ij
Di
xij 0
ij
where:
MIT Center for
Di = Demand
for product i in units
Optimal Solution
Ethiopia
Tanzania
Nigeria
Ginko
0
375
175
Kola
425
25
0
Total min cost = 22,637.50

14
MIT Center for

15
plant has a different fixed and variable cost structure and capacity for
manufacturing the different juices. The fixed cost only applies if the plant
produces any juice. Also, each juice has an expected demand.
Cost/Unit
Ethiopia
Tanzania
Nigeria
Ginko
21.00
22.50
23.00
Kola
22.50
24.50
25.50
Capacity
Units/Month
Ethiopia
425
Tanzania
400
Nigeria
750
Fixed (/Month)
1,500
2,000
3,000
Demand Units/Month
Ginko
550
Kola
450
MIT Center for
16
Formulating GoNuts 2

yj = 1 if plant j is opened; = 0 otherwise
Step 2. Formulate Objective Function

Min z = 21xG,E + 22.5xK,E + 22.5xG,T + 24.5xK,T + 23xG,N + 25.5xK,N + 1500yE + 2000yT + 3000yN
Min
where:
z =
cij xij + f j y j
j

yj = 1 if plant j is opened; = 0 o.w.
fj = Fixed cost per month if plant j is used
MIT Center for

17

Demand Capacity
xG,E + xK,E 425

xG,T + xK,T 400
xG,N + xK,N 750
x
x
i
ij
Cj
ij
Di
where:
Is this enough? Try solving it!

n
You need to ensure that if a plant produces product, then

it is actually opened!
If Then conditions require both a
w Binary Variable
w Linking Constraint
MIT Center for

18
If Then Conditions
MIT Center for

19
If-Then Condition
Looking at the Nigeria Plant . . .
yN
Feasible Values
1
How do yN, xGN and xKN interact?

0
IF . . .
THEN
xGN+xKN
yN = 0 yN = 1
CN
= 0
YES
YES
xGN + xKN 750 yN
>0 and CN
NO
YES
IF xGN+xKN =
x
i
ij
My j
where:
M = a big number (such as Cj in this case)
MIT Center for
xGN + xKN
THEN yN =
0 or 1
0 750 yN
99
99 750 yN
1 750 yN
If the X values >0, then Y MUST be

equal to 1! Otherwise, it would
violate the constraint.
20
Linking Demand Capacity

xG,E + xK,E 425
xG,T + xK,T 400
xG,N + xK,N 750
xG,E + xK,E - 425yE 0
xG,T + xK,T - 400yT 0
xG,N + xK,N - 750yN 0
MIT Center for

x C
x D
x My
i
ij
ij
ij
0 j
where:
21
GoNuts Juice Company Model 2:

With If Then Conditions
MIT Center for

22
Formulation of GoNuts Model 2
subject to
xG,E + xK,E 425

xG,T + xK,T 400
xG,N + xK,N 750
xG,E + xK,E - 425yE 0
xG,T + xK,T - 400yT 0
xG,N + xK,N - 750yN 0
xG,E, xK,E, xG,T, xK,T, xG,N, xK,N 0
yE, yT, yN = {0, 1}
Demand Di
i=1 550
i=2 450
Capacity
j=1
j=2
j=3
Cj
fj
425 1500
400 2000
MIT Center for
750 &3000
Transportation
Logistics
cij
j=1
j=2
j=3
i=1
21.00
22.50
23.00
Min
z = cij xij + f j y j
i
s.t.
x
x
x
i
ij
Cj
ij
Di
ij
My j 0
xij 0
y = {0,1}
ij
j
where:
i=2
22.50
24.50
23
25.50
Solution: GoNuts Models 1 & 2

Model 1 only variable costs
z*= 22,637.50
Ethiopia
Tanzania
Nigeria
Ginko
0
375
175
MIT Center for

Model 2 with fixed plant costs

z*= 27,350.00
Kola
425
25
0
Ethiopia
Tanzania
Nigeria
Ginko
0
0
550
Kola
425
0
25
24

Adding Either Or Conditions
MIT Center for

25
plant has a different fixed and variable cost structure and both minimum and
maximum capacities for manufacturing the different juices if the plant opens.
The fixed cost only applies if the plant produces any juice. Also, each juice has
an expected demand.
Cost/Unit
Ethiopia
Tanzania
Nigeria
Ginko
21.00
22.50
23.00
Kola
22.50
24.50
25.50
Capacity
Max
Min
(units/Month) Capacity Capacity
Ethiopia
425
100
Tanzania
400
250
Nigeria
750
600
Demand Units/Month
Ginko
550
Kola
450
Fixed
(/Month)
1,500
2,000
3,000
If the Nigeria plant opens, it

must produce at least 600 units
MIT Center for
26
No Change!

Step 2. Formulate Objective Function No Change!

Min
where:
z =
cij xij + f j y j
j

fj = Fixed cost per month if plant j is used
MIT Center for

27

xG,E + xK,E 425
xG,T + xK,T 400
xG,N + xK,N 750
xG,E + xK,E - 425yE 0
xG,T + xK,T - 400yT 0
xG,N + xK,N - 750yN 0
x C
x D
x My
i
ij
ij
ij
0 j
where:
Cj = Maximum capacity in units at plant j
Lj = Minimum level of production at plant j
We need to add a constraint that ensures that if we DO use plant j, that

the volume is between the minimum allowable level, Lj, and the maximum
capacity, Cj. This is sometimes called an Either-Or condition.
MIT Center for
28
Either Or Condition
Looking at the Nigeria Plant . . .
yN
Feasible
Values
How do yN, xGN and xKN interact?

0
IF . . .
THEN
xGN+xKN
yN = 0 yN = 1
= 0
YES
NO
>0 and < LN
NO
NO
LN and CN
NO
YES
ij
My j
ij
Lj y j
x
x
i
i
where:
MIT Center for level of production at plant j
Lj = Minimum
LN
600
CN
750
xGN + xKN
IF
xGN+xKN
THEN yN
750yN
THEN yN
600yN
0 or 1
200
600
0 or 1
If the X values >0, then they must be

L, the lower limit, and C, the
maximum capacity!
29
subject to
xG,E + xK,E 425

xG,T + xK,T 400
xG,N + xK,N 750
xG,E + xK,E - 425yE 0
xG,T + xK,T - 400yT 0
xG,N + xK,N - 750yN 0
xG,E + xK,E - 100yE 0

xG,T + xK,T - 250yT 0
xG,N + xK,N - 600yN 0
xG,E, xK,E, xG,T, xK,T ,
Products Di
xG,N, xK,N 0
i=1
550
yE, yT, yN = {0, 1}
i=2
450
cij
j=1
j=2
j=3
Plants
i=1
i=2
j=1
21.00 22.50
j=2
22.50 24.50
MIT Center for
j=3
& Logistics
23.00 Transportation
25.50
Min
z = cij xij + f j y j
i
s.t.
x
x
x
x
i
ij
Cj
ij
Di
ij
My j 0
ij
Lj y j 0
xij 0
ij
y j = {0,1}
where:
Lj = Minimum level of production at plant j
Cj
Lj
fj
425 100 1500 M = a big number (such as C in this case)
j
400 250 2000
750 600 3000
30
Solution: GoNuts Models 1, 2, & 3

z*= 22,637.50
Ethiopia
Tanzania
Nigeria
Ginko
0
375
175

z*= 27,350.00
Kola
425
25
0
Ethiopia
Tanzania
Nigeria
Ginko
0
0
550
Kola
425
0
25

and minimum production levels
z*= 27,425.00
Ethiopia
Tanzania
Nigeria
Ginko
0
0
550
MIT Center for

Kola
400
0
50
31

Adding Select From Conditions
MIT Center for

32
plant has a different variable cost structure and a maximum capacity. GoNuts
can only operate 2 plants at a maximum. Also, each juice has an expected
demand.
Cost/Unit
Ethiopia
Tanzania
Nigeria
Ginko
21.00
22.50
23.00
Kola
22.50
24.50
25.50
Capacity
Max
(units/Month) Capacity
Ethiopia
425
Tanzania
400
Nigeria
750
Demand Units/Month
Ginko
550
Kola
450
MIT Center for
33
No Change!

Step 2. Formulate Objective Function Slight Change!

Min z = 21xG,E + 22.5xK,E + 22.5xG,T + 24.5xK,T + 23xG,N + 25.5xK,N
Min
where:
z =
cij xij

MIT Center for

34
Max
Plants

xG,E + xK,E 425
xG,T + xK,T 400
xG,N + xK,N 750
xG,E + xK,E - 425yE 0
xG,T + xK,T - 400yT 0
xG,N + xK,N - 750yN 0
yE + yT + yN 2
x C
x D
x My
y N
i
ij
ij
ij
0 j
where:
N = Number of plants allowed to be opened
We need to add a constraint that ensures that only N plants are used! We will use the
Binary Variables, yj, the Linking Constraints, and a new constraint that says the sum of the
MIT Center for
Binary Variables
must not exceed N. This is sometimes called an Select-From condition.
35
Min z = 21xG,E + 22.5xK,E + 22.5xG,T + 24.5xK,T + 23xG,N + 25.5xK,N

subject to
xG,E + xK,E 425

xG,T + xK,T 400
xG,N + xK,N 750
xG,E + xK,E - 425yE 0
xG,T + xK,T - 400yT 0
xG,N + xK,N - 750yN 0
yE + yT + yN 2
xG,E, xK,E, xG,T, xK,T , xG,N, xK,N 0
yE, yT, yN = {0, 1}
Products Di
i=1
550
i=2
450
cij
j=1
j=2
j=3
N = 2
Plants
i=1
i=2
j=1
21.00 22.50
j=2
22.50 24.50
MIT Center for
j=3
& Logistics
23.00 Transportation
25.50
Cj
425
400
750
Min
z = cij xij
i
s.t.
x C
x D
x My
y N
i
ij
ij
ij
xij 0
ij
y = {0,1}
j
where:
N = Number of plants allowed to be opened
36
Solution: GoNuts All Models

z*= 22,637.50
Ethiopia
Tanzania
Nigeria
Ginko
0
375
175
z*= 27,350.00
Kola
425
25
0

and minimum production levels
z*= 27,425.00
Ethiopia
Tanzania
Nigeria
Ginko
0
0
550
MIT Center for

Ethiopia
Tanzania
Nigeria
Ginko
0
0
550
Kola
425
0
25
Model 4 only variable costs but

with maximum number of plants allowed
z*= 22,850.50
Kola
400
0
50
Ethiopia
Tanzania
Nigeria
Ginko
0
0
550
Kola
425
0
25
37
Key Points from Lesson
MIT Center for

38
Key Points from Lesson (1/2)

IPs and MILPs are different from LPs
n
n
Much harder to solve since solution space expands!

Formulations
w LPs a correct formulation is generally a good formulation
w For IPs a correct formulation is necessary but not sufficient to
guarantee solvability
IPs require solving multiple LPs to establish bounds

relaxing the Integer constraints
Cant just round the LP solution might not be feasible
When using integer (not binary) variables, solve the

LP first to see if it is sufficient.
MIT Center for
39
Key Points from Lesson (2/2)

Binary variables are very powerful and can be used for
modeling logical conditions
n
If Then links continuous to binary variables
x
i
My j 0 j
Either Or ensures a minimum level if used at all
x
i
ij
ij
My j 0 j
x
i
ij
L j y j 0 j
Select From picks the best X of Y choices (min or max)
x
i
MIT Center for

ij
My j 0 j
yj N
40
Questions, Comments, Suggestions?

Use the Discussion Forum!
Athena before and after completing the MITx MicroMasters Credential.

(photos courtesy of Lana Scott)
MIT Center for

MIT Center for

caplice@mit.edu
ctl.mit.edu

Sc0x w3l1 Clean Final

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sc0x w3l1 Clean Final

Uploaded by

Copyright:

Available Formats

Integer

and Mixed Integer

MIT Center for

Numbers, Numbers, Everywhere!

B = Binary Integers = {0,1}

Why the heck do we care?

By Mortalmoth (Own work) [Public domain], via Wikimedia Commons

When its physically impossible to have fractional solutions

Allows for modeling logical conditions (Binary)

If we have product leaving plant A then we must open it

Why do we have to treat them differently?

Banner Chemicals II: IP Example

MIT Center for

Motivating Problem Banner Chemicals II

Banner Chemicals manufactures specialty chemicals. One of their products

MIT Center for

Max z(XH, XS) = 80XH + 200XS

3XH + 2XS 300

Feasible region becomes a collection of points, no longer a convex hull

How to find solution to IP?

the closest acceptable integer values?

Max z(XH, XS) = 80XH + 200XS

3XH + 2XS 300

Rounding to closest 10 barrel solution for (XH, XS):

zLOT(30, 90) = $20,400 but it is infeasible (Plant constraint)

So, using this approach z*LOT= $18,400 with XH=30, XS=80

Lets solve all of the points to make sure!

Max z(XH, XS) = 80XH + 200XS

3XH + 2XS 300

Each cell shows z = 80XH + 200XS

Formulation Changes . . . not much!

3XH + 2XS 300

In order to solve in integer values of

Convert Decision Variables

Scale the coefficients and constraint RHS

Indicate that the new DVs are Integers

MIT Center for

GoNuts Juice Company: Model 1

MIT Center for

GoNuts Juice Company

Image CC0 Public Domain from https://pixabay.com

Step 1. Determine Decision Variables

Step 2. Formulate Objective Function

xij = Number of units of product i made in plant j

Step 3. Formulate Constraints

Minimize z = Cost = 21xG,E + 22.5xK,E + 22.5xG,T + 24.5xK,T + 23xG,N + 25.5xK,N

Total min cost = 22,637.50

GoNuts Juice Company: Model 2

MIT Center for

GoNuts Juice Company: Model 2

Image CC0 Public Domain from https://pixabay.com

Step 1. Determine Decision Variables

Step 2. Formulate Objective Function

xij = Number of units of product i made in plant j

MIT Center for

Step 3. Formulate Constraints

xG,E + xK,E 425

Is this enough? Try solving it!

You need to ensure that if a plant produces product, then

MIT Center for

MIT Center for

How do yN, xGN and xKN interact?

xGN + xKN 750 yN

If the X values >0, then Y MUST be