1

Algebraic Formulations
Usually in class, we describe linear programs by writing them out fully. This is
fine for small linear programs, but it does not work when the linear programs
are very large. In that case, it helps to use algebraic formulations.

Algebraic formulations sound hard. But

they are not so hard. However, they do In this tutorial, we will explain
take a while to get used to. algebraic formulations with
some examples. Algebraic

2
 On Creating Algebraic Formulations
When we create algebraic formulations, we rely on substituting notations for
some of the coefficients. Let’s start with an example of a linear program.

min 500 x1 + 200 x2 + 250 x3 + 125 x4

s.t. 50,000 x1 + 25,000 x2 + 20,000 x3 + 15,000 x4 ≥ 1,500,000
0 ≤ x1 ≤ 20

0 ≤ x2 ≤ 15

0 ≤ x3 ≤ 10

0 ≤ x4 ≤ 15

This is the MSR example which is described on the next slide.

3
 MSR Marketing Inc.
adapted from Frontline Systems

•Need to choose ads to reach at least 1.5 million people

•Minimize Cost
•Upper bound on number of ads of each type

TV Radio Mail Newspaper

Audience Size 50,000 25,000 20,000 15,000

Cost/Impression $500 $200 $250 $125

Max # of ads 20 15 10 15

Decision variables:
• x1 is the number of TV ads.
• x2 is the number of radio ads.
• x3 is the number of mail ads.
• x4 is the number of newspaper ads.
4
 The LP Formulation again

min 500 x1 + 200 x2 + 250 x3 + 125 x4

s.t. 50,000 x1 + 25,000 x2 + 20,000 x3 + 15,000 x4 ≥ 1,500,000

0 ≤ x1 ≤ 20

0 ≤ x2 ≤ 15

0 ≤ x3 ≤ 10

0 ≤ x4 ≤ 15

Illustration of the objective function and constraints:

• The objective is to minimize the cost of advertising.
• The first constraint says that the number of people who see the
ads is at least 1.5 million.
• The remaining four constraints give upper and lower bounds
on the number of showings of each of the four ads.
5
 Transforming into an algebraic problem
We’ll transform this problem into an algebraic version in a couple
of stages. Then we’ll show how to do it all at once.

So, let’s start with the four upper bound constraints. Suppose that
we let d = (d1, d2, d3, d4) = (20, 15, 10, 15). We can then write
the linear program as follows:
min 500 x1 + 200 x2 + 250 x3 + 125 x4

s.t. 50,000 x1 + 25,000 x2 + 20,000 x3 + 15,000 x4 ≥ 1,500,000

0 ≤ x1 ≤ 20; 00 ≤≤ xxj 2≤≤ d15;
j for j0= ≤
1 to ≤ 10; 0 ≤ x4 ≤ 15
x3 4.

Is dj decision variable?

It looks like dj is a variable, but it isn’t. It’s called a

“parameter” and it means that there is an associated value
stored for it somewhere, perhaps in a spreadsheet, perhaps
in a database.
6
 Parameters versus decision variables

Most students (and others) find this

I don’t get it. confusing at the beginning. But after a
The d’s don’t while, one gets used to it.
look like For students who have seen linear
numbers to me.
algebra, it’s pretty similar to when one
first sees a system of linear equations
expressed as Ax = b.

Remember that even in Algebra 1, the equation for a line

is often represented as
“ax + by = c.”

7
 More on the algebraic formulation

min 500 x1 + 200 x2 + 250 x3 + 125 x4

s.t. 50,000 x1 + 25,000 x2 + 20,000 x3 + 15,000 x4 ≥ 1,500,000

0 ≤ xj ≤ dj for j = 1 to 4.

The key advantage of the algebraic formulation is that

the formulation becomes “independent” of the data.
Algebraic formulation For example, if we were to change the upper bounds
seems hard and I do
not get what is the on the x’s, this more algebraic version would still be
advantage of doing it valid.You will see more advantages in the next slides.
in this form.
Actually, it won’t be the algebraic version until we get
rid of almost all of the numbers. We will permit the
number 0 at times, plus numbers for the indices. The
above formulation is not yet the algebraic formulation.
We next make the remaining constraints more
algebraic.

8
 Making the remaining constraint more algebraic

Let aj be the audience size of the j-th ad type, which is the coefficient of
xj in the constraint. And let b denote the required number of people
reached by the ads. We then can rewrite the constraint.

min 500 x1 + 200 x2 + 250 x3 + 125 x4

s.t. 50,000
a1 xx
1 1++ 25,000
a2 x2x+
2 + 20,000 a4 x4x4≥ ≥ 1,500,000
a3 x3x+3 + 15,000 b
0 ≤ xj ≤ dj for j = 1 to 4.

It doesn’t look simpler than the old version. But it

would if there were 1000 variables instead of just 4.

It is not yet an algebraic formulation, and we still need

to write the objective function in an algebraic form.
9
 Transforming the cost coefficients

Let cj be the cost of an ad of type j, which is the cost coefficient of xj. We

now rewrite the objective.

min 500
c1 x11 + 200
c2 x22 ++ 250
c3 xx33++ 125
c4 xx44

s.t. a1 x1 + a2 x2 + a3 x3 + a4 x4 ≥ b
0 ≤ xj ≤ dj for j = 1 to 4.

This is a valid algebraic representation of the problem

if we know that there are exactly four variables. But
we can carry it a step further.

10
 Using Summation Notation

Next we use summation notation and rewrite the LP formulation as

follows:

min c1 x1 + c2 x2 + c3 x3 + c4 x4

s.t. a1 x1 + a2 x2 + a3 x3 + a4 x4 ≥ b
0 ≤ xj ≤ dj for j = 1 to 4.

11
 Replacing the number of variables.

Minimize

subject to

Next Finally, we use n to

represent the number of Yes. I know it looks
much more abstract
variables. than the original
formulation. But the
abstraction means
that this formulation
is correct for many
different possible
choices of the data.

12
 Summary of the transformation
• Let xj be the number of ads purchased of type j for j = 1 to n.
• Let aj be the number of persons who view one ad of type j
• Let b be the required number of viewers to see the ads. (That is, the total number of
viewers must be at least b)
• Let dj be an upper bound on the number of ads purchased of type j.

Minimize

subject to

Minimize 500 x1 + 200 x2 + 250 x3 + 125 x4

subject to 50,000 x1 + 25,000 x2 + 20,000 x3 + 15,000 x4 ≥ 1,500,000

0 ≤ x1 ≤ 20; 0 ≤ x2 ≤ 15; 0 ≤ x3 ≤ 10; 0 ≤ x4 ≤ 15;

13
 On the reason for algebraic formulations

• Remember that the advantage of algebraic formulations is in their ability to describe very
large problems in a very compact manner. This is critical if one is to model large
problems, involving thousands or perhaps millions of variables.

• For small problems, it seems unnecessarily cumbersome and difficult.

The notation is also

very useful when we
describe the simplex
algorithm for linear
programs.

We next formulate the problem

for DTC, David’s Tool Company

14
 Formulation of the DTC Problem
(David’s Tool Company)
We will write the linear program as if up the constraints are broken into two
parts, the demand constraints and the resource constraints.

Maximize z = 3 K + 5 S

subject to 2 K + 3 S δ 100 Gathering time:

Resource
K + 2 S δ 60 Smoothing time: Constraints
K + S δ 50 Delivery time:

K δ 40 Slingshot demand:
Bounds on
S δ 30 Shield demand: variables.
K,S ≥ 0 Non-negativity:

Decision Variables:
• K is number of kits made
• S is number of shields made
15
 Before checking out
the next page, try it
for yourself. There
are several correct
Don’t ask me. I have answers. We will
no clue what is going show one soon.
on. But I like
watching others
explain math.

16
 The algebraic formulation
Maximize z = 3 x
K1 ++ 55 Sx2

subject to 2 K + 3 S δ 100 Gathering time:

Resource
K + 2 S δ 60 Smoothing time: Constraints
K + S δ 50 Delivery time:

K δ 40 Slingshot demand:
Bounds on
S δ 30 Shield demand: variables.
K,S ≥ 0 Non-negativity:

Let xj be the number of items of type i that are produced. In the above
formulation we have replaced K by x1 and S by x2. This will make it more easily
described using algebraic notation.

17
 Some hints

For now, we will keep the number of variables as 2. Later on, we will write the
formulation so that the number of variables is n. This will be more general.

In linear programming, “n” is often used to represent the number of decision

variables. And “m” usually represents the number of constraints (excluding the
“≥ 0” constraints).

Also, the variables are often represented by letters near the end of
the alphabet such as w, x, y, and z. This convention is not always
followed, but it is used a lot.

18
 Resource Constraints

There are three resources: gathering time, smoothing time, and delivery time. We will let
the limits (upper bounds) on these three resources be denoted as b1, b2, and b3.

We let ai1 be the amount of resource i used in the making of one Kit. We let ai2 be the
amount of resource i used in making one shield.

subject to a
211K x1+ + a312Sx2 δ≤ b100
1 Gathering time:
Resource
a21K x+
1 + 2
a22S x2 δ≤ b260 Smoothing time: Constraints
a31K x+
1 + a32
S x2 δ≤ b50
3 Delivery time:

19
 Resource Constraints

a11 x1 + a12 x2 ≤ b1 Gathering

Gathering
time:
time:
Resource
Resource
a21 x1 + a22 x2 ≤ b2 Smoothing
Smoothing
time:
time: Constraints
Constraints
a31 x1 + a32 x2 ≤ b3 Delivery
Delivery
time:
time:

My personal preference is
My head
to write them using
hurts.
summation notation. It
gets even more concise,
and is easy once you get
used to it.

20
 The complete algebraic formulation
Maximize z = 3 x 1 + 5 x2

subject to 2 x1 + 3 x2 ≤ 100 Gathering time:

Resource
x1 + 2 x2 ≤ 60 Smoothing time: Constraints
x1 + x2 ≤ 50 Delivery time:

x1 ≤ 40 Slingshot demand:
0 ≤ xxj2≤≤ dj 30 Bounds on
Shield demand: variables.
for j = 1 to n.
x1, x2 ≥ 0 Non-negativity:

In this formulation:
dj : an upper bound on the demand for item j.
n : the number of items.
aij : the amount of resource i used up by one unit of item j.
m: the number of different resources.
pj : the profit from making one unit of item j
21
 Another Practice Example
The following example called “Charging a Blast Furnace” is from Section 1.3 of Applied
Mathematical Programming.

An iron foundry has a firm order to produce 1000 pounds of castings containing at least
0.45 percent manganese and between 3.25 percent and 5.50 percent silicon. As these
particular castings are a special order, there are no suitable castings on hand. The castings
sell for $0.45 per pound. The foundry has three types of pig iron available in essentially
unlimited amounts, with the following properties:
Before going to the
next slide, try to
formulate the linear
program.

• Pure manganese can be purchased at $8/pound.

• The cost of melting pig iron is .5 cents per pound.
• Let xA, xB, xC denote the amount of pig iron A, B, and C used.
• Let xM be the amount of manganese used.

22
 The LP formulation

maximize profit = revenue minus cost =

Objective
450 − 26x A − 30xB − 20xC − 8xM.

To see the s.t. The mixture weighs 1000 pounds.

model, just Weight
1000xA + 1000x constraint
B + 1000xB + xM = 1000.
keep on
clicking.
The mixture has at least 4.5 pounds of manganese.
Manganese
4.5xA + 5.0x constraint
B + 4.0xC + xM ≥ 4.5.

The amount of silicon is between 32.5 and 55 pounds.

40xA + 10xSilicon
B + 6xCconstraints
≤ 32.5
40xA + 10xB + 6xC ≥ 55

Each variable is nonnegative.

The constraints that must always be remembered
xA ≥ ,0 xB ≥ 0, xC ≥ 0, xM ≥ 0

23
 An Algebraic Version

Let’s consider an algebraic version of the example.

An iron foundry has a firm order to produce P pounds of castings

containing at least bj pounds of material j and at most uj pounds of
the material j for j = 1 to m. The castings sell for $d per pound.
The foundry has n types of pig iron available in essentially unlimited
Before going to the
amounts, with the following properties: Pig iron i costs ci dollars next slide, try to
per pound and the percentage of material j in the iron is aij. In formulate the linear
addition, the firm can purchase material j in its pure form for mi program.
dollars per bound. The cost of melting pig iron is $p per pound
regardless of the type of pig iron.

• Let xi be the amount of pig iron i used in the mixture.

• Let Mj be the amount of pure material j used that is
purchased and used in the mixture

24
 The LP formulation

Objective
maximize profit = Pd − Pp - ∑i ci xi - ∑j mj Mj

To see the s.t. The mixture weighs

WeightPconstraint
pounds.
model, just ∑i xi + ∑j Mj = P
keep on
clicking.
The mixture has at least bj pounds of material j
and at most uMaterial
j poundsconstraints
of material j.
∑i aij xi + ∑ Mj ≥ bj for j = 1 to m
∑i aij xi + ∑ Mj ≤ uj for j = 1 to m

Each variable is nonnegative.

≥ 0constraints
xiThe for i = 1 tothat
n must always be remembered
Mj ≥ 0 for j = 1 to m

25
 I hope that you are
getting the hang of If it helps, you can
this now. If not, all represent the
it takes is some more notation so that it is
practice. easier to remember
(but longer to write).
We’ll do this on the
next slide.

This is the manner in which it

is usually described to a
“modeling language”, which then
rewrites the LP in a format
that can be solved by a
computer. We won’t be using
modeling languages, but it is
worth knowing that they exist.
2
6
26
 SETS:
IRONS: Set of pig irons
MATERIALS: Set of materials

VARIABLES:
IronUsed(j): amount of iron i used, for j ∈ IRONS.
Purchased(i) : = amount of material j purchased, for i ∈ MATERIALS

PARAMETERS (Data)
CastingRevenue: The price per pound for selling castings
PigIronCost(j): Cost per pound of pig iron j for j ∈ IRONS
MaterialCost(i): Cost per pound of material i for i ∈ MATERIALS
MeltingCost = Cost per pound of melting any of the pig irons
TotalCastings = number of pounds of castings to be sold Materials_Per_Iron(i,
j) = The amount of material i in pig iron j
for i ∈ MATERIALS and j ∈ IRONS.
LowerLimit(i): The minimum fraction of Material i needed in the
mixture
UpperLimit(i): The maximum fraction of material i allowedin the
2
mixture
7
27
 The LP formulation, for the last time

max TotalCastings × CastingRevenue

− ∑ Objective
[ MeltingCost(j) + PigIronCost(j)] IronUsed(j)
j∈IRONS
− ∑i∈MATERIALS MaterialCost(i) × Purchased(i)

s.t. ∑j∈IRONS IronUsed(j) Weight

+ ∑i∈MATERIALS Purchased(i) = TotalCastings
constraint

LowerLimit(i) ≤ ∑j∈IRONS Materials_Per_Iron(i, j) × IronUsed(j) +

Material
∑i∈MATERIALS constraints
Purchased(i) ≤ UpperLimit(i)
for i ∈ Materials

IronUsed(j) ≥ 0 for j ∈ IRONS

The constraints that must always be remembered
Purchased(i) ≥ 0 for i ∈ MATERIALS

28
28
 The huge advantage of the
previous formulation is that it is
much easier to debug and
extremely flexible. Notation is
consistently used. Sets are well
defined. Sets, Variables, and
Parameters are all defined using
easily understood terms.

The presumption is that the data If I don’t understand things

is all stored in a database that the in three different ways, am
“modeling language” can directly I doing better or worse?
access.

2
9
29
 Notation for linear programs in standard form

• Finally, we show some conventions that are used in describing a linear

programming in standard form. The conventions are used in 15.053.

• There are usually n variables and m equality constraints

• The variables are usually x1, …, xn.
• The cost coefficients are usually c1, …, cn. (Objective function coefficients
are often called cost coefficients even if one is maximizing profit. It is
widely agreed that this is a weird convention, but it is commonly done in
any case.)
• The coefficient for xj in constraint i is aij. The RHS is bi.
• Then the LP in “standard form” can be written as follows:

30
 Last slide

You will have a chance to

In case you were
practice algebraic
wondering, there are
formulations on the homework
different ways of
sets.
writing algebraic
formulations. You can
choose notation And that’s the end of this
differently, and you tutorial. I hope it was of value
can combine groups of to you. Bye!
constraints
differently.

31
31
MIT OpenCourseWare
http://ocw.mit.edu

15.053 Optimization Methods in Management Science

Spring 2013

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.