Octave with Spice

Or a Gentle Introduction to GNU Octave Towards

Linear Programming

Tommi Sottinen

[email protected]
www.uwasa.fi/∼tsottine/spicy or/octave with spice.pdf

August 30, 2021

October 31, 2022
Preface

These notes are an introduction on how to use GNU Octave for the six week 5 ECTS course
ORMS1020 Operations Research in the University of Vaasa. I expect that it will take two
weeks to go through the material presented here in the course. The remaining four weeks will
then be from the companion notes “Linear Programming with Spice”.
The first three chapters of these notes are the gentle introduction to Octave and the fourth
chapter is the spice. Indeed, the first three chapters should give a student with minimal
experience of programming the necessary background to understand chapter 4, where we show
how Linear Programming problems (LP’s) can be solved by using GNU Linear Programming
Kit (GLPK) in GNU Octave.

T.S.
Vaasa October 31, 2022
Contents

1 Getting GNU Octave 3

Installing GNU Octave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Getting Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 GNU Octave as Calculator 7

Simple Calculations and Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Matrices and Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Working with m-Files 16

Script m-Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Function m-Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Solving LP’s with GLPK 21

Product-Mix Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Solving Product-Mix Problem with GLPK . . . . . . . . . . . . . . . . . . . . . . . . 24
Chapter 1

Getting GNU Octave

These instructions refer to GNU Octave version 6.2.0 released in 2021-02-20. Instructions for
newer versions (e.g. 6.3.0 released in 2021-07-11) should be pretty much the same. If you have
already an older GNU Octave installed, it will probably work just fine. There is most likely no
reason to update your GNU Octave for the purpose of these notes and the ORMS1020 course.
On the other hand, there is absolutely no reason not to update your GNU Octave to the latest
stable version.

Installing GNU Octave

GNU Octave is open source free alternative to the famous mathematical programming language
Matlab. The home page of GNU Octave is https://www.gnu.org/software/octave/index, where
you can download it and find its documentations.
For Windows users there are installers for GNU Octave. A direct link to the installers
is https://www.gnu.org/software/octave/download#ms-windows. It is recommended to use
Windows-64 installer.
If you are a macOS user and familiar with Homebrew or MacPorts, you can find an in-
staller that way. There are also instructions for macOS users in the GNU Octave wiki page
https://wiki.octave.org/Octave for macOS.
Linux users find help on how to install GNU Octave from the GNU Octave wiki page
https://wiki.octave.org/Octave for GNU/Linux.
Finally, since GNU Octave is open source, you can also compile it yourself. The source code
can be found here: https://ftpmirror.gnu.org/octave.
After installation, depending on your system, you should have Octave GUI (Graphical
User Interface) and maybe also Octave CLI (Command Line Interface) installed. I strongly
recommend using the GUI version. Once you open the GUI version of GNU Octave, you
should have something like the following window in your screen.
 Chapter 1 Installing GNU Octave 4

Note that depending on your system setup, you Octave GUI window may look slightly
different.
The following is our first exercise. The exercises are littered all around the notes and not
collected to separate sections. They also follow the general running numbering of subsections.
This makes finding the exercises slightly inconvenient and the numbering may look a bit con-
fusing. The reason for this inconvenience is that in this way you are encouraged to read the
material and you will also know how the exercises are related to the material. The first Exercise
1.1 is simple. Its almost only point is to check that your installation of GNU Octave works.

1.1 Exercise (Hello, World!)

Type help printf and help disp in the GNU Octave Command Window. Then make GNU
Octave say ”Hello, World!” to you.

1.2 Remark (Octave Online)

If you failed to install GNU Octave to your computer, or just for some reason do not want to
do it, you can use GNU Octave online in https://octave-online.net/. In Octave Online you
 Chapter 1 Getting Help 5

can use the command line (a.k.a. console) immediately, but using script files (a.k.a. m-files)
requires signing in to the system. You can simply sign in by using your e-mail.
Octave Online seems convenient, but I strongly discourage using it. Besides the obvious
problems with online systems, it seems that Octave Online is very slow compared to installing
GNU Octave on your own computer. Also, it seems that Octave Online is not comfortable
with big data. Finally, I have limited experience in using Octave Online, so I may not be able
to help with the problems you will experience in using it.

Getting Help
The most obvious way of getting help on GNU Octave is to switch to the Documenta-
tion tab next to the Command Window tab at the bottom of your GNU Octave GUI
window. This will open the GNU Octave Manual. You can also find this manual in
https://octave.org/doc/v6.2.0/. (This is the manual for version 6.2.0. I leave it to you to
guess where you can find the manual for version 6.3.0.)
GNU Octave also has Wiki pages https://wiki.octave.org/GNU Octave Wiki with FAQ
https://wiki.octave.org/FAQ. This is a good place to start if you have problems.
There is also a GNU Octave community https://octave.discourse.group/, where you can ask
for help. This is a good place to ask if you have specific advanced problems or you think you
have found a bug.
Finally, if you already know what you want to do, but do not know or remember the details,
you can type help <topic> on the Command Window. The next Example 1.3 explains how
to use help.

1.3 Example (Help tan)

Suppose you want to solve the equation

(1.4) tan x = 0.5.

Formally, a solution is

x = tan−1 0.5,

where tan−1 denotes an inverse of the function tan. Unfortunately, the inverses of trigono-
metric functions have funny names. Also, some might interpret tan−1 y as 1/ tan y . So, how
do we solve (1.4) with GNU Octave? The best way to start is to see what GNU Octave has
to say about the tan function. So, we type help tan in the Command Window:
1 >> help tan
2 ’ tan ’ i s a b u i l t −i n function from t h e f i l e l i b i n t e r p / c o r e f c n / mappers . c c
3
4 −− tan ( Z )
5 Compute t h e t a n g e n t f o r each e l e m e n t o f X i n r a d i a n s .
6
7 See a l s o : atan , tand , tanh .
 Chapter 1 Getting Help 6

8
9 A d d i t i o n a l help f o r b u i l t −i n f u n c t i o n s and o p e r a t o r s i s
10 a v a i l a b l e i n t h e o n l i n e version o f t h e manual . Use t h e command
11 ’ doc <t o p i c > ’ t o s e a r c h t h e manual i n d e x .
12
13 Help and i n f o r m a t i o n about Octave i s a l s o a v a i l a b l e on t h e WWW
14 a t h t t p s : / /www. o c t a v e . o r g and v i a t h e help@octave . o r g
15 mailing l i s t .

In the answer we spot (look at line 7) something vaguely familiar from our days in high school:
atan. This looks like arcus tan or arctan, which is one name for the inverse of tan. So, let us
ask help for that
1 >> help atan
2 ’ atan ’ i s a b u i l t −i n function from t h e f i l e l i b i n t e r p / c o r e f c n / mappers . c c
3
4 −− atan (X)
5 Compute t h e i n v e r s e t a n g e n t i n r a d i a n s f o r each e l e m e n t o f X.
6
7 See a l s o : tan , atand .
8
9 A d d i t i o n a l help f o r b u i l t −i n f u n c t i o n s and o p e r a t o r s i s
10 a v a i l a b l e i n t h e o n l i n e version o f t h e manual . Use t h e command
11 ’ doc <t o p i c > ’ t o s e a r c h t h e manual i n d e x .
12
13 Help and i n f o r m a t i o n about Octave i s a l s o a v a i l a b l e on t h e WWW
14 a t h t t p s : / /www. o c t a v e . o r g and v i a t h e help@octave . o r g
15 mailing l i s t .

Looks correct. So, the solution for (1.4) is given by the following Command Window conver-
sation.
1 >> atan ( 0 . 5 )
2 ans = 0 . 4 6 3 6

So, the answer to (1.4) is x = 0.4636.

 Chapter 2

GNU Octave as Calculator

GNU Octave is a very convenient calculator. It can be used as a simple calculator (with
variables), but its true power is in how it handles matrices and vectors.

Simple Calculations and Variables

Basic arithmetics with GNU Octave is pretty similar to any computing language: + and − signs
are what they are, multiplication sign is ∗, division sign is /, and ˆ denote the exponentiation.
Precedences follow some more or less natural rule, that may or may not be the same as you
have learned in school. Thus, the use of parentheses is strongly encouraged.

2.1 Example (What is Precedence?)

Suppose we want to calculate
2
(2.2) 1.4 × 10−2 + .
7+3
This can be easily calculated by hand. Indeed, we obtain
2 1.4 2
1.4 × 10−2 + = +
7+3 100 10
1.4 + 20
=
100
= 0.214.
Let us then calculate the same with GNU Octave. Let us give a wrong answer first:
1 >> 1.4∗10ˆ −2 + 2 / 7+3
2 ans = 3 . 2 9 9 7

This answer is wrong, because 2/7+3 ̸= 2/(7+3). The human reader might read the command
line 1 above differently because of all the extra spaces, but GNU Octave does not care about
extra spaces. Also, while 10ˆ−2 is technically correct, I strongly encourage to use parentheses
there to avoid mistakes. Here is the correct way to calculate (2.2):
1 >> 1.4∗10ˆ( −2) + 2/(7+3)
2 ans = 0 . 2 1 4 0
 Chapter 2 Simple Calculations and Variables 8

2.3 Exercise (Viral Precedence Problem)

Calculate the following viral math problem

8 ÷ 2(2 + 2)

with GNU Octave.

2.4 Exercise (Rounding Errors)

Computers are fast, but they still make errors. One typical source of errors are rounding errors
due to the way computers understand numbers. Make GNU Octave give a wrong aswer to
you (due to rounding errors).
Hint: 1/3 is difficult for us humans using decimal system. Computers use binary system.
For them 0.2 is difficult.

Let us then consider calculation with variables. Like most programming languages, GNU
Octave can handle many different types of variables. GNU Octave uses weak typing meaning
that we do not have to declare variables or their type like in Java, or C. If you don’t know what
this means, you are lucky. So, don’t worry about it. In this section we only consider (real)
numbers. In the next section we consider (real) vectors and (real) matrices. Later in the course
we will also consider strings and a little bit struct (structure) type variables. Other variable
types are not used in this course.

2.5 Example (Variables)

To assing a value 13 to a variable x we can simply write
1 >> x = 1/3
2 x = 0.3333

Now we should see in the Workspace panel a line that says something like x under Name and
double under Class. We can also ask what variables we have in our workspace by asking who
1 >> who
2 V a r i a b l e s v i s i b l e from t h e c u r r e n t s c o p e :
3
4 x

Since x is defined we can use it in calculations. For example, we can calculate 2x = 21/3 as
1 >> 2ˆ x
2 ans = 1 . 2 5 9 9

GNU Octave calculated the answer and assinged its value to the variable named ans. Indeed,
if we now ask who, we obtain
1 >> who
2 V a r i a b l e s v i s i b l e from t h e c u r r e n t s c o p e :
3
4 ans x
 Chapter 2 Simple Calculations and Variables 9

To see the values of the variables that are defined we can simply type them on the Command
Window
1 >> ans
2 ans = 1 . 2 5 9 9
3 >> x
4 x = 0.3333

2.6 Remark (Variable ans)

The variable ans that appeared in Example 2.5 above is used by GNU Octave to assing the
value of the latest answer, unless we assign its value to a different variable. Example 2.7 below
illustrates this point.

2.7 Example (clc, who, clear, and ans)

In the following discussion we start by clearing the visible Command Window by using the
command clc (you can also press CTRL+L), and then clear all the variables in the workspace
by using the command clear, and then assing some values to some variables.
1 >> c l c
2 >> who
3 V a r i a b l e s v i s i b l e from t h e c u r r e n t s c o p e :
4
5 ans x
6
7 >> c l e a r
8 >> who
9 >> x
10 error : ’ x ’ u n d e f i n e d n e a r l i n e 1 , column 1
11 >> x = 1/3
12 x = 0.3333
13 >> y = 2 ˆ ( 1 / 3 )
14 y = 1.2599
15 >> who
16 V a r i a b l e s v i s i b l e from t h e c u r r e n t s c o p e :
17
18 x y

So, after line 1 the Command Window becomes empty, but the variables are not cleared as is
shown by the who command in line 2. Lines 7 and 8 show that after the clear command the
workspace is empty of variables. Indeed, asking for x in line 9 gives a more or less obvious
error message. Now, in line 11 we define x and GNU Octave annoyingly echoes back what
happened. In line 13 we define y to be the calculation we did earlier in Example 2.5, when we
did not assing a variable to the solution. Since we now asked the answer to be stored in the
variable y, GNU Octave does not use the auxiliary variable ans as in Example 2.5. This can
be seen by asking who in line 15. We obtain the result that there are two variables, x and y,
defined. The variable ans is not defined.
 Chapter 2 Matrices and Vectors 10

In Example 2.7 we noticed the annoying thing that GNU Octave echoes back (almost)
everything we do. Quite often this is not what we want. Indeed, we usually want GNU Octave
to perform the calculations quietly and only give the answer when we ask for it. We can tell
GNU Octave to be quiet by ending our commands with semicolon. So, the semicolon works
very differently in GNU Octave than in Java or C!

2.8 Example (Shut-Up Semicolon)

Compare this with Example 2.7:
1 >> c l c
2 >> c l e a r
3 >> who
4 >> x = 1 / 3 ;
5 >> y = 2ˆ x ;
6 >> who
7 V a r i a b l e s v i s i b l e from t h e c u r r e n t s c o p e :
8
9 x y
10
11 >> x
12 x = 0.3333
13 >> y
14 y = 1.2599

2.9 Remark (Characters in Names)

Finally, a word of warning: GNU Octave, like most programming languages, is case sensitive.
This means that variables x and X are not the same. Also, there are some restrictions on
the variable names. As a general rule, do not start a name with a number and do not use
non-English letters like ä or ø. Obviously, you should not use spaces or special characters
(except the underscore ) in variable names.

Matrices and Vectors

GNU Octave, being a Matlab clone, is matrix oriented. This means that all the basic arithmetics
is designed to work for matrices.
A matrix is defined in GNU Octave by presenting its elements inside brackets row-by-row.
Elements in the row are separated by a comma or simply by a space. The rows are separated
by a semicolon (this is the second use of semicolon in GNU Octave). If A is a matrix, then
A(i, j ) is the element in its ith row and j th column.

2.10 Example (Matrices and Their Elements)

Let us consider the matrix
 
11 12 13 14
A = .
21 22 23 24
 Chapter 2 Matrices and Vectors 11

To introduce this matrix to GNU Octave we can assing

1 >> A = [ 1 1 12 13 1 4 ; 21 22 23 2 4 ]
2 A =
3
4 11 12 13 14
5 21 22 23 24

Now, for A = [Aij ] we have A13 = A1,3 = 13 and A31 = A3,1 does not exist. The following
conversation is how GNU Octave sees this:
1 >> A( 1 , 3 )
2 ans = 13
3 >> A( 3 , 1 )
4 error : A( 3 , ) : out o f bound 2 ( d i m e n s i o n s a r e 2 x4 )

A vector is simply a matrix that has either one row or one column. If it has only one column,
it is a column vector. If it has only one row, it is a row vector. If not otherwise stated, we will
assume that all vectors are column vectors. Also note that vectors that have only one row and
one column are called numbers.
Matrix algebra works with GNU Octave with the obvious and convenient way: if A and
B are matrices (with suitable dimensions), then the transpose A′ = A⊤ of A is in GNU
Octave A’. Sums like A + B and A − B are obvious, the product (when defined) AB is A∗B.
Matrix power An is obviously Aˆn, the inverse matrix A−1 can be written as Aˆ(−1) or inv(A).
Finally, the matrix equation (a.k.a. a system of linear equations) Ax = b can be solved by the
left-division \ as x=A\b. It is left for the reader to figure out what the right-division / does.
The n-by-n identity matrix I = In can be constructed with GNU Octave by using the
(terrible) pun eye(n) as the following console command shows:
1 >> eye ( 4 )
2 ans =
3
4 D i a g o n a l Matrix
5
6 1 0 0 0
7 0 1 0 0
8 0 0 1 0
9 0 0 0 1

In this course we do not use matrix algebra much as such. We will however use block
matrices a lot. This means that we will cut matrices apart and paste them together. This
can become very complicated very quickly. So, instead of giving the full exposure, we will just
consider some cases now. More complicated constructions will be given later.
Let A be some matrix. If we want to refer to its row i, a typical mathematical notation is
Ai• . One way to understand this is that the row i is fixed, but the columns “run free”. With
GNU Octave colon takes the place of the bullet: Ai• becomes A(i ,:) In the same way, A•j and
A(:, j ) refer to the fixed column j of the matrix A.
 Chapter 2 Matrices and Vectors 12

2.11 Example (Matrix Rows and Columns via Colon)

The following console conversation should explain what is going on with the bullet/colon
notation.
1 >> A = [ 1 0 0 2 0 ; 3 6 0 0 7 ]
2 A =
3
4 1 0 0 2 0
5 3 6 0 0 7
6
7 >> A( 1 , : )
8 ans =
9
10 1 0 0 2 0
11
12 >> A( 2 , : )
13 ans =
14
15 3 6 0 0 7
16
17 >> A( : , 1 )
18 ans =
19
20 1
21 3
22
23 >> A( : , 2 )
24 ans =
25
26 0
27 6
28
29 >> A( : , 6 )
30 error : A( , 6 ) : out o f bound 5 ( d i m e n s i o n s a r e 2 x5 )

We can also take out a block of matrix. In GNU Octave this means that we use A(v1,v2),
where v1 and v2 are vectors that indicate which rows and columns we take.

2.12 Example (Matrix Blocks)

The following discussion should explain how the cutting block construct A(v1,v2) works.
1 >> A = [ 1 1 12 13 1 4 ; 21 22 23 2 4 ; 31 32 33 3 4 ]
2 A =
3
4 11 12 13 14
5 21 22 23 24
6 31 32 33 34
7
8 >> v1 = [ 1 3 ]
9 v1 =
10
 Chapter 2 Matrices and Vectors 13

11 1 3
12
13 >> v2 = [ 3 4 ]
14 v2 =
15
16 3 4
17
18 >> A( v1 , v2 )
19 ans =
20
21 13 14
22 33 34
23
24 >> A( v1 , v2 ) = [ 0 0 ; 0 0 ]
25 A =
26
27 11 12 0 0
28 21 22 23 24
29 31 32 0 0

So here we first, in line 1, defined the matrix A. In line 8 we defined the (row or column,
it does not matter) vector v1 to correspond rows 1 and 3. In line 13 we defined the (row or
column, it does not matter) vector v2 to correspond columns 3 and 4. In line 19 we asked
what is the submatrix of the matrix A, where only rows 1 and 3 and columns 3 and 4 are
considered. Finally, in line 25 we asked GNU Octave to replace the 2-by-2 submatrix A(v1,v2)
with a 2-by-2 zero matrix.

Submatrices, that is cutting matrices, as considered above can be tricky. Block matrices,
that is pasting matrices, is arguably much simpler. Again the following console discussion of
Example 2.13 should explain how to paste matrices.

2.13 Example (Block Matrix)

Let
 
2
     4   
11 12 13 760   1 0
A= , b= , c= 0 , and I = .
21 22 23 −98   0 1
 4 
0

We want to construct the block matrix

 
 ′
 2 4 0 4 0 1
c 1  11 12 13 1 0 760  .
T = =
A I b
21 22 23 0 1 −98

This is done with the following console discussion

1 >> A = [ 1 1 12 1 3 ; 21 22 2 3 ]
2 A =
 Chapter 2 Matrices and Vectors 14

3
4 11 12 13
5 21 22 23
6
7 >> b = [ 7 6 0 −98] ’
8 b =
9
10 760
11 −98
12
13 >> I = eye ( 2 )
14 I =
15
16 D i a g o n a l Matrix
17
18 1 0
19 0 1
20
21 >> c = [ 2 4 0 4 0 ] ’
22 c =
23
24 2
25 4
26 0
27 4
28 0
29
30 >> T = [ c ’ 1 ; A I b ]
31 T =
32
33 2 4 0 4 0 1
34 11 12 13 1 0 760
35 21 22 23 0 1 −98

2.14 Exercise (Matrix Block)

Let
 
11 12 13 14 15 16
A =  21 22 23 24 25 26  .
31 32 33 34 35 36

What is A([2 3], [1 6]) ?

2.15 Exercise (Marix Block with Colon Notation)

In GNU Octave there is a construct n:m that builds a vector of successive indices in the
following way: n:m=[n n+1 n+2 ... m]. So, for example, 2:5 = [2 3 4 5] .
Chapter 2 Matrices and Vectors 15

Let
 
11 12 13 14 15 16
 21 22 23 24 25 26 
A = 
 31
.
32 33 34 35 36 
41 42 43 44 45 46

What is A(2:4, [1 6]) ?

 Chapter 3

Working with m-Files

So far we have used the Command Window when discussing with GNU Octave. This becomes
very quickly very clumsy. In this chapter we introduce the so-called m-files that makes this
discussion much more fluent.
There are two types of m-files: script files and function files.

Script m-Files
A script type m-file is simply a list of console commands. The advantage of using script m-files
instead of the console are twofold:

(i) You can edit and debug the file and run all its command at once without the need executes
many commands in a row.
(ii) You can format the file so that, for example, matrices can be adjusted by rows and
columns.

Suppose we want to calculate the net present value of a cash-flow under a given interest
rate. As all students with loans should know, the formula for this is
T
X ct
(3.1) npv = ,
t=1
(1 + r)t

where ct is the cash received at the end of period t, T is the number of periods, and r is the
(fixed) rate of interest during the periods.

3.2 Problem (Net Present Value)

Suppose we want to calculate the net present value of a cash-flow gives us 1 000 EUR, 5 000
EUR and 2 500 EUR respectively at the end of the following 3 years. We assume that the
annual interest rate is fixed 0.5%. Thus the cash-flow is the vector c = [1000 5000 2500]′ and
the interest rate is r = 0.005. Calculating this with Console Window by using the formula
(3.1) would be quite painful. So, we should do something else. We give two solutions by using
script m-files in this section. The first one is quick and dirty and the second one is elegant
but tedious. In both solutions we create a text file (m-files are technically text files having the
 Chapter 3 Script m-Files 17

extension .m in their names) that is in the Current Directory of our GNU session. You can
set the Current Directory in the field near the top of your Octave GUI window. The script
file is executed by typing its name (without the .m extension) in the Command Window.

3.3 Solution (Net Present Value, Quick and Dirty)

Below is the code of a quick and dirty solution to the npv-problem 3.2 (indeed, it’s a one-
liner). The code is the contents of the m-file npv qd.m, which can be downloaded from
www.uwasa.fi/∼tsottine/spicy or/npv qd.m
1 npv = 1 0 0 0 / ( ( 1 + 0 . 0 0 5 ) ˆ 1 ) + 5 0 0 0 / ( ( 1 + 0 . 0 0 5 ) ˆ 2 ) + 2 5 0 0 / ( ( 1 + 0 . 0 0 5 ) ˆ 3 )

As you can see, this is very quick and very dirty. To get the solution, make sure that the
file npv qd.m is in your Current Directory. You should be able to see it listed in the File
Browser panel (top left panel typically). To execute the file simply state its name without the
.m extension. Here is the related Console Window discussion
1 >> npv qd
2 npv = 8 4 0 8 . 3

I strongly discourage using quick and dirty m files like in Example 3.3. One of the dirtiest
and stupidest thing we did was that we plugged in the numbers in the formula. You should
absolutely never ever do this! What we should have done is to give the numbers as variables
and then implement the formula for GNU Octave in a general form. Of course, if you are in a
hurry and do not intend to use the m file later, you can be quick and dirty. However, if you
intend to use the m file later, I encourage you to do something like the more elegant solution
in Solution 3.4 below.

3.4 Solution (Net Present Value, Tedious and Elegant)

Below is the lengthy code of an elegant but tedious solution for Problem 3.2. The
code is the contents of the m-file npv elegant.m, which can be downloaded here:
www.uwasa.fi/∼tsottine/spicy or/npv elegant.m
1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2 %%
3 %% Net P r e s e n t Value ( Problem 3 . 2 from Octave with S p i c e ;
4 %% S o l u t i o n 3 . 4 − Tedious and E l e g a n t )
5 %%
6 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
7
8 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
9 %% Data
10 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
11
12 c f = [ 1 0 0 0 5000 2500] ’; %% Cash−f l o w .
13 r = 0.005; %% Fixed r a t e o f i n t e r e s t .
14
 Chapter 3 Script m-Files 18

15 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
16 %% C a l c u l a t i o n s
17 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
18
19 T = length ( c f ) ; %% Number o f p e r i o d s .
20
21 %% C a l c u l a t e t h e npv sum i t e r a t i v e l y
22 npv = 0 ;
23 fo r t =1:T
24 npv = npv + c f ( t ) / ( (1+ r ) ˆ t ) ;
25 end
26
27 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
28 %% Output Net P r e s e n t Value ( npv )
29 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
30 npv

Compared to the quick and dirty solution of Example 3.3 this one is certainly tedious. But it
is also elegant.
Let us analyze the code above on a block level:
The code is divided into 4 blocks: The 1st block, lines 1–7 just describes what this code
is supposed to do. The 2nd block, lines 8–14, set the data. The key point here is that this
code will also work with different data, i.e., if you want to change the cash-flow or the interest
rate, you can do it here and the rest will work just fine. The 3rd block, lines 15–26 will do
the actual calculations, so that’s where the beef is. The final 4th block, lines 27–30, is for the
output.
A line-by-line analysis of the code is:
The lines 1–11 do absolutety nothing. They are there for the readers’ convenience. Indeed,
the percentage symbol % at the begining of the lines 1–6 and 8–10 means that these lines
are comments. GNU Octave will simply ignore these lines. GNU Octave will also ignore the
empty lines 7 and 11. Also, the use of double comment signs %% is not necessary; one %
would be enough. This is a matter of style. Indeed, a double comment sign %% should mean
a permanent comment, while a single comment sign % should mean something that is used
for debugging or testing purposes.
Line 12 defines the (column) vector cf to contain the cash-flow. There is a comment sign
%% at the end of the line, which GNU Octave will ignore. This comment is here for the
readers’ convenience. Line 13 defines the interest rate in the same way.
Lines 14–18 do absolutely nothing.
Line 19 gets the number of periods from the vector cf . The idea here is that we infer the
number of periods T from the length of the vector cf , and do not set it by hand. Indeed, this
way the m-file will work just fine if we change the parameters in line 12–13.
Lines 20–21 do absolutely nothing.
Lines 22–25 are the beef of the code. They calculate the sum in the formula (3.1) in a
for-loop. Type help for in GNU Octave to unerstand what is going on!
Again, lines 26–29, do absolutely nothing.
Finally, in line 30 we call the variable npv without semicolon. Thus GNU Octave will echo
the value of the variable npv to the console, showing us the result.
 Chapter 3 Function m-Files 19

Now, calling the script m-file npv elegant.m in the console (make su you have the m-file in
your Current Directory) solves the Net Present Value problem 3.2:
1 >> n p v e l e g a n t
2 npv = 8 4 0 8 . 3

3.5 Exercise
Calculate the Net Present Value (npv) of having 100 000 EUR after 10 years when the annual
interest rate is 50%.

Function m-Files
Suppose you want to calculate the net present value (3.1) for many different values of cash-
flows and/or interest rates. Then it makes sense to have the npv as a function, and not to make
separate elegant script m-files for all the different cf and r values. To do this, you should make
a function m-file for npv.
A function m-file is an m-file that starts with the keyword function. So, GNU Octave will
know not to try to parse this file as a script m-file. Instead, GNU Octave will only parse this
file, if it will be called as a function with given parameters.
Below is the code for the function m-file for npv, that can be downloaded from
www.uwasa.fi/∼tsottine/spicy or/npv.m
1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2 %% Function v = npv ( c f , r ) r e t u r n s t h e Net P r e s e n t Value ( npv ) o f t h e c a s h f l o w
3 %% c f . The c a s h f l o w c f i s r e c e i v e d a t t h e end o f each p e r i o d . The r a t e o f
4 %% r e t u r n o v e r t h e p e r i o d i s r . The parameter r i s s c a l a r . The c a s h f l o w c f
5 %% i s a ( column ) v e c t o r .
6 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
7
8 function v = npv ( c f , r )
9 T = length ( c f ) ; %% The number o f p e r i o d s .
10 pv = zeros (T, 1 ) ; %% I n i t i a l i z e p r e s e n t v a l u e s ( pv ) t o z e r o .
11 f o r t =1:T
12 pv ( t ) = c f ( t ) / (1+ r ) ˆ t ; %% S e t t h e pv ’ s .
13 end
14 v = sum( pv ) ; %% npv i s t h e sum o f pv ’ s .
15 end

Trying to call the function as it were a script m-file will lead to the following unsatisfying
discussion with GNU Octave console:
1 >> npv
2 error : ’ c f ’ u n d e f i n e d n e a r l i n e 9 , column 9
3 error : c a l l e d from
4 npv a t l i n e 9 column 7

A more fruitful conversation with GNU Octave console and npv-function could be
 Chapter 3 Function m-Files 20

1 >> r = 0 . 0 5
2 r = 0.050000
3 >> c f = [ 0 0 0 1 ]
4 cf =
5
6 0 0 0 1
7
8 >> npv ( c f , r )
9 ans = 0 . 8 2 2 7

So, what happened here?

In the first discussion GNU Octave just tried to understand the m-file npv.m as a script
type m-file. It skipped the lines starting with % signs and empty lines. Then it accepted the
keyword function and all its wrappings. But at line 9 column 9, GNU Octave should consider
length of the variable cf . But the variable cf has no value assigned to it. Thus, GNU Octave,
lacking abstract thinking, just quit.
The second discussion is quite different. There the variables r and cf are given values. Then
the function npv is called with variables that have values assigned to them. Thus GNU Octave
just passes the values of r and cf to the code written in the function m-file npv, and everything
works just fine.

3.6 Exercise (Variable Interest Rate Net Present Value)

Modify the function npv.m so that it assumes the interest rate be non-constant, i.e. the interest
rate for period t ir rt .
 Chapter 4

Solving LP’s with GLPK

A Linear Program (LP) is a linear optimization problem, where all the “things” are linear.
This means that that the objective function you want to optimize (maximize or minimize) is a
linear function of your decision variables, and so are also your constraints.
In this chapter we consider an LP called the Product-Mix Problem or Product-
Selection Problem. We first introduce the problem. Then we model it mathematically, and
finally we solve it by using the GNU Octave function glpk (GNU Linear Programming Kit).

Product-Mix Problem

4.1 Problem (Muad’Dib Bakery)

Muad’Dib Bakery produces three types of Fremen cakes: Atreides, Corrino and Harkonnen.
Each cake is made out of nutrient powders that are: fat, sugar, protein, water, and spice
Melange. The nutrient contents and selling prices (in solaris) of the cakes are:

Atreides Cake Fat 70 g, sugar 500 g, protein 280 g, water 600 ml, spice Melange 8 mg.
Price: 120 solaris.
Corrino Cake Fat 50 g, sugar 300 g, protein 180 g, water 800 ml, spice Melange 35 mg.
Price: 170 solaris.
Harkonnen Cake Fat 150 g, sugar 500 g, protein 50 g, water 500 ml, spice Melange 10 mg.
Price: 100 solaris.

Muad’Dib Bakery has the following amount of nutrient powders at its disposal daily:

Fat 100 kg
Sugar 400 kg
Protein 150 kg
Water 300 l
Spice Melange 10 g

Muad’Dib Bakery wants to maximize daily revenues. What should Muad’Dib Bakery do?
 Chapter 4 Product-Mix Problem 22

To model mathematically optimization problems like 4.1 one may follow the following three-step
procedure:

4.2 Algorithm (Optimization modeling)

Step 1 Find the decision variables, i.e. find out what are the variables whose values you
can choose.
Step 2 Find the objective function, i.e. find out how your objective to be minimized or
maximized depends on the decision variables.
Step 3 Find the constraints, i.e. find out the (in)equalities that the decision variables must
satisfy. (Don’t forget the possible sign constraints!)

Let us then start the mathematical modeling phase following Algorithm 4.2 above.
Muad’Dib Bakery produces three different types of cakes. So, the decision variables are:

x1 = number of Atreides cakes produced daily,

x2 = number of Corrino cakes produced daily,
x3 = number of Harkonnen cakes produced daily.

Once the decision variables x = [x1 x2 x3 ]′ are known, the objective function z = z(x)
of this problem is simply the revenue

z = 120x1 + 170x2 + 100x3

Note that the profit z depends linearly on the number of produced cakes x1 , x2 , x3 . So, this
is a linear problem so far (the constraints must turn out to be linear, too).
It may seem at first glance that the profit can be maximized by simply increasing x1 , x2
and x2 . Well, if life were that easy, let’s all start manufacturing Fremen cakes and become
infinitely rich! Unfortunately, there are constraints that limit the decisions (or else the model
is very likely to be wrong). Indeed, we need nutrient powders, and these are scarce resources.
Let us consider the constraint induced by the limited supply of fat powder. If we decide
to produce x1 Atreides cakes, x2 Corrino cakes, and x3 Harkonnen cakes, then the amount (in
grams) of fat powder we have used is 70x1 + 50x2 + 150x3 grams. Since we only have 100 kg
of fat powder, we have the constraint (note the unit change 1 kg = 1 000 g)

70x1 + 50x2 + 150x3 ≤ 100 000.

So, this is the constraint induced by the scarcity of fat powder. Note that this is a linear
constraint. This is good. Indeed, if it were not a linear constraint we wouldn’t be able to solve
it with GNU Octave’s function glpk.
Similarly, for sugar powder we have the linear constraint

500x1 + 300x2 + 500x3 ≤ 400 000,

since there is only 400 kg = 400 000 g of sugar powder available daily, and the requirement per
cake for Atreides, Corrino, and Harkonnen type cakes are 500 g, 300 g and 500 g, respectively.
 Chapter 4 Product-Mix Problem 23

The daily supply for protein powder is 150 kg = 150 000 g. Consequently, we obtain the
linear constraint

280x1 + 180x2 + 50x3 ≤ 150 000,

since Atreides cake needs 280 g of protein, Corrino cake need 180 g of protein, and Harkonnen
cake needs 50 g of protein.
For water powder we obtain the linear constraint

600x1 + 800x2 + 500x3 ≤ 300 000,

since 300 l = 300 000 ml, and the requirements per cake types are 600 ml, 800 ml, and 500 ml.
Finally, for spice Melange the linear constraint is

8x1 + 35x2 + 10x3 ≤ 10 000,

for reasons that should be obvious by now.

The mathematical model for Muad’Dib Bakery is now ready. Indeed, putting all that we
have found out together in a compact form, we have
max z = 120x1 + 170x2 + 100x3 (revenue)
s.t. 70x1 + 50x2 + 150x3 ≤ 100 000 (fat)
500x1 + 300x2 + 500x3 ≤ 400 000 (sugar)
(4.3) 280x1 + 180x2 + 50x3 ≤ 150 000 (protein)
600x1 + 800x2 + 500x3 ≤ 300 000 (water)
8x1 + 35x2 + 10x3 ≤ 10 000 (spice)
x1 , x2 , x3 ≥ 0 (sign constraints)
Note the last sign constraints. They ensure that the values of the decision variables will
always be positive (Muad’Dib Bakery sells cakes; it does not buy them). The problem does not
state this explicitly, but it’s still important (and obvious).

4.4 Remark (Muad’Dib LP with Matrices)

The compact form (4.3) can be written even more compactly by using matrices. Indeed, let x =
[x1 x2 x3 ]′ be the column vector containing the three decision variables. Let c = [120 170 100]′
be the column vector containing the prices associated with the decision variables. Then, the
objective function z can identified with the vector c as

z = c′ x.

Let then
 
70 50 150

 500 300 500 

A = 
 280 180 50 .

 600 800 500 
8 35 10

This means that A is the technology matrix that corresponds to how the different products
(cakes) are produced from the resources (nutrient powders). The final piece of data is the
 Chapter 4 Solving Product-Mix Problem with GLPK 24

column vector b = [10000 400000 150000 300000 10000]′ that tells the number of available
resources. With c, A, and b the Muad’Dib Bakery LP (4.3) can be written very compactly
as
max z = c′ x
(4.5) s.t. Ax ≤ b .
x ≥ 0

4.6 Exercise (Limited Atreides, Part I)

Suppose that there is only limited demand for Atreides cakes: only 100 Atreides cakes can be
sold daily. Model the Muad’Dib Bakery problem so that it takes into account this limitation.

4.7 Exercise (Harkonnen Promise, Part I)

Consider the Muad’Dib Bakery problem 4.1. Assume that the bakery has promised to bake
at least one Harkonnen cake each day. Model this new problem mathematically, i.e., find the
LP model similar to (4.3) for this problem.
Hint: You will have a new constraint with a lower bound instead of an upper bound.

Now that the mathematical modeling phase is finished, there is the implementation
phase, i.e. we have to tell the problem to a computer so that it can solve it. In the next
section we show how to do this with GNU Octave and its GLPK solver.

Solving Product-Mix Problem with GLPK

There are many LP (Linear Program) and MILP (Mixed Integer Linear Program) solvers in
the world. In GNU Octave, one has GLPK (GNU Linear Programming Kit). Here you have
its help (version 6.2.0):
1 >> help g l p k
2 ’ g l p k ’ i s a function from t h e f i l e / snap / o c t a v e /78/ s h a r e / o c t a v e / 6 . 2 . 0 /m/
o p t i m i z a t i o n / g l p k .m
3
4 −− [XOPT, FMIN, ERRNUM, EXTRA] = g l p k (C, A, B, LB, UB, CTYPE,
5 VARTYPE, SENSE, PARAM)
6 S o l v e a l i n e a r program u s i n g t h e GNU GLPK l i b r a r y .
7
8 Given t h r e e arguments , ’ g l p k ’ s o l v e s t h e f o l l o w i n g s t a n d a r d LP :
9
10 min C’ ∗ x
11
12 s u b j e c t t o
13
 Chapter 4 Solving Product-Mix Problem with GLPK 25

14 A∗x = b
15 x >= 0
16
17 but may a l s o s o l v e problems o f t h e form
18
19 [ min | max ] C’ ∗ x
20
21 s u b j e c t to
22
23 A∗x [ ”=” | ”<=” | ”>=” ] b
24 x >= LB
25 x <= UB
26
27 Input arguments :
28
29 C
30 A column a r r a y c o n t a i n i n g t h e o b j e c t i v e function c o e f f i c i e n t s .
31
32 A
33 A matrix c o n t a i n i n g t h e c o n s t r a i n t s c o e f f i c i e n t s .
34
35 B
36 A column a r r a y c o n t a i n i n g t h e r i g h t −hand s i d e v a l u e f o r each
37 c o n s t r a i n t i n t h e c o n s t r a i n t matrix .
38
39 LB
40 An a r r a y c o n t a i n i n g t h e lower bound on each o f t h e v a r i a b l e s .
41 I f LB i s not s u p p l i e d , t h e d e f a u l t lower bound f o r t h e
42 variables i s zero .
43
44 UB
45 An a r r a y c o n t a i n i n g t h e upper bound on each o f t h e v a r i a b l e s .
46 I f UB i s not s u p p l i e d , t h e d e f a u l t upper bound i s assumed t o
47 be i n f i n i t e .
48
49 CTYPE
50 An a r r a y o f c h a r a c t e r s c o n t a i n i n g t h e s e n s e o f each c o n s t r a i n t
51 i n t h e c o n s t r a i n t matrix . Each e l e m e n t o f t h e a r r a y may be
52 one o f t h e f o l l o w i n g v a l u e s
53
54 ”F”
55 A f r e e ( unbounded ) c o n s t r a i n t ( t h e c o n s t r a i n t i s
56 ignored ) .
57
58 ”U”
59 An i n e q u a l i t y c o n s t r a i n t with an upper bound ( ’A( i , : ) ∗x
60 <= b ( i ) ’ ) .
61
62 ”S”
63 An e q u a l i t y c o n s t r a i n t ( ’A( i , : ) ∗x = b ( i ) ’ ) .
64
65 ”L”
66 An i n e q u a l i t y with a lower bound ( ’A( i , : ) ∗x >= b ( i ) ’ ) .
67
68 ”D”
69 An i n e q u a l i t y c o n s t r a i n t with both upper and lower bounds
70 ( ’A( i , : ) ∗x >= −b ( i ) ’ ) a n d ( ’A( i , : ) ∗x <= b ( i ) ’ ) .
 Chapter 4 Solving Product-Mix Problem with GLPK 26

71
72 VARTYPE
73 A column a r r a y c o n t a i n i n g t h e t y p e s o f t h e v a r i a b l e s .
74
75 ”C”
76 A continuous variable .
77
78 ”I”
79 An i n t e g e r v a r i a b l e .
80
81 SENSE
82 I f SENSE i s 1 , t h e problem i s a m i n i m i z a t i o n . I f SENSE i s −1,
83 t h e problem i s a maximization . The d e f a u l t v a l u e i s 1 .
84
85 PARAM
86 A s t r u c t u r e c o n t a i n i n g t h e f o l l o w i n g p a r a m e t e r s used t o d e f i n e
87 the behavior of s o l v e r . Missing elements in the s t r u c t u r e
88 t a k e on d e f a u l t v a l u e s , s o you o n l y need t o set t h e e l e m e n t s
89 t h a t you wish t o change from t h e d e f a u l t .
90
91 I n t e g e r parameters :
92
93 ’ msglev ( d e f a u l t : 1 ) ’
94 L e v e l o f m es sa ges output by s o l v e r r o u t i n e s :
95
96 0 ( ’GLP MSG OFF ’ )
97 No output .
98
99 1 ( ’GLP MSG ERR ’ )
100 E r r o r and warning me ssa ge s o n l y .
101
102 2 ( ’GLP MSG ON ’ )
103 Normal output .
104
105 3 ( ’GLP MSG ALL ’ )
106 F u l l output ( i n c l u d e s i n f o r m a t i o n a l m es sa ges ) .
107
108 ’ s c a l e ( d e f a u l t : 16) ’
109 S c a l i n g o p t i o n . The v a l u e s can be combined with t h e
110 b i t w i s e OR o p e r a t o r and may be t h e f o l l o w i n g :
111
112 1 ( ’GLP SF GM ’ )
113 Geometric mean s c a l i n g .
114
115 16 ( ’ GLP SF EQ ’ )
116 Equilibration scaling .
117
118 32 ( ’ GLP SF 2N ’ )
119 Round s c a l e f a c t o r s t o power o f two .
120
121 64 ( ’ GLP SF SKIP ’ )
122 Skip i f problem i s w e l l s c a l e d .
123
124 A l t e r n a t i v e l y , a v a l u e o f 128 ( ’GLP SF AUTO ’ ) may be a l s o
125 s p e c i f i e d , i n which case t h e r o u t i n e c h o o s e s t h e s c a l i n g
126 options automatically .
127
 Chapter 4 Solving Product-Mix Problem with GLPK 27

128 ’ dual ( d e f a u l t : 1) ’
129 Simplex method o p t i o n :
130
131 1 ( ’GLP PRIMAL ’ )
132 Use two−phase p r i m a l s i m p l e x .
133
134 2 ( ’GLP DUALP ’ )
135 Use two−phase d u a l simplex , and i f i t f a i l s , switch
136 to the primal simplex .
137
138 3 ( ’GLP DUAL ’ )
139 Use two−phase d u a l s i m p l e x .
140
141 ’ p r i c e ( d e f a u l t : 34) ’
142 P r i c i n g o p t i o n ( f o r both p r i m a l and d u a l s i m p l e x ) :
143
144 17 ( ’GLP PT STD ’ )
145 Textbook p r i c i n g .
146
147 34 ( ’GLP PT PSE ’ )
148 S t e e p e s t edge p r i c i n g .
149
150 ’ i t l i m ( d e f a u l t : intmax ) ’
151 Simplex i t e r a t i o n s l i m i t . I t i s d e c r e a s e d by one each
152 time when one s i m p l e x i t e r a t i o n has been performed , and
153 reaching zero value s i g n a l s the s o l v e r to stop the
154 search .
155
156 ’ o u t f r q ( d e f a u l t : 200) ’
157 Output f r e q u e n c y , i n i t e r a t i o n s . This parameter
158 s p e c i f i e s how f r e q u e n t l y t h e s o l v e r s e n d s i n f o r m a t i o n
159 about t h e s o l u t i o n t o t h e s t a n d a r d output .
160
161 ’ branch ( d e f a u l t : 4 ) ’
162 Branching t e c h n i q u e o p t i o n ( f o r MIP o n l y ) :
163
164 1 ( ’GLP BR FFV ’ )
165 First fractional variable .
166
167 2 ( ’GLP BR LFV ’ )
168 Last f r a c t i o n a l v a r i a b l e .
169
170 3 ( ’GLP BR MFV ’ )
171 Most f r a c t i o n a l v a r i a b l e .
172
173 4 ( ’GLP BR DTH ’ )
174 H e u r i s t i c by D r i e b e c k and Tomlin .
175
176 5 ( ’GLP BR PCH ’ )
177 Hybrid p s e u d o c o s t h e u r i s t i c .
178
179 ’ btrack ( d e f a u l t : 4) ’
180 B a c k t r a c k i n g t e c h n i q u e o p t i o n ( f o r MIP o n l y ) :
181
182 1 ( ’GLP BT DFS ’ )
183 Depth f i r s t s e a r c h .
184
 Chapter 4 Solving Product-Mix Problem with GLPK 28

185 2 ( ’GLP BT BFS ’ )

186 Breadth f i r s t s e a r c h .
187
188 3 ( ’GLP BT BLB ’ )
189 Best l o c a l bound .
190
191 4 ( ’GLP BT BPH ’ )
192 Best p r o j e c t i o n h e u r i s t i c .
193
194 ’ p r e s o l ( d e f a u l t : 1) ’
195 I f t h i s f l a g i s set , t h e s i m p l e x s o l v e r u s e s t h e b u i l t −i n
196 LP p r e s o l v e r . O t h e r w i s e t h e LP p r e s o l v e r i s not used .
197
198 ’ l p s o l v e r ( d e f a u l t : 1) ’
199 S e l e c t which s o l v e r t o u s e . I f t h e problem i s a MIP
200 problem t h i s f l a g w i l l be i g n o r e d .
201
202 1
203 R e v i s e d s i m p l e x method .
204
205 2
206 I n t e r i o r p o i n t method .
207
208 ’ r t e s t ( d e f a u l t : 34) ’
209 Ra ti o t e s t t e c h n i q u e :
210
211 17 ( ’GLP RT STD ’ )
212 Standard ( ” t e x t b o o k ” ) .
213
214 34 ( ’GLP RT HAR ’ )
215 H a r r i s ’ two−p a s s r a t i o t e s t .
216
217 ’ tmlim ( d e f a u l t : intmax ) ’
218 S e a r c h i n g time l i m i t , i n m i l l i s e c o n d s .
219
220 ’ outdly ( d e f a u l t : 0) ’
221 Output dela y , i n s e c o n d s . This parameter s p e c i f i e s how
222 l o n g t h e s o l v e r s h o u l d d e l a y s e n d i n g i n f o r m a t i o n about
223 t h e s o l u t i o n t o t h e s t a n d a r d output .
224
225 ’ save ( d e f a u l t : 0) ’
226 I f t h i s parameter i s nonzero , save a copy o f t h e problem
227 i n CPLEX LP format t o t h e f i l e ’ ” outpb . l p ” ’ . There i s
228 c u r r e n t l y no way t o change t h e name o f t h e output f i l e .
229
230 Real p a r a m e t e r s :
231
232 ’ t o l b n d ( d e f a u l t : 1 e −7) ’
233 R e l a t i v e t o l e r a n c e used t o check i f t h e c u r r e n t b a s i c
234 s o l u t i o n i s primal f e a s i b l e . I t i s not recommended t h a t
235 you change t h i s parameter u n l e s s you have a d e t a i l e d
236 understanding o f i t s purpose .
237
238 ’ t o l d j ( d e f a u l t : 1 e −7) ’
239 A b s o l u t e t o l e r a n c e used t o check i f t h e c u r r e n t b a s i c
240 s o l u t i o n i s dual f e a s i b l e . I t i s not recommended t h a t
241 you change t h i s parameter u n l e s s you have a d e t a i l e d
 Chapter 4 Solving Product-Mix Problem with GLPK 29

242 understanding o f i t s purpose .

243
244 ’ t o l p i v ( d e f a u l t : 1 e −10) ’
245 R e l a t i v e t o l e r a n c e used t o c h o o s e e l i g i b l e p i v o t a l
246 elements of the simplex t a b l e . I t i s not recommended
247 t h a t you change t h i s parameter u n l e s s you have a d e t a i l e d
248 understanding o f i t s purpose .
249
250 ’ o b j l l ( d e f a u l t : −DBL MAX) ’
251 Lower l i m i t o f t h e o b j e c t i v e function . I f the o b j e c t i v e
252 function r e a c h e s t h i s l i m i t and c o n t i n u e s d e c r e a s i n g , t h e
253 s o l v e r s t o p s t h e s e a r c h . This parameter i s used i n t h e
254 d u a l s i m p l e x method o n l y .
255
256 ’ o b j u l ( d e f a u l t : +DBL MAX) ’
257 Upper l i m i t o f t h e o b j e c t i v e function . I f the o b j e c t i v e
258 function r e a c h e s t h i s l i m i t and c o n t i n u e s i n c r e a s i n g , t h e
259 s o l v e r s t o p s t h e s e a r c h . This parameter i s used i n t h e
260 dual simplex only .
261
262 ’ t o l i n t ( d e f a u l t : 1 e −5) ’
263 R e l a t i v e t o l e r a n c e used t o check i f t h e c u r r e n t b a s i c
264 solution is integer feasible . I t i s not recommended t h a t
265 you change t h i s parameter u n l e s s you have a d e t a i l e d
266 understanding o f i t s purpose .
267
268 ’ t o l o b j ( d e f a u l t : 1 e −7) ’
269 R e l a t i v e t o l e r a n c e used t o check i f t h e v a l u e o f t h e
270 o b j e c t i v e function i s not b e t t e r than i n t h e b e s t known
271 integer feasible solution . I t i s not recommended t h a t
272 you change t h i s parameter u n l e s s you have a d e t a i l e d
273 understanding o f i t s purpose .
274
275 Output v a l u e s :
276
277 XOPT
278 The o p t i m i z e r ( t h e v a l u e o f t h e d e c i s i o n v a r i a b l e s a t t h e
279 optimum ) .
280
281 FOPT
282 The optimum v a l u e o f t h e o b j e c t i v e function .
283
284 ERRNUM
285 E r r o r code .
286
287 0
288 No error .
289
290 1 ( ’GLP EBADB ’ )
291 Invalid basis .
292
293 2 ( ’GLP ESING ’ )
294 S i n g u l a r matrix .
295
296 3 ( ’GLP ECOND ’ )
297 I l l −c o n d i t i o n e d matrix .
298
 Chapter 4 Solving Product-Mix Problem with GLPK 30

299 4 ( ’GLP EBOUND ’ )

300 I n v a l i d bounds .
301
302 5 ( ’GLP EFAIL ’ )
303 Solver f a i l e d .
304
305 6 ( ’GLP EOBJLL ’ )
306 O b j e c t i v e function lower l i m i t r e a c h e d .
307
308 7 ( ’GLP EOBJUL ’ )
309 O b j e c t i v e function upper l i m i t r e a c h e d .
310
311 8 ( ’GLP EITLIM ’ )
312 I t e r a t i o n s l i m i t exhausted .
313
314 9 ( ’GLP ETMLIM ’ )
315 Time l i m i t e x h a u s t e d .
316
317 10 ( ’GLP ENOPFS ’ )
318 No p r i m a l f e a s i b l e s o l u t i o n .
319
320 11 ( ’GLP ENODFS ’ )
321 No d u a l f e a s i b l e s o l u t i o n .
322
323 12 ( ’GLP EROOT ’ )
324 Root LP optimum not p r o v i d e d .
325
326 13 ( ’GLP ESTOP ’ )
327 S e a r c h t e r m i n a t e d by a p p l i c a t i o n .
328
329 14 ( ’GLP EMIPGAP ’ )
330 R e l a t i v e MIP gap t o l e r a n c e r e a c h e d .
331
332 15 ( ’GLP ENOFEAS ’ )
333 No p r i m a l / d u a l f e a s i b l e s o l u t i o n .
334
335 16 ( ’GLP ENOCVG ’ )
336 No c o n v e r g e n c e .
337
338 17 ( ’GLP EINSTAB ’ )
339 Numerical i n s t a b i l i t y .
340
341 18 ( ’GLP EDATA ’ )
342 I n v a l i d data .
343
344 19 ( ’GLP ERANGE ’ )
345 R e s u l t out o f r a n g e .
346
347 EXTRA
348 A data s t r u c t u r e c o n t a i n i n g t h e f o l l o w i n g f i e l d s :
349
350 ’ lambda ’
351 Dual v a r i a b l e s .
352
353 ’ redcosts ’
354 Reduced C o s ts .
355
 Chapter 4 Solving Product-Mix Problem with GLPK 31

356 ’ time ’
357 Time ( i n s e c o n d s ) used f o r s o l v i n g LP/MIP problem .
358
359 ’ status ’
360 Status of the optimization .
361
362 1 ( ’GLP UNDEF ’ )
363 Solution status i s undefined .
364
365 2 ( ’GLP FEAS ’ )
366 Solution i s f e a s i b l e .
367
368 3 ( ’GLP INFEAS ’ )
369 Solution i s i n f e a s i b l e .
370
371 4 ( ’GLP NOFEAS ’ )
372 Problem has no f e a s i b l e s o l u t i o n .
373
374 5 ( ’GLP OPT ’ )
375 Solution i s optimal .
376
377 6 ( ’GLP UNBND ’ )
378 Problem has no unbounded s o l u t i o n .
379
380 Example :
381
382 c = [10 , 6 , 4] ’;
383 A = [ 1 , 1 , 1;
384 10 , 4 , 5 ;
385 2, 2, 6];
386 b = [ 1 0 0 , 600 , 3 0 0 ] ’ ;
387 lb = [0 , 0 , 0 ] ’ ;
388 ub = [ ] ;
389 c t y p e = ”UUU” ;
390 v a r t y p e = ”CCC” ;
391 s = −1;
392
393 param . msglev = 1 ;
394 param . i t l i m = 1 0 0 ;
395
396 [ xmin , fmin , s t a t u s , e x t r a ] = . . .
397 g l p k ( c , A, b , lb , ub , ctype , vartype , s , param ) ;
398
399 A d d i t i o n a l help f o r b u i l t −i n f u n c t i o n s and o p e r a t o r s i s
400 a v a i l a b l e i n t h e o n l i n e version o f t h e manual . Use t h e command
401 ’ doc <t o p i c > ’ t o s e a r c h t h e manual i n d e x .
402
403 Help and i n f o r m a t i o n about Octave i s a l s o a v a i l a b l e on t h e WWW
404 a t h t t p s : / /www. o c t a v e . o r g and v i a t h e help@octave . o r g
405 mailing l i s t .

Quite a lot, eh!

If you look at the lines 4–5 of the help listing above, you’ll find that there are three main
ingredients in the (Muad’Dib Bakery) LP problem:

• The objective function, typically denoted by c.

 Chapter 4 Solving Product-Mix Problem with GLPK 32

• The technology matrix, typically denoted by A.

• The constraint (upper) bounds, or available resources, typically denoted by b.

Additionally, we must tell the function glpk what are the lower bounds lb, the upper bounds
ub, the types of the constraints ctype, and the sense (max or min) of the problem sense. Finally
there is the input parameter param, but this is optional. For now, we do not care about it. The
parameters c, A and b are given in Remark 4.4. As for lb, the only lower bounds we have are the
sign constraints. From lines 39–42 we see that the default lower bound are the sign constraints.
So, we use the default. This means that we give the nothing for the lower bounds. In GNU
Octave the nothing is the empty matrix [] . As for the upper bounds ub, we do not have any.
Indeed, the upper bounds b are not for the decision variables, but for the technology matrix.
So, we give the nothing [] for the upper bounds also. Then we have the ctype, that is the type
of bounds associated with the technology matrix A. They are all upper bounds, i.e. of type
≤. There are 5 of them. Thus, we will have ctype=”UUUUU”. Then we have to tell the types of
our 3 decision variables x1 , x2 and x3 . We will assume that they are continuous. Indeed, we
will almost always assume this. So, vartype=”CCC”. Finally, we must tell glpk whether we are
maximizing or minimizing. We are maximizing. So according to lines 81–83 of the help listing
we set sense=−1.
Having set the parameters c, A and b, we can solve Problem 4.1 by calling the function glpk
as
1 [ x max , z max ] = g l p k ( c , A, b , [ ] , [ ] , ”UUUUU” , ”CCC” , −1) .

Then the optimal decision will be in the vector x max and the optimal revenue will be in the
scalar z max.

4.8 Solution (Muad’Dib Bakery)

Here is the script m-file that solves the Muad’Dib Bakery problems 4.1 (you can downdoad it
from www.uwasa.fi/∼tsottine/spicy or/muaddib.m)
1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2 %%
3 %% Muad ’ Dib Bakery . ( Problem 4 . 1 from Octave with S p i c e )
4 %%
5 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6
7 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
8 %% Data
9 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10
11 %% P r i c e s f o r A t r e i d e s , Corrino , Harkonnen ( S o l a r i )
12 c = [ 1 2 0 170 1 0 0 ] ’ ;
13
14 %% N u t r i e n t powder c o m p o s i t i o n o f t h e c a k e s A t r e i d e s , Corrino , Harkonnen
15 A = [ 70 50 1 5 0 ; %% Fat ( g )
16 500 300 5 0 0 ; %% Sugar ( g )
17 280 180 50; %% P r o t e i n ( g )
18 600 800 5 0 0 ; %% Water ( ml )
19 8 35 10 ] ; %% S p i c e Melange (mg)
20
21 %% A v a i l a b l e n u t r i e n t s ( n o t e u n i t s c a l e )
22 b = 1 0 0 0 ∗ [ 1 0 0 400 150 300 1 0 ] ’ ;
 Chapter 4 Solving Product-Mix Problem with GLPK 33

23
24 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
25 %% S o l u t i o n with GLPK
26 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
27
28 [ x max , z max ] = g l p k ( c , A, b , [ ] , [ ] , ”UUUUU” , ”CCC” , −1) ;
29
30 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
31 %% Output : x max f o r t h e o p t i m a l d e c i s i o n and z max f o r t h e o p t i m a l v a l u e .
32 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
33 x max
34 z max
35
36 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
37 %% Test r e s u l t :
38 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
39 %%>> muaddib
40 %%x max =
41 %%
42 %% 171.2329
43 %% 246.5753
44 %% 0
45 %%
46 %%z max = 6 . 2 4 6 6 e+04

Let us explain a little bit what happens in the m-file muaddib.m.

Lines 1–5 are comment lines that do absolutely nothing. They are there for the human
reader so that they see what this m-file is all about. You might also want to type help muaddib
in the GNU Octave console, just for fun.
Empty lines, like line 6, do absolute nothing.
Lines 7–10 do absolutely nothing, except they tell the human reader that next we have a
block where the data of the problem is given.
Line 11 is a comment line that tells that next the prices of the cakes are given.
In line 12 the prices of the cakes are stored to the vector c. The apostrophe makes the
vertor a column vector and the semicolon in the end suppresses the output, i.e., GNU Octave
will not echo to the console what just happened.
Line 14 is a comment line that tells the human reader that next the technology matrix is
given.
The technology matrix A is given in lines 15–19. Each line ends with a comment for the
human readers’ convenience. Note the semicolon in line 19, which tell the GNU Octave to
shut up instead of shouting what just happens in the console.
You should be able to guess by now what the line 21 does.
The resource vector b in given in line 22. Note the change of scale: each element in the
vector is multiplied by 1 000.
Lines 24–26 tell the human reader that the data section is finished and next the solution
is calculated.
The solution is calculated in the one-liner 28. The result is stored in the variables x max
and z max. Note the semicolon in the end of line 28. This means that the solution is not
echoed to the console.
 Chapter 4 Solving Product-Mix Problem with GLPK 34

Lines 30-32 tells the human reader that next we will ask GNU Octave to print the solutions.
Indeed, so far we have told GNU OCtave to keep quiet. If we just type a variable name, its
value will be printed in the console.
Finally, the commented section 36–46 is there just for test purposes. Indeed, you sould
run the m-file muaddib.m and check that you get the same result as given there.

Let us then run the m-file muaddib.m in the GNU Octave console (make sure that the m-file
is in you current directory). We get the solution:
1 >> muaddib
2 x max =
3
4 171.2329
5 246.5753
6 0
7
8 z max = 6 . 2 4 6 6 e+04

So, Muad’Dib Bakery should produce 171.239 Atreides cakes, 246.5753 Corrino cakes, and
no Harkonnen cakes at all. With this choise the daily revenue is maximized and it is 62 466
solaris.

4.9 Exercise (LP with No Story)

Consider the following LP that has no story:

max z = 12x1 + 4x2 + 8x3

s.t. 7x1 + 6x2 + 3x3 − 2x4 ≤ 20
5x2 + 4x3 + 3x4 ≤ 16
3x1 + x2 = 9 .
2x1 + x2 − x4 ≥ 2
− x2 + x3 ≥ 1
x1 , x2 , x3 , x4 ≥ 0

Solve it with glpk.

4.10 Exercise (Limited Atreides, Part II)

Solve the problem of Exercise 4.6 with glpk.

4.11 Exercise (Harkonnen Promise, Part II)

Solve the problem of Exercise 4.7 with glpk.
 Chapter 4 Solving Product-Mix Problem with GLPK 35

4.12 Remark (LP vs. IP vs. MILP)

The optimal solution 4.8 of Muad’Dib Bakery problem 4.1 turned out to be fractional: The
number of cakes produced are not whole numbers (or integers). It was not clearly stated that
the cakes should be whole. If we insist that the all the cake types produced must be whole
cakes, then instead of an LP (Linear Program) we have an IP (Integer Progam). If we only
insist that some of the cake types (Atreides, say) must be whole cakes, then we are dealing
with a MILP (Mixed Integer Linear Program).

4.13 Exercise (Muad’Dib Bakery with Integrity)

Solve the Muad’Dib Bakery problem 4.1 when only whole cakes can be sold.
Hint: Look at the glpk’s 7th input parameter vartype.

4.14 Remark (Muad’Dib Kwisatz Haderach)

Let us take a sneak peek into what we are going to learn during the Operations Research
course. At this point there is no need to fully understand what is going on here. Everything
will be explained in full detail later.
So far we have asked glpk only to tell us x and z the optimal decision and the optimal value
(max or min), i.e, we have used glpk in the form
1 [ x , z ] = g l p k ( c , A, b , lb , ub , ctype , vartype , s e n s e )

We have omitted the optional input parameter param, and we shall usually continue to do so.
Here, we use it a little bit, just to a have a small flavor of it. The parameter param (just p
later) is actually a tuning paramer. It is of struct type which means it has different fields
that can be set to different values. To go through all the possible tuning possibilities would
take too long, and would probably be almost pointless. We will consider only to fields here:

• msglev controls how much glpk will tell us what it is doing. Value 0 means blissful
quietness. Value 3 means annoying babbling.
• presol set whether glpk uses a so-called presolver (by default it does) or not. The idea of
the presolver is to try some easy tricks before going to a “full” solver.

In the example below we will set p.msglev=3 so that glpk will tell us “everything” it is doing.
We will also set p. presol=0, i.e., we will tell glpk not to use a presolver.
Let us consider the optional output parameters e and xtra. That is, let us consider calling
glpk as
1 [ x , z , e , x t r a ] = g l p k ( c , A, b , lb , ub , ctype , vartype , s e n s e , p )

The third output parameter e is of no interest to us, except in the case if something went
wrong. The fourth output parameter xtra, on the other hand, is very interesting. The output
parameter xtra is of struct type. This means that it is a structural variable that contains
different fields. Its fields are:
 Chapter 4 Solving Product-Mix Problem with GLPK 36

• lambda for the dual variables a.k.a. shadow prices a.k.a. marginal prices.
• redcosts for the reduced costs a.k.a. opportunity costs.
• time for the time it took to calculate the solution.
• status for the status of the solution. (This is a funny field, since it seems to be redundant
given the return value e. The author is truly confused why this even exitsts.)

Here is a solution to the Muad’Dib Bakery Problem 4.1 where we have tuned
glpk so that it babbles a lot and does not use a presolver. Also, we asked
for full output x, z, e, xtra. You can donwload the solution m-file muaddib kh here:
www.uwasa.fi/∼tsottine/spicy or/muaddib kh.m
1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2 %%
3 %% Muad ’ Dib Kwisatz Haderach . ( Remark 4 . 1 4 from Octave with S p i c e )
4 %%
5 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6
7 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
8 %% Data
9 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10
11 %% P r i c e s f o r A t r e i d e s , Corrino , Harkonnen ( S o l a r i )
12 c = [ 1 2 0 170 1 0 0 ] ’ ;
13
14 %% N u t r i e n t powder c o m p o s i t i o n of t h e c a k e s A t r e i d e s , Corrino , Harkonnen
15 A = [ 70 50 1 5 0 ; %% Fat ( g )
16 500 300 5 0 0 ; %% Sugar ( g )
17 280 180 5 0 ; %% Protein (g)
18 600 800 5 0 0 ; %% Water ( ml )
19 8 35 10 ] ; %% S p i c e Melange (mg)
20
21 %% A v a i l a b l e n u t r i e n t s ( n o t e u n i t s c a l e )
22 b = 1 0 0 0 ∗ [ 1 0 0 400 150 300 1 0 ] ’ ;
23
24 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
25 %% S o l u t i o n with GLPK with o u t p u t s
26 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
27
28 p . msglev = 3 ;
29 p . presol = 0;
30 [ x , z , e , x t r a ] = g l p k ( c , A, b , [ ] , [ ] , ”UUUUU” , ”CCC” , −1, p )

Running the m-file muaddib kh.m will give hte following output:
1 >> muaddib kh
2 Scaling . . .
3 A: min | a i j | = 8 . 0 0 0 e+00 max| a i j | = 8 . 0 0 0 e+02 r a t i o = 1 . 0 0 0 e+02
4 EQ: min | a i j | = 1 . 7 8 6 e −01 max| a i j | = 1 . 0 0 0 e+00 r a t i o = 5 . 6 0 0 e+00
5 Constructing i n i t i a l basis . . .
6 Size of t r i a n g u l a r part i s 5
7 GLPK Simplex Optimizer , v4 . 6 5
8 5 rows , 3 columns , 15 non−zeros
9 ∗ 0 : o b j = −0.000000000 e+00 i n f = 0 . 0 0 0 e+00 ( 3 )
10 ∗ 2: obj = 6 . 2 4 6 5 7 5 3 4 2 e+04 i n f = 0 . 0 0 0 e+00 ( 0 )
11 OPTIMAL LP SOLUTION FOUND
12 x =
 Chapter 4 Solving Product-Mix Problem with GLPK 37

13
14 171.2329
15 246.5753
16 0
17
18 z = 6 . 2 4 6 6 e+04
19 e = 0
20 xtra =
21
22 s c a l a r s t r u c t u r e containing the f i e l d s :
23
24 lambda =
25
26 0
27 0
28 0
29 0.1945
30 0.4110
31
32 redcosts =
33
34 0
35 0
36 −1.3699
37
38 time = 0
39 status = 5

Here the output lines 2–11 are there because we asked glpk to babble. If we would have let glpk
to use the presolver these lines would have been slightly different. In thelines 12–18 we have
the usual result. The interesting new information is in lines 20–36. There we see two vectors:
• xtra.lambda
• xtra. redcosts
The shadow prices λ=xtra.lambda are related to the constraints or to the resources b. The
shadow price associated with a resource tells you how much more profit you would get by
increasing the amount of that resource by one unit. So, since the shadow prices for fat, sugar
and protein are all zero, it means that Muad’Dib Bakery does not gain anything for additional
fat, sugar or proteig powder. Shadow prices for water and spice Melange are positive. For
sexampe spice Melange’s shadow price λ5 = 0.4110. This means that if someone is willing to
sell spice Melange to Muad’Dib Bakery with price less than 0.4110 sol/mg, Muad’Dib Bakery
should buy the powder. If the price is more than 0.4110 sol/mg, it does not make sense for
Muad’Dib bakery to buy more spice Melange.
The reduced costs u=xtra. redcosts are related to the decision variables, or to the objective
coefficients c. The reduced costs and the decision variables are complementary in the sense
that if xk > 0 then the associated reduced cost uk = 0, and vice versa. One way of interpreting
the reduced cost is to say that it is the amount of how much profit gained from decision variable
must increase so that it will become optimal to produce it. So, for Muad’Dib the reduced costs
for Atreides and Corrino are both 0, since it is optimal to make them. It is not optimal to
make Harkonnen cakes. The reduced cost u3 = −1.3699 tells us that the price of Harkonnen
cake must be increase to 100+1.37 = 101.37 before it is optimal to make them.