Matlab Ohio
Matlab Ohio
Matlab Ohio
8:27 a.m.)
A MATLAB Tutorial
Ed Overman
Department of Mathematics
The Ohio State University
Introduction
. . . . . . . . . . . . . . .
1 Scalar Calculations . . . . . . . . . . . .
. . . . . . .
1.1 Simple Arithmetical Operations
1.2 Variables . . . . . . . . . . . . . .
1.3 Round-off Errors . . . . . . . . . . . .
1.4 Formatting Printing . . . . . . . . . . .
1.5 Common Mathematical Functions
. . . . . .
. . . . . . . . . . .
1.6 Complex Numbers
1.7 Script M-files . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
1.8 Help!
1.9 Be Able To Do
. . . . . . . . . . . .
2 Arrays: Vector and Matrix Calculations . . . . .
2.1 Generating Matrices . . . . . . . . . . .
2.2 The Colon Operator . . . . . . . . . . .
2.3 Manipulating Matrices . . . . . . . . . .
2.4 Simple Arithmetical Operations
. . . . . . .
2.5 Operator Precedence
. . . . . . . . . .
2.6 Be Careful!
. . . . . . . . . . . . .
2.7 Common Mathematical Functions
. . . . . .
. . . . . . .
2.8 Data Manipulation Commands
2.9 Advanced Topic: Multidimensional Arrays . . . .
2.10 Be Able To Do
. . . . . . . . . . . .
3 Anonymous Functions, Strings, and Other Data Types
3.1 Anonymous Functions . . . . . . . . . .
3.2 Passing Functions as Arguments . . . . . . .
3.3 Strings . . . . . . . . . . . . . . .
3.4 Cell Arrays and Structures . . . . . . . . .
3.5 Advanced Topic: Data Types and Classes
. . . .
3.6 Be Able To Do
. . . . . . . . . . . .
4 Graphics
. . . . . . . . . . . . . . .
4.1 Two-Dimensional Graphics . . . . . . . . .
4.2 Three-Dimensional Graphics
. . . . . . . .
4.3 Advanced Topic: Commands . . . . . . . .
4.4 Advanced Topic: Handles and Properties
. . . .
.
4.5 Advanced Topic: GUIs (Graphical User Interfaces)
4.6 Advanced Topic: Making Movies . . . . . . .
4.7 Be Able To Do
. . . . . . . . . . . .
5 Solving Linear Systems of Equations . . . . . .
5.1 Square Linear Systems . . . . . . . . . .
5.2 Catastrophic Round-Off Errors
. . . . . . .
5.3 Overdetermined and Underdetermined Linear Systems
6 File Input-Output
. . . . . . . . . . . .
7 Some Useful Linear Algebra Functions
. . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
6
6
7
9
10
11
13
14
15
17
17
18
22
23
28
31
32
34
34
37
38
39
40
41
42
43
46
48
49
49
58
61
66
68
74
77
77
78
81
82
83
85
Programming in MATLAB . . . . . . . .
8.1 Flow Control and Logical Variables . . . . .
8.2 Matrix Relational Operators and Logical Operators
8.3 Function M-files . . . . . . . . . . .
8.4 Odds and Ends
. . . . . . . . . . .
8.5 Advanced Topic: Vectorizing Code
. . . . .
9 Sparse Matrices . . . . . . . . . . . .
10 Initial-Value Ordinary Differential Equations . .
10.1 Basic Commands . . . . . . . . . . .
10.2 Advanced Commands
. . . . . . . . .
11 Boundary-Value Ordinary Differential Equations
12 Polynomials and Polynomial Functions . . . .
. . . . .
13 Numerical Operations on Functions
14 Discrete Fourier Transform . . . . . . . .
15 Mathematical Functions Applied to Matrices . .
Appendix: Reference Tables . . . . . . . . .
Arithmetical Operators
. . . . . . . . . .
Special Characters . . . . . . . . . . . .
Getting Help
. . . . . . . . . . . . .
Predefined Variables
. . . . . . . . . . .
Format Options . . . . . . . . . . . . .
Some Common Mathematical Functions
. . . . .
Input-Output Functions
. . . . . . . . . .
Arithmetical Matrix Operations . . . . . . . .
Elementary Matrices
. . . . . . . . . . .
Specialized Matrices
. . . . . . . . . . .
Elementary Matrix Operations
. . . . . . . .
Manipulating Matrices . . . . . . . . . . .
Odds and Ends . . . . . . . . . . . . .
Two-Dimensional Graphics
. . . . . . . . .
Three-Dimensional Graphics . . . . . . . . .
Advanced Graphics Features . . . . . . . . .
String Functions, Cell Arrays, Structures, and Classes .
Data Manipulation Commands . . . . . . . .
Some Useful Functions in Linear Algebra . . . . .
Logical and Relational Operators
. . . . . . .
Flow Control
. . . . . . . . . . . . .
Logical Functions
. . . . . . . . . . . .
Programming Language Functions . . . . . . .
Debugging Commands . . . . . . . . . . .
Discrete Fourier Transform
. . . . . . . . .
Sparse Matrix Functions . . . . . . . . . .
Time Evolution ODE Solvers . . . . . . . . .
Boundary-Value Solver
. . . . . . . . . .
Numerical Operations on Functions . . . . . . .
Numerical Operations on Polynomials . . . . . .
Matrix Functions
. . . . . . . . . . . .
Solutions To Exercises . . . . . . . . . . .
ASCII Table
. . . . . . . . . . . . . .
Index . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
91
91
96
100
113
114
117
120
120
125
131
135
138
140
147
149
149
149
150
150
150
151
152
152
153
153
153
154
154
155
155
156
157
157
158
158
158
159
159
160
160
160
161
161
161
162
162
163
167
169
Introduction
MATLAB is an interactive software package which was developed to perform numerical calculations
on vectors and matrices. Initially, it was simply a MATrix LABoratory. However, today it is much more
powerful:
It can do quite sophisticated graphics in two and three dimensions.
It contains a high-level programming language (a baby C) which makes it quite easy to code complicated algorithms involving vectors and matrices.
It can numerically solve nonlinear initial-value ordinary differential equations.
It can numerically solve nonlinear boundary-value ordinary differential equations.
It contains a wide variety of toolboxes which allow it to perform a wide range of applications from science and engineering. Since users can write their own toolboxes, the breadth of applications is quite
amazing.
Mathematics is the basic building block of science and engineering, and MATLAB makes it easy to handle
many of the computations involved. You should not think of MATLAB as another complication programming language, but as a powerful calculator that gives you fingertip access to exploring interesting problems in science, engineering, and mathematics. And this access is available by using only a small number
of commands and function because MATLABs basic data element is a matrix (or an array).
This is a crucial feature of MATLAB it was designed to group large amounts of data in arrays and
to perform mathematical operations on this data as individual arrays rather than as groups of data. This
makes it very easy to apply complicated operations to the data, and it make it very difficult to do it
wrong. In high-level computer languages you would usually have to work on each piece of data separately
and use loops to cycle over all the pieces. In MATLAB this can frequently do complicated things in
one, or a few, statements (and no loops). In addition, in a high-level language many mathematical operations require the use of sophisticated software packages, which you have to find and, much worse, to
understand since the interfaces to these packages are frequently quite complicated and the documentation
must be read and mastered. In MATLAB, on the other hand, these operations have simple and consistent
interfaces which are quite easy to master. For an overview of the capabilities of MATLAB, type
>> demo
in the Help Navigator and click on MATLAB.
This tutorial is designed to be a concise introduction to many of the capabilities of MATLAB. It makes
no attempt to cover either the range of topics or the depth of detail that you can find in a reference manual, such as Mastering MATLAB 7 by Duane Hanselman and Bruce Littlefield (which is over 850 pages
long) or MATLAB Guide, 2nd edition by Desmond and Nicholas Higham (which is almost 400 pages long).
This tutorial was initially written to provide students with a free basic overview of commands which
are useful in an undergraduate course on linear algebra. Over the years it has grown to include courses in
ordinary differential equations, mathematical modelling, and numerical analysis. It also includes an introduction to two- and three-dimensional graphics because graphics is often the preferred way to present the
results of calculations.
In this tutorial MATLAB is first introduced as a calculator and then as a plotting package. Only afterwards are more technical topics discussed. We take this approach because most people are quite familiar
with calculators, and it is only a small step to understand how to apply these same techniques to matrices
rather than individual numbers or varibles. In addition, by viewing MATLAB as a simple but powerful
calculater, rather than as a complicated software package or computer language, you will be in the correct
frame of mind to use MATLAB.
You should view MATLAB as a tool that you are playing with trying ideas out and seeing how
There is a technical distinction between a command and a function in MATLAB: input arguments to commands are not enclosed in parentheses (they are separated by spaces) and there are no output arguments (i.e., a
command cannot be on the right-hand side of an equal sign). In reality, this is a very fine distinction since many
commands can be written as functions by putting the arguments between parentheses and separating them with
commas. We will generally use the terms interchangably.
they work. If an idea works, fine; if it doesnt, investigate further and figure out why. Maybe you misunderstood some MATLAB command, or maybe your idea needs some refinement. Play around interactively and figure it out. There are no hard and fast rules for figuring it out try things and see
what happens. Dont be afraid to make mistakes; MATLAB wont call you an idiot for making a mistake.
When you first learned to ride a bicycle, you fell down a lot and you looked pretty silly. But you kept
at it until you didnt fall down. You didnt study Newtons laws of motion and try to analyze the motion
of a bicycle; you didnt take classes in how to ride a bicycle; you didnt get videos from the library on how
to ride a bicycle. You just kept at it, possibly with the assistance of someone who steadied the bicycle and
gave you a little push to get you started. This is how you should learn MATLAB.
However, this tutorial is not designed for playing around. It is very ordered, because it has been designed as a brief introduction to all the basic topics that I consider important and then as a reference
manual. It would be very useful for students to have a document which uses this play around approach
so you would learn topics by using them in exploring some exercise. This is how workbooks should be
written: present some exercise for students to investigate, and let them investigate it themselves. And
these exercises should be interesting, having some connection to physical or mathematical models that
the students or at least a reasonable fraction thereof have some knowledge of and some interest in.
This tutorial is designed to be a reference manual that could be used alongside such a workbook if only
someone would write it.
Summary of Contents
We have tried to make this tutorial as linear as possible so that the building blocks necessary for a section are contained in preceding sections. This is not the best way to learn MATLAB, but it is a good way
to document it. In addition, we try to separate these building blocks and put them in short subsections so
that they are are easy to find and to understand. Next, we collect all the commands discussed in a subsection and put them in a box at the end along with a very brief discussion to make it easy to remember
these commands. Finally, we collect all these commands and put them in the appendix again boxed up by
topic. MATLAB has a
number of commands and functions and this is one way to collect them
for easy reference.
Warning: Usually we do not discuss the complete behavior of these commands, but only their most \useful" behavior. Typing
>> help <command>
or
>> doc <command>
gives you complete information about the command.
Notation: help <command> means to enter whatever command you desire (without the braces).
help command means to type these two words as written.
Section 1 of this tutorial discusses how to use MATLAB as a scalar calculator, and Section 2 how to
use it as a matrix calculator. Following this, you will be able to set up and solve the matrix equation
Ax = b where A is a square nonsingular matrix.
Section 4 discusses how to plot curves in two and three dimensions and how to plot surfaces in three dimensions. These three sections provide a basic introduction to MATLAB. At the end of each of these
three sections there is a subsection entitled Be Able To Do which contains sample exercises to make sure
you understand the basic commands discussed. (Solutions are included.)
You have hopefully noticed that we skipped section 3. It discusses a number of minor topics. Since they
are useful in generating two- and three-dimensional plots, we have included it here.
The following sections delve more deeply into particular topics. Section 5 discusses how to find any and
all solutions of Ax = b where A Cmn need not be a square matrix; there might be no solutions, one
solution, or an infinite number to this linear system. When no solution exists, it discusses how to calculate a least-squares solution (i.e., the best approximation to a solution). In addition, it discusses how
round-off errors can corrupt the solution, and how to determine if this is likely to occur.
Section 6 is quite brief and discusses advanced commands to input data into MATLAB and output it
to a file. (The basic commands are discussed in Section 4.1.) This is useful if the data is being shared
HUGE
1. Scalar Calculations
1.1.
MATLAB can be used as a scientific calculator. To begin a MATLAB session, click on a MATLAB
icon or type matlab in a terminal and wait for the prompt, i.e., >> , to appear. (To exit MATLAB,
click on Exit MATLAB in the File menu item or type exit or quit.) You are now in the MATLAB
workspace.
You can calculate 3.17 5.7 + 17/3 by entering
>> 3.17*5.7 + 17/3
and 220 by entering
>> 2 20
P12
And
j=1 1/j can be entered as
>> 1 + 1/2 + 1/3 + 1/4 + 1/5 + 1/6 + 1/7 + 1/8 + 1/9 + 1/10 + 1/11 + 1/12
You can enter a number in scientific notation using the operator. For example, you can enter
21020 by
>> 2*10 -20
MATLAB, however, uses e to represent 10 so that MATLAB displays
2.0000e-20
The standard way to input 21020 is as 2e-20 or 2E-20 or 2.e-20 or 2.E-20 (even
2.0000000e-00020 is acceptable).
Warning: 1020 cannot be input as e-20, but must be input as 1e-20 or 1E-20 or 1.e-20 or 1.E-20
or . . . .
MATLAB can also handle complex numbers, where i or j represents 1 . For example, 5i can be
input as 5i or as 5*i, while 51030 i can be input as 5e30i or as 5e30*i or as 5*10 30*i, but not
as 5*10 30i (which MATLAB considers to be 51030i ). To calculate (2 + 2i)4 , enter
>> (2 + 2i) 4
and MATLAB returns 64.
You can also save all of your input to MATLAB and most of the output (plots are not saved) by using
the diary command. This archive of your work can be invaluable when you are solving homework problems. You can later use an editor to extract the part you want to turn in, while burying all the false
starts and typing mistakes that occur. Conversely, if you are involved in a continuing project, this archive
can be invaluable in keeping a record of your progress.
If you do not specify a file, this archive is saved to the file diary (no extension) in the present directory. If the file already exists, this is appended to the end of the file (i.e., the file is not overwritten).
Because of this feature you can use the diary command without fear that crucial work will be overwritten.
If you are entering a line and make a mistake, there are a number of ways you can correct your error:
you can use the backspace or delete key to erase all the text back to your mistake,
you can use the left-arrow key, i.e., , and the right-arrow key, i.e., , to move back and forth
in the line, or
you can use the mouse to move back and forth in the line.
Frequently, you will want to reexecute the previous line, or another previous line. For example, you
might have made a mistake in the previous line and so it did not execute, or did not execute correctly.
Of course, you can just retype the line but, if it is very long, this can get very time-consuming. Instead,
you can use the up-arrow key, i.e., , to move backward, one statement at a time (or the down-arrow
key, i.e., to move forward). Then hit the enter (or the return) key to execute the line.
1.2. Variables
Arithmetical Operations
a+b
a-b
a*b
Addition.
Subtraction.
Multiplication.
a/b
a\b
a b
diary
1.2.
Division.
Left division, (this is exactly the same as
b/a ).
Exponentiation (i.e., ab ).
Saves your input to MATLAB and most of the output to disk. This command toggles
diary on and off. (If no file is given, it is saved to the file diary in the current directory.)
diary on turns the diary on.
diary off turns the diary off.
diary 0 <file name> 0 saves to the named file.
The up-arrow key moves backward in the MATLAB workspace, one line at a time.
Variables
1.2. Variables
x =
-23.6000
It is sometimes useful to display the value of a variable or an expression or a character string without
displaying the name of the variable or ans. This is done by using disp. For example,
>> disp(x)
>> disp(pi 3)
>> disp( 0 And now for something completely different 0 )
>> disp( 0 ------------------------------------------ 0 )
displays
-23.6000
31.0063
And now for something completely different
-----------------------------------------(The command fprintf, which will be discussed in Section 6, allows much finer formatting of variables.)
Note: When disp displays a variable or an array or an expression, it follows with a blank line. However,
when it displays a string or a string variable, it does not.
Incidentally, a valid name for a MATLAB variable is a character string containing letters (upper or
lower case), digits, and underscores where the first character must be a letter. The maximum length of a
name is too long to worry about. However, there are a few names which are reserved because they have
special meanings. The reserved words, called keywords, are
break
case
catch
classdef
continue
else
elseif
end
for
function
global
if
otherwise
parfor
persistent
return
(Of course, you can still use End or END but you probably shouldnt.)
Variables can also be deleted by using clear. For example, to delete x type
>> clear x
Warning: This is a very dangerous command because it is so easy to lose
clear x
>>
clear
spmd
switch
try
while
you will delete all the variables you have created in the workspace!
Predefined Variables
ans
pi
eps
Inf
NaN
i
j
realmin
realmax
The default variable name when one has not been specified.
.
Approximately the smallest positive real number on the computer such that
1 + eps 6= 1 .
(as in 1/0 ). You can also type inf.
Not-a-Number (as in 0/0 ). You can also type nan.
1 .
1 (the same as i because engineers often use these interchangeably).
The smallest usable positive real number on the computer. This is approximately
the smallest positive real number that can be represented on the computer (on some
computer realmin/2 returns 0 ).
The largest usable positive real number on the computer. This is approximately the
largest positive real number that can be represented on the computer (on most computer
2*realmax returns Inf ).
About Variables
are case sensitive (so xa is not the same as Xa ).
can contain up many, many characters (but this is certainly overkill).
must start with a letter, and can then be followed by any number of letters, numbers,
and/or underscores (so z 0
is allowed).
do not need to be declared or typed.
To display a variable, type it alone on a line.
To delete a variable, type clear <variable>.
Variables:
disp(X)
,
;
1.3.
Displays a variable (including an array) or a string without printing the variable name or
ans.
Separates multiple statements on the same line. The results appear on the screen.
When this ends a MATLAB command, the result is not printed on the screen. This can
also separate multiple statements on the same line.
Round-off Errors
The most important principle for you to understand about computers is the following.
Principle 1.1.
Computers do integer arithmetic correctly (as long as the numbers are not too large to be stored in the
computer). However, computers cannot store most floating-point numbers (i.e., real numbers) correctly.
For example, the fraction 1/3 is equal to the real number 0.3333 . . . Since a computer cannot store this
infinite sequence of threes, the number has to be truncated.
eps is close to the difference between the exact number 1/3 and the approximation to 1/3 used in
MATLAB. 1 + eps is the smallest floating-point number after 1 which can be stored precisely in the
computer. For example, in MATLAB 1 + 0.1 is clearly greater than 1; however, on our computer
1 + 1e-40 is not. To see this, when we enter
>> (1 + .1) - 1
we obtain 0.1000 as expected.
Note: MATLAB guarantees that the expression in parentheses is evaluated first, and then 1 is subtracted
from the result.
9
Request Input
input( 0 <prompt> 0 )
1.4.
Displays the prompt on the screen and waits for you to input whatever is desired. The optional second argument of 0 s 0 allows you to enter a string
(including spaces) without using quote marks.
Formatting Printing
3
The reason that (n1/3 ) n can be nonzero numerically is that MATLAB only stores real numbers to a
certain number of digits of accuracy. Type
>> log10(1/eps)
and remember the integer part of this number. This is approximately the maximum number of digits of accuracy of any calculation performed in MATLAB. For example, if you type 1/3 in MATLAB the
result is only accurate to approximately this number of digits. You do not see the decimal representation
of 1/3 to this number of digits because on start-up MATLAB only prints the result to four decimal digits
or five significant digits if scientific notation is used (e.g., the calculation 1/30000 is displayed in scientific notation). To change how the results are printed out, use the format command in MATLAB. Use
10
each of these six format functions and then type in 1/3 to see how the result is printed out.
Format Options
format short
format long
format short e
format long e
format short g
format long g
1.5.
MATLAB contains a large number of mathematical functions. Most are entered exactly as you would
write them mathematically. For example,
>> sin(3)
>> exp(2)
>> log(10)
return exactly what you would expect. As is common in programming languages, the trig functions are
evaluated in radians. However, there are corresponding functions which are evaluated in degrees.
Almost all the functions shown here are built-in functions. That is, they are coded in C so they execute
very quickly. The one exception is the factorial function, i.e., n! = 1 2 3 n, which is calculated by
>> factorial(n)
Note: This function is actually calculated by generating the vector (1, 2, . . . , n) and them multiplying all
its elements together by prod([1:n]). (We discuss the colon operator in Section 2.2.)
There is an important principle to remember about computer arithmetic in MATLAB.
If all the numbers you enter into MATLAB to do some calculation are \reasonably large" and the result of this calculation is one or more numbers which are \close to"
eps , it is very likely that the number or numbers should be zero.
Principle 1.2.
As an example, enter
>> deg = pi/180; th = 40; 1 - ( cos(th*deg) 2 + sin(th*deg) 2 )
The result is 1.1102e-16. Clearly, all the numbers entered into this calculation are reasonable and the
result is approximately eps. Obviously, the result is supposed to be zero since, from the Pythagorean
theorem
cos2 + sin2 = 1
for all angles . MATLAB tries to calculate the correct result, but it cannot quite. It is up to you to
interpret what MATLAB is trying to tell you.
Note: If you obtained zero for the above calculation, try
>> th = input( 0 angle = 0 ); 1 - ( cosd(th) 2 + sind(th) 2 )
for various angles. Some of these calculations should be nonzero.
There are a number of occasions in this overview where we reiterate that MATLAB cannot usually calculate results exactly. Sometimes these errors are small and unimportant other times they are very
important. In fact, MATLAB has introduced two functions to reduce round-off errors. The relative error
in the calculation of ex 1, i.e., (exp(x) - 1)/exp(x) can be very large when x 1 since
ex 1 =
1+
x
x2
x3
x
x2
x3
+
+
+ 1 =
+
+
+ ;
1!
2!
3!
1!
2!
3!
11
the term within parentheses is very close to 1 and so subtracting by 1 causes a loss of many digits in the
result. For example,
>> exp(1.e-8) - 1 = 9.999999939225290e-09
>> expm1(1.e-8) = 1.000000005000000e-08
>> exp(1.e-20) - 1 = 0
>> expm1(1.e-20) = 9.999999999999999e-21
Similarly,
log z =
z 1 (z 1)2
(z 1)3
(z 1)4
+
1
2
3
4
so if z 1 then accuracy is lost in the calculation of z 1. This can be avoided by entering x directly
in
x3
x4
x x2
+
+ ,
log(1 + x) =
1
2
3
4
which is evaluated by log1p(x).
Warning: There is one technical detail about functions that will trip you up occasionally: how does MATLAB determine whether a word you enter is a variable or a function? The answer is that MATLAB first checks if the word is a variable and, only if it fails, does it check if the word is a
function. For example, suppose you enter
>> sin = 20
by mistake (possibly you meant bin = 20 but were thinking about something else). If you now
type
>> sin(3)
MATLAB will reply
??? Index exceeds matrix dimensions.
because it recognizes that sin is a variable. Since MATLAB considers a variable to be a vector
of length one, its complaint is that you are asking for the value of the third element of the vector sin (which only has one element). Similarly, if you enter
>> sin(.25*pi)
MATLAB will reply
Warning: Subscript indices must be integer values.
because it thinks you are asking for the .25-th element of the vector sin. The way to undo
your mistake is by typing
>> clear sin
12
1.6.
exp(x)
expm1(x)
factorial(n)
fix(x)
floor(x)
heaviside(x)
log(x)
log10(x)
log1p(x)
mod(x, y)
rem(x, y)
round(x)
sec(x)
secd(x)
sech(x)
sign(x)
sin(x)
sind(x)
sinh(x)
sqrt(x)
tan(x)
tand(x)
tanh(x)
ex .
ex 1.
n! for n a non-negative
integer.
If x 0 this is the largest integer which is x.
If x < 0 this is the smallest
integer which is x.
This is the largest integer
which is x.
If x > 0 this returns 1,
if x < 0 this returns 0, and
if x = 0 this returns 1/2 .
The natural log of x, i.e.,
loge x.
The common log of x, i.e.,
log10 x.
log(x + 1).
The modulus after division.
That is, x n y where
n = floor(x/y).
The remainder of x/y. This
is the same as mod(x, y) if
x, y > 0.
Warning: be careful if x < 0.
The integer which is closest to
x.
sec x.
sec x where x is in degrees.
sech x.
If x > 0 this returns +1,
if x < 0 this returns 1, and
if x = 0 this returns 0.
sin x.
sin x where x is in degrees.
sinh
x.
x.
tan x.
tan x where x is in degrees.
tanh x.
Complex Numbers
MATLAB can work with complex numbers as easily as with real numbers. For example, to find the
roots of the quadratic polynomial x2 + 2x + 5 enter
>> a = 1; b = 2; c = 5;
>> x1 = ( -b + sqrt( b 2 - 4*a*c ) ) / (2*a)
>> x2 = ( -b - sqrt( b 2 - 4*a*c ) ) / (2*a)
The output is
-1.0000 + 2.0000i
This function is in the symbolic math toolbox. If it is not on your computer, the code is shown on page 104.
13
and
-1.0000 - 2.0000i
As another example, to calculate ei/2 enter
>> exp(1i*pi/2)
and obtain
0.0000 + 1.0000i
There are standard commands for obtaining the real part, the imaginary part, and the complex conjugate of a complex number or variable. For example,
>> x = 3 - 5i
>> real(x)
>> imag(x)
>> conj(x)
returns 3, -5, and 3.0000 + 5.0000i respectively.
Note that many of the common mathematical functions can take complex arguments. Above, MATLAB
has calculated ei/2 , which is evaluated using the formula
ez = ex+iy = ex (cos y + i sin y) .
Similarly,
cos z =
eiz + eiz
2
and
sin z =
eiz eiz
.
2i
1.7.
conj(z)
imag(z)
real(z)
z = x iy.
The imaginary part of z, i.e., y.
The real part of z, i.e., x.
Script M-files
So far we have always entered MATLAB statements directly into the text window so that they are executed immediately. However, if we want to repeatedly execute a number of statements we have to put
them all on one line and reexecute the whole line. This line can get very l o o o n n n g! The solution is to
type the sequence of statements in a separate file named <file name>.m. It is easy to edit this file to
remove any errors, and the sequence can be executed whenever desired by typing
>> <file name>
The MATLAB statements themselves are not printed out, but the result of each statement is, unless
a semicolon ends it. This type of file is called a script m-file: when MATLAB executes the command
<file name> the contents of the file <file name>.m are executed just as if you had typed them into
into the text window. We will not emphasize script m-files further, but you will find many occasions where
they are very helpful.
You can easily work on a script m-file by clicking on the menu item File and then on New to create a
new m-file and finally, if it asks, click on Script. Or click on File and then Open to open an already existing one (if you want to modify it). You can also create a new m-file or open an already existing one
by
>> edit <file name>.m
Warning: The name of the file includes the extension .m, i.e., <file name>.m, but you execute it in
MATLAB by typing <file name>, i.e., without the extension.
Warning: The file name can consist of (almost any number of) letters (lowercase and/or uppercase), numbers, and underscores, i.e. .
If a is a complex number, then its complex conjugate, denoted by a is obtained by changing the sign of i
whenever it appears in the expression for a. For example, if a = 3 + 17i, then a = 3 17i; if a = ei/4 , then
a = ei/4 ; if a = (2 + 3i) sin(1 + 3i)/(3 5 i), then a = (2 3i) sin(1 3i)/(3 + 5 i).
14
1.8. Help!
the same name as yours! (The m-file it executes depends on the order in which directories are
searched for m-files see path for more details.) To check this, you can enter
>> type <file name>
before you save your m-file. This will type out the entire file if it is written in MATLAB or
type out
<file name> is a built-in function
if it is written in C or Fortran and so cannot be viewed directly. If the file name doesnt exist,
MATLAB returns
??? Undefined function or variable 0 <file name> 0 .
If MATLAB returns your m-file, it means you have already saved it. In this case enter
>> doc <file name>
(which is discussed in the next section), which returns useful information about a MATLAB
function, i.e., not one of yours. If it cannot find this particular function, you are safe.
A long expression can be continued to a new line by typing three periods followed by the enter (or
P20
return) key. For example,
j=1 1/j can be entered as
>> 1 + 1/2 + 1/3 + 1/4 + 1/5 + 1/6 + 1/7 + 1/8 + 1/9 + 1/10 + 1/11 + 1/12 + ...
1/13 + 1/14 + 1/15 + 1/16 + 1/17 + 1/18 + 1/19 + 1/20
although there are much better ways to obtain this same expression with many fewer keystrokes (as you
will see in Section 2.8). Lines can also be continued in the MATLAB workspace by using three periods,
but it is much more common to use continuation in an m-file.
If your m-file is very long, it is often valuable to include comments to explain what you are doing. Each
line of comments must begin with the percent character, i.e., %. Comments can appear alone on a line
or they can follow a statement that you have entered.
Odds and Ends
edit
type
...
%
1.8.
Help!
Before discussing how to obtain help in MATLAB, here is a good place to discuss a very frustrating
situation where you desperately need help: how do you abort a MATLAB statement which is presently
executing. The answer is simply to type C (that is, hold down the control key and type c).
The on-line help facility in MATLAB is quite extensive. If you type
>> help
you will get a list of all the topics that you can peruse further by typing help followed by the name of
the topic. If you want help on a specific command, simply type help followed by the name of the command, i.e.,
>> help <command>
For example, if you forget the exact form of the format command, just type
>> help format
and you will see all the various ways that the output can be formatted.
Note: Typing
>> help ?
gives you lots of information about arithmetical and relational and logical operators and special
15
1.8. Help!
characters.
There is a more general command that can help you determine which commands might be of use. The
command lookfor searches through the first line of all MATLAB help entries for a particular string. It is
case insensitive so capital letters need not be used. For example,
>> lookfor plot
returns all the MATLAB commands that have something to do with plots. (There are over one hundred.)
This command may be useful or it may not be. However, it is worth a try if you cannot remember the
name of the command you want to use.
Warning: All of the thousands of MATLAB commands have to be checked, so this command might run
slowly.
Note: The string need not be a complete word. For example, the string compl is contained in the words
complement, complex, complete, completion, and incomplete and in the capitals of all
these words.
If you want to find out more about a specific command, enter
>> type <command>
or
>> edit <command>
If the command is written in MATLABs programming language (as discussed in Section 8), the entire
function will be typed out for the former and appear in a new window in the latter. These commands cannot return anything useful on internal MATLAB commands, called built-in function, which are coded in
C.)
MATLAB also has an entire reference manual on-line which can be accessed by entering
>> doc
or
>> helpbrowser
This hypertext documentation is displayed using your Web browser. It generally gives much more information than the help command, and in a more easily understood format.
After working for a while, you may well forget what variables you have defined in the workspace. Simply type who or whos to get a list of all your variables (but not their values). who simply returns the
names of the variables you have defined, while whos also returns the size and type of each variable. To
see what a variable contains, simply type the name of the variable on a line.
By the way, the demonstrations available by running demo show many of the capabilities of MATLAB
and include the actual code used. This is always a good place to look if you are not sure how do do something.
Two commands that dont quite fit in any category are save and load. However, since these commands are occasionally very helpful, this is a good place to discuss them. Occasionally, you might need to
save one or more MATLAB variables: it might have taken you some time to generate these variables and
you might have to quit your MATLAB session without finishing your work or you just might be afraid
that you will overwrite some of them by mistake. The save command saves the contents of all your variables to the file matlab.mat. Use doc to learn how to save all the variables to a file of your own choice
and how to save just some of the variables. The load command loads all the saved variables back into
your MATLAB session. (As we discuss in Section 4.1, the load command can also be used to input our
own data into MATLAB.)
These variables are saved in binary format; when loaded back in using load the variables will be exactly the
same as before. The contents of this file can be viewed by the user with an editor but the contents will appear
to be gibberish. The contents can only be interpreted by the load command.
16
Getting Help
help
doc
helpbrowser
type <command>
lookfor <keyword>
who
whos
demo
save
load
C
1.9.
On-line help.
help lists all the primary help topics.
help <command> displays information about the command.
On-line help hypertext reference manual.
doc accesses the manual.
doc <command> displays information about the command.
Accesses the main page of the on-line reference manual.
Displays the actual MATLAB code for this command.
Searches all MATLAB commands for this keyword.
Lists all the current variables.
Lists all the current variables in more detail than who.
Runs demonstrations of many of the capabilities of MATLAB.
Saves all of your variables.
Loads back all of the variables which have been saved previously.
Abort the command which is currently executing (i.e., hold down the control
key and type c).
Be Able To Do
After reading this section you should be able to do the following exercises. The solutions are given on
page 163.
1. Consider a triangle with sides a, b, and c and corresponding angles ab, ac, and bc.
(a) Use the law of cosines, i.e.,
c2 = a2 + b2 2ab cos ab ,
to calculate c if a = 3.7, b = 5.7, and ab = 79 .
(b) Then show c to its full accuracy.
(c) Use the law of sines, i.e.,
sin ab
sin ac
=
,
c
b
to calculate ac in degrees and show it in scientific notation.
(d) What MATLAB command should you have used first if you wanted to save these results to the file
triangle.ans?
ters.
This makes it much easier to understand MATLAB operations. This is also a good practice
17
2.1.
Generating Matrices
1
A = 4
7
2
5
8
3
6
9
in MATLAB type
>> A = [1 2 3; 4 5 6; 7 8 9]
(where denotes one or more spaces) or
>> A = [ 1 2 3 ; 4 5 6 ; 7 8 9]
or
>> A = [1,2,3; 4,5,6; 7,8,9]
or
>> A = [ 1 , 2 , 3 ; 4 , 5 , 6 ; 7 , 8 , 9 ]
In other words, either spaces or commas can be used to delineate the elements of each row of a matrix;
semicolons are required to separate rows. (Any number of spaces can be put around commas or semicolons
to improve the readability of the expression.)
Notation: Since we prefer spaces, we will generally use them rather than commas to separate elements in a
row.
Rows can also be separated by beginning each on a separate line. For example, the matrix A can also
be entered by
>> A = [1,2,3
4,5,6
7,8,9]
However, we do not use it because there is no way to correct an element on a previous line which you have
just noticed has been entered incorrectly. The more complicated matrix
3 sin 1
1 2+ 3
17/3
+3
C = e2
1/3 2 3 7 cos /7
18
We have now discussed how to enter matrices into MATLAB by using square parentheses, i.e., [...].
You work with individual elements of a matrix by using round parentheses, i.e., (...). For example, the
element aij of the matrix A is A(i,j) in MATLAB. Suppose you want to create the matrix
1
B = 4
7
2
5
8
3
6
10
without having to enter all nine elements. If A (see the beginning of this section) has already been generated, the simplest way is to type
>> B = A; B(3,3) = 10
Also, the element xi of the vector x is x(i) in MATLAB. For example, to create the column vector
x = (1, 2, 3, . . . , 47, 48, 49, 51)T R50
enter
>> x = [1:50] 0 ; x(50) = 51
or
>> x = [1:50] 0 ; x(50) = x(50) + 1
or
>> x = [1:50] 0 ; x(length(x)) = x(length(x)) + 1
where length returns the number of elements in a vector.
MATLAB also has a number of commands that can generate matrices. For example,
>> C = zeros(5)
or
>> C = zeros(5, 5)
generates a 55 zero matrix. Also,
>> C = zeros(5, 8)
generates a 58 zero matrix. Finally, you can generate a zero matrix C with the same size as an already
existing matrix, such as A, by
>> C = zeros(size(A))
where size(A) is a row vector consisting of the number of rows and columns of A. This command is frequently used to preallocate a matrix of a given size so that MATLAB does not need to guess how large
to make it.
Similarly, you can generate a matrix with all ones by ones(n) or ones(m, n) or ones(size(D)).
You can also generate the identity matrix, i.e., the matrix with ones on the main diagonal and zeroes off of
it, by using the command eye with the same arguments as above.
Another useful matrix is a random matrix, that is, a matrix whose elements are all random numbers.
The two most commonly used random numbers are uniformly distributed random numbers and normally
distributed random numbers. Uniformly distributed random numbers in [0, 1) are generated by the rand
function, which takes the same arguments as above, i.e.,
>> r = rand % or rand(1)
>> R = rand(n) % or rand(m,n)
To generate uniformly distributed random numbers in [a, b) use
>> R = a + (b - a)*rand(m, n)
To be precise, these are pseudorandom numbers because they are calculated by a deterministic formula
which begins with an initial seed which is called the state , not the seed. Every time that a new
MATLAB session is started, the default seed is set, and so the same sequence of random numbers will be
generated. However, every time that this command is executed during a session, a different sequence of
random numbers is generated. If desired, a different seed can be set at any time by entering
>> rng(<non-negative integer state number>)
To use a different sequence of random numbers every time you run MATLAB, begin your session with
>> rng( 0 shuffle 0 )
To general normally distributed random numbers with mean 0 and standard deviation 1 use
>> R = randn(m, n)
20
21
Elementary Matrices
zeros(n)
zeros(m, n)
zeros(size(A))
ones
eye
rand
randi
randn
rng
randperm(n)
size(A)
length(x)
numel(A)
A. 0
2.2.
Transpose, i.e., AT .
2.3.
Manipulating Matrices
For specificity in this subsection we will mainly work with the 56 matrix
1
7
E = 13
19
25
2
8
14
20
26
3
9
15
21
27
4 5 6
10 11 12
16 17 18 ,
22 23 24
28 29 30
2 3 4 5
F = 14 15 16 17 .
26 27 28 29
You can generate this submatrix more easily by typing
>> F = E( 1:2:5 , 2:5 )
There is an additional shortcut you can use: in a matrix a colon by itself represents an entire row or
column. For example, the second column of F is F(:,2) and the second row is F(2,:). To replace the
second column of F by two times the present second column minus four times the fourth column enter
>> F(:,2) = 2*F(:,2) - 4*F(:,4)
And suppose you now want to double all the elements in the last two columns of F. Simply type
>> F(:,3:4) = 2*F(:,3:4)
There is a last-additional shortcut you can use. Suppose you want the matrix F to consist of the odd
rows of E and the second to the last column, as we did above. You might be changing the size of E and
not want to have to remember how large it is. This can be easily done by
>> F = E( [1:2:end] , [2:end] )
The keyword end designates the last element of the dimension: 5 for the rows of E and 6 for the
columns.
Note: The keyword end has a number of meanings. It also ends a block of code begun with a if, for,
while, or switch (as we will see later). Finally, it can also terminate a primary function or a
This possiblity is much more real in the programming language C. For example, the statement
for ( i = 0.02; i <= 0.98; i = i + .001 )
generates successive values of i by adding 0.001 to the preceding value. It is possible that when i should have
the value 0.98, due to round-off errors the value will be slightly larger; the condition i <= 0.98 will be false
and the loop will not be evaluated when i should be 0.98.
23
subfunction, and it must terminate a nested function (as we will also see later).
Returning to :, entering E(:,:) prints out exactly the same matrix as entering E (as does
E(1:end,1:end) ). This is not a very useful way of entering E, but it shows how the colon operator
can work. On the other hand, entering
>> G = E( : , 6:-1:1 )
generates a matrix with the same size as E but with the columns reversed, i.e.,
6
12
G = 18
24
30
5
11
17
23
29
4
10
16
22
28
3
9
15
21
27
2
8
14
20
26
1
7
13 .
19
25
It is also very easy to switch rows in a matrix. For example, to switch the third and fifth rows of G, enter
the single line
>> G([5 3],:) = G([3 5],:)
which is much simpler, and more transparent, than the three lines
>> temp = G(3,:)
>> G(3,:) = G(5,:)
>> G(5,:) = temp
Note: There is a more general function which can reverse two variables: scalars or vector or matrices. For
example,
>> [y, x] = deal(x, y);
% or [Y, X] = deal(X, Y);
reverses the values in these two variables. They can even have different sizes!
(This is a very specialized function, but it is annoying to need three statements to reverse two variables.)
Finally, there is one more use of a colon. Entering
>> f = E(:)
generates a column vector consisting of the columns of E (i.e., the first five elements of f are the first
column of E, the next five elements of f are the second column of E, etc.).
Note: On the right side of an equation, E(:) is a column vector with the elements being the columns of
E in order. On the left side of an equation, E(:) reshapes a matrix. However, we will not discuss
this reshaping further because the reshape command described below is easier to understand.
The colon operator works on rows and/or columns of a matrix. A different command is needed to work
on the diagonals of a matrix. For example, you extract the main diagonal of E by typing
>> d = diag(E)
(so d is the column vector (1, 8, 15, 22, 29)T ), one above the main diagonal by typing
>> d1 = diag(E, 1)
(so d1 is the column vector (2, 9, 16, 23, 30)T ), and two below the main diagonal by typing
>> d2 = diag(E, -2)
(so d2 is the column vector (13, 20, 27)T ).
The MATLAB function diag transforms a matrix (i.e., a non-vector) into a column vector. The converse also holds: when diag is applied to a vector, it generates a symmetric matrix. The command
>> F = diag(d)
generates a 55 matrix whose main diagonal elements are the elements of d, i.e., 1, 8, 15, 22, 29, and
whose off-diagonal elements are zero. Similarly, entering
>> F1 = diag(d1, 1)
generates a 66 matrix whose first diagonal elements (i.e., one above the main diagonal) are the elements
of d1, i.e., 2, 9, 16, 23, 30, and whose other elements are zero, that is,
0
0
0
F1 =
0
0
0
2
0
0
0
0
0
0
9
0
0
0
0
24
0
0
16
0
0
0
0
0
0
23
0
0
0
0
0
.
0
30
0
Finally, typing
>> F2 = diag(d2, -2)
generates a 55 matrix whose 2nd diagonal elements (i.e., two below the main diagonal) are the elements of d2, i.e., 13, 20, 27, and whose other elements are zero, i.e.,
0
0
F2 = 13
0
0
0
0
0
20
0
0 0
0 0
0 0
0 0
27 0
0
0
0 .
0
0
The Toeplitz matrix is a very special matrix whose values are constant along each diagonal. For example,
7
4
3 1
4 3
2 7
5 2 7 4
1 5 2 7
is generated by
toeplitz([7 -2 -5 1], [7 4 3 1])
If the Topelitz matrix is symmetric, only the row or the column elements need be entered (i.e., only one
argument is required).
You can also extract the upper triangular or the lower triangular part of a matrix. For example,
>> G1 = triu(E)
constructs a matrix which is the same size as E and which contains the same elements as E on and above
the main diagonal; the other elements of G1 are zero. This command can also be applied to any of the
diagonals of a matrix. For example,
>> G2 = triu(E, 1)
constructs a matrix which is the same size as E and which contains the same elements as E on and above
the first diagonal, i.e.,
0 2 3 4 5 6
0 0 9 10 11 12
G2 = 0 0 0 16 17 18 .
0 0 0 0 23 24
0 0 0 0 0 30
The similar command tril extracts the lower triangular part of a matrix.
As an example of the relationship between these three commands, consider the square random matrix F
generated by
>> F = rand(6)
All the following MATLAB commands calculate F anew:
>> triu(F) + tril(F) - diag(diag(F))
>> triu(F, 1) + diag(diag(F)) + tril(F, -1)
>> triu(F) + tril(F, -1)
>> triu(F, 2) + diag(diag(F, 1), 1) + tril(F)
Note: Numerically the first command might not generate exactly the same matrix as the following three
because of round-off errors.
It is important to note that diag , triu and tril cannot appear on the left-hand side of an
equation. Instead, to zero out all the diagonals above the main diagonal of F enter
>> F = F - triu(F, 1)
and to zero out just the first diagonal above the main diagonal enter
>> F = F - tril(triu(F, 1), 1)
What if you want to insert numbers from the upper right-hand corner of a matrix to the lower left-hand
corner? There is no explicit function which does this but there are a number of indirect functions:
fliplr(A) flips the matrix from left to right, i.e., reverses the columns of the matrix;
flipud(A) flips the matrix up and down, i.e., reverses the rows of the matrix;
25
6 4 2
12 10 8
G = 18 16 14 .
24 22 20
30 28 26
Also, occasionally, there is a need to replicate or tile a matrix to form a larger matrix. The command
>> B = repmat(A, m, n)
generates a matrix B which contains m rows and n columns of copies of A. (If n = m then repmat(A,
m) is sufficient.) If A is a p by q matrix, then B Rmpnq . This even works if A is a scalar, in which
case this is the same as
>> B = A*ones(m, n)
(but it is much faster if m and n are large since no multiplication is involved).
One frequent use of repmat is when a specific operation is to be applied to each row or to each column of a matrix. For example, suppose that the column vectors { x1 , x2 , . . . , xn } have been combined
into the matrix X and we want to calculate the corresponding matrix for the vectors yj = xj + a for all
j N[1, n]. This can be easily done by
>> Y = X + repmat(a, 1, n);
which, unfortunately, requires that the new matrix A = repmat(a, 1, n) be created. We would prefer to
simply enter
>> Y = X + a
(WRONG);
However, we can enter
26
Plus
Minus
Array multiply
Array right division
Array left division
Array power
Binary maximum
Binary minimum
Remainder
Modulus
Arctangent
@atan2d
@hypot
@eq
@ne
@lt
@le
@gt
@ge
@and
@or
@xor
Arctangent (degrees)
Hypotenuse
Equal
Not equal
Less than
Less than or equal to
Greater than
Greater than or equal to
Logical AND
Logical OR
Logical exclusive OR
Of course, it is possible to write your own function which inputs either two column vectors of the same
size or one column vector and one scalar; it then outputs a column vector of the same size as the input.
27
Manipulating Matrices
A(i,j)
A(:,j)
A(i,:)
A(:,:)
A(?1,?2)
A(:)
o
diag(A)
diag(A, k)
o
diag(d)
diag(d, k)
o
triu(A)
triu(A, k)
o
tril(A)
tril(A, k)
fliplr(A)
flipud(A)
o
rot90(A)
rot90(A, k)
repmat(A, m, n)
bsxfun(<fnc>, A, b)
reshape(A, m, n)
[]
toeplitz(c,r)
2.4.
ai,j .
the j th column of A.
the ith row of A.
A itself.
There are many more choices than we care to describe:
?1 can be i or i1:i2 or i1:i3:i2 or : or [i1 i2 ... ir] and
?2 can be j or j1:j2 or j1:j3:j2 or : or [j1 j2 ... jr].
On the right-hand side of an equation, this is a column vector containing
the columns of A one after the other.
A column vector of the k th diagonal of the matrix (i.e., non-vector) A. If k
is not given, then k = 0.
A square matrix with the k th diagonal being the vector d. If k is not
given, then k = 0.
A matrix which is the same size as A and consists of the elements on and
above the k th diagonal of A. If k is not given, then k = 0.
The same as the command triu except it uses the elements on and below
the k th diagonalof A. If k is not given, then k = 0.
Flips a matrix left to right.
Flips a matrix up and down.
Rotates a matrix k90 . If k is not given, then k = 1.
Generates a matrix with m rows and n columns of copies of A. (If n = m
the third argument is not needed.)
Perform the operation given by the function handle on all the columns of
the matrix A using the column vector b or on all the rows using the row
vector b
Generates an mn matrix whose elements are taken columnwise from A.
Note: The number of elements in A must be mn.
The null matrix. This is also useful for deleting elements of a vector and
rows or columns of a matrix.
Generates a Toeplitz matrix where the elements along each diagonal are
constant. c and r are the values on the first diagonal and the first row
respectively.
Matrix Addition:
If A, B Cmn then the MATLAB operation
>> A + B
means A + B = (aij ) + (bij ) = (aij + bij ). That is, the (i, j)th element of A + B is aij + bij .
Matrix Subtraction:
If A, B Cmn then the MATLAB operation
>> A - B
means A B = (aij ) (bij ) = (aij bij ).
Matrix Multiplication by a scalar:
If A Cmn then for any scalar c the MATLAB operation
>> c*A
means cA = c(aij ) = (caij ). For example, the matrix q = (0, .1, .2, .3, .4, .5)T can be generated by
>> q = [ 0 : .1*pi : .5*pi ] 0
28
p times
29
Where needed in these arithmetic operations, MATLAB checks that the matrices have the correct size.
For example,
>> A + B
will return an error message if A and B have different sizes, and
>> A*B
will return an error message if the number of columns of A is not the same as the number of rows of B.
Note: There is one exception to this rule. When a scalar is added to a matrix, as in A + c, the scalar is
promoted to the matrix cJ where J has the same size as A and all its elements are 1. That is,
>> A + c
is evaluated as
>> A + c*ones(size(A))
This is not a legitimate expression in linear algebra, but it is a very useful expression in MATLAB.
For example, you can represent the function
y = 2 sin(3x + 4) 5
for x [2, 3]
In linear algebra this is called the inner product and is defined for vectors a, b Rn by aT b. It is calculated by
>> a 0 *b
(If a, b Cn the inner product is aH b and is calculated by a 0 *b.) The outer product of these two vectors is defined to be abT and is calculated by
>> a*b 0
(If a, b are complex the outer product is abH and is calculated by a*b 0 .) It is important to keep these
two products separate: the inner product is a scalar, i.e., aT b R (if complex, aH b C ), while the outer
product is an nn matrix, i.e., abT Rnn (if complex, abH Cnn ).
In linear algebra we often work with large matrices and are interested in the amount of work required to perform some operation. In the distant past, MATLAB kept track of the number of flops, i.e.,
the number of fl oating-pointoperations, performed during the MATLAB session. Unfortunately, this dis30
appeared in version 6. Instead, we can calculate the amount of CPU time required to execute a command
by using cputime. This command returns the CPU time in seconds that have been used since you began
your MATLAB session. This time is frequently difficult to calculate, and is seldom more accurate than to
1/
100 -th of a second. Here is a simple example to determine the CPU time required to invert a matrix.
>> n = input( 0 n = 0 ); time = cputime; inv(rand(n)); cputime - time
Warning: Remember that you have to subtract the CPU time used before the operation from the CPU
time used after the operation.
You can also calculate the wall clock time required for some sequence of commands by using tic and
toc . For example,
>> tic; <sequence of commands>; toc
returns the time in seconds for this sequence of commands to be performed.
Note: This is very different from using cputime. tic followed by toc is exactly the same as if you had
used a stopwatch to determine the time. Since a timesharing computer can be running many different processes at the same time, the elapsed time might be much greater than the CPU time. On
the other hand, on a multiprocessor computer, the elapsed time might be much less.
Arithmetical Matrix Operations
A+B
A-B
A*B
A n
A\b
A\B
b/A
B/A
Matrix addition.
Matrix subtraction.
Matrix multiplication.
Matrix exponentiation.
The solution to Ax = b by Gaussian
elimination when A is a square nonsingular matrix.
The solution to AX = B by Gaussian
elimination.
The solution to xA = b where x and
b are row vectors.
The solution to XA = B by Gaussian
elimination.
cputime
tic, toc
2.5.
A.*B
A. p
p. A
A. B
A./B
B.\A
Elementwise multiplication.
Elementwise exponentiation.
Elementwise division.
Elementwise left division, i.e., B.\A is
exactly the same as A./B.
Approximately the amount of CPU time (in seconds) used during this session.
Returns the elapsed time between these two commands.
Operator Precedence
It is important to list the precedence for MATLAB operators. That is, if an expression uses two or more
MATLAB operators, in which order does MATLAB do the calculations? For example, what is 1:n+1? Is
it (1:n)+1 or is it 1:(n+1)? And if we solve ACx = b by A*C\b, does MATLAB do (A*C)\b or A*(C\b)?
The former is C1 A1 b while the latter is AC1 b and these are completely different. The following table shows the precedence of all MATLAB operators, that is, the order in which it evaluates an expression.
The precedence is from highest to lowest. Operators with the same precedence are evaluated from left to
right in an expression.
The CPU, Central Processing Unit, is the guts of the computer, that is, the hardware that executes the
instructions and operates on the data.
31
2.6. Be Careful!
( ,, )
. 0 ,, . ,, 0 ,,
+ [unary plus] ,, [unary minus] ,,
. ,, ./ ,, .\ ,, ,, / ,, \
+ [addition] ,, [subtraction]
:
< ,, <= ,, > ,, >= ,, == ,,
&
|
&&
||
The unary plus and minus are the plus and minus signs in x = +1 and x = 1. The plus and minus signs for addition and subtraction are, for example, x = 5 + 1 and x = 10 13. Thus, 1:n+1 is
1:(n+1) because + has higher precedence than :. Also, A*C\b = (A*C)\b because * and \
have the same precedence and so the operations are evaluated from left to right.
2.6.
Be Careful!
Be very careful: occasionally you might misinterpret how MATLAB displays the elements
of a vector or matrix. For example, the MATLAB command eig calculates the eigenvalues of a square
matrix. (We discuss eigenvalues in Section 7.)
der 5, i.e.,
1
1/2
1/3
1/
4
1/
5
1/
3
1/
4
1/
5
1/
6
1/
7
1/
4
1/
5
1/
6
1/
7
1/
8
1/
5
1/
6
1/
7
1/
8
1/
9
On the other hand, in the statistical computer languages R and S (which are somewhat similar to MATLAB),
: has higher precedence than + and so 1:n+1 is (1:n)+1 2:(n+1).
32
2.6. Be Careful!
displays
ans =
3.2879e-06
3.0590e-04
1.1407e-02
2.0853e-01
1.5671e+00
which makes it clear that the smallest eigenvalue is far from zero.
On the other hand, if you enter
>> format short
>> A = [1 2 3; 4 5 6; 7 8 9]
>> eig(A)
MATLAB displays
ans =
16.1168
-1.1168
-0.0000
It might appear from our previous discussion that the last eigenvalue is not zero, but is simply too small
to appear in this format. However, entering
>> format short e
>> ans
displays
ans =
1.6117e+01
-1.1168e+00
-8.0463e-16
Since the last eigenvalue is close to eps, but all the numbers in the matrix A are of reasonable size,
you can safely assume that this eigenvalue is zero analytically. It only appears to be nonzero when calculated by MATLAB because computers cannot add, subtract, multiply, or divide correctly!
As another example of how you might misinterpret the display of a matrix, consider the Hilbert matrix
of order two
1 1/2
H= 1
.
/2 1/3
We write H100 as
H100 1010
1.5437
0.8262
0.8262
0.4421
,
1.0e+10 *
1.5437
0.8262
0.8262
0.4421
It is very easy to miss the term 1.0e+10 * because it stands apart from the elements of the matrix.
Similarly, entering
33
>>
>>
>>
format short
H = hilb(2)
( H (1/2) ) 2 - H
should result in the zero matrix, since (H1/2 ) = H. However, MATLAB displays
ans =
1.0e-15 *
0.2220
0
0
0
where, again, it is easy to miss the term 1.e-15 * and not realize that this matrix is very small in
fact, it should be zero.
Be careful: MATLAB has nite memory. You should have no problem creating a matrix by
>> A = zeros(1000)
but you might well have a problem if you enter
>> A = zeros(10000)
The amount of memory available is dependent on the computer and the operating system and is very hard
to determine. Frequently it is much larger than the amount of physical memory on your computer. But,
even if you have sufficient memory, MATLAB may slow to a crawl and become unusable. The whos command will tell you how much memory you are using and show you the size of all your variables. If you
have large matrices which are no longer needed, you can reduce their sizes by equating them to the null
matrix, i.e., [], or remove them entirely by using clear.
Warning: Recall that the clear command is very dangerous because clear A deletes the variable A
but clear (without anything following) deletes all variables!
2.7.
In linear algebra mathematical functions cannot usually be applied to matrices. For example, eA and
sin A have no meaning unless A is a square matrix. (We will discuss their mathematical definitions in
Section 15.)
Here we are interested in how MATLAB applies common mathematical functions to matrices and vectors. For example, you might want to take the sine of every element of the matrix A = (aij ) Cmn , i.e.,
B = (sin aij ). This is easily done in MATLAB by
>> B = sin(A)
Similarly, if you want C = (eaij ), enter
>> C = exp(A)
p
Also, if you want D =
aij type
>> C = sqrt(A)
or
>> C = A. (1/2)
All the common mathematical functions in the table entitled Some Common Real Mathematical Functions in Section 1.5 can be used in this way.
As we will see in the section on graphics, this new interpretation of mathematical functions makes it
easy in MATLAB to graph functions without having to use the MATLAB programming language.
2.8.
MATLAB has a number of simple commands which are used quite frequently. Since many of them
are quite useful in analyzing data, we have grouped them around this common theme.
To calculate the maximum value of the vector x, type
>> m = max(x)
If you also want to know the element of the vector which contains this maximum value, type
>> [m, i] = max(x)
If the elements of the vector are all real, the result of this command is the element which has the maximum value. However, if any of the elements of x are complex (i.e., non-real), this command has no math34
ematical meaning. MATLAB defines this command to determine the element of the vector which has the
maximum absolute value of the elements of x.
Warning: Make sure you understand the description of max if you every apply it to non-real vectors. For
example, if x = (2, 1)T then max(x) returns 1 as expected. However, if x = (2, i)T then
max(x) returns 2. This is because the element which has the largest absolute value is 2.
Thus, if x is a non-real vector, then max(x) is not the same as max(abs(x)).
Since the columns of the matrix A can be considered to be vectors in their own right, this command
can also be applied to matrices. Thus,
>> max(A)
returns a row vector of the maximum element in each of the columns of A if all the elements of A are
real. If any of the elements of A are non-real, this command returns the element in each column which
has the maximum absolute value of all the elements in that column.
To find the maximum value of an entire real matrix, type
>> max(max(A))
or
>> max(A(:))
and to find the maximum absolute value of an entire real or complex matrix, type
>> max(max(abs(A)))
or
>> max(abs(A(:)))
There is also another use for max. If A and B are matrices which either have the same size, or one or
both is a scalar, then
>> max(A, B)
returns a matrix which is the same size as A and B (or the size of the larger if one is a scalar) and which
contains the larger of the values in each element. For example,
>> A = max(A, 0)
replaces all negative elements of A with zeroes.
Note: If max has one argument, then it determines the maximum value of a vector or the maximum value
in each column of a matrix. If it has two arguments, it determines the maximum value of each element of the two matrices.
Not surprisingly, the command min acts similarly to max except that it finds the minimum value (or
element with the minimum absolute value) of the elements of a vector or the columns of a matrix.
To calculate the sum of the elements of the vector x, type
>> sum(x)
sum behaves similarly to max when applied to a matrix. That is, it returns the row vector of the sums
of each column of the matrix. This command is sometimes useful in adding a deterministic series. For
example,
>> 1 + 1/2 + 1/3 + 1/4 + 1/5 + 1/6 + 1/7 + 1/8 + 1/9 + 1/10 + 1/11 + 1/12 + ...
1/13 + 1/14 + 1/15 + 1/16 + 1/17 + 1/18 + 1/19 + 1/20
is entered much more easily as
>> sum(ones(1, 20)./[1:20])
or even as
>> sum(1./[1:20])
The mean, or average, of these elements is calculated by
>> mean(x)
where mean(x) = sum(x)/length(x).
std calculates the standard deviation of the elements of a vector. The standard deviation is a measure
of how much a set of numbers vary and is defined as
std(x) =
n
2
1X
xi h x i
n i=1
1/2
>> sort(x)
If the vector is non-real, the elements are sorted in increasing absolute value. (If two elements have the
same absolute value, the one with the smaller absolute angle in polar coordinates is used.)
The MATLAB function diff calculates the difference between successive elements of a vector. For
example, if x Rn then the command
>> s = diff(x)
generates the vector s Rn1 which is defined by si = xi+1 xi . There are a number of uses for this
command. For example,
if s has been sorted, then if any element of diff(s) is 0, an element of s has been repeated
and we can even determine the number of times it has been repeated.
similarly, if all the elements of diff(x) are positive, then all the elements of s are monotonically
increasing.
a numerical approximation to the derivative of y = f (x) can be calculated by diff(y)./diff(x).
(The functions any and all , which are discussed in Section 8.2, are used to determine if any the elements
of a vector satisfy some property and/or if all the elements satisfy it.)
The MATLAB function which is almost the inverse of diff is cumsum. It calculates the cumulative sum of the elements of a vector or matrix. For example, if s Rn1 has been generated by s =
diff(x), then
>> c = cumsum(s)
Pi
generates the vector c Rn1 where ci = j=1 sj . We can recover x by
>> xrecovered = zeros(size(x))
>> xrecovered(1) = x(1)
>> xrecovered(2:length(x)) = x(1) + c
There is also a cumulative product command, namely cumprod. Thus can be used to generate
(1, x, x2 , . . . , xn ) by
>> [1, cumprod(x*ones(1,n))]
and (1!, 2!, 3!, . . . , n!) by
>> cumprod(1:n)
Note: All of these commands can be applied to matrices, in which case they act on each column of the
matrix separately. However, there is always an optional argument (usually the second) which can
change the command so that it acts on each row of the matrix.
There are also a number of MATLAB commands which are particularly designed to plot data. The
commands we have just discussed, such as the average and standard deviation, give a coarse measure of
the distribution of the data. To actually see what the data looks like, it has to be plotted. Two particularly useful types of plots are histograms (which show the distribution of the data) and plots of data which
include error bars. These are both discussed in Section 4.1.
pAlthough it does not quite fit here, sometimes you want to know the length of a vector x, which is
x21 + x22 + x2n . (Note that this is not length(x) which returns the number of elements in x, i.e.,
n.) This length, which is often called the Euclidean length, can be calculated by entering
>> sqrt( x 0 *x )
but it can be entered more easily by
>> norm(x)
(As we discuss in Section 7, the norm of a vector is a more general concept than simply the Euclidean
length.)
Warning: The number of elements in the vector x is calculated by length(x) while the (Pythagorean)
length of the vector is calculated by norm(x).
36
max(A)
max(A,B)
o
min(x)
min(A)
o
mean(x)
mean(A)
norm(x)
norm(A)
o
prod(x)
prod(A)
o
sort(x)
sort(A)
o
std(x)
std(A)
o
sum(x)
sum(A)
o
diff(x)
diff(A)
o
cumsum(x)
cumsum(A)
o
cumprod(x)
cumprod(A)
2.9.
We have already discussed 1-D arrays (i.e., vectors) and 2-D arrays (i.e., matrices). Since these are two
of the most fundamental objects in linear algebra, there are many operations and functions which can be
applied to them. In MATLAB you can also use multidimensional arrays (i.e., n-D arrays).
A common use for multidimensional arrays is simply to hold data. For example, suppose a company
produces three products and we know the amount of each product produced each quarter; the data naturally fits in a 2-D array, i.e., (product, amount). Now suppose the company has five sales regions so we
split the amount of each product into these regions; the data naturally fits in a 3-D array, i.e., (product,
region, amount). Finally, suppose that each product comes in four colors; the data naturally fits in a 4-D
array, i.e., (product, color, region, amount).
For another example, a 3-D array might be the time evolution of 2-D data. Suppose we record a grey
scale digital image of an experiment every minute for an hour. Each image is stored as a matrix M with
mi,j denoting the value of the pixel positioned at (xi , yj ). The 3-D array Mall can contain all these images: Mall(i,j,k) denotes the value of the pixel positioned at (xi , yj ) in the k th image. The entire k th
image is Mall(:,:,k) and it is filled with the k th image M by
>> Mall(:,:,k) = M
If you want to multiply M by another matrix A, you can use M*A or Mall(:,:,k)*A; if you want to
average the first two images you can use .5*(Mall(:,:,1)+Mall(:,:,2)).
Many MATLAB functions can be used in n-D, such as ones, rand, sum, and size. The cat func37
2.10. Be Able To Do
tion is particularly useful in generating higher-dimensional arrays. For example, suppose we have four
matrices A, B, C, and D R27 which we want to put into a three-dimensional array. This is easily done
by
>> ABCD = cat(3, A, B, C, D)
which concatenates the four matrices using the third dimension of ABCD. (The 3 denotes the third dimension of ABCD.) And it is much easier than entering
>> ABCD(:,:,1) = A;
>> ABCD(:,:,2) = B;
>> ABCD(:,:,3) = C;
>> ABCD(:,:,4) = D;
If instead, we enter
>> ABCD = cat(j, A, B, C, D)
then the four matrices are concatenated along the j th dimension of ABCD. That is, cat(1, A, B, C, D)
is the same as [A, B, C, D] and cat(2, A, B, C, D) is the same as [A; B; C; D].
Another useful command is squeeze which squeezes out dimensions which only have one element. For
example, if we enter
>> E = ABCD(:,2,:)
(where the array ABCD was created above), then we might think that E is a matrix whose columns consist of the second columns of A, B, C, and D. However, size(E) = 2 1 4 so that E is a threedimensional array, not a two-dimensional array. We obtain a two-dimensional array by squeeze(E).
The command permute reorders the dimensions of a matrix. For example,
>> ABCD = cat(3, A, B, C, D)
>> BCDA = permute(ABCD, [2 3 4 1])
is the same as
>> BCDA = cat(3, B, C, D, A)
That is, the second argument of permute shows where the original ordering of the dimensions, i.e., 1, 2,
. . . , n, are to be placed in the new ordering. ipermute is the inverse of permute so, for example,
>> BCDA = cat(3, B, C, D, A)
>> ABCD = ipermute( BCDA, [2 3 4 1] )
Multidimensional Array Functions
cat
ipermute
permute
squeeze
2.10.
Concatenates arrays; this is useful for putting arrays into a higher-dimensional array.
The inverse of permute.
Reorders the dimensions of a multidimensional array.
Removes (i.e., squeezes out) dimensions which only have one element.
Be Able To Do
After reading this section you should be able to do the following exercises. The solutions are given on
page 163.
1. Consider the matrix
1 2 3 4
5 6 7 8
A=
.
9 10 11 12
13 14 15 16
A=
1
4
1
1
4
. .
.
1
. .
.
1
0
. .
.
4
1
1
4
Rnn
A=
1
4
e2
1
9
..
.
1 1 1
1 1 1
1 1 1
A=
1 1 1
1 1 1
5 1 1
1
..
.
en1
0
..
.
(n 1)2
en
1
n2
Rnn
into MATLAB.
1 0 0 0 0 0
1 0 0 0 0 0
1 0 0 0 0 0
1 0 0 0 0 0
1 0 0 0 0 0
1 0 0 0 0 0
5
0
0
.
0
0
0
(a) Enter it using as few keystrokes as possible. (In other words, dont enter the elements individually.)
(b) Zero out all the elements of A below the diagonal.
5. Enter the column vector
x = (0, 1, 4, 9, 16, 25, . . . , 841, 900)T
using as few keystrokes as possible. (In other words, dont enter the elements individually.)
6. (a) Generate a random 55 matrix R.
(b) Determine the largest value in each row of R and the element in which this value occurs.
(c) Determine the average value of all the elements of R.
(d) Generate the matrix S where every element of S is the sine of the corresponding element of R.
(e) Put the diagonal elements of R into the vector r.
7. Generate the matrix
1 2 3
A = 4 5 6 .
7 8 10
(a) Calculate a matrix B which is the square root of A. That is, B2 = A. Also, calculate a matrix C each
of whose elements is the square root of the corresponding element of A.
(b) Show that the matrices you have obtained in (a) are correct by substituting the results back into the
original formulas.
anonymous functions allow us to easily define a function which we can then plot. Second, some graphics
functions require that the name of a function be passed as an argument. Third, character strings are necessary in labelling plots. And, finally, cell arrays are occasionally helpful in labelling plots. Cell arrays are
generally used to manage data and since structures are also used to manage data we also include them
here. Another reason is that there are a number of data types in MATLAB including floating-point variables, integers, text variables, cell arrays, structures, and logical variables We might as well get all these
out of the way at once.
3.1.
Anonymous Functions
In MATLAB it is common to define a mathematical function in a separate file as we discuss in Section 8.3. (This is similar to writing a function or subroutine or subprogram in a high-level computer
language.) However, if the mathematical function is particularly simple, that is, it can be written as one
simple expression, we can define it in MATLAB using an anonymous function. If our function is
f (< arg1 >, < arg2 >, . . .) = < expression >
the MATLAB statement is
>> f = @(<arg1>, <arg2>, ...) <expression>
For example, we can define the function
f (t) = t5 e2t cos(3t)
by
>> f = @(t) t. 5 .* exp(-2*t) .* cos(3*t)
and then evaluate it by
>> x = [0:.01:1] 0
>> fx = f(x)
>> A = rand(5)
>> fA = f(A)
More generally, we can define
g(x, y, a, b, c) = xa ebx cos(cy)
by
>> g = @(x, y, a, b, c) x. a .* exp(-b.*x) .* cos(c.*y)
in which case any of the input arguments can be in R or in Rn . It is also possible although probably
not very useful to let g have one vector argument, say x = (x, y, a, b, c)T by
>> g = @(x) x(1) x(3) * exp(-x(4)*x(1)) * cos(x(5)*x(2))
(In this example there is no advantage to using .* or . .)
Warning: It is quite easy to forget to put dots (i.e., .) before the mathematical operations of multiplication (i.e., * ), division (i.e., / ), and exponentiation (i.e., ). For example, if f is defined by
>> f = @(t) t 5 * exp(-2*t) * cos(3*t)
then
>> f(3)
is allowed, but not
>> f([1:10])
Be careful!
The syntax for defining an anonymous function is
>> @(<argument list>) <expression>
(Since there is no left-hand side to this expression, the name of this function is ans.) The symbol @ is the
MATLAB operator that constructs a function handle. This is similar to a pointer in C which gives the address of a variable or of a function. The name handle is used in MATLAB to denote a variable which
refers to some object which has been created. Thus, we can think of an anonymous function as being
created by
(<argument list>) <expression>
and the handle to the function (in C, the address of the function) being returned by using @. By the way,
we can create a function handle to a MATLAB function by, for example,
40
>> f = @cos
so that f(3) is the same as cos(3). We give an example where this is very useful in Section 3.4
It is even possible to define a piecewise function in this way. For example, the piecewise function
t(x) =
i.e., an isosceles triangle with the length of the base 2 and the height 1, is
>> t = @(x) (heaviside(x + 1) - heaviside(x - 1)).*(1 - abs(x));
in MATLAB.
Note: The x in @(x) is a dummy variable, i.e., the x is independent of any other x which appears in
the code. Or, in other words, the function could have been defined equally well by
>> t = @(w o ) = heaviside(w o + 1) - heaviside(w o - 1).*(1 - abs(w o ));
It is important to understand that all user-defined variables which appear in <expression> must eiIf the variable does not
appear in the argument list, then its value is xed when the function is defined. For example, if a very simple function is defined by
>> r = 10
>> h = @(x) r*x
then the function is h(x) = 10x even if r is modified later. Thus,
>> h(5)
returns 50 and so does
>> r = 0
>> h(5)
Warning: Dont forget that if a variable does not appear in the argument list, then its value is xed when
the function is defined.
A function can also be defined but don't do it by the inline command. For example, the function f defined above can also be defined by
>> f = inline( 0 t. 5 .* exp(-2*t) .* cos(3*t) 0 , 0 t 0 )
In general, if our function is
ther appear in the argument list or be defined before the function is defined.
3.2.
Warning: This is a very important section read it carefully and understand it. If you try to pass a
function as an argument to another function, something like
>> ezplot(sin)
(you are trying to generate an easy plot of the sine function) and you get a strange error
message, something like
41
3.3. Strings
3.3.
Strings
Character strings are a very minor part of MATLAB, which is mainly designed to perform numerical
calculations. However, they perform some very useful tasks which are worth discussing now.
It is often important to combine text and numbers on a plot. Since we discuss graphics in the next section, now is a good time to discuss how characters are stored in MATLAB variables. A string variable,
such as
>> str = 0 And now for something completely different 0
is simply a row vector with each character (actually its ASCII representation as shown on page 167) being
a single element. MATLAB knows that this is a text variable, not a regular row vector, and so converts
the numerical value in each element into the corresponding character when it is printed out. For example,
to see what is actually contained in the vector str enter
>> str + 0
or
>> 1*str
Since str is a row vector, a substring can be easily extracted. For example,
>> str(1:7)
returns And now and
>> str([9:11, 34])
returns ford.
Character variables are handled the same as vectors or matrices. For example, to generate a new text
variable which adds by Monty Python to str, i.e., to concatenate the two strings, enter
>> str2 = [str 0 - by Monty Python 0 ]
or
>> str2 = [str, 0 - by Monty Python 0 ]
(which might be easier to read). To convert a scalar variable, or even a vector or a matrix, to a character
variable use the function num2str. For example, suppose you enter
>> x = linspace(0, 2*pi, 100) 0
>> c1 = 2
>> c2 = -3
>> y = c1*sin(x) + c2*cos(x)
and want to put a description of the function into a variable. This can be done by
>> s = [ num2str(c1), 0 *sin(x) + 0 , num2str(c2), 0 *cos(x) 0 ]
42
without explicitly having to enter the values of c1 and c2. (An optional second argument to
num2str determines exactly how the number or numbers are to be printed, but this is usually not needed.)
A text variable can also contain more than one line if it is created as a matrix. For example,
0
>> Str = [ 0 And
0
0
now
0
0
for
0
something 0
0
completely 0
0
different 0 ]
is four lines long. Since str is a matrix, each row must have the same number of elements and so we have
to pad all but the longest row. (Using cell arrays, we will shortly show how to avoid this requirement.)
Note: We do not usually enter matrices this way, i.e., one column per line. Instead, we simply use ; to
separate columns. However here we need to make sure that each row has exactly the same number
of characters or else a fatal error message will be generated.
If desired, you can have more control over how data is stored in strings by using the sprintf command
which behaves very similarly to the C commands sprintf, fprintf, and printf. It is also very similar
to the fprintf command in MATLAB which is discussed in detail in Section 6. Note that the data can
be displayed directly on the screen by using disp. That is, sprintf(...) generates a character string
and disp(sprintf(...)) displays it on the screen.
There also is a str2num command to convert a text variable to a number and sscanf to do the same
with more control over how the data is read. (This is also very similar to the C command, as discussed in
Section 6.)
Occasionally, there may be a worry that a string has leading or trailing blanks. These can be removed
by
>> strtrim(<string>)
This can also be used with string matrices if there leading or trailing blanks in all the rows. (It can also
be used with cell arrays of strings, which we discuss next.)
Some Useful String Commands
num2str(x)
str2num(str)
strtrim(str)
sscanf
sprintf
3.4.
It is occasionally useful in MATLAB to have a single variable contain all the data which is related to
a specific task and this data might well consist of scalars, vectors and/or matrices, and text variables.
One simple reason for this is that it is easier to pass all the data into and out of functions. A cell array
generalizes the standard arrays which were discussed in the previous section. The elements of a standard array are numbers, either real or complex, whereas the elements of a cell array can be any data
type. The primary difference between a cell array and a structure is that in a structure the elements are
named rather than numbered. We consider this an advanced topic not because it is complicated, but because it is so seldom necessary.
A simple example of a cell array is
>> C = {2+3i, 0 go cells 0 ; [1 2 3] 0 , hilb(5) }
43
44
scalar 0
text 0
0
vector 0
0
matrix 0
and
45
3.5.
A MATLAB variable can have a large number of different types of values. These values used to be
called data types, but they are now commonly called classes which is a central concept in objectoriented programming (OOP). The OOP capabilities of MATLAB are similar to those in C++ and
Java. We will discuss some of these capabilities at the end of this subsection. However, first we discuss
the fundamental classes in MATLAB. An important point to remember is that a variable in most programming languages is a single quantity, whereas in MATLAB it is a vector or a matrix or an array of
quantities.
We have already described a number of fundamental classes, and we first discuss those which occur
frequently in this tutorial.
double: By default any variable that is given a numerical value is a double precision floating-point
number. For example, the variable x which is defined by x = 1 is an instance of the
double class, not an integer class.
char: All strings are instances of the char class. Each character in a string is represented by two
bytes because it can represent any Unicode UTF-16 character (although here we are only
interested in ASCII characters).
cell: A cell variable itself is an instance of the cell class, but it can contain any number of elements, which can all be instances of different classes.
struct: Similarly, a structure variable is an instance of the struct class, but it can contain any
number of fields, which can all be instances of different classes.
function handle: This provides a means to call a function indirectly.
logical: We will discuss logical variables in Section 8.1, but a simple example is
>> A = rand(2)
>> C = (A > .5)
Each element of C R22 has the logical value 1, i.e., true, if the corresponding element of
46
A is > .5 and value 0, i.e., false, otherwise. These logical values are each 1 byte.
MATLAB has 15 fundamental classes, each in the form of a matrix (from a 00 matrix up to an n
dimensional matrix for any n ). For completeness, we now list them all.
Fundamental Classes
double
single
int8
uint8
int16
uint16
int32
uint32
int64
uint64
char
logical
cell
struct
function handle
For example, to obtain an instance of the single class you can enter
A = single(rand(5))
The same technique holds for all the numerical classes.
Warning: Caveat Emptor! The procedures for combining different numerical classes is very different from
other programming languages. For example, the results of
>> a = 5.5
>> i = int32(3)
>> j = int8(127)
>> ai = a + i
>> aj = a + j
are ai = 9, which is an instance of the int32 class, and aj = 127, which is an instance of
the int8 class.
To determine the class of a variable, use the function class. You can also determine if a variable has a
particular class by using isa. Continuing the previous example,
>> class(i)
returns int32 and
>> isa(i, 0 int8 0 )
returns 0.
In addition, MATLAB has user-defined classes, similar to classes in object-oriented programming languages. A simple template for generating a new class is
47
3.6. Be Able To Do
3.6.
Be Able To Do
After reading this section you should be able to do the following exercises. The solutions are given on
page 163.
1. Generate a structure with the fields name, rank, and serial number. Put something appropriate in
each in two ways:
(1) directly, i.e., s.name = ???, and
(2) using the struct command.
Then add one to the serial number.
2. (a) Generate a 23 cell array with the following elements:
(1,1): a uniform random matrix of size 5.
(2,1): the string Hilbert.
(1,2): 10.
(2,2): the function handle for the function sin e x.
(1,3): this is the square of the matrix in (1, 1) .
(2,3): the logical value true.
(b) Square the value of the element (1, 2) which is in the cell element (1, 1).
48
4. Graphics
A very useful feature of MATLAB is its ability to generate high quality two- and three-dimensional
plots using simple and flexible commands. All graphical images are generated in a graphics window,
which is completely separate from the text window in which MATLAB commands are typed. Thus,
non-graphical and graphical commands can be completely intermixed.
Graphical images can be generated both from data calculated in MATLAB and from data which has
been generated outside of MATLAB. In addition, these images can be output from MATLAB and printed
on a wide variety of output devices, including color ink-jet printers and black-and-white and color laser
printers.
There are a number of demonstrations of the graphical capabilities in MATLAB which are invoked by
>> demo
Since the MATLAB commands which generate the plots are also shown, this demo makes it quite easy to
generate your own graphics. You also can have very fine control over the appearance of the plots. We begin by considering only the basic commands; more advanced graphics commands are discussed in the next
section.
Note: Most MATLAB commands which take vectors as arguments will accept either row or column vectors.
4.1.
Two-Dimensional Graphics
The MATLAB function plot is used to constructing basic two-dimensional plots. For example, suppose you want to plot the functions y1 = sin x and y2 = ecos x for x [0, 2]; also, you want to plot
y3 = sin cos(x2 x) for x [0, 8]. First, generate n data points on the curve by
>> n = 100;
>> x = 2*pi*[0:n-1] 0 /(n-1);
>> y1 = sin(x);
>> y2 = exp(cos(x));
>> xx = 8*[0:n-1]/(n-1);
>> y3 = sin( cos( xx. 2 - xx ) );
We plot these data points by
>> plot(x, y1)
>> plot(x, y2)
>> plot(xx, y3)
Note that the axes are changed for every plot so that the curve just fits inside the axes. We can generate
the x coordinates of the data points more easily by
>> x = linspace(0, 2*pi, n);
>> xx = linspace(0, 8, n);
The linspace command has two advantages over the colon operator:
(1) the endpoints of the axis and the number of points are entered directly as
>> x = linspace(<first point>, <last point>, <number of points>)
so it is much harder to make a mistake; and
(2) round-off errors are minimalized so you are guaranteed that x has exactly n elements, and its first
49
and last elements are exactly the values entered into the command.
To put all the curves on one plot, type
>> plot(x, y1, x, y2, xx, y3)
Each curve will be a different color but this will not be visible on a black-and-white output device.
Instead, you can change the type of lines by
>> plot(x, y1, x, y2, 0 -- 0 , xx, y3, 0 : 0 )
where -- means a dashed line and : means a dotted line. (We list all these symbols in the following
table.) In addition, you can use small asterisks to show the locations of the data points for the y3 curve
by
>> plot(x, y1, x, y2, 0 -- 0 , xx, y3, 0 :* 0 )
These strings are used to modify the color of the line, to put markers at the nodes, and to modify the type
of line as shown in the table below. (As we discuss later in this section, the colors are defined by giving
the intensities of the red, green, and blue components in that order.)
Note: The plot function can even have only one argument. For example, entering
>> plot(y1)
will result in a plot which is equivalent to
>> plot([1:length(y1)], y1)
red (1 0 0)
green (0 1 0)
blue (0 0 1)
yellow (1 1 0)
magenta (1 0 1)
(a deep purplish red)
cyan (0 1 1)
(greenish blue)
white (1 1 1)
black (0 0 0)
Marker Description
+
o
*
.
x
s
d
v
>
<
p
h
plus sign
circle
asterisk
point
cross
square
diamond
upward pointing
triangle
downward pointing
triangle
right pointing triangle
left pointing triangle
pentagram
hexagram
For example,
>> plot(x, y1, 0 r 0 , x, y2, 0 g--o 0 , x, y3, 0 mp 0 )
plots three curves: the first is a red, solid line; the second is a a green, dashed line with circles at the data
points; the third has magenta pentagrams at the data points but no line connecting the points.
We can also plot the first curve, and then add the second, and then the third by
As we discussed previously, it is very unlikely (but it is possible) that round-off errors might cause the
statement
>> x = [0: 2*pi/(n-1): 2*pi] 0 ;
to return n 1 elements rather than n. (For example, the output of [0 : 0.01 : 0.02-eps] is 0 0.0100.)
This is why we used the statement
>> x = 2*pi*[0:n-1] 0 /(n-1);
above, which does not suffer from round-off errors because the colon operator is only applied to integers.
50
10
X
ak e
k=1
for x [0, L]. This is a sum of k modes, each of which has a Gaussian shape with half-width w, where
the k th mode has amplitude ak and speed ck . The code entitled running guassians (which is contained in the accompanying zip file) is
If there is no graphics window then this command is equivalent to figure, while if there is a graphics window,
then this command is equivalent to figure(gcf). We discuss figure and gcf in advanced subsections.
51
Alternatively, you can use the pause command (discussed in Section 8.3) to make all the plots visible.
drawnow draws each new plot as fast as possible, whereas pause(time) pauses execution of the code for the given
time.
52
>>
axis equal
or
>> axis image
and to have the same length by
>> axis square
To learn about all the options for these commands, use the doc command.
Note: The command axis is generally only in effect for one plot. Every new plot turns it off, so it must
be called for every plot (unless hold on has been invoked).
The plot command generates linear axes. To generate logarithmic axes use semilogx for a logarithmic axis in x and a linear axis in y, semilogy for a linear axis in x and a logarithmic axis in y, and
loglog for logarithmic axes in both x and y.
MATLAB has two different commands to plot a function directly rather than plotting a set of points.
Warning: These commands do not always generate the correct curve (or curves) because they know nothing
of the actual behavior of the function. They can have problems with sharp peaks and asymptotes and other strange behavior. We will show some examples shortly.
The first command we discuss is fplot, which can be executing by simply entering
>> fplot(<function handle>, <limits>)
where the function is usually generated as an anonymous function or a MATLAB function or a user generated function m-file (as described in Section 8.3). The limits are either
[xmin xmax]
in which case the y-axis just encloses the curve or
[xmin xmax ymin ymax]
in which case you are also specifying the endpoints on the y-axis.
Note: Recall in Section 3.2 we discussed how to pass a function as an argument.
This function uses adaptive step control to generate as many data points as it considers necessary to
plot the function accurately. You can also store the data points calculated by
>> [x, y] = fplot(<function handle>, <limits>)
rather than having the function plotted directly. You then have complete control over how to plot the
curve using the plot function.
The other command which can plot a function is ezplot, which is more general than fplot. To plot a
function on the interval [2, +2] enter
>> ezplot(<function handle>)
To include limits (as with fplot ) enter
>> ezplot(<function handle>, <limits>)
In addition, a parametrically defined function can be plotted by
>> ezplot(<fnc 1>, <fnc 2>, <limits>)
Finally, this command can also plot an implicitly defined function, i.e., f (x, y) = 0, by
>> ezplot(<2D fnc>, <limits>)
For example,
>> f = @(x, y) (x 2 + y 2) 2 - (x 2 - y 2);
>> ezplot(f)
plots the lemniscate of Bernoulli (basically an symbol).
Warning: Be particularly careful when plotting implicit functions because they can be really nasty and
occasionally ezplot may not get it right.
There is an important difference between
>> fplot(f, [-5 5])
and
>> ezplot(f, [-5 5])
In the former f(x) is only evaluated for scalar values of x, while in the latter f(x) is evaluated for vector values of x. Thus, when using ezplot care must be taken if f is evaluated in a function m-file. If
f(x) cannot be evaluated for vector values, the error message
Warning: Function failed to evaluate on array inputs; vectorizing the function may
speed up its evaluation and avoid the need to loop over array elements.
will be generated
53
fplot and ezplot do not always generate exactly the same curves. For example, in
>> f = @(x) log(x) + 1;
>> fplot(f, [-2*pi 2*pi])
>> ezplot(f)
fplot generates a spurious plot for x [2, 0) where it plots the real part of log x while ezplot only
plots the function for x (0, 2]. Also, in
>> f = @(x) x ./ (x. 2 + 0.01);
>> fplot(f, [-2*pi +2*pi])
>> ezplot(f)
the vertical axes are different and ezplot is missing part of the curve. Finally, in
f = @(x) x 3/(x 2 + 3*x - 10);
ezplot(f, [-10 +10])
the function blows up at x = 5 and 2 and part of the curve for x (5, 2) is not shown.
Polar plots can also be generated by the polar command. There is also an easy command for generating polar plots, namely ezpolar.
Since you often want to label the axes and put a title on the plot, there are specific commands for each
of these. Entering
>> xlabel(<string>)
>> ylabel(<string>)
>> title(<string>)
put labels on the x-axis, on the y-axis, and on top of the plot, respectively. Note that a title can contain
more than one line as was discussed in Section [Macro:[text: cell]chap].
For example, typing title(t) where
0
t = [ 0 The Dead
0
Parrot Sketch 0 ]
or
t = { 0 The Dead 0
0
Parrot Sketch 0 }
or
t = { 0 The Dead 0 ; 0 Parrot Sketch 0 }
results in a two-line title. The first uses a standard array and so requires all the rows to have the same
number of columns, whereas the latter two use a cell array and so each row can have a different length.
There are also a number of ways to plot data, in addition to the commands discussed above. The two
we discuss here are histograms and error bars. To plot a histogram of the data stored in the vector x,
type
>> hist(x)
which draws ten bins between the minimum and maximum values of the elements in x. For example, to
see how uniform the distribution of random numbers generated by rand is, type
>> x = rand(100000, 1);
>> hist(x)
To draw a histogram with a different number of bins, type
>> hist(x, <number of bins>)
and to draw a histogram with the centers of the bins given by the vector c, type
>> hist(x, c)
As another example, to see how uniform the distribution of Gaussian random numbers generated by
randn is, type
>> x = randn(100000, 1);
>> hist(x)
which generates random numbers with mean 0 and standard deviation 1. Clearly you need more random
numbers to get a good histogram but, at the moment, we are interested in a different point. If you
rerun this command a number of times, you will find that the endpoints of the histogram fluctuate. To
avoid this instability, you can fix the endpoints of the histogram by
54
>> xmax = 4;
>> nrbin = 20;
>> nrdata = 100000;
>> c = xmax*[ -1+1/nrbin : 2/nrbin : 1-1/nrbin ];
>> x = randn(nrdata, 1);
>> hist(x, c)
Note that c contains the midpoints of each bin and not their endpoints. Another way to calculate c,
which might be clearer, is
>> c = linspace(-xmax+xmax/nrbin, xmax-xmax/nrbin, nrbin);
Of course, to get a good histogram you should increase nrbin, say to 100, and nrdata, say to
100,000. If you now rerun this code you will see a much smoother histogram.
A histogram shows the frequency of values in a vector, say x again, but suppose we want to compare
this histogram to an actual probability density function. For example, we have just discussed the Gaussian
distribution. If it has mean and standard deviation , then the density function is
(x )2
1
exp
.
f (x) =
2 2
2 2
R +
where f (x) dx = 1. To generate a random sequence use
>> mu = 5;
>> sig = 2;
>> nrdata = 100000; >> x = mu + sig*randn(100000, 1);
>> hist(x)
In order to compare the histogram with the density function, we must rescale the histogram so that its
area is 1. To find its area enter
>> nr bin = 1000; >> [nr each bin, x center] = hist(x, nr bin);
which returns the number of elements in each bin as well as its midpoint. (The histogram is not plotted
because the results are output.) Thus,
>> dx = x center(2) - x center(1); >> area = sum(nr each bin)*dx % or area =
nrdata*dx
To plot this rescaled histogram, enter
>> bar(x center, nr each bin/area, 0 hist 0 )
Note that output variables from hist are reversed when input to bar.
A histogram shows a graphical image of discrete data by stacking rectangles next to each other. A
stairstep graph is similar in that it draws the continuous curve describing the maximum of the histogram,
but it does not complete the rectangle. The command
>> stairs(x, y)
which can also be considered to be the plot of a step function, draws such a curve.
We have already seen how to plot the vector x vs. the vector y by using the plot command. If, additionally, you have an error bar of size ei for each point yi , you can plot the curve connecting the data
points along with the error bars by
>> errorbar(x, y, e)
Sometimes the error bars are not symmetric about the y values. In this case, you need vectors l and u
where at xi the error bars extend from yi li to yi + ui . This is done by
>> errorbar(x, y, l, u)
Note: All the elements of l and u are non-negative.
Data can also be entered into MATLAB from a separate data file. For example,
>> M = csvread( 0 <file name> 0 )
reads in data from a file one row per line of input. The numbers in each line must be separated by commas. The data can then be plotted as desired. The command csvwrite writes the elements of a matrix
into a file using the same format. (If desired, you can have much more control over how data is input and
output by using the fscanf and fprintf commands, which are similar to their C counterparts. These
commands are discussed in detail in Section 6.)
The load command can also be used to read a matrix into MATLAB from a separate data file. The
data must be stored in the data file one row per line. The difference between this command and csvread
is that the numbers can be separated by commas or semicolons or by spaces. The matrix is input by
55
entering
>> load <file name>
or
>> load 0 <file name> 0
or
>> load( 0 <file name> 0 )
and it is stored in the matrix named <file name-no extension> (i.e., drop the extension, if any, in the
file name). Or you can enter
>> M = load( 0 <file name> 0 )
and the data is stored in the matrix M. This can also be done by
>> M = importdata( 0 <file name> 0 )
which had the added advantage that there does not need to be an equal number of data in each row. If
not, the number of rows of M is the largest of the number of elements in any row of the file and missing
data are replaced by NaNs.
Graphics can also be easily printed from within MATLAB. You can print directly from the graphics
window by going into the File menu item. If desired, the plot can be sent to a file rather than to an
output device. You can also store the plot in the text window by using the command print. There are
an innumerable number of printer specific formats that can be used. (See help print or doc print for
details.) If you want to save a file in postscript, you can save it in black-and-white postscript by
>> print -deps <file name b&w>
or in color postscript by
>> print -depsc <file name color>
There is a minor, but important, difference between these two files if they are printed on a black-andwhite laser printer. When the black-and-white file is printed, all the non-white colors in the plot become
black. However, when the color file is printed, the colors are converted to different grayscales. This makes
it possible to differentiate lines and/or regions.
Note: The print command is also a MATLAB function where it is called by
>> print( 0 -deps 0 , 0 <file name b&w> 0 )
The advantage of using the print function is that the arguments can be variables. An
oversimplified example is
>> device = 0 -deps 0 ;
>> file = 0 <file name b&w> 0 ;
>> print(device, file)
The load command is a little tricky because it can read in files generated both by MATLAB (using the save
command) and by the user. For example,
>> save allvariables;
>> clear
or
>> save allvariables.mat;
>> clear
saves all the variables to the file allvariables.mat in binary format and then deletes all the variables. Entering
>> load allvariables
or
>> load allvariables.mat
loads all these variables back into MATLAB using the binary format. On the other hand, if you create a file, say
mymatrix.dat, containing the elements of a matrix and enter it into MATLAB using
>> load mymatrix.dat
you obtain a new matrix, called mymatrix, which contains these elements. Thus, the load command determines
how to read a file depending on the extension.
56
It is oversimplifed because there is no need to use three lines when one will do. However, if many
plots are to be printed then the print device can be changed once rather than in every print
command. Also, if you are printing many plots then you can easily modify the file names as in
>> i = 1;
>> file = [ 0 fiddlededum 0 , num2str(i), 0 .eps 0 ];
>> print(device, file)
>> . . .
>> i = i + 1;
>> file = [ 0 fiddlededum 0 , num2str(i), 0 .eps 0 ];
>> print(device, file)
>> . . .
Input-Output
csvread( 0 <file name> 0 )
load( 0 <file name> 0 )
Reads data into MATLAB from the named file, one row per line of
input; the numbers in each line must be separated by commas.
Reads data into MATLAB from the named file, one row per line
of input; the numbers in each line can be separated by spaces or
commas. The name of the resulting matrix is <file name>.
Similar to load but there need not be the same number of elements
in each row.
Writes out the elements of a matrix to the named file using the
same format as csvread.
Prints a plot or saves it in a file using various printer specific formats.
For example, print -deps <file name> saves the plot in the file
using encapsulated PostScript (so it can be plotted on a PostScript
laser printer).
57
Two-Dimensional Graphics
plot(x, y)
semilogx
semilogy
loglog
fplot(<function handle>,
<limits>)
)
ezplot(<function handle>
ezplot(<fnc 1>, <fnc 2>)
ezplot(<2D fnc>)
polar(r, theta)
ezpolar(<function handle>)
xlabel(<string>)
ylabel(<string>)
title(<string>)
axis
hold
linspace(a, b, n)
logspace(a, b, n)
hist(x)
stairs(x,y)
o
errorbar(x, y, e)
errorbar(x, y, l, u)
subplot(m, n, p)
shg
drawnow
4.2.
Three-Dimensional Graphics
The MATLAB command plot3 plots curves in three-dimensions. For example, to generate a helix
58
enter
>> t = linspace(0, 20*pi, 1001);
>> c = cos(t);
>> s = sin(t);
>> plot3(c, s, t)
and to generate a conical helix enter
>> t = linspace(0, 20*pi, 2001);
>> c = cos(t);
>> s = sin(t);
>> plot3(t.*c, t.*s, t)
Also, you can put a label on the z-axis by
>> zlabel(<string>)
There is also an easy plot3 command. It generates the curve x(t), y(t), z(t) for t (0, 2) by
>> ezplot3(x, y, z)
if x, y, and z have been defined using anonymous functions. Again, you change the domain of t by
specifying the additional argument [tmin, tmax].
MATLAB also plots surfaces z = f(x, y) in three-dimensions with the hidden surfaces removed. First,
the underlying mesh must be created. The easiest way is to use the command meshgrid. This combines
a discretization of the x axis, i.e., { x1 , x2 , . . . , xm }, and
the y axis, i.e., { y1 , y2 , . . . , yn }, into the rectangular mesh (xi , yj ) i = 1, 2, . . . , m , j = 1, 2, . . . , n in the xy plane. The function f can then be
evaluated at these mesh nodes. For example,
>> x = linspace(-3, 3, 61) 0 ;
>> y = linspace(-2, 2, 41) 0 ;
>> [X, Y] = meshgrid(x, y);
>> F = (X + Y).*exp( -X.*X - 2*Y.*Y );
>> mesh(X, Y, F)
generates a colored, wire-frame surface whereas
>> surf(X, Y, F)
generates a colored, filled-in surface. We discuss how to change the colors, and even how to use the colors
as another variable, in the next section.
You can change the view of a three-dimensional plot by clicking on the menu item which shows a counterclockwise rotation. Then put the mouse in the plot, hold down the left button, and begin moving it.
You can also change it by using the view command, which can be called in either of two ways:
First, you can give the angles from the origin of the plot to your eye by
view(<azimuth>, <elevation>)
where the azimuth is the angle in degrees in the xy plane measured from the y axis (so 0 is the
y axis, 90 is the x axis, 180 is the y axis, etc.) and the elevation is the angle in degrees up
from the xy plane toward the +z axis (so 0 is in the xy plane, 90 is on the +z axis, etc.).
Second, you can give the coordinates of a vector pointing from the origin of the plot to your eye by
view([x y z]), where you enter the coordinates of the vector.
If you type
>> contour(X, Y, F)
you will see contour plots of the surface. That is, you will be looking down the z axis at curves which
represent lines of constant elevation (i.e., constant z values). If we type
>> contour3(X, Y, F)
you will see contour plots of the surface in three dimensions. You can again change your view of these
curves by using the view command. These contour lines are labelled by
>> [C, h] = contour(X, Y, F)
>> clabel(C, h)
Also, contour lines are plotted at specific values by
>> contour(X, Y, F, v)
where v is a vector of the values. To obtain a filled contour plot enter
>> contourf(X, Y, F)
If you do not want to bother with generating the mesh explicitly, you can generate easy plots by
59
f (r, ) =
r + e2r sin 2
1.2 r cos 3
for r 1
is
>> f = @(r, th) ( r + exp(2*r.*sin(2*th)) ) ./ ( 1.2 - r.*cos(3*th) );
>> r = linspace(0, 1, 51);
>> th = linspace(0, 2*pi, 61);
>> [R, Th] = meshgrid(r, th);
>> [X, Y] = pol2cart(Th, R);
>> surf(X, Y, f(R, Th))
The function pol2cart transforms the polar coordinates into cartesian coordinates which can be understood by surf, mesh, or contour.
We close with an additional detail about meshgrid. It can also generate a grid in three dimensions by,
for example,
>> x = linspace(-3, 3, 61) 0 ;
>> y = linspace(-2, 2, 41) 0 ;
>> z = linspace(0, 1, 11) 0 ;
>> [X, Y, Z] = meshgrid(x, y, z);
Three dimensions is the hightest we can go with meshgrid. However, a multidimensional grid can also be
generated by
>> [X, Y, Z] = ndgrid(x, y, z);
and ndgrid can be used in any number of dimensions. The difference between the two functions is that
the order of the first two arguments is reversed. For example,
>> [X, Y] = meshgrid(1:3, 4:7)
returns
4 4 4
1 2 3
5 5 5
1 2 3
X=
, Y=
6 6 6
1 2 3
7 7 7
1 2 3
while
>> [X, Y] = ndgrid(1:3, 4:7)
returns
1
X = 2
3
1
2
3
1
2
3
4
1
2 , Y = 4
3
4
60
5
5
5
6
6
6
7
7
7
Three-Dimensional Graphics
plot3(x, y, z)
ezplot3(<fnc 1>,
<fnc 2>, <fnc 3>)
mesh(X, Y, Z)
ezmesh(<2D fnc>)
surf(X, Y, Z)
ezsurf(<2D fnc>)
view
meshgrid(x, y)
ndgrid(x, y)
pol2cart(Th, R)
zlabel(<string>)
axis
contour(X, Y, Z)
contourf(X, Y, Z)
ezcontour(<2D fnc>)
contour3(X, Y, Z)
ezcontour3(<2D fnc>)
clabel
subplot(m, n, p)
4.3.
Plots the data points in Cartesian coordinates. The general form of this
command is plot(x1, y1, z1, s1, x2, y2, z2, s2, ...) where s1,
s2, . . . are optional character strings containing information about the
type of line, mark, and color to be used.
Generates an easy plot in 3-D.
Plots a 3-D surface using a wire mesh.
Generates an easy 3-D surface using a wire mesh.
Plots a 3-D filled-in surface.
Generates an easy 3-D filled-in surface.
Changes the viewpoint of a 3-D surface plot by
view(<azimuth>, <elevation>) or view([x y z]).
Generates a 2-D grid given the x-coordinates and the y-coordinates of the
mesh lines.
Same as meshgrid except that the two arguments are reversed.
convert polar to cartesian coordinates.
Puts a label on the z-axis.
Controls the scaling and the appearance of the axes.
axis([xmin xmax ymin ymax zmin zmax]) changes the endpoints of the
axes.
Plots a contour looking down the z axis.
Plots a filled contour.
Generates an easy contour looking down the z axis.
Plots a contour in 3-D.
Generates an easy contour in 3-D.
Label contour lines generated by contour or contour3.
Remember than subplot can also be called in 3-D to put a number of
plots in one graphics window.
In the previous subsections we have discussed how to use simple graphics commands to generate basic plots. MATLAB can also do much more interesting graphics, and even publication quality graphics.
Here we discuss some of the more useful advanced features. We divide the topic into two subsections: the
first discusses the commands themselves and the second discusses how to change some of the properties of
these commands.
Note: The demonstration program shows many more of the graphics capabilities of MATLAB. Enter
demo
and then in Help Navigator click on Graphics.
First, however, we want to return to the plot command. We have already discussed plot(x, y) and
plot(x, y, LineSpec) where LineSpec contains one or more symbols which customize the line. Additionally, you can use
>> plot(x, y, 0 PropertyName 0 , PropertyValue, ...)
or
>> plot(x, y, LineSpec, 0 PropertyName 0 , PropertyValue, ...)
where ... indicates that there can be more property names and values. There are a huge number of properties which can be used. The three names and values we discuss refer to the color of the line, its width,
and the size of any markers.
0 Color 0 , ? where ? is a single character denoting one of the colors in the table.
0
Color 0 , ?...? where ?...? is the long name of one of the colors.
0
Color 0 , [r, g, b] where this denotes the color by giving its red, green, and blue intensities in the
61
62
probably should have the title include a definition of the function and you should not have to modify
the title by hand every time you change the coefficients. This can be done by
>> str = [num2str(c1), 0 *sin 0 , num2str(p1), 0 (x) + 0 , num2str(c2), ...
0
*cos 0 , num2str(p2), 0 (x) 0 ]
>> title(str)
where we use the text variable t, rather than putting the string directly into title, simply to make
the example easier to read. There is now a permanent record of the function which generated the curve.
(Alright, this isnt a great example, but its better than nothing.)
You can also put plots in a new graphics window by entering
>> figure
where the figures are numbered consecutively starting with one (and appear at the top of the window). Or
enter
>> figure(n)
and the figure will have the specific number n. This creates a new window, makes it visible, and makes it
the current target for graphics commands. You can bounce between graphics windows by entering
>> figure(n)
where n is the number of the graphics window you want to make current. New plots will now appear in
this figure. In this way much more information can be generated and viewed on the computer terminal.
Occasionally, it is useful to clear a figure. For example, suppose you divide a window into a 22 array of plotting regions and use subplot to put a plot into each region; you then save the figure into a
file. Next, you only want to put plots into two of these four regions. The difficulty is that the other two
regions will still contain the previous plots. You can avoid this difficulty by clearing the figure using
>> clf
which clears the current figure. You can clear a particular figure by clf(n) or clf(<handle>). (Handle
graphics is discussed in the next subsection.) In addition, you can clear the current figure by
>> close
or a particular figure by close(<handle>). You can also clear all the figures by
>> close all
All the above MATLAB commands can be used for 3-D graphics except for gtext. The text
command is the same as described above except that the position of the text requires three coordinates,
i.e.,
>> text(x, y, z, <string>)
As we discussed in the previous subsection, the mesh and surf commands allow us to plot a surface
in three dimensions where the colors on the surface represent its height. We can add a rectangle which
contains the correspondence between the color and the height of the surface by adding
>> colorbar
We can also let the colors represent a separate quantity C, which is also defined at each mesh point, by
changing the command to
>> mesh(X, Y, F, C)
or
>> surf(X, Y, F, C)
Each graphics window has a separate color map associated with it. This color map is simply an n3
matrix, where each element is a real number between 0 and 1 inclusive. In each row the first column
gives the intensity of the color red, the second column green, and the third column blue; these are called
the RGB components of a color. For example, we show the RGB components of cyan, magenta, yellow,
red, blue, green, white, and black in the table Customizing Lines and Markers at the beginning of this
section; for further information, enter doc colorspec. The value input to this color map is the row representing the desired color.
For mesh or surf the value of F (or of C if there is a fourth argument) is linearly rescaled so its minimum value is 1 and its maximum value is n. To see the current color map, enter
>> colormap
To change the color map, enter
>> colormap(<color map>)
where <color map> can be an explicit n3 matrix of the desired RGB components or it can be a string
63
containing the name of an existing color map. The existing color maps can be found by typing
>> doc graph3d
A useful color map for outputting to laser printers is 0 gray 0 . In this colormap all three components of
each row have the same value so that the colors change gradually from black (RGB components [0 0 0])
through gray [.5 .5 .5]) to white [1 1 1]).
MATLAB can also fill-in two-dimensional polygons using fill or three-dimensional polygons using
fill3. For example, to draw a red circle surrounding a yellow square, enter
>> t = linspace(0, 2*pi, 100);
>> s = 0.5;
>> xsquare = [-s s s -s] 0 ;
>> ysquare = [-s -s s s] 0 ;
>> fill(cos(t), sin(t), 0 r 0 , xsquare, ysquare, 0 y 0 )
>> axis equal;
To obtain a more interesting pattern replace the above fill command by
>> colormap( 0 hsv 0 );
>> fill(cos(t), sin(t), [1:100], xsquare, ysquare, [100:10:130])
Rather than entering polygons sequentially in the argument list, you can enter
>> fill(X, Y, <color>)
where each column of X and Y contain the endpoints of a different polygon. Of course, in this case the
number of endpoints of each polygon must be the same, by padding if necessary. For example, to draw a
cube with all the faces having a different solid color, input the matrices
0
1
X=
1
0
1
1
1
1
1
0
0
1
0
0
0
0
0
1
1
0
0
0
1
0
, Y=
0
1
0
0
0
1
1
0
1
1
1
1
1
0
0
1
0
0
1
1
0
0
0
0
, Z=
1
1
1
1
0
0
1
1
1
1
1
1
0
0
1
1
0
0
0
0
1
1
.
1
1
Then enter
>> fill3(X, Y, Z, [1:6])
>> axis equal
Change your orientation using view to see all six faces. Read the documentation on fill and fill3 for
more details.
We include an example which shows in detail how to modify the colormap directly when using
contourf to generate a filled contour plot. Suppose you have an interesting function which takes on
both positive and negative values. You want the more positive values to be redder and the more negative
values to be greener and a zero value to be white. This is easily done by
%%%%% script m-file: colormap example
n = 101;
% 1
f = @(x,y) .5*(sin(2*pi*(x - y. 2)). 2 + 1.5*sin(2*pi*(x. 2 + y))) ./ ...
(1 + abs(x) + abs(y));
% 2
x = linspace(-1, 1, n);
% 3
y = x;
% 4
[X, Y] = meshgrid(x, y);
% 5
F = f(X, Y);
% 6
color scale = [0:.05:1] 0 ;
% 7
C green = [color scale, ones(size(color scale)), color scale];
% 8
color scale = flipud(color scale(2:end));
% 9
C red = [ones(size(color scale)), color scale, color scale];
% 10
C = [C green; C red];
% 11
colormap(C)
% 12
contourf(X, Y, F, 20)
% 13
caxis([-1 1])
% 14
colorbar
% 15
The amusing function is defined in line 2. The array C green goes from green, i.e., (0 1 0), to white,
i.e., (1 1 1), in steps of 0.05. The array C red then goes from almost white, i.e., (1 0.95 0.95), to
64
red, i.e., (1 0 0), also in steps of 0.05. The complete array C is calculated in line 11 and the colormap
changed in line 12. The filled contour is calculated in line 13. However, the zero value of F does not correspond to white because the colors in the colormap change linearly from the minimum value of F, i.e.,
min(F(:)) = 0.58, to the maximum value of F, i.e., max(F(:)) = 0.89. Thus, the zero value of F has a
value of (1, 0.15, 0.15). This is corrected in line 14 where caxis changes the endpoints used in the colormap to a minimum of 1 and a maximum of +1; thus a value of 0 corresponds to the middle row of
C which is, in fact, white. Finally, in line 15 we attach the color map to the plot so that we can determine
the values in the contour plot.
Note: There is a linear scaling between the value of F and the corresponding color in the color map C
Rn,3 . For caxis([f min, f max]) the value f is first modified to lie in the interval [f min, f max]
by f mod = max([min([f, f max]), f min]). Then the row ic of C is calculated by
ic = fix
(f mod f min)(n 1)
f max f min
+1.
MATLAB can also plot a two-dimensional image (i.e., a picture) which is represented by a matrix
X Rmn . The (i, j)th element of X specifies the color to use in the current color map. This color appear in the (i, j)th rectilinear patch in the plot. For example, to display the color image of a clown enter
>> load clown
>> image(X);
>> colormap(map)
The image command inputs the matrix X and the colormap map from clown.mat. Then the image is
displayed using the new color map. Similarly,
>> load earth
>> image(X);
>> colormap(map);
>> axis image
displays an image of the earth. (The axis command forces the earth to be round, rather than elliptical.)
(In the demonstration program, after clicking on Visualization double-click on Image colormaps to see
the images which you can access in MATLAB and the existing color maps.)
Incidentally, imagesc is a similar command, but with one important difference. They both use the
same colormap, i.e., the current colormap, but the matrix A Rmn , which should have integer elements,
has a different meaning. In
>> image(C)
each element of C, say cij , corresponds to this entry in the colormap, and let it have ` colors, i.e., it is
a R`3 matrix. If cij < 1 or cij > `, this command treats them as if cij = 1 or cij = `, respectively.
However,
>> imagesc(C)
first determines the minimum element m1 and the maximum element m2 in C. The minimum value is
scaled linearly to the first color in the colormap, and the maximum value to the last color.
Incidentally, many grayscale or color images can be imported into MATLAB by
>> X = imread(<filename>);
or
>> [X, map] = imread(<filename>);
where map is the associated colormap for the image. The types of files which can be imported are: bmp,
cur, fts, fits, gif, hdf, ico, j2c, j2k, jpf, jpx, jpg, jpeg, pmb, pcx,
pgm, png, pnm, pps, ras, tif, tiff, or wxd. They can be shown by image or imshow.
They can be exported by imwrite.
65
clf
colorbar
colormap
demo
figure
fill(x, y, <color>)
fill3(x, y, z, <color>)
image
imagesc
imread
imshow
imwrite
Change the scaling used in the color map so that the value of
v min corresponds to the first row of the colormap and v max to the
last row. Values outside this interval map to the closest endpoint.
Clear a figure (i.e., delete everything in the figure)
Adds a color bar showing the correspondence between the value and
the color.
Determines the current color map or choose a new one.
Runs demonstrations of many of the capabilities of MATLAB.
Creates a new graphics window and makes it the current target.
figure(n) makes the nth graphics window the current target.
Fills one or more polygons with the color or colors specified by the
vector or string <color>.
Fills one or more 3D polygons with the color or colors specified by the
vector or string <color>.
Plots a two-dimensional matrix.
Plots a two-dimensional matrix and scales the colors
Import an image from a graphics file.
Display an image.
Export an image to a graphics file.
4.4.
In this subsection we briefly discuss handle graphics. This is a collection of low-level graphics commands
which do the actual work of generating graphics. In the previous parts of this section we have mainly
discussed high-level graphics commands which allow us to create useful and high quality graphical images very easily. The low-level commands allow us to customize these graphical images, but at the cost of
having to get much more involved in how graphical images are actually created. This subsection will be
quite short because we do not want to get bogged down in this complicated subject. Instead, we will only
discuss a few of what we consider to be the more useful customizations.
In handle graphics we consider every component of a graphical image to be an object, such as a subplot,
an axis, a piece of text, a line, a surface, etc. Each object has properties and we customize an object by
changing its properties. Of course, we have to be able to refer to a particular object to change its properties, and a handle is the unique identifier which refers to a particular object. (Each handle is a unique
floating-point number.)
We will use a small number of examples to explain handle graphics. There are many properties of the
text that can be changed in the text command by
>> text(xpt, ypt, <string>, 0 <Prop 1> 0 , <Value 1>, 0 <Prop 2> 0 , <Value 2>, . . . )
or
>> h = text(xpt, ypt, <string>);
>> set(h, 0 <Prop 1> 0 , <Value 1>, 0 <Prop 2> 0 , <Value 2>, . . . )
where <Prop ?> is the name of one of the properties for the text object and <Value ?> is one of the
66
allowed values. (We show some names and values in the following table.) We have shown two ways to customize the properties. In the former all the properties are set in the text command. In the latter the
text command creates an object, using its default properties, with handle h. The set command then
changes some of the properties of the object whose handle is h. For example, entering
>> set(h, 0 Color 0 , 0 r 0 , 0 FontSize 0 , 16, 0 Rotation 0 , 90)
results in a large, red text which is rotated 90 . You can also change the default properties for gtext,
xlabel, ylabel, zlabel, and title.
Text Properties
Clipping
Color
FontName
FontSize
HorizontalAlignment
Rotation
VerticalAlignment
on (default) Any portion of the text that extends outside the axes rectangle is clipped
off No clipping is done.
A three-element vector specifying a color in terms of its red, blue, and
green components, or a string of the predefined colors.
The name of the font to use. (The default is Helvetica.)
The font point size. (The default is 10 point.)
left (default) Text is left-justified
center Text is centered.
right Text is right justified.
The text orientation. The property value is the angle in degrees.
top The top of the text rectangle is at the point.
cap The top of a capital letter is at the point.
center (default) The text is centered vertically at the point.
baseline The baseline of the text is placed at the point.
bottom The bottom of the text rectangle is placed at the point.
The more common way of customizing parameters is by using the set command. The two functions
get and set are used to obtain the value of one parameter and to set one or more parameters. For example, to get the font which is presenting being used enter
>> s = get(h, 0 FontName 0 )
and the string s now contains the name of the font. The two arguments to get are the handle of the
object desired and the name of the property.
There are two other commands which can obtain a handle:
>> hf = gcf
returns the handle of the current figure and
>> ha = gca
returns the handle of the current axes in the current figure. For a simple example which uses handle
graphics, suppose we want to plot the function y = esin x for x [0, 2] and we want the horizontal axis
to have major tick marks at every /2 and we want these tick marks labelled. We do this by
>> x = linspace(0,2*pi,101);
>> plot(x,exp(sin(x)))
>> set(gca, 0 XTick 0 , [0:pi/2:2*pi])
>> set(gca, 0 XTickLabel 0 , { 0 0 0 , 0 .5*pi 0 , 0 pi 0 , 0 1.5*pi 0 , 0 2*pi 0 })
>> set(gca, 0 XMinorTick 0 , 0 on 0 )
>> set(gca, 0 TickDir 0 , 0 out 0 )
where the third line puts major tick marks at multiples of /2, the fourth line puts the labels shown
at each tick mark, the fifth line adds small tick marks between the labelled ticks, and the last line puts
the tick marks outside the plot boxed area. (At present, we cannot use TEXcommands in XTickLabel to
generate a Greek .)
There is one case where we frequently use handle graphics. When a figure is printed, the graphical images do not fill the entire page. The default size is approximately 6.5 inches wide and 5.5 inches high.
When we want to use the full size of a sheet of paper we use
67
>>
or
>> figure(n)
>> set(gcf, 0 PositionPaper 0 , [0 0 8.5 11])
since the default units for this property are inches. This sets the graphical images to use the full paper
size (the position is given as [left bottom width height] ) with a one inch border. This is frequently
useful if subplot is being used to put a number of plots on a page.
Finally, if subplot is being used, it is sometimes useful to put a title on the entire page, not just in
each subplot. This can be done by
>> axes handle = axes( 0 Position 0 , [0 0 1 0.95], 0 Visible 0 , 0 off 0 );
>> title handle = get(axes handle, 0 Title 0 );
>> set(title handle, 0 String 0 , <title>, 0 Visible 0 , 0 on 0 );
The first line specifies a rectangle for the axes in normalized units (so that
[left bottom width height] = [0 0 1 1] is the full size of the figure). The axes are invisible because
they are only being created so that a title can go on top. The second line gets the handle for the title
object of the new axes. The third line puts <title> into the title object and makes it visible.
Advanced Topic: Properties
get(<handle>, 0 <Prop> 0 )
set(<handle>, 0 <Prop 1> 0 ,
<Value 1>, ...)
gca
gcf
4.5.
Return the current value of the property of the object with this
handle. Note: Case is unimportant for property names.
Set the property, or properties, of the object with this handle. Note:
Case is unimportant for property names.
The current axes handle.
The current figure handle.
Warning: The codes in this subsection use for loops and if tests which are discussed in Section 8.1.
They also contain primary functions, subfunctions, and nested functions which are discussed in
Section 8.3.
MATLAB makes it very easy to create a graphical user interface for a program. Of course, it is usually
even easier for the programmer to have the users interact with a program by entering data (i.e., typing) in
the workspace. However, this is often not easier for the users. Thus, the programmer must make the decision whether or not to spend the time required to write a good GUI. Note that this not only includes the
time required to code the GUI but also the possibly much longer time required to design the GUI.
In this subsection we are only concerned with how to write a GUI not how to write a good one.
A GUI is a graphical display which allows a user to interact pictorally with a program. It usually
consists of three elements:
(1) Components: These are the tools which enable the user to interact with a program, and include push
bottoms, sliders, radio buttons, check boxes, editable text, pop-up menus, listboxes, and toggle buttons. Such an interaction is called an event, and a program which responds to events is event driven.
Components can also include plots and/or tables which allow the program to interact with the user,
i.e., show results of the program.
(2) Figures: All of these components must be arranged within a figure, which is a graphical window
which is separate from the MATLAB window which arises when MATLAB is first executed.
(3) Callbacks: These are pieces of code which enable the user to interact with the program. For example, when a user uses a mouse to click a button, this triggers an event, but it is not directly connected
with the program. It is the responsibility of the programmer to write the code which specifies how
how each event affects the program.
The simple way to write a GUI is to type
>> guide
This launches a GUI which you can use to write your own GUI. The largest part of this window is a rectangular and gridded canvas into which the various components are placed. To the left of this canvas is a
68
list of all the possible components, either using icons or their actual names. You build your own GUI by
positioning any or all of these components in the canvas. When done, you save your GUI under its own
name. You have created two files, the fig file (i.e., the extension is fig) and the function file (i.e., the
extension is m). So far you have only created the first two elements of a GUI: the components and the figure. The third part, i.e., the callbacks, are your responsibility. Your program must be integrated into this
function file and you must also write the code to connect an event with its effect on your now integrated
program.
We will not discuss guide any further, and we will not discuss how to integrate your program and
code the callbacks. This is well-documented in MATLAB, including an explicit example which is called
simple gui. Read this documentation carefully and work through this explicit example yourself.
What we will discuss is how to use the non-simple way to write a GUI, namely by writing all the actual
commands which setup the figure, the components, and the callbacks. Since this can get quite complicated, we only show two very simple examples. We will not write any GUIs from scratch. Instead, we will
show how to take a simple figure which is created by the plot command and add something to it. Thus,
we will not attempt to describe all the commands which are involved in generating a GUI, but only a very
small subset.
For the first example consider the function
function rippling(a, b)
n = 101;
x = linspace(-2, 2, n);
y = x;
[X, Y] = meshgrid(x, y);
for t = 0:.01:10
at = a + .8*t;
bt = b + .4*t;
Yxy = Y + .25*sin(2*(exp(X/2)-1));
Z = sin((at + .1*X + .2*Yxy).*X. 2 + ...
bt*(sin(2*Yxy) + .2*Yxy).* Yxy. 2);
surf(X, Y, Z);
view([-45, 60])
xlabel( 0 x 0 )
ylabel( 0 y 0 )
zlabel( 0 z 0 )
title(t)
drawnow
end
69
The button could have been positioned in a more pleasing location in the GUI because there is lots of avail
space in a three-dimensional plot. However, in a two-dimensional plot this is a safe location.
70
71
The function button is also a nested function, whereas it was a subfunction in pause rippling. Since it does
not modify a variable which is needed by the primary function, it can be either. Since edit a and edit b need to
72
tion, i.e., the function rippling is executed, and the label is changed to Pause.
73
GUI
guide
uicontrol(<handle>, 0 <Prop 1> 0 ,
<Value 1>, ...)
uiwait
uiresume
4.6.
There are a number of different ways to make movies in MATLAB. We show some of them in the following code which is a modification of the MATLAB function life (which is copyrighted by The MathWorks, Inc). The documentation inside life follows.
LIFE MATLABs version of Conways Game of Life.
Life is a cellular automaton invented by John Conway that involves live and dead cells in a
rectangular, two-dimensional universe. In MATLAB, the universe is a sparse matrix that is
initially all zero.
Whether cells stay alive, die, or generate new cells depends upon how many of
their eight possible neighbors are alive. By using sparse matrices, the calculations required
become astonishingly simple. We use periodic (torus) boundary conditions at the edges of the
universe. Pressing the Start button automatically seeds this universe with several small
random communities. Some will succeed and some will fail.
C. Moler, 7-11-92, 8-7-92.
Adapted by Ned Gulley, 6-21-93
Copyright 1984-2004 The MathWorks, Inc.
$Revision: 5.10.4.1 $ $Date: 2004/08/16 01:38:30 $
This code strips out the part of life which does the actual iterations and shows how to save the results in an avi movie. Two different methods are shown explicitly in the code. The first initializes the
movie, adds frames explicitly to the movie, and then closes it. The second only saves each frame in a jpg
file; an external software package must then be used to combine these files into a movie. A third method,
which is a slight modification of the first, is then discussed.
The advantage of the second method is that no additional memory is required; the disadvantage is that
each frame must be individually saved to disk. Another disadvantage of the first and third methods in
Linux and Mac is that the movie file which is generated is not compressed. An external software package
74
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
16
17
18
19
20
22
21
22
23
24
25
26
27
28
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
76
Movies
avifile
addframe
getframe
close (<avifile>)
movie
movie2avi
saveas
4.7.
Be Able To Do
After reading this section you should be able to do the following exercises. The solutions are given on
page 163.
1. Plot ex and one of its Taylor series approximations.
(a) Begin by plotting ex for x [1, +1].
(b) Then plot
p3 (x) = 1 +
x
x2
x3
+
+
1!
2!
3!
5.1.
a11
a
21
...
am1
x2
xn
a12
a22
..
.
am2
...
...
..
.
...
a1n
a2n
.
..
amn
rhs
b1
b2
.
..
.
bm
Warning: It is very important to realize that an augmented matrix is not a matrix (because the
operations we apply to augmented matrices are not the operations we apply to matrices). It is
simply a linear system of equations written in shorthand: the first column is the coefficients of
the x1 term, the second column is the coefficients of the x2 term, etc., and the last column is
the coefficients on the right-hand side. The vertical line between the last two columns represents the equal sign. Normally, an augmented matrix is written without explicitly writing the
header information; however, the vertical line representing the equal sign should be included to
explicitly indicate that this is an augmented matrix.
78
rref operates on this augmented matrix to make as many of the elements as possible zero by using allowed operations on linear equations these operations are not allowed on matrices, but only on linear
systems of equations. The result is an augmented matrix which, when written back out as a linear system
of equations, is particularly easy to solve. For example, consider the system of equations
x1 + 2x2 + 3x3 = 1
4x1 + 5x2 + 6x3 = 1
7x1 + 8x2 + 10x3 = 0 ,
which is equivalent to the matrix equation
1
A = 4
7
Ax = b where
2 3
1
5 6 and b = 1 .
8 10
0
x2
2
5
8
x3
3
6
10
rhs
1
1 .
0
(We have included the header information for the last time.) Entering
>> rref([A b])
returns the augmented matrix
1 0 0
2
0 1 0
3 .
0 0 1
1
Clearly, the solution of the linear system is x1 = 2, x2 = 3, and x3 = 1.
Of course, you could just as easily have found the solution by
>> x = A\b
so let us now consider the slightly different linear system
x1 + 2x2 + 3x3 = 1
4x1 + 5x2 + 6x3 = 1
7x1 + 6x2 + 9x3 = 1 ,
This is equivalent to the matrix equation Ax = b where
1
1 2 3
A = 4 5 6 and b = 1 .
7 8 9
1
Since A is a singular matrix, the linear system has
The augmented matrix for this linear system is
1 2
4 5
7 8
1
1 .
0
Entering
>> rref([A b])
returns the augmented matrix
1 0
0 1
0 0
1
2
0
79
1
1 ,
0
so the solution of the linear system is x1 = 1 + x3 and x2 = 1 2x3 for any x3 R (or C if desired).
In vector form, the solution is
x1
1 + x3
1
x3
1
1
x2 = 1 2x3 = 1 + 2x3 = 1 + x3 2 .
x3
x3
0
x3
0
1
Suppose you modify the matrix equation slightly by letting b = (1, 1, 0)T . Now entering
>> rref([A b])
results in the augmented matrix
1 0 1
1
0 1 2
1 .
0 0 0
1
Since the third equation is 0 = 1, there is clearly no solution to the linear system.
Warning: The command rref does not always give correct results. For example, if
0.95 0.03
C=
0.05 0.97
then the matrix I C is singular (where I is the identity matrix). However, if you solve
(I C)x = 0 by
>> C = [0.95 0.03; 0.05 0.97];
>> rref([eye(size(C))-C, [0 0] 0 ])
MATLAB displays
ans =
1
0
0
0
1
0
which indicates that the only solution is x = 0. On the other hand, if you enter
>> C = [0.95 0.03; 0.05 0.97]; b = 1;
>> rref([eye(size(C))-C [b 0] 0 ])
then MATLAB realizes that I C is singular. Clearly there is some value of b between 0
and 1 where MATLAB switches between believing that I C is non-singular and singular.
Solving Linear Systems
linsolve(A, b, <properties>)
rref
To understand this switch, look at the actual coding of rref. It uses the variable tol to determine
whether an element of the augmented matrix
b1
0.05 0.03
0.05 0.03
b2
is small enough that it should be set to 0. tol is (essentially) calculated by
tol = max(size(<augmented matrix>)) * eps * norm(<augmented matrix>, inf);
The maximum of the number of rows and columns of the augmented matrix, i.e., max(size(...)), is multiplied
by eps and this is multiplied by the size of the augmented matrix. ( norm in Section 7.) Since b is the last
column of the augmented matrix, the size of this matrix depends on the size of the elements of b. Thus, the
determination whether a number should be set to 0 depends on the magnitude of the elements of b.
You can obtain the correct answer to the homogeneous equation by entering
>> rref([eye(size(C))-C [0 0] 0 ], eps)
which decreases the tolerance to eps.
80
5.2.
We have mentioned repeatedly that computers cannot add, subtract, multiply, or divide corUp until now, the errors that have resulted have been very small. Now we present two examples
where the errors are very large.
In this first example, the reason for the large errors is easy to understand. Consider the matrix
rectly!
1
A = 4
7
3
6 ,
9+
2
5
8
which is singular when = 0 and nonsingular otherwise. But how well does MATLAB do when 1?
Enter
>> ep = input( 0 ep = 0 ); A = [1 2 3; 4 5 6; 7 8 9+ep]; inv(A)*A - eye(size(A))
so that the final matrix should be O. (Do not use eps for the name of this variable or you will change
the predefined variable eps. Other possible names are epsi or epsilon.) Begin by letting = 0 and
observe that the result displayed is nowhere close to the zero matrix! However, note that MATLAB is
warning you that it thinks something is wrong with the statement
Warning: Matrix is close to singular or badly scaled.
Results may be inaccurate. RCOND = 1.541976e-18.
( RCOND is its estimate of the inverse of the condition number. See cond in Section 7 for more details.)
Now choose some small nonzero values for and see what happens. How small can be before MATLAB warns you that the matrix is close to singular or badly scaled? In this example, you know that the
matrix is close to singular if is small (but nonzero) even if MATLAB does not. The next example is
more interesting.
For the second example, consider the Hilbert matrix of order n, i.e.,
1
1/2
1/3
Hn =
.
.
.
1/n
1/2
1/3
1/4
..
.
1/3
1/4
1/5
..
.
1/(n + 1)
1/(n + 2)
.
.
.
.
. .
. .
. .
. .
. . .
1/n
1/(n + 1)
1/(n + 2)
..
.
1/(2n 1)
Hx = b
for a given b. How close is the numerical solution to the exact solution? Of course, the problem is: how
can you know what the analytical solution is for a given b? The answer is to begin with x and calculate b by b = Hx. Then solve Hx = b for x and compare the final and initial values of x. Do this in
MATLAB by
>> n = 10; x = rand(n, 1); b = H*x; xnum = H\b
and compare x with xnum by calculating their difference, i.e.,
>> x - xnum
The result is not very satisfactory: the maximum difference in the elements of the two vectors is usually
somewhere between 105 and 103 . That is, even though all the calculations have been done to approximately 16 significant digits, the result is only accurate to three to ve significant digits! (To see how
81
much worse the result can be, repeat the above commands for n = 12.)
It is important to realize that most calculations in MATLAB are very accurate. It is not that solving
a matrix equation necessarily introduces lots of round-off errors; instead, Hilbert matrices are very unstable matrices working with them can lead to inaccurate results. On the other hand, most matrices
are quite stable. For example, if you repeat the above sequence of steps with a random matrix, you find
that the results are quite accurate. For example, enter
>> n = 1000; R = rand(n); x = rand(n, 1); b = R*x; xnum = R\b; max(abs(x - xnum))
The results are much more reassuring, even though n is 100 times as large for this random matrix as for
the Hilbert matrix and even though there are over 600,000 times as many floating point operations
needed to calculate x by Gaussian elimination for this random matrix!
Note: By entering all the commands on one line, it is easy to repeat this experiment many times for different random numbers by simply rerunning this one line.
5.3.
If A Cmn where m > n, Ax = b is called an overdetermined system because there are more equations than unknowns. In general, there are no solutions to this linear equation. However, you can find a
best approximation by finding the solution for which the vector
r = Ax b
which is called the residual, is smallest in Euclidean length; that is,
norm(r)
X
n
ri2
1/2
i=1
is minimized. (The norm function is discussed in Sections 2.8 and 7.) This is called the least-squares
solution. This best approximation is calculated in MATLAB by typing
>> A\b
Analytically, the approximation can be calculated by solving
AT Ax = AT b .
However, numerically this is less accurate than the method used in MATLAB.
Note that this is the same command used to find the solution to a square linear system. This cannot be
the intent here since A is not a square matrix. Instead, MATLAB interprets this command as asking for
the least-squares solution. Again, this command only makes sense if there is a unique solution which minimizes the length of the vector Ax b. If there are an infinite number of least-squares solutions, MATLAB
warns you of this fact and then returns one of the solutions. For example, if
2
1 2 3
1
4 5 6
A=
and b =
2
7 8 9
4
10 11 12
then Ax = b has no solutions, but has an infinite number of least-square approximations. If you enter
>> A\b
the response is
Warning: Rank deficient, rank = 2 tol = 1.4594e-14.
It also returns the solution (1/4 , 0, 29/60 )T (after using the MATLAB command rats which we discuss
below), which is one particular least-squares approximation. (If A is complex, solve AH Ax = AH b.)
Occasionally, if there are an infinite number of least-squares approximations, the solution desired is the
smallest one, i.e., the x for which the length of the vector x is minimized. This can be calculated using the pseudoinverse of A, also called the Moore-Penrose pseudoinverse, which is denoted by A+ . Since
A is not square, it cannot have an inverse. However, the pseudoinverse is the unique nm matrix which
satisfies the Moore-Penrose conditions:
AA+ A = A
82
6. File Input-Output
A+ AA+ = A+
(AA+ )T = AA+
(A+ A)T = A+ A
In particular, if A is a square nonsingular matrix, then A+ is precisely A1 . This pseudoinverse is calculated in MATLAB by entering
>> pinv(A)
The reason for mentioning the pseudoinverse of A is that the least-squares approximation to Ax = b can
also be calculated by
>> pinv(A)*b
If there are an infinite number of least-squares approximations, this returns the one with the smallest
length. (In the previous example it is (13/45, 7/90, 4/9)T .)
Next, suppose that A Cmn with m < n. Ax = b is called an underdetermined system because there
are less equations than unknowns. In general, there are an infinite number of solutions to this equation.
We can find one particular solution by entering
>> A\b
This solution will have many of its elements being 0. We can also find the solution with the smallest
length by entering
>> pinv(A)*b
Warning: It is possible for an overdetermined system to have one or even an infinite number of solutions
(not least-squares approximations). It is also possible for an underdetermined system to have
no solutions.
One command which is occasionally useful is rats. If all the elements of A and b are rational numbers, then the solution and/or approximation obtained is usually a rational number, although stored as
a floating-point number. This command displays a close rational approximation to the floating-point
number, which may or may not be the exact answer. For example, entering
>> rats(1/3 - 1/17 + 1/5)
results in the text variable 121/255, which is the correct answer.
Warning: Be careful when using this command. rats(sqrt(2)) makes no sense (as was known to
Pythagoras).
Solving Linear Systems
A\b
pinv(A)
rats(x)
6. File Input-Output
In Section 4.1 we discussed the csvread and csvwrite commands which allow simple input from and
output to a file. The MATLAB commands fscanf and fprintf, which behave very similarly to their C
counterparts, allow much finer control over input and output. Before using them a file has to be opened
by
>> fid = fopen( 0 <file name> 0 , <permission string>)
where the file identifier fid is a unique nonnegative integer attached to the file. (Three file identifiers always exist as in C: 0 is the standard input, 1 is the standard output, and 2 is the standard error.) The
permission string specifies how the file is to be accessed:
83
6. File Input-Output
0
r0
w0
x =
-23.6000
x = -23.6000
-2.360000e+01
Note: It is easy to print the matrix A in the MATLAB workspace as we just described. However, it is
a little more difficult to print it to a file. The following works and can be entered on one line, although it is actually a number of statements.
>> Str=num2str(A);for i = [1:size(Str, 1)] fprintf(fid, 0 %s\n 0 , Str(i,:));end
To read formatted data from a file, enter
>> A = fscanf(fid, <format string>, <size>)
The data is read from the file specified in a previous fopen command according to the format string and
put into the matrix A. The size argument, which puts an upper limit on the amount of data to be read, is
optional. If it is a scalar, or is not used at all, A is actually a vector. If it is [m n], then A is a matrix of
this size.
84
Advanced Input-Output
fopen( 0 <file name> 0 ,
<permission string>)
fclose(fid)
fscanf(fid, <format string>)
fprintf(fid, <format string>,
<variable 1>,...)
fprintf(<format string>,
<variable 1>,...)
Opens the file with the permission string determining how the
file is to be accessed. The function returns the file identifier,
which is a unique nonnegative integer attached to the file.
Closes the file with the given file identifier.
Behaves very similarly to the C command in reading data from
a file using any desired format.
Behaves very similarly to the C command in writing data to a
file using any desired format.
Behaves very similarly to the C command in displaying data on
the screen using any desired format.
for p [1, ]
or
condF (A) = kAkF kA1 kF .
85
It is calculated in MATLAB by
>> cond(A, p)
where p is 1, 2, Inf, or 0 fro 0 . If p = 2 the command can be shortened to
>> cond(A)
Note that the calculation of the condition number of A requires the calculation of the inverse of A.
The MATLAB command condest approximates the condition number without having to calculate this
inverse. See the discussion of this command below for further information on when it might be preferable.
Note: Sometimes we want to solve, or find the best approximation to, Ax = b when A Cmn is not a
square matrix. (This is discussed in detail in Section 5.3.) Since we still want to know the accuracy
of any solution, we want to generalize the condition number to nonsquare matrices. This is done by
defining the condition number of a nonsquare matrix in the 2-norm to be the ratio of the largest to
the smallest singular value of A, i.e., 1 /min{m,n} .
condest
Note: Read the discussion on cond above first.
The calculation of the condition number of A Cnn requires the calculation of its inverse. There are
two reasons this might be inadvisable.
The calculation of A1 requires approximately 2n3 flops, which might take too long if n is very
large.
If A is a sparse matrix (i.e., most of its elements are zero), we discuss in Section 9 how to store only
the nonzero elements of A to conserve storage. (For example, if n = 10,000 and A is tridiagonal ,
the number of nonzero elements in A is approximately 30,000 but the total number of elements in A
is 100,000,000.) Since the inverse of a sparse matrix is generally much less sparse (in fact it may have
no zero elements at all), MATLAB may not be able to store A1 .
The command condest calculates a lower bound to the condition number of a matrix in the 1-norm without having to determine its inverse. This approximation is almost always within a factor of ten of the
exact value.
When MATLAB calculates A\b or inv(A), it also calculates condest(A). It checks if its estimate
of the condition number is large enough that A is likely to be singular. If so, it returns an error message
such as
Warning: Matrix is close to singular or badly scaled.
Results may be inaccurate. RCOND = 2.055969e-18.
where RCOND is the inverse of condest(A).
det
Let A Cnn . The determinant of A is calculated by
>> det(A)
det(A) = 0 if and only if A is singular. However, due to round-off errors it is very unlikely that you will
obtain 0 numerically unless all the entries to A are integers. For example, consider the matrix
C=
0.95
0.05
0.03
0.97
.
A matrix is tridiagonal if its only nonzero elements occur on the main diagonal or on the first diagonal above
or below the main diagonal
86
returns 10100 , but it is also not singular. (The singular value decomposition, which is described below, is
a much better method for determining if a square matrix is singular.)
eig
Let A Cnn . A scalar C is an eigenvalue of A if there exists a nonzero vector v Cn such that
Av = v ;
v is called the eigenvector corresponding to . There are always n eigenvalues of A, although they need
not all be distinct. MATLAB will very happily calculate all the eigenvalues of A by
>> eig(A)
It will also calculate all the eigenvectors by
>> [V, D] = eig(A)
D Cnn is a diagonal matrix containing the n eigenvalues on its diagonal and the corresponding eigenvectors are found in the same columns of the matrix V Cnn .
Note: This is the first time we have had a function return more than one argument. We discuss this notation in detail in Section 8.3. For now, we simply state that when [V, D] occurs on the right side
of the equal sign it means the matrix whose first columns come from V and whose last columns
come from D. However, on the left side of the equal sign it means that the function returns two
arguments where the first is stored in the variable V and the second in D.
eig can also calculate all the eigenvalues of the generalized eigenvalue problem
Ax = Bx
by
>> eig(A, B)
A matrix is defective if it has less eigenvectors than eigenvalues. MATLAB normally cannot determine
when this occurs. For example, the matrix
B=
1
0
1
1
is defective since it has two eigenvalues, both of which are 1, but it only has one eigenvector, namely
(1, 0)T . If you enter
>> B = [1 1; 0 1]; [V, D] = eig(B)
MATLAB calculates the two eigenvalues correctly, but it finds the two eigenvectors (1, 0)T and
T
(1, 2.21016 ) . Clearly the latter eigenvector should be (1, 0)T so that, in fact, there is only one
eigenvector.
Note: If A is a sparse matrix, you cannot use eig. You either have to use the function eigs or do
eig(full(A)).
eigs
Note: Read the discussion on eig above first.
Frequently, you do not need all the eigenvalues of a matrix. For example, you might only need the
largest ten in magnitude, or the five with the largest real part, or the one which is smallest in magnitude,
or . . . . Or you might only need a few of the generalized eigenvalues of Ax = Bx. eigs can do all of this.
Of course, this means that there are numerous possible arguments to this function so read the documentation carefully.
Why not just use eig anyway? Calculating all the eigenvalues of a nonsymmetric A Rnn requires
(very) approximately 10n3 flops, which can take a very long time if n is very large. On the other hand,
calculating only a few eigenvalues requires many, many fewer flops. If A is a full matrix, it requires cn2
flops where c is of reasonable size; if A is a sparse matrix, it requires cn flops.
Note: If A is sparse, you cannot use eig you will first have to do eig(full(A)).
87
Also, this command generates lots of diagnostic output. To calculate the largest 3 eigenvalues of A in
magnitude without generating any diagnostics, enter
>> op.disp = 0
>> eigs(A, 3, 0 LM 0 , op)
( op.disp is a structure, which was discussed in Section 3.4.)
inv
To calculate the inverse of the square matrix A Cnn enter
>> inv(A)
The inverse of A, denoted by A1 , is a matrix such that AA1 = A1 A = I, where I Rnn is the identity matrix. If such a matrix exists, it must be unique.
MATLAB cannot always tell whether this matrix does, in fact, exist. For example, the matrix
1 2
A=
2 4
does not have an inverse. If you try to take the inverse of this matrix, MATLAB will complain that
Warning: Matrix is singular to working precision.
It will display the inverse matrix, but all the entries will be Inf.
The above matrix was very simple. The matrix
1 2 3
A = 4 5 6
7 8 9
(7.1)
also does not have an inverse. If you ask MATLAB to calculate the inverse of A, it will complain that
Warning: Matrix is close to singular or badly scaled.
Results may be inaccurate. RCOND = 2.055969e-18.
( RCOND is the inverse of a numerical approximation to the condition number of A; see condest above.)
That is, MATLAB is not positive that A is singular, because of round-off errors, but it thinks it is likely.
However, MATLAB still does try to calculate the inverse. Of course, if you multiply this matrix by A
the result is nowhere close to I. (Try it!) In other words, be careful and read (and understand) all
warning messages.
lu
Let A Cnn . Then there exists an upper triangular matrix U, a unit lower triangular matrix L , and
a permutation matrix P such that
LU = PA .
The MATLAB command lu calculates these matrices by entering
>> [L, U, P] = lu(A)
If A is invertible, all the elements of U on the main diagonal are nonzero. If you enter
>> A = [1 2 3; 4 5 6; 7 8 9]; [L, U, P] = lu(A)
where A is the singular matrix defined earlier, u33 should be zero. Entering
>> U(3,3)
displays 1.1102e-16, which clearly should be zero as we discussed in Section 1.5.
norm
The norm of a vector or matrix is a nonnegative real number which gives some measure of the size of
the vector or matrix. (It was briefly discussed in Section 2.8.) The pth norm of a vector is defined by
1/p
n
X
|xi |
if p [1, )
i=1
kxkp =
if p = .
max |xi |
1in
A unit lower triangular matrix is lower triangular and, in addition, all the elements on the main diagonal
are 1.
88
for p [1, ]
X
m X
n
|aij |
1/2
i=1 j=1
89
It is calculated by
>> [Q, R] = qr(A, 0)
This is equivalent to applying the Gram-Schmidt algorithm to A.
pinv
The Moore-Penrose pseudoinverse has already been discussed in Section 5.3. We include it here for
completeness. It is calculated by using the singular value decomposition, which we discuss below.
rank
Let A Cmn . The rank of A is the number of linearly independent columns of A and is calculated by
>> rank(A)
This number is calculated by using the singular value decomposition, which we discuss below.
svd
Let A Rmn . A can be decomposed into
A = UVT
where U Rmm and V Rnn are orthogonal matrices and Rmn is a diagonal matrix (although
not necessarily square) with real nonnegative elements in decreasing order. That is,
1 2 min{m,n} 0 .
(If A Cmn then U Cmm and V Cnn are unitary matrices and Rmn is again a diagonal
matrix with real nonnegative elements in decreasing order.) These matrices are calculated by
>> [U, S, V] = svd(A)
The diagonal elements of are called the singular values of A. Although A need not be a square matrix, both AT A Rnn and AAT Rmm are square symmetric matrices. (If A is complex, AH A and
AAH are both square Hermitian matrices.) Thus, their eigenvalues are nonnegative. Their nonzero eigenvalues are the squares of the singular values of A. In addition, the eigenvectors of AT A are the columns of
V and those of AAT are the columns of U. (If A is complex, the eigenvectors of AH A are the columns of
V and those of AAH are the columns of U.)
The best numerical method to determine the rank of A is to use its singular values. For example, to see
that
1 2 3 4
A = 5 6 7 8
9 10 11 12
has rank 2, use the svd command to find that the singular values of A are 25.4368, 1.7226,
and 8.10791016 . Clearly the third singular value should be 0 and so A has 2 nonzero singular values
and so has a rank of 2. On the other hand, the Hilbert matrix of order 15 has singular values
1.8100 , 4.3101 , 5.7102 , 5.6103 , 4.3104 , 2.7105 , 1.3106 , 5.5108 ,
1.8109 , 4.71011 , 9.31013 , 1.41014 , 1.41016 , 1.21017 , and 2.41018
according to MATLAB. Following Principle 1.2, you can see there is no separation between the singular
values which are clearly not zero and the ones which are close to eps. Thus, you cannot conclude that
any of these singular values should be set to 0. Our best guess is that the rank of this matrix is 15.
The eigenvalues of a real square symmetric matrix are nonnegative. (The eigenvalues of a complex square
Hermitian matrix are real and nonnegative.)
For example, if m > n there are n singular values and their squares are the eigenvalues of AT A. The m
eigenvalues of AAT consist of the squares of these n singular values and mn additional zero eigenvalues.
In fact, it can be proven that the Hilbert matrix of order n is nonsingular for all n, and so its rank is
truly n. However, if you enter
>> rank( hilb(15) )
you obtain 12, so that MATLAB is off by three.
90
inv(A)
lu(A)
norm(v)
norm(A)
null(A)
orth(A)
qr(A)
rank(A)
svd(A)
8. Programming in MATLAB
Using the commands we have already discussed, MATLAB can do very complicated matrix operations.
However, sometimes there is a need for finer control over the elements of matrices and the ability to test,
and branch on, logical conditions. Although prior familiarity with a high-level programming language is
useful, MATLABs programming language is so simple that it can be learned quite easily and quickly.
8.1.
MATLAB has four flow control and/or branching instructions: for loops, while loops, if-else
branching tests, and switch branching tests.
Notation: All of these instructions end with an end statement, and it is frequently difficult to determine
the extent of these instructions. Thus, it is very important to use indentation to indicate the
structure of a code, as we do in the remainder of this tutorial. This greatly increases the readability of the code for human beings.
The general form of the for loop is
for <variable> = <expression>
<statement>
...
<statement>
end
where the variable is often called the loop index. The elements of the row vector <expression> are stored
one at a time in the variable and then the statements up to the end statement are executed. For exam
<expression> can be a matrix in which case each column vector is stored one at a time in i.
91
Warning: In using i as the index of the for loop, i has just been redefined to be n instead of 1 .
Caveat emptor!
A more practical example of the use of a for loop is the generation of the Hilbert matrix of order n,
which we have already discussed a number of times. This is easily done using two for loops by
H = zeros(n);
for i = 1:n
for j = 1:n
H(i,j) = 1/(i + j - 1);
end
end
Warning: In using i and j as the indices of the for loops, i and j have just been redefined to be n
instead of 1 . Caveat emptor!
for loops often have branches in them. For this we need the if branch, which we now describe. The
simplest form of the if statement is
if <logical expression>
<statement>
...
<statement>
end
where the statements are evaluated as long as the <logical expression> is true. The
<logical expression> is generally of the form
92
enter
>> strcmp(a, b)
The result is true if the two character strings are identical and false otherwise.
Relational Operators
<
<=
==
Less than.
Less than or equal to.
Equal.
>
>=
=
strcmp(a, b)
Greater than.
Greater than or equal to.
Not equal to.
Compares strings.
93
Logical Operators
A&B
A|B
A
xor(A, B)
AND.
OR.
NOT.
EXCLUSIVE OR.
a && b
a || b
The second MATLAB loop structure is the while statement. The general form of the while loop is
while <logical expression>
<statement>
...
<statement>
end
where the statements are executed repeatedly as long as the <logical expression> is true. For example, eps can be calculated by
ep = 1;
while 1 + ep > 1
ep = ep/2;
end
eps = 2*ep
It is possible to break out of a for loop or a while loop from inside the loop by using the break
94
command as in C. This terminates the execution of the innermost for loop or while loop.
The continue statement is related to break. It causes the next iteration of the for or while loop
to begin immediately.
The switch command executes particular statements based on the value of a variable or an expression.
Its general form is
switch <variable or expression>
case <Value 1>
<statement group 1>
case {<Value 2a>, <Value 2b>, <Value 2c>, ..., <Value 2m>}
<statement group 2>
...
case <value n>
<statement group r>
otherwise
<statement group r+1>
end
where statement group 1 is evaluated if the variable or expression has <Value 1>, where statement group
2 is evaluated if the variable or expression has values <Value 2a> or <Value 2b> or <Value 2c>, etc.
(Note that if a case has more than one value, then all the values must be surrounded by curly brackets.)
The final otherwise is not required. If it occurs and if none of the values match the variable or expression, then statement group r+1 is evaluated. If it does not occur and if none of the values match, then
none of the statement groups are executed.
Warning: The switch command is different in MATLAB than in C in two ways:
First, in MATLAB the case statement can contain more than one value; in C it can only contain one.
And, second, in MATLAB only the statements between the selected case and the following one
or the following otherwise or end (whichever occurs first) are executed; in C all the statements following the selected case are executed up to the next break or the end of the block.
Flow Control
break
case
continue
else
elseif
end
for
if
otherwise
switch
while
Generates a logical matrix with all elements having the logical value true.
Use true or true(n) or true(m, n).
Generates a logical matrix with all elements having the logical value false.
Use false or false(n) or false(m, n).
95
8.2.
Although MATLAB does have a quite powerful programming language, it is needed much less frequently than in typical high-level languages. Many of the operations and functions that can only be
applied to scalar quantities in other languages can be applied to vector and matrices in MATLAB. For
example, MATLABs relational and logical operators can also be applied to vectors and matrices. In this
way, algorithms that would normally require flow control for coding in most programming languages can
be coded using simple MATLAB commands.
If A, B Rmn then the relational equation
>> C = A rop B
is evaluated as cij = aij rop bij , where rop is one of the relational operators defined previously. C is a
logical array, that is, its data type is logical not numeric. The elements of C are all 0 or 1: 0 if
aij rop bij is a false statement and 1 if it is a true one. Also, the relational equation
>> C = A rop c
is defined when c is a scalar. It is evaluated as if we had entered
>> C = A rop c*ones(size(A))
Similar behavior holds for logical operators:
>> C = A & B
means cij = aij & bij ,
>> C = A | B
means cij = aij | bij ,
>> C = A
means cij = aij , and
>> C = xor(A, B)
means cij = xor(aij , bij ). Again the elements of C are all 0 or 1.
To show the power of these MATLAB commands, suppose we have entered
>> F = rand(m, n)
and now we want to know how many elements of F are greater than 0.5. We can code this as
nr elements = 0;
for i = 1:m
for j = 1:n
if F(i,j) > 0.5
nr elements = nr elements + 1;
end
end
end
nr elements
However, it can be coded much more simply, quickly, and efficiently since the relational expression
>> C = F > 0.5;
or, to make the meaning clearer,
>> C = (F > 0.5);
generates the matrix C where
n
1 if fij > 0.5
cij =
0 otherwise.
the desired elements of F. We can also replace all the elements of F which are 0.5 by using
>> C = (F > 0.5)
>> F = F.*C - pi*( C)
Shortly we will present two easier ways to do this.
There is even a MATLAB function which determines the locations of the elements of a vector or a matrix where some property is satisfied. The command
>> find(x)
generates a column vector containing the indices of x which are nonzero. (Recall that nonzero can also
mean TRUE so that this command finds the elements where some condition is true.) For example, if
x = (0, 4, 0, 1, 1, 0, )T then the resulting vector is (2, 4, 5, 7)T . We can add 10 to every nonzero element of x by
>> ix = find(x);
>> x(ix) = x(ix) + 10;
Note: If no element of the vector x is nonzero, the result is the empty matrix [] and the following statement is not executed.
Note: There is a similar function which finds if a substring is contained in a string.
>> k = strfind(str, pattern)
returns the starting index for any and all occurrence ofs pattern in str.
find can also be applied to a matrix. The command
>> find(A)
first transforms A to a column vector (i.e., A(:) ) and then determines the locations of the nonzero elements. As a simple example of the power of this command we can add 10 to every nonzero element of A
by
>> ijA = find(A);
>> A(ijA) = A(ijA) + 10
Or we can work with the matrix directly by entering
>> [iA, jA] = find(A)
The two column vectors iA and jA contain the rows and columns, respectively, of the nonzero elements.
We can also find the locations of the nonzero elements and their values by
>> [iA, jA, valueA] = find(A)
Now is as good a time as any to discuss a fact about matrices which might be confusing. We can access
and/or modify elements of a matrix A using either one index or two. Suppose that A R64 . Then the
element A(3,1) is also A(3) and the element A(3,2) is also A(9). If we use two indexes we are treating
A as a matrix, while if we use one we are treating A as the column vector A(:). And we can switch back
and forth between the two. For example, when we enter
>> [m, n] = size(A);
>> ijA = find(A);
>> [iA, jA] = find(A)
we have the elements of A(:) stored in ijA and the elements of A stored in iA and jA. Knowing one, we
can calculate the other by
>> ijA from iA jA = sub2ind(size(A), iA, jA);
>> [iA from ijA, jA from ijA] = ind2sub(m, ijA);
And, as a specific example, we can choose one column randomly from each row of A by
>> r = randi(n, m, 1);
>> ij = sub2ind(size(A), [1:m], r);
>> random columnsA = A(ij);
We can also find the elements of a vector or a matrix which satisfy a more general property than being
nonzero. For example, to find the locations of all the elements of x which are greater than 5 enter
>> find(x > 5)
and to find the locations of all the elements of x which are greater than 5 and less than 8 enter
>> find( (x > 5) & (x < 8) )
We can find the number of elements which satisfy this last property by entering
>> length( find( (x > 5) & (x < 8) ) )
Previously, we showed how to replace all the elements of F which are 0.5 by . A method which
97
99
Logical Functions
all
any
exist( 0 <name> 0 )
isequal
find
ind2sub
sub2ind
logical
ischar
isempty
isfinite
isinf
islogical
isnan
strfind
8.3.
True if all the elements of a vector are true; operates on the columns of a
matrix.
True if any of the elements of a vector are true; operates on the columns of a
matrix.
False if this name is not the name of a variable or a file. If it is, this function
returns:
1 if this is the name of a variable,
2 if this is the name of an m-file,
5 if this is the name of a built-in MATLAB function.
Tests if two (or more) arrays have the same contents.
The indices of a vector or matrix which are nonzero.
Converts indices of a matrix A from A(:), the column vector form, to A, the
matrix form.
Converts indices of a matrix A from A, the matrix for, to A(:), the column
vector form.
Converts a numeric variable to a logical one.
True for a character variable or array.
True if the matrix is empty, i.e., [].
Generates a matrix with 1 in all the elements which are finite (i.e., not Inf or
NaN ) and 0 otherwise.
Generates a matrix with 1 in all the elements which are Inf and 0 otherwise.
True for a logical variable or array.
Generates a matrix with 1 in all the elements which are NaN and 0 otherwise.
Find any and all occurrences of a substring in a string.
Function M-files
We have already discussed script m-files, which are simply an easy way to collect a number of statements and execute them all at once. Function m-files, on the other hand, are similar to functions or procedures or subroutines or subprograms in other programming languages. Ordinarily, variables which are
created in a function file exist only inside the file and disappear when the execution of the file is completed
these are called local variables. Thus you do not need to understand the internal workings of a function
file; you only need to understand what the input and output arguments represent.
Note: The generic term for script files and function files is m-files, because the extension is m.
Unlike script files, function files must be constructed in a specific way. The first line of the file
<file name>.m must begin with the keyword function. Without this word, the file is a script file. The
complete first line, called the function definition line, is
function <out> = <function name>(<in 1>, ..., <in n>)
or
function [<out 1>, ..., <out m>] = <file name>(<in 1>, ..., <in n>)
where the name of the function must be the same as the name of the file (but without the extension). The
input arguments are <in 1>, <in 2>, . . . . The output arguments must appear to the left of the equal
sign: if there is only one output argument, i.e., <out>, it appears by itself; if there is more than one, i.e.,
<out 1>, etc., they must be separated by commas and must be enclosed in square brackets.
Variables in MATLAB are stored in a part of memory called a workspace. The base workspace contains
all the variables created during the interactive MATLAB session, which includes all variables created in
script m-files which have been executed. Each function m-file contains its own function workspace which
is independent of the base workspace and every other function workspace. The only way to connect
these workspaces is through the arguments of a function or by using the global command (which we will
discuss shortly).
There is great flexibility in the number and type of input and output arguments; we discuss this topic in
100
great detail later. The only detail we want to mention now is that the input arguments are all passed by
value as in C. (That is, the values of the input arguments are stored in temporary variables which are
local to the function.) Thus, the input arguments can be modified in the function without affecting any
input variables in the calling statement.
Warning: The name of the le and the name of the function must agree. This is also the name of the com-
Comment lines should immediately follow. A comment line begins with the percent character, i.e., %.
All comment lines which immediately follow the function definition line constitute the documentation for
this function; these lines are called the online help entry for the function. When you type
>> help <function name>
all these lines of documentation are typed out. If you type
type <function name>
the entire file is printed out. In addition, the first line of documentation, i.e., the second line of the file,
can be searched for keywords by entering
>> lookfor <keyword>
Make sure this first comment line contains the name of the command and important keywords which describe its purpose.
Note: Comments can be placed anywhere in an m-file, including on a line following a MATLAB statement. The initial comment lines in a script file and the comment lines in a function file which
immediately follow the first line are special: they appear on the screen when you type
>> help <function name>
Before discussing functions at great length, there is one technical detail it is important to consider before it trips you up: how does MATLAB find the m-files you have created? Since MATLAB contains
thousands of functions, this is not an easy task. Once MATLAB has determined that the word is not a
variable, it searches for the function in a particular order. We show the order here and then discuss the
items in detail throughout this subsection.
(1) It checks if <function name> is a built-in function (i.e., coded in C).
(2) It checks if <function name> is a function, i.e., the primary function, a subfunction, or a nested
function in the current scope, in the current file. (We discuss all these terms shortly.)
(3) It checks if the file <function name>.m exists in the current directory.
(4) It checks if the current directory has a subdirectory called private; if it does, MATLAB checks if
the file <function name>.m exists in this subdirectory.
(5) It searches the directories in the search path for the file <function name>.m.
Note from (3) that MATLAB searches in the current directory for the function by searching for the mfile with the same name. If the m-file is not in the current directory, the simplest way to enable MATLAB
to find it is have the subdirectory in your search path. If you type
>> path
you will see all the directories that are searched. If you have created a subdirectory called matlab in
your main directory, this is usually the first directory searched (unless the search path has been modified).
Thus, you can put your m-files in this subdirectory and be sure that MATLAB will find them. You can
also add directories to the search path by
>> path( 0 new directory 0 , path)
or
>> path(path, 0 new directory 0 )
(The former puts new directory at the beginning of the search path while the latter puts it at the end.)
Alternately, you can add one or more directories at the beginning of the search path by
>> addpath( 0 new directory #1 0 , 0 new directory #2 0 , ...)
Warning: When you begin a MATLAB session, it always checks if the subdirectory matlab exists in
If you are worried because passing arguments by value might drastically increase the execution time of the
function, we want to reassure you that this does not happen. To be precise, MATLAB does not actually pass all
the input arguments by value. Instead, an input variable is only passed by value if it is modified by the function. If an input variable is not modified, it is passed by reference. (That is, the input argument is the actual
variable used in the calling statement and not a local copy.) In this way you get the benefit of call by value
without any unnecessary overhead. And how does MATLAB know if an input argument is modified? It can only
be modified if it appears on the left-hand side of an equal sign inside the function!
101
your main directory. If you create this subdirectory after you start a MATLAB session, it will
not be in the search path.
Now we return to our discussion of creating functions. We begin with a simple example of a function file
which constructs the Hilbert matrix (which we have already used a number of times).
function H = hilb local(n)
% hilb local: Hilbert matrix of order n (not from MATLAB)
% hilb local(n) constructs the n by n matrix with elements 1/(i+j-1).
% This is one of the most famous examples of a matrix which is
% nonsingular, but which is very badly conditioned.
H = zeros(n);
for i = 1:n
for j = 1:n
H(i,j) = 1/(i+j-1);
end
end
The input argument is n and the output argument is H. The first line of the documentation includes the
name of the function as well as a brief description that lookfor uses. The following lines of documentation also appear on the screen if we enter
>> help hilb local
Note: The above code is not presently used in MATLAB (although it was in early versions.) The actual
MATLAB code for this function is shown in Section 8.5.
We follow by defining H to be an nn matrix. Although not essential, this statement can greatly increase the speed of the function because space can be preallocated for the matrix. For example, consider
the following code.
function a = prealloc(n, initialize)
% prealloc: testing how well preallocating a vector works
% n = the size of the vector
% initialize = true - preallocate the vector
%
= false - do not
if initialize == true
a = zeros(n,1);
end
a(1) = 1;
for i = 2:n
a(i) = a(i-1) + 1;
end
If initialize = false the vector a is not preallocated, while if initialize = true it is. We find
that
>> prealloc(100000, true);
runs over 400 (thats right, four hundred) times as fast as
>> prealloc(100000, false);
Note that i and j are redefined from 1 since they appear as for loop indices. However, since i
and j are local to this function, this does not have any effect on the calling code when this command is
executed. Also, the variable H is local to the function. If we type
>> Z = hilb local(12)
then the matrix Z contains the Hilbert matrix and H is undefined.
Normally functions are completed when the end of the file is reached (as above). If the flow control in a
function file is complicated enough, this might be difficult to accomplish. Instead, you can use the return
command, which can appear anywhere in the function and force an immediate end to the function.
Note: The return function also works in script m-files
Alternately, you can force the function to abort by entering
error(<string>)
If the string is not empty, the string is displayed on the terminal and the function is aborted; on the other
hand, if the string is empty, the statement is ignored.
One feature of function files which is occasionally very useful is that they can have a variable number of
102
input and output variables. For example, the norm of a vector x can be calculated by entering
>> norm(x, p)
if p = 1, 2, or Inf or, more simply, by
>> norm(x)
if p = 2. Similarly, if only the eigenvalues of a matrix A Cnn are desired, enter
>> eigval = eig(A)
However, if both the eigenvalues and eigenvectors are desired, enter
>> [V, D] = eig(A)
where D Cnn is a diagonal matrix containing the n eigenvalues on its diagonal and the corresponding
eigenvectors are found in the same columns of the matrix V Cnn .
Note: On the right side of an equation, [V D] or [V, D] is the matrix whose initial columns come from
V and whose final columns come from D. This requires that V and D be matrices which have the
same number of rows. On the left side, [V, D] denotes the two output arguments which are returned by a function. V and D can be completely different variables. For example, one can be a
character variable and the other a matrix.
MATLAB can also determine the number of input and output arguments: nargin returns the number
of input arguments and nargout returns the number of output arguments. For example, suppose we want
to create a function file which calculates
2
and all the values can be calculated in one call to spruce. If, on the other hand, x, , and a are all vectors, then the function is
2
for i = 1, 2, . . . , n ,
and, again, all the values can be calculated in one call to spruce.
A common error which writing a function m-file is forgetting that the argument(s) might be vectors or
103
matrices. For example, we mentioned the Heaviside step function in Some Common Real Mathematical
Functions on page 13, but pointed out that it is in a toolbox which you might not have. We could write it
as
function Y = myheaviside scalar(x)
if x < 0
y = 0;
elseif x == 0
y = 0.5;
else
y = 1;
end
but this will only work if x is a scalar. For example,
>> myheaviside scalar([-3:3])
returns
ans =
1
Note that the function does not return an error message it simply returns an incorrect result. (Incidentally, myheaviside scalar([-6:0]) also returns 1.) The following function, however, will work in all
cases.
function Y = myheaviside(X)
Y = zeros(size(X));
Y(X>0) = 1;
Y(X==0) = 0.5;
(The input and output arguments are capitalized to indicate that they can be matrices.) For example,
>> myheaviside([-3:3])
returns
ans =
0
0
0
0.5000
1.0000
1.0000
1.0000
We have now presented all the essential features of the MATLAB programming language, and it certainly is a minimal language. MATLAB can get away with this because most matrix operations can
be performed directly unlike in most other programming languages. You only need to write your own
function if MATLAB cannot already do what you want. If you want to become proficient in this language,
simply use the type command to look at the coding of some functions.
Now that we have discussed the essentials of programming in MATLAB, it is necessary to discuss how
to program correctly. When you are entering one statement at a time in the text window, you immediately
see the result of your calculation and you can determine whether or not it is correct. However, in an m-file
you have a sequence of statements which normally end with semicolons so that you do not see the intermediate calculations. What do you do if the result is incorrect? In other words, how do you debug your
m-file?
There are a number of simple techniques you can use and we discuss them in turn. In a script m-file
intermediate calculations are normally not printed out, but they are still available to look at. This can frequently lead to an understanding of where the calculation first went wrong. However, this is not true of
function m-files since all the local variables in the function disappear when the function ends. Of course,
with any m-file you can selectively remove semicolons so that intermediate results are printed out. This is
probably the most common method of debugging programs no matter what programming language is
being used.
When loops are involved (either using for or while statements), the printed output can seem to be
endless (and it is endless if you are in an infinite loop). And it is usually impossible to read the output since it is zipping by at (what appears to be) nearly the speed of light! The pause command can
slow down or even stop this output. By itself pause stops the m-file until some key is pressed while
pause(<floating-point number>) pauses execution for this many seconds or fractions of a second.
(This is computer dependent, but pause(0.01) should be supported on most platforms.) You can even
turn these pauses on and off by using pause on and pause off.
Note: Occasionally, you will forget that you have put the command
104
pause
into your code and wonder why it is taking so long to execute. The alternative
input( 0 Press Enter to continue 0 , 0 s 0 );
pauses the code and also reminds you that it has been paused.
The echo command is also useful for debugging script and function m-files, especially when if statements are involved. Typing
>> echo on
turns on the echoing of statements in all script files (but not printing the results if the statements end
with semicolons), and echo off turns echoing off again. However, this does not affect function files. To
turn echoing on for a particular function, type
>> echo <function name> on
and to turn echoing on for all functions, type
>> echo on all
The keyboard command is also very useful for debugging m-files. It stops execution of the m-file,
similar to the pause command. However, it returns complete control to the user to enter any and all
MATLAB commands. In particular, you can examine any variables in the functions workspace. If desired, you can also change the value of any of these variables. The only way you will recognize this is not a
standard MATLAB session is that the prompt is
K>>
for Keyboard. To terminate the keyboard session and return control to the m-file, enter
K>> return
To terminate both the keyboard session and the execution of the m-file, enter
K>> dbquit
When using the debugger, you are not running your program. Instead, you are running the debugger
and it is running your program. Thus, many of the commands you enter are commands for the debugger;
to distinguish these commands from normal MATLAB commands they begin with db. There are two
ways to run the debugger: you can type the debugger commands into the workspace, or in a MATLAB
window you can use the mouse and click on the commands.
In addition, you are still in the workspace so many MATLAB commands can still be executed. For example, to see the values of variables just type the variable name into the workspace. (Alternatively, you
can move the mouse over the variable name in the window and its description and value(s) will be shown.)
Also, you can run most MATLAB commands as long as you do not try to create new variables (but you
can modify existing variables).
We will not discuss the commands in this debugger in detail, but only provide a brief description of
each one, because these are similar to commands in any debugger. If you have experience with using a
debugger, doc will give you complete details.
105
Debugging Commands
keyboard
dbstep
dbstep n
dbstep in
dbstep out
dbcont
dbstop
dbclear
dbup
dbdown
dbstack
dbstatus
dbtype
dbquit
return
Now we want to discuss the arguments in a MATLAB function, since they are are used somewhat differently than in other programming languages. For example, in
function out = funct1(a, t)
a and t are the input arguments and out is the output argument. Any and all input variables are local to the function and so can be modified without affecting the arguments when the function funct1 is
called. (This is true no matter what type of variables they are.) In
function [out1, out2, out3] = funct2(z)
z is the only input argument and there are three output arguments, each of which can be any type of
variable. There is no requirement that all three of these output arguments actually be used. For example,
the calling statement might be any of the following:
>> art = funct2(1.5)
>> [physics, chemistry] = funct2([1 2 3])
>> [math, philosophy, horticulture] = funct2(reshape([1:30], 6, 5))
(just to be somewhat silly).
That is, in MATLAB input arguments occur on the right side of the equal sign and output arguments
occur on the left. Arguments which are to be modified by the function must occur on both sides of the
equal sign in the calling statement. For example, in funct2 if z is modified and returned in out1 then
the calling sequence should be
>> [z, b, c] = funct2(z)
where z appears on both sides of the equal sign. (There is an alternative to this awkward use of parameters which are modified by the function: you can make a variable global, as we discuss at the end of this
section. However, this is not usually a good idea.)
There is another difference between MATLAB and most other programming languages where the type
of each variables has to be declared, either explicitly or implicitly. For example, a variable might be an
integer, a single-precision floating-point number, a double-precision floating-point number, a character
string, etc. In MATLAB, on the other hand, there is no such requirement. For example, the following
statements can follow one another in order and define x to be a string variable, then a vector, then a
scalar, and finally a matrix.
>> x = 0 WOW? 0
>> x = x + 0
>> x = sum(x)
>> x = x*[1 2; 3 4]
It is particularly important to understand this typelessness when considering output arguments. For example, there are three output arguments to funct2 and any of them can contain any type of variable. In
fact, you can let the type of these arguments depend on the value or type of the input argument. This is
106
probably not something you should want to do frequently, but it is sometimes very useful.
Occasionally, there is a need to pass values from the workspace to a function or to pass values between
different functions without using the input arguments. (As we discussed earlier, this may be desirable if a
variable is modified by a function.) In C this is done by using global variables. MATLAB also has global
variables which are defined by declaring the variables to be global using
>> global <variable 1> <variable 2> <variable 3> ...
By the way, a variable is a global variable if it appears in a global statement and a local variable if it does
not. (Note that a variable can be a local variable in one function and a global variable in another.) This
statement must appear in every function which is to share the variables. If the workspace is also to share
these variables, you must type this statement (or be put into a script file which you execute) before these
variables are used.
Warning: Spaces, not commas, must separate the variables in a global statement
Instead of using a global variable, it is frequently preferable to save the value of a local variable between
calls to the function. Normally, local variables come into existence when the function is called and disappear when the function ends. Sometimes it is very convenient to be able to save the value of a local
variable so that it will still be in existence when the function is next called. In C, this is done by declaring
the variable static. In MATLAB it is done by declaring the variable persistent using
>> persistent <variable 1> <variable 2> <variable 3> ...
Warning: Spaces, not commas, must separate the variables.
Note: The first time you enter the function, a persistent variable will be empty, i.e., [], and you can test
for this by using isempty.
We now present a simple example where persistent variables are very helpful. Suppose we want to write
a function m-file to evaluate
y2
h(y) =
y1 (1 y12 ) y2 + cos t
where , , , and are parameters which will be set initially and then left unchanged during a run.
(The parameter Gamma is capitalized even though it is not a matrix because is very different from .)
We might be studying a mathematical model where this function will be evaluated many, many times
for different values of y. For each experiment these parameters will be fixed, but they will be different
for each experiment. We do not want to hardcode the values in the function because we would have to
repeatedly change the function which is very undesirable. Certainly we can write the function as
function z = fncz1(y, alpha, beta, Gamma, omega)
z = [ y(2) ; -y(1)*(1-beta*y(1) 2)-alpha*y(2)+Gamma*cos(omega*t) ];
but then we have to include these four parameters in each call to the function. We can always simplify
this function by combining the four parameters into one structure by
function z = fncz2(y, param)
z = [ y(2) ; -y(1)*(1-param.beta*y(1) 2)-param.alpha*y(2)+...
param.Gamma*cos(param.omega*t) ];
but then it is harder to read the equation. (If this function was more complicated it would be much
harder to read.) To make this last function easier to read we could write it as
function z = fncz3(y, param)
alpha = param.alpha
beta = param.beta
Gamma = param.Gamma
omega = param.omega
z = [ y(2) ; -y(1)*(1-beta*y(1) 2)-alpha*y(2)+Gamma*cos(omega*t) ];
but we want to propose two other alternatives.
There are many reasons why evaluating h(y) using fncz1, fncz2, or fncz3 might not be desirable
or even practical. For instance, this function might be called repeatedly inside a general purpose function m-file, say general, which we have written. In general we only want to call the function as z =
fncz?(y) and not have to worry about how parameters are passed to the function.
The first way to reduce the number of parameters is to write the function m-file for h(y) as
107
A closure is a complicated term to explain. In this context it means that the parameters used when the function is defined are saved and can be referenced when the function is later used.
108
tion with the command end, but this is not necessary because MATLAB recognizes that a function
has ended when it encounters the next function statement. However, if a nested function is used
then it and all other functions must end with the end statement.
First, we discuss subfunctions, which are quite simple. They are placed following the primary function
and between or following other subfunctions. For example,
function primary function
% code for the primary function
function subfunction1
% code for the first subfunction
function subfunction2
% code for the second subfunction
They are only visible to the primary function and to other subfunctions in the same file. Thus, different
m-files can contain subfunctions with the same name. Also, the help, lookfor, and type commands
can only access the primary file. It is crucial to understand that variables in the primary function or in a
subfunction are local to that function and unknown outside it (unless they are declared to be global ).
The only way to pass variables between these functions is through the argument list.
Usually, subfunctions are sufficient and they are much easier to describe. When they are not sufficient, we have nested functions. Its main advantage (as far as we are concerned) is that variables can be
passed into and out of a nested function without being in the argument list. Nested functions are more
complicated than subfunctions and we will only provide a brief discussion.
To make this discussion specific, consider the following function m-file.
They are similar to internal functions in Fortran 95, and they are somewhat related to inner classes in Java
but not in C++.
109
%
%
%
%
%
%
1
2
3
4
5
6
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
%
%
%
%
%
27
28
29
30
31
%
%
%
%
%
%
32
33
34
35
36
37
A nested function is within another function. For example, the nested functions nest 1 and nest 2 are
nested within the primary function nested ex, and the nested function nest 3 is nested within the subfunction sub 1. (Nested functions can have other nested functions within them, but enough is enough.)
The important concept to understand when using nested functions is the scope of variables in the function m-file. The scope of a variable is the context within which it is defined, i.e., where it can be set, modified, and used. Now let us consider a function workspace. The workspace of the primary function is also
independent of the workspace of each subfunction. However, since a nested function is within one or more
other functions, it is within the workspace of this function or these functions. In the function nested ex
the nested functions nest 1 (lines 715) and nest 2 (lines 1623) have access to the variables p1, p2,
and p3 of the primary function (lines 126). They also have access to the subfunction sub 1 (lines 27
37) (but not any of its variables) as shown in line 13. Note that nest 2 also has access to nest 1 as
shown in line 14.
To begin, a nested function can have local variables. Any variable in the argument list of a nested func110
tion is local to that function, and the same is true for any variable which contains values returned by the
function. Thus, the variable out1 is local to the function nest 1 and out2 is local to nest 2. Also,
the variable r is local to nest 1 and also local to nest 2. The variable r which is defined on line 4 is
unchanged by the calls to the two nested function. The value returned in line 24 will always be 5. Similarly, the variables t and out3 are local to nest 3.
What about the remaining variables? The variables n1a and p are local to nest 1 (as shown in
line 18) and n2a is local to nest 2 because the outer function, i.e., nested ex does not define or use
them. Similarly, n3a is local to nest 3. Also, n1a cannot be accessed in nest 2 as shown in line 17.
(If n1a really needs to be passed to nest 2, then it must be in the workspace of nested ex. (This
could be done by adding n1a = 0 after line 3.) On the other hand, the variables p1 and p2 are defined
in nested ex and so can be used in nest 1 and nest 2. This is also true for p3 which is an output
variable in nested ex. Finally, note that n1b is not used in nested ex until line 25, after nest 1 and
nest 2 have been called, but it can still be used in nest 2 even though it was defined in nest 1.
CDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDB
FHEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEHG
FG
FG
FG
FG
WARNING
FG
FG
FG A common mistake when using nested functions is defining what you think is a
FG
FG local variable in a nested function and forgetting that the same variable name
FG
FG
FG
FG is used in the outer function. It is hard to imagine making that mistake here
FG
FG because the code is so short but it can easily happen in a real code. One soFG
FG
FG
FG lution is to append a special character to all local variables in nested functions
FG
FG
FG
FG (for example, append an underscore, i.e., , to the end of the name of each local FG
FG variable).
FG
FG
FG
FG
FG
FHDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDHG
@EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEA
Just for completeness, none of the variables in nested ex can be accessed in sub 1 and vice versa.
For instance, p2 cannot be accessed in the subfunction sub 1 as we show on line 30.
Now let us return to the topic of how MATLAB finds a function. As we stated previously (but did not
discuss), when a function is called from within an m-file, MATLAB first checks if the function named is
the primary function or a subfunction in the current file. If it is not, MATLAB searches for the m-file in
the current directory. Then MATLAB searches for a private function by the same name (described below).
Only if all this fails does MATLAB use your search path to find the function. Because of the way that
MATLAB searches for functions, you can replace a MATLAB function by a subfunction in the current
m-file but make sure you have a good reason for doing so!
In the previous paragraph we described how to create a subfunction to replace one function by another
of the same name. There is another, more general, way to handle this replacement: you can create a subdirectory in your current directory with the special name private. Any m-files in this subdirectory are
visible only to functions in the current directory. The functions in this subdirectory are called private
functions. For example, suppose we are working in the directory personal and have created a number of
files which use rref to solve linear systems. And suppose we have written our own version of this command, because we think we can calculate the reduced row echelon of a matrix more accurately. The usual
way to test our new function would be to give it a new name, say myrref, and to change the call to rref
in every file in this directory to myrref. This would be quite time-consuming, and we might well miss
some. Instead, we can code and debug our new function in the subdirectory private, letting the name of
our new function be rref and the name of the m-file be rref.m. All calls in the directory to rref will
use the new function we are testing in the subdirectory private, rather than MATLABs function. Even
more important, any function in any other directory which calls rref will use the MATLAB function and
not our new, improved version.
The final topic we will briefly discuss involves recursion. It is possible and sometimes useful for a
Since MATLAB contains thousands of functions, this means you do not have to worry about one of your subfunctions being hijacked by an already existing function. When you think up a name for a primary function
(and, thus, for the name of the m-file) it is important to check that the name is not already in use. However,
when breaking a function up into a primary function plus subfunctions, it would be very annoying if the name of
every subfunction had to be checked especially since these subfunctions are not visible outside the m-file.
111
for n 0
and f1 = 1 .
This sequence, i.e., 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, . . ., can be coded as
function y = fibonacci(n)
% WARNING: VERY VERY INEFFICIENT CODE
% ***** n must be a nonnegative integer *****
if n == 0
y = 1;
% no recursion if n = 0
elseif n == 1
y = 1;
% no recursion if n = 1
else
y = fibonacci(n-1) + fibonacci(n-2);
% two recursive calls for
end
n > 1
A recursive code must be able to stop, and this code stops if n = 0 or n = 1. For larger values of n, the
function is called recursively.
Warning: This code is very, very, very inefficient. We have provided it simply to show how recursion can
lead to a very compact code. The reason this code is very inefficient is because it uses an incredibly large amount of CPU time for large n (and 50 is very, very, very large). In fact,
in order to calculate fn the function fibonacci is called recursively 2fn 2 times and fn
grows exponentially with n.
There are examples for which recursion is advantageous. However, our aim is simply to show how to use
it not whether to use it.
Function Commands
function
end
error( 0 <message> 0 )
echo
global
persistent
return
nargin
nargout
pause
varargin
varargout
112
8.4.
In MATLAB it is possible for a program to create or modify statemennts on the fly, i.e., as the program is running. Entering
>> eval(<string>)
executes whatever statement or statements are contained in the string. For example, entering
>> s = 0 x = linspace(0, 10, n); y = x.*sin(x).*exp(x/5); plot(x, y) 0
>> eval(s)
executes all three statements contained in the string s. In addition, if an executed statement generates
output, this is the output of eval. For example, if we type
>> A = zeros(5,6);
>> [m, n] = eval( 0 size(A) 0 );
then m is 5 and n is 6.
There is a very practical applications for this command since it can combine a number of statements
into one. For example, suppose we want to work with the columns of the Hilbert matrix of size n and
we want to create variables to hold each column, rather than using H(:,i). We can do this by hand by
typing
>> c1=H(:,1);
>> c2=H(:,2);
...
which gets tiring very quickly. Instead, we can do this by typing
>> for i = 1:n
eval( [ 0 c 0 num2str(i) 0 =H(:,i) 0 ] )
end
This requires some explanation. It might be a little clearer if we separate the statement inside the for
loop into two statements by
s = [ 0 c 0 , num2str(i), 0 =H(:,i) 0 ]
eval(s)
(where we include commas in the first statement for readability). s is a text variable which contains
c1=H(:,1) the first time the loop is executed, then c2=H(:,2) the second time, etc. (To understand
how s is created, recall that s is really just a row vector with each element containing the ASCII representation of the corresponding character.)
Finally, there is a very esoteric application for this command that allows it to catch errors. This is similar to the catch and throw commands in C++ and Java. To use this feature of eval, call it using
two arguments as
>> eval(<try string>, <catch string>)
The function executes the contents of <try string> and ignores the second argument if the execution
succeeds. However, if it fails then the contents of <catch string> are executed. (This might be a call
to a function which can handle the error.) If there is an error, the command lasterr returns a string
containing the error message generated by MATLAB.
A MATLAB command which is occasionally useful in a function is feval. It executes a function, usually defined by an m-file, whose name is contained in a string by
>> feval(<string>, x1, x2, ..., xn)
(See below for other ways to pass the function in the argument list.) Here x1, x2, ..., xn are the
arguments to the function. For example, the following two statements are equivalent
>> A = zeros(5,6)
>> A = feval( 0 zeros 0 , 5, 6)
Suppose that in the body of one function, say sample, we want to execute another function whose
name we do not know. Instead, the name of the function is to be passed as an argument to sample.
Then feval can be used to execute this text variable. For example, suppose in function sample we want
to generate either linear or logarithmic plots. We can input the type of plot to use by
113
-- 0 , xx, y3,
:0 )
There are two common ways to pass the function <type of plot> in the argument list:
(1) use a character string, e.g., feval( 0 loglog 0 , ...), or
(2) use a function handle, e.g., feval(@logval, ...), or
Note: eval and feval serve similar purposes since they both evaluate something. In fact, feval can
always be replaced by eval since, for example, feval( 0 zeros 0 , 5, 6) can always be replaced by
eval( 0 zeros(5,6) 0 ). However, there is a fundamental difference between them: eval requires
the MATLAB interpreter to completely evaluate the string, whereas feval only requires MATLAB
to evaluate an already existing function. feval is much more efficient, especially if the string must
be evaluated many times inside a loop.
8.5.
As long as your MATLAB code executes quickly, there is no need to try to make it faster. However, if
your code is executing slowly, you might be willing to spend some time trying to speed it up. There are
three standard methods to speed up a code:
(0) Preallocate matrices as shown in the function prealloc on page 102. This is very simple and very
effective if the matrices are large.
(1) Use MATLAB functions, whenever possible, rather than writing your own. If a MATLAB function is
built-in, then it has been written in C and is faster than anything you can do. Even if it is not, much
time has been spent optimizing the functions that come with MATLAB; you are unlikely to do better.
(2) Replace flow control instructions with vector operations. We have already discussed this topic at
length in Section 8.2. Here we will focus on some advanced techniques.
Of course, we should first determine how much CPU time is really being expended in our program
and that is the profile command. It will happily show you the percentage of CPU time spent in
each line of each function in your program! And it will return the results in a very readable fashion! The
commands
>> profile on
>> ..........
>> profile viewer
begin and end profiling and display the results in the Profiler window.
As a simple example of method (0), consider the function hilb on page 102. hilb local(2000) runs
over 300 times slower if the line H = zeros(n) is omitted.
We have put quickly and slowly in quotes because this is quite subjective. Remember that your time is
valuable: if it takes you longer to optimize your code than you will save in running it more quickly, stifle the urge
to muck around with it. Also remember that the amount of time it actually takes to optimize a code is usually a
factor of two or three or . . . longer than the time you think it will take before you get started.
114
Continuing with this example, currently the MATLAB function hilb is written as
function H = hilb2(n)
J = 1:n;
% J is a row vector
J = J(ones(n, 1),:);
% J is now an n by n matrix with each row being 1:n
I = J0 ;
% I is an n by n matrix with each column being 1:n
E = ones(n, n);
H = E./(I+J-1);
as can be seen by entering
>> edit hilb
In the past this code ran nearly 20 times as fast as hilb local. However, now hilb local they are
roughly comparable depending on the computer and operating system. The reason is that MATLAB
has greatly improved its handling of for and while statements. Thus, it is frequently not necessary to
convert simple loops into complicated vector code.
As a realistic example of method (2), suppose you have a large vector y which is the discretization of
a smooth function and you want to know some information about it. In particular, consider the intervals in y where yi > R. What is the average length of these intervals and what is their standard deviation? Also, only include intervals which lie completely within y (i.e., ignore any intervals which begin or
end y ). It is not difficult to write such a code using flow control statements:
115
116
9. Sparse Matrices
Profile the execution time of a MATLAB code. This is very useful for improving the
performance of a code by determining where most of the CPU time is spent.
9. Sparse Matrices
Many matrices that arise in applications only have a small proportion of nonzero elements. For example, if T Cnn is a tridiagonal matrix, then the maximum number of nonzero elements is 3n2. This
is certainly a small proportion of the total number of elements, i.e., n2 , if n is large (which commonly
means in the hundreds or thousands or . . . ).
For full matrices (i.e., most of the elements are nonzero) MATLAB stores all the elements, while for
sparse matrices (i.e., most of the elements are zero) MATLAB only stores the nonzero elements: their
locations (i.e., their row numbers and column numbers) and their values. Thus, sparse matrices require
much less storage space in the computer. In addition, the computation time for matrix operations is significantly reduced because zero elements can be ignored.
Once sparse matrices are generated, MATLAB is completely responsible for handling all the details
of their use: there are no special commands needed to work with sparse matrices. However, there are a
number of commands which are inappropriate for sparse matrices, and MATLAB generally generates a
warning message and refers you to more appropriate commands. For example, cond(S) has to calculate
S1 , which is generally a full matrix; instead, you can use condest which estimates the condition number
by using Gaussian elimination. You have two alternatives: first, use full to generate a full matrix and
use the desired command; or, second, use the recommended alternative command.
There are three common commands in MATLAB for creating sparse matrices. The first is to use speye
to create a sparse identity matrix instead of using eye which creates a full identity matrix. The second is
to enter all the nonzero elements of S Cmn individually by
>> S = sparse(i, j, s, m, n)
where i and j are vectors which contain the row and column indices of nonzero elements and s is the
vector which contains the corresponding values. For example, the square bidiagonal matrix
S=
2
n1
4
n2
6
..
.
0
117
0
..
.
2
2n + 2
1
9. Sparse Matrices
si,j
si,j
1
2
3
..
.
1
2
3
..
.
n
n1
n2
..
.
1
2
3
..
.
2
3
4
..
.
2
4
6
..
.
n1
n
n1
n
2
1
n2
n1
n1
n
2n + 4
2n + 2
0 0 6 0 0
1 0 0 7 0
S1 = 0 2 0 0 8
0 0 3 0 0
0 0 0 4 0
can be
>>
>>
or by
>>
>>
9. Sparse Matrices
In the latter case note that the columns of B have to be padded with zeroes so that each column has five elements, whereas in the former case the vector which becomes the particular
diagonal precisely fits into the diagonal. The element s1,3 of S1 contains the value 6. It appears in the 3rd row of B because it occurs in the 3rd column of S1. Note that the element bn,2
is not used since it would go into the element sn,n+1 .
A slight variation of the above command is
>> T = spdiags(B, d, S)
where T is equated to S and then the columns of B are placed in the diagonals of T specified by d.
Thus, a third way to generate the matrix S given above is
>> S = spdiags([n:-1:1] 0 , [0], n, n)
>> S = spdiags([0:-2:-2*n+2] 0 , [1], S)
Just as with the diag command, we can also extract the diagonals of a sparse matrix by using
spdiags. For example, to extract the main diagonal of S, enter
>> B = spdiags(S, [0])
The number of nonzero elements in the sparse matrix S are calculated by
>> nnz(S)
(Note that this is not necessarily the number of elements stored in S because all these elements are
checked to see if they are nonzero.) The locations and values of the nonzero elements can be obtained
by
>> [iA, jA, valueA] = find(A)
The locations of the nonzero elements is shown in the graphics window by entering
>> spy(S)
These locations are returned as dots in a rectangular box representing the matrix which shows any structure in their positions.
All of MATLABs intrinsic arithmetic and logical operations can be applied to sparse matrices as well
as full ones. In addition, sparse and full matrices can be mixed together. The type of the resulting matrix
depends on the particular operation which is performed, although usually the result is a full matrix. In
addition, intrinsic MATLAB functions often preserve sparseness.
You can generate sparse random patrices by sprand and sparse, normally distributed random matrices by sprandn. There are a number of different arguments for these functions. For example, you can
generate a random matrix with the same sparsity structure as S by
>> sprand(S)
or you can generate an mn matrix with the number of nonzero random elements being approximately
mn by
>> sprand(m, n, rho)
Finally, you can generate sparse random symmetric matrices by sprandsym; if desired, the matrix will also be positive definite. (There is no equivalent command for non-sparse matrices so use
full(sprandsym(...))
Additionally, sparse matrices can be input from a data file with the spconvert command. Use
csvread or load to input the sparsity pattern from a data file into the matrix <sparsity matrix>.
This data file should contain three columns: the first two columns contain the row and column indices of
the nonzero elements, and the third column contains the corresponding values. Then type
>> S = spconvert(<sparsity matrix>)
to generate the sparse matrix S. Note that the size of S is determined from the maximum row and the
maximum column given in <sparsity matrix>. If this is not the size desired, one row in the data file
should be m n 0 where the desired size of S is mn. (This element will not be used, since its value is
zero, but the size of the matrix will be adjusted.)
119
for t t0
where y Rn with initial condition y(t0 ) = y0 . The basic MATLAB commands are easily learned. However, the commands become more involved if we want to explore the trajectories in more detail. Thus, we
divide this section into the really basic commands which are needed to generate a simple trajectory and
into a more advanced section that goes into many technical details. We also provide a large number of examples, many more than in other sections of this overview, to provide a template of how to actually use
the advanced features.
10.1.
Basic Commands
y1
y2
0
=
y2
y1 (1 y12 ) y2 + cos t
120
.
Note: This same trick can be applied to an nth order by defining y1 = y, y2 = y10 , y3 = y20 , . . . ,
0
yn = yn1
.
Before discussing how to solve Duffings equation specifically, we discuss the commands which solve
time-evolution odes. To obtain a numerical solution to a time-evolution first-order ode system, enter
>> <ode solver>(<function handle>, tspan, y0)
or
>> [t, Y] = <ode solver>(<function handle>, tspan, y0)
or
>> sol = <ode solver>(<function handle>, tspan, y0)
First, we have to choose which ode solver to use; this is discussed in detail shortly. It would be possible
for MATLAB itself to decide which numerical method to employ. However, there are good reasons why
the decision should be left in the hand of the user.
Warning: Make sure you understand how to enter the name of the function handle. This is discussed at
length in Section 3.2, and we also briefly discuss it below.
All of the solvers use the same input and output arguments, which we now discuss. The input parameters are:
function The name of the function handle that calculates f(t, y).
tspan The vector that specifies the time interval over which the solution is to be calculated. If this
vector contains two elements, these are the initial time and the final time; in this case the
ode solver determines the times at which the solution is output. If this vector contains more
than two elements, these are the only times at which the solution is output.
Note: the final time can be less than the initial time, in which case the trajectory is moving
backwards in time.
y0 The vector of the initial conditions for the ode.
If there are no output parameters, the individual elements of the solution, i.e., y1 (t), y2 (t), . . . , yn (t),
are plotted vs. t on a single plot. The circles on the trajectories show the actual times at which the solution is calculated.
If there are two output parameters, these are:
t The column vector of the times at which the solution is calculated.
Y The matrix which contains the numerical solution at the times corresponding to t. The first
column of Y contains y1 , the second column y2 , etc.
If there is one output parameter, then it is a structure. The output is now
sol.x The column vector of the times at which the solution is calculated.
sol.y The matrix which contains the numerical solution at the times corresponding to t.
There are seven distinct ode solvers which can be used, as shown in the table below. All these ode
solvers use an adaptive step size to control the error in the numerical solution. Each time step is chosen
to try to keep the local error within the prescribed bounds as determined by the relative error and the
absolute error tolerances (although it does not always succeed). That is, ei , which is the error in yi , is
supposed to satisfy
ei max{ RelTol |yi |, AbsTol(i) }
where the default value of RelTol is 103 and of the vector AbsTol is 106 for each element. (However, there is no guarantee that the error in the numerical calculation actually satisfies this bound.)
We have capitalized the Y in [t, Y] to indicate that the output is a matrix whereas the argument y is a
vector in the function.
121
ODE Solvers
ode45
ode23
ode113
ode15s
ode23s
ode23t
ode23tb
It is up to you to decide which ode solver to use. As a general rule, unless you believe that the ode is
stiff (which we discuss in the next paragraph), try ode45 or ode113. For a given level of accuracy, these
methods should run reasonably fast. (Which one runs faster is very dependent on the ode.) If you know
(or believe) that the ode is stiff, or if these two non-stiff solvers fail, then try ode15s.
And what is a stiff ode? There is no precise definition. Instead, we say it is stiff if the time step required to obtain a stable and accurate solution is unreasonably small. The best way to explain this
rather vague impression is through some simple examples.
Consider the second-order time-evolution ode
y 00 + 999y 0 + 1000y = 0
for t 0
with the initial conditions y(0) = 1 and y 0 (0) = 2 . The solution to this ode is
y(t) = c1 et + c2 e1000t
where
1
1
(1 2 ) and c2 =
(10001 + 2 ) .
1001
1001
There are two time scales in this solution: there is a rapid decay due to the e1000t term and there is a
slow growth due to the et term. Initially, the time step will be very small so that the rapid decay is
calculated accurately (i.e., t 1/1000 ). However, soon it will be negligible and the time step should increase so that it calculates the slow growth accurately (i.e., t 1 ). However, if a non-stiff solver, such
as ode45 or ode23, is used, the time step must always be very small. That is, it must accurately track
the rapidly decaying term even after this term has disappeared in the numerical solution. The reason
is that a numerical instability will cause the trajectory to blow up if the time step increases. However, if
a stiff solver is used, the time step can increase by many orders of magnitude when the rapidly decaying
term has disappeared.
The same is true for the ode
y 00 + 1001y 0 + 1000y = 0
c1 =
whose solution is
y(t) = c1 et + c2 e1000t .
Initially, the time step will be very small so that the rapid decay is calculated accurately (i.e.,
t 1/1000 ). However, soon it will be negligible and the time step should increase so that it calculates
the slowly decaying mode accurately (i.e., t 1 ).
On the other hand, consider the ode
y 00 1001y 0 + 1000y = 0
whose solution is
y(t) = c1 et + c2 e1000t .
the time step must always be very small so that the rapidly growing mode e1000t is calculated accurately (i.e., t 1/1000 ). Thus, this is not a stiff ode.
The above examples are very simple. They are only designed to show that an ode is stiff if there is a
rapidly decaying mode and any growth in the solution occurs on a much slower time scale. (This frequently happens in chemical reaction models, where some reactions occur on a very fast time scale and
122
and other occur on a much slower time scale.) In the next subsection we discuss van der Pols equation, a
second-order ode which is either non-stiff or stiff depending on the value of one parameter. You can plot
the solution and observe the separation of the fast scale and the slow scale as this parameter increases.
One difficulty with a stiff ode solver is that you might have to supply the Jacobian of the ode yourself if
the ode is really nasty. The Jacobian of f(t, y) is the nn matrix
fi
J(t, y) =
(t, y) ,
yj
i.e., the element in the ith row and j th column of J is
fi
.
yj
Any of the stiff methods can approximate this matrix numerically. However, if the ode is bad enough,
this may not be enough. You may have to calculate all these partial derivatives yourself and include them
in your function file. (We show an example of this later.)
The reason for this large choice of ode solvers is that some odes are very, very, very nasty. It is possible
that most of the ode solvers will fail and only one, or maybe two, will succeed. SAY MORE???
To conclude this subsection, we return to Duffings equation. Suppose we want to solve the ode for
t [0, 100] with initial conditions y = (2, 1)T and plot the results. Since this is a very well-behaved ode
for the parameters given, we can use ode45. The simplest approach is to use an anonymous function to
input the right-hand side.
>> alpha = 0.05;
>> beta = 1.0;
>> Gamma = 0.5;
>> omega = 1.0;
>> duffing a = @(t, y)[y(2) ; y(1)*(1-beta*y(1) 2)-alpha*y(2)+Gamma*cos(omega*t)];
>> ode45(duffing a, [0 100], [2 1]);
(The a denotes the fact that duffing a is an anonymous function handle.) The solution will now be
plotted as y1 and y2 vs. t. (This plot is rather cluttered because, not only is the trajectory plotted,
but in addition markers are put at each of the points of the numerical solution.)
Warning: There are a number of parameters which are needed by the function and these must be defined before the function is created. Also, the function handle duffing a will always use these
parameters, even if they are later changed.
Note: Since duffing a is already a function handle, we merely need to use its name as the first argument
to ode45.
To obtain complete control over what is plotted, you should let ode45 output the trajectory and do the
plots yourself. This is easily accomplished by changing the last line of the previous code to
>> [t, Y] = ode45(duffing a, [0 100], [2 1]);
>> figure(1)
>> subplot(2, 1, 1)
>> plot(t, Y(:,1))
>> subplot(2, 1, 2)
>> plot(t, Y(:,2))
>> figure(2)
>> plot(Y(:,1), Y(:,2))
This results in a plot of y vs. t and a separate plot of y 0 vs. t, so that both plots are visible even if
they have vastly different scales. There is also a separate plot of y 0 vs. y, which is called a phase plane.
The next simplest approach is to use a nested function, and so there must also be a primary function.
function duffing ode(alpha, beta, Gamma, omega, y0, final time)
ode45(@duffing n, [0 final time], y0);
%%%%% nested function follows
function deriv = duffing n(t, y)
deriv = [ y(2) ; y(1)*(1-beta*y(1) 2)-alpha*y(2)+Gamma*cos(omega*t) ];
end
end
(The n denotes the fact that duffing n is a nested function.) Note that the parameters are input to
123
the primary function and so are immediately accessible to the nested function. Clearly, this second approach (of using a nested function) requires more coding that the first approach (of using an anonymous
function). However, the first approach only works if the right-hand side can be defined using one MATLAB statement. If the right-hand side is more complicated, then a nested function is the simplest choice.
Note: Since duffing n is a function, and not a function handle, we have to include @ before the name
of the function.
The third, and oldest, approach is to create a separate function m-file (i.e., a primary function) which
calculates the right hand side of this ode system.
function deriv = duffing p(t, y)
% duffing p: Duffings equation, primary function
alpha = 0.05;
beta = 1.0;
Gamma = 0.5;
omega = 1.0;
deriv = [ y(2) ; y(1)*(1-beta*y(1) 2)-alpha*y(2)+Gamma*cos(omega*t) ];
(The p denotes the fact that duffing p is a primary function.) Note that all the parameters are defined in the m-file so that it will have to be modified whenever we want to modify the parameters. This is
a very bad approach because this file will have to be repeatedly modified.
Note: Since duffing p is a function, and not a function handle, we have to include @ before the name
of the function.
Finally, it is very inconvenient that the parameters in Duffings equation are determined in the function
itself. We should be able to explore the rich behavior of Duffings equation without having to constantly
modify the function in fact, once we have the function exactly as we want it, we should never touch
it again. (This is not only true for esthetic reasons; the more we fool around with the function, the more
likely we are to screw it up!)
This is easily done by adding parameters to the function file.
function deriv = duffing p2(t, y, alpha, beta, Gamma, omega)
% duffing p2: Duffings equation, primary function
% with coefficients passed through the argument list
deriv = [ y(2) ; y(1)*(1-beta*y(1) 2)-alpha*y(2)+Gamma*cos(omega*t) ];
(The p2 denotes the fact that duffing p2 is another primary function.) However, this function cannot
be called directly by the ode solver. Instead it is called indirectly using closure by
%%%%% script m-file: duffing closure
alpha = 0.05;
beta = 1.0;
Gamma = 0.5;
omega = 1.0;
duffing c = @(t, y) duffing p2(t, y, alpha, beta, Gamma, omega);
ode45(duffing c, [0 100], [2 1]);
(which is contained in the accompanying zip file). Notice that the function duffing c takes only two
arguments: t and y. But the function it invokes is duffing p2 which takes six arguments. Thus,
ode45 thinks it is only passing two arguments to duffing c, but it is actually passing six arguments to
duffing p2.
To see a sampling of the different type of behavior in Duffings equation, let = 0.15, = 1, = 0.3
and = 1, and let the initial condition be y(0) = (0, 1)T . After a short initial transient, the solution settles down and appears to be regular by t = 100: in fact, it appears to be exactly periodic with a period
of 2 due to the 0.3 cos t term. (In fact, to the accuracy of the computer it is exactly periodic.) However, if we merely change the initial condition to y = (1, 0)T the behavior appears to be chaotic, even at
t = 1000. Here is an example of a ode which has periodic motion for one initial condition and is chaotic
for another! If we now change from 0.15 to 0.22 we find periodic motion with a period of 6. This is
just a sampling of the behavior of Duffings equation in different parameter regions.
By the way, to separate the initial transient behavior from the long-time behavior, you can use the
script m-file
124
10.2.
Advanced Commands
There are a number of parameters that we can use to tune the particular ode solver we choose. The
MATLAB function odeset is used to change these parameters from their default values by
>> params = odeset( 0 <Prop 1> 0 , <Value 1>, 0 <Prop 2> 0 , <Value 2>, ...)
where each parameter has a particular name and it is followed by the desired value. The result of this
command is that the parameters are contained in the variable params. You include these parameters in
the ode solver by adding this variable to the argument list of the ode solver function as
>> [t, Y] = <ode solver>(<function handle>, tspan, y0, params)
Some of the more common parameters are shown in the table below; they will be discussed further later.
To determine all the parameters, their possible values and the default value, enter
>> odeset
125
Assigns values to properties; these are passed to the ode solver when it is
executed.
The absolute error tolerance. This can be a scalar in which case it applies to all the
elements of y or it can be a vector where each element applies to the corresponding
element of y. (Default value: 106 .)
Events
A handle to a function which determines when an event occurs.
Jacobian A handle to a function which returns the Jacobian.
JPattern A sparse matrix whose nonzero elements (which should be 1 ) correspond to the possible nonzero elements of the Jacobian. This is only used when the Jacobian is calculated
numerically, i.e., when the Jacobian property is not used.
OutputFcn A handle to a function which is called after each successful time step. For example, a
plot of the trajectory can be generated automatically as it is being calculated.
Useful MATLAB functions are:
0
odeplot 0 which generates a plot of time versus all the components of the trajectory,
i.e., t vs. y1 , y2 , . . . , yn ;
0
odephas2 0 which generates a plot of y1 vs. y2 , i.e., Y(:,1) vs. Y(:,2);
0
odephas3 0 which generates a plot of y1 vs. y2 vs. y3 , i.e., Y(:,1) vs. Y(:,2)
vs. Y(:,3).
It is possible to plot different components of y using OutputSel.
OutputSel A vector containing the components of Y which are to be passed to the function specified by the OutputFcn parameter.
Refine
Refines the times which are output in t. This integer value increases the number of
times by this factor. (Default value: 1 for all ode solvers except ode45, 4 for ode45.)
RelTol
The relative error tolerance. (Default value: 103 ).
Stats
Whether statistics about the run are output on the terminal (value: 0 on 0 ) after the
trajectory is calculated or they are not (value: 0 off 0 ). (Default value: 0 off 0 .)
AbsTol
For example, if you want to use ode45 with the relative error tolerance set to 106 for Duffings equation, enter
>> params = odeset( 0 RelTol 0 , 1.e-6);
>> [t, Y] = ode45(duffing a, tspan, y0, params);
The trajectory will be more accurate but the command will run slower. If you also want the statistics
on the performance of the particular ode solver used, enter
>> params = odeset( 0 RelTol 0 , 1.e-6, 0 Stats 0 , 0 on 0 );
>> [t, Y] = ode45(@duffing a, tspan, y0, params);
and the number of successful steps, the number of failed steps, and the number of times f(t, y) was evaluated will be printed on the terminal. This might be useful in optimizing the performance of the ode
solver if the command seems to be running excessively slowly. For implicit methods where the Jacobian
needs to be calculated, the number of times the Jacobian was evaluated, the number of LU decompositions,
and the number of times the linear system was solved will also be returned.
The ode solver can also record the time and the location when the trajectory satisfies a particular condition: this is called an event. For example, if we are calculating the motion of the earth around the sun,
we can determine the position of the earth when it is closest to the sun and/or farthest away; or, if we are
following the motion of a ball, we can end the calculation when the ball hits the ground or we can let it
continue bouncing. Enter
>> ballode
to see a simple example.
For example, suppose we want to record where and when a trajectory of Duffings equation passes
through y1 = 0.5. That is, we define an event to be whenever the first component of y passes
through 0.5 or +0.5. This can be done by modifying the primary function duffing ode and replac126
y1
y2
0
0
2y1 y2 1
=
y2
(1 y12 )y2 y1
1
(1 y12 )
.
The right-hand side can be coded as a nested function inside a primary function by
function vdp ode(mu, y0, final time)
ode45(@vdp n, [0 final time], y0);
%%%%% nested function follows
function deriv = vdp n(t, y)
deriv = [y(2); mu*(1-y(1) 2)*y(2)-y(1)];
end
end
This is not stiff unless is large. For example, let = 1 and solve the ode with initial conditions
y(0) = 1 and y 0 (0) = 0 for t [0, 100] using ode45. Then, plot the result and note the number of elements in t. Repeat this procedure using = 10 and increase the final time, if necessary, so that you still
see a few complete oscillations. Then let = 100, etc., until the time required to plot a few oscillations
becomes very large. Then use ode15s and note the huge difference in the time required.
There is no need to use the ode solver parameters JPattern or Jacobian in this example because this
ode is so nice. However, since they might be needed for a nastier ode, we include them by using
Vdp pattern = sparse([1 2 2], [2 1 2], [1 1 1], 2, 2);
params = odeset( 0 Jacobian 0 , @vdpj n, 0 JPattern 0 , Vdp pattern):
[t, Y] = <ode solver>(@vdp n, tspan, y0, opt);
where the Jacobian is calculated numerically using the nested function
function J = vdpj n(t, y)
% vdpj n: Jacobian for van der Pols equation
J = [ 0 1; -2*mu*y(1)*y(2)-1 mu*(1-y(1) 2) ];
end
for the elements determined by Vdp pattern. By the way, if we use the property JPattern but not
Jacobian then the Jacobian is calculated numerically just for the elements determined by the sparse matrix.
Note: Plotting the trajectory by
plot(t, Y)
is not very instructive. Instead, use
subplot(2,1,1)
plot(t, Y(:,1))
subplot(2,1,2)
plot(t, Y(:,2))
Our final example is slightly more complicated. Suppose we kick a ball into the air with initial speed
s and at an angle of , and we want to follow its motion until it hits the ground. Let the x axis be the
horizontal axis along the direction of flight and z be the vertical axis. Using Newtons laws we obtain the
ode system
x00 = 0 and z 00 = g
where g = 9.8 meters/second is the acceleration on the ball due to the earths gravity. The initial conditions are
x(0) = 0 , x0 (0) = s cos , z(0) = 0, and z 0 (0) = s sin
where we assume, without loss of generality, that the center of our coordinate system is the initial location
of the ball. We also want to determine four events in the balls flight: the highest point of the trajectory
of the ball and the time it occurs, the distance it travels and the time it hits the ground, and the x values and times when the ball reaches the height h > 0. But beware because the ball may never attain this
128
height!
Although these odes can be solved analytically (consult any calculus book), our aim is to give an example of how to use many of the advanced features of MATLABs ode solvers. (If we would include the
effects of air resistance on the ball, then these odes would become nonlinear and would not be solvable
analytically.) We convert Newtons laws to the first-order system
0
y1
y2
y2 0
=
y3
y4
y4
g
One complication with solving this system numerically is that we do not know when the ball will hit
the ground, so we cannot give the final time. Instead, we use a time, 10s/g which is much greater than
needed and we let the program stop itself when the ball hits the ground. In addition, we want the relative
error to be 106 . Finally, we want the trajectory (i.e., z vs. x ) to be plotted automatically.
The following is a completely self-contained example using nested functions.
129
Jn (t) and, for specificity, we will concentrate on n = 1. At t = 0 the ode reduces to y(0) = 0 and so
we require y(0) = 0. The free initial condition is y 0 (0) and for this example we choose y 0 (0) = 1.
If we write Bessels equation as
1 0
n2
y (t) + y (t) + 1 2 y(t) = 0
t
t
00
(10.2)
we clearly have a problem at t = 0 and for t 0. The ode solvers we discussed previously can handle (10.2) for t 1 with the initial conditions that y(1) and y 0 (1) are given. However, a completely
different method of solution is required for t 0 and the form (10.1) is preferred to (10.2).
When we convert Bessels equation to the first order system we again let y1 (t) = y(t) and y2 (t) = y 0 (t)
and leave the t2 in the numerator to obtain
0
y2
y1
=
t2 y20
ty2 (t2 1)y1
Previously, we have always written the first-order system as y0 = f(t, y), but this form has a problem
when t = 0. Instead, we write it as g(t, y, y0 ) = 0 so that
g(t, y, y0 ) =
y10 y2
2 0
t y2 + ty2 + (t2 1)y1
.
Finally, we not only have to input the initial condition y(0) = (0, 1)T , but we also have to input
T
y0 (0) = y10 (0), y20 (0) . It is easy to calculate y10 (0) = y2 (0), but y20 (0) y 00 (0) is more complicated.
Differentiate (10.1) with respect to t to obtain
t2 y 000 (t) + 3ty 00 (t) + t2 y 0 (t) + 2ty(t) = 0
and differentiate it again to obtain
t2 y 0000 (t) + 5ty 000 (t) + (t2 + 3)y 00 (t) + 4ty 0 (t) + 2y(t) = 0 .
Now set
>>
>>
>>
>>
>>
>>
for x [a, b]
where y Rn but conditions are given at both ends of the interval. If the boundary conditions are separated, then k conditions are given at x = a and n k other conditions are given at x = b. If the
boundary conditions are non-separated, then the conditions at x = a and at x = b are related. To allow
any of these boundary conditions we write the boundary conditions as y(a), y(b) = 0 where Rn .
For simplicity, we will only consider two closely related second-order odes, i.e., n = 2. This example
should enable you to study any boundary-value ode. Consider the two nonlinear boundary-value ordinary
differential equations
dy
d2 y
(x) + 2 (x) + ey(x) = 0
2
dx
dx
(11.1a)
dy
d2 y
(x) + 2 (x) + ey(x) = 0
2
dx
dx
(11.1b)
and
(11.2)
which are called Dirichlet boundary conditions. These two odes are quite simple, but also quite interesting
and challenging to solve for certain intervals in .
We could use the Neumann boundary conditions y 0 (0) = 4 and y 0 (1) = 7 by
0
y (0) 4
y(0), y(1) =
=0.
(11.3)
y 0 (1) + 7
Or we could use the mixed boundary conditions y(0) y 0 (0) = 1 and y(1) + 2y 0 (1) = 3 by
y(0) y 0 (0) 1
y(0), y(1) =
=0.
y(1) + 2y 0 (1) 3
Finally, we could use periodic boundary conditions, which are non-separated, by
y(1) y(0)
y(0), y(1) =
=0.
y 0 (1) y 0 (0)
(11.4)
(11.5)
The primary MATLAB function is bvp4c. However, the functions bvpinit and deval are also
needed. We solve the boundary value problem by
>> sol = bvp4c(<right-hand side>, <boundary conditions>, <initial guess>)
There are two functions we need to write: odefun is f(x, y) and bcfun is the boundary conditions. For
our example the ode given by
function yp = nlode(x, y)
global which ode eps
if which ode == 1
yp = [y(2); -eps*exp(y(1))-2*y(2)];
else
yp = [y(2); -(exp(y(1))+2*y(2))/eps];
end
where we use global to input which ode to use and . The boundary condition is given by
function bc = nlode bc(ya, yb)
bc = [ya(1); yb(1)];
Since these boundary conditions are particularly simple, we also include the function
function bc = nlode bc2(ya, yb)
bc = [ya(1)-ya(2)-1; yb(1)+2*yb(2)-3];
for mixed boundary conditions (11.4). In addition, we have to choose an initial guess for y(x) using
bvpinit by either
>> bvpinit(x, y init)
132
or
>> bvpinit(x, <initial guess function>)
For example, if we want the initial iterate to be a parabola which is zero at x = 0 and 1 and has maximum value A then y(x) = y1 (x) = 4Ax(1 x) and y 0 (x) = y2 (x) = 4A(1 2x) then we can write
>> x = linspace(0, 1, 21);
>> solinit = bvpinit(x, @nlode y ic);
where nrode y ic is written as
function y ic = nlode y ic(x)
global A
y ic = [4*A*x.*(1 - x); 4*A*(1-2*x)];
The only alternative is to write
>> x = linspace(0, 1, 21);
>> y1 val = ???;
>> y2 val = ???;
>> solinit = bvpinit(x, [y1 val; y2 val]);
where y1 val and y2 val are scalar values. Thus the initial guess is y1 = y1 val*ones(size(x)) and y2
= y2 val*ones(size(x)). This is often unacceptable because constant initial guesses may be so far from
the solution that convergence cannot be obtained. What we would like to do is
>> x = linspace(0, 1, 21);
>> y1 = 4*A*x.*(1 - x);
>> y2 = 4*A*(1 - 2*x);
>> solinit = bvpinit(x, [y1; y2]); % WRONG
This fails because y1 and y2 must be scalar variables and not vectors. If you really, really need y1 and
y2 to be vectors, then do not use bvpinit. Instead, specify the structure solinit directly by entering
>> x = linspace(0, 1, 21);
>> y1 = 4*A*x.*(1 - x);
>> y2 = 4*A*(1 - 2*x);
>> solinit.x = x;
>> solinit.y = [y1;y2];
Warning: This is dangerous because future versions of Matlab might change the fieldnames of the structure solinit. However, it works for now.
We are finally ready to solve this ode by
%%%%% script m-file: nlode1 >> global which ode e
>> global A
>> which ode = 1;
>> A = 1;
>> e = 3;
>> x = linspace(0, 1, 21);
>> solinit = bvpinit(x, @nlode y ic);
>> sol = bvp4c(@nlode, @nlode bc, solinit);
The solution is contained in sol and is extracted by deval. For example, if xi = (i 1)x where
x1 = 0 and xn = 1 then we determine, and plot, the numerical solution y by
%%%%% script m-file: nlode2 >> xpt = linspace(0, 1, 101);
>> Ypt = deval(sol, xpt);
>> plot(xpt, Ypt(1,:), xpt, Ypt(2,:), 0 r 0 )
Having done all this work, we now combine everything into the function m-file nlode all to show how
much easier it is to use nested functions and to combine everything into one primary function.
133
with a large value of , say = 1, and choose a reasonable initial guess. (Since the two odes are
identical when = 1 you can use the solution you found to ode (11.1a).) Then slowly decrease . For
example, when = 0.01 we have y 0 (0) 130 and when = 0.001 we have y 0 (0) 1300. In conclusion,
we want to remind you that for the odes we have discussed here almost any halfway reasonable initial
choice for the ode (11.1a) will converge to one of the two solutions and for the ode (11.1b) will converge to
the single solution. However, you might well find an ode for which this is not true.
Boundary-Value Solver
bvp4c(<right-hand side>,
<boundary conditions>,
<initial guess>)
bvpset
o
bvpinit(x,y)
bvpinit(x, <initial guess function>)
deval(x,y soln)
y1
x1 1
x2 1 c1
y2
.
=
..
..
.. .
. c2
.
xn 1
yn
135
In general, there is no solution to this overdetermined linear system. Instead, we find the least-squares
solution c = (c1 , c2 )T by
>> c = [x ones(n, 1)] \ y
We can plot the data points along with this straight line by
>> xx = linspace(min(x), max(x), 100);
>> yy = polyval(c, xx);
>> plot(xx, yy, x, y, 0 o 0 )
We can find the best fit by a polynomial of degree m < n, i.e., y = c1 xm + c2 xm1 + + cm+1 , by
calculating the least-squares solution to
Vc = y
where
xm
1
xm
2
V=
..
.
xm
n
x1m1
xm1
2
..
.
x1
x2
..
.
1
1
..
xm1
n
xn
c1
c2
and c =
... .
cn
point have the same first and second derivatives at this point.
pchip 0 : Piecewise cubic Hermite polynomials connect each pair of adjacent data points. This is
similar to spline but the second derivatives need not be continuous at the interior data
points. Instead, this interpolation is better at preserving the shape of the data. In particular, on intervals where the data is monotonic so is the piecewise polynomial, and on
intervals where the data is concave up or down so is the piecewise polynomial.
0
cubic 0 : The same as pchip.
An alternate way to interpolate these points is by using the two commands
>> pp = spline(x, y)
>> yvalues = ppval(pp, xvalues)
to generate and interpolate the cubic spline or
>> pp = pchip(x, y)
>> yvalues = ppval(pp, xvalues)
to generate and interpolate the piecewise cubic Hermite polynomials. The first command generates the
structure pp which contains all the information required to obtain a unique piecewise polynomial. The
second command interpolates the piecewise polynomial at the x values given by the vector xvalues.
Interpolation really means interpolation. If a value lies outside the interval [x1 , xn ] then, by default,
NaN is returned. This can be changed by adding a fifth argument:
If the fifth argument is a number, this value is returned whenever the value lies outside the interval.
If the fifth argument is 0 extrap 0 , extrapolation (using the same method) is used.
The command spline can be used instead of using interp1 with the method spline. With it you
can specify precisely the boundary conditions to use. Similarly, the command pchip can be used instead
of using interp1 with the method pchip or cubic.
0
Polynomial Functions
interp1(x, y, xvalues, <method>)
interp2
interp3
interpn
pchip
poly(<roots>)
polyder(q)
polyfit(x, y, n)
polyint(q)
polyval(q, x)
polyvalm(q, A)
ppval
roots(q)
spline
vander
137
f(x) =
x1 + x2 + sin(x1 x2 )
x1 x2 + 2 cos(x1 + x2 )
.
Instead, we can find a minimum of g(x) = f12 (x) + f22 (x). If the minimum value is 0, we have found
a zero of f if it is not zero, we have not found a zero of f. For example, if f is defined an an anonymous function the result of
>> xmin = fminsearch(f, [0 0])
is xmin = (.1324 . . . , 1.0627 . . .). We are not done since we still have to calculate g(xmin ). This is
2.4109 which is small but is it small enough? We can decrease the termination tolerance by
>> opt = optimset( 0 TolX 0 , 1.e-8, 0 TolFun 0 , 1.e-8)
>> xmin = fminsearch(f, [0 0], opt)
Since g(xmin ) = 2.31017 we can assume that we have found a zero of f.
MATLAB can also calculate definite integrals using three commands. The first is quad which uses
Rb
adaptive Simpsons method. To evaluate a f (x) dx by Simpsons method enter
>> quad(<function handle>, a, b)
The full argument list is
>> quad(<function handle>, a, b, tol, trace)
where tol sets the relative tolerance for the convergence test and information about each iterate is
printed if trace is non-zero.
The second is quadl which uses adaptive Gauss-Lobatto quadrature, which is a variant of Gauss
quadrature.
quadl uses the more accurate formula and so should require many fewer function evaluations. For example, quad calculates the exact integral (up to round-off errors) for polynomials of degree five whereas
quadl calculates the exact integral (up to round-off errors) for polynomials of degree nine.
The third is quadgk which uses adaptive Gauss-Kronrod quadrature. This command is more general
than the previous two because it is much more general:
The interval can be half-infinite (i.e., a = or b = + ) or fully infinute ( a = and
b = + ).
In addition the integrand can have an integrable singularity.
For example, the error in
>> f = @(x) 1./(1 + x. 2);
>> quadgk(f, 0, inf)
is 8.88181016 (the actual value is 2 ) and the error in
139
ymin
by
>> dblquad(<function handle>, xmin, xmax, ymin, ymax)
It can also calculate the triple integral
Z xmax Z ymax Z zmax
f (x, y, z) dxdy dz
xmin
ymin
zmin
by
>>
dblquad(<function handle>, a, b, c, d)
fminbnd(<function handle>, xmin, xmax)
quadl(<function handle>, a, b)
140
where
1
a0 =
T
2
ak =
T
2
bk =
T
f (t) dt
0
T
f (t) cos kt dt
for k N[1, ) .
f (t) sin kt dt
The coefficients a0 , a1 , a2 , . . . and b1 , b2 , . . . are called the real Fourier coefficients of f , and ak and bk
are the coefficients of the k th mode. The power of the function f (t) is
P =
1
T
f (t) 2 dt
so that
P = |a0 |2 +
1X
|ak |2 + |bk |2 .
2
k=1
|a0 |2
1
2
if k = 0
2
|ak | + |bk |
if k > 0
and the frequency of the k th mode is k/T cycles per unit time.
Since
eit + eit
eit eit
cos t =
and sin t =
,
2
2i
we can rewrite the real Fourier series as the complex Fourier series
f (t) = a0 +
X
1
2 (ak
k=1
so that
f (t) =
ck e2ikt/T
(14.2)
k=
where
The term power is a misnomer because the function f need not be related to a physical quantity for which
the power makes any sense. However, we will stick to the common usage.
To understand the physical significance of power, we begin with the definition of work. Consider a particle
~ . If the particle moves from the point P0 to P1 then the
which is under the influence of the constant force F
q
~
work done to the particle is F ~r , where ~r is the vector from P0 to P1 . The power of the particle is the work
~ q~v where ~v = ~r/t.
done per unit time, i.e., F
Next, consider a charge q which is moving between two terminals having a potential difference of V . The
work done on the charge is W = qV = ItV , where I is the current and t is the time it takes for the charge to
move between the two terminals. If R is the resistance in the circuit, V = IR and the power is
W
V2
= IV = I 2 R =
.
t
R
Thus, if we consider f (t) to be the voltage or the current of some signal, the instantaneous power in the signal is
proportional to f 2 (t) and the average power is proportional to
Z
2
1 T
f (t) dt .
T 0
P =
141
c0 = a 0
ck = 12 (ak ibk )
)
for k > 0 .
ck = 21 (ak + ibk )
(14.3)
The coefficients . . . , c2 , c1 , c0 , c1 , c2 , . . . are called the complex Fourier coefficients of f , and ck and
ck are the coefficients of the k th mode. (Note that these Fourier coefficients are generally complex.) We
can also calculate ck directly from f by
ck =
1
T
f (t)e2ikt/T dt
for k = . . . , 2, 1, 0, 1, 2, . . .
Note that if f is real, then ck = ck (by replacing k by k in the above equation). The power of f (t)
is
X
P = |c0 |2 +
|ck |2 + |ck |2
k=1
|c0 |2
if k = 0
2
|ck | + |ck |
if k > 0 .
We can only calculate a finite number of Fourier coefficients numerically and so we truncate the infinite
series at the M th mode. We should choose M large enough that
f (t)
M
X
ck e2ikt/T
k=M
f (tj ) =
k e2iktj /T
for j = 0, 1, 2, . . . , N 1
k e2ijk/N
for j = 0, 1, 2, . . . , N 1
k=M
M
X
(14.4)
k=M
for k = M, M + 1, . . . , M .
(14.5)
Note that tN is not used because f (tN ) has the same value as f (t0 ) and so does not provide us with an
independent equation.
142
Note: To repeat: the discrete Fourier coefficient k is a function of M , i.e., k (M ), and is generally not
equal to the continuous Fourier coefficient ck . However, as M we have k (M ) ck . For a
fixed M we generally only have k (M ) ck as long as |k| is much less than M . Of course, it
takes practice and experimentation to determine what much less than means.
We define the discrete Fourier series by
fFS (t) =
M
X
k e2ikt/T
k=M
It is our responsibility (using our experience) to choose M large enough that f (t) fFS (t). Given
f = (f0 , f1 , f2 , . . . , fN 1 )T , the Fourier coefficients are calculated in MATLAB by
>> fc = fft(f)/N
where the coefficients of the discrete Fourier transform are contained in fc in the order
T
0 , 1 , . . . , M 1 , M , M , M +1 , . . . , 2 , 1 .
The command fftshift changes the order to
M , M +1 , . . . , 2 , 1 , 0 , 1 , . . . , M 1 , M
T
not
M , M +1 , . . . , 2 , 1 0 , 1 , . . . , M 1 , M .
There is only one difficulty with our presentation. As we have already stated, the vector f has
N = 2M + 1 elements, which is an odd number. The Fast Fourier Transform (FFT, for short), which
is the method used to calculate the discrete Fourier coefficients by fft and also to recover the original
function by ifft, generally works faster if the number of elements of f is even, and is particularly fast if
it a power of 2.
The figure below shows the cputime needed to calculate fft(f) as a function of N . Since the vertical axis is logarithmic, it is clear that there is a huge difference in the time required as we vary N .The
dashed lines show the minimum and maximum asymptotic times as cn log2 n.
10
10
10
10
10
200
400
600
800
1000
n
1200
1400
1600
For N to be even, we have to drop one coefficient, and the one we drop is M . Now
N = 2M
is even. The discrete complex Fourier series is
M
1
X
fFS (t) =
k e2ikt/T
k=M
144
1800
2000
N 1
1 X
fj e2ijk/N
N j=0
for k = M, M + 1, . . . , M 2, M 1 .
T
T
145
(which is contained in the accompanying zip file). ( ychop is the real part of N*ifft(fcchop) because,
due to round-off errors, the inverse Fourier transform returns a slightly complex result.) The result is
remarkably good considering the size of the perturbation.
If f (t) is an odd function in t, i.e., f (t) = f (t) for all t, then the real trigonometric series can be
simplified to
X
2kt
f (t) =
bk sin
T
k=1
for t [0, T ] or for t [1/2 T, +1/2 T ]. Choosing the latter interval, we only need define f (t) for
t [0, 1/2 T ] and, additionally, state that it is an odd function. We discretize this infinite series by
fdst (t) =
n
X
bk sin
k=1
146
2kt
T
for t [0, 1/2 T ] and we discretize this interval by 0 = t0 < t1 < < tn < tn+1 = 1/2 T where t =
1/ T /(n + 1). We immediately have f (t ) = f (t
This leaves
0
2
n+1 ) = 0 so these two nodes
are notneeded.
us with the n coefficients bi i = 1, 2, . . . , n and the n data points
ti , f (ti ) i = 1, 2, . . . n .
T
Defining the vectors b = (b1 , b2 , . . . , bn )T and f = f (t1 ), f (t2 ), . . . , f (tn ) , the MATLAB functions
dst and idst switch between them by
>> f = dst(b)
>> b = idst(f)
Discrete Fourier Transform
fft(f)
ifft(fc)
fftshift(fc)
ifftshift(cf)
dst(b)
idst(f)
p p
a11
a11
p
a22
a22
. .
. .
Ap =
.
=
.
.
p
an1,n1
an1,n1
apnn
ann
Thus,
e =
ea11
ea22
. .
.
ean1,n1
eann
is
y(t) = yic eat .
Similarly, the solution of
y (t)
a11
1
y2 (t) a21
d
. = .
dt .. ..
an1
yn (t)
a12
a22
..
.
..
.
an2
y (t)
a1n
1
a2n y2 (t)
.
..
. ..
ann
yn (t)
Calculates
Calculates
Calculates
Calculates
matrix.
148
These tables summarize the functions and operations described in this tutorial. The number (or numbers) shown give the page number of the table where this entry is discussed.
Arithmetical Operators
+
*
.*
/
./
\
\
.\
Addition.(p. 7, 31)
Subtraction.(p. 7, 31)
Scalar or matrix multiplication.(p. 7, 31)
Elementwise multiplication of matrices.(p. 31)
Scalar division.(p. 7, 31)
Elementwise division of matrices.(p. 31)
Scalar left division, i.e., b\a = a/b. (p. 7)
The solution to Ax = b for A Cmn : when m = n and A is nonsingular this
is the solution Gaussian elimination; when m > n this is the least-squares approximation of the overdetermined system; when m < n this is a solution of the
underdetermined system.(p. 31, 83)
Elementwise left division of matrices i.e., B.\A = A./B. (p. 31)
Scalar or matrix exponentiation.(p. 7, 31)
Elementwise exponentiation of matrices.(p. 31)
Special Characters
:
;
,
...
%
149
Getting Help
demo
doc
help
helpbrowser
load
lookfor
profile
save
type
who
whos
C
Predefined Variables
ans
pi
eps
Inf
NaN
i
j
realmin
realmax
The default variable name when one has not been specified.(p. 9)
. (p. 9)
Approximately the smallest positive real number on the computer such that
1 + eps 6= 1. (p. 9)
(as in 1/0 ).(p. 9)
Not-a-Number
(as in 0/0 ).(p. 9)
1
.
(p.
9)
1 . (p. 9)
The smallest usable positive real number on the computer.(p. 9)
The largest usable positive real number on the computer.(p. 9)
Format Options
The default setting.(p. 11)
Results are printed to approximately the maximum number of digits of accuracy in
MATLAB.(p. 11)
format short e Results are printed in scientific notation.(p. 11)
format long e Results are printed in scientific notation to approximately the maximum number of
digits of accuracy in MATLAB.(p. 11)
format short g Results are printed in the best of either format short or format short e.
(p. 11)
format long g Results are printed in the best of either format long or format long e. (p. 11)
format short
format long
150
atan2d
atanh
ceil
conj
cos
cosd
cosh
Cotangent.(p. 13)
Cotangent (argument in degrees).(p. 13)
coth
Hyperbolic cotangent.(p. 13)
csc
Cosecant.(p. 13)
cscd
Cosecant (argument in degrees).(p. 13)
csch
Hyperbolic cosecant.(p. 13)
exp
Exponential function.(p. 13)
expm1(x) ex 1. (p. 13)
factorial Factorial function.(p. 13)
fix
Round toward zero to the nearest
integer.(p. 13)
floor
Round downward to the nearest
integer.(p. 13)
heaviside The Heaviside step function.(p. 13)
imag
The imaginary part of a complex
number.(p. 14)
log
The natural logarithm, i.e., to the
base e. (p. 13)
log10
The common logarithm, i.e., to the
base 10. (p. 13)
log1p(x) log(x + 1). (p. 13)
mod
The modulus after division.(p. 13)
real
The real part of a complex number.(p. 14)
rem
The remainder after division.(p. 13)
round
Round to the closest integer.(p. 13)
sec
Secant.(p. 13)
secd
Secant (argument in degrees).(p. 13)
sech
Hyperbolic secant.(p. 13)
sign
The sign of the real number.(p. 13)
sin
Sine.(p. 13)
sind
Sine (argument in degrees).(p. 13)
sinh
Hyperbolic sine.(p. 13)
sqrt
Square root.(p. 13)
tan
Tangent.(p. 13)
tand
Tangent (argument in degrees).(p. 13)
tanh
Hyperbolic tangent.(p. 13)
cot
cotd
151
Input-Output Functions
csvread
csvwrite
diary
fopen
fclose
fscanf
fprintf
input
load
importdata
Reads data into MATLAB from the named file, one row per line of input.(p. 57)
Writes out the elements of a matrix to the named file using the same format as
csvread. (p. 57)
Saves your input to MATLAB and most of the output from MATLAB to
disk.(p. 7)
Opens the file with the permission string determining how the file is to be accessed.(p. 85)
Closes the file.(p. 85)
Behaves very similarly to the C command in reading data from a file using any
desired format.(p. 85)
Behaves very similarly to the C command in writing data to a file using any desired format.
It can also be used to display data on the screen.(p. 85)
Displays the prompt on the screen and waits for you to enter whatever is desired.(p. 10)
Reads data into MATLAB from the named file, one row per line of input.(p. 57)
Similar to load but there need not be an equal number of elements in each
row.(p. 57)
A\B
b/A
B/A
A.*B
A. p
p. A
A. B
A./B
B.\A
152
Elementary Matrices
eye
false
ones
rand
randi
randn
rng
randperm(n)
speye
sprand
sprandsym
sprandn
true
zeros
Specialized Matrices
hilb
vander
toeplitz
The
The
The
The
The
153
Manipulating Matrices
cat
clear
diag
fliplr
flipud
ind2sub
ipermute
permute
spdiags
repmat
bsxfun
reshape
rot90
squeeze
sub2ind
triu
tril
[]
154
Two-Dimensional Graphics
plot
fill
semilogx
semilogy
loglog
ezplot
polar
ezpolar
linspace
logspace
xlabel
ylabel
title
axis
hold
hist
bar
stairs
errorbar
subplot
shg
Three-Dimensional Graphics
plot3
ezplot3
fill3
mesh
ezmesh
surf
ezsurf
view
meshgrid
ndgrid
pol2cart
zlabel
axis
contour
contourf
ezcontour
contour3
ezcontour3
clabel
subplot
colorbar
colormap
shg
drawnow
155
156
min
mean
norm
prod
sort
std
sum
157
chol
cond
condest
det
eig
eigs
inv
linsolve
lu
norm
null
orth
pinv
qr
rank
rref
svd
xor
&&
||
<
<=
==
>
>=
=
strcmp
Flow Control
break
case
continue
else
elseif
end
error
for
if
otherwise
switch
while
158
Logical Functions
all
any
exist
isequal
find
ischar
isempty
isfinite
isinf
islogical
isnan
logical
True if all the elements of a vector are true; operates on the columns of a matrix.(p. 100)
True if any of the elements of a vector are true; operates on the columns of a matrix.(p. 100)
False if this name is not the name of a variable or a file.(p. 100)
Tests if two (or more) arrays have the same contents.(p. 100)
The indices of a vector or matrix which are nonzero.(p. 100)
True if a vector or array contains character elements.(p. 100)
True if the matrix is empty, i.e., []. (p. 100)
Generates a matrix with 1 in all the elements which are finite (i.e., not Inf or
NaN ) and 0 otherwise.(p. 100)
Generates a matrix with 1 in all the elements which are Inf and 0 otherwise.(p. 100)
True for a logical variable or array.(p. 100)
Generates a matrix with 1 in all the elements which are NaN and 0 otherwise.(p. 100)
Converts a numeric variable to a logical one.(p. 100)
159
Debugging Commands
keyboard
dbstep
dbcont
dbstop
dbclear
dbup
dbdown
dbstack
dbstatus
dbtype
dbquit
return
160
Non-stiff ode solver; fourth-order, one-step method for the ode y0 = f(t, y).
(p. 122)
Non-stiff ode solver; second-order, one-step method.(p. 122)
Non-stiff ode solver; variable-order, multi-step method.(p. 122)
Stiff ode solver; variable-order, multi-step method.(p. 122)
Stiff ode solver; second-order, one-step method.(p. 122)
Stiff ode solver; trapezoidal method.(p. 122)
Stiff ode solver; second-order, one-step method.(p. 122)
Stiff ode solver; variable-order, multi-step method for the fully implicit ode
f(t, y, y0 ) = 0. (p. 131)
Assigns values to properties of the ode solver.(p. 126)
Boundary-Value Solver
bvp4c
bvpset
bvpinit
deval
Numerically solves y0 (x) = f(x, y) for x [a, b] with given boundary conditions
and an initial guess for y. (p. 135)
Assigns values to properites of bvp4c. (p. 135)
Calculates the initial guess either by giving y directly or by using a function
y = initial guess function(x). (p. 135)
Interpolate to determine the solution desired points.(p. 135)
quadl
161
Matrix Functions
expm
funm
logm
sqrtm
162
Solutions To Exercises
These are the solutions to the exercises given in subsections 1.9, 2.10, 3.6, and 4.7.
1.9.1a)
>>
>>
or
>>
answer:
b)
c = sqrt(a 2
6.1751
b 2
- 2*a*b*cosd(ab))
c)
>>
>>
format short e
asin( (b/c)*sin(ab*deg) ) / deg
or
>> asind( (b/c)*sind(ab) )
answer: 4.9448e+01
d)
1.9.2)
1.9.3)
>>
diary
triangle.ans 0
>>
>>
>>
>>
A = [1 2 3 4; 5 6 7 8; 9 10 11 12; 13 14 15 16]
A = [1:4; 5:8; 9:12; 13:16]
A = [ [1:4:13] 0 [2:4:14] 0 [3:4:15] 0 [4:4:16] 0 ]
b)
>>
2.10.2)
>>
2.10.3)
>>
2.10.4a)
>>
b)
>>
A = A - tril(A,-1)
2.10.5)
>>
x = [0:30] 0 . 2
2.10.6a)
>>
R = rand(5)
b)
>>
c)
>>
mean(mean(R))
d)
>>
S = sin(R)
e)
>>
r = diag(R)
2.10.7a)
>>
>>
>>
A = [1 2 3; 4 5 6; 7 8 10]
B = A .5
% or B = sqrtm(A)
C = A. .5
% or C = sqrt(A)
>>
>>
A A -
b)
% or
% or
x = [0:30]. 2 0
mean(R(:))
B 2
C. 2
163
Solutions To Exercises
3.6.1)
3.6.2a)
bandit 0 ,
>>
>>
>>
>>
>>
>>
c{1,1}=
c{2,1}=
c{1,2}=
c{2,2}=
c{1,3}=
c{2,3}=
b)
>>
>>
c{1,1}(1,2)
c{2,2}(5)
3.6.3a)
>>
b)
>>
c)
>>
4.7.1a)
>>
>>
>>
b)
>>
>>
>>
z = 1 + x + x. 2 /2 + x. 6 /6
hold on
plot(x, z)
c)
>>
plot(x, y-z)
d)
>>
>>
>>
>>
>>
>>
hold off
plot(x, y, 0 r 0 , x, z, 0 g 0 , x, y-z,
axis equal
xlabel( 0 x 0 )
ylabel( 0 y 0 )
title( 0 e i\pi = -1 is profound 0 )
>>
>>
>>
>>
>>
>>
>>
>>
>>
subplot(2, 1, 1)
hold off
plot(x, y, 0 r 0 , x, z, 0 g 0 )
axis equal
xlabel( 0 x 0 )
ylabel( 0 y 0 )
title( 0 e i\pi = -1 is profound 0 )
subplot(2, 1, 2)
plot(x, y-z)
>>
>>
>>
>>
>>
x = linspace(-3, 3, 91);
y = x;
[X, Y] = meshgrid(x, y);
% or just do [X, Y] = meshgrid(x, x);
Z = (X. 2 + 4* Y. 2) .* sin(2*pi*X) .* sin(2*pi*Y);
surf(X, Y, Z);
e)
4.7.2a)
rand(5)
0
uniform 0
pi 10
@(x) sin(exp(x))
c{1,2} 2
true
m0 )
164
Solutions To Exercises
% or
view([63 66])
4.7.3) x is a row vector containing all the points at which the function is to be evaluated. c is
a column vector of the speeds. We can consider these to be the horizontal P
and vertical
n
axes in two dimensions. We define h(x, c, t) = f (x ct) so that g(x, t) = k=1 h(x, ck , t).
meshgrid then turns these vectors into a mesh so that h(xj , ci , t) is evaluated by
h(X(i, j), C(i, j), t), where i refers to the vertical axis and j the horizontal one. (Since the
matrices X and C never change, they would not need to be actual arguments to h.) Thus,
h(t) (short for h(X, C, t) ) is a matrix whose (i, j)th element is f (xj ci t). We need to
multiply this by ai and then sum over i. This is done by multiplying the matrix R, whose
(i, j)th element is ai , elementwise by the matrix h(X, C, t). The sum command then adds all
the rows in each column.
165
Solutions To Exercises
166
ASCII Table
ASCII Table
Octal
Decimal
000
001
002
003
004
005
006
007
010
011
012
013
014
015
016
017
020
021
022
023
024
025
026
027
030
031
032
033
034
035
036
037
040
041
042
043
044
045
046
047
050
051
052
053
054
055
056
057
060
061
062
063
064
065
066
067
070
071
072
073
074
075
076
077
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Control
Sequence
@
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
[
\
]
Description
Octal
Decimal
Null character
Start of header
Start of text
End of text
End of transmission
Enquiry
Acknowledgment
Bell
Backspace
Horizontal tab
Line feed
Vertical tab
Form feed
Carriage return
Shift out
Shift in
Data link escape
Device control 1 (often XON)
Device control 2
Device control 3 (often XOFF)
Device control 4
Negative acknowledgement
Synchronous idle
End of transmissions block
Cancel
End of medium
Substitute
Escape
File separator
Group separator
Record separator
Unit separator
Space
!
Double quote
# Numer sign
$
Dollar sign
% Percent
& Ampersand
100
101
102
103
104
105
106
107
110
111
112
113
114
115
116
117
120
121
122
123
124
125
126
127
130
131
132
133
134
135
136
137
140
141
142
143
144
145
146
147
150
151
152
153
154
155
156
157
160
161
162
163
164
165
166
167
170
171
172
173
174
175
176
177
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
Control
Sequence
Description
@
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
[
\
]
AT symbol
a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p
q
r
s
t
u
v
w
x
y
z
{
|
}
Left bracket
Back slash
Right bracket
Caret
Underscore
Opening single quote
Left brace
Vertical bar
Right brace
Tilde
Delete
American Standard Code for Information Interchange (ASCII) specifies a correspondence between bit patterns and character symbols. The octal and decimal representations of the bit patterns are shown along
167
ASCII Table
with a description of the character symbol. The first 32 codes (numbers 031 decimal) as well as the last
(number 127 decimal) are non-printing characters which were initially intended to control devices or provide meta-information about data streams. For example, decimal 10 ended a line on a line printer and
decimal 8 backspaced one character so that the preceding character would be overstruck. The control
sequence column shows the traditional key sequences for inputting these non-printing characters where
the caret ( ) represents the Control or Ctrl key which must be held down while the following key is
depressed.
168
Index
In this index MATLAB commands come first, followed by symbols, and then followed by the function m-files and
named script files which are coded in this tutorial and contained in the corresponding zip file. Only then does the
index begin with the letter A.
Note: All words shown in typewriter font are MATLAB commands or predefined variables unless it is specifically
stated that they are defined locally (i.e., in this document).
Note: If an item is a primary topic of a section, an appendix, or a subsection, this is indicated as well as the page
number (in parentheses).
Note: When an item appears in a box at the end of a subsection, or in the reference tables in the appendix, it is
underlined.
MATLAB functions
abs, 13, 14, 151
acos, 13, 151
acosd, 13, 151
acosh, 13, 151
acot, 13, 151
acotd, 13, 151
acoth, 13, 151
acsc, 13, 151
acscd, 13, 151
acsch, 13, 151
addframe, 77, 156
addpath, 101, 112, 154
all, 36, 99, 100, 159
angle, 14, 151
any, 36, 98, 100, 159
asec, 151
asecd, 151
asech, 151
asin, 13, 151
asind, 13, 151
asinh, 13, 151
atan, 13, 151
atan2, 13, 151
atan2d, 13, 151
atand, 13, 151
atanh, 13, 151
avifile, 77, 156
axis, 52, 58, 61, 65, 155
ballode, 126
bar, 55, 155, 157
break, 95, 158
bsxfun, 27, 28, 154
bvp4c, 132, 133, 135, 161
bvpinit, 133, 135, 161
bvpset, 134
case, 95, 158
different than in C, 95
cat, 37, 38, 154
caxis, 64, 66, 156
ceil, 13, 151
cell, 44, 46, 47, 157
celldisp, 44, 46, 157
char, 47
169
170
171
Symbols
+, 7, 28, 31, 149, 152
exception to, 30
-, 7, 28, 31, 149, 152
*, 7, 28, 29, 31, 149, 152
.*, 29, 31, 149, 152
/, 7, 29, 31, 149, 152
warning about matrix division, 29
./, 30, 31, 149, 152
\, 7, 31, 78, 79, 82, 83, 149, 152
.\, 30, 31, 152
, 6, 7, 29, 31, 149, 152
. , 30, 31, 152
0
, 7, 19, 22, 153
. 0 , 19, 22, 153
..., 15, 149
%, 15, 149
,, 7, 9, 18, 28, 149
;, 7, 9, 18, 28, 149
:, 19, Subsect. 2.2 (22), Subsect. 2.3 (23), 28, 149
<, 93, 158
A
AH , See Conjugate transpose
AT , See Transpose
A+ , See Matrix, pseudoinverse of
Abort current statement, 15
abs, 13, 14, 151
Accuracy, 10
principle, 11
acos, 13, 151
172
B
Ball, 128130
ballode, 126
C
15, 17, 150
CC,(programming
language),
173
Data (cont.)
D
Data
best polynomial fit to, 135
closing files, 84
manipulation, Subsect. 2.8 (34), 157
opening files, 83, 84
E
ez , 14
Earth, 65
echo, 105, 112, 159
edit, 14, 15, 16, 154
eig, 32, 87, 90, 103, 158
Eigenvalues, 32, 85, 87, 90, 103
definition of, 87
Eigenvectors, 87, 90, 103
eigs, 87, 90, 158
else, 93, 95, 158
elseif, 93, 95, 158
end, 23, 91, 92, 94, 95, 109, 112, 158, 159
174
F
factorial, 11, 13, 151
Factorial function, 11
false, 94, 95, 153
FALSE (result of logical expression), 93
Fast Fourier transform, See Fourier transform
fclose, 84, 85, 152
feval, 113, 114, 117, 159
fft, 143, 147, 160
fftshift, 143, 145, 147, 160
Fibonacci sequence, 112
Field, See Structure
fieldnames, 46, 157
figure, 63, 66, 156
fill, 64, 66, 155
fill3, 64, 66, 155
find, 97, 98, 100, 119, 120, 159, 160
Finite differences, 36
fix, 13, 151
fliplr, 25, 28, 154
flipud, 25, 28, 154
Floating-point numbers, 9, 23
Floating-point operations, See Flops
floor, 13, 151
Flops (fl oating-point operations), 30
Flow control, See Programming language
fminbnd, 138, 140, 161
fminsearch, 139, 140, 161
fopen, 83, 84, 85, 152
for, 23, 91, 95, 158
format, 10, 11, Subsect. 2.6 (32), 150
Format options (in format command), 10, 11, 150
Format specifications (in fprintf, fscanf, sprintf, and
sscanf), 84
Fourier series, Sect. 14 (140)
complex, 141
real, 140
Fourier transform, Sect. 14 (140)
discrete, Sect. 14 (140), 147, 160
discrete sine, 146
fast (FFT), 144
fplot, 42, 53, 66, 156
fprintf, 8, 43, 55, 83, 85, 152
printing a matrix, 84
specifications (format), 84
Frequency, See Power
fscanf, 55, 83, 84, 85, 152
specifications (format), 84
full, 118, 120, 160
function, 100, 108, 112, 159
Function, 23, Subsect. 8.3 (100)
anonymous, Subsect. 3.1 (40), 46
warning, 41
built-in, 11, 15, 16
commands in, 106, 112, 159, 160
comments in, 101
conflict between function and variable name, 12
debugging, 104, 105
definition line, 100
differences from command, 3
end statement, 108, 109
ending, 109
error, 102, 112, 158, 159
example using multiple input and output arguments,
103
function (required word), 100, 108, 112, 159
inline, 41
warning, 41
input and output arguments, 100, 106
grouping together, 108
pass by reference, 101
pass by value, 101
variable number of, 102
name of, 14
warning about user-defined m-files, 15
nested, 101, 108, 109
order in which MATLAB searches for functions, 101,
111
passing arguments indirectly, See Closure
passing function name in argument list, Subsect. 3.2
(41), 113, 114
piecewise, 41
primary, 108
private, 111
return, 102, 105, 106, 112, 159, 160
saving parameters in, 107108
subfunctions in, 101, 108
Function handle, 40, 46
function handle, 47
Function m-file, Subsect. 8.3 (100)
debugging, 104, 105
names of, 15, 100
passing name into, 42
recursive, 111
175
Functions (mathematical)
See also Polynomials
common mathematical, Subsect. 1.5 (11)
definite integrals of, 139, 140
hijacked, 111
local minimum of, 138
numerical operations on, 137, 140, 161
zeroes of, 138, 139
funm, 148, 162
fzero, 138, 140, 161
G
Gauss-Kronrod quadrature (for numerical integration),
139
Gauss-Lobatto quadrature (for numerical integration),
139
Gaussian elimination, 78, 82
gca, 68, 156
gcf, 68, 156
Generalized eigenvalue problem, 87
get, 67, 68, 156
getfield, 46, 157
getframe, 77, 156
ginput, 62, 66, 156
global, 107, 112, 159
Gram-Schmidt algorithm, 90
Graphical image, 65
export, 65
extensions, 65
import, 65
show, 65
Graphics, Sect. 4 (49)
advanced techniques, Subsect. 4.3 (61), 156
changing endpoints, 52
customizing lines and markers, 50
demonstration, 49
handle, Subsect. 4.4 (66)
holding the current plot, 50
labelling, 6266
text properties, 67
using TEX commands, 62, 67
multiple plots, 52
multiple windows, 63
object, 66
handle for an, 66
printing, 56, 57, 152
properties, Subsect. 4.4 (66)
saving to a file, 56, 57, 152
three-dimensional, Subsect. 4.2 (58)
two-dimensional, Subsect. 4.1 (49), 155
window, 49
Gravity, 128
gtext, 62, 63, 66, 67, 156
GUI (Graphical User Interface), Subsect. 4.5 (68)
guide, 68, 74, 156
H
H
I
I, See Identity matrix
i, 6, 9, 150
Identity matrix, 20
See also eye
idst, 147, 160
if, 23, 92, 95, 158
ifft, 143, 147, 160
ifftshift, 143, 145, 147, 160
imag, 14, 151
image, 65, 66, 156
imagesc, 65, 66, 156
Imaginary numbers, 6, 9, 150
importdata, 56, 57, 152
imread, 65, 66, 156
imshow, 65, 66, 156
imwrite, 65, 66, 156
ind2sub, 97, 100, 154
Inf, 9, 52, 150
Initial-value ordinary differential equations,
(120)
Bessels equation, 130
Duffings equation, 120127
first-order system, 120
with constant coefficients, 147
solvers, 121, 122, 161
absolute error, 121
adaptive step size, 121
events, 127
ode15i, 130, 131, 161
ode15s, 122, 161
ode23, 122, 161
ode23s, 122, 161
ode23t, 122, 161
ode23tb, 122, 161
ode45, 122, 161
ode113, 122, 161
passing parameters to, 124
properties of, 126
relative error, 121
statistics for, 126
stiff, 122, 128
Van der Pols equation, 127128
inline, 41
176
Sect. 10
j, 6, 9, 150
Java (programming language), 113
K
keyboard, 105, 106, 159, 160
Keywords, 8
Kill current statement, 15
L
lasterr, 113, 114, 117, 159
Left division, See \
legend, 62, 66, 156
Lemniscate of Bernoulli, 53
length (number of elements in), 20, 22, 36, 97, 153
Length of a vector (i.e., Euclidean length), 36
See also norm
Life (Conways game of), 74
Line properties, 61
Linear splines, See Interpolation
Linear system of equations, Sect. 5 (77), Subsect. 5.3
(82)
least-squares solution, 82, 136
overdetermined, Subsect. 5.3 (82), 136
solving by \, 29, 78, 82
solving by linsolve, 78
solving by rref, Sect. 5 (77)
underdetermined, Subsect. 5.3 (82)
M-files, 100
See also Function m-file and Script m-file
Machine epsilon (eps), 9, 150
calculation of, 94
Mathematical functions, Subsect. 1.5 (11), 14, Subsect.
2.7 (34), 151
Matrix
as column vector, 24, 97
augmented, 7780
is not a matrix, 78
Cholesky decomposition, 85
condition number, 85
approximation to, 86
defective, 87
deleting rows or columns, 26
determinant of, See Determinant
diagonals of, 24, 118, 119
elementary, 22, 153
elementary operations, 153
empty, See null (below)
extracting submatrices, 23
full, 117
generating, Subsect. 2.1 (18), Subsect. 2.3 (23)
by submatrices, 21
individual elements, 20
Hermitian, 19
Hilbert, See Hilbert matrix
177
Matrix (cont.)
identity, 20
inverse of, 88
Jacobian, 123, 128
lower triangular part of, 25, 88
unit, 88
LU decomposition, 88
manipulation, Subsect. 2.3 (23), 154
masking elements of, 98
maximum value, 35
minimum value, 35
multidimensional, Subsect. 2.9 (37)
null, 26, 28, 34, 154
orthogonal, 89
positive definite, 119
preallocation of, 20, 44, 102
pseudoinverse of, 82, 90
QR decomposition, 89
replicating, 26
reshaping, 24, 26
singular, 79, 80, 81, 85, 86
warning of, 88
singular value decomposition, 90
sparse, Sect. 9 (117), 160
specialized, 153
sum of elements, 35
SVD, See singular value decomposition (above)
symmetric, 19, 119
Toeplitz, 25
tridiagonal, 86, 117
unitary, 89
upper triangular part of, 25
Vandermonde, See Vandermonde matrix
max, 34, 37, 157
Maximum value, 34
mean, 35, 37, 157
Mean value, 35
Memory (of variables), 34
mesh, 59, 61, 63, 155
meshgrid, 59, 60, 61, 155
methods, 48
min, 35, 37, 157
Minimum value, 35
mod, 13, 151
Monotonicity, test for, 36
Monty Python, 42
Moore-Penrose conditions, 82
Moore-Penrose pseudoinverse, See Matrix,
pseudoinverse of
Mouse location, See ginput
movie, 77, 156
movie2avi, 77, 156
Multidimensional arrays, Subsect. 2.9 (37)
generate grid, 60
permute order, 38
N
NaN, 9, 150
nargin, 103, 112, 159
nargout, 103, 112, 159
O
Ode, See Initial-value ordinary differential equations
ode15i, 130, 131, 161
ode15s, 122, 161
ode23, 122, 161
ode23s, 122, 161
ode23t, 122, 161
ode23tb, 122, 161
ode45, 122, 161
ode113, 122, 161
odeset, 125, 126, 130, 134, 161
ones, 20, 22, 153
Operator precedence, Subsect. 2.5 (31)
optimset, 138, 140, 161
OR (logical operator), 93, 94, 96, 158
orderfields, 46, 157
Ordinary differential equations, See Initial-value
ordinary differential equations and Boundary-value
ordinary differential equations
orth, 89, 90, 158
Orthonormal basis, 89
otherwise, 95, 158
Outer product, 30
Overdetermined system, See Linear system of equations
P
Parentheses, 9
path, 101, 112, 154
Path, See Search path
pause, 52, 104, 112, 154
pchip, 137, 162
permute, 38, 154
persistent, 107, 112, 159
uses of, 107
Phase plane, See Plotting
pi, 8, 9, 150
Piecewise polynomials, See Interpolation
pinv, 83, 90, 158
plot, 49, 50, 53, 58, 61, 136, 155
line properties, See Line properties
using set rather than plot, 75, 76
Plot, generating a, See Graphics
178
Q
qr, 89, 90, 158
QR decomposition, 89
quad, 139, 140, 161
quadgk, 139, 140, 161
quadl, 139, 140, 161
Quadratic polynomial, roots of,
Quote mark, 7
13
R
rand, 20, 22, 54, 81, 82, 153
randi, 21, 22, 153
randn, 20, 22, 54, 153
Random matrix, 20, 25, 81, 82, 120, 160
Random numbers, 20
Gaussian distribution, 20, 54
normal distribution, 20
pseudorandom numbers, 20
initial seed, 20
recommended procedure
state, 20
uniform distribution, 20, 54
randperm, 21, 22, 153
rank, 90, 158
Rank of matrix, 90
Rational approximation to floating-point number, 83,
154
rats, 83, 154
RCOND, 81, 86, 88
real, 14, 151
realmax, 9, 150
realmin, 9, 10, 150
Recursion, 111
Reduced row echelon form, 78
round-off errors in, 80
Relational operators, 93, 158
<, 93, 158
<=, 93, 158
>, 93, 158
>=, 93, 158
==, 93, 158
=, 93, 158
matrix, Subsect. 8.2 (96)
result of, 96
rem, 13, 151
Remainder, 13, 151
repmat, 26, 28
Request input, 10
reshape, 24, 26, 28, 154
return, 102, 105, 106, 112, 159, 160
Reverse two variables, 24
RGB components (of a color), 63
rmfield, 46, 157
randn, 20, 22, 54, 153
rng, 20, 22, 153
roots, 135, 137, 162
rot90, 28, 154
round, 13, 151
Round-off errors, Subsect. 1.3 (9), 11, 23, 25, 29, 33, 34,
179
Statements (cont.)
rerunning previous, 10
separating on a line, 7, 9, 18, 28, 149
std, 35, 37, 157
Step function, plotting a, 55
Stiff ode, 122, 128
str2num, 43, 157
strcmp, 93, 157, 158
strfind, 97, 100, 157
String, See Character string
strtrim, 43, 157
struct, 45, 46, 47, 157
Structure, 40, 45, 138
equal (are two structures), 99
field, 45
sub2ind, 97, 100, 154
Subfunctions, See Function m-file
subplot, 52, 58, 61, 155
warning, 52
sum, 37, 96, 157
surf, 59, 61, 63, 155
Surface plot, 59
changing view, 59
filled-in, 59
wire-frame, 59
svd, 90, 158
SVD, See Singular value decomposition
switch, 23, 95, 158
S
save, 16, 17, 150
Save terminal commands, 6
Save work, 6
saveas, 77, 156
Scientific notation, 6
Scope, See Variables
Script m-file, 14, 101, 105
debugging, 104, 105
names of, 14, 15
Search path, 101, 111
sec, 13, 151
secd, 13, 151
sech, 13, 151
semilogx, 53, 58, 155
semilogy, 53, 58, 155
set, 66, 67, 68, 156
setfield, 46, 157
shg, 51, 58, 155, 156
Short circuiting (logical operators), See LOGICAL
AND and LOGICAL OR
sign, 13, 151
Simpsons method (of numerical integration), 139
sin, 13, 151
sin z, 14
sind, 13, 151
single, 47
Singular value decomposition, 90
sinh, 13, 151
size, 20, 22, 153
sort, 36, 37, 45, 157
Sort numbers, 35
sparse, 117, 118, 120, 160
spconvert, 119, 120, 160
spdiags, 118, 120, 160
differences from diag, 118
speye, 120, 153, 160
spfun, 120, 160
spline, 137, 162
Splines, See Interpolation
sprand, 119, 120, 153, 160
sprandn, 119, 120, 153, 160
sprandsym, 119, 120, 153
sprintf, 43, 157
specifications (format), 84
spy, 120, 160
sqrt, 13, 34, 151
sqrtm, 29, 148, 162
squeeze, 38, 154
sscanf, 43, 157
specifications (format), 84
stairs, 55, 58, 155
Stairstep graph, 55
Standard deviation, 35
Statements
executing in text variables, 113
T
T
, See Transpose
tan, 13, 151
tand, 13, 151
tanh, 13, 151
Taylor series expansion, 147
TEX, See Character string
text, 62, 63, 66, 156
Text properties, 67
Text window, 49
tic, 31, 152, 154
Time, See cputime and tic and toc
title, 54, 58, 67, 155
multiline, See Character string, multiline
Title
for entire figure, 68
Toeplitz matrix, See Matrix
toeplitz, 25, 28, 153
toc, 31, 152, 154
Transpose, 19, 22, 153
conjugate, 19, 22, 153
Trigonometric functions, Subsect. 1.5 (11), Subsect. 2.7
(34)
tril, 25, 28, 154
triplequad, 140
triu, 25, 28, 154
true, 94, 95, 153
TRUE (result of logical expression), 93
type, 15, 16, 17, 101, 109, 150
180
Vector (cont.)
U
uicontrol, 70, 74, 156
uint8, 47
uint16, 47
uint32, 47
uint64, 47
uipanel, 70, 74, 156
uiresume, 74, 156
uiwait, 74, 156
Underdetermined system,
equations
V
Van der Pols equation, See Initial-value ordinary
differential equations
vander, 136, 137, 153
Vandermonde matrix, 136
varargin, 108, 112, 159
varargout, 108, 112, 159
Variables, Subsect. 1.2 (7)
about, 9
case sensitive, 9
conflict between variable and function name, 12
defined, 99
deleting, 9
global, 106, 107
inputting, 10
list of, 16
loading, 16
local, 100, 106, 107
logical, 98
modifying, 106, 107
overwriting, 7
persistent, 107
predefined, 7, 9, 150
ans, 8, 9, 150
eps, 9, 10, 94, 150
i, 6, 9, 150
Inf, 9, 52, 150
j, 6, 9, 150
NaN, 9, 150
overwriting, 8, 92
pi, 8, 9, 150
realmax, 9, 150
realmin, 9, 10, 150
reverse two, 24
saving, 16
saving local variables in functions, 107
scope of, 110
special cases of vectors or matrices, 7
static, 107
string, 7, Subsect. 3.3 (42)
See also Character string
typeless, 7, 106
valid names, 8
Vector
average value of elements, 35
column vs. row, 18
deleting elements, 26
W
while, 23, 94, 95, 158
who, 16, 17, 150
whos, 16, 17, 150
Workspace, 6, 105
X
xlabel, 54, 58, 62, 67, 155
xor, 93, 94, 96, 158
Y
ylabel, 54, 58, 62, 67, 155
Z
zeros, 20, 22, 153
zlabel, 59, 61, 67, 155
181
Index
182