Matlab tutorial by manish myst, ssgbcoet

Crash course in MATLAB
c Tobin A. Driscoll.∗ All rights reserved.

June 23, 2006

The purpose of this document is to give a medium-length introduction to the essentials of
MATLAB and how to use it well. I’m aiming for a document that’s somewhere between a two-
page introduction and the excellent all-inclusive books available. I use it in a week-long “camp”
for graduate students at the University of Delaware in the summer after their first year of study.
No advance knowledge of MATLAB is necessary to read this document, but a working fa-
miliarity with basic linear algebra is expected, and general knowledge of programming is a big
help. The first four sections cover the basics needed to solve even simple exercises. Section 5 is a
bare-bones introduction to the big subject of graphics in MATLAB. The next three sections discuss
matters than an intermediate MATLAB programmer ought to know. The remaining sections go
into advanced issues in varying degrees of detail.
The version of MATLAB at this writing is 7.1.

Contents
1 Introduction 3
1.1 The fifty-cent tour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Graphical versus command-line usage . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Basic commands and syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Saving work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Arrays and matrices 9
2.1 Building arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Referencing elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Matrix operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Array operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Sparse matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
∗
Department of Mathematical Sciences, Ewing Hall, University of Delaware, Newark, DE 19716;
driscoll@math.udel.edu.

1

CONTENTS 2

3 Scripts and functions 25
3.1 Using scripts effectively . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Conditionals: if and switch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Loops: for and while . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5 Debugging and profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 More on functions 32
4.1 Subfunctions and nested functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 Anonymous functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3 Function functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 Errors and warnings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5 Graphics 38
5.1 2D plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2 3D plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.3 Annotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.4 Handles and properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.5 Color . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.6 Saving and exporting figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

6 Speed, style, and trickery 47
6.1 Functions or scripts? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Memory preallocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.3 Vectorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.4 Masking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.5 Scoping exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

7 Advanced data structures 55
7.1 Strings and formatted output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2 Cell arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7.3 Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

8 Scientific computing 60
8.1 Linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
8.2 Iterative linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.3 Rootfinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.4 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.5 Fitting data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.6 Quadrature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

1 INTRODUCTION 3

8.7 Initial-value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.8 Boundary-value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.9 Initial-boundary value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

1 Introduction
MATLAB is a software package for computation in engineering, science, and applied mathemat-
ics. It offers a powerful programming language, excellent graphics, and a wide range of expert
knowledge. MATLAB is published by and a trademark of The MathWorks, Inc.
The focus in MATLAB is on computation, not mathematics: Symbolic expressions and ma-
nipulations are not possible (except through the optional Symbolic Toolbox, a clever interface to
Maple). All results are not only numerical but inexact, thanks to the rounding errors inherent in
computer arithmetic. The limitation to numerical computation can be seen as a drawback, but it
is a source of strength too: MATLAB is much preferred to Maple, Mathematica, and the like when
it comes to numerics.
On the other hand, compared to other numerically oriented languages like C++ and FOR-
TRAN, MATLAB is much easier to use and comes with a huge standard library.1 The unfavorable
comparison here is a gap in execution speed. This gap is not always as dramatic as popular lore has
it, and it can often be narrowed or closed with good MATLAB programming (see section 6). More-
over, one can link other codes into MATLAB, or vice versa, and MATLAB now optionally supports
parallel computing. Still, MATLAB is usually not the tool of choice for maximum-performance
computing.
The MATLAB niche is numerical computation on workstations for non-experts in computa-
tion. This is a huge niche—one way to tell is to look at the number of MATLAB-related books
on mathworks.com. Even for supercomputer users, MATLAB can be a valuable environment in
which to explore and ﬁne-tune algorithms before more laborious coding in another language.
Most successful computing languages and environments acquire a distinctive character or cul-
ture. In MATLAB, that culture contains several elements: an experimental and graphical bias,
resulting from the interactive environment and compression of the write-compile-link-execute-
analyze cycle; an emphasis on syntax that is compact and friendly to the interactive mode, rather
than tightly constrained and verbose; a kitchen-sink mentality for providing functionality; and a
high degree of openness and transparency (though not to the extent of being open source soft-
ware).

1.1 The ﬁfty-cent tour
When you start MATLAB, you get a multipaneled desktop. The layout and behavior of the desk-
top and its components are highly customizable (and may in fact already be customized for your
site). The component that is the heart of MATLAB is called the Command Window, located on the
1
Here and elsewhere I am thinking of the “old FORTRAN,” FORTRAN 77. This is not a commentary on the useful-
ness of FORTRAN 90 but on my ignorance of it.

1 INTRODUCTION 4

right by default. Here you can give MATLAB commands typed at the prompt, >>. Unlike FOR-
TRAN and other compiled computer languages, MATLAB is an interpreted environment—you
give a command, and MATLAB tries to execute it right away before asking for another.
At the top left you can see the Current Directory. In general MATLAB is aware only of files
in the current directory (folder) and on its path, which can be customized. Commands for work-
ing with the directory and path include cd, what, addpath, and editpath (or you can choose
“File/Set path. . . ” from the menus). You can add files to a directory on the path and thereby add
commands to MATLAB; we will return to this subject in section 3.
Next to the Current Directory tab is the Workspace tab. The workspace shows you what vari-
able names are currently defined and some information about their contents. (At start-up it is,
naturally, empty.) This represents another break from compiled environments: variables created
in the workspace persist for you to examine and modify, even after code execution stops.
Below the Command Window/Workspace window is the Command History window. As you
enter commands, they are recorded here. This record persists across different MATLAB sessions,
and commands or blocks of commands can be copied from here or saved to files.
As you explore MATLAB, you will soon encounter some toolboxes. These are individually
packaged sets of capabilities that provide in-depth expertise on particular subject areas. There is
no need to load them explicitly—once installed, they are always available transparently. You may
also encounter Simulink, which is a semi-independent graphical control-engineering package not
covered in this document.

1.2 Graphical versus command-line usage
MATLAB was originally entirely a command-line environment, and it retains that orientation.
But it is now possible to access a great deal of the functionality from graphical interfaces—menus,
buttons, and so on. These interfaces are especially useful to beginners, because they lay out the
available choices clearly.2
As a rule, graphical interfaces can be more natural for certain types of interactive work, such as
annotating a graph or debugging a program, whereas typed commands remain better for complex,
precise, repeated, or reproducible tasks. One does not always need to make a choice, though; for
instance, it is possible to save a figure’s styles as a template that can be used with different data
by pointing and clicking. Moreover, you can package code you want to distribute with your
own graphical interface, one that itself may be designed with a combination of graphical and
command-oriented tools. In the end, an advanced MATLAB user should be able to exploit both
modes of work to be productive.
That said, the focus of this document is on typed commands. In many (most?) cases these have
graphical interface equivalents, even if I don’t explicitly point them out.
2
In particular, feel free to right-click (on Control-click on a Mac) on various objects to see what you might be able to
do to them.

1 INTRODUCTION 5

1.3 Help
MATLAB is huge. Nobody can tell you everything that you personally will need to know, nor
could you remember it all anyway. It is essential that you become familiar with the online help.
There are two levels of help:

• If you need quick help on the syntax of a command, use help. For example, help plot
shows right in the Commmand Window all the ways in which you can use the plot com-
mand. Typing help by itself gives you a list of categories that themselves yield lists of
commands.

• Typing doc followed by a command name brings up more extensive help in a separate win-
dow.3 For example, doc plot is better formatted and more informative than help plot.
In the left panel one sees a hierarchical, browsable display of all the online documentation.
Typing doc alone or selecting Help from the menu brings up the window at a “root” page.

The Getting Started with MATLAB manual is a very good place to get a more gentle and thor-
ough introduction than the one to follow here. Depending on your installation, the documentation
may be available in PDF form for printing and ofﬂine reading.

1.4 Basic commands and syntax

If you type in a valid expression and press Enter, MATLAB will immediately execute it and return
the result, just like a calculator.

>> 2+2

ans =
4

>> 4ˆ2

ans =
16

>> sin(pi/2)

ans =
1

>> 1/0
Warning: Divide by zero.

ans =
Inf
3
Indeed, the help browser is a fully functional web browser as well.

1 INTRODUCTION 6

>> exp(i*pi)

ans =
-1.0000 + 0.0000i
√
Notice some of the special expressions here: pi for π, Inf for ∞, and i for −1.4 Another
special value is NaN, which stands for not a number. NaN is used to express an undeﬁned value.
For example,

>> Inf/Inf

ans =
NaN

NaNs can be tricky. For example, two NaN values are unequal by deﬁnition.
You can assign values to variables with alphanumeric names.

>> x = sqrt(3)

x =
1.7321

>> days_since_birth = floor(now) - datenum(1969,05,06)

days_since_birth =
12810

>> 3*z
??? Undefined function or variable ’z’.

Observe that variables must have values before they can be used. When an expression returns a single
result that is not assigned to a variable, this result is assigned to ans, which can then be used like
any other variable.

>> atan(x)

ans =
1.0472

>> pi/ans

ans =
3
4
√
j is also understood for −1. Both names can be reassigned, however, and it’s safer to use always 1i or 1j to refer
to the imaginary unit.

1 INTRODUCTION 7

In floating-point arithmetic, you should not expect “equal” values to have a difference of ex-
actly zero. The built-in number eps tells you the maximum error in arithmetic on your particular
machine.5 For simple operations, the relative error should be close to this number. For instance,

>> eps

ans =
2.2204e-16

>> exp(log(10)) - 10

ans =
1.7764e-15

>> ans/10

ans =
1.7764e-16

Here are a few other demonstration statements.

>> % Anything after a % sign is a comment.
>> x = rand(100,100); % ; means "don’t print out result"
>> s = ’Hello world’; % single quotes enclose a string
>> t = 1 + 2 + 3 + ...
4 + 5 + 6 % ... continues a line

t =
21

Once variables have been defined, they exist in the workspace. You can see what’s in the
workspace from its desktop window, or by typing

>> who

Your variables are:

ans s t x

1.5 Saving work
If you enter save myfile, all the variables in the workspace will be saved to a file called myfile.mat
in the current directory. You can also select the variables to be saved by typing them after the
filename argument. If you later enter load myfile, the saved variables are returned to the
workspace (overwriting any values with the same names).
5
Like other names, eps can be reassigned, but doing so has no effect on the roundoff precision.

1 INTRODUCTION 8

If you highlight commands in the Command History window, right-click, and select “Create
M-File”, you can save the typed commands to a text ﬁle. This can be very helpful for recreating
what you have done. Also see section 3.1.

1.6 Exercises
1. Evaluate the following mathematical expressions in MATLAB.

(a) tanh(e) (b) log10 (2)
−1
(c) sin −1
2 (d) 123456 mod 789 (remainder after division)

2. What is the name of the built-in function that MATLAB uses to:

(a) Compute a Bessel function of the second kind?
(b) Test the primality of an integer?
(c) Multiply two polynomials together?
(d) Plot a vector ﬁeld?
(e) Report the current date and time?

2 ARRAYS AND MATRICES 9

2 Arrays and matrices
The heart and soul of MATLAB is linear algebra. In fact, MATLAB was originally a contraction of
“matrix laboratory.” More so than any other language, MATLAB encourages and expects you to
make heavy use of arrays, vectors, and matrices.
Some jargon: An array is a collection of numbers, called elements or entries, referenced by
one or more indices running over different index sets. In MATLAB, the index sets are always
sequential integers starting with 1. The dimension of the array is the number of indices needed to
specify an element. The size of an array is a list of the sizes of the index sets.
A matrix is a two-dimensional array with special rules for addition, multiplication, and other
operations. It represents a mathematical linear transformation. The two dimensions are called the
rows and the columns. A vector is a matrix for which one dimension has only the index 1. A row
vector has only one row and a column vector has only one column.
Although an array is much more general and less mathematical than a matrix, the terms are
often used interchangeably. What’s more, in MATLAB there is really no formal distinction—
sometimes, not even between a scalar and a 1 × 1 matrix. The commands below are sorted ac-
cording to the array/matrix distinction, but MATLAB will usually let you mix them freely. The
idea (here as elsewhere) is that MATLAB keeps the language simple and natural. It’s up to you to
stay out of trouble.

2.1 Building arrays

The simplest way to construct a small array is by enclosing its elements in square brackets.

>> A = [1 2 3; 4 5 6; 7 8 9]

A =
1 2 3
4 5 6
7 8 9

>> b = [0;1;0]

b =
0
1
0

Separate columns by spaces or commas, and rows by semicolons or new lines. Information about
size and dimension is stored with the array.6

>> size(A)

6
Because of this, array sizes are not usually passed explicitly to functions as they are in FORTRAN.


ans =
3 3

>> ndims(A)

ans =
2

>> size(b)

ans =
3 1

>> ndims(b)

ans =
2

Notice that there is really no such thing as a one-dimensional array in MATLAB. Even vectors are
technically two-dimensional, with a trivial dimension.7 Table 1 lists more commands for obtaining
information about an array.

Table 1: Matrix information commands.
size size in each dimension
length size of longest dimension (esp. for vectors)
ndims number of dimensions
find indices of nonzero elements

Arrays can be built out of other arrays, as long as the sizes are compatible.

>> [A b]

ans =
1 2 3 0
4 5 6 1
7 8 9 0

>> [A;b]
??? Error using ==> vertcat
All rows in the bracketed expression must have the same
number of columns.

>> B = [ [1 2;3 4] [5;6] ]
7
What’s more, the distinction between row and column vectors is often, but not always, important.


B =
1 2 5
3 4 6

One special array is the empty matrix, which is entered as [].
Bracket constructions are suitable only for very small matrices. For larger ones, there are many
useful functions, some of which are shown in Table 2.

Table 2: Commands for building arrays and matrices.
eye identity matrix
zeros all zeros
ones all ones
diag diagonal matrix (or, extract a diagonal)
toeplitz constant on each diagonal
triu upper triangle
tril lower triangle
rand, randn random entries
linspace evenly spaced entries
cat concatenate along a given dimension
repmat duplicate vector across a dimension

An especially important array constructor is the colon operator.

>> 1:8

ans =
1 2 3 4 5 6 7 8

>> 0:2:10

ans =
0 2 4 6 8 10

>> 1:-.5:-1

ans =
1.0000 0.5000 0 -0.5000 -1.0000

The format is first:step:last. The result is always a row vector, or the empty matrix if last
< first.


2.2 Referencing elements

It is frequently necessary to access one or more of the elements of a matrix. Each dimension is
given a single index or vector of indices. The result is a block extracted from the matrix. Some
examples using the deﬁnitions above:

>> A(2,3)

ans =
6

>> b(2) % b is a vector

ans =
1

>> b([1 3]) % multiple elements

ans =
0
0

>> A(1:2,2:3) % a submatrix

ans =
2 3
5 6

>> B(1,2:end) % special keyword

ans =
2 5

>> B(:,3) % "include all" syntax

ans =
5
6

>> b(:,[1 1 1 1])

ans =
0 0 0 0
1 1 1 1
0 0 0 0

The colon is often a useful way to construct these indices. There are some special syntaxes: end
means the largest index in a dimension, and : alone is short for 1:end—i.e. everything in that


dimension. Note too from the last example that the result need not be a subset of the original
array.
Vectors, unsurprisingly, can be given a single subscript. But in fact any array can be accessed
via a single subscript. Multidimensional arrays are actually stored linearly in memory, varying
over the ﬁrst dimension, then the second, and so on. (Think of the columns of a table being stacked
on top of each other.) In this sense the array is equivalent to a vector, and a single subscript will
be interpreted in this context. (See sub2ind and ind2sub for more details.)

>> A

A =
1 2 3
4 5 6
7 8 9

>> A(2)

ans =
4

>> A(7)

ans =
3

>> A([1 2 3 4])

ans =
1 4 7 2

>> A([1;2;3;4])

ans =
1
4
7
2

>> A(:)

ans =
1
4
7
2
5
8


3
6
9

The output of this type of index is in the same shape as the index. The potentially ambiguous
A(:) is always a column vector.
Subscript referencing can be used on either side of assignments.

>> B(1,:) = A(1,:)

B =
1 2 3
3 4 6

>> C = rand(2,5)

C =
0.8125 0.4054 0.4909 0.5909 0.5943
0.2176 0.5699 0.1294 0.8985 0.3020

>> C(:,4) = [] % delete elements

C =
0.8125 0.4054 0.4909 0.5943
0.2176 0.5699 0.1294 0.3020

>> C(2,:) = 0 % expand the scalar into the submatrix

C =
0.8125 0.4054 0.4909 0.5943
0 0 0 0

>> C(3,1) = 3 % create a new row to make space

C =
0.8125 0.4054 0.4909 0.5943
0 0 0 0
3.0000 0 0 0

An array is resized automatically if you delete elements or make assignments outside the current
size. (Any new undeﬁned elements are made zero.) This can be highly convenient, but it can also
cause hard-to-ﬁnd mistakes.
A different kind of indexing is logical indexing. Logical indices usually arise from a relational
operator (see Table 3). The result of applying a relational operator is a logical array, whose ele-
ments are 0 and 1 with interpretation as “false” and “true.”8 Using a logical array as an index
8
The commands false and true can be used to create logical arrays.


returns those values where the index is 1 (in the single-index sense above).

>> B>3

ans =
0 0 0
0 1 1

>> B(ans)

ans =
4
6

>> b(b==0)

ans =
0
0

>> b([1 1 1]) % first element, three copies

ans =
0
0
0

>> b(logical([1 1 1])) % every element

ans =
0
1
0

Table 3: Relational operators.
== equal to ˜= not equal to
< less than > greater than
<= less than or equal to >= greater than or equal to


2.3 Matrix operations

The arithmetic operators +,-,*,ˆ are interpreted in a matrix sense. When appropriate, scalars are
“expanded” to match a matrix.9

>> A+A

ans =
2 4 6
8 10 12
14 16 18

>> ans-1

ans =
1 3 5
7 9 11
13 15 17

>> 3*B

ans =
3 6 9
9 12 18

>> A*b

ans =
2
5
8

>> B*A

ans =
30 36 42
61 74 87

>> A*B
??? Error using ==> *
Inner matrix dimensions must agree.

>> Aˆ2

ans =
9
One consequence is that adding a scalar to a square matrix is not the same as adding a scalar times the identity
matrix.


30 36 42
66 81 96
102 126 150

The apostrophe ’ produces the complex-conjugate transpose of a matrix. This corresponds to the
mathematical adjoint of the linear operator represented by the matrix.

>> A*B’-(B*A’)’

ans =
0 0
0 0
0 0

>> b’*b

ans =
1

>> b*b’

ans =
0 0 0
0 1 0
0 0 0

A special operator, (backslash), is used to solve linear systems of equations.

>> C = [1 3 -1; 2 4 0; 6 0 1];
>> x = Cb

x =
-0.1364
0.3182
0.8182

>> C*x - b

ans =
1.0e-16 *

0.5551
0
0

Several functions from linear algebra are listed in Table 4; there are many, many others.


Table 4: Functions from linear algebra.
solve linear system (or least squares)
rank rank
det determinant
norm norm (2-norm, by default)
expm matrix exponential
lu LU factorization (Gaussian elimination)
qr QR factorization
chol Cholesky factorization
eig eigenvalue decomposition
svd singular value decomposition

2.4 Array operations

Array operations simply act identically on each element of an array. We have already seen some
array operations, namely + and -. But the operators *, ’, ˆ, and / have particular matrix in-
terpretations. To get elementwise behavior appropriate for an array, precede the operator with a
dot.

>> A

A =
1 2 3
4 5 6
7 8 9

>> C

C =
1 3 -1
2 4 0
6 0 1

>> A.*C

ans =
1 6 -3
8 20 0
42 0 9

>> A*C

ans =
23 11 2


50 32 2
77 53 2

>> A./A

ans =
1 1 1
1 1 1
1 1 1

>> (B+i)’
ans =
-1.0000 - 1.0000i 3.0000 - 1.0000i
-2.0000 - 1.0000i 4.0000 - 1.0000i
-3.0000 - 1.0000i 6.0000 - 1.0000i

>> (B+i).’
ans =
-1.0000 + 1.0000i 3.0000 + 1.0000i
-2.0000 + 1.0000i 4.0000 + 1.0000i
-3.0000 + 1.0000i 6.0000 + 1.0000i

There is no difference between ’ and .’ for real-valued arrays.
Most elementary functions, such as sin, exp, etc., act elementwise.

>> B

B =
1 2 3
3 4 6

>> cos(pi*B)
ans =
-1 1 -1
-1 1 1

>> exp(A)
ans =
1.0e+03 *
0.0027 0.0074 0.0201
0.0546 0.1484 0.4034
1.0966 2.9810 8.1031

>> expm(A)
ans =
1.0e+06 *
1.1189 1.3748 1.6307
2.5339 3.1134 3.6929


3.9489 4.8520 5.7552

It’s easy to forget that exp(A) is an array function. Use expm(A) to get the matrix exponential
I + A + A2 /2 + A3 /6 + · · · .
Elementwise operators are often useful in functional expressions. Consider evaluating a Taylor
approximation to sin(t) at several values of t:

>> t = (0:0.25:1)*pi/2

t =
0 0.3927 0.7854 1.1781 1.5708

>> t - t.ˆ3/6 + t.ˆ5/120

ans =
0 0.3827 0.7071 0.9245 1.0045

This is easier and better than writing a loop for the calculation. (See section 6.3.) In this particular
case, polyval([1/120,0,-1/6,0,1,0],t) is arguably even better.
Another kind of array operation works in parallel along one dimension of the array, returning
a result that is one dimension smaller.

>> C

C =
1 3 -1
2 4 0
6 0 1

>> sum(C,1)

ans =
9 7 0

>> sum(C,2)

ans =
3
6
7

Other functions that behave this way are shown in Table 5.


Table 5: Reducing functions.
max sum mean any
min diff median all
sort prod std cumsum

2.5 Sparse matrices

It’s natural to think of a matrix as a complete rectangular table of numbers. However, many real-
world matrices are both extremely large and very sparse, meaning that most entries are zero.10 In
such cases it’s wasteful or downright impossible to store every entry. Instead one can keep a list
of nonzero entries and their locations. MATLAB has a sparse data type for this purpose. The
sparse and full commands convert back and forth and lay bare the storage difference.

>> A = vander(1:3);
>> sparse(A)

ans =
(1,1) 1
(2,1) 4
(3,1) 9
(1,2) 1
(2,2) 2
(3,2) 3
(1,3) 1
(2,3) 1
(3,3) 1

>> full(ans)
ans =

1 1 1
4 2 1
9 3 1

Sparsifying a standard full matrix is usually not the right way to create a sparse matrix—you
should avoid creating very large full matrices, even temporarily. One alternative is to give sparse
the raw data required by the format.

>> sparse(1:4,8:-2:2,[2 3 5 7])

10
This is often the result of a “nearest neighbor interaction” that the matrix is modeling. For instance, the PageRank
algorithm used by Google starts with an adjacency matrix in which ai j is nonzero if page j links to page i. Obviously
any page links to a tiny fraction of the more than 3 billion and counting public pages!


ans =
(4,2) 7
(3,4) 5
(2,6) 3
(1,8) 2

(This is the functional inverse of the find command.) Alternatively, you can create an empty
sparse matrix with space to hold a speciﬁed number of nonzeros, and then ﬁll in using standard
subscript assignments. Another useful sparse building command is spdiags, which builds along
the diagonals of the matrix.

>> M = ones(6,1)*[-20 Inf 10]

M =
-20 Inf 10
-20 Inf 10
-20 Inf 10
-20 Inf 10
-20 Inf 10
-20 Inf 10

>> full( spdiags( M,[-2 0 1],6,6 ) )

ans =
Inf 10 0 0 0 0
0 Inf 10 0 0 0
-20 0 Inf 10 0 0
0 -20 0 Inf 10 0
0 0 -20 0 Inf 10
0 0 0 -20 0 Inf

The nnz command tells how many nonzeros are in a given sparse matrix. Since it’s impractical
to view directly all the entries (even just the nonzeros) of a realistically sized sparse matrix, the spy
command helps by producing a plot in which the locations of nonzeros are shown. For instance,
spy(bucky) shows the pattern of bonds among the 60 carbon atoms in a buckyball.
MATLAB has a lot of ability to work intelligently with sparse matrices. The arithmetic opera-
tors +, -, *, and ˆ use sparse-aware algorithms and produce sparse results when applied to sparse
inputs. The backslash uses sparse-appropriate linear system algorithms automatically as well.
There are also functions for the iterative solution of linear equations, eigenvalues, and singular
values.


2.6 Exercises

1. (a) Check the help for diag and use it (maybe more than once) to build the 16 × 16 matrix
−2 1 0 ···
 
0 0 1
 1 −2 1 0 ··· 0 0 
 
 0
 1 −2 1 0 ··· 0 
D= . .. .. .. .. .. . 
 . . 
 . . . . . . . 
 0 ··· 0 1 −2 1 0 
 
 0 0 ··· 0 1 −2 1 
1 0 0 ··· 0 1 −2

(b) Now read about toeplitz and use it to build D.
(c) Use toeplitz and whatever else you need to build
 
1 1 1

1 2 3 4
 1 2 3 4 
4 3 2 1

1 1 1
0 1 2 3 2 1 2 3 3 2 1 2
     
0 0 1 2 1 1 1 1 2 1 2 3
3 2 2
0 0 0 1 1 1 1 1 2 3 4
4 3 2 1

Do not just enter the elements directly—your solutions should be just as easy to use if
the matrices were 100 × 100.
2. Find a MATLAB one-line expression to create the n × n matrix A satisfying

1, if i − j is prime
ai j =
0, otherwise

3. Let A be a random 8 × 8 matrix. Find the maximum values (a) in each column, (b) in each
row, and (c) overall. Also (d) find the row and column indices of all elements that are larger
than 0.25.
4. Suppose A is a matrix whose entries are all positive numbers. Write one line that will multi-
ply each column of A by a scalar so that in the resulting matrix, every column sums to 1. (To
make it more difficult, suppose that zero entries are allowed and leave a column that sums
to zero unchanged.)
5. A magic square is an n × n matrix in which each integer 1, 2, . . . , n2 appears once and for
which all the row, column, and diagonal sums are identical. MATLAB has a command
magic that returns magic squares. Check its output at a few sizes and use MATLAB to
verify the summation property. (The “antidiagonal” sum will be the trickiest.)
6. Suppose we represent a standard deck of playing cards by a vector v containing one copy of
each integer from 1 to 52. Show how to “shuffle” v by rearranging its contents in a random
order. (Note: There is one very easy answer to this problem—if you look hard enough.)


7. Examine the eigenvalues of the family of matrices

−2 1 ···
 
0 0 1
 1 −2 1 0 ···  0
 
 0 1 −2 1 ···  0
DN = − N 2 
 
.. .. ..  .
.

 . . . 
 .
 0 0 ··· 1 −2 1 
1 0 0 ··· 1 −2

where D N is N × N, for several growing values of N; for example, N = 4, 8, 16, 32. (This
is one approximate representation of the second-derivative operator for periodic functions.
The smallest eigenvalues are near integer multiples of a simple number.)

8. Use several random instances of an m × n matrix A to convince yourself that
K
2
A F = ∑ σi2 ,
i =1

where K = min{m, n}, {σ1 , . . . , σK } are the singular values of A, and · F is the Frobenius
norm (root-mean-square of the elements of A).

9. Let B=bucky and make a series of spy plots of B2 , B3 , etc. to see the phenomenon of ﬁll-in:
many operations, including multiplication, increase the density of nonzeros. (Can you see
why the (i, j) entry of Bn is the number of paths of length n between nodes i and j?) What
ﬁll-in do you see with inv(B)?

3 SCRIPTS AND FUNCTIONS 25

3 Scripts and functions

An M-file is a plain text file containing MATLAB commands and saved with the filename exten-
sion .m. There are two types, scripts and functions. MATLAB comes with a pretty good editor
that is tightly integrated into the environment; start it by typing edit with a file name. However,
you are free to use any text editor.
An M-file should be saved in the path in order to be executed. The path is just a list of direc-
tories (folders) in which MATLAB will look for files. Use editpath or menus to see and change
the path.
There is no need to compile either type of M-file. Simply type in the name of the file (without the
extension) in order to run it. Changes that are saved to disk will be included in the next call to the
function or script. (You can alter this behavior with mlock.)
One important type of statement in an M-file is a comment, which is indicated by a percent
sign %. Any text on the same line after a percent sign is ignored (unless % appears as part of
a string in quotes). Furthermore, the first contiguous block of comments in an M-file serves as
documentation for the file and will be typed out in the command window if help is used on the
file. For instance, say the following is saved as myscript.m on the path:

% This script solves the nasty homework problem assigned by
% Professor Driscoll.

x = rand(1); % He’ll never notice.

Then at the prompt one would find

>> help myscript

This script solves the nasty homework problem assigned by
Professor Driscoll.

3.1 Using scripts effectively

The contents of a script file are literally interpreted as though they were typed at the prompt. In
fact, some people prefer to use MATLAB exclusively by typing all commands into scripts and
running them. Good reasons to use scripts are

• Creating or revising a sequence of more than four or five lines.
• Reproducing or rereading your work at a later time.
• Running a CPU-intensive job in the background, allowing you to log off.

The last point here refers specifically to UNIX. For example, suppose you wrote a script called
run.m that said:


result = execute_big_routine(1);
result = another_big_routine(result);
result = an_even_bigger_routine(result);
save rundata result

At the UNIX prompt in the directory of run.m, you would enter (using csh style)

nice +19 matlab < run.m >! run.log &

which would cause your script to run in the background with low priority. The job will continue
to run until ﬁnished, even if you log off. The output that would have been typed to the screen is
redirected to run.log. You will usually need at least one save command to save your results.
Use it often in the script in case of a crash or other interruption. Also take pains to avoid taking
huge chunks of memory or disk space when running in an unattended mode. It’s advisable to
start the process and monitor it using top in the UNIX shell for a little while to make sure it does
not gobble up resources unexpectedly.

3.2 Functions

Functions are the main way to extend the capabilities of MATLAB. Compared to scripts, they are
much better at compartmentalizing tasks. Each function starts with a line such as

function [out1,out2] = myfun(in1,in2,in3)

The variables in1, etc. are input arguments, and out1 etc. are output arguments. You can have
as many as you like of each type (including zero) and call them whatever you want. The name
myfun should match the name of the disk ﬁle.
Here is a function that implements (badly, it turns out) the quadratic formula.

function [x1,x2] = quadform(a,b,c)

d = sqrt(bˆ2 - 4*a*c);
x1 = (-b + d) / (2*a);
x2 = (-b - d) / (2*a);

From MATLAB you could call

>> [r1,r2] = quadform(1,-2,1)

r1 =
1

r2 =
1


One of the most important features of a function is its local workspace. Any arguments or
other variables created while the function executes are available only within the function. Con-
versely, the variables available to the command line (the so-called base workspace) are normally
not visible to the function. If during the function execution, other functions are called, each of
those calls also sets up a private workspace. These restrictions are called scoping, and they make
it possible to write complex programs without worrying about name clashes. The values of the
input arguments are copies of the original data, so any changes you make to them will not change
anything outside the function’s scope.11 In general, the only communication between a function
and its caller is through the input and output arguments (though see section 6.5 for exceptions).
You can always see the variables defined in the current workspace by typing who or whos.
Another important aspect of function M-files is that most of the functions built into MATLAB
(except core math functions) are themselves M-files that you can read and copy. This is an excellent
way to learn good programming practice—and dirty tricks.

3.3 Conditionals: if and switch

Often a function needs to branch based on runtime conditions. MATLAB offers structures for this
similar to those in most languages.
Here is an example illustrating most of the features of if.

if isinf(x) | ˜isreal(x)
disp(’Bad input!’)
y = NaN;
elseif (x == round(x)) && (x > 0)
y = prod(1:x-1);
else
y = gamma(x);
end

The conditions for if statements may involve the relational operators of Table 3, or functions
such as isinf that return logical values. Numerical values can also be used, with nonzero mean-
ing true, but “if x˜=0” is better practice than “if x”. Note also that if the if clause is a non-
scalar array, it is taken as true only when all the elements are true/nonzero. To avoid confusion,
it’s best to use any or all to reduce array logic to scalar values.
Individual conditions can be combined using

& (logical AND) | (logical OR) ˜ (logical NOT)

Compound logic in if statements can be short-circuited. As a condition is evaluated from left
to right, it may become obvious before the end that truth or falsity is assured. At that point,
evaluation of the condition is halted. This makes it convenient to write things like
11
MATLAB does avoid copying a function argument (i.e., it “passes by reference” as in FORTRAN) if the function
never alters the data.


if (length(x) > 2) & (x(3)==1) ...

that otherwise could create errors or be awkward to write.12
The if/elseif construct is ﬁne when only a few options are present. When a large number
of options are possible, it’s customary to use switch instead. For instance:

switch units
case ’length’
disp(’meters’)
case ’volume’
disp(’liters’)
case ’time’
disp(’seconds’)
otherwise
disp(’I give up’)
end

The switch expression can be a string or a number. The ﬁrst matching case has its commands
executed.13 If otherwise is present, it gives a default option if no case matches.

3.4 Loops: for and while

Many programs require iteration, or repetitive execution of a block of statements. Again, MATLAB
is similar to other languages here.
This code illustrates the most common type of for loop:

>> f = [1 1];
>> for n = 3:10
f(n) = f(n-1) + f(n-2);
end

You can have as many statements as you like in the body of the loop. The value of the index n
will change from 3 to 10, with an execution of the body after each assignment. But remember that
3:10 is really just a row vector. In fact, you can use any row vector in a for loop, not just one
created by a colon. For example,

>> x = 1:100; s = 0;
>> for j = find(isprime(x))
s = s + x(j);
end
12
If you want short-circuit behavior for logical operations outside if and while statements, you must use the special
operators || and &&.
13
Execution does not “fall through” as in C.


This finds the sum of all primes less than 100. (For a better version, though, see page 51.)
A warning: If you are using complex numbers, you might want to avoid using i as the loop
√
index. Once assigned a value by the loop, i will no longer equal −1. However, you can always
use 1i for the imaginary unit.
It is sometimes necessary to repeat statements based on a condition rather than a fixed number
of times. This is done with while.

while x > 1
x = x/2;
end

The condition is evaluated before the body is executed, so it is possible to get zero iterations. It’s
often a good idea to limit the number of repetitions, to avoid infinite loops (as could happen above
if x is infinite). This can be done using break.

n = 0;
while x > 1
x = x/2;
n = n+1;
if n > 50, break, end
end

A break immediately jumps execution to the first statement after the loop.

3.5 Debugging and profiling

To debug a program that doesn’t work,14 you can set breakpoints in one or more functions. (See
the Debug menu in the Editor.) When MATLAB reaches a breakpoint, it halts and lets you in-
spect and modify all the variables currently in scope—in fact, you can do anything at all from the
command line. You can then continue execution normally or step by step. It’s also possible to set
non-specific breakpoints for error and warning conditions. See the help on debug (or explore the
menus) for all the details.
Sometimes a program spends most of its running time on just a few lines of code. These lines
are then obvious candidates for optimization. You can find such lines by profiling, which keeps
track of time spent on every line of every function. Profiling is also a great way to determine
function dependencies (who calls whom). Get started by entering profile viewer, or find the
Profiler under the Desktop menu.
14
The term “debugging” is perhaps too limited, since the debugging tools are also very helpful in getting to under-
stand working code that someone else has written.


3.6 Exercises

1. Write a function quadform2 that implements the quadratic formula differently from quadform
above (page 26). Once d is computed, use it to find

−b − sign(b)d
x1 = ,
2a
which is the root of largest magnitude, and then use the identity x1 x2 = c/ a to find x2 .
Use both quadform and quadform2 to find the roots of x2 − (107 + 10−7 ) x + 1. Do you see
why quadform2 is better?

2. The degree-n Chebyshev polynomial is defined by

Tn ( x) = cos n cos−1 ( x) , −1 ≤ x ≤ 1.

We have T0 ( x) = 1, T1 ( x) = x, and a recursion relation:

Tn+1 ( x) = 2xTn ( x) − Tn−1 ( x), n ≥ 1.

Write a function chebeval(x,N) that evaluates all the Chebyshev polynomials of degree
less than or equal to N at all of the points in column vector x. The result should be a matrix
of size length(x) by N+1.

3. One way to compute the exponential function e x is to use its Taylor series expansion around
x = 0. Unfortunately, many terms are required if | x| is large. But a special property of the
exponential is that e2x = (e x )2 . This leads to a scaling and squaring method: Divide x by
2 repeatedly until | x| < 1/2, use a Taylor series (16 terms should be more than enough),
and square the result repeatedly. Write a function expss(x) that does this. (The function
polyval can help with evaluating the Taylor expansion.) Test your function on x values
−30, −3, 3, 30.
4. Let x and y be column vectors describing the vertices of a polygon (given in order). Write
functions polyperim(x,y) and polyarea(x,y) that compute the perimeter and area of
the polygon. For the area, use a formula based on Green’s theorem:

n
1
A=
2 ∑ xk yk+1 − xk+1 yk .
k=1

Here n is the number of vertices and it’s understood that xn+1 = x1 and yn+1 = y1 . Test
your function on a square and an equilateral triangle.

5. If a data source produces symbol k with probability pk , the first-order entropy of the source is
defined as
H1 = − ∑ pk log2 pk .
k


Essentially H1 is the number of bits needed per symbol to encode a long message; i.e., it mea-
sures the amount of information content (and therefore the potential success of compression
strategies). The value H1 = 0 corresponds to one symbol only—no information—while for
M symbols of equal probability, H1 = log2 M.
Write a function [H,M] = entropy(v) that computes entropy for a vector v. The prob-
abilities should be computed empirically, by counting occurrences of each unique symbol
and dividing by the length of v. (The built-in functions find and unique may be help-
ful.) Try your function on some carefully selected examples with different levels of informa-
tion content. Once source of data is to use load clown; v = X(:);. You can also ﬁnd
data sources from help gallery and (if you have the Image Processing Toolbox) help
imdemos. You might also want to stick with integer data by using round.

4 MORE ON FUNCTIONS 32

4 More on functions

In the last section we covered the bare minimum on functions. Here we elaborate on more con-
cepts and tools that at some point you will probably find essential.

4.1 Subfunctions and nested functions

A single M-file may hold more than one function definition. The function header line at the top of
a file defines the primary function of the file. Two other types of functions can be in the same file:
subfunctions and nested functions.
A subfunction is mostly a convenient way to avoid unnecessary files. The subfunction begins
with a new header line after the end of the primary function. Every subfunction in the file is
available to be called by the primary function and the other subfunctions, but they have private
workspaces and otherwise behave like functions in separate files. The difference is that only the
primary function, not the subfunctions, can be called from sources outside the file.15 As a silly
example, consider

d = discrim(a,b,c);
x1 = (-b + d) / (2*a);
x2 = (-b - d) / (2*a);
end % quadform()

function D = discrim(a,b,c)
D = sqrt(bˆ2 - 4*a*c);
end % discrim()

(The end line is optional for single-function files, but it is a good idea when subfunctions are in-
volved and mandatory when using nested functions.) Since the subfunction discrim has its own
workspace, changes made to a inside discrim would not propagate into the rest of quadform;
nor would any other variables in the primary function be available to discrim.
Nested functions behave a little differently. They are defined within the scope of another func-
tion, and they share access to the containing function’s workspace. For example, we can recast our
quadratic formula yet again:


function discrim
d = sqrt(bˆ2 - 4*a*c);
end % discrim()

15
However, see section 4.3.


discrim;
x1 = (-b + d) / (2*a);
x2 = (-b - d) / (2*a);
end % quadform()

(The indentation, and placement of the nested function before its use, are optional. However, the
nested function definition does have to be completed before the end of its parent.) As a nested
function, discrim can read the values a, b, and c without having them passed as input, and its
assignment to d makes a change visible to quadform.
The purpose of nested functions is not made clear in this example—in fact, the nested function
version is inferior. But nested functions make certain important programming tasks a lot easier,
as explained in section 4.3.

4.2 Anonymous functions

Sometimes you may need a quick, short function definition that doesn’t seem to merit a named file
on disk, or even a named subfunction. The “old-fashioned” way to do this is by using an inline
function. Starting in MATLAB 7, a better alternative became available: the anonymous function.
A simple example of an anonymous function is

sincos = @(x) sin(x) + cos(x);

One can easily have multiple variables:

w = @(x,t,c) cos(x-c*t);

More interestingly, anonymous functions can “capture” the values of variables that are de-
fined at the time of the anonymous function’s creation. As an illustration, consider the process
of interpolation, i. e. the creation of a continuous function that passes through some given data.
The built-in function interp1 provides several methods for this purpose. Each call to interp1
requires passing in the data to be interpolated. Using anonymous functions we can hide, or least
encapsulate, this process:

x = 1:10; y = rand(1,10);
f_lin = @(t) interp1(x,y,t,’linear’);
f_spl = @(t) interp1(x,y,t,’spline’);
plot(x,y,’o’), hold on
y = rand(1,10); plot(x,y,’rs’)
g = @(t) interp1(x,y,t,’pchip’);
fplot(f_lin,[1 10],’b’)
fplot(f_spl,[1 10],’k’)
fplot(g,[1 10],’r’)


The two variants of f always interpolate the first data set—even after the variable y has been
changed—while g interpolates the second. Each of them is a function of just one independent
variable, the one that is usually of interest. Reproducing the effect of the above script by other
means is possible but considerably more awkward. A more advanced use of anonymous functions
can be found in section 4.3.
Nested functions and anonymous functions have similarities but are not quite identical. Nested
functions always share scope with their enclosing parent; their behavior can be changed by and
create changes within that scope. An anonymous function can be influenced by variables that
were in scope at its creation time, but thereafter it acts autonomously.

4.3 Function functions

Many problems in scientific computation involve operating on functions—finding roots, approx-
imating integrals (quadrature), and solving differential equations are three of the most common
examples. In many cases one can use or write a “black box” to solve this type of problem in many
contexts. Since these black box functions operate on other functions, MATLAB glibly calls them
function functions. (See the help topic funfun for a complete list.)
A function function needs at least one function as an input. One way to do this is by using an
anonymous function (section 4.2). For example, the built-in fzero finds a root of a scalar function
of one variable. We could say

>> f = @(x) sin(x);
>> fzero(f,3)

ans =
3.1416

This example is needlessly indirect, however—we can pass in the sin function directly, if we refer
to it using a special syntax:

>> fzero(@sin,3)

ans =
3.1416

The name @sin is called a function handle and is a way to refer to a function abstractly, rather
than invoking it immediately. By contrast, the syntax fzero(sin,3) would cause an error, be-
cause the sin function would be called first and with no input arguments.
Function handles can be created for any function that is in scope, including subfunctions and
nested functions. This can be extremely handy! Suppose for instance that we want to write a func-
tion that integrates an interpolant for given data (see section 4.2). The implementation is compli-
cated somewhat by the fact that the built-in interpolator interp1 accepts four arguments: vectors


of the x and y coordinates of the data to be interpolated, the value(s) t where the interpolant is to
be evaluated, and an optional choice of the method of interpolation. For our integration prob-
lem, the data points and the method should remain ﬁxed while we integrate with respect to the
variable t. Here is an implementation that embodies that idea:

function I = interpquad(x,y)

method = ’pchip’;
I = quad(@interpolant,x(1),x(end));

function f = interpolant(t)
f = interp1(x,y,t,method);
end % interpolant()

end % interpquad()

The quad function has to repeatedly call the integrand in order to do its job. We supply the inte-
grand as a function handle to the nested function interpolant. Each call of interpolantgets
a new value of t but also has access to the values x, y and method of its parent, which remain
ﬁxed throughout the course of the integration.16
Sooner or later you will probably have to write function functions of your own. This requires
that you be able to call a function that was passed as an argument. For example, suppose for
the sake of argument that instead of quad for integration, we want to use a much cruder type of
Riemann sum as an approximation.

function I = quadrs(integrand,a,b,n)

h = (b-a)/n; % subinterval width
x = a + (0:n)*h; % endpoints of subintervals
f = integrand(x(1:n));
I = h*sum(f);

Here integrand is an anonymous function or a function handle passed in as an argument.17

4.4 Errors and warnings

MATLAB functions may encounter statements that are impossible to execute (for example, multi-
plication of incompatible matrices). In that case an error is thrown: execution of the function halts,
a message is displayed, and the output arguments of the function are ignored. You can throw er-
rors in your own functions with the error statement, called with a string that is displayed as
the message. Similar to an error is a warning, which displays a message but allows execution to
continue. You can create these using warning.
16
For this simple case, an anonymous function could have been used instead of the nested function.
17
Past versions of MATLAB required feval to call functions in this context, but this syntax is no longer necessary.


Sometimes you would like the ability to recover from an error in a subroutine and continue
with a contingency plan. This can be done using the try–catch construct. For example, the
following will continue asking for a statement until you give it one that executes successfully.

done = false;
while ˜done
state = input(’Enter a valid statement: ’,’s’);
try
eval(state);
done = true;
catch
disp(’That was not a valid statement! Look:’)
disp(lasterr)
end
end

Within the catch block you can find the most recent error message using lasterr.

4.5 Exercises

1. Write a function newton(fdf,x0,tol) that implements Newton’s iteration for rootfind-
ing on a scalar function:
f ( xn )
xn+1 = xn −
f ( xn )
The first input is a handle to a function computing f and f , and the second input is an
initial root estimate. Continue the iteration until either | f ( xn+1 )| or | xn+1 − xn | is less than
tol. You might want a “safety valve” as well to avoid an infinite loop.
2. Modify newton from the previous exercise so that it works on a system of equations F(x).
The function fdf now returns the vector F and the Jacobian matrix J, and the update is
written mathematically as
xn+1 = xn − J −1 F ( xn ) ,
although in numerical practice one does not compute the inverse of the Jacobian but solves
a linear system of equations in which F(xn ) is the right-hand side.
3. Write a function I=trap(f,a,b,n) that implements the trapezoidal quadrature rule:
b h
f ( x) dx ≈ f ( x0 ) + 2 f ( x1 ) + 2 f ( x2 ) + · · · + 2 f ( xn−1 ) + f ( xn ) ,
a 2
where h = (b − a)/n and xi = a + ih. Test your function on sin( x) + cos( x) for 0 ≤ x ≤ π /3.
For a greater challenge, write a function simp for Simpson’s rule,
b h
f ( x) dx ≈ f ( x0 ) + 4 f ( x1 ) + 2 f ( x2 ) + 4 f ( x3 ) + 2 f ( x4 ) + · · · + 4 f ( xn−1 ) + f ( xn ) .
a 3
(This formula requires n to be even. You may choose to check the input for this.)


4. The implementation of interpquad on page 35 is not optimal in terms of execution speed.
The reason is that each call to interp1 has to do some preliminary calculations based on
x, y, and the method in order to compute interpolants. These preliminary calculations are
independent of t, so computing them anew each time is very wasteful.
Read the help on interp1 on the piecewise polynomial option for output, and use it to
modify interpquad to remove this problem.

5. Many simple financial instruments that have regular equal payments, such as car loans or
investment annuities, can be modeled by the equation

(1 + r)t − 1
F=P ,
r

where P is the regular payment, r is a fixed interest rate (say, r = 0.05 for 5% interest), t is the
number of payment intervals elapsed, and F (t) is the accumulated value of the instrument
at time t. This equation is not easily solved for r.
Write a script that finds r when P = 200, t = 30, and F takes the values 10000, 15000, . . . ,
40000. The simplest approach is to write a loop that creates an anonymous function and
calls fzero.

6. Lambert’s W function is defined as the inverse of the function f ( x) = xe x . It has no simple
analytic expression. Write a function W = lambert(x) that evaluates Lambert’s W at any
value of x > 0. For best style, lambert should have a subfunction that is passed in the call
to fzero. (Hint: You are to solve the expression x = ye y for y given x. Use fzero.)

5 GRAPHICS 38

5 Graphics

Graphical display is one of MATLAB’s greatest strengths—and most complicated subjects. The
basics are quite simple, but you also get complete control over practically every aspect of each
graph, and with that power comes complexity.
Graphics are represented by objects that are rigidly classified by type. The available types can
be found in the online documentation. The most important object types are the figure, which is
a window onscreen, the axes, into which data-related objects are drawn, and graphics primitives
such as line, surface, text, and others. These types obey a hierarchy. For instance, a figure
can be parent to one or more axes, each of which has graphical primitives as children, and so on.
Each container type provides a local coordinate system for its children to refer to. Thus, objects
drawn within an axes remain correctly rendered if the axes are moved inside its parent figure.
The most important aspect of all this is that graphically rendered objects have named prop-
erties that control their display. You can access or change these properties by point-and-click or
through the commands, as described briefly in section 5.4. Hence, you can control the display of
a graphic not only at creation time but at any later time too. The simplest and most common take
on this is to add annotations such as title, legend, tick marks, axis labels, and the like to an axes
containing data objects, but the idea can be carried much further.

5.1 2D plots

The most fundamental plotting command is plot. Normally it uses line segments to connect
points given by two vectors of x and y coordinates. Here is a simple example.

>> t = pi*(0:0.02:2);
>> plot(t,sin(t))

A line object is drawn in the current axes of the current figure, both of which are created if not yet
present. The line may appear to be a smooth, continuous curve. However, it’s really just a game
of connect-the-dots, as you can see clearly by entering

>> plot(t,sin(t),’o-’)

Now a circle is drawn at each of the given data points. Just as t and sin(t) are really vectors,
not functions, curves in MATLAB are really joined line segments.18
If you now say

>> plot(t,cos(t),’r’)
18
A significant difference from Maple and other packages is that if the viewpoint is rescaled to zoom in, the “dots”
are not recomputed to give a smooth curve.

5 GRAPHICS 39

you will get a red curve representing cos(t). The curve you drew earlier (in fact, its parent axes
too) is erased. To add curves, rather than replacing them, first enter hold on.

>> plot(t,sin(t),’b’)
>> hold on
>> plot(t,cos(t),’r’)

You can also do multiple curves in one go, if you use column vectors:

>> hold off
>> t = (0:0.01:1)’;
>> plot(t,[t t.ˆ2 t.ˆ3])

Since it can be a drag to find points yourself that will make a nice graph, there is an alternative
plotting command for a mathematical expression or any function handle (section 4.3), e. g.,

>> ezplot( @(x) exp(3*sin(x)-cos(2*x)), [0 4] )

Other useful 2D plotting commands are given in Table 6. See a bunch more by typing help
graph2d.

Table 6: 2D plotting commands
figure Open a new figure window.
subplot Multiple axes in one figure.
semilogx, semilogy, loglog Logarithmic axis scaling.
axis, xlim, ylim Axes limits.
legend Legend for multiple curves.
print Send to printer.

You may zoom in to particular portions of a plot by clicking on the magnifying glass icon in
the figure and drawing a rectangle.

5.2 3D plots

Plots of surfaces and such for functions f ( x, y) also operate on the connect-the-dots principle, but
the details are more difficult. The first step is to create a grid of points in the xy plane. These are
the points where f is evaluated to get the “dots” in 3D.
Here is a typical example:

>> x = pi*(0:0.02:1);
>> y = 2*x;
>> [X,Y] = meshgrid(x,y);
>> surf(X,Y,sin(X.ˆ2+Y))

5 GRAPHICS 40

Once a 3D plot has been made, you can use the rotation button in the figure window to manipulate
the 3D viewpoint. There are additional menus that give you much more control of the view, too.
The surf plot begins by using meshgrid to make an x-y grid that is stored in the arrays X
and Y. To see the grid graphically, use

plot(X(:),Y(:),’k.’)

With the grid so defined, the expression sin(X.ˆ2+Y) is actually an array of values of f ( x, y) =
sin( x2 + y) on the grid. (This array could be assigned to another variable if you wish.) Finally,
surf makes a solid-looking surface in which color and apparent height describe the given values
of f . An alternative command mesh is similar to surf but makes a “wireframe” surface. The
most common 3D plotting commands are shown in Table 7. There are also ezsurf, ezcontour,
and the like when you don’t want to find the data points manually.

Table 7: 3D plotting commands
surf, mesh, waterfall Surfaces in 3D.
colorbar Show color scaling.
plot3 Curves in space.
pcolor Top view of a colored surface.
contour, contourf Contour plot.

5.3 Annotation

Graphs of data usually need labels and maybe a title. For example,

>> t = 2*pi*(0:0.01:1);
>> plot(t,sin(t))
>> xlabel(’time’)
>> ylabel(’amplitude’)
>> title(’Simple Harmonic Oscillator’)

You can also add legends, text, arrows, or text/arrow combinations to help label diffrent data. See
the Insert menu on the figure window.

5.4 Handles and properties

Every rendered object has an identifier or handle. The functions gcf, gca, and gco return the
handles to the active figure, axes, and object (usually the most recently drawn).

5 GRAPHICS 41

Properties can be accessed and changed by the functions get and set, or graphically (see the
Plot Edit Toolbar, Plot Browser, and Property Editor in the figure’s View menu). Here is just a taste
of what you can do:

>> h = plot(t,sin(t))
>> set(h,’color’,’m’,’linewidth’,2,’marker’,’s’)
>> set(gca,’pos’,[0 0 1 1],’visible’,’off’)

Here is a way to make a “dynamic” graph or simple animation:

>> clf, axis([-2 2 -2 2]), axis equal
>> h = line(NaN,NaN,’marker’,’o’,’linestyle’,’-’,’erasemode’,’none’);
>> t = 6*pi*(0:0.02:1);
>> for n = 1:length(t)
set(h,’XData’,2*cos(t(1:n)),’YData’,sin(t(1:n)))
pause(0.05)
end

From this example you see that even the data displayed by an object are settable properties (XData
and YData). Among other things, this means you can extract the precise values used to plot any
object from the object itself.
Because of the way handles are used, plots in MATLAB are usually created first in a basic form
and then modified to look exactly as you want. However, it can be useful to change the default
property values that are first used to render an object. You can do this by resetting the defaults at
any level above the target object’s type. For instance, to make sure that all future Text objects in
the current figure have font size 10, enter

>> set(gcf,’defaulttextfontsize’,10)

All figures are also considered to be children of a root object that has handle 0, so this can be
used to create global defaults. Finally, you can make it a global default for all future MATLAB
sessions by placing the set command in a file called startup.m in a certain directory (see the
documentation).

5.5 Color

The coloring of lines and text is easy to understand. Each object has a Color property that can be
assigned an RGB (red, green, blue) vector whose entries are between zero and one. In addition
many one-letter string abbreviations are understood (see help plot).
Surfaces are different. You specify color data (CData) at all the points of your surface. In
between the points the color is determined by shading. In flat shading, each face or mesh line
has constant color determined by one boundary point. In interpolated shading, the color is

5 GRAPHICS 42

determined by interpolation of the boundary values. While interpolated shading makes much
smoother and prettier pictures, it can be very slow to render, particularly on printers.19 Finally,
there is faceted shading which uses flat shading for the faces and black for the edges. You select
the shading of a surface by calling shading after the surface is created.
Furthermore, there are two models for setting color data:

Indirect Also called indexed. The colors are not assigned directly, but instead by indexes in a
lookup table called a colormap. This is how things work by default.
Direct Also called truecolor. You specify RGB values at each point of the data.

Direct color is more straightforward conceptually, but it produces bigger files and is most suitable
for photos and similar images.
Here’s how indirect mapping works. Just as a surface has XData, YData, and ZData properties,
with axes limits in each dimension, it also has a CData property and “color axis” limits. The color
axis is mapped linearly to the colormap, which is a 64 × 3 list of RGB values stored in the figure.
A point’s CData value is located relative to the color axis limits in order to look up its color in
the colormap. By changing the figure’s colormap, you can change all the surface colors instantly.
Consider these examples:

>> [X,Y,Z] = peaks; % some built-in data
>> surf(X,Y,Z), colorbar
>> caxis % current color axis limits

ans =
-6.5466 8.0752

>> caxis([-8 8]), colorbar % a symmetric scheme
>> shading interp
>> colormap pink
>> colormap gray
>> colormap(flipud(gray)) % reverse order

By default, the CData of a surface is equal to its ZData. But you can make it different and
thereby display more information. One use of this is for functions of a complex variable.

>> [T,R] = meshgrid(2*pi*(0:0.02:1),0:0.05:1);
>> [X,Y] = pol2cart(T,R);
>> Z = X + 1i*Y;
>> W = Z.ˆ2;
>> surf(X,Y,abs(W),angle(W)/pi)
>> axis equal, colorbar
>> colormap hsv % ideal for this situation
19
In fact it’s often faster on a printer to interpolate the data yourself and print it with flat shading. See interp2 to
get started on this.

5 GRAPHICS 43

5.6 Saving and exporting figures

It often happens that a figure needs to be changed long after its creation. You can save the figure-
creation commands in a script (section 3.1), but this has drawbacks. If the plot data take a long
time to generate, re-running the script will waste time. Also, any edits made through menus and
buttons will be lost.
Instead you can save a figure in a format that allows it to be recreated in future sessions. Just
enter

>> saveas(gcf,’myfig.fig’)

to save the current figure in a file myfig.fig. (This is equivalent to “Save as. . . ” on the figure’s
File menu.) Later you can enter openfig myfig to recreate it. You can also use “Generate M-
file. . . ” on the File menu to create an M-file that will apply the same look as in the current graph
to new data.
Ultimately, you will want to save some figures for inclusion in a presentation or publication.
Understand that what you see on the screen is not necessarily what you get on paper, because
MATLAB tweaks graphs depending on the output device. There are three major issues here: file
format, graphics size and position, and color.
The big difference in graphics file formats is between vector (representing the lines in an image)
and bitmap (a pixel-by-pixel snapshot) graphics. Bitmaps, including GIF, JPEG, PNG, and TIFF, are
ideal for photographs, but for most other scientific applications they are a bad idea. These formats
fix the resolution of your image forever, whereas the resolution of your screen, your printer, and
a journal’s printer are all very different. Vector formats are usually a much better choice. They
include EPS and WMF. The choice here depends somewhat on your platform and word processor.
EPS files (Encapsulated PostScript) are the best choice for documents in L TEX.20 21 For example,
A
to export the current MATLAB figure to file myfig.eps, use

>> saveas(gcf,’myfig.eps’)

or

>> print -deps myfig

For color output use -depsc in the second position. EPS also works with MS Word, if you print
output on a postscript printer. In this case it’s handy to use

>> print -deps -tiff myfig
20
However, for LTEXdocuments (such as this one) outputting directly to PDF, one must use a bitmap format, prefer-
A

ably PNG.
21
Check out the LTEXpackage psfrag for putting any text or math formula anywhere in an EPS graphic.
A

5 GRAPHICS 44

in which case Word will be able to show a crude version of the graph in the document on screen.
A common problem with publishing MATLAB graphs has to do with size. By default, MAT-
LAB figures are rendered at 8 inches by 6 inches on paper. This is great for an isolated output but
much too large for most journal papers. It’s easy in L TEXand other word processors to rescale the
A
image to a more reasonable size. However, this reduces everything, including the thickness of lines
and the size of text, and the result may be unreadable. Instead you should consider rescaling the
graph before exporting it. Here are two versions of an annotated graph. On the left, the figure was
saved at default size and then rescaled in L TEX. On the right, the figure was rescaled first.
A

In practice, a compromise between these two would probably be best.
To scale the size a figure will be after export, before saving you need to enter

>> set(gcf,’paperpos’,[0 0 3 2.25])

where the units are in inches. (Or see the File/Page Setup menu of the figure.) Unfortunately,
sometimes the axes or other elements need to be repositioned. To make the display match the
paper output, you should enter

>> set(gcf,’unit’,’inch’,’pos’,[0 0 3 2.25])

Most journals do not normally accept color figures. Colored lines are automatically converted
to black when saved in a non-color format, so you should distinguish them by other features, such
as symbol or line style. The colors of surfaces are converted to grayscale, which presents a prob-
lem. By default, the colors on a surface range from deep blue to deep red, which in grayscale
look similar. You should consider using colormap(gray) or colormap(flipud(gray)),
whichever gives less total black, before exporting the figure. Finally, the edges of wireframe sur-
faces created by mesh are also converted to gray, often with poor results. Make them all the lines
black by entering

5 GRAPHICS 45

>> set(findobj(gcf,’type’,’surface’),’edgecolor’,’k’)

or by pointing and clicking.
The “File/Export setup. . . ” option on any figure window provides a mechanism for impos-
ing entire sets of changes in figure appearance, including size and color conversion, for different
contexts (print, presentation, etc.).
Finally, it’s worth remembering that figures can be saved in fig format and recreated later.
It’s a good idea to save every figure both as fig and EPS in the same directory.

5.7 Exercises

1. Recall the identity
n
1
e = lim rn , rn = 1+ .
n→∞ n
Make a standard and a log-log plot of e − rn for n = 5, 10, 15, . . . , 500. What does the log-log
plot say about the asymptotic behavior of e − rn ?

2. Here are two different ways of plotting a sawtooth wave. Explain concisely why they behave
differently.

x = [ 0:7; 1:8 ]; y = [ zeros(1,8); ones(1,8) ];
subplot(121), plot(x,y,’b’), axis equal
subplot(122), plot(x(:),y(:),’b’), axis equal

(The first version is more mathematically proper, but the second is more likely to appear in
a journal article.)

3. Play the “chaos game.” Let P1 , P2 , and P3 be the vertices of an equilateral triangle. Start
with a point anywhere inside the triangle. At random, pick one of the three vertices and
move halfway toward it. Repeat indefinitely. If you plot all the points obtained, a very clear
pattern will emerge. (Hint: This is particularly easy to do if you use complex numbers. If z
is complex, then plot(z) is equivalent to plot(real(z),imag(z)).)

4. Make surface plots of the following functions over the given ranges.
2 − y2
(a) ( x2 + 3y2 )e−x , −3 ≤ x ≤ 3, −3 ≤ y ≤ 3
(b) −3y/( x2 + y2 + 1), | x| ≤ 2, | y| ≤ 4
(c) | x| + | y|, | x| ≤ 1, | y| ≤ 1

5. Make contour plots of the functions in the previous exercise.

5 GRAPHICS 46

6. Make a contour plot of
2 +2y2 ) 2 +2y2 )
f ( x, y) = e−(4x cos(8x) + e−3((2x+1/2)

for −1.5 < x < 1.5, −2.5 < y < 2.5, showing only the contour at the level f ( x, y) = 0.001.
You should see a friendly message.

7. Parametric surfaces are easily done in MATLAB. Plot the surface represented by

x = u(3 + cos(v)) cos(2u), y = u(3 + cos(v)) sin(2u), z = u sin(v) − 3u

for 0 ≤ u ≤ 2π, 0 ≤ v ≤ 2π. (Define U and V as a grid over the specified limits, use them to
define X, Y, and Z, and then use surf(X,Y,Z).)

6 SPEED, STYLE, AND TRICKERY 47

6 Speed, style, and trickery

MATLAB is oriented towards minimizing development and interaction time, not computational
time. In some cases even the best MATLAB code may not keep up with good C code, but the gap
is not always wide. In fact, on core linear algebra routines such as matrix multiplication and linear
system solution, there is very little practical difference in performance. By writing good MATLAB
programs—starting with profiling your code (section 3.5) to find out where the bottlenecks are—
you can often nearly recover the speed of compiled code.22
Since you normally spend quite a bit of time with your code, it pays to write it well. MATLAB’s
language has features that can make certain operations, most commonly those involving loops
in C or FORTRAN, quite compact. There are also some conventions and quirks that are worth
knowing about.

6.1 Functions or scripts?

Scripts are always interpreted and executed one line at a time. No matter how many times you
execute the same script, MATLAB must spend time parsing your syntax. By contrast, functions
are effectively compiled into memory when called for the first time (or modified). Subsequent
invocations skip the interpretation step.
Most of your programming should be done in functions, which also require you to carefully
state your input and output assumptions. Use scripts for drivers or attention-free execution, as
described in section 3.1. As a rule of thumb, call scripts only from the command line, and do not
call other scripts from within a script.

6.2 Memory preallocation

MATLAB hides the tedious process of allocating memory for variables. This generosity can cause
you to waste a lot of runtime, though. Consider an implementation of Euler’s method for the
vector differential equation y = Ay in which we keep the value at every time step:

A = rand(100);
y = ones(100,1);
dt = 0.001;
for n = 1:(1/dt)
y(:,n+1) = y(:,n) + dt*A*y(:,n);
end

This takes about 7.3 seconds on a certain PC. Almost all of this time, though, is spent on a non-
computational task.
22
In a pinch, you can write a time-consuming function alone in C and link the compiled code into MATLAB. See the
online help under “External interfaces.”


When MATLAB encounters the statement y = ones(100,1) representing the initial condi-
tion, it asks the operating system for a block of memory to hold 100 numbers. On the ﬁrst execu-
tion of the loop, it becomes clear that we actually need space to hold 200 numbers, so a new block
of this size is requested. On the next iteration, this also becomes obsolete, and more memory is
allocated. The little program above requires 1001 memory allocations of increasing size, and this
task occupies most of the execution time. Also, the obsoleted space affects your other software,
because (at least in UNIX-like environments) MATLAB can’t give memory back to the operating
system until it exits.
Changing the second line to y = ones(100,1001); changes none of the mathematics but
does all the required memory allocation at once. This is called preallocation. With preallocation
the program takes about 0.4 seconds on the same computer as before.

6.3 Vectorization

Vectorization refers to the removal of loops (for and while). As an example, suppose x is a column
vector and you want to compute a matrix D such that di j = xi − x j . The standard implementation
would involve two nested loops:

n = length(x);
D = zeros(n); % preallocation
for j = 1:n
for i = 1:n
D(i,j) = x(i) - x(j);
end
end

The loops could be written in either order here. But the innermost loop is easily replaced by a
vector operation.

n = length(x);
D = zeros(n); % preallocation
for j = 1:n
D(:,j) = x - x(j);
end

We can get rid of the remaining loop, too, by upgrading from vectors to two-dimensional arrays.
This is a bit more subtle.

n = length(x);
X = x(:,ones(n,1)); % copy columns to make n by n
D = X - X.’;

The second line here is a trick that was introduced in section 2.2 (and in MATLAB circles is called
“Tony’s trick”). Note too the use of .’ in the last line to be compatible with a complex input.


At this point you might ask, why vectorize? There are two answers: speed and style. Neither is
a simple issue. Until around 2002, careful vectorization always yielded tremendous improvements
in execution speed—often orders of magnitude. But this is changing due a technique called JIT
acceleration. Acceleration, which is applied automatically, can remove the speed penalty that
MATLAB traditionally experiences with loops. At this writing, not every loop can be optimized,
but it’s clear that code vectorization is no longer speed-critical in every case. For example, with
n = 200 the the times in milliseconds for each of the three methods above were 1.73, 0.97, and
2.80, respectively. In this case a medium level of vectorization proved fastest. (Observe also that
the ﬁrst two versions, unlike the third, are easily adjusted to account for the obvious antisymmetry
in the result.)
Style is a subjective matter, of course. To get a better idea of the trade-offs involved, consider
the classic example of Gaussian elimination. Here is a basic implementation without any vector-
ization.

n = length(A);
for k = 1:n-1
for i = k+1:n
s = A(i,k)/A(k,k);
for j = k:n
A(i,j) = A(i,j) - s*A(k,j);
end
end
end

Again we start vectorization with the innermost loop, on j. Each iteration of this loop is indepen-
dent of all the others. This parallelism is a big hint that we can use a vector operation:

n = length(A);
for k = 1:n-1
for i = k+1:n
s = A(i,k)/A(k,k);
cols = k:n;
A(i,cols) = A(i,cols) - s*A(k,cols);
end
end

This version is actually makes the linear algebra idea of a row operation much more apparent
and, in my view, is clearly preferable. However, the innermost of the remaining loops is also
vectorizable.

n = length(A);
for k = 1:n-1
rows = k+1:n;
cols = k:n;
s = A(rows,k)/A(k,k);


A(rows,cols) = A(rows,cols) - s*A(k,cols);
end

You have to flex your linear algebra muscles a bit to see that the vector outer product in the next-
to-last line is appropriate. This is an interesting insight, but does not lead to an unquestionable
improvement in style.
To compare the speed of these three versions, they were run 20 times each for six different
values of the matrix size n on a 2.5GHz Linux machine in MATLAB 7.0.1. The results, per factor-
ization in milliseconds:

n = 50 100 150 200 250 300
Three loops 1.5 14.5 41 108.5 231 419.5
Two loops 17.5 99.5 258.5 521 918 1465
One loop 3 12 76.5 197.5 439.5 672

The version that works naturally at the vector level is now by far the slowest. Besides pointing out
the pitfalls of vectorization, this experiment illustrates the limitations of comparing algorithms by
counting their floating-point operations. In the two-loop version the growth in time is less than
the asymptotically expected O(n3 ), indicating that non-arithmetic operations are playing a major
role.
An uncontroversial application of vectorization is the use of array operators in the evalua-
tion of a mathematical expression such as t sin(t2 ) at many values of t. It’s hard to deny that
t.*sin(t.ˆ2) is plainer and shorter than an equivalent loop. Similarly, once you understand
the cumsum function for cumulative summation, computing partial sums of a geometric series by

s = cumsum( (1/3).ˆ(0:100) );

is a little shorter and more transparent than the equivalent loop, which also requires a prealloca-
tion (section 6.2):

s = ones(1,101);
for j = 1:100,
s(j+1) = s(j) + (1/3)ˆj;
end

Functions like sum and diff listed in Table 5 can make code more readable in some cases.
The bottom line: Loops should not be written casually. Writing code that works at a vector
level is easy and natural in most cases. But if profiling indicates that a lot of time is being spent
in a place where the level of vectorization is selectable, experimentation may be the only way to
see what level is best. In the future, MATLAB may become even more forgiving when it comes to
loops.


6.4 Masking

An advanced type of vectorization is called masking. Let’s say that we have a vector x of values
at which we want to evaluate the piecewise-defined function

1 + cos(2π x), | x| ≤ 1
2
f ( x) =
0, | x| > 1 .
2

Here is the standard loop method:

f = zeros(size(x));
for j = 1:length(x)
if abs(x(j)) <= 0.5
f(j) = 1 + cos(2*pi*x(j));
end
end

The shorter way is to use a mask.

f = zeros(size(x));
mask = (abs(x) < 0.5);
f(mask) = 1 + cos(2*pi*x(mask));

The mask is a logical index into x (see page 14). You could refer, if needed, to the unmasked points
by using ˜mask.
Consider a new version of the sum-of-primes idea from page 28. Here’s how we could count
the number of primes less than 100 and add them up:

isprm = isprime(1:100);
sum( isprm )
sum( find(isprm) )

Here find converts a logical index into an absolute one. The only disadvantage of doing so in
general is that referring to the unmasked elements becomes more difficult.
Generally, when a graphical function encounters a NaN value, it quietly omits drawing that
point. This can be useful when combined with masking. For example, the commands

L = membrane(1,18,8,8);
surf(L)

plots a surface defined over an L-shaped region. The part outside the L is flat at zero, which works
OK since the function goes to zero at the boundary. But it looks better if we erase that part of the
surface using a mask:


[X,Y] = meshgrid(0:36);
outside = (x < 18) & (y > 18); % logical mask
L(outside) = NaN;
surf(X,Y,L)

6.5 Scoping exceptions

Once in a while the scoping rules for functions get in the way. Although you can almost always
do what you need within the rules, it’s nice to know how to bend them.
The least useful and most potentially troublesome violation of scoping comes from global
variables. Any function (or the user at the command line) can declares a variable to be global
before assigning it a value. Then any other workspace may also declare it global and see or change
its value. At one time global values were more or less necessary in MATLAB, but that is no longer
the case. They should not be used, for example, to pass extra parameters into “function functions.”
As described in section 4.3, there are better and more stable means of doing so. The primary
problem with global variables is that it becomes possible to have conflicting names, or to lose
track of what functions may modify a value. Input and output parameters make this information
much more apparent.
A more interesting type of variable is called persistent. One use of persistent variables is to
compute some preliminary data that needs to be used on subsequent calls. Although the data
could be returned to the caller and passed back in to the function, that is inconvenient when such
data are meaningless to the caller. Consider this example for the Fibonacci numbers.

function y = fib(n)

persistent f
if length(f) < 2
f = [1 1];
end
for k = length(f)+1:n
f(k) = f(k-2) + f(k-1);
end
y = f(1:n);

The first time this function is called, f will be empty.23 Immediately f will then be given the first
two values in the sequence. (The first invocation is the only one in which the conditional of line
5 will be true.) Then, f is extended as needed to get the first n values. So far this is nothing
unusual, if it is a bit indirect. But upon exit of the function, the persistent f is not destroyed.
Thus future calls will begin with an f that has as many entries as the largest n encountered in all
calls. The same effect could be achieved with a global variable, but a persistent variable is
accessible only to the function that created it. In fact, different functions can use the same name
for a persistent variable without interference.
23
persistent variables, unlike others, are given an initial value: the empty matrix.


6.6 Exercises

1. Rewrite trap or simp (page 36) so that it does not use any loops. (Hint: Set up a vector
of all x values and evaluate f for it. Then set up a vector of quadrature weights and use an
inner product.)

2. Explore the issue of whether the order of the loops affects the execution time of either of the
first two versions of the vector differencing code given on page 48.

3. Suppose x is a column vector. Compute, without using loops or conditionals, the matrix A
given by
1
( xi − x j )2
, if i = j,
ai j =
1, otherwise.
(One way to do this is by direct assignment of the diagonal elements of A. This is impos-
sible without a loop using row and column indices, but easy using the single-index linear
addressing model.)

4. Reconsider the function chebeval (page 30) that evaluates Chebyshev polynomials at mul-
tiple points. Write the function so that it performs as efficiently as you can make it. Among
other things, you have to choose whether to use the definition or the recursion.

5. Consider “shuffling” a vector of integers from 1 to 52 using a physical interpretation of a
card shuffle. Divide the cards into two equal stacks, and merge the stacks together such that
each time a pair of cards is to fall off the “bottom” of each stack, a random decision is made
as to which falls first. A loop-based implementation would be

function s = shuffle(x)
n = length(x)/2;
s = [];
for i = 1:n
if rand(1) > 0.5
s = [s x([2*i-1 2*i])];
else
s = [s x([2*i 2*i-1])];
end
end

Rewrite this function without loops. (This can be done in as little as four statements using
resizing tricks.) It can be interesting to start with a perfectly ordered deck and see how many
shuffles it takes to “randomize” it. One crude measure of randomness is the (1,2) element of
corrcoef(1:52,v), which is expected to be zero if v is random.

6. Rewrite the function entropy on page 30 without any loops using sort, diff, find, and
(perhaps) sum.


7. In the function newton (page 36), suppose that input x0 is actually a vector of initial guesses,
and you want to run Newton’s method on each. Keep track of an error vector and use
masking to rewrite newton so that it still uses only one loop.

8. Rewrite expss from page 30 so that it works on a vector input and as fast as possible. (You
should consider, for example, whether making use of persistent variables might help.)

9. Different Fibonacci sequences can be produced by changing the ﬁrst two members of the
sequence. Rewrite fib from page 52 so that it accepts these seed values and recomputes the
sequence only when necessary.

7 ADVANCED DATA STRUCTURES 55

7 Advanced data structures

Not long ago, MATLAB viewed every variable as a matrix or array. While this point of view is
ideal for simplicity, it is too limited for some tasks. A few additional data types are provided.

7.1 Strings and formatted output

As we have seen, a string in MATLAB is enclosed in single forward quotes. In fact a string is
really just a row vector of character codes. Because of this, strings can be concatenated using
matrix concatenation.

>> str = ’Hello world’;
>> str(1:5)

ans =

Hello

>> double(str)

ans =
72 101 108 108 111 32 119 111 114 108 100

>> char(ans)

ans =

Hello world

>> [’Hello’,’ ’,’world’]

ans =

Hello world

You can convert a string such as ’3.14’ into its numerical meaning (not its character codes) by
using eval or str2num on it. Conversely, you can convert a number to string representation
using num2str or the much more powerful sprintf (see below). If you want a quote character
within a string, use two quotes, as in ’It’’s Cleve’’s fault’.
Multiple strings can be stored as rows in an array using str2mat, which pads strings with
blanks so that they all have the same length. However, a better way to collect strings is to use cell
arrays (section 7.2).
There are lots of string handling functions. See the help on strfun. Here are a few:


>> upper(str)

ans =

HELLO WORLD

>> strcmp(str,’Hello world’)

ans =

1

>> findstr(’world’,str)

ans =

7

For the best control over conversion of numbers to strings, use sprintf or fprintf. These
are closely based on the C function printf, with the important vectorization enhancement that
format speciﬁers are “recycled” through all the elements of a vector or matrix (in the usual row-
ﬁrst order).
For example, here’s a script that prints out successive Taylor approximations for e1/4 .

x=0.25; n=1:6; c=1./cumprod([1 n]);
for k=1:7, T(k)=polyval(c(k:-1:1),x); end
fprintf(’n T_n(x) |T_n(x)-exp(x)|n’);
fprintf(’----------------------------------n’);
fprintf(’%15.12f %8.3en’, [T;abs(T-exp(x))] )

T_n(x) |T_n(x)-exp(x)|
----------------------------------
1.000000000000 2.840e-01
1.250000000000 3.403e-02
1.281250000000 2.775e-03
1.283854166667 1.713e-04
1.284016927083 8.490e-06
1.284025065104 3.516e-07
1.284025404188 1.250e-08

Use sprintf if you want to save the result as a string rather than have it output immediately.

7.2 Cell arrays

Collecting objects of different sizes is a common chore. For instance, suppose you want to tabu-
late the Chebyshev polynomials 1, x, 2x2 − 1, 4x3 − 3x, and so on. In MATLAB one expresses a


polynomial as a vector (highest degree first) of its coefficients. The number of coefficients needed
grows with the degree of the polynomial. Although you can put all the Chebyshev coefficients
into a triangular array, this is an inconvenient complication.
Cell arrays are used to gather dissimilar objects into one variable. They are indexed like reg-
ular numeric arrays, but their elements can be absolutely anything. A cell array is created or
referenced using curly braces {} rather than parentheses.

>> str = { ’Goodbye’, ’cruel’, ’world’ }

str =

’Goodbye’ ’cruel’ ’world’

>> str{2}

ans =

cruel

>> T = cell(1,9);
>> T(1:2) = { [1], [1 0] };
>> for n = 2:8, T{n+1} = [2*T{n} 0] - [0 0 T{n-1}]; end
>> T

T =
Columns 1 through 5

[1] [1x2 double] [1x3 double] [1x4 double] [1x5 double]

Columns 6 through 9

[1x6 double] [1x7 double] [1x8 double] [1x9 double]

>> T{4}

ans =
4 0 -3 0

Cell arrays can have any size and dimension, and their elements do not need to be of the same
size or type. Cell arrays may even be nested. Because of their generality, cell arrays are mostly just
containers; they do not support any sort of arithmetic.
One special cell syntax is quite useful. The idiom C{:} for cell array C is interpreted as a
comma-separated list of the elements of C, just as if they had been typed. For example,

>> str2mat(str{:}) % same as str2mat(’Goodbye’,’cruel’,’world’)


ans =
Goodbye
cruel
world

The special cell array varargin is used to pass optional arguments into functions. For exam-
ple, reconsider the function interpquad on page 35 for integrating an interpolant. The built-in
interp1 used there to perform the interpolation accepts an optional input parameter to select the
type of interpolant used. We can update interpquad to produce exactly the same behavior, as
follows:

function I = interpquad(x,y,varargin)

I = quad(@interpolant,x(1),x(end));

function f = interpolant(t)
f = interp1(x,y,t,varargin{:});
end % interpolant()

end % interpquad()

The nested function has access to varargin. The syntax varargin{:} puts all input argu-
ments after the ﬁrst two into each call to interp1. Thus, for example,

>> x=pi*sort(rand(100,1));
>> interpquad(x,sin(x)) % default choice
ans =
1.9987

>> interpquad(x,sin(x),’linear’)
ans =
1.9987

>> interpquad(x,sin(x),’pchip’)
ans =
1.9994

Note how an empty varargin is valid too and gives the default interp1 behavior (linear).

7.3 Structures

Structures are much like cell arrays, but they are indexed by names rather than by numbers.
Say you are keeping track of the grades of students in a class. You might start by creating a
student struct (structure) as follows:


>> student.name = ’Moe’;
>> student.SSN = 123456789;
>> student.homework = [10 10 7 9 10];
>> student.exam = [98 94];
>> student

student =

name: ’Moe’
SSN: 123456789
homework: [10 10 7 9 10]
exam: [98 94]

The name of the structure is student. Data is stored in the structure according to named fields,
which are accessed using the dot notation above. The field values can be anything.
Probably you have more students.

>> student(2).name = ’Curly’;
>> student(2).SSN = 987654321;
>> student(2).homework = [4 6 7 3 0];
>> student(2).exam = [53 66];
>> student

student =
1x2 struct array with fields:
name
SSN
homework
exam

Now you have an array of structures. This array can have any size and dimension. However, all
elements of the array must have the same fields.
Struct arrays make it easy to extract data into cells:

>> [roster{1:2}] = deal(student.name)

roster =

’Moe’ ’Curly’

In fact, deal is a very handy function for converting between cells and structs. See online help for
many examples.

8 SCIENTIFIC COMPUTING 60

8 Scientific computing

Certain computational problems appear repeatedly in applications motivated by science and en-
gineering. Naturally, MATLAB is thoroughly equipped for such problems. In this section are
reviews of some of these methods from a “recipe” point of view. Deeper mathematical under-
standing, which is vital when things do not work out according to plan, can be found in texts and
papers on numerical analysis.

8.1 Linear algebra

The most common tasks in numerical linear algebra are solving linear systems Ax = b or overde-
termined least-squares systems Ax ≈ b, and finding factorizations such as the eigenvalue or sin-
gular value decompositions. The greatest dichotomy in methods for these problems is full versus
sparse matrices. (For least-squares problems, however, sparse situations seem to be relatively rare
and less supported.) See section 2.5 for more on sparse matrices.
For square or rectangular linear systems the method of first resort is the backslash operator.
The solution to either Ax = b or Ax ≈ b is implemented in MATLAB as x=Ab. For square
matrices, in fact, the idiom A is mathematically equivalent and computationally preferred to left-
multiplication by A−1 .24 The backslash does not apply a single algorithm but rather represents
an expert system that applies the best of several methods depending on detectable properties of
A (such as sparsity, triangularity, and positive definiteness). These are spelled out in the docu-
mentation page for mldivide. One consequence is that you must set up A properly to get the
fastest solution—for instance, if A is tridiagonal, you should make it formally sparse to get an
O(n) solution time.
If you need to solve multiple systems Axi = bi , for i = 1, . . . , k, you can use the identity

b1 b2 · · · bk = Ax1 Ax2 · · · Axk = A x1 x2 · · · xk .

In MATLAB terms one lets B be the matrix whose columns are b1 , . . . , bk , and then X=AB has the
solutions x1 , . . . , xk as its columns. But if the right-hand side vectors bi are not all known at once
(e. g., they are produced by an iteration), then backslash is much less useful, because it has to
repeat costly work each time. In this case you have to do the key factorization step yourself if you
want efficiency. For instance, the so-called inverse iteration for an eigenvector of A could look like

[L,U] = lu(A);
v = randn(length(A),1); v = v/v(1);
for k = 1:50
v = U (Lv);
v = v/v(1);
end
24
To right-multiply by A−1 , use a forward slash, as in x=b/A.


Note that U (Lv) is equivalent to U −1 ( L−1 v) = ( LU )−1 v = A−1 v. This version is fine for
a generic A but less than ideal if A has one of the special properties detected by the backslash. At
this point one really needs to know some numerical linear algebra to make further improvements.
Eigenvalue and singular value decompositions can be found using eig and svd, respectively.
Both functions have two output formats:

lambda = eig(A); sigma = svd(A);
[V,D] = eig(A); [U,S,V] = svd(A);

The top line returns just a vector of eigenvalues or singular values; the bottom line returns the full
decomposition. For algorithmic reasons, the value-only forms can be substantially faster than the
full versions, so use the full versions only when you need them.

8.2 Iterative linear algebra

The methods of section 8.1 are appropriate for matrices of size in the several thousands for linear
systems, or perhaps about a thousand for the decompositions. For truly large problems these
direct methods, which give answers at the end of a finite number of operations25 , are less useful
than iterative methods, which produce sequences of improving estimates. Iterative methods also
do not factor or modify the given matrix, which is ideal for sparse problems.
For linear systems there are many choices of iterative methods (see the help for sparfun). All
of these tend to be much better than classical methods such as Gauss–Seidel or SOR. If the sys-
tem matrix is positive definite, the usual choice is the conjugate gradient method, implemented
by pcg. For more general matrices, bicgstab and gmres tend to be quite popular, but there is
no “best” method. In practice the choice of method is often less critical than the choice of a pre-
conditioner, which is a matrix M that is somehow “close to” A yet allows very fast solutions of
Mx = b.
For the decompositions the choices are simpler: eigs for eigenvalues and sigs for singular
values. Both are much preferred to the power iteration and related classical methods. Usually one
does not seek the full decomposition (which might be impractical even to store), but a few of the
largest or smallest values and perhaps their associated vectors.
All of these iterative methods have a surprising additional flexibility: they don’t require matri-
ces! They can instead be given a function that returns the vector Au for any given vector u. This is
more useful than it may sound, for example in applications where one can use multipole or other
“fast” approximate methods for the matrix-times-vector operation.
25
Technically, all eigenvalue and singular value methods must be iterative. However, the classical methods used by
eig and svd converge very quickly and in practice spend most of their time in a finite preprocessing step.


8.3 Rootfinding

One of the oldest and most common numerical problems is trying to find a root or zero of a scalar
function f ( x) of one variable. (This is equivalent to solving f ( x) = c for any constant c, since
in that case we can find a root of the function f ( x) − c.) Usually one learns Newton’s method
for this problem, but it turns out not to be the best general-purpose method because it needs an
explicit derivative calculation and has erratic convergence. The MATLAB function fzero is much
better than a handwritten algorithm for most problems. It can use a starting point or (preferably)
a starting interval containing a root, and requires a function only to compute f ( x), not f ( x).
Roots of polynomials are handled differently using roots. You should be aware that polyno-
mial roots can be very sensitive to perturbations in the coefficients, so interpret the results with
some caution.
For solving multidimensional systems of nonlinear equations F ( x) = 0, MATLAB offers fsolve
in the Optimization Toolbox. You only need to supply a function for computing F and a starting
point, but you might greatly improve the convergence if you also supply a function computing
the Jacobian matrix of F, or at least giving the sparsity pattern of the Jacobian. There are several
algorithms to choose from, so read the documentation if you have trouble on a problem. All of the
methods expect F to be reasonably smooth, and they benefit from good starting guesses.

8.4 Optimization

To find the minimum of a scalar function f ( x) of one variable, use fminbnd. It requires a function
computing f and an interval on which to optimize. (To maximize f ( x), you minimize − f ( x).)
For minimization with multiple variables there are two main choices: fminunc (in the Opti-
mization Toolbox) and fminsearch. For smooth functions fminunc is usually faster, but fminsearch
does not use any derivative information and may be better for nonsmooth problems. In the special
case of nonlinear least squares,

f ( x) = f 1 ( x)2 + f 2 ( x)2 + · · · f m ( x)2 ,

one should use lsqnonlin instead of either of these. (In fact, this is one of the methods fsolve
uses for solving systems.)
Optimization in engineering often comes with side constraints on the variables. None of the
multidimensional methods named here respects such constraints. See the Optimization Toolbox
for methods for several variants of these problems.

8.5 Fitting data

A common task is to make sense of data from noisy observations of a presumably simple rela-
tionship. Here “noise” can refer to experimental error or uncertainty, or neglected higher-order


effects. In data fitting one posits a form for the underlying relationship involving some unknown
parameters, then uses the data to find the best parameters. This leads to an optimization.
When the form of the relationship is a linear combination of basis functions and the parameters
are the coefficients, the resulting optimization is a linear least-squares problem and can be solved
using the backslash. For example,

t = (1:100)’; y = 4 + 0.1*cos(t) - 0.2*sin(t);
y = y + 0.1*randn(100,1);
A = [ ones(100,1) cos(t) sin(t) ];
Ay

ans =
4.0048
0.0938
-0.1951

When the functional form is a polynomial, the entire process is automated by polyfit. Also,
a graphical data fitting tool for many linear fits is available from the Tools menu of every figure
window.
If the coefficients appear nonlinearly, a nonlinear optimization results. For instance, MATLAB
ships with U. S. census data for the national population every ten years since 1790. To fit these
data with an exponential curve of the form c1 + c2 ec3 t , we could proceed as follows:

load census
t = (cdate-1790)/10;
residual = @(param) sum( (param(1) + param(2)*exp(param(3)*t) - pop).ˆ2 );
fminsearch(residual,[1 1 1])

ans =
-39.1124 35.8209 0.1053

You can check graphically that this is a pretty good fit. (Note that if we set c1 = 0 in this example,
then by taking logs we get a fit to a linear function.)

8.6 Quadrature

Numerical approximation of definite integrals is called quadrature in one dimension and cubature
if the integral is over multiple dimensions. The basic use of quad for quadrature was shown
in section 4.3. An alternative quadl may be faster for high accuracy with smooth integrands.
MATLAB offers dblquad for cubature over rectangular domains. More general domains can be
implemented literally using iterated integration.
In truth, MATLAB’s integration offerings fall a bit short in accuracy, reliability, generality, and
speed. If you need to do anything beyond some basic one-dimensional integrands, you might do
well to look up CUBPACK or QUADPACK on the web.


8.7 Initial-value problems

MATLAB has an excellent library of solvers for the initial value problem (IVP) x (t) = f (t, x(t)),
x( a) = x0 , where x and f are understood to be vectors. Any higher-order differential equation
must ﬁrst be written in ﬁrst-order form. The primary division between IVP solvers is stiff versus
nonstiff. Loosely speaking, stiff problems have simultaneous phenomena over very different time
scales, and they require methods that are much more expensive for each time step taken, especially
in the face of high-dimensional nonlinearity. The logical extreme of stiffness is the differential–
algebraic equation, which has side constraints on the variables in addition to the dynamics.
Basic ODE solution takes the form

output = odexxx( odefunc, time, init, options )

Here the odefunc is a function accepting scalar t and vector x, returning the vector f (t, x); time is a
vector of two or more times spanning the time interval of solution; init is the initial condition x0 ;
options controls aspects of the solution behavior, including error tolerances, and is created using
odeset; and the solver’s name is one of ode45, ode113, ode15s, or the other solvers listed in
the documentation.
There are two forms for the output. If two names [t,y] are given, then t is a vector of times
and each row of y is the solution vector at one of those times. If the time input is a vector of length
greater than two, then t is identical to it; otherwise, t is the vector of time steps that were chosen
automatically.26 The other form of output is to give a single variable soln, which can be used in
deval to evaluate the solution at any desired time values.
For example, consider the well-known predator-prey model

dx1
= x1 − αx1 x2
dt
dx2
= βx1 x2 − x2 .
dt
This can be solved by

f = @(t,x) [ x(1)-0.2*x(1)*x(2); 0.5*x(1)*x(2) - x(2) ];
[t,x] = ode45(f, 0:0.05:20, [8;1]);
subplot(2,1,1), plot(t,x)
subplot(2,1,2), plot(x(:,1),x(:,2))

The IVP solvers offer lots of customization and additional features. Investigate online help if
you are doing any more than the basics.
26
The only difference between the two cases is in the output—the time steps chosen by the algorithm in the solver
are unaffected.


8.8 Boundary-value problems

A boundary-value problem also starts with a first-order ordinary differential equation, y ( x) =
f ( x, y), for a ≤ x ≤ b, but unlike an initial-value problem, it has side conditions involving two
points: g( y( a), y(b)) = 0. Existence and uniqueness theory is much less general than for IVPs,
and the numerical solution process is less straightforward. Here we just give a simple case study,
the Allen–Cahn equation

y + y(1 − y2 ) = 0, 0 ≤ x ≤ 1, y(0) = −1, y(1) = 1.

Both y ≡ −1 and y ≡ 1 are constant solutions, and this equation models a “phase change”
between them. If is small, this change is abrupt.
Three elements are needed to attempt a numerical solution: a function for the ODE, a function
for the boundary conditions, and an initial guess to the solution. Although in general this is
probably best done in an M-file, here we use anonymous functions. The ODE must be cast in first
order form by introducing y1 = y, y2 = y :

f = @(x,y) [ y(2); (y(1)ˆ3-y(1))/0.01 ];

Here we set = 0.01. Next, we create a function that should return all zeros when the boundary
conditions are satisfied:

g = @(ya,yb) [ ya(1)+1; yb(1)-1 ];

Finally, we must create an initial guess to the solution using bvpinit. We choose a linear inter-
polation on equally spaced nodes between the boundary values.

x = linspace(0,1,10)’;
y0 = @(x) [ -1+2*x; 2 ];
ACinit = bvpinit(x,y0);

The initial guess may be crucial to obtaining a solution in difficult problems. Finally, we call bvp4c
to create the solution and use deval to evaluate it for a plot:

soln = bvp4c(f,g,ACinit);
xp=0:0.05:1; plot(xp,deval(soln,xp,1))

(The third argument to deval asks for just the first component of the solution, the original variable
y.) If you inspect the output soln, you will see that it is a structure, and that the original 10 nodes
in x have been replaced by 35 unequally spaced ones.
If we attempt the same process with = 10−3 , the solver appears to hang. The problem is that
the linear initial guess does not give a sufficient clue about the shape of the solution. In this case
one gets better results by using the solution with = 10−2 as the initial guess, a crude form of a
process known as continuation. The syntax of bvp4c makes this easy.


soln = bvp4c(f,g,soln)

ans =
x: [1x63 double]
y: [2x63 double]
yp: [2x63 double]
solver: ’bvp4c’

Now we see that 63 nodes are being used, and a plot confirms that the transition region near
x = 1/2 is much steeper.

8.9 Initial-boundary value problems

An initial-boundary value problem is a time-dependent partial differential equation with side con-
ditions at an initial time and on geometric boundaries. While MATLAB does have an automatic
facility for certain PDEs with one space and one time dimension (pdepe), here we study a semi-
automated process known as semidiscretization or the method of lines. In the method of lines
ones discretizes space first and then time independently. At a practical level this means the orig-
inal PDE is replaced by a (typically large) system of ordinary differential equations. In principle,
and often in practice too, one can solve the ODE system using the IVP solvers from section 8.7.
For example, consider the problem

ut = u xx + u2 , 0 ≤ x ≤ 1, u(t, 0) = u(t, 1) = 0.

This problem mixes blowup in finite time (via ut = u2 ) with diffusion. If the initial condition is
large enough, blowup wins. If we use a simple second-order finite difference for the u xx term, this
problem can be attacked using the remarkably simple script

N = 100; h = 1/N;
x = h*(1:N-1)’;
D = sparse(toeplitz([-2 1 zeros(1,N-3)]/hˆ2));
f = @(t,u) D*u + u.ˆ2;
u0 = 30*sin(pi*x);
[t,u] = ode113(f,[0 .1],u0);

Upon execution one receives the message

Warning: Failure at t=4.561109e-02. Unable to meet integration
tolerances without reducing the step size below the smallest value
allowed (2.220446e-16) at time t.

This is an indication of the finite-time blowup. By increasing N or using more accurate finite dif-
ferences, one could quickly determine that at least two digits of the blowup time have apparently
been found accurately.

Matlab tutorial by manish myst, ssgbcoet

More Related Content

Similar to Matlab tutorial by manish myst, ssgbcoet (20)

More from Manish Myst (12)

Recently uploaded (20)

Matlab tutorial by manish myst, ssgbcoet