Introduction to MATLAB

Stefan Güttel

October 15, 2020

Contents
1 Introduction 2

2 Matrices and Arrays 2

3 Expressions 3

4 Basic Linear Algebra commands 4

5 Graphics 5

6 Programming Scripts 6

7 Functions 7

8 Numerical Integration and Differential Equations 7

9 Useful tools and tricks 9

10 Assignment 11

1
1 Introduction
In this course of 4 hours we will give an introduction to MATLAB. This course material
is based on the MATLAB documentation which can be accessed by typing doc in the
MATLAB command window. Another useful command is help. In combination with a
function name it returns a help text for that function. Example: help sum
Apart from the excellent MATLAB documentation, there are many MATLAB tutori-
als and books available. Here are a few:

• The MathWorks website https://uk.mathworks.com/ contains a well-managed

support section with documentation, examples, and an active user community.
Most importantly, MatlabCentral is a way to share useful MATLAB functions
between users:
https://uk.mathworks.com/matlabcentral/

• Cleve Moler, one of the MATLAB authors, has written two interesting books on
numerical computing with MATLAB. These books are freely available at
https://uk.mathworks.com/moler.html

• The MATLAB Guide by D. J. Higham and N. J. Higham (SIAM 2017) is another

useful book to learn about MATLAB.

Familiarize yourself with the command window, workspace, command history, and
navigation in the MATLAB documentation. MATLAB code can be stored in so-called
m-files and it is convenient to use these files also during the development phase of a
project. Everything that follows the character % is treated as a comment. Use the cursor
keys to navigate up and down the history in the command window. Use the Tab key for
auto-completion.
You can leave MATLAB by typing exit or closing the window. Long-running MAT-
LAB codes can (often) be interrupted by pressing Ctrl+C. Example: while 1, end

Useful commands: doc, help, clc, clear, clear all

2 Matrices and Arrays

Learn how to create a matrix, the fundamental data structure in MATLAB, as an explicit
list of elements, via index assignment, and using MATLAB commands. Learn how to
suppress outputs to the command line. Understand the colon : operator. Understand
how to concatenate matrices and how to delete rows/columns.

Commands: [], [1,2,3], [1;2;3], A(1,2), +, -, *, /, .*, ./, sum, prod,

transpose, diag, :, linspace, rand, randn, end, zeros, ones, eye, gallery,
sparse, full, size, length

2
Exercise 1: Create the matrix
A =
0 0 0 0 0
0 0 0 0 0
5 4 3 2 1
with a single line of MATLAB code. Then assign random numbers to its 2nd row. Then
assign the 1st row such that all columns sum to 0. Then reorder the columns reversely
and transpose the matrix. (4 commands)

Exercise 2: Create the matrix

B =
1 2 3 4 5
-1 -2 -3 -4 -5
1 2 3 4 5
-1 -2 -3 -4 -5
1 2 3 4 5
with a single line of MATLAB code.

Exercise 3: When accessing elements of a matrix with just a single index, the elements
will be addressed columnwise starting with the first column. For example, B(6) with
the above 5 × 5 matrix will return 2, the element in the (1, 2) position. Create a sparse
zero matrix C of size n × n via
n = 4;
C = sparse (n , n );
and populate its entries with 3 lines of MATLAB code such that
full ( C )
ans =
2 -1 0 0
-1 2 -1 0
0 -1 2 -1
0 0 -1 2
Do you recognize this matrix?

3 Expressions
In MATLAB numbers are just 1 × 1 matrices. There are some special numbers like i, pi,
eps, realmin, realmax, Inf, NaN (the latter two are not case-sensitive). MATLAB allows
to redefine these numbers as variables, which can cause confusion. So it is better to avoid
such variable names. Nevertheless, it is good practice to write 1i for the imaginary unit,
as this will give the expected result even if i has been redefined. Variables do not need

3
to be declared; they are instantiated with their first definition using the assignment
operator = .

Example 4:
rho = 1 .6180 ;

a = abs (3+4 i )
a =
5

z = sqrt ( besselk (4/3 , rho -1 i ))

z =
0 .3730 + 0 .3214i

huge = exp ( log ( realmax ))

huge =
1 .7977e +308

toobig = pi * huge
toobig =
Inf
Note how the final ; in the first line suppresses the output. In order to refer to the
last output use the ans variable.

4 Basic Linear Algebra commands

MATLAB is a very powerful tool for Linear Algebra calculations. Matrices (and vectors)
can be added and subtracted (+ and -) and multiplied (*). Remember that matrix–
matrix multiplication is different from element-wise multiplication (.*)!
One of the most important commands is the backslash \ (or equivalently, mldivide)
for the solution of linear systems or least squares problems.

Other commands: cond, rank, null, norm

Exercise 5: Create a magic square via A = magic(5) and a vector b = ones(5,1).

Solve the linear system Ax = b, obtaining the numerical solution xe. What do you notice
about the solution vector xe and can you explain it? What is the exact solution x? Check
numerically if the following inequalities are satisfied:

1 kb − Ax
ek2 kx − x
ek2 kb − Axe k2
≤ ≤ cond2 (A) .
cond2 (A) kbk2 kxk2 kbk2

More commands: qr, chol, schur, eig, eigs, svd, svds, fft

4
Exercise 6: Familiarize yourself with the eig command and compute the eigenvectors
and eigenvalues of the matrix A from Exercise 5. The Perron theorem implies that all
eigenvalues λj of a matrix with positive entries and equal row sums σ satisfy |λj | ≤ σ,
and that there is exactly one eigenvalue λk = σ. Verify this numerically.

5 Graphics
It is very easy to generate nice plots and animations with MATLAB. There are many
functions for 2D/3D plots and images and we will only list a few of them.

Commands: figure, close, hold, spy, plot, semilogx, semilogx, loglog,

bar, hist, meshgrid, contour, contourf, surf, pcolor, image, imagesc

Example 7: Plot the function f (x) = sin(x) and its zeros on [0, 2π].
figure (1)
x = linspace (0 ,2* pi ,200); % 200 typically enough for plotting
fx = sin ( x );
plot (x , fx , 'b - ') % this plots a ( b ) lue line ( -)
hold on
plot ([0 , pi ,2* pi ] ,[0 ,0 ,0] , ' ro ') % this plots ( r ) ed circles ( o )
There are also commands for changing the axes, and adding legends or titles. Some
of the most useful ones are listed below.

Commands: axis, xlim, ylim, xlabel, ylabel, legend, title, grid

Example 8: Following the above example, we can make our plot more descriptive.
figure (1)
xlabel ( 'x ' ); ylabel ( 'f ( x ) ' );
legend ( ' sin ( x ) ' , ' zeros ')
title ( 'a simple graph ')
axis ([0 ,2* pi , -1 .2 ,1 .2 ])
grid on
For generating a surface plot of a function f (x, y) we need to create a parametrization
of its (x, y) domain. This can be done conveniently using the meshgrid command.

Example 9: Plot | sin(x + iy)| over the domain [−1, 7] × [−1, 1].
figure (2)
x = linspace ( -1 ,7 ,200); y = linspace ( -1 ,1 ,200);
[X , Y ] = meshgrid (x , y );
Z = X + 1i*Y;
F = sin ( Z );

5
surf (X ,Y , abs ( F ) , ' LineStyle ' , ' none ')
axis tight
colorbar

6 Programming Scripts
MATLAB offers a very intuitive and clear programming language. Programs are typi-
cally written as m-files in the editor. When working with MATLAB interactively (as we
have done so far), it is recommended to type everything which requires more than one
line in the editor. This allows one to modify and debug the code very efficiently. Saved
m-files can be called by their name from the MATLAB command line or by pressing
Ctrl+Enter.

Keywords: for, while, if else, continue, break, switch case

Logical: ==, &&, ||, ~, all, any, isempty

Example 10: Fibonacci sequence (fibonacci.m)

% % clear workspace and close figures
close all ; clear all ; clc

% % initialize variable f to store Fibonacci numbers

f = zeros (20 ,1); f (1) = 0; f (2) = 1;

% % Fibonacci sequence
for j = 3: length ( f )
f ( j ) = f (j -2) + f (j -1);
if f ( j ) <= 10
str = sprintf ( ' %2 drd Fibonacci number is smaller / equal to 10 ' ,j );
else
str = sprintf ( ' %2 drd Fibonacci number is greater than 10 ' ,j );
end
if j == 19 , str = [ str ' --- puh , almost done... ' ]; end
disp ( str )
end

% % semilogy plot of sequence and golden section growth

semilogy ( f ); hold on
q = (1 + sqrt (5))/2;
semilogy ( q. ^(1: length ( f )) , 'r - - ')
xlabel ( 'n ')
legend ( ' Fibonacci sequence f_n ' , '1 .6180 ^ n ' , ' Location ' , ' NorthWest ')

6
Exercise 11: Use meshgrid to create a grid of complex numbers z = x + iy with
x ∈ [−2, 1] and y ∈ [−1, 1]. For each number z in this grid run 200 iterations of the
Mandelbrot recursion rn+1 = rn2 + z starting with r0 = 0. Use the pcolor function to
visualize |r200 | for each number z. Challenge for MATLAB gurus: no more than 10 lines!

7 Functions
Essentially, MATLAB functions are m-files that accept input arguments and produce
outputs. In its simplest form, a function is defined in an m-file with the same name.

Example 12: If this is the content of an m-file with name quadroots.m:

function [ z1 , z2 ] = quadroots (p , q )
% QUADROOTS Compute the solutions of z ^2 + p * z + q = 0 .
% Given the scalar inputs p , q this functions returns
% the roots z1 and z2 of a quadratic equation.
z1 = -p /2 - sqrt ( p ^2/4 - q );
z2 = -p /2 + sqrt ( p ^2/4 - q );
return % not necessarily required
end
then we can call it by its name, e.g., like
z1 = quadroots (0 ,1); % z2 will be ignored
[ z1 , z2 ] = quadroots (0 ,1);
Note that help quadroots will output the comments following the function keyword.
Many MATLAB functions are implemented as m-files and can be viewed by using the
edit command. For example, edit magic will open the implementation of the function
we used above for generating magic squares.
Another powerful construct are so-called inline functions, which are single-line func-
tions returning a single output. They are defined using the @ operator. For example, by
grouping the two roots z1 , z2 in a two-dimensional vector, we can rewrite quadroots as
quadroots1 = @ (p , q ) [ -p /2 - sqrt ( p ^2/4 - q ) , -p /2+ sqrt ( p ^2/4 - q ) ];
Or even shorter:
quadroots2 = @ (p , q ) -p /2 + [ -1 ,1]* sqrt ( p ^2/4 - q );

8 Numerical Integration and Differential Equations

MATLAB provides many routines for standard tasks in computing, ranging from el-
ementary math operations, over linear algebra, statistics and random numbers, inter-
polation, optimization, Fourier analysis and filtering, sparse matrix computation, to
computational geometry. More information can be found, as always, in the MATLAB

7
documentation (type doc). In this section we will discuss with the help of two examples
some functions for numerical integration and the solution of differential equations.
There are a number of routines for calculation definite integrals ab f (x) dx and their
R

multi-dimensional extensions. These MATLAB function require as their first input a

function handle to f (x) like, e.g., an inline function as discussed above. Note that this
function should be vectorized, i.e., it should be evaluatable for vector arguments x.

Commands: integral, integral2, integral3 (see also quad)

Example 13: Integrate f (x) = exp(−x2 ) log(x)2 from 0 to +∞:

f = @ ( x ) exp ( - x. ^2) . * log ( x ) . ^2; % vectorized !
q = integral (f ,0 , Inf , ' AbsTol ' ,1e -6 , ' RelTol ' ,1e -6)

MATLAB also provides a collection of integrators for ordinary differential equations.

Commands: ode45, ode15s, ode23, odeset, odeexamples

Exercise 14: Solve the 1D heat equation on an interval, discretized in space by second-
order finite differences, and in time by MATLAB’s ode15s. More precisely, solve

ut (t, x) = uxx (t, x), x ∈ [0, L], t ∈ [0, T ],

with given initial and boundary data

u(t = 0, x) = b(x), u(t, x = 0) = 0, u0 (t, x = L) = 0.

Let us define equispaced spatial grid points xj = jh, h = L/n, j = 1, . . . , n. A finite

difference approximation of the second derivative of a function u(x) at x = xj is

uj+1 − 2uj + uj−1

u00 (xj ) ≈ , j = 1, . . . , n − 1,
h2
with uj denoting approximations to u(xj ) and the convention that u0 = 0. For x = xn
we can use the approximation
−2uj + 2uj−1
u00 (xn ) ≈ .
h2
1. Define variables L = 1, T = 1, n = 100, and an inline function b(x) = x(L − x)2 .

2. Create a column vector xx of the grid points [x1 , x2 , . . . , xn ]T using the colon
operator : . Define a vector u0 of the initial data, [b(x1 ), b(x2 ), . . . , b(xn )]T . (If the
inline function for b(x) is vectorized this can be done via u0 = b(xx).)

8
 3. Create the n × n matrix A which maps column vectors u to the finite difference ap-
proximations of the second derivative (i.e., Au will be an approximation for u00 (x) at
the grid points x1 , x2 , . . . , xn ). You can use the command gallery('tridiag',n),
which will generate a sparse tridiagonal matrix
 
2 −1
 .. 
 −1 2 . 
 ∈ Rn×n .
 

.. ..

 . . −1 

−1 2

4. Define the inline function f = @(t,u) A*u and a vector tt of 100 equidistant time
points in [0, T ]. Then call
[ tt , uu ] = ode15s (f , tt , u0 );
figure
for j = 1: length ( tt )
plot ( xx , uu (j ,:));
title ([ 't = ' num2str ( tt ( j ))])
axis ([0 , L ,0 , norm ( u0 , inf )])
shg ; pause (0 .2 )
end

This will create an animation of the solution of the heat equation.

5. You may also want to try other MATLAB integrators like, e.g., ode45, and compare
which one works best for this problem.

9 Useful tools and tricks

Timings and profiler: The pause command will pause the program execution until a key
is pressed. With pause(1) one can pause a program for 1 second. The commands tic and
toc can be used to measure time that passed in between. Example: tic; pause(1); toc
The MATLAB profiler is an extremely valuable tool for speeding up complex programs,
as it allows one to identify the computationally most intensive parts of a code. It is
activated via profile on. After program execution one types profile report to obtain
the report. The profiler is cleared or deactivated via profile clear or profile off,
respectively.

Vectorization and memory allocation: Operations on arrays should be vectorized

whenever possible. Somewhat oversimplified this boils down to “avoiding for loops”,
because these can extremely slow down your code. Also, it is important to allocate
memory for large arrays in advance, as a dynamic resize may cause noticeable overhead.

9
Example 15: Compute the square root of 5 · 106 random numbers.
clear all
v = rand (5 e6 ,1);

tic ; % no memory allocation and no vectorization

for j = 1: length ( v )
s1 ( j ) = sqrt ( v ( j ));
end
t1 = toc ;

tic ; % with memory allocation but without vectorization

s2 = zeros ( size ( v ));
for j = 1: length ( v )
s2 ( j ) = sqrt ( v ( j ));
end
t2 = toc ;

tic ; % with memory allocation and vectorization

s3 = sqrt ( v );
t3 = toc ;

% compare the timings

bar ([ t1 , t2 , t3 ])

Exercise 16: Run the above code through the profiler and find the line where most of
computation time is spent on.

Publishing MATLAB code: When publishing MATLAB code in LATEX inside the
verbatim environment, it is recommended to include the package upquote to make sure
all quotes ' can be copied-and-pasted correctly into the MATLAB command window
(this is what one gets without this package: ’ ).
Another option for publishing MATLAB in LATEX is the mcode package, which high-
lights keywords, comments, and strings in different colors (as is done in this document).

Graphics workflow: When writing a paper or thesis, often a lot of time is spent on
generating graphics. It is therefore worth spending some time in advance for optimizing
the workflow of including figures in your work. I recommend to generate all graphics with
a dedicated script and to include these directly into the LATEX file without generating
copies. (Warning: multiple instances of the same file may cause severe headaches!)

10
10 Assignment
Task: Implement and test the Runge-Kutta method and numerically solve an ordinary
differential equation. Make sure to comment your work by adding MATLAB comments.

1. Copy and paste the following MATLAB code into an m-file called rk4.m:
function [ tt , yy ] = rk4 (f , tt , y0 )
% RK4 Solve y ' = f (t , y ) with initial value y0 using Runge - Kutta.
% The argument f is a function handle with input parameters (t , y ) .
% The vector tt contains the time points at which to solve the ODE.
% The row vector y0 is the initial value at tt (1) .
% Each column of the output matrix yy corresponds to the Runge - Kutta
% approximation of y at each time point in tt.

% ... ADD YOUR CODE HERE ...

end

2. Modify rk4.m for implementing the classical Runge-Kutta method defined via
k1 = f (tn , yn )
k2 = f (tn + h2 , yn + h2 k1 )
k3 = f (tn + h2 , yn + h2 k2 )
k4 = f (tn + h, yn + hk3 )
k1 + 2k2 + 2k3 + k4
yn+1 = yn + h ,
6
where at h = tn+1 − tn at each time step. (This can be done in less than 10 lines
of code, but doesn’t need to!)
3. Create a new m-file called test1.m to test your Runge-Kutta implementation. To
this end consider the simple initial value problem
y 0 (t) = −y, y(0) = 1, (1)
to be solved at equispaced time points t = 0, 0.1, 0.2, . . . , 2. Visually compare
your numerical solution ye(t) (the output yy) with the exact solution y(t) = e−t
for t ∈ [0, 2] by plotting both in the same figure. (Hint: Both curves should look
almost identical.)
4. Create a new m-file called test2.m to solve the same test problem (1) over the
time interval [0, 2] and measure the error of the computed solution ye(t) at t = 2
compared to the exact solution y(t) = e−t , i.e., evaluate err = |ye(2) − y(2)|. Do
this for n = 5, 10, 15, . . . , 50 equal time steps in [0, 2] and plot the error versus n.
It’s best to use a loglog plot for this. (Hint: You should see a straight line and
the error should decrease approximately like C/n4 for some constant C.)

11
5. Attach the three files rk4.m, test1.m, test2.m to an email and

send to [email protected] by November 13th, 2020.

You should receive a confirmation of submission via email until November 16th.
If you do not receive this confirmation, you need to get in contact immediately.

Note that the submission deadline is absolutely strict.

Aim to submit your work a week before that deadline.
Late submissions will not be accepted!

12